linux-pikaos-template/patches/cachyos/0001-cachyos-base-all.patch

From 2b7dc22b0a950292985c4d5118c5eeaa51ea2918 Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:06:09 +0200
Subject: [PATCH 1/8] aex-xts

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 arch/x86/Kconfig.assembler           |  10 +
 arch/x86/crypto/Makefile             |   3 +-
 arch/x86/crypto/aes-xts-avx-x86_64.S | 838 +++++++++++++++++++++++++++
 arch/x86/crypto/aesni-intel_glue.c   | 270 ++++++++-
 4 files changed, 1118 insertions(+), 3 deletions(-)
 create mode 100644 arch/x86/crypto/aes-xts-avx-x86_64.S

diff --git a/arch/x86/Kconfig.assembler b/arch/x86/Kconfig.assembler
index 8ad41da301e5..59aedf32c4ea 100644
--- a/arch/x86/Kconfig.assembler
+++ b/arch/x86/Kconfig.assembler
@@ -25,6 +25,16 @@ config AS_GFNI
 	help
 	  Supported by binutils >= 2.30 and LLVM integrated assembler
 
+config AS_VAES
+	def_bool $(as-instr,vaesenc %ymm0$(comma)%ymm1$(comma)%ymm2)
+	help
+	  Supported by binutils >= 2.30 and LLVM integrated assembler
+
+config AS_VPCLMULQDQ
+	def_bool $(as-instr,vpclmulqdq \$0x10$(comma)%ymm0$(comma)%ymm1$(comma)%ymm2)
+	help
+	  Supported by binutils >= 2.30 and LLVM integrated assembler
+
 config AS_WRUSS
 	def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
 	help
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index 9aa46093c91b..9c5ce5613738 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -48,7 +48,8 @@ chacha-x86_64-$(CONFIG_AS_AVX512) += chacha-avx512vl-x86_64.o
 
 obj-$(CONFIG_CRYPTO_AES_NI_INTEL) += aesni-intel.o
 aesni-intel-y := aesni-intel_asm.o aesni-intel_glue.o
-aesni-intel-$(CONFIG_64BIT) += aesni-intel_avx-x86_64.o aes_ctrby8_avx-x86_64.o
+aesni-intel-$(CONFIG_64BIT) += aesni-intel_avx-x86_64.o \
+	aes_ctrby8_avx-x86_64.o aes-xts-avx-x86_64.o
 
 obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
 sha1-ssse3-y := sha1_avx2_x86_64_asm.o sha1_ssse3_asm.o sha1_ssse3_glue.o
diff --git a/arch/x86/crypto/aes-xts-avx-x86_64.S b/arch/x86/crypto/aes-xts-avx-x86_64.S
new file mode 100644
index 000000000000..b8005d0205f8
--- /dev/null
+++ b/arch/x86/crypto/aes-xts-avx-x86_64.S
@@ -0,0 +1,838 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * AES-XTS for modern x86_64 CPUs
+ *
+ * Copyright 2024 Google LLC
+ *
+ * Author: Eric Biggers <ebiggers@google.com>
+ */
+
+/*
+ * This file implements AES-XTS for modern x86_64 CPUs.  To handle the
+ * complexities of coding for x86 SIMD, e.g. where every vector length needs
+ * different code, it uses a macro to generate several implementations that
+ * share similar source code but are targeted at different CPUs, listed below:
+ *
+ * AES-NI + AVX
+ *    - 128-bit vectors (1 AES block per vector)
+ *    - VEX-coded instructions
+ *    - xmm0-xmm15
+ *    - This is for older CPUs that lack VAES but do have AVX.
+ *
+ * VAES + VPCLMULQDQ + AVX2
+ *    - 256-bit vectors (2 AES blocks per vector)
+ *    - VEX-coded instructions
+ *    - ymm0-ymm15
+ *    - This is for CPUs that have VAES but lack AVX512 or AVX10,
+ *      e.g. Intel's Alder Lake and AMD's Zen 3.
+ *
+ * VAES + VPCLMULQDQ + AVX10/256 + BMI2
+ *    - 256-bit vectors (2 AES blocks per vector)
+ *    - EVEX-coded instructions
+ *    - ymm0-ymm31
+ *    - This is for CPUs that have AVX512 but where using zmm registers causes
+ *      downclocking, and for CPUs that have AVX10/256 but not AVX10/512.
+ *    - By "AVX10/256" we really mean (AVX512BW + AVX512VL) || AVX10/256.
+ *      To avoid confusion with 512-bit, we just write AVX10/256.
+ *
+ * VAES + VPCLMULQDQ + AVX10/512 + BMI2
+ *    - Same as the previous one, but upgrades to 512-bit vectors
+ *      (4 AES blocks per vector) in zmm0-zmm31.
+ *    - This is for CPUs that have good AVX512 or AVX10/512 support.
+ *
+ * This file doesn't have an implementation for AES-NI alone (without AVX), as
+ * the lack of VEX would make all the assembly code different.
+ *
+ * When we use VAES, we also use VPCLMULQDQ to parallelize the computation of
+ * the XTS tweaks.  This avoids a bottleneck.  Currently there don't seem to be
+ * any CPUs that support VAES but not VPCLMULQDQ.  If that changes, we might
+ * need to start also providing an implementation using VAES alone.
+ *
+ * The AES-XTS implementations in this file support everything required by the
+ * crypto API, including support for arbitrary input lengths and multi-part
+ * processing.  However, they are most heavily optimized for the common case of
+ * power-of-2 length inputs that are processed in a single part (disk sectors).
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+
+.section .rodata
+.p2align 4
+.Lgf_poly:
+	// The low 64 bits of this value represent the polynomial x^7 + x^2 + x
+	// + 1.  It is the value that must be XOR'd into the low 64 bits of the
+	// tweak each time a 1 is carried out of the high 64 bits.
+	//
+	// The high 64 bits of this value is just the internal carry bit that
+	// exists when there's a carry out of the low 64 bits of the tweak.
+	.quad	0x87, 1
+
+	// This table contains constants for vpshufb and vpblendvb, used to
+	// handle variable byte shifts and blending during ciphertext stealing
+	// on CPUs that don't support AVX10-style masking.
+.Lcts_permute_table:
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07
+	.byte	0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+.text
+
+// Function parameters
+.set	KEY,		%rdi	// Initially points to crypto_aes_ctx, then is
+				// advanced to point directly to the round keys
+.set	SRC,		%rsi	// Pointer to next source data
+.set	DST,		%rdx	// Pointer to next destination data
+.set	LEN,		%rcx	// Remaining length in bytes
+.set	TWEAK,		%r8	// Pointer to next tweak
+
+// %r9d holds the AES key length in bytes.
+.set	KEYLEN,		%r9d
+
+// %rax and %r10-r11 are available as temporaries.
+
+.macro	_define_Vi	i
+.if VL == 16
+	.set	V\i,		%xmm\i
+.elseif VL == 32
+	.set	V\i,		%ymm\i
+.elseif VL == 64
+	.set	V\i,		%zmm\i
+.else
+	.error "Unsupported Vector Length (VL)"
+.endif
+.endm
+
+.macro _define_aliases
+	// Define register aliases V0-V15, or V0-V31 if all 32 SIMD registers
+	// are available, that map to the xmm, ymm, or zmm registers according
+	// to the selected Vector Length (VL).
+	_define_Vi	0
+	_define_Vi	1
+	_define_Vi	2
+	_define_Vi	3
+	_define_Vi	4
+	_define_Vi	5
+	_define_Vi	6
+	_define_Vi	7
+	_define_Vi	8
+	_define_Vi	9
+	_define_Vi	10
+	_define_Vi	11
+	_define_Vi	12
+	_define_Vi	13
+	_define_Vi	14
+	_define_Vi	15
+.if USE_AVX10
+	_define_Vi	16
+	_define_Vi	17
+	_define_Vi	18
+	_define_Vi	19
+	_define_Vi	20
+	_define_Vi	21
+	_define_Vi	22
+	_define_Vi	23
+	_define_Vi	24
+	_define_Vi	25
+	_define_Vi	26
+	_define_Vi	27
+	_define_Vi	28
+	_define_Vi	29
+	_define_Vi	30
+	_define_Vi	31
+.endif
+
+	// V0-V3 hold the data blocks during the main loop, or temporary values
+	// otherwise.  V4-V5 hold temporary values.
+
+	// V6-V9 hold XTS tweaks.  Each 128-bit lane holds one tweak.
+	.set	TWEAK0_XMM,	%xmm6
+	.set	TWEAK0,		V6
+	.set	TWEAK1_XMM,	%xmm7
+	.set	TWEAK1,		V7
+	.set	TWEAK2,		V8
+	.set	TWEAK3,		V9
+
+	// V10-V13 are used for computing the next values of TWEAK[0-3].
+	.set	NEXT_TWEAK0,	V10
+	.set	NEXT_TWEAK1,	V11
+	.set	NEXT_TWEAK2,	V12
+	.set	NEXT_TWEAK3,	V13
+
+	// V14 holds the constant from .Lgf_poly, copied to all 128-bit lanes.
+	.set	GF_POLY_XMM,	%xmm14
+	.set	GF_POLY,	V14
+
+	// V15 holds the first AES round key, copied to all 128-bit lanes.
+	.set	KEY0_XMM,	%xmm15
+	.set	KEY0,		V15
+
+	// If 32 SIMD registers are available, then V16-V29 hold the remaining
+	// AES round keys, copied to all 128-bit lanes.
+.if USE_AVX10
+	.set	KEY1_XMM,	%xmm16
+	.set	KEY1,		V16
+	.set	KEY2_XMM,	%xmm17
+	.set	KEY2,		V17
+	.set	KEY3_XMM,	%xmm18
+	.set	KEY3,		V18
+	.set	KEY4_XMM,	%xmm19
+	.set	KEY4,		V19
+	.set	KEY5_XMM,	%xmm20
+	.set	KEY5,		V20
+	.set	KEY6_XMM,	%xmm21
+	.set	KEY6,		V21
+	.set	KEY7_XMM,	%xmm22
+	.set	KEY7,		V22
+	.set	KEY8_XMM,	%xmm23
+	.set	KEY8,		V23
+	.set	KEY9_XMM,	%xmm24
+	.set	KEY9,		V24
+	.set	KEY10_XMM,	%xmm25
+	.set	KEY10,		V25
+	.set	KEY11_XMM,	%xmm26
+	.set	KEY11,		V26
+	.set	KEY12_XMM,	%xmm27
+	.set	KEY12,		V27
+	.set	KEY13_XMM,	%xmm28
+	.set	KEY13,		V28
+	.set	KEY14_XMM,	%xmm29
+	.set	KEY14,		V29
+.endif
+	// V30-V31 are currently unused.
+.endm
+
+// Move a vector between memory and a register.
+.macro	_vmovdqu	src, dst
+.if VL < 64
+	vmovdqu		\src, \dst
+.else
+	vmovdqu8	\src, \dst
+.endif
+.endm
+
+// Broadcast a 128-bit value into a vector.
+.macro	_vbroadcast128	src, dst
+.if VL == 16 && !USE_AVX10
+	vmovdqu		\src, \dst
+.elseif VL == 32 && !USE_AVX10
+	vbroadcasti128	\src, \dst
+.else
+	vbroadcasti32x4	\src, \dst
+.endif
+.endm
+
+// XOR two vectors together.
+.macro	_vpxor	src1, src2, dst
+.if USE_AVX10
+	vpxord		\src1, \src2, \dst
+.else
+	vpxor		\src1, \src2, \dst
+.endif
+.endm
+
+// XOR three vectors together.
+.macro	_xor3	src1, src2, src3_and_dst
+.if USE_AVX10
+	// vpternlogd with immediate 0x96 is a three-argument XOR.
+	vpternlogd	$0x96, \src1, \src2, \src3_and_dst
+.else
+	vpxor		\src1, \src3_and_dst, \src3_and_dst
+	vpxor		\src2, \src3_and_dst, \src3_and_dst
+.endif
+.endm
+
+// Given a 128-bit XTS tweak in the xmm register \src, compute the next tweak
+// (by multiplying by the polynomial 'x') and write it to \dst.
+.macro	_next_tweak	src, tmp, dst
+	vpshufd		$0x13, \src, \tmp
+	vpaddq		\src, \src, \dst
+	vpsrad		$31, \tmp, \tmp
+	vpand		GF_POLY_XMM, \tmp, \tmp
+	vpxor		\tmp, \dst, \dst
+.endm
+
+// Given the XTS tweak(s) in the vector \src, compute the next vector of
+// tweak(s) (by multiplying by the polynomial 'x^(VL/16)') and write it to \dst.
+//
+// If VL > 16, then there are multiple tweaks, and we use vpclmulqdq to compute
+// all tweaks in the vector in parallel.  If VL=16, we just do the regular
+// computation without vpclmulqdq, as it's the faster method for a single tweak.
+.macro	_next_tweakvec	src, tmp1, tmp2, dst
+.if VL == 16
+	_next_tweak	\src, \tmp1, \dst
+.else
+	vpsrlq		$64 - VL/16, \src, \tmp1
+	vpclmulqdq	$0x01, GF_POLY, \tmp1, \tmp2
+	vpslldq		$8, \tmp1, \tmp1
+	vpsllq		$VL/16, \src, \dst
+	_xor3		\tmp1, \tmp2, \dst
+.endif
+.endm
+
+// Given the first XTS tweak at (TWEAK), compute the first set of tweaks and
+// store them in the vector registers TWEAK0-TWEAK3.  Clobbers V0-V5.
+.macro	_compute_first_set_of_tweaks
+	vmovdqu		(TWEAK), TWEAK0_XMM
+	_vbroadcast128	.Lgf_poly(%rip), GF_POLY
+.if VL == 16
+	// With VL=16, multiplying by x serially is fastest.
+	_next_tweak	TWEAK0, %xmm0, TWEAK1
+	_next_tweak	TWEAK1, %xmm0, TWEAK2
+	_next_tweak	TWEAK2, %xmm0, TWEAK3
+.else
+.if VL == 32
+	// Compute the second block of TWEAK0.
+	_next_tweak	TWEAK0_XMM, %xmm0, %xmm1
+	vinserti128	$1, %xmm1, TWEAK0, TWEAK0
+.elseif VL == 64
+	// Compute the remaining blocks of TWEAK0.
+	_next_tweak	TWEAK0_XMM, %xmm0, %xmm1
+	_next_tweak	%xmm1, %xmm0, %xmm2
+	_next_tweak	%xmm2, %xmm0, %xmm3
+	vinserti32x4	$1, %xmm1, TWEAK0, TWEAK0
+	vinserti32x4	$2, %xmm2, TWEAK0, TWEAK0
+	vinserti32x4	$3, %xmm3, TWEAK0, TWEAK0
+.endif
+	// Compute TWEAK[1-3] from TWEAK0.
+	vpsrlq		$64 - 1*VL/16, TWEAK0, V0
+	vpsrlq		$64 - 2*VL/16, TWEAK0, V2
+	vpsrlq		$64 - 3*VL/16, TWEAK0, V4
+	vpclmulqdq	$0x01, GF_POLY, V0, V1
+	vpclmulqdq	$0x01, GF_POLY, V2, V3
+	vpclmulqdq	$0x01, GF_POLY, V4, V5
+	vpslldq		$8, V0, V0
+	vpslldq		$8, V2, V2
+	vpslldq		$8, V4, V4
+	vpsllq		$1*VL/16, TWEAK0, TWEAK1
+	vpsllq		$2*VL/16, TWEAK0, TWEAK2
+	vpsllq		$3*VL/16, TWEAK0, TWEAK3
+.if USE_AVX10
+	vpternlogd	$0x96, V0, V1, TWEAK1
+	vpternlogd	$0x96, V2, V3, TWEAK2
+	vpternlogd	$0x96, V4, V5, TWEAK3
+.else
+	vpxor		V0, TWEAK1, TWEAK1
+	vpxor		V2, TWEAK2, TWEAK2
+	vpxor		V4, TWEAK3, TWEAK3
+	vpxor		V1, TWEAK1, TWEAK1
+	vpxor		V3, TWEAK2, TWEAK2
+	vpxor		V5, TWEAK3, TWEAK3
+.endif
+.endif
+.endm
+
+// Do one step in computing the next set of tweaks using the method of just
+// multiplying by x repeatedly (the same method _next_tweak uses).
+.macro	_tweak_step_mulx	i
+.if \i == 0
+	.set PREV_TWEAK, TWEAK3
+	.set NEXT_TWEAK, NEXT_TWEAK0
+.elseif \i == 5
+	.set PREV_TWEAK, NEXT_TWEAK0
+	.set NEXT_TWEAK, NEXT_TWEAK1
+.elseif \i == 10
+	.set PREV_TWEAK, NEXT_TWEAK1
+	.set NEXT_TWEAK, NEXT_TWEAK2
+.elseif \i == 15
+	.set PREV_TWEAK, NEXT_TWEAK2
+	.set NEXT_TWEAK, NEXT_TWEAK3
+.endif
+.if \i < 20 && \i % 5 == 0
+	vpshufd		$0x13, PREV_TWEAK, V5
+.elseif \i < 20 && \i % 5 == 1
+	vpaddq		PREV_TWEAK, PREV_TWEAK, NEXT_TWEAK
+.elseif \i < 20 && \i % 5 == 2
+	vpsrad		$31, V5, V5
+.elseif \i < 20 && \i % 5 == 3
+	vpand		GF_POLY, V5, V5
+.elseif \i < 20 && \i % 5 == 4
+	vpxor		V5, NEXT_TWEAK, NEXT_TWEAK
+.elseif \i == 1000
+	vmovdqa		NEXT_TWEAK0, TWEAK0
+	vmovdqa		NEXT_TWEAK1, TWEAK1
+	vmovdqa		NEXT_TWEAK2, TWEAK2
+	vmovdqa		NEXT_TWEAK3, TWEAK3
+.endif
+.endm
+
+// Do one step in computing the next set of tweaks using the VPCLMULQDQ method
+// (the same method _next_tweakvec uses for VL > 16).  This means multiplying
+// each tweak by x^(4*VL/16) independently.  Since 4*VL/16 is a multiple of 8
+// when VL > 16 (which it is here), the needed shift amounts are byte-aligned,
+// which allows the use of vpsrldq and vpslldq to do 128-bit wide shifts.
+.macro	_tweak_step_pclmul	i
+.if \i == 2
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK0, NEXT_TWEAK0
+.elseif \i == 4
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK1, NEXT_TWEAK1
+.elseif \i == 6
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK2, NEXT_TWEAK2
+.elseif \i == 8
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK3, NEXT_TWEAK3
+.elseif \i == 10
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK0, NEXT_TWEAK0
+.elseif \i == 12
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK1, NEXT_TWEAK1
+.elseif \i == 14
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK2, NEXT_TWEAK2
+.elseif \i == 16
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK3, NEXT_TWEAK3
+.elseif \i == 1000
+	vpslldq		$(4*VL/16) / 8, TWEAK0, TWEAK0
+	vpslldq		$(4*VL/16) / 8, TWEAK1, TWEAK1
+	vpslldq		$(4*VL/16) / 8, TWEAK2, TWEAK2
+	vpslldq		$(4*VL/16) / 8, TWEAK3, TWEAK3
+	_vpxor		NEXT_TWEAK0, TWEAK0, TWEAK0
+	_vpxor		NEXT_TWEAK1, TWEAK1, TWEAK1
+	_vpxor		NEXT_TWEAK2, TWEAK2, TWEAK2
+	_vpxor		NEXT_TWEAK3, TWEAK3, TWEAK3
+.endif
+.endm
+
+// _tweak_step does one step of the computation of the next set of tweaks from
+// TWEAK[0-3].  To complete all steps, this must be invoked with \i values 0
+// through at least 19, then 1000 which signals the last step.
+//
+// This is used to interleave the computation of the next set of tweaks with the
+// AES en/decryptions, which increases performance in some cases.
+.macro	_tweak_step	i
+.if VL == 16
+	_tweak_step_mulx	\i
+.else
+	_tweak_step_pclmul	\i
+.endif
+.endm
+
+// Load the round keys: just the first one if !USE_AVX10, otherwise all of them.
+.macro	_load_round_keys
+	_vbroadcast128	0*16(KEY), KEY0
+.if USE_AVX10
+	_vbroadcast128	1*16(KEY), KEY1
+	_vbroadcast128	2*16(KEY), KEY2
+	_vbroadcast128	3*16(KEY), KEY3
+	_vbroadcast128	4*16(KEY), KEY4
+	_vbroadcast128	5*16(KEY), KEY5
+	_vbroadcast128	6*16(KEY), KEY6
+	_vbroadcast128	7*16(KEY), KEY7
+	_vbroadcast128	8*16(KEY), KEY8
+	_vbroadcast128	9*16(KEY), KEY9
+	_vbroadcast128	10*16(KEY), KEY10
+	// Note: if it's AES-128 or AES-192, the last several round keys won't
+	// be used.  We do the loads anyway to save a conditional jump.
+	_vbroadcast128	11*16(KEY), KEY11
+	_vbroadcast128	12*16(KEY), KEY12
+	_vbroadcast128	13*16(KEY), KEY13
+	_vbroadcast128	14*16(KEY), KEY14
+.endif
+.endm
+
+// Do a single round of AES encryption (if \enc==1) or decryption (if \enc==0)
+// on the block(s) in \data using the round key(s) in \key.  The register length
+// determines the number of AES blocks en/decrypted.
+.macro	_vaes	enc, last, key, data
+.if \enc
+.if \last
+	vaesenclast	\key, \data, \data
+.else
+	vaesenc		\key, \data, \data
+.endif
+.else
+.if \last
+	vaesdeclast	\key, \data, \data
+.else
+	vaesdec		\key, \data, \data
+.endif
+.endif
+.endm
+
+// Do a single round of AES en/decryption on the block(s) in \data, using the
+// same key for all block(s).  The round key is loaded from the appropriate
+// register or memory location for round \i.  May clobber V4.
+.macro _vaes_1x		enc, last, i, xmm_suffix, data
+.if USE_AVX10
+	_vaes		\enc, \last, KEY\i\xmm_suffix, \data
+.else
+.ifnb \xmm_suffix
+	_vaes		\enc, \last, \i*16(KEY), \data
+.else
+	_vbroadcast128	\i*16(KEY), V4
+	_vaes		\enc, \last, V4, \data
+.endif
+.endif
+.endm
+
+// Do a single round of AES en/decryption on the blocks in registers V0-V3,
+// using the same key for all blocks.  The round key is loaded from the
+// appropriate register or memory location for round \i.  In addition, does step
+// \i of the computation of the next set of tweaks.  May clobber V4.
+.macro	_vaes_4x	enc, last, i
+.if USE_AVX10
+	_tweak_step	(2*(\i-1))
+	_vaes		\enc, \last, KEY\i, V0
+	_vaes		\enc, \last, KEY\i, V1
+	_tweak_step	(2*(\i-1) + 1)
+	_vaes		\enc, \last, KEY\i, V2
+	_vaes		\enc, \last, KEY\i, V3
+.else
+	_vbroadcast128	\i*16(KEY), V4
+	_tweak_step	(2*(\i-1))
+	_vaes		\enc, \last, V4, V0
+	_vaes		\enc, \last, V4, V1
+	_tweak_step	(2*(\i-1) + 1)
+	_vaes		\enc, \last, V4, V2
+	_vaes		\enc, \last, V4, V3
+.endif
+.endm
+
+// Do tweaked AES en/decryption (i.e., XOR with \tweak, then AES en/decrypt,
+// then XOR with \tweak again) of the block(s) in \data.  To process a single
+// block, use xmm registers and set \xmm_suffix=_XMM.  To process a vector of
+// length VL, use V* registers and leave \xmm_suffix empty.  May clobber V4.
+.macro	_aes_crypt	enc, xmm_suffix, tweak, data
+	_xor3		KEY0\xmm_suffix, \tweak, \data
+	_vaes_1x	\enc, 0, 1, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 2, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 3, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 4, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 5, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 6, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 7, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 8, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 9, \xmm_suffix, \data
+	cmp		$24, KEYLEN
+	jle		.Laes_128_or_192\@
+	_vaes_1x	\enc, 0, 10, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 11, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 12, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 13, \xmm_suffix, \data
+	_vaes_1x	\enc, 1, 14, \xmm_suffix, \data
+	jmp		.Laes_done\@
+.Laes_128_or_192\@:
+	je		.Laes_192\@
+	_vaes_1x	\enc, 1, 10, \xmm_suffix, \data
+	jmp		.Laes_done\@
+.Laes_192\@:
+	_vaes_1x	\enc, 0, 10, \xmm_suffix, \data
+	_vaes_1x	\enc, 0, 11, \xmm_suffix, \data
+	_vaes_1x	\enc, 1, 12, \xmm_suffix, \data
+.Laes_done\@:
+	_vpxor		\tweak, \data, \data
+.endm
+
+.macro	_aes_xts_crypt	enc
+	_define_aliases
+
+	// Load the AES key length: 16 (AES-128), 24 (AES-192), or 32 (AES-256).
+	movl		480(KEY), KEYLEN
+
+	// If decrypting, advance KEY to the decryption round keys.
+.if !\enc
+	add		$240, KEY
+.endif
+
+	// Check whether the data length is a multiple of the AES block length.
+	test		$15, LEN
+	jnz		.Lneed_cts\@
+.Lxts_init\@:
+
+	// Cache as many round keys as possible.
+	_load_round_keys
+
+	// Compute the first set of tweaks TWEAK[0-3].
+	_compute_first_set_of_tweaks
+
+	sub		$4*VL, LEN
+	jl		.Lhandle_remainder\@
+
+.Lmain_loop\@:
+	// This is the main loop, en/decrypting 4*VL bytes per iteration.
+
+	// XOR each source block with its tweak and the first round key.
+.if USE_AVX10
+	vmovdqu8	0*VL(SRC), V0
+	vmovdqu8	1*VL(SRC), V1
+	vmovdqu8	2*VL(SRC), V2
+	vmovdqu8	3*VL(SRC), V3
+	vpternlogd	$0x96, TWEAK0, KEY0, V0
+	vpternlogd	$0x96, TWEAK1, KEY0, V1
+	vpternlogd	$0x96, TWEAK2, KEY0, V2
+	vpternlogd	$0x96, TWEAK3, KEY0, V3
+.else
+	vpxor		0*VL(SRC), KEY0, V0
+	vpxor		1*VL(SRC), KEY0, V1
+	vpxor		2*VL(SRC), KEY0, V2
+	vpxor		3*VL(SRC), KEY0, V3
+	vpxor		TWEAK0, V0, V0
+	vpxor		TWEAK1, V1, V1
+	vpxor		TWEAK2, V2, V2
+	vpxor		TWEAK3, V3, V3
+.endif
+	// Do all the AES rounds on the data blocks, interleaved with
+	// the computation of the next set of tweaks.
+	_vaes_4x	\enc, 0, 1
+	_vaes_4x	\enc, 0, 2
+	_vaes_4x	\enc, 0, 3
+	_vaes_4x	\enc, 0, 4
+	_vaes_4x	\enc, 0, 5
+	_vaes_4x	\enc, 0, 6
+	_vaes_4x	\enc, 0, 7
+	_vaes_4x	\enc, 0, 8
+	_vaes_4x	\enc, 0, 9
+	// Try to optimize for AES-256 by keeping the code for AES-128 and
+	// AES-192 out-of-line.
+	cmp		$24, KEYLEN
+	jle		.Lencrypt_4x_aes_128_or_192\@
+	_vaes_4x	\enc, 0, 10
+	_vaes_4x	\enc, 0, 11
+	_vaes_4x	\enc, 0, 12
+	_vaes_4x	\enc, 0, 13
+	_vaes_4x	\enc, 1, 14
+.Lencrypt_4x_done\@:
+
+	// XOR in the tweaks again.
+	_vpxor		TWEAK0, V0, V0
+	_vpxor		TWEAK1, V1, V1
+	_vpxor		TWEAK2, V2, V2
+	_vpxor		TWEAK3, V3, V3
+
+	// Store the destination blocks.
+	_vmovdqu	V0, 0*VL(DST)
+	_vmovdqu	V1, 1*VL(DST)
+	_vmovdqu	V2, 2*VL(DST)
+	_vmovdqu	V3, 3*VL(DST)
+
+	// Finish computing the next set of tweaks.
+	_tweak_step	1000
+
+	add		$4*VL, SRC
+	add		$4*VL, DST
+	sub		$4*VL, LEN
+	jge		.Lmain_loop\@
+
+	// Check for the uncommon case where the data length isn't a multiple of
+	// 4*VL.  Handle it out-of-line in order to optimize for the common
+	// case.  In the common case, just fall through to the ret.
+	test		$4*VL-1, LEN
+	jnz		.Lhandle_remainder\@
+.Ldone\@:
+	// Store the next tweak back to *TWEAK to support continuation calls.
+	vmovdqu		TWEAK0_XMM, (TWEAK)
+.if VL > 16
+	vzeroupper
+.endif
+	RET
+
+.Lhandle_remainder\@:
+	add		$4*VL, LEN	// Undo the extra sub from earlier.
+
+	// En/decrypt any remaining full blocks, one vector at a time.
+.if VL > 16
+	sub		$VL, LEN
+	jl		.Lvec_at_a_time_done\@
+.Lvec_at_a_time\@:
+	_vmovdqu	(SRC), V0
+	_aes_crypt	\enc, , TWEAK0, V0
+	_vmovdqu	V0, (DST)
+	_next_tweakvec	TWEAK0, V0, V1, TWEAK0
+	add		$VL, SRC
+	add		$VL, DST
+	sub		$VL, LEN
+	jge		.Lvec_at_a_time\@
+.Lvec_at_a_time_done\@:
+	add		$VL-16, LEN	// Undo the extra sub from earlier.
+.else
+	sub		$16, LEN
+.endif
+
+	// En/decrypt any remaining full blocks, one at a time.
+	jl		.Lblock_at_a_time_done\@
+.Lblock_at_a_time\@:
+	vmovdqu		(SRC), %xmm0
+	_aes_crypt	\enc, _XMM, TWEAK0_XMM, %xmm0
+	vmovdqu		%xmm0, (DST)
+	_next_tweak	TWEAK0_XMM, %xmm0, TWEAK0_XMM
+	add		$16, SRC
+	add		$16, DST
+	sub		$16, LEN
+	jge		.Lblock_at_a_time\@
+.Lblock_at_a_time_done\@:
+	add		$16, LEN	// Undo the extra sub from earlier.
+
+.Lfull_blocks_done\@:
+	// Now 0 <= LEN <= 15.  If LEN is nonzero, do ciphertext stealing to
+	// process the last 16 + LEN bytes.  If LEN is zero, we're done.
+	test		LEN, LEN
+	jnz		.Lcts\@
+	jmp		.Ldone\@
+
+	// Out-of-line handling of AES-128 and AES-192
+.Lencrypt_4x_aes_128_or_192\@:
+	jz		.Lencrypt_4x_aes_192\@
+	_vaes_4x	\enc, 1, 10
+	jmp		.Lencrypt_4x_done\@
+.Lencrypt_4x_aes_192\@:
+	_vaes_4x	\enc, 0, 10
+	_vaes_4x	\enc, 0, 11
+	_vaes_4x	\enc, 1, 12
+	jmp		.Lencrypt_4x_done\@
+
+.Lneed_cts\@:
+	// The data length isn't a multiple of the AES block length, so
+	// ciphertext stealing (CTS) will be needed.  Subtract one block from
+	// LEN so that the main loop doesn't process the last full block.  The
+	// CTS step will process it specially along with the partial block.
+	sub		$16, LEN
+	jmp		.Lxts_init\@
+
+.Lcts\@:
+	// Do ciphertext stealing (CTS) to en/decrypt the last full block and
+	// the partial block.  CTS needs two tweaks.  TWEAK0_XMM contains the
+	// next tweak; compute the one after that.  Decryption uses these two
+	// tweaks in reverse order, so also define aliases to handle that.
+	_next_tweak	TWEAK0_XMM, %xmm0, TWEAK1_XMM
+.if \enc
+	.set		CTS_TWEAK0,	TWEAK0_XMM
+	.set		CTS_TWEAK1,	TWEAK1_XMM
+.else
+	.set		CTS_TWEAK0,	TWEAK1_XMM
+	.set		CTS_TWEAK1,	TWEAK0_XMM
+.endif
+
+	// En/decrypt the last full block.
+	vmovdqu		(SRC), %xmm0
+	_aes_crypt	\enc, _XMM, CTS_TWEAK0, %xmm0
+
+.if USE_AVX10
+	// Create a mask that has the first LEN bits set.
+	mov		$-1, %rax
+	bzhi		LEN, %rax, %rax
+	kmovq		%rax, %k1
+
+	// Swap the first LEN bytes of the above result with the partial block.
+	// Note that to support in-place en/decryption, the load from the src
+	// partial block must happen before the store to the dst partial block.
+	vmovdqa		%xmm0, %xmm1
+	vmovdqu8	16(SRC), %xmm0{%k1}
+	vmovdqu8	%xmm1, 16(DST){%k1}
+.else
+	lea		.Lcts_permute_table(%rip), %rax
+
+	// Load the src partial block, left-aligned.  Note that to support
+	// in-place en/decryption, this must happen before the store to the dst
+	// partial block.
+	vmovdqu		(SRC, LEN, 1), %xmm1
+
+	// Shift the first LEN bytes of the en/decryption of the last full block
+	// to the end of a register, then store it to DST+LEN.  This stores the
+	// dst partial block.  It also writes to the second part of the dst last
+	// full block, but that part is overwritten later.
+	vpshufb		(%rax, LEN, 1), %xmm0, %xmm2
+	vmovdqu		%xmm2, (DST, LEN, 1)
+
+	// Make xmm3 contain [16-LEN,16-LEN+1,...,14,15,0x80,0x80,...].
+	sub		LEN, %rax
+	vmovdqu		32(%rax), %xmm3
+
+	// Shift the src partial block to the beginning of its register.
+	vpshufb		%xmm3, %xmm1, %xmm1
+
+	// Do a blend to generate the src partial block followed by the second
+	// part of the en/decryption of the last full block.
+	vpblendvb	%xmm3, %xmm0, %xmm1, %xmm0
+.endif
+	// En/decrypt again and store the last full block.
+	_aes_crypt	\enc, _XMM, CTS_TWEAK1, %xmm0
+	vmovdqu		%xmm0, (DST)
+	jmp		.Ldone\@
+.endm
+
+// void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+//			   u8 iv[AES_BLOCK_SIZE]);
+SYM_FUNC_START(aes_xts_encrypt_iv)
+	vmovdqu		(%rsi), %xmm0
+	vpxor		0*16(%rdi), %xmm0, %xmm0
+	vaesenc		1*16(%rdi), %xmm0, %xmm0
+	vaesenc		2*16(%rdi), %xmm0, %xmm0
+	vaesenc		3*16(%rdi), %xmm0, %xmm0
+	vaesenc		4*16(%rdi), %xmm0, %xmm0
+	vaesenc		5*16(%rdi), %xmm0, %xmm0
+	vaesenc		6*16(%rdi), %xmm0, %xmm0
+	vaesenc		7*16(%rdi), %xmm0, %xmm0
+	vaesenc		8*16(%rdi), %xmm0, %xmm0
+	vaesenc		9*16(%rdi), %xmm0, %xmm0
+	cmpl		$24, 480(%rdi)
+	jle		.Lencrypt_iv_aes_128_or_192
+	vaesenc		10*16(%rdi), %xmm0, %xmm0
+	vaesenc		11*16(%rdi), %xmm0, %xmm0
+	vaesenc		12*16(%rdi), %xmm0, %xmm0
+	vaesenc		13*16(%rdi), %xmm0, %xmm0
+	vaesenclast	14*16(%rdi), %xmm0, %xmm0
+.Lencrypt_iv_done:
+	vmovdqu		%xmm0, (%rsi)
+	RET
+
+	// Out-of-line handling of AES-128 and AES-192
+.Lencrypt_iv_aes_128_or_192:
+	jz		.Lencrypt_iv_aes_192
+	vaesenclast	10*16(%rdi), %xmm0, %xmm0
+	jmp		.Lencrypt_iv_done
+.Lencrypt_iv_aes_192:
+	vaesenc		10*16(%rdi), %xmm0, %xmm0
+	vaesenc		11*16(%rdi), %xmm0, %xmm0
+	vaesenclast	12*16(%rdi), %xmm0, %xmm0
+	jmp		.Lencrypt_iv_done
+SYM_FUNC_END(aes_xts_encrypt_iv)
+
+// Below are the actual AES-XTS encryption and decryption functions,
+// instantiated from the above macro.  They all have the following prototype:
+//
+// void (*xts_asm_func)(const struct crypto_aes_ctx *key,
+//			const u8 *src, u8 *dst, size_t len,
+//			u8 tweak[AES_BLOCK_SIZE]);
+//
+// |key| is the data key.  |tweak| contains the next tweak; the encryption of
+// the original IV with the tweak key was already done.  This function supports
+// incremental computation, but |len| must always be >= 16 (AES_BLOCK_SIZE), and
+// |len| must be a multiple of 16 except on the last call.  If |len| is a
+// multiple of 16, then this function updates |tweak| to contain the next tweak.
+
+.set	VL, 16
+.set	USE_AVX10, 0
+SYM_TYPED_FUNC_START(aes_xts_encrypt_aesni_avx)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_aesni_avx)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_aesni_avx)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_aesni_avx)
+
+#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
+.set	VL, 32
+.set	USE_AVX10, 0
+SYM_TYPED_FUNC_START(aes_xts_encrypt_vaes_avx2)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_vaes_avx2)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_vaes_avx2)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_vaes_avx2)
+
+.set	VL, 32
+.set	USE_AVX10, 1
+SYM_TYPED_FUNC_START(aes_xts_encrypt_vaes_avx10_256)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_vaes_avx10_256)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_vaes_avx10_256)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_vaes_avx10_256)
+
+.set	VL, 64
+.set	USE_AVX10, 1
+SYM_TYPED_FUNC_START(aes_xts_encrypt_vaes_avx10_512)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_vaes_avx10_512)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_vaes_avx10_512)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_vaes_avx10_512)
+#endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index b1d90c25975a..0855ace8659c 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -1137,7 +1137,264 @@ static struct skcipher_alg aesni_xctr = {
 };
 
 static struct simd_skcipher_alg *aesni_simd_xctr;
-#endif /* CONFIG_X86_64 */
+
+asmlinkage void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+				   u8 iv[AES_BLOCK_SIZE]);
+
+typedef void (*xts_asm_func)(const struct crypto_aes_ctx *key,
+			     const u8 *src, u8 *dst, size_t len,
+			     u8 tweak[AES_BLOCK_SIZE]);
+
+/* This handles cases where the source and/or destination span pages. */
+static noinline int
+xts_crypt_slowpath(struct skcipher_request *req, xts_asm_func asm_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	int tail = req->cryptlen % AES_BLOCK_SIZE;
+	struct scatterlist sg_src[2], sg_dst[2];
+	struct skcipher_request subreq;
+	struct skcipher_walk walk;
+	struct scatterlist *src, *dst;
+	int err;
+
+	/*
+	 * If the message length isn't divisible by the AES block size, then
+	 * separate off the last full block and the partial block.  This ensures
+	 * that they are processed in the same call to the assembly function,
+	 * which is required for ciphertext stealing.
+	 */
+	if (tail) {
+		skcipher_request_set_tfm(&subreq, tfm);
+		skcipher_request_set_callback(&subreq,
+					      skcipher_request_flags(req),
+					      NULL, NULL);
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   req->cryptlen - tail - AES_BLOCK_SIZE,
+					   req->iv);
+		req = &subreq;
+	}
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while (walk.nbytes) {
+		unsigned int nbytes = walk.nbytes;
+
+		if (nbytes < walk.total)
+			nbytes = round_down(nbytes, AES_BLOCK_SIZE);
+
+		kernel_fpu_begin();
+		(*asm_func)(&ctx->crypt_ctx, walk.src.virt.addr,
+			    walk.dst.virt.addr, nbytes, req->iv);
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
+	}
+
+	if (err || !tail)
+		return err;
+
+	/* Do ciphertext stealing with the last full block and partial block. */
+
+	dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
+	if (req->dst != req->src)
+		dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
+
+	skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
+				   req->iv);
+
+	err = skcipher_walk_virt(&walk, req, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	(*asm_func)(&ctx->crypt_ctx, walk.src.virt.addr, walk.dst.virt.addr,
+		    walk.nbytes, req->iv);
+	kernel_fpu_end();
+
+	return skcipher_walk_done(&walk, 0);
+}
+
+/* __always_inline to avoid indirect call in fastpath */
+static __always_inline int
+xts_crypt2(struct skcipher_request *req, xts_asm_func asm_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	const unsigned int cryptlen = req->cryptlen;
+	struct scatterlist *src = req->src;
+	struct scatterlist *dst = req->dst;
+
+	if (unlikely(cryptlen < AES_BLOCK_SIZE))
+		return -EINVAL;
+
+	kernel_fpu_begin();
+	aes_xts_encrypt_iv(&ctx->tweak_ctx, req->iv);
+
+	/*
+	 * In practice, virtually all XTS plaintexts and ciphertexts are either
+	 * 512 or 4096 bytes, aligned such that they don't span page boundaries.
+	 * To optimize the performance of these cases, and also any other case
+	 * where no page boundary is spanned, the below fast-path handles
+	 * single-page sources and destinations as efficiently as possible.
+	 */
+	if (likely(src->length >= cryptlen && dst->length >= cryptlen &&
+		   src->offset + cryptlen <= PAGE_SIZE &&
+		   dst->offset + cryptlen <= PAGE_SIZE)) {
+		struct page *src_page = sg_page(src);
+		struct page *dst_page = sg_page(dst);
+		void *src_virt = kmap_local_page(src_page) + src->offset;
+		void *dst_virt = kmap_local_page(dst_page) + dst->offset;
+
+		(*asm_func)(&ctx->crypt_ctx, src_virt, dst_virt, cryptlen,
+			    req->iv);
+		kunmap_local(dst_virt);
+		kunmap_local(src_virt);
+		kernel_fpu_end();
+		return 0;
+	}
+	kernel_fpu_end();
+	return xts_crypt_slowpath(req, asm_func);
+}
+
+#define DEFINE_XTS_ALG(suffix, driver_name, priority)			       \
+									       \
+asmlinkage void aes_xts_encrypt_##suffix(const struct crypto_aes_ctx *key,     \
+					 const u8 *src, u8 *dst, size_t len,   \
+					 u8 tweak[AES_BLOCK_SIZE]);	       \
+asmlinkage void aes_xts_decrypt_##suffix(const struct crypto_aes_ctx *key,     \
+					 const u8 *src, u8 *dst, size_t len,   \
+					 u8 tweak[AES_BLOCK_SIZE]);	       \
+									       \
+static int xts_encrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt2(req, aes_xts_encrypt_##suffix);		       \
+}									       \
+									       \
+static int xts_decrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt2(req, aes_xts_decrypt_##suffix);		       \
+}									       \
+									       \
+static struct skcipher_alg aes_xts_alg_##suffix = {			       \
+	.base = {							       \
+		.cra_name		= "__xts(aes)",			       \
+		.cra_driver_name	= "__" driver_name,		       \
+		.cra_priority		= priority,			       \
+		.cra_flags		= CRYPTO_ALG_INTERNAL,		       \
+		.cra_blocksize		= AES_BLOCK_SIZE,		       \
+		.cra_ctxsize		= XTS_AES_CTX_SIZE,		       \
+		.cra_module		= THIS_MODULE,			       \
+	},								       \
+	.min_keysize	= 2 * AES_MIN_KEY_SIZE,				       \
+	.max_keysize	= 2 * AES_MAX_KEY_SIZE,				       \
+	.ivsize		= AES_BLOCK_SIZE,				       \
+	.walksize	= 2 * AES_BLOCK_SIZE,				       \
+	.setkey		= xts_aesni_setkey,				       \
+	.encrypt	= xts_encrypt_##suffix,				       \
+	.decrypt	= xts_decrypt_##suffix,				       \
+};									       \
+									       \
+static struct simd_skcipher_alg *aes_xts_simdalg_##suffix
+
+DEFINE_XTS_ALG(aesni_avx, "xts-aes-aesni-avx", 500);
+#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
+DEFINE_XTS_ALG(vaes_avx2, "xts-aes-vaes-avx2", 600);
+DEFINE_XTS_ALG(vaes_avx10_256, "xts-aes-vaes-avx10_256", 700);
+DEFINE_XTS_ALG(vaes_avx10_512, "xts-aes-vaes-avx10_512", 800);
+#endif
+
+/*
+ * This is a list of CPU models that are known to suffer from downclocking when
+ * zmm registers (512-bit vectors) are used.  On these CPUs, the AES-XTS
+ * implementation with zmm registers won't be used by default.  An
+ * implementation with ymm registers (256-bit vectors) will be used instead.
+ */
+static const struct x86_cpu_id zmm_exclusion_list[] = {
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_SKYLAKE_X },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_ICELAKE_X },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_ICELAKE_D },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_ICELAKE },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_ICELAKE_L },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_ICELAKE_NNPI },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_TIGERLAKE_L },
+	{ .vendor = X86_VENDOR_INTEL, .family = 6, .model = INTEL_FAM6_TIGERLAKE },
+	/* Allow Rocket Lake and later, and Sapphire Rapids and later. */
+	/* Also allow AMD CPUs (starting with Zen 4, the first with AVX-512). */
+	{},
+};
+
+static int __init register_xts_algs(void)
+{
+	int err;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX))
+		return 0;
+	err = simd_register_skciphers_compat(&aes_xts_alg_aesni_avx, 1,
+					     &aes_xts_simdalg_aesni_avx);
+	if (err)
+		return err;
+#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
+	if (!boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_VAES) ||
+	    !boot_cpu_has(X86_FEATURE_VPCLMULQDQ) ||
+	    !boot_cpu_has(X86_FEATURE_PCLMULQDQ) ||
+	    !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL))
+		return 0;
+	err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx2, 1,
+					     &aes_xts_simdalg_vaes_avx2);
+	if (err)
+		return err;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX512BW) ||
+	    !boot_cpu_has(X86_FEATURE_AVX512VL) ||
+	    !boot_cpu_has(X86_FEATURE_BMI2) ||
+	    !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
+			       XFEATURE_MASK_AVX512, NULL))
+		return 0;
+
+	err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_256, 1,
+					     &aes_xts_simdalg_vaes_avx10_256);
+	if (err)
+		return err;
+
+	if (x86_match_cpu(zmm_exclusion_list))
+		aes_xts_alg_vaes_avx10_512.base.cra_priority = 1;
+
+	err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_512, 1,
+					     &aes_xts_simdalg_vaes_avx10_512);
+	if (err)
+		return err;
+#endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
+	return 0;
+}
+
+static void unregister_xts_algs(void)
+{
+	if (aes_xts_simdalg_aesni_avx)
+		simd_unregister_skciphers(&aes_xts_alg_aesni_avx, 1,
+					  &aes_xts_simdalg_aesni_avx);
+#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
+	if (aes_xts_simdalg_vaes_avx2)
+		simd_unregister_skciphers(&aes_xts_alg_vaes_avx2, 1,
+					  &aes_xts_simdalg_vaes_avx2);
+	if (aes_xts_simdalg_vaes_avx10_256)
+		simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_256, 1,
+					  &aes_xts_simdalg_vaes_avx10_256);
+	if (aes_xts_simdalg_vaes_avx10_512)
+		simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_512, 1,
+					  &aes_xts_simdalg_vaes_avx10_512);
+#endif
+}
+#else /* CONFIG_X86_64 */
+static int __init register_xts_algs(void)
+{
+	return 0;
+}
+
+static void unregister_xts_algs(void)
+{
+}
+#endif /* !CONFIG_X86_64 */
 
 #ifdef CONFIG_X86_64
 static int generic_gcmaes_set_key(struct crypto_aead *aead, const u8 *key,
@@ -1276,13 +1533,21 @@ static int __init aesni_init(void)
 		goto unregister_aeads;
 #endif /* CONFIG_X86_64 */
 
+	err = register_xts_algs();
+	if (err)
+		goto unregister_xts;
+
 	return 0;
 
+unregister_xts:
+	unregister_xts_algs();
 #ifdef CONFIG_X86_64
+	if (aesni_simd_xctr)
+		simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
 unregister_aeads:
+#endif /* CONFIG_X86_64 */
 	simd_unregister_aeads(aesni_aeads, ARRAY_SIZE(aesni_aeads),
 				aesni_simd_aeads);
-#endif /* CONFIG_X86_64 */
 
 unregister_skciphers:
 	simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
@@ -1303,6 +1568,7 @@ static void __exit aesni_exit(void)
 	if (boot_cpu_has(X86_FEATURE_AVX))
 		simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
 #endif /* CONFIG_X86_64 */
+	unregister_xts_algs();
 }
 
 late_initcall(aesni_init);
-- 
2.44.0

From 4a47b09deb67c3854ac102bcb18ef0df00aae437 Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:06:20 +0200
Subject: [PATCH 2/8] amd-pstate

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 .../admin-guide/kernel-parameters.txt         |   5 +
 Documentation/admin-guide/pm/amd-pstate.rst   |  70 ++-
 arch/x86/Kconfig                              |   5 +-
 arch/x86/include/asm/msr-index.h              |   2 +
 drivers/acpi/cppc_acpi.c                      |  17 +-
 drivers/acpi/processor_driver.c               |   6 +
 drivers/cpufreq/acpi-cpufreq.c                |   2 -
 drivers/cpufreq/amd-pstate-ut.c               |   2 +-
 drivers/cpufreq/amd-pstate.c                  | 499 +++++++++++++++---
 include/acpi/cppc_acpi.h                      |   5 +
 include/linux/amd-pstate.h                    |  32 +-
 include/linux/cpufreq.h                       |   1 +
 12 files changed, 560 insertions(+), 86 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index d2150bd3acc5..71ed7f1b0f9b 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -374,6 +374,11 @@
 			  selects a performance level in this range and appropriate
 			  to the current workload.
 
+	amd_prefcore=
+			[X86]
+			disable
+			  Disable amd-pstate preferred core.
+
 	amijoy.map=	[HW,JOY] Amiga joystick support
 			Map of devices attached to JOY0DAT and JOY1DAT
 			Format: <a>,<b>
diff --git a/Documentation/admin-guide/pm/amd-pstate.rst b/Documentation/admin-guide/pm/amd-pstate.rst
index 9eb26014d34b..82fbd01da658 100644
--- a/Documentation/admin-guide/pm/amd-pstate.rst
+++ b/Documentation/admin-guide/pm/amd-pstate.rst
@@ -300,8 +300,8 @@ platforms. The AMD P-States mechanism is the more performance and energy
 efficiency frequency management method on AMD processors.
 
 
-AMD Pstate Driver Operation Modes
-=================================
+``amd-pstate`` Driver Operation Modes
+======================================
 
 ``amd_pstate`` CPPC has 3 operation modes: autonomous (active) mode,
 non-autonomous (passive) mode and guided autonomous (guided) mode.
@@ -353,6 +353,48 @@ is activated.  In this mode, driver requests minimum and maximum performance
 level and the platform autonomously selects a performance level in this range
 and appropriate to the current workload.
 
+``amd-pstate`` Preferred Core
+=================================
+
+The core frequency is subjected to the process variation in semiconductors.
+Not all cores are able to reach the maximum frequency respecting the
+infrastructure limits. Consequently, AMD has redefined the concept of
+maximum frequency of a part. This means that a fraction of cores can reach
+maximum frequency. To find the best process scheduling policy for a given
+scenario, OS needs to know the core ordering informed by the platform through
+highest performance capability register of the CPPC interface.
+
+``amd-pstate`` preferred core enables the scheduler to prefer scheduling on
+cores that can achieve a higher frequency with lower voltage. The preferred
+core rankings can dynamically change based on the workload, platform conditions,
+thermals and ageing.
+
+The priority metric will be initialized by the ``amd-pstate`` driver. The ``amd-pstate``
+driver will also determine whether or not ``amd-pstate`` preferred core is
+supported by the platform.
+
+``amd-pstate`` driver will provide an initial core ordering when the system boots.
+The platform uses the CPPC interfaces to communicate the core ranking to the
+operating system and scheduler to make sure that OS is choosing the cores
+with highest performance firstly for scheduling the process. When ``amd-pstate``
+driver receives a message with the highest performance change, it will
+update the core ranking and set the cpu's priority.
+
+``amd-pstate`` Preferred Core Switch
+=====================================
+Kernel Parameters
+-----------------
+
+``amd-pstate`` peferred core`` has two states: enable and disable.
+Enable/disable states can be chosen by different kernel parameters.
+Default enable ``amd-pstate`` preferred core.
+
+``amd_prefcore=disable``
+
+For systems that support ``amd-pstate`` preferred core, the core rankings will
+always be advertised by the platform. But OS can choose to ignore that via the
+kernel parameter ``amd_prefcore=disable``.
+
 User Space Interface in ``sysfs`` - General
 ===========================================
 
@@ -385,6 +427,30 @@ control its functionality at the system level.  They are located in the
         to the operation mode represented by that string - or to be
         unregistered in the "disable" case.
 
+``prefcore``
+	Preferred core state of the driver: "enabled" or "disabled".
+
+	"enabled"
+		Enable the ``amd-pstate`` preferred core.
+
+	"disabled"
+		Disable the ``amd-pstate`` preferred core
+
+
+        This attribute is read-only to check the state of preferred core set
+        by the kernel parameter.
+
+``cpb_boost``
+        Specifies whether core performance boost is requested to be enabled or disabled
+        If core performance boost is disabled while a core is in a boosted P-state, the
+        core automatically transitions to the highest performance non-boosted P-state.
+        AMD Core Performance Boost(CPB) is controlled by this new attribute file which
+        allow user to change all cores frequency boosting state. It supports both
+        ``active``, ``passive`` and ``guided`` mode control with below value write to it.
+
+        "0" Disable Core Performance Boosting
+        "1" Enable  Core Performance Boosting
+
 ``cpupower`` tool support for ``amd-pstate``
 ===============================================
 
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 184730705650..70732a76171f 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1054,8 +1054,9 @@ config SCHED_MC
 
 config SCHED_MC_PRIO
 	bool "CPU core priorities scheduler support"
-	depends on SCHED_MC && CPU_SUP_INTEL
-	select X86_INTEL_PSTATE
+	depends on SCHED_MC
+	select X86_INTEL_PSTATE if CPU_SUP_INTEL
+	select X86_AMD_PSTATE if CPU_SUP_AMD && ACPI
 	select CPU_FREQ
 	default y
 	help
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index d1b5edaf6c34..bfe139eb75b6 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -744,6 +744,8 @@
 #define MSR_K7_HWCR_IRPERF_EN		BIT_ULL(MSR_K7_HWCR_IRPERF_EN_BIT)
 #define MSR_K7_FID_VID_CTL		0xc0010041
 #define MSR_K7_FID_VID_STATUS		0xc0010042
+#define MSR_K7_HWCR_CPB_DIS_BIT		25
+#define MSR_K7_HWCR_CPB_DIS		BIT_ULL(MSR_K7_HWCR_CPB_DIS_BIT)
 
 /* K6 MSRs */
 #define MSR_K6_WHCR			0xc0000082
diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
index d155a86a8614..e23a84f4a50a 100644
--- a/drivers/acpi/cppc_acpi.c
+++ b/drivers/acpi/cppc_acpi.c
@@ -679,8 +679,10 @@ int acpi_cppc_processor_probe(struct acpi_processor *pr)
 
 	if (!osc_sb_cppc2_support_acked) {
 		pr_debug("CPPC v2 _OSC not acked\n");
-		if (!cpc_supported_by_cpu())
+		if (!cpc_supported_by_cpu()) {
+			pr_debug("CPPC is not supported by the CPU\n");
 			return -ENODEV;
+		}
 	}
 
 	/* Parse the ACPI _CPC table for this CPU. */
@@ -1157,6 +1159,19 @@ int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
 	return cppc_get_perf(cpunum, NOMINAL_PERF, nominal_perf);
 }
 
+/**
+ * cppc_get_highest_perf - Get the highest performance register value.
+ * @cpunum: CPU from which to get highest performance.
+ * @highest_perf: Return address.
+ *
+ * Return: 0 for success, -EIO otherwise.
+ */
+int cppc_get_highest_perf(int cpunum, u64 *highest_perf)
+{
+	return cppc_get_perf(cpunum, HIGHEST_PERF, highest_perf);
+}
+EXPORT_SYMBOL_GPL(cppc_get_highest_perf);
+
 /**
  * cppc_get_epp_perf - Get the epp register value.
  * @cpunum: CPU from which to get epp preference value.
diff --git a/drivers/acpi/processor_driver.c b/drivers/acpi/processor_driver.c
index 4bd16b3f0781..67db60eda370 100644
--- a/drivers/acpi/processor_driver.c
+++ b/drivers/acpi/processor_driver.c
@@ -27,6 +27,7 @@
 #define ACPI_PROCESSOR_NOTIFY_PERFORMANCE 0x80
 #define ACPI_PROCESSOR_NOTIFY_POWER	0x81
 #define ACPI_PROCESSOR_NOTIFY_THROTTLING	0x82
+#define ACPI_PROCESSOR_NOTIFY_HIGEST_PERF_CHANGED	0x85
 
 MODULE_AUTHOR("Paul Diefenbaugh");
 MODULE_DESCRIPTION("ACPI Processor Driver");
@@ -83,6 +84,11 @@ static void acpi_processor_notify(acpi_handle handle, u32 event, void *data)
 		acpi_bus_generate_netlink_event(device->pnp.device_class,
 						  dev_name(&device->dev), event, 0);
 		break;
+	case ACPI_PROCESSOR_NOTIFY_HIGEST_PERF_CHANGED:
+		cpufreq_update_limits(pr->id);
+		acpi_bus_generate_netlink_event(device->pnp.device_class,
+						  dev_name(&device->dev), event, 0);
+		break;
 	default:
 		acpi_handle_debug(handle, "Unsupported event [0x%x]\n", event);
 		break;
diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 37f1cdf46d29..2fc82831bddd 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -50,8 +50,6 @@ enum {
 #define AMD_MSR_RANGE		(0x7)
 #define HYGON_MSR_RANGE		(0x7)
 
-#define MSR_K7_HWCR_CPB_DIS	(1ULL << 25)
-
 struct acpi_cpufreq_data {
 	unsigned int resume;
 	unsigned int cpu_feature;
diff --git a/drivers/cpufreq/amd-pstate-ut.c b/drivers/cpufreq/amd-pstate-ut.c
index f04ae67dda37..b3601b0e6dd3 100644
--- a/drivers/cpufreq/amd-pstate-ut.c
+++ b/drivers/cpufreq/amd-pstate-ut.c
@@ -226,7 +226,7 @@ static void amd_pstate_ut_check_freq(u32 index)
 			goto skip_test;
 		}
 
-		if (cpudata->boost_supported) {
+		if (amd_pstate_global_params.cpb_boost) {
 			if ((policy->max == cpudata->max_freq) ||
 					(policy->max == cpudata->nominal_freq))
 				amd_pstate_ut_cases[index].result = AMD_PSTATE_UT_RESULT_PASS;
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 07f341995439..651055df1710 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -37,6 +37,7 @@
 #include <linux/uaccess.h>
 #include <linux/static_call.h>
 #include <linux/amd-pstate.h>
+#include <linux/topology.h>
 
 #include <acpi/processor.h>
 #include <acpi/cppc_acpi.h>
@@ -64,6 +65,10 @@ static struct cpufreq_driver amd_pstate_driver;
 static struct cpufreq_driver amd_pstate_epp_driver;
 static int cppc_state = AMD_PSTATE_UNDEFINED;
 static bool cppc_enabled;
+static bool amd_pstate_prefcore = true;
+static struct quirk_entry *quirks;
+struct amd_pstate_global_params amd_pstate_global_params;
+EXPORT_SYMBOL_GPL(amd_pstate_global_params);
 
 /*
  * AMD Energy Preference Performance (EPP)
@@ -108,6 +113,41 @@ static unsigned int epp_values[] = {
 
 typedef int (*cppc_mode_transition_fn)(int);
 
+static struct quirk_entry quirk_amd_7k62 = {
+	.nominal_freq = 2600,
+	.lowest_freq = 550,
+};
+
+static int __init dmi_matched_7k62_bios_bug(const struct dmi_system_id *dmi)
+{
+	/**
+	 * match the broken bios for family 17h processor support CPPC V2
+	 * broken BIOS lack of nominal_freq and lowest_freq capabilities
+	 * definition in ACPI tables
+	 */
+	if (boot_cpu_has(X86_FEATURE_ZEN2)) {
+		quirks = dmi->driver_data;
+		pr_info("Overriding nominal and lowest frequencies for %s\n", dmi->ident);
+		return 1;
+	}
+
+	return 0;
+}
+
+static const struct dmi_system_id amd_pstate_quirks_table[] __initconst = {
+	{
+		.callback = dmi_matched_7k62_bios_bug,
+		.ident = "AMD EPYC 7K62",
+		.matches = {
+			DMI_MATCH(DMI_BIOS_VERSION, "5.14"),
+			DMI_MATCH(DMI_BIOS_RELEASE, "12/12/2019"),
+		},
+		.driver_data = &quirk_amd_7k62,
+	},
+	{}
+};
+MODULE_DEVICE_TABLE(dmi, amd_pstate_quirks_table);
+
 static inline int get_mode_idx_from_str(const char *str, size_t size)
 {
 	int i;
@@ -291,16 +331,20 @@ static int pstate_init_perf(struct amd_cpudata *cpudata)
 {
 	u64 cap1;
 	u32 highest_perf;
+	struct cppc_perf_caps cppc_perf;
+	int ret;
 
-	int ret = rdmsrl_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
+	ret = rdmsrl_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
 				     &cap1);
 	if (ret)
 		return ret;
 
-	/*
-	 * TODO: Introduce AMD specific power feature.
-	 *
-	 * CPPC entry doesn't indicate the highest performance in some ASICs.
+	ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
+	if (ret)
+		return ret;
+
+	/* Some CPUs have different highest_perf from others, it is safer
+	 * to read it than to assume some erroneous value, leading to performance issues.
 	 */
 	highest_perf = amd_get_highest_perf();
 	if (highest_perf > AMD_CPPC_HIGHEST_PERF(cap1))
@@ -311,7 +355,11 @@ static int pstate_init_perf(struct amd_cpudata *cpudata)
 	WRITE_ONCE(cpudata->nominal_perf, AMD_CPPC_NOMINAL_PERF(cap1));
 	WRITE_ONCE(cpudata->lowest_nonlinear_perf, AMD_CPPC_LOWNONLIN_PERF(cap1));
 	WRITE_ONCE(cpudata->lowest_perf, AMD_CPPC_LOWEST_PERF(cap1));
+	WRITE_ONCE(cpudata->prefcore_ranking, AMD_CPPC_HIGHEST_PERF(cap1));
 	WRITE_ONCE(cpudata->min_limit_perf, AMD_CPPC_LOWEST_PERF(cap1));
+	WRITE_ONCE(cpudata->lowest_freq, cppc_perf.lowest_freq);
+	WRITE_ONCE(cpudata->nominal_freq, cppc_perf.nominal_freq);
+
 	return 0;
 }
 
@@ -319,11 +367,15 @@ static int cppc_init_perf(struct amd_cpudata *cpudata)
 {
 	struct cppc_perf_caps cppc_perf;
 	u32 highest_perf;
+	int ret;
 
-	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
+	ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
 	if (ret)
 		return ret;
 
+	/* Some CPUs have different highest_perf from others, it is safer
+	 * to read it than to assume some erroneous value, leading to performance issues.
+	 */
 	highest_perf = amd_get_highest_perf();
 	if (highest_perf > cppc_perf.highest_perf)
 		highest_perf = cppc_perf.highest_perf;
@@ -334,7 +386,10 @@ static int cppc_init_perf(struct amd_cpudata *cpudata)
 	WRITE_ONCE(cpudata->lowest_nonlinear_perf,
 		   cppc_perf.lowest_nonlinear_perf);
 	WRITE_ONCE(cpudata->lowest_perf, cppc_perf.lowest_perf);
+	WRITE_ONCE(cpudata->prefcore_ranking, cppc_perf.highest_perf);
 	WRITE_ONCE(cpudata->min_limit_perf, cppc_perf.lowest_perf);
+	WRITE_ONCE(cpudata->lowest_freq, cppc_perf.lowest_freq);
+	WRITE_ONCE(cpudata->nominal_freq, cppc_perf.nominal_freq);
 
 	if (cppc_state == AMD_PSTATE_ACTIVE)
 		return 0;
@@ -430,7 +485,10 @@ static inline bool amd_pstate_sample(struct amd_cpudata *cpudata)
 static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
 			      u32 des_perf, u32 max_perf, bool fast_switch, int gov_flags)
 {
+	unsigned long max_freq;
+	struct cpufreq_policy *policy = cpufreq_cpu_get(cpudata->cpu);
 	u64 prev = READ_ONCE(cpudata->cppc_req_cached);
+	u32 nominal_perf = READ_ONCE(cpudata->nominal_perf);
 	u64 value = prev;
 
 	min_perf = clamp_t(unsigned long, min_perf, cpudata->min_limit_perf,
@@ -439,6 +497,9 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
 			cpudata->max_limit_perf);
 	des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
 
+	max_freq = READ_ONCE(cpudata->max_limit_freq);
+	policy->cur = div_u64(des_perf * max_freq, max_perf);
+
 	if ((cppc_state == AMD_PSTATE_GUIDED) && (gov_flags & CPUFREQ_GOV_DYNAMIC_SWITCHING)) {
 		min_perf = des_perf;
 		des_perf = 0;
@@ -450,6 +511,10 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
 	value &= ~AMD_CPPC_DES_PERF(~0L);
 	value |= AMD_CPPC_DES_PERF(des_perf);
 
+	/* limit the max perf when core performance boost feature is disabled */
+	if (!amd_pstate_global_params.cpb_boost)
+		max_perf = min_t(unsigned long, nominal_perf, max_perf);
+
 	value &= ~AMD_CPPC_MAX_PERF(~0L);
 	value |= AMD_CPPC_MAX_PERF(max_perf);
 
@@ -477,12 +542,19 @@ static int amd_pstate_verify(struct cpufreq_policy_data *policy)
 
 static int amd_pstate_update_min_max_limit(struct cpufreq_policy *policy)
 {
-	u32 max_limit_perf, min_limit_perf;
+	u32 max_limit_perf, min_limit_perf, lowest_perf;
 	struct amd_cpudata *cpudata = policy->driver_data;
 
 	max_limit_perf = div_u64(policy->max * cpudata->highest_perf, cpudata->max_freq);
 	min_limit_perf = div_u64(policy->min * cpudata->highest_perf, cpudata->max_freq);
 
+	lowest_perf = READ_ONCE(cpudata->lowest_perf);
+	if (min_limit_perf < lowest_perf)
+		min_limit_perf = lowest_perf;
+
+	if (max_limit_perf < min_limit_perf)
+		max_limit_perf = min_limit_perf;
+
 	WRITE_ONCE(cpudata->max_limit_perf, max_limit_perf);
 	WRITE_ONCE(cpudata->min_limit_perf, min_limit_perf);
 	WRITE_ONCE(cpudata->max_limit_freq, policy->max);
@@ -553,10 +625,9 @@ static void amd_pstate_adjust_perf(unsigned int cpu,
 				   unsigned long capacity)
 {
 	unsigned long max_perf, min_perf, des_perf,
-		      cap_perf, lowest_nonlinear_perf, max_freq;
+		      cap_perf, lowest_nonlinear_perf;
 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 	struct amd_cpudata *cpudata = policy->driver_data;
-	unsigned int target_freq;
 
 	if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)
 		amd_pstate_update_min_max_limit(policy);
@@ -564,7 +635,6 @@ static void amd_pstate_adjust_perf(unsigned int cpu,
 
 	cap_perf = READ_ONCE(cpudata->highest_perf);
 	lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf);
-	max_freq = READ_ONCE(cpudata->max_freq);
 
 	des_perf = cap_perf;
 	if (target_perf < capacity)
@@ -582,8 +652,6 @@ static void amd_pstate_adjust_perf(unsigned int cpu,
 		max_perf = min_perf;
 
 	des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
-	target_freq = div_u64(des_perf * max_freq, max_perf);
-	policy->cur = target_freq;
 
 	amd_pstate_update(cpudata, min_perf, des_perf, max_perf, true,
 			policy->governor->flags);
@@ -592,30 +660,30 @@ static void amd_pstate_adjust_perf(unsigned int cpu,
 
 static int amd_get_min_freq(struct amd_cpudata *cpudata)
 {
-	struct cppc_perf_caps cppc_perf;
+	u32 lowest_freq;
 
-	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
-	if (ret)
-		return ret;
+	if (quirks && quirks->lowest_freq)
+		lowest_freq = quirks->lowest_freq;
+	else
+		lowest_freq = READ_ONCE(cpudata->lowest_freq);
 
 	/* Switch to khz */
-	return cppc_perf.lowest_freq * 1000;
+	return lowest_freq * 1000;
 }
 
 static int amd_get_max_freq(struct amd_cpudata *cpudata)
 {
-	struct cppc_perf_caps cppc_perf;
 	u32 max_perf, max_freq, nominal_freq, nominal_perf;
 	u64 boost_ratio;
 
-	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
-	if (ret)
-		return ret;
-
-	nominal_freq = cppc_perf.nominal_freq;
+	nominal_freq = READ_ONCE(cpudata->nominal_freq);
 	nominal_perf = READ_ONCE(cpudata->nominal_perf);
 	max_perf = READ_ONCE(cpudata->highest_perf);
 
+	/* when boost is off, the highest perf will be limited to nominal_perf */
+	if (!amd_pstate_global_params.cpb_boost)
+		max_perf = nominal_perf;
+
 	boost_ratio = div_u64(max_perf << SCHED_CAPACITY_SHIFT,
 			      nominal_perf);
 
@@ -627,31 +695,25 @@ static int amd_get_max_freq(struct amd_cpudata *cpudata)
 
 static int amd_get_nominal_freq(struct amd_cpudata *cpudata)
 {
-	struct cppc_perf_caps cppc_perf;
+	u32 nominal_freq;
 
-	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
-	if (ret)
-		return ret;
+	if (quirks && quirks->nominal_freq)
+		nominal_freq = quirks->nominal_freq;
+	else
+		nominal_freq = READ_ONCE(cpudata->nominal_freq);
 
-	/* Switch to khz */
-	return cppc_perf.nominal_freq * 1000;
+	return nominal_freq;
 }
 
 static int amd_get_lowest_nonlinear_freq(struct amd_cpudata *cpudata)
 {
-	struct cppc_perf_caps cppc_perf;
 	u32 lowest_nonlinear_freq, lowest_nonlinear_perf,
 	    nominal_freq, nominal_perf;
 	u64 lowest_nonlinear_ratio;
 
-	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
-	if (ret)
-		return ret;
-
-	nominal_freq = cppc_perf.nominal_freq;
+	nominal_freq = READ_ONCE(cpudata->nominal_freq);
 	nominal_perf = READ_ONCE(cpudata->nominal_perf);
-
-	lowest_nonlinear_perf = cppc_perf.lowest_nonlinear_perf;
+	lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf);
 
 	lowest_nonlinear_ratio = div_u64(lowest_nonlinear_perf << SCHED_CAPACITY_SHIFT,
 					 nominal_perf);
@@ -662,48 +724,164 @@ static int amd_get_lowest_nonlinear_freq(struct amd_cpudata *cpudata)
 	return lowest_nonlinear_freq * 1000;
 }
 
-static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
+static int amd_pstate_boost_init(struct amd_cpudata *cpudata)
 {
-	struct amd_cpudata *cpudata = policy->driver_data;
+	u64 boost_val;
 	int ret;
 
-	if (!cpudata->boost_supported) {
-		pr_err("Boost mode is not supported by this processor or SBIOS\n");
-		return -EINVAL;
+	ret = rdmsrl_on_cpu(cpudata->cpu, MSR_K7_HWCR, &boost_val);
+	if (ret) {
+		pr_err_once("failed to read initial CPU boost state!\n");
+		return ret;
 	}
 
-	if (state)
-		policy->cpuinfo.max_freq = cpudata->max_freq;
-	else
-		policy->cpuinfo.max_freq = cpudata->nominal_freq;
+	amd_pstate_global_params.cpb_supported = !(boost_val & MSR_K7_HWCR_CPB_DIS);
+	amd_pstate_global_params.cpb_boost = amd_pstate_global_params.cpb_supported;
 
-	policy->max = policy->cpuinfo.max_freq;
+	return ret;
+}
 
-	ret = freq_qos_update_request(&cpudata->req[1],
-				      policy->cpuinfo.max_freq);
-	if (ret < 0)
-		return ret;
+static void amd_perf_ctl_reset(unsigned int cpu)
+{
+	wrmsrl_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
+}
 
-	return 0;
+/*
+ * Set amd-pstate preferred core enable can't be done directly from cpufreq callbacks
+ * due to locking, so queue the work for later.
+ */
+static void amd_pstste_sched_prefcore_workfn(struct work_struct *work)
+{
+	sched_set_itmt_support();
 }
+static DECLARE_WORK(sched_prefcore_work, amd_pstste_sched_prefcore_workfn);
 
-static void amd_pstate_boost_init(struct amd_cpudata *cpudata)
+/*
+ * Get the highest performance register value.
+ * @cpu: CPU from which to get highest performance.
+ * @highest_perf: Return address.
+ *
+ * Return: 0 for success, -EIO otherwise.
+ */
+static int amd_pstate_get_highest_perf(int cpu, u32 *highest_perf)
 {
-	u32 highest_perf, nominal_perf;
+	int ret;
 
-	highest_perf = READ_ONCE(cpudata->highest_perf);
-	nominal_perf = READ_ONCE(cpudata->nominal_perf);
+	if (boot_cpu_has(X86_FEATURE_CPPC)) {
+		u64 cap1;
 
-	if (highest_perf <= nominal_perf)
+		ret = rdmsrl_safe_on_cpu(cpu, MSR_AMD_CPPC_CAP1, &cap1);
+		if (ret)
+			return ret;
+		WRITE_ONCE(*highest_perf, AMD_CPPC_HIGHEST_PERF(cap1));
+	} else {
+		u64 cppc_highest_perf;
+
+		ret = cppc_get_highest_perf(cpu, &cppc_highest_perf);
+		if (ret)
+			return ret;
+		WRITE_ONCE(*highest_perf, cppc_highest_perf);
+	}
+
+	return (ret);
+}
+
+#define CPPC_MAX_PERF	U8_MAX
+
+static void amd_pstate_init_prefcore(struct amd_cpudata *cpudata)
+{
+	int ret, prio;
+	u32 highest_perf;
+
+	ret = amd_pstate_get_highest_perf(cpudata->cpu, &highest_perf);
+	if (ret)
 		return;
 
-	cpudata->boost_supported = true;
-	current_pstate_driver->boost_enabled = true;
+	cpudata->hw_prefcore = true;
+	/* check if CPPC preferred core feature is enabled*/
+	if (highest_perf < CPPC_MAX_PERF)
+		prio = (int)highest_perf;
+	else {
+		pr_debug("AMD CPPC preferred core is unsupported!\n");
+		cpudata->hw_prefcore = false;
+		return;
+	}
+
+	if (!amd_pstate_prefcore)
+		return;
+
+	/*
+	 * The priorities can be set regardless of whether or not
+	 * sched_set_itmt_support(true) has been called and it is valid to
+	 * update them at any time after it has been called.
+	 */
+	sched_set_itmt_core_prio(prio, cpudata->cpu);
+
+	schedule_work(&sched_prefcore_work);
 }
 
-static void amd_perf_ctl_reset(unsigned int cpu)
+static void amd_pstate_update_limits(unsigned int cpu)
 {
-	wrmsrl_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
+	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+	struct amd_cpudata *cpudata = policy->driver_data;
+	u32 prev_high = 0, cur_high = 0;
+	int ret;
+	bool highest_perf_changed = false;
+
+	mutex_lock(&amd_pstate_driver_lock);
+	if ((!amd_pstate_prefcore) || (!cpudata->hw_prefcore))
+		goto free_cpufreq_put;
+
+	ret = amd_pstate_get_highest_perf(cpu, &cur_high);
+	if (ret)
+		goto free_cpufreq_put;
+
+	prev_high = READ_ONCE(cpudata->prefcore_ranking);
+	if (prev_high != cur_high) {
+		highest_perf_changed = true;
+		WRITE_ONCE(cpudata->prefcore_ranking, cur_high);
+
+		if (cur_high < CPPC_MAX_PERF)
+			sched_set_itmt_core_prio((int)cur_high, cpu);
+	}
+
+free_cpufreq_put:
+	cpufreq_cpu_put(policy);
+
+	if (!highest_perf_changed)
+		cpufreq_update_policy(cpu);
+
+	mutex_unlock(&amd_pstate_driver_lock);
+}
+
+/**
+ * Get pstate transition delay time from ACPI tables that firmware set
+ * instead of using hardcode value directly.
+ */
+static u32 amd_pstate_get_transition_delay_us(unsigned int cpu)
+{
+	u32 transition_delay_ns;
+
+	transition_delay_ns = cppc_get_transition_latency(cpu);
+	if (transition_delay_ns == CPUFREQ_ETERNAL)
+		return AMD_PSTATE_TRANSITION_DELAY;
+
+	return transition_delay_ns / NSEC_PER_USEC;
+}
+
+/**
+ * Get pstate transition latency value from ACPI tables that firmware set
+ * instead of using hardcode value directly.
+ */
+static u32 amd_pstate_get_transition_latency(unsigned int cpu)
+{
+	u32 transition_latency;
+
+	transition_latency = cppc_get_transition_latency(cpu);
+	if (transition_latency  == CPUFREQ_ETERNAL)
+		return AMD_PSTATE_TRANSITION_LATENCY;
+
+	return transition_latency;
 }
 
 static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
@@ -727,24 +905,30 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
 
 	cpudata->cpu = policy->cpu;
 
+	amd_pstate_init_prefcore(cpudata);
+
 	ret = amd_pstate_init_perf(cpudata);
 	if (ret)
 		goto free_cpudata1;
 
+	/* initialize cpu cores boot state */
+	amd_pstate_boost_init(cpudata);
+
 	min_freq = amd_get_min_freq(cpudata);
-	max_freq = amd_get_max_freq(cpudata);
 	nominal_freq = amd_get_nominal_freq(cpudata);
+	cpudata->nominal_freq = nominal_freq;
+	max_freq = amd_get_max_freq(cpudata);
 	lowest_nonlinear_freq = amd_get_lowest_nonlinear_freq(cpudata);
 
-	if (min_freq < 0 || max_freq < 0 || min_freq > max_freq) {
-		dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n",
-			min_freq, max_freq);
+	if (min_freq < 0 || max_freq < 0 || min_freq > max_freq || nominal_freq == 0) {
+		dev_err(dev, "min_freq(%d) or max_freq(%d) or nominal_freq(%d) is incorrect\n",
+			min_freq, max_freq, nominal_freq);
 		ret = -EINVAL;
 		goto free_cpudata1;
 	}
 
-	policy->cpuinfo.transition_latency = AMD_PSTATE_TRANSITION_LATENCY;
-	policy->transition_delay_us = AMD_PSTATE_TRANSITION_DELAY;
+	policy->cpuinfo.transition_latency = amd_pstate_get_transition_latency(policy->cpu);
+	policy->transition_delay_us = amd_pstate_get_transition_delay_us(policy->cpu);
 
 	policy->min = min_freq;
 	policy->max = max_freq;
@@ -777,12 +961,10 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
 	cpudata->min_freq = min_freq;
 	cpudata->max_limit_freq = max_freq;
 	cpudata->min_limit_freq = min_freq;
-	cpudata->nominal_freq = nominal_freq;
 	cpudata->lowest_nonlinear_freq = lowest_nonlinear_freq;
 
 	policy->driver_data = cpudata;
 
-	amd_pstate_boost_init(cpudata);
 	if (!current_pstate_driver->adjust_perf)
 		current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
 
@@ -877,6 +1059,28 @@ static ssize_t show_amd_pstate_highest_perf(struct cpufreq_policy *policy,
 	return sysfs_emit(buf, "%u\n", perf);
 }
 
+static ssize_t show_amd_pstate_prefcore_ranking(struct cpufreq_policy *policy,
+						char *buf)
+{
+	u32 perf;
+	struct amd_cpudata *cpudata = policy->driver_data;
+
+	perf = READ_ONCE(cpudata->prefcore_ranking);
+
+	return sysfs_emit(buf, "%u\n", perf);
+}
+
+static ssize_t show_amd_pstate_hw_prefcore(struct cpufreq_policy *policy,
+					   char *buf)
+{
+	bool hw_prefcore;
+	struct amd_cpudata *cpudata = policy->driver_data;
+
+	hw_prefcore = READ_ONCE(cpudata->hw_prefcore);
+
+	return sysfs_emit(buf, "%s\n", str_enabled_disabled(hw_prefcore));
+}
+
 static ssize_t show_energy_performance_available_preferences(
 				struct cpufreq_policy *policy, char *buf)
 {
@@ -1074,18 +1278,125 @@ static ssize_t status_store(struct device *a, struct device_attribute *b,
 	return ret < 0 ? ret : count;
 }
 
+static ssize_t prefcore_show(struct device *dev,
+			     struct device_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "%s\n", str_enabled_disabled(amd_pstate_prefcore));
+}
+
+static int amd_cpu_boost_update(struct amd_cpudata *cpudata, u32 on)
+{
+	struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpudata->cpu);
+	struct cppc_perf_ctrls perf_ctrls;
+	u32 highest_perf, nominal_perf;
+	int ret;
+
+	if (!policy)
+		return -ENODATA;
+
+	highest_perf = READ_ONCE(cpudata->highest_perf);
+	nominal_perf = READ_ONCE(cpudata->nominal_perf);
+
+	if (boot_cpu_has(X86_FEATURE_CPPC)) {
+		u64 value = READ_ONCE(cpudata->cppc_req_cached);
+
+		value &= ~GENMASK_ULL(7, 0);
+		value |= on ? highest_perf : nominal_perf;
+		WRITE_ONCE(cpudata->cppc_req_cached, value);
+
+		wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
+
+	} else {
+		perf_ctrls.max_perf = on ? highest_perf : nominal_perf;
+		ret = cppc_set_epp_perf(cpudata->cpu, &perf_ctrls, 1);
+		if (ret) {
+			pr_debug("failed to set energy perf value (%d)\n", ret);
+			return ret;
+		}
+	}
+
+	if (on)
+		policy->cpuinfo.max_freq = cpudata->max_freq;
+	else
+		policy->cpuinfo.max_freq = cpudata->nominal_freq * 1000;
+
+	policy->max = policy->cpuinfo.max_freq;
+
+	if (cppc_state == AMD_PSTATE_PASSIVE) {
+		ret = freq_qos_update_request(&cpudata->req[1],
+				      policy->cpuinfo.max_freq);
+	}
+
+	cpufreq_cpu_release(policy);
+
+	return ret;
+}
+
+static ssize_t cpb_boost_show(struct device *dev,
+			   struct device_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "%u\n", amd_pstate_global_params.cpb_boost);
+}
+
+static ssize_t cpb_boost_store(struct device *dev, struct device_attribute *b,
+			    const char *buf, size_t count)
+{
+	bool new_state;
+	ssize_t ret;
+	int cpu;
+
+	mutex_lock(&amd_pstate_driver_lock);
+	if (!amd_pstate_global_params.cpb_supported) {
+		pr_err("Boost mode is not supported by this processor or SBIOS\n");
+		return -EINVAL;
+	}
+
+	ret = kstrtobool(buf, &new_state);
+	if (ret)
+		return -EINVAL;
+
+	amd_pstate_global_params.cpb_boost = !!new_state;
+
+	for_each_online_cpu(cpu) {
+
+		struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+		struct amd_cpudata *cpudata = policy->driver_data;
+
+		if (!cpudata) {
+			pr_err("cpudata is NULL\n");
+			ret = -ENODATA;
+			cpufreq_cpu_put(policy);
+			goto err_exit;
+		}
+
+		amd_cpu_boost_update(cpudata, amd_pstate_global_params.cpb_boost);
+		refresh_frequency_limits(policy);
+		cpufreq_cpu_put(policy);
+	}
+
+err_exit:
+	mutex_unlock(&amd_pstate_driver_lock);
+	return ret < 0 ? ret : count;
+}
+
 cpufreq_freq_attr_ro(amd_pstate_max_freq);
 cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
 
 cpufreq_freq_attr_ro(amd_pstate_highest_perf);
+cpufreq_freq_attr_ro(amd_pstate_prefcore_ranking);
+cpufreq_freq_attr_ro(amd_pstate_hw_prefcore);
 cpufreq_freq_attr_rw(energy_performance_preference);
 cpufreq_freq_attr_ro(energy_performance_available_preferences);
 static DEVICE_ATTR_RW(status);
+static DEVICE_ATTR_RO(prefcore);
+static DEVICE_ATTR_RW(cpb_boost);
 
 static struct freq_attr *amd_pstate_attr[] = {
 	&amd_pstate_max_freq,
 	&amd_pstate_lowest_nonlinear_freq,
 	&amd_pstate_highest_perf,
+	&amd_pstate_prefcore_ranking,
+	&amd_pstate_hw_prefcore,
 	NULL,
 };
 
@@ -1093,6 +1404,8 @@ static struct freq_attr *amd_pstate_epp_attr[] = {
 	&amd_pstate_max_freq,
 	&amd_pstate_lowest_nonlinear_freq,
 	&amd_pstate_highest_perf,
+	&amd_pstate_prefcore_ranking,
+	&amd_pstate_hw_prefcore,
 	&energy_performance_preference,
 	&energy_performance_available_preferences,
 	NULL,
@@ -1100,6 +1413,8 @@ static struct freq_attr *amd_pstate_epp_attr[] = {
 
 static struct attribute *pstate_global_attributes[] = {
 	&dev_attr_status.attr,
+	&dev_attr_prefcore.attr,
+	&dev_attr_cpb_boost.attr,
 	NULL
 };
 
@@ -1151,17 +1466,23 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
 	cpudata->cpu = policy->cpu;
 	cpudata->epp_policy = 0;
 
+	amd_pstate_init_prefcore(cpudata);
+
 	ret = amd_pstate_init_perf(cpudata);
 	if (ret)
 		goto free_cpudata1;
 
+	/* initialize cpu cores boot state */
+	amd_pstate_boost_init(cpudata);
+
 	min_freq = amd_get_min_freq(cpudata);
-	max_freq = amd_get_max_freq(cpudata);
 	nominal_freq = amd_get_nominal_freq(cpudata);
+	cpudata->nominal_freq = nominal_freq;
+	max_freq = amd_get_max_freq(cpudata);
 	lowest_nonlinear_freq = amd_get_lowest_nonlinear_freq(cpudata);
-	if (min_freq < 0 || max_freq < 0 || min_freq > max_freq) {
-		dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n",
-				min_freq, max_freq);
+	if (min_freq < 0 || max_freq < 0 || min_freq > max_freq || nominal_freq == 0) {
+		dev_err(dev, "min_freq(%d) or max_freq(%d) or nominal_freq(%d) is incorrect\n",
+				min_freq, max_freq, nominal_freq);
 		ret = -EINVAL;
 		goto free_cpudata1;
 	}
@@ -1174,7 +1495,6 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
 	/* Initial processor data capability frequencies */
 	cpudata->max_freq = max_freq;
 	cpudata->min_freq = min_freq;
-	cpudata->nominal_freq = nominal_freq;
 	cpudata->lowest_nonlinear_freq = lowest_nonlinear_freq;
 
 	policy->driver_data = cpudata;
@@ -1205,7 +1525,6 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
 			return ret;
 		WRITE_ONCE(cpudata->cppc_cap1_cached, value);
 	}
-	amd_pstate_boost_init(cpudata);
 
 	return 0;
 
@@ -1232,6 +1551,12 @@ static void amd_pstate_epp_update_limit(struct cpufreq_policy *policy)
 	max_limit_perf = div_u64(policy->max * cpudata->highest_perf, cpudata->max_freq);
 	min_limit_perf = div_u64(policy->min * cpudata->highest_perf, cpudata->max_freq);
 
+	if (min_limit_perf < min_perf)
+		min_limit_perf = min_perf;
+
+	if (max_limit_perf < min_limit_perf)
+		max_limit_perf = min_limit_perf;
+
 	WRITE_ONCE(cpudata->max_limit_perf, max_limit_perf);
 	WRITE_ONCE(cpudata->min_limit_perf, min_limit_perf);
 
@@ -1294,6 +1619,12 @@ static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy)
 
 	amd_pstate_epp_update_limit(policy);
 
+	/*
+	 * policy->cur is never updated with the amd_pstate_epp driver, but it
+	 * is used as a stale frequency value. So, keep it within limits.
+	 */
+	policy->cur = policy->min;
+
 	return 0;
 }
 
@@ -1431,7 +1762,7 @@ static struct cpufreq_driver amd_pstate_driver = {
 	.exit		= amd_pstate_cpu_exit,
 	.suspend	= amd_pstate_cpu_suspend,
 	.resume		= amd_pstate_cpu_resume,
-	.set_boost	= amd_pstate_set_boost,
+	.update_limits	= amd_pstate_update_limits,
 	.name		= "amd-pstate",
 	.attr		= amd_pstate_attr,
 };
@@ -1446,6 +1777,7 @@ static struct cpufreq_driver amd_pstate_epp_driver = {
 	.online		= amd_pstate_epp_cpu_online,
 	.suspend	= amd_pstate_epp_suspend,
 	.resume		= amd_pstate_epp_resume,
+	.update_limits	= amd_pstate_update_limits,
 	.name		= "amd-pstate-epp",
 	.attr		= amd_pstate_epp_attr,
 };
@@ -1486,6 +1818,11 @@ static int __init amd_pstate_init(void)
 	if (cpufreq_get_current_driver())
 		return -EEXIST;
 
+	quirks = NULL;
+
+	/* check if this machine need CPPC quirks */
+	dmi_check_system(amd_pstate_quirks_table);
+
 	switch (cppc_state) {
 	case AMD_PSTATE_UNDEFINED:
 		/* Disable on the following configs by default:
@@ -1567,7 +1904,17 @@ static int __init amd_pstate_param(char *str)
 
 	return amd_pstate_set_driver(mode_idx);
 }
+
+static int __init amd_prefcore_param(char *str)
+{
+	if (!strcmp(str, "disable"))
+		amd_pstate_prefcore = false;
+
+	return 0;
+}
+
 early_param("amd_pstate", amd_pstate_param);
+early_param("amd_prefcore", amd_prefcore_param);
 
 MODULE_AUTHOR("Huang Rui <ray.huang@amd.com>");
 MODULE_DESCRIPTION("AMD Processor P-state Frequency Driver");
diff --git a/include/acpi/cppc_acpi.h b/include/acpi/cppc_acpi.h
index 3a0995f8bce8..930b6afba6f4 100644
--- a/include/acpi/cppc_acpi.h
+++ b/include/acpi/cppc_acpi.h
@@ -139,6 +139,7 @@ struct cppc_cpudata {
 #ifdef CONFIG_ACPI_CPPC_LIB
 extern int cppc_get_desired_perf(int cpunum, u64 *desired_perf);
 extern int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf);
+extern int cppc_get_highest_perf(int cpunum, u64 *highest_perf);
 extern int cppc_get_perf_ctrs(int cpu, struct cppc_perf_fb_ctrs *perf_fb_ctrs);
 extern int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls);
 extern int cppc_set_enable(int cpu, bool enable);
@@ -167,6 +168,10 @@ static inline int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
 {
 	return -ENOTSUPP;
 }
+static inline int cppc_get_highest_perf(int cpunum, u64 *highest_perf)
+{
+	return -ENOTSUPP;
+}
 static inline int cppc_get_perf_ctrs(int cpu, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
 {
 	return -ENOTSUPP;
diff --git a/include/linux/amd-pstate.h b/include/linux/amd-pstate.h
index 6ad02ad9c7b4..e89cf1249715 100644
--- a/include/linux/amd-pstate.h
+++ b/include/linux/amd-pstate.h
@@ -39,11 +39,16 @@ struct amd_aperf_mperf {
  * @cppc_req_cached: cached performance request hints
  * @highest_perf: the maximum performance an individual processor may reach,
  *		  assuming ideal conditions
+ *		  For platforms that do not support the preferred core feature, the
+ *		  highest_pef may be configured with 166 or 255, to avoid max frequency
+ *		  calculated wrongly. we take the fixed value as the highest_perf.
  * @nominal_perf: the maximum sustained performance level of the processor,
  *		  assuming ideal operating conditions
  * @lowest_nonlinear_perf: the lowest performance level at which nonlinear power
  *			   savings are achieved
  * @lowest_perf: the absolute lowest performance level of the processor
+ * @prefcore_ranking: the preferred core ranking, the higher value indicates a higher
+ * 		  priority.
  * @max_freq: the frequency that mapped to highest_perf
  * @min_freq: the frequency that mapped to lowest_perf
  * @nominal_freq: the frequency that mapped to nominal_perf
@@ -51,7 +56,9 @@ struct amd_aperf_mperf {
  * @cur: Difference of Aperf/Mperf/tsc count between last and current sample
  * @prev: Last Aperf/Mperf/tsc count value read from register
  * @freq: current cpu frequency value
- * @boost_supported: check whether the Processor or SBIOS supports boost mode
+ * @hw_prefcore: check whether HW supports preferred core featue.
+ * 		  Only when hw_prefcore and early prefcore param are true,
+ * 		  AMD P-State driver supports preferred core featue.
  * @epp_policy: Last saved policy used to set energy-performance preference
  * @epp_cached: Cached CPPC energy-performance preference value
  * @policy: Cpufreq policy value
@@ -70,6 +77,7 @@ struct amd_cpudata {
 	u32	nominal_perf;
 	u32	lowest_nonlinear_perf;
 	u32	lowest_perf;
+	u32     prefcore_ranking;
 	u32     min_limit_perf;
 	u32     max_limit_perf;
 	u32     min_limit_freq;
@@ -79,12 +87,13 @@ struct amd_cpudata {
 	u32	min_freq;
 	u32	nominal_freq;
 	u32	lowest_nonlinear_freq;
+	u32	lowest_freq;
 
 	struct amd_aperf_mperf cur;
 	struct amd_aperf_mperf prev;
 
 	u64	freq;
-	bool	boost_supported;
+	bool	hw_prefcore;
 
 	/* EPP feature related attributes*/
 	s16	epp_policy;
@@ -114,4 +123,23 @@ static const char * const amd_pstate_mode_string[] = {
 	[AMD_PSTATE_GUIDED]      = "guided",
 	NULL,
 };
+
+struct quirk_entry {
+	u32 nominal_freq;
+	u32 lowest_freq;
+};
+
+/**
+ * struct amd_pstate_global_params - Global parameters, mostly tunable via sysfs.
+ * @cpb_boost:		Whether or not to use boost CPU P-states.
+ * @cpb_supported:	Whether or not CPU boost P-states are available
+ *			based on the MSR_K7_HWCR bit[25] state
+ */
+struct amd_pstate_global_params {
+	bool cpb_boost;
+	bool cpb_supported;
+};
+
+extern struct amd_pstate_global_params amd_pstate_global_params;
+
 #endif /* _LINUX_AMD_PSTATE_H */
diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h
index 320fab7d2e94..3129411fa978 100644
--- a/include/linux/cpufreq.h
+++ b/include/linux/cpufreq.h
@@ -263,6 +263,7 @@ static inline bool cpufreq_supports_freq_invariance(void)
 	return false;
 }
 static inline void disable_cpufreq(void) { }
+static inline void cpufreq_update_limits(unsigned int cpu) { }
 #endif
 
 #ifdef CONFIG_CPU_FREQ_STAT
-- 
2.44.0

From 7f2e4860d7405f71337e99ea74b84ebcd2c3b90c Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:06:31 +0200
Subject: [PATCH 3/8] bbr3

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 include/linux/tcp.h                |    4 +-
 include/net/inet_connection_sock.h |    4 +-
 include/net/tcp.h                  |   72 +-
 include/uapi/linux/inet_diag.h     |   23 +
 include/uapi/linux/rtnetlink.h     |    4 +-
 include/uapi/linux/tcp.h           |    1 +
 net/ipv4/Kconfig                   |   21 +-
 net/ipv4/bpf_tcp_ca.c              |    9 +-
 net/ipv4/tcp.c                     |    3 +
 net/ipv4/tcp_bbr.c                 | 2230 +++++++++++++++++++++-------
 net/ipv4/tcp_cong.c                |    1 +
 net/ipv4/tcp_input.c               |   40 +-
 net/ipv4/tcp_minisocks.c           |    2 +
 net/ipv4/tcp_output.c              |   48 +-
 net/ipv4/tcp_rate.c                |   30 +-
 net/ipv4/tcp_timer.c               |    1 +
 16 files changed, 1940 insertions(+), 553 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index a1c47a6d69b0..9e63e5580dc5 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -369,7 +369,9 @@ struct tcp_sock {
 	u8	compressed_ack;
 	u8	dup_ack_counter:2,
 		tlp_retrans:1,	/* TLP is a retransmission */
-		unused:5;
+		fast_ack_mode:2, /* which fast ack mode ? */
+		tlp_orig_data_app_limited:1, /* app-limited before TLP rtx? */
+		unused:2;
 	u8	thin_lto    : 1,/* Use linear timeouts for thin streams */
 		recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
 		fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h
index 9ab4bf704e86..f681cfdb2164 100644
--- a/include/net/inet_connection_sock.h
+++ b/include/net/inet_connection_sock.h
@@ -137,8 +137,8 @@ struct inet_connection_sock {
 	u32			  icsk_probes_tstamp;
 	u32			  icsk_user_timeout;
 
-	u64			  icsk_ca_priv[104 / sizeof(u64)];
-#define ICSK_CA_PRIV_SIZE	  sizeof_field(struct inet_connection_sock, icsk_ca_priv)
+#define ICSK_CA_PRIV_SIZE      (144)
+	u64			  icsk_ca_priv[ICSK_CA_PRIV_SIZE / sizeof(u64)];
 };
 
 #define ICSK_TIME_RETRANS	1	/* Retransmit timer */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index f6eba9652d01..3998a5f145ad 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -381,6 +381,8 @@ static inline void tcp_dec_quickack_mode(struct sock *sk)
 #define	TCP_ECN_QUEUE_CWR	2
 #define	TCP_ECN_DEMAND_CWR	4
 #define	TCP_ECN_SEEN		8
+#define	TCP_ECN_LOW		16
+#define	TCP_ECN_ECT_PERMANENT	32
 
 enum tcp_tw_status {
 	TCP_TW_SUCCESS = 0,
@@ -737,6 +739,15 @@ static inline void tcp_fast_path_check(struct sock *sk)
 
 u32 tcp_delack_max(const struct sock *sk);
 
+static inline void tcp_set_ecn_low_from_dst(struct sock *sk,
+					    const struct dst_entry *dst)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	if (dst_feature(dst, RTAX_FEATURE_ECN_LOW))
+		tp->ecn_flags |= TCP_ECN_LOW;
+}
+
 /* Compute the actual rto_min value */
 static inline u32 tcp_rto_min(const struct sock *sk)
 {
@@ -842,6 +853,11 @@ static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
 	return max_t(s64, t1 - t0, 0);
 }
 
+static inline u32 tcp_stamp32_us_delta(u32 t1, u32 t0)
+{
+	return max_t(s32, t1 - t0, 0);
+}
+
 /* provide the departure time in us unit */
 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
 {
@@ -930,9 +946,14 @@ struct tcp_skb_cb {
 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
 			__u32 delivered;
 			/* start of send pipeline phase */
-			u64 first_tx_mstamp;
+			u32 first_tx_mstamp;
 			/* when we reached the "delivered" count */
-			u64 delivered_mstamp;
+			u32 delivered_mstamp;
+#define TCPCB_IN_FLIGHT_BITS 20
+#define TCPCB_IN_FLIGHT_MAX ((1U << TCPCB_IN_FLIGHT_BITS) - 1)
+			u32 in_flight:20,   /* packets in flight at transmit */
+			    unused2:12;
+			u32 lost;	/* packets lost so far upon tx of skb */
 		} tx;   /* only used for outgoing skbs */
 		union {
 			struct inet_skb_parm	h4;
@@ -1036,6 +1057,7 @@ enum tcp_ca_event {
 	CA_EVENT_LOSS,		/* loss timeout */
 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
+	CA_EVENT_TLP_RECOVERY,	/* a lost segment was repaired by TLP probe */
 };
 
 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
@@ -1058,7 +1080,11 @@ enum tcp_ca_ack_event_flags {
 #define TCP_CONG_NON_RESTRICTED 0x1
 /* Requires ECN/ECT set on all packets */
 #define TCP_CONG_NEEDS_ECN	0x2
-#define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
+/* Wants notification of CE events (CA_EVENT_ECN_IS_CE, CA_EVENT_ECN_NO_CE). */
+#define TCP_CONG_WANTS_CE_EVENTS	0x4
+#define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | \
+			 TCP_CONG_NEEDS_ECN | \
+			 TCP_CONG_WANTS_CE_EVENTS)
 
 union tcp_cc_info;
 
@@ -1078,10 +1104,13 @@ struct ack_sample {
  */
 struct rate_sample {
 	u64  prior_mstamp; /* starting timestamp for interval */
+	u32  prior_lost;	/* tp->lost at "prior_mstamp" */
 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
 	u32  prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
+	u32 tx_in_flight;	/* packets in flight at starting timestamp */
+	s32  lost;		/* number of packets lost over interval */
 	s32  delivered;		/* number of packets delivered over interval */
-	s32  delivered_ce;	/* number of packets delivered w/ CE marks*/
+	s32  delivered_ce;	/* packets delivered w/ CE mark over interval */
 	long interval_us;	/* time for tp->delivered to incr "delivered" */
 	u32 snd_interval_us;	/* snd interval for delivered packets */
 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
@@ -1092,7 +1121,9 @@ struct rate_sample {
 	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
 	bool is_retrans;	/* is sample from retransmission? */
+	bool is_acking_tlp_retrans_seq;  /* ACKed a TLP retransmit sequence? */
 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
+	bool is_ece;		/* did this ACK have ECN marked? */
 };
 
 struct tcp_congestion_ops {
@@ -1116,8 +1147,11 @@ struct tcp_congestion_ops {
 	/* hook for packet ack accounting (optional) */
 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
 
-	/* override sysctl_tcp_min_tso_segs */
-	u32 (*min_tso_segs)(struct sock *sk);
+	/* pick target number of segments per TSO/GSO skb (optional): */
+	u32 (*tso_segs)(struct sock *sk, unsigned int mss_now);
+
+	/* react to a specific lost skb (optional) */
+	void (*skb_marked_lost)(struct sock *sk, const struct sk_buff *skb);
 
 	/* call when packets are delivered to update cwnd and pacing rate,
 	 * after all the ca_state processing. (optional)
@@ -1183,6 +1217,14 @@ static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
 }
 #endif
 
+static inline bool tcp_ca_wants_ce_events(const struct sock *sk)
+{
+	const struct inet_connection_sock *icsk = inet_csk(sk);
+
+	return icsk->icsk_ca_ops->flags & (TCP_CONG_NEEDS_ECN |
+					   TCP_CONG_WANTS_CE_EVENTS);
+}
+
 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
@@ -1202,6 +1244,7 @@ static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
 
 /* From tcp_rate.c */
+void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb);
 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
 			    struct rate_sample *rs);
@@ -1214,6 +1257,21 @@ static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
 	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
 }
 
+/* If a retransmit failed due to local qdisc congestion or other local issues,
+ * then we may have called tcp_set_skb_tso_segs() to increase the number of
+ * segments in the skb without increasing the tx.in_flight. In all other cases,
+ * the tx.in_flight should be at least as big as the pcount of the sk_buff.  We
+ * do not have the state to know whether a retransmit failed due to local qdisc
+ * congestion or other local issues, so to avoid spurious warnings we consider
+ * that any skb marked lost may have suffered that fate.
+ */
+static inline bool tcp_skb_tx_in_flight_is_suspicious(u32 skb_pcount,
+						      u32 skb_sacked_flags,
+						      u32 tx_in_flight)
+{
+	return (skb_pcount > tx_in_flight) && !(skb_sacked_flags & TCPCB_LOST);
+}
+
 /* These functions determine how the current flow behaves in respect of SACK
  * handling. SACK is negotiated with the peer, and therefore it can vary
  * between different flows.
@@ -2373,7 +2431,7 @@ struct tcp_plb_state {
 	u8	consec_cong_rounds:5, /* consecutive congested rounds */
 		unused:3;
 	u32	pause_until; /* jiffies32 when PLB can resume rerouting */
-};
+} __attribute__ ((__packed__));
 
 static inline void tcp_plb_init(const struct sock *sk,
 				struct tcp_plb_state *plb)
diff --git a/include/uapi/linux/inet_diag.h b/include/uapi/linux/inet_diag.h
index 50655de04c9b..82f8bd8f0d16 100644
--- a/include/uapi/linux/inet_diag.h
+++ b/include/uapi/linux/inet_diag.h
@@ -229,6 +229,29 @@ struct tcp_bbr_info {
 	__u32	bbr_min_rtt;		/* min-filtered RTT in uSec */
 	__u32	bbr_pacing_gain;	/* pacing gain shifted left 8 bits */
 	__u32	bbr_cwnd_gain;		/* cwnd gain shifted left 8 bits */
+	__u32	bbr_bw_hi_lsb;		/* lower 32 bits of bw_hi */
+	__u32	bbr_bw_hi_msb;		/* upper 32 bits of bw_hi */
+	__u32	bbr_bw_lo_lsb;		/* lower 32 bits of bw_lo */
+	__u32	bbr_bw_lo_msb;		/* upper 32 bits of bw_lo */
+	__u8	bbr_mode;		/* current bbr_mode in state machine */
+	__u8	bbr_phase;		/* current state machine phase */
+	__u8	unused1;		/* alignment padding; not used yet */
+	__u8	bbr_version;		/* BBR algorithm version */
+	__u32	bbr_inflight_lo;	/* lower short-term data volume bound */
+	__u32	bbr_inflight_hi;	/* higher long-term data volume bound */
+	__u32	bbr_extra_acked;	/* max excess packets ACKed in epoch */
+};
+
+/* TCP BBR congestion control bbr_phase as reported in netlink/ss stats. */
+enum tcp_bbr_phase {
+	BBR_PHASE_INVALID		= 0,
+	BBR_PHASE_STARTUP		= 1,
+	BBR_PHASE_DRAIN			= 2,
+	BBR_PHASE_PROBE_RTT		= 3,
+	BBR_PHASE_PROBE_BW_UP		= 4,
+	BBR_PHASE_PROBE_BW_DOWN		= 5,
+	BBR_PHASE_PROBE_BW_CRUISE	= 6,
+	BBR_PHASE_PROBE_BW_REFILL	= 7,
 };
 
 union tcp_cc_info {
diff --git a/include/uapi/linux/rtnetlink.h b/include/uapi/linux/rtnetlink.h
index 3b687d20c9ed..a7c30c243b54 100644
--- a/include/uapi/linux/rtnetlink.h
+++ b/include/uapi/linux/rtnetlink.h
@@ -507,12 +507,14 @@ enum {
 #define RTAX_FEATURE_TIMESTAMP		(1 << 2) /* unused */
 #define RTAX_FEATURE_ALLFRAG		(1 << 3) /* unused */
 #define RTAX_FEATURE_TCP_USEC_TS	(1 << 4)
+#define RTAX_FEATURE_ECN_LOW		(1 << 5)
 
 #define RTAX_FEATURE_MASK	(RTAX_FEATURE_ECN |		\
 				 RTAX_FEATURE_SACK |		\
 				 RTAX_FEATURE_TIMESTAMP |	\
 				 RTAX_FEATURE_ALLFRAG |		\
-				 RTAX_FEATURE_TCP_USEC_TS)
+				 RTAX_FEATURE_TCP_USEC_TS |	\
+				 RTAX_FEATURE_ECN_LOW)
 
 struct rta_session {
 	__u8	proto;
diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
index c07e9f90c084..5c88336ced60 100644
--- a/include/uapi/linux/tcp.h
+++ b/include/uapi/linux/tcp.h
@@ -176,6 +176,7 @@ enum tcp_fastopen_client_fail {
 #define TCPI_OPT_ECN_SEEN	16 /* we received at least one packet with ECT */
 #define TCPI_OPT_SYN_DATA	32 /* SYN-ACK acked data in SYN sent or rcvd */
 #define TCPI_OPT_USEC_TS	64 /* usec timestamps */
+#define TCPI_OPT_ECN_LOW	128 /* Low-latency ECN configured at init */
 
 /*
  * Sender's congestion state indicating normal or abnormal situations
diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index 8e94ed7c56a0..50dc9970cad2 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -668,15 +668,18 @@ config TCP_CONG_BBR
 	default n
 	help
 
-	  BBR (Bottleneck Bandwidth and RTT) TCP congestion control aims to
-	  maximize network utilization and minimize queues. It builds an explicit
-	  model of the bottleneck delivery rate and path round-trip propagation
-	  delay. It tolerates packet loss and delay unrelated to congestion. It
-	  can operate over LAN, WAN, cellular, wifi, or cable modem links. It can
-	  coexist with flows that use loss-based congestion control, and can
-	  operate with shallow buffers, deep buffers, bufferbloat, policers, or
-	  AQM schemes that do not provide a delay signal. It requires the fq
-	  ("Fair Queue") pacing packet scheduler.
+	  BBR (Bottleneck Bandwidth and RTT) TCP congestion control is a
+	  model-based congestion control algorithm that aims to maximize
+	  network utilization, keep queues and retransmit rates low, and to be
+	  able to coexist with Reno/CUBIC in common scenarios. It builds an
+	  explicit model of the network path.  It tolerates a targeted degree
+	  of random packet loss and delay. It can operate over LAN, WAN,
+	  cellular, wifi, or cable modem links, and can use shallow-threshold
+	  ECN signals. It can coexist to some degree with flows that use
+	  loss-based congestion control, and can operate with shallow buffers,
+	  deep buffers, bufferbloat, policers, or AQM schemes that do not
+	  provide a delay signal. It requires pacing, using either TCP internal
+	  pacing or the fq ("Fair Queue") pacing packet scheduler.
 
 choice
 	prompt "Default TCP congestion control"
diff --git a/net/ipv4/bpf_tcp_ca.c b/net/ipv4/bpf_tcp_ca.c
index ae8b15e6896f..beb040e80b6f 100644
--- a/net/ipv4/bpf_tcp_ca.c
+++ b/net/ipv4/bpf_tcp_ca.c
@@ -296,11 +296,15 @@ static void bpf_tcp_ca_pkts_acked(struct sock *sk, const struct ack_sample *samp
 {
 }
 
-static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk)
+static u32 bpf_tcp_ca_tso_segs(struct sock *sk, unsigned int mss_now)
 {
 	return 0;
 }
 
+static void bpf_tcp_ca_skb_marked_lost(struct sock *sk, const struct sk_buff *skb)
+{
+}
+
 static void bpf_tcp_ca_cong_control(struct sock *sk, const struct rate_sample *rs)
 {
 }
@@ -330,7 +334,8 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = {
 	.cwnd_event = bpf_tcp_ca_cwnd_event,
 	.in_ack_event = bpf_tcp_ca_in_ack_event,
 	.pkts_acked = bpf_tcp_ca_pkts_acked,
-	.min_tso_segs = bpf_tcp_ca_min_tso_segs,
+	.tso_segs = bpf_tcp_ca_tso_segs,
+	.skb_marked_lost = bpf_tcp_ca_skb_marked_lost,
 	.cong_control = bpf_tcp_ca_cong_control,
 	.undo_cwnd = bpf_tcp_ca_undo_cwnd,
 	.sndbuf_expand = bpf_tcp_ca_sndbuf_expand,
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index a4f418592314..58469fe5195e 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3089,6 +3089,7 @@ int tcp_disconnect(struct sock *sk, int flags)
 	tp->rx_opt.dsack = 0;
 	tp->rx_opt.num_sacks = 0;
 	tp->rcv_ooopack = 0;
+	tp->fast_ack_mode = 0;
 
 
 	/* Clean up fastopen related fields */
@@ -3815,6 +3816,8 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 		info->tcpi_options |= TCPI_OPT_ECN;
 	if (tp->ecn_flags & TCP_ECN_SEEN)
 		info->tcpi_options |= TCPI_OPT_ECN_SEEN;
+	if (tp->ecn_flags & TCP_ECN_LOW)
+		info->tcpi_options |= TCPI_OPT_ECN_LOW;
 	if (tp->syn_data_acked)
 		info->tcpi_options |= TCPI_OPT_SYN_DATA;
 	if (tp->tcp_usec_ts)
diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c
index 22358032dd48..cd6bef71bf4c 100644
--- a/net/ipv4/tcp_bbr.c
+++ b/net/ipv4/tcp_bbr.c
@@ -1,18 +1,19 @@
-/* Bottleneck Bandwidth and RTT (BBR) congestion control
+/* BBR (Bottleneck Bandwidth and RTT) congestion control
  *
- * BBR congestion control computes the sending rate based on the delivery
- * rate (throughput) estimated from ACKs. In a nutshell:
+ * BBR is a model-based congestion control algorithm that aims for low queues,
+ * low loss, and (bounded) Reno/CUBIC coexistence. To maintain a model of the
+ * network path, it uses measurements of bandwidth and RTT, as well as (if they
+ * occur) packet loss and/or shallow-threshold ECN signals. Note that although
+ * it can use ECN or loss signals explicitly, it does not require either; it
+ * can bound its in-flight data based on its estimate of the BDP.
  *
- *   On each ACK, update our model of the network path:
- *      bottleneck_bandwidth = windowed_max(delivered / elapsed, 10 round trips)
- *      min_rtt = windowed_min(rtt, 10 seconds)
- *   pacing_rate = pacing_gain * bottleneck_bandwidth
- *   cwnd = max(cwnd_gain * bottleneck_bandwidth * min_rtt, 4)
- *
- * The core algorithm does not react directly to packet losses or delays,
- * although BBR may adjust the size of next send per ACK when loss is
- * observed, or adjust the sending rate if it estimates there is a
- * traffic policer, in order to keep the drop rate reasonable.
+ * The model has both higher and lower bounds for the operating range:
+ *   lo: bw_lo, inflight_lo: conservative short-term lower bound
+ *   hi: bw_hi, inflight_hi: robust long-term upper bound
+ * The bandwidth-probing time scale is (a) extended dynamically based on
+ * estimated BDP to improve coexistence with Reno/CUBIC; (b) bounded by
+ * an interactive wall-clock time-scale to be more scalable and responsive
+ * than Reno and CUBIC.
  *
  * Here is a state transition diagram for BBR:
  *
@@ -65,6 +66,13 @@
 #include <linux/random.h>
 #include <linux/win_minmax.h>
 
+#include <trace/events/tcp.h>
+#include "tcp_dctcp.h"
+
+#define BBR_VERSION		3
+
+#define bbr_param(sk,name)	(bbr_ ## name)
+
 /* Scale factor for rate in pkt/uSec unit to avoid truncation in bandwidth
  * estimation. The rate unit ~= (1500 bytes / 1 usec / 2^24) ~= 715 bps.
  * This handles bandwidths from 0.06pps (715bps) to 256Mpps (3Tbps) in a u32.
@@ -85,36 +93,41 @@ enum bbr_mode {
 	BBR_PROBE_RTT,	/* cut inflight to min to probe min_rtt */
 };
 
+/* How does the incoming ACK stream relate to our bandwidth probing? */
+enum bbr_ack_phase {
+	BBR_ACKS_INIT,		  /* not probing; not getting probe feedback */
+	BBR_ACKS_REFILLING,	  /* sending at est. bw to fill pipe */
+	BBR_ACKS_PROBE_STARTING,  /* inflight rising to probe bw */
+	BBR_ACKS_PROBE_FEEDBACK,  /* getting feedback from bw probing */
+	BBR_ACKS_PROBE_STOPPING,  /* stopped probing; still getting feedback */
+};
+
 /* BBR congestion control block */
 struct bbr {
 	u32	min_rtt_us;	        /* min RTT in min_rtt_win_sec window */
 	u32	min_rtt_stamp;	        /* timestamp of min_rtt_us */
 	u32	probe_rtt_done_stamp;   /* end time for BBR_PROBE_RTT mode */
-	struct minmax bw;	/* Max recent delivery rate in pkts/uS << 24 */
-	u32	rtt_cnt;	    /* count of packet-timed rounds elapsed */
+	u32	probe_rtt_min_us;	/* min RTT in probe_rtt_win_ms win */
+	u32	probe_rtt_min_stamp;	/* timestamp of probe_rtt_min_us*/
 	u32     next_rtt_delivered; /* scb->tx.delivered at end of round */
 	u64	cycle_mstamp;	     /* time of this cycle phase start */
-	u32     mode:3,		     /* current bbr_mode in state machine */
+	u32     mode:2,		     /* current bbr_mode in state machine */
 		prev_ca_state:3,     /* CA state on previous ACK */
-		packet_conservation:1,  /* use packet conservation? */
 		round_start:1,	     /* start of packet-timed tx->ack round? */
+		ce_state:1,          /* If most recent data has CE bit set */
+		bw_probe_up_rounds:5,   /* cwnd-limited rounds in PROBE_UP */
+		try_fast_path:1,	/* can we take fast path? */
 		idle_restart:1,	     /* restarting after idle? */
 		probe_rtt_round_done:1,  /* a BBR_PROBE_RTT round at 4 pkts? */
-		unused:13,
-		lt_is_sampling:1,    /* taking long-term ("LT") samples now? */
-		lt_rtt_cnt:7,	     /* round trips in long-term interval */
-		lt_use_bw:1;	     /* use lt_bw as our bw estimate? */
-	u32	lt_bw;		     /* LT est delivery rate in pkts/uS << 24 */
-	u32	lt_last_delivered;   /* LT intvl start: tp->delivered */
-	u32	lt_last_stamp;	     /* LT intvl start: tp->delivered_mstamp */
-	u32	lt_last_lost;	     /* LT intvl start: tp->lost */
+		init_cwnd:7,         /* initial cwnd */
+		unused_1:10;
 	u32	pacing_gain:10,	/* current gain for setting pacing rate */
 		cwnd_gain:10,	/* current gain for setting cwnd */
 		full_bw_reached:1,   /* reached full bw in Startup? */
 		full_bw_cnt:2,	/* number of rounds without large bw gains */
-		cycle_idx:3,	/* current index in pacing_gain cycle array */
+		cycle_idx:2,	/* current index in pacing_gain cycle array */
 		has_seen_rtt:1, /* have we seen an RTT sample yet? */
-		unused_b:5;
+		unused_2:6;
 	u32	prior_cwnd;	/* prior cwnd upon entering loss recovery */
 	u32	full_bw;	/* recent bw, to estimate if pipe is full */
 
@@ -124,19 +137,67 @@ struct bbr {
 	u32	ack_epoch_acked:20,	/* packets (S)ACKed in sampling epoch */
 		extra_acked_win_rtts:5,	/* age of extra_acked, in round trips */
 		extra_acked_win_idx:1,	/* current index in extra_acked array */
-		unused_c:6;
+	/* BBR v3 state: */
+		full_bw_now:1,		/* recently reached full bw plateau? */
+		startup_ecn_rounds:2,	/* consecutive hi ECN STARTUP rounds */
+		loss_in_cycle:1,	/* packet loss in this cycle? */
+		ecn_in_cycle:1,		/* ECN in this cycle? */
+		unused_3:1;
+	u32	loss_round_delivered; /* scb->tx.delivered ending loss round */
+	u32	undo_bw_lo;	     /* bw_lo before latest losses */
+	u32	undo_inflight_lo;    /* inflight_lo before latest losses */
+	u32	undo_inflight_hi;    /* inflight_hi before latest losses */
+	u32	bw_latest;	 /* max delivered bw in last round trip */
+	u32	bw_lo;		 /* lower bound on sending bandwidth */
+	u32	bw_hi[2];	 /* max recent measured bw sample */
+	u32	inflight_latest; /* max delivered data in last round trip */
+	u32	inflight_lo;	 /* lower bound of inflight data range */
+	u32	inflight_hi;	 /* upper bound of inflight data range */
+	u32	bw_probe_up_cnt; /* packets delivered per inflight_hi incr */
+	u32	bw_probe_up_acks;  /* packets (S)ACKed since inflight_hi incr */
+	u32	probe_wait_us;	 /* PROBE_DOWN until next clock-driven probe */
+	u32	prior_rcv_nxt;	/* tp->rcv_nxt when CE state last changed */
+	u32	ecn_eligible:1,	/* sender can use ECN (RTT, handshake)? */
+		ecn_alpha:9,	/* EWMA delivered_ce/delivered; 0..256 */
+		bw_probe_samples:1,    /* rate samples reflect bw probing? */
+		prev_probe_too_high:1, /* did last PROBE_UP go too high? */
+		stopped_risky_probe:1, /* last PROBE_UP stopped due to risk? */
+		rounds_since_probe:8,  /* packet-timed rounds since probed bw */
+		loss_round_start:1,    /* loss_round_delivered round trip? */
+		loss_in_round:1,       /* loss marked in this round trip? */
+		ecn_in_round:1,	       /* ECN marked in this round trip? */
+		ack_phase:3,	       /* bbr_ack_phase: meaning of ACKs */
+		loss_events_in_round:4,/* losses in STARTUP round */
+		initialized:1;	       /* has bbr_init() been called? */
+	u32	alpha_last_delivered;	 /* tp->delivered    at alpha update */
+	u32	alpha_last_delivered_ce; /* tp->delivered_ce at alpha update */
+
+	u8	unused_4;		/* to preserve alignment */
+	struct tcp_plb_state plb;
 };
 
-#define CYCLE_LEN	8	/* number of phases in a pacing gain cycle */
+struct bbr_context {
+	u32 sample_bw;
+};
 
-/* Window length of bw filter (in rounds): */
-static const int bbr_bw_rtts = CYCLE_LEN + 2;
 /* Window length of min_rtt filter (in sec): */
 static const u32 bbr_min_rtt_win_sec = 10;
 /* Minimum time (in ms) spent at bbr_cwnd_min_target in BBR_PROBE_RTT mode: */
 static const u32 bbr_probe_rtt_mode_ms = 200;
-/* Skip TSO below the following bandwidth (bits/sec): */
-static const int bbr_min_tso_rate = 1200000;
+/* Window length of probe_rtt_min_us filter (in ms), and consequently the
+ * typical interval between PROBE_RTT mode entries. The default is 5000ms.
+ * Note that bbr_probe_rtt_win_ms must be <= bbr_min_rtt_win_sec * MSEC_PER_SEC
+ */
+static const u32 bbr_probe_rtt_win_ms = 5000;
+/* Proportion of cwnd to estimated BDP in PROBE_RTT, in units of BBR_UNIT: */
+static const u32 bbr_probe_rtt_cwnd_gain = BBR_UNIT * 1 / 2;
+
+/* Use min_rtt to help adapt TSO burst size, with smaller min_rtt resulting
+ * in bigger TSO bursts. We cut the RTT-based allowance in half
+ * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance
+ * is below 1500 bytes after 6 * ~500 usec = 3ms.
+ */
+static const u32 bbr_tso_rtt_shift = 9;
 
 /* Pace at ~1% below estimated bw, on average, to reduce queue at bottleneck.
  * In order to help drive the network toward lower queues and low latency while
@@ -146,13 +207,15 @@ static const int bbr_min_tso_rate = 1200000;
  */
 static const int bbr_pacing_margin_percent = 1;
 
-/* We use a high_gain value of 2/ln(2) because it's the smallest pacing gain
+/* We use a startup_pacing_gain of 4*ln(2) because it's the smallest value
  * that will allow a smoothly increasing pacing rate that will double each RTT
  * and send the same number of packets per RTT that an un-paced, slow-starting
  * Reno or CUBIC flow would:
  */
-static const int bbr_high_gain  = BBR_UNIT * 2885 / 1000 + 1;
-/* The pacing gain of 1/high_gain in BBR_DRAIN is calculated to typically drain
+static const int bbr_startup_pacing_gain = BBR_UNIT * 277 / 100 + 1;
+/* The gain for deriving startup cwnd: */
+static const int bbr_startup_cwnd_gain = BBR_UNIT * 2;
+/* The pacing gain in BBR_DRAIN is calculated to typically drain
  * the queue created in BBR_STARTUP in a single round:
  */
 static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885;
@@ -160,13 +223,17 @@ static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885;
 static const int bbr_cwnd_gain  = BBR_UNIT * 2;
 /* The pacing_gain values for the PROBE_BW gain cycle, to discover/share bw: */
 static const int bbr_pacing_gain[] = {
-	BBR_UNIT * 5 / 4,	/* probe for more available bw */
-	BBR_UNIT * 3 / 4,	/* drain queue and/or yield bw to other flows */
-	BBR_UNIT, BBR_UNIT, BBR_UNIT,	/* cruise at 1.0*bw to utilize pipe, */
-	BBR_UNIT, BBR_UNIT, BBR_UNIT	/* without creating excess queue... */
+	BBR_UNIT * 5 / 4,	/* UP: probe for more available bw */
+	BBR_UNIT * 91 / 100,	/* DOWN: drain queue and/or yield bw */
+	BBR_UNIT,		/* CRUISE: try to use pipe w/ some headroom */
+	BBR_UNIT,		/* REFILL: refill pipe to estimated 100% */
+};
+enum bbr_pacing_gain_phase {
+	BBR_BW_PROBE_UP		= 0,  /* push up inflight to probe for bw/vol */
+	BBR_BW_PROBE_DOWN	= 1,  /* drain excess inflight from the queue */
+	BBR_BW_PROBE_CRUISE	= 2,  /* use pipe, w/ headroom in queue/pipe */
+	BBR_BW_PROBE_REFILL	= 3,  /* v2: refill the pipe again to 100% */
 };
-/* Randomize the starting gain cycling phase over N phases: */
-static const u32 bbr_cycle_rand = 7;
 
 /* Try to keep at least this many packets in flight, if things go smoothly. For
  * smooth functioning, a sliding window protocol ACKing every other packet
@@ -174,24 +241,12 @@ static const u32 bbr_cycle_rand = 7;
  */
 static const u32 bbr_cwnd_min_target = 4;
 
-/* To estimate if BBR_STARTUP mode (i.e. high_gain) has filled pipe... */
+/* To estimate if BBR_STARTUP or BBR_BW_PROBE_UP has filled pipe... */
 /* If bw has increased significantly (1.25x), there may be more bw available: */
 static const u32 bbr_full_bw_thresh = BBR_UNIT * 5 / 4;
 /* But after 3 rounds w/o significant bw growth, estimate pipe is full: */
 static const u32 bbr_full_bw_cnt = 3;
 
-/* "long-term" ("LT") bandwidth estimator parameters... */
-/* The minimum number of rounds in an LT bw sampling interval: */
-static const u32 bbr_lt_intvl_min_rtts = 4;
-/* If lost/delivered ratio > 20%, interval is "lossy" and we may be policed: */
-static const u32 bbr_lt_loss_thresh = 50;
-/* If 2 intervals have a bw ratio <= 1/8, their bw is "consistent": */
-static const u32 bbr_lt_bw_ratio = BBR_UNIT / 8;
-/* If 2 intervals have a bw diff <= 4 Kbit/sec their bw is "consistent": */
-static const u32 bbr_lt_bw_diff = 4000 / 8;
-/* If we estimate we're policed, use lt_bw for this many round trips: */
-static const u32 bbr_lt_bw_max_rtts = 48;
-
 /* Gain factor for adding extra_acked to target cwnd: */
 static const int bbr_extra_acked_gain = BBR_UNIT;
 /* Window length of extra_acked window. */
@@ -201,8 +256,121 @@ static const u32 bbr_ack_epoch_acked_reset_thresh = 1U << 20;
 /* Time period for clamping cwnd increment due to ack aggregation */
 static const u32 bbr_extra_acked_max_us = 100 * 1000;
 
+/* Flags to control BBR ECN-related behavior... */
+
+/* Ensure ACKs only ACK packets with consistent ECN CE status? */
+static const bool bbr_precise_ece_ack = true;
+
+/* Max RTT (in usec) at which to use sender-side ECN logic.
+ * Disabled when 0 (ECN allowed at any RTT).
+ */
+static const u32 bbr_ecn_max_rtt_us = 5000;
+
+/* On losses, scale down inflight and pacing rate by beta scaled by BBR_SCALE.
+ * No loss response when 0.
+ */
+static const u32 bbr_beta = BBR_UNIT * 30 / 100;
+
+/* Gain factor for ECN mark ratio samples, scaled by BBR_SCALE (1/16 = 6.25%) */
+static const u32 bbr_ecn_alpha_gain = BBR_UNIT * 1 / 16;
+
+/* The initial value for ecn_alpha; 1.0 allows a flow to respond quickly
+ * to congestion if the bottleneck is congested when the flow starts up.
+ */
+static const u32 bbr_ecn_alpha_init = BBR_UNIT;
+
+/* On ECN, cut inflight_lo to (1 - ecn_factor * ecn_alpha) scaled by BBR_SCALE.
+ * No ECN based bounding when 0.
+ */
+static const u32 bbr_ecn_factor = BBR_UNIT * 1 / 3;	 /* 1/3 = 33% */
+
+/* Estimate bw probing has gone too far if CE ratio exceeds this threshold.
+ * Scaled by BBR_SCALE. Disabled when 0.
+ */
+static const u32 bbr_ecn_thresh = BBR_UNIT * 1 / 2;  /* 1/2 = 50% */
+
+/* If non-zero, if in a cycle with no losses but some ECN marks, after ECN
+ * clears then make the first round's increment to inflight_hi the following
+ * fraction of inflight_hi.
+ */
+static const u32 bbr_ecn_reprobe_gain = BBR_UNIT * 1 / 2;
+
+/* Estimate bw probing has gone too far if loss rate exceeds this level. */
+static const u32 bbr_loss_thresh = BBR_UNIT * 2 / 100;  /* 2% loss */
+
+/* Slow down for a packet loss recovered by TLP? */
+static const bool bbr_loss_probe_recovery = true;
+
+/* Exit STARTUP if number of loss marking events in a Recovery round is >= N,
+ * and loss rate is higher than bbr_loss_thresh.
+ * Disabled if 0.
+ */
+static const u32 bbr_full_loss_cnt = 6;
+
+/* Exit STARTUP if number of round trips with ECN mark rate above ecn_thresh
+ * meets this count.
+ */
+static const u32 bbr_full_ecn_cnt = 2;
+
+/* Fraction of unutilized headroom to try to leave in path upon high loss. */
+static const u32 bbr_inflight_headroom = BBR_UNIT * 15 / 100;
+
+/* How much do we increase cwnd_gain when probing for bandwidth in
+ * BBR_BW_PROBE_UP? This specifies the increment in units of
+ * BBR_UNIT/4. The default is 1, meaning 0.25.
+ * The min value is 0 (meaning 0.0); max is 3 (meaning 0.75).
+ */
+static const u32 bbr_bw_probe_cwnd_gain = 1;
+
+/* Max number of packet-timed rounds to wait before probing for bandwidth.  If
+ * we want to tolerate 1% random loss per round, and not have this cut our
+ * inflight too much, we must probe for bw periodically on roughly this scale.
+ * If low, limits Reno/CUBIC coexistence; if high, limits loss tolerance.
+ * We aim to be fair with Reno/CUBIC up to a BDP of at least:
+ *  BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets
+ */
+static const u32 bbr_bw_probe_max_rounds = 63;
+
+/* Max amount of randomness to inject in round counting for Reno-coexistence.
+ */
+static const u32 bbr_bw_probe_rand_rounds = 2;
+
+/* Use BBR-native probe time scale starting at this many usec.
+ * We aim to be fair with Reno/CUBIC up to an inter-loss time epoch of at least:
+ *  BDP*RTT = 25Mbps * .030sec /(1514bytes) * 0.030sec = 1.9 secs
+ */
+static const u32 bbr_bw_probe_base_us = 2 * USEC_PER_SEC;  /* 2 secs */
+
+/* Use BBR-native probes spread over this many usec: */
+static const u32 bbr_bw_probe_rand_us = 1 * USEC_PER_SEC;  /* 1 secs */
+
+/* Use fast path if app-limited, no loss/ECN, and target cwnd was reached? */
+static const bool bbr_fast_path = true;
+
+/* Use fast ack mode? */
+static const bool bbr_fast_ack_mode = true;
+
+static u32 bbr_max_bw(const struct sock *sk);
+static u32 bbr_bw(const struct sock *sk);
+static void bbr_exit_probe_rtt(struct sock *sk);
+static void bbr_reset_congestion_signals(struct sock *sk);
+static void bbr_run_loss_probe_recovery(struct sock *sk);
+
 static void bbr_check_probe_rtt_done(struct sock *sk);
 
+/* This connection can use ECN if both endpoints have signaled ECN support in
+ * the handshake and the per-route settings indicated this is a
+ * shallow-threshold ECN environment, meaning both:
+ *  (a) ECN CE marks indicate low-latency/shallow-threshold congestion, and
+ *  (b) TCP endpoints provide precise ACKs that only ACK data segments
+ *      with consistent ECN CE status
+ */
+static bool bbr_can_use_ecn(const struct sock *sk)
+{
+	return (tcp_sk(sk)->ecn_flags & TCP_ECN_OK) &&
+	       (tcp_sk(sk)->ecn_flags & TCP_ECN_LOW);
+}
+
 /* Do we estimate that STARTUP filled the pipe? */
 static bool bbr_full_bw_reached(const struct sock *sk)
 {
@@ -214,17 +382,17 @@ static bool bbr_full_bw_reached(const struct sock *sk)
 /* Return the windowed max recent bandwidth sample, in pkts/uS << BW_SCALE. */
 static u32 bbr_max_bw(const struct sock *sk)
 {
-	struct bbr *bbr = inet_csk_ca(sk);
+	const struct bbr *bbr = inet_csk_ca(sk);
 
-	return minmax_get(&bbr->bw);
+	return max(bbr->bw_hi[0], bbr->bw_hi[1]);
 }
 
 /* Return the estimated bandwidth of the path, in pkts/uS << BW_SCALE. */
 static u32 bbr_bw(const struct sock *sk)
 {
-	struct bbr *bbr = inet_csk_ca(sk);
+	const struct bbr *bbr = inet_csk_ca(sk);
 
-	return bbr->lt_use_bw ? bbr->lt_bw : bbr_max_bw(sk);
+	return min(bbr_max_bw(sk), bbr->bw_lo);
 }
 
 /* Return maximum extra acked in past k-2k round trips,
@@ -241,15 +409,23 @@ static u16 bbr_extra_acked(const struct sock *sk)
  * The order here is chosen carefully to avoid overflow of u64. This should
  * work for input rates of up to 2.9Tbit/sec and gain of 2.89x.
  */
-static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain)
+static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain,
+				  int margin)
 {
 	unsigned int mss = tcp_sk(sk)->mss_cache;
 
 	rate *= mss;
 	rate *= gain;
 	rate >>= BBR_SCALE;
-	rate *= USEC_PER_SEC / 100 * (100 - bbr_pacing_margin_percent);
-	return rate >> BW_SCALE;
+	rate *= USEC_PER_SEC / 100 * (100 - margin);
+	rate >>= BW_SCALE;
+	rate = max(rate, 1ULL);
+	return rate;
+}
+
+static u64 bbr_bw_bytes_per_sec(struct sock *sk, u64 rate)
+{
+	return bbr_rate_bytes_per_sec(sk, rate, BBR_UNIT, 0);
 }
 
 /* Convert a BBR bw and gain factor to a pacing rate in bytes per second. */
@@ -257,12 +433,13 @@ static unsigned long bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain)
 {
 	u64 rate = bw;
 
-	rate = bbr_rate_bytes_per_sec(sk, rate, gain);
+	rate = bbr_rate_bytes_per_sec(sk, rate, gain,
+				      bbr_pacing_margin_percent);
 	rate = min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate));
 	return rate;
 }
 
-/* Initialize pacing rate to: high_gain * init_cwnd / RTT. */
+/* Initialize pacing rate to: startup_pacing_gain * init_cwnd / RTT. */
 static void bbr_init_pacing_rate_from_rtt(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
@@ -279,7 +456,7 @@ static void bbr_init_pacing_rate_from_rtt(struct sock *sk)
 	bw = (u64)tcp_snd_cwnd(tp) * BW_UNIT;
 	do_div(bw, rtt_us);
 	WRITE_ONCE(sk->sk_pacing_rate,
-		   bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain));
+		   bbr_bw_to_pacing_rate(sk, bw, bbr_param(sk, startup_pacing_gain)));
 }
 
 /* Pace using current bw estimate and a gain factor. */
@@ -295,26 +472,48 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain)
 		WRITE_ONCE(sk->sk_pacing_rate, rate);
 }
 
-/* override sysctl_tcp_min_tso_segs */
-__bpf_kfunc static u32 bbr_min_tso_segs(struct sock *sk)
+/* Return the number of segments BBR would like in a TSO/GSO skb, given a
+ * particular max gso size as a constraint. TODO: make this simpler and more
+ * consistent by switching bbr to just call tcp_tso_autosize().
+ */
+static u32 bbr_tso_segs_generic(struct sock *sk, unsigned int mss_now,
+				u32 gso_max_size)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 segs, r;
+	u64 bytes;
+
+	/* Budget a TSO/GSO burst size allowance based on bw (pacing_rate). */
+	bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift);
+
+	/* Budget a TSO/GSO burst size allowance based on min_rtt. For every
+	 * K = 2^tso_rtt_shift microseconds of min_rtt, halve the burst.
+	 * The min_rtt-based burst allowance is: 64 KBytes / 2^(min_rtt/K)
+	 */
+	if (bbr_param(sk, tso_rtt_shift)) {
+		r = bbr->min_rtt_us >> bbr_param(sk, tso_rtt_shift);
+		if (r < BITS_PER_TYPE(u32))   /* prevent undefined behavior */
+			bytes += GSO_LEGACY_MAX_SIZE >> r;
+	}
+
+	bytes = min_t(u32, bytes, gso_max_size - 1 - MAX_TCP_HEADER);
+	segs = max_t(u32, bytes / mss_now,
+		     sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs);
+	return segs;
+}
+
+/* Custom tcp_tso_autosize() for BBR, used at transmit time to cap skb size. */
+__bpf_kfunc static u32 bbr_tso_segs(struct sock *sk, unsigned int mss_now)
 {
-	return READ_ONCE(sk->sk_pacing_rate) < (bbr_min_tso_rate >> 3) ? 1 : 2;
+	return bbr_tso_segs_generic(sk, mss_now, sk->sk_gso_max_size);
 }
 
+/* Like bbr_tso_segs(), using mss_cache, ignoring driver's sk_gso_max_size. */
 static u32 bbr_tso_segs_goal(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
-	u32 segs, bytes;
-
-	/* Sort of tcp_tso_autosize() but ignoring
-	 * driver provided sk_gso_max_size.
-	 */
-	bytes = min_t(unsigned long,
-		      READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift),
-		      GSO_LEGACY_MAX_SIZE - 1 - MAX_TCP_HEADER);
-	segs = max_t(u32, bytes / tp->mss_cache, bbr_min_tso_segs(sk));
 
-	return min(segs, 0x7FU);
+	return bbr_tso_segs_generic(sk, tp->mss_cache, GSO_LEGACY_MAX_SIZE);
 }
 
 /* Save "last known good" cwnd so we can restore it after losses or PROBE_RTT */
@@ -334,7 +533,9 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event)
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
 
-	if (event == CA_EVENT_TX_START && tp->app_limited) {
+	if (event == CA_EVENT_TX_START) {
+		if (!tp->app_limited)
+			return;
 		bbr->idle_restart = 1;
 		bbr->ack_epoch_mstamp = tp->tcp_mstamp;
 		bbr->ack_epoch_acked = 0;
@@ -345,6 +546,16 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event)
 			bbr_set_pacing_rate(sk, bbr_bw(sk), BBR_UNIT);
 		else if (bbr->mode == BBR_PROBE_RTT)
 			bbr_check_probe_rtt_done(sk);
+	} else if ((event == CA_EVENT_ECN_IS_CE ||
+		    event == CA_EVENT_ECN_NO_CE) &&
+		   bbr_can_use_ecn(sk) &&
+		   bbr_param(sk, precise_ece_ack)) {
+		u32 state = bbr->ce_state;
+		dctcp_ece_ack_update(sk, event, &bbr->prior_rcv_nxt, &state);
+		bbr->ce_state = state;
+	} else if (event == CA_EVENT_TLP_RECOVERY &&
+		   bbr_param(sk, loss_probe_recovery)) {
+		bbr_run_loss_probe_recovery(sk);
 	}
 }
 
@@ -367,10 +578,10 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain)
 	 * default. This should only happen when the connection is not using TCP
 	 * timestamps and has retransmitted all of the SYN/SYNACK/data packets
 	 * ACKed so far. In this case, an RTO can cut cwnd to 1, in which
-	 * case we need to slow-start up toward something safe: TCP_INIT_CWND.
+	 * case we need to slow-start up toward something safe: initial cwnd.
 	 */
 	if (unlikely(bbr->min_rtt_us == ~0U))	 /* no valid RTT samples yet? */
-		return TCP_INIT_CWND;  /* be safe: cap at default initial cwnd*/
+		return bbr->init_cwnd;  /* be safe: cap at initial cwnd */
 
 	w = (u64)bw * bbr->min_rtt_us;
 
@@ -387,23 +598,23 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain)
  *   - one skb in sending host Qdisc,
  *   - one skb in sending host TSO/GSO engine
  *   - one skb being received by receiver host LRO/GRO/delayed-ACK engine
- * Don't worry, at low rates (bbr_min_tso_rate) this won't bloat cwnd because
- * in such cases tso_segs_goal is 1. The minimum cwnd is 4 packets,
+ * Don't worry, at low rates this won't bloat cwnd because
+ * in such cases tso_segs_goal is small. The minimum cwnd is 4 packets,
  * which allows 2 outstanding 2-packet sequences, to try to keep pipe
  * full even with ACK-every-other-packet delayed ACKs.
  */
 static u32 bbr_quantization_budget(struct sock *sk, u32 cwnd)
 {
 	struct bbr *bbr = inet_csk_ca(sk);
+	u32 tso_segs_goal;
 
-	/* Allow enough full-sized skbs in flight to utilize end systems. */
-	cwnd += 3 * bbr_tso_segs_goal(sk);
-
-	/* Reduce delayed ACKs by rounding up cwnd to the next even number. */
-	cwnd = (cwnd + 1) & ~1U;
+	tso_segs_goal = 3 * bbr_tso_segs_goal(sk);
 
+	/* Allow enough full-sized skbs in flight to utilize end systems. */
+	cwnd = max_t(u32, cwnd, tso_segs_goal);
+	cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target));
 	/* Ensure gain cycling gets inflight above BDP even for small BDPs. */
-	if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == 0)
+	if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP)
 		cwnd += 2;
 
 	return cwnd;
@@ -458,10 +669,10 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk)
 {
 	u32 max_aggr_cwnd, aggr_cwnd = 0;
 
-	if (bbr_extra_acked_gain && bbr_full_bw_reached(sk)) {
+	if (bbr_param(sk, extra_acked_gain)) {
 		max_aggr_cwnd = ((u64)bbr_bw(sk) * bbr_extra_acked_max_us)
 				/ BW_UNIT;
-		aggr_cwnd = (bbr_extra_acked_gain * bbr_extra_acked(sk))
+		aggr_cwnd = (bbr_param(sk, extra_acked_gain) * bbr_extra_acked(sk))
 			     >> BBR_SCALE;
 		aggr_cwnd = min(aggr_cwnd, max_aggr_cwnd);
 	}
@@ -469,66 +680,27 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk)
 	return aggr_cwnd;
 }
 
-/* An optimization in BBR to reduce losses: On the first round of recovery, we
- * follow the packet conservation principle: send P packets per P packets acked.
- * After that, we slow-start and send at most 2*P packets per P packets acked.
- * After recovery finishes, or upon undo, we restore the cwnd we had when
- * recovery started (capped by the target cwnd based on estimated BDP).
- *
- * TODO(ycheng/ncardwell): implement a rate-based approach.
- */
-static bool bbr_set_cwnd_to_recover_or_restore(
-	struct sock *sk, const struct rate_sample *rs, u32 acked, u32 *new_cwnd)
+/* Returns the cwnd for PROBE_RTT mode. */
+static u32 bbr_probe_rtt_cwnd(struct sock *sk)
 {
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct bbr *bbr = inet_csk_ca(sk);
-	u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state;
-	u32 cwnd = tcp_snd_cwnd(tp);
-
-	/* An ACK for P pkts should release at most 2*P packets. We do this
-	 * in two steps. First, here we deduct the number of lost packets.
-	 * Then, in bbr_set_cwnd() we slow start up toward the target cwnd.
-	 */
-	if (rs->losses > 0)
-		cwnd = max_t(s32, cwnd - rs->losses, 1);
-
-	if (state == TCP_CA_Recovery && prev_state != TCP_CA_Recovery) {
-		/* Starting 1st round of Recovery, so do packet conservation. */
-		bbr->packet_conservation = 1;
-		bbr->next_rtt_delivered = tp->delivered;  /* start round now */
-		/* Cut unused cwnd from app behavior, TSQ, or TSO deferral: */
-		cwnd = tcp_packets_in_flight(tp) + acked;
-	} else if (prev_state >= TCP_CA_Recovery && state < TCP_CA_Recovery) {
-		/* Exiting loss recovery; restore cwnd saved before recovery. */
-		cwnd = max(cwnd, bbr->prior_cwnd);
-		bbr->packet_conservation = 0;
-	}
-	bbr->prev_ca_state = state;
-
-	if (bbr->packet_conservation) {
-		*new_cwnd = max(cwnd, tcp_packets_in_flight(tp) + acked);
-		return true;	/* yes, using packet conservation */
-	}
-	*new_cwnd = cwnd;
-	return false;
+	return max_t(u32, bbr_param(sk, cwnd_min_target),
+		     bbr_bdp(sk, bbr_bw(sk), bbr_param(sk, probe_rtt_cwnd_gain)));
 }
 
 /* Slow-start up toward target cwnd (if bw estimate is growing, or packet loss
  * has drawn us down below target), or snap down to target if we're above it.
  */
 static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs,
-			 u32 acked, u32 bw, int gain)
+			 u32 acked, u32 bw, int gain, u32 cwnd,
+			 struct bbr_context *ctx)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
-	u32 cwnd = tcp_snd_cwnd(tp), target_cwnd = 0;
+	u32 target_cwnd = 0;
 
 	if (!acked)
 		goto done;  /* no packet fully ACKed; just apply caps */
 
-	if (bbr_set_cwnd_to_recover_or_restore(sk, rs, acked, &cwnd))
-		goto done;
-
 	target_cwnd = bbr_bdp(sk, bw, gain);
 
 	/* Increment the cwnd to account for excess ACKed data that seems
@@ -537,74 +709,26 @@ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs,
 	target_cwnd += bbr_ack_aggregation_cwnd(sk);
 	target_cwnd = bbr_quantization_budget(sk, target_cwnd);
 
-	/* If we're below target cwnd, slow start cwnd toward target cwnd. */
-	if (bbr_full_bw_reached(sk))  /* only cut cwnd if we filled the pipe */
-		cwnd = min(cwnd + acked, target_cwnd);
-	else if (cwnd < target_cwnd || tp->delivered < TCP_INIT_CWND)
-		cwnd = cwnd + acked;
-	cwnd = max(cwnd, bbr_cwnd_min_target);
+	/* Update cwnd and enable fast path if cwnd reaches target_cwnd. */
+	bbr->try_fast_path = 0;
+	if (bbr_full_bw_reached(sk)) { /* only cut cwnd if we filled the pipe */
+		cwnd += acked;
+		if (cwnd >= target_cwnd) {
+			cwnd = target_cwnd;
+			bbr->try_fast_path = 1;
+		}
+	} else if (cwnd < target_cwnd || cwnd  < 2 * bbr->init_cwnd) {
+		cwnd += acked;
+	} else {
+		bbr->try_fast_path = 1;
+	}
 
+	cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target));
 done:
-	tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp));	/* apply global cap */
+	tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp));  /* global cap */
 	if (bbr->mode == BBR_PROBE_RTT)  /* drain queue, refresh min_rtt */
-		tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), bbr_cwnd_min_target));
-}
-
-/* End cycle phase if it's time and/or we hit the phase's in-flight target. */
-static bool bbr_is_next_cycle_phase(struct sock *sk,
-				    const struct rate_sample *rs)
-{
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct bbr *bbr = inet_csk_ca(sk);
-	bool is_full_length =
-		tcp_stamp_us_delta(tp->delivered_mstamp, bbr->cycle_mstamp) >
-		bbr->min_rtt_us;
-	u32 inflight, bw;
-
-	/* The pacing_gain of 1.0 paces at the estimated bw to try to fully
-	 * use the pipe without increasing the queue.
-	 */
-	if (bbr->pacing_gain == BBR_UNIT)
-		return is_full_length;		/* just use wall clock time */
-
-	inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight);
-	bw = bbr_max_bw(sk);
-
-	/* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at
-	 * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is
-	 * small (e.g. on a LAN). We do not persist if packets are lost, since
-	 * a path with small buffers may not hold that much.
-	 */
-	if (bbr->pacing_gain > BBR_UNIT)
-		return is_full_length &&
-			(rs->losses ||  /* perhaps pacing_gain*BDP won't fit */
-			 inflight >= bbr_inflight(sk, bw, bbr->pacing_gain));
-
-	/* A pacing_gain < 1.0 tries to drain extra queue we added if bw
-	 * probing didn't find more bw. If inflight falls to match BDP then we
-	 * estimate queue is drained; persisting would underutilize the pipe.
-	 */
-	return is_full_length ||
-		inflight <= bbr_inflight(sk, bw, BBR_UNIT);
-}
-
-static void bbr_advance_cycle_phase(struct sock *sk)
-{
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	bbr->cycle_idx = (bbr->cycle_idx + 1) & (CYCLE_LEN - 1);
-	bbr->cycle_mstamp = tp->delivered_mstamp;
-}
-
-/* Gain cycling: cycle pacing gain to converge to fair share of available bw. */
-static void bbr_update_cycle_phase(struct sock *sk,
-				   const struct rate_sample *rs)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	if (bbr->mode == BBR_PROBE_BW && bbr_is_next_cycle_phase(sk, rs))
-		bbr_advance_cycle_phase(sk);
+		tcp_snd_cwnd_set(tp, min_t(u32, tcp_snd_cwnd(tp),
+					   bbr_probe_rtt_cwnd(sk)));
 }
 
 static void bbr_reset_startup_mode(struct sock *sk)
@@ -614,191 +738,49 @@ static void bbr_reset_startup_mode(struct sock *sk)
 	bbr->mode = BBR_STARTUP;
 }
 
-static void bbr_reset_probe_bw_mode(struct sock *sk)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	bbr->mode = BBR_PROBE_BW;
-	bbr->cycle_idx = CYCLE_LEN - 1 - get_random_u32_below(bbr_cycle_rand);
-	bbr_advance_cycle_phase(sk);	/* flip to next phase of gain cycle */
-}
-
-static void bbr_reset_mode(struct sock *sk)
-{
-	if (!bbr_full_bw_reached(sk))
-		bbr_reset_startup_mode(sk);
-	else
-		bbr_reset_probe_bw_mode(sk);
-}
-
-/* Start a new long-term sampling interval. */
-static void bbr_reset_lt_bw_sampling_interval(struct sock *sk)
-{
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	bbr->lt_last_stamp = div_u64(tp->delivered_mstamp, USEC_PER_MSEC);
-	bbr->lt_last_delivered = tp->delivered;
-	bbr->lt_last_lost = tp->lost;
-	bbr->lt_rtt_cnt = 0;
-}
-
-/* Completely reset long-term bandwidth sampling. */
-static void bbr_reset_lt_bw_sampling(struct sock *sk)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	bbr->lt_bw = 0;
-	bbr->lt_use_bw = 0;
-	bbr->lt_is_sampling = false;
-	bbr_reset_lt_bw_sampling_interval(sk);
-}
-
-/* Long-term bw sampling interval is done. Estimate whether we're policed. */
-static void bbr_lt_bw_interval_done(struct sock *sk, u32 bw)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-	u32 diff;
-
-	if (bbr->lt_bw) {  /* do we have bw from a previous interval? */
-		/* Is new bw close to the lt_bw from the previous interval? */
-		diff = abs(bw - bbr->lt_bw);
-		if ((diff * BBR_UNIT <= bbr_lt_bw_ratio * bbr->lt_bw) ||
-		    (bbr_rate_bytes_per_sec(sk, diff, BBR_UNIT) <=
-		     bbr_lt_bw_diff)) {
-			/* All criteria are met; estimate we're policed. */
-			bbr->lt_bw = (bw + bbr->lt_bw) >> 1;  /* avg 2 intvls */
-			bbr->lt_use_bw = 1;
-			bbr->pacing_gain = BBR_UNIT;  /* try to avoid drops */
-			bbr->lt_rtt_cnt = 0;
-			return;
-		}
-	}
-	bbr->lt_bw = bw;
-	bbr_reset_lt_bw_sampling_interval(sk);
-}
-
-/* Token-bucket traffic policers are common (see "An Internet-Wide Analysis of
- * Traffic Policing", SIGCOMM 2016). BBR detects token-bucket policers and
- * explicitly models their policed rate, to reduce unnecessary losses. We
- * estimate that we're policed if we see 2 consecutive sampling intervals with
- * consistent throughput and high packet loss. If we think we're being policed,
- * set lt_bw to the "long-term" average delivery rate from those 2 intervals.
+/* See if we have reached next round trip. Upon start of the new round,
+ * returns packets delivered since previous round start plus this ACK.
  */
-static void bbr_lt_bw_sampling(struct sock *sk, const struct rate_sample *rs)
-{
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct bbr *bbr = inet_csk_ca(sk);
-	u32 lost, delivered;
-	u64 bw;
-	u32 t;
-
-	if (bbr->lt_use_bw) {	/* already using long-term rate, lt_bw? */
-		if (bbr->mode == BBR_PROBE_BW && bbr->round_start &&
-		    ++bbr->lt_rtt_cnt >= bbr_lt_bw_max_rtts) {
-			bbr_reset_lt_bw_sampling(sk);    /* stop using lt_bw */
-			bbr_reset_probe_bw_mode(sk);  /* restart gain cycling */
-		}
-		return;
-	}
-
-	/* Wait for the first loss before sampling, to let the policer exhaust
-	 * its tokens and estimate the steady-state rate allowed by the policer.
-	 * Starting samples earlier includes bursts that over-estimate the bw.
-	 */
-	if (!bbr->lt_is_sampling) {
-		if (!rs->losses)
-			return;
-		bbr_reset_lt_bw_sampling_interval(sk);
-		bbr->lt_is_sampling = true;
-	}
-
-	/* To avoid underestimates, reset sampling if we run out of data. */
-	if (rs->is_app_limited) {
-		bbr_reset_lt_bw_sampling(sk);
-		return;
-	}
-
-	if (bbr->round_start)
-		bbr->lt_rtt_cnt++;	/* count round trips in this interval */
-	if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts)
-		return;		/* sampling interval needs to be longer */
-	if (bbr->lt_rtt_cnt > 4 * bbr_lt_intvl_min_rtts) {
-		bbr_reset_lt_bw_sampling(sk);  /* interval is too long */
-		return;
-	}
-
-	/* End sampling interval when a packet is lost, so we estimate the
-	 * policer tokens were exhausted. Stopping the sampling before the
-	 * tokens are exhausted under-estimates the policed rate.
-	 */
-	if (!rs->losses)
-		return;
-
-	/* Calculate packets lost and delivered in sampling interval. */
-	lost = tp->lost - bbr->lt_last_lost;
-	delivered = tp->delivered - bbr->lt_last_delivered;
-	/* Is loss rate (lost/delivered) >= lt_loss_thresh? If not, wait. */
-	if (!delivered || (lost << BBR_SCALE) < bbr_lt_loss_thresh * delivered)
-		return;
-
-	/* Find average delivery rate in this sampling interval. */
-	t = div_u64(tp->delivered_mstamp, USEC_PER_MSEC) - bbr->lt_last_stamp;
-	if ((s32)t < 1)
-		return;		/* interval is less than one ms, so wait */
-	/* Check if can multiply without overflow */
-	if (t >= ~0U / USEC_PER_MSEC) {
-		bbr_reset_lt_bw_sampling(sk);  /* interval too long; reset */
-		return;
-	}
-	t *= USEC_PER_MSEC;
-	bw = (u64)delivered * BW_UNIT;
-	do_div(bw, t);
-	bbr_lt_bw_interval_done(sk, bw);
-}
-
-/* Estimate the bandwidth based on how fast packets are delivered */
-static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs)
+static u32 bbr_update_round_start(struct sock *sk,
+		const struct rate_sample *rs, struct bbr_context *ctx)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
-	u64 bw;
+	u32 round_delivered = 0;
 
 	bbr->round_start = 0;
-	if (rs->delivered < 0 || rs->interval_us <= 0)
-		return; /* Not a valid observation */
 
 	/* See if we've reached the next RTT */
-	if (!before(rs->prior_delivered, bbr->next_rtt_delivered)) {
+	if (rs->interval_us > 0 &&
+	    !before(rs->prior_delivered, bbr->next_rtt_delivered)) {
+		round_delivered = tp->delivered - bbr->next_rtt_delivered;
 		bbr->next_rtt_delivered = tp->delivered;
-		bbr->rtt_cnt++;
 		bbr->round_start = 1;
-		bbr->packet_conservation = 0;
 	}
+	return round_delivered;
+}
 
-	bbr_lt_bw_sampling(sk, rs);
+/* Calculate the bandwidth based on how fast packets are delivered */
+static void bbr_calculate_bw_sample(struct sock *sk,
+			const struct rate_sample *rs, struct bbr_context *ctx)
+{
+	u64 bw = 0;
 
 	/* Divide delivered by the interval to find a (lower bound) bottleneck
 	 * bandwidth sample. Delivered is in packets and interval_us in uS and
 	 * ratio will be <<1 for most connections. So delivered is first scaled.
+	 * Round up to allow growth at low rates, even with integer division.
 	 */
-	bw = div64_long((u64)rs->delivered * BW_UNIT, rs->interval_us);
-
-	/* If this sample is application-limited, it is likely to have a very
-	 * low delivered count that represents application behavior rather than
-	 * the available network rate. Such a sample could drag down estimated
-	 * bw, causing needless slow-down. Thus, to continue to send at the
-	 * last measured network rate, we filter out app-limited samples unless
-	 * they describe the path bw at least as well as our bw model.
-	 *
-	 * So the goal during app-limited phase is to proceed with the best
-	 * network rate no matter how long. We automatically leave this
-	 * phase when app writes faster than the network can deliver :)
-	 */
-	if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) {
-		/* Incorporate new sample into our max bw filter. */
-		minmax_running_max(&bbr->bw, bbr_bw_rtts, bbr->rtt_cnt, bw);
+	if (rs->interval_us > 0) {
+		if (WARN_ONCE(rs->delivered < 0,
+			      "negative delivered: %d interval_us: %ld\n",
+			      rs->delivered, rs->interval_us))
+			return;
+
+		bw = DIV_ROUND_UP_ULL((u64)rs->delivered * BW_UNIT, rs->interval_us);
 	}
+
+	ctx->sample_bw = bw;
 }
 
 /* Estimates the windowed max degree of ack aggregation.
@@ -812,7 +794,7 @@ static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs)
  *
  * Max extra_acked is clamped by cwnd and bw * bbr_extra_acked_max_us (100 ms).
  * Max filter is an approximate sliding window of 5-10 (packet timed) round
- * trips.
+ * trips for non-startup phase, and 1-2 round trips for startup.
  */
 static void bbr_update_ack_aggregation(struct sock *sk,
 				       const struct rate_sample *rs)
@@ -820,15 +802,19 @@ static void bbr_update_ack_aggregation(struct sock *sk,
 	u32 epoch_us, expected_acked, extra_acked;
 	struct bbr *bbr = inet_csk_ca(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
+	u32 extra_acked_win_rtts_thresh = bbr_param(sk, extra_acked_win_rtts);
 
-	if (!bbr_extra_acked_gain || rs->acked_sacked <= 0 ||
+	if (!bbr_param(sk, extra_acked_gain) || rs->acked_sacked <= 0 ||
 	    rs->delivered < 0 || rs->interval_us <= 0)
 		return;
 
 	if (bbr->round_start) {
 		bbr->extra_acked_win_rtts = min(0x1F,
 						bbr->extra_acked_win_rtts + 1);
-		if (bbr->extra_acked_win_rtts >= bbr_extra_acked_win_rtts) {
+		if (!bbr_full_bw_reached(sk))
+			extra_acked_win_rtts_thresh = 1;
+		if (bbr->extra_acked_win_rtts >=
+		    extra_acked_win_rtts_thresh) {
 			bbr->extra_acked_win_rtts = 0;
 			bbr->extra_acked_win_idx = bbr->extra_acked_win_idx ?
 						   0 : 1;
@@ -862,49 +848,6 @@ static void bbr_update_ack_aggregation(struct sock *sk,
 		bbr->extra_acked[bbr->extra_acked_win_idx] = extra_acked;
 }
 
-/* Estimate when the pipe is full, using the change in delivery rate: BBR
- * estimates that STARTUP filled the pipe if the estimated bw hasn't changed by
- * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited
- * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the
- * higher rwin, 3: we get higher delivery rate samples. Or transient
- * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar
- * design goal, but uses delay and inter-ACK spacing instead of bandwidth.
- */
-static void bbr_check_full_bw_reached(struct sock *sk,
-				      const struct rate_sample *rs)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-	u32 bw_thresh;
-
-	if (bbr_full_bw_reached(sk) || !bbr->round_start || rs->is_app_limited)
-		return;
-
-	bw_thresh = (u64)bbr->full_bw * bbr_full_bw_thresh >> BBR_SCALE;
-	if (bbr_max_bw(sk) >= bw_thresh) {
-		bbr->full_bw = bbr_max_bw(sk);
-		bbr->full_bw_cnt = 0;
-		return;
-	}
-	++bbr->full_bw_cnt;
-	bbr->full_bw_reached = bbr->full_bw_cnt >= bbr_full_bw_cnt;
-}
-
-/* If pipe is probably full, drain the queue and then enter steady-state. */
-static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs)
-{
-	struct bbr *bbr = inet_csk_ca(sk);
-
-	if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) {
-		bbr->mode = BBR_DRAIN;	/* drain queue we created */
-		tcp_sk(sk)->snd_ssthresh =
-				bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT);
-	}	/* fall through to check if in-flight is already small: */
-	if (bbr->mode == BBR_DRAIN &&
-	    bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <=
-	    bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT))
-		bbr_reset_probe_bw_mode(sk);  /* we estimate queue is drained */
-}
-
 static void bbr_check_probe_rtt_done(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
@@ -914,9 +857,9 @@ static void bbr_check_probe_rtt_done(struct sock *sk)
 	      after(tcp_jiffies32, bbr->probe_rtt_done_stamp)))
 		return;
 
-	bbr->min_rtt_stamp = tcp_jiffies32;  /* wait a while until PROBE_RTT */
+	bbr->probe_rtt_min_stamp = tcp_jiffies32; /* schedule next PROBE_RTT */
 	tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd));
-	bbr_reset_mode(sk);
+	bbr_exit_probe_rtt(sk);
 }
 
 /* The goal of PROBE_RTT mode is to have BBR flows cooperatively and
@@ -942,23 +885,35 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
-	bool filter_expired;
+	bool probe_rtt_expired, min_rtt_expired;
+	u32 expire;
 
-	/* Track min RTT seen in the min_rtt_win_sec filter window: */
-	filter_expired = after(tcp_jiffies32,
-			       bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ);
+	/* Track min RTT in probe_rtt_win_ms to time next PROBE_RTT state. */
+	expire = bbr->probe_rtt_min_stamp +
+		 msecs_to_jiffies(bbr_param(sk, probe_rtt_win_ms));
+	probe_rtt_expired = after(tcp_jiffies32, expire);
 	if (rs->rtt_us >= 0 &&
-	    (rs->rtt_us < bbr->min_rtt_us ||
-	     (filter_expired && !rs->is_ack_delayed))) {
-		bbr->min_rtt_us = rs->rtt_us;
-		bbr->min_rtt_stamp = tcp_jiffies32;
+	    (rs->rtt_us < bbr->probe_rtt_min_us ||
+	     (probe_rtt_expired && !rs->is_ack_delayed))) {
+		bbr->probe_rtt_min_us = rs->rtt_us;
+		bbr->probe_rtt_min_stamp = tcp_jiffies32;
+	}
+	/* Track min RTT seen in the min_rtt_win_sec filter window: */
+	expire = bbr->min_rtt_stamp + bbr_param(sk, min_rtt_win_sec) * HZ;
+	min_rtt_expired = after(tcp_jiffies32, expire);
+	if (bbr->probe_rtt_min_us <= bbr->min_rtt_us ||
+	    min_rtt_expired) {
+		bbr->min_rtt_us = bbr->probe_rtt_min_us;
+		bbr->min_rtt_stamp = bbr->probe_rtt_min_stamp;
 	}
 
-	if (bbr_probe_rtt_mode_ms > 0 && filter_expired &&
+	if (bbr_param(sk, probe_rtt_mode_ms) > 0 && probe_rtt_expired &&
 	    !bbr->idle_restart && bbr->mode != BBR_PROBE_RTT) {
 		bbr->mode = BBR_PROBE_RTT;  /* dip, drain queue */
 		bbr_save_cwnd(sk);  /* note cwnd so we can restore it */
 		bbr->probe_rtt_done_stamp = 0;
+		bbr->ack_phase = BBR_ACKS_PROBE_STOPPING;
+		bbr->next_rtt_delivered = tp->delivered;
 	}
 
 	if (bbr->mode == BBR_PROBE_RTT) {
@@ -967,9 +922,9 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs)
 			(tp->delivered + tcp_packets_in_flight(tp)) ? : 1;
 		/* Maintain min packets in flight for max(200 ms, 1 round). */
 		if (!bbr->probe_rtt_done_stamp &&
-		    tcp_packets_in_flight(tp) <= bbr_cwnd_min_target) {
+		    tcp_packets_in_flight(tp) <= bbr_probe_rtt_cwnd(sk)) {
 			bbr->probe_rtt_done_stamp = tcp_jiffies32 +
-				msecs_to_jiffies(bbr_probe_rtt_mode_ms);
+				msecs_to_jiffies(bbr_param(sk, probe_rtt_mode_ms));
 			bbr->probe_rtt_round_done = 0;
 			bbr->next_rtt_delivered = tp->delivered;
 		} else if (bbr->probe_rtt_done_stamp) {
@@ -990,18 +945,20 @@ static void bbr_update_gains(struct sock *sk)
 
 	switch (bbr->mode) {
 	case BBR_STARTUP:
-		bbr->pacing_gain = bbr_high_gain;
-		bbr->cwnd_gain	 = bbr_high_gain;
+		bbr->pacing_gain = bbr_param(sk, startup_pacing_gain);
+		bbr->cwnd_gain	 = bbr_param(sk, startup_cwnd_gain);
 		break;
 	case BBR_DRAIN:
-		bbr->pacing_gain = bbr_drain_gain;	/* slow, to drain */
-		bbr->cwnd_gain	 = bbr_high_gain;	/* keep cwnd */
+		bbr->pacing_gain = bbr_param(sk, drain_gain);  /* slow, to drain */
+		bbr->cwnd_gain	 = bbr_param(sk, startup_cwnd_gain);  /* keep cwnd */
 		break;
 	case BBR_PROBE_BW:
-		bbr->pacing_gain = (bbr->lt_use_bw ?
-				    BBR_UNIT :
-				    bbr_pacing_gain[bbr->cycle_idx]);
-		bbr->cwnd_gain	 = bbr_cwnd_gain;
+		bbr->pacing_gain = bbr_pacing_gain[bbr->cycle_idx];
+		bbr->cwnd_gain	 = bbr_param(sk, cwnd_gain);
+		if (bbr_param(sk, bw_probe_cwnd_gain) &&
+		    bbr->cycle_idx == BBR_BW_PROBE_UP)
+			bbr->cwnd_gain +=
+				BBR_UNIT * bbr_param(sk, bw_probe_cwnd_gain) / 4;
 		break;
 	case BBR_PROBE_RTT:
 		bbr->pacing_gain = BBR_UNIT;
@@ -1013,144 +970,1387 @@ static void bbr_update_gains(struct sock *sk)
 	}
 }
 
-static void bbr_update_model(struct sock *sk, const struct rate_sample *rs)
+__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk)
 {
-	bbr_update_bw(sk, rs);
-	bbr_update_ack_aggregation(sk, rs);
-	bbr_update_cycle_phase(sk, rs);
-	bbr_check_full_bw_reached(sk, rs);
-	bbr_check_drain(sk, rs);
-	bbr_update_min_rtt(sk, rs);
-	bbr_update_gains(sk);
+	/* Provision 3 * cwnd since BBR may slow-start even during recovery. */
+	return 3;
 }
 
-__bpf_kfunc static void bbr_main(struct sock *sk, const struct rate_sample *rs)
+/* Incorporate a new bw sample into the current window of our max filter. */
+static void bbr_take_max_bw_sample(struct sock *sk, u32 bw)
 {
 	struct bbr *bbr = inet_csk_ca(sk);
-	u32 bw;
-
-	bbr_update_model(sk, rs);
 
-	bw = bbr_bw(sk);
-	bbr_set_pacing_rate(sk, bw, bbr->pacing_gain);
-	bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain);
+	bbr->bw_hi[1] = max(bw, bbr->bw_hi[1]);
 }
 
-__bpf_kfunc static void bbr_init(struct sock *sk)
+/* Keep max of last 1-2 cycles. Each PROBE_BW cycle, flip filter window. */
+static void bbr_advance_max_bw_filter(struct sock *sk)
 {
-	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
 
-	bbr->prior_cwnd = 0;
-	tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
-	bbr->rtt_cnt = 0;
-	bbr->next_rtt_delivered = tp->delivered;
-	bbr->prev_ca_state = TCP_CA_Open;
-	bbr->packet_conservation = 0;
-
-	bbr->probe_rtt_done_stamp = 0;
-	bbr->probe_rtt_round_done = 0;
-	bbr->min_rtt_us = tcp_min_rtt(tp);
-	bbr->min_rtt_stamp = tcp_jiffies32;
-
-	minmax_reset(&bbr->bw, bbr->rtt_cnt, 0);  /* init max bw to 0 */
+	if (!bbr->bw_hi[1])
+		return;  /* no samples in this window; remember old window */
+	bbr->bw_hi[0] = bbr->bw_hi[1];
+	bbr->bw_hi[1] = 0;
+}
 
-	bbr->has_seen_rtt = 0;
-	bbr_init_pacing_rate_from_rtt(sk);
+/* Reset the estimator for reaching full bandwidth based on bw plateau. */
+static void bbr_reset_full_bw(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
 
-	bbr->round_start = 0;
-	bbr->idle_restart = 0;
-	bbr->full_bw_reached = 0;
 	bbr->full_bw = 0;
 	bbr->full_bw_cnt = 0;
-	bbr->cycle_mstamp = 0;
-	bbr->cycle_idx = 0;
-	bbr_reset_lt_bw_sampling(sk);
-	bbr_reset_startup_mode(sk);
+	bbr->full_bw_now = 0;
+}
 
-	bbr->ack_epoch_mstamp = tp->tcp_mstamp;
-	bbr->ack_epoch_acked = 0;
-	bbr->extra_acked_win_rtts = 0;
-	bbr->extra_acked_win_idx = 0;
-	bbr->extra_acked[0] = 0;
-	bbr->extra_acked[1] = 0;
+/* How much do we want in flight? Our BDP, unless congestion cut cwnd. */
+static u32 bbr_target_inflight(struct sock *sk)
+{
+	u32 bdp = bbr_inflight(sk, bbr_bw(sk), BBR_UNIT);
 
-	cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED);
+	return min(bdp, tcp_sk(sk)->snd_cwnd);
 }
 
-__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk)
+static bool bbr_is_probing_bandwidth(struct sock *sk)
 {
-	/* Provision 3 * cwnd since BBR may slow-start even during recovery. */
-	return 3;
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	return (bbr->mode == BBR_STARTUP) ||
+		(bbr->mode == BBR_PROBE_BW &&
+		 (bbr->cycle_idx == BBR_BW_PROBE_REFILL ||
+		  bbr->cycle_idx == BBR_BW_PROBE_UP));
+}
+
+/* Has the given amount of time elapsed since we marked the phase start? */
+static bool bbr_has_elapsed_in_phase(const struct sock *sk, u32 interval_us)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	const struct bbr *bbr = inet_csk_ca(sk);
+
+	return tcp_stamp_us_delta(tp->tcp_mstamp,
+				  bbr->cycle_mstamp + interval_us) > 0;
+}
+
+static void bbr_handle_queue_too_high_in_startup(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 bdp;  /* estimated BDP in packets, with quantization budget */
+
+	bbr->full_bw_reached = 1;
+
+	bdp = bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT);
+	bbr->inflight_hi = max(bdp, bbr->inflight_latest);
+}
+
+/* Exit STARTUP upon N consecutive rounds with ECN mark rate > ecn_thresh. */
+static void bbr_check_ecn_too_high_in_startup(struct sock *sk, u32 ce_ratio)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr_full_bw_reached(sk) || !bbr->ecn_eligible ||
+	    !bbr_param(sk, full_ecn_cnt) || !bbr_param(sk, ecn_thresh))
+		return;
+
+	if (ce_ratio >= bbr_param(sk, ecn_thresh))
+		bbr->startup_ecn_rounds++;
+	else
+		bbr->startup_ecn_rounds = 0;
+
+	if (bbr->startup_ecn_rounds >= bbr_param(sk, full_ecn_cnt)) {
+		bbr_handle_queue_too_high_in_startup(sk);
+		return;
+	}
+}
+
+/* Updates ecn_alpha and returns ce_ratio. -1 if not available. */
+static int bbr_update_ecn_alpha(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct net *net = sock_net(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	s32 delivered, delivered_ce;
+	u64 alpha, ce_ratio;
+	u32 gain;
+	bool want_ecn_alpha;
+
+	/* See if we should use ECN sender logic for this connection. */
+	if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) &&
+	    bbr_param(sk, ecn_factor) &&
+	    (bbr->min_rtt_us <= bbr_ecn_max_rtt_us ||
+	     !bbr_ecn_max_rtt_us))
+		bbr->ecn_eligible = 1;
+
+	/* Skip updating alpha only if not ECN-eligible and PLB is disabled. */
+	want_ecn_alpha = (bbr->ecn_eligible ||
+			  (bbr_can_use_ecn(sk) &&
+			   READ_ONCE(net->ipv4.sysctl_tcp_plb_enabled)));
+	if (!want_ecn_alpha)
+		return -1;
+
+	delivered = tp->delivered - bbr->alpha_last_delivered;
+	delivered_ce = tp->delivered_ce - bbr->alpha_last_delivered_ce;
+
+	if (delivered == 0 ||		/* avoid divide by zero */
+	    WARN_ON_ONCE(delivered < 0 || delivered_ce < 0))  /* backwards? */
+		return -1;
+
+	BUILD_BUG_ON(BBR_SCALE != TCP_PLB_SCALE);
+	ce_ratio = (u64)delivered_ce << BBR_SCALE;
+	do_div(ce_ratio, delivered);
+
+	gain = bbr_param(sk, ecn_alpha_gain);
+	alpha = ((BBR_UNIT - gain) * bbr->ecn_alpha) >> BBR_SCALE;
+	alpha += (gain * ce_ratio) >> BBR_SCALE;
+	bbr->ecn_alpha = min_t(u32, alpha, BBR_UNIT);
+
+	bbr->alpha_last_delivered = tp->delivered;
+	bbr->alpha_last_delivered_ce = tp->delivered_ce;
+
+	bbr_check_ecn_too_high_in_startup(sk, ce_ratio);
+	return (int)ce_ratio;
 }
 
-/* In theory BBR does not need to undo the cwnd since it does not
- * always reduce cwnd on losses (see bbr_main()). Keep it for now.
+/* Protective Load Balancing (PLB). PLB rehashes outgoing data (to a new IPv6
+ * flow label) if it encounters sustained congestion in the form of ECN marks.
  */
-__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk)
+static void bbr_plb(struct sock *sk, const struct rate_sample *rs, int ce_ratio)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr->round_start && ce_ratio >= 0)
+		tcp_plb_update_state(sk, &bbr->plb, ce_ratio);
+
+	tcp_plb_check_rehash(sk, &bbr->plb);
+}
+
+/* Each round trip of BBR_BW_PROBE_UP, double volume of probing data. */
+static void bbr_raise_inflight_hi_slope(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 growth_this_round, cnt;
+
+	/* Calculate "slope": packets S/Acked per inflight_hi increment. */
+	growth_this_round = 1 << bbr->bw_probe_up_rounds;
+	bbr->bw_probe_up_rounds = min(bbr->bw_probe_up_rounds + 1, 30);
+	cnt = tcp_snd_cwnd(tp) / growth_this_round;
+	cnt = max(cnt, 1U);
+	bbr->bw_probe_up_cnt = cnt;
+}
+
+/* In BBR_BW_PROBE_UP, not seeing high loss/ECN/queue, so raise inflight_hi. */
+static void bbr_probe_inflight_hi_upward(struct sock *sk,
+					  const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 delta;
+
+	if (!tp->is_cwnd_limited || tcp_snd_cwnd(tp) < bbr->inflight_hi)
+		return;  /* not fully using inflight_hi, so don't grow it */
+
+	/* For each bw_probe_up_cnt packets ACKed, increase inflight_hi by 1. */
+	bbr->bw_probe_up_acks += rs->acked_sacked;
+	if (bbr->bw_probe_up_acks >=  bbr->bw_probe_up_cnt) {
+		delta = bbr->bw_probe_up_acks / bbr->bw_probe_up_cnt;
+		bbr->bw_probe_up_acks -= delta * bbr->bw_probe_up_cnt;
+		bbr->inflight_hi += delta;
+		bbr->try_fast_path = 0;  /* Need to update cwnd */
+	}
+
+	if (bbr->round_start)
+		bbr_raise_inflight_hi_slope(sk);
+}
+
+/* Does loss/ECN rate for this sample say inflight is "too high"?
+ * This is used by both the bbr_check_loss_too_high_in_startup() function,
+ * which can be used in either v1 or v2, and the PROBE_UP phase of v2, which
+ * uses it to notice when loss/ECN rates suggest inflight is too high.
+ */
+static bool bbr_is_inflight_too_high(const struct sock *sk,
+				      const struct rate_sample *rs)
+{
+	const struct bbr *bbr = inet_csk_ca(sk);
+	u32 loss_thresh, ecn_thresh;
+
+	if (rs->lost > 0 && rs->tx_in_flight) {
+		loss_thresh = (u64)rs->tx_in_flight * bbr_param(sk, loss_thresh) >>
+				BBR_SCALE;
+		if (rs->lost > loss_thresh) {
+			return true;
+		}
+	}
+
+	if (rs->delivered_ce > 0 && rs->delivered > 0 &&
+	    bbr->ecn_eligible && bbr_param(sk, ecn_thresh)) {
+		ecn_thresh = (u64)rs->delivered * bbr_param(sk, ecn_thresh) >>
+				BBR_SCALE;
+		if (rs->delivered_ce > ecn_thresh) {
+			return true;
+		}
+	}
+
+	return false;
+}
+
+/* Calculate the tx_in_flight level that corresponded to excessive loss.
+ * We find "lost_prefix" segs of the skb where loss rate went too high,
+ * by solving for "lost_prefix" in the following equation:
+ *   lost                     /  inflight                     >= loss_thresh
+ *  (lost_prev + lost_prefix) / (inflight_prev + lost_prefix) >= loss_thresh
+ * Then we take that equation, convert it to fixed point, and
+ * round up to the nearest packet.
+ */
+static u32 bbr_inflight_hi_from_lost_skb(const struct sock *sk,
+					  const struct rate_sample *rs,
+					  const struct sk_buff *skb)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	u32 loss_thresh  = bbr_param(sk, loss_thresh);
+	u32 pcount, divisor, inflight_hi;
+	s32 inflight_prev, lost_prev;
+	u64 loss_budget, lost_prefix;
+
+	pcount = tcp_skb_pcount(skb);
+
+	/* How much data was in flight before this skb? */
+	inflight_prev = rs->tx_in_flight - pcount;
+	if (inflight_prev < 0) {
+		WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious(
+				  pcount,
+				  TCP_SKB_CB(skb)->sacked,
+				  rs->tx_in_flight),
+			  "tx_in_flight: %u pcount: %u reneg: %u",
+			  rs->tx_in_flight, pcount, tcp_sk(sk)->is_sack_reneg);
+		return ~0U;
+	}
+
+	/* How much inflight data was marked lost before this skb? */
+	lost_prev = rs->lost - pcount;
+	if (WARN_ONCE(lost_prev < 0,
+		      "cwnd: %u ca: %d out: %u lost: %u pif: %u "
+		      "tx_in_flight: %u tx.lost: %u tp->lost: %u rs->lost: %d "
+		      "lost_prev: %d pcount: %d seq: %u end_seq: %u reneg: %u",
+		      tcp_snd_cwnd(tp), inet_csk(sk)->icsk_ca_state,
+		      tp->packets_out, tp->lost_out, tcp_packets_in_flight(tp),
+		      rs->tx_in_flight, TCP_SKB_CB(skb)->tx.lost, tp->lost,
+		      rs->lost, lost_prev, pcount,
+		      TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
+		      tp->is_sack_reneg))
+		return ~0U;
+
+	/* At what prefix of this lost skb did losss rate exceed loss_thresh? */
+	loss_budget = (u64)inflight_prev * loss_thresh + BBR_UNIT - 1;
+	loss_budget >>= BBR_SCALE;
+	if (lost_prev >= loss_budget) {
+		lost_prefix = 0;   /* previous losses crossed loss_thresh */
+	} else {
+		lost_prefix = loss_budget - lost_prev;
+		lost_prefix <<= BBR_SCALE;
+		divisor = BBR_UNIT - loss_thresh;
+		if (WARN_ON_ONCE(!divisor))  /* loss_thresh is 8 bits */
+			return ~0U;
+		do_div(lost_prefix, divisor);
+	}
+
+	inflight_hi = inflight_prev + lost_prefix;
+	return inflight_hi;
+}
+
+/* If loss/ECN rates during probing indicated we may have overfilled a
+ * buffer, return an operating point that tries to leave unutilized headroom in
+ * the path for other flows, for fairness convergence and lower RTTs and loss.
+ */
+static u32 bbr_inflight_with_headroom(const struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 headroom, headroom_fraction;
+
+	if (bbr->inflight_hi == ~0U)
+		return ~0U;
+
+	headroom_fraction = bbr_param(sk, inflight_headroom);
+	headroom = ((u64)bbr->inflight_hi * headroom_fraction) >> BBR_SCALE;
+	headroom = max(headroom, 1U);
+	return max_t(s32, bbr->inflight_hi - headroom,
+		     bbr_param(sk, cwnd_min_target));
+}
+
+/* Bound cwnd to a sensible level, based on our current probing state
+ * machine phase and model of a good inflight level (inflight_lo, inflight_hi).
+ */
+static void bbr_bound_cwnd_for_inflight_model(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 cap;
+
+	/* tcp_rcv_synsent_state_process() currently calls tcp_ack()
+	 * and thus cong_control() without first initializing us(!).
+	 */
+	if (!bbr->initialized)
+		return;
+
+	cap = ~0U;
+	if (bbr->mode == BBR_PROBE_BW &&
+	    bbr->cycle_idx != BBR_BW_PROBE_CRUISE) {
+		/* Probe to see if more packets fit in the path. */
+		cap = bbr->inflight_hi;
+	} else {
+		if (bbr->mode == BBR_PROBE_RTT ||
+		    (bbr->mode == BBR_PROBE_BW &&
+		     bbr->cycle_idx == BBR_BW_PROBE_CRUISE))
+			cap = bbr_inflight_with_headroom(sk);
+	}
+	/* Adapt to any loss/ECN since our last bw probe. */
+	cap = min(cap, bbr->inflight_lo);
+
+	cap = max_t(u32, cap, bbr_param(sk, cwnd_min_target));
+	tcp_snd_cwnd_set(tp, min(cap, tcp_snd_cwnd(tp)));
+}
+
+/* How should we multiplicatively cut bw or inflight limits based on ECN? */
+static u32 bbr_ecn_cut(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	return BBR_UNIT -
+		((bbr->ecn_alpha * bbr_param(sk, ecn_factor)) >> BBR_SCALE);
+}
+
+/* Init lower bounds if have not inited yet. */
+static void bbr_init_lower_bounds(struct sock *sk, bool init_bw)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (init_bw && bbr->bw_lo == ~0U)
+		bbr->bw_lo = bbr_max_bw(sk);
+	if (bbr->inflight_lo == ~0U)
+		bbr->inflight_lo = tcp_snd_cwnd(tp);
+}
+
+/* Reduce bw and inflight to (1 - beta). */
+static void bbr_loss_lower_bounds(struct sock *sk, u32 *bw, u32 *inflight)
+{
+	struct bbr* bbr = inet_csk_ca(sk);
+	u32 loss_cut = BBR_UNIT - bbr_param(sk, beta);
+
+	*bw = max_t(u32, bbr->bw_latest,
+		    (u64)bbr->bw_lo * loss_cut >> BBR_SCALE);
+	*inflight = max_t(u32, bbr->inflight_latest,
+			  (u64)bbr->inflight_lo * loss_cut >> BBR_SCALE);
+}
+
+/* Reduce inflight to (1 - alpha*ecn_factor). */
+static void bbr_ecn_lower_bounds(struct sock *sk, u32 *inflight)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 ecn_cut = bbr_ecn_cut(sk);
+
+	*inflight = (u64)bbr->inflight_lo * ecn_cut >> BBR_SCALE;
+}
+
+/* Estimate a short-term lower bound on the capacity available now, based
+ * on measurements of the current delivery process and recent history. When we
+ * are seeing loss/ECN at times when we are not probing bw, then conservatively
+ * move toward flow balance by multiplicatively cutting our short-term
+ * estimated safe rate and volume of data (bw_lo and inflight_lo). We use a
+ * multiplicative decrease in order to converge to a lower capacity in time
+ * logarithmic in the magnitude of the decrease.
+ *
+ * However, we do not cut our short-term estimates lower than the current rate
+ * and volume of delivered data from this round trip, since from the current
+ * delivery process we can estimate the measured capacity available now.
+ *
+ * Anything faster than that approach would knowingly risk high loss, which can
+ * cause low bw for Reno/CUBIC and high loss recovery latency for
+ * request/response flows using any congestion control.
+ */
+static void bbr_adapt_lower_bounds(struct sock *sk,
+				    const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 ecn_inflight_lo = ~0U;
+
+	/* We only use lower-bound estimates when not probing bw.
+	 * When probing we need to push inflight higher to probe bw.
+	 */
+	if (bbr_is_probing_bandwidth(sk))
+		return;
+
+	/* ECN response. */
+	if (bbr->ecn_in_round && bbr_param(sk, ecn_factor)) {
+		bbr_init_lower_bounds(sk, false);
+		bbr_ecn_lower_bounds(sk, &ecn_inflight_lo);
+	}
+
+	/* Loss response. */
+	if (bbr->loss_in_round) {
+		bbr_init_lower_bounds(sk, true);
+		bbr_loss_lower_bounds(sk, &bbr->bw_lo, &bbr->inflight_lo);
+	}
+
+	/* Adjust to the lower of the levels implied by loss/ECN. */
+	bbr->inflight_lo = min(bbr->inflight_lo, ecn_inflight_lo);
+	bbr->bw_lo = max(1U, bbr->bw_lo);
+}
+
+/* Reset any short-term lower-bound adaptation to congestion, so that we can
+ * push our inflight up.
+ */
+static void bbr_reset_lower_bounds(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->bw_lo = ~0U;
+	bbr->inflight_lo = ~0U;
+}
+
+/* After bw probing (STARTUP/PROBE_UP), reset signals before entering a state
+ * machine phase where we adapt our lower bound based on congestion signals.
+ */
+static void bbr_reset_congestion_signals(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->loss_in_round = 0;
+	bbr->ecn_in_round = 0;
+	bbr->loss_in_cycle = 0;
+	bbr->ecn_in_cycle = 0;
+	bbr->bw_latest = 0;
+	bbr->inflight_latest = 0;
+}
+
+static void bbr_exit_loss_recovery(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd));
+	bbr->try_fast_path = 0; /* bound cwnd using latest model */
+}
+
+/* Update rate and volume of delivered data from latest round trip. */
+static void bbr_update_latest_delivery_signals(
+	struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->loss_round_start = 0;
+	if (rs->interval_us <= 0 || !rs->acked_sacked)
+		return; /* Not a valid observation */
+
+	bbr->bw_latest       = max_t(u32, bbr->bw_latest,       ctx->sample_bw);
+	bbr->inflight_latest = max_t(u32, bbr->inflight_latest, rs->delivered);
+
+	if (!before(rs->prior_delivered, bbr->loss_round_delivered)) {
+		bbr->loss_round_delivered = tp->delivered;
+		bbr->loss_round_start = 1;  /* mark start of new round trip */
+	}
+}
+
+/* Once per round, reset filter for latest rate and volume of delivered data. */
+static void bbr_advance_latest_delivery_signals(
+	struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	/* If ACK matches a TLP retransmit, persist the filter. If we detect
+	 * that a TLP retransmit plugged a tail loss, we'll want to remember
+	 * how much data the path delivered before the tail loss.
+	 */
+	if (bbr->loss_round_start && !rs->is_acking_tlp_retrans_seq) {
+		bbr->bw_latest = ctx->sample_bw;
+		bbr->inflight_latest = rs->delivered;
+	}
+}
+
+/* Update (most of) our congestion signals: track the recent rate and volume of
+ * delivered data, presence of loss, and EWMA degree of ECN marking.
+ */
+static void bbr_update_congestion_signals(
+	struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx)
 {
 	struct bbr *bbr = inet_csk_ca(sk);
+	u64 bw;
+
+	if (rs->interval_us <= 0 || !rs->acked_sacked)
+		return; /* Not a valid observation */
+	bw = ctx->sample_bw;
 
-	bbr->full_bw = 0;   /* spurious slow-down; reset full pipe detection */
+	if (!rs->is_app_limited || bw >= bbr_max_bw(sk))
+		bbr_take_max_bw_sample(sk, bw);
+
+	bbr->loss_in_round |= (rs->losses > 0);
+
+	if (!bbr->loss_round_start)
+		return;		/* skip the per-round-trip updates */
+	/* Now do per-round-trip updates. */
+	bbr_adapt_lower_bounds(sk, rs);
+
+	bbr->loss_in_round = 0;
+	bbr->ecn_in_round  = 0;
+}
+
+/* Bandwidth probing can cause loss. To help coexistence with loss-based
+ * congestion control we spread out our probing in a Reno-conscious way. Due to
+ * the shape of the Reno sawtooth, the time required between loss epochs for an
+ * idealized Reno flow is a number of round trips that is the BDP of that
+ * flow. We count packet-timed round trips directly, since measured RTT can
+ * vary widely, and Reno is driven by packet-timed round trips.
+ */
+static bool bbr_is_reno_coexistence_probe_time(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 rounds;
+
+	/* Random loss can shave some small percentage off of our inflight
+	 * in each round. To survive this, flows need robust periodic probes.
+	 */
+	rounds = min_t(u32, bbr_param(sk, bw_probe_max_rounds), bbr_target_inflight(sk));
+	return bbr->rounds_since_probe >= rounds;
+}
+
+/* How long do we want to wait before probing for bandwidth (and risking
+ * loss)? We randomize the wait, for better mixing and fairness convergence.
+ *
+ * We bound the Reno-coexistence inter-bw-probe time to be 62-63 round trips.
+ * This is calculated to allow fairness with a 25Mbps, 30ms Reno flow,
+ * (eg 4K video to a broadband user):
+ *   BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets
+ *
+ * We bound the BBR-native inter-bw-probe wall clock time to be:
+ *  (a) higher than 2 sec: to try to avoid causing loss for a long enough time
+ *      to allow Reno at 30ms to get 4K video bw, the inter-bw-probe time must
+ *      be at least: 25Mbps * .030sec / (1514bytes) * 0.030sec = 1.9secs
+ *  (b) lower than 3 sec: to ensure flows can start probing in a reasonable
+ *      amount of time to discover unutilized bw on human-scale interactive
+ *      time-scales (e.g. perhaps traffic from a web page download that we
+ *      were competing with is now complete).
+ */
+static void bbr_pick_probe_wait(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	/* Decide the random round-trip bound for wait until probe: */
+	bbr->rounds_since_probe =
+		get_random_u32_below(bbr_param(sk, bw_probe_rand_rounds));
+	/* Decide the random wall clock bound for wait until probe: */
+	bbr->probe_wait_us = bbr_param(sk, bw_probe_base_us) +
+			     get_random_u32_below(bbr_param(sk, bw_probe_rand_us));
+}
+
+static void bbr_set_cycle_idx(struct sock *sk, int cycle_idx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->cycle_idx = cycle_idx;
+	/* New phase, so need to update cwnd and pacing rate. */
+	bbr->try_fast_path = 0;
+}
+
+/* Send at estimated bw to fill the pipe, but not queue. We need this phase
+ * before PROBE_UP, because as soon as we send faster than the available bw
+ * we will start building a queue, and if the buffer is shallow we can cause
+ * loss. If we do not fill the pipe before we cause this loss, our bw_hi and
+ * inflight_hi estimates will underestimate.
+ */
+static void bbr_start_bw_probe_refill(struct sock *sk, u32 bw_probe_up_rounds)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr_reset_lower_bounds(sk);
+	bbr->bw_probe_up_rounds = bw_probe_up_rounds;
+	bbr->bw_probe_up_acks = 0;
+	bbr->stopped_risky_probe = 0;
+	bbr->ack_phase = BBR_ACKS_REFILLING;
+	bbr->next_rtt_delivered = tp->delivered;
+	bbr_set_cycle_idx(sk, BBR_BW_PROBE_REFILL);
+}
+
+/* Now probe max deliverable data rate and volume. */
+static void bbr_start_bw_probe_up(struct sock *sk, struct bbr_context *ctx)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->ack_phase = BBR_ACKS_PROBE_STARTING;
+	bbr->next_rtt_delivered = tp->delivered;
+	bbr->cycle_mstamp = tp->tcp_mstamp;
+	bbr_reset_full_bw(sk);
+	bbr->full_bw = ctx->sample_bw;
+	bbr_set_cycle_idx(sk, BBR_BW_PROBE_UP);
+	bbr_raise_inflight_hi_slope(sk);
+}
+
+/* Start a new PROBE_BW probing cycle of some wall clock length. Pick a wall
+ * clock time at which to probe beyond an inflight that we think to be
+ * safe. This will knowingly risk packet loss, so we want to do this rarely, to
+ * keep packet loss rates low. Also start a round-trip counter, to probe faster
+ * if we estimate a Reno flow at our BDP would probe faster.
+ */
+static void bbr_start_bw_probe_down(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr_reset_congestion_signals(sk);
+	bbr->bw_probe_up_cnt = ~0U;     /* not growing inflight_hi any more */
+	bbr_pick_probe_wait(sk);
+	bbr->cycle_mstamp = tp->tcp_mstamp;		/* start wall clock */
+	bbr->ack_phase = BBR_ACKS_PROBE_STOPPING;
+	bbr->next_rtt_delivered = tp->delivered;
+	bbr_set_cycle_idx(sk, BBR_BW_PROBE_DOWN);
+}
+
+/* Cruise: maintain what we estimate to be a neutral, conservative
+ * operating point, without attempting to probe up for bandwidth or down for
+ * RTT, and only reducing inflight in response to loss/ECN signals.
+ */
+static void bbr_start_bw_probe_cruise(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr->inflight_lo != ~0U)
+		bbr->inflight_lo = min(bbr->inflight_lo, bbr->inflight_hi);
+
+	bbr_set_cycle_idx(sk, BBR_BW_PROBE_CRUISE);
+}
+
+/* Loss and/or ECN rate is too high while probing.
+ * Adapt (once per bw probe) by cutting inflight_hi and then restarting cycle.
+ */
+static void bbr_handle_inflight_too_high(struct sock *sk,
+					  const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	const u32 beta = bbr_param(sk, beta);
+
+	bbr->prev_probe_too_high = 1;
+	bbr->bw_probe_samples = 0;  /* only react once per probe */
+	/* If we are app-limited then we are not robustly
+	 * probing the max volume of inflight data we think
+	 * might be safe (analogous to how app-limited bw
+	 * samples are not known to be robustly probing bw).
+	 */
+	if (!rs->is_app_limited) {
+		bbr->inflight_hi = max_t(u32, rs->tx_in_flight,
+					 (u64)bbr_target_inflight(sk) *
+					 (BBR_UNIT - beta) >> BBR_SCALE);
+	}
+	if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP)
+		bbr_start_bw_probe_down(sk);
+}
+
+/* If we're seeing bw and loss samples reflecting our bw probing, adapt
+ * using the signals we see. If loss or ECN mark rate gets too high, then adapt
+ * inflight_hi downward. If we're able to push inflight higher without such
+ * signals, push higher: adapt inflight_hi upward.
+ */
+static bool bbr_adapt_upper_bounds(struct sock *sk,
+				    const struct rate_sample *rs,
+				    struct bbr_context *ctx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	/* Track when we'll see bw/loss samples resulting from our bw probes. */
+	if (bbr->ack_phase == BBR_ACKS_PROBE_STARTING && bbr->round_start)
+		bbr->ack_phase = BBR_ACKS_PROBE_FEEDBACK;
+	if (bbr->ack_phase == BBR_ACKS_PROBE_STOPPING && bbr->round_start) {
+		/* End of samples from bw probing phase. */
+		bbr->bw_probe_samples = 0;
+		bbr->ack_phase = BBR_ACKS_INIT;
+		/* At this point in the cycle, our current bw sample is also
+		 * our best recent chance at finding the highest available bw
+		 * for this flow. So now is the best time to forget the bw
+		 * samples from the previous cycle, by advancing the window.
+		 */
+		if (bbr->mode == BBR_PROBE_BW && !rs->is_app_limited)
+			bbr_advance_max_bw_filter(sk);
+		/* If we had an inflight_hi, then probed and pushed inflight all
+		 * the way up to hit that inflight_hi without seeing any
+		 * high loss/ECN in all the resulting ACKs from that probing,
+		 * then probe up again, this time letting inflight persist at
+		 * inflight_hi for a round trip, then accelerating beyond.
+		 */
+		if (bbr->mode == BBR_PROBE_BW &&
+		    bbr->stopped_risky_probe && !bbr->prev_probe_too_high) {
+			bbr_start_bw_probe_refill(sk, 0);
+			return true;  /* yes, decided state transition */
+		}
+	}
+	if (bbr_is_inflight_too_high(sk, rs)) {
+		if (bbr->bw_probe_samples)  /*  sample is from bw probing? */
+			bbr_handle_inflight_too_high(sk, rs);
+	} else {
+		/* Loss/ECN rate is declared safe. Adjust upper bound upward. */
+
+		if (bbr->inflight_hi == ~0U)
+			return false;   /* no excess queue signals yet */
+
+		/* To be resilient to random loss, we must raise bw/inflight_hi
+		 * if we observe in any phase that a higher level is safe.
+		 */
+		if (rs->tx_in_flight > bbr->inflight_hi) {
+			bbr->inflight_hi = rs->tx_in_flight;
+		}
+
+		if (bbr->mode == BBR_PROBE_BW &&
+		    bbr->cycle_idx == BBR_BW_PROBE_UP)
+			bbr_probe_inflight_hi_upward(sk, rs);
+	}
+
+	return false;
+}
+
+/* Check if it's time to probe for bandwidth now, and if so, kick it off. */
+static bool bbr_check_time_to_probe_bw(struct sock *sk,
+					const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 n;
+
+	/* If we seem to be at an operating point where we are not seeing loss
+	 * but we are seeing ECN marks, then when the ECN marks cease we reprobe
+	 * quickly (in case cross-traffic has ceased and freed up bw).
+	 */
+	if (bbr_param(sk, ecn_reprobe_gain) && bbr->ecn_eligible &&
+	    bbr->ecn_in_cycle && !bbr->loss_in_cycle &&
+	    inet_csk(sk)->icsk_ca_state == TCP_CA_Open) {
+		/* Calculate n so that when bbr_raise_inflight_hi_slope()
+		 * computes growth_this_round as 2^n it will be roughly the
+		 * desired volume of data (inflight_hi*ecn_reprobe_gain).
+		 */
+		n = ilog2((((u64)bbr->inflight_hi *
+			    bbr_param(sk, ecn_reprobe_gain)) >> BBR_SCALE));
+		bbr_start_bw_probe_refill(sk, n);
+		return true;
+	}
+
+	if (bbr_has_elapsed_in_phase(sk, bbr->probe_wait_us) ||
+	    bbr_is_reno_coexistence_probe_time(sk)) {
+		bbr_start_bw_probe_refill(sk, 0);
+		return true;
+	}
+	return false;
+}
+
+/* Is it time to transition from PROBE_DOWN to PROBE_CRUISE? */
+static bool bbr_check_time_to_cruise(struct sock *sk, u32 inflight, u32 bw)
+{
+	/* Always need to pull inflight down to leave headroom in queue. */
+	if (inflight > bbr_inflight_with_headroom(sk))
+		return false;
+
+	return inflight <= bbr_inflight(sk, bw, BBR_UNIT);
+}
+
+/* PROBE_BW state machine: cruise, refill, probe for bw, or drain? */
+static void bbr_update_cycle_phase(struct sock *sk,
+				    const struct rate_sample *rs,
+				    struct bbr_context *ctx)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	bool is_bw_probe_done = false;
+	u32 inflight, bw;
+
+	if (!bbr_full_bw_reached(sk))
+		return;
+
+	/* In DRAIN, PROBE_BW, or PROBE_RTT, adjust upper bounds. */
+	if (bbr_adapt_upper_bounds(sk, rs, ctx))
+		return;		/* already decided state transition */
+
+	if (bbr->mode != BBR_PROBE_BW)
+		return;
+
+	inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight);
+	bw = bbr_max_bw(sk);
+
+	switch (bbr->cycle_idx) {
+	/* First we spend most of our time cruising with a pacing_gain of 1.0,
+	 * which paces at the estimated bw, to try to fully use the pipe
+	 * without building queue. If we encounter loss/ECN marks, we adapt
+	 * by slowing down.
+	 */
+	case BBR_BW_PROBE_CRUISE:
+		if (bbr_check_time_to_probe_bw(sk, rs))
+			return;		/* already decided state transition */
+		break;
+
+	/* After cruising, when it's time to probe, we first "refill": we send
+	 * at the estimated bw to fill the pipe, before probing higher and
+	 * knowingly risking overflowing the bottleneck buffer (causing loss).
+	 */
+	case BBR_BW_PROBE_REFILL:
+		if (bbr->round_start) {
+			/* After one full round trip of sending in REFILL, we
+			 * start to see bw samples reflecting our REFILL, which
+			 * may be putting too much data in flight.
+			 */
+			bbr->bw_probe_samples = 1;
+			bbr_start_bw_probe_up(sk, ctx);
+		}
+		break;
+
+	/* After we refill the pipe, we probe by using a pacing_gain > 1.0, to
+	 * probe for bw. If we have not seen loss/ECN, we try to raise inflight
+	 * to at least pacing_gain*BDP; note that this may take more than
+	 * min_rtt if min_rtt is small (e.g. on a LAN).
+	 *
+	 * We terminate PROBE_UP bandwidth probing upon any of the following:
+	 *
+	 * (1) We've pushed inflight up to hit the inflight_hi target set in the
+	 *     most recent previous bw probe phase. Thus we want to start
+	 *     draining the queue immediately because it's very likely the most
+	 *     recently sent packets will fill the queue and cause drops.
+	 * (2) If inflight_hi has not limited bandwidth growth recently, and
+	 *     yet delivered bandwidth has not increased much recently
+	 *     (bbr->full_bw_now).
+	 * (3) Loss filter says loss rate is "too high".
+	 * (4) ECN filter says ECN mark rate is "too high".
+	 *
+	 * (1) (2) checked here, (3) (4) checked in bbr_is_inflight_too_high()
+	 */
+	case BBR_BW_PROBE_UP:
+		if (bbr->prev_probe_too_high &&
+		    inflight >= bbr->inflight_hi) {
+			bbr->stopped_risky_probe = 1;
+			is_bw_probe_done = true;
+		} else {
+			if (tp->is_cwnd_limited &&
+			    tcp_snd_cwnd(tp) >= bbr->inflight_hi) {
+				/* inflight_hi is limiting bw growth */
+				bbr_reset_full_bw(sk);
+				bbr->full_bw = ctx->sample_bw;
+			} else if (bbr->full_bw_now) {
+				/* Plateau in estimated bw. Pipe looks full. */
+				is_bw_probe_done = true;
+			}
+		}
+		if (is_bw_probe_done) {
+			bbr->prev_probe_too_high = 0;  /* no loss/ECN (yet) */
+			bbr_start_bw_probe_down(sk);  /* restart w/ down */
+		}
+		break;
+
+	/* After probing in PROBE_UP, we have usually accumulated some data in
+	 * the bottleneck buffer (if bw probing didn't find more bw). We next
+	 * enter PROBE_DOWN to try to drain any excess data from the queue. To
+	 * do this, we use a pacing_gain < 1.0. We hold this pacing gain until
+	 * our inflight is less then that target cruising point, which is the
+	 * minimum of (a) the amount needed to leave headroom, and (b) the
+	 * estimated BDP. Once inflight falls to match the target, we estimate
+	 * the queue is drained; persisting would underutilize the pipe.
+	 */
+	case BBR_BW_PROBE_DOWN:
+		if (bbr_check_time_to_probe_bw(sk, rs))
+			return;		/* already decided state transition */
+		if (bbr_check_time_to_cruise(sk, inflight, bw))
+			bbr_start_bw_probe_cruise(sk);
+		break;
+
+	default:
+		WARN_ONCE(1, "BBR invalid cycle index %u\n", bbr->cycle_idx);
+	}
+}
+
+/* Exiting PROBE_RTT, so return to bandwidth probing in STARTUP or PROBE_BW. */
+static void bbr_exit_probe_rtt(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr_reset_lower_bounds(sk);
+	if (bbr_full_bw_reached(sk)) {
+		bbr->mode = BBR_PROBE_BW;
+		/* Raising inflight after PROBE_RTT may cause loss, so reset
+		 * the PROBE_BW clock and schedule the next bandwidth probe for
+		 * a friendly and randomized future point in time.
+		 */
+		bbr_start_bw_probe_down(sk);
+		/* Since we are exiting PROBE_RTT, we know inflight is
+		 * below our estimated BDP, so it is reasonable to cruise.
+		 */
+		bbr_start_bw_probe_cruise(sk);
+	} else {
+		bbr->mode = BBR_STARTUP;
+	}
+}
+
+/* Exit STARTUP based on loss rate > 1% and loss gaps in round >= N. Wait until
+ * the end of the round in recovery to get a good estimate of how many packets
+ * have been lost, and how many we need to drain with a low pacing rate.
+ */
+static void bbr_check_loss_too_high_in_startup(struct sock *sk,
+						const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr_full_bw_reached(sk))
+		return;
+
+	/* For STARTUP exit, check the loss rate at the end of each round trip
+	 * of Recovery episodes in STARTUP. We check the loss rate at the end
+	 * of the round trip to filter out noisy/low loss and have a better
+	 * sense of inflight (extent of loss), so we can drain more accurately.
+	 */
+	if (rs->losses && bbr->loss_events_in_round < 0xf)
+		bbr->loss_events_in_round++;  /* update saturating counter */
+	if (bbr_param(sk, full_loss_cnt) && bbr->loss_round_start &&
+	    inet_csk(sk)->icsk_ca_state == TCP_CA_Recovery &&
+	    bbr->loss_events_in_round >= bbr_param(sk, full_loss_cnt) &&
+	    bbr_is_inflight_too_high(sk, rs)) {
+		bbr_handle_queue_too_high_in_startup(sk);
+		return;
+	}
+	if (bbr->loss_round_start)
+		bbr->loss_events_in_round = 0;
+}
+
+/* Estimate when the pipe is full, using the change in delivery rate: BBR
+ * estimates bw probing filled the pipe if the estimated bw hasn't changed by
+ * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited
+ * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the
+ * higher rwin, 3: we get higher delivery rate samples. Or transient
+ * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar
+ * design goal, but uses delay and inter-ACK spacing instead of bandwidth.
+ */
+static void bbr_check_full_bw_reached(struct sock *sk,
+				       const struct rate_sample *rs,
+				       struct bbr_context *ctx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 bw_thresh, full_cnt, thresh;
+
+	if (bbr->full_bw_now || rs->is_app_limited)
+		return;
+
+	thresh = bbr_param(sk, full_bw_thresh);
+	full_cnt = bbr_param(sk, full_bw_cnt);
+	bw_thresh = (u64)bbr->full_bw * thresh >> BBR_SCALE;
+	if (ctx->sample_bw >= bw_thresh) {
+		bbr_reset_full_bw(sk);
+		bbr->full_bw = ctx->sample_bw;
+		return;
+	}
+	if (!bbr->round_start)
+		return;
+	++bbr->full_bw_cnt;
+	bbr->full_bw_now = bbr->full_bw_cnt >= full_cnt;
+	bbr->full_bw_reached |= bbr->full_bw_now;
+}
+
+/* If pipe is probably full, drain the queue and then enter steady-state. */
+static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs,
+			    struct bbr_context *ctx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) {
+		bbr->mode = BBR_DRAIN;	/* drain queue we created */
+		/* Set ssthresh to export purely for monitoring, to signal
+		 * completion of initial STARTUP by setting to a non-
+		 * TCP_INFINITE_SSTHRESH value (ssthresh is not used by BBR).
+		 */
+		tcp_sk(sk)->snd_ssthresh =
+				bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT);
+		bbr_reset_congestion_signals(sk);
+	}	/* fall through to check if in-flight is already small: */
+	if (bbr->mode == BBR_DRAIN &&
+	    bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <=
+	    bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) {
+		bbr->mode = BBR_PROBE_BW;
+		bbr_start_bw_probe_down(sk);
+	}
+}
+
+static void bbr_update_model(struct sock *sk, const struct rate_sample *rs,
+			      struct bbr_context *ctx)
+{
+	bbr_update_congestion_signals(sk, rs, ctx);
+	bbr_update_ack_aggregation(sk, rs);
+	bbr_check_loss_too_high_in_startup(sk, rs);
+	bbr_check_full_bw_reached(sk, rs, ctx);
+	bbr_check_drain(sk, rs, ctx);
+	bbr_update_cycle_phase(sk, rs, ctx);
+	bbr_update_min_rtt(sk, rs);
+}
+
+/* Fast path for app-limited case.
+ *
+ * On each ack, we execute bbr state machine, which primarily consists of:
+ * 1) update model based on new rate sample, and
+ * 2) update control based on updated model or state change.
+ *
+ * There are certain workload/scenarios, e.g. app-limited case, where
+ * either we can skip updating model or we can skip update of both model
+ * as well as control. This provides signifcant softirq cpu savings for
+ * processing incoming acks.
+ *
+ * In case of app-limited, if there is no congestion (loss/ecn) and
+ * if observed bw sample is less than current estimated bw, then we can
+ * skip some of the computation in bbr state processing:
+ *
+ * - if there is no rtt/mode/phase change: In this case, since all the
+ *   parameters of the network model are constant, we can skip model
+ *   as well control update.
+ *
+ * - else we can skip rest of the model update. But we still need to
+ *   update the control to account for the new rtt/mode/phase.
+ *
+ * Returns whether we can take fast path or not.
+ */
+static bool bbr_run_fast_path(struct sock *sk, bool *update_model,
+		const struct rate_sample *rs, struct bbr_context *ctx)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 prev_min_rtt_us, prev_mode;
+
+	if (bbr_param(sk, fast_path) && bbr->try_fast_path &&
+	    rs->is_app_limited && ctx->sample_bw < bbr_max_bw(sk) &&
+	    !bbr->loss_in_round && !bbr->ecn_in_round ) {
+		prev_mode = bbr->mode;
+		prev_min_rtt_us = bbr->min_rtt_us;
+		bbr_check_drain(sk, rs, ctx);
+		bbr_update_cycle_phase(sk, rs, ctx);
+		bbr_update_min_rtt(sk, rs);
+
+		if (bbr->mode == prev_mode &&
+		    bbr->min_rtt_us == prev_min_rtt_us &&
+		    bbr->try_fast_path) {
+			return true;
+		}
+
+		/* Skip model update, but control still needs to be updated */
+		*update_model = false;
+	}
+	return false;
+}
+
+__bpf_kfunc static void bbr_main(struct sock *sk, const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	struct bbr_context ctx = { 0 };
+	bool update_model = true;
+	u32 bw, round_delivered;
+	int ce_ratio = -1;
+
+	round_delivered = bbr_update_round_start(sk, rs, &ctx);
+	if (bbr->round_start) {
+		bbr->rounds_since_probe =
+			min_t(s32, bbr->rounds_since_probe + 1, 0xFF);
+		ce_ratio = bbr_update_ecn_alpha(sk);
+	}
+	bbr_plb(sk, rs, ce_ratio);
+
+	bbr->ecn_in_round  |= (bbr->ecn_eligible && rs->is_ece);
+	bbr_calculate_bw_sample(sk, rs, &ctx);
+	bbr_update_latest_delivery_signals(sk, rs, &ctx);
+
+	if (bbr_run_fast_path(sk, &update_model, rs, &ctx))
+		goto out;
+
+	if (update_model)
+		bbr_update_model(sk, rs, &ctx);
+
+	bbr_update_gains(sk);
+	bw = bbr_bw(sk);
+	bbr_set_pacing_rate(sk, bw, bbr->pacing_gain);
+	bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain,
+		     tcp_snd_cwnd(tp), &ctx);
+	bbr_bound_cwnd_for_inflight_model(sk);
+
+out:
+	bbr_advance_latest_delivery_signals(sk, rs, &ctx);
+	bbr->prev_ca_state = inet_csk(sk)->icsk_ca_state;
+	bbr->loss_in_cycle |= rs->lost > 0;
+	bbr->ecn_in_cycle  |= rs->delivered_ce > 0;
+}
+
+__bpf_kfunc static void bbr_init(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->initialized = 1;
+
+	bbr->init_cwnd = min(0x7FU, tcp_snd_cwnd(tp));
+	bbr->prior_cwnd = tp->prior_cwnd;
+	tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
+	bbr->next_rtt_delivered = tp->delivered;
+	bbr->prev_ca_state = TCP_CA_Open;
+
+	bbr->probe_rtt_done_stamp = 0;
+	bbr->probe_rtt_round_done = 0;
+	bbr->probe_rtt_min_us = tcp_min_rtt(tp);
+	bbr->probe_rtt_min_stamp = tcp_jiffies32;
+	bbr->min_rtt_us = tcp_min_rtt(tp);
+	bbr->min_rtt_stamp = tcp_jiffies32;
+
+	bbr->has_seen_rtt = 0;
+	bbr_init_pacing_rate_from_rtt(sk);
+
+	bbr->round_start = 0;
+	bbr->idle_restart = 0;
+	bbr->full_bw_reached = 0;
+	bbr->full_bw = 0;
 	bbr->full_bw_cnt = 0;
-	bbr_reset_lt_bw_sampling(sk);
-	return tcp_snd_cwnd(tcp_sk(sk));
+	bbr->cycle_mstamp = 0;
+	bbr->cycle_idx = 0;
+
+	bbr_reset_startup_mode(sk);
+
+	bbr->ack_epoch_mstamp = tp->tcp_mstamp;
+	bbr->ack_epoch_acked = 0;
+	bbr->extra_acked_win_rtts = 0;
+	bbr->extra_acked_win_idx = 0;
+	bbr->extra_acked[0] = 0;
+	bbr->extra_acked[1] = 0;
+
+	bbr->ce_state = 0;
+	bbr->prior_rcv_nxt = tp->rcv_nxt;
+	bbr->try_fast_path = 0;
+
+	cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED);
+
+	/* Start sampling ECN mark rate after first full flight is ACKed: */
+	bbr->loss_round_delivered = tp->delivered + 1;
+	bbr->loss_round_start = 0;
+	bbr->undo_bw_lo = 0;
+	bbr->undo_inflight_lo = 0;
+	bbr->undo_inflight_hi = 0;
+	bbr->loss_events_in_round = 0;
+	bbr->startup_ecn_rounds = 0;
+	bbr_reset_congestion_signals(sk);
+	bbr->bw_lo = ~0U;
+	bbr->bw_hi[0] = 0;
+	bbr->bw_hi[1] = 0;
+	bbr->inflight_lo = ~0U;
+	bbr->inflight_hi = ~0U;
+	bbr_reset_full_bw(sk);
+	bbr->bw_probe_up_cnt = ~0U;
+	bbr->bw_probe_up_acks = 0;
+	bbr->bw_probe_up_rounds = 0;
+	bbr->probe_wait_us = 0;
+	bbr->stopped_risky_probe = 0;
+	bbr->ack_phase = BBR_ACKS_INIT;
+	bbr->rounds_since_probe = 0;
+	bbr->bw_probe_samples = 0;
+	bbr->prev_probe_too_high = 0;
+	bbr->ecn_eligible = 0;
+	bbr->ecn_alpha = bbr_param(sk, ecn_alpha_init);
+	bbr->alpha_last_delivered = 0;
+	bbr->alpha_last_delivered_ce = 0;
+	bbr->plb.pause_until = 0;
+
+	tp->fast_ack_mode = bbr_fast_ack_mode ? 1 : 0;
+
+	if (bbr_can_use_ecn(sk))
+		tp->ecn_flags |= TCP_ECN_ECT_PERMANENT;
+}
+
+/* BBR marks the current round trip as a loss round. */
+static void bbr_note_loss(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	/* Capture "current" data over the full round trip of loss, to
+	 * have a better chance of observing the full capacity of the path.
+	 */
+	if (!bbr->loss_in_round)  /* first loss in this round trip? */
+		bbr->loss_round_delivered = tp->delivered;  /* set round trip */
+	bbr->loss_in_round = 1;
+	bbr->loss_in_cycle = 1;
 }
 
-/* Entering loss recovery, so save cwnd for when we exit or undo recovery. */
+/* Core TCP stack informs us that the given skb was just marked lost. */
+__bpf_kfunc static void bbr_skb_marked_lost(struct sock *sk,
+					    const struct sk_buff *skb)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
+	struct rate_sample rs = {};
+
+	bbr_note_loss(sk);
+
+	if (!bbr->bw_probe_samples)
+		return;  /* not an skb sent while probing for bandwidth */
+	if (unlikely(!scb->tx.delivered_mstamp))
+		return;  /* skb was SACKed, reneged, marked lost; ignore it */
+	/* We are probing for bandwidth. Construct a rate sample that
+	 * estimates what happened in the flight leading up to this lost skb,
+	 * then see if the loss rate went too high, and if so at which packet.
+	 */
+	rs.tx_in_flight = scb->tx.in_flight;
+	rs.lost = tp->lost - scb->tx.lost;
+	rs.is_app_limited = scb->tx.is_app_limited;
+	if (bbr_is_inflight_too_high(sk, &rs)) {
+		rs.tx_in_flight = bbr_inflight_hi_from_lost_skb(sk, &rs, skb);
+		bbr_handle_inflight_too_high(sk, &rs);
+	}
+}
+
+static void bbr_run_loss_probe_recovery(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	struct rate_sample rs = {0};
+
+	bbr_note_loss(sk);
+
+	if (!bbr->bw_probe_samples)
+		return;  /* not sent while probing for bandwidth */
+	/* We are probing for bandwidth. Construct a rate sample that
+	 * estimates what happened in the flight leading up to this
+	 * loss, then see if the loss rate went too high.
+	 */
+	rs.lost = 1;	/* TLP probe repaired loss of a single segment */
+	rs.tx_in_flight = bbr->inflight_latest + rs.lost;
+	rs.is_app_limited = tp->tlp_orig_data_app_limited;
+	if (bbr_is_inflight_too_high(sk, &rs))
+		bbr_handle_inflight_too_high(sk, &rs);
+}
+
+/* Revert short-term model if current loss recovery event was spurious. */
+__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr_reset_full_bw(sk); /* spurious slow-down; reset full bw detector */
+	bbr->loss_in_round = 0;
+
+	/* Revert to cwnd and other state saved before loss episode. */
+	bbr->bw_lo = max(bbr->bw_lo, bbr->undo_bw_lo);
+	bbr->inflight_lo = max(bbr->inflight_lo, bbr->undo_inflight_lo);
+	bbr->inflight_hi = max(bbr->inflight_hi, bbr->undo_inflight_hi);
+	bbr->try_fast_path = 0;  /* take slow path to set proper cwnd, pacing */
+	return bbr->prior_cwnd;
+}
+
+/* Entering loss recovery, so save state for when we undo recovery. */
 __bpf_kfunc static u32 bbr_ssthresh(struct sock *sk)
 {
+	struct bbr *bbr = inet_csk_ca(sk);
+
 	bbr_save_cwnd(sk);
+	/* For undo, save state that adapts based on loss signal. */
+	bbr->undo_bw_lo		= bbr->bw_lo;
+	bbr->undo_inflight_lo	= bbr->inflight_lo;
+	bbr->undo_inflight_hi	= bbr->inflight_hi;
 	return tcp_sk(sk)->snd_ssthresh;
 }
 
+static enum tcp_bbr_phase bbr_get_phase(struct bbr *bbr)
+{
+	switch (bbr->mode) {
+	case BBR_STARTUP:
+		return BBR_PHASE_STARTUP;
+	case BBR_DRAIN:
+		return BBR_PHASE_DRAIN;
+	case BBR_PROBE_BW:
+		break;
+	case BBR_PROBE_RTT:
+		return BBR_PHASE_PROBE_RTT;
+	default:
+		return BBR_PHASE_INVALID;
+	}
+	switch (bbr->cycle_idx) {
+	case BBR_BW_PROBE_UP:
+		return BBR_PHASE_PROBE_BW_UP;
+	case BBR_BW_PROBE_DOWN:
+		return BBR_PHASE_PROBE_BW_DOWN;
+	case BBR_BW_PROBE_CRUISE:
+		return BBR_PHASE_PROBE_BW_CRUISE;
+	case BBR_BW_PROBE_REFILL:
+		return BBR_PHASE_PROBE_BW_REFILL;
+	default:
+		return BBR_PHASE_INVALID;
+	}
+}
+
 static size_t bbr_get_info(struct sock *sk, u32 ext, int *attr,
-			   union tcp_cc_info *info)
+			    union tcp_cc_info *info)
 {
 	if (ext & (1 << (INET_DIAG_BBRINFO - 1)) ||
 	    ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
-		struct tcp_sock *tp = tcp_sk(sk);
 		struct bbr *bbr = inet_csk_ca(sk);
-		u64 bw = bbr_bw(sk);
-
-		bw = bw * tp->mss_cache * USEC_PER_SEC >> BW_SCALE;
-		memset(&info->bbr, 0, sizeof(info->bbr));
-		info->bbr.bbr_bw_lo		= (u32)bw;
-		info->bbr.bbr_bw_hi		= (u32)(bw >> 32);
-		info->bbr.bbr_min_rtt		= bbr->min_rtt_us;
-		info->bbr.bbr_pacing_gain	= bbr->pacing_gain;
-		info->bbr.bbr_cwnd_gain		= bbr->cwnd_gain;
+		u64 bw = bbr_bw_bytes_per_sec(sk, bbr_bw(sk));
+		u64 bw_hi = bbr_bw_bytes_per_sec(sk, bbr_max_bw(sk));
+		u64 bw_lo = bbr->bw_lo == ~0U ?
+			~0ULL : bbr_bw_bytes_per_sec(sk, bbr->bw_lo);
+		struct tcp_bbr_info *bbr_info = &info->bbr;
+
+		memset(bbr_info, 0, sizeof(*bbr_info));
+		bbr_info->bbr_bw_lo		= (u32)bw;
+		bbr_info->bbr_bw_hi		= (u32)(bw >> 32);
+		bbr_info->bbr_min_rtt		= bbr->min_rtt_us;
+		bbr_info->bbr_pacing_gain	= bbr->pacing_gain;
+		bbr_info->bbr_cwnd_gain		= bbr->cwnd_gain;
+		bbr_info->bbr_bw_hi_lsb		= (u32)bw_hi;
+		bbr_info->bbr_bw_hi_msb		= (u32)(bw_hi >> 32);
+		bbr_info->bbr_bw_lo_lsb		= (u32)bw_lo;
+		bbr_info->bbr_bw_lo_msb		= (u32)(bw_lo >> 32);
+		bbr_info->bbr_mode		= bbr->mode;
+		bbr_info->bbr_phase		= (__u8)bbr_get_phase(bbr);
+		bbr_info->bbr_version		= (__u8)BBR_VERSION;
+		bbr_info->bbr_inflight_lo	= bbr->inflight_lo;
+		bbr_info->bbr_inflight_hi	= bbr->inflight_hi;
+		bbr_info->bbr_extra_acked	= bbr_extra_acked(sk);
 		*attr = INET_DIAG_BBRINFO;
-		return sizeof(info->bbr);
+		return sizeof(*bbr_info);
 	}
 	return 0;
 }
 
 __bpf_kfunc static void bbr_set_state(struct sock *sk, u8 new_state)
 {
+	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
 
 	if (new_state == TCP_CA_Loss) {
-		struct rate_sample rs = { .losses = 1 };
 
 		bbr->prev_ca_state = TCP_CA_Loss;
-		bbr->full_bw = 0;
-		bbr->round_start = 1;	/* treat RTO like end of a round */
-		bbr_lt_bw_sampling(sk, &rs);
+		tcp_plb_update_state_upon_rto(sk, &bbr->plb);
+		/* The tcp_write_timeout() call to sk_rethink_txhash() likely
+		 * repathed this flow, so re-learn the min network RTT on the
+		 * new path:
+		 */
+		bbr_reset_full_bw(sk);
+		if (!bbr_is_probing_bandwidth(sk) && bbr->inflight_lo == ~0U) {
+			/* bbr_adapt_lower_bounds() needs cwnd before
+			 * we suffered an RTO, to update inflight_lo:
+			 */
+			bbr->inflight_lo =
+				max(tcp_snd_cwnd(tp), bbr->prior_cwnd);
+		}
+	} else if (bbr->prev_ca_state == TCP_CA_Loss &&
+		   new_state != TCP_CA_Loss) {
+		bbr_exit_loss_recovery(sk);
 	}
 }
 
+
 static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = {
-	.flags		= TCP_CONG_NON_RESTRICTED,
+	.flags		= TCP_CONG_NON_RESTRICTED | TCP_CONG_WANTS_CE_EVENTS,
 	.name		= "bbr",
 	.owner		= THIS_MODULE,
 	.init		= bbr_init,
 	.cong_control	= bbr_main,
 	.sndbuf_expand	= bbr_sndbuf_expand,
+	.skb_marked_lost = bbr_skb_marked_lost,
 	.undo_cwnd	= bbr_undo_cwnd,
 	.cwnd_event	= bbr_cwnd_event,
 	.ssthresh	= bbr_ssthresh,
-	.min_tso_segs	= bbr_min_tso_segs,
+	.tso_segs	= bbr_tso_segs,
 	.get_info	= bbr_get_info,
 	.set_state	= bbr_set_state,
 };
@@ -1161,10 +2361,11 @@ BTF_SET8_START(tcp_bbr_check_kfunc_ids)
 BTF_ID_FLAGS(func, bbr_init)
 BTF_ID_FLAGS(func, bbr_main)
 BTF_ID_FLAGS(func, bbr_sndbuf_expand)
+BTF_ID_FLAGS(func, bbr_skb_marked_lost)
 BTF_ID_FLAGS(func, bbr_undo_cwnd)
 BTF_ID_FLAGS(func, bbr_cwnd_event)
 BTF_ID_FLAGS(func, bbr_ssthresh)
-BTF_ID_FLAGS(func, bbr_min_tso_segs)
+BTF_ID_FLAGS(func, bbr_tso_segs)
 BTF_ID_FLAGS(func, bbr_set_state)
 #endif
 #endif
@@ -1199,5 +2400,12 @@ MODULE_AUTHOR("Van Jacobson <vanj@google.com>");
 MODULE_AUTHOR("Neal Cardwell <ncardwell@google.com>");
 MODULE_AUTHOR("Yuchung Cheng <ycheng@google.com>");
 MODULE_AUTHOR("Soheil Hassas Yeganeh <soheil@google.com>");
+MODULE_AUTHOR("Priyaranjan Jha <priyarjha@google.com>");
+MODULE_AUTHOR("Yousuk Seung <ysseung@google.com>");
+MODULE_AUTHOR("Kevin Yang <yyd@google.com>");
+MODULE_AUTHOR("Arjun Roy <arjunroy@google.com>");
+MODULE_AUTHOR("David Morley <morleyd@google.com>");
+
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_DESCRIPTION("TCP BBR (Bottleneck Bandwidth and RTT)");
+MODULE_VERSION(__stringify(BBR_VERSION));
diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
index 1b34050a7538..66d40449b3f4 100644
--- a/net/ipv4/tcp_cong.c
+++ b/net/ipv4/tcp_cong.c
@@ -241,6 +241,7 @@ void tcp_init_congestion_control(struct sock *sk)
 	struct inet_connection_sock *icsk = inet_csk(sk);
 
 	tcp_sk(sk)->prior_ssthresh = 0;
+	tcp_sk(sk)->fast_ack_mode = 0;
 	if (icsk->icsk_ca_ops->init)
 		icsk->icsk_ca_ops->init(sk);
 	if (tcp_ca_needs_ecn(sk))
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index df7b13f0e5e0..8415aa41524e 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -364,7 +364,7 @@ static void __tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb)
 			tcp_enter_quickack_mode(sk, 2);
 		break;
 	case INET_ECN_CE:
-		if (tcp_ca_needs_ecn(sk))
+		if (tcp_ca_wants_ce_events(sk))
 			tcp_ca_event(sk, CA_EVENT_ECN_IS_CE);
 
 		if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) {
@@ -375,7 +375,7 @@ static void __tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb)
 		tp->ecn_flags |= TCP_ECN_SEEN;
 		break;
 	default:
-		if (tcp_ca_needs_ecn(sk))
+		if (tcp_ca_wants_ce_events(sk))
 			tcp_ca_event(sk, CA_EVENT_ECN_NO_CE);
 		tp->ecn_flags |= TCP_ECN_SEEN;
 		break;
@@ -1112,7 +1112,12 @@ static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
  */
 static void tcp_notify_skb_loss_event(struct tcp_sock *tp, const struct sk_buff *skb)
 {
+	struct sock *sk = (struct sock *)tp;
+	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
+
 	tp->lost += tcp_skb_pcount(skb);
+	if (ca_ops->skb_marked_lost)
+		ca_ops->skb_marked_lost(sk, skb);
 }
 
 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb)
@@ -1493,6 +1498,17 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *prev,
 	WARN_ON_ONCE(tcp_skb_pcount(skb) < pcount);
 	tcp_skb_pcount_add(skb, -pcount);
 
+	/* Adjust tx.in_flight as pcount is shifted from skb to prev. */
+	if (WARN_ONCE(TCP_SKB_CB(skb)->tx.in_flight < pcount,
+		      "prev in_flight: %u skb in_flight: %u pcount: %u",
+		      TCP_SKB_CB(prev)->tx.in_flight,
+		      TCP_SKB_CB(skb)->tx.in_flight,
+		      pcount))
+		TCP_SKB_CB(skb)->tx.in_flight = 0;
+	else
+		TCP_SKB_CB(skb)->tx.in_flight -= pcount;
+	TCP_SKB_CB(prev)->tx.in_flight += pcount;
+
 	/* When we're adding to gso_segs == 1, gso_size will be zero,
 	 * in theory this shouldn't be necessary but as long as DSACK
 	 * code can come after this skb later on it's better to keep
@@ -3761,7 +3777,8 @@ static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq)
 /* This routine deals with acks during a TLP episode and ends an episode by
  * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack
  */
-static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
+static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag,
+				struct rate_sample *rs)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
@@ -3778,6 +3795,7 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
 		/* ACK advances: there was a loss, so reduce cwnd. Reset
 		 * tlp_high_seq in tcp_init_cwnd_reduction()
 		 */
+		tcp_ca_event(sk, CA_EVENT_TLP_RECOVERY);
 		tcp_init_cwnd_reduction(sk);
 		tcp_set_ca_state(sk, TCP_CA_CWR);
 		tcp_end_cwnd_reduction(sk);
@@ -3788,6 +3806,11 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
 			     FLAG_NOT_DUP | FLAG_DATA_SACKED))) {
 		/* Pure dupack: original and TLP probe arrived; no loss */
 		tp->tlp_high_seq = 0;
+	} else {
+		/* This ACK matches a TLP retransmit. We cannot yet tell if
+		 * this ACK is for the original or the TLP retransmit.
+		 */
+		rs->is_acking_tlp_retrans_seq = 1;
 	}
 }
 
@@ -3896,6 +3919,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 	prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una;
 	rs.prior_in_flight = tcp_packets_in_flight(tp);
+	tcp_rate_check_app_limited(sk);
 
 	/* ts_recent update must be made after we are sure that the packet
 	 * is in window.
@@ -3970,7 +3994,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	tcp_rack_update_reo_wnd(sk, &rs);
 
 	if (tp->tlp_high_seq)
-		tcp_process_tlp_ack(sk, ack, flag);
+		tcp_process_tlp_ack(sk, ack, flag, &rs);
 
 	if (tcp_ack_is_dubious(sk, flag)) {
 		if (!(flag & (FLAG_SND_UNA_ADVANCED |
@@ -3994,6 +4018,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	delivered = tcp_newly_delivered(sk, delivered, flag);
 	lost = tp->lost - lost;			/* freshly marked lost */
 	rs.is_ack_delayed = !!(flag & FLAG_ACK_MAYBE_DELAYED);
+	rs.is_ece = !!(flag & FLAG_ECE);
 	tcp_rate_gen(sk, delivered, lost, is_sack_reneg, sack_state.rate);
 	tcp_cong_control(sk, ack, delivered, flag, sack_state.rate);
 	tcp_xmit_recovery(sk, rexmit);
@@ -4013,7 +4038,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	tcp_ack_probe(sk);
 
 	if (tp->tlp_high_seq)
-		tcp_process_tlp_ack(sk, ack, flag);
+		tcp_process_tlp_ack(sk, ack, flag, &rs);
 	return 1;
 
 old_ack:
@@ -5664,13 +5689,14 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 
 	    /* More than one full frame received... */
 	if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&
+	     (tp->fast_ack_mode == 1 ||
 	     /* ... and right edge of window advances far enough.
 	      * (tcp_recvmsg() will send ACK otherwise).
 	      * If application uses SO_RCVLOWAT, we want send ack now if
 	      * we have not received enough bytes to satisfy the condition.
 	      */
-	    (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||
-	     __tcp_select_window(sk) >= tp->rcv_wnd)) ||
+	      (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||
+	       __tcp_select_window(sk) >= tp->rcv_wnd))) ||
 	    /* We ACK each frame or... */
 	    tcp_in_quickack_mode(sk) ||
 	    /* Protocol state mandates a one-time immediate ACK */
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 0ecc7311dc6c..82622782486a 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -460,6 +460,8 @@ void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
 	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
 	bool ca_got_dst = false;
 
+	tcp_set_ecn_low_from_dst(sk, dst);
+
 	if (ca_key != TCP_CA_UNSPEC) {
 		const struct tcp_congestion_ops *ca;
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index e3167ad96567..08fe7a626be1 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -332,10 +332,9 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
 	bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk);
 	bool use_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn) == 1 ||
 		tcp_ca_needs_ecn(sk) || bpf_needs_ecn;
+	const struct dst_entry *dst = __sk_dst_get(sk);
 
 	if (!use_ecn) {
-		const struct dst_entry *dst = __sk_dst_get(sk);
-
 		if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
 			use_ecn = true;
 	}
@@ -347,6 +346,9 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
 		tp->ecn_flags = TCP_ECN_OK;
 		if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn)
 			INET_ECN_xmit(sk);
+
+		if (dst)
+			tcp_set_ecn_low_from_dst(sk, dst);
 	}
 }
 
@@ -384,7 +386,8 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
 				th->cwr = 1;
 				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 			}
-		} else if (!tcp_ca_needs_ecn(sk)) {
+		} else if (!(tp->ecn_flags & TCP_ECN_ECT_PERMANENT) &&
+			!tcp_ca_needs_ecn(sk)) {
 			/* ACK or retransmitted segment: clear ECT|CE */
 			INET_ECN_dontxmit(sk);
 		}
@@ -1593,7 +1596,7 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *buff;
-	int old_factor;
+	int old_factor, inflight_prev;
 	long limit;
 	int nlen;
 	u8 flags;
@@ -1668,6 +1671,30 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
 
 		if (diff)
 			tcp_adjust_pcount(sk, skb, diff);
+
+		inflight_prev = TCP_SKB_CB(skb)->tx.in_flight - old_factor;
+		if (inflight_prev < 0) {
+			WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious(
+					  old_factor,
+					  TCP_SKB_CB(skb)->sacked,
+					  TCP_SKB_CB(skb)->tx.in_flight),
+				  "inconsistent: tx.in_flight: %u "
+				  "old_factor: %d mss: %u sacked: %u "
+				  "1st pcount: %d 2nd pcount: %d "
+				  "1st len: %u 2nd len: %u ",
+				  TCP_SKB_CB(skb)->tx.in_flight, old_factor,
+				  mss_now, TCP_SKB_CB(skb)->sacked,
+				  tcp_skb_pcount(skb), tcp_skb_pcount(buff),
+				  skb->len, buff->len);
+			inflight_prev = 0;
+		}
+		/* Set 1st tx.in_flight as if 1st were sent by itself: */
+		TCP_SKB_CB(skb)->tx.in_flight = inflight_prev +
+						 tcp_skb_pcount(skb);
+		/* Set 2nd tx.in_flight with new 1st and 2nd pcounts: */
+		TCP_SKB_CB(buff)->tx.in_flight = inflight_prev +
+						 tcp_skb_pcount(skb) +
+						 tcp_skb_pcount(buff);
 	}
 
 	/* Link BUFF into the send queue. */
@@ -2025,13 +2052,12 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
 {
 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
-	u32 min_tso, tso_segs;
-
-	min_tso = ca_ops->min_tso_segs ?
-			ca_ops->min_tso_segs(sk) :
-			READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs);
+	u32 tso_segs;
 
-	tso_segs = tcp_tso_autosize(sk, mss_now, min_tso);
+	tso_segs = ca_ops->tso_segs ?
+		ca_ops->tso_segs(sk, mss_now) :
+		tcp_tso_autosize(sk, mss_now,
+				 sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs);
 	return min_t(u32, tso_segs, sk->sk_gso_max_segs);
 }
 
@@ -2731,6 +2757,7 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
 			skb_set_delivery_time(skb, tp->tcp_wstamp_ns, true);
 			list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
 			tcp_init_tso_segs(skb, mss_now);
+			tcp_set_tx_in_flight(sk, skb);
 			goto repair; /* Skip network transmission */
 		}
 
@@ -2944,6 +2971,7 @@ void tcp_send_loss_probe(struct sock *sk)
 	if (WARN_ON(!skb || !tcp_skb_pcount(skb)))
 		goto rearm_timer;
 
+	tp->tlp_orig_data_app_limited = TCP_SKB_CB(skb)->tx.is_app_limited;
 	if (__tcp_retransmit_skb(sk, skb, 1))
 		goto rearm_timer;
 
diff --git a/net/ipv4/tcp_rate.c b/net/ipv4/tcp_rate.c
index a8f6d9d06f2e..8737f2134648 100644
--- a/net/ipv4/tcp_rate.c
+++ b/net/ipv4/tcp_rate.c
@@ -34,6 +34,24 @@
  * ready to send in the write queue.
  */
 
+void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	u32 in_flight;
+
+	/* Check, sanitize, and record packets in flight after skb was sent. */
+	in_flight = tcp_packets_in_flight(tp) + tcp_skb_pcount(skb);
+	if (WARN_ONCE(in_flight > TCPCB_IN_FLIGHT_MAX,
+		      "insane in_flight %u cc %s mss %u "
+		      "cwnd %u pif %u %u %u %u\n",
+		      in_flight, inet_csk(sk)->icsk_ca_ops->name,
+		      tp->mss_cache, tp->snd_cwnd,
+		      tp->packets_out, tp->retrans_out,
+		      tp->sacked_out, tp->lost_out))
+		in_flight = TCPCB_IN_FLIGHT_MAX;
+	TCP_SKB_CB(skb)->tx.in_flight = in_flight;
+}
+
 /* Snapshot the current delivery information in the skb, to generate
  * a rate sample later when the skb is (s)acked in tcp_rate_skb_delivered().
  */
@@ -66,7 +84,9 @@ void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb)
 	TCP_SKB_CB(skb)->tx.delivered_mstamp	= tp->delivered_mstamp;
 	TCP_SKB_CB(skb)->tx.delivered		= tp->delivered;
 	TCP_SKB_CB(skb)->tx.delivered_ce	= tp->delivered_ce;
+	TCP_SKB_CB(skb)->tx.lost		= tp->lost;
 	TCP_SKB_CB(skb)->tx.is_app_limited	= tp->app_limited ? 1 : 0;
+	tcp_set_tx_in_flight(sk, skb);
 }
 
 /* When an skb is sacked or acked, we fill in the rate sample with the (prior)
@@ -91,18 +111,21 @@ void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
 	if (!rs->prior_delivered ||
 	    tcp_skb_sent_after(tx_tstamp, tp->first_tx_mstamp,
 			       scb->end_seq, rs->last_end_seq)) {
+		rs->prior_lost	     = scb->tx.lost;
 		rs->prior_delivered_ce  = scb->tx.delivered_ce;
 		rs->prior_delivered  = scb->tx.delivered;
 		rs->prior_mstamp     = scb->tx.delivered_mstamp;
 		rs->is_app_limited   = scb->tx.is_app_limited;
 		rs->is_retrans	     = scb->sacked & TCPCB_RETRANS;
+		rs->tx_in_flight     = scb->tx.in_flight;
 		rs->last_end_seq     = scb->end_seq;
 
 		/* Record send time of most recently ACKed packet: */
 		tp->first_tx_mstamp  = tx_tstamp;
 		/* Find the duration of the "send phase" of this window: */
-		rs->interval_us = tcp_stamp_us_delta(tp->first_tx_mstamp,
-						     scb->tx.first_tx_mstamp);
+		rs->interval_us      = tcp_stamp32_us_delta(
+						tp->first_tx_mstamp,
+						scb->tx.first_tx_mstamp);
 
 	}
 	/* Mark off the skb delivered once it's sacked to avoid being
@@ -144,6 +167,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
 		return;
 	}
 	rs->delivered   = tp->delivered - rs->prior_delivered;
+	rs->lost        = tp->lost - rs->prior_lost;
 
 	rs->delivered_ce = tp->delivered_ce - rs->prior_delivered_ce;
 	/* delivered_ce occupies less than 32 bits in the skb control block */
@@ -155,7 +179,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
 	 * longer phase.
 	 */
 	snd_us = rs->interval_us;				/* send phase */
-	ack_us = tcp_stamp_us_delta(tp->tcp_mstamp,
+	ack_us = tcp_stamp32_us_delta(tp->tcp_mstamp,
 				    rs->prior_mstamp); /* ack phase */
 	rs->interval_us = max(snd_us, ack_us);
 
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index d1ad20ce1c8c..ef74f33c7905 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -678,6 +678,7 @@ void tcp_write_timer_handler(struct sock *sk)
 		return;
 	}
 
+	tcp_rate_check_app_limited(sk);
 	tcp_mstamp_refresh(tcp_sk(sk));
 	event = icsk->icsk_pending;
 
-- 
2.44.0

From 71b4361aff469d7e31d2260c0f689a976a1a89d0 Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:06:41 +0200
Subject: [PATCH 4/8] block

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 block/bfq-iosched.c | 120 ++++++++++++++++++++++++++++++++++++--------
 block/bfq-iosched.h |  16 +++++-
 block/mq-deadline.c | 114 +++++++++++++++++++++++++++++++++--------
 3 files changed, 205 insertions(+), 45 deletions(-)

diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 3cce6de464a7..9bd57baa4b0b 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -467,6 +467,21 @@ static struct bfq_io_cq *bfq_bic_lookup(struct request_queue *q)
 	return icq;
 }
 
+static struct bfq_io_cq *bfq_bic_try_lookup(struct request_queue *q)
+{
+	if (!current->io_context)
+		return NULL;
+	if (spin_trylock_irq(&q->queue_lock)) {
+		struct bfq_io_cq *icq;
+
+		icq = icq_to_bic(ioc_lookup_icq(q));
+		spin_unlock_irq(&q->queue_lock);
+		return icq;
+	}
+
+	return NULL;
+}
+
 /*
  * Scheduler run of queue, if there are requests pending and no one in the
  * driver that will restart queueing.
@@ -2454,10 +2469,21 @@ static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
 	 * returned by bfq_bic_lookup does not go away before
 	 * bfqd->lock is taken.
 	 */
-	struct bfq_io_cq *bic = bfq_bic_lookup(q);
+	struct bfq_io_cq *bic = bfq_bic_try_lookup(q);
 	bool ret;
 
-	spin_lock_irq(&bfqd->lock);
+	/*
+	 * bio merging is called for every bio queued, and it's very easy
+	 * to run into contention because of that. If we fail getting
+	 * the dd lock, just skip this merge attempt. For related IO, the
+	 * plug will be the successful merging point. If we get here, we
+	 * already failed doing the obvious merge. Chances of actually
+	 * getting a merge off this path is a lot slimmer, so skipping an
+	 * occassional lookup that will most likely not succeed anyway should
+	 * not be a problem.
+	 */
+	if (!spin_trylock_irq(&bfqd->lock))
+		return false;
 
 	if (bic) {
 		/*
@@ -5148,6 +5174,10 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
 {
 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
 
+	if (!list_empty_careful(&bfqd->at_head) ||
+	    !list_empty_careful(&bfqd->at_tail))
+		return true;
+
 	/*
 	 * Avoiding lock: a race on bfqd->queued should cause at
 	 * most a call to dispatch for nothing
@@ -5297,15 +5327,61 @@ static inline void bfq_update_dispatch_stats(struct request_queue *q,
 					     bool idle_timer_disabled) {}
 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
 
+static void bfq_insert_request(struct request_queue *q, struct request *rq,
+			       blk_insert_t flags, struct list_head *free);
+
+static void __bfq_do_insert(struct request_queue *q, blk_insert_t flags,
+			    struct list_head *list, struct list_head *free)
+{
+	while (!list_empty(list)) {
+		struct request *rq;
+
+		rq = list_first_entry(list, struct request, queuelist);
+		list_del_init(&rq->queuelist);
+		bfq_insert_request(q, rq, flags, free);
+	}
+}
+
+static void bfq_do_insert(struct request_queue *q, struct list_head *free)
+{
+	struct bfq_data *bfqd = q->elevator->elevator_data;
+	LIST_HEAD(at_head);
+	LIST_HEAD(at_tail);
+
+	spin_lock(&bfqd->insert_lock);
+	list_splice_init(&bfqd->at_head, &at_head);
+	list_splice_init(&bfqd->at_tail, &at_tail);
+	spin_unlock(&bfqd->insert_lock);
+
+	__bfq_do_insert(q, BLK_MQ_INSERT_AT_HEAD, &at_head, free);
+	__bfq_do_insert(q, 0, &at_tail, free);
+}
+
 static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
 {
-	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
+	struct request_queue *q = hctx->queue;
+	struct bfq_data *bfqd = q->elevator->elevator_data;
 	struct request *rq;
 	struct bfq_queue *in_serv_queue;
 	bool waiting_rq, idle_timer_disabled = false;
+	LIST_HEAD(free);
+
+	/*
+	 * If someone else is already dispatching, skip this one. This will
+	 * defer the next dispatch event to when something completes, and could
+	 * potentially lower the queue depth for contended cases.
+	 *
+	 * See the logic in blk_mq_do_dispatch_sched(), which loops and
+	 * retries if nothing is dispatched.
+	 */
+	if (test_bit(BFQ_DISPATCHING, &bfqd->run_state) ||
+	    test_and_set_bit_lock(BFQ_DISPATCHING, &bfqd->run_state))
+		return NULL;
 
 	spin_lock_irq(&bfqd->lock);
 
+	bfq_do_insert(hctx->queue, &free);
+
 	in_serv_queue = bfqd->in_service_queue;
 	waiting_rq = in_serv_queue && bfq_bfqq_wait_request(in_serv_queue);
 
@@ -5315,7 +5391,9 @@ static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
 			waiting_rq && !bfq_bfqq_wait_request(in_serv_queue);
 	}
 
+	clear_bit_unlock(BFQ_DISPATCHING, &bfqd->run_state);
 	spin_unlock_irq(&bfqd->lock);
+	blk_mq_free_requests(&free);
 	bfq_update_dispatch_stats(hctx->queue, rq,
 			idle_timer_disabled ? in_serv_queue : NULL,
 				idle_timer_disabled);
@@ -6236,27 +6314,21 @@ static inline void bfq_update_insert_stats(struct request_queue *q,
 
 static struct bfq_queue *bfq_init_rq(struct request *rq);
 
-static void bfq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
-			       blk_insert_t flags)
+static void bfq_insert_request(struct request_queue *q, struct request *rq,
+			       blk_insert_t flags, struct list_head *free)
 {
-	struct request_queue *q = hctx->queue;
 	struct bfq_data *bfqd = q->elevator->elevator_data;
 	struct bfq_queue *bfqq;
 	bool idle_timer_disabled = false;
 	blk_opf_t cmd_flags;
-	LIST_HEAD(free);
 
 #ifdef CONFIG_BFQ_GROUP_IOSCHED
 	if (!cgroup_subsys_on_dfl(io_cgrp_subsys) && rq->bio)
 		bfqg_stats_update_legacy_io(q, rq);
 #endif
-	spin_lock_irq(&bfqd->lock);
 	bfqq = bfq_init_rq(rq);
-	if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
-		spin_unlock_irq(&bfqd->lock);
-		blk_mq_free_requests(&free);
+	if (blk_mq_sched_try_insert_merge(q, rq, free))
 		return;
-	}
 
 	trace_block_rq_insert(rq);
 
@@ -6286,8 +6358,6 @@ static void bfq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
 	 * merge).
 	 */
 	cmd_flags = rq->cmd_flags;
-	spin_unlock_irq(&bfqd->lock);
-
 	bfq_update_insert_stats(q, bfqq, idle_timer_disabled,
 				cmd_flags);
 }
@@ -6296,13 +6366,15 @@ static void bfq_insert_requests(struct blk_mq_hw_ctx *hctx,
 				struct list_head *list,
 				blk_insert_t flags)
 {
-	while (!list_empty(list)) {
-		struct request *rq;
+	struct request_queue *q = hctx->queue;
+	struct bfq_data *bfqd = q->elevator->elevator_data;
 
-		rq = list_first_entry(list, struct request, queuelist);
-		list_del_init(&rq->queuelist);
-		bfq_insert_request(hctx, rq, flags);
-	}
+	spin_lock_irq(&bfqd->insert_lock);
+	if (flags & BLK_MQ_INSERT_AT_HEAD)
+		list_splice_init(list, &bfqd->at_head);
+	else
+		list_splice_init(list, &bfqd->at_tail);
+	spin_unlock_irq(&bfqd->insert_lock);
 }
 
 static void bfq_update_hw_tag(struct bfq_data *bfqd)
@@ -7211,6 +7283,12 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
 	q->elevator = eq;
 	spin_unlock_irq(&q->queue_lock);
 
+	spin_lock_init(&bfqd->lock);
+	spin_lock_init(&bfqd->insert_lock);
+
+	INIT_LIST_HEAD(&bfqd->at_head);
+	INIT_LIST_HEAD(&bfqd->at_tail);
+
 	/*
 	 * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues.
 	 * Grab a permanent reference to it, so that the normal code flow
@@ -7329,8 +7407,6 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
 	/* see comments on the definition of next field inside bfq_data */
 	bfqd->actuator_load_threshold = 4;
 
-	spin_lock_init(&bfqd->lock);
-
 	/*
 	 * The invocation of the next bfq_create_group_hierarchy
 	 * function is the head of a chain of function calls
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 467e8cfc41a2..f44f5d4ec2f4 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -504,12 +504,26 @@ struct bfq_io_cq {
 	unsigned int requests;	/* Number of requests this process has in flight */
 };
 
+enum {
+	BFQ_DISPATCHING	= 0,
+};
+
 /**
  * struct bfq_data - per-device data structure.
  *
  * All the fields are protected by @lock.
  */
 struct bfq_data {
+	struct {
+		spinlock_t lock;
+		spinlock_t insert_lock;
+	} ____cacheline_aligned_in_smp;
+
+	unsigned long run_state;
+
+	struct list_head at_head;
+	struct list_head at_tail;
+
 	/* device request queue */
 	struct request_queue *queue;
 	/* dispatch queue */
@@ -795,8 +809,6 @@ struct bfq_data {
 	/* fallback dummy bfqq for extreme OOM conditions */
 	struct bfq_queue oom_bfqq;
 
-	spinlock_t lock;
-
 	/*
 	 * bic associated with the task issuing current bio for
 	 * merging. This and the next field are used as a support to
diff --git a/block/mq-deadline.c b/block/mq-deadline.c
index 02a916ba62ee..8bf621316a9e 100644
--- a/block/mq-deadline.c
+++ b/block/mq-deadline.c
@@ -79,10 +79,24 @@ struct dd_per_prio {
 	struct io_stats_per_prio stats;
 };
 
+enum {
+	DD_DISPATCHING	= 0,
+};
+
 struct deadline_data {
 	/*
 	 * run time data
 	 */
+	struct {
+		spinlock_t lock;
+		spinlock_t insert_lock;
+		spinlock_t zone_lock;
+	} ____cacheline_aligned_in_smp;
+
+	unsigned long run_state;
+
+	struct list_head at_head;
+	struct list_head at_tail;
 
 	struct dd_per_prio per_prio[DD_PRIO_COUNT];
 
@@ -100,9 +114,6 @@ struct deadline_data {
 	int front_merges;
 	u32 async_depth;
 	int prio_aging_expire;
-
-	spinlock_t lock;
-	spinlock_t zone_lock;
 };
 
 /* Maps an I/O priority class to a deadline scheduler priority. */
@@ -113,6 +124,9 @@ static const enum dd_prio ioprio_class_to_prio[] = {
 	[IOPRIO_CLASS_IDLE]	= DD_IDLE_PRIO,
 };
 
+static void dd_insert_request(struct request_queue *q, struct request *rq,
+			      blk_insert_t flags, struct list_head *free);
+
 static inline struct rb_root *
 deadline_rb_root(struct dd_per_prio *per_prio, struct request *rq)
 {
@@ -585,6 +599,33 @@ static struct request *dd_dispatch_prio_aged_requests(struct deadline_data *dd,
 	return NULL;
 }
 
+static void __dd_do_insert(struct request_queue *q, blk_insert_t flags,
+			   struct list_head *list, struct list_head *free)
+{
+	while (!list_empty(list)) {
+		struct request *rq;
+
+		rq = list_first_entry(list, struct request, queuelist);
+		list_del_init(&rq->queuelist);
+		dd_insert_request(q, rq, flags, free);
+	}
+}
+
+static void dd_do_insert(struct request_queue *q, struct list_head *free)
+{
+	struct deadline_data *dd = q->elevator->elevator_data;
+	LIST_HEAD(at_head);
+	LIST_HEAD(at_tail);
+
+	spin_lock(&dd->insert_lock);
+	list_splice_init(&dd->at_head, &at_head);
+	list_splice_init(&dd->at_tail, &at_tail);
+	spin_unlock(&dd->insert_lock);
+
+	__dd_do_insert(q, BLK_MQ_INSERT_AT_HEAD, &at_head, free);
+	__dd_do_insert(q, 0, &at_tail, free);
+}
+
 /*
  * Called from blk_mq_run_hw_queue() -> __blk_mq_sched_dispatch_requests().
  *
@@ -595,12 +636,27 @@ static struct request *dd_dispatch_prio_aged_requests(struct deadline_data *dd,
  */
 static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
 {
-	struct deadline_data *dd = hctx->queue->elevator->elevator_data;
+	struct request_queue *q = hctx->queue;
+	struct deadline_data *dd = q->elevator->elevator_data;
 	const unsigned long now = jiffies;
 	struct request *rq;
 	enum dd_prio prio;
+	LIST_HEAD(free);
+
+	/*
+	 * If someone else is already dispatching, skip this one. This will
+	 * defer the next dispatch event to when something completes, and could
+	 * potentially lower the queue depth for contended cases.
+	 *
+	 * See the logic in blk_mq_do_dispatch_sched(), which loops and
+	 * retries if nothing is dispatched.
+	 */
+	if (test_bit(DD_DISPATCHING, &dd->run_state) ||
+	    test_and_set_bit_lock(DD_DISPATCHING, &dd->run_state))
+		return NULL;
 
 	spin_lock(&dd->lock);
+	dd_do_insert(q, &free);
 	rq = dd_dispatch_prio_aged_requests(dd, now);
 	if (rq)
 		goto unlock;
@@ -616,8 +672,10 @@ static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
 	}
 
 unlock:
+	clear_bit_unlock(DD_DISPATCHING, &dd->run_state);
 	spin_unlock(&dd->lock);
 
+	blk_mq_free_requests(&free);
 	return rq;
 }
 
@@ -705,6 +763,13 @@ static int dd_init_sched(struct request_queue *q, struct elevator_type *e)
 
 	eq->elevator_data = dd;
 
+	spin_lock_init(&dd->lock);
+	spin_lock_init(&dd->insert_lock);
+	spin_lock_init(&dd->zone_lock);
+
+	INIT_LIST_HEAD(&dd->at_head);
+	INIT_LIST_HEAD(&dd->at_tail);
+
 	for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
 		struct dd_per_prio *per_prio = &dd->per_prio[prio];
 
@@ -721,8 +786,6 @@ static int dd_init_sched(struct request_queue *q, struct elevator_type *e)
 	dd->last_dir = DD_WRITE;
 	dd->fifo_batch = fifo_batch;
 	dd->prio_aging_expire = prio_aging_expire;
-	spin_lock_init(&dd->lock);
-	spin_lock_init(&dd->zone_lock);
 
 	/* We dispatch from request queue wide instead of hw queue */
 	blk_queue_flag_set(QUEUE_FLAG_SQ_SCHED, q);
@@ -778,7 +841,19 @@ static bool dd_bio_merge(struct request_queue *q, struct bio *bio,
 	struct request *free = NULL;
 	bool ret;
 
-	spin_lock(&dd->lock);
+	/*
+	 * bio merging is called for every bio queued, and it's very easy
+	 * to run into contention because of that. If we fail getting
+	 * the dd lock, just skip this merge attempt. For related IO, the
+	 * plug will be the successful merging point. If we get here, we
+	 * already failed doing the obvious merge. Chances of actually
+	 * getting a merge off this path is a lot slimmer, so skipping an
+	 * occassional lookup that will most likely not succeed anyway should
+	 * not be a problem.
+	 */
+	if (!spin_trylock(&dd->lock))
+		return false;
+
 	ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
 	spin_unlock(&dd->lock);
 
@@ -791,10 +866,9 @@ static bool dd_bio_merge(struct request_queue *q, struct bio *bio,
 /*
  * add rq to rbtree and fifo
  */
-static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
+static void dd_insert_request(struct request_queue *q, struct request *rq,
 			      blk_insert_t flags, struct list_head *free)
 {
-	struct request_queue *q = hctx->queue;
 	struct deadline_data *dd = q->elevator->elevator_data;
 	const enum dd_data_dir data_dir = rq_data_dir(rq);
 	u16 ioprio = req_get_ioprio(rq);
@@ -866,19 +940,13 @@ static void dd_insert_requests(struct blk_mq_hw_ctx *hctx,
 {
 	struct request_queue *q = hctx->queue;
 	struct deadline_data *dd = q->elevator->elevator_data;
-	LIST_HEAD(free);
 
-	spin_lock(&dd->lock);
-	while (!list_empty(list)) {
-		struct request *rq;
-
-		rq = list_first_entry(list, struct request, queuelist);
-		list_del_init(&rq->queuelist);
-		dd_insert_request(hctx, rq, flags, &free);
-	}
-	spin_unlock(&dd->lock);
-
-	blk_mq_free_requests(&free);
+	spin_lock(&dd->insert_lock);
+	if (flags & BLK_MQ_INSERT_AT_HEAD)
+		list_splice_init(list, &dd->at_head);
+	else
+		list_splice_init(list, &dd->at_tail);
+	spin_unlock(&dd->insert_lock);
 }
 
 /* Callback from inside blk_mq_rq_ctx_init(). */
@@ -957,6 +1025,10 @@ static bool dd_has_work(struct blk_mq_hw_ctx *hctx)
 	struct deadline_data *dd = hctx->queue->elevator->elevator_data;
 	enum dd_prio prio;
 
+	if (!list_empty_careful(&dd->at_head) ||
+	    !list_empty_careful(&dd->at_tail))
+		return true;
+
 	for (prio = 0; prio <= DD_PRIO_MAX; prio++)
 		if (dd_has_work_for_prio(&dd->per_prio[prio]))
 			return true;
-- 
2.44.0

From b667355ece89a997a7b8508e6d6f1b5be46d3833 Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:06:52 +0200
Subject: [PATCH 5/8] cachy

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 .../admin-guide/kernel-parameters.txt         |   9 +
 Documentation/admin-guide/sysctl/vm.rst       |  72 ++
 Makefile                                      | 162 ++++-
 arch/arm/Makefile                             |  56 +-
 arch/x86/Kconfig.cpu                          | 426 +++++++++++-
 arch/x86/Makefile                             |  19 +-
 arch/x86/Makefile_32.cpu                      |  41 --
 arch/x86/include/asm/pci.h                    |   6 +
 arch/x86/include/asm/vermagic.h               |  74 ++
 arch/x86/pci/common.c                         |   7 +-
 block/bfq-iosched.c                           |   6 +
 block/elevator.c                              |  10 +
 drivers/ata/ahci.c                            |  23 +-
 drivers/cpufreq/Kconfig.x86                   |   2 -
 drivers/gpu/drm/amd/amdgpu/amdgpu.h           |   1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c       |  10 +
 drivers/gpu/drm/amd/display/Kconfig           |   6 +
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c |   2 +-
 .../amd/display/amdgpu_dm/amdgpu_dm_color.c   |   2 +-
 .../amd/display/amdgpu_dm/amdgpu_dm_crtc.c    |   6 +-
 .../amd/display/amdgpu_dm/amdgpu_dm_plane.c   |   6 +-
 drivers/gpu/drm/amd/pm/amdgpu_pm.c            |   3 +
 drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c     |  14 +-
 drivers/i2c/busses/Kconfig                    |   9 +
 drivers/i2c/busses/Makefile                   |   1 +
 drivers/i2c/busses/i2c-nct6775.c              | 648 ++++++++++++++++++
 drivers/i2c/busses/i2c-piix4.c                |   4 +-
 drivers/input/evdev.c                         |  19 +-
 drivers/md/dm-crypt.c                         |   5 +
 drivers/pci/controller/Makefile               |   6 +
 drivers/pci/controller/intel-nvme-remap.c     | 462 +++++++++++++
 drivers/pci/quirks.c                          | 101 +++
 drivers/platform/x86/Kconfig                  |  14 +
 drivers/platform/x86/Makefile                 |   3 +
 drivers/platform/x86/steamdeck.c              | 523 ++++++++++++++
 include/linux/mm.h                            |  10 +-
 include/linux/pagemap.h                       |   2 +-
 include/linux/user_namespace.h                |   4 +
 init/Kconfig                                  |  26 +
 kernel/Kconfig.hz                             |  24 +
 kernel/fork.c                                 |  14 +
 kernel/sched/fair.c                           |  13 +
 kernel/sched/sched.h                          |   2 +-
 kernel/sysctl.c                               |  46 ++
 kernel/user_namespace.c                       |   7 +
 mm/Kconfig                                    |  65 +-
 mm/compaction.c                               |   4 +
 mm/huge_memory.c                              |   4 +
 mm/mm_init.c                                  |   1 +
 mm/page-writeback.c                           |   8 +
 mm/page_alloc.c                               |  27 +-
 mm/swap.c                                     |   5 +
 mm/vmpressure.c                               |   4 +
 mm/vmscan.c                                   | 178 ++++-
 54 files changed, 3020 insertions(+), 182 deletions(-)
 create mode 100644 drivers/i2c/busses/i2c-nct6775.c
 create mode 100644 drivers/pci/controller/intel-nvme-remap.c
 create mode 100644 drivers/platform/x86/steamdeck.c

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 71ed7f1b0f9b..fbfaea49cbed 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -4394,6 +4394,15 @@
 		nomsi		[MSI] If the PCI_MSI kernel config parameter is
 				enabled, this kernel boot option can be used to
 				disable the use of MSI interrupts system-wide.
+		pcie_acs_override =
+					[PCIE] Override missing PCIe ACS support for:
+				downstream
+					All downstream ports - full ACS capabilities
+				multfunction
+					All multifunction devices - multifunction ACS subset
+				id:nnnn:nnnn
+					Specfic device - full ACS capabilities
+					Specified as vid:did (vendor/device ID) in hex
 		noioapicquirk	[APIC] Disable all boot interrupt quirks.
 				Safety option to keep boot IRQs enabled. This
 				should never be necessary.
diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst
index c59889de122b..468ae7dec1e1 100644
--- a/Documentation/admin-guide/sysctl/vm.rst
+++ b/Documentation/admin-guide/sysctl/vm.rst
@@ -25,6 +25,9 @@ files can be found in mm/swap.c.
 Currently, these files are in /proc/sys/vm:
 
 - admin_reserve_kbytes
+- anon_min_ratio
+- clean_low_ratio
+- clean_min_ratio
 - compact_memory
 - compaction_proactiveness
 - compact_unevictable_allowed
@@ -106,6 +109,67 @@ On x86_64 this is about 128MB.
 Changing this takes effect whenever an application requests memory.
 
 
+anon_min_ratio
+==============
+
+This knob provides *hard* protection of anonymous pages. The anonymous pages
+on the current node won't be reclaimed under any conditions when their amount
+is below vm.anon_min_ratio.
+
+This knob may be used to prevent excessive swap thrashing when anonymous
+memory is low (for example, when memory is going to be overfilled by
+compressed data of zram module).
+
+Setting this value too high (close to 100) can result in inability to
+swap and can lead to early OOM under memory pressure.
+
+The unit of measurement is the percentage of the total memory of the node.
+
+The default value is 15.
+
+
+clean_low_ratio
+================
+
+This knob provides *best-effort* protection of clean file pages. The file pages
+on the current node won't be reclaimed under memory pressure when the amount of
+clean file pages is below vm.clean_low_ratio *unless* we threaten to OOM.
+
+Protection of clean file pages using this knob may be used when swapping is
+still possible to
+  - prevent disk I/O thrashing under memory pressure;
+  - improve performance in disk cache-bound tasks under memory pressure.
+
+Setting it to a high value may result in a early eviction of anonymous pages
+into the swap space by attempting to hold the protected amount of clean file
+pages in memory.
+
+The unit of measurement is the percentage of the total memory of the node.
+
+The default value is 0.
+
+
+clean_min_ratio
+================
+
+This knob provides *hard* protection of clean file pages. The file pages on the
+current node won't be reclaimed under memory pressure when the amount of clean
+file pages is below vm.clean_min_ratio.
+
+Hard protection of clean file pages using this knob may be used to
+  - prevent disk I/O thrashing under memory pressure even with no free swap space;
+  - improve performance in disk cache-bound tasks under memory pressure;
+  - avoid high latency and prevent livelock in near-OOM conditions.
+
+Setting it to a high value may result in a early out-of-memory condition due to
+the inability to reclaim the protected amount of clean file pages when other
+types of pages cannot be reclaimed.
+
+The unit of measurement is the percentage of the total memory of the node.
+
+The default value is 15.
+
+
 compact_memory
 ==============
 
@@ -910,6 +974,14 @@ be 133 (x + 2x = 200, 2x = 133.33).
 At 0, the kernel will not initiate swap until the amount of free and
 file-backed pages is less than the high watermark in a zone.
 
+This knob has no effect if the amount of clean file pages on the current
+node is below vm.clean_low_ratio or vm.clean_min_ratio. In this case,
+only anonymous pages can be reclaimed.
+
+If the number of anonymous pages on the current node is below
+vm.anon_min_ratio, then only file pages can be reclaimed with
+any vm.swappiness value.
+
 
 unprivileged_userfaultfd
 ========================
diff --git a/Makefile b/Makefile
index a78379891d22..e58a4e647e7d 100644
--- a/Makefile
+++ b/Makefile
@@ -808,9 +808,164 @@ endif # need-config
 
 KBUILD_CFLAGS	+= -fno-delete-null-pointer-checks
 
+# This selects which ARM instruction set is used.
+arch-$(CONFIG_CPU_32v7M)	:=-march=armv7-m
+arch-$(CONFIG_CPU_32v7)		:=-march=armv7-a
+arch-$(CONFIG_CPU_32v6)		:=-march=armv6
+# Only override the compiler option if ARMv6. The ARMv6K extensions are
+# always available in ARMv7
+ifeq ($(CONFIG_CPU_32v6),y)
+arch-$(CONFIG_CPU_32v6K)	:=-march=armv6k
+endif
+arch-$(CONFIG_CPU_32v5)		:=-march=armv5te
+arch-$(CONFIG_CPU_32v4T)	:=-march=armv4t
+arch-$(CONFIG_CPU_32v4)		:=-march=armv4
+arch-$(CONFIG_CPU_32v3)		:=-march=armv3m
+
+# Note that GCC does not numerically define an architecture version
+# macro, but instead defines a whole series of macros which makes
+# testing for a specific architecture or later rather impossible.
+cpp-$(CONFIG_CPU_32v7M)		:=-D__LINUX_ARM_ARCH__=7
+cpp-$(CONFIG_CPU_32v7)		:=-D__LINUX_ARM_ARCH__=7
+cpp-$(CONFIG_CPU_32v6)		:=-D__LINUX_ARM_ARCH__=6
+# Only override the compiler option if ARMv6. The ARMv6K extensions are
+# always available in ARMv7
+ifeq ($(CONFIG_CPU_32v6),y)
+cpp-$(CONFIG_CPU_32v6K)		:=-D__LINUX_ARM_ARCH__=6
+endif
+cpp-$(CONFIG_CPU_32v5)		:=-D__LINUX_ARM_ARCH__=5
+cpp-$(CONFIG_CPU_32v4T)		:=-D__LINUX_ARM_ARCH__=4
+cpp-$(CONFIG_CPU_32v4)		:=-D__LINUX_ARM_ARCH__=4
+cpp-$(CONFIG_CPU_32v3)		:=-D__LINUX_ARM_ARCH__=3
+
+# This selects how we optimise for the processor.
+tune-$(CONFIG_CPU_ARM7TDMI)	:=-mtune=arm7tdmi
+tune-$(CONFIG_CPU_ARM720T)	:=-mtune=arm7tdmi
+tune-$(CONFIG_CPU_ARM740T)	:=-mtune=arm7tdmi
+tune-$(CONFIG_CPU_ARM9TDMI)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_ARM940T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_ARM946E)	:=-mtune=arm9e
+tune-$(CONFIG_CPU_ARM920T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_ARM922T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_ARM925T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_ARM926T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_FA526)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_SA110)	:=-mtune=strongarm110
+tune-$(CONFIG_CPU_SA1100)	:=-mtune=strongarm1100
+tune-$(CONFIG_CPU_XSCALE)	:=-mtune=xscale
+tune-$(CONFIG_CPU_XSC3)		:=-mtune=xscale
+tune-$(CONFIG_CPU_FEROCEON)	:=-mtune=xscale
+tune-$(CONFIG_CPU_V6)		:=-mtune=arm1136j-s
+tune-$(CONFIG_CPU_V6K)		:=-mtune=arm1136j-s
+
+KBUILD_CPPFLAGS	+=$(cpp-y)
+KBUILD_CFLAGS	+=$(arch-y) $(tune-y)
+KBUILD_AFLAGS	+=$(arch-y) $(tune-y)
+
+# This selects which x86 instruction set is used.
+cflags-$(CONFIG_M486SX)		+= -march=i486
+cflags-$(CONFIG_M486)		+= -march=i486
+cflags-$(CONFIG_M586)		+= -march=i586
+cflags-$(CONFIG_M586TSC)	+= -march=i586
+cflags-$(CONFIG_M586MMX)	+= -march=pentium-mmx
+cflags-$(CONFIG_M686)		+= -march=i686
+cflags-$(CONFIG_MPENTIUMII)	+= -march=i686 $(call tune,pentium2)
+cflags-$(CONFIG_MPENTIUMIII)	+= -march=i686 $(call tune,pentium3)
+cflags-$(CONFIG_MPENTIUMM)	+= -march=i686 $(call tune,pentium3)
+cflags-$(CONFIG_MPENTIUM4)	+= -march=i686 $(call tune,pentium4)
+cflags-$(CONFIG_MK6)		+= -march=k6
+# Please note, that patches that add -march=athlon-xp and friends are pointless.
+# They make zero difference whatsosever to performance at this time.
+cflags-$(CONFIG_MK7)		+= -march=athlon
+cflags-$(CONFIG_MK8)		+= $(call cc-option,-march=k8,-march=athlon)
+cflags-$(CONFIG_MCRUSOE)	+= -march=i686 $(align)
+cflags-$(CONFIG_MEFFICEON)	+= -march=i686 $(call tune,pentium3) $(align)
+cflags-$(CONFIG_MWINCHIPC6)	+= $(call cc-option,-march=winchip-c6,-march=i586)
+cflags-$(CONFIG_MWINCHIP3D)	+= $(call cc-option,-march=winchip2,-march=i586)
+cflags-$(CONFIG_MCYRIXIII)	+= $(call cc-option,-march=c3,-march=i486) $(align)
+cflags-$(CONFIG_MVIAC3_2)	+= $(call cc-option,-march=c3-2,-march=i686)
+cflags-$(CONFIG_MVIAC7)		+= -march=i686
+cflags-$(CONFIG_MCORE2)		+= -march=i686 $(call tune,core2)
+cflags-$(CONFIG_MATOM)		+= $(call cc-option,-march=atom,$(call cc-option,-march=core2,-march=i686)) \
+$(call cc-option,-mtune=atom,$(call cc-option,-mtune=generic))
+
+# AMD Elan support
+cflags-$(CONFIG_MELAN)		+= -march=i486
+
+# Geode GX1 support
+cflags-$(CONFIG_MGEODEGX1)	+= -march=pentium-mmx
+cflags-$(CONFIG_MGEODE_LX)	+= $(call cc-option,-march=geode,-march=pentium-mmx)
+# add at the end to overwrite eventual tuning options from earlier
+# cpu entries
+cflags-$(CONFIG_X86_GENERIC) 	+= $(call tune,generic,$(call tune,i686))
+
+# Bug fix for binutils: this option is required in order to keep
+# binutils from generating NOPL instructions against our will.
+ifneq ($(CONFIG_X86_P6_NOP),y)
+cflags-y			+= $(call cc-option,-Wa$(comma)-mtune=generic32,)
+endif
+
+# x86_64 instruction set
+cflags64-$(CONFIG_MK8)		+= -march=k8
+cflags64-$(CONFIG_MPSC)		+= -march=nocona
+cflags64-$(CONFIG_MK8SSE3)	+= -march=k8-sse3
+cflags64-$(CONFIG_MK10) 		+= -march=amdfam10
+cflags64-$(CONFIG_MBARCELONA) 	+= -march=barcelona
+cflags64-$(CONFIG_MBOBCAT) 	+= -march=btver1
+cflags64-$(CONFIG_MJAGUAR) 	+= -march=btver2
+cflags64-$(CONFIG_MBULLDOZER) 	+= -march=bdver1
+cflags64-$(CONFIG_MPILEDRIVER)	+= -march=bdver2 -mno-tbm
+cflags64-$(CONFIG_MSTEAMROLLER) 	+= -march=bdver3 -mno-tbm
+cflags64-$(CONFIG_MEXCAVATOR) 	+= -march=bdver4 -mno-tbm
+cflags64-$(CONFIG_MZEN) 		+= -march=znver1
+cflags64-$(CONFIG_MZEN2) 	+= -march=znver2
+cflags64-$(CONFIG_MZEN3) 	+= -march=znver3
+cflags64-$(CONFIG_MZEN4) 	+= -march=znver4
+cflags64-$(CONFIG_MNATIVE_INTEL) += -march=native
+cflags64-$(CONFIG_MNATIVE_AMD) 	+= -march=native
+cflags64-$(CONFIG_MATOM) 	+= -march=bonnell
+cflags64-$(CONFIG_MCORE2) 	+= -march=core2
+cflags64-$(CONFIG_MNEHALEM) 	+= -march=nehalem
+cflags64-$(CONFIG_MWESTMERE) 	+= -march=westmere
+cflags64-$(CONFIG_MSILVERMONT) 	+= -march=silvermont
+cflags64-$(CONFIG_MGOLDMONT) 	+= -march=goldmont
+cflags64-$(CONFIG_MGOLDMONTPLUS) += -march=goldmont-plus
+cflags64-$(CONFIG_MSANDYBRIDGE) 	+= -march=sandybridge
+cflags64-$(CONFIG_MIVYBRIDGE) 	+= -march=ivybridge
+cflags64-$(CONFIG_MHASWELL) 	+= -march=haswell
+cflags64-$(CONFIG_MBROADWELL) 	+= -march=broadwell
+cflags64-$(CONFIG_MSKYLAKE) 	+= -march=skylake
+cflags64-$(CONFIG_MSKYLAKEX) 	+= -march=skylake-avx512
+cflags64-$(CONFIG_MCANNONLAKE) 	+= -march=cannonlake
+cflags64-$(CONFIG_MICELAKE) 	+= -march=icelake-client
+cflags64-$(CONFIG_MCASCADELAKE) 	+= -march=cascadelake
+cflags64-$(CONFIG_MCOOPERLAKE) 	+= -march=cooperlake
+cflags64-$(CONFIG_MTIGERLAKE) 	+= -march=tigerlake
+cflags64-$(CONFIG_MSAPPHIRERAPIDS) += -march=sapphirerapids
+cflags64-$(CONFIG_MROCKETLAKE) 	+= -march=rocketlake
+cflags64-$(CONFIG_MALDERLAKE) 	+= -march=alderlake
+cflags64-$(CONFIG_MRAPTORLAKE) 	+= -march=raptorlake
+cflags64-$(CONFIG_MMETEORLAKE) 	+= -march=meteorlake
+cflags64-$(CONFIG_MEMERALDRAPIDS)	+= -march=emeraldrapids
+cflags64-$(CONFIG_GENERIC_CPU2) 	+= -march=x86-64-v2
+cflags64-$(CONFIG_GENERIC_CPU3) 	+= -march=x86-64-v3
+cflags64-$(CONFIG_GENERIC_CPU4) 	+= -march=x86-64-v4
+cflags64-$(CONFIG_GENERIC_CPU)	+= -mtune=generic
+KBUILD_CFLAGS += $(cflags64-y)
+
+rustflags64-$(CONFIG_MK8)		+= -Ctarget-cpu=k8
+rustflags64-$(CONFIG_MPSC)	+= -Ctarget-cpu=nocona
+rustflags64-$(CONFIG_MCORE2)	+= -Ctarget-cpu=core2
+rustflags64-$(CONFIG_MATOM)	+= -Ctarget-cpu=atom
+rustflags64-$(CONFIG_GENERIC_CPU)	+= -Ztune-cpu=generic
+KBUILD_RUSTFLAGS += $(rustflags64-y)
+
 ifdef CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE
 KBUILD_CFLAGS += -O2
 KBUILD_RUSTFLAGS += -Copt-level=2
+else ifdef CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE_O3
+KBUILD_CFLAGS += -O3
+KBUILD_RUSTFLAGS += -Copt-level=3
 else ifdef CONFIG_CC_OPTIMIZE_FOR_SIZE
 KBUILD_CFLAGS += -Os
 KBUILD_RUSTFLAGS += -Copt-level=s
@@ -990,15 +1145,18 @@ KBUILD_CFLAGS += $(call cc-option, -fstrict-flex-arrays=3)
 KBUILD_CFLAGS-$(CONFIG_CC_NO_STRINGOP_OVERFLOW) += $(call cc-option, -Wno-stringop-overflow)
 KBUILD_CFLAGS-$(CONFIG_CC_STRINGOP_OVERFLOW) += $(call cc-option, -Wstringop-overflow)
 
+# disable GCC vectorization on trees
+KBUILD_CFLAGS	+= $(call cc-option, -fno-tree-vectorize)
+
 # disable invalid "can't wrap" optimizations for signed / pointers
 KBUILD_CFLAGS	+= -fno-strict-overflow
 
 # Make sure -fstack-check isn't enabled (like gentoo apparently did)
 KBUILD_CFLAGS  += -fno-stack-check
 
-# conserve stack if available
+# conserve stack, ivopts and modulo-sched if available
 ifdef CONFIG_CC_IS_GCC
-KBUILD_CFLAGS   += -fconserve-stack
+KBUILD_CFLAGS   += -fconserve-stack -fivopts -fmodulo-sched
 endif
 
 # change __FILE__ to the relative path from the srctree
diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 473280d5adce..c7596c898625 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -59,56 +59,6 @@ endif
 #
 KBUILD_CFLAGS	+= $(call cc-option,-fno-ipa-sra)
 
-# This selects which instruction set is used.
-arch-$(CONFIG_CPU_32v7M)	:=-march=armv7-m
-arch-$(CONFIG_CPU_32v7)		:=-march=armv7-a
-arch-$(CONFIG_CPU_32v6)		:=-march=armv6
-# Only override the compiler option if ARMv6. The ARMv6K extensions are
-# always available in ARMv7
-ifeq ($(CONFIG_CPU_32v6),y)
-arch-$(CONFIG_CPU_32v6K)	:=-march=armv6k
-endif
-arch-$(CONFIG_CPU_32v5)		:=-march=armv5te
-arch-$(CONFIG_CPU_32v4T)	:=-march=armv4t
-arch-$(CONFIG_CPU_32v4)		:=-march=armv4
-arch-$(CONFIG_CPU_32v3)		:=-march=armv3m
-
-# Note that GCC does not numerically define an architecture version
-# macro, but instead defines a whole series of macros which makes
-# testing for a specific architecture or later rather impossible.
-cpp-$(CONFIG_CPU_32v7M)		:=-D__LINUX_ARM_ARCH__=7
-cpp-$(CONFIG_CPU_32v7)		:=-D__LINUX_ARM_ARCH__=7
-cpp-$(CONFIG_CPU_32v6)		:=-D__LINUX_ARM_ARCH__=6
-# Only override the compiler option if ARMv6. The ARMv6K extensions are
-# always available in ARMv7
-ifeq ($(CONFIG_CPU_32v6),y)
-cpp-$(CONFIG_CPU_32v6K)		:=-D__LINUX_ARM_ARCH__=6
-endif
-cpp-$(CONFIG_CPU_32v5)		:=-D__LINUX_ARM_ARCH__=5
-cpp-$(CONFIG_CPU_32v4T)		:=-D__LINUX_ARM_ARCH__=4
-cpp-$(CONFIG_CPU_32v4)		:=-D__LINUX_ARM_ARCH__=4
-cpp-$(CONFIG_CPU_32v3)		:=-D__LINUX_ARM_ARCH__=3
-
-# This selects how we optimise for the processor.
-tune-$(CONFIG_CPU_ARM7TDMI)	:=-mtune=arm7tdmi
-tune-$(CONFIG_CPU_ARM720T)	:=-mtune=arm7tdmi
-tune-$(CONFIG_CPU_ARM740T)	:=-mtune=arm7tdmi
-tune-$(CONFIG_CPU_ARM9TDMI)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_ARM940T)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_ARM946E)	:=-mtune=arm9e
-tune-$(CONFIG_CPU_ARM920T)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_ARM922T)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_ARM925T)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_ARM926T)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_FA526)	:=-mtune=arm9tdmi
-tune-$(CONFIG_CPU_SA110)	:=-mtune=strongarm110
-tune-$(CONFIG_CPU_SA1100)	:=-mtune=strongarm1100
-tune-$(CONFIG_CPU_XSCALE)	:=-mtune=xscale
-tune-$(CONFIG_CPU_XSC3)		:=-mtune=xscale
-tune-$(CONFIG_CPU_FEROCEON)	:=-mtune=xscale
-tune-$(CONFIG_CPU_V6)		:=-mtune=arm1136j-s
-tune-$(CONFIG_CPU_V6K)		:=-mtune=arm1136j-s
-
 ifeq ($(CONFIG_AEABI),y)
 CFLAGS_ABI	:=-mabi=aapcs-linux -mfpu=vfp
 else
@@ -140,9 +90,9 @@ AFLAGS_ISA	:=$(CFLAGS_ISA)
 endif
 
 # Need -Uarm for gcc < 3.x
-KBUILD_CPPFLAGS	+=$(cpp-y)
-KBUILD_CFLAGS	+=$(CFLAGS_ABI) $(CFLAGS_ISA) $(arch-y) $(tune-y) $(call cc-option,-mshort-load-bytes,$(call cc-option,-malignment-traps,)) -msoft-float -Uarm
-KBUILD_AFLAGS	+=$(CFLAGS_ABI) $(AFLAGS_ISA) -Wa,$(arch-y) $(tune-y) -include asm/unified.h -msoft-float
+
+KBUILD_CFLAGS	+=$(CFLAGS_ABI) $(CFLAGS_ISA) $(call cc-option,-mshort-load-bytes,$(call cc-option,-malignment-traps,)) -msoft-float -Uarm
+KBUILD_AFLAGS	+=$(CFLAGS_ABI) $(AFLAGS_ISA) -Wa,-include asm/unified.h -msoft-float
 
 CHECKFLAGS	+= -D__arm__
 
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index 2a7279d80460..b6a64a959e09 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -157,7 +157,7 @@ config MPENTIUM4
 
 
 config MK6
-	bool "K6/K6-II/K6-III"
+	bool "AMD K6/K6-II/K6-III"
 	depends on X86_32
 	help
 	  Select this for an AMD K6-family processor.  Enables use of
@@ -165,7 +165,7 @@ config MK6
 	  flags to GCC.
 
 config MK7
-	bool "Athlon/Duron/K7"
+	bool "AMD Athlon/Duron/K7"
 	depends on X86_32
 	help
 	  Select this for an AMD Athlon K7-family processor.  Enables use of
@@ -173,12 +173,106 @@ config MK7
 	  flags to GCC.
 
 config MK8
-	bool "Opteron/Athlon64/Hammer/K8"
+	bool "AMD Opteron/Athlon64/Hammer/K8"
 	help
 	  Select this for an AMD Opteron or Athlon64 Hammer-family processor.
 	  Enables use of some extended instructions, and passes appropriate
 	  optimization flags to GCC.
 
+config MK8SSE3
+	bool "AMD Opteron/Athlon64/Hammer/K8 with SSE3"
+	help
+	  Select this for improved AMD Opteron or Athlon64 Hammer-family processors.
+	  Enables use of some extended instructions, and passes appropriate
+	  optimization flags to GCC.
+
+config MK10
+	bool "AMD 61xx/7x50/PhenomX3/X4/II/K10"
+	help
+	  Select this for an AMD 61xx Eight-Core Magny-Cours, Athlon X2 7x50,
+	  Phenom X3/X4/II, Athlon II X2/X3/X4, or Turion II-family processor.
+	  Enables use of some extended instructions, and passes appropriate
+	  optimization flags to GCC.
+
+config MBARCELONA
+	bool "AMD Barcelona"
+	help
+	  Select this for AMD Family 10h Barcelona processors.
+
+	  Enables -march=barcelona
+
+config MBOBCAT
+	bool "AMD Bobcat"
+	help
+	  Select this for AMD Family 14h Bobcat processors.
+
+	  Enables -march=btver1
+
+config MJAGUAR
+	bool "AMD Jaguar"
+	help
+	  Select this for AMD Family 16h Jaguar processors.
+
+	  Enables -march=btver2
+
+config MBULLDOZER
+	bool "AMD Bulldozer"
+	help
+	  Select this for AMD Family 15h Bulldozer processors.
+
+	  Enables -march=bdver1
+
+config MPILEDRIVER
+	bool "AMD Piledriver"
+	help
+	  Select this for AMD Family 15h Piledriver processors.
+
+	  Enables -march=bdver2
+
+config MSTEAMROLLER
+	bool "AMD Steamroller"
+	help
+	  Select this for AMD Family 15h Steamroller processors.
+
+	  Enables -march=bdver3
+
+config MEXCAVATOR
+	bool "AMD Excavator"
+	help
+	  Select this for AMD Family 15h Excavator processors.
+
+	  Enables -march=bdver4
+
+config MZEN
+	bool "AMD Zen"
+	help
+	  Select this for AMD Family 17h Zen processors.
+
+	  Enables -march=znver1
+
+config MZEN2
+	bool "AMD Zen 2"
+	help
+	  Select this for AMD Family 17h Zen 2 processors.
+
+	  Enables -march=znver2
+
+config MZEN3
+	bool "AMD Zen 3"
+	depends on (CC_IS_GCC && GCC_VERSION >= 100300) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	help
+	  Select this for AMD Family 19h Zen 3 processors.
+
+	  Enables -march=znver3
+
+config MZEN4
+	bool "AMD Zen 4"
+	depends on (CC_IS_GCC && GCC_VERSION >= 130000) || (CC_IS_CLANG && CLANG_VERSION >= 160000)
+	help
+	  Select this for AMD Family 19h Zen 4 processors.
+
+	  Enables -march=znver4
+
 config MCRUSOE
 	bool "Crusoe"
 	depends on X86_32
@@ -270,7 +364,7 @@ config MPSC
 	  in /proc/cpuinfo. Family 15 is an older Xeon, Family 6 a newer one.
 
 config MCORE2
-	bool "Core 2/newer Xeon"
+	bool "Intel Core 2"
 	help
 
 	  Select this for Intel Core 2 and newer Core 2 Xeons (Xeon 51xx and
@@ -278,6 +372,8 @@ config MCORE2
 	  family in /proc/cpuinfo. Newer ones have 6 and older ones 15
 	  (not a typo)
 
+	  Enables -march=core2
+
 config MATOM
 	bool "Intel Atom"
 	help
@@ -287,6 +383,212 @@ config MATOM
 	  accordingly optimized code. Use a recent GCC with specific Atom
 	  support in order to fully benefit from selecting this option.
 
+config MNEHALEM
+	bool "Intel Nehalem"
+	select X86_P6_NOP
+	help
+
+	  Select this for 1st Gen Core processors in the Nehalem family.
+
+	  Enables -march=nehalem
+
+config MWESTMERE
+	bool "Intel Westmere"
+	select X86_P6_NOP
+	help
+
+	  Select this for the Intel Westmere formerly Nehalem-C family.
+
+	  Enables -march=westmere
+
+config MSILVERMONT
+	bool "Intel Silvermont"
+	select X86_P6_NOP
+	help
+
+	  Select this for the Intel Silvermont platform.
+
+	  Enables -march=silvermont
+
+config MGOLDMONT
+	bool "Intel Goldmont"
+	select X86_P6_NOP
+	help
+
+	  Select this for the Intel Goldmont platform including Apollo Lake and Denverton.
+
+	  Enables -march=goldmont
+
+config MGOLDMONTPLUS
+	bool "Intel Goldmont Plus"
+	select X86_P6_NOP
+	help
+
+	  Select this for the Intel Goldmont Plus platform including Gemini Lake.
+
+	  Enables -march=goldmont-plus
+
+config MSANDYBRIDGE
+	bool "Intel Sandy Bridge"
+	select X86_P6_NOP
+	help
+
+	  Select this for 2nd Gen Core processors in the Sandy Bridge family.
+
+	  Enables -march=sandybridge
+
+config MIVYBRIDGE
+	bool "Intel Ivy Bridge"
+	select X86_P6_NOP
+	help
+
+	  Select this for 3rd Gen Core processors in the Ivy Bridge family.
+
+	  Enables -march=ivybridge
+
+config MHASWELL
+	bool "Intel Haswell"
+	select X86_P6_NOP
+	help
+
+	  Select this for 4th Gen Core processors in the Haswell family.
+
+	  Enables -march=haswell
+
+config MBROADWELL
+	bool "Intel Broadwell"
+	select X86_P6_NOP
+	help
+
+	  Select this for 5th Gen Core processors in the Broadwell family.
+
+	  Enables -march=broadwell
+
+config MSKYLAKE
+	bool "Intel Skylake"
+	select X86_P6_NOP
+	help
+
+	  Select this for 6th Gen Core processors in the Skylake family.
+
+	  Enables -march=skylake
+
+config MSKYLAKEX
+	bool "Intel Skylake X"
+	select X86_P6_NOP
+	help
+
+	  Select this for 6th Gen Core processors in the Skylake X family.
+
+	  Enables -march=skylake-avx512
+
+config MCANNONLAKE
+	bool "Intel Cannon Lake"
+	select X86_P6_NOP
+	help
+
+	  Select this for 8th Gen Core processors
+
+	  Enables -march=cannonlake
+
+config MICELAKE
+	bool "Intel Ice Lake"
+	select X86_P6_NOP
+	help
+
+	  Select this for 10th Gen Core processors in the Ice Lake family.
+
+	  Enables -march=icelake-client
+
+config MCASCADELAKE
+	bool "Intel Cascade Lake"
+	select X86_P6_NOP
+	help
+
+	  Select this for Xeon processors in the Cascade Lake family.
+
+	  Enables -march=cascadelake
+
+config MCOOPERLAKE
+	bool "Intel Cooper Lake"
+	depends on (CC_IS_GCC && GCC_VERSION > 100100) || (CC_IS_CLANG && CLANG_VERSION >= 100000)
+	select X86_P6_NOP
+	help
+
+	  Select this for Xeon processors in the Cooper Lake family.
+
+	  Enables -march=cooperlake
+
+config MTIGERLAKE
+	bool "Intel Tiger Lake"
+	depends on  (CC_IS_GCC && GCC_VERSION > 100100) || (CC_IS_CLANG && CLANG_VERSION >= 100000)
+	select X86_P6_NOP
+	help
+
+	  Select this for third-generation 10 nm process processors in the Tiger Lake family.
+
+	  Enables -march=tigerlake
+
+config MSAPPHIRERAPIDS
+	bool "Intel Sapphire Rapids"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	select X86_P6_NOP
+	help
+
+	  Select this for fourth-generation 10 nm process processors in the Sapphire Rapids family.
+
+	  Enables -march=sapphirerapids
+
+config MROCKETLAKE
+	bool "Intel Rocket Lake"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	select X86_P6_NOP
+	help
+
+	  Select this for eleventh-generation processors in the Rocket Lake family.
+
+	  Enables -march=rocketlake
+
+config MALDERLAKE
+	bool "Intel Alder Lake"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	select X86_P6_NOP
+	help
+
+	  Select this for twelfth-generation processors in the Alder Lake family.
+
+	  Enables -march=alderlake
+
+config MRAPTORLAKE
+	bool "Intel Raptor Lake"
+	depends on (CC_IS_GCC && GCC_VERSION >= 130000) || (CC_IS_CLANG && CLANG_VERSION >= 150500)
+	select X86_P6_NOP
+	help
+
+	  Select this for thirteenth-generation processors in the Raptor Lake family.
+
+	  Enables -march=raptorlake
+
+config MMETEORLAKE
+	bool "Intel Meteor Lake"
+	depends on (CC_IS_GCC && GCC_VERSION >= 130000) || (CC_IS_CLANG && CLANG_VERSION >= 150500)
+	select X86_P6_NOP
+	help
+
+	  Select this for fourteenth-generation processors in the Meteor Lake family.
+
+	  Enables -march=meteorlake
+
+config MEMERALDRAPIDS
+	bool "Intel Emerald Rapids"
+	depends on (CC_IS_GCC && GCC_VERSION > 130000) || (CC_IS_CLANG && CLANG_VERSION >= 150500)
+	select X86_P6_NOP
+	help
+
+	  Select this for fifth-generation 10 nm process processors in the Emerald Rapids family.
+
+	  Enables -march=emeraldrapids
+
 config GENERIC_CPU
 	bool "Generic-x86-64"
 	depends on X86_64
@@ -294,6 +596,50 @@ config GENERIC_CPU
 	  Generic x86-64 CPU.
 	  Run equally well on all x86-64 CPUs.
 
+config GENERIC_CPU2
+	bool "Generic-x86-64-v2"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	depends on X86_64
+	help
+	  Generic x86-64 CPU.
+	  Run equally well on all x86-64 CPUs with min support of x86-64-v2.
+
+config GENERIC_CPU3
+	bool "Generic-x86-64-v3"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	depends on X86_64
+	help
+	  Generic x86-64-v3 CPU with v3 instructions.
+	  Run equally well on all x86-64 CPUs with min support of x86-64-v3.
+
+config GENERIC_CPU4
+	bool "Generic-x86-64-v4"
+	depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000)
+	depends on X86_64
+	help
+	  Generic x86-64 CPU with v4 instructions.
+	  Run equally well on all x86-64 CPUs with min support of x86-64-v4.
+
+config MNATIVE_INTEL
+	bool "Intel-Native optimizations autodetected by the compiler"
+	help
+
+	  Clang 3.8, GCC 4.2 and above support -march=native, which automatically detects
+	  the optimum settings to use based on your processor. Do NOT use this
+	  for AMD CPUs.  Intel Only!
+
+	  Enables -march=native
+
+config MNATIVE_AMD
+	bool "AMD-Native optimizations autodetected by the compiler"
+	help
+
+	  Clang 3.8, GCC 4.2 and above support -march=native, which automatically detects
+	  the optimum settings to use based on your processor. Do NOT use this
+	  for Intel CPUs.  AMD Only!
+
+	  Enables -march=native
+
 endchoice
 
 config X86_GENERIC
@@ -318,9 +664,17 @@ config X86_INTERNODE_CACHE_SHIFT
 config X86_L1_CACHE_SHIFT
 	int
 	default "7" if MPENTIUM4 || MPSC
-	default "6" if MK7 || MK8 || MPENTIUMM || MCORE2 || MATOM || MVIAC7 || X86_GENERIC || GENERIC_CPU
+	default "6" if MK7 || MK8 || MPENTIUMM || MCORE2 || MATOM || MVIAC7 || MK8SSE3 || MK10 \
+	|| MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER || MPILEDRIVER || MSTEAMROLLER \
+	|| MEXCAVATOR || MZEN || MZEN2 || MZEN3 || MZEN4 || MNEHALEM || MWESTMERE || MSILVERMONT \
+	|| MGOLDMONT || MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL \
+	|| MSKYLAKE || MSKYLAKEX || MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE \
+	|| MTIGERLAKE || MSAPPHIRERAPIDS || MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MMETEORLAKE \
+	|| MEMERALDRAPIDS || MNATIVE_INTEL || MNATIVE_AMD || X86_GENERIC || GENERIC_CPU || GENERIC_CPU2 \
+	|| GENERIC_CPU3 || GENERIC_CPU4
 	default "4" if MELAN || M486SX || M486 || MGEODEGX1
-	default "5" if MWINCHIP3D || MWINCHIPC6 || MCRUSOE || MEFFICEON || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2 || MGEODE_LX
+	default "5" if MWINCHIP3D || MWINCHIPC6 || MCRUSOE || MEFFICEON || MCYRIXIII || MK6 || MPENTIUMIII \
+	|| MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2 || MGEODE_LX
 
 config X86_F00F_BUG
 	def_bool y
@@ -332,15 +686,27 @@ config X86_INVD_BUG
 
 config X86_ALIGNMENT_16
 	def_bool y
-	depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MELAN || MK6 || M586MMX || M586TSC || M586 || M486SX || M486 || MVIAC3_2 || MGEODEGX1
+	depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MELAN || MK6 || M586MMX || M586TSC \
+	|| M586 || M486SX || M486 || MVIAC3_2 || MGEODEGX1
 
 config X86_INTEL_USERCOPY
 	def_bool y
-	depends on MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7 || MEFFICEON || MCORE2
+	depends on MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC \
+	|| MK8 || MK7 || MEFFICEON || MCORE2 || MNEHALEM || MWESTMERE || MSILVERMONT || MGOLDMONT \
+	|| MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL || MSKYLAKE || MSKYLAKEX \
+	|| MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE || MTIGERLAKE || MSAPPHIRERAPIDS \
+	|| MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS || MNATIVE_INTEL
 
 config X86_USE_PPRO_CHECKSUM
 	def_bool y
-	depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MVIAC7 || MEFFICEON || MGEODE_LX || MCORE2 || MATOM
+	depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM \
+	|| MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MVIAC7 || MEFFICEON || MGEODE_LX \
+	|| MCORE2 || MATOM || MK8SSE3 || MK10 || MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER \
+	|| MPILEDRIVER || MSTEAMROLLER || MEXCAVATOR || MZEN || MZEN2 || MZEN3 || MZEN4 || MNEHALEM \
+	|| MWESTMERE || MSILVERMONT || MGOLDMONT || MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE \
+	|| MHASWELL || MBROADWELL || MSKYLAKE || MSKYLAKEX || MCANNONLAKE || MICELAKE \
+	|| MCASCADELAKE || MCOOPERLAKE || MTIGERLAKE || MSAPPHIRERAPIDS || MROCKETLAKE \
+	|| MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS || MNATIVE_INTEL || MNATIVE_AMD
 
 #
 # P6_NOPs are a relatively minor optimization that require a family >=
@@ -356,11 +722,22 @@ config X86_USE_PPRO_CHECKSUM
 config X86_P6_NOP
 	def_bool y
 	depends on X86_64
-	depends on (MCORE2 || MPENTIUM4 || MPSC)
+	depends on (MCORE2 || MPENTIUM4 || MPSC || MNEHALEM || MWESTMERE || MSILVERMONT || MGOLDMONT \
+	|| MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL || MSKYLAKE \
+	|| MSKYLAKEX || MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE || MTIGERLAKE \
+	|| MSAPPHIRERAPIDS || MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS \
+	|| MNATIVE_INTEL)
 
 config X86_TSC
 	def_bool y
-	depends on (MWINCHIP3D || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2 || MVIAC7 || MGEODEGX1 || MGEODE_LX || MCORE2 || MATOM) || X86_64
+	depends on (MWINCHIP3D || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM \
+	|| MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2 || MVIAC7 || MGEODEGX1 \
+	|| MGEODE_LX || MCORE2 || MATOM || MK8SSE3 || MK10 || MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER \
+	|| MPILEDRIVER || MSTEAMROLLER || MEXCAVATOR || MZEN || MZEN2 || MZEN3 || MZEN4 || MNEHALEM \
+	|| MWESTMERE || MSILVERMONT || MGOLDMONT || MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL \
+	|| MBROADWELL || MSKYLAKE || MSKYLAKEX || MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE \
+	|| MTIGERLAKE || MSAPPHIRERAPIDS || MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS \
+	|| MNATIVE_INTEL || MNATIVE_AMD) || X86_64
 
 config X86_HAVE_PAE
 	def_bool y
@@ -368,24 +745,43 @@ config X86_HAVE_PAE
 
 config X86_CMPXCHG64
 	def_bool y
-	depends on X86_HAVE_PAE || M586TSC || M586MMX || MK6 || MK7
+	depends on X86_HAVE_PAE || M586TSC || M586MMX || MK6 || MK7 || MK8 || MK8SSE3 || MK10 \
+	|| MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER || MPILEDRIVER || MSTEAMROLLER || MEXCAVATOR || MZEN \
+	|| MZEN2 || MZEN3 || MZEN4 || MNEHALEM || MWESTMERE || MSILVERMONT || MGOLDMONT || MGOLDMONTPLUS \
+	|| MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL || MSKYLAKE || MSKYLAKEX || MCANNONLAKE \
+	|| MICELAKE || MCASCADELAKE || MCOOPERLAKE || MTIGERLAKE || MSAPPHIRERAPIDS || MROCKETLAKE \
+	|| MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS || MNATIVE_INTEL || MNATIVE_AMD
 
 # this should be set for all -march=.. options where the compiler
 # generates cmov.
 config X86_CMOV
 	def_bool y
-	depends on (MK8 || MK7 || MCORE2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || MCRUSOE || MEFFICEON || X86_64 || MATOM || MGEODE_LX)
+	depends on (MK8 || MK7 || MCORE2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 \
+	|| MVIAC3_2 || MVIAC7 || MCRUSOE || MEFFICEON || X86_64 || MATOM || MGEODE_LX || MK8SSE3 || MK10 \
+	|| MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER || MPILEDRIVER || MSTEAMROLLER || MEXCAVATOR \
+	|| MZEN || MZEN2 || MZEN3 || MZEN4 || MNEHALEM || MWESTMERE || MSILVERMONT || MGOLDMONT \
+	|| MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL || MSKYLAKE || MSKYLAKEX \
+	|| MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE || MTIGERLAKE || MSAPPHIRERAPIDS \
+	|| MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MMETEORLAKE || MEMERALDRAPIDS || MNATIVE_INTEL || MNATIVE_AMD)
 
 config X86_MINIMUM_CPU_FAMILY
 	int
 	default "64" if X86_64
-	default "6" if X86_32 && (MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || MEFFICEON || MATOM || MCORE2 || MK7 || MK8)
+	default "6" if X86_32 && (MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 \
+	|| MVIAC3_2 || MVIAC7 || MEFFICEON || MATOM || MCORE2 || MK7 || MK8 ||  MK8SSE3 \
+	|| MK10 || MBARCELONA || MBOBCAT || MJAGUAR || MBULLDOZER || MPILEDRIVER || MSTEAMROLLER \
+	|| MEXCAVATOR || MZEN || MZEN2 || MZEN3 || MZEN4 || MNEHALEM || MWESTMERE || MSILVERMONT \
+	|| MGOLDMONT || MGOLDMONTPLUS || MSANDYBRIDGE || MIVYBRIDGE || MHASWELL || MBROADWELL \
+	|| MSKYLAKE || MSKYLAKEX || MCANNONLAKE || MICELAKE || MCASCADELAKE || MCOOPERLAKE \
+	|| MTIGERLAKE || MSAPPHIRERAPIDS || MROCKETLAKE || MALDERLAKE || MRAPTORLAKE || MRAPTORLAKE \
+	|| MNATIVE_INTEL || MNATIVE_AMD)
 	default "5" if X86_32 && X86_CMPXCHG64
 	default "4"
 
 config X86_DEBUGCTLMSR
 	def_bool y
-	depends on !(MK6 || MWINCHIPC6 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 || M486SX || M486) && !UML
+	depends on !(MK6 || MWINCHIPC6 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 \
+	|| M486SX || M486) && !UML
 
 config IA32_FEAT_CTL
 	def_bool y
diff --git a/arch/x86/Makefile b/arch/x86/Makefile
index da8f3caf2781..adf396b23669 100644
--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -67,8 +67,8 @@ export BITS
 #
 #    https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53383
 #
-KBUILD_CFLAGS += -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -mno-avx
-KBUILD_RUSTFLAGS += -Ctarget-feature=-sse,-sse2,-sse3,-ssse3,-sse4.1,-sse4.2,-avx,-avx2
+KBUILD_CFLAGS += -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -mno-avx -mno-avx2 -mno-avx512f
+KBUILD_RUSTFLAGS += -Ctarget-feature=-sse,-sse2,-sse3,-ssse3,-sse4.1,-sse4.2,-avx,-avx2,-avx512f
 
 ifeq ($(CONFIG_X86_KERNEL_IBT),y)
 #
@@ -149,21 +149,6 @@ else
 	# Use -mskip-rax-setup if supported.
 	KBUILD_CFLAGS += $(call cc-option,-mskip-rax-setup)
 
-        # FIXME - should be integrated in Makefile.cpu (Makefile_32.cpu)
-        cflags-$(CONFIG_MK8)		+= -march=k8
-        cflags-$(CONFIG_MPSC)		+= -march=nocona
-        cflags-$(CONFIG_MCORE2)		+= -march=core2
-        cflags-$(CONFIG_MATOM)		+= -march=atom
-        cflags-$(CONFIG_GENERIC_CPU)	+= -mtune=generic
-        KBUILD_CFLAGS += $(cflags-y)
-
-        rustflags-$(CONFIG_MK8)		+= -Ctarget-cpu=k8
-        rustflags-$(CONFIG_MPSC)	+= -Ctarget-cpu=nocona
-        rustflags-$(CONFIG_MCORE2)	+= -Ctarget-cpu=core2
-        rustflags-$(CONFIG_MATOM)	+= -Ctarget-cpu=atom
-        rustflags-$(CONFIG_GENERIC_CPU)	+= -Ztune-cpu=generic
-        KBUILD_RUSTFLAGS += $(rustflags-y)
-
         KBUILD_CFLAGS += -mno-red-zone
         KBUILD_CFLAGS += -mcmodel=kernel
         KBUILD_RUSTFLAGS += -Cno-redzone=y
diff --git a/arch/x86/Makefile_32.cpu b/arch/x86/Makefile_32.cpu
index 94834c4b5e5e..81923b4afdf8 100644
--- a/arch/x86/Makefile_32.cpu
+++ b/arch/x86/Makefile_32.cpu
@@ -10,44 +10,3 @@ else
 align		:= -falign-functions=0 -falign-jumps=0 -falign-loops=0
 endif
 
-cflags-$(CONFIG_M486SX)		+= -march=i486
-cflags-$(CONFIG_M486)		+= -march=i486
-cflags-$(CONFIG_M586)		+= -march=i586
-cflags-$(CONFIG_M586TSC)	+= -march=i586
-cflags-$(CONFIG_M586MMX)	+= -march=pentium-mmx
-cflags-$(CONFIG_M686)		+= -march=i686
-cflags-$(CONFIG_MPENTIUMII)	+= -march=i686 $(call tune,pentium2)
-cflags-$(CONFIG_MPENTIUMIII)	+= -march=i686 $(call tune,pentium3)
-cflags-$(CONFIG_MPENTIUMM)	+= -march=i686 $(call tune,pentium3)
-cflags-$(CONFIG_MPENTIUM4)	+= -march=i686 $(call tune,pentium4)
-cflags-$(CONFIG_MK6)		+= -march=k6
-# Please note, that patches that add -march=athlon-xp and friends are pointless.
-# They make zero difference whatsosever to performance at this time.
-cflags-$(CONFIG_MK7)		+= -march=athlon
-cflags-$(CONFIG_MK8)		+= $(call cc-option,-march=k8,-march=athlon)
-cflags-$(CONFIG_MCRUSOE)	+= -march=i686 $(align)
-cflags-$(CONFIG_MEFFICEON)	+= -march=i686 $(call tune,pentium3) $(align)
-cflags-$(CONFIG_MWINCHIPC6)	+= $(call cc-option,-march=winchip-c6,-march=i586)
-cflags-$(CONFIG_MWINCHIP3D)	+= $(call cc-option,-march=winchip2,-march=i586)
-cflags-$(CONFIG_MCYRIXIII)	+= $(call cc-option,-march=c3,-march=i486) $(align)
-cflags-$(CONFIG_MVIAC3_2)	+= $(call cc-option,-march=c3-2,-march=i686)
-cflags-$(CONFIG_MVIAC7)		+= -march=i686
-cflags-$(CONFIG_MCORE2)		+= -march=i686 $(call tune,core2)
-cflags-$(CONFIG_MATOM)		+= $(call cc-option,-march=atom,$(call cc-option,-march=core2,-march=i686)) \
-	$(call cc-option,-mtune=atom,$(call cc-option,-mtune=generic))
-
-# AMD Elan support
-cflags-$(CONFIG_MELAN)		+= -march=i486
-
-# Geode GX1 support
-cflags-$(CONFIG_MGEODEGX1)	+= -march=pentium-mmx
-cflags-$(CONFIG_MGEODE_LX)	+= $(call cc-option,-march=geode,-march=pentium-mmx)
-# add at the end to overwrite eventual tuning options from earlier
-# cpu entries
-cflags-$(CONFIG_X86_GENERIC) 	+= $(call tune,generic,$(call tune,i686))
-
-# Bug fix for binutils: this option is required in order to keep
-# binutils from generating NOPL instructions against our will.
-ifneq ($(CONFIG_X86_P6_NOP),y)
-cflags-y			+= $(call cc-option,-Wa$(comma)-mtune=generic32,)
-endif
diff --git a/arch/x86/include/asm/pci.h b/arch/x86/include/asm/pci.h
index b40c462b4af3..c4e66e60d559 100644
--- a/arch/x86/include/asm/pci.h
+++ b/arch/x86/include/asm/pci.h
@@ -27,6 +27,7 @@ struct pci_sysdata {
 #if IS_ENABLED(CONFIG_VMD)
 	struct pci_dev	*vmd_dev;	/* VMD Device if in Intel VMD domain */
 #endif
+	struct pci_dev	*nvme_remap_dev;	/* AHCI Device if NVME remapped bus */
 };
 
 extern int pci_routeirq;
@@ -70,6 +71,11 @@ static inline bool is_vmd(struct pci_bus *bus)
 #define is_vmd(bus)		false
 #endif /* CONFIG_VMD */
 
+static inline bool is_nvme_remap(struct pci_bus *bus)
+{
+	return to_pci_sysdata(bus)->nvme_remap_dev != NULL;
+}
+
 /* Can be used to override the logic in pci_scan_bus for skipping
    already-configured bus numbers - to be used for buggy BIOSes
    or architectures with incomplete PCI setup by the loader */
diff --git a/arch/x86/include/asm/vermagic.h b/arch/x86/include/asm/vermagic.h
index 75884d2cdec3..02c1386eb653 100644
--- a/arch/x86/include/asm/vermagic.h
+++ b/arch/x86/include/asm/vermagic.h
@@ -17,6 +17,54 @@
 #define MODULE_PROC_FAMILY "586MMX "
 #elif defined CONFIG_MCORE2
 #define MODULE_PROC_FAMILY "CORE2 "
+#elif defined CONFIG_MNATIVE_INTEL
+#define MODULE_PROC_FAMILY "NATIVE_INTEL "
+#elif defined CONFIG_MNATIVE_AMD
+#define MODULE_PROC_FAMILY "NATIVE_AMD "
+#elif defined CONFIG_MNEHALEM
+#define MODULE_PROC_FAMILY "NEHALEM "
+#elif defined CONFIG_MWESTMERE
+#define MODULE_PROC_FAMILY "WESTMERE "
+#elif defined CONFIG_MSILVERMONT
+#define MODULE_PROC_FAMILY "SILVERMONT "
+#elif defined CONFIG_MGOLDMONT
+#define MODULE_PROC_FAMILY "GOLDMONT "
+#elif defined CONFIG_MGOLDMONTPLUS
+#define MODULE_PROC_FAMILY "GOLDMONTPLUS "
+#elif defined CONFIG_MSANDYBRIDGE
+#define MODULE_PROC_FAMILY "SANDYBRIDGE "
+#elif defined CONFIG_MIVYBRIDGE
+#define MODULE_PROC_FAMILY "IVYBRIDGE "
+#elif defined CONFIG_MHASWELL
+#define MODULE_PROC_FAMILY "HASWELL "
+#elif defined CONFIG_MBROADWELL
+#define MODULE_PROC_FAMILY "BROADWELL "
+#elif defined CONFIG_MSKYLAKE
+#define MODULE_PROC_FAMILY "SKYLAKE "
+#elif defined CONFIG_MSKYLAKEX
+#define MODULE_PROC_FAMILY "SKYLAKEX "
+#elif defined CONFIG_MCANNONLAKE
+#define MODULE_PROC_FAMILY "CANNONLAKE "
+#elif defined CONFIG_MICELAKE
+#define MODULE_PROC_FAMILY "ICELAKE "
+#elif defined CONFIG_MCASCADELAKE
+#define MODULE_PROC_FAMILY "CASCADELAKE "
+#elif defined CONFIG_MCOOPERLAKE
+#define MODULE_PROC_FAMILY "COOPERLAKE "
+#elif defined CONFIG_MTIGERLAKE
+#define MODULE_PROC_FAMILY "TIGERLAKE "
+#elif defined CONFIG_MSAPPHIRERAPIDS
+#define MODULE_PROC_FAMILY "SAPPHIRERAPIDS "
+#elif defined CONFIG_ROCKETLAKE
+#define MODULE_PROC_FAMILY "ROCKETLAKE "
+#elif defined CONFIG_MALDERLAKE
+#define MODULE_PROC_FAMILY "ALDERLAKE "
+#elif defined CONFIG_MRAPTORLAKE
+#define MODULE_PROC_FAMILY "RAPTORLAKE "
+#elif defined CONFIG_MMETEORLAKE
+#define MODULE_PROC_FAMILY "METEORLAKE "
+#elif defined CONFIG_MEMERALDRAPIDS
+#define MODULE_PROC_FAMILY "EMERALDRAPIDS "
 #elif defined CONFIG_MATOM
 #define MODULE_PROC_FAMILY "ATOM "
 #elif defined CONFIG_M686
@@ -35,6 +83,32 @@
 #define MODULE_PROC_FAMILY "K7 "
 #elif defined CONFIG_MK8
 #define MODULE_PROC_FAMILY "K8 "
+#elif defined CONFIG_MK8SSE3
+#define MODULE_PROC_FAMILY "K8SSE3 "
+#elif defined CONFIG_MK10
+#define MODULE_PROC_FAMILY "K10 "
+#elif defined CONFIG_MBARCELONA
+#define MODULE_PROC_FAMILY "BARCELONA "
+#elif defined CONFIG_MBOBCAT
+#define MODULE_PROC_FAMILY "BOBCAT "
+#elif defined CONFIG_MBULLDOZER
+#define MODULE_PROC_FAMILY "BULLDOZER "
+#elif defined CONFIG_MPILEDRIVER
+#define MODULE_PROC_FAMILY "PILEDRIVER "
+#elif defined CONFIG_MSTEAMROLLER
+#define MODULE_PROC_FAMILY "STEAMROLLER "
+#elif defined CONFIG_MJAGUAR
+#define MODULE_PROC_FAMILY "JAGUAR "
+#elif defined CONFIG_MEXCAVATOR
+#define MODULE_PROC_FAMILY "EXCAVATOR "
+#elif defined CONFIG_MZEN
+#define MODULE_PROC_FAMILY "ZEN "
+#elif defined CONFIG_MZEN2
+#define MODULE_PROC_FAMILY "ZEN2 "
+#elif defined CONFIG_MZEN3
+#define MODULE_PROC_FAMILY "ZEN3 "
+#elif defined CONFIG_MZEN4
+#define MODULE_PROC_FAMILY "ZEN4 "
 #elif defined CONFIG_MELAN
 #define MODULE_PROC_FAMILY "ELAN "
 #elif defined CONFIG_MCRUSOE
diff --git a/arch/x86/pci/common.c b/arch/x86/pci/common.c
index ddb798603201..7c20387d8202 100644
--- a/arch/x86/pci/common.c
+++ b/arch/x86/pci/common.c
@@ -723,12 +723,15 @@ int pci_ext_cfg_avail(void)
 		return 0;
 }
 
-#if IS_ENABLED(CONFIG_VMD)
 struct pci_dev *pci_real_dma_dev(struct pci_dev *dev)
 {
+#if IS_ENABLED(CONFIG_VMD)
 	if (is_vmd(dev->bus))
 		return to_pci_sysdata(dev->bus)->vmd_dev;
+#endif
+
+	if (is_nvme_remap(dev->bus))
+		return to_pci_sysdata(dev->bus)->nvme_remap_dev;
 
 	return dev;
 }
-#endif
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 9bd57baa4b0b..efe818271cf7 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -7703,6 +7703,7 @@ MODULE_ALIAS("bfq-iosched");
 static int __init bfq_init(void)
 {
 	int ret;
+	char msg[60] = "BFQ I/O-scheduler: BFQ-CachyOS v6.8";
 
 #ifdef CONFIG_BFQ_GROUP_IOSCHED
 	ret = blkcg_policy_register(&blkcg_policy_bfq);
@@ -7734,6 +7735,11 @@ static int __init bfq_init(void)
 	if (ret)
 		goto slab_kill;
 
+#ifdef CONFIG_BFQ_GROUP_IOSCHED
+	strcat(msg, " (with cgroups support)");
+#endif
+	pr_info("%s", msg);
+
 	return 0;
 
 slab_kill:
diff --git a/block/elevator.c b/block/elevator.c
index 5ff093cb3cf8..1c93fe91b006 100644
--- a/block/elevator.c
+++ b/block/elevator.c
@@ -574,9 +574,19 @@ static struct elevator_type *elevator_get_default(struct request_queue *q)
 
 	if (q->nr_hw_queues != 1 &&
 	    !blk_mq_is_shared_tags(q->tag_set->flags))
+#if defined(CONFIG_CACHY) && defined(CONFIG_MQ_IOSCHED_KYBER)
+		return elevator_find_get(q, "kyber");
+#elif defined(CONFIG_CACHY)
+		return elevator_find_get(q, "mq-deadline");
+#else
 		return NULL;
+#endif
 
+#if defined(CONFIG_CACHY) && defined(CONFIG_IOSCHED_BFQ)
+	return elevator_find_get(q, "bfq");
+#else
 	return elevator_find_get(q, "mq-deadline");
+#endif
 }
 
 /*
diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
index df3fd6474bf2..4303eb5fe11b 100644
--- a/drivers/ata/ahci.c
+++ b/drivers/ata/ahci.c
@@ -1547,7 +1547,7 @@ static irqreturn_t ahci_thunderx_irq_handler(int irq, void *dev_instance)
 }
 #endif
 
-static void ahci_remap_check(struct pci_dev *pdev, int bar,
+static int ahci_remap_check(struct pci_dev *pdev, int bar,
 		struct ahci_host_priv *hpriv)
 {
 	int i;
@@ -1560,7 +1560,7 @@ static void ahci_remap_check(struct pci_dev *pdev, int bar,
 	    pci_resource_len(pdev, bar) < SZ_512K ||
 	    bar != AHCI_PCI_BAR_STANDARD ||
 	    !(readl(hpriv->mmio + AHCI_VSCAP) & 1))
-		return;
+		return 0;
 
 	cap = readq(hpriv->mmio + AHCI_REMAP_CAP);
 	for (i = 0; i < AHCI_MAX_REMAP; i++) {
@@ -1575,18 +1575,11 @@ static void ahci_remap_check(struct pci_dev *pdev, int bar,
 	}
 
 	if (!hpriv->remapped_nvme)
-		return;
-
-	dev_warn(&pdev->dev, "Found %u remapped NVMe devices.\n",
-		 hpriv->remapped_nvme);
-	dev_warn(&pdev->dev,
-		 "Switch your BIOS from RAID to AHCI mode to use them.\n");
+		return 0;
 
-	/*
-	 * Don't rely on the msi-x capability in the remap case,
-	 * share the legacy interrupt across ahci and remapped devices.
-	 */
-	hpriv->flags |= AHCI_HFLAG_NO_MSI;
+	/* Abort probe, allowing intel-nvme-remap to step in when available */
+	dev_info(&pdev->dev, "Device will be handled by intel-nvme-remap.\n");
+	return -ENODEV;
 }
 
 static int ahci_get_irq_vector(struct ata_host *host, int port)
@@ -1806,7 +1799,9 @@ static int ahci_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	hpriv->mmio = pcim_iomap_table(pdev)[ahci_pci_bar];
 
 	/* detect remapped nvme devices */
-	ahci_remap_check(pdev, ahci_pci_bar, hpriv);
+	rc = ahci_remap_check(pdev, ahci_pci_bar, hpriv);
+	if (rc)
+		return rc;
 
 	sysfs_add_file_to_group(&pdev->dev.kobj,
 				&dev_attr_remapped_nvme.attr,
diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86
index 438c9e75a04d..1bbfeca5f01e 100644
--- a/drivers/cpufreq/Kconfig.x86
+++ b/drivers/cpufreq/Kconfig.x86
@@ -9,7 +9,6 @@ config X86_INTEL_PSTATE
 	select ACPI_PROCESSOR if ACPI
 	select ACPI_CPPC_LIB if X86_64 && ACPI && SCHED_MC_PRIO
 	select CPU_FREQ_GOV_PERFORMANCE
-	select CPU_FREQ_GOV_SCHEDUTIL if SMP
 	help
 	  This driver provides a P state for Intel core processors.
 	  The driver implements an internal governor and will become
@@ -39,7 +38,6 @@ config X86_AMD_PSTATE
 	depends on X86 && ACPI
 	select ACPI_PROCESSOR
 	select ACPI_CPPC_LIB if X86_64
-	select CPU_FREQ_GOV_SCHEDUTIL if SMP
 	help
 	  This driver adds a CPUFreq driver which utilizes a fine grain
 	  processor performance frequency control range instead of legacy
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 79827a6dcd7f..ee85a2352771 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -153,6 +153,7 @@ struct amdgpu_watchdog_timer
  */
 extern int amdgpu_modeset;
 extern unsigned int amdgpu_vram_limit;
+extern int amdgpu_ignore_min_pcap;
 extern int amdgpu_vis_vram_limit;
 extern int amdgpu_gart_size;
 extern int amdgpu_gtt_size;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 586f4d03039d..a2524615b696 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -132,6 +132,7 @@ enum AMDGPU_DEBUG_MASK {
 };
 
 unsigned int amdgpu_vram_limit = UINT_MAX;
+int amdgpu_ignore_min_pcap = 0; /* do not ignore by default */
 int amdgpu_vis_vram_limit;
 int amdgpu_gart_size = -1; /* auto */
 int amdgpu_gtt_size = -1; /* auto */
@@ -241,6 +242,15 @@ struct amdgpu_watchdog_timer amdgpu_watchdog_timer = {
 	.period = 0x0, /* default to 0x0 (timeout disable) */
 };
 
+/**
+ * DOC: ignore_min_pcap (int)
+ * Ignore the minimum power cap.
+ * Useful on graphics cards where the minimum power cap is very high.
+ * The default is 0 (Do not ignore).
+ */
+MODULE_PARM_DESC(ignore_min_pcap, "Ignore the minimum power cap");
+module_param_named(ignore_min_pcap, amdgpu_ignore_min_pcap, int, 0600);
+
 /**
  * DOC: vramlimit (int)
  * Restrict the total amount of VRAM in MiB for testing.  The default is 0 (Use full VRAM).
diff --git a/drivers/gpu/drm/amd/display/Kconfig b/drivers/gpu/drm/amd/display/Kconfig
index 901d1961b739..05c49141f580 100644
--- a/drivers/gpu/drm/amd/display/Kconfig
+++ b/drivers/gpu/drm/amd/display/Kconfig
@@ -51,4 +51,10 @@ config DRM_AMD_SECURE_DISPLAY
 	  This option enables the calculation of crc of specific region via
 	  debugfs. Cooperate with specific DMCU FW.
 
+config AMD_PRIVATE_COLOR
+	bool "Enable KMS color management by AMD for AMD"
+	default n
+	help
+	  This option extends the KMS color management API with AMD driver-specific properties to enhance the color management support on AMD Steam Deck.
+
 endmenu
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 1eb0f82e9dfa..5e0c551759ab 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -4072,7 +4072,7 @@ static int amdgpu_dm_mode_config_init(struct amdgpu_device *adev)
 		return r;
 	}
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 	if (amdgpu_dm_create_color_properties(adev))
 		return -ENOMEM;
 #endif
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c
index c87b64e464ed..6fe07243adc3 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c
@@ -97,7 +97,7 @@ static inline struct fixed31_32 amdgpu_dm_fixpt_from_s3132(__u64 x)
 	return val;
 }
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 /* Pre-defined Transfer Functions (TF)
  *
  * AMD driver supports pre-defined mathematical functions for transferring
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
index 6e715ef3a556..11c7199ec3b3 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
@@ -290,7 +290,7 @@ static int amdgpu_dm_crtc_late_register(struct drm_crtc *crtc)
 }
 #endif
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 /**
  * dm_crtc_additional_color_mgmt - enable additional color properties
  * @crtc: DRM CRTC
@@ -372,7 +372,7 @@ static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = {
 #if defined(CONFIG_DEBUG_FS)
 	.late_register = amdgpu_dm_crtc_late_register,
 #endif
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 	.atomic_set_property = amdgpu_dm_atomic_crtc_set_property,
 	.atomic_get_property = amdgpu_dm_atomic_crtc_get_property,
 #endif
@@ -551,7 +551,7 @@ int amdgpu_dm_crtc_init(struct amdgpu_display_manager *dm,
 
 	drm_mode_crtc_set_gamma_size(&acrtc->base, MAX_COLOR_LEGACY_LUT_ENTRIES);
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 	dm_crtc_additional_color_mgmt(&acrtc->base);
 #endif
 	return 0;
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c
index 8a4c40b4c27e..779880c64575 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c
@@ -1468,7 +1468,7 @@ static void amdgpu_dm_plane_drm_plane_destroy_state(struct drm_plane *plane,
 	drm_atomic_helper_plane_destroy_state(plane, state);
 }
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 static void
 dm_atomic_plane_attach_color_mgmt_properties(struct amdgpu_display_manager *dm,
 					     struct drm_plane *plane)
@@ -1659,7 +1659,7 @@ static const struct drm_plane_funcs dm_plane_funcs = {
 	.atomic_duplicate_state = amdgpu_dm_plane_drm_plane_duplicate_state,
 	.atomic_destroy_state = amdgpu_dm_plane_drm_plane_destroy_state,
 	.format_mod_supported = amdgpu_dm_plane_format_mod_supported,
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 	.atomic_set_property = dm_atomic_plane_set_property,
 	.atomic_get_property = dm_atomic_plane_get_property,
 #endif
@@ -1742,7 +1742,7 @@ int amdgpu_dm_plane_init(struct amdgpu_display_manager *dm,
 
 	drm_plane_helper_add(plane, &dm_plane_helper_funcs);
 
-#ifdef AMD_PRIVATE_COLOR
+#ifdef CONFIG_AMD_PRIVATE_COLOR
 	dm_atomic_plane_attach_color_mgmt_properties(dm, plane);
 #endif
 	/* Create (reset) the plane state */
diff --git a/drivers/gpu/drm/amd/pm/amdgpu_pm.c b/drivers/gpu/drm/amd/pm/amdgpu_pm.c
index 39c5e1dfa275..ee97bb26a8ef 100644
--- a/drivers/gpu/drm/amd/pm/amdgpu_pm.c
+++ b/drivers/gpu/drm/amd/pm/amdgpu_pm.c
@@ -3034,6 +3034,9 @@ static ssize_t amdgpu_hwmon_show_power_cap_min(struct device *dev,
 					 struct device_attribute *attr,
 					 char *buf)
 {
+	if (amdgpu_ignore_min_pcap)
+		return sysfs_emit(buf, "%i\n", 0);
+
 	return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_MIN);
 }
 
diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 0ad947df777a..7b82e3ef7c91 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -2695,7 +2695,10 @@ int smu_get_power_limit(void *handle,
 			*limit = smu->max_power_limit;
 			break;
 		case SMU_PPT_LIMIT_MIN:
-			*limit = smu->min_power_limit;
+			if (amdgpu_ignore_min_pcap)
+				*limit = 0;
+			else
+				*limit = smu->min_power_limit;
 			break;
 		default:
 			return -EINVAL;
@@ -2719,7 +2722,14 @@ static int smu_set_power_limit(void *handle, uint32_t limit)
 		if (smu->ppt_funcs->set_power_limit)
 			return smu->ppt_funcs->set_power_limit(smu, limit_type, limit);
 
-	if ((limit > smu->max_power_limit) || (limit < smu->min_power_limit)) {
+	if (amdgpu_ignore_min_pcap) {
+		if ((limit > smu->max_power_limit)) {
+			dev_err(smu->adev->dev,
+				"New power limit (%d) is over the max allowed %d\n",
+				limit, smu->max_power_limit);
+			return -EINVAL;
+		}
+	} else if ((limit > smu->max_power_limit) || (limit < smu->min_power_limit)) {
 		dev_err(smu->adev->dev,
 			"New power limit (%d) is out of range [%d,%d]\n",
 			limit, smu->min_power_limit, smu->max_power_limit);
diff --git a/drivers/i2c/busses/Kconfig b/drivers/i2c/busses/Kconfig
index 28eb48dd5b32..1cf4c700b108 100644
--- a/drivers/i2c/busses/Kconfig
+++ b/drivers/i2c/busses/Kconfig
@@ -230,6 +230,15 @@ config I2C_CHT_WC
 	  combined with a FUSB302 Type-C port-controller as such it is advised
 	  to also select CONFIG_TYPEC_FUSB302=m.
 
+config I2C_NCT6775
+	tristate "Nuvoton NCT6775 and compatible SMBus controller"
+	help
+		If you say yes to this option, support will be included for the
+		Nuvoton NCT6775 and compatible SMBus controllers.
+
+		This driver can also be built as a module.  If so, the module
+		will be called i2c-nct6775.
+
 config I2C_NFORCE2
 	tristate "Nvidia nForce2, nForce3 and nForce4"
 	depends on PCI
diff --git a/drivers/i2c/busses/Makefile b/drivers/i2c/busses/Makefile
index aa0ee8ecd6f2..020714113e9a 100644
--- a/drivers/i2c/busses/Makefile
+++ b/drivers/i2c/busses/Makefile
@@ -20,6 +20,7 @@ obj-$(CONFIG_I2C_CHT_WC)	+= i2c-cht-wc.o
 obj-$(CONFIG_I2C_I801)		+= i2c-i801.o
 obj-$(CONFIG_I2C_ISCH)		+= i2c-isch.o
 obj-$(CONFIG_I2C_ISMT)		+= i2c-ismt.o
+obj-$(CONFIG_I2C_NCT6775)   += i2c-nct6775.o
 obj-$(CONFIG_I2C_NFORCE2)	+= i2c-nforce2.o
 obj-$(CONFIG_I2C_NFORCE2_S4985)	+= i2c-nforce2-s4985.o
 obj-$(CONFIG_I2C_NVIDIA_GPU)	+= i2c-nvidia-gpu.o
diff --git a/drivers/i2c/busses/i2c-nct6775.c b/drivers/i2c/busses/i2c-nct6775.c
new file mode 100644
index 000000000000..e919d1e10c51
--- /dev/null
+++ b/drivers/i2c/busses/i2c-nct6775.c
@@ -0,0 +1,648 @@
+/*
+ * i2c-nct6775 - Driver for the SMBus master functionality of
+ *	       Nuvoton NCT677x Super-I/O chips
+ *
+ * Copyright (C) 2019  Adam Honse <calcprogrammer1@gmail.com>
+ *
+ * Derived from nct6775 hwmon driver
+ * Copyright (C) 2012  Guenter Roeck <linux@roeck-us.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/jiffies.h>
+#include <linux/platform_device.h>
+#include <linux/hwmon.h>
+#include <linux/hwmon-sysfs.h>
+#include <linux/hwmon-vid.h>
+#include <linux/err.h>
+#include <linux/mutex.h>
+#include <linux/delay.h>
+#include <linux/ioport.h>
+#include <linux/i2c.h>
+#include <linux/acpi.h>
+#include <linux/bitops.h>
+#include <linux/dmi.h>
+#include <linux/io.h>
+#include <linux/nospec.h>
+
+#define DRVNAME "i2c-nct6775"
+
+/* Nuvoton SMBus address offsets */
+#define SMBHSTDAT       (0 + nuvoton_nct6793d_smba)
+#define SMBBLKSZ        (1 + nuvoton_nct6793d_smba)
+#define SMBHSTCMD       (2 + nuvoton_nct6793d_smba)
+#define SMBHSTIDX       (3 + nuvoton_nct6793d_smba)  //Index field is the Command field on other controllers
+#define SMBHSTCTL       (4 + nuvoton_nct6793d_smba)
+#define SMBHSTADD       (5 + nuvoton_nct6793d_smba)
+#define SMBHSTERR       (9 + nuvoton_nct6793d_smba)
+#define SMBHSTSTS       (0xE + nuvoton_nct6793d_smba)
+
+/* Command register */
+#define NCT6793D_READ_BYTE      0
+#define NCT6793D_READ_WORD      1
+#define NCT6793D_READ_BLOCK     2
+#define NCT6793D_BLOCK_WRITE_READ_PROC_CALL 3
+#define NCT6793D_PROC_CALL      4
+#define NCT6793D_WRITE_BYTE     8
+#define NCT6793D_WRITE_WORD     9
+#define NCT6793D_WRITE_BLOCK    10
+
+/* Control register */
+#define NCT6793D_MANUAL_START   128
+#define NCT6793D_SOFT_RESET     64
+
+/* Error register */
+#define NCT6793D_NO_ACK         32
+
+/* Status register */
+#define NCT6793D_FIFO_EMPTY     1
+#define NCT6793D_FIFO_FULL      2
+#define NCT6793D_MANUAL_ACTIVE  4
+
+#define NCT6775_LD_SMBUS		0x0B
+
+/* Other settings */
+#define MAX_RETRIES		400
+
+enum kinds { nct6106, nct6775, nct6776, nct6779, nct6791, nct6792, nct6793,
+	     nct6795, nct6796, nct6798 };
+
+struct nct6775_sio_data {
+	int sioreg;
+	enum kinds kind;
+};
+
+/* used to set data->name = nct6775_device_names[data->sio_kind] */
+static const char * const nct6775_device_names[] = {
+	"nct6106",
+	"nct6775",
+	"nct6776",
+	"nct6779",
+	"nct6791",
+	"nct6792",
+	"nct6793",
+	"nct6795",
+	"nct6796",
+	"nct6798",
+};
+
+static const char * const nct6775_sio_names[] __initconst = {
+	"NCT6106D",
+	"NCT6775F",
+	"NCT6776D/F",
+	"NCT6779D",
+	"NCT6791D",
+	"NCT6792D",
+	"NCT6793D",
+	"NCT6795D",
+	"NCT6796D",
+	"NCT6798D",
+};
+
+#define SIO_REG_LDSEL		0x07	/* Logical device select */
+#define SIO_REG_DEVID		0x20	/* Device ID (2 bytes) */
+#define SIO_REG_SMBA		0x62	/* SMBus base address register */
+
+#define SIO_NCT6106_ID		0xc450
+#define SIO_NCT6775_ID		0xb470
+#define SIO_NCT6776_ID		0xc330
+#define SIO_NCT6779_ID		0xc560
+#define SIO_NCT6791_ID		0xc800
+#define SIO_NCT6792_ID		0xc910
+#define SIO_NCT6793_ID		0xd120
+#define SIO_NCT6795_ID		0xd350
+#define SIO_NCT6796_ID		0xd420
+#define SIO_NCT6798_ID		0xd428
+#define SIO_ID_MASK			0xFFF0
+
+static inline void
+superio_outb(int ioreg, int reg, int val)
+{
+	outb(reg, ioreg);
+	outb(val, ioreg + 1);
+}
+
+static inline int
+superio_inb(int ioreg, int reg)
+{
+	outb(reg, ioreg);
+	return inb(ioreg + 1);
+}
+
+static inline void
+superio_select(int ioreg, int ld)
+{
+	outb(SIO_REG_LDSEL, ioreg);
+	outb(ld, ioreg + 1);
+}
+
+static inline int
+superio_enter(int ioreg)
+{
+	/*
+	 * Try to reserve <ioreg> and <ioreg + 1> for exclusive access.
+	 */
+	if (!request_muxed_region(ioreg, 2, DRVNAME))
+		return -EBUSY;
+
+	outb(0x87, ioreg);
+	outb(0x87, ioreg);
+
+	return 0;
+}
+
+static inline void
+superio_exit(int ioreg)
+{
+	outb(0xaa, ioreg);
+	outb(0x02, ioreg);
+	outb(0x02, ioreg + 1);
+	release_region(ioreg, 2);
+}
+
+/*
+ * ISA constants
+ */
+
+#define IOREGION_ALIGNMENT	(~7)
+#define IOREGION_LENGTH		2
+#define ADDR_REG_OFFSET		0
+#define DATA_REG_OFFSET		1
+
+#define NCT6775_REG_BANK	0x4E
+#define NCT6775_REG_CONFIG	0x40
+
+static struct i2c_adapter *nct6775_adapter;
+
+struct i2c_nct6775_adapdata {
+	unsigned short smba;
+};
+
+/* Return negative errno on error. */
+static s32 nct6775_access(struct i2c_adapter * adap, u16 addr,
+		 unsigned short flags, char read_write,
+		 u8 command, int size, union i2c_smbus_data * data)
+{
+	struct i2c_nct6775_adapdata *adapdata = i2c_get_adapdata(adap);
+	unsigned short nuvoton_nct6793d_smba = adapdata->smba;
+	int i, len, cnt;
+	union i2c_smbus_data tmp_data;
+	int timeout = 0;
+
+	tmp_data.word = 0;
+	cnt = 0;
+	len = 0;
+
+	outb_p(NCT6793D_SOFT_RESET, SMBHSTCTL);
+
+	switch (size) {
+		case I2C_SMBUS_QUICK:
+			outb_p((addr << 1) | read_write,
+			       SMBHSTADD);
+			break;
+		case I2C_SMBUS_BYTE_DATA:
+			tmp_data.byte = data->byte;
+			fallthrough;
+		case I2C_SMBUS_BYTE:
+			outb_p((addr << 1) | read_write,
+			       SMBHSTADD);
+			outb_p(command, SMBHSTIDX);
+			if (read_write == I2C_SMBUS_WRITE) {
+				outb_p(tmp_data.byte, SMBHSTDAT);
+				outb_p(NCT6793D_WRITE_BYTE, SMBHSTCMD);
+			}
+			else {
+				outb_p(NCT6793D_READ_BYTE, SMBHSTCMD);
+			}
+			break;
+		case I2C_SMBUS_WORD_DATA:
+			outb_p((addr << 1) | read_write,
+			       SMBHSTADD);
+			outb_p(command, SMBHSTIDX);
+			if (read_write == I2C_SMBUS_WRITE) {
+				outb_p(data->word & 0xff, SMBHSTDAT);
+				outb_p((data->word & 0xff00) >> 8, SMBHSTDAT);
+				outb_p(NCT6793D_WRITE_WORD, SMBHSTCMD);
+			}
+			else {
+				outb_p(NCT6793D_READ_WORD, SMBHSTCMD);
+			}
+			break;
+		case I2C_SMBUS_BLOCK_DATA:
+			outb_p((addr << 1) | read_write,
+			       SMBHSTADD);
+			outb_p(command, SMBHSTIDX);
+			if (read_write == I2C_SMBUS_WRITE) {
+				len = data->block[0];
+				if (len == 0 || len > I2C_SMBUS_BLOCK_MAX)
+					return -EINVAL;
+				outb_p(len, SMBBLKSZ);
+
+				cnt = 1;
+				if (len >= 4) {
+					for (i = cnt; i <= 4; i++) {
+						outb_p(data->block[i], SMBHSTDAT);
+					}
+
+					len -= 4;
+					cnt += 4;
+				}
+				else {
+					for (i = cnt; i <= len; i++ ) {
+						outb_p(data->block[i], SMBHSTDAT);
+					}
+
+					len = 0;
+				}
+
+				outb_p(NCT6793D_WRITE_BLOCK, SMBHSTCMD);
+			}
+			else {
+				return -ENOTSUPP;
+			}
+			break;
+		default:
+			dev_warn(&adap->dev, "Unsupported transaction %d\n", size);
+			return -EOPNOTSUPP;
+	}
+
+	outb_p(NCT6793D_MANUAL_START, SMBHSTCTL);
+
+	while ((size == I2C_SMBUS_BLOCK_DATA) && (len > 0)) {
+		if (read_write == I2C_SMBUS_WRITE) {
+			timeout = 0;
+			while ((inb_p(SMBHSTSTS) & NCT6793D_FIFO_EMPTY) == 0)
+			{
+				if(timeout > MAX_RETRIES)
+				{
+					return -ETIMEDOUT;
+				}
+				usleep_range(250, 500);
+				timeout++;
+			}
+
+			//Load more bytes into FIFO
+			if (len >= 4) {
+				for (i = cnt; i <= (cnt + 4); i++) {
+					outb_p(data->block[i], SMBHSTDAT);
+				}
+
+				len -= 4;
+				cnt += 4;
+			}
+			else {
+				for (i = cnt; i <= (cnt + len); i++) {
+					outb_p(data->block[i], SMBHSTDAT);
+				}
+
+				len = 0;
+			}
+		}
+		else {
+			return -ENOTSUPP;
+		}
+		
+	}
+
+	//wait for manual mode to complete
+	timeout = 0;
+	while ((inb_p(SMBHSTSTS) & NCT6793D_MANUAL_ACTIVE) != 0)
+	{
+		if(timeout > MAX_RETRIES)
+		{
+			return -ETIMEDOUT;
+		}
+		usleep_range(250, 500);
+		timeout++;
+	}
+
+	if ((inb_p(SMBHSTERR) & NCT6793D_NO_ACK) != 0) {    	
+		return -ENXIO;
+	}
+	else if ((read_write == I2C_SMBUS_WRITE) || (size == I2C_SMBUS_QUICK)) {
+		return 0;
+	}
+
+	switch (size) {
+		case I2C_SMBUS_QUICK:
+		case I2C_SMBUS_BYTE_DATA:
+			data->byte = inb_p(SMBHSTDAT);
+			break;
+		case I2C_SMBUS_WORD_DATA:
+			data->word = inb_p(SMBHSTDAT) + (inb_p(SMBHSTDAT) << 8);
+			break;
+	}
+	return 0;
+}
+
+static u32 nct6775_func(struct i2c_adapter *adapter)
+{
+	return I2C_FUNC_SMBUS_QUICK | I2C_FUNC_SMBUS_BYTE |
+	    I2C_FUNC_SMBUS_BYTE_DATA | I2C_FUNC_SMBUS_WORD_DATA |
+	    I2C_FUNC_SMBUS_BLOCK_DATA;
+}
+
+static const struct i2c_algorithm smbus_algorithm = {
+	.smbus_xfer	= nct6775_access,
+	.functionality	= nct6775_func,
+};
+
+static int nct6775_add_adapter(unsigned short smba, const char *name, struct i2c_adapter **padap)
+{
+	struct i2c_adapter *adap;
+	struct i2c_nct6775_adapdata *adapdata;
+	int retval;
+
+	adap = kzalloc(sizeof(*adap), GFP_KERNEL);
+	if (adap == NULL) {
+		return -ENOMEM;
+	}
+
+	adap->owner = THIS_MODULE;
+	adap->class = I2C_CLASS_HWMON | I2C_CLASS_SPD;
+	adap->algo = &smbus_algorithm;
+
+	adapdata = kzalloc(sizeof(*adapdata), GFP_KERNEL);
+	if (adapdata == NULL) {
+		kfree(adap);
+		return -ENOMEM;
+	}
+
+	adapdata->smba = smba;
+
+	snprintf(adap->name, sizeof(adap->name),
+		"SMBus NCT67xx adapter%s at %04x", name, smba);
+
+	i2c_set_adapdata(adap, adapdata);
+
+	retval = i2c_add_adapter(adap);
+	if (retval) {
+		kfree(adapdata);
+		kfree(adap);
+		return retval;
+	}
+
+	*padap = adap;
+	return 0;
+}
+
+static void nct6775_remove_adapter(struct i2c_adapter *adap)
+{
+	struct i2c_nct6775_adapdata *adapdata = i2c_get_adapdata(adap);
+
+	if (adapdata->smba) {
+		i2c_del_adapter(adap);
+		kfree(adapdata);
+		kfree(adap);
+	}
+}
+
+//static SIMPLE_DEV_PM_OPS(nct6775_dev_pm_ops, nct6775_suspend, nct6775_resume);
+
+/*
+ * when Super-I/O functions move to a separate file, the Super-I/O
+ * bus will manage the lifetime of the device and this module will only keep
+ * track of the nct6775 driver. But since we use platform_device_alloc(), we
+ * must keep track of the device
+ */
+static struct platform_device *pdev[2];
+
+static int nct6775_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct nct6775_sio_data *sio_data = dev_get_platdata(dev);
+	struct resource *res;
+
+	res = platform_get_resource(pdev, IORESOURCE_IO, 0);
+	if (!devm_request_region(&pdev->dev, res->start, IOREGION_LENGTH,
+				 DRVNAME))
+		return -EBUSY;
+
+	switch (sio_data->kind) {
+	case nct6791:
+	case nct6792:
+	case nct6793:
+	case nct6795:
+	case nct6796:
+	case nct6798:
+		nct6775_add_adapter(res->start, "", &nct6775_adapter);
+		break;
+	default:
+		return -ENODEV;
+	}
+
+	return 0;
+}
+/*
+static void nct6791_enable_io_mapping(int sioaddr)
+{
+	int val;
+
+	val = superio_inb(sioaddr, NCT6791_REG_HM_IO_SPACE_LOCK_ENABLE);
+	if (val & 0x10) {
+		pr_info("Enabling hardware monitor logical device mappings.\n");
+		superio_outb(sioaddr, NCT6791_REG_HM_IO_SPACE_LOCK_ENABLE,
+			     val & ~0x10);
+	}
+}*/
+
+static struct platform_driver i2c_nct6775_driver = {
+	.driver = {
+		.name	= DRVNAME,
+//		.pm	= &nct6775_dev_pm_ops,
+	},
+	.probe		= nct6775_probe,
+};
+
+static void __exit i2c_nct6775_exit(void)
+{
+	int i;
+
+	if(nct6775_adapter)
+		nct6775_remove_adapter(nct6775_adapter);
+
+	for (i = 0; i < ARRAY_SIZE(pdev); i++) {
+		if (pdev[i])
+			platform_device_unregister(pdev[i]);
+	}
+	platform_driver_unregister(&i2c_nct6775_driver);
+}
+
+/* nct6775_find() looks for a '627 in the Super-I/O config space */
+static int __init nct6775_find(int sioaddr, struct nct6775_sio_data *sio_data)
+{
+	u16 val;
+	int err;
+	int addr;
+
+	err = superio_enter(sioaddr);
+	if (err)
+		return err;
+
+	val = (superio_inb(sioaddr, SIO_REG_DEVID) << 8) |
+		superio_inb(sioaddr, SIO_REG_DEVID + 1);
+
+	switch (val & SIO_ID_MASK) {
+	case SIO_NCT6106_ID:
+		sio_data->kind = nct6106;
+		break;
+	case SIO_NCT6775_ID:
+		sio_data->kind = nct6775;
+		break;
+	case SIO_NCT6776_ID:
+		sio_data->kind = nct6776;
+		break;
+	case SIO_NCT6779_ID:
+		sio_data->kind = nct6779;
+		break;
+	case SIO_NCT6791_ID:
+		sio_data->kind = nct6791;
+		break;
+	case SIO_NCT6792_ID:
+		sio_data->kind = nct6792;
+		break;
+	case SIO_NCT6793_ID:
+		sio_data->kind = nct6793;
+		break;
+	case SIO_NCT6795_ID:
+		sio_data->kind = nct6795;
+		break;
+	case SIO_NCT6796_ID:
+		sio_data->kind = nct6796;
+		break;
+	case SIO_NCT6798_ID:
+		sio_data->kind = nct6798;
+		break;
+	default:
+		if (val != 0xffff)
+			pr_debug("unsupported chip ID: 0x%04x\n", val);
+		superio_exit(sioaddr);
+		return -ENODEV;
+	}
+
+	/* We have a known chip, find the SMBus I/O address */
+	superio_select(sioaddr, NCT6775_LD_SMBUS);
+	val = (superio_inb(sioaddr, SIO_REG_SMBA) << 8)
+	    | superio_inb(sioaddr, SIO_REG_SMBA + 1);
+	addr = val & IOREGION_ALIGNMENT;
+	if (addr == 0) {
+		pr_err("Refusing to enable a Super-I/O device with a base I/O port 0\n");
+		superio_exit(sioaddr);
+		return -ENODEV;
+	}
+
+	//if (sio_data->kind == nct6791 || sio_data->kind == nct6792 ||
+	//    sio_data->kind == nct6793 || sio_data->kind == nct6795 ||
+	//    sio_data->kind == nct6796)
+	//	nct6791_enable_io_mapping(sioaddr);
+
+	superio_exit(sioaddr);
+	pr_info("Found %s or compatible chip at %#x:%#x\n",
+		nct6775_sio_names[sio_data->kind], sioaddr, addr);
+	sio_data->sioreg = sioaddr;
+
+	return addr;
+}
+
+static int __init i2c_nct6775_init(void)
+{
+	int i, err;
+	bool found = false;
+	int address;
+	struct resource res;
+	struct nct6775_sio_data sio_data;
+	int sioaddr[2] = { 0x2e, 0x4e };
+
+	err = platform_driver_register(&i2c_nct6775_driver);
+	if (err)
+		return err;
+
+	/*
+	 * initialize sio_data->kind and sio_data->sioreg.
+	 *
+	 * when Super-I/O functions move to a separate file, the Super-I/O
+	 * driver will probe 0x2e and 0x4e and auto-detect the presence of a
+	 * nct6775 hardware monitor, and call probe()
+	 */
+	for (i = 0; i < ARRAY_SIZE(pdev); i++) {
+		address = nct6775_find(sioaddr[i], &sio_data);
+		if (address <= 0)
+			continue;
+
+		found = true;
+
+		pdev[i] = platform_device_alloc(DRVNAME, address);
+		if (!pdev[i]) {
+			err = -ENOMEM;
+			goto exit_device_unregister;
+		}
+
+		err = platform_device_add_data(pdev[i], &sio_data,
+					       sizeof(struct nct6775_sio_data));
+		if (err)
+			goto exit_device_put;
+
+		memset(&res, 0, sizeof(res));
+		res.name = DRVNAME;
+		res.start = address;
+		res.end = address + IOREGION_LENGTH - 1;
+		res.flags = IORESOURCE_IO;
+
+		err = acpi_check_resource_conflict(&res);
+		if (err) {
+			platform_device_put(pdev[i]);
+			pdev[i] = NULL;
+			continue;
+		}
+
+		err = platform_device_add_resources(pdev[i], &res, 1);
+		if (err)
+			goto exit_device_put;
+
+		/* platform_device_add calls probe() */
+		err = platform_device_add(pdev[i]);
+		if (err)
+			goto exit_device_put;
+	}
+	if (!found) {
+		err = -ENODEV;
+		goto exit_unregister;
+	}
+
+	return 0;
+
+exit_device_put:
+	platform_device_put(pdev[i]);
+exit_device_unregister:
+	while (--i >= 0) {
+		if (pdev[i])
+			platform_device_unregister(pdev[i]);
+	}
+exit_unregister:
+	platform_driver_unregister(&i2c_nct6775_driver);
+	return err;
+}
+
+MODULE_AUTHOR("Adam Honse <calcprogrammer1@gmail.com>");
+MODULE_DESCRIPTION("SMBus driver for NCT6775F and compatible chips");
+MODULE_LICENSE("GPL");
+
+module_init(i2c_nct6775_init);
+module_exit(i2c_nct6775_exit);
diff --git a/drivers/i2c/busses/i2c-piix4.c b/drivers/i2c/busses/i2c-piix4.c
index 6a0392172b2f..e7dd007bf6b1 100644
--- a/drivers/i2c/busses/i2c-piix4.c
+++ b/drivers/i2c/busses/i2c-piix4.c
@@ -568,11 +568,11 @@ static int piix4_transaction(struct i2c_adapter *piix4_adapter)
 	if (srvrworks_csb5_delay) /* Extra delay for SERVERWORKS_CSB5 */
 		usleep_range(2000, 2100);
 	else
-		usleep_range(250, 500);
+		usleep_range(25, 50);
 
 	while ((++timeout < MAX_TIMEOUT) &&
 	       ((temp = inb_p(SMBHSTSTS)) & 0x01))
-		usleep_range(250, 500);
+		usleep_range(25, 50);
 
 	/* If the SMBus is still busy, we give up */
 	if (timeout == MAX_TIMEOUT) {
diff --git a/drivers/input/evdev.c b/drivers/input/evdev.c
index 51e0c4954600..35c3ad741870 100644
--- a/drivers/input/evdev.c
+++ b/drivers/input/evdev.c
@@ -46,6 +46,7 @@ struct evdev_client {
 	struct fasync_struct *fasync;
 	struct evdev *evdev;
 	struct list_head node;
+	struct rcu_head rcu;
 	enum input_clock_type clk_type;
 	bool revoked;
 	unsigned long *evmasks[EV_CNT];
@@ -377,13 +378,22 @@ static void evdev_attach_client(struct evdev *evdev,
 	spin_unlock(&evdev->client_lock);
 }
 
+static void evdev_reclaim_client(struct rcu_head *rp)
+{
+	struct evdev_client *client = container_of(rp, struct evdev_client, rcu);
+	unsigned int i;
+	for (i = 0; i < EV_CNT; ++i)
+		bitmap_free(client->evmasks[i]);
+	kvfree(client);
+}
+
 static void evdev_detach_client(struct evdev *evdev,
 				struct evdev_client *client)
 {
 	spin_lock(&evdev->client_lock);
 	list_del_rcu(&client->node);
 	spin_unlock(&evdev->client_lock);
-	synchronize_rcu();
+	call_rcu(&client->rcu, evdev_reclaim_client);
 }
 
 static int evdev_open_device(struct evdev *evdev)
@@ -436,7 +446,6 @@ static int evdev_release(struct inode *inode, struct file *file)
 {
 	struct evdev_client *client = file->private_data;
 	struct evdev *evdev = client->evdev;
-	unsigned int i;
 
 	mutex_lock(&evdev->mutex);
 
@@ -448,11 +457,6 @@ static int evdev_release(struct inode *inode, struct file *file)
 
 	evdev_detach_client(evdev, client);
 
-	for (i = 0; i < EV_CNT; ++i)
-		bitmap_free(client->evmasks[i]);
-
-	kvfree(client);
-
 	evdev_close_device(evdev);
 
 	return 0;
@@ -495,7 +499,6 @@ static int evdev_open(struct inode *inode, struct file *file)
 
  err_free_client:
 	evdev_detach_client(evdev, client);
-	kvfree(client);
 	return error;
 }
 
diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c
index 59445763e55a..568f85414c85 100644
--- a/drivers/md/dm-crypt.c
+++ b/drivers/md/dm-crypt.c
@@ -3271,6 +3271,11 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 			goto bad;
 	}
 
+#ifdef CONFIG_CACHY
+	set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
+	set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
+#endif
+
 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
 	if (ret < 0)
 		goto bad;
diff --git a/drivers/pci/controller/Makefile b/drivers/pci/controller/Makefile
index f2b19e6174af..4fef4b174321 100644
--- a/drivers/pci/controller/Makefile
+++ b/drivers/pci/controller/Makefile
@@ -1,4 +1,10 @@
 # SPDX-License-Identifier: GPL-2.0
+ifdef CONFIG_X86_64
+ifdef CONFIG_SATA_AHCI
+obj-y += intel-nvme-remap.o
+endif
+endif
+
 obj-$(CONFIG_PCIE_CADENCE) += cadence/
 obj-$(CONFIG_PCI_FTPCI100) += pci-ftpci100.o
 obj-$(CONFIG_PCI_IXP4XX) += pci-ixp4xx.o
diff --git a/drivers/pci/controller/intel-nvme-remap.c b/drivers/pci/controller/intel-nvme-remap.c
new file mode 100644
index 000000000000..e105e6f5cc91
--- /dev/null
+++ b/drivers/pci/controller/intel-nvme-remap.c
@@ -0,0 +1,462 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Intel remapped NVMe device support.
+ *
+ * Copyright (c) 2019 Endless Mobile, Inc.
+ * Author: Daniel Drake <drake@endlessm.com>
+ *
+ * Some products ship by default with the SATA controller in "RAID" or
+ * "Intel RST Premium With Intel Optane System Acceleration" mode. Under this
+ * mode, which we refer to as "remapped NVMe" mode, any installed NVMe
+ * devices disappear from the PCI bus, and instead their I/O memory becomes
+ * available within the AHCI device BARs.
+ *
+ * This scheme is understood to be a way of avoiding usage of the standard
+ * Windows NVMe driver under that OS, instead mandating usage of Intel's
+ * driver instead, which has better power management, and presumably offers
+ * some RAID/disk-caching solutions too.
+ *
+ * Here in this driver, we support the remapped NVMe mode by claiming the
+ * AHCI device and creating a fake PCIe root port. On the new bus, the
+ * original AHCI device is exposed with only minor tweaks. Then, fake PCI
+ * devices corresponding to the remapped NVMe devices are created. The usual
+ * ahci and nvme drivers are then expected to bind to these devices and
+ * operate as normal.
+ *
+ * The PCI configuration space for the NVMe devices is completely
+ * unavailable, so we fake a minimal one and hope for the best.
+ *
+ * Interrupts are shared between the AHCI and NVMe devices. For simplicity,
+ * we only support the legacy interrupt here, although MSI support
+ * could potentially be added later.
+ */
+
+#define MODULE_NAME "intel-nvme-remap"
+
+#include <linux/ahci-remap.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+
+#define AHCI_PCI_BAR_STANDARD 5
+
+struct nvme_remap_dev {
+	struct pci_dev		*dev;		/* AHCI device */
+	struct pci_bus		*bus;		/* our fake PCI bus */
+	struct pci_sysdata	sysdata;
+	int			irq_base;	/* our fake interrupts */
+
+	/*
+	 * When we detect an all-ones write to a BAR register, this flag
+	 * is set, so that we return the BAR size on the next read (a
+	 * standard PCI behaviour).
+	 * This includes the assumption that an all-ones BAR write is
+	 * immediately followed by a read of the same register.
+	 */
+	bool			bar_sizing;
+
+	/*
+	 * Resources copied from the AHCI device, to be regarded as
+	 * resources on our fake bus.
+	 */
+	struct resource		ahci_resources[PCI_NUM_RESOURCES];
+
+	/* Resources corresponding to the NVMe devices. */
+	struct resource		remapped_dev_mem[AHCI_MAX_REMAP];
+
+	/* Number of remapped NVMe devices found. */
+	int			num_remapped_devices;
+};
+
+static inline struct nvme_remap_dev *nrdev_from_bus(struct pci_bus *bus)
+{
+	return container_of(bus->sysdata, struct nvme_remap_dev, sysdata);
+}
+
+
+/******** PCI configuration space **********/
+
+/*
+ * Helper macros for tweaking returned contents of PCI configuration space.
+ *
+ * value contains len bytes of data read from reg.
+ * If fixup_reg is included in that range, fix up the contents of that
+ * register to fixed_value.
+ */
+#define NR_FIX8(fixup_reg, fixed_value) do { \
+		if (reg <= fixup_reg && fixup_reg < reg + len) \
+			((u8 *) value)[fixup_reg - reg] = (u8) (fixed_value); \
+	} while (0)
+
+#define NR_FIX16(fixup_reg, fixed_value) do { \
+		NR_FIX8(fixup_reg, fixed_value); \
+		NR_FIX8(fixup_reg + 1, fixed_value >> 8); \
+	} while (0)
+
+#define NR_FIX24(fixup_reg, fixed_value) do { \
+		NR_FIX8(fixup_reg, fixed_value); \
+		NR_FIX8(fixup_reg + 1, fixed_value >> 8); \
+		NR_FIX8(fixup_reg + 2, fixed_value >> 16); \
+	} while (0)
+
+#define NR_FIX32(fixup_reg, fixed_value) do { \
+		NR_FIX16(fixup_reg, (u16) fixed_value); \
+		NR_FIX16(fixup_reg + 2, fixed_value >> 16); \
+	} while (0)
+
+/*
+ * Read PCI config space of the slot 0 (AHCI) device.
+ * We pass through the read request to the underlying device, but
+ * tweak the results in some cases.
+ */
+static int nvme_remap_pci_read_slot0(struct pci_bus *bus, int reg,
+				     int len, u32 *value)
+{
+	struct nvme_remap_dev *nrdev = nrdev_from_bus(bus);
+	struct pci_bus *ahci_dev_bus = nrdev->dev->bus;
+	int ret;
+
+	ret = ahci_dev_bus->ops->read(ahci_dev_bus, nrdev->dev->devfn,
+				      reg, len, value);
+	if (ret)
+		return ret;
+
+	/*
+	 * Adjust the device class, to prevent this driver from attempting to
+	 * additionally probe the device we're simulating here.
+	 */
+	NR_FIX24(PCI_CLASS_PROG, PCI_CLASS_STORAGE_SATA_AHCI);
+
+	/*
+	 * Unset interrupt pin, otherwise ACPI tries to find routing
+	 * info for our virtual IRQ, fails, and complains.
+	 */
+	NR_FIX8(PCI_INTERRUPT_PIN, 0);
+
+	/*
+	 * Truncate the AHCI BAR to not include the region that covers the
+	 * hidden devices. This will cause the ahci driver to successfully
+	 * probe th new device (instead of handing it over to this driver).
+	 */
+	if (nrdev->bar_sizing) {
+		NR_FIX32(PCI_BASE_ADDRESS_5, ~(SZ_16K - 1));
+		nrdev->bar_sizing = false;
+	}
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+/*
+ * Read PCI config space of a remapped device.
+ * Since the original PCI config space is inaccessible, we provide a minimal,
+ * fake config space instead.
+ */
+static int nvme_remap_pci_read_remapped(struct pci_bus *bus, unsigned int port,
+					int reg, int len, u32 *value)
+{
+	struct nvme_remap_dev *nrdev = nrdev_from_bus(bus);
+	struct resource *remapped_mem;
+
+	if (port > nrdev->num_remapped_devices)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+
+	*value = 0;
+	remapped_mem = &nrdev->remapped_dev_mem[port - 1];
+
+	/* Set a Vendor ID, otherwise Linux assumes no device is present */
+	NR_FIX16(PCI_VENDOR_ID, PCI_VENDOR_ID_INTEL);
+
+	/* Always appear on & bus mastering */
+	NR_FIX16(PCI_COMMAND, PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
+
+	/* Set class so that nvme driver probes us */
+	NR_FIX24(PCI_CLASS_PROG, PCI_CLASS_STORAGE_EXPRESS);
+
+	if (nrdev->bar_sizing) {
+		NR_FIX32(PCI_BASE_ADDRESS_0,
+			 ~(resource_size(remapped_mem) - 1));
+		nrdev->bar_sizing = false;
+	} else {
+		resource_size_t mem_start = remapped_mem->start;
+
+		mem_start |= PCI_BASE_ADDRESS_MEM_TYPE_64;
+		NR_FIX32(PCI_BASE_ADDRESS_0, mem_start);
+		mem_start >>= 32;
+		NR_FIX32(PCI_BASE_ADDRESS_1, mem_start);
+	}
+
+	return PCIBIOS_SUCCESSFUL;
+}
+
+/* Read PCI configuration space. */
+static int nvme_remap_pci_read(struct pci_bus *bus, unsigned int devfn,
+			       int reg, int len, u32 *value)
+{
+	if (PCI_SLOT(devfn) == 0)
+		return nvme_remap_pci_read_slot0(bus, reg, len, value);
+	else
+		return nvme_remap_pci_read_remapped(bus, PCI_SLOT(devfn),
+						    reg, len, value);
+}
+
+/*
+ * Write PCI config space of the slot 0 (AHCI) device.
+ * Apart from the special case of BAR sizing, we disable all writes.
+ * Otherwise, the ahci driver could make changes (e.g. unset PCI bus master)
+ * that would affect the operation of the NVMe devices.
+ */
+static int nvme_remap_pci_write_slot0(struct pci_bus *bus, int reg,
+				      int len, u32 value)
+{
+	struct nvme_remap_dev *nrdev = nrdev_from_bus(bus);
+	struct pci_bus *ahci_dev_bus = nrdev->dev->bus;
+
+	if (reg >= PCI_BASE_ADDRESS_0 && reg <= PCI_BASE_ADDRESS_5) {
+		/*
+		 * Writing all-ones to a BAR means that the size of the
+		 * memory region is being checked. Flag this so that we can
+		 * reply with an appropriate size on the next read.
+		 */
+		if (value == ~0)
+			nrdev->bar_sizing = true;
+
+		return ahci_dev_bus->ops->write(ahci_dev_bus,
+						nrdev->dev->devfn,
+						reg, len, value);
+	}
+
+	return PCIBIOS_SET_FAILED;
+}
+
+/*
+ * Write PCI config space of a remapped device.
+ * Since the original PCI config space is inaccessible, we reject all
+ * writes, except for the special case of BAR probing.
+ */
+static int nvme_remap_pci_write_remapped(struct pci_bus *bus,
+					 unsigned int port,
+					 int reg, int len, u32 value)
+{
+	struct nvme_remap_dev *nrdev = nrdev_from_bus(bus);
+
+	if (port > nrdev->num_remapped_devices)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+
+	/*
+	 * Writing all-ones to a BAR means that the size of the memory
+	 * region is being checked. Flag this so that we can reply with
+	 * an appropriate size on the next read.
+	 */
+	if (value == ~0 && reg >= PCI_BASE_ADDRESS_0
+			&& reg <= PCI_BASE_ADDRESS_5) {
+		nrdev->bar_sizing = true;
+		return PCIBIOS_SUCCESSFUL;
+	}
+
+	return PCIBIOS_SET_FAILED;
+}
+
+/* Write PCI configuration space. */
+static int nvme_remap_pci_write(struct pci_bus *bus, unsigned int devfn,
+				int reg, int len, u32 value)
+{
+	if (PCI_SLOT(devfn) == 0)
+		return nvme_remap_pci_write_slot0(bus, reg, len, value);
+	else
+		return nvme_remap_pci_write_remapped(bus, PCI_SLOT(devfn),
+						     reg, len, value);
+}
+
+static struct pci_ops nvme_remap_pci_ops = {
+	.read	= nvme_remap_pci_read,
+	.write	= nvme_remap_pci_write,
+};
+
+
+/******** Initialization & exit **********/
+
+/*
+ * Find a PCI domain ID to use for our fake bus.
+ * Start at 0x10000 to not clash with ACPI _SEG domains (16 bits).
+ */
+static int find_free_domain(void)
+{
+	int domain = 0xffff;
+	struct pci_bus *bus = NULL;
+
+	while ((bus = pci_find_next_bus(bus)) != NULL)
+		domain = max_t(int, domain, pci_domain_nr(bus));
+
+	return domain + 1;
+}
+
+static int find_remapped_devices(struct nvme_remap_dev *nrdev,
+				 struct list_head *resources)
+{
+	void __iomem *mmio;
+	int i, count = 0;
+	u32 cap;
+
+	mmio = pcim_iomap(nrdev->dev, AHCI_PCI_BAR_STANDARD,
+			  pci_resource_len(nrdev->dev,
+					   AHCI_PCI_BAR_STANDARD));
+	if (!mmio)
+		return -ENODEV;
+
+	/* Check if this device might have remapped nvme devices. */
+	if (pci_resource_len(nrdev->dev, AHCI_PCI_BAR_STANDARD) < SZ_512K ||
+	    !(readl(mmio + AHCI_VSCAP) & 1))
+		return -ENODEV;
+
+	cap = readq(mmio + AHCI_REMAP_CAP);
+	for (i = AHCI_MAX_REMAP-1; i >= 0; i--) {
+		struct resource *remapped_mem;
+
+		if ((cap & (1 << i)) == 0)
+			continue;
+		if (readl(mmio + ahci_remap_dcc(i))
+				!= PCI_CLASS_STORAGE_EXPRESS)
+			continue;
+
+		/* We've found a remapped device */
+		remapped_mem = &nrdev->remapped_dev_mem[count++];
+		remapped_mem->start =
+			pci_resource_start(nrdev->dev, AHCI_PCI_BAR_STANDARD)
+			+ ahci_remap_base(i);
+		remapped_mem->end = remapped_mem->start
+			+ AHCI_REMAP_N_SIZE - 1;
+		remapped_mem->flags = IORESOURCE_MEM | IORESOURCE_PCI_FIXED;
+		pci_add_resource(resources, remapped_mem);
+	}
+
+	pcim_iounmap(nrdev->dev, mmio);
+
+	if (count == 0)
+		return -ENODEV;
+
+	nrdev->num_remapped_devices = count;
+	dev_info(&nrdev->dev->dev, "Found %d remapped NVMe devices\n",
+		 nrdev->num_remapped_devices);
+	return 0;
+}
+
+static void nvme_remap_remove_root_bus(void *data)
+{
+	struct pci_bus *bus = data;
+
+	pci_stop_root_bus(bus);
+	pci_remove_root_bus(bus);
+}
+
+static int nvme_remap_probe(struct pci_dev *dev,
+			    const struct pci_device_id *id)
+{
+	struct nvme_remap_dev *nrdev;
+	LIST_HEAD(resources);
+	int i;
+	int ret;
+	struct pci_dev *child;
+
+	nrdev = devm_kzalloc(&dev->dev, sizeof(*nrdev), GFP_KERNEL);
+	nrdev->sysdata.domain = find_free_domain();
+	nrdev->sysdata.nvme_remap_dev = dev;
+	nrdev->dev = dev;
+	pci_set_drvdata(dev, nrdev);
+
+	ret = pcim_enable_device(dev);
+	if (ret < 0)
+		return ret;
+
+	pci_set_master(dev);
+
+	ret = find_remapped_devices(nrdev, &resources);
+	if (ret)
+		return ret;
+
+	/* Add resources from the original AHCI device */
+	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
+		struct resource *res = &dev->resource[i];
+
+		if (res->start) {
+			struct resource *nr_res = &nrdev->ahci_resources[i];
+
+			nr_res->start = res->start;
+			nr_res->end = res->end;
+			nr_res->flags = res->flags;
+			pci_add_resource(&resources, nr_res);
+		}
+	}
+
+	/* Create virtual interrupts */
+	nrdev->irq_base = devm_irq_alloc_descs(&dev->dev, -1, 0,
+					       nrdev->num_remapped_devices + 1,
+					       0);
+	if (nrdev->irq_base < 0)
+		return nrdev->irq_base;
+
+	/* Create and populate PCI bus */
+	nrdev->bus = pci_create_root_bus(&dev->dev, 0, &nvme_remap_pci_ops,
+					 &nrdev->sysdata, &resources);
+	if (!nrdev->bus)
+		return -ENODEV;
+
+	if (devm_add_action_or_reset(&dev->dev, nvme_remap_remove_root_bus,
+				     nrdev->bus))
+		return -ENOMEM;
+
+	/* We don't support sharing MSI interrupts between these devices */
+	nrdev->bus->bus_flags |= PCI_BUS_FLAGS_NO_MSI;
+
+	pci_scan_child_bus(nrdev->bus);
+
+	list_for_each_entry(child, &nrdev->bus->devices, bus_list) {
+		/*
+		 * Prevent PCI core from trying to move memory BARs around.
+		 * The hidden NVMe devices are at fixed locations.
+		 */
+		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
+			struct resource *res = &child->resource[i];
+
+			if (res->flags & IORESOURCE_MEM)
+				res->flags |= IORESOURCE_PCI_FIXED;
+		}
+
+		/* Share the legacy IRQ between all devices */
+		child->irq = dev->irq;
+	}
+
+	pci_assign_unassigned_bus_resources(nrdev->bus);
+	pci_bus_add_devices(nrdev->bus);
+
+	return 0;
+}
+
+static const struct pci_device_id nvme_remap_ids[] = {
+	/*
+	 * Match all Intel RAID controllers.
+	 *
+	 * There's overlap here with the set of devices detected by the ahci
+	 * driver, but ahci will only successfully probe when there
+	 * *aren't* any remapped NVMe devices, and this driver will only
+	 * successfully probe when there *are* remapped NVMe devices that
+	 * need handling.
+	 */
+	{
+		PCI_VDEVICE(INTEL, PCI_ANY_ID),
+		.class = PCI_CLASS_STORAGE_RAID << 8,
+		.class_mask = 0xffffff00,
+	},
+	{0,}
+};
+MODULE_DEVICE_TABLE(pci, nvme_remap_ids);
+
+static struct pci_driver nvme_remap_drv = {
+	.name		= MODULE_NAME,
+	.id_table	= nvme_remap_ids,
+	.probe		= nvme_remap_probe,
+};
+module_pci_driver(nvme_remap_drv);
+
+MODULE_AUTHOR("Daniel Drake <drake@endlessm.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index eff7f5df08e2..cfb099dbeb5f 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -3732,6 +3732,106 @@ static void quirk_no_bus_reset(struct pci_dev *dev)
 	dev->dev_flags |= PCI_DEV_FLAGS_NO_BUS_RESET;
 }
 
+static bool acs_on_downstream;
+static bool acs_on_multifunction;
+
+#define NUM_ACS_IDS 16
+struct acs_on_id {
+	unsigned short vendor;
+	unsigned short device;
+};
+static struct acs_on_id acs_on_ids[NUM_ACS_IDS];
+static u8 max_acs_id;
+
+static __init int pcie_acs_override_setup(char *p)
+{
+	if (!p)
+		return -EINVAL;
+
+	while (*p) {
+		if (!strncmp(p, "downstream", 10))
+			acs_on_downstream = true;
+		if (!strncmp(p, "multifunction", 13))
+			acs_on_multifunction = true;
+		if (!strncmp(p, "id:", 3)) {
+			char opt[5];
+			int ret;
+			long val;
+
+			if (max_acs_id >= NUM_ACS_IDS - 1) {
+				pr_warn("Out of PCIe ACS override slots (%d)\n",
+						NUM_ACS_IDS);
+				goto next;
+			}
+
+			p += 3;
+			snprintf(opt, 5, "%s", p);
+			ret = kstrtol(opt, 16, &val);
+			if (ret) {
+				pr_warn("PCIe ACS ID parse error %d\n", ret);
+				goto next;
+			}
+			acs_on_ids[max_acs_id].vendor = val;
+
+			p += strcspn(p, ":");
+			if (*p != ':') {
+				pr_warn("PCIe ACS invalid ID\n");
+				goto next;
+			}
+
+			p++;
+			snprintf(opt, 5, "%s", p);
+			ret = kstrtol(opt, 16, &val);
+			if (ret) {
+				pr_warn("PCIe ACS ID parse error %d\n", ret);
+				goto next;
+			}
+			acs_on_ids[max_acs_id].device = val;
+			max_acs_id++;
+		}
+next:
+		p += strcspn(p, ",");
+		if (*p == ',')
+			p++;
+	}
+
+	if (acs_on_downstream || acs_on_multifunction || max_acs_id)
+		pr_warn("Warning: PCIe ACS overrides enabled; This may allow non-IOMMU protected peer-to-peer DMA\n");
+
+	return 0;
+}
+early_param("pcie_acs_override", pcie_acs_override_setup);
+
+static int pcie_acs_overrides(struct pci_dev *dev, u16 acs_flags)
+{
+	int i;
+
+	/* Never override ACS for legacy devices or devices with ACS caps */
+	if (!pci_is_pcie(dev) ||
+		pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS))
+			return -ENOTTY;
+
+	for (i = 0; i < max_acs_id; i++)
+		if (acs_on_ids[i].vendor == dev->vendor &&
+			acs_on_ids[i].device == dev->device)
+				return 1;
+
+	switch (pci_pcie_type(dev)) {
+	case PCI_EXP_TYPE_DOWNSTREAM:
+	case PCI_EXP_TYPE_ROOT_PORT:
+		if (acs_on_downstream)
+			return 1;
+		break;
+	case PCI_EXP_TYPE_ENDPOINT:
+	case PCI_EXP_TYPE_UPSTREAM:
+	case PCI_EXP_TYPE_LEG_END:
+	case PCI_EXP_TYPE_RC_END:
+		if (acs_on_multifunction && dev->multifunction)
+			return 1;
+	}
+
+	return -ENOTTY;
+}
 /*
  * Some NVIDIA GPU devices do not work with bus reset, SBR needs to be
  * prevented for those affected devices.
@@ -5143,6 +5243,7 @@ static const struct pci_dev_acs_enabled {
 	{ PCI_VENDOR_ID_ZHAOXIN, PCI_ANY_ID, pci_quirk_zhaoxin_pcie_ports_acs },
 	/* Wangxun nics */
 	{ PCI_VENDOR_ID_WANGXUN, PCI_ANY_ID, pci_quirk_wangxun_nic_acs },
+	{ PCI_ANY_ID, PCI_ANY_ID, pcie_acs_overrides },
 	{ 0 }
 };
 
diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig
index bdd302274b9a..0afc3e4c1880 100644
--- a/drivers/platform/x86/Kconfig
+++ b/drivers/platform/x86/Kconfig
@@ -1127,6 +1127,20 @@ config SEL3350_PLATFORM
 	  To compile this driver as a module, choose M here: the module
 	  will be called sel3350-platform.
 
+config STEAMDECK
+	tristate "Valve Steam Deck platform driver"
+	depends on X86_64
+	help
+	  Driver exposing various bits and pieces of functionality
+	  provided by Steam Deck specific VLV0100 device presented by
+	  EC firmware. This includes but not limited to:
+	     - CPU/device's fan control
+	     - Read-only access to DDIC registers
+	     - Battery tempreature measurements
+	     - Various display related control knobs
+	     - USB Type-C connector event notification
+	  Say N unless you are running on a Steam Deck.
+
 endif # X86_PLATFORM_DEVICES
 
 config P2SB
diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile
index 1de432e8861e..59bfbd2649eb 100644
--- a/drivers/platform/x86/Makefile
+++ b/drivers/platform/x86/Makefile
@@ -144,3 +144,6 @@ obj-$(CONFIG_WINMATE_FM07_KEYS)		+= winmate-fm07-keys.o
 
 # SEL
 obj-$(CONFIG_SEL3350_PLATFORM)		+= sel3350-platform.o
+
+# Steam Deck
+obj-$(CONFIG_STEAMDECK)			+= steamdeck.o
diff --git a/drivers/platform/x86/steamdeck.c b/drivers/platform/x86/steamdeck.c
new file mode 100644
index 000000000000..77a6677ec19e
--- /dev/null
+++ b/drivers/platform/x86/steamdeck.c
@@ -0,0 +1,523 @@
+// SPDX-License-Identifier: GPL-2.0+
+
+/*
+ * Steam Deck ACPI platform driver
+ *
+ * Copyright (C) 2021-2022 Valve Corporation
+ *
+ */
+#include <linux/acpi.h>
+#include <linux/hwmon.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/extcon-provider.h>
+
+#define ACPI_STEAMDECK_NOTIFY_STATUS	0x80
+
+/* 0 - port connected, 1 -port disconnected */
+#define ACPI_STEAMDECK_PORT_CONNECT	BIT(0)
+/* 0 - Upstream Facing Port, 1 - Downdstream Facing Port */
+#define ACPI_STEAMDECK_CUR_DATA_ROLE	BIT(3)
+/*
+ * Debouncing delay to allow negotiation process to settle. 2s value
+ * was arrived at via trial and error.
+ */
+#define STEAMDECK_ROLE_SWITCH_DELAY	(msecs_to_jiffies(2000))
+
+struct steamdeck {
+	struct acpi_device *adev;
+	struct device *hwmon;
+	void *regmap;
+	long fan_target;
+	struct delayed_work role_work;
+	struct extcon_dev *edev;
+	struct device *dev;
+};
+
+static ssize_t
+steamdeck_simple_store(struct device *dev, const char *buf, size_t count,
+		       const char *method,
+		       unsigned long upper_limit)
+{
+	struct steamdeck *fan = dev_get_drvdata(dev);
+	unsigned long value;
+
+	if (kstrtoul(buf, 10, &value) || value >= upper_limit)
+		return -EINVAL;
+
+	if (ACPI_FAILURE(acpi_execute_simple_method(fan->adev->handle,
+						    (char *)method, value)))
+		return -EIO;
+
+	return count;
+}
+
+#define STEAMDECK_ATTR_WO(_name, _method, _upper_limit)			\
+	static ssize_t _name##_store(struct device *dev,		\
+				     struct device_attribute *attr,	\
+				     const char *buf, size_t count)	\
+	{								\
+		return steamdeck_simple_store(dev, buf, count,		\
+					    _method,			\
+					    _upper_limit);		\
+	}								\
+	static DEVICE_ATTR_WO(_name)
+
+STEAMDECK_ATTR_WO(target_cpu_temp, "STCT", U8_MAX / 2);
+STEAMDECK_ATTR_WO(gain, "SGAN", U16_MAX);
+STEAMDECK_ATTR_WO(ramp_rate, "SFRR", U8_MAX);
+STEAMDECK_ATTR_WO(hysteresis, "SHTS",  U16_MAX);
+STEAMDECK_ATTR_WO(maximum_battery_charge_rate, "CHGR", U16_MAX);
+STEAMDECK_ATTR_WO(recalculate, "SCHG", U16_MAX);
+
+STEAMDECK_ATTR_WO(led_brightness, "CHBV", U8_MAX);
+STEAMDECK_ATTR_WO(content_adaptive_brightness, "CABC", U8_MAX);
+STEAMDECK_ATTR_WO(gamma_set, "GAMA", U8_MAX);
+STEAMDECK_ATTR_WO(display_brightness, "WDBV", U8_MAX);
+STEAMDECK_ATTR_WO(ctrl_display, "WCDV", U8_MAX);
+STEAMDECK_ATTR_WO(cabc_minimum_brightness, "WCMB", U8_MAX);
+STEAMDECK_ATTR_WO(memory_data_access_control, "MDAC", U8_MAX);
+
+#define STEAMDECK_ATTR_WO_NOARG(_name, _method)				\
+	static ssize_t _name##_store(struct device *dev,		\
+				     struct device_attribute *attr,	\
+				     const char *buf, size_t count)	\
+	{								\
+		struct steamdeck *fan = dev_get_drvdata(dev);		\
+									\
+		if (ACPI_FAILURE(acpi_evaluate_object(fan->adev->handle, \
+						      _method, NULL, NULL))) \
+			return -EIO;					\
+									\
+		return count;						\
+	}								\
+	static DEVICE_ATTR_WO(_name)
+
+STEAMDECK_ATTR_WO_NOARG(power_cycle_display, "DPCY");
+STEAMDECK_ATTR_WO_NOARG(display_normal_mode_on, "NORO");
+STEAMDECK_ATTR_WO_NOARG(display_inversion_off, "INOF");
+STEAMDECK_ATTR_WO_NOARG(display_inversion_on, "INON");
+STEAMDECK_ATTR_WO_NOARG(idle_mode_on, "WRNE");
+
+#define STEAMDECK_ATTR_RO(_name, _method)				\
+	static ssize_t _name##_show(struct device *dev,			\
+				    struct device_attribute *attr,	\
+				    char *buf)				\
+	{								\
+		struct steamdeck *jup = dev_get_drvdata(dev);		\
+		unsigned long long val;					\
+									\
+		if (ACPI_FAILURE(acpi_evaluate_integer(			\
+					 jup->adev->handle,		\
+					 _method, NULL, &val)))		\
+			return -EIO;					\
+									\
+		return sprintf(buf, "%llu\n", val);			\
+	}								\
+	static DEVICE_ATTR_RO(_name)
+
+STEAMDECK_ATTR_RO(firmware_version, "PDFW");
+STEAMDECK_ATTR_RO(board_id, "BOID");
+STEAMDECK_ATTR_RO(pdcs, "PDCS");
+
+static umode_t
+steamdeck_is_visible(struct kobject *kobj, struct attribute *attr, int index)
+{
+	return attr->mode;
+}
+
+static struct attribute *steamdeck_attributes[] = {
+	&dev_attr_target_cpu_temp.attr,
+	&dev_attr_gain.attr,
+	&dev_attr_ramp_rate.attr,
+	&dev_attr_hysteresis.attr,
+	&dev_attr_maximum_battery_charge_rate.attr,
+	&dev_attr_recalculate.attr,
+	&dev_attr_power_cycle_display.attr,
+
+	&dev_attr_led_brightness.attr,
+	&dev_attr_content_adaptive_brightness.attr,
+	&dev_attr_gamma_set.attr,
+	&dev_attr_display_brightness.attr,
+	&dev_attr_ctrl_display.attr,
+	&dev_attr_cabc_minimum_brightness.attr,
+	&dev_attr_memory_data_access_control.attr,
+
+	&dev_attr_display_normal_mode_on.attr,
+	&dev_attr_display_inversion_off.attr,
+	&dev_attr_display_inversion_on.attr,
+	&dev_attr_idle_mode_on.attr,
+
+	&dev_attr_firmware_version.attr,
+	&dev_attr_board_id.attr,
+	&dev_attr_pdcs.attr,
+
+	NULL
+};
+
+static const struct attribute_group steamdeck_group = {
+	.attrs = steamdeck_attributes,
+	.is_visible = steamdeck_is_visible,
+};
+
+static const struct attribute_group *steamdeck_groups[] = {
+	&steamdeck_group,
+	NULL
+};
+
+static int steamdeck_read_fan_speed(struct steamdeck *jup, long *speed)
+{
+	unsigned long long val;
+
+	if (ACPI_FAILURE(acpi_evaluate_integer(jup->adev->handle,
+					       "FANR", NULL, &val)))
+		return -EIO;
+
+	*speed = val;
+	return 0;
+}
+
+static int
+steamdeck_hwmon_read(struct device *dev, enum hwmon_sensor_types type,
+		     u32 attr, int channel, long *out)
+{
+	struct steamdeck *sd = dev_get_drvdata(dev);
+	unsigned long long val;
+
+	switch (type) {
+	case hwmon_temp:
+		if (attr != hwmon_temp_input)
+			return -EOPNOTSUPP;
+
+		if (ACPI_FAILURE(acpi_evaluate_integer(sd->adev->handle,
+						       "BATT", NULL, &val)))
+			return -EIO;
+		/*
+		 * Assuming BATT returns deg C we need to mutiply it
+		 * by 1000 to convert to mC
+		 */
+		*out = val * 1000;
+		break;
+	case hwmon_fan:
+		switch (attr) {
+		case hwmon_fan_input:
+			return steamdeck_read_fan_speed(sd, out);
+		case hwmon_fan_target:
+			*out = sd->fan_target;
+			break;
+		case hwmon_fan_fault:
+			if (ACPI_FAILURE(acpi_evaluate_integer(
+						 sd->adev->handle,
+						 "FANC", NULL, &val)))
+				return -EIO;
+			/*
+			 * FANC (Fan check):
+			 * 0: Abnormal
+			 * 1: Normal
+			 */
+			*out = !val;
+			break;
+		default:
+			return -EOPNOTSUPP;
+		}
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
+static int
+steamdeck_hwmon_read_string(struct device *dev, enum hwmon_sensor_types type,
+			    u32 attr, int channel, const char **str)
+{
+	switch (type) {
+	case hwmon_temp:
+		*str = "Battery Temp";
+		break;
+	case hwmon_fan:
+		*str = "System Fan";
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
+static int
+steamdeck_hwmon_write(struct device *dev, enum hwmon_sensor_types type,
+		      u32 attr, int channel, long val)
+{
+	struct steamdeck *sd = dev_get_drvdata(dev);
+
+	if (type != hwmon_fan ||
+	    attr != hwmon_fan_target)
+		return -EOPNOTSUPP;
+
+	if (val > U16_MAX)
+		return -EINVAL;
+
+	sd->fan_target = val;
+
+	if (ACPI_FAILURE(acpi_execute_simple_method(sd->adev->handle,
+						    "FANS", val)))
+		return -EIO;
+
+	return 0;
+}
+
+static umode_t
+steamdeck_hwmon_is_visible(const void *data, enum hwmon_sensor_types type,
+			   u32 attr, int channel)
+{
+	if (type == hwmon_fan &&
+	    attr == hwmon_fan_target)
+		return 0644;
+
+	return 0444;
+}
+
+static const struct hwmon_channel_info *steamdeck_info[] = {
+	HWMON_CHANNEL_INFO(temp,
+			   HWMON_T_INPUT | HWMON_T_LABEL),
+	HWMON_CHANNEL_INFO(fan,
+			   HWMON_F_INPUT | HWMON_F_LABEL |
+			   HWMON_F_TARGET | HWMON_F_FAULT),
+	NULL
+};
+
+static const struct hwmon_ops steamdeck_hwmon_ops = {
+	.is_visible = steamdeck_hwmon_is_visible,
+	.read = steamdeck_hwmon_read,
+	.read_string = steamdeck_hwmon_read_string,
+	.write = steamdeck_hwmon_write,
+};
+
+static const struct hwmon_chip_info steamdeck_chip_info = {
+	.ops = &steamdeck_hwmon_ops,
+	.info = steamdeck_info,
+};
+
+#define STEAMDECK_STA_OK			\
+	(ACPI_STA_DEVICE_ENABLED |		\
+	 ACPI_STA_DEVICE_PRESENT |		\
+	 ACPI_STA_DEVICE_FUNCTIONING)
+
+static int
+steamdeck_ddic_reg_read(void *context, unsigned int reg, unsigned int *val)
+{
+	union acpi_object obj = { .type = ACPI_TYPE_INTEGER };
+	struct acpi_object_list arg_list = { .count = 1, .pointer = &obj, };
+	struct steamdeck *sd = context;
+	unsigned long long _val;
+
+	obj.integer.value = reg;
+
+	if (ACPI_FAILURE(acpi_evaluate_integer(sd->adev->handle,
+					       "RDDI", &arg_list, &_val)))
+		return -EIO;
+
+	*val = _val;
+	return 0;
+}
+
+static int steamdeck_read_pdcs(struct steamdeck *sd, unsigned long long *pdcs)
+{
+	acpi_status status;
+
+	status = acpi_evaluate_integer(sd->adev->handle, "PDCS", NULL, pdcs);
+	if (ACPI_FAILURE(status)) {
+		dev_err(sd->dev, "PDCS evaluation failed: %s\n",
+			acpi_format_exception(status));
+		return -EIO;
+	}
+
+	return 0;
+}
+
+static void steamdeck_usb_role_work(struct work_struct *work)
+{
+	struct steamdeck *sd =
+		container_of(work, struct steamdeck, role_work.work);
+	unsigned long long pdcs;
+	bool usb_host;
+
+	if (steamdeck_read_pdcs(sd, &pdcs))
+		return;
+
+	/*
+	 * We only care about these two
+	 */
+	pdcs &= ACPI_STEAMDECK_PORT_CONNECT | ACPI_STEAMDECK_CUR_DATA_ROLE;
+
+	/*
+	 * For "connect" events our role is determined by a bit in
+	 * PDCS, for "disconnect" we switch to being a gadget
+	 * unconditionally. The thinking for the latter is we don't
+	 * want to start acting as a USB host until we get
+	 * confirmation from the firmware that we are a USB host
+	 */
+	usb_host = (pdcs & ACPI_STEAMDECK_PORT_CONNECT) ?
+		pdcs & ACPI_STEAMDECK_CUR_DATA_ROLE : false;
+
+	WARN_ON(extcon_set_state_sync(sd->edev, EXTCON_USB_HOST,
+				      usb_host));
+	dev_dbg(sd->dev, "USB role is %s\n", usb_host ? "host" : "device");
+}
+
+static void steamdeck_notify(acpi_handle handle, u32 event, void *context)
+{
+	struct device *dev = context;
+	struct steamdeck *sd = dev_get_drvdata(dev);
+	unsigned long long pdcs;
+	unsigned long delay;
+
+	switch (event) {
+	case ACPI_STEAMDECK_NOTIFY_STATUS:
+		if (steamdeck_read_pdcs(sd, &pdcs))
+			return;
+		/*
+		 * We process "disconnect" events immediately and
+		 * "connect" events with a delay to give the HW time
+		 * to settle. For example attaching USB hub (at least
+		 * for HW used for testing) will generate intermediary
+		 * event with "host" bit not set, followed by the one
+		 * that does have it set.
+		 */
+		delay = (pdcs & ACPI_STEAMDECK_PORT_CONNECT) ?
+			STEAMDECK_ROLE_SWITCH_DELAY : 0;
+
+		queue_delayed_work(system_long_wq, &sd->role_work, delay);
+		break;
+	default:
+		dev_err(dev, "Unsupported event [0x%x]\n", event);
+	}
+}
+
+static void steamdeck_remove_notify_handler(void *data)
+{
+	struct steamdeck *sd = data;
+
+	acpi_remove_notify_handler(sd->adev->handle, ACPI_DEVICE_NOTIFY,
+				   steamdeck_notify);
+	cancel_delayed_work_sync(&sd->role_work);
+}
+
+static const unsigned int steamdeck_extcon_cable[] = {
+	EXTCON_USB,
+	EXTCON_USB_HOST,
+	EXTCON_CHG_USB_SDP,
+	EXTCON_CHG_USB_CDP,
+	EXTCON_CHG_USB_DCP,
+	EXTCON_CHG_USB_ACA,
+	EXTCON_NONE,
+};
+
+static int steamdeck_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct steamdeck *sd;
+	acpi_status status;
+	unsigned long long sta;
+	int ret;
+
+	static const struct regmap_config regmap_config = {
+		.reg_bits = 8,
+		.val_bits = 8,
+		.max_register = 255,
+		.cache_type = REGCACHE_NONE,
+		.reg_read = steamdeck_ddic_reg_read,
+	};
+
+	sd = devm_kzalloc(dev, sizeof(*sd), GFP_KERNEL);
+	if (!sd)
+		return -ENOMEM;
+	sd->adev = ACPI_COMPANION(&pdev->dev);
+	sd->dev = dev;
+	platform_set_drvdata(pdev, sd);
+	INIT_DELAYED_WORK(&sd->role_work, steamdeck_usb_role_work);
+
+	status = acpi_evaluate_integer(sd->adev->handle, "_STA",
+				       NULL, &sta);
+	if (ACPI_FAILURE(status)) {
+		dev_err(dev, "Status check failed (0x%x)\n", status);
+		return -EINVAL;
+	}
+
+	if ((sta & STEAMDECK_STA_OK) != STEAMDECK_STA_OK) {
+		dev_err(dev, "Device is not ready\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * Our ACPI interface doesn't expose a method to read current
+	 * fan target, so we use current fan speed as an
+	 * approximation.
+	 */
+	if (steamdeck_read_fan_speed(sd, &sd->fan_target))
+		dev_warn(dev, "Failed to read fan speed");
+
+	sd->hwmon = devm_hwmon_device_register_with_info(dev,
+							 "steamdeck",
+							 sd,
+							 &steamdeck_chip_info,
+							 steamdeck_groups);
+	if (IS_ERR(sd->hwmon)) {
+		dev_err(dev, "Failed to register HWMON device");
+		return PTR_ERR(sd->hwmon);
+	}
+
+	sd->regmap = devm_regmap_init(dev, NULL, sd, &regmap_config);
+	if (IS_ERR(sd->regmap))
+		dev_err(dev, "Failed to register REGMAP");
+
+	sd->edev = devm_extcon_dev_allocate(dev, steamdeck_extcon_cable);
+	if (IS_ERR(sd->edev))
+		return -ENOMEM;
+
+	ret = devm_extcon_dev_register(dev, sd->edev);
+	if (ret < 0) {
+		dev_err(dev, "Failed to register extcon device: %d\n", ret);
+		return ret;
+	}
+
+	/*
+	 * Set initial role value
+	 */
+	queue_delayed_work(system_long_wq, &sd->role_work, 0);
+	flush_delayed_work(&sd->role_work);
+
+	status = acpi_install_notify_handler(sd->adev->handle,
+					     ACPI_DEVICE_NOTIFY,
+					     steamdeck_notify,
+					     dev);
+	if (ACPI_FAILURE(status)) {
+		dev_err(dev, "Error installing ACPI notify handler\n");
+		return -EIO;
+	}
+
+	ret = devm_add_action_or_reset(dev, steamdeck_remove_notify_handler,
+				       sd);
+	return ret;
+}
+
+static const struct acpi_device_id steamdeck_device_ids[] = {
+	{ "VLV0100", 0 },
+	{ "", 0 },
+};
+MODULE_DEVICE_TABLE(acpi, steamdeck_device_ids);
+
+static struct platform_driver steamdeck_driver = {
+	.probe = steamdeck_probe,
+	.driver = {
+		.name = "steamdeck",
+		.acpi_match_table = steamdeck_device_ids,
+	},
+};
+module_platform_driver(steamdeck_driver);
+
+MODULE_AUTHOR("Andrey Smirnov <andrew.smirnov@gmail.com>");
+MODULE_DESCRIPTION("Steam Deck ACPI platform driver");
+MODULE_LICENSE("GPL");
diff --git a/include/linux/mm.h b/include/linux/mm.h
index f5a97dec5169..397ad6f1ac39 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -191,10 +191,18 @@ static inline void __mm_zero_struct_page(struct page *page)
  * that.
  */
 #define MAPCOUNT_ELF_CORE_MARGIN	(5)
-#define DEFAULT_MAX_MAP_COUNT	(USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
+#define DEFAULT_MAX_MAP_COUNT	(INT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
 
 extern int sysctl_max_map_count;
 
+extern bool sysctl_workingset_protection;
+extern u8 sysctl_anon_min_ratio;
+extern u8 sysctl_clean_low_ratio;
+extern u8 sysctl_clean_min_ratio;
+int vm_workingset_protection_update_handler(
+	struct ctl_table *table, int write,
+	void __user *buffer, size_t *lenp, loff_t *ppos);
+
 extern unsigned long sysctl_user_reserve_kbytes;
 extern unsigned long sysctl_admin_reserve_kbytes;
 
diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 2df35e65557d..a52bd9f4b632 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -1268,7 +1268,7 @@ struct readahead_control {
 		._index = i,						\
 	}
 
-#define VM_READAHEAD_PAGES	(SZ_128K / PAGE_SIZE)
+#define VM_READAHEAD_PAGES	(SZ_8M / PAGE_SIZE)
 
 void page_cache_ra_unbounded(struct readahead_control *,
 		unsigned long nr_to_read, unsigned long lookahead_count);
diff --git a/include/linux/user_namespace.h b/include/linux/user_namespace.h
index 6030a8235617..60b7fe5fa74a 100644
--- a/include/linux/user_namespace.h
+++ b/include/linux/user_namespace.h
@@ -156,6 +156,8 @@ static inline void set_userns_rlimit_max(struct user_namespace *ns,
 
 #ifdef CONFIG_USER_NS
 
+extern int unprivileged_userns_clone;
+
 static inline struct user_namespace *get_user_ns(struct user_namespace *ns)
 {
 	if (ns)
@@ -189,6 +191,8 @@ extern bool current_in_userns(const struct user_namespace *target_ns);
 struct ns_common *ns_get_owner(struct ns_common *ns);
 #else
 
+#define unprivileged_userns_clone 0
+
 static inline struct user_namespace *get_user_ns(struct user_namespace *ns)
 {
 	return &init_user_ns;
diff --git a/init/Kconfig b/init/Kconfig
index bee58f7468c3..9ea39297f149 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -132,6 +132,10 @@ config THREAD_INFO_IN_TASK
 
 menu "General setup"
 
+config CACHY
+    bool "Some kernel tweaks by CachyOS"
+    default y
+
 config BROKEN
 	bool
 
@@ -1247,6 +1251,22 @@ config USER_NS
 
 	  If unsure, say N.
 
+config USER_NS_UNPRIVILEGED
+	bool "Allow unprivileged users to create namespaces"
+	default y
+	depends on USER_NS
+	help
+	  When disabled, unprivileged users will not be able to create
+	  new namespaces. Allowing users to create their own namespaces
+	  has been part of several recent local privilege escalation
+	  exploits, so if you need user namespaces but are
+	  paranoid^Wsecurity-conscious you want to disable this.
+
+	  This setting can be overridden at runtime via the
+	  kernel.unprivileged_userns_clone sysctl.
+
+	  If unsure, say Y.
+
 config PID_NS
 	bool "PID Namespaces"
 	default y
@@ -1389,6 +1409,12 @@ config CC_OPTIMIZE_FOR_PERFORMANCE
 	  with the "-O2" compiler flag for best performance and most
 	  helpful compile-time warnings.
 
+config CC_OPTIMIZE_FOR_PERFORMANCE_O3
+	bool "Optimize more for performance (-O3)"
+	help
+	  Choosing this option will pass "-O3" to your compiler to optimize
+	  the kernel yet more for performance.
+
 config CC_OPTIMIZE_FOR_SIZE
 	bool "Optimize for size (-Os)"
 	help
diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz
index 38ef6d06888e..0f78364efd4f 100644
--- a/kernel/Kconfig.hz
+++ b/kernel/Kconfig.hz
@@ -40,6 +40,27 @@ choice
 	 on SMP and NUMA systems and exactly dividing by both PAL and
 	 NTSC frame rates for video and multimedia work.
 
+	config HZ_500
+		bool "500 HZ"
+	help
+	 500 Hz is a balanced timer frequency. Provides fast interactivity
+	 on desktops with good smoothness without increasing CPU power
+	 consumption and sacrificing the battery life on laptops.
+
+	config HZ_600
+		bool "600 HZ"
+	help
+	 600 Hz is a balanced timer frequency. Provides fast interactivity
+	 on desktops with good smoothness without increasing CPU power
+	 consumption and sacrificing the battery life on laptops.
+
+	config HZ_750
+		bool "750 HZ"
+	help
+	 750 Hz is a balanced timer frequency. Provides fast interactivity
+	 on desktops with good smoothness without increasing CPU power
+	 consumption and sacrificing the battery life on laptops.
+
 	config HZ_1000
 		bool "1000 HZ"
 	help
@@ -53,6 +74,9 @@ config HZ
 	default 100 if HZ_100
 	default 250 if HZ_250
 	default 300 if HZ_300
+	default 500 if HZ_500
+	default 600 if HZ_600
+	default 750 if HZ_750
 	default 1000 if HZ_1000
 
 config SCHED_HRTICK
diff --git a/kernel/fork.c b/kernel/fork.c
index 0d944e92a43f..5449c990a91a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -102,6 +102,10 @@
 #include <linux/iommu.h>
 #include <linux/rseq.h>
 
+#ifdef CONFIG_USER_NS
+#include <linux/user_namespace.h>
+#endif
+
 #include <asm/pgalloc.h>
 #include <linux/uaccess.h>
 #include <asm/mmu_context.h>
@@ -2260,6 +2264,10 @@ __latent_entropy struct task_struct *copy_process(
 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
 		return ERR_PTR(-EINVAL);
 
+	if ((clone_flags & CLONE_NEWUSER) && !unprivileged_userns_clone)
+		if (!capable(CAP_SYS_ADMIN))
+			return ERR_PTR(-EPERM);
+
 	/*
 	 * Thread groups must share signals as well, and detached threads
 	 * can only be started up within the thread group.
@@ -3406,6 +3414,12 @@ int ksys_unshare(unsigned long unshare_flags)
 	if (unshare_flags & CLONE_NEWNS)
 		unshare_flags |= CLONE_FS;
 
+	if ((unshare_flags & CLONE_NEWUSER) && !unprivileged_userns_clone) {
+		err = -EPERM;
+		if (!capable(CAP_SYS_ADMIN))
+			goto bad_unshare_out;
+	}
+
 	err = check_unshare_flags(unshare_flags);
 	if (err)
 		goto bad_unshare_out;
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e2b4e0396af8..97983b041e9d 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -73,10 +73,19 @@ unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG;
  *
  * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
  */
+#ifdef CONFIG_CACHY
+unsigned int sysctl_sched_base_slice			= 350000ULL;
+static unsigned int normalized_sysctl_sched_base_slice	= 350000ULL;
+#else
 unsigned int sysctl_sched_base_slice			= 750000ULL;
 static unsigned int normalized_sysctl_sched_base_slice	= 750000ULL;
+#endif
 
+#ifdef CONFIG_CACHY
+const_debug unsigned int sysctl_sched_migration_cost	= 300000UL;
+#else
 const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
+#endif
 
 int sched_thermal_decay_shift;
 static int __init setup_sched_thermal_decay_shift(char *str)
@@ -127,8 +136,12 @@ int __weak arch_asym_cpu_priority(int cpu)
  *
  * (default: 5 msec, units: microseconds)
  */
+#ifdef CONFIG_CACHY
+static unsigned int sysctl_sched_cfs_bandwidth_slice		= 3000UL;
+#else
 static unsigned int sysctl_sched_cfs_bandwidth_slice		= 5000UL;
 #endif
+#endif
 
 #ifdef CONFIG_NUMA_BALANCING
 /* Restrict the NUMA promotion throughput (MB/s) for each target node. */
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 001fe047bd5d..ed5c758c7368 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2542,7 +2542,7 @@ extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
 
 extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
 
-#ifdef CONFIG_PREEMPT_RT
+#if defined(CONFIG_PREEMPT_RT) || defined(CONFIG_CACHY)
 #define SCHED_NR_MIGRATE_BREAK 8
 #else
 #define SCHED_NR_MIGRATE_BREAK 32
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 157f7ce2942d..aa55ebba2ec3 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -95,6 +95,9 @@ EXPORT_SYMBOL_GPL(sysctl_long_vals);
 #ifdef CONFIG_PERF_EVENTS
 static const int six_hundred_forty_kb = 640 * 1024;
 #endif
+#ifdef CONFIG_USER_NS
+#include <linux/user_namespace.h>
+#endif
 
 
 static const int ngroups_max = NGROUPS_MAX;
@@ -1623,6 +1626,15 @@ static struct ctl_table kern_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
+#ifdef CONFIG_USER_NS
+	{
+		.procname	= "unprivileged_userns_clone",
+		.data		= &unprivileged_userns_clone,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+#endif
 #ifdef CONFIG_PROC_SYSCTL
 	{
 		.procname	= "tainted",
@@ -2204,6 +2216,40 @@ static struct ctl_table vm_table[] = {
 		.extra1		= SYSCTL_ZERO,
 	},
 #endif
+	{
+		.procname	= "workingset_protection",
+		.data		= &sysctl_workingset_protection,
+		.maxlen		= sizeof(bool),
+		.mode		= 0644,
+		.proc_handler	= &proc_dobool,
+	},
+	{
+		.procname	= "anon_min_ratio",
+		.data		= &sysctl_anon_min_ratio,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= &vm_workingset_protection_update_handler,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE_HUNDRED,
+	},
+	{
+		.procname	= "clean_low_ratio",
+		.data		= &sysctl_clean_low_ratio,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= &vm_workingset_protection_update_handler,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE_HUNDRED,
+	},
+	{
+		.procname	= "clean_min_ratio",
+		.data		= &sysctl_clean_min_ratio,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= &vm_workingset_protection_update_handler,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE_HUNDRED,
+	},
 	{
 		.procname	= "user_reserve_kbytes",
 		.data		= &sysctl_user_reserve_kbytes,
diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index ce4d99df5f0e..8272e2e359f1 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -22,6 +22,13 @@
 #include <linux/bsearch.h>
 #include <linux/sort.h>
 
+/* sysctl */
+#ifdef CONFIG_USER_NS_UNPRIVILEGED
+int unprivileged_userns_clone = 1;
+#else
+int unprivileged_userns_clone;
+#endif
+
 static struct kmem_cache *user_ns_cachep __ro_after_init;
 static DEFINE_MUTEX(userns_state_mutex);
 
diff --git a/mm/Kconfig b/mm/Kconfig
index ffc3a2ba3a8c..002f48b655de 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -486,6 +486,69 @@ config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
 config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
 	bool
 
+config ANON_MIN_RATIO
+	int "Default value for vm.anon_min_ratio"
+	depends on SYSCTL
+	range 0 100
+	default 15
+	help
+	  This option sets the default value for vm.anon_min_ratio sysctl knob.
+
+	  The vm.anon_min_ratio sysctl knob provides *hard* protection of
+	  anonymous pages. The anonymous pages on the current node won't be
+	  reclaimed under any conditions when their amount is below
+	  vm.anon_min_ratio. This knob may be used to prevent excessive swap
+	  thrashing when anonymous memory is low (for example, when memory is
+	  going to be overfilled by compressed data of zram module).
+
+	  Setting this value too high (close to MemTotal) can result in
+	  inability to swap and can lead to early OOM under memory pressure.
+
+config CLEAN_LOW_RATIO
+	int "Default value for vm.clean_low_ratio"
+	depends on SYSCTL
+	range 0 100
+	default 0
+	help
+	  This option sets the default value for vm.clean_low_ratio sysctl knob.
+
+	  The vm.clean_low_ratio sysctl knob provides *best-effort*
+	  protection of clean file pages. The file pages on the current node
+	  won't be reclaimed under memory pressure when the amount of clean file
+	  pages is below vm.clean_low_ratio *unless* we threaten to OOM.
+	  Protection of clean file pages using this knob may be used when
+	  swapping is still possible to
+	    - prevent disk I/O thrashing under memory pressure;
+	    - improve performance in disk cache-bound tasks under memory
+	      pressure.
+
+	  Setting it to a high value may result in a early eviction of anonymous
+	  pages into the swap space by attempting to hold the protected amount
+	  of clean file pages in memory.
+
+config CLEAN_MIN_RATIO
+	int "Default value for vm.clean_min_ratio"
+	depends on SYSCTL
+	range 0 100
+	default 15
+	help
+	  This option sets the default value for vm.clean_min_ratio sysctl knob.
+
+	  The vm.clean_min_ratio sysctl knob provides *hard* protection of
+	  clean file pages. The file pages on the current node won't be
+	  reclaimed under memory pressure when the amount of clean file pages is
+	  below vm.clean_min_ratio. Hard protection of clean file pages using
+	  this knob may be used to
+	    - prevent disk I/O thrashing under memory pressure even with no free
+	      swap space;
+	    - improve performance in disk cache-bound tasks under memory
+	      pressure;
+	    - avoid high latency and prevent livelock in near-OOM conditions.
+
+	  Setting it to a high value may result in a early out-of-memory condition
+	  due to the inability to reclaim the protected amount of clean file pages
+	  when other types of pages cannot be reclaimed.
+
 config HAVE_MEMBLOCK_PHYS_MAP
 	bool
 
@@ -630,7 +693,7 @@ config COMPACTION
 config COMPACT_UNEVICTABLE_DEFAULT
 	int
 	depends on COMPACTION
-	default 0 if PREEMPT_RT
+	default 0 if PREEMPT_RT || CACHY
 	default 1
 
 #
diff --git a/mm/compaction.c b/mm/compaction.c
index b961db601df4..91d627e8a93d 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -1830,7 +1830,11 @@ static int sysctl_compact_unevictable_allowed __read_mostly = CONFIG_COMPACT_UNE
  * aggressively the kernel should compact memory in the
  * background. It takes values in the range [0, 100].
  */
+#ifdef CONFIG_CACHY
+static unsigned int __read_mostly sysctl_compaction_proactiveness;
+#else
 static unsigned int __read_mostly sysctl_compaction_proactiveness = 20;
+#endif
 static int sysctl_extfrag_threshold = 500;
 static int __read_mostly sysctl_compact_memory;
 
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 94c958f7ebb5..2f9974f305ee 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -62,7 +62,11 @@ unsigned long transparent_hugepage_flags __read_mostly =
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE
 	(1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)|
 #endif
+#ifdef CONFIG_CACHY
+	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG)|
+#else
 	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG)|
+#endif
 	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)|
 	(1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
 
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 2c19f5515e36..419ba5ac7c52 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2749,6 +2749,7 @@ static void __init mem_init_print_info(void)
 		, K(totalhigh_pages())
 #endif
 		);
+	printk(KERN_INFO "le9 Unofficial (le9uo) working set protection 1.4 by Masahito Suzuki (forked from hakavlad's original le9 patch)");
 }
 
 /*
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 3f255534986a..01b3e5cb8da1 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -71,7 +71,11 @@ static long ratelimit_pages = 32;
 /*
  * Start background writeback (via writeback threads) at this percentage
  */
+#ifdef CONFIG_CACHY
+static int dirty_background_ratio = 5;
+#else
 static int dirty_background_ratio = 10;
+#endif
 
 /*
  * dirty_background_bytes starts at 0 (disabled) so that it is a function of
@@ -99,7 +103,11 @@ static unsigned long vm_dirty_bytes;
 /*
  * The interval between `kupdate'-style writebacks
  */
+#ifdef CONFIG_CACHY
+unsigned int dirty_writeback_interval = 10 * 100; /* centiseconds */
+#else
 unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */
+#endif
 
 EXPORT_SYMBOL_GPL(dirty_writeback_interval);
 
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index a663202045dc..7c48b114331b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -287,7 +287,11 @@ const char * const migratetype_names[MIGRATE_TYPES] = {
 
 int min_free_kbytes = 1024;
 int user_min_free_kbytes = -1;
+#ifdef CONFIG_CACHY
+static int watermark_boost_factor __read_mostly;
+#else
 static int watermark_boost_factor __read_mostly = 15000;
+#endif
 static int watermark_scale_factor = 10;
 
 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
@@ -2116,16 +2120,17 @@ __rmqueue(struct zone *zone, unsigned int order, int migratetype,
 }
 
 /*
- * Obtain a specified number of elements from the buddy allocator, all under
- * a single hold of the lock, for efficiency.  Add them to the supplied list.
- * Returns the number of new pages which were placed at *list.
+ * Obtain a specified number of elements from the buddy allocator, and relax the
+ * zone lock when needed. Add them to the supplied list. Returns the number of
+ * new pages which were placed at *list.
  */
 static int rmqueue_bulk(struct zone *zone, unsigned int order,
 			unsigned long count, struct list_head *list,
 			int migratetype, unsigned int alloc_flags)
 {
+	const bool can_resched = !preempt_count() && !irqs_disabled();
 	unsigned long flags;
-	int i;
+	int i, last_mod = 0;
 
 	spin_lock_irqsave(&zone->lock, flags);
 	for (i = 0; i < count; ++i) {
@@ -2134,6 +2139,18 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
 		if (unlikely(page == NULL))
 			break;
 
+		/* Reschedule and ease the contention on the lock if needed */
+		if (i + 1 < count && ((can_resched && need_resched()) ||
+				      spin_needbreak(&zone->lock))) {
+			__mod_zone_page_state(zone, NR_FREE_PAGES,
+					      -((i + 1 - last_mod) << order));
+			last_mod = i + 1;
+			spin_unlock_irqrestore(&zone->lock, flags);
+			if (can_resched)
+				cond_resched();
+			spin_lock_irqsave(&zone->lock, flags);
+		}
+
 		/*
 		 * Split buddy pages returned by expand() are received here in
 		 * physical page order. The page is added to the tail of
@@ -2150,7 +2167,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
 					      -(1 << order));
 	}
 
-	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
+	__mod_zone_page_state(zone, NR_FREE_PAGES, -((i - last_mod) << order));
 	spin_unlock_irqrestore(&zone->lock, flags);
 
 	return i;
diff --git a/mm/swap.c b/mm/swap.c
index cd8f0150ba3a..42c405a4f114 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -1090,6 +1090,10 @@ void folio_batch_remove_exceptionals(struct folio_batch *fbatch)
  */
 void __init swap_setup(void)
 {
+#ifdef CONFIG_CACHY
+	/* Only swap-in pages requested, avoid readahead */
+	page_cluster = 0;
+#else
 	unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
 
 	/* Use a smaller cluster for small-memory machines */
@@ -1101,4 +1105,5 @@ void __init swap_setup(void)
 	 * Right now other parts of the system means that we
 	 * _really_ don't want to cluster much more
 	 */
+#endif
 }
diff --git a/mm/vmpressure.c b/mm/vmpressure.c
index bd5183dfd879..3a410f53a07c 100644
--- a/mm/vmpressure.c
+++ b/mm/vmpressure.c
@@ -43,7 +43,11 @@ static const unsigned long vmpressure_win = SWAP_CLUSTER_MAX * 16;
  * essence, they are percents: the higher the value, the more number
  * unsuccessful reclaims there were.
  */
+#ifdef CONFIG_CACHY
+static const unsigned int vmpressure_level_med = 65;
+#else
 static const unsigned int vmpressure_level_med = 60;
+#endif
 static const unsigned int vmpressure_level_critical = 95;
 
 /*
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 4255619a1a31..62f42e92964f 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -133,6 +133,15 @@ struct scan_control {
 	/* The file folios on the current node are dangerously low */
 	unsigned int file_is_tiny:1;
 
+	/* The anonymous pages on the current node are below vm.anon_min_ratio */
+	unsigned int anon_below_min:1;
+
+	/* The clean file pages on the current node are below vm.clean_low_ratio */
+	unsigned int clean_below_low:1;
+
+	/* The clean file pages on the current node are below vm.clean_min_ratio */
+	unsigned int clean_below_min:1;
+
 	/* Always discard instead of demoting to lower tier memory */
 	unsigned int no_demotion:1;
 
@@ -182,10 +191,23 @@ struct scan_control {
 #define prefetchw_prev_lru_folio(_folio, _base, _field) do { } while (0)
 #endif
 
+bool sysctl_workingset_protection __read_mostly = true;
+u8 sysctl_anon_min_ratio  __read_mostly = CONFIG_ANON_MIN_RATIO;
+u8 sysctl_clean_low_ratio __read_mostly = CONFIG_CLEAN_LOW_RATIO;
+u8 sysctl_clean_min_ratio __read_mostly = CONFIG_CLEAN_MIN_RATIO;
+static u64 sysctl_anon_min_ratio_kb  __read_mostly = 0;
+static u64 sysctl_clean_low_ratio_kb __read_mostly = 0;
+static u64 sysctl_clean_min_ratio_kb __read_mostly = 0;
+static u64 workingset_protection_prev_totalram __read_mostly = 0;
+
 /*
  * From 0 .. 200.  Higher means more swappy.
  */
+#ifdef CONFIG_CACHY
+int vm_swappiness = 20;
+#else
 int vm_swappiness = 60;
+#endif
 
 #ifdef CONFIG_MEMCG
 
@@ -1052,6 +1074,9 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,
 		    folio_mapped(folio) && folio_test_referenced(folio))
 			goto keep_locked;
 
+		if (folio_is_file_lru(folio) ? sc->clean_below_min : sc->anon_below_min)
+			goto keep_locked;
+
 		/*
 		 * The number of dirty pages determines if a node is marked
 		 * reclaim_congested. kswapd will stall and start writing
@@ -2353,6 +2378,23 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
 		goto out;
 	}
 
+	/*
+	 * Force-scan the other type if anon/clean pages is
+	 * under vm.{anon,clean}_{low,min}_ratio, respectively.
+	 */
+	if (sc->clean_below_min) {
+		scan_balance = SCAN_ANON;
+		goto out;
+	}
+	if (sc->anon_below_min) {
+		scan_balance = SCAN_FILE;
+		goto out;
+	}
+	if (sc->clean_below_low) {
+		scan_balance = SCAN_ANON;
+		goto out;
+	}
+
 	/*
 	 * Do not apply any pressure balancing cleverness when the
 	 * system is close to OOM, scan both anon and file equally
@@ -2515,6 +2557,14 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
 			BUG();
 		}
 
+		/*
+		 * Hard protection of the working set.
+		 * Don't reclaim anon/file pages when the amount is
+		 * below the watermark of the same type.
+		 */
+		if (file ? sc->clean_below_min : sc->anon_below_min)
+			scan = 0;
+
 		nr[lru] = scan;
 	}
 }
@@ -3922,7 +3972,28 @@ static bool lruvec_is_reclaimable(struct lruvec *lruvec, struct scan_control *sc
 }
 
 /* to protect the working set of the last N jiffies */
+#ifdef CONFIG_CACHY
+static unsigned long lru_gen_min_ttl __read_mostly = 1000;
+#else
 static unsigned long lru_gen_min_ttl __read_mostly;
+#endif
+
+static void do_invoke_oom(struct scan_control *sc, bool try_memcg) {
+	struct oom_control oc = {
+		.gfp_mask = sc->gfp_mask,
+		.order = sc->order,
+	};
+
+	if (try_memcg && mem_cgroup_oom_synchronize(true))
+		return;
+
+	if (!mutex_trylock(&oom_lock))
+		return;
+	out_of_memory(&oc);
+	mutex_unlock(&oom_lock);
+}
+#define invoke_oom(sc)         do_invoke_oom(sc, true)
+#define invoke_oom_nomemcg(sc) do_invoke_oom(sc, false)
 
 static void lru_gen_age_node(struct pglist_data *pgdat, struct scan_control *sc)
 {
@@ -3952,14 +4023,96 @@ static void lru_gen_age_node(struct pglist_data *pgdat, struct scan_control *sc)
 	 * younger than min_ttl. However, another possibility is all memcgs are
 	 * either too small or below min.
 	 */
-	if (mutex_trylock(&oom_lock)) {
-		struct oom_control oc = {
-			.gfp_mask = sc->gfp_mask,
-		};
+	invoke_oom_nomemcg(sc);
+}
+
+int vm_workingset_protection_update_handler(struct ctl_table *table, int write,
+		void __user *buffer, size_t *lenp, loff_t *ppos)
+{
+	int ret = proc_dou8vec_minmax(table, write, buffer, lenp, ppos);
+	if (ret || !write)
+		return ret;
+
+	workingset_protection_prev_totalram = 0;
+
+	return 0;
+}
+
+static void prepare_workingset_protection(pg_data_t *pgdat, struct scan_control *sc)
+{
+	unsigned long node_mem_total;
+	struct sysinfo i;
+
+	if (!(sysctl_workingset_protection)) {
+		sc->anon_below_min = 0;
+		sc->clean_below_low = 0;
+		sc->clean_below_min = 0;
+		return;
+	}
+
+	if (likely(sysctl_anon_min_ratio  ||
+	           sysctl_clean_low_ratio ||
+		       sysctl_clean_min_ratio)) {
+#ifdef CONFIG_NUMA
+		si_meminfo_node(&i, pgdat->node_id);
+#else //CONFIG_NUMA
+		si_meminfo(&i);
+#endif //CONFIG_NUMA
+		node_mem_total = i.totalram;
+
+		if (unlikely(workingset_protection_prev_totalram != node_mem_total)) {
+			sysctl_anon_min_ratio_kb  =
+				node_mem_total * sysctl_anon_min_ratio  / 100;
+			sysctl_clean_low_ratio_kb =
+				node_mem_total * sysctl_clean_low_ratio / 100;
+			sysctl_clean_min_ratio_kb =
+				node_mem_total * sysctl_clean_min_ratio / 100;
+			workingset_protection_prev_totalram = node_mem_total;
+		}
+	}
 
-		out_of_memory(&oc);
+	/*
+	 * Check the number of anonymous pages to protect them from
+	 * reclaiming if their amount is below the specified.
+	 */
+	if (sysctl_anon_min_ratio) {
+		unsigned long reclaimable_anon;
+
+		reclaimable_anon =
+			node_page_state(pgdat, NR_ACTIVE_ANON) +
+			node_page_state(pgdat, NR_INACTIVE_ANON) +
+			node_page_state(pgdat, NR_ISOLATED_ANON);
+
+		sc->anon_below_min = reclaimable_anon < sysctl_anon_min_ratio_kb;
+	} else
+		sc->anon_below_min = 0;
+
+	/*
+	 * Check the number of clean file pages to protect them from
+	 * reclaiming if their amount is below the specified.
+	 */
+	if (sysctl_clean_low_ratio || sysctl_clean_min_ratio) {
+		unsigned long reclaimable_file, dirty, clean;
+
+		reclaimable_file =
+			node_page_state(pgdat, NR_ACTIVE_FILE) +
+			node_page_state(pgdat, NR_INACTIVE_FILE) +
+			node_page_state(pgdat, NR_ISOLATED_FILE);
+		dirty = node_page_state(pgdat, NR_FILE_DIRTY);
+		/*
+		 * node_page_state() sum can go out of sync since
+		 * all the values are not read at once.
+		 */
+		if (likely(reclaimable_file > dirty))
+			clean = reclaimable_file - dirty;
+		else
+			clean = 0;
 
-		mutex_unlock(&oom_lock);
+		sc->clean_below_low = clean < sysctl_clean_low_ratio_kb;
+		sc->clean_below_min = clean < sysctl_clean_min_ratio_kb;
+	} else {
+		sc->clean_below_low = 0;
+		sc->clean_below_min = 0;
 	}
 }
 
@@ -4462,6 +4615,12 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	 */
 	if (!swappiness)
 		type = LRU_GEN_FILE;
+	else if (sc->clean_below_min)
+		type = LRU_GEN_ANON;
+	else if (sc->anon_below_min)
+		type = LRU_GEN_FILE;
+	else if (sc->clean_below_low)
+		type = LRU_GEN_ANON;
 	else if (min_seq[LRU_GEN_ANON] < min_seq[LRU_GEN_FILE])
 		type = LRU_GEN_ANON;
 	else if (swappiness == 1)
@@ -4471,7 +4630,7 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	else
 		type = get_type_to_scan(lruvec, swappiness, &tier);
 
-	for (i = !swappiness; i < ANON_AND_FILE; i++) {
+	for (i = 0; i < ANON_AND_FILE; i++) {
 		if (tier < 0)
 			tier = get_tier_idx(lruvec, type);
 
@@ -4749,6 +4908,7 @@ static int shrink_one(struct lruvec *lruvec, struct scan_control *sc)
 	struct mem_cgroup *memcg = lruvec_memcg(lruvec);
 	struct pglist_data *pgdat = lruvec_pgdat(lruvec);
 
+	prepare_workingset_protection(pgdat, sc);
 	mem_cgroup_calculate_protection(NULL, memcg);
 
 	if (mem_cgroup_below_min(NULL, memcg))
@@ -5899,6 +6059,8 @@ static void shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 
 	prepare_scan_control(pgdat, sc);
 
+	prepare_workingset_protection(pgdat, sc);
+
 	shrink_node_memcgs(pgdat, sc);
 
 	flush_reclaim_state(sc);
@@ -5987,6 +6149,8 @@ static void shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 	 */
 	if (reclaimable)
 		pgdat->kswapd_failures = 0;
+	else if (sc->clean_below_min && !sc->priority)
+		invoke_oom(sc);
 }
 
 /*
-- 
2.44.0

From 3719b448ce6ae6e6df7f49a99ef30eeb0bf2117d Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:43:37 +0200
Subject: [PATCH] Revert "le9uo"

This reverts commit 9bb31a68ef456524c4370323e1c19b07fc0633df.
---
 Documentation/admin-guide/sysctl/vm.rst |  72 ----------
 include/linux/mm.h                      |   8 --
 kernel/sysctl.c                         |  34 -----
 mm/Kconfig                              |  63 ---------
 mm/mm_init.c                            |   1 -
 mm/vmscan.c                             | 170 +-----------------------
 6 files changed, 7 insertions(+), 341 deletions(-)

diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst
index 468ae7dec1e1..c59889de122b 100644
--- a/Documentation/admin-guide/sysctl/vm.rst
+++ b/Documentation/admin-guide/sysctl/vm.rst
@@ -25,9 +25,6 @@ files can be found in mm/swap.c.
 Currently, these files are in /proc/sys/vm:
 
 - admin_reserve_kbytes
-- anon_min_ratio
-- clean_low_ratio
-- clean_min_ratio
 - compact_memory
 - compaction_proactiveness
 - compact_unevictable_allowed
@@ -109,67 +106,6 @@ On x86_64 this is about 128MB.
 Changing this takes effect whenever an application requests memory.
 
 
-anon_min_ratio
-==============
-
-This knob provides *hard* protection of anonymous pages. The anonymous pages
-on the current node won't be reclaimed under any conditions when their amount
-is below vm.anon_min_ratio.
-
-This knob may be used to prevent excessive swap thrashing when anonymous
-memory is low (for example, when memory is going to be overfilled by
-compressed data of zram module).
-
-Setting this value too high (close to 100) can result in inability to
-swap and can lead to early OOM under memory pressure.
-
-The unit of measurement is the percentage of the total memory of the node.
-
-The default value is 15.
-
-
-clean_low_ratio
-================
-
-This knob provides *best-effort* protection of clean file pages. The file pages
-on the current node won't be reclaimed under memory pressure when the amount of
-clean file pages is below vm.clean_low_ratio *unless* we threaten to OOM.
-
-Protection of clean file pages using this knob may be used when swapping is
-still possible to
-  - prevent disk I/O thrashing under memory pressure;
-  - improve performance in disk cache-bound tasks under memory pressure.
-
-Setting it to a high value may result in a early eviction of anonymous pages
-into the swap space by attempting to hold the protected amount of clean file
-pages in memory.
-
-The unit of measurement is the percentage of the total memory of the node.
-
-The default value is 0.
-
-
-clean_min_ratio
-================
-
-This knob provides *hard* protection of clean file pages. The file pages on the
-current node won't be reclaimed under memory pressure when the amount of clean
-file pages is below vm.clean_min_ratio.
-
-Hard protection of clean file pages using this knob may be used to
-  - prevent disk I/O thrashing under memory pressure even with no free swap space;
-  - improve performance in disk cache-bound tasks under memory pressure;
-  - avoid high latency and prevent livelock in near-OOM conditions.
-
-Setting it to a high value may result in a early out-of-memory condition due to
-the inability to reclaim the protected amount of clean file pages when other
-types of pages cannot be reclaimed.
-
-The unit of measurement is the percentage of the total memory of the node.
-
-The default value is 15.
-
-
 compact_memory
 ==============
 
@@ -974,14 +910,6 @@ be 133 (x + 2x = 200, 2x = 133.33).
 At 0, the kernel will not initiate swap until the amount of free and
 file-backed pages is less than the high watermark in a zone.
 
-This knob has no effect if the amount of clean file pages on the current
-node is below vm.clean_low_ratio or vm.clean_min_ratio. In this case,
-only anonymous pages can be reclaimed.
-
-If the number of anonymous pages on the current node is below
-vm.anon_min_ratio, then only file pages can be reclaimed with
-any vm.swappiness value.
-
 
 unprivileged_userfaultfd
 ========================
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 397ad6f1ac39..c9fb00c56844 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -195,14 +195,6 @@ static inline void __mm_zero_struct_page(struct page *page)
 
 extern int sysctl_max_map_count;
 
-extern bool sysctl_workingset_protection;
-extern u8 sysctl_anon_min_ratio;
-extern u8 sysctl_clean_low_ratio;
-extern u8 sysctl_clean_min_ratio;
-int vm_workingset_protection_update_handler(
-	struct ctl_table *table, int write,
-	void __user *buffer, size_t *lenp, loff_t *ppos);
-
 extern unsigned long sysctl_user_reserve_kbytes;
 extern unsigned long sysctl_admin_reserve_kbytes;
 
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index aa55ebba2ec3..c92d8a4b23fb 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -2216,40 +2216,6 @@ static struct ctl_table vm_table[] = {
 		.extra1		= SYSCTL_ZERO,
 	},
 #endif
-	{
-		.procname	= "workingset_protection",
-		.data		= &sysctl_workingset_protection,
-		.maxlen		= sizeof(bool),
-		.mode		= 0644,
-		.proc_handler	= &proc_dobool,
-	},
-	{
-		.procname	= "anon_min_ratio",
-		.data		= &sysctl_anon_min_ratio,
-		.maxlen		= sizeof(u8),
-		.mode		= 0644,
-		.proc_handler	= &vm_workingset_protection_update_handler,
-		.extra1		= SYSCTL_ZERO,
-		.extra2		= SYSCTL_ONE_HUNDRED,
-	},
-	{
-		.procname	= "clean_low_ratio",
-		.data		= &sysctl_clean_low_ratio,
-		.maxlen		= sizeof(u8),
-		.mode		= 0644,
-		.proc_handler	= &vm_workingset_protection_update_handler,
-		.extra1		= SYSCTL_ZERO,
-		.extra2		= SYSCTL_ONE_HUNDRED,
-	},
-	{
-		.procname	= "clean_min_ratio",
-		.data		= &sysctl_clean_min_ratio,
-		.maxlen		= sizeof(u8),
-		.mode		= 0644,
-		.proc_handler	= &vm_workingset_protection_update_handler,
-		.extra1		= SYSCTL_ZERO,
-		.extra2		= SYSCTL_ONE_HUNDRED,
-	},
 	{
 		.procname	= "user_reserve_kbytes",
 		.data		= &sysctl_user_reserve_kbytes,
diff --git a/mm/Kconfig b/mm/Kconfig
index 002f48b655de..0e440573033c 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -486,69 +486,6 @@ config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
 config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
 	bool
 
-config ANON_MIN_RATIO
-	int "Default value for vm.anon_min_ratio"
-	depends on SYSCTL
-	range 0 100
-	default 15
-	help
-	  This option sets the default value for vm.anon_min_ratio sysctl knob.
-
-	  The vm.anon_min_ratio sysctl knob provides *hard* protection of
-	  anonymous pages. The anonymous pages on the current node won't be
-	  reclaimed under any conditions when their amount is below
-	  vm.anon_min_ratio. This knob may be used to prevent excessive swap
-	  thrashing when anonymous memory is low (for example, when memory is
-	  going to be overfilled by compressed data of zram module).
-
-	  Setting this value too high (close to MemTotal) can result in
-	  inability to swap and can lead to early OOM under memory pressure.
-
-config CLEAN_LOW_RATIO
-	int "Default value for vm.clean_low_ratio"
-	depends on SYSCTL
-	range 0 100
-	default 0
-	help
-	  This option sets the default value for vm.clean_low_ratio sysctl knob.
-
-	  The vm.clean_low_ratio sysctl knob provides *best-effort*
-	  protection of clean file pages. The file pages on the current node
-	  won't be reclaimed under memory pressure when the amount of clean file
-	  pages is below vm.clean_low_ratio *unless* we threaten to OOM.
-	  Protection of clean file pages using this knob may be used when
-	  swapping is still possible to
-	    - prevent disk I/O thrashing under memory pressure;
-	    - improve performance in disk cache-bound tasks under memory
-	      pressure.
-
-	  Setting it to a high value may result in a early eviction of anonymous
-	  pages into the swap space by attempting to hold the protected amount
-	  of clean file pages in memory.
-
-config CLEAN_MIN_RATIO
-	int "Default value for vm.clean_min_ratio"
-	depends on SYSCTL
-	range 0 100
-	default 15
-	help
-	  This option sets the default value for vm.clean_min_ratio sysctl knob.
-
-	  The vm.clean_min_ratio sysctl knob provides *hard* protection of
-	  clean file pages. The file pages on the current node won't be
-	  reclaimed under memory pressure when the amount of clean file pages is
-	  below vm.clean_min_ratio. Hard protection of clean file pages using
-	  this knob may be used to
-	    - prevent disk I/O thrashing under memory pressure even with no free
-	      swap space;
-	    - improve performance in disk cache-bound tasks under memory
-	      pressure;
-	    - avoid high latency and prevent livelock in near-OOM conditions.
-
-	  Setting it to a high value may result in a early out-of-memory condition
-	  due to the inability to reclaim the protected amount of clean file pages
-	  when other types of pages cannot be reclaimed.
-
 config HAVE_MEMBLOCK_PHYS_MAP
 	bool
 
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 419ba5ac7c52..2c19f5515e36 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2749,7 +2749,6 @@ static void __init mem_init_print_info(void)
 		, K(totalhigh_pages())
 #endif
 		);
-	printk(KERN_INFO "le9 Unofficial (le9uo) working set protection 1.4 by Masahito Suzuki (forked from hakavlad's original le9 patch)");
 }
 
 /*
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 346810e1b69d..fd1d9b4194e3 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -133,15 +133,6 @@ struct scan_control {
 	/* The file folios on the current node are dangerously low */
 	unsigned int file_is_tiny:1;
 
-	/* The anonymous pages on the current node are below vm.anon_min_ratio */
-	unsigned int anon_below_min:1;
-
-	/* The clean file pages on the current node are below vm.clean_low_ratio */
-	unsigned int clean_below_low:1;
-
-	/* The clean file pages on the current node are below vm.clean_min_ratio */
-	unsigned int clean_below_min:1;
-
 	/* Always discard instead of demoting to lower tier memory */
 	unsigned int no_demotion:1;
 
@@ -191,15 +182,6 @@ struct scan_control {
 #define prefetchw_prev_lru_folio(_folio, _base, _field) do { } while (0)
 #endif
 
-bool sysctl_workingset_protection __read_mostly = true;
-u8 sysctl_anon_min_ratio  __read_mostly = CONFIG_ANON_MIN_RATIO;
-u8 sysctl_clean_low_ratio __read_mostly = CONFIG_CLEAN_LOW_RATIO;
-u8 sysctl_clean_min_ratio __read_mostly = CONFIG_CLEAN_MIN_RATIO;
-static u64 sysctl_anon_min_ratio_kb  __read_mostly = 0;
-static u64 sysctl_clean_low_ratio_kb __read_mostly = 0;
-static u64 sysctl_clean_min_ratio_kb __read_mostly = 0;
-static u64 workingset_protection_prev_totalram __read_mostly = 0;
-
 /*
  * From 0 .. 200.  Higher means more swappy.
  */
@@ -1074,9 +1056,6 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,
 		    folio_mapped(folio) && folio_test_referenced(folio))
 			goto keep_locked;
 
-		if (folio_is_file_lru(folio) ? sc->clean_below_min : sc->anon_below_min)
-			goto keep_locked;
-
 		/*
 		 * The number of dirty pages determines if a node is marked
 		 * reclaim_congested. kswapd will stall and start writing
@@ -2378,23 +2357,6 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
 		goto out;
 	}
 
-	/*
-	 * Force-scan the other type if anon/clean pages is
-	 * under vm.{anon,clean}_{low,min}_ratio, respectively.
-	 */
-	if (sc->clean_below_min) {
-		scan_balance = SCAN_ANON;
-		goto out;
-	}
-	if (sc->anon_below_min) {
-		scan_balance = SCAN_FILE;
-		goto out;
-	}
-	if (sc->clean_below_low) {
-		scan_balance = SCAN_ANON;
-		goto out;
-	}
-
 	/*
 	 * Do not apply any pressure balancing cleverness when the
 	 * system is close to OOM, scan both anon and file equally
@@ -2557,14 +2519,6 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
 			BUG();
 		}
 
-		/*
-		 * Hard protection of the working set.
-		 * Don't reclaim anon/file pages when the amount is
-		 * below the watermark of the same type.
-		 */
-		if (file ? sc->clean_below_min : sc->anon_below_min)
-			scan = 0;
-
 		nr[lru] = scan;
 	}
 }
@@ -3978,23 +3932,6 @@ static unsigned long lru_gen_min_ttl __read_mostly = 1000;
 static unsigned long lru_gen_min_ttl __read_mostly;
 #endif
 
-static void do_invoke_oom(struct scan_control *sc, bool try_memcg) {
-	struct oom_control oc = {
-		.gfp_mask = sc->gfp_mask,
-		.order = sc->order,
-	};
-
-	if (try_memcg && mem_cgroup_oom_synchronize(true))
-		return;
-
-	if (!mutex_trylock(&oom_lock))
-		return;
-	out_of_memory(&oc);
-	mutex_unlock(&oom_lock);
-}
-#define invoke_oom(sc)         do_invoke_oom(sc, true)
-#define invoke_oom_nomemcg(sc) do_invoke_oom(sc, false)
-
 static void lru_gen_age_node(struct pglist_data *pgdat, struct scan_control *sc)
 {
 	struct mem_cgroup *memcg;
@@ -4023,96 +3960,14 @@ static void lru_gen_age_node(struct pglist_data *pgdat, struct scan_control *sc)
 	 * younger than min_ttl. However, another possibility is all memcgs are
 	 * either too small or below min.
 	 */
-	invoke_oom_nomemcg(sc);
-}
-
-int vm_workingset_protection_update_handler(struct ctl_table *table, int write,
-		void __user *buffer, size_t *lenp, loff_t *ppos)
-{
-	int ret = proc_dou8vec_minmax(table, write, buffer, lenp, ppos);
-	if (ret || !write)
-		return ret;
-
-	workingset_protection_prev_totalram = 0;
-
-	return 0;
-}
-
-static void prepare_workingset_protection(pg_data_t *pgdat, struct scan_control *sc)
-{
-	unsigned long node_mem_total;
-	struct sysinfo i;
-
-	if (!(sysctl_workingset_protection)) {
-		sc->anon_below_min = 0;
-		sc->clean_below_low = 0;
-		sc->clean_below_min = 0;
-		return;
-	}
-
-	if (likely(sysctl_anon_min_ratio  ||
-	           sysctl_clean_low_ratio ||
-		       sysctl_clean_min_ratio)) {
-#ifdef CONFIG_NUMA
-		si_meminfo_node(&i, pgdat->node_id);
-#else //CONFIG_NUMA
-		si_meminfo(&i);
-#endif //CONFIG_NUMA
-		node_mem_total = i.totalram;
-
-		if (unlikely(workingset_protection_prev_totalram != node_mem_total)) {
-			sysctl_anon_min_ratio_kb  =
-				node_mem_total * sysctl_anon_min_ratio  / 100;
-			sysctl_clean_low_ratio_kb =
-				node_mem_total * sysctl_clean_low_ratio / 100;
-			sysctl_clean_min_ratio_kb =
-				node_mem_total * sysctl_clean_min_ratio / 100;
-			workingset_protection_prev_totalram = node_mem_total;
-		}
-	}
-
-	/*
-	 * Check the number of anonymous pages to protect them from
-	 * reclaiming if their amount is below the specified.
-	 */
-	if (sysctl_anon_min_ratio) {
-		unsigned long reclaimable_anon;
-
-		reclaimable_anon =
-			node_page_state(pgdat, NR_ACTIVE_ANON) +
-			node_page_state(pgdat, NR_INACTIVE_ANON) +
-			node_page_state(pgdat, NR_ISOLATED_ANON);
+	if (mutex_trylock(&oom_lock)) {
+		struct oom_control oc = {
+			.gfp_mask = sc->gfp_mask,
+		};
 
-		sc->anon_below_min = reclaimable_anon < sysctl_anon_min_ratio_kb;
-	} else
-		sc->anon_below_min = 0;
+		out_of_memory(&oc);
 
-	/*
-	 * Check the number of clean file pages to protect them from
-	 * reclaiming if their amount is below the specified.
-	 */
-	if (sysctl_clean_low_ratio || sysctl_clean_min_ratio) {
-		unsigned long reclaimable_file, dirty, clean;
-
-		reclaimable_file =
-			node_page_state(pgdat, NR_ACTIVE_FILE) +
-			node_page_state(pgdat, NR_INACTIVE_FILE) +
-			node_page_state(pgdat, NR_ISOLATED_FILE);
-		dirty = node_page_state(pgdat, NR_FILE_DIRTY);
-		/*
-		 * node_page_state() sum can go out of sync since
-		 * all the values are not read at once.
-		 */
-		if (likely(reclaimable_file > dirty))
-			clean = reclaimable_file - dirty;
-		else
-			clean = 0;
-
-		sc->clean_below_low = clean < sysctl_clean_low_ratio_kb;
-		sc->clean_below_min = clean < sysctl_clean_min_ratio_kb;
-	} else {
-		sc->clean_below_low = 0;
-		sc->clean_below_min = 0;
+		mutex_unlock(&oom_lock);
 	}
 }
 
@@ -4615,12 +4470,6 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	 */
 	if (!swappiness)
 		type = LRU_GEN_FILE;
-	else if (sc->clean_below_min)
-		type = LRU_GEN_ANON;
-	else if (sc->anon_below_min)
-		type = LRU_GEN_FILE;
-	else if (sc->clean_below_low)
-		type = LRU_GEN_ANON;
 	else if (min_seq[LRU_GEN_ANON] < min_seq[LRU_GEN_FILE])
 		type = LRU_GEN_ANON;
 	else if (swappiness == 1)
@@ -4630,7 +4479,7 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	else
 		type = get_type_to_scan(lruvec, swappiness, &tier);
 
-	for (i = 0; i < ANON_AND_FILE; i++) {
+	for (i = !swappiness; i < ANON_AND_FILE; i++) {
 		if (tier < 0)
 			tier = get_tier_idx(lruvec, type);
 
@@ -4908,7 +4757,6 @@ static int shrink_one(struct lruvec *lruvec, struct scan_control *sc)
 	struct mem_cgroup *memcg = lruvec_memcg(lruvec);
 	struct pglist_data *pgdat = lruvec_pgdat(lruvec);
 
-	prepare_workingset_protection(pgdat, sc);
 	mem_cgroup_calculate_protection(NULL, memcg);
 
 	if (mem_cgroup_below_min(NULL, memcg))
@@ -6059,8 +5907,6 @@ static void shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 
 	prepare_scan_control(pgdat, sc);
 
-	prepare_workingset_protection(pgdat, sc);
-
 	shrink_node_memcgs(pgdat, sc);
 
 	flush_reclaim_state(sc);
@@ -6149,8 +5995,6 @@ static void shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 	 */
 	if (reclaimable)
 		pgdat->kswapd_failures = 0;
-	else if (sc->clean_below_min && !sc->priority)
-		invoke_oom(sc);
 }
 
 /*
-- 
2.44.0

From 4833f48c9738d6bb475df2e4c16be2ea26a7d91d Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:07:02 +0200
Subject: [PATCH 6/8] fixes

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 .../ABI/testing/sysfs-driver-hid-asus         |   85 +
 arch/Kconfig                                  |    4 +-
 drivers/hid/Makefile                          |    2 +
 drivers/hid/{hid-asus.c => hid-asus-core.c}   |  193 +--
 drivers/hid/hid-asus-rog.c                    | 1468 +++++++++++++++++
 drivers/hid/hid-asus-rog.h                    |  482 ++++++
 drivers/hid/hid-asus.h                        |   58 +
 drivers/hid/hid-ids.h                         |    1 +
 8 files changed, 2174 insertions(+), 119 deletions(-)
 create mode 100644 Documentation/ABI/testing/sysfs-driver-hid-asus
 rename drivers/hid/{hid-asus.c => hid-asus-core.c} (89%)
 create mode 100644 drivers/hid/hid-asus-rog.c
 create mode 100644 drivers/hid/hid-asus-rog.h
 create mode 100644 drivers/hid/hid-asus.h

diff --git a/Documentation/ABI/testing/sysfs-driver-hid-asus b/Documentation/ABI/testing/sysfs-driver-hid-asus
new file mode 100644
index 000000000000..df5b0c5b0702
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-driver-hid-asus
@@ -0,0 +1,85 @@
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/gamepad_mode
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the mode the ROG Ally xpad operates in:
+		- 1 = Game mode
+		- 2 = WASD mode
+		- 3 = Mouse mode
+		This setting applies instantly and applies settings that were previously changed
+		under that mode which are:
+		- deadzones
+		- anti-deadzones
+		- button mapping
+		- button turbo settings
+		- response curves
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/apply
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Apply the settings that have been stored in attributes so far. Because there are
+		many individual settings across a dozen packets this separation is required to
+		prevent spamming the MCU when userspace applications apply many changes at once.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/reset_btn_mapping
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Reset a gamepad mode to its default button mapping.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y/z>_<left/right>/deadzone
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the inner and outer deadzones of joysticks and triggers. These settings are not
+		written to the MCU until `apply` is set.
+		- range 0-64 (corresponds to 0-100%)
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y/z>_<left/right>/deadzone_index
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Descriptive labels for joystick deadzone array.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y>_<left/right>/anti-deadzone
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the joystick anti-deadzone feature:
+		- range 0-32 (corresponds to 0-50%)
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y/z>_<left/right>/calibration
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Calibration values for the joysticks and trigger analogues. There are no default
+		values as the calibration is determined in userspace.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y/z>_<left/right>/calibration_index
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Descriptive labels for joystick and triggers calibration array.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y>_<left/right>/rc_point<n>
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the joystick response curve. There are 4 points available with 1 being the lowest
+		point and 4 being the highest point.
+		- range 0-64 (corresponds to 0-100%)
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/axis_<x/y>_<left/right>/rc_point_index
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Descriptive labels for joystick response curve points.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/btn_<label>/turbo
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the turbo mode of the button:
+		- 0 = no turbo, a separate press and release is registered on press and release
+		- 1-16 = interval between presses if button held down in steps of 1000ms/16
+		These settings are not written to the MCU until `apply` is set.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/vibration_intensity
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Set the vibration intensity for left and right haptics. Applies instantly.
+
+What:		/sys/bus/usb/devices/1-3:1.0/0003:0B05:1ABE.0001/vibration_intensity_index
+Date:		December 2023
+Contact:	linux-input@vger.kernel.org
+Description:	Descriptive labels for index points of vibration_intensity.
\ No newline at end of file
diff --git a/arch/Kconfig b/arch/Kconfig
index a5af0edd3eb8..0731bc203aa9 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -1025,7 +1025,7 @@ config ARCH_MMAP_RND_BITS
 	int "Number of bits to use for ASLR of mmap base address" if EXPERT
 	range ARCH_MMAP_RND_BITS_MIN ARCH_MMAP_RND_BITS_MAX
 	default ARCH_MMAP_RND_BITS_DEFAULT if ARCH_MMAP_RND_BITS_DEFAULT
-	default ARCH_MMAP_RND_BITS_MIN
+	default ARCH_MMAP_RND_BITS_MAX
 	depends on HAVE_ARCH_MMAP_RND_BITS
 	help
 	  This value can be used to select the number of bits to use to
@@ -1059,7 +1059,7 @@ config ARCH_MMAP_RND_COMPAT_BITS
 	int "Number of bits to use for ASLR of mmap base address for compatible applications" if EXPERT
 	range ARCH_MMAP_RND_COMPAT_BITS_MIN ARCH_MMAP_RND_COMPAT_BITS_MAX
 	default ARCH_MMAP_RND_COMPAT_BITS_DEFAULT if ARCH_MMAP_RND_COMPAT_BITS_DEFAULT
-	default ARCH_MMAP_RND_COMPAT_BITS_MIN
+	default ARCH_MMAP_RND_COMPAT_BITS_MAX
 	depends on HAVE_ARCH_MMAP_RND_COMPAT_BITS
 	help
 	  This value can be used to select the number of bits to use to
diff --git a/drivers/hid/Makefile b/drivers/hid/Makefile
index 082a728eac60..7eefb548e33a 100644
--- a/drivers/hid/Makefile
+++ b/drivers/hid/Makefile
@@ -23,6 +23,8 @@ hid-logitech-$(CONFIG_LOGIWHEELS_FF)	+= hid-lg4ff.o
 hid-wiimote-y		:= hid-wiimote-core.o hid-wiimote-modules.o
 hid-wiimote-$(CONFIG_DEBUG_FS)	+= hid-wiimote-debug.o
 
+hid-asus-y		:= hid-asus-core.o hid-asus-rog.o
+
 obj-$(CONFIG_HID_A4TECH)	+= hid-a4tech.o
 obj-$(CONFIG_HID_ACCUTOUCH)	+= hid-accutouch.o
 obj-$(CONFIG_HID_ALPS)		+= hid-alps.o
diff --git a/drivers/hid/hid-asus.c b/drivers/hid/hid-asus-core.c
similarity index 89%
rename from drivers/hid/hid-asus.c
rename to drivers/hid/hid-asus-core.c
index 78cdfb8b9a7a..026705c43ee1 100644
--- a/drivers/hid/hid-asus.c
+++ b/drivers/hid/hid-asus-core.c
@@ -20,9 +20,8 @@
  *  Copyright (c) 2016 Frederik Wenigwieser <frederik.wenigwieser@gmail.com>
  */
 
-/*
- */
-
+#include <asm-generic/errno-base.h>
+#include <asm-generic/errno.h>
 #include <linux/dmi.h>
 #include <linux/hid.h>
 #include <linux/module.h>
@@ -32,6 +31,7 @@
 #include <linux/power_supply.h>
 #include <linux/leds.h>
 
+#include "hid-asus.h"
 #include "hid-ids.h"
 
 MODULE_AUTHOR("Yusuke Fujimaki <usk.fujimaki@gmail.com>");
@@ -47,10 +47,6 @@ MODULE_DESCRIPTION("Asus HID Keyboard and TouchPad");
 #define T100CHI_MOUSE_REPORT_ID 0x06
 #define FEATURE_REPORT_ID 0x0d
 #define INPUT_REPORT_ID 0x5d
-#define FEATURE_KBD_REPORT_ID 0x5a
-#define FEATURE_KBD_REPORT_SIZE 16
-#define FEATURE_KBD_LED_REPORT_ID1 0x5d
-#define FEATURE_KBD_LED_REPORT_ID2 0x5e
 
 #define SUPPORT_KBD_BACKLIGHT BIT(0)
 
@@ -71,20 +67,6 @@ MODULE_DESCRIPTION("Asus HID Keyboard and TouchPad");
 #define	BATTERY_STAT_CHARGING	(1)
 #define	BATTERY_STAT_FULL	(2)
 
-#define QUIRK_FIX_NOTEBOOK_REPORT	BIT(0)
-#define QUIRK_NO_INIT_REPORTS		BIT(1)
-#define QUIRK_SKIP_INPUT_MAPPING	BIT(2)
-#define QUIRK_IS_MULTITOUCH		BIT(3)
-#define QUIRK_NO_CONSUMER_USAGES	BIT(4)
-#define QUIRK_USE_KBD_BACKLIGHT		BIT(5)
-#define QUIRK_T100_KEYBOARD		BIT(6)
-#define QUIRK_T100CHI			BIT(7)
-#define QUIRK_G752_KEYBOARD		BIT(8)
-#define QUIRK_T90CHI			BIT(9)
-#define QUIRK_MEDION_E1239T		BIT(10)
-#define QUIRK_ROG_NKEY_KEYBOARD		BIT(11)
-#define QUIRK_ROG_CLAYMORE_II_KEYBOARD BIT(12)
-
 #define I2C_KEYBOARD_QUIRKS			(QUIRK_FIX_NOTEBOOK_REPORT | \
 						 QUIRK_NO_INIT_REPORTS | \
 						 QUIRK_NO_CONSUMER_USAGES)
@@ -113,22 +95,6 @@ struct asus_touchpad_info {
 	int report_size;
 };
 
-struct asus_drvdata {
-	unsigned long quirks;
-	struct hid_device *hdev;
-	struct input_dev *input;
-	struct input_dev *tp_kbd_input;
-	struct asus_kbd_leds *kbd_backlight;
-	const struct asus_touchpad_info *tp;
-	bool enable_backlight;
-	struct power_supply *battery;
-	struct power_supply_desc battery_desc;
-	int battery_capacity;
-	int battery_stat;
-	bool battery_in_query;
-	unsigned long battery_next_query;
-};
-
 static int asus_report_battery(struct asus_drvdata *, u8 *, int);
 
 static const struct asus_touchpad_info asus_i2c_tp = {
@@ -329,42 +295,36 @@ static int asus_raw_event(struct hid_device *hdev,
 	if (drvdata->battery && data[0] == BATTERY_REPORT_ID)
 		return asus_report_battery(drvdata, data, size);
 
+	// TODO: remove after debugging
+	// if (data[0] == 0x5a || data[0] == 0x5d || data[0] == 0x5e){
+	// 	 for (int i = 0; i < size; i++) {
+	// 		if (i == 0)
+	// 			printk(KERN_INFO "GOT: %02x,", data[i]);
+	// 		else
+	// 			printk(KERN_CONT "%02x,", data[i]);
+	// 	 }
+	// }
+
 	if (drvdata->tp && data[0] == INPUT_REPORT_ID)
 		return asus_report_input(drvdata, data, size);
 
 	if (drvdata->quirks & QUIRK_MEDION_E1239T)
 		return asus_e1239t_event(drvdata, data, size);
 
-	if (drvdata->quirks & QUIRK_USE_KBD_BACKLIGHT) {
+	/*
+	 * Skip these report ID, the device emits a continuous stream associated
+	 * with the AURA mode it is in which looks like an 'echo'.
+	*/
+	if (report->id == FEATURE_KBD_LED_REPORT_ID1 || report->id == FEATURE_KBD_LED_REPORT_ID2)
+		return -1;
+	if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD) {
 		/*
-		 * Skip these report ID, the device emits a continuous stream associated
-		 * with the AURA mode it is in which looks like an 'echo'.
+		 * G713 and G733 send these codes on some keypresses, depending on
+		 * the key pressed it can trigger a shutdown event if not caught.
 		*/
-		if (report->id == FEATURE_KBD_LED_REPORT_ID1 ||
-				report->id == FEATURE_KBD_LED_REPORT_ID2) {
+		if(data[0] == 0x02 && data[1] == 0x30) {
 			return -1;
-		/* Additional report filtering */
-		} else if (report->id == FEATURE_KBD_REPORT_ID) {
-			/*
-			 * G14 and G15 send these codes on some keypresses with no
-			 * discernable reason for doing so. We'll filter them out to avoid
-			 * unmapped warning messages later.
-			*/
-			if (data[1] == 0xea || data[1] == 0xec || data[1] == 0x02 ||
-					data[1] == 0x8a || data[1] == 0x9e) {
-				return -1;
-			}
-		}
-		if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD) {
-			/*
-			 * G713 and G733 send these codes on some keypresses, depending on
-			 * the key pressed it can trigger a shutdown event if not caught.
-			*/
-			if(data[0] == 0x02 && data[1] == 0x30) {
-				return -1;
-			}
 		}
-
 	}
 
 	if (drvdata->quirks & QUIRK_ROG_CLAYMORE_II_KEYBOARD) {
@@ -381,7 +341,7 @@ static int asus_raw_event(struct hid_device *hdev,
 	return 0;
 }
 
-static int asus_kbd_set_report(struct hid_device *hdev, const u8 *buf, size_t buf_size)
+int asus_kbd_set_report(struct hid_device *hdev, const u8 *buf, size_t buf_size)
 {
 	unsigned char *dmabuf;
 	int ret;
@@ -402,9 +362,16 @@ static int asus_kbd_set_report(struct hid_device *hdev, const u8 *buf, size_t bu
 	return ret;
 }
 
-static int asus_kbd_init(struct hid_device *hdev)
+int asus_kbd_get_report(struct hid_device *hdev, u8 *out_buf, size_t out_buf_size)
+{
+	return hid_hw_raw_request(hdev, FEATURE_KBD_REPORT_ID, out_buf,
+				 out_buf_size, HID_FEATURE_REPORT,
+				 HID_REQ_GET_REPORT);
+}
+
+static int asus_kbd_init(struct hid_device *hdev, u8 report_id)
 {
-	const u8 buf[] = { FEATURE_KBD_REPORT_ID, 0x41, 0x53, 0x55, 0x53, 0x20, 0x54,
+	const u8 buf[] = { report_id, 0x41, 0x53, 0x55, 0x53, 0x20, 0x54,
 		     0x65, 0x63, 0x68, 0x2e, 0x49, 0x6e, 0x63, 0x2e, 0x00 };
 	int ret;
 
@@ -416,9 +383,10 @@ static int asus_kbd_init(struct hid_device *hdev)
 }
 
 static int asus_kbd_get_functions(struct hid_device *hdev,
-				  unsigned char *kbd_func)
+				  unsigned char *kbd_func,
+				  u8 report_id)
 {
-	const u8 buf[] = { FEATURE_KBD_REPORT_ID, 0x05, 0x20, 0x31, 0x00, 0x08 };
+	const u8 buf[] = { report_id, 0x05, 0x20, 0x31, 0x00, 0x08 };
 	u8 *readbuf;
 	int ret;
 
@@ -447,51 +415,6 @@ static int asus_kbd_get_functions(struct hid_device *hdev,
 	return ret;
 }
 
-static int rog_nkey_led_init(struct hid_device *hdev)
-{
-	const u8 buf_init_start[] = { FEATURE_KBD_LED_REPORT_ID1, 0xB9 };
-	u8 buf_init2[] = { FEATURE_KBD_LED_REPORT_ID1, 0x41, 0x53, 0x55, 0x53, 0x20,
-				0x54, 0x65, 0x63, 0x68, 0x2e, 0x49, 0x6e, 0x63, 0x2e, 0x00 };
-	u8 buf_init3[] = { FEATURE_KBD_LED_REPORT_ID1,
-						0x05, 0x20, 0x31, 0x00, 0x08 };
-	int ret;
-
-	hid_info(hdev, "Asus initialise N-KEY Device");
-	/* The first message is an init start */
-	ret = asus_kbd_set_report(hdev, buf_init_start, sizeof(buf_init_start));
-	if (ret < 0) {
-		hid_warn(hdev, "Asus failed to send init start command: %d\n", ret);
-		return ret;
-	}
-	/* Followed by a string */
-	ret = asus_kbd_set_report(hdev, buf_init2, sizeof(buf_init2));
-	if (ret < 0) {
-		hid_warn(hdev, "Asus failed to send init command 1.0: %d\n", ret);
-		return ret;
-	}
-	/* Followed by a string */
-	ret = asus_kbd_set_report(hdev, buf_init3, sizeof(buf_init3));
-	if (ret < 0) {
-		hid_warn(hdev, "Asus failed to send init command 1.1: %d\n", ret);
-		return ret;
-	}
-
-	/* begin second report ID with same data */
-	buf_init2[0] = FEATURE_KBD_LED_REPORT_ID2;
-	buf_init3[0] = FEATURE_KBD_LED_REPORT_ID2;
-
-	ret = asus_kbd_set_report(hdev, buf_init2, sizeof(buf_init2));
-	if (ret < 0) {
-		hid_warn(hdev, "Asus failed to send init command 2.0: %d\n", ret);
-		return ret;
-	}
-	ret = asus_kbd_set_report(hdev, buf_init3, sizeof(buf_init3));
-	if (ret < 0)
-		hid_warn(hdev, "Asus failed to send init command 2.1: %d\n", ret);
-
-	return ret;
-}
-
 static void asus_schedule_work(struct asus_kbd_leds *led)
 {
 	unsigned long flags;
@@ -574,17 +497,27 @@ static int asus_kbd_register_leds(struct hid_device *hdev)
 	int ret;
 
 	if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD) {
-		ret = rog_nkey_led_init(hdev);
+		/* Initialize keyboard */
+		ret = asus_kbd_init(hdev, FEATURE_KBD_REPORT_ID);
+		if (ret < 0)
+			return ret;
+
+		/* The LED endpoint is initialised in two HID */
+		ret = asus_kbd_init(hdev, FEATURE_KBD_LED_REPORT_ID1);
+		if (ret < 0)
+			return ret;
+
+		ret = asus_kbd_init(hdev, FEATURE_KBD_LED_REPORT_ID2);
 		if (ret < 0)
 			return ret;
 	} else {
 		/* Initialize keyboard */
-		ret = asus_kbd_init(hdev);
+		ret = asus_kbd_init(hdev, FEATURE_KBD_REPORT_ID);
 		if (ret < 0)
 			return ret;
 
 		/* Get keyboard functions */
-		ret = asus_kbd_get_functions(hdev, &kbd_func);
+		ret = asus_kbd_get_functions(hdev, &kbd_func, FEATURE_KBD_REPORT_ID);
 		if (ret < 0)
 			return ret;
 
@@ -896,7 +829,11 @@ static int asus_input_mapping(struct hid_device *hdev,
 		case 0xb2: asus_map_key_clear(KEY_PROG2);	break; /* Fn+Left previous aura */
 		case 0xb3: asus_map_key_clear(KEY_PROG3);	break; /* Fn+Left next aura */
 		case 0x6a: asus_map_key_clear(KEY_F13);		break; /* Screenpad toggle */
-		case 0x4b: asus_map_key_clear(KEY_F14);		break; /* Arrows/Pg-Up/Dn toggle */
+		case 0x4b: asus_map_key_clear(KEY_F14);		break; /* Arrows/Pg-Up/Dn toggle, Ally M1 */
+		case 0xa5: asus_map_key_clear(KEY_F15);		break; /* ROG Ally M2 */
+		case 0xa6: asus_map_key_clear(KEY_F16);		break; /* ROG Ally QAM button */
+		case 0xa7: asus_map_key_clear(KEY_F17);		break; /* ROG Ally ROG long-press */
+		case 0xa8: asus_map_key_clear(KEY_F18);		break; /* ROG Ally ROG long-press-release */
 
 
 		default:
@@ -1109,6 +1046,10 @@ static int asus_probe(struct hid_device *hdev, const struct hid_device_id *id)
 		}
 	}
 
+	/* all ROG devices have this HID interface but we will focus on Ally for now */
+	if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD && hid_is_usb(hdev))
+		rog_ally.probe(hdev, &rog_ally);
+
 	ret = hid_parse(hdev);
 	if (ret) {
 		hid_err(hdev, "Asus hid parse failed: %d\n", ret);
@@ -1158,6 +1099,8 @@ static void asus_remove(struct hid_device *hdev)
 		cancel_work_sync(&drvdata->kbd_backlight->work);
 	}
 
+	rog_ally.remove(hdev, &rog_ally);
+
 	hid_hw_stop(hdev);
 }
 
@@ -1250,6 +1193,19 @@ static __u8 *asus_report_fixup(struct hid_device *hdev, __u8 *rdesc,
 		rdesc[205] = 0x01;
 	}
 
+	/* match many more n-key devices */
+	if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD) {
+		for (int i = 0; i < *rsize + 1; i++) {
+			/* offset to the count from 0x5a report part always 14 */
+			if (rdesc[i] == 0x85 && rdesc[i + 1] == 0x5a &&
+			    rdesc[i + 14] == 0x95 && rdesc[i + 15] == 0x05) {
+				hid_info(hdev, "Fixing up Asus N-Key report descriptor\n");
+				rdesc[i + 15] = 0x01;
+				break;
+			}
+		}
+	}
+
 	return rdesc;
 }
 
@@ -1276,6 +1232,9 @@ static const struct hid_device_id asus_devices[] = {
 	{ HID_USB_DEVICE(USB_VENDOR_ID_ASUSTEK,
 	    USB_DEVICE_ID_ASUSTEK_ROG_NKEY_KEYBOARD3),
 	  QUIRK_USE_KBD_BACKLIGHT | QUIRK_ROG_NKEY_KEYBOARD },
+	{ HID_USB_DEVICE(USB_VENDOR_ID_ASUSTEK,
+	    USB_DEVICE_ID_ASUSTEK_ROG_NKEY_ALLY),
+	  QUIRK_USE_KBD_BACKLIGHT | QUIRK_ROG_NKEY_KEYBOARD | QUIRK_ROG_ALLY_XPAD },
 	{ HID_USB_DEVICE(USB_VENDOR_ID_ASUSTEK,
 	    USB_DEVICE_ID_ASUSTEK_ROG_CLAYMORE_II_KEYBOARD),
 	  QUIRK_ROG_CLAYMORE_II_KEYBOARD },
diff --git a/drivers/hid/hid-asus-rog.c b/drivers/hid/hid-asus-rog.c
new file mode 100644
index 000000000000..584aefd0915a
--- /dev/null
+++ b/drivers/hid/hid-asus-rog.c
@@ -0,0 +1,1468 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  HID driver for Asus ROG laptops and Ally
+ *
+ *  Copyright (c) 2023 Luke Jones <luke@ljones.dev>
+ */
+
+#include <linux/hid.h>
+#include <linux/types.h>
+#include <linux/usb.h>
+
+#include "hid-asus.h"
+#include "hid-asus-rog.h"
+
+/* ROG Ally has many settings related to the gamepad, all using the same n-key endpoint */
+struct asus_rog_ally {
+	enum xpad_mode mode;
+	/*
+	 * index: [joysticks/triggers][left(2 bytes), right(2 bytes)]
+	 * joysticks: 2 bytes: inner, outer
+	 * triggers: 2 bytes: lower, upper
+	 * min/max: 0-64
+	 */
+	u8 deadzones[xpad_mode_mouse][2][4];
+	/*
+	 * index: left, right
+	 * max: 64
+	 */
+	u8 vibration_intensity[xpad_mode_mouse][2];
+	/*
+	 * index: [joysticks][2 byte stepping per point]
+	 * - 4 points of 2 bytes each
+	 * - byte 0 of pair = stick move %
+	 * - byte 1 of pair = stick response %
+	 * - min/max: 1-63
+	 */
+	bool supports_response_curves;
+	u8 response_curve[xpad_mode_mouse][2][8];
+	/*
+	 * left = byte 0, right = byte 1
+	 */
+	bool supports_anti_deadzones;
+	u8 anti_deadzones[xpad_mode_mouse][2];
+	/*
+	 * index: [mode][phys pair][b1, b1 secondary, b2, b2 secondary, blocks of 11]
+	*/
+	u8 key_mapping[xpad_mode_mouse][btn_pair_lt_rt][MAPPING_BLOCK_LEN];
+	/*
+	 *
+	*/
+	u8 turbo_btns[xpad_mode_mouse][TURBO_BLOCK_LEN];
+	/*
+	*/
+	u32 js_calibrations[2][6];
+	u32 tr_calibrations[2][2];
+};
+
+static struct asus_rog_ally *__rog_ally_data(struct device *raw_dev)
+{
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	return ((struct asus_drvdata *)hid_get_drvdata(hdev))->rog_ally_data;
+}
+
+#define STR_TO_CODE_IF(_idx, _code, _label) \
+	if (!strcmp(buf, _label))           \
+		out[_idx] = _code;
+
+#define STR_TO_CODE_ELIF(_idx, _code, _label) else if (!strcmp(buf, _label)) out[_idx] = _code;
+
+/* writes the bytes for a requested key/function in to the out buffer */
+const static int __string_to_key_code(const char *buf, u8 *out, int out_len)
+{
+	u8 *save_buf;
+
+	if (out_len != BTN_CODE_LEN)
+		return -EINVAL;
+
+	save_buf = kzalloc(out_len, GFP_KERNEL);
+	if (!save_buf)
+		return -ENOMEM;
+	memcpy(save_buf, out, out_len);
+	memset(out, 0, out_len); // always clear before adjusting
+
+	// Allow clearing
+	if (!strcmp(buf, " ") || !strcmp(buf, "\n"))
+		goto success;
+
+	// set group xpad
+	out[0] = 0x01;
+	STR_TO_CODE_IF(1, 0x01, PAD_A)
+	STR_TO_CODE_ELIF(1, 0x02, PAD_B)
+	STR_TO_CODE_ELIF(1, 0x03, PAD_X)
+	STR_TO_CODE_ELIF(1, 0x04, PAD_Y)
+	STR_TO_CODE_ELIF(1, 0x05, PAD_LB)
+	STR_TO_CODE_ELIF(1, 0x06, PAD_RB)
+	STR_TO_CODE_ELIF(1, 0x07, PAD_LS)
+	STR_TO_CODE_ELIF(1, 0x08, PAD_RS)
+	STR_TO_CODE_ELIF(1, 0x09, PAD_DPAD_UP)
+	STR_TO_CODE_ELIF(1, 0x0a, PAD_DPAD_DOWN)
+	STR_TO_CODE_ELIF(1, 0x0b, PAD_DPAD_LEFT)
+	STR_TO_CODE_ELIF(1, 0x0c, PAD_DPAD_RIGHT)
+	STR_TO_CODE_ELIF(1, 0x11, PAD_VIEW)
+	STR_TO_CODE_ELIF(1, 0x12, PAD_MENU)
+	STR_TO_CODE_ELIF(1, 0x13, PAD_XBOX)
+	if (out[1])
+		goto success;
+
+	// set group keyboard
+	out[0] = 0x02;
+	STR_TO_CODE_IF(2, 0x8f, KB_M1)
+	STR_TO_CODE_ELIF(2, 0x8e, KB_M2)
+
+	STR_TO_CODE_ELIF(2, 0x76, KB_ESC)
+	STR_TO_CODE_ELIF(2, 0x50, KB_F1)
+	STR_TO_CODE_ELIF(2, 0x60, KB_F2)
+	STR_TO_CODE_ELIF(2, 0x40, KB_F3)
+	STR_TO_CODE_ELIF(2, 0x0c, KB_F4)
+	STR_TO_CODE_ELIF(2, 0x03, KB_F5)
+	STR_TO_CODE_ELIF(2, 0x0b, KB_F6)
+	STR_TO_CODE_ELIF(2, 0x80, KB_F7)
+	STR_TO_CODE_ELIF(2, 0x0a, KB_F8)
+	STR_TO_CODE_ELIF(2, 0x01, KB_F9)
+	STR_TO_CODE_ELIF(2, 0x09, KB_F10)
+	STR_TO_CODE_ELIF(2, 0x78, KB_F11)
+	STR_TO_CODE_ELIF(2, 0x07, KB_F12)
+	STR_TO_CODE_ELIF(2, 0x10, KB_F14)
+	STR_TO_CODE_ELIF(2, 0x18, KB_F15)
+
+	STR_TO_CODE_ELIF(2, 0x0e, KB_BACKTICK)
+	STR_TO_CODE_ELIF(2, 0x16, KB_1)
+	STR_TO_CODE_ELIF(2, 0x1e, KB_2)
+	STR_TO_CODE_ELIF(2, 0x26, KB_3)
+	STR_TO_CODE_ELIF(2, 0x25, KB_4)
+	STR_TO_CODE_ELIF(2, 0x2e, KB_5)
+	STR_TO_CODE_ELIF(2, 0x36, KB_6)
+	STR_TO_CODE_ELIF(2, 0x3d, KB_7)
+	STR_TO_CODE_ELIF(2, 0x3e, KB_8)
+	STR_TO_CODE_ELIF(2, 0x46, KB_9)
+	STR_TO_CODE_ELIF(2, 0x45, KB_0)
+	STR_TO_CODE_ELIF(2, 0x4e, KB_HYPHEN)
+	STR_TO_CODE_ELIF(2, 0x55, KB_EQUALS)
+	STR_TO_CODE_ELIF(2, 0x66, KB_BACKSPACE)
+
+	STR_TO_CODE_ELIF(2, 0x0d, KB_TAB)
+	STR_TO_CODE_ELIF(2, 0x15, KB_Q)
+	STR_TO_CODE_ELIF(2, 0x1d, KB_W)
+	STR_TO_CODE_ELIF(2, 0x24, KB_E)
+	STR_TO_CODE_ELIF(2, 0x2d, KB_R)
+	STR_TO_CODE_ELIF(2, 0x2d, KB_T)
+	STR_TO_CODE_ELIF(2, 0x35, KB_Y)
+	STR_TO_CODE_ELIF(2, 0x3c, KB_U)
+	STR_TO_CODE_ELIF(2, 0x43, KB_I)
+	STR_TO_CODE_ELIF(2, 0x44, KB_O)
+	STR_TO_CODE_ELIF(2, 0x4d, KB_P)
+	STR_TO_CODE_ELIF(2, 0x54, KB_LBRACKET)
+	STR_TO_CODE_ELIF(2, 0x5b, KB_RBRACKET)
+	STR_TO_CODE_ELIF(2, 0x5d, KB_BACKSLASH)
+
+	STR_TO_CODE_ELIF(2, 0x58, KB_CAPS)
+	STR_TO_CODE_ELIF(2, 0x1c, KB_A)
+	STR_TO_CODE_ELIF(2, 0x1b, KB_S)
+	STR_TO_CODE_ELIF(2, 0x23, KB_D)
+	STR_TO_CODE_ELIF(2, 0x2b, KB_F)
+	STR_TO_CODE_ELIF(2, 0x34, KB_G)
+	STR_TO_CODE_ELIF(2, 0x33, KB_H)
+	STR_TO_CODE_ELIF(2, 0x3b, KB_J)
+	STR_TO_CODE_ELIF(2, 0x42, KB_K)
+	STR_TO_CODE_ELIF(2, 0x4b, KB_L)
+	STR_TO_CODE_ELIF(2, 0x4c, KB_SEMI)
+	STR_TO_CODE_ELIF(2, 0x52, KB_QUOTE)
+	STR_TO_CODE_ELIF(2, 0x5a, KB_RET)
+
+	STR_TO_CODE_ELIF(2, 0x88, KB_LSHIFT)
+	STR_TO_CODE_ELIF(2, 0x1a, KB_Z)
+	STR_TO_CODE_ELIF(2, 0x22, KB_X)
+	STR_TO_CODE_ELIF(2, 0x21, KB_C)
+	STR_TO_CODE_ELIF(2, 0x2a, KB_V)
+	STR_TO_CODE_ELIF(2, 0x32, KB_B)
+	STR_TO_CODE_ELIF(2, 0x31, KB_N)
+	STR_TO_CODE_ELIF(2, 0x3a, KB_M)
+	STR_TO_CODE_ELIF(2, 0x41, KB_COMMA)
+	STR_TO_CODE_ELIF(2, 0x49, KB_PERIOD)
+	STR_TO_CODE_ELIF(2, 0x4a, KB_FWDSLASH)
+	STR_TO_CODE_ELIF(2, 0x89, KB_RSHIFT)
+
+	STR_TO_CODE_ELIF(2, 0x8c, KB_LCTL)
+	STR_TO_CODE_ELIF(2, 0x82, KB_META)
+	STR_TO_CODE_ELIF(2, 0xba, KB_LALT)
+	STR_TO_CODE_ELIF(2, 0x29, KB_SPACE)
+	STR_TO_CODE_ELIF(2, 0x8b, KB_RALT)
+	STR_TO_CODE_ELIF(2, 0x84, KB_MENU)
+	STR_TO_CODE_ELIF(2, 0x8d, KB_RCTL)
+
+	STR_TO_CODE_ELIF(2, 0xc3, KB_PRNTSCN)
+	STR_TO_CODE_ELIF(2, 0x7e, KB_SCRLCK)
+	STR_TO_CODE_ELIF(2, 0x91, KB_PAUSE)
+	STR_TO_CODE_ELIF(2, 0xc2, KB_INS)
+	STR_TO_CODE_ELIF(2, 0x94, KB_HOME)
+	STR_TO_CODE_ELIF(2, 0x96, KB_PGUP)
+	STR_TO_CODE_ELIF(2, 0xc0, KB_DEL)
+	STR_TO_CODE_ELIF(2, 0x95, KB_END)
+	STR_TO_CODE_ELIF(2, 0x97, KB_PGDWN)
+
+	STR_TO_CODE_ELIF(2, 0x99, KB_UP_ARROW)
+	STR_TO_CODE_ELIF(2, 0x98, KB_DOWN_ARROW)
+	STR_TO_CODE_ELIF(2, 0x91, KB_LEFT_ARROW)
+	STR_TO_CODE_ELIF(2, 0x9b, KB_RIGHT_ARROW)
+
+	STR_TO_CODE_ELIF(2, 0x77, NUMPAD_LOCK)
+	STR_TO_CODE_ELIF(2, 0x90, NUMPAD_FWDSLASH)
+	STR_TO_CODE_ELIF(2, 0x7c, NUMPAD_ASTERISK)
+	STR_TO_CODE_ELIF(2, 0x7b, NUMPAD_HYPHEN)
+	STR_TO_CODE_ELIF(2, 0x70, NUMPAD_0)
+	STR_TO_CODE_ELIF(2, 0x69, NUMPAD_1)
+	STR_TO_CODE_ELIF(2, 0x72, NUMPAD_2)
+	STR_TO_CODE_ELIF(2, 0x7a, NUMPAD_3)
+	STR_TO_CODE_ELIF(2, 0x6b, NUMPAD_4)
+	STR_TO_CODE_ELIF(2, 0x73, NUMPAD_5)
+	STR_TO_CODE_ELIF(2, 0x74, NUMPAD_6)
+	STR_TO_CODE_ELIF(2, 0x6c, NUMPAD_7)
+	STR_TO_CODE_ELIF(2, 0x75, NUMPAD_8)
+	STR_TO_CODE_ELIF(2, 0x7d, NUMPAD_9)
+	STR_TO_CODE_ELIF(2, 0x79, NUMPAD_PLUS)
+	STR_TO_CODE_ELIF(2, 0x81, NUMPAD_ENTER)
+	STR_TO_CODE_ELIF(2, 0x71, NUMPAD_PERIOD)
+	if (out[2])
+		goto success;
+
+	out[0] = 0x03;
+	STR_TO_CODE_IF(4, 0x01, RAT_LCLICK)
+	STR_TO_CODE_ELIF(4, 0x02, RAT_RCLICK)
+	STR_TO_CODE_ELIF(4, 0x03, RAT_MCLICK)
+	STR_TO_CODE_ELIF(4, 0x04, RAT_WHEEL_UP)
+	STR_TO_CODE_ELIF(4, 0x05, RAT_WHEEL_DOWN)
+	if (out[4] != 0)
+		goto success;
+
+	out[0] = 0x05;
+	STR_TO_CODE_IF(3, 0x16, MEDIA_SCREENSHOT)
+	STR_TO_CODE_ELIF(3, 0x19, MEDIA_SHOW_KEYBOARD)
+	STR_TO_CODE_ELIF(3, 0x1c, MEDIA_SHOW_DESKTOP)
+	STR_TO_CODE_ELIF(3, 0x1e, MEDIA_START_RECORDING)
+	STR_TO_CODE_ELIF(3, 0x01, MEDIA_MIC_OFF)
+	STR_TO_CODE_ELIF(3, 0x02, MEDIA_VOL_DOWN)
+	STR_TO_CODE_ELIF(3, 0x03, MEDIA_VOL_UP)
+	if (out[3])
+		goto success;
+
+	// restore bytes if invalid input
+	memcpy(out, save_buf, out_len);
+	kfree(save_buf);
+	return -EINVAL;
+
+success:
+	kfree(save_buf);
+	return 0;
+}
+
+#define CODE_TO_STR_IF(_idx, _code, _label) \
+	if (btn_block[_idx] == _code)       \
+		return _label;
+
+const static char *__btn_map_to_string(struct device *raw_dev, enum btn_pair pair,
+				       enum btn_pair_side side, bool secondary)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	u8 *btn_block;
+	int offs;
+
+	// TODO: this little block is common
+	offs = side ? MAPPING_BLOCK_LEN / 2 : 0;
+	offs = secondary ? offs + BTN_CODE_LEN : offs;
+	btn_block = rog_ally->key_mapping[rog_ally->mode - 1][pair - 1] + offs;
+
+	if (btn_block[0] == 0x01) {
+		CODE_TO_STR_IF(1, 0x01, PAD_A)
+		CODE_TO_STR_IF(1, 0x02, PAD_B)
+		CODE_TO_STR_IF(1, 0x03, PAD_X)
+		CODE_TO_STR_IF(1, 0x04, PAD_Y)
+		CODE_TO_STR_IF(1, 0x05, PAD_LB)
+		CODE_TO_STR_IF(1, 0x06, PAD_RB)
+		CODE_TO_STR_IF(1, 0x07, PAD_LS)
+		CODE_TO_STR_IF(1, 0x08, PAD_RS)
+		CODE_TO_STR_IF(1, 0x09, PAD_DPAD_UP)
+		CODE_TO_STR_IF(1, 0x0a, PAD_DPAD_DOWN)
+		CODE_TO_STR_IF(1, 0x0b, PAD_DPAD_LEFT)
+		CODE_TO_STR_IF(1, 0x0c, PAD_DPAD_RIGHT)
+		CODE_TO_STR_IF(1, 0x11, PAD_VIEW)
+		CODE_TO_STR_IF(1, 0x12, PAD_MENU)
+		CODE_TO_STR_IF(1, 0x13, PAD_XBOX)
+	}
+
+	if (btn_block[0] == 0x02) {
+		CODE_TO_STR_IF(2, 0x8f, KB_M1)
+		CODE_TO_STR_IF(2, 0x8e, KB_M2)
+		CODE_TO_STR_IF(2, 0x76, KB_ESC)
+		CODE_TO_STR_IF(2, 0x50, KB_F1)
+		CODE_TO_STR_IF(2, 0x60, KB_F2)
+		CODE_TO_STR_IF(2, 0x40, KB_F3)
+		CODE_TO_STR_IF(2, 0x0c, KB_F4)
+		CODE_TO_STR_IF(2, 0x03, KB_F5)
+		CODE_TO_STR_IF(2, 0x0b, KB_F6)
+		CODE_TO_STR_IF(2, 0x80, KB_F7)
+		CODE_TO_STR_IF(2, 0x0a, KB_F8)
+		CODE_TO_STR_IF(2, 0x01, KB_F9)
+		CODE_TO_STR_IF(2, 0x09, KB_F10)
+		CODE_TO_STR_IF(2, 0x78, KB_F11)
+		CODE_TO_STR_IF(2, 0x07, KB_F12)
+		CODE_TO_STR_IF(2, 0x10, KB_F14)
+		CODE_TO_STR_IF(2, 0x18, KB_F15)
+
+		CODE_TO_STR_IF(2, 0x0e, KB_BACKTICK)
+		CODE_TO_STR_IF(2, 0x16, KB_1)
+		CODE_TO_STR_IF(2, 0x1e, KB_2)
+		CODE_TO_STR_IF(2, 0x26, KB_3)
+		CODE_TO_STR_IF(2, 0x25, KB_4)
+		CODE_TO_STR_IF(2, 0x2e, KB_5)
+		CODE_TO_STR_IF(2, 0x36, KB_6)
+		CODE_TO_STR_IF(2, 0x3d, KB_7)
+		CODE_TO_STR_IF(2, 0x3e, KB_8)
+		CODE_TO_STR_IF(2, 0x46, KB_9)
+		CODE_TO_STR_IF(2, 0x45, KB_0)
+		CODE_TO_STR_IF(2, 0x4e, KB_HYPHEN)
+		CODE_TO_STR_IF(2, 0x55, KB_EQUALS)
+		CODE_TO_STR_IF(2, 0x66, KB_BACKSPACE)
+
+		CODE_TO_STR_IF(2, 0x0d, KB_TAB)
+		CODE_TO_STR_IF(2, 0x15, KB_Q)
+		CODE_TO_STR_IF(2, 0x1d, KB_W)
+		CODE_TO_STR_IF(2, 0x24, KB_E)
+		CODE_TO_STR_IF(2, 0x2d, KB_R)
+		CODE_TO_STR_IF(2, 0x2d, KB_T)
+		CODE_TO_STR_IF(2, 0x35, KB_Y)
+		CODE_TO_STR_IF(2, 0x3c, KB_U)
+		CODE_TO_STR_IF(2, 0x43, KB_I)
+		CODE_TO_STR_IF(2, 0x44, KB_O)
+		CODE_TO_STR_IF(2, 0x4d, KB_P)
+		CODE_TO_STR_IF(2, 0x54, KB_LBRACKET)
+		CODE_TO_STR_IF(2, 0x5b, KB_RBRACKET)
+		CODE_TO_STR_IF(2, 0x5d, KB_BACKSLASH)
+
+		CODE_TO_STR_IF(2, 0x58, KB_CAPS)
+		CODE_TO_STR_IF(2, 0x1c, KB_A)
+		CODE_TO_STR_IF(2, 0x1b, KB_S)
+		CODE_TO_STR_IF(2, 0x23, KB_D)
+		CODE_TO_STR_IF(2, 0x2b, KB_F)
+		CODE_TO_STR_IF(2, 0x34, KB_G)
+		CODE_TO_STR_IF(2, 0x33, KB_H)
+		CODE_TO_STR_IF(2, 0x3b, KB_J)
+		CODE_TO_STR_IF(2, 0x42, KB_K)
+		CODE_TO_STR_IF(2, 0x4b, KB_L)
+		CODE_TO_STR_IF(2, 0x4c, KB_SEMI)
+		CODE_TO_STR_IF(2, 0x52, KB_QUOTE)
+		CODE_TO_STR_IF(2, 0x5a, KB_RET)
+
+		CODE_TO_STR_IF(2, 0x88, KB_LSHIFT)
+		CODE_TO_STR_IF(2, 0x1a, KB_Z)
+		CODE_TO_STR_IF(2, 0x22, KB_X)
+		CODE_TO_STR_IF(2, 0x21, KB_C)
+		CODE_TO_STR_IF(2, 0x2a, KB_V)
+		CODE_TO_STR_IF(2, 0x32, KB_B)
+		CODE_TO_STR_IF(2, 0x31, KB_N)
+		CODE_TO_STR_IF(2, 0x3a, KB_M)
+		CODE_TO_STR_IF(2, 0x41, KB_COMMA)
+		CODE_TO_STR_IF(2, 0x49, KB_PERIOD)
+		CODE_TO_STR_IF(2, 0x4a, KB_FWDSLASH)
+		CODE_TO_STR_IF(2, 0x89, KB_RSHIFT)
+
+		CODE_TO_STR_IF(2, 0x8c, KB_LCTL)
+		CODE_TO_STR_IF(2, 0x82, KB_META)
+		CODE_TO_STR_IF(2, 0xba, KB_LALT)
+		CODE_TO_STR_IF(2, 0x29, KB_SPACE)
+		CODE_TO_STR_IF(2, 0x8b, KB_RALT)
+		CODE_TO_STR_IF(2, 0x84, KB_MENU)
+		CODE_TO_STR_IF(2, 0x8d, KB_RCTL)
+
+		CODE_TO_STR_IF(2, 0xc3, KB_PRNTSCN)
+		CODE_TO_STR_IF(2, 0x7e, KB_SCRLCK)
+		CODE_TO_STR_IF(2, 0x91, KB_PAUSE)
+		CODE_TO_STR_IF(2, 0xc2, KB_INS)
+		CODE_TO_STR_IF(2, 0x94, KB_HOME)
+		CODE_TO_STR_IF(2, 0x96, KB_PGUP)
+		CODE_TO_STR_IF(2, 0xc0, KB_DEL)
+		CODE_TO_STR_IF(2, 0x95, KB_END)
+		CODE_TO_STR_IF(2, 0x97, KB_PGDWN)
+
+		CODE_TO_STR_IF(2, 0x99, KB_UP_ARROW)
+		CODE_TO_STR_IF(2, 0x98, KB_DOWN_ARROW)
+		CODE_TO_STR_IF(2, 0x91, KB_LEFT_ARROW)
+		CODE_TO_STR_IF(2, 0x9b, KB_RIGHT_ARROW)
+
+		CODE_TO_STR_IF(2, 0x77, NUMPAD_LOCK)
+		CODE_TO_STR_IF(2, 0x90, NUMPAD_FWDSLASH)
+		CODE_TO_STR_IF(2, 0x7c, NUMPAD_ASTERISK)
+		CODE_TO_STR_IF(2, 0x7b, NUMPAD_HYPHEN)
+		CODE_TO_STR_IF(2, 0x70, NUMPAD_0)
+		CODE_TO_STR_IF(2, 0x69, NUMPAD_1)
+		CODE_TO_STR_IF(2, 0x72, NUMPAD_2)
+		CODE_TO_STR_IF(2, 0x7a, NUMPAD_3)
+		CODE_TO_STR_IF(2, 0x6b, NUMPAD_4)
+		CODE_TO_STR_IF(2, 0x73, NUMPAD_5)
+		CODE_TO_STR_IF(2, 0x74, NUMPAD_6)
+		CODE_TO_STR_IF(2, 0x6c, NUMPAD_7)
+		CODE_TO_STR_IF(2, 0x75, NUMPAD_8)
+		CODE_TO_STR_IF(2, 0x7d, NUMPAD_9)
+		CODE_TO_STR_IF(2, 0x79, NUMPAD_PLUS)
+		CODE_TO_STR_IF(2, 0x81, NUMPAD_ENTER)
+		CODE_TO_STR_IF(2, 0x71, NUMPAD_PERIOD)
+	}
+
+	if (btn_block[0] == 0x03) {
+		CODE_TO_STR_IF(4, 0x01, RAT_LCLICK)
+		CODE_TO_STR_IF(4, 0x02, RAT_RCLICK)
+		CODE_TO_STR_IF(4, 0x03, RAT_MCLICK)
+		CODE_TO_STR_IF(4, 0x04, RAT_WHEEL_UP)
+		CODE_TO_STR_IF(4, 0x05, RAT_WHEEL_DOWN)
+	}
+
+	if (btn_block[0] == 0x05) {
+		CODE_TO_STR_IF(3, 0x16, MEDIA_SCREENSHOT)
+		CODE_TO_STR_IF(3, 0x19, MEDIA_SHOW_KEYBOARD)
+		CODE_TO_STR_IF(3, 0x1c, MEDIA_SHOW_DESKTOP)
+		CODE_TO_STR_IF(3, 0x1e, MEDIA_START_RECORDING)
+		CODE_TO_STR_IF(3, 0x01, MEDIA_MIC_OFF)
+		CODE_TO_STR_IF(3, 0x02, MEDIA_VOL_DOWN)
+		CODE_TO_STR_IF(3, 0x03, MEDIA_VOL_UP)
+	}
+
+	return "";
+}
+
+/* ASUS ROG Ally device specific attributes */
+
+/* This should be called before any attempts to set device functions */
+static int __gamepad_check_ready(struct hid_device *hdev)
+{
+	u8 *hidbuf;
+	int ret, count;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	for (count = 0; count < 3; count++) {
+		hidbuf[0] = FEATURE_KBD_REPORT_ID;
+		hidbuf[1] = 0xD1;
+		hidbuf[2] = xpad_cmd_check_ready;
+		hidbuf[3] = 01;
+		ret = asus_kbd_set_report(hdev, hidbuf,
+					  FEATURE_ROG_ALLY_REPORT_SIZE);
+		if (ret < 0)
+			hid_warn(hdev, "ROG Ally check failed set report: %d\n", ret);
+
+		hidbuf[0] = hidbuf[1] = hidbuf[2] = hidbuf[3] = 0;
+		ret = asus_kbd_get_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+		if (ret < 0)
+			hid_warn(hdev, "ROG Ally check failed get report: %d\n", ret);
+
+		ret = hidbuf[2] == xpad_cmd_check_ready;
+		if (!ret)
+			hid_warn(hdev, "ROG Ally not ready, retry %d\n", count);
+		else
+			break;
+	}
+
+	if (count == 3)
+		hid_err(hdev, "ROG Ally never responded with a ready\n");
+
+	kfree(hidbuf);
+	return ret;
+}
+
+/********** BUTTON REMAPPING *********************************************************************/
+static void __btn_pair_to_pkt(struct device *raw_dev, enum btn_pair pair, u8 *out, int out_len)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+
+	out[0] = FEATURE_KBD_REPORT_ID;
+	out[1] = 0xD1;
+	out[2] = xpad_cmd_set_mapping;
+	out[3] = pair;
+	out[4] = 0x2c; //length
+	memcpy(&out[5], &rog_ally->key_mapping[rog_ally->mode - 1][pair - 1], MAPPING_BLOCK_LEN);
+}
+
+/* Store the button setting in driver data. Does not apply to device until __gamepad_set_mapping */
+static int __gamepad_mapping_store(struct device *raw_dev, const char *buf, enum btn_pair pair,
+				   int side, bool secondary)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	u8 *key_code;
+	int offs;
+
+	offs = side ? MAPPING_BLOCK_LEN / 2 : 0;
+	offs = secondary ? offs + BTN_CODE_LEN : offs;
+	key_code = rog_ally->key_mapping[rog_ally->mode - 1][pair - 1] + offs;
+
+	return __string_to_key_code(buf, key_code, BTN_CODE_LEN);
+}
+
+/* Apply the mapping pair to the device */
+static int __gamepad_set_mapping(struct device *raw_dev, enum btn_pair pair)
+{
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	__btn_pair_to_pkt(raw_dev, pair, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	kfree(hidbuf);
+
+	return ret;
+}
+
+static ssize_t btn_mapping_apply_store(struct device *raw_dev, struct device_attribute *attr,
+				       const char *buf, size_t count)
+{
+	int ret = __gamepad_write_all_to_mcu(raw_dev);
+	if (ret < 0)
+		return ret;
+	return count;
+}
+ALLY_DEVICE_ATTR_WO(btn_mapping_apply, apply);
+
+/********** BUTTON TURBO *************************************************************************/
+static int __gamepad_turbo_index(enum btn_pair pair, int side)
+{
+	return (pair - 1) * (2 * TURBO_BLOCK_STEP) + (side * TURBO_BLOCK_STEP);
+};
+
+static int __gamepad_turbo_show(struct device *raw_dev, enum btn_pair pair, int side)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	return rog_ally->turbo_btns[rog_ally->mode - 1][__gamepad_turbo_index(pair, side)];
+};
+
+static int __gamepad_turbo_store(struct device *raw_dev, const char *buf, enum btn_pair pair,
+				 int side)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int ret, val;
+
+	ret = kstrtoint(buf, 0, &val);
+	if (ret)
+		return ret;
+	if (val < 0 || val > 16)
+		return -EINVAL;
+
+	rog_ally->turbo_btns[rog_ally->mode - 1][__gamepad_turbo_index(pair, side)] = val;
+
+	return 0;
+};
+
+/* button map attributes, regular and macro*/
+ALLY_BTN_MAPPING(m2, btn_pair_m1_m2, btn_pair_side_left);
+ALLY_BTN_MAPPING(m1, btn_pair_m1_m2, btn_pair_side_right);
+ALLY_BTN_MAPPING(a, btn_pair_a_b, btn_pair_side_left);
+ALLY_BTN_MAPPING(b, btn_pair_a_b, btn_pair_side_right);
+ALLY_BTN_MAPPING(x, btn_pair_x_y, btn_pair_side_left);
+ALLY_BTN_MAPPING(y, btn_pair_x_y, btn_pair_side_right);
+ALLY_BTN_MAPPING(lb, btn_pair_lb_rb, btn_pair_side_left);
+ALLY_BTN_MAPPING(rb, btn_pair_lb_rb, btn_pair_side_right);
+ALLY_BTN_MAPPING(ls, btn_pair_ls_rs, btn_pair_side_left);
+ALLY_BTN_MAPPING(rs, btn_pair_ls_rs, btn_pair_side_right);
+ALLY_BTN_MAPPING(lt, btn_pair_lt_rt, btn_pair_side_left);
+ALLY_BTN_MAPPING(rt, btn_pair_lt_rt, btn_pair_side_right);
+ALLY_BTN_MAPPING(dpad_u, btn_pair_dpad_u_d, btn_pair_side_left);
+ALLY_BTN_MAPPING(dpad_d, btn_pair_dpad_u_d, btn_pair_side_right);
+ALLY_BTN_MAPPING(dpad_l, btn_pair_dpad_l_r, btn_pair_side_left);
+ALLY_BTN_MAPPING(dpad_r, btn_pair_dpad_l_r, btn_pair_side_right);
+ALLY_BTN_MAPPING(view, btn_pair_view_menu, btn_pair_side_left);
+ALLY_BTN_MAPPING(menu, btn_pair_view_menu, btn_pair_side_right);
+
+static void __gamepad_mapping_xpad_default(struct asus_rog_ally *rog_ally)
+{
+	memcpy(&rog_ally->key_mapping[0][0], &XPAD_DEF1, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][1], &XPAD_DEF2, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][2], &XPAD_DEF3, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][3], &XPAD_DEF4, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][4], &XPAD_DEF5, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][5], &XPAD_DEF6, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][6], &XPAD_DEF7, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][7], &XPAD_DEF8, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[0][8], &XPAD_DEF9, MAPPING_BLOCK_LEN);
+}
+
+static void __gamepad_mapping_wasd_default(struct asus_rog_ally *rog_ally)
+{
+	memcpy(&rog_ally->key_mapping[1][0], &WASD_DEF1, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][1], &WASD_DEF2, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][2], &WASD_DEF3, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][3], &WASD_DEF4, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][4], &WASD_DEF5, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][5], &WASD_DEF6, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][6], &WASD_DEF7, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][7], &WASD_DEF8, MAPPING_BLOCK_LEN);
+	memcpy(&rog_ally->key_mapping[1][8], &WASD_DEF9, MAPPING_BLOCK_LEN);
+}
+
+static ssize_t btn_mapping_reset_store(struct device *raw_dev, struct device_attribute *attr,
+				       const char *buf, size_t count)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	switch (rog_ally->mode) {
+	case xpad_mode_game:
+		__gamepad_mapping_xpad_default(rog_ally);
+		break;
+	case xpad_mode_wasd:
+		__gamepad_mapping_wasd_default(rog_ally);
+		break;
+	default:
+		__gamepad_mapping_xpad_default(rog_ally);
+		break;
+	}
+
+	return count;
+}
+
+ALLY_DEVICE_ATTR_WO(btn_mapping_reset, reset_btn_mapping);
+
+/********** GAMEPAD MODE *************************************************************************/
+static ssize_t __gamepad_set_mode(struct device *raw_dev, int val)
+{
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_mode;
+	hidbuf[3] = 0x01;
+	hidbuf[4] = val;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		goto report_fail;
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+	ret = __gamepad_write_all_to_mcu(raw_dev);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t gamepad_mode_show(struct device *raw_dev, struct device_attribute *attr, char *buf)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	return sysfs_emit(buf, "%d\n", rog_ally->mode);
+}
+
+static ssize_t gamepad_mode_store(struct device *raw_dev, struct device_attribute *attr,
+				  const char *buf, size_t count)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int ret, val;
+
+	ret = kstrtoint(buf, 0, &val);
+	if (ret)
+		return ret;
+
+	if (val < xpad_mode_game || val > xpad_mode_mouse)
+		return -EINVAL;
+
+	rog_ally->mode = val;
+
+	ret = __gamepad_set_mode(raw_dev, val);
+	if (ret < 0)
+		return ret;
+
+	return count;
+}
+
+DEVICE_ATTR_RW(gamepad_mode);
+
+/********** VIBRATION INTENSITY ******************************************************************/
+static ssize_t gamepad_vibration_intensity_index_show(struct device *raw_dev,
+						      struct device_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "left right\n");
+}
+
+ALLY_DEVICE_ATTR_RO(gamepad_vibration_intensity_index, vibration_intensity_index);
+
+static ssize_t __gamepad_write_vibe_intensity_to_mcu(struct device *raw_dev)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_vibe_intensity;
+	hidbuf[3] = 0x02; // length
+	hidbuf[4] = rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_left];
+	hidbuf[5] = rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_right];
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		goto report_fail;
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t gamepad_vibration_intensity_show(struct device *raw_dev,
+						struct device_attribute *attr, char *buf)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	return sysfs_emit(buf, "%d %d\n",
+			  rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_left],
+			  rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_right]);
+}
+
+static ssize_t gamepad_vibration_intensity_store(struct device *raw_dev,
+						 struct device_attribute *attr, const char *buf,
+						 size_t count)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	u32 left, right;
+	int ret;
+
+	if (sscanf(buf, "%d %d", &left, &right) != 2)
+		return -EINVAL;
+
+	if (left > 64 || right > 64)
+		return -EINVAL;
+
+	rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_left] = left;
+	rog_ally->vibration_intensity[rog_ally->mode - 1][btn_pair_side_right] = right;
+
+	ret = __gamepad_write_vibe_intensity_to_mcu(raw_dev);
+	if (ret < 0)
+		return ret;
+
+	return count;
+}
+
+ALLY_DEVICE_ATTR_RW(gamepad_vibration_intensity, vibration_intensity);
+
+/********** ROOT LEVEL ATTRS **********************************************************************/
+static struct attribute *gamepad_device_attrs[] = { &dev_attr_gamepad_mode.attr,
+						    &dev_attr_btn_mapping_reset.attr,
+						    &dev_attr_btn_mapping_apply.attr,
+						    &dev_attr_gamepad_vibration_intensity.attr,
+						    &dev_attr_gamepad_vibration_intensity_index.attr,
+						    NULL };
+
+static const struct attribute_group ally_controller_attr_group = {
+	.attrs = gamepad_device_attrs,
+};
+
+/********** ANALOGUE DEADZONES ********************************************************************/
+static ssize_t __gamepad_set_deadzones(struct device *raw_dev)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_js_dz;
+	hidbuf[3] = 0x04; // length
+	hidbuf[4] = rog_ally->deadzones[rog_ally->mode - 1][0][0];
+	hidbuf[5] = rog_ally->deadzones[rog_ally->mode - 1][0][1];
+	hidbuf[6] = rog_ally->deadzones[rog_ally->mode - 1][0][2];
+	hidbuf[7] = rog_ally->deadzones[rog_ally->mode - 1][0][3];
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto end;
+
+	hidbuf[2] = xpad_cmd_set_tr_dz;
+	hidbuf[4] = rog_ally->deadzones[rog_ally->mode - 1][1][0];
+	hidbuf[5] = rog_ally->deadzones[rog_ally->mode - 1][1][1];
+	hidbuf[6] = rog_ally->deadzones[rog_ally->mode - 1][1][2];
+	hidbuf[7] = rog_ally->deadzones[rog_ally->mode - 1][1][3];
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto end;
+
+end:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t __gamepad_store_deadzones(struct device *raw_dev, enum xpad_axis axis,
+					 const char *buf)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int cmd, side, is_tr;
+	u32 inner, outer;
+
+	if (sscanf(buf, "%d %d", &inner, &outer) != 2)
+		return -EINVAL;
+
+	if (inner > 64 || outer > 64 || inner > outer)
+		return -EINVAL;
+
+	is_tr = axis > xpad_axis_xy_right;
+	side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 2 : 0;
+	cmd = is_tr ? xpad_cmd_set_js_dz : xpad_cmd_set_tr_dz;
+
+	rog_ally->deadzones[rog_ally->mode - 1][is_tr][side] = inner;
+	rog_ally->deadzones[rog_ally->mode - 1][is_tr][side + 1] = outer;
+
+	return 0;
+}
+
+static ssize_t axis_xyz_deadzone_index_show(struct device *raw_dev, struct device_attribute *attr,
+					    char *buf)
+{
+	return sysfs_emit(buf, "inner outer\n");
+}
+
+ALLY_DEVICE_ATTR_RO(axis_xyz_deadzone_index, deadzone_index);
+
+ALLY_AXIS_DEADZONE(xpad_axis_xy_left, deadzone);
+ALLY_AXIS_DEADZONE(xpad_axis_xy_right, deadzone);
+ALLY_AXIS_DEADZONE(xpad_axis_z_left, deadzone);
+ALLY_AXIS_DEADZONE(xpad_axis_z_right, deadzone);
+
+/********** ANTI-DEADZONES ***********************************************************************/
+static ssize_t __gamepad_write_js_ADZ_to_mcu(struct device *raw_dev)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_adz;
+	hidbuf[3] = 0x02; // length
+	hidbuf[4] = rog_ally->anti_deadzones[rog_ally->mode - 1][btn_pair_side_left];
+	hidbuf[5] = rog_ally->anti_deadzones[rog_ally->mode - 1][btn_pair_side_right];
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		goto report_fail;
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t __gamepad_js_ADZ_store(struct device *raw_dev, const char *buf,
+				      enum btn_pair_side side)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int ret, val;
+
+	ret = kstrtoint(buf, 0, &val);
+	if (ret)
+		return ret;
+
+	if (val < 0 || val > 32)
+		return -EINVAL;
+
+	rog_ally->anti_deadzones[rog_ally->mode - 1][side] = val;
+
+	return ret;
+}
+
+static ssize_t xpad_axis_xy_left_ADZ_show(struct device *raw_dev, struct device_attribute *attr,
+					  char *buf)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	return sysfs_emit(buf, "%d\n",
+			  rog_ally->anti_deadzones[rog_ally->mode - 1][btn_pair_side_left]);
+}
+
+static ssize_t xpad_axis_xy_left_ADZ_store(struct device *raw_dev, struct device_attribute *attr,
+					   const char *buf, size_t count)
+{
+	int ret = __gamepad_js_ADZ_store(raw_dev, buf, btn_pair_side_left);
+	if (ret)
+		return ret;
+
+	return count;
+}
+
+ALLY_DEVICE_ATTR_RW(xpad_axis_xy_left_ADZ, anti_deadzone);
+
+static ssize_t xpad_axis_xy_right_ADZ_show(struct device *raw_dev, struct device_attribute *attr,
+					   char *buf)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	return sysfs_emit(buf, "%d\n",
+			  rog_ally->anti_deadzones[rog_ally->mode - 1][btn_pair_side_right]);
+}
+
+static ssize_t xpad_axis_xy_right_ADZ_store(struct device *raw_dev, struct device_attribute *attr,
+					    const char *buf, size_t count)
+{
+	int ret = __gamepad_js_ADZ_store(raw_dev, buf, btn_pair_side_right);
+	if (ret)
+		return ret;
+
+	return count;
+}
+
+ALLY_DEVICE_ATTR_RW(xpad_axis_xy_right_ADZ, anti_deadzone);
+
+/********** JS RESPONSE CURVES *******************************************************************/
+static ssize_t rc_point_index_show(struct device *raw_dev, struct device_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "move response\n");
+}
+
+ALLY_DEVICE_ATTR_RO(rc_point_index, rc_point_index);
+
+static ssize_t __gamepad_write_response_curves_to_mcu(struct device *raw_dev)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_response_curve;
+	hidbuf[3] = 0x09; // length
+	hidbuf[4] = 0x01;
+	memcpy(&hidbuf[5], &rog_ally->response_curve[rog_ally->mode - 1][btn_pair_side_left], 8);
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		goto report_fail;
+
+	hidbuf[4] = 0x02;
+	memcpy(&hidbuf[5], &rog_ally->response_curve[rog_ally->mode - 1][btn_pair_side_right], 8);
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		goto report_fail;
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t __gamepad_store_response_curve(struct device *raw_dev, const char *buf,
+					      enum btn_pair_side side, int point)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int idx = (point - 1) * 2;
+	u32 move, response;
+
+	if (sscanf(buf, "%d %d", &move, &response) != 2)
+		return -EINVAL;
+
+	if (move > 64 || response > 64)
+		return -EINVAL;
+
+	rog_ally->response_curve[rog_ally->mode - 1][side][idx] = move;
+	rog_ally->response_curve[rog_ally->mode - 1][side][idx + 1] = response;
+
+	return 0;
+}
+
+ALLY_JS_RC_POINT(left, 1, rc_point_);
+ALLY_JS_RC_POINT(left, 2, rc_point_);
+ALLY_JS_RC_POINT(left, 3, rc_point_);
+ALLY_JS_RC_POINT(left, 4, rc_point_);
+
+ALLY_JS_RC_POINT(right, 1, rc_point_);
+ALLY_JS_RC_POINT(right, 2, rc_point_);
+ALLY_JS_RC_POINT(right, 3, rc_point_);
+ALLY_JS_RC_POINT(right, 4, rc_point_);
+
+/********** CALIBRATIONS *************************************************************************/
+static ssize_t __gamepad_write_cal_to_mcu(struct device *raw_dev, enum xpad_axis axis)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	u8 *c, side, pkt_len, data_len;
+	int ret, cal, checksum = 0;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 1 : 0;
+	pkt_len = axis > xpad_axis_xy_right ? 0x06 : 0x0E;
+	data_len = axis > xpad_axis_xy_right ? 2 : 6;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_calibration;
+	hidbuf[3] = pkt_len;
+	hidbuf[4] = 0x01; // second command (set)
+	hidbuf[5] = axis;
+	c = &hidbuf[6]; // pointer
+
+	for (size_t i = 0; i < data_len; i++) {
+		cal = rog_ally->js_calibrations[side][i];
+		*c = (u8)((cal & 0xff00) >> 8);
+		checksum += *c;
+		c += 1;
+		*c = (u8)(cal & 0xff);
+		checksum += *c;
+		c += 1;
+	}
+
+	hidbuf[6 + data_len * 2] = checksum;
+
+	// TODO: debug if
+	printk("CAL: ");
+	for (size_t i = 0; i < 19; i++) {
+		printk(KERN_CONT "%02x,", hidbuf[i]);
+	}
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+	memset(hidbuf, 0, FEATURE_ROG_ALLY_REPORT_SIZE);
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_calibration;
+	hidbuf[3] = 0x01; // pkt len
+	hidbuf[4] = 0x03; // second command (set)
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+static ssize_t __gamepad_cal_store(struct device *raw_dev, const char *buf, enum xpad_axis axis)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	u32 x_stable, x_min, x_max, y_stable, y_min, y_max, side;
+
+	if (axis == xpad_axis_xy_left || axis == xpad_axis_xy_right) {
+		if (sscanf(buf, "%d %d %d %d %d %d", &x_stable, &x_min, &x_max, &y_stable, &y_min,
+			   &y_max) != 6)
+			return -EINVAL;
+		//TODO: validate input
+
+		side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 1 : 0;
+		/* stored in reverse order for easy copy to packet */
+		rog_ally->js_calibrations[side][0] = y_stable;
+		rog_ally->js_calibrations[side][1] = y_min;
+		rog_ally->js_calibrations[side][2] = y_max;
+		rog_ally->js_calibrations[side][3] = x_stable;
+		rog_ally->js_calibrations[side][4] = x_min;
+		rog_ally->js_calibrations[side][5] = x_max;
+
+		return __gamepad_write_cal_to_mcu(raw_dev, axis);
+	} else {
+		if (sscanf(buf, "%d %d", &x_stable, &x_max) != 2)
+			return -EINVAL;
+		//TODO: validate input
+
+		side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 1 : 0;
+		/* stored in reverse order for easy copy to packet */
+		rog_ally->tr_calibrations[side][0] = x_stable;
+		rog_ally->tr_calibrations[side][1] = x_max;
+
+		return __gamepad_write_cal_to_mcu(raw_dev, axis);
+	}
+}
+
+static ssize_t __gamepad_cal_show(struct device *raw_dev, char *buf, enum xpad_axis axis)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	int side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 1 : 0;
+
+	if (axis == xpad_axis_xy_left || axis == xpad_axis_xy_right) {
+		return sysfs_emit(buf, "%d %d %d %d %d %d\n", rog_ally->js_calibrations[side][3],
+				  rog_ally->js_calibrations[side][4],
+				  rog_ally->js_calibrations[side][5],
+				  rog_ally->js_calibrations[side][0],
+				  rog_ally->js_calibrations[side][1],
+				  rog_ally->js_calibrations[side][2]);
+	} else {
+		return sysfs_emit(buf, "%d %d\n", rog_ally->tr_calibrations[side][0],
+				  rog_ally->tr_calibrations[side][1]);
+	}
+}
+
+ALLY_CAL_ATTR(xpad_axis_xy_left_cal, xpad_axis_xy_left, calibration);
+ALLY_CAL_ATTR(xpad_axis_xy_right_cal, xpad_axis_xy_right, calibration);
+ALLY_CAL_ATTR(xpad_axis_z_left_cal, xpad_axis_z_left, calibration);
+ALLY_CAL_ATTR(xpad_axis_z_right_cal, xpad_axis_z_right, calibration);
+
+static ssize_t xpad_axis_xy_cal_index_show(struct device *raw_dev, struct device_attribute *attr,
+					   char *buf)
+{
+	return sysfs_emit(buf, "x_stable x_min x_max y_stable y_min y_max\n");
+}
+
+ALLY_DEVICE_ATTR_RO(xpad_axis_xy_cal_index, calibration_index);
+
+static ssize_t xpad_axis_z_cal_index_show(struct device *raw_dev, struct device_attribute *attr,
+					  char *buf)
+{
+	return sysfs_emit(buf, "z_stable z_max\n");
+}
+
+ALLY_DEVICE_ATTR_RO(xpad_axis_z_cal_index, calibration_index);
+
+static ssize_t __gamepad_cal_reset(struct device *raw_dev, const char *buf, enum xpad_axis axis)
+{
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	u8 side;
+	int ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+
+	side = axis == xpad_axis_xy_right || axis == xpad_axis_z_right ? 1 : 0;
+
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_calibration;
+	hidbuf[3] = 0x02; // pkt len
+	hidbuf[4] = 0x02; // second command (reset)
+	hidbuf[5] = axis;
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+	memset(hidbuf, 0, FEATURE_ROG_ALLY_REPORT_SIZE);
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_calibration;
+	hidbuf[3] = 0x01; // pkt len
+	hidbuf[4] = 0x03; // second command (set)
+
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+	if (ret < 0)
+		goto report_fail;
+
+report_fail:
+	kfree(hidbuf);
+	return ret;
+}
+
+ALLY_CAL_RESET_ATTR(xpad_axis_xy_left_cal_reset, xpad_axis_xy_left, calibration_reset);
+ALLY_CAL_RESET_ATTR(xpad_axis_xy_right_cal_reset, xpad_axis_xy_right, calibration_reset);
+ALLY_CAL_RESET_ATTR(xpad_axis_z_left_cal_reset, xpad_axis_z_left, calibration_reset);
+ALLY_CAL_RESET_ATTR(xpad_axis_z_right_cal_reset, xpad_axis_z_right, calibration_reset);
+
+static struct attribute *gamepad_axis_xy_left_attrs[] = { &dev_attr_xpad_axis_xy_left_deadzone.attr,
+							  &dev_attr_axis_xyz_deadzone_index.attr,
+							  &dev_attr_xpad_axis_xy_left_ADZ.attr,
+							  &dev_attr_xpad_axis_xy_left_cal_reset.attr,
+							  &dev_attr_xpad_axis_xy_left_cal.attr,
+							  &dev_attr_xpad_axis_xy_cal_index.attr,
+							  &dev_attr_rc_point_left_1.attr,
+							  &dev_attr_rc_point_left_2.attr,
+							  &dev_attr_rc_point_left_3.attr,
+							  &dev_attr_rc_point_left_4.attr,
+							  &dev_attr_rc_point_index.attr,
+							  NULL };
+static const struct attribute_group ally_controller_axis_xy_left_attr_group = {
+	.name = "axis_xy_left",
+	.attrs = gamepad_axis_xy_left_attrs,
+};
+
+static struct attribute *gamepad_axis_xy_right_attrs[] = {
+	&dev_attr_xpad_axis_xy_right_deadzone.attr,
+	&dev_attr_axis_xyz_deadzone_index.attr,
+	&dev_attr_xpad_axis_xy_right_ADZ.attr,
+	&dev_attr_xpad_axis_xy_right_cal_reset.attr,
+	&dev_attr_xpad_axis_xy_right_cal.attr,
+	&dev_attr_xpad_axis_xy_cal_index.attr,
+	&dev_attr_rc_point_right_1.attr,
+	&dev_attr_rc_point_right_2.attr,
+	&dev_attr_rc_point_right_3.attr,
+	&dev_attr_rc_point_right_4.attr,
+	&dev_attr_rc_point_index.attr,
+	NULL
+};
+static const struct attribute_group ally_controller_axis_xy_right_attr_group = {
+	.name = "axis_xy_right",
+	.attrs = gamepad_axis_xy_right_attrs,
+};
+
+static struct attribute *gamepad_axis_z_left_attrs[] = {
+	&dev_attr_xpad_axis_z_left_deadzone.attr,  &dev_attr_axis_xyz_deadzone_index.attr,
+	&dev_attr_xpad_axis_z_left_cal.attr,	   &dev_attr_xpad_axis_z_cal_index.attr,
+	&dev_attr_xpad_axis_z_left_cal_reset.attr, NULL
+};
+static const struct attribute_group ally_controller_axis_z_left_attr_group = {
+	.name = "axis_z_left",
+	.attrs = gamepad_axis_z_left_attrs,
+};
+
+static struct attribute *gamepad_axis_z_right_attrs[] = {
+	&dev_attr_xpad_axis_z_right_deadzone.attr,  &dev_attr_axis_xyz_deadzone_index.attr,
+	&dev_attr_xpad_axis_z_right_cal.attr,	    &dev_attr_xpad_axis_z_cal_index.attr,
+	&dev_attr_xpad_axis_z_right_cal_reset.attr, NULL
+};
+static const struct attribute_group ally_controller_axis_z_right_attr_group = {
+	.name = "axis_z_right",
+	.attrs = gamepad_axis_z_right_attrs,
+};
+
+static const struct attribute_group *gamepad_device_attr_groups[] = {
+	&ally_controller_attr_group,
+	&ally_controller_axis_xy_left_attr_group,
+	&ally_controller_axis_xy_right_attr_group,
+	&ally_controller_axis_z_left_attr_group,
+	&ally_controller_axis_z_right_attr_group,
+	&btn_mapping_m1_attr_group,
+	&btn_mapping_m2_attr_group,
+	&btn_mapping_a_attr_group,
+	&btn_mapping_b_attr_group,
+	&btn_mapping_x_attr_group,
+	&btn_mapping_y_attr_group,
+	&btn_mapping_lb_attr_group,
+	&btn_mapping_rb_attr_group,
+	&btn_mapping_ls_attr_group,
+	&btn_mapping_rs_attr_group,
+	&btn_mapping_dpad_u_attr_group,
+	&btn_mapping_dpad_d_attr_group,
+	&btn_mapping_dpad_l_attr_group,
+	&btn_mapping_dpad_r_attr_group,
+	&btn_mapping_view_attr_group,
+	&btn_mapping_menu_attr_group,
+	NULL
+};
+
+static int __gamepad_write_all_to_mcu(struct device *raw_dev)
+{
+	struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);
+	struct hid_device *hdev = to_hid_device(raw_dev);
+	u8 *hidbuf;
+	int ret = 0;
+
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_dpad_u_d);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_dpad_l_r);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_ls_rs);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_lb_rb);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_a_b);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_x_y);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_view_menu);
+	if (ret < 0)
+		return ret;
+	ret = __gamepad_set_mapping(&hdev->dev, btn_pair_m1_m2);
+	if (ret < 0)
+		return ret;
+	__gamepad_set_mapping(&hdev->dev, btn_pair_lt_rt);
+	if (ret < 0)
+		return ret;
+	__gamepad_set_deadzones(raw_dev);
+	if (ret < 0)
+		return ret;
+	__gamepad_write_js_ADZ_to_mcu(raw_dev);
+	if (ret < 0)
+		return ret;
+	__gamepad_write_vibe_intensity_to_mcu(raw_dev);
+	if (ret < 0)
+		return ret;
+	__gamepad_write_response_curves_to_mcu(raw_dev);
+	if (ret < 0)
+		return ret;
+
+	ret = __gamepad_check_ready(hdev);
+	if (ret < 0)
+		return ret;
+
+	/* set turbo */
+	hidbuf = kzalloc(FEATURE_ROG_ALLY_REPORT_SIZE, GFP_KERNEL);
+	if (!hidbuf)
+		return -ENOMEM;
+	hidbuf[0] = FEATURE_KBD_REPORT_ID;
+	hidbuf[1] = 0xD1;
+	hidbuf[2] = xpad_cmd_set_turbo;
+	hidbuf[3] = 0x20; // length
+	memcpy(&hidbuf[4], rog_ally->turbo_btns[rog_ally->mode - 1], TURBO_BLOCK_LEN);
+	ret = asus_kbd_set_report(hdev, hidbuf, FEATURE_ROG_ALLY_REPORT_SIZE);
+
+	kfree(hidbuf);
+	return ret;
+}
+
+static int asus_rog_ally_probe(struct hid_device *hdev, const struct rog_ops *ops)
+{
+	struct asus_drvdata *drvdata = hid_get_drvdata(hdev);
+	int ret = 0;
+
+	/* all ROG devices have this HID interface but we will focus on Ally for now */
+	if (drvdata->quirks & QUIRK_ROG_NKEY_KEYBOARD && hid_is_usb(hdev)) {
+		struct usb_interface *intf = to_usb_interface(hdev->dev.parent);
+
+		if (intf->altsetting->desc.bInterfaceNumber == 0) {
+			hid_info(hdev, "Setting up ROG USB interface\n");
+			/* initialise and set up USB, common to ROG */
+			// TODO:
+
+			/* initialise the Ally data */
+			if (drvdata->quirks & QUIRK_ROG_ALLY_XPAD) {
+				hid_info(hdev, "Setting up ROG Ally interface\n");
+
+				drvdata->rog_ally_data = devm_kzalloc(
+					&hdev->dev, sizeof(*drvdata->rog_ally_data), GFP_KERNEL);
+				if (!drvdata->rog_ally_data) {
+					hid_err(hdev, "Can't alloc Asus ROG USB interface\n");
+					ret = -ENOMEM;
+					goto err_stop_hw;
+				}
+				// TODO: move these to functions
+				drvdata->rog_ally_data->mode = xpad_mode_game;
+				for (int i = 0; i < xpad_mode_mouse; i++) {
+					drvdata->rog_ally_data->deadzones[i][0][1] = 64;
+					drvdata->rog_ally_data->deadzones[i][0][3] = 64;
+					drvdata->rog_ally_data->deadzones[i][1][1] = 64;
+					drvdata->rog_ally_data->deadzones[i][1][3] = 64;
+
+					drvdata->rog_ally_data->response_curve[i][0][0] = 0x14;
+					drvdata->rog_ally_data->response_curve[i][0][1] = 0x14;
+					drvdata->rog_ally_data->response_curve[i][0][2] = 0x28;
+					drvdata->rog_ally_data->response_curve[i][0][3] = 0x28;
+					drvdata->rog_ally_data->response_curve[i][0][4] = 0x3c;
+					drvdata->rog_ally_data->response_curve[i][0][5] = 0x3c;
+					drvdata->rog_ally_data->response_curve[i][0][6] = 0x50;
+					drvdata->rog_ally_data->response_curve[i][0][7] = 0x50;
+
+					drvdata->rog_ally_data->response_curve[i][1][0] = 0x14;
+					drvdata->rog_ally_data->response_curve[i][1][1] = 0x14;
+					drvdata->rog_ally_data->response_curve[i][1][2] = 0x28;
+					drvdata->rog_ally_data->response_curve[i][1][3] = 0x28;
+					drvdata->rog_ally_data->response_curve[i][1][4] = 0x3c;
+					drvdata->rog_ally_data->response_curve[i][1][5] = 0x3c;
+					drvdata->rog_ally_data->response_curve[i][1][6] = 0x50;
+					drvdata->rog_ally_data->response_curve[i][1][7] = 0x50;
+
+					drvdata->rog_ally_data->vibration_intensity[i][0] = 64;
+					drvdata->rog_ally_data->vibration_intensity[i][1] = 64;
+				}
+
+				/* ignore all errors for this as they are related to USB HID I/O */
+				__gamepad_mapping_xpad_default(drvdata->rog_ally_data);
+				__gamepad_mapping_wasd_default(drvdata->rog_ally_data);
+				// these calls will never error so ignore the return
+				__gamepad_mapping_store(&hdev->dev, "kb_f14", btn_pair_m1_m2,
+							btn_pair_side_left, false); // M2
+				__gamepad_mapping_store(&hdev->dev, "kb_f15", btn_pair_m1_m2,
+							btn_pair_side_right, false); // M1
+				__gamepad_set_mapping(&hdev->dev, btn_pair_m1_m2);
+				__gamepad_set_mode(&hdev->dev, xpad_mode_game);
+			}
+
+			if (sysfs_create_groups(&hdev->dev.kobj, gamepad_device_attr_groups))
+				goto err_stop_hw;
+		}
+	}
+
+	return 0;
+err_stop_hw:
+	hid_hw_stop(hdev);
+	return ret;
+}
+
+void asus_rog_ally_remove(struct hid_device *hdev, const struct rog_ops *ops)
+{
+	struct asus_drvdata *drvdata = hid_get_drvdata(hdev);
+	if (drvdata->rog_ally_data) {
+		__gamepad_set_mode(&hdev->dev, xpad_mode_mouse);
+		sysfs_remove_groups(&hdev->dev.kobj, gamepad_device_attr_groups);
+	}
+}
+
+const struct rog_ops rog_ally = {
+	.probe = asus_rog_ally_probe,
+	.remove = asus_rog_ally_remove,
+};
diff --git a/drivers/hid/hid-asus-rog.h b/drivers/hid/hid-asus-rog.h
new file mode 100644
index 000000000000..efad0b041d5d
--- /dev/null
+++ b/drivers/hid/hid-asus-rog.h
@@ -0,0 +1,482 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  HID driver for Asus ROG laptops and Ally
+ *
+ *  Copyright (c) 2023 Luke Jones <luke@ljones.dev>
+ */
+
+/* data that is private to the hid-asus-rog module */
+
+#include <linux/hid.h>
+#include <linux/types.h>
+
+#define BTN_CODE_LEN 11
+#define MAPPING_BLOCK_LEN 44
+
+#define TURBO_BLOCK_LEN 32
+#define TURBO_BLOCK_STEP 2
+
+#define PAD_A "pad_a"
+#define PAD_B "pad_b"
+#define PAD_X "pad_x"
+#define PAD_Y "pad_y"
+#define PAD_LB "pad_lb"
+#define PAD_RB "pad_rb"
+#define PAD_LS "pad_ls"
+#define PAD_RS "pad_rs"
+#define PAD_DPAD_UP "pad_dpad_up"
+#define PAD_DPAD_DOWN "pad_dpad_down"
+#define PAD_DPAD_LEFT "pad_dpad_left"
+#define PAD_DPAD_RIGHT "pad_dpad_right"
+#define PAD_VIEW "pad_view"
+#define PAD_MENU "pad_menu"
+#define PAD_XBOX "pad_xbox"
+
+#define KB_M1 "kb_m1"
+#define KB_M2 "kb_m2"
+#define KB_ESC "kb_esc"
+#define KB_F1 "kb_f1"
+#define KB_F2 "kb_f2"
+#define KB_F3 "kb_f3"
+#define KB_F4 "kb_f4"
+#define KB_F5 "kb_f5"
+#define KB_F6 "kb_f6"
+#define KB_F7 "kb_f7"
+#define KB_F8 "kb_f8"
+#define KB_F9 "kb_f9"
+#define KB_F10 "kb_f10"
+#define KB_F11 "kb_f11"
+#define KB_F12 "kb_f12"
+#define KB_F14 "kb_f14"
+#define KB_F15 "kb_f15"
+
+#define KB_BACKTICK "kb_backtick"
+#define KB_1 "kb_1"
+#define KB_2 "kb_2"
+#define KB_3 "kb_3"
+#define KB_4 "kb_4"
+#define KB_5 "kb_5"
+#define KB_6 "kb_6"
+#define KB_7 "kb_7"
+#define KB_8 "kb_8"
+#define KB_9 "kb_9"
+#define KB_0 "kb_0"
+#define KB_HYPHEN "kb_hyphen"
+#define KB_EQUALS "kb_equals"
+#define KB_BACKSPACE "kb_backspace"
+
+#define KB_TAB "kb_tab"
+#define KB_Q "kb_q"
+#define KB_W "kb_w"
+#define KB_E "kb_e"
+#define KB_R "kb_r"
+#define KB_T "kb_t"
+#define KB_Y "kb_y"
+#define KB_U "kb_u"
+#define KB_I "kb_i"
+#define KB_O "kb_o"
+#define KB_P "kb_p"
+#define KB_LBRACKET "kb_lbracket"
+#define KB_RBRACKET "kb_rbracket"
+#define KB_BACKSLASH "kb_bkslash"
+
+#define KB_CAPS "kb_caps"
+#define KB_A "kb_a"
+#define KB_S "kb_s"
+#define KB_D "kb_d"
+#define KB_F "kb_f"
+#define KB_G "kb_g"
+#define KB_H "kb_h"
+#define KB_J "kb_j"
+#define KB_K "kb_k"
+#define KB_L "kb_l"
+#define KB_SEMI "kb_semicolon"
+#define KB_QUOTE "kb_quote"
+#define KB_RET "kb_enter"
+
+#define KB_LSHIFT "kb_lshift"
+#define KB_Z "kb_z"
+#define KB_X "kb_x"
+#define KB_C "kb_c"
+#define KB_V "kb_v"
+#define KB_B "kb_b"
+#define KB_N "kb_n"
+#define KB_M "kb_m"
+#define KB_COMMA "kb_comma"
+#define KB_PERIOD "kb_period"
+#define KB_FWDSLASH "kb_fwdslash"
+#define KB_RSHIFT "kb_rshift"
+
+#define KB_LCTL "kb_lctl"
+#define KB_META "kb_meta"
+#define KB_LALT "kb_lalt"
+#define KB_SPACE "kb_space"
+#define KB_RALT "kb_ralt"
+#define KB_MENU "kb_menu"
+#define KB_RCTL "kb_rctl"
+
+#define KB_PRNTSCN "kb_prntscn"
+#define KB_SCRLCK "kb_scrlck"
+#define KB_PAUSE "kb_pause"
+#define KB_INS "kb_ins"
+#define KB_HOME "kb_home"
+#define KB_PGUP "kb_pgup"
+#define KB_DEL "kb_del"
+#define KB_END "kb_end"
+#define KB_PGDWN "kb_pgdwn"
+
+#define KB_UP_ARROW "kb_up_arrow"
+#define KB_DOWN_ARROW "kb_down_arrow"
+#define KB_LEFT_ARROW "kb_left_arrow"
+#define KB_RIGHT_ARROW "kb_right_arrow"
+
+#define NUMPAD_LOCK "numpad_lock"
+#define NUMPAD_FWDSLASH "numpad_fwdslash"
+#define NUMPAD_ASTERISK "numpad_asterisk"
+#define NUMPAD_HYPHEN "numpad_hyphen"
+#define NUMPAD_0 "numpad_0"
+#define NUMPAD_1 "numpad_1"
+#define NUMPAD_2 "numpad_2"
+#define NUMPAD_3 "numpad_3"
+#define NUMPAD_4 "numpad_4"
+#define NUMPAD_5 "numpad_5"
+#define NUMPAD_6 "numpad_6"
+#define NUMPAD_7 "numpad_7"
+#define NUMPAD_8 "numpad_8"
+#define NUMPAD_9 "numpad_9"
+#define NUMPAD_PLUS "numpad_plus"
+#define NUMPAD_ENTER "numpad_enter"
+#define NUMPAD_PERIOD "numpad_."
+
+#define RAT_LCLICK "rat_lclick"
+#define RAT_RCLICK "rat_rclick"
+#define RAT_MCLICK "rat_mclick"
+#define RAT_WHEEL_UP "rat_wheel_up"
+#define RAT_WHEEL_DOWN "rat_wheel_down"
+
+#define MEDIA_SCREENSHOT "media_screenshot"
+#define MEDIA_SHOW_KEYBOARD "media_show_keyboard"
+#define MEDIA_SHOW_DESKTOP "media_show_desktop"
+#define MEDIA_START_RECORDING "media_start_recording"
+#define MEDIA_MIC_OFF "media_mic_off"
+#define MEDIA_VOL_DOWN "media_vol_down"
+#define MEDIA_VOL_UP "media_vol_up"
+
+/* required so we can have nested attributes with same name but different functions */
+#define ALLY_DEVICE_ATTR_RW(_name, _sysfs_name)    \
+	struct device_attribute dev_attr_##_name = \
+		__ATTR(_sysfs_name, 0644, _name##_show, _name##_store)
+
+#define ALLY_DEVICE_ATTR_RO(_name, _sysfs_name) \
+	struct device_attribute dev_attr_##_name = __ATTR(_sysfs_name, 0444, _name##_show, NULL)
+
+#define ALLY_DEVICE_ATTR_WO(_name, _sysfs_name) \
+	struct device_attribute dev_attr_##_name = __ATTR(_sysfs_name, 0200, NULL, _name##_store)
+
+/* response curve macros */
+#define ALLY_RESP_CURVE_SHOW(_name, _point_n)                                                    \
+	static ssize_t _name##_show(struct device *raw_dev, struct device_attribute *attr,       \
+				    char *buf)                                                   \
+	{                                                                                        \
+		struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);                       \
+		int idx = (_point_n - 1) * 2;                                                    \
+		return sysfs_emit(                                                               \
+			buf, "%d %d\n",                                                          \
+			rog_ally->response_curve[rog_ally->mode][btn_pair_side_left][idx],       \
+			rog_ally->response_curve[rog_ally->mode][btn_pair_side_right][idx + 1]); \
+	}
+
+#define ALLY_RESP_CURVE_STORE(_name, _side, _point_n)                                         \
+	static ssize_t _name##_store(struct device *raw_dev, struct device_attribute *attr,   \
+				     const char *buf, size_t count)                           \
+	{                                                                                     \
+		int ret = __gamepad_store_response_curve(raw_dev, buf, btn_pair_side_##_side, \
+							 _point_n);                           \
+		if (ret < 0)                                                                  \
+			return ret;                                                           \
+		return count;                                                                 \
+	}
+
+/* _point_n must start at 1 */
+#define ALLY_JS_RC_POINT(_side, _point_n, _sysfs_label)                        \
+	ALLY_RESP_CURVE_SHOW(rc_point_##_side##_##_point_n, _point_n);         \
+	ALLY_RESP_CURVE_STORE(rc_point_##_side##_##_point_n, _side, _point_n); \
+	ALLY_DEVICE_ATTR_RW(rc_point_##_side##_##_point_n, _sysfs_label##_point_n);
+
+/* deadzone macros */
+#define ALLY_AXIS_DEADZONE_SHOW(_axis)                                                    \
+	static ssize_t _axis##_deadzone_show(struct device *raw_dev,                      \
+					     struct device_attribute *attr, char *buf)    \
+	{                                                                                 \
+		struct asus_rog_ally *rog_ally = __rog_ally_data(raw_dev);                \
+		int side, is_tr;                                                          \
+                                                                                          \
+		is_tr = _axis > xpad_axis_xy_right;                                       \
+		side = _axis == xpad_axis_xy_right || _axis == xpad_axis_z_right ? 2 : 0; \
+                                                                                          \
+		return sysfs_emit(buf, "%d %d\n",                                         \
+				  rog_ally->deadzones[rog_ally->mode][is_tr][side],       \
+				  rog_ally->deadzones[rog_ally->mode][is_tr][side + 1]);  \
+	}
+
+#define ALLY_AXIS_DEADZONE_STORE(_axis)                                                       \
+	static ssize_t _axis##_deadzone_store(struct device *raw_dev,                         \
+					      struct device_attribute *attr, const char *buf, \
+					      size_t count)                                   \
+	{                                                                                     \
+		int ret = __gamepad_store_deadzones(raw_dev, _axis, buf);                     \
+		if (ret < 0)                                                                  \
+			return ret;                                                           \
+		return count;                                                                 \
+	}
+
+#define ALLY_AXIS_DEADZONE(_axis, _sysfs_label) \
+	ALLY_AXIS_DEADZONE_SHOW(_axis);         \
+	ALLY_AXIS_DEADZONE_STORE(_axis);        \
+	ALLY_DEVICE_ATTR_RW(_axis##_deadzone, _sysfs_label);
+
+/* button specific macros */
+#define ALLY_BTN_SHOW(_fname, _pair, _side, _secondary)                                     \
+	static ssize_t _fname##_show(struct device *raw_dev, struct device_attribute *attr, \
+				     char *buf)                                             \
+	{                                                                                   \
+		return sysfs_emit(buf, "%s\n",                                              \
+				  __btn_map_to_string(raw_dev, _pair, _side, _secondary));  \
+	}
+
+#define ALLY_BTN_STORE(_fname, _pair, _side, _secondary)                                     \
+	static ssize_t _fname##_store(struct device *raw_dev, struct device_attribute *attr, \
+				      const char *buf, size_t count)                         \
+	{                                                                                    \
+		int ret = __gamepad_mapping_store(raw_dev, buf, _pair, _side, _secondary);   \
+		if (ret < 0)                                                                 \
+			return ret;                                                          \
+		return count;                                                                \
+	}
+
+#define ALLY_BTN_TURBO_SHOW(_fname, _pair, _side)                                                 \
+	static ssize_t _fname##_turbo_show(struct device *raw_dev, struct device_attribute *attr, \
+					   char *buf)                                             \
+	{                                                                                         \
+		return sysfs_emit(buf, "%d\n", __gamepad_turbo_show(raw_dev, _pair, _side));      \
+	}
+
+#define ALLY_BTN_TURBO_STORE(_fname, _pair, _side)                                                 \
+	static ssize_t _fname##_turbo_store(struct device *raw_dev, struct device_attribute *attr, \
+					    const char *buf, size_t count)                         \
+	{                                                                                          \
+		int ret = __gamepad_turbo_store(raw_dev, buf, _pair, _side);                       \
+		if (ret < 0)                                                                       \
+			return ret;                                                                \
+		return count;                                                                      \
+	}
+
+#define ALLY_BTN_ATTRS_GROUP(_name, _fname)                                                    \
+	static struct attribute *_fname##_attrs[] = { &dev_attr_##_fname.attr,                 \
+						      &dev_attr_##_fname##_macro.attr,         \
+						      &dev_attr_##_fname##_turbo.attr, NULL }; \
+	static const struct attribute_group _fname##_attr_group = {                            \
+		.name = __stringify(_name),                                                    \
+		.attrs = _fname##_attrs,                                                       \
+	};
+
+#define ALLY_BTN_MAPPING(_fname, _pair, _side)                            \
+	ALLY_BTN_SHOW(btn_mapping_##_fname, _pair, _side, false);         \
+	ALLY_BTN_STORE(btn_mapping_##_fname, _pair, _side, false);        \
+                                                                          \
+	ALLY_BTN_SHOW(btn_mapping_##_fname##_macro, _pair, _side, true);  \
+	ALLY_BTN_STORE(btn_mapping_##_fname##_macro, _pair, _side, true); \
+                                                                          \
+	ALLY_BTN_TURBO_SHOW(btn_mapping_##_fname, _pair, _side);          \
+	ALLY_BTN_TURBO_STORE(btn_mapping_##_fname, _pair, _side);         \
+                                                                          \
+	ALLY_DEVICE_ATTR_RW(btn_mapping_##_fname, remap);                 \
+	ALLY_DEVICE_ATTR_RW(btn_mapping_##_fname##_macro, macro_remap);   \
+	ALLY_DEVICE_ATTR_RW(btn_mapping_##_fname##_turbo, turbo);         \
+                                                                          \
+	ALLY_BTN_ATTRS_GROUP(btn_##_fname, btn_mapping_##_fname);
+
+/* calibration macros */
+#define ALLY_CAL_STORE(_fname, _axis)                                                  \
+	static ssize_t _fname##_store(struct device *raw_dev, struct device_attribute *attr, \
+				      const char *buf, size_t count)                         \
+	{                                                                                    \
+		int ret = __gamepad_cal_store(raw_dev, buf, _axis);                          \
+		if (ret < 0)                                                                 \
+			return ret;                                                          \
+		return count;                                                                \
+	};
+
+#define ALLY_CAL_SHOW(_fname, _axis)                                     \
+	static ssize_t _fname##_show(struct device *raw_dev, struct device_attribute *attr, \
+				     char *buf)                                             \
+	{                                                                                   \
+		return __gamepad_cal_show(raw_dev, buf, _axis);  \
+	}
+
+#define ALLY_CAL_ATTR(_fname, _axis, _sysfs_label) \
+	ALLY_CAL_STORE(_fname, _axis);             \
+	ALLY_CAL_SHOW(_fname, _axis);             \
+	ALLY_DEVICE_ATTR_RW(_fname, _sysfs_label);
+
+#define ALLY_CAL_RESET_STORE(_fname, _axis)                                                  \
+	static ssize_t _fname##_store(struct device *raw_dev, struct device_attribute *attr, \
+				      const char *buf, size_t count)                         \
+	{                                                                                    \
+		int ret = __gamepad_cal_reset(raw_dev, buf, _axis);                          \
+		if (ret < 0)                                                                 \
+			return ret;                                                          \
+		return count;                                                                \
+	};
+
+#define ALLY_CAL_RESET_ATTR(_fname, _axis, _sysfs_label) \
+	ALLY_CAL_RESET_STORE(_fname, _axis);             \
+	ALLY_DEVICE_ATTR_WO(_fname, _sysfs_label);
+
+/* Default blocks for the xpad mode */
+static const u8 XPAD_DEF1[MAPPING_BLOCK_LEN] = {
+	0x01, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x19,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x0a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x03, 0x8c, 0x88, 0x76, 0x00, 0x00
+};
+static const u8 XPAD_DEF2[MAPPING_BLOCK_LEN] = {
+	0x01, 0x0b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+	0x00, 0x02, 0x82, 0x23, 0x00, 0x00, 0x00, 0x01, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x02, 0x82, 0x0d, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF3[MAPPING_BLOCK_LEN] = {
+	0x01, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF4[MAPPING_BLOCK_LEN] = {
+	0x01, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x06, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF5[MAPPING_BLOCK_LEN] = {
+	0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x16,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x02, 0x82, 0x31, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF6[MAPPING_BLOCK_LEN] = {
+	0x01, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+	0x00, 0x02, 0x82, 0x4d, 0x00, 0x00, 0x00, 0x01, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF7[MAPPING_BLOCK_LEN] = {
+	0x01, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF8[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x8e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x8e, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x8f, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x02, 0x00, 0x8f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 XPAD_DEF9[MAPPING_BLOCK_LEN] = {
+	0x01, 0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x0e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+/* default blocks for the wasd mode */
+static const u8 WASD_DEF1[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x98, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x19,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x99, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x03, 0x8c, 0x88, 0x76, 0x00, 0x00
+};
+static const u8 WASD_DEF2[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x9a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+	0x00, 0x02, 0x82, 0x23, 0x00, 0x00, 0x00, 0x02, 0x00, 0x9b, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x02, 0x82, 0x0d, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF3[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x88, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF4[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF5[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x5a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x16,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x02, 0x82, 0x31, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF6[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x97, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+	0x00, 0x02, 0x82, 0x4d, 0x00, 0x00, 0x00, 0x02, 0x00, 0x96, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF7[MAPPING_BLOCK_LEN] = {
+	0x01, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF8[MAPPING_BLOCK_LEN] = {
+	0x02, 0x00, 0x8e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x8e, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x8f, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x02, 0x00, 0x8f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+static const u8 WASD_DEF9[MAPPING_BLOCK_LEN] = {
+	0x04, 0x00, 0x00, 0x00, 0x00, 0x02, 0x88, 0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+/*
+ * the xpad_mode is used inside the mode setting packet and is used
+ * for indexing (xpad_mode - 1)
+ */
+enum xpad_mode {
+	xpad_mode_game = 0x01,
+	xpad_mode_wasd = 0x02,
+	xpad_mode_mouse = 0x03,
+};
+
+/* the xpad_cmd determines which feature is set or queried */
+enum xpad_cmd {
+	xpad_cmd_set_mode = 0x01,
+	xpad_cmd_set_mapping = 0x02,
+	xpad_cmd_set_js_dz = 0x04, /* deadzones */
+	xpad_cmd_set_tr_dz = 0x05, /* deadzones */
+	xpad_cmd_set_vibe_intensity = 0x06,
+	xpad_cmd_check_ready = 0x0A,
+	xpad_cmd_set_calibration = 0x0D,
+	xpad_cmd_set_turbo = 0x0F,
+	xpad_cmd_set_response_curve = 0x13,
+	xpad_cmd_set_adz = 0x18,
+};
+
+/*
+ * the xpad_mode is used in various set and query HID packets and is
+ * used for indexing (xpad_axis - 1)
+ */
+enum xpad_axis {
+	xpad_axis_xy_left = 0x01,
+	xpad_axis_xy_right = 0x02,
+	xpad_axis_z_left = 0x03,
+	xpad_axis_z_right = 0x04,
+};
+
+enum btn_pair {
+	btn_pair_dpad_u_d = 0x01,
+	btn_pair_dpad_l_r = 0x02,
+	btn_pair_ls_rs = 0x03,
+	btn_pair_lb_rb = 0x04,
+	btn_pair_a_b = 0x05,
+	btn_pair_x_y = 0x06,
+	btn_pair_view_menu = 0x07,
+	btn_pair_m1_m2 = 0x08,
+	btn_pair_lt_rt = 0x09,
+};
+
+enum btn_pair_side {
+	btn_pair_side_left = 0x00,
+	btn_pair_side_right = 0x01,
+};
+
+static int __gamepad_write_all_to_mcu(struct device *raw_dev);
\ No newline at end of file
diff --git a/drivers/hid/hid-asus.h b/drivers/hid/hid-asus.h
new file mode 100644
index 000000000000..18317cad7110
--- /dev/null
+++ b/drivers/hid/hid-asus.h
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  HID driver for Asus ROG laptops and Ally
+ *
+ *  Copyright (c) 2023 Luke Jones <luke@ljones.dev>
+ */
+
+#include <linux/hid.h>
+#include <linux/types.h>
+
+#define FEATURE_KBD_REPORT_ID		0x5a
+#define FEATURE_KBD_REPORT_SIZE		16
+#define FEATURE_KBD_LED_REPORT_ID1	0x5d
+#define FEATURE_KBD_LED_REPORT_ID2	0x5e
+#define FEATURE_ROG_ALLY_REPORT_SIZE	64
+
+#define QUIRK_FIX_NOTEBOOK_REPORT	BIT(0)
+#define QUIRK_NO_INIT_REPORTS		BIT(1)
+#define QUIRK_SKIP_INPUT_MAPPING	BIT(2)
+#define QUIRK_IS_MULTITOUCH		BIT(3)
+#define QUIRK_NO_CONSUMER_USAGES	BIT(4)
+#define QUIRK_USE_KBD_BACKLIGHT		BIT(5)
+#define QUIRK_T100_KEYBOARD		BIT(6)
+#define QUIRK_T100CHI			BIT(7)
+#define QUIRK_G752_KEYBOARD		BIT(8)
+#define QUIRK_T90CHI			BIT(9)
+#define QUIRK_MEDION_E1239T		BIT(10)
+#define QUIRK_ROG_NKEY_KEYBOARD		BIT(11)
+#define QUIRK_ROG_CLAYMORE_II_KEYBOARD	BIT(12)
+#define QUIRK_ROG_ALLY_XPAD		BIT(13)
+
+struct asus_drvdata {
+	unsigned long quirks;
+	struct hid_device *hdev;
+	struct input_dev *input;
+	struct input_dev *tp_kbd_input;
+	struct asus_kbd_leds *kbd_backlight;
+	const struct asus_touchpad_info *tp;
+	bool enable_backlight;
+	struct power_supply *battery;
+	struct power_supply_desc battery_desc;
+	int battery_capacity;
+	int battery_stat;
+	bool battery_in_query;
+	unsigned long battery_next_query;
+	struct asus_rog_ally *rog_ally_data;
+};
+
+extern int asus_kbd_set_report(struct hid_device *hdev, const u8 *buf, size_t buf_size);
+
+extern int asus_kbd_get_report(struct hid_device *hdev, u8 *out_buf, size_t out_buf_size);
+
+struct rog_ops {
+	int (*probe) (struct hid_device *hdev, const struct rog_ops *ops);
+	void (*remove) (struct hid_device *hdev, const struct rog_ops *ops);
+};
+
+extern const struct rog_ops rog_ally;
\ No newline at end of file
diff --git a/drivers/hid/hid-ids.h b/drivers/hid/hid-ids.h
index 828a5c022c64..79ae30212725 100644
--- a/drivers/hid/hid-ids.h
+++ b/drivers/hid/hid-ids.h
@@ -208,6 +208,7 @@
 #define USB_DEVICE_ID_ASUSTEK_ROG_NKEY_KEYBOARD	0x1866
 #define USB_DEVICE_ID_ASUSTEK_ROG_NKEY_KEYBOARD2	0x19b6
 #define USB_DEVICE_ID_ASUSTEK_ROG_NKEY_KEYBOARD3	0x1a30
+#define USB_DEVICE_ID_ASUSTEK_ROG_NKEY_ALLY		0x1abe
 #define USB_DEVICE_ID_ASUSTEK_ROG_CLAYMORE_II_KEYBOARD	0x196b
 #define USB_DEVICE_ID_ASUSTEK_FX503VD_KEYBOARD	0x1869
 
-- 
2.44.0

From a9a585447d5ab5667a7139052240a9abd6eb20b2 Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:07:11 +0200
Subject: [PATCH 7/8] ksm

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 arch/alpha/kernel/syscalls/syscall.tbl      |   3 +
 arch/arm/tools/syscall.tbl                  |   3 +
 arch/arm64/include/asm/unistd.h             |   2 +-
 arch/arm64/include/asm/unistd32.h           |   6 +
 arch/m68k/kernel/syscalls/syscall.tbl       |   3 +
 arch/microblaze/kernel/syscalls/syscall.tbl |   3 +
 arch/mips/kernel/syscalls/syscall_n32.tbl   |   3 +
 arch/mips/kernel/syscalls/syscall_n64.tbl   |   3 +
 arch/mips/kernel/syscalls/syscall_o32.tbl   |   3 +
 arch/parisc/kernel/syscalls/syscall.tbl     |   3 +
 arch/powerpc/kernel/syscalls/syscall.tbl    |   3 +
 arch/s390/kernel/syscalls/syscall.tbl       |   3 +
 arch/sh/kernel/syscalls/syscall.tbl         |   3 +
 arch/sparc/kernel/syscalls/syscall.tbl      |   3 +
 arch/x86/entry/syscalls/syscall_32.tbl      |   3 +
 arch/x86/entry/syscalls/syscall_64.tbl      |   3 +
 arch/xtensa/kernel/syscalls/syscall.tbl     |   3 +
 include/linux/syscalls.h                    |   3 +
 include/uapi/asm-generic/unistd.h           |  11 +-
 kernel/sys.c                                | 147 ++++++++++++++++++++
 kernel/sys_ni.c                             |   3 +
 21 files changed, 215 insertions(+), 2 deletions(-)

diff --git a/arch/alpha/kernel/syscalls/syscall.tbl b/arch/alpha/kernel/syscalls/syscall.tbl
index 8ff110826ce2..ad64d9b5ce62 100644
--- a/arch/alpha/kernel/syscalls/syscall.tbl
+++ b/arch/alpha/kernel/syscalls/syscall.tbl
@@ -501,3 +501,6 @@
 569	common	lsm_get_self_attr		sys_lsm_get_self_attr
 570	common	lsm_set_self_attr		sys_lsm_set_self_attr
 571	common	lsm_list_modules		sys_lsm_list_modules
+572	common	process_ksm_enable		sys_process_ksm_enable
+573	common	process_ksm_disable		sys_process_ksm_disable
+574	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/arm/tools/syscall.tbl b/arch/arm/tools/syscall.tbl
index b6c9e01e14f5..524a0429016a 100644
--- a/arch/arm/tools/syscall.tbl
+++ b/arch/arm/tools/syscall.tbl
@@ -475,3 +475,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/arm64/include/asm/unistd.h b/arch/arm64/include/asm/unistd.h
index 491b2b9bd553..8bdc6380b21d 100644
--- a/arch/arm64/include/asm/unistd.h
+++ b/arch/arm64/include/asm/unistd.h
@@ -39,7 +39,7 @@
 #define __ARM_NR_compat_set_tls		(__ARM_NR_COMPAT_BASE + 5)
 #define __ARM_NR_COMPAT_END		(__ARM_NR_COMPAT_BASE + 0x800)
 
-#define __NR_compat_syscalls		462
+#define __NR_compat_syscalls		465
 #endif
 
 #define __ARCH_WANT_SYS_CLONE
diff --git a/arch/arm64/include/asm/unistd32.h b/arch/arm64/include/asm/unistd32.h
index 7118282d1c79..af409ce2d8af 100644
--- a/arch/arm64/include/asm/unistd32.h
+++ b/arch/arm64/include/asm/unistd32.h
@@ -929,6 +929,12 @@ __SYSCALL(__NR_lsm_get_self_attr, sys_lsm_get_self_attr)
 __SYSCALL(__NR_lsm_set_self_attr, sys_lsm_set_self_attr)
 #define __NR_lsm_list_modules 461
 __SYSCALL(__NR_lsm_list_modules, sys_lsm_list_modules)
+#define __NR_process_ksm_enable 462
+__SYSCALL(__NR_process_ksm_enable, sys_process_ksm_enable)
+#define __NR_process_ksm_disable 463
+__SYSCALL(__NR_process_ksm_disable, sys_process_ksm_disable)
+#define __NR_process_ksm_status 464
+__SYSCALL(__NR_process_ksm_status, sys_process_ksm_status)
 
 /*
  * Please add new compat syscalls above this comment and update
diff --git a/arch/m68k/kernel/syscalls/syscall.tbl b/arch/m68k/kernel/syscalls/syscall.tbl
index 7fd43fd4c9f2..3374d91d8665 100644
--- a/arch/m68k/kernel/syscalls/syscall.tbl
+++ b/arch/m68k/kernel/syscalls/syscall.tbl
@@ -461,3 +461,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/microblaze/kernel/syscalls/syscall.tbl b/arch/microblaze/kernel/syscalls/syscall.tbl
index b00ab2cabab9..f44872df0f9c 100644
--- a/arch/microblaze/kernel/syscalls/syscall.tbl
+++ b/arch/microblaze/kernel/syscalls/syscall.tbl
@@ -467,3 +467,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/mips/kernel/syscalls/syscall_n32.tbl b/arch/mips/kernel/syscalls/syscall_n32.tbl
index 83cfc9eb6b88..7a5a02c29175 100644
--- a/arch/mips/kernel/syscalls/syscall_n32.tbl
+++ b/arch/mips/kernel/syscalls/syscall_n32.tbl
@@ -400,3 +400,6 @@
 459	n32	lsm_get_self_attr		sys_lsm_get_self_attr
 460	n32	lsm_set_self_attr		sys_lsm_set_self_attr
 461	n32	lsm_list_modules		sys_lsm_list_modules
+462	n32	process_ksm_enable		sys_process_ksm_enable
+463	n32	process_ksm_disable		sys_process_ksm_disable
+464	n32	process_ksm_status		sys_process_ksm_status
diff --git a/arch/mips/kernel/syscalls/syscall_n64.tbl b/arch/mips/kernel/syscalls/syscall_n64.tbl
index 532b855df589..bc960ce89fac 100644
--- a/arch/mips/kernel/syscalls/syscall_n64.tbl
+++ b/arch/mips/kernel/syscalls/syscall_n64.tbl
@@ -376,3 +376,6 @@
 459	n64	lsm_get_self_attr		sys_lsm_get_self_attr
 460	n64	lsm_set_self_attr		sys_lsm_set_self_attr
 461	n64	lsm_list_modules		sys_lsm_list_modules
+462	n64	process_ksm_enable		sys_process_ksm_enable
+463	n64	process_ksm_disable		sys_process_ksm_disable
+464	n64	process_ksm_status		sys_process_ksm_status
diff --git a/arch/mips/kernel/syscalls/syscall_o32.tbl b/arch/mips/kernel/syscalls/syscall_o32.tbl
index f45c9530ea93..1179ed8891be 100644
--- a/arch/mips/kernel/syscalls/syscall_o32.tbl
+++ b/arch/mips/kernel/syscalls/syscall_o32.tbl
@@ -449,3 +449,6 @@
 459	o32	lsm_get_self_attr		sys_lsm_get_self_attr
 460	o32	lsm_set_self_attr		sys_lsm_set_self_attr
 461	o32	lsm_list_modules		sys_lsm_list_modules
+462	o32	process_ksm_enable		sys_process_ksm_enable
+463	o32	process_ksm_disable		sys_process_ksm_disable
+464	o32	process_ksm_status		sys_process_ksm_status
diff --git a/arch/parisc/kernel/syscalls/syscall.tbl b/arch/parisc/kernel/syscalls/syscall.tbl
index b236a84c4e12..d3157f90ef49 100644
--- a/arch/parisc/kernel/syscalls/syscall.tbl
+++ b/arch/parisc/kernel/syscalls/syscall.tbl
@@ -460,3 +460,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/powerpc/kernel/syscalls/syscall.tbl b/arch/powerpc/kernel/syscalls/syscall.tbl
index 17173b82ca21..bf7920c3de18 100644
--- a/arch/powerpc/kernel/syscalls/syscall.tbl
+++ b/arch/powerpc/kernel/syscalls/syscall.tbl
@@ -548,3 +548,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/s390/kernel/syscalls/syscall.tbl b/arch/s390/kernel/syscalls/syscall.tbl
index 095bb86339a7..25421b092cbc 100644
--- a/arch/s390/kernel/syscalls/syscall.tbl
+++ b/arch/s390/kernel/syscalls/syscall.tbl
@@ -464,3 +464,6 @@
 459  common	lsm_get_self_attr	sys_lsm_get_self_attr		sys_lsm_get_self_attr
 460  common	lsm_set_self_attr	sys_lsm_set_self_attr		sys_lsm_set_self_attr
 461  common	lsm_list_modules	sys_lsm_list_modules		sys_lsm_list_modules
+462  common	process_ksm_enable	sys_process_ksm_enable		sys_process_ksm_enable
+463  common	process_ksm_disable	sys_process_ksm_disable		sys_process_ksm_disable
+464  common	process_ksm_status	sys_process_ksm_status		sys_process_ksm_status
diff --git a/arch/sh/kernel/syscalls/syscall.tbl b/arch/sh/kernel/syscalls/syscall.tbl
index 86fe269f0220..d6c06877a69a 100644
--- a/arch/sh/kernel/syscalls/syscall.tbl
+++ b/arch/sh/kernel/syscalls/syscall.tbl
@@ -464,3 +464,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/sparc/kernel/syscalls/syscall.tbl b/arch/sparc/kernel/syscalls/syscall.tbl
index b23d59313589..b749293082d9 100644
--- a/arch/sparc/kernel/syscalls/syscall.tbl
+++ b/arch/sparc/kernel/syscalls/syscall.tbl
@@ -507,3 +507,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/arch/x86/entry/syscalls/syscall_32.tbl b/arch/x86/entry/syscalls/syscall_32.tbl
index 5f8591ce7f25..f652771d6e94 100644
--- a/arch/x86/entry/syscalls/syscall_32.tbl
+++ b/arch/x86/entry/syscalls/syscall_32.tbl
@@ -466,3 +466,6 @@
 459	i386	lsm_get_self_attr	sys_lsm_get_self_attr
 460	i386	lsm_set_self_attr	sys_lsm_set_self_attr
 461	i386	lsm_list_modules	sys_lsm_list_modules
+462	i386	process_ksm_enable		sys_process_ksm_enable
+463	i386	process_ksm_disable		sys_process_ksm_disable
+464	i386	process_ksm_status		sys_process_ksm_status
diff --git a/arch/x86/entry/syscalls/syscall_64.tbl b/arch/x86/entry/syscalls/syscall_64.tbl
index 7e8d46f4147f..0e86c5b0fd39 100644
--- a/arch/x86/entry/syscalls/syscall_64.tbl
+++ b/arch/x86/entry/syscalls/syscall_64.tbl
@@ -383,6 +383,9 @@
 459	common	lsm_get_self_attr	sys_lsm_get_self_attr
 460	common	lsm_set_self_attr	sys_lsm_set_self_attr
 461	common	lsm_list_modules	sys_lsm_list_modules
+462	common	process_ksm_enable	sys_process_ksm_enable
+463	common	process_ksm_disable	sys_process_ksm_disable
+464	common	process_ksm_status	sys_process_ksm_status
 
 #
 # Due to a historical design error, certain syscalls are numbered differently
diff --git a/arch/xtensa/kernel/syscalls/syscall.tbl b/arch/xtensa/kernel/syscalls/syscall.tbl
index dd116598fb25..28e59bfe9474 100644
--- a/arch/xtensa/kernel/syscalls/syscall.tbl
+++ b/arch/xtensa/kernel/syscalls/syscall.tbl
@@ -432,3 +432,6 @@
 459	common	lsm_get_self_attr		sys_lsm_get_self_attr
 460	common	lsm_set_self_attr		sys_lsm_set_self_attr
 461	common	lsm_list_modules		sys_lsm_list_modules
+462	common	process_ksm_enable		sys_process_ksm_enable
+463	common	process_ksm_disable		sys_process_ksm_disable
+464	common	process_ksm_status		sys_process_ksm_status
diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
index e619ac10cd23..ad49152558ac 100644
--- a/include/linux/syscalls.h
+++ b/include/linux/syscalls.h
@@ -818,6 +818,9 @@ asmlinkage long sys_madvise(unsigned long start, size_t len, int behavior);
 asmlinkage long sys_process_madvise(int pidfd, const struct iovec __user *vec,
 			size_t vlen, int behavior, unsigned int flags);
 asmlinkage long sys_process_mrelease(int pidfd, unsigned int flags);
+asmlinkage long sys_process_ksm_enable(int pidfd, unsigned int flags);
+asmlinkage long sys_process_ksm_disable(int pidfd, unsigned int flags);
+asmlinkage long sys_process_ksm_status(int pidfd, unsigned int flags);
 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
 			unsigned long prot, unsigned long pgoff,
 			unsigned long flags);
diff --git a/include/uapi/asm-generic/unistd.h b/include/uapi/asm-generic/unistd.h
index 75f00965ab15..c46daa8bda1e 100644
--- a/include/uapi/asm-generic/unistd.h
+++ b/include/uapi/asm-generic/unistd.h
@@ -842,8 +842,17 @@ __SYSCALL(__NR_lsm_set_self_attr, sys_lsm_set_self_attr)
 #define __NR_lsm_list_modules 461
 __SYSCALL(__NR_lsm_list_modules, sys_lsm_list_modules)
 
+#define __NR_process_ksm_enable 462
+__SYSCALL(__NR_process_ksm_enable, sys_process_ksm_enable)
+
+#define __NR_process_ksm_disable 463
+__SYSCALL(__NR_process_ksm_disable, sys_process_ksm_disable)
+
+#define __NR_process_ksm_status 464
+__SYSCALL(__NR_process_ksm_status, sys_process_ksm_status)
+
 #undef __NR_syscalls
-#define __NR_syscalls 462
+#define __NR_syscalls 465
 
 /*
  * 32 bit systems traditionally used different
diff --git a/kernel/sys.c b/kernel/sys.c
index 8bb106a56b3a..34b78c8e7bc6 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -2767,6 +2767,153 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 	return error;
 }
 
+#ifdef CONFIG_KSM
+enum pkc_action {
+	PKSM_ENABLE = 0,
+	PKSM_DISABLE,
+	PKSM_STATUS,
+};
+
+static long do_process_ksm_control(int pidfd, enum pkc_action action)
+{
+	long ret;
+	struct pid *pid;
+	struct task_struct *task;
+	struct mm_struct *mm;
+	unsigned int f_flags;
+
+	pid = pidfd_get_pid(pidfd, &f_flags);
+	if (IS_ERR(pid)) {
+		ret = PTR_ERR(pid);
+		goto out;
+	}
+
+	task = get_pid_task(pid, PIDTYPE_PID);
+	if (!task) {
+		ret = -ESRCH;
+		goto put_pid;
+	}
+
+	/* Require PTRACE_MODE_READ to avoid leaking ASLR metadata. */
+	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
+	if (IS_ERR_OR_NULL(mm)) {
+		ret = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
+		goto release_task;
+	}
+
+	/* Require CAP_SYS_NICE for influencing process performance. */
+	if (!capable(CAP_SYS_NICE)) {
+		ret = -EPERM;
+		goto release_mm;
+	}
+
+	if (mmap_write_lock_killable(mm)) {
+		ret = -EINTR;
+		goto release_mm;
+	}
+
+	switch (action) {
+		case PKSM_ENABLE:
+			ret = ksm_enable_merge_any(mm);
+			break;
+		case PKSM_DISABLE:
+			ret = ksm_disable_merge_any(mm);
+			break;
+		case PKSM_STATUS:
+			ret = !!test_bit(MMF_VM_MERGE_ANY, &mm->flags);
+			break;
+	}
+
+	mmap_write_unlock(mm);
+
+release_mm:
+	mmput(mm);
+release_task:
+	put_task_struct(task);
+put_pid:
+	put_pid(pid);
+out:
+	return ret;
+}
+#endif /* CONFIG_KSM */
+
+SYSCALL_DEFINE2(process_ksm_enable, int, pidfd, unsigned int, flags)
+{
+#ifdef CONFIG_KSM
+	if (flags != 0)
+		return -EINVAL;
+
+	return do_process_ksm_control(pidfd, PKSM_ENABLE);
+#else /* CONFIG_KSM */
+	return -ENOSYS;
+#endif /* CONFIG_KSM */
+}
+
+SYSCALL_DEFINE2(process_ksm_disable, int, pidfd, unsigned int, flags)
+{
+#ifdef CONFIG_KSM
+	if (flags != 0)
+		return -EINVAL;
+
+	return do_process_ksm_control(pidfd, PKSM_DISABLE);
+#else /* CONFIG_KSM */
+	return -ENOSYS;
+#endif /* CONFIG_KSM */
+}
+
+SYSCALL_DEFINE2(process_ksm_status, int, pidfd, unsigned int, flags)
+{
+#ifdef CONFIG_KSM
+	if (flags != 0)
+		return -EINVAL;
+
+	return do_process_ksm_control(pidfd, PKSM_STATUS);
+#else /* CONFIG_KSM */
+	return -ENOSYS;
+#endif /* CONFIG_KSM */
+}
+
+#ifdef CONFIG_KSM
+static ssize_t process_ksm_enable_show(struct kobject *kobj,
+		struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%u\n", __NR_process_ksm_enable);
+}
+static struct kobj_attribute process_ksm_enable_attr = __ATTR_RO(process_ksm_enable);
+
+static ssize_t process_ksm_disable_show(struct kobject *kobj,
+		struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%u\n", __NR_process_ksm_disable);
+}
+static struct kobj_attribute process_ksm_disable_attr = __ATTR_RO(process_ksm_disable);
+
+static ssize_t process_ksm_status_show(struct kobject *kobj,
+		struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%u\n", __NR_process_ksm_status);
+}
+static struct kobj_attribute process_ksm_status_attr = __ATTR_RO(process_ksm_status);
+
+static struct attribute *process_ksm_sysfs_attrs[] = {
+	&process_ksm_enable_attr.attr,
+	&process_ksm_disable_attr.attr,
+	&process_ksm_status_attr.attr,
+	NULL,
+};
+
+static const struct attribute_group process_ksm_sysfs_attr_group = {
+	.attrs = process_ksm_sysfs_attrs,
+	.name = "process_ksm",
+};
+
+static int __init process_ksm_sysfs_init(void)
+{
+	return sysfs_create_group(kernel_kobj, &process_ksm_sysfs_attr_group);
+}
+subsys_initcall(process_ksm_sysfs_init);
+#endif /* CONFIG_KSM */
+
 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
 		struct getcpu_cache __user *, unused)
 {
diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c
index faad00cce269..c7c9eb656468 100644
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -188,6 +188,9 @@ COND_SYSCALL(mincore);
 COND_SYSCALL(madvise);
 COND_SYSCALL(process_madvise);
 COND_SYSCALL(process_mrelease);
+COND_SYSCALL(process_ksm_enable);
+COND_SYSCALL(process_ksm_disable);
+COND_SYSCALL(process_ksm_status);
 COND_SYSCALL(remap_file_pages);
 COND_SYSCALL(mbind);
 COND_SYSCALL(get_mempolicy);
-- 
2.44.0

From ef7086e8c29b16dc397fcf7e29f4608865f41cad Mon Sep 17 00:00:00 2001
From: Peter Jung <admin@ptr1337.dev>
Date: Wed, 3 Apr 2024 17:07:41 +0200
Subject: [PATCH 8/8] zstd

Signed-off-by: Peter Jung <admin@ptr1337.dev>
---
 include/linux/zstd.h                          |    2 +-
 include/linux/zstd_errors.h                   |   23 +-
 include/linux/zstd_lib.h                      |  850 +++++--
 lib/zstd/Makefile                             |    2 +-
 lib/zstd/common/allocations.h                 |   56 +
 lib/zstd/common/bits.h                        |  149 ++
 lib/zstd/common/bitstream.h                   |  127 +-
 lib/zstd/common/compiler.h                    |  134 +-
 lib/zstd/common/cpu.h                         |    3 +-
 lib/zstd/common/debug.c                       |    9 +-
 lib/zstd/common/debug.h                       |   34 +-
 lib/zstd/common/entropy_common.c              |   42 +-
 lib/zstd/common/error_private.c               |   12 +-
 lib/zstd/common/error_private.h               |   84 +-
 lib/zstd/common/fse.h                         |   94 +-
 lib/zstd/common/fse_decompress.c              |  130 +-
 lib/zstd/common/huf.h                         |  237 +-
 lib/zstd/common/mem.h                         |    3 +-
 lib/zstd/common/portability_macros.h          |   28 +-
 lib/zstd/common/zstd_common.c                 |   38 +-
 lib/zstd/common/zstd_deps.h                   |   16 +-
 lib/zstd/common/zstd_internal.h               |  109 +-
 lib/zstd/compress/clevels.h                   |    3 +-
 lib/zstd/compress/fse_compress.c              |   74 +-
 lib/zstd/compress/hist.c                      |    3 +-
 lib/zstd/compress/hist.h                      |    3 +-
 lib/zstd/compress/huf_compress.c              |  441 ++--
 lib/zstd/compress/zstd_compress.c             | 2111 ++++++++++++-----
 lib/zstd/compress/zstd_compress_internal.h    |  359 ++-
 lib/zstd/compress/zstd_compress_literals.c    |  155 +-
 lib/zstd/compress/zstd_compress_literals.h    |   25 +-
 lib/zstd/compress/zstd_compress_sequences.c   |    7 +-
 lib/zstd/compress/zstd_compress_sequences.h   |    3 +-
 lib/zstd/compress/zstd_compress_superblock.c  |  376 ++-
 lib/zstd/compress/zstd_compress_superblock.h  |    3 +-
 lib/zstd/compress/zstd_cwksp.h                |  169 +-
 lib/zstd/compress/zstd_double_fast.c          |  143 +-
 lib/zstd/compress/zstd_double_fast.h          |   17 +-
 lib/zstd/compress/zstd_fast.c                 |  596 +++--
 lib/zstd/compress/zstd_fast.h                 |    6 +-
 lib/zstd/compress/zstd_lazy.c                 |  732 +++---
 lib/zstd/compress/zstd_lazy.h                 |  138 +-
 lib/zstd/compress/zstd_ldm.c                  |   21 +-
 lib/zstd/compress/zstd_ldm.h                  |    3 +-
 lib/zstd/compress/zstd_ldm_geartab.h          |    3 +-
 lib/zstd/compress/zstd_opt.c                  |  497 ++--
 lib/zstd/compress/zstd_opt.h                  |   41 +-
 lib/zstd/decompress/huf_decompress.c          |  887 ++++---
 lib/zstd/decompress/zstd_ddict.c              |    9 +-
 lib/zstd/decompress/zstd_ddict.h              |    3 +-
 lib/zstd/decompress/zstd_decompress.c         |  358 ++-
 lib/zstd/decompress/zstd_decompress_block.c   |  708 +++---
 lib/zstd/decompress/zstd_decompress_block.h   |   10 +-
 .../decompress/zstd_decompress_internal.h     |    9 +-
 lib/zstd/decompress_sources.h                 |    2 +-
 lib/zstd/zstd_common_module.c                 |    5 +-
 lib/zstd/zstd_compress_module.c               |    2 +-
 lib/zstd/zstd_decompress_module.c             |    4 +-
 58 files changed, 6577 insertions(+), 3531 deletions(-)
 create mode 100644 lib/zstd/common/allocations.h
 create mode 100644 lib/zstd/common/bits.h

diff --git a/include/linux/zstd.h b/include/linux/zstd.h
index 113408eef6ec..f109d49f43f8 100644
--- a/include/linux/zstd.h
+++ b/include/linux/zstd.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/include/linux/zstd_errors.h b/include/linux/zstd_errors.h
index 58b6dd45a969..6d5cf55f0bf3 100644
--- a/include/linux/zstd_errors.h
+++ b/include/linux/zstd_errors.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -17,8 +18,17 @@
 
 
 /* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
-#define ZSTDERRORLIB_VISIBILITY 
-#define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBILITY
+#define ZSTDERRORLIB_VISIBLE 
+
+#ifndef ZSTDERRORLIB_HIDDEN
+#  if (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDERRORLIB_HIDDEN
+#  endif
+#endif
+
+#define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBLE
 
 /*-*********************************************
  *  Error codes list
@@ -43,14 +53,17 @@ typedef enum {
   ZSTD_error_frameParameter_windowTooLarge = 16,
   ZSTD_error_corruption_detected = 20,
   ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_literals_headerWrong = 24,
   ZSTD_error_dictionary_corrupted      = 30,
   ZSTD_error_dictionary_wrong          = 32,
   ZSTD_error_dictionaryCreation_failed = 34,
   ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_combination_unsupported = 41,
   ZSTD_error_parameter_outOfBound    = 42,
   ZSTD_error_tableLog_tooLarge       = 44,
   ZSTD_error_maxSymbolValue_tooLarge = 46,
   ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_stabilityCondition_notRespected = 50,
   ZSTD_error_stage_wrong       = 60,
   ZSTD_error_init_missing      = 62,
   ZSTD_error_memory_allocation = 64,
@@ -58,11 +71,15 @@ typedef enum {
   ZSTD_error_dstSize_tooSmall = 70,
   ZSTD_error_srcSize_wrong    = 72,
   ZSTD_error_dstBuffer_null   = 74,
+  ZSTD_error_noForwardProgress_destFull = 80,
+  ZSTD_error_noForwardProgress_inputEmpty = 82,
   /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
   ZSTD_error_frameIndex_tooLarge = 100,
   ZSTD_error_seekableIO          = 102,
   ZSTD_error_dstBuffer_wrong     = 104,
   ZSTD_error_srcBuffer_wrong     = 105,
+  ZSTD_error_sequenceProducer_failed = 106,
+  ZSTD_error_externalSequences_invalid = 107,
   ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
 } ZSTD_ErrorCode;
 
diff --git a/include/linux/zstd_lib.h b/include/linux/zstd_lib.h
index 79d55465d5c1..6320fedcf8a4 100644
--- a/include/linux/zstd_lib.h
+++ b/include/linux/zstd_lib.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,23 +12,42 @@
 #ifndef ZSTD_H_235446
 #define ZSTD_H_235446
 
-/* ======   Dependency   ======*/
+/* ======   Dependencies   ======*/
 #include <linux/limits.h>   /* INT_MAX */
 #include <linux/types.h>   /* size_t */
 
 
 /* =====   ZSTDLIB_API : control library symbols visibility   ===== */
-#ifndef ZSTDLIB_VISIBLE
+#define ZSTDLIB_VISIBLE 
+
+#ifndef ZSTDLIB_HIDDEN
 #  if (__GNUC__ >= 4) && !defined(__MINGW32__)
-#    define ZSTDLIB_VISIBLE __attribute__ ((visibility ("default")))
 #    define ZSTDLIB_HIDDEN __attribute__ ((visibility ("hidden")))
 #  else
-#    define ZSTDLIB_VISIBLE
 #    define ZSTDLIB_HIDDEN
 #  endif
 #endif
+
 #define ZSTDLIB_API ZSTDLIB_VISIBLE
 
+/* Deprecation warnings :
+ * Should these warnings be a problem, it is generally possible to disable them,
+ * typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
+ * Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
+ */
+#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
+#  define ZSTD_DEPRECATED(message) /* disable deprecation warnings */
+#else
+#  if (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif (__GNUC__ >= 3)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated))
+#  else
+#    pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
+#    define ZSTD_DEPRECATED(message)
+#  endif
+#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
+
 
 /* *****************************************************************************
   Introduction
@@ -65,7 +85,7 @@
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
 #define ZSTD_VERSION_MINOR    5
-#define ZSTD_VERSION_RELEASE  2
+#define ZSTD_VERSION_RELEASE  6
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
 
 /*! ZSTD_versionNumber() :
@@ -107,7 +127,8 @@ ZSTDLIB_API const char* ZSTD_versionString(void);
 ***************************************/
 /*! ZSTD_compress() :
  *  Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
- *  Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
+ *  NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
+ *        enough space to successfully compress the data.
  *  @return : compressed size written into `dst` (<= `dstCapacity),
  *            or an error code if it fails (which can be tested using ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
@@ -156,7 +177,9 @@ ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t
  *  "empty", "unknown" and "error" results to the same return value (0),
  *  while ZSTD_getFrameContentSize() gives them separate return values.
  * @return : decompressed size of `src` frame content _if known and not empty_, 0 otherwise. */
-ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+ZSTD_DEPRECATED("Replaced by ZSTD_getFrameContentSize")
+ZSTDLIB_API
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
 
 /*! ZSTD_findFrameCompressedSize() : Requires v1.4.0+
  * `src` should point to the start of a ZSTD frame or skippable frame.
@@ -168,8 +191,30 @@ ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize)
 
 
 /*======  Helper functions  ======*/
-#define ZSTD_COMPRESSBOUND(srcSize)   ((srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
-ZSTDLIB_API size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_compressBound() :
+ * maximum compressed size in worst case single-pass scenario.
+ * When invoking `ZSTD_compress()` or any other one-pass compression function,
+ * it's recommended to provide @dstCapacity >= ZSTD_compressBound(srcSize)
+ * as it eliminates one potential failure scenario,
+ * aka not enough room in dst buffer to write the compressed frame.
+ * Note : ZSTD_compressBound() itself can fail, if @srcSize > ZSTD_MAX_INPUT_SIZE .
+ *        In which case, ZSTD_compressBound() will return an error code
+ *        which can be tested using ZSTD_isError().
+ *
+ * ZSTD_COMPRESSBOUND() :
+ * same as ZSTD_compressBound(), but as a macro.
+ * It can be used to produce constants, which can be useful for static allocation,
+ * for example to size a static array on stack.
+ * Will produce constant value 0 if srcSize too large.
+ */
+#define ZSTD_MAX_INPUT_SIZE ((sizeof(size_t)==8) ? 0xFF00FF00FF00FF00ULL : 0xFF00FF00U)
+#define ZSTD_COMPRESSBOUND(srcSize)   (((size_t)(srcSize) >= ZSTD_MAX_INPUT_SIZE) ? 0 : (srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+ZSTDLIB_API size_t ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_isError() :
+ * Most ZSTD_* functions returning a size_t value can be tested for error,
+ * using ZSTD_isError().
+ * @return 1 if error, 0 otherwise
+ */
 ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
 ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
 ZSTDLIB_API int         ZSTD_minCLevel(void);               /*!< minimum negative compression level allowed, requires v1.4.0+ */
@@ -183,7 +228,7 @@ ZSTDLIB_API int         ZSTD_defaultCLevel(void);           /*!< default compres
 /*= Compression context
  *  When compressing many times,
  *  it is recommended to allocate a context just once,
- *  and re-use it for each successive compression operation.
+ *  and reuse it for each successive compression operation.
  *  This will make workload friendlier for system's memory.
  *  Note : re-using context is just a speed / resource optimization.
  *         It doesn't change the compression ratio, which remains identical.
@@ -196,9 +241,9 @@ ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);  /* accept NULL pointer *
 
 /*! ZSTD_compressCCtx() :
  *  Same as ZSTD_compress(), using an explicit ZSTD_CCtx.
- *  Important : in order to behave similarly to `ZSTD_compress()`,
- *  this function compresses at requested compression level,
- *  __ignoring any other parameter__ .
+ *  Important : in order to mirror `ZSTD_compress()` behavior,
+ *  this function compresses at the requested compression level,
+ *  __ignoring any other advanced parameter__ .
  *  If any advanced parameter was set using the advanced API,
  *  they will all be reset. Only `compressionLevel` remains.
  */
@@ -210,7 +255,7 @@ ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
 /*= Decompression context
  *  When decompressing many times,
  *  it is recommended to allocate a context only once,
- *  and re-use it for each successive compression operation.
+ *  and reuse it for each successive compression operation.
  *  This will make workload friendlier for system's memory.
  *  Use one context per thread for parallel execution. */
 typedef struct ZSTD_DCtx_s ZSTD_DCtx;
@@ -220,7 +265,7 @@ ZSTDLIB_API size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);  /* accept NULL pointer *
 /*! ZSTD_decompressDCtx() :
  *  Same as ZSTD_decompress(),
  *  requires an allocated ZSTD_DCtx.
- *  Compatible with sticky parameters.
+ *  Compatible with sticky parameters (see below).
  */
 ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
                                        void* dst, size_t dstCapacity,
@@ -236,12 +281,12 @@ ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
  *   using ZSTD_CCtx_set*() functions.
  *   Pushed parameters are sticky : they are valid for next compressed frame, and any subsequent frame.
  *   "sticky" parameters are applicable to `ZSTD_compress2()` and `ZSTD_compressStream*()` !
- *   __They do not apply to "simple" one-shot variants such as ZSTD_compressCCtx()__ .
+ *   __They do not apply to one-shot variants such as ZSTD_compressCCtx()__ .
  *
  *   It's possible to reset all parameters to "default" using ZSTD_CCtx_reset().
  *
  *   This API supersedes all other "advanced" API entry points in the experimental section.
- *   In the future, we expect to remove from experimental API entry points which are redundant with this API.
+ *   In the future, we expect to remove API entry points from experimental which are redundant with this API.
  */
 
 
@@ -324,6 +369,19 @@ typedef enum {
                               * The higher the value of selected strategy, the more complex it is,
                               * resulting in stronger and slower compression.
                               * Special: value 0 means "use default strategy". */
+
+    ZSTD_c_targetCBlockSize=130, /* v1.5.6+
+                                  * Attempts to fit compressed block size into approximatively targetCBlockSize.
+                                  * Bound by ZSTD_TARGETCBLOCKSIZE_MIN and ZSTD_TARGETCBLOCKSIZE_MAX.
+                                  * Note that it's not a guarantee, just a convergence target (default:0).
+                                  * No target when targetCBlockSize == 0.
+                                  * This is helpful in low bandwidth streaming environments to improve end-to-end latency,
+                                  * when a client can make use of partial documents (a prominent example being Chrome).
+                                  * Note: this parameter is stable since v1.5.6.
+                                  * It was present as an experimental parameter in earlier versions,
+                                  * but it's not recommended using it with earlier library versions
+                                  * due to massive performance regressions.
+                                  */
     /* LDM mode parameters */
     ZSTD_c_enableLongDistanceMatching=160, /* Enable long distance matching.
                                      * This parameter is designed to improve compression ratio
@@ -403,7 +461,6 @@ typedef enum {
      * ZSTD_c_forceMaxWindow
      * ZSTD_c_forceAttachDict
      * ZSTD_c_literalCompressionMode
-     * ZSTD_c_targetCBlockSize
      * ZSTD_c_srcSizeHint
      * ZSTD_c_enableDedicatedDictSearch
      * ZSTD_c_stableInBuffer
@@ -412,6 +469,9 @@ typedef enum {
      * ZSTD_c_validateSequences
      * ZSTD_c_useBlockSplitter
      * ZSTD_c_useRowMatchFinder
+     * ZSTD_c_prefetchCDictTables
+     * ZSTD_c_enableSeqProducerFallback
+     * ZSTD_c_maxBlockSize
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -421,7 +481,7 @@ typedef enum {
      ZSTD_c_experimentalParam3=1000,
      ZSTD_c_experimentalParam4=1001,
      ZSTD_c_experimentalParam5=1002,
-     ZSTD_c_experimentalParam6=1003,
+     /* was ZSTD_c_experimentalParam6=1003; is now ZSTD_c_targetCBlockSize */
      ZSTD_c_experimentalParam7=1004,
      ZSTD_c_experimentalParam8=1005,
      ZSTD_c_experimentalParam9=1006,
@@ -430,7 +490,11 @@ typedef enum {
      ZSTD_c_experimentalParam12=1009,
      ZSTD_c_experimentalParam13=1010,
      ZSTD_c_experimentalParam14=1011,
-     ZSTD_c_experimentalParam15=1012
+     ZSTD_c_experimentalParam15=1012,
+     ZSTD_c_experimentalParam16=1013,
+     ZSTD_c_experimentalParam17=1014,
+     ZSTD_c_experimentalParam18=1015,
+     ZSTD_c_experimentalParam19=1016
 } ZSTD_cParameter;
 
 typedef struct {
@@ -493,7 +557,7 @@ typedef enum {
  *                  They will be used to compress next frame.
  *                  Resetting session never fails.
  *  - The parameters : changes all parameters back to "default".
- *                  This removes any reference to any dictionary too.
+ *                  This also removes any reference to any dictionary or external sequence producer.
  *                  Parameters can only be changed between 2 sessions (i.e. no compression is currently ongoing)
  *                  otherwise the reset fails, and function returns an error value (which can be tested using ZSTD_isError())
  *  - Both : similar to resetting the session, followed by resetting parameters.
@@ -502,11 +566,13 @@ ZSTDLIB_API size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset);
 
 /*! ZSTD_compress2() :
  *  Behave the same as ZSTD_compressCCtx(), but compression parameters are set using the advanced API.
+ *  (note that this entry point doesn't even expose a compression level parameter).
  *  ZSTD_compress2() always starts a new frame.
  *  Should cctx hold data from a previously unfinished frame, everything about it is forgotten.
  *  - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_set*()
  *  - The function is always blocking, returns when compression is completed.
- *  Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
+ *  NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
+ *        enough space to successfully compress the data, though it is possible it fails for other reasons.
  * @return : compressed size written into `dst` (<= `dstCapacity),
  *           or an error code if it fails (which can be tested using ZSTD_isError()).
  */
@@ -543,13 +609,17 @@ typedef enum {
      * ZSTD_d_stableOutBuffer
      * ZSTD_d_forceIgnoreChecksum
      * ZSTD_d_refMultipleDDicts
+     * ZSTD_d_disableHuffmanAssembly
+     * ZSTD_d_maxBlockSize
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly
      */
      ZSTD_d_experimentalParam1=1000,
      ZSTD_d_experimentalParam2=1001,
      ZSTD_d_experimentalParam3=1002,
-     ZSTD_d_experimentalParam4=1003
+     ZSTD_d_experimentalParam4=1003,
+     ZSTD_d_experimentalParam5=1004,
+     ZSTD_d_experimentalParam6=1005
 
 } ZSTD_dParameter;
 
@@ -604,14 +674,14 @@ typedef struct ZSTD_outBuffer_s {
 *  A ZSTD_CStream object is required to track streaming operation.
 *  Use ZSTD_createCStream() and ZSTD_freeCStream() to create/release resources.
 *  ZSTD_CStream objects can be reused multiple times on consecutive compression operations.
-*  It is recommended to re-use ZSTD_CStream since it will play nicer with system's memory, by re-using already allocated memory.
+*  It is recommended to reuse ZSTD_CStream since it will play nicer with system's memory, by re-using already allocated memory.
 *
 *  For parallel execution, use one separate ZSTD_CStream per thread.
 *
 *  note : since v1.3.0, ZSTD_CStream and ZSTD_CCtx are the same thing.
 *
 *  Parameters are sticky : when starting a new compression on the same context,
-*  it will re-use the same sticky parameters as previous compression session.
+*  it will reuse the same sticky parameters as previous compression session.
 *  When in doubt, it's recommended to fully initialize the context before usage.
 *  Use ZSTD_CCtx_reset() to reset the context and ZSTD_CCtx_setParameter(),
 *  ZSTD_CCtx_setPledgedSrcSize(), or ZSTD_CCtx_loadDictionary() and friends to
@@ -700,6 +770,11 @@ typedef enum {
  *            only ZSTD_e_end or ZSTD_e_flush operations are allowed.
  *            Before starting a new compression job, or changing compression parameters,
  *            it is required to fully flush internal buffers.
+ *  - note: if an operation ends with an error, it may leave @cctx in an undefined state.
+ *          Therefore, it's UB to invoke ZSTD_compressStream2() of ZSTD_compressStream() on such a state.
+ *          In order to be re-employed after an error, a state must be reset,
+ *          which can be done explicitly (ZSTD_CCtx_reset()),
+ *          or is sometimes implied by methods starting a new compression job (ZSTD_initCStream(), ZSTD_compressCCtx())
  */
 ZSTDLIB_API size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
                                          ZSTD_outBuffer* output,
@@ -728,8 +803,6 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /*< recommended size for output
  * This following is a legacy streaming API, available since v1.0+ .
  * It can be replaced by ZSTD_CCtx_reset() and ZSTD_compressStream2().
  * It is redundant, but remains fully supported.
- * Streaming in combination with advanced parameters and dictionary compression
- * can only be used through the new API.
  ******************************************************************************/
 
 /*!
@@ -738,6 +811,9 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /*< recommended size for output
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
  *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+ *
+ * Note that ZSTD_initCStream() clears any previously set dictionary. Use the new API
+ * to compress with a dictionary.
  */
 ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
 /*!
@@ -758,7 +834,7 @@ ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
 *
 *  A ZSTD_DStream object is required to track streaming operations.
 *  Use ZSTD_createDStream() and ZSTD_freeDStream() to create/release resources.
-*  ZSTD_DStream objects can be re-used multiple times.
+*  ZSTD_DStream objects can be reused multiple times.
 *
 *  Use ZSTD_initDStream() to start a new decompression operation.
 * @return : recommended first input size
@@ -788,13 +864,37 @@ ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);  /* accept NULL pointer
 
 /*===== Streaming decompression functions =====*/
 
-/* This function is redundant with the advanced API and equivalent to:
+/*! ZSTD_initDStream() :
+ * Initialize/reset DStream state for new decompression operation.
+ * Call before new decompression operation using same DStream.
  *
+ * Note : This function is redundant with the advanced API and equivalent to:
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
  *     ZSTD_DCtx_refDDict(zds, NULL);
  */
 ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
 
+/*! ZSTD_decompressStream() :
+ * Streaming decompression function.
+ * Call repetitively to consume full input updating it as necessary.
+ * Function will update both input and output `pos` fields exposing current state via these fields:
+ * - `input.pos < input.size`, some input remaining and caller should provide remaining input
+ *   on the next call.
+ * - `output.pos < output.size`, decoder finished and flushed all remaining buffers.
+ * - `output.pos == output.size`, potentially uncflushed data present in the internal buffers,
+ *   call ZSTD_decompressStream() again to flush remaining data to output.
+ * Note : with no additional input, amount of data flushed <= ZSTD_BLOCKSIZE_MAX.
+ *
+ * @return : 0 when a frame is completely decoded and fully flushed,
+ *           or an error code, which can be tested using ZSTD_isError(),
+ *           or any other value > 0, which means there is some decoding or flushing to do to complete current frame.
+ *
+ * Note: when an operation returns with an error code, the @zds state may be left in undefined state.
+ *       It's UB to invoke `ZSTD_decompressStream()` on such a state.
+ *       In order to re-use such a state, it must be first reset,
+ *       which can be done explicitly (`ZSTD_DCtx_reset()`),
+ *       or is implied for operations starting some new decompression job (`ZSTD_initDStream`, `ZSTD_decompressDCtx()`, `ZSTD_decompress_usingDict()`)
+ */
 ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 
 ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
@@ -913,7 +1013,7 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
  *  If @return == 0, the dictID could not be decoded.
  *  This could for one of the following reasons :
  *  - The frame does not require a dictionary to be decoded (most common case).
- *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+ *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden piece of information.
  *    Note : this use case also happens when using a non-conformant dictionary.
  *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
  *  - This is not a Zstandard frame.
@@ -925,9 +1025,11 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
  * Advanced dictionary and prefix API (Requires v1.4.0+)
  *
  * This API allows dictionaries to be used with ZSTD_compress2(),
- * ZSTD_compressStream2(), and ZSTD_decompressDCtx(). Dictionaries are sticky, and
- * only reset with the context is reset with ZSTD_reset_parameters or
- * ZSTD_reset_session_and_parameters. Prefixes are single-use.
+ * ZSTD_compressStream2(), and ZSTD_decompressDCtx().
+ * Dictionaries are sticky, they remain valid when same context is reused,
+ * they only reset when the context is reset
+ * with ZSTD_reset_parameters or ZSTD_reset_session_and_parameters.
+ * In contrast, Prefixes are single-use.
  ******************************************************************************/
 
 
@@ -937,8 +1039,9 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special: Loading a NULL (or 0-size) dictionary invalidates previous dictionary,
  *           meaning "return to no-dictionary mode".
- *  Note 1 : Dictionary is sticky, it will be used for all future compressed frames.
- *           To return to "no-dictionary" situation, load a NULL dictionary (or reset parameters).
+ *  Note 1 : Dictionary is sticky, it will be used for all future compressed frames,
+ *           until parameters are reset, a new dictionary is loaded, or the dictionary
+ *           is explicitly invalidated by loading a NULL dictionary.
  *  Note 2 : Loading a dictionary involves building tables.
  *           It's also a CPU consuming operation, with non-negligible impact on latency.
  *           Tables are dependent on compression parameters, and for this reason,
@@ -947,11 +1050,15 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
  *           Use experimental ZSTD_CCtx_loadDictionary_byReference() to reference content instead.
  *           In such a case, dictionary buffer must outlive its users.
  *  Note 4 : Use ZSTD_CCtx_loadDictionary_advanced()
- *           to precisely select how dictionary content must be interpreted. */
+ *           to precisely select how dictionary content must be interpreted.
+ *  Note 5 : This method does not benefit from LDM (long distance mode).
+ *           If you want to employ LDM on some large dictionary content,
+ *           prefer employing ZSTD_CCtx_refPrefix() described below.
+ */
 ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
 
 /*! ZSTD_CCtx_refCDict() : Requires v1.4.0+
- *  Reference a prepared dictionary, to be used for all next compressed frames.
+ *  Reference a prepared dictionary, to be used for all future compressed frames.
  *  Note that compression parameters are enforced from within CDict,
  *  and supersede any compression parameter previously set within CCtx.
  *  The parameters ignored are labelled as "superseded-by-cdict" in the ZSTD_cParameter enum docs.
@@ -970,6 +1077,7 @@ ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
  *  Decompression will need same prefix to properly regenerate data.
  *  Compressing with a prefix is similar in outcome as performing a diff and compressing it,
  *  but performs much faster, especially during decompression (compression speed is tunable with compression level).
+ *  This method is compatible with LDM (long distance mode).
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
  *  Note 1 : Prefix buffer is referenced. It **must** outlive compression.
@@ -986,9 +1094,9 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
                                  const void* prefix, size_t prefixSize);
 
 /*! ZSTD_DCtx_loadDictionary() : Requires v1.4.0+
- *  Create an internal DDict from dict buffer,
- *  to be used to decompress next frames.
- *  The dictionary remains valid for all future frames, until explicitly invalidated.
+ *  Create an internal DDict from dict buffer, to be used to decompress all future frames.
+ *  The dictionary remains valid for all future frames, until explicitly invalidated, or
+ *  a new dictionary is loaded.
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
  *            meaning "return to no-dictionary mode".
@@ -1012,9 +1120,10 @@ ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, s
  *  The memory for the table is allocated on the first call to refDDict, and can be
  *  freed with ZSTD_freeDCtx().
  *
+ *  If called with ZSTD_d_refMultipleDDicts disabled (the default), only one dictionary
+ *  will be managed, and referencing a dictionary effectively "discards" any previous one.
+ *
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
- *  Note 1 : Currently, only one dictionary can be managed.
- *           Referencing a new dictionary effectively "discards" any previous one.
  *  Special: referencing a NULL DDict means "return to no-dictionary mode".
  *  Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
  */
@@ -1071,24 +1180,6 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTDLIB_STATIC_API ZSTDLIB_VISIBLE
 #endif
 
-/* Deprecation warnings :
- * Should these warnings be a problem, it is generally possible to disable them,
- * typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
- * Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
- */
-#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
-#  define ZSTD_DEPRECATED(message) ZSTDLIB_STATIC_API  /* disable deprecation warnings */
-#else
-#  if (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_STATIC_API __attribute__((deprecated(message)))
-#  elif (__GNUC__ >= 3)
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_STATIC_API __attribute__((deprecated))
-#  else
-#    pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_STATIC_API
-#  endif
-#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
-
 /* **************************************************************************************
  *   experimental API (static linking only)
  ****************************************************************************************
@@ -1123,6 +1214,7 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTD_TARGETLENGTH_MIN     0   /* note : comparing this constant to an unsigned results in a tautological test */
 #define ZSTD_STRATEGY_MIN        ZSTD_fast
 #define ZSTD_STRATEGY_MAX        ZSTD_btultra2
+#define ZSTD_BLOCKSIZE_MAX_MIN (1 << 10) /* The minimum valid max blocksize. Maximum blocksizes smaller than this make compressBound() inaccurate. */
 
 
 #define ZSTD_OVERLAPLOG_MIN       0
@@ -1146,7 +1238,7 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTD_LDM_HASHRATELOG_MAX (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
 
 /* Advanced parameter bounds */
-#define ZSTD_TARGETCBLOCKSIZE_MIN   64
+#define ZSTD_TARGETCBLOCKSIZE_MIN   1340 /* suitable to fit into an ethernet / wifi / 4G transport frame */
 #define ZSTD_TARGETCBLOCKSIZE_MAX   ZSTD_BLOCKSIZE_MAX
 #define ZSTD_SRCSIZEHINT_MIN        0
 #define ZSTD_SRCSIZEHINT_MAX        INT_MAX
@@ -1303,7 +1395,7 @@ typedef enum {
 } ZSTD_paramSwitch_e;
 
 /* *************************************
-*  Frame size functions
+*  Frame header and size functions
 ***************************************/
 
 /*! ZSTD_findDecompressedSize() :
@@ -1350,29 +1442,122 @@ ZSTDLIB_STATIC_API unsigned long long ZSTD_decompressBound(const void* src, size
  *           or an error code (if srcSize is too small) */
 ZSTDLIB_STATIC_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
 
+typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
+typedef struct {
+    unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
+    unsigned long long windowSize;       /* can be very large, up to <= frameContentSize */
+    unsigned blockSizeMax;
+    ZSTD_frameType_e frameType;          /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
+    unsigned headerSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+    unsigned _reserved1;
+    unsigned _reserved2;
+} ZSTD_frameHeader;
+
+/*! ZSTD_getFrameHeader() :
+ *  decode Frame Header, or requires larger `srcSize`.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /*< doesn't consume input */
+/*! ZSTD_getFrameHeader_advanced() :
+ *  same as ZSTD_getFrameHeader(),
+ *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
+
+/*! ZSTD_decompressionMargin() :
+ * Zstd supports in-place decompression, where the input and output buffers overlap.
+ * In this case, the output buffer must be at least (Margin + Output_Size) bytes large,
+ * and the input buffer must be at the end of the output buffer.
+ *
+ *  _______________________ Output Buffer ________________________
+ * |                                                              |
+ * |                                        ____ Input Buffer ____|
+ * |                                       |                      |
+ * v                                       v                      v
+ * |---------------------------------------|-----------|----------|
+ * ^                                                   ^          ^
+ * |___________________ Output_Size ___________________|_ Margin _|
+ *
+ * NOTE: See also ZSTD_DECOMPRESSION_MARGIN().
+ * NOTE: This applies only to single-pass decompression through ZSTD_decompress() or
+ * ZSTD_decompressDCtx().
+ * NOTE: This function supports multi-frame input.
+ *
+ * @param src The compressed frame(s)
+ * @param srcSize The size of the compressed frame(s)
+ * @returns The decompression margin or an error that can be checked with ZSTD_isError().
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_decompressionMargin(const void* src, size_t srcSize);
+
+/*! ZSTD_DECOMPRESS_MARGIN() :
+ * Similar to ZSTD_decompressionMargin(), but instead of computing the margin from
+ * the compressed frame, compute it from the original size and the blockSizeLog.
+ * See ZSTD_decompressionMargin() for details.
+ *
+ * WARNING: This macro does not support multi-frame input, the input must be a single
+ * zstd frame. If you need that support use the function, or implement it yourself.
+ *
+ * @param originalSize The original uncompressed size of the data.
+ * @param blockSize    The block size == MIN(windowSize, ZSTD_BLOCKSIZE_MAX).
+ *                     Unless you explicitly set the windowLog smaller than
+ *                     ZSTD_BLOCKSIZELOG_MAX you can just use ZSTD_BLOCKSIZE_MAX.
+ */
+#define ZSTD_DECOMPRESSION_MARGIN(originalSize, blockSize) ((size_t)(                                              \
+        ZSTD_FRAMEHEADERSIZE_MAX                                                              /* Frame header */ + \
+        4                                                                                         /* checksum */ + \
+        ((originalSize) == 0 ? 0 : 3 * (((originalSize) + (blockSize) - 1) / blockSize)) /* 3 bytes per block */ + \
+        (blockSize)                                                                    /* One block of margin */   \
+    ))
+
 typedef enum {
   ZSTD_sf_noBlockDelimiters = 0,         /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
   ZSTD_sf_explicitBlockDelimiters = 1    /* Representation of ZSTD_Sequence contains explicit block delimiters */
 } ZSTD_sequenceFormat_e;
 
+/*! ZSTD_sequenceBound() :
+ * `srcSize` : size of the input buffer
+ *  @return : upper-bound for the number of sequences that can be generated
+ *            from a buffer of srcSize bytes
+ *
+ *  note : returns number of sequences - to get bytes, multiply by sizeof(ZSTD_Sequence).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_sequenceBound(size_t srcSize);
+
 /*! ZSTD_generateSequences() :
- * Generate sequences using ZSTD_compress2, given a source buffer.
+ * WARNING: This function is meant for debugging and informational purposes ONLY!
+ * Its implementation is flawed, and it will be deleted in a future version.
+ * It is not guaranteed to succeed, as there are several cases where it will give
+ * up and fail. You should NOT use this function in production code.
+ *
+ * This function is deprecated, and will be removed in a future version.
+ *
+ * Generate sequences using ZSTD_compress2(), given a source buffer.
+ *
+ * @param zc The compression context to be used for ZSTD_compress2(). Set any
+ *           compression parameters you need on this context.
+ * @param outSeqs The output sequences buffer of size @p outSeqsSize
+ * @param outSeqsSize The size of the output sequences buffer.
+ *                    ZSTD_sequenceBound(srcSize) is an upper bound on the number
+ *                    of sequences that can be generated.
+ * @param src The source buffer to generate sequences from of size @p srcSize.
+ * @param srcSize The size of the source buffer.
  *
  * Each block will end with a dummy sequence
  * with offset == 0, matchLength == 0, and litLength == length of last literals.
  * litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
  * simply acts as a block delimiter.
  *
- * zc can be used to insert custom compression params.
- * This function invokes ZSTD_compress2
- *
- * The output of this function can be fed into ZSTD_compressSequences() with CCtx
- * setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
- * @return : number of sequences generated
+ * @returns The number of sequences generated, necessarily less than
+ *          ZSTD_sequenceBound(srcSize), or an error code that can be checked
+ *          with ZSTD_isError().
  */
-
-ZSTDLIB_STATIC_API size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
-                                          size_t outSeqsSize, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("For debugging only, will be replaced by ZSTD_extractSequences()")
+ZSTDLIB_STATIC_API size_t
+ZSTD_generateSequences(ZSTD_CCtx* zc,
+                       ZSTD_Sequence* outSeqs, size_t outSeqsSize,
+                       const void* src, size_t srcSize);
 
 /*! ZSTD_mergeBlockDelimiters() :
  * Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
@@ -1388,7 +1573,9 @@ ZSTDLIB_STATIC_API size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* o
 ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
 
 /*! ZSTD_compressSequences() :
- * Compress an array of ZSTD_Sequence, generated from the original source buffer, into dst.
+ * Compress an array of ZSTD_Sequence, associated with @src buffer, into dst.
+ * @src contains the entire input (not just the literals).
+ * If @srcSize > sum(sequence.length), the remaining bytes are considered all literals
  * If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
  * The entire source is compressed into a single frame.
  *
@@ -1413,11 +1600,12 @@ ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, si
  * Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
  * Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
  *         and cannot emit an RLE block that disagrees with the repcode history
- * @return : final compressed size or a ZSTD error.
+ * @return : final compressed size, or a ZSTD error code.
  */
-ZSTDLIB_STATIC_API size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstSize,
-                                  const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
-                                  const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t
+ZSTD_compressSequences( ZSTD_CCtx* cctx, void* dst, size_t dstSize,
+                        const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                        const void* src, size_t srcSize);
 
 
 /*! ZSTD_writeSkippableFrame() :
@@ -1464,48 +1652,59 @@ ZSTDLIB_API unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
 /*! ZSTD_estimate*() :
  *  These functions make it possible to estimate memory usage
  *  of a future {D,C}Ctx, before its creation.
+ *  This is useful in combination with ZSTD_initStatic(),
+ *  which makes it possible to employ a static buffer for ZSTD_CCtx* state.
  *
  *  ZSTD_estimateCCtxSize() will provide a memory budget large enough
- *  for any compression level up to selected one.
- *  Note : Unlike ZSTD_estimateCStreamSize*(), this estimate
- *         does not include space for a window buffer.
- *         Therefore, the estimation is only guaranteed for single-shot compressions, not streaming.
+ *  to compress data of any size using one-shot compression ZSTD_compressCCtx() or ZSTD_compress2()
+ *  associated with any compression level up to max specified one.
  *  The estimate will assume the input may be arbitrarily large,
  *  which is the worst case.
  *
+ *  Note that the size estimation is specific for one-shot compression,
+ *  it is not valid for streaming (see ZSTD_estimateCStreamSize*())
+ *  nor other potential ways of using a ZSTD_CCtx* state.
+ *
  *  When srcSize can be bound by a known and rather "small" value,
- *  this fact can be used to provide a tighter estimation
- *  because the CCtx compression context will need less memory.
- *  This tighter estimation can be provided by more advanced functions
+ *  this knowledge can be used to provide a tighter budget estimation
+ *  because the ZSTD_CCtx* state will need less memory for small inputs.
+ *  This tighter estimation can be provided by employing more advanced functions
  *  ZSTD_estimateCCtxSize_usingCParams(), which can be used in tandem with ZSTD_getCParams(),
  *  and ZSTD_estimateCCtxSize_usingCCtxParams(), which can be used in tandem with ZSTD_CCtxParams_setParameter().
  *  Both can be used to estimate memory using custom compression parameters and arbitrary srcSize limits.
  *
- *  Note 2 : only single-threaded compression is supported.
+ *  Note : only single-threaded compression is supported.
  *  ZSTD_estimateCCtxSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
  */
-ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int maxCompressionLevel);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateDCtxSize(void);
 
 /*! ZSTD_estimateCStreamSize() :
- *  ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
- *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  ZSTD_estimateCStreamSize() will provide a memory budget large enough for streaming compression
+ *  using any compression level up to the max specified one.
+ *  It will also consider src size to be arbitrarily "large", which is a worst case scenario.
  *  If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
  *  ZSTD_estimateCStreamSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
  *  ZSTD_estimateCStreamSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParams_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_c_nbWorkers is >= 1.
  *  Note : CStream size estimation is only correct for single-threaded compression.
- *  ZSTD_DStream memory budget depends on window Size.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *  Note 2 : ZSTD_estimateCStreamSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+ *  Size estimates assume that no external sequence producer is registered.
+ *
+ *  ZSTD_DStream memory budget depends on frame's window Size.
  *  This information can be passed manually, using ZSTD_estimateDStreamSize,
  *  or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
+ *  Any frame requesting a window size larger than max specified one will be rejected.
  *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
  *         an internal ?Dict will be created, which additional size is not estimated here.
- *         In this case, get total size by adding ZSTD_estimate?DictSize */
-ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ *         In this case, get total size by adding ZSTD_estimate?DictSize
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int maxCompressionLevel);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
-ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t maxWindowSize);
 ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
 
 /*! ZSTD_estimate?DictSize() :
@@ -1649,22 +1848,45 @@ ZSTDLIB_STATIC_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
  *  This function never fails (wide contract) */
 ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
 
+/*! ZSTD_CCtx_setCParams() :
+ *  Set all parameters provided within @p cparams into the working @p cctx.
+ *  Note : if modifying parameters during compression (MT mode only),
+ *         note that changes to the .windowLog parameter will be ignored.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ *         On failure, no parameters are updated.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams);
+
+/*! ZSTD_CCtx_setFParams() :
+ *  Set all parameters provided within @p fparams into the working @p cctx.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams);
+
+/*! ZSTD_CCtx_setParams() :
+ *  Set all parameters provided within @p params into the working @p cctx.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params);
+
 /*! ZSTD_compress_advanced() :
  *  Note : this function is now DEPRECATED.
  *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_setParameter() and other parameter setters.
  *  This prototype will generate compilation warnings. */
 ZSTD_DEPRECATED("use ZSTD_compress2")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_advanced(ZSTD_CCtx* cctx,
-                                          void* dst, size_t dstCapacity,
-                                    const void* src, size_t srcSize,
-                                    const void* dict,size_t dictSize,
-                                          ZSTD_parameters params);
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                        const void* dict,size_t dictSize,
+                              ZSTD_parameters params);
 
 /*! ZSTD_compress_usingCDict_advanced() :
  *  Note : this function is now DEPRECATED.
  *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_loadDictionary() and other parameter setters.
  *  This prototype will generate compilation warnings. */
 ZSTD_DEPRECATED("use ZSTD_compress2 with ZSTD_CCtx_loadDictionary")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
                                               void* dst, size_t dstCapacity,
                                         const void* src, size_t srcSize,
@@ -1737,11 +1959,6 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  */
 #define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
 
-/* Tries to fit compressed block size to be around targetCBlockSize.
- * No target when targetCBlockSize == 0.
- * There is no guarantee on compressed block size (default:0) */
-#define ZSTD_c_targetCBlockSize ZSTD_c_experimentalParam6
-
 /* User's best guess of source size.
  * Hint is not valid when srcSizeHint == 0.
  * There is no guarantee that hint is close to actual source size,
@@ -1808,13 +2025,16 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  * Experimental parameter.
  * Default is 0 == disabled. Set to 1 to enable.
  *
- * Tells the compressor that the ZSTD_inBuffer will ALWAYS be the same
- * between calls, except for the modifications that zstd makes to pos (the
- * caller must not modify pos). This is checked by the compressor, and
- * compression will fail if it ever changes. This means the only flush
- * mode that makes sense is ZSTD_e_end, so zstd will error if ZSTD_e_end
- * is not used. The data in the ZSTD_inBuffer in the range [src, src + pos)
- * MUST not be modified during compression or you will get data corruption.
+ * Tells the compressor that input data presented with ZSTD_inBuffer
+ * will ALWAYS be the same between calls.
+ * Technically, the @src pointer must never be changed,
+ * and the @pos field can only be updated by zstd.
+ * However, it's possible to increase the @size field,
+ * allowing scenarios where more data can be appended after compressions starts.
+ * These conditions are checked by the compressor,
+ * and compression will fail if they are not respected.
+ * Also, data in the ZSTD_inBuffer within the range [src, src + pos)
+ * MUST not be modified during compression or it will result in data corruption.
  *
  * When this flag is enabled zstd won't allocate an input window buffer,
  * because the user guarantees it can reference the ZSTD_inBuffer until
@@ -1822,18 +2042,15 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  * large enough to fit a block (see ZSTD_c_stableOutBuffer). This will also
  * avoid the memcpy() from the input buffer to the input window buffer.
  *
- * NOTE: ZSTD_compressStream2() will error if ZSTD_e_end is not used.
- * That means this flag cannot be used with ZSTD_compressStream().
- *
  * NOTE: So long as the ZSTD_inBuffer always points to valid memory, using
  * this flag is ALWAYS memory safe, and will never access out-of-bounds
- * memory. However, compression WILL fail if you violate the preconditions.
+ * memory. However, compression WILL fail if conditions are not respected.
  *
- * WARNING: The data in the ZSTD_inBuffer in the range [dst, dst + pos) MUST
- * not be modified during compression or you will get data corruption. This
- * is because zstd needs to reference data in the ZSTD_inBuffer to find
+ * WARNING: The data in the ZSTD_inBuffer in the range [src, src + pos) MUST
+ * not be modified during compression or it will result in data corruption.
+ * This is because zstd needs to reference data in the ZSTD_inBuffer to find
  * matches. Normally zstd maintains its own window buffer for this purpose,
- * but passing this flag tells zstd to use the user provided buffer.
+ * but passing this flag tells zstd to rely on user provided buffer instead.
  */
 #define ZSTD_c_stableInBuffer ZSTD_c_experimentalParam9
 
@@ -1878,7 +2095,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  * Without validation, providing a sequence that does not conform to the zstd spec will cause
  * undefined behavior, and may produce a corrupted block.
  *
- * With validation enabled, a if sequence is invalid (see doc/zstd_compression_format.md for
+ * With validation enabled, if sequence is invalid (see doc/zstd_compression_format.md for
  * specifics regarding offset/matchlength requirements) then the function will bail out and
  * return an error.
  *
@@ -1928,6 +2145,79 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  */
 #define ZSTD_c_deterministicRefPrefix ZSTD_c_experimentalParam15
 
+/* ZSTD_c_prefetchCDictTables
+ * Controlled with ZSTD_paramSwitch_e enum. Default is ZSTD_ps_auto.
+ *
+ * In some situations, zstd uses CDict tables in-place rather than copying them
+ * into the working context. (See docs on ZSTD_dictAttachPref_e above for details).
+ * In such situations, compression speed is seriously impacted when CDict tables are
+ * "cold" (outside CPU cache). This parameter instructs zstd to prefetch CDict tables
+ * when they are used in-place.
+ *
+ * For sufficiently small inputs, the cost of the prefetch will outweigh the benefit.
+ * For sufficiently large inputs, zstd will by default memcpy() CDict tables
+ * into the working context, so there is no need to prefetch. This parameter is
+ * targeted at a middle range of input sizes, where a prefetch is cheap enough to be
+ * useful but memcpy() is too expensive. The exact range of input sizes where this
+ * makes sense is best determined by careful experimentation.
+ *
+ * Note: for this parameter, ZSTD_ps_auto is currently equivalent to ZSTD_ps_disable,
+ * but in the future zstd may conditionally enable this feature via an auto-detection
+ * heuristic for cold CDicts.
+ * Use ZSTD_ps_disable to opt out of prefetching under any circumstances.
+ */
+#define ZSTD_c_prefetchCDictTables ZSTD_c_experimentalParam16
+
+/* ZSTD_c_enableSeqProducerFallback
+ * Allowed values are 0 (disable) and 1 (enable). The default setting is 0.
+ *
+ * Controls whether zstd will fall back to an internal sequence producer if an
+ * external sequence producer is registered and returns an error code. This fallback
+ * is block-by-block: the internal sequence producer will only be called for blocks
+ * where the external sequence producer returns an error code. Fallback parsing will
+ * follow any other cParam settings, such as compression level, the same as in a
+ * normal (fully-internal) compression operation.
+ *
+ * The user is strongly encouraged to read the full Block-Level Sequence Producer API
+ * documentation (below) before setting this parameter. */
+#define ZSTD_c_enableSeqProducerFallback ZSTD_c_experimentalParam17
+
+/* ZSTD_c_maxBlockSize
+ * Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
+ * The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
+ *
+ * This parameter can be used to set an upper bound on the blocksize
+ * that overrides the default ZSTD_BLOCKSIZE_MAX. It cannot be used to set upper
+ * bounds greater than ZSTD_BLOCKSIZE_MAX or bounds lower than 1KB (will make
+ * compressBound() inaccurate). Only currently meant to be used for testing.
+ *
+ */
+#define ZSTD_c_maxBlockSize ZSTD_c_experimentalParam18
+
+/* ZSTD_c_searchForExternalRepcodes
+ * This parameter affects how zstd parses external sequences, such as sequences
+ * provided through the compressSequences() API or from an external block-level
+ * sequence producer.
+ *
+ * If set to ZSTD_ps_enable, the library will check for repeated offsets in
+ * external sequences, even if those repcodes are not explicitly indicated in
+ * the "rep" field. Note that this is the only way to exploit repcode matches
+ * while using compressSequences() or an external sequence producer, since zstd
+ * currently ignores the "rep" field of external sequences.
+ *
+ * If set to ZSTD_ps_disable, the library will not exploit repeated offsets in
+ * external sequences, regardless of whether the "rep" field has been set. This
+ * reduces sequence compression overhead by about 25% while sacrificing some
+ * compression ratio.
+ *
+ * The default value is ZSTD_ps_auto, for which the library will enable/disable
+ * based on compression level.
+ *
+ * Note: for now, this param only has an effect if ZSTD_c_blockDelimiters is
+ * set to ZSTD_sf_explicitBlockDelimiters. That may change in the future.
+ */
+#define ZSTD_c_searchForExternalRepcodes ZSTD_c_experimentalParam19
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.
@@ -2084,7 +2374,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParamete
  * in the range [dst, dst + pos) MUST not be modified during decompression
  * or you will get data corruption.
  *
- * When this flags is enabled zstd won't allocate an output buffer, because
+ * When this flag is enabled zstd won't allocate an output buffer, because
  * it can write directly to the ZSTD_outBuffer, but it will still allocate
  * an input buffer large enough to fit any compressed block. This will also
  * avoid the memcpy() from the internal output buffer to the ZSTD_outBuffer.
@@ -2137,6 +2427,33 @@ ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParamete
  */
 #define ZSTD_d_refMultipleDDicts ZSTD_d_experimentalParam4
 
+/* ZSTD_d_disableHuffmanAssembly
+ * Set to 1 to disable the Huffman assembly implementation.
+ * The default value is 0, which allows zstd to use the Huffman assembly
+ * implementation if available.
+ *
+ * This parameter can be used to disable Huffman assembly at runtime.
+ * If you want to disable it at compile time you can define the macro
+ * ZSTD_DISABLE_ASM.
+ */
+#define ZSTD_d_disableHuffmanAssembly ZSTD_d_experimentalParam5
+
+/* ZSTD_d_maxBlockSize
+ * Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
+ * The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
+ *
+ * Forces the decompressor to reject blocks whose content size is
+ * larger than the configured maxBlockSize. When maxBlockSize is
+ * larger than the windowSize, the windowSize is used instead.
+ * This saves memory on the decoder when you know all blocks are small.
+ *
+ * This option is typically used in conjunction with ZSTD_c_maxBlockSize.
+ *
+ * WARNING: This causes the decoder to reject otherwise valid frames
+ * that have block sizes larger than the configured maxBlockSize.
+ */
+#define ZSTD_d_maxBlockSize ZSTD_d_experimentalParam6
+
 
 /*! ZSTD_DCtx_setFormat() :
  *  This function is REDUNDANT. Prefer ZSTD_DCtx_setParameter().
@@ -2145,6 +2462,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParamete
  *  such ZSTD_f_zstd1_magicless for example.
  * @return : 0, or an error code (which can be tested using ZSTD_isError()). */
 ZSTD_DEPRECATED("use ZSTD_DCtx_setParameter() instead")
+ZSTDLIB_STATIC_API
 size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
 
 /*! ZSTD_decompressStream_simpleArgs() :
@@ -2181,6 +2499,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_decompressStream_simpleArgs (
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
                          int compressionLevel,
                          unsigned long long pledgedSrcSize);
@@ -2198,17 +2517,15 @@ size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
                      const void* dict, size_t dictSize,
                            int compressionLevel);
 
 /*! ZSTD_initCStream_advanced() :
- * This function is DEPRECATED, and is approximately equivalent to:
+ * This function is DEPRECATED, and is equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
- *     // Pseudocode: Set each zstd parameter and leave the rest as-is.
- *     for ((param, value) : params) {
- *         ZSTD_CCtx_setParameter(zcs, param, value);
- *     }
+ *     ZSTD_CCtx_setParams(zcs, params);
  *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
  *     ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
  *
@@ -2218,6 +2535,7 @@ size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
                     const void* dict, size_t dictSize,
                           ZSTD_parameters params,
@@ -2232,15 +2550,13 @@ size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
 
 /*! ZSTD_initCStream_usingCDict_advanced() :
- *   This function is DEPRECATED, and is approximately equivalent to:
+ *   This function is DEPRECATED, and is equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
- *     // Pseudocode: Set each zstd frame parameter and leave the rest as-is.
- *     for ((fParam, value) : fParams) {
- *         ZSTD_CCtx_setParameter(zcs, fParam, value);
- *     }
+ *     ZSTD_CCtx_setFParams(zcs, fParams);
  *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
  *     ZSTD_CCtx_refCDict(zcs, cdict);
  *
@@ -2250,6 +2566,7 @@ size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
                                const ZSTD_CDict* cdict,
                                      ZSTD_frameParameters fParams,
@@ -2264,7 +2581,7 @@ size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
  *       explicitly specified.
  *
  *  start a new frame, using same parameters from previous frame.
- *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place.
+ *  This is typically useful to skip dictionary loading stage, since it will reuse it in-place.
  *  Note that zcs must be init at least once before using ZSTD_resetCStream().
  *  If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
  *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
@@ -2274,6 +2591,7 @@ size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
  *  This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
 
 
@@ -2319,8 +2637,8 @@ ZSTDLIB_STATIC_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
  *     ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
  *
  * note: no dictionary will be used if dict == NULL or dictSize < 8
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_loadDictionary, see zstd.h for detailed instructions")
 ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
 
 /*!
@@ -2330,8 +2648,8 @@ ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const vo
  *     ZSTD_DCtx_refDDict(zds, ddict);
  *
  * note : ddict is referenced, it must outlive decompression session
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_refDDict, see zstd.h for detailed instructions")
 ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
 
 /*!
@@ -2339,18 +2657,202 @@ ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const Z
  *
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
  *
- * re-use decompression parameters from previous init; saves dictionary loading
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
+ * reuse decompression parameters from previous init; saves dictionary loading
  */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset, see zstd.h for detailed instructions")
 ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 
 
+/* ********************* BLOCK-LEVEL SEQUENCE PRODUCER API *********************
+ *
+ * *** OVERVIEW ***
+ * The Block-Level Sequence Producer API allows users to provide their own custom
+ * sequence producer which libzstd invokes to process each block. The produced list
+ * of sequences (literals and matches) is then post-processed by libzstd to produce
+ * valid compressed blocks.
+ *
+ * This block-level offload API is a more granular complement of the existing
+ * frame-level offload API compressSequences() (introduced in v1.5.1). It offers
+ * an easier migration story for applications already integrated with libzstd: the
+ * user application continues to invoke the same compression functions
+ * ZSTD_compress2() or ZSTD_compressStream2() as usual, and transparently benefits
+ * from the specific advantages of the external sequence producer. For example,
+ * the sequence producer could be tuned to take advantage of known characteristics
+ * of the input, to offer better speed / ratio, or could leverage hardware
+ * acceleration not available within libzstd itself.
+ *
+ * See contrib/externalSequenceProducer for an example program employing the
+ * Block-Level Sequence Producer API.
+ *
+ * *** USAGE ***
+ * The user is responsible for implementing a function of type
+ * ZSTD_sequenceProducer_F. For each block, zstd will pass the following
+ * arguments to the user-provided function:
+ *
+ *   - sequenceProducerState: a pointer to a user-managed state for the sequence
+ *     producer.
+ *
+ *   - outSeqs, outSeqsCapacity: an output buffer for the sequence producer.
+ *     outSeqsCapacity is guaranteed >= ZSTD_sequenceBound(srcSize). The memory
+ *     backing outSeqs is managed by the CCtx.
+ *
+ *   - src, srcSize: an input buffer for the sequence producer to parse.
+ *     srcSize is guaranteed to be <= ZSTD_BLOCKSIZE_MAX.
+ *
+ *   - dict, dictSize: a history buffer, which may be empty, which the sequence
+ *     producer may reference as it parses the src buffer. Currently, zstd will
+ *     always pass dictSize == 0 into external sequence producers, but this will
+ *     change in the future.
+ *
+ *   - compressionLevel: a signed integer representing the zstd compression level
+ *     set by the user for the current operation. The sequence producer may choose
+ *     to use this information to change its compression strategy and speed/ratio
+ *     tradeoff. Note: the compression level does not reflect zstd parameters set
+ *     through the advanced API.
+ *
+ *   - windowSize: a size_t representing the maximum allowed offset for external
+ *     sequences. Note that sequence offsets are sometimes allowed to exceed the
+ *     windowSize if a dictionary is present, see doc/zstd_compression_format.md
+ *     for details.
+ *
+ * The user-provided function shall return a size_t representing the number of
+ * sequences written to outSeqs. This return value will be treated as an error
+ * code if it is greater than outSeqsCapacity. The return value must be non-zero
+ * if srcSize is non-zero. The ZSTD_SEQUENCE_PRODUCER_ERROR macro is provided
+ * for convenience, but any value greater than outSeqsCapacity will be treated as
+ * an error code.
+ *
+ * If the user-provided function does not return an error code, the sequences
+ * written to outSeqs must be a valid parse of the src buffer. Data corruption may
+ * occur if the parse is not valid. A parse is defined to be valid if the
+ * following conditions hold:
+ *   - The sum of matchLengths and literalLengths must equal srcSize.
+ *   - All sequences in the parse, except for the final sequence, must have
+ *     matchLength >= ZSTD_MINMATCH_MIN. The final sequence must have
+ *     matchLength >= ZSTD_MINMATCH_MIN or matchLength == 0.
+ *   - All offsets must respect the windowSize parameter as specified in
+ *     doc/zstd_compression_format.md.
+ *   - If the final sequence has matchLength == 0, it must also have offset == 0.
+ *
+ * zstd will only validate these conditions (and fail compression if they do not
+ * hold) if the ZSTD_c_validateSequences cParam is enabled. Note that sequence
+ * validation has a performance cost.
+ *
+ * If the user-provided function returns an error, zstd will either fall back
+ * to an internal sequence producer or fail the compression operation. The user can
+ * choose between the two behaviors by setting the ZSTD_c_enableSeqProducerFallback
+ * cParam. Fallback compression will follow any other cParam settings, such as
+ * compression level, the same as in a normal compression operation.
+ *
+ * The user shall instruct zstd to use a particular ZSTD_sequenceProducer_F
+ * function by calling
+ *         ZSTD_registerSequenceProducer(cctx,
+ *                                       sequenceProducerState,
+ *                                       sequenceProducer)
+ * This setting will persist until the next parameter reset of the CCtx.
+ *
+ * The sequenceProducerState must be initialized by the user before calling
+ * ZSTD_registerSequenceProducer(). The user is responsible for destroying the
+ * sequenceProducerState.
+ *
+ * *** LIMITATIONS ***
+ * This API is compatible with all zstd compression APIs which respect advanced parameters.
+ * However, there are three limitations:
+ *
+ * First, the ZSTD_c_enableLongDistanceMatching cParam is not currently supported.
+ * COMPRESSION WILL FAIL if it is enabled and the user tries to compress with a block-level
+ * external sequence producer.
+ *   - Note that ZSTD_c_enableLongDistanceMatching is auto-enabled by default in some
+ *     cases (see its documentation for details). Users must explicitly set
+ *     ZSTD_c_enableLongDistanceMatching to ZSTD_ps_disable in such cases if an external
+ *     sequence producer is registered.
+ *   - As of this writing, ZSTD_c_enableLongDistanceMatching is disabled by default
+ *     whenever ZSTD_c_windowLog < 128MB, but that's subject to change. Users should
+ *     check the docs on ZSTD_c_enableLongDistanceMatching whenever the Block-Level Sequence
+ *     Producer API is used in conjunction with advanced settings (like ZSTD_c_windowLog).
+ *
+ * Second, history buffers are not currently supported. Concretely, zstd will always pass
+ * dictSize == 0 to the external sequence producer (for now). This has two implications:
+ *   - Dictionaries are not currently supported. Compression will *not* fail if the user
+ *     references a dictionary, but the dictionary won't have any effect.
+ *   - Stream history is not currently supported. All advanced compression APIs, including
+ *     streaming APIs, work with external sequence producers, but each block is treated as
+ *     an independent chunk without history from previous blocks.
+ *
+ * Third, multi-threading within a single compression is not currently supported. In other words,
+ * COMPRESSION WILL FAIL if ZSTD_c_nbWorkers > 0 and an external sequence producer is registered.
+ * Multi-threading across compressions is fine: simply create one CCtx per thread.
+ *
+ * Long-term, we plan to overcome all three limitations. There is no technical blocker to
+ * overcoming them. It is purely a question of engineering effort.
+ */
+
+#define ZSTD_SEQUENCE_PRODUCER_ERROR ((size_t)(-1))
+
+typedef size_t (*ZSTD_sequenceProducer_F) (
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+);
+
+/*! ZSTD_registerSequenceProducer() :
+ * Instruct zstd to use a block-level external sequence producer function.
+ *
+ * The sequenceProducerState must be initialized by the caller, and the caller is
+ * responsible for managing its lifetime. This parameter is sticky across
+ * compressions. It will remain set until the user explicitly resets compression
+ * parameters.
+ *
+ * Sequence producer registration is considered to be an "advanced parameter",
+ * part of the "advanced API". This means it will only have an effect on compression
+ * APIs which respect advanced parameters, such as compress2() and compressStream2().
+ * Older compression APIs such as compressCCtx(), which predate the introduction of
+ * "advanced parameters", will ignore any external sequence producer setting.
+ *
+ * The sequence producer can be "cleared" by registering a NULL function pointer. This
+ * removes all limitations described above in the "LIMITATIONS" section of the API docs.
+ *
+ * The user is strongly encouraged to read the full API documentation (above) before
+ * calling this function. */
+ZSTDLIB_STATIC_API void
+ZSTD_registerSequenceProducer(
+  ZSTD_CCtx* cctx,
+  void* sequenceProducerState,
+  ZSTD_sequenceProducer_F sequenceProducer
+);
+
+/*! ZSTD_CCtxParams_registerSequenceProducer() :
+ * Same as ZSTD_registerSequenceProducer(), but operates on ZSTD_CCtx_params.
+ * This is used for accurate size estimation with ZSTD_estimateCCtxSize_usingCCtxParams(),
+ * which is needed when creating a ZSTD_CCtx with ZSTD_initStaticCCtx().
+ *
+ * If you are using the external sequence producer API in a scenario where ZSTD_initStaticCCtx()
+ * is required, then this function is for you. Otherwise, you probably don't need it.
+ *
+ * See tests/zstreamtest.c for example usage. */
+ZSTDLIB_STATIC_API void
+ZSTD_CCtxParams_registerSequenceProducer(
+  ZSTD_CCtx_params* params,
+  void* sequenceProducerState,
+  ZSTD_sequenceProducer_F sequenceProducer
+);
+
+
 /* *******************************************************************
-*  Buffer-less and synchronous inner streaming functions
+*  Buffer-less and synchronous inner streaming functions (DEPRECATED)
+*
+*  This API is deprecated, and will be removed in a future version.
+*  It allows streaming (de)compression with user allocated buffers.
+*  However, it is hard to use, and not as well tested as the rest of
+*  our API.
 *
-*  This is an advanced API, giving full control over buffer management, for users which need direct control over memory.
-*  But it's also a complex one, with several restrictions, documented below.
-*  Prefer normal streaming API for an easier experience.
+*  Please use the normal streaming API instead: ZSTD_compressStream2,
+*  and ZSTD_decompressStream.
+*  If there is functionality that you need, but it doesn't provide,
+*  please open an issue on our GitHub.
 ********************************************************************* */
 
 /*
@@ -2358,11 +2860,10 @@ ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 
   A ZSTD_CCtx object is required to track streaming operations.
   Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
-  ZSTD_CCtx object can be re-used multiple times within successive compression operations.
+  ZSTD_CCtx object can be reused multiple times within successive compression operations.
 
   Start by initializing a context.
   Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression.
-  It's also possible to duplicate a reference context which has already been initialized, using ZSTD_copyCCtx()
 
   Then, consume your input using ZSTD_compressContinue().
   There are some important considerations to keep in mind when using this advanced function :
@@ -2380,36 +2881,46 @@ ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
   It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
   Without last block mark, frames are considered unfinished (hence corrupted) by compliant decoders.
 
-  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
+  `ZSTD_CCtx` object can be reused (ZSTD_compressBegin()) to compress again.
 */
 
 /*=====   Buffer-less streaming compression functions  =====*/
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /*< note: fails if cdict==NULL */
-ZSTDLIB_STATIC_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /*<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
 
+ZSTD_DEPRECATED("This function will likely be removed in a future release. It is misleading and has very limited utility.")
+ZSTDLIB_STATIC_API
+size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /*<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 /* The ZSTD_compressBegin_advanced() and ZSTD_compressBegin_usingCDict_advanced() are now DEPRECATED and will generate a compiler warning */
 ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /*< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
 ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
 /*
   Buffer-less streaming decompression (synchronous mode)
 
   A ZSTD_DCtx object is required to track streaming operations.
   Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
-  A ZSTD_DCtx object can be re-used multiple times.
+  A ZSTD_DCtx object can be reused multiple times.
 
   First typical operation is to retrieve frame parameters, using ZSTD_getFrameHeader().
   Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
   Data fragment must be large enough to ensure successful decoding.
  `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
-  @result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
-           >0 : `srcSize` is too small, please provide at least @result bytes on next attempt.
+  result  : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
+           >0 : `srcSize` is too small, please provide at least result bytes on next attempt.
            errorCode, which can be tested using ZSTD_isError().
 
   It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
@@ -2428,7 +2939,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
 
   The most memory efficient way is to use a round buffer of sufficient size.
   Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
-  which can @return an error code if required value is too large for current system (in 32-bits mode).
+  which can return an error code if required value is too large for current system (in 32-bits mode).
   In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
   up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
   which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
@@ -2448,7 +2959,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
   ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
   ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
 
- @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
   It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
   It can also be an error code, which can be tested with ZSTD_isError().
 
@@ -2471,27 +2982,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
 */
 
 /*=====   Buffer-less streaming decompression functions  =====*/
-typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
-typedef struct {
-    unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
-    unsigned long long windowSize;       /* can be very large, up to <= frameContentSize */
-    unsigned blockSizeMax;
-    ZSTD_frameType_e frameType;          /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
-    unsigned headerSize;
-    unsigned dictID;
-    unsigned checksumFlag;
-} ZSTD_frameHeader;
 
-/*! ZSTD_getFrameHeader() :
- *  decode Frame Header, or requires larger `srcSize`.
- * @return : 0, `zfhPtr` is correctly filled,
- *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
- *           or an error code, which can be tested using ZSTD_isError() */
-ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /*< doesn't consume input */
-/*! ZSTD_getFrameHeader_advanced() :
- *  same as ZSTD_getFrameHeader(),
- *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
-ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
 ZSTDLIB_STATIC_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /*< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
 
 ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
@@ -2502,6 +2993,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
 ZSTDLIB_STATIC_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 /* misc */
+ZSTD_DEPRECATED("This function will likely be removed in the next minor release. It is misleading and has very limited utility.")
 ZSTDLIB_STATIC_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
 typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
 ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
@@ -2509,11 +3001,23 @@ ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 
 
 
-/* ============================ */
-/*       Block level API       */
-/* ============================ */
+/* ========================================= */
+/*       Block level API (DEPRECATED)       */
+/* ========================================= */
 
 /*!
+
+    This API is deprecated in favor of the regular compression API.
+    You can get the frame header down to 2 bytes by setting:
+      - ZSTD_c_format = ZSTD_f_zstd1_magicless
+      - ZSTD_c_contentSizeFlag = 0
+      - ZSTD_c_checksumFlag = 0
+      - ZSTD_c_dictIDFlag = 0
+
+    This API is not as well tested as our normal API, so we recommend not using it.
+    We will be removing it in a future version. If the normal API doesn't provide
+    the functionality you need, please open a GitHub issue.
+
     Block functions produce and decode raw zstd blocks, without frame metadata.
     Frame metadata cost is typically ~12 bytes, which can be non-negligible for very small blocks (< 100 bytes).
     But users will have to take in charge needed metadata to regenerate data, such as compressed and content sizes.
@@ -2524,7 +3028,6 @@ ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
     - It is necessary to init context before starting
       + compression : any ZSTD_compressBegin*() variant, including with dictionary
       + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
-      + copyCCtx() and copyDCtx() can be used too
     - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
       + If input is larger than a block size, it's necessary to split input data into multiple blocks
       + For inputs larger than a single block, consider using regular ZSTD_compress() instead.
@@ -2541,11 +3044,14 @@ ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 */
 
 /*=====   Raw zstd block functions  =====*/
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
 ZSTDLIB_STATIC_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /*< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
 
-
 #endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
 
diff --git a/lib/zstd/Makefile b/lib/zstd/Makefile
index 20f08c644b71..464c410b2768 100644
--- a/lib/zstd/Makefile
+++ b/lib/zstd/Makefile
@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/common/allocations.h b/lib/zstd/common/allocations.h
new file mode 100644
index 000000000000..16c3d08e8d1a
--- /dev/null
+++ b/lib/zstd/common/allocations.h
@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This file provides custom allocation primitives
+ */
+
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"   /* ZSTD_malloc, ZSTD_calloc, ZSTD_free, ZSTD_memset */
+
+#include "compiler.h" /* MEM_STATIC */
+#define ZSTD_STATIC_LINKING_ONLY
+#include <linux/zstd.h> /* ZSTD_customMem */
+
+#ifndef ZSTD_ALLOCATIONS_H
+#define ZSTD_ALLOCATIONS_H
+
+/* custom memory allocation functions */
+
+MEM_STATIC void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc)
+        return customMem.customAlloc(customMem.opaque, size);
+    return ZSTD_malloc(size);
+}
+
+MEM_STATIC void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc) {
+        /* calloc implemented as malloc+memset;
+         * not as efficient as calloc, but next best guess for custom malloc */
+        void* const ptr = customMem.customAlloc(customMem.opaque, size);
+        ZSTD_memset(ptr, 0, size);
+        return ptr;
+    }
+    return ZSTD_calloc(1, size);
+}
+
+MEM_STATIC void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
+{
+    if (ptr!=NULL) {
+        if (customMem.customFree)
+            customMem.customFree(customMem.opaque, ptr);
+        else
+            ZSTD_free(ptr);
+    }
+}
+
+#endif /* ZSTD_ALLOCATIONS_H */
diff --git a/lib/zstd/common/bits.h b/lib/zstd/common/bits.h
new file mode 100644
index 000000000000..aa3487ec4b6a
--- /dev/null
+++ b/lib/zstd/common/bits.h
@@ -0,0 +1,149 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_BITS_H
+#define ZSTD_BITS_H
+
+#include "mem.h"
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32_fallback(U32 val)
+{
+    assert(val != 0);
+    {
+        static const U32 DeBruijnBytePos[32] = {0, 1, 28, 2, 29, 14, 24, 3,
+                                                30, 22, 20, 15, 25, 17, 4, 8,
+                                                31, 27, 13, 23, 21, 19, 16, 7,
+                                                26, 12, 18, 6, 11, 5, 10, 9};
+        return DeBruijnBytePos[((U32) ((val & -(S32) val) * 0x077CB531U)) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if (__GNUC__ >= 4)
+        return (unsigned)__builtin_ctz(val);
+#   else
+        return ZSTD_countTrailingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) {
+    assert(val != 0);
+    {
+        static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29,
+                                            11, 14, 16, 18, 22, 25, 3, 30,
+                                            8, 12, 20, 28, 15, 17, 24, 7,
+                                            19, 27, 23, 6, 26, 5, 4, 31};
+        val |= val >> 1;
+        val |= val >> 2;
+        val |= val >> 4;
+        val |= val >> 8;
+        val |= val >> 16;
+        return 31 - DeBruijnClz[(val * 0x07C4ACDDU) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if (__GNUC__ >= 4)
+        return (unsigned)__builtin_clz(val);
+#   else
+        return ZSTD_countLeadingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if (__GNUC__ >= 4) && defined(__LP64__)
+        return (unsigned)__builtin_ctzll(val);
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (leastSignificantWord == 0) {
+                return 32 + ZSTD_countTrailingZeros32(mostSignificantWord);
+            } else {
+                return ZSTD_countTrailingZeros32(leastSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if (__GNUC__ >= 4)
+        return (unsigned)(__builtin_clzll(val));
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (mostSignificantWord == 0) {
+                return 32 + ZSTD_countLeadingZeros32(leastSignificantWord);
+            } else {
+                return ZSTD_countLeadingZeros32(mostSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+            return ZSTD_countTrailingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countTrailingZeros32((U32)val) >> 3;
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+            return ZSTD_countLeadingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countLeadingZeros32((U32)val) >> 3;
+        }
+    }
+}
+
+MEM_STATIC unsigned ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
+{
+    assert(val != 0);
+    return 31 - ZSTD_countLeadingZeros32(val);
+}
+
+/* ZSTD_rotateRight_*():
+ * Rotates a bitfield to the right by "count" bits.
+ * https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
+ */
+MEM_STATIC
+U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
+    assert(count < 64);
+    count &= 0x3F; /* for fickle pattern recognition */
+    return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
+}
+
+MEM_STATIC
+U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
+    assert(count < 32);
+    count &= 0x1F; /* for fickle pattern recognition */
+    return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
+}
+
+MEM_STATIC
+U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
+    assert(count < 16);
+    count &= 0x0F; /* for fickle pattern recognition */
+    return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
+}
+
+#endif /* ZSTD_BITS_H */
diff --git a/lib/zstd/common/bitstream.h b/lib/zstd/common/bitstream.h
index feef3a1b1d60..6a13f1f0f1e8 100644
--- a/lib/zstd/common/bitstream.h
+++ b/lib/zstd/common/bitstream.h
@@ -1,7 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /* ******************************************************************
  * bitstream
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -27,6 +28,7 @@
 #include "compiler.h"       /* UNLIKELY() */
 #include "debug.h"          /* assert(), DEBUGLOG(), RAWLOG() */
 #include "error_private.h"  /* error codes and messages */
+#include "bits.h"           /* ZSTD_highbit32 */
 
 
 /*=========================================
@@ -79,19 +81,20 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
 /*-********************************************
 *  bitStream decoding API (read backward)
 **********************************************/
+typedef size_t BitContainerType;
 typedef struct {
-    size_t   bitContainer;
+    BitContainerType bitContainer;
     unsigned bitsConsumed;
     const char* ptr;
     const char* start;
     const char* limitPtr;
 } BIT_DStream_t;
 
-typedef enum { BIT_DStream_unfinished = 0,
-               BIT_DStream_endOfBuffer = 1,
-               BIT_DStream_completed = 2,
-               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
-               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+typedef enum { BIT_DStream_unfinished = 0,  /* fully refilled */
+               BIT_DStream_endOfBuffer = 1, /* still some bits left in bitstream */
+               BIT_DStream_completed = 2,   /* bitstream entirely consumed, bit-exact */
+               BIT_DStream_overflow = 3     /* user requested more bits than present in bitstream */
+    } BIT_DStream_status;  /* result of BIT_reloadDStream() */
 
 MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
 MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
@@ -101,7 +104,7 @@ MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
 
 /* Start by invoking BIT_initDStream().
 *  A chunk of the bitStream is then stored into a local register.
-*  Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+*  Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (BitContainerType).
 *  You can then retrieve bitFields stored into the local register, **in reverse order**.
 *  Local register is explicitly reloaded from memory by the BIT_reloadDStream() method.
 *  A reload guarantee a minimum of ((8*sizeof(bitD->bitContainer))-7) bits when its result is BIT_DStream_unfinished.
@@ -122,33 +125,6 @@ MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC);
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 /* faster, but works only if nbBits >= 1 */
 
-
-
-/*-**************************************************************
-*  Internal functions
-****************************************************************/
-MEM_STATIC unsigned BIT_highbit32 (U32 val)
-{
-    assert(val != 0);
-    {
-#   if (__GNUC__ >= 3)   /* Use GCC Intrinsic */
-        return __builtin_clz (val) ^ 31;
-#   else   /* Software version */
-        static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
-                                                 11, 14, 16, 18, 22, 25,  3, 30,
-                                                  8, 12, 20, 28, 15, 17, 24,  7,
-                                                 19, 27, 23,  6, 26,  5,  4, 31 };
-        U32 v = val;
-        v |= v >> 1;
-        v |= v >> 2;
-        v |= v >> 4;
-        v |= v >> 8;
-        v |= v >> 16;
-        return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
-#   endif
-    }
-}
-
 /*=====    Local Constants   =====*/
 static const unsigned BIT_mask[] = {
     0,          1,         3,         7,         0xF,       0x1F,
@@ -178,6 +154,12 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
     return 0;
 }
 
+FORCE_INLINE_TEMPLATE size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+{
+    assert(nbBits < BIT_MASK_SIZE);
+    return bitContainer & BIT_mask[nbBits];
+}
+
 /*! BIT_addBits() :
  *  can add up to 31 bits into `bitC`.
  *  Note : does not check for register overflow ! */
@@ -187,7 +169,7 @@ MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
     DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
     assert(nbBits < BIT_MASK_SIZE);
     assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
-    bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
+    bitC->bitContainer |= BIT_getLowerBits(value, nbBits) << bitC->bitPos;
     bitC->bitPos += nbBits;
 }
 
@@ -266,35 +248,35 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
         { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
+          bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
           if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
     } else {
         bitD->ptr   = bitD->start;
         bitD->bitContainer = *(const BYTE*)(bitD->start);
         switch(srcSize)
         {
-        case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
+        case 7: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
                 ZSTD_FALLTHROUGH;
 
-        case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
+        case 6: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
                 ZSTD_FALLTHROUGH;
 
-        case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
+        case 5: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
                 ZSTD_FALLTHROUGH;
 
-        case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
+        case 4: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[3]) << 24;
                 ZSTD_FALLTHROUGH;
 
-        case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
+        case 3: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[2]) << 16;
                 ZSTD_FALLTHROUGH;
 
-        case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
+        case 2: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[1]) <<  8;
                 ZSTD_FALLTHROUGH;
 
         default: break;
         }
         {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-            bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+            bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
             if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
         }
         bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
@@ -303,12 +285,12 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
     return srcSize;
 }
 
-MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
+FORCE_INLINE_TEMPLATE size_t BIT_getUpperBits(BitContainerType bitContainer, U32 const start)
 {
     return bitContainer >> start;
 }
 
-MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
+FORCE_INLINE_TEMPLATE size_t BIT_getMiddleBits(BitContainerType bitContainer, U32 const start, U32 const nbBits)
 {
     U32 const regMask = sizeof(bitContainer)*8 - 1;
     /* if start > regMask, bitstream is corrupted, and result is undefined */
@@ -325,19 +307,13 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 c
 #endif
 }
 
-MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
-{
-    assert(nbBits < BIT_MASK_SIZE);
-    return bitContainer & BIT_mask[nbBits];
-}
-
 /*! BIT_lookBits() :
  *  Provides next n bits from local register.
  *  local register is not modified.
  *  On 32-bits, maxNbBits==24.
  *  On 64-bits, maxNbBits==56.
  * @return : value extracted */
-MEM_STATIC  FORCE_INLINE_ATTR size_t BIT_lookBits(const BIT_DStream_t*  bitD, U32 nbBits)
+FORCE_INLINE_TEMPLATE size_t BIT_lookBits(const BIT_DStream_t*  bitD, U32 nbBits)
 {
     /* arbitrate between double-shift and shift+mask */
 #if 1
@@ -360,7 +336,7 @@ MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
     return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
 }
 
-MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+FORCE_INLINE_TEMPLATE void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
 {
     bitD->bitsConsumed += nbBits;
 }
@@ -369,7 +345,7 @@ MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
  *  Read (consume) next n bits from local register and update.
  *  Pay attention to not read more than nbBits contained into local register.
  * @return : extracted value. */
-MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
+FORCE_INLINE_TEMPLATE size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
     BIT_skipBits(bitD, nbBits);
@@ -377,7 +353,7 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned n
 }
 
 /*! BIT_readBitsFast() :
- *  unsafe version; only works only if nbBits >= 1 */
+ *  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBitsFast(bitD, nbBits);
@@ -386,6 +362,21 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
     return value;
 }
 
+/*! BIT_reloadDStream_internal() :
+ *  Simple variant of BIT_reloadDStream(), with two conditions:
+ *  1. bitstream is valid : bitsConsumed <= sizeof(bitD->bitContainer)*8
+ *  2. look window is valid after shifted down : bitD->ptr >= bitD->start
+ */
+MEM_STATIC BIT_DStream_status BIT_reloadDStream_internal(BIT_DStream_t* bitD)
+{
+    assert(bitD->bitsConsumed <= sizeof(bitD->bitContainer)*8);
+    bitD->ptr -= bitD->bitsConsumed >> 3;
+    assert(bitD->ptr >= bitD->start);
+    bitD->bitsConsumed &= 7;
+    bitD->bitContainer = MEM_readLEST(bitD->ptr);
+    return BIT_DStream_unfinished;
+}
+
 /*! BIT_reloadDStreamFast() :
  *  Similar to BIT_reloadDStream(), but with two differences:
  *  1. bitsConsumed <= sizeof(bitD->bitContainer)*8 must hold!
@@ -396,31 +387,35 @@ MEM_STATIC BIT_DStream_status BIT_reloadDStreamFast(BIT_DStream_t* bitD)
 {
     if (UNLIKELY(bitD->ptr < bitD->limitPtr))
         return BIT_DStream_overflow;
-    assert(bitD->bitsConsumed <= sizeof(bitD->bitContainer)*8);
-    bitD->ptr -= bitD->bitsConsumed >> 3;
-    bitD->bitsConsumed &= 7;
-    bitD->bitContainer = MEM_readLEST(bitD->ptr);
-    return BIT_DStream_unfinished;
+    return BIT_reloadDStream_internal(bitD);
 }
 
 /*! BIT_reloadDStream() :
  *  Refill `bitD` from buffer previously set in BIT_initDStream() .
- *  This function is safe, it guarantees it will not read beyond src buffer.
+ *  This function is safe, it guarantees it will not never beyond src buffer.
  * @return : status of `BIT_DStream_t` internal register.
  *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
-MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+FORCE_INLINE_TEMPLATE BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 {
-    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
+    /* note : once in overflow mode, a bitstream remains in this mode until it's reset */
+    if (UNLIKELY(bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))) {
+        static const BitContainerType zeroFilled = 0;
+        bitD->ptr = (const char*)&zeroFilled; /* aliasing is allowed for char */
+        /* overflow detected, erroneous scenario or end of stream: no update */
         return BIT_DStream_overflow;
+    }
+
+    assert(bitD->ptr >= bitD->start);
 
     if (bitD->ptr >= bitD->limitPtr) {
-        return BIT_reloadDStreamFast(bitD);
+        return BIT_reloadDStream_internal(bitD);
     }
     if (bitD->ptr == bitD->start) {
+        /* reached end of bitStream => no update */
         if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
         return BIT_DStream_completed;
     }
-    /* start < ptr < limitPtr */
+    /* start < ptr < limitPtr => cautious update */
     {   U32 nbBytes = bitD->bitsConsumed >> 3;
         BIT_DStream_status result = BIT_DStream_unfinished;
         if (bitD->ptr - nbBytes < bitD->start) {
diff --git a/lib/zstd/common/compiler.h b/lib/zstd/common/compiler.h
index c42d39faf9bd..508ee25537bb 100644
--- a/lib/zstd/common/compiler.h
+++ b/lib/zstd/common/compiler.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,6 +12,8 @@
 #ifndef ZSTD_COMPILER_H
 #define ZSTD_COMPILER_H
 
+#include <linux/types.h>
+
 #include "portability_macros.h"
 
 /*-*******************************************************
@@ -41,12 +44,15 @@
 */
 #define WIN_CDECL
 
+/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
+#define UNUSED_ATTR __attribute__((unused))
+
 /*
  * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
  * parameters. They must be inlined for the compiler to eliminate the constant
  * branches.
  */
-#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR UNUSED_ATTR
 /*
  * HINT_INLINE is used to help the compiler generate better code. It is *not*
  * used for "templates", so it can be tweaked based on the compilers
@@ -61,11 +67,21 @@
 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
 #  define HINT_INLINE static INLINE_KEYWORD
 #else
-#  define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
+#  define HINT_INLINE FORCE_INLINE_TEMPLATE
 #endif
 
-/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
-#define UNUSED_ATTR __attribute__((unused))
+/* "soft" inline :
+ * The compiler is free to select if it's a good idea to inline or not.
+ * The main objective is to silence compiler warnings
+ * when a defined function in included but not used.
+ *
+ * Note : this macro is prefixed `MEM_` because it used to be provided by `mem.h` unit.
+ * Updating the prefix is probably preferable, but requires a fairly large codemod,
+ * since this name is used everywhere.
+ */
+#ifndef MEM_STATIC  /* already defined in Linux Kernel mem.h */
+#define MEM_STATIC static __inline UNUSED_ATTR
+#endif
 
 /* force no inlining */
 #define FORCE_NOINLINE static __attribute__((__noinline__))
@@ -86,23 +102,24 @@
 #  define PREFETCH_L1(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
 #  define PREFETCH_L2(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
 #elif defined(__aarch64__)
-#  define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
-#  define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
+#  define PREFETCH_L1(ptr)  do { __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))); } while (0)
+#  define PREFETCH_L2(ptr)  do { __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))); } while (0)
 #else
-#  define PREFETCH_L1(ptr) (void)(ptr)  /* disabled */
-#  define PREFETCH_L2(ptr) (void)(ptr)  /* disabled */
+#  define PREFETCH_L1(ptr) do { (void)(ptr); } while (0)  /* disabled */
+#  define PREFETCH_L2(ptr) do { (void)(ptr); } while (0)  /* disabled */
 #endif  /* NO_PREFETCH */
 
 #define CACHELINE_SIZE 64
 
-#define PREFETCH_AREA(p, s)  {            \
-    const char* const _ptr = (const char*)(p);  \
-    size_t const _size = (size_t)(s);     \
-    size_t _pos;                          \
-    for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) {  \
-        PREFETCH_L2(_ptr + _pos);         \
-    }                                     \
-}
+#define PREFETCH_AREA(p, s)                              \
+    do {                                                 \
+        const char* const _ptr = (const char*)(p);       \
+        size_t const _size = (size_t)(s);                \
+        size_t _pos;                                     \
+        for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \
+            PREFETCH_L2(_ptr + _pos);                    \
+        }                                                \
+    } while (0)
 
 /* vectorization
  * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
@@ -126,9 +143,9 @@
 #define UNLIKELY(x) (__builtin_expect((x), 0))
 
 #if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)))
-#  define ZSTD_UNREACHABLE { assert(0), __builtin_unreachable(); }
+#  define ZSTD_UNREACHABLE do { assert(0), __builtin_unreachable(); } while (0)
 #else
-#  define ZSTD_UNREACHABLE { assert(0); }
+#  define ZSTD_UNREACHABLE do { assert(0); } while (0)
 #endif
 
 /* disable warnings */
@@ -179,6 +196,85 @@
 *  Sanitizer
 *****************************************************************/
 
+/*
+ * Zstd relies on pointer overflow in its decompressor.
+ * We add this attribute to functions that rely on pointer overflow.
+ */
+#ifndef ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+#  if __has_attribute(no_sanitize)
+#    if !defined(__clang__) && defined(__GNUC__) && __GNUC__ < 8
+       /* gcc < 8 only has signed-integer-overlow which triggers on pointer overflow */
+#      define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("signed-integer-overflow")))
+#    else
+       /* older versions of clang [3.7, 5.0) will warn that pointer-overflow is ignored. */
+#      define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("pointer-overflow")))
+#    endif
+#  else
+#    define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+#  endif
+#endif
+
+/*
+ * Helper function to perform a wrapped pointer difference without trigging
+ * UBSAN.
+ *
+ * @returns lhs - rhs with wrapping
+ */
+MEM_STATIC
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+ptrdiff_t ZSTD_wrappedPtrDiff(unsigned char const* lhs, unsigned char const* rhs)
+{
+    return lhs - rhs;
+}
+
+/*
+ * Helper function to perform a wrapped pointer add without triggering UBSAN.
+ *
+ * @return ptr + add with wrapping
+ */
+MEM_STATIC
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+unsigned char const* ZSTD_wrappedPtrAdd(unsigned char const* ptr, ptrdiff_t add)
+{
+    return ptr + add;
+}
+
+/*
+ * Helper function to perform a wrapped pointer subtraction without triggering
+ * UBSAN.
+ *
+ * @return ptr - sub with wrapping
+ */
+MEM_STATIC
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+unsigned char const* ZSTD_wrappedPtrSub(unsigned char const* ptr, ptrdiff_t sub)
+{
+    return ptr - sub;
+}
+
+/*
+ * Helper function to add to a pointer that works around C's undefined behavior
+ * of adding 0 to NULL.
+ *
+ * @returns `ptr + add` except it defines `NULL + 0 == NULL`.
+ */
+MEM_STATIC
+unsigned char* ZSTD_maybeNullPtrAdd(unsigned char* ptr, ptrdiff_t add)
+{
+    return add > 0 ? ptr + add : ptr;
+}
+
+/* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an
+ * abundance of caution, disable our custom poisoning on mingw. */
+#ifdef __MINGW32__
+#ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE
+#define ZSTD_ASAN_DONT_POISON_WORKSPACE 1
+#endif
+#ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE
+#define ZSTD_MSAN_DONT_POISON_WORKSPACE 1
+#endif
+#endif
+
 
 
 #endif /* ZSTD_COMPILER_H */
diff --git a/lib/zstd/common/cpu.h b/lib/zstd/common/cpu.h
index 0db7b42407ee..d8319a2bef4c 100644
--- a/lib/zstd/common/cpu.h
+++ b/lib/zstd/common/cpu.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/common/debug.c b/lib/zstd/common/debug.c
index bb863c9ea616..8eb6aa9a3b20 100644
--- a/lib/zstd/common/debug.c
+++ b/lib/zstd/common/debug.c
@@ -1,7 +1,8 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * debug
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -21,4 +22,10 @@
 
 #include "debug.h"
 
+#if (DEBUGLEVEL>=2)
+/* We only use this when DEBUGLEVEL>=2, but we get -Werror=pedantic errors if a
+ * translation unit is empty. So remove this from Linux kernel builds, but
+ * otherwise just leave it in.
+ */
 int g_debuglevel = DEBUGLEVEL;
+#endif
diff --git a/lib/zstd/common/debug.h b/lib/zstd/common/debug.h
index 6dd88d1fbd02..226ba3c57ec3 100644
--- a/lib/zstd/common/debug.h
+++ b/lib/zstd/common/debug.h
@@ -1,7 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /* ******************************************************************
  * debug
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -82,18 +83,27 @@ extern int g_debuglevel; /* the variable is only declared,
                             It's useful when enabling very verbose levels
                             on selective conditions (such as position in src) */
 
-#  define RAWLOG(l, ...) {                                       \
-                if (l<=g_debuglevel) {                           \
-                    ZSTD_DEBUG_PRINT(__VA_ARGS__);               \
-            }   }
-#  define DEBUGLOG(l, ...) {                                     \
-                if (l<=g_debuglevel) {                           \
-                    ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
-                    ZSTD_DEBUG_PRINT(" \n");                     \
-            }   }
+#  define RAWLOG(l, ...)                   \
+    do {                                   \
+        if (l<=g_debuglevel) {             \
+            ZSTD_DEBUG_PRINT(__VA_ARGS__); \
+        }                                  \
+    } while (0)
+
+#define STRINGIFY(x) #x
+#define TOSTRING(x) STRINGIFY(x)
+#define LINE_AS_STRING TOSTRING(__LINE__)
+
+#  define DEBUGLOG(l, ...)                               \
+    do {                                                 \
+        if (l<=g_debuglevel) {                           \
+            ZSTD_DEBUG_PRINT(__FILE__ ":" LINE_AS_STRING ": " __VA_ARGS__); \
+            ZSTD_DEBUG_PRINT(" \n");                     \
+        }                                                \
+    } while (0)
 #else
-#  define RAWLOG(l, ...)      {}    /* disabled */
-#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#  define RAWLOG(l, ...)   do { } while (0)    /* disabled */
+#  define DEBUGLOG(l, ...) do { } while (0)    /* disabled */
 #endif
 
 
diff --git a/lib/zstd/common/entropy_common.c b/lib/zstd/common/entropy_common.c
index fef67056f052..6cdd82233fb5 100644
--- a/lib/zstd/common/entropy_common.c
+++ b/lib/zstd/common/entropy_common.c
@@ -1,6 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * Common functions of New Generation Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -19,8 +20,8 @@
 #include "error_private.h"       /* ERR_*, ERROR */
 #define FSE_STATIC_LINKING_ONLY  /* FSE_MIN_TABLELOG */
 #include "fse.h"
-#define HUF_STATIC_LINKING_ONLY  /* HUF_TABLELOG_ABSOLUTEMAX */
 #include "huf.h"
+#include "bits.h"                /* ZSDT_highbit32, ZSTD_countTrailingZeros32 */
 
 
 /*===   Version   ===*/
@@ -38,23 +39,6 @@ const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-static U32 FSE_ctz(U32 val)
-{
-    assert(val != 0);
-    {
-#   if (__GNUC__ >= 3)   /* GCC Intrinsic */
-        return __builtin_ctz(val);
-#   else   /* Software version */
-        U32 count = 0;
-        while ((val & 1) == 0) {
-            val >>= 1;
-            ++count;
-        }
-        return count;
-#   endif
-    }
-}
-
 FORCE_INLINE_TEMPLATE
 size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
                            const void* headerBuffer, size_t hbSize)
@@ -102,7 +86,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
              * repeat.
              * Avoid UB by setting the high bit to 1.
              */
-            int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+            int repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
             while (repeats >= 12) {
                 charnum += 3 * 12;
                 if (LIKELY(ip <= iend-7)) {
@@ -113,7 +97,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
                     ip = iend - 4;
                 }
                 bitStream = MEM_readLE32(ip) >> bitCount;
-                repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+                repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
             }
             charnum += 3 * repeats;
             bitStream >>= 2 * repeats;
@@ -178,7 +162,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
                  * know that threshold > 1.
                  */
                 if (remaining <= 1) break;
-                nbBits = BIT_highbit32(remaining) + 1;
+                nbBits = ZSTD_highbit32(remaining) + 1;
                 threshold = 1 << (nbBits - 1);
             }
             if (charnum >= maxSV1) break;
@@ -253,7 +237,7 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      const void* src, size_t srcSize)
 {
     U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
-    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* bmi2 */ 0);
+    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* flags */ 0);
 }
 
 FORCE_INLINE_TEMPLATE size_t
@@ -301,14 +285,14 @@ HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
     if (weightTotal == 0) return ERROR(corruption_detected);
 
     /* get last non-null symbol weight (implied, total must be 2^n) */
-    {   U32 const tableLog = BIT_highbit32(weightTotal) + 1;
+    {   U32 const tableLog = ZSTD_highbit32(weightTotal) + 1;
         if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
         *tableLogPtr = tableLog;
         /* determine last weight */
         {   U32 const total = 1 << tableLog;
             U32 const rest = total - weightTotal;
-            U32 const verif = 1 << BIT_highbit32(rest);
-            U32 const lastWeight = BIT_highbit32(rest) + 1;
+            U32 const verif = 1 << ZSTD_highbit32(rest);
+            U32 const lastWeight = ZSTD_highbit32(rest) + 1;
             if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
             huffWeight[oSize] = (BYTE)lastWeight;
             rankStats[lastWeight]++;
@@ -345,13 +329,13 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize,
                      void* workSpace, size_t wkspSize,
-                     int bmi2)
+                     int flags)
 {
 #if DYNAMIC_BMI2
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
         return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
     }
 #endif
-    (void)bmi2;
+    (void)flags;
     return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
 }
diff --git a/lib/zstd/common/error_private.c b/lib/zstd/common/error_private.c
index 6d1135f8c373..a4062d30d170 100644
--- a/lib/zstd/common/error_private.c
+++ b/lib/zstd/common/error_private.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -27,9 +28,11 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(version_unsupported): return "Version not supported";
     case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
     case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
-    case PREFIX(corruption_detected): return "Corrupted block detected";
+    case PREFIX(corruption_detected): return "Data corruption detected";
     case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(literals_headerWrong): return "Header of Literals' block doesn't respect format specification";
     case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_combination_unsupported): return "Unsupported combination of parameters";
     case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
     case PREFIX(init_missing): return "Context should be init first";
     case PREFIX(memory_allocation): return "Allocation error : not enough memory";
@@ -38,17 +41,22 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
     case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
     case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
+    case PREFIX(stabilityCondition_notRespected): return "pledged buffer stability condition is not respected";
     case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
     case PREFIX(dictionary_wrong): return "Dictionary mismatch";
     case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
     case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
     case PREFIX(srcSize_wrong): return "Src size is incorrect";
     case PREFIX(dstBuffer_null): return "Operation on NULL destination buffer";
+    case PREFIX(noForwardProgress_destFull): return "Operation made no progress over multiple calls, due to output buffer being full";
+    case PREFIX(noForwardProgress_inputEmpty): return "Operation made no progress over multiple calls, due to input being empty";
         /* following error codes are not stable and may be removed or changed in a future version */
     case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
     case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
     case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
     case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
+    case PREFIX(sequenceProducer_failed): return "Block-level external sequence producer returned an error code";
+    case PREFIX(externalSequences_invalid): return "External sequences are not valid";
     case PREFIX(maxCode):
     default: return notErrorCode;
     }
diff --git a/lib/zstd/common/error_private.h b/lib/zstd/common/error_private.h
index ca5101e542fa..0410ca415b54 100644
--- a/lib/zstd/common/error_private.h
+++ b/lib/zstd/common/error_private.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -49,8 +50,13 @@ ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
 ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) return (ERR_enum)0; return (ERR_enum) (0-code); }
 
 /* check and forward error code */
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
-#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+#define CHECK_V_F(e, f)     \
+    size_t const e = f;     \
+    do {                    \
+        if (ERR_isError(e)) \
+            return e;       \
+    } while (0)
+#define CHECK_F(f)   do { CHECK_V_F(_var_err__, f); } while (0)
 
 
 /*-****************************************
@@ -84,10 +90,12 @@ void _force_has_format_string(const char *format, ...) {
  * We want to force this function invocation to be syntactically correct, but
  * we don't want to force runtime evaluation of its arguments.
  */
-#define _FORCE_HAS_FORMAT_STRING(...) \
-  if (0) { \
-    _force_has_format_string(__VA_ARGS__); \
-  }
+#define _FORCE_HAS_FORMAT_STRING(...)              \
+    do {                                           \
+        if (0) {                                   \
+            _force_has_format_string(__VA_ARGS__); \
+        }                                          \
+    } while (0)
 
 #define ERR_QUOTE(str) #str
 
@@ -98,48 +106,50 @@ void _force_has_format_string(const char *format, ...) {
  * In order to do that (particularly, printing the conditional that failed),
  * this can't just wrap RETURN_ERROR().
  */
-#define RETURN_ERROR_IF(cond, err, ...) \
-  if (cond) { \
-    RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
-           __FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
-    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-    RAWLOG(3, ": " __VA_ARGS__); \
-    RAWLOG(3, "\n"); \
-    return ERROR(err); \
-  }
+#define RETURN_ERROR_IF(cond, err, ...)                                        \
+    do {                                                                       \
+        if (cond) {                                                            \
+            RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s",          \
+                  __FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
+            _FORCE_HAS_FORMAT_STRING(__VA_ARGS__);                             \
+            RAWLOG(3, ": " __VA_ARGS__);                                       \
+            RAWLOG(3, "\n");                                                   \
+            return ERROR(err);                                                 \
+        }                                                                      \
+    } while (0)
 
 /*
  * Unconditionally return the specified error.
  *
  * In debug modes, prints additional information.
  */
-#define RETURN_ERROR(err, ...) \
-  do { \
-    RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
-           __FILE__, __LINE__, ERR_QUOTE(ERROR(err))); \
-    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-    RAWLOG(3, ": " __VA_ARGS__); \
-    RAWLOG(3, "\n"); \
-    return ERROR(err); \
-  } while(0);
+#define RETURN_ERROR(err, ...)                                               \
+    do {                                                                     \
+        RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
+              __FILE__, __LINE__, ERR_QUOTE(ERROR(err)));                    \
+        _FORCE_HAS_FORMAT_STRING(__VA_ARGS__);                               \
+        RAWLOG(3, ": " __VA_ARGS__);                                         \
+        RAWLOG(3, "\n");                                                     \
+        return ERROR(err);                                                   \
+    } while(0)
 
 /*
  * If the provided expression evaluates to an error code, returns that error code.
  *
  * In debug modes, prints additional information.
  */
-#define FORWARD_IF_ERROR(err, ...) \
-  do { \
-    size_t const err_code = (err); \
-    if (ERR_isError(err_code)) { \
-      RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
-             __FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
-      _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-      RAWLOG(3, ": " __VA_ARGS__); \
-      RAWLOG(3, "\n"); \
-      return err_code; \
-    } \
-  } while(0);
+#define FORWARD_IF_ERROR(err, ...)                                                 \
+    do {                                                                           \
+        size_t const err_code = (err);                                             \
+        if (ERR_isError(err_code)) {                                               \
+            RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s",                 \
+                  __FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
+            _FORCE_HAS_FORMAT_STRING(__VA_ARGS__);                                 \
+            RAWLOG(3, ": " __VA_ARGS__);                                           \
+            RAWLOG(3, "\n");                                                       \
+            return err_code;                                                       \
+        }                                                                          \
+    } while(0)
 
 
 #endif /* ERROR_H_MODULE */
diff --git a/lib/zstd/common/fse.h b/lib/zstd/common/fse.h
index 4507043b2287..2185a578617d 100644
--- a/lib/zstd/common/fse.h
+++ b/lib/zstd/common/fse.h
@@ -1,7 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /* ******************************************************************
  * FSE : Finite State Entropy codec
  * Public Prototypes declaration
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -50,34 +51,6 @@
 FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /*< library version number; to be used when checking dll version */
 
 
-/*-****************************************
-*  FSE simple functions
-******************************************/
-/*! FSE_compress() :
-    Compress content of buffer 'src', of size 'srcSize', into destination buffer 'dst'.
-    'dst' buffer must be already allocated. Compression runs faster is dstCapacity >= FSE_compressBound(srcSize).
-    @return : size of compressed data (<= dstCapacity).
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
-                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.
-                     if FSE_isError(return), compression failed (more details using FSE_getErrorName())
-*/
-FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,
-                             const void* src, size_t srcSize);
-
-/*! FSE_decompress():
-    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
-    into already allocated destination buffer 'dst', of size 'dstCapacity'.
-    @return : size of regenerated data (<= maxDstSize),
-              or an error code, which can be tested using FSE_isError() .
-
-    ** Important ** : FSE_decompress() does not decompress non-compressible nor RLE data !!!
-    Why ? : making this distinction requires a header.
-    Header management is intentionally delegated to the user layer, which can better manage special cases.
-*/
-FSE_PUBLIC_API size_t FSE_decompress(void* dst,  size_t dstCapacity,
-                               const void* cSrc, size_t cSrcSize);
-
-
 /*-*****************************************
 *  Tool functions
 ******************************************/
@@ -88,20 +61,6 @@ FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return
 FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
 
 
-/*-*****************************************
-*  FSE advanced functions
-******************************************/
-/*! FSE_compress2() :
-    Same as FSE_compress(), but allows the selection of 'maxSymbolValue' and 'tableLog'
-    Both parameters can be defined as '0' to mean : use default value
-    @return : size of compressed data
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within cSrc !!!
-                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.
-                     if FSE_isError(return), it's an error code.
-*/
-FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
-
-
 /*-*****************************************
 *  FSE detailed API
 ******************************************/
@@ -161,8 +120,6 @@ FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
 typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
-FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);
-FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
 
 /*! FSE_buildCTable():
     Builds `ct`, which must be already allocated, using FSE_createCTable().
@@ -238,23 +195,7 @@ FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
                            unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
                            const void* rBuffer, size_t rBuffSize, int bmi2);
 
-/*! Constructor and Destructor of FSE_DTable.
-    Note that its size depends on 'tableLog' */
 typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
-FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);
-FSE_PUBLIC_API void        FSE_freeDTable(FSE_DTable* dt);
-
-/*! FSE_buildDTable():
-    Builds 'dt', which must be already allocated, using FSE_createDTable().
-    return : 0, or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
-
-/*! FSE_decompress_usingDTable():
-    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
-    into `dst` which must be already allocated.
-    @return : size of regenerated data (necessarily <= `dstCapacity`),
-              or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
 
 /*!
 Tutorial :
@@ -286,6 +227,7 @@ If there is an error, the function will return an error code, which can be teste
 
 #endif  /* FSE_H */
 
+
 #if !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
 #define FSE_H_FSE_STATIC_LINKING_ONLY
 
@@ -317,16 +259,6 @@ If there is an error, the function will return an error code, which can be teste
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
 /*< same as FSE_optimalTableLog(), which used `minus==2` */
 
-/* FSE_compress_wksp() :
- * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
- * FSE_COMPRESS_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
- */
-#define FSE_COMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
-size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
-
-size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
-/*< build a fake FSE_CTable, designed for a flat distribution, where each symbol uses nbBits */
-
 size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
 /*< build a fake FSE_CTable, designed to compress always the same symbolValue */
 
@@ -344,19 +276,11 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
 FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 /*< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
 
-size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
-/*< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
-
-size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
-/*< build a fake FSE_DTable, designed to always generate the same symbolValue */
-
-#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + 1 + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
 #define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize);
-/*< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)` */
-
 size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
-/*< Same as FSE_decompress_wksp() but with dynamic BMI2 support. Pass 1 if your CPU supports BMI2 or 0 if it doesn't. */
+/*< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)`.
+ * Set bmi2 to 1 if your CPU supports BMI2 or 0 if it doesn't */
 
 typedef enum {
    FSE_repeat_none,  /*< Cannot use the previous table */
@@ -539,20 +463,20 @@ MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, un
     FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
     const U16* const stateTable = (const U16*)(statePtr->stateTable);
     U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
-    BIT_addBits(bitC, statePtr->value, nbBitsOut);
+    BIT_addBits(bitC,  (size_t)statePtr->value, nbBitsOut);
     statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
 }
 
 MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)
 {
-    BIT_addBits(bitC, statePtr->value, statePtr->stateLog);
+    BIT_addBits(bitC, (size_t)statePtr->value, statePtr->stateLog);
     BIT_flushBits(bitC);
 }
 
 
 /* FSE_getMaxNbBits() :
  * Approximate maximum cost of a symbol, in bits.
- * Fractional get rounded up (i.e : a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * Fractional get rounded up (i.e. a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
  * note 1 : assume symbolValue is valid (<= maxSymbolValue)
  * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
 MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
diff --git a/lib/zstd/common/fse_decompress.c b/lib/zstd/common/fse_decompress.c
index 8dcb8ca39767..3a17e84f27bf 100644
--- a/lib/zstd/common/fse_decompress.c
+++ b/lib/zstd/common/fse_decompress.c
@@ -1,6 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * FSE : Finite State Entropy decoder
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -22,8 +23,8 @@
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #include "error_private.h"
-#define ZSTD_DEPS_NEED_MALLOC
-#include "zstd_deps.h"
+#include "zstd_deps.h"  /* ZSTD_memcpy */
+#include "bits.h"       /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -55,19 +56,6 @@
 #define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
 #define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
 
-
-/* Function templates */
-FSE_DTable* FSE_createDTable (unsigned tableLog)
-{
-    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
-    return (FSE_DTable*)ZSTD_malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
-}
-
-void FSE_freeDTable (FSE_DTable* dt)
-{
-    ZSTD_free(dt);
-}
-
 static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
 {
     void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
@@ -96,7 +84,7 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
                     symbolNext[s] = 1;
                 } else {
                     if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
-                    symbolNext[s] = normalizedCounter[s];
+                    symbolNext[s] = (U16)normalizedCounter[s];
         }   }   }
         ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
     }
@@ -111,8 +99,7 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
          * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
          * our buffer to handle the over-write.
          */
-        {
-            U64 const add = 0x0101010101010101ull;
+        {   U64 const add = 0x0101010101010101ull;
             size_t pos = 0;
             U64 sv = 0;
             U32 s;
@@ -123,14 +110,13 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
                 for (i = 8; i < n; i += 8) {
                     MEM_write64(spread + pos + i, sv);
                 }
-                pos += n;
-            }
-        }
+                pos += (size_t)n;
+        }   }
         /* Now we spread those positions across the table.
-         * The benefit of doing it in two stages is that we avoid the the
+         * The benefit of doing it in two stages is that we avoid the
          * variable size inner loop, which caused lots of branch misses.
          * Now we can run through all the positions without any branch misses.
-         * We unroll the loop twice, since that is what emperically worked best.
+         * We unroll the loop twice, since that is what empirically worked best.
          */
         {
             size_t position = 0;
@@ -166,7 +152,7 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
         for (u=0; u<tableSize; u++) {
             FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
             U32 const nextState = symbolNext[symbol]++;
-            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
             tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
     }   }
 
@@ -184,49 +170,6 @@ size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsi
 /*-*******************************************************
 *  Decompression (Byte symbols)
 *********************************************************/
-size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
-{
-    void* ptr = dt;
-    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
-    void* dPtr = dt + 1;
-    FSE_decode_t* const cell = (FSE_decode_t*)dPtr;
-
-    DTableH->tableLog = 0;
-    DTableH->fastMode = 0;
-
-    cell->newState = 0;
-    cell->symbol = symbolValue;
-    cell->nbBits = 0;
-
-    return 0;
-}
-
-
-size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
-{
-    void* ptr = dt;
-    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
-    void* dPtr = dt + 1;
-    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
-    const unsigned tableSize = 1 << nbBits;
-    const unsigned tableMask = tableSize - 1;
-    const unsigned maxSV1 = tableMask+1;
-    unsigned s;
-
-    /* Sanity checks */
-    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
-
-    /* Build Decoding Table */
-    DTableH->tableLog = (U16)nbBits;
-    DTableH->fastMode = 1;
-    for (s=0; s<maxSV1; s++) {
-        dinfo[s].newState = 0;
-        dinfo[s].symbol = (BYTE)s;
-        dinfo[s].nbBits = (BYTE)nbBits;
-    }
-
-    return 0;
-}
 
 FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
           void* dst, size_t maxDstSize,
@@ -287,32 +230,12 @@ FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
             break;
     }   }
 
-    return op-ostart;
-}
-
-
-size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
-                            const void* cSrc, size_t cSrcSize,
-                            const FSE_DTable* dt)
-{
-    const void* ptr = dt;
-    const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
-    const U32 fastMode = DTableH->fastMode;
-
-    /* select fast mode (static) */
-    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
-    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
-}
-
-
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
-{
-    return FSE_decompress_wksp_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, /* bmi2 */ 0);
+    assert(op >= ostart);
+    return (size_t)(op-ostart);
 }
 
 typedef struct {
     short ncount[FSE_MAX_SYMBOL_VALUE + 1];
-    FSE_DTable dtable[]; /* Dynamically sized */
 } FSE_DecompressWksp;
 
 
@@ -327,13 +250,18 @@ FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
     unsigned tableLog;
     unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
     FSE_DecompressWksp* const wksp = (FSE_DecompressWksp*)workSpace;
+    size_t const dtablePos = sizeof(FSE_DecompressWksp) / sizeof(FSE_DTable);
+    FSE_DTable* const dtable = (FSE_DTable*)workSpace + dtablePos;
 
-    DEBUG_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
+    FSE_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
     if (wkspSize < sizeof(*wksp)) return ERROR(GENERIC);
 
+    /* correct offset to dtable depends on this property */
+    FSE_STATIC_ASSERT(sizeof(FSE_DecompressWksp) % sizeof(FSE_DTable) == 0);
+
     /* normal FSE decoding mode */
-    {
-        size_t const NCountLength = FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
+    {   size_t const NCountLength =
+            FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
         if (FSE_isError(NCountLength)) return NCountLength;
         if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
         assert(NCountLength <= cSrcSize);
@@ -342,19 +270,20 @@ FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
     }
 
     if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
-    workSpace = wksp->dtable + FSE_DTABLE_SIZE_U32(tableLog);
+    assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
+    workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
     wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
 
-    CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
+    CHECK_F( FSE_buildDTable_internal(dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
 
     {
-        const void* ptr = wksp->dtable;
+        const void* ptr = dtable;
         const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
         const U32 fastMode = DTableH->fastMode;
 
         /* select fast mode (static) */
-        if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 1);
-        return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 0);
+        if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, dtable, 1);
+        return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, dtable, 0);
     }
 }
 
@@ -382,9 +311,4 @@ size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc,
     return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
 }
 
-
-typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
-
-
-
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/zstd/common/huf.h b/lib/zstd/common/huf.h
index 5042ff870308..57462466e188 100644
--- a/lib/zstd/common/huf.h
+++ b/lib/zstd/common/huf.h
@@ -1,7 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /* ******************************************************************
  * huff0 huffman codec,
  * part of Finite State Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -18,99 +19,22 @@
 
 /* *** Dependencies *** */
 #include "zstd_deps.h"    /* size_t */
-
-
-/* *** library symbols visibility *** */
-/* Note : when linking with -fvisibility=hidden on gcc, or by default on Visual,
- *        HUF symbols remain "private" (internal symbols for library only).
- *        Set macro FSE_DLL_EXPORT to 1 if you want HUF symbols visible on DLL interface */
-#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
-#  define HUF_PUBLIC_API __attribute__ ((visibility ("default")))
-#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
-#  define HUF_PUBLIC_API __declspec(dllexport)
-#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
-#  define HUF_PUBLIC_API __declspec(dllimport)  /* not required, just to generate faster code (saves a function pointer load from IAT and an indirect jump) */
-#else
-#  define HUF_PUBLIC_API
-#endif
-
-
-/* ========================== */
-/* ***  simple functions  *** */
-/* ========================== */
-
-/* HUF_compress() :
- *  Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
- * 'dst' buffer must be already allocated.
- *  Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
- * `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
- * @return : size of compressed data (<= `dstCapacity`).
- *  Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
- *                   if HUF_isError(return), compression failed (more details using HUF_getErrorName())
- */
-HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
-                             const void* src, size_t srcSize);
-
-/* HUF_decompress() :
- *  Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
- *  into already allocated buffer 'dst', of minimum size 'dstSize'.
- * `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
- *  Note : in contrast with FSE, HUF_decompress can regenerate
- *         RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
- *         because it knows size to regenerate (originalSize).
- * @return : size of regenerated data (== originalSize),
- *           or an error code, which can be tested using HUF_isError()
- */
-HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
-                               const void* cSrc, size_t cSrcSize);
+#include "mem.h"          /* U32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
 
 
 /* ***   Tool functions *** */
-#define HUF_BLOCKSIZE_MAX (128 * 1024)                  /*< maximum input size for a single block compressed with HUF_compress */
-HUF_PUBLIC_API size_t HUF_compressBound(size_t size);   /*< maximum compressed size (worst case) */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)   /*< maximum input size for a single block compressed with HUF_compress */
+size_t HUF_compressBound(size_t size);   /*< maximum compressed size (worst case) */
 
 /* Error Management */
-HUF_PUBLIC_API unsigned    HUF_isError(size_t code);       /*< tells if a return value is an error code */
-HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /*< provides error code string (useful for debugging) */
-
+unsigned    HUF_isError(size_t code);       /*< tells if a return value is an error code */
+const char* HUF_getErrorName(size_t code);  /*< provides error code string (useful for debugging) */
 
-/* ***   Advanced function   *** */
 
-/* HUF_compress2() :
- *  Same as HUF_compress(), but offers control over `maxSymbolValue` and `tableLog`.
- * `maxSymbolValue` must be <= HUF_SYMBOLVALUE_MAX .
- * `tableLog` must be `<= HUF_TABLELOG_MAX` . */
-HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
-                               const void* src, size_t srcSize,
-                               unsigned maxSymbolValue, unsigned tableLog);
-
-/* HUF_compress4X_wksp() :
- *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
- * `workspace` must be at least as large as HUF_WORKSPACE_SIZE */
 #define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
 #define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
-HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     unsigned maxSymbolValue, unsigned tableLog,
-                                     void* workSpace, size_t wkspSize);
-
-#endif   /* HUF_H_298734234 */
-
-/* ******************************************************************
- *  WARNING !!
- *  The following section contains advanced and experimental definitions
- *  which shall never be used in the context of a dynamic library,
- *  because they are not guaranteed to remain stable in the future.
- *  Only consider them in association with static linking.
- * *****************************************************************/
-#if !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
-#define HUF_H_HUF_STATIC_LINKING_ONLY
-
-/* *** Dependencies *** */
-#include "mem.h"   /* U32 */
-#define FSE_STATIC_LINKING_ONLY
-#include "fse.h"
-
 
 /* *** Constants *** */
 #define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
@@ -151,25 +75,49 @@ typedef U32 HUF_DTable;
 /* ****************************************
 *  Advanced decompression functions
 ******************************************/
-size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< double-symbols decoder */
-#endif
 
-size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< decodes RLE and uncompressed */
-size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /*< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /*< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< single-symbol decoder */
-size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /*< single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< double-symbols decoder */
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /*< double-symbols decoder */
-#endif
+/*
+ * Huffman flags bitset.
+ * For all flags, 0 is the default value.
+ */
+typedef enum {
+    /*
+     * If compiled with DYNAMIC_BMI2: Set flag only if the CPU supports BMI2 at runtime.
+     * Otherwise: Ignored.
+     */
+    HUF_flags_bmi2 = (1 << 0),
+    /*
+     * If set: Test possible table depths to find the one that produces the smallest header + encoded size.
+     * If unset: Use heuristic to find the table depth.
+     */
+    HUF_flags_optimalDepth = (1 << 1),
+    /*
+     * If set: If the previous table can encode the input, always reuse the previous table.
+     * If unset: If the previous table can encode the input, reuse the previous table if it results in a smaller output.
+     */
+    HUF_flags_preferRepeat = (1 << 2),
+    /*
+     * If set: Sample the input and check if the sample is uncompressible, if it is then don't attempt to compress.
+     * If unset: Always histogram the entire input.
+     */
+    HUF_flags_suspectUncompressible = (1 << 3),
+    /*
+     * If set: Don't use assembly implementations
+     * If unset: Allow using assembly implementations
+     */
+    HUF_flags_disableAsm = (1 << 4),
+    /*
+     * If set: Don't use the fast decoding loop, always use the fallback decoding loop.
+     * If unset: Use the fast decoding loop when possible.
+     */
+    HUF_flags_disableFast = (1 << 5)
+} HUF_flags_e;
 
 
 /* ****************************************
  *  HUF detailed API
  * ****************************************/
+#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
 
 /*! HUF_compress() does the following:
  *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
@@ -182,12 +130,12 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
  *  For example, it's possible to compress several blocks using the same 'CTable',
  *  or to save and regenerate 'CTable' using external methods.
  */
-unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
-size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);   /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
-size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
+unsigned HUF_minTableLog(unsigned symbolCardinality);
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue);
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace,
+ size_t wkspSize, HUF_CElt* table, const unsigned* count, int flags); /* table is used as scratch space for building and testing tables, not a return value */
 size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
-size_t HUF_compress4X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2);
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
 size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
 int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
 
@@ -196,6 +144,7 @@ typedef enum {
    HUF_repeat_check, /*< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
    HUF_repeat_valid  /*< Can use the previous table and it is assumed to be valid */
  } HUF_repeat;
+
 /* HUF_compress4X_repeat() :
  *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
  *  If it uses hufTable it does not modify hufTable or repeat.
@@ -206,13 +155,13 @@ size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
                        const void* src, size_t srcSize,
                        unsigned maxSymbolValue, unsigned tableLog,
                        void* workSpace, size_t wkspSize,    /*< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
-                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible);
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
 
 /* HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
  * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
  */
-#define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 ((4 * (HUF_SYMBOLVALUE_MAX + 1)) + 192)
 #define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
 size_t HUF_buildCTable_wksp (HUF_CElt* tree,
                        const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
@@ -238,7 +187,7 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
                           U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                           const void* src, size_t srcSize,
                           void* workspace, size_t wkspSize,
-                          int bmi2);
+                          int flags);
 
 /* HUF_readCTable() :
  *  Loading a CTable saved with HUF_writeCTable() */
@@ -246,9 +195,22 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
 
 /* HUF_getNbBitsFromCTable() :
  *  Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
- *  Note 1 : is not inlined, as HUF_CElt definition is private */
+ *  Note 1 : If symbolValue > HUF_readCTableHeader(symbolTable).maxSymbolValue, returns 0
+ *  Note 2 : is not inlined, as HUF_CElt definition is private
+ */
 U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
 
+typedef struct {
+    BYTE tableLog;
+    BYTE maxSymbolValue;
+    BYTE unused[sizeof(size_t) - 2];
+} HUF_CTableHeader;
+
+/* HUF_readCTableHeader() :
+ *  @returns The header from the CTable specifying the tableLog and the maxSymbolValue.
+ */
+HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable);
+
 /*
  * HUF_decompress() does the following:
  * 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
@@ -276,32 +238,12 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 #define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
 #define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
 
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_readDTableX1 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX1_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-#endif
-
-size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress4X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
-
 
 /* ====================== */
 /* single stream variants */
 /* ====================== */
 
-size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
-size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /*< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U64 U64 */
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
-size_t HUF_compress1X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2);
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
 /* HUF_compress1X_repeat() :
  *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
  *  If it uses hufTable it does not modify hufTable or repeat.
@@ -312,47 +254,28 @@ size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
                        const void* src, size_t srcSize,
                        unsigned maxSymbolValue, unsigned tableLog,
                        void* workSpace, size_t wkspSize,   /*< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
-                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible);
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
 
-size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
-#endif
-
-size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
-size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< single-symbol decoder */
-size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /*< single-symbol decoder */
-#endif
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /*< double-symbols decoder */
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /*< double-symbols decoder */
-#endif
-
-size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /*< automatic selection of sing or double symbol decoder, based on DTable */
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);   /*< double-symbols decoder */
 #endif
 
 /* BMI2 variants.
  * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
  */
-size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #endif
-size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
-size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
 #endif
 #ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_readDTableX2_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
 #endif
 
-#endif /* HUF_STATIC_LINKING_ONLY */
+#endif   /* HUF_H_298734234 */
 
diff --git a/lib/zstd/common/mem.h b/lib/zstd/common/mem.h
index 1d9cc03924ca..2e91e7780c1f 100644
--- a/lib/zstd/common/mem.h
+++ b/lib/zstd/common/mem.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -24,6 +24,7 @@
 /*-****************************************
 *  Compiler specifics
 ******************************************/
+#undef MEM_STATIC /* may be already defined from common/compiler.h */
 #define MEM_STATIC static inline
 
 /*-**************************************************************
diff --git a/lib/zstd/common/portability_macros.h b/lib/zstd/common/portability_macros.h
index 0e3b2c0a527d..f08638cced6c 100644
--- a/lib/zstd/common/portability_macros.h
+++ b/lib/zstd/common/portability_macros.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,7 +13,7 @@
 #define ZSTD_PORTABILITY_MACROS_H
 
 /*
- * This header file contains macro defintions to support portability.
+ * This header file contains macro definitions to support portability.
  * This header is shared between C and ASM code, so it MUST only
  * contain macro definitions. It MUST not contain any C code.
  *
@@ -45,6 +46,8 @@
 /* Mark the internal assembly functions as hidden  */
 #ifdef __ELF__
 # define ZSTD_HIDE_ASM_FUNCTION(func) .hidden func
+#elif defined(__APPLE__)
+# define ZSTD_HIDE_ASM_FUNCTION(func) .private_extern func
 #else
 # define ZSTD_HIDE_ASM_FUNCTION(func)
 #endif
@@ -65,7 +68,7 @@
 #endif
 
 /*
- * Only enable assembly for GNUC comptabile compilers,
+ * Only enable assembly for GNUC compatible compilers,
  * because other platforms may not support GAS assembly syntax.
  *
  * Only enable assembly for Linux / MacOS, other platforms may
@@ -90,4 +93,23 @@
  */
 #define ZSTD_ENABLE_ASM_X86_64_BMI2 0
 
+/*
+ * For x86 ELF targets, add .note.gnu.property section for Intel CET in
+ * assembly sources when CET is enabled.
+ *
+ * Additionally, any function that may be called indirectly must begin
+ * with ZSTD_CET_ENDBRANCH.
+ */
+#if defined(__ELF__) && (defined(__x86_64__) || defined(__i386__)) \
+    && defined(__has_include)
+# if __has_include(<cet.h>)
+#  include <cet.h>
+#  define ZSTD_CET_ENDBRANCH _CET_ENDBR
+# endif
+#endif
+
+#ifndef ZSTD_CET_ENDBRANCH
+# define ZSTD_CET_ENDBRANCH
+#endif
+
 #endif /* ZSTD_PORTABILITY_MACROS_H */
diff --git a/lib/zstd/common/zstd_common.c b/lib/zstd/common/zstd_common.c
index 3d7e35b309b5..44b95b25344a 100644
--- a/lib/zstd/common/zstd_common.c
+++ b/lib/zstd/common/zstd_common.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,7 +15,6 @@
 *  Dependencies
 ***************************************/
 #define ZSTD_DEPS_NEED_MALLOC
-#include "zstd_deps.h"   /* ZSTD_malloc, ZSTD_calloc, ZSTD_free, ZSTD_memset */
 #include "error_private.h"
 #include "zstd_internal.h"
 
@@ -47,37 +47,3 @@ ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
 /*! ZSTD_getErrorString() :
  *  provides error code string from enum */
 const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
-
-
-
-/*=**************************************************************
-*  Custom allocator
-****************************************************************/
-void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
-{
-    if (customMem.customAlloc)
-        return customMem.customAlloc(customMem.opaque, size);
-    return ZSTD_malloc(size);
-}
-
-void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
-{
-    if (customMem.customAlloc) {
-        /* calloc implemented as malloc+memset;
-         * not as efficient as calloc, but next best guess for custom malloc */
-        void* const ptr = customMem.customAlloc(customMem.opaque, size);
-        ZSTD_memset(ptr, 0, size);
-        return ptr;
-    }
-    return ZSTD_calloc(1, size);
-}
-
-void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
-{
-    if (ptr!=NULL) {
-        if (customMem.customFree)
-            customMem.customFree(customMem.opaque, ptr);
-        else
-            ZSTD_free(ptr);
-    }
-}
diff --git a/lib/zstd/common/zstd_deps.h b/lib/zstd/common/zstd_deps.h
index 2c34e8a33a1c..f931f7d0e294 100644
--- a/lib/zstd/common/zstd_deps.h
+++ b/lib/zstd/common/zstd_deps.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -105,3 +105,17 @@ static uint64_t ZSTD_div64(uint64_t dividend, uint32_t divisor) {
 
 #endif /* ZSTD_DEPS_IO */
 #endif /* ZSTD_DEPS_NEED_IO */
+
+/*
+ * Only requested when MSAN is enabled.
+ * Need:
+ * intptr_t
+ */
+#ifdef ZSTD_DEPS_NEED_STDINT
+#ifndef ZSTD_DEPS_STDINT
+#define ZSTD_DEPS_STDINT
+
+/* intptr_t already provided by ZSTD_DEPS_COMMON */
+
+#endif /* ZSTD_DEPS_STDINT */
+#endif /* ZSTD_DEPS_NEED_STDINT */
diff --git a/lib/zstd/common/zstd_internal.h b/lib/zstd/common/zstd_internal.h
index 93305d9b41bb..11da1233e890 100644
--- a/lib/zstd/common/zstd_internal.h
+++ b/lib/zstd/common/zstd_internal.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -28,7 +29,6 @@
 #include <linux/zstd.h>
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
 #include <linux/xxhash.h>                /* XXH_reset, update, digest */
 #define ZSTD_TRACE 0
@@ -83,9 +83,9 @@ typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
 #define ZSTD_FRAMECHECKSUMSIZE 4
 
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
-#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */)   /* for a non-null block */
+#define MIN_LITERALS_FOR_4_STREAMS 6
 
-#define HufLog 12
 typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingType_e;
 
 #define LONGNBSEQ 0x7F00
@@ -93,6 +93,7 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 #define MINMATCH 3
 
 #define Litbits  8
+#define LitHufLog 11
 #define MaxLit ((1<<Litbits) - 1)
 #define MaxML   52
 #define MaxLL   35
@@ -103,6 +104,8 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 #define LLFSELog    9
 #define OffFSELog   8
 #define MaxFSELog  MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
+#define MaxMLBits 16
+#define MaxLLBits 16
 
 #define ZSTD_MAX_HUF_HEADER_SIZE 128 /* header + <= 127 byte tree description */
 /* Each table cannot take more than #symbols * FSELog bits */
@@ -166,7 +169,7 @@ static void ZSTD_copy8(void* dst, const void* src) {
     ZSTD_memcpy(dst, src, 8);
 #endif
 }
-#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+#define COPY8(d,s) do { ZSTD_copy8(d,s); d+=8; s+=8; } while (0)
 
 /* Need to use memmove here since the literal buffer can now be located within
    the dst buffer. In circumstances where the op "catches up" to where the
@@ -186,7 +189,7 @@ static void ZSTD_copy16(void* dst, const void* src) {
     ZSTD_memcpy(dst, copy16_buf, 16);
 #endif
 }
-#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
+#define COPY16(d,s) do { ZSTD_copy16(d,s); d+=16; s+=16; } while (0)
 
 #define WILDCOPY_OVERLENGTH 32
 #define WILDCOPY_VECLEN 16
@@ -215,7 +218,7 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
     if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
         /* Handle short offset copies. */
         do {
-            COPY8(op, ip)
+            COPY8(op, ip);
         } while (op < oend);
     } else {
         assert(diff >= WILDCOPY_VECLEN || diff <= -WILDCOPY_VECLEN);
@@ -225,12 +228,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
          * one COPY16() in the first call. Then, do two calls per loop since
          * at that point it is more likely to have a high trip count.
          */
-#ifdef __aarch64__
-        do {
-            COPY16(op, ip);
-        }
-        while (op < oend);
-#else
         ZSTD_copy16(op, ip);
         if (16 >= length) return;
         op += 16;
@@ -240,7 +237,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
             COPY16(op, ip);
         }
         while (op < oend);
-#endif
     }
 }
 
@@ -289,11 +285,11 @@ typedef enum {
 typedef struct {
     seqDef* sequencesStart;
     seqDef* sequences;      /* ptr to end of sequences */
-    BYTE* litStart;
-    BYTE* lit;              /* ptr to end of literals */
-    BYTE* llCode;
-    BYTE* mlCode;
-    BYTE* ofCode;
+    BYTE*  litStart;
+    BYTE*  lit;             /* ptr to end of literals */
+    BYTE*  llCode;
+    BYTE*  mlCode;
+    BYTE*  ofCode;
     size_t maxNbSeq;
     size_t maxNbLit;
 
@@ -301,8 +297,8 @@ typedef struct {
      * in the seqStore that has a value larger than U16 (if it exists). To do so, we increment
      * the existing value of the litLength or matchLength by 0x10000.
      */
-    ZSTD_longLengthType_e   longLengthType;
-    U32                     longLengthPos;  /* Index of the sequence to apply long length modification to */
+    ZSTD_longLengthType_e longLengthType;
+    U32                   longLengthPos;  /* Index of the sequence to apply long length modification to */
 } seqStore_t;
 
 typedef struct {
@@ -321,10 +317,10 @@ MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore
     seqLen.matchLength = seq->mlBase + MINMATCH;
     if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
         if (seqStore->longLengthType == ZSTD_llt_literalLength) {
-            seqLen.litLength += 0xFFFF;
+            seqLen.litLength += 0x10000;
         }
         if (seqStore->longLengthType == ZSTD_llt_matchLength) {
-            seqLen.matchLength += 0xFFFF;
+            seqLen.matchLength += 0x10000;
         }
     }
     return seqLen;
@@ -337,72 +333,13 @@ MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore
  *          `decompressedBound != ZSTD_CONTENTSIZE_ERROR`
  */
 typedef struct {
+    size_t nbBlocks;
     size_t compressedSize;
     unsigned long long decompressedBound;
 } ZSTD_frameSizeInfo;   /* decompress & legacy */
 
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
-void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
-
-/* custom memory allocation functions */
-void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem);
-void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem);
-void ZSTD_customFree(void* ptr, ZSTD_customMem customMem);
-
-
-MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
-{
-    assert(val != 0);
-    {
-#   if (__GNUC__ >= 3)   /* GCC Intrinsic */
-        return __builtin_clz (val) ^ 31;
-#   else   /* Software version */
-        static const U32 DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
-        U32 v = val;
-        v |= v >> 1;
-        v |= v >> 2;
-        v |= v >> 4;
-        v |= v >> 8;
-        v |= v >> 16;
-        return DeBruijnClz[(v * 0x07C4ACDDU) >> 27];
-#   endif
-    }
-}
-
-/*
- * Counts the number of trailing zeros of a `size_t`.
- * Most compilers should support CTZ as a builtin. A backup
- * implementation is provided if the builtin isn't supported, but
- * it may not be terribly efficient.
- */
-MEM_STATIC unsigned ZSTD_countTrailingZeros(size_t val)
-{
-    if (MEM_64bits()) {
-#       if (__GNUC__ >= 4)
-            return __builtin_ctzll((U64)val);
-#       else
-            static const int DeBruijnBytePos[64] = {  0,  1,  2,  7,  3, 13,  8, 19,
-                                                      4, 25, 14, 28,  9, 34, 20, 56,
-                                                      5, 17, 26, 54, 15, 41, 29, 43,
-                                                      10, 31, 38, 35, 21, 45, 49, 57,
-                                                      63,  6, 12, 18, 24, 27, 33, 55,
-                                                      16, 53, 40, 42, 30, 37, 44, 48,
-                                                      62, 11, 23, 32, 52, 39, 36, 47,
-                                                      61, 22, 51, 46, 60, 50, 59, 58 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-    } else { /* 32 bits */
-#       if (__GNUC__ >= 3)
-            return __builtin_ctz((U32)val);
-#       else
-            static const int DeBruijnBytePos[32] = {  0,  1, 28,  2, 29, 14, 24,  3,
-                                                     30, 22, 20, 15, 25, 17,  4,  8,
-                                                     31, 27, 13, 23, 21, 19, 16,  7,
-                                                     26, 12, 18,  6, 11,  5, 10,  9 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-    }
-}
+int ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
 
 
 /* ZSTD_invalidateRepCodes() :
@@ -420,13 +357,13 @@ typedef struct {
 
 /*! ZSTD_getcBlockSize() :
  *  Provides the size of compressed block from block header `src` */
-/* Used by: decompress, fullbench (does not get its definition from here) */
+/*  Used by: decompress, fullbench */
 size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
                           blockProperties_t* bpPtr);
 
 /*! ZSTD_decodeSeqHeaders() :
  *  decode sequence header from src */
-/* Used by: decompress, fullbench (does not get its definition from here) */
+/*  Used by: zstd_decompress_block, fullbench */
 size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                        const void* src, size_t srcSize);
 
diff --git a/lib/zstd/compress/clevels.h b/lib/zstd/compress/clevels.h
index d9a76112ec3a..6ab8be6532ef 100644
--- a/lib/zstd/compress/clevels.h
+++ b/lib/zstd/compress/clevels.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/compress/fse_compress.c b/lib/zstd/compress/fse_compress.c
index ec5b1ca6d71a..44a3c10becf2 100644
--- a/lib/zstd/compress/fse_compress.c
+++ b/lib/zstd/compress/fse_compress.c
@@ -1,6 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * FSE : Finite State Entropy encoder
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -25,7 +26,8 @@
 #include "../common/error_private.h"
 #define ZSTD_DEPS_NEED_MALLOC
 #define ZSTD_DEPS_NEED_MATH64
-#include "../common/zstd_deps.h"  /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
+#include "../common/zstd_deps.h"  /* ZSTD_memset */
+#include "../common/bits.h" /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -90,7 +92,7 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     assert(tableLog < 16);   /* required for threshold strategy to work */
 
     /* For explanations on how to distribute symbol values over the table :
-     * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+     * https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
 
      #ifdef __clang_analyzer__
      ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
@@ -191,7 +193,7 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
                 break;
             default :
                 assert(normalizedCounter[s] > 1);
-                {   U32 const maxBitsOut = tableLog - BIT_highbit32 ((U32)normalizedCounter[s]-1);
+                {   U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
                     U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
                     symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
                     symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
@@ -224,8 +226,8 @@ size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
     size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
                                    + 4 /* bitCount initialized at 4 */
                                    + 2 /* first two symbols may use one additional bit each */) / 8)
-                                    + 1 /* round up to whole nb bytes */
-                                    + 2 /* additional two bytes for bitstream flush */;
+                                   + 1 /* round up to whole nb bytes */
+                                   + 2 /* additional two bytes for bitstream flush */;
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
@@ -254,7 +256,7 @@ FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     /* Init */
     remaining = tableSize+1;   /* +1 for extra accuracy */
     threshold = tableSize;
-    nbBits = tableLog+1;
+    nbBits = (int)tableLog+1;
 
     while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
         if (previousIs0) {
@@ -273,7 +275,7 @@ FSE_writeNCount_generic (void* header, size_t headerBufferSize,
             }
             while (symbol >= start+3) {
                 start+=3;
-                bitStream += 3 << bitCount;
+                bitStream += 3U << bitCount;
                 bitCount += 2;
             }
             bitStream += (symbol-start) << bitCount;
@@ -293,7 +295,7 @@ FSE_writeNCount_generic (void* header, size_t headerBufferSize,
             count++;   /* +1 for extra accuracy */
             if (count>=threshold)
                 count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
-            bitStream += count << bitCount;
+            bitStream += (U32)count << bitCount;
             bitCount  += nbBits;
             bitCount  -= (count<max);
             previousIs0  = (count==1);
@@ -321,7 +323,8 @@ FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     out[1] = (BYTE)(bitStream>>8);
     out+= (bitCount+7) /8;
 
-    return (out-ostart);
+    assert(out >= ostart);
+    return (size_t)(out-ostart);
 }
 
 
@@ -342,21 +345,11 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize,
 *  FSE Compression Code
 ****************************************************************/
 
-FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
-{
-    size_t size;
-    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
-    size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-    return (FSE_CTable*)ZSTD_malloc(size);
-}
-
-void FSE_freeCTable (FSE_CTable* ct) { ZSTD_free(ct); }
-
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 {
-    U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
-    U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
+    U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
+    U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
     U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
     return minBits;
@@ -364,7 +357,7 @@ static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
 {
-    U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
+    U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
     U32 tableLog = maxTableLog;
     U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
@@ -532,40 +525,6 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
     return tableLog;
 }
 
-
-/* fake FSE_CTable, for raw (uncompressed) input */
-size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
-{
-    const unsigned tableSize = 1 << nbBits;
-    const unsigned tableMask = tableSize - 1;
-    const unsigned maxSymbolValue = tableMask;
-    void* const ptr = ct;
-    U16* const tableU16 = ( (U16*) ptr) + 2;
-    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1);   /* assumption : tableLog >= 1 */
-    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
-    unsigned s;
-
-    /* Sanity checks */
-    if (nbBits < 1) return ERROR(GENERIC);             /* min size */
-
-    /* header */
-    tableU16[-2] = (U16) nbBits;
-    tableU16[-1] = (U16) maxSymbolValue;
-
-    /* Build table */
-    for (s=0; s<tableSize; s++)
-        tableU16[s] = (U16)(tableSize + s);
-
-    /* Build Symbol Transformation Table */
-    {   const U32 deltaNbBits = (nbBits << 16) - (1 << nbBits);
-        for (s=0; s<=maxSymbolValue; s++) {
-            symbolTT[s].deltaNbBits = deltaNbBits;
-            symbolTT[s].deltaFindState = s-1;
-    }   }
-
-    return 0;
-}
-
 /* fake FSE_CTable, for rle input (always same symbol) */
 size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
 {
@@ -664,5 +623,4 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
-
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/zstd/compress/hist.c b/lib/zstd/compress/hist.c
index 3ddc6dfb6894..0b12587cc14b 100644
--- a/lib/zstd/compress/hist.c
+++ b/lib/zstd/compress/hist.c
@@ -1,7 +1,8 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * hist : Histogram functions
  * part of Finite State Entropy project
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/lib/zstd/compress/hist.h b/lib/zstd/compress/hist.h
index fc1830abc9c6..f7687b0fc20a 100644
--- a/lib/zstd/compress/hist.h
+++ b/lib/zstd/compress/hist.h
@@ -1,7 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /* ******************************************************************
  * hist : Histogram functions
  * part of Finite State Entropy project
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/lib/zstd/compress/huf_compress.c b/lib/zstd/compress/huf_compress.c
index 74ef0db47621..0b229f5d2ae2 100644
--- a/lib/zstd/compress/huf_compress.c
+++ b/lib/zstd/compress/huf_compress.c
@@ -1,6 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * Huffman encoder, part of New Generation Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -26,9 +27,9 @@
 #include "hist.h"
 #define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
 #include "../common/fse.h"        /* header compression */
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/error_private.h"
+#include "../common/bits.h"       /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -39,13 +40,67 @@
 
 
 /* **************************************************************
-*  Utils
+*  Required declarations
 ****************************************************************/
-unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+
+#if DEBUGLEVEL >= 2
+
+static size_t showU32(const U32* arr, size_t size)
 {
-    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", arr[u]); (void)arr;
+    }
+    RAWLOG(6, " \n");
+    return size;
 }
 
+static size_t HUF_getNbBits(HUF_CElt elt);
+
+static size_t showCTableBits(const HUF_CElt* ctable, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %zu", HUF_getNbBits(ctable[u])); (void)ctable;
+    }
+    RAWLOG(6, " \n");
+    return size;
+
+}
+
+static size_t showHNodeSymbols(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].byte); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+static size_t showHNodeBits(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].nbBits); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+#endif
+
 
 /* *******************************************************
 *  HUF : Huffman block compression
@@ -86,7 +141,10 @@ typedef struct {
     S16 norm[HUF_TABLELOG_MAX+1];
 } HUF_CompressWeightsWksp;
 
-static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightTable, size_t wtSize, void* workspace, size_t workspaceSize)
+static size_t
+HUF_compressWeights(void* dst, size_t dstSize,
+              const void* weightTable, size_t wtSize,
+                    void* workspace, size_t workspaceSize)
 {
     BYTE* const ostart = (BYTE*) dst;
     BYTE* op = ostart;
@@ -137,7 +195,7 @@ static size_t HUF_getNbBitsFast(HUF_CElt elt)
 
 static size_t HUF_getValue(HUF_CElt elt)
 {
-    return elt & ~0xFF;
+    return elt & ~(size_t)0xFF;
 }
 
 static size_t HUF_getValueFast(HUF_CElt elt)
@@ -160,6 +218,25 @@ static void HUF_setValue(HUF_CElt* elt, size_t value)
     }
 }
 
+HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable)
+{
+    HUF_CTableHeader header;
+    ZSTD_memcpy(&header, ctable, sizeof(header));
+    return header;
+}
+
+static void HUF_writeCTableHeader(HUF_CElt* ctable, U32 tableLog, U32 maxSymbolValue)
+{
+    HUF_CTableHeader header;
+    HUF_STATIC_ASSERT(sizeof(ctable[0]) == sizeof(header));
+    ZSTD_memset(&header, 0, sizeof(header));
+    assert(tableLog < 256);
+    header.tableLog = (BYTE)tableLog;
+    assert(maxSymbolValue < 256);
+    header.maxSymbolValue = (BYTE)maxSymbolValue;
+    ZSTD_memcpy(ctable, &header, sizeof(header));
+}
+
 typedef struct {
     HUF_CompressWeightsWksp wksp;
     BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
@@ -175,6 +252,11 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
     U32 n;
     HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
 
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
+
+    assert(HUF_readCTableHeader(CTable).maxSymbolValue == maxSymbolValue);
+    assert(HUF_readCTableHeader(CTable).tableLog == huffLog);
+
     /* check conditions */
     if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
     if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
@@ -204,16 +286,6 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
     return ((maxSymbolValue+1)/2) + 1;
 }
 
-/*! HUF_writeCTable() :
-    `CTable` : Huffman tree to save, using huf representation.
-    @return : size of saved CTable */
-size_t HUF_writeCTable (void* dst, size_t maxDstSize,
-                        const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog)
-{
-    HUF_WriteCTableWksp wksp;
-    return HUF_writeCTable_wksp(dst, maxDstSize, CTable, maxSymbolValue, huffLog, &wksp, sizeof(wksp));
-}
-
 
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
 {
@@ -231,7 +303,9 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
     if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
 
-    CTable[0] = tableLog;
+    *maxSymbolValuePtr = nbSymbols - 1;
+
+    HUF_writeCTableHeader(CTable, tableLog, *maxSymbolValuePtr);
 
     /* Prepare base value per rank */
     {   U32 n, nextRankStart = 0;
@@ -263,74 +337,71 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
         { U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
     }
 
-    *maxSymbolValuePtr = nbSymbols - 1;
     return readSize;
 }
 
 U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
 {
-    const HUF_CElt* ct = CTable + 1;
+    const HUF_CElt* const ct = CTable + 1;
     assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
+    if (symbolValue > HUF_readCTableHeader(CTable).maxSymbolValue)
+        return 0;
     return (U32)HUF_getNbBits(ct[symbolValue]);
 }
 
 
-typedef struct nodeElt_s {
-    U32 count;
-    U16 parent;
-    BYTE byte;
-    BYTE nbBits;
-} nodeElt;
-
 /*
  * HUF_setMaxHeight():
- * Enforces maxNbBits on the Huffman tree described in huffNode.
+ * Try to enforce @targetNbBits on the Huffman tree described in @huffNode.
  *
- * It sets all nodes with nbBits > maxNbBits to be maxNbBits. Then it adjusts
- * the tree to so that it is a valid canonical Huffman tree.
+ * It attempts to convert all nodes with nbBits > @targetNbBits
+ * to employ @targetNbBits instead. Then it adjusts the tree
+ * so that it remains a valid canonical Huffman tree.
  *
  * @pre               The sum of the ranks of each symbol == 2^largestBits,
  *                    where largestBits == huffNode[lastNonNull].nbBits.
  * @post              The sum of the ranks of each symbol == 2^largestBits,
- *                    where largestBits is the return value <= maxNbBits.
+ *                    where largestBits is the return value (expected <= targetNbBits).
  *
- * @param huffNode    The Huffman tree modified in place to enforce maxNbBits.
+ * @param huffNode    The Huffman tree modified in place to enforce targetNbBits.
+ *                    It's presumed sorted, from most frequent to rarest symbol.
  * @param lastNonNull The symbol with the lowest count in the Huffman tree.
- * @param maxNbBits   The maximum allowed number of bits, which the Huffman tree
+ * @param targetNbBits  The allowed number of bits, which the Huffman tree
  *                    may not respect. After this function the Huffman tree will
- *                    respect maxNbBits.
- * @return            The maximum number of bits of the Huffman tree after adjustment,
- *                    necessarily no more than maxNbBits.
+ *                    respect targetNbBits.
+ * @return            The maximum number of bits of the Huffman tree after adjustment.
  */
-static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
 {
     const U32 largestBits = huffNode[lastNonNull].nbBits;
-    /* early exit : no elt > maxNbBits, so the tree is already valid. */
-    if (largestBits <= maxNbBits) return largestBits;
+    /* early exit : no elt > targetNbBits, so the tree is already valid. */
+    if (largestBits <= targetNbBits) return largestBits;
+
+    DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
 
     /* there are several too large elements (at least >= 2) */
     {   int totalCost = 0;
-        const U32 baseCost = 1 << (largestBits - maxNbBits);
+        const U32 baseCost = 1 << (largestBits - targetNbBits);
         int n = (int)lastNonNull;
 
-        /* Adjust any ranks > maxNbBits to maxNbBits.
+        /* Adjust any ranks > targetNbBits to targetNbBits.
          * Compute totalCost, which is how far the sum of the ranks is
          * we are over 2^largestBits after adjust the offending ranks.
          */
-        while (huffNode[n].nbBits > maxNbBits) {
+        while (huffNode[n].nbBits > targetNbBits) {
             totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
-            huffNode[n].nbBits = (BYTE)maxNbBits;
+            huffNode[n].nbBits = (BYTE)targetNbBits;
             n--;
         }
-        /* n stops at huffNode[n].nbBits <= maxNbBits */
-        assert(huffNode[n].nbBits <= maxNbBits);
-        /* n end at index of smallest symbol using < maxNbBits */
-        while (huffNode[n].nbBits == maxNbBits) --n;
+        /* n stops at huffNode[n].nbBits <= targetNbBits */
+        assert(huffNode[n].nbBits <= targetNbBits);
+        /* n end at index of smallest symbol using < targetNbBits */
+        while (huffNode[n].nbBits == targetNbBits) --n;
 
-        /* renorm totalCost from 2^largestBits to 2^maxNbBits
+        /* renorm totalCost from 2^largestBits to 2^targetNbBits
          * note : totalCost is necessarily a multiple of baseCost */
-        assert((totalCost & (baseCost - 1)) == 0);
-        totalCost >>= (largestBits - maxNbBits);
+        assert(((U32)totalCost & (baseCost - 1)) == 0);
+        totalCost >>= (largestBits - targetNbBits);
         assert(totalCost > 0);
 
         /* repay normalized cost */
@@ -339,19 +410,19 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
 
             /* Get pos of last (smallest = lowest cum. count) symbol per rank */
             ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
-            {   U32 currentNbBits = maxNbBits;
+            {   U32 currentNbBits = targetNbBits;
                 int pos;
                 for (pos=n ; pos >= 0; pos--) {
                     if (huffNode[pos].nbBits >= currentNbBits) continue;
-                    currentNbBits = huffNode[pos].nbBits;   /* < maxNbBits */
-                    rankLast[maxNbBits-currentNbBits] = (U32)pos;
+                    currentNbBits = huffNode[pos].nbBits;   /* < targetNbBits */
+                    rankLast[targetNbBits-currentNbBits] = (U32)pos;
             }   }
 
             while (totalCost > 0) {
                 /* Try to reduce the next power of 2 above totalCost because we
                  * gain back half the rank.
                  */
-                U32 nBitsToDecrease = BIT_highbit32((U32)totalCost) + 1;
+                U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
                 for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
                     U32 const highPos = rankLast[nBitsToDecrease];
                     U32 const lowPos = rankLast[nBitsToDecrease-1];
@@ -391,7 +462,7 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
                     rankLast[nBitsToDecrease] = noSymbol;
                 else {
                     rankLast[nBitsToDecrease]--;
-                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
                         rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
                 }
             }   /* while (totalCost > 0) */
@@ -403,11 +474,11 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
              * TODO.
              */
             while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
-                /* special case : no rank 1 symbol (using maxNbBits-1);
-                 * let's create one from largest rank 0 (using maxNbBits).
+                /* special case : no rank 1 symbol (using targetNbBits-1);
+                 * let's create one from largest rank 0 (using targetNbBits).
                  */
                 if (rankLast[1] == noSymbol) {
-                    while (huffNode[n].nbBits == maxNbBits) n--;
+                    while (huffNode[n].nbBits == targetNbBits) n--;
                     huffNode[n+1].nbBits--;
                     assert(n >= 0);
                     rankLast[1] = (U32)(n+1);
@@ -421,7 +492,7 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
         }   /* repay normalized cost */
     }   /* there are several too large elements (at least >= 2) */
 
-    return maxNbBits;
+    return targetNbBits;
 }
 
 typedef struct {
@@ -429,7 +500,7 @@ typedef struct {
     U16 curr;
 } rankPos;
 
-typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
+typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
 
 /* Number of buckets available for HUF_sort() */
 #define RANK_POSITION_TABLE_SIZE 192
@@ -448,8 +519,8 @@ typedef struct {
  * Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
  */
 #define RANK_POSITION_MAX_COUNT_LOG 32
-#define RANK_POSITION_LOG_BUCKETS_BEGIN (RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */
-#define RANK_POSITION_DISTINCT_COUNT_CUTOFF RANK_POSITION_LOG_BUCKETS_BEGIN + BIT_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */
+#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
+#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
 
 /* Return the appropriate bucket index for a given count. See definition of
  * RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
@@ -457,7 +528,7 @@ typedef struct {
 static U32 HUF_getIndex(U32 const count) {
     return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
         ? count
-        : BIT_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
+        : ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
 }
 
 /* Helper swap function for HUF_quickSortPartition() */
@@ -580,7 +651,7 @@ static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSy
 
     /* Sort each bucket. */
     for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
-        U32 const bucketSize = rankPosition[n].curr-rankPosition[n].base;
+        int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
         U32 const bucketStartIdx = rankPosition[n].base;
         if (bucketSize > 1) {
             assert(bucketStartIdx < maxSymbolValue1);
@@ -591,6 +662,7 @@ static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSy
     assert(HUF_isSorted(huffNode, maxSymbolValue1));
 }
 
+
 /* HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
  *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
@@ -611,6 +683,7 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
     int lowS, lowN;
     int nodeNb = STARTNODE;
     int n, nodeRoot;
+    DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
     /* init for parents */
     nonNullRank = (int)maxSymbolValue;
     while(huffNode[nonNullRank].count == 0) nonNullRank--;
@@ -637,6 +710,8 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
     for (n=0; n<=nonNullRank; n++)
         huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
 
+    DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
+
     return nonNullRank;
 }
 
@@ -671,31 +746,40 @@ static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, i
         HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits);   /* push nbBits per symbol, symbol order */
     for (n=0; n<alphabetSize; n++)
         HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++);   /* assign value within rank, symbol order */
-    CTable[0] = maxNbBits;
+
+    HUF_writeCTableHeader(CTable, maxNbBits, maxSymbolValue);
 }
 
-size_t HUF_buildCTable_wksp (HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+size_t
+HUF_buildCTable_wksp(HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
+                     void* workSpace, size_t wkspSize)
 {
-    HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
+    HUF_buildCTable_wksp_tables* const wksp_tables =
+        (HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
     nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
     nodeElt* const huffNode = huffNode0+1;
     int nonNullRank;
 
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
+
+    DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
+
     /* safety checks */
     if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
-      return ERROR(workSpace_tooSmall);
+        return ERROR(workSpace_tooSmall);
     if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
     if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
-      return ERROR(maxSymbolValue_tooLarge);
+        return ERROR(maxSymbolValue_tooLarge);
     ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
 
     /* sort, decreasing order */
     HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
+    DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
 
     /* build tree */
     nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
 
-    /* enforce maxTableLog */
+    /* determine and enforce maxTableLog */
     maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
     if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
 
@@ -716,13 +800,20 @@ size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count,
 }
 
 int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
-  HUF_CElt const* ct = CTable + 1;
-  int bad = 0;
-  int s;
-  for (s = 0; s <= (int)maxSymbolValue; ++s) {
-    bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
-  }
-  return !bad;
+    HUF_CTableHeader header = HUF_readCTableHeader(CTable);
+    HUF_CElt const* ct = CTable + 1;
+    int bad = 0;
+    int s;
+
+    assert(header.tableLog <= HUF_TABLELOG_ABSOLUTEMAX);
+
+    if (header.maxSymbolValue < maxSymbolValue)
+        return 0;
+
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
+    }
+    return !bad;
 }
 
 size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
@@ -804,7 +895,7 @@ FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int id
 #if DEBUGLEVEL >= 1
     {
         size_t const nbBits = HUF_getNbBits(elt);
-        size_t const dirtyBits = nbBits == 0 ? 0 : BIT_highbit32((U32)nbBits) + 1;
+        size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
         (void)dirtyBits;
         /* Middle bits are 0. */
         assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
@@ -884,7 +975,7 @@ static size_t HUF_closeCStream(HUF_CStream_t* bitC)
     {
         size_t const nbBits = bitC->bitPos[0] & 0xFF;
         if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
-        return (bitC->ptr - bitC->startPtr) + (nbBits > 0);
+        return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
     }
 }
 
@@ -964,17 +1055,17 @@ HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
                                    const void* src, size_t srcSize,
                                    const HUF_CElt* CTable)
 {
-    U32 const tableLog = (U32)CTable[0];
+    U32 const tableLog = HUF_readCTableHeader(CTable).tableLog;
     HUF_CElt const* ct = CTable + 1;
     const BYTE* ip = (const BYTE*) src;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
-    BYTE* op = ostart;
     HUF_CStream_t bitC;
 
     /* init */
     if (dstSize < 8) return 0;   /* not enough space to compress */
-    { size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
+    { BYTE* op = ostart;
+      size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
       if (HUF_isError(initErr)) return 0; }
 
     if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
@@ -1045,9 +1136,9 @@ HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
 static size_t
 HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, const int bmi2)
+                              const HUF_CElt* CTable, const int flags)
 {
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
         return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
     }
     return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
@@ -1058,28 +1149,23 @@ HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
 static size_t
 HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, const int bmi2)
+                              const HUF_CElt* CTable, const int flags)
 {
-    (void)bmi2;
+    (void)flags;
     return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
 }
 
 #endif
 
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
 {
-    return HUF_compress1X_usingCTable_bmi2(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
-}
-
-size_t HUF_compress1X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2)
-{
-    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, bmi2);
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
 }
 
 static size_t
 HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, int bmi2)
+                              const HUF_CElt* CTable, int flags)
 {
     size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
     const BYTE* ip = (const BYTE*) src;
@@ -1093,7 +1179,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
     op += 6;   /* jumpTable */
 
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
         if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart, (U16)cSize);
         op += cSize;
@@ -1101,7 +1187,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
 
     ip += segmentSize;
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
         if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart+2, (U16)cSize);
         op += cSize;
@@ -1109,7 +1195,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
 
     ip += segmentSize;
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
         if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart+4, (U16)cSize);
         op += cSize;
@@ -1118,7 +1204,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
     ip += segmentSize;
     assert(op <= oend);
     assert(ip <= iend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, bmi2) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
         if (cSize == 0 || cSize > 65535) return 0;
         op += cSize;
     }
@@ -1126,14 +1212,9 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
     return (size_t)(op-ostart);
 }
 
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
-{
-    return HUF_compress4X_usingCTable_bmi2(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
-}
-
-size_t HUF_compress4X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2)
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
 {
-    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, bmi2);
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
 }
 
 typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
@@ -1141,11 +1222,11 @@ typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
 static size_t HUF_compressCTable_internal(
                 BYTE* const ostart, BYTE* op, BYTE* const oend,
                 const void* src, size_t srcSize,
-                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int bmi2)
+                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
 {
     size_t const cSize = (nbStreams==HUF_singleStream) ?
-                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2) :
-                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2);
+                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
+                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
     if (HUF_isError(cSize)) { return cSize; }
     if (cSize==0) { return 0; }   /* uncompressible */
     op += cSize;
@@ -1168,6 +1249,81 @@ typedef struct {
 #define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
 #define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10  /* Must be >= 2 */
 
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
+{
+    unsigned cardinality = 0;
+    unsigned i;
+
+    for (i = 0; i < maxSymbolValue + 1; i++) {
+        if (count[i] != 0) cardinality += 1;
+    }
+
+    return cardinality;
+}
+
+unsigned HUF_minTableLog(unsigned symbolCardinality)
+{
+    U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
+    return minBitsSymbols;
+}
+
+unsigned HUF_optimalTableLog(
+            unsigned maxTableLog,
+            size_t srcSize,
+            unsigned maxSymbolValue,
+            void* workSpace, size_t wkspSize,
+            HUF_CElt* table,
+      const unsigned* count,
+            int flags)
+{
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
+
+    if (!(flags & HUF_flags_optimalDepth)) {
+        /* cheap evaluation, based on FSE */
+        return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    }
+
+    {   BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
+        size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
+        size_t hSize, newSize;
+        const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
+        const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
+        size_t optSize = ((size_t) ~0) - 1;
+        unsigned optLog = maxTableLog, optLogGuess;
+
+        DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
+
+        /* Search until size increases */
+        for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
+            DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
+
+            {   size_t maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
+                if (ERR_isError(maxBits)) continue;
+
+                if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
+
+                hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
+            }
+
+            if (ERR_isError(hSize)) continue;
+
+            newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
+
+            if (newSize > optSize + 1) {
+                break;
+            }
+
+            if (newSize < optSize) {
+                optSize = newSize;
+                optLog = optLogGuess;
+            }
+        }
+        assert(optLog <= HUF_TABLELOG_MAX);
+        return optLog;
+    }
+}
+
 /* HUF_compress_internal() :
  * `workSpace_align4` must be aligned on 4-bytes boundaries,
  * and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
@@ -1177,14 +1333,14 @@ HUF_compress_internal (void* dst, size_t dstSize,
                        unsigned maxSymbolValue, unsigned huffLog,
                        HUF_nbStreams_e nbStreams,
                        void* workSpace, size_t wkspSize,
-                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
-                 const int bmi2, unsigned suspectUncompressible)
+                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
 {
     HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
 
+    DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
     HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
 
     /* checks & inits */
@@ -1198,16 +1354,17 @@ HUF_compress_internal (void* dst, size_t dstSize,
     if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
 
     /* Heuristic : If old table is valid, use it for small inputs */
-    if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
         return HUF_compressCTable_internal(ostart, op, oend,
                                            src, srcSize,
-                                           nbStreams, oldHufTable, bmi2);
+                                           nbStreams, oldHufTable, flags);
     }
 
     /* If uncompressible data is suspected, do a smaller sampling first */
     DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
-    if (suspectUncompressible && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
+    if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
         size_t largestTotal = 0;
+        DEBUGLOG(5, "input suspected incompressible : sampling to check");
         {   unsigned maxSymbolValueBegin = maxSymbolValue;
             CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
             largestTotal += largestBegin;
@@ -1224,6 +1381,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
         if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
+    DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
 
     /* Check validity of previous table */
     if ( repeat
@@ -1232,25 +1390,20 @@ HUF_compress_internal (void* dst, size_t dstSize,
         *repeat = HUF_repeat_none;
     }
     /* Heuristic : use existing table for small inputs */
-    if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
         return HUF_compressCTable_internal(ostart, op, oend,
                                            src, srcSize,
-                                           nbStreams, oldHufTable, bmi2);
+                                           nbStreams, oldHufTable, flags);
     }
 
     /* Build Huffman Tree */
-    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
     {   size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
                                             maxSymbolValue, huffLog,
                                             &table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
         CHECK_F(maxBits);
         huffLog = (U32)maxBits;
-    }
-    /* Zero unused symbols in CTable, so we can check it for validity */
-    {
-        size_t const ctableSize = HUF_CTABLE_SIZE_ST(maxSymbolValue);
-        size_t const unusedSize = sizeof(table->CTable) - ctableSize * sizeof(HUF_CElt);
-        ZSTD_memset(table->CTable + ctableSize, 0, unusedSize);
+        DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
     }
 
     /* Write table description header */
@@ -1263,7 +1416,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
             if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
                 return HUF_compressCTable_internal(ostart, op, oend,
                                                    src, srcSize,
-                                                   nbStreams, oldHufTable, bmi2);
+                                                   nbStreams, oldHufTable, flags);
         }   }
 
         /* Use the new huffman table */
@@ -1275,61 +1428,35 @@ HUF_compress_internal (void* dst, size_t dstSize,
     }
     return HUF_compressCTable_internal(ostart, op, oend,
                                        src, srcSize,
-                                       nbStreams, table->CTable, bmi2);
-}
-
-
-size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
-                      const void* src, size_t srcSize,
-                      unsigned maxSymbolValue, unsigned huffLog,
-                      void* workSpace, size_t wkspSize)
-{
-    return HUF_compress_internal(dst, dstSize, src, srcSize,
-                                 maxSymbolValue, huffLog, HUF_singleStream,
-                                 workSpace, wkspSize,
-                                 NULL, NULL, 0, 0 /*bmi2*/, 0);
+                                       nbStreams, table->CTable, flags);
 }
 
 size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat,
-                      int bmi2, unsigned suspectUncompressible)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
 {
+    DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
     return HUF_compress_internal(dst, dstSize, src, srcSize,
                                  maxSymbolValue, huffLog, HUF_singleStream,
                                  workSpace, wkspSize, hufTable,
-                                 repeat, preferRepeat, bmi2, suspectUncompressible);
-}
-
-/* HUF_compress4X_repeat():
- * compress input using 4 streams.
- * provide workspace to generate compression tables */
-size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
-                      const void* src, size_t srcSize,
-                      unsigned maxSymbolValue, unsigned huffLog,
-                      void* workSpace, size_t wkspSize)
-{
-    return HUF_compress_internal(dst, dstSize, src, srcSize,
-                                 maxSymbolValue, huffLog, HUF_fourStreams,
-                                 workSpace, wkspSize,
-                                 NULL, NULL, 0, 0 /*bmi2*/, 0);
+                                 repeat, flags);
 }
 
 /* HUF_compress4X_repeat():
  * compress input using 4 streams.
  * consider skipping quickly
- * re-use an existing huffman compression table */
+ * reuse an existing huffman compression table */
 size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
 {
+    DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
     return HUF_compress_internal(dst, dstSize, src, srcSize,
                                  maxSymbolValue, huffLog, HUF_fourStreams,
                                  workSpace, wkspSize,
-                                 hufTable, repeat, preferRepeat, bmi2, suspectUncompressible);
+                                 hufTable, repeat, flags);
 }
-
diff --git a/lib/zstd/compress/zstd_compress.c b/lib/zstd/compress/zstd_compress.c
index f620cafca633..0d139727cd39 100644
--- a/lib/zstd/compress/zstd_compress.c
+++ b/lib/zstd/compress/zstd_compress.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,12 +12,12 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
 #include "../common/zstd_deps.h"  /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
 #include "../common/mem.h"
 #include "hist.h"           /* HIST_countFast_wksp */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "zstd_compress_internal.h"
 #include "zstd_compress_sequences.h"
@@ -27,6 +28,7 @@
 #include "zstd_opt.h"
 #include "zstd_ldm.h"
 #include "zstd_compress_superblock.h"
+#include  "../common/bits.h"      /* ZSTD_highbit32, ZSTD_rotateRight_U64 */
 
 /* ***************************************************************
 *  Tuning parameters
@@ -55,14 +57,17 @@
 *  Helper functions
 ***************************************/
 /* ZSTD_compressBound()
- * Note that the result from this function is only compatible with the "normal"
- * full-block strategy.
- * When there are a lot of small blocks due to frequent flush in streaming mode
- * the overhead of headers can make the compressed data to be larger than the
- * return value of ZSTD_compressBound().
+ * Note that the result from this function is only valid for
+ * the one-pass compression functions.
+ * When employing the streaming mode,
+ * if flushes are frequently altering the size of blocks,
+ * the overhead from block headers can make the compressed data larger
+ * than the return value of ZSTD_compressBound().
  */
 size_t ZSTD_compressBound(size_t srcSize) {
-    return ZSTD_COMPRESSBOUND(srcSize);
+    size_t const r = ZSTD_COMPRESSBOUND(srcSize);
+    if (r==0) return ERROR(srcSize_wrong);
+    return r;
 }
 
 
@@ -168,15 +173,13 @@ static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
 
 size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
 {
+    DEBUGLOG(3, "ZSTD_freeCCtx (address: %p)", (void*)cctx);
     if (cctx==NULL) return 0;   /* support free on NULL */
     RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
                     "not compatible with static CCtx");
-    {
-        int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
+    {   int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
         ZSTD_freeCCtxContent(cctx);
-        if (!cctxInWorkspace) {
-            ZSTD_customFree(cctx, cctx->customMem);
-        }
+        if (!cctxInWorkspace) ZSTD_customFree(cctx, cctx->customMem);
     }
     return 0;
 }
@@ -257,9 +260,9 @@ static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
     return forDDSDict || ((strategy != ZSTD_fast) && !ZSTD_rowMatchFinderUsed(strategy, useRowMatchFinder));
 }
 
-/* Returns 1 if compression parameters are such that we should
+/* Returns ZSTD_ps_enable if compression parameters are such that we should
  * enable long distance matching (wlog >= 27, strategy >= btopt).
- * Returns 0 otherwise.
+ * Returns ZSTD_ps_disable otherwise.
  */
 static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
                                  const ZSTD_compressionParameters* const cParams) {
@@ -267,6 +270,34 @@ static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
     return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
 }
 
+static int ZSTD_resolveExternalSequenceValidation(int mode) {
+    return mode;
+}
+
+/* Resolves maxBlockSize to the default if no value is present. */
+static size_t ZSTD_resolveMaxBlockSize(size_t maxBlockSize) {
+    if (maxBlockSize == 0) {
+        return ZSTD_BLOCKSIZE_MAX;
+    } else {
+        return maxBlockSize;
+    }
+}
+
+static ZSTD_paramSwitch_e ZSTD_resolveExternalRepcodeSearch(ZSTD_paramSwitch_e value, int cLevel) {
+    if (value != ZSTD_ps_auto) return value;
+    if (cLevel < 10) {
+        return ZSTD_ps_disable;
+    } else {
+        return ZSTD_ps_enable;
+    }
+}
+
+/* Returns 1 if compression parameters are such that CDict hashtable and chaintable indices are tagged.
+ * If so, the tags need to be removed in ZSTD_resetCCtx_byCopyingCDict. */
+static int ZSTD_CDictIndicesAreTagged(const ZSTD_compressionParameters* const cParams) {
+    return cParams->strategy == ZSTD_fast || cParams->strategy == ZSTD_dfast;
+}
+
 static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
         ZSTD_compressionParameters cParams)
 {
@@ -284,6 +315,10 @@ static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
     }
     cctxParams.useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.useBlockSplitter, &cParams);
     cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+    cctxParams.validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams.validateSequences);
+    cctxParams.maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams.maxBlockSize);
+    cctxParams.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams.searchForExternalRepcodes,
+                                                                             cctxParams.compressionLevel);
     assert(!ZSTD_checkCParams(cParams));
     return cctxParams;
 }
@@ -329,10 +364,13 @@ size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel)
 #define ZSTD_NO_CLEVEL 0
 
 /*
- * Initializes the cctxParams from params and compressionLevel.
+ * Initializes `cctxParams` from `params` and `compressionLevel`.
  * @param compressionLevel If params are derived from a compression level then that compression level, otherwise ZSTD_NO_CLEVEL.
  */
-static void ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams, ZSTD_parameters const* params, int compressionLevel)
+static void
+ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams,
+                        const ZSTD_parameters* params,
+                              int compressionLevel)
 {
     assert(!ZSTD_checkCParams(params->cParams));
     ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
@@ -345,6 +383,9 @@ static void ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams, ZSTD_par
     cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, &params->cParams);
     cctxParams->useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->useBlockSplitter, &params->cParams);
     cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, &params->cParams);
+    cctxParams->validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams->validateSequences);
+    cctxParams->maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams->maxBlockSize);
+    cctxParams->searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams->searchForExternalRepcodes, compressionLevel);
     DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
                 cctxParams->useRowMatchFinder, cctxParams->useBlockSplitter, cctxParams->ldmParams.enableLdm);
 }
@@ -359,7 +400,7 @@ size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_paramete
 
 /*
  * Sets cctxParams' cParams and fParams from params, but otherwise leaves them alone.
- * @param param Validated zstd parameters.
+ * @param params Validated zstd parameters.
  */
 static void ZSTD_CCtxParams_setZstdParams(
         ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
@@ -455,8 +496,8 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         return bounds;
 
     case ZSTD_c_enableLongDistanceMatching:
-        bounds.lowerBound = 0;
-        bounds.upperBound = 1;
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
         return bounds;
 
     case ZSTD_c_ldmHashLog:
@@ -549,6 +590,26 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = 1;
         return bounds;
 
+    case ZSTD_c_prefetchCDictTables:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_maxBlockSize:
+        bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+        bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
     default:
         bounds.error = ERROR(parameter_unsupported);
         return bounds;
@@ -567,10 +628,11 @@ static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
     return 0;
 }
 
-#define BOUNDCHECK(cParam, val) { \
-    RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val), \
-                    parameter_outOfBound, "Param out of bounds"); \
-}
+#define BOUNDCHECK(cParam, val)                                       \
+    do {                                                              \
+        RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val),        \
+                        parameter_outOfBound, "Param out of bounds"); \
+    } while (0)
 
 
 static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
@@ -613,6 +675,10 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
     default:
         return 0;
     }
@@ -625,7 +691,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
         if (ZSTD_isUpdateAuthorized(param)) {
             cctx->cParamsChanged = 1;
         } else {
-            RETURN_ERROR(stage_wrong, "can only set params in ctx init stage");
+            RETURN_ERROR(stage_wrong, "can only set params in cctx init stage");
     }   }
 
     switch(param)
@@ -668,6 +734,10 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
         break;
 
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
@@ -723,12 +793,12 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
     case ZSTD_c_minMatch :
         if (value!=0)   /* 0 => use default */
             BOUNDCHECK(ZSTD_c_minMatch, value);
-        CCtxParams->cParams.minMatch = value;
+        CCtxParams->cParams.minMatch = (U32)value;
         return CCtxParams->cParams.minMatch;
 
     case ZSTD_c_targetLength :
         BOUNDCHECK(ZSTD_c_targetLength, value);
-        CCtxParams->cParams.targetLength = value;
+        CCtxParams->cParams.targetLength = (U32)value;
         return CCtxParams->cParams.targetLength;
 
     case ZSTD_c_strategy :
@@ -741,12 +811,12 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         /* Content size written in frame header _when known_ (default:1) */
         DEBUGLOG(4, "set content size flag = %u", (value!=0));
         CCtxParams->fParams.contentSizeFlag = value != 0;
-        return CCtxParams->fParams.contentSizeFlag;
+        return (size_t)CCtxParams->fParams.contentSizeFlag;
 
     case ZSTD_c_checksumFlag :
         /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
         CCtxParams->fParams.checksumFlag = value != 0;
-        return CCtxParams->fParams.checksumFlag;
+        return (size_t)CCtxParams->fParams.checksumFlag;
 
     case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
         DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
@@ -755,18 +825,18 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
 
     case ZSTD_c_forceMaxWindow :
         CCtxParams->forceWindow = (value != 0);
-        return CCtxParams->forceWindow;
+        return (size_t)CCtxParams->forceWindow;
 
     case ZSTD_c_forceAttachDict : {
         const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
-        BOUNDCHECK(ZSTD_c_forceAttachDict, pref);
+        BOUNDCHECK(ZSTD_c_forceAttachDict, (int)pref);
         CCtxParams->attachDictPref = pref;
         return CCtxParams->attachDictPref;
     }
 
     case ZSTD_c_literalCompressionMode : {
         const ZSTD_paramSwitch_e lcm = (ZSTD_paramSwitch_e)value;
-        BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, (int)lcm);
         CCtxParams->literalCompressionMode = lcm;
         return CCtxParams->literalCompressionMode;
     }
@@ -789,47 +859,50 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
 
     case ZSTD_c_enableDedicatedDictSearch :
         CCtxParams->enableDedicatedDictSearch = (value!=0);
-        return CCtxParams->enableDedicatedDictSearch;
+        return (size_t)CCtxParams->enableDedicatedDictSearch;
 
     case ZSTD_c_enableLongDistanceMatching :
+        BOUNDCHECK(ZSTD_c_enableLongDistanceMatching, value);
         CCtxParams->ldmParams.enableLdm = (ZSTD_paramSwitch_e)value;
         return CCtxParams->ldmParams.enableLdm;
 
     case ZSTD_c_ldmHashLog :
         if (value!=0)   /* 0 ==> auto */
             BOUNDCHECK(ZSTD_c_ldmHashLog, value);
-        CCtxParams->ldmParams.hashLog = value;
+        CCtxParams->ldmParams.hashLog = (U32)value;
         return CCtxParams->ldmParams.hashLog;
 
     case ZSTD_c_ldmMinMatch :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
-        CCtxParams->ldmParams.minMatchLength = value;
+        CCtxParams->ldmParams.minMatchLength = (U32)value;
         return CCtxParams->ldmParams.minMatchLength;
 
     case ZSTD_c_ldmBucketSizeLog :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
-        CCtxParams->ldmParams.bucketSizeLog = value;
+        CCtxParams->ldmParams.bucketSizeLog = (U32)value;
         return CCtxParams->ldmParams.bucketSizeLog;
 
     case ZSTD_c_ldmHashRateLog :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
-        CCtxParams->ldmParams.hashRateLog = value;
+        CCtxParams->ldmParams.hashRateLog = (U32)value;
         return CCtxParams->ldmParams.hashRateLog;
 
     case ZSTD_c_targetCBlockSize :
-        if (value!=0)   /* 0 ==> default */
+        if (value!=0) {  /* 0 ==> default */
+            value = MAX(value, ZSTD_TARGETCBLOCKSIZE_MIN);
             BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
-        CCtxParams->targetCBlockSize = value;
+        }
+        CCtxParams->targetCBlockSize = (U32)value;
         return CCtxParams->targetCBlockSize;
 
     case ZSTD_c_srcSizeHint :
         if (value!=0)    /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_srcSizeHint, value);
         CCtxParams->srcSizeHint = value;
-        return CCtxParams->srcSizeHint;
+        return (size_t)CCtxParams->srcSizeHint;
 
     case ZSTD_c_stableInBuffer:
         BOUNDCHECK(ZSTD_c_stableInBuffer, value);
@@ -849,7 +922,7 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
     case ZSTD_c_validateSequences:
         BOUNDCHECK(ZSTD_c_validateSequences, value);
         CCtxParams->validateSequences = value;
-        return CCtxParams->validateSequences;
+        return (size_t)CCtxParams->validateSequences;
 
     case ZSTD_c_useBlockSplitter:
         BOUNDCHECK(ZSTD_c_useBlockSplitter, value);
@@ -864,7 +937,28 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
     case ZSTD_c_deterministicRefPrefix:
         BOUNDCHECK(ZSTD_c_deterministicRefPrefix, value);
         CCtxParams->deterministicRefPrefix = !!value;
-        return CCtxParams->deterministicRefPrefix;
+        return (size_t)CCtxParams->deterministicRefPrefix;
+
+    case ZSTD_c_prefetchCDictTables:
+        BOUNDCHECK(ZSTD_c_prefetchCDictTables, value);
+        CCtxParams->prefetchCDictTables = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->prefetchCDictTables;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        BOUNDCHECK(ZSTD_c_enableSeqProducerFallback, value);
+        CCtxParams->enableMatchFinderFallback = value;
+        return (size_t)CCtxParams->enableMatchFinderFallback;
+
+    case ZSTD_c_maxBlockSize:
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_maxBlockSize, value);
+        CCtxParams->maxBlockSize = value;
+        return CCtxParams->maxBlockSize;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        BOUNDCHECK(ZSTD_c_searchForExternalRepcodes, value);
+        CCtxParams->searchForExternalRepcodes = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->searchForExternalRepcodes;
 
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
@@ -980,6 +1074,18 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_deterministicRefPrefix:
         *value = (int)CCtxParams->deterministicRefPrefix;
         break;
+    case ZSTD_c_prefetchCDictTables:
+        *value = (int)CCtxParams->prefetchCDictTables;
+        break;
+    case ZSTD_c_enableSeqProducerFallback:
+        *value = CCtxParams->enableMatchFinderFallback;
+        break;
+    case ZSTD_c_maxBlockSize:
+        *value = (int)CCtxParams->maxBlockSize;
+        break;
+    case ZSTD_c_searchForExternalRepcodes:
+        *value = (int)CCtxParams->searchForExternalRepcodes;
+        break;
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
     return 0;
@@ -1006,9 +1112,47 @@ size_t ZSTD_CCtx_setParametersUsingCCtxParams(
     return 0;
 }
 
+size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(cparams) == 7 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setCParams");
+    /* only update if all parameters are valid */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(cparams), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, cparams.windowLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_chainLog, cparams.chainLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_hashLog, cparams.hashLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_searchLog, cparams.searchLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, cparams.minMatch), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_targetLength, cparams.targetLength), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_strategy, cparams.strategy), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(fparams) == 3 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setFParams");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, fparams.contentSizeFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, fparams.checksumFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_dictIDFlag, fparams.noDictIDFlag == 0), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParams");
+    /* First check cParams, because we want to update all or none. */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    /* Next set fParams, because this could fail if the cctx isn't in init stage. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setFParams(cctx, params.fParams), "");
+    /* Finally set cParams, which should succeed. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setCParams(cctx, params.cParams), "");
+    return 0;
+}
+
 size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
 {
-    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %llu bytes", pledgedSrcSize);
     RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
                     "Can't set pledgedSrcSize when not in init stage.");
     cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
@@ -1024,9 +1168,9 @@ static void ZSTD_dedicatedDictSearch_revertCParams(
         ZSTD_compressionParameters* cParams);
 
 /*
- * Initializes the local dict using the requested parameters.
- * NOTE: This does not use the pledged src size, because it may be used for more
- * than one compression.
+ * Initializes the local dictionary using requested parameters.
+ * NOTE: Initialization does not employ the pledged src size,
+ * because the dictionary may be used for multiple compressions.
  */
 static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
 {
@@ -1039,8 +1183,8 @@ static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
         return 0;
     }
     if (dl->cdict != NULL) {
-        assert(cctx->cdict == dl->cdict);
         /* Local dictionary already initialized. */
+        assert(cctx->cdict == dl->cdict);
         return 0;
     }
     assert(dl->dictSize > 0);
@@ -1060,26 +1204,30 @@ static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
 }
 
 size_t ZSTD_CCtx_loadDictionary_advanced(
-        ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
-        ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+        ZSTD_CCtx* cctx,
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType)
 {
-    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
-                    "Can't load a dictionary when ctx is not in init stage.");
     DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
-    ZSTD_clearAllDicts(cctx);  /* in case one already exists */
-    if (dict == NULL || dictSize == 0)  /* no dictionary mode */
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't load a dictionary when cctx is not in init stage.");
+    ZSTD_clearAllDicts(cctx);  /* erase any previously set dictionary */
+    if (dict == NULL || dictSize == 0)  /* no dictionary */
         return 0;
     if (dictLoadMethod == ZSTD_dlm_byRef) {
         cctx->localDict.dict = dict;
     } else {
+        /* copy dictionary content inside CCtx to own its lifetime */
         void* dictBuffer;
         RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
-                        "no malloc for static CCtx");
+                        "static CCtx can't allocate for an internal copy of dictionary");
         dictBuffer = ZSTD_customMalloc(dictSize, cctx->customMem);
-        RETURN_ERROR_IF(!dictBuffer, memory_allocation, "NULL pointer!");
+        RETURN_ERROR_IF(dictBuffer==NULL, memory_allocation,
+                        "allocation failed for dictionary content");
         ZSTD_memcpy(dictBuffer, dict, dictSize);
-        cctx->localDict.dictBuffer = dictBuffer;
-        cctx->localDict.dict = dictBuffer;
+        cctx->localDict.dictBuffer = dictBuffer;  /* owned ptr to free */
+        cctx->localDict.dict = dictBuffer;        /* read-only reference */
     }
     cctx->localDict.dictSize = dictSize;
     cctx->localDict.dictContentType = dictContentType;
@@ -1149,7 +1297,7 @@ size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
     if ( (reset == ZSTD_reset_parameters)
       || (reset == ZSTD_reset_session_and_parameters) ) {
         RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
-                        "Can't reset parameters only when not in init stage.");
+                        "Reset parameters is only possible during init stage.");
         ZSTD_clearAllDicts(cctx);
         return ZSTD_CCtxParams_reset(&cctx->requestedParams);
     }
@@ -1178,11 +1326,12 @@ size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
 static ZSTD_compressionParameters
 ZSTD_clampCParams(ZSTD_compressionParameters cParams)
 {
-#   define CLAMP_TYPE(cParam, val, type) {                                \
-        ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);         \
-        if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
-        else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
-    }
+#   define CLAMP_TYPE(cParam, val, type)                                      \
+        do {                                                                  \
+            ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);         \
+            if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
+            else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
+        } while (0)
 #   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
     CLAMP(ZSTD_c_windowLog, cParams.windowLog);
     CLAMP(ZSTD_c_chainLog,  cParams.chainLog);
@@ -1247,12 +1396,55 @@ static ZSTD_compressionParameters
 ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
                             unsigned long long srcSize,
                             size_t dictSize,
-                            ZSTD_cParamMode_e mode)
+                            ZSTD_cParamMode_e mode,
+                            ZSTD_paramSwitch_e useRowMatchFinder)
 {
     const U64 minSrcSize = 513; /* (1<<9) + 1 */
     const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
     assert(ZSTD_checkCParams(cPar)==0);
 
+    /* Cascade the selected strategy down to the next-highest one built into
+     * this binary. */
+#ifdef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btultra2) {
+        cPar.strategy = ZSTD_btultra;
+    }
+    if (cPar.strategy == ZSTD_btultra) {
+        cPar.strategy = ZSTD_btopt;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btopt) {
+        cPar.strategy = ZSTD_btlazy2;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btlazy2) {
+        cPar.strategy = ZSTD_lazy2;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_lazy2) {
+        cPar.strategy = ZSTD_lazy;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_lazy) {
+        cPar.strategy = ZSTD_greedy;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_greedy) {
+        cPar.strategy = ZSTD_dfast;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_dfast) {
+        cPar.strategy = ZSTD_fast;
+        cPar.targetLength = 0;
+    }
+#endif
+
     switch (mode) {
     case ZSTD_cpm_unknown:
     case ZSTD_cpm_noAttachDict:
@@ -1281,8 +1473,8 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
     }
 
     /* resize windowLog if input is small enough, to use less memory */
-    if ( (srcSize < maxWindowResize)
-      && (dictSize < maxWindowResize) )  {
+    if ( (srcSize <= maxWindowResize)
+      && (dictSize <= maxWindowResize) )  {
         U32 const tSize = (U32)(srcSize + dictSize);
         static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
         U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
@@ -1300,6 +1492,42 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
     if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
         cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* minimum wlog required for valid frame header */
 
+    /* We can't use more than 32 bits of hash in total, so that means that we require:
+     * (hashLog + 8) <= 32 && (chainLog + 8) <= 32
+     */
+    if (mode == ZSTD_cpm_createCDict && ZSTD_CDictIndicesAreTagged(&cPar)) {
+        U32 const maxShortCacheHashLog = 32 - ZSTD_SHORT_CACHE_TAG_BITS;
+        if (cPar.hashLog > maxShortCacheHashLog) {
+            cPar.hashLog = maxShortCacheHashLog;
+        }
+        if (cPar.chainLog > maxShortCacheHashLog) {
+            cPar.chainLog = maxShortCacheHashLog;
+        }
+    }
+
+
+    /* At this point, we aren't 100% sure if we are using the row match finder.
+     * Unless it is explicitly disabled, conservatively assume that it is enabled.
+     * In this case it will only be disabled for small sources, so shrinking the
+     * hash log a little bit shouldn't result in any ratio loss.
+     */
+    if (useRowMatchFinder == ZSTD_ps_auto)
+        useRowMatchFinder = ZSTD_ps_enable;
+
+    /* We can't hash more than 32-bits in total. So that means that we require:
+     * (hashLog - rowLog + 8) <= 32
+     */
+    if (ZSTD_rowMatchFinderUsed(cPar.strategy, useRowMatchFinder)) {
+        /* Switch to 32-entry rows if searchLog is 5 (or more) */
+        U32 const rowLog = BOUNDED(4, cPar.searchLog, 6);
+        U32 const maxRowHashLog = 32 - ZSTD_ROW_HASH_TAG_BITS;
+        U32 const maxHashLog = maxRowHashLog + rowLog;
+        assert(cPar.hashLog >= rowLog);
+        if (cPar.hashLog > maxHashLog) {
+            cPar.hashLog = maxHashLog;
+        }
+    }
+
     return cPar;
 }
 
@@ -1310,7 +1538,7 @@ ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
 {
     cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
     if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
-    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown);
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown, ZSTD_ps_auto);
 }
 
 static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
@@ -1341,7 +1569,7 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
     ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
     assert(!ZSTD_checkCParams(cParams));
     /* srcSizeHint == 0 means 0 */
-    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode);
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode, CCtxParams->useRowMatchFinder);
 }
 
 static size_t
@@ -1367,10 +1595,10 @@ ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
       + ZSTD_cwksp_aligned_alloc_size((MaxLL+1) * sizeof(U32))
       + ZSTD_cwksp_aligned_alloc_size((MaxOff+1) * sizeof(U32))
       + ZSTD_cwksp_aligned_alloc_size((1<<Litbits) * sizeof(U32))
-      + ZSTD_cwksp_aligned_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t))
-      + ZSTD_cwksp_aligned_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
+      + ZSTD_cwksp_aligned_alloc_size(ZSTD_OPT_SIZE * sizeof(ZSTD_match_t))
+      + ZSTD_cwksp_aligned_alloc_size(ZSTD_OPT_SIZE * sizeof(ZSTD_optimal_t));
     size_t const lazyAdditionalSpace = ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)
-                                            ? ZSTD_cwksp_aligned_alloc_size(hSize*sizeof(U16))
+                                            ? ZSTD_cwksp_aligned_alloc_size(hSize)
                                             : 0;
     size_t const optSpace = (forCCtx && (cParams->strategy >= ZSTD_btopt))
                                 ? optPotentialSpace
@@ -1386,6 +1614,13 @@ ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
     return tableSpace + optSpace + slackSpace + lazyAdditionalSpace;
 }
 
+/* Helper function for calculating memory requirements.
+ * Gives a tighter bound than ZSTD_sequenceBound() by taking minMatch into account. */
+static size_t ZSTD_maxNbSeq(size_t blockSize, unsigned minMatch, int useSequenceProducer) {
+    U32 const divider = (minMatch==3 || useSequenceProducer) ? 3 : 4;
+    return blockSize / divider;
+}
+
 static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
         const ZSTD_compressionParameters* cParams,
         const ldmParams_t* ldmParams,
@@ -1393,12 +1628,13 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
         const ZSTD_paramSwitch_e useRowMatchFinder,
         const size_t buffInSize,
         const size_t buffOutSize,
-        const U64 pledgedSrcSize)
+        const U64 pledgedSrcSize,
+        int useSequenceProducer,
+        size_t maxBlockSize)
 {
     size_t const windowSize = (size_t) BOUNDED(1ULL, 1ULL << cParams->windowLog, pledgedSrcSize);
-    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
-    U32    const divider = (cParams->minMatch==3) ? 3 : 4;
-    size_t const maxNbSeq = blockSize / divider;
+    size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(maxBlockSize), windowSize);
+    size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, cParams->minMatch, useSequenceProducer);
     size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
                             + ZSTD_cwksp_aligned_alloc_size(maxNbSeq * sizeof(seqDef))
                             + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
@@ -1417,6 +1653,11 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
 
     size_t const cctxSpace = isStatic ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
 
+    size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+    size_t const externalSeqSpace = useSequenceProducer
+        ? ZSTD_cwksp_aligned_alloc_size(maxNbExternalSeq * sizeof(ZSTD_Sequence))
+        : 0;
+
     size_t const neededSpace =
         cctxSpace +
         entropySpace +
@@ -1425,7 +1666,8 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
         ldmSeqSpace +
         matchStateSize +
         tokenSpace +
-        bufferSpace;
+        bufferSpace +
+        externalSeqSpace;
 
     DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
     return neededSpace;
@@ -1443,7 +1685,7 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
      * be needed. However, we still allocate two 0-sized buffers, which can
      * take space under ASAN. */
     return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
-        &cParams, &params->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN);
+        &cParams, &params->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
 }
 
 size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
@@ -1493,7 +1735,7 @@ size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
     RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
     {   ZSTD_compressionParameters const cParams =
                 ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
+        size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(params->maxBlockSize), (size_t)1 << cParams.windowLog);
         size_t const inBuffSize = (params->inBufferMode == ZSTD_bm_buffered)
                 ? ((size_t)1 << cParams.windowLog) + blockSize
                 : 0;
@@ -1504,7 +1746,7 @@ size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
 
         return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
             &cParams, &params->ldmParams, 1, useRowMatchFinder, inBuffSize, outBuffSize,
-            ZSTD_CONTENTSIZE_UNKNOWN);
+            ZSTD_CONTENTSIZE_UNKNOWN, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
     }
 }
 
@@ -1637,6 +1879,19 @@ typedef enum {
     ZSTD_resetTarget_CCtx
 } ZSTD_resetTarget_e;
 
+/* Mixes bits in a 64 bits in a value, based on XXH3_rrmxmx */
+static U64 ZSTD_bitmix(U64 val, U64 len) {
+    val ^= ZSTD_rotateRight_U64(val, 49) ^ ZSTD_rotateRight_U64(val, 24);
+    val *= 0x9FB21C651E98DF25ULL;
+    val ^= (val >> 35) + len ;
+    val *= 0x9FB21C651E98DF25ULL;
+    return val ^ (val >> 28);
+}
+
+/* Mixes in the hashSalt and hashSaltEntropy to create a new hashSalt */
+static void ZSTD_advanceHashSalt(ZSTD_matchState_t* ms) {
+    ms->hashSalt = ZSTD_bitmix(ms->hashSalt, 8) ^ ZSTD_bitmix((U64) ms->hashSaltEntropy, 4);
+}
 
 static size_t
 ZSTD_reset_matchState(ZSTD_matchState_t* ms,
@@ -1664,6 +1919,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
     }
 
     ms->hashLog3 = hashLog3;
+    ms->lazySkipping = 0;
 
     ZSTD_invalidateMatchState(ms);
 
@@ -1685,22 +1941,19 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
         ZSTD_cwksp_clean_tables(ws);
     }
 
-    /* opt parser space */
-    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
-        DEBUGLOG(4, "reserving optimal parser space");
-        ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<<Litbits) * sizeof(unsigned));
-        ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned));
-        ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned));
-        ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned));
-        ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t));
-        ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
-    }
-
     if (ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)) {
-        {   /* Row match finder needs an additional table of hashes ("tags") */
-            size_t const tagTableSize = hSize*sizeof(U16);
-            ms->tagTable = (U16*)ZSTD_cwksp_reserve_aligned(ws, tagTableSize);
-            if (ms->tagTable) ZSTD_memset(ms->tagTable, 0, tagTableSize);
+        /* Row match finder needs an additional table of hashes ("tags") */
+        size_t const tagTableSize = hSize;
+        /* We want to generate a new salt in case we reset a Cctx, but we always want to use
+         * 0 when we reset a Cdict */
+        if(forWho == ZSTD_resetTarget_CCtx) {
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned_init_once(ws, tagTableSize);
+            ZSTD_advanceHashSalt(ms);
+        } else {
+            /* When we are not salting we want to always memset the memory */
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned(ws, tagTableSize);
+            ZSTD_memset(ms->tagTable, 0, tagTableSize);
+            ms->hashSalt = 0;
         }
         {   /* Switch to 32-entry rows if searchLog is 5 (or more) */
             U32 const rowLog = BOUNDED(4, cParams->searchLog, 6);
@@ -1709,6 +1962,17 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
         }
     }
 
+    /* opt parser space */
+    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<<Litbits) * sizeof(unsigned));
+        ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned));
+        ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned));
+        ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned));
+        ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, ZSTD_OPT_SIZE * sizeof(ZSTD_match_t));
+        ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, ZSTD_OPT_SIZE * sizeof(ZSTD_optimal_t));
+    }
+
     ms->cParams = *cParams;
 
     RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
@@ -1768,6 +2032,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
     assert(params->useRowMatchFinder != ZSTD_ps_auto);
     assert(params->useBlockSplitter != ZSTD_ps_auto);
     assert(params->ldmParams.enableLdm != ZSTD_ps_auto);
+    assert(params->maxBlockSize != 0);
     if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
         /* Adjust long distance matching parameters */
         ZSTD_ldm_adjustParameters(&zc->appliedParams.ldmParams, &params->cParams);
@@ -1776,9 +2041,8 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
     }
 
     {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params->cParams.windowLog), pledgedSrcSize));
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
-        U32    const divider = (params->cParams.minMatch==3) ? 3 : 4;
-        size_t const maxNbSeq = blockSize / divider;
+        size_t const blockSize = MIN(params->maxBlockSize, windowSize);
+        size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, params->cParams.minMatch, ZSTD_hasExtSeqProd(params));
         size_t const buffOutSize = (zbuff == ZSTDb_buffered && params->outBufferMode == ZSTD_bm_buffered)
                 ? ZSTD_compressBound(blockSize) + 1
                 : 0;
@@ -1795,8 +2059,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         size_t const neededSpace =
             ZSTD_estimateCCtxSize_usingCCtxParams_internal(
                 &params->cParams, &params->ldmParams, zc->staticSize != 0, params->useRowMatchFinder,
-                buffInSize, buffOutSize, pledgedSrcSize);
-        int resizeWorkspace;
+                buffInSize, buffOutSize, pledgedSrcSize, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
 
         FORWARD_IF_ERROR(neededSpace, "cctx size estimate failed!");
 
@@ -1805,7 +2068,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         {   /* Check if workspace is large enough, alloc a new one if needed */
             int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
             int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
-            resizeWorkspace = workspaceTooSmall || workspaceWasteful;
+            int resizeWorkspace = workspaceTooSmall || workspaceWasteful;
             DEBUGLOG(4, "Need %zu B workspace", neededSpace);
             DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
 
@@ -1838,6 +2101,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
         /* init params */
         zc->blockState.matchState.cParams = params->cParams;
+        zc->blockState.matchState.prefetchCDictTables = params->prefetchCDictTables == ZSTD_ps_enable;
         zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
         zc->consumedSrcSize = 0;
         zc->producedCSize = 0;
@@ -1854,13 +2118,46 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
         ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
 
+        FORWARD_IF_ERROR(ZSTD_reset_matchState(
+                &zc->blockState.matchState,
+                ws,
+                &params->cParams,
+                params->useRowMatchFinder,
+                crp,
+                needsIndexReset,
+                ZSTD_resetTarget_CCtx), "");
+
+        zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef));
+
+        /* ldm hash table */
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+            /* TODO: avoid memset? */
+            size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
+            zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
+            ZSTD_memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
+            zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            ZSTD_window_init(&zc->ldmState.window);
+            zc->ldmState.loadedDictEnd = 0;
+        }
+
+        /* reserve space for block-level external sequences */
+        if (ZSTD_hasExtSeqProd(params)) {
+            size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+            zc->extSeqBufCapacity = maxNbExternalSeq;
+            zc->extSeqBuf =
+                (ZSTD_Sequence*)ZSTD_cwksp_reserve_aligned(ws, maxNbExternalSeq * sizeof(ZSTD_Sequence));
+        }
+
+        /* buffers */
+
         /* ZSTD_wildcopy() is used to copy into the literals buffer,
          * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
          */
         zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH);
         zc->seqStore.maxNbLit = blockSize;
 
-        /* buffers */
         zc->bufferedPolicy = zbuff;
         zc->inBuffSize = buffInSize;
         zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
@@ -1883,32 +2180,9 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
         zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
         zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
-        zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef));
-
-        FORWARD_IF_ERROR(ZSTD_reset_matchState(
-            &zc->blockState.matchState,
-            ws,
-            &params->cParams,
-            params->useRowMatchFinder,
-            crp,
-            needsIndexReset,
-            ZSTD_resetTarget_CCtx), "");
-
-        /* ldm hash table */
-        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
-            /* TODO: avoid memset? */
-            size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
-            zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
-            ZSTD_memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
-            zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
-            zc->maxNbLdmSequences = maxNbLdmSeq;
-
-            ZSTD_window_init(&zc->ldmState.window);
-            zc->ldmState.loadedDictEnd = 0;
-        }
 
         DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
-        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace, resizeWorkspace));
+        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace));
 
         zc->initialized = 1;
 
@@ -1980,7 +2254,8 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
         }
 
         params.cParams = ZSTD_adjustCParams_internal(adjusted_cdict_cParams, pledgedSrcSize,
-                                                     cdict->dictContentSize, ZSTD_cpm_attachDict);
+                                                     cdict->dictContentSize, ZSTD_cpm_attachDict,
+                                                     params.useRowMatchFinder);
         params.cParams.windowLog = windowLog;
         params.useRowMatchFinder = cdict->useRowMatchFinder;    /* cdict overrides */
         FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, &params, pledgedSrcSize,
@@ -2019,6 +2294,22 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
     return 0;
 }
 
+static void ZSTD_copyCDictTableIntoCCtx(U32* dst, U32 const* src, size_t tableSize,
+                                        ZSTD_compressionParameters const* cParams) {
+    if (ZSTD_CDictIndicesAreTagged(cParams)){
+        /* Remove tags from the CDict table if they are present.
+         * See docs on "short cache" in zstd_compress_internal.h for context. */
+        size_t i;
+        for (i = 0; i < tableSize; i++) {
+            U32 const taggedIndex = src[i];
+            U32 const index = taggedIndex >> ZSTD_SHORT_CACHE_TAG_BITS;
+            dst[i] = index;
+        }
+    } else {
+        ZSTD_memcpy(dst, src, tableSize * sizeof(U32));
+    }
+}
+
 static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
                             const ZSTD_CDict* cdict,
                             ZSTD_CCtx_params params,
@@ -2054,21 +2345,23 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
                                                             : 0;
         size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
 
-        ZSTD_memcpy(cctx->blockState.matchState.hashTable,
-               cdict->matchState.hashTable,
-               hSize * sizeof(U32));
+        ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.hashTable,
+                                cdict->matchState.hashTable,
+                                hSize, cdict_cParams);
+
         /* Do not copy cdict's chainTable if cctx has parameters such that it would not use chainTable */
         if (ZSTD_allocateChainTable(cctx->appliedParams.cParams.strategy, cctx->appliedParams.useRowMatchFinder, 0 /* forDDSDict */)) {
-            ZSTD_memcpy(cctx->blockState.matchState.chainTable,
-               cdict->matchState.chainTable,
-               chainSize * sizeof(U32));
+            ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.chainTable,
+                                    cdict->matchState.chainTable,
+                                    chainSize, cdict_cParams);
         }
         /* copy tag table */
         if (ZSTD_rowMatchFinderUsed(cdict_cParams->strategy, cdict->useRowMatchFinder)) {
-            size_t const tagTableSize = hSize*sizeof(U16);
+            size_t const tagTableSize = hSize;
             ZSTD_memcpy(cctx->blockState.matchState.tagTable,
-                cdict->matchState.tagTable,
-                tagTableSize);
+                        cdict->matchState.tagTable,
+                        tagTableSize);
+            cctx->blockState.matchState.hashSalt = cdict->matchState.hashSalt;
         }
     }
 
@@ -2147,6 +2440,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
         params.useBlockSplitter = srcCCtx->appliedParams.useBlockSplitter;
         params.ldmParams = srcCCtx->appliedParams.ldmParams;
         params.fParams = fParams;
+        params.maxBlockSize = srcCCtx->appliedParams.maxBlockSize;
         ZSTD_resetCCtx_internal(dstCCtx, &params, pledgedSrcSize,
                                 /* loadedDictSize */ 0,
                                 ZSTDcrp_leaveDirty, zbuff);
@@ -2294,7 +2588,7 @@ static void ZSTD_reduceIndex (ZSTD_matchState_t* ms, ZSTD_CCtx_params const* par
 
 /* See doc/zstd_compression_format.md for detailed format description */
 
-void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
+int ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
 {
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     BYTE* const llCodeTable = seqStorePtr->llCode;
@@ -2302,18 +2596,24 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
     BYTE* const mlCodeTable = seqStorePtr->mlCode;
     U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     U32 u;
+    int longOffsets = 0;
     assert(nbSeq <= seqStorePtr->maxNbSeq);
     for (u=0; u<nbSeq; u++) {
         U32 const llv = sequences[u].litLength;
+        U32 const ofCode = ZSTD_highbit32(sequences[u].offBase);
         U32 const mlv = sequences[u].mlBase;
         llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
-        ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offBase);
+        ofCodeTable[u] = (BYTE)ofCode;
         mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
+        assert(!(MEM_64bits() && ofCode >= STREAM_ACCUMULATOR_MIN));
+        if (MEM_32bits() && ofCode >= STREAM_ACCUMULATOR_MIN)
+            longOffsets = 1;
     }
     if (seqStorePtr->longLengthType==ZSTD_llt_literalLength)
         llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
     if (seqStorePtr->longLengthType==ZSTD_llt_matchLength)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+    return longOffsets;
 }
 
 /* ZSTD_useTargetCBlockSize():
@@ -2347,6 +2647,7 @@ typedef struct {
     U32 MLtype;
     size_t size;
     size_t lastCountSize; /* Accounts for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
+    int longOffsets;
 } ZSTD_symbolEncodingTypeStats_t;
 
 /* ZSTD_buildSequencesStatistics():
@@ -2357,11 +2658,13 @@ typedef struct {
  * entropyWkspSize must be of size at least ENTROPY_WORKSPACE_SIZE - (MaxSeq + 1)*sizeof(U32)
  */
 static ZSTD_symbolEncodingTypeStats_t
-ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
-                        const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
-                              BYTE* dst, const BYTE* const dstEnd,
-                              ZSTD_strategy strategy, unsigned* countWorkspace,
-                              void* entropyWorkspace, size_t entropyWkspSize) {
+ZSTD_buildSequencesStatistics(
+                const seqStore_t* seqStorePtr, size_t nbSeq,
+                const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
+                      BYTE* dst, const BYTE* const dstEnd,
+                      ZSTD_strategy strategy, unsigned* countWorkspace,
+                      void* entropyWorkspace, size_t entropyWkspSize)
+{
     BYTE* const ostart = dst;
     const BYTE* const oend = dstEnd;
     BYTE* op = ostart;
@@ -2375,7 +2678,7 @@ ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
 
     stats.lastCountSize = 0;
     /* convert length/distances into codes */
-    ZSTD_seqToCodes(seqStorePtr);
+    stats.longOffsets = ZSTD_seqToCodes(seqStorePtr);
     assert(op <= oend);
     assert(nbSeq != 0); /* ZSTD_selectEncodingType() divides by nbSeq */
     /* build CTable for Literal Lengths */
@@ -2480,22 +2783,22 @@ ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
  */
 #define SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO 20
 MEM_STATIC size_t
-ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
-                          const ZSTD_entropyCTables_t* prevEntropy,
-                                ZSTD_entropyCTables_t* nextEntropy,
-                          const ZSTD_CCtx_params* cctxParams,
-                                void* dst, size_t dstCapacity,
-                                void* entropyWorkspace, size_t entropyWkspSize,
-                          const int bmi2)
+ZSTD_entropyCompressSeqStore_internal(
+                        const seqStore_t* seqStorePtr,
+                        const ZSTD_entropyCTables_t* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                        const ZSTD_CCtx_params* cctxParams,
+                              void* dst, size_t dstCapacity,
+                              void* entropyWorkspace, size_t entropyWkspSize,
+                        const int bmi2)
 {
-    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     ZSTD_strategy const strategy = cctxParams->cParams.strategy;
     unsigned* count = (unsigned*)entropyWorkspace;
     FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
     FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
     FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
     const seqDef* const sequences = seqStorePtr->sequencesStart;
-    const size_t nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    const size_t nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     const BYTE* const ofCodeTable = seqStorePtr->ofCode;
     const BYTE* const llCodeTable = seqStorePtr->llCode;
     const BYTE* const mlCodeTable = seqStorePtr->mlCode;
@@ -2503,29 +2806,31 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
     BYTE* const oend = ostart + dstCapacity;
     BYTE* op = ostart;
     size_t lastCountSize;
+    int longOffsets = 0;
 
     entropyWorkspace = count + (MaxSeq + 1);
     entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
 
-    DEBUGLOG(4, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu)", nbSeq);
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu, dstCapacity=%zu)", nbSeq, dstCapacity);
     ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
     assert(entropyWkspSize >= HUF_WORKSPACE_SIZE);
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
-        size_t const numSequences = seqStorePtr->sequences - seqStorePtr->sequencesStart;
-        size_t const numLiterals = seqStorePtr->lit - seqStorePtr->litStart;
+        size_t const numSequences = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        size_t const numLiterals = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
         /* Base suspicion of uncompressibility on ratio of literals to sequences */
         unsigned const suspectUncompressible = (numSequences == 0) || (numLiterals / numSequences >= SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO);
         size_t const litSize = (size_t)(seqStorePtr->lit - literals);
+
         size_t const cSize = ZSTD_compressLiterals(
-                                    &prevEntropy->huf, &nextEntropy->huf,
-                                    cctxParams->cParams.strategy,
-                                    ZSTD_literalsCompressionIsDisabled(cctxParams),
                                     op, dstCapacity,
                                     literals, litSize,
                                     entropyWorkspace, entropyWkspSize,
-                                    bmi2, suspectUncompressible);
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                    suspectUncompressible, bmi2);
         FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
         assert(cSize <= dstCapacity);
         op += cSize;
@@ -2551,11 +2856,10 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
         ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
         return (size_t)(op - ostart);
     }
-    {
-        ZSTD_symbolEncodingTypeStats_t stats;
-        BYTE* seqHead = op++;
+    {   BYTE* const seqHead = op++;
         /* build stats for sequences */
-        stats = ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+        const ZSTD_symbolEncodingTypeStats_t stats =
+                ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
                                              &prevEntropy->fse, &nextEntropy->fse,
                                               op, oend,
                                               strategy, count,
@@ -2564,6 +2868,7 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
         *seqHead = (BYTE)((stats.LLtype<<6) + (stats.Offtype<<4) + (stats.MLtype<<2));
         lastCountSize = stats.lastCountSize;
         op += stats.size;
+        longOffsets = stats.longOffsets;
     }
 
     {   size_t const bitstreamSize = ZSTD_encodeSequences(
@@ -2598,14 +2903,15 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
 }
 
 MEM_STATIC size_t
-ZSTD_entropyCompressSeqStore(seqStore_t* seqStorePtr,
-                       const ZSTD_entropyCTables_t* prevEntropy,
-                             ZSTD_entropyCTables_t* nextEntropy,
-                       const ZSTD_CCtx_params* cctxParams,
-                             void* dst, size_t dstCapacity,
-                             size_t srcSize,
-                             void* entropyWorkspace, size_t entropyWkspSize,
-                             int bmi2)
+ZSTD_entropyCompressSeqStore(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          void* dst, size_t dstCapacity,
+                          size_t srcSize,
+                          void* entropyWorkspace, size_t entropyWkspSize,
+                          int bmi2)
 {
     size_t const cSize = ZSTD_entropyCompressSeqStore_internal(
                             seqStorePtr, prevEntropy, nextEntropy, cctxParams,
@@ -2615,15 +2921,21 @@ ZSTD_entropyCompressSeqStore(seqStore_t* seqStorePtr,
     /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
      * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
      */
-    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity)) {
+        DEBUGLOG(4, "not enough dstCapacity (%zu) for ZSTD_entropyCompressSeqStore_internal()=> do not compress block", dstCapacity);
         return 0;  /* block not compressed */
+    }
     FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSeqStore_internal failed");
 
     /* Check compressibility */
     {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
         if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
-    DEBUGLOG(4, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    /* libzstd decoder before  > v1.5.4 is not compatible with compressed blocks of size ZSTD_BLOCKSIZE_MAX exactly.
+     * This restriction is indirectly already fulfilled by respecting ZSTD_minGain() condition above.
+     */
+    assert(cSize < ZSTD_BLOCKSIZE_MAX);
     return cSize;
 }
 
@@ -2635,40 +2947,43 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramS
     static const ZSTD_blockCompressor blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
         { ZSTD_compressBlock_fast  /* default for 0 */,
           ZSTD_compressBlock_fast,
-          ZSTD_compressBlock_doubleFast,
-          ZSTD_compressBlock_greedy,
-          ZSTD_compressBlock_lazy,
-          ZSTD_compressBlock_lazy2,
-          ZSTD_compressBlock_btlazy2,
-          ZSTD_compressBlock_btopt,
-          ZSTD_compressBlock_btultra,
-          ZSTD_compressBlock_btultra2 },
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST,
+          ZSTD_COMPRESSBLOCK_GREEDY,
+          ZSTD_COMPRESSBLOCK_LAZY,
+          ZSTD_COMPRESSBLOCK_LAZY2,
+          ZSTD_COMPRESSBLOCK_BTLAZY2,
+          ZSTD_COMPRESSBLOCK_BTOPT,
+          ZSTD_COMPRESSBLOCK_BTULTRA,
+          ZSTD_COMPRESSBLOCK_BTULTRA2
+        },
         { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
           ZSTD_compressBlock_fast_extDict,
-          ZSTD_compressBlock_doubleFast_extDict,
-          ZSTD_compressBlock_greedy_extDict,
-          ZSTD_compressBlock_lazy_extDict,
-          ZSTD_compressBlock_lazy2_extDict,
-          ZSTD_compressBlock_btlazy2_extDict,
-          ZSTD_compressBlock_btopt_extDict,
-          ZSTD_compressBlock_btultra_extDict,
-          ZSTD_compressBlock_btultra_extDict },
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT,
+          ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT,
+          ZSTD_COMPRESSBLOCK_LAZY_EXTDICT,
+          ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT
+        },
         { ZSTD_compressBlock_fast_dictMatchState  /* default for 0 */,
           ZSTD_compressBlock_fast_dictMatchState,
-          ZSTD_compressBlock_doubleFast_dictMatchState,
-          ZSTD_compressBlock_greedy_dictMatchState,
-          ZSTD_compressBlock_lazy_dictMatchState,
-          ZSTD_compressBlock_lazy2_dictMatchState,
-          ZSTD_compressBlock_btlazy2_dictMatchState,
-          ZSTD_compressBlock_btopt_dictMatchState,
-          ZSTD_compressBlock_btultra_dictMatchState,
-          ZSTD_compressBlock_btultra_dictMatchState },
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE
+        },
         { NULL  /* default for 0 */,
           NULL,
           NULL,
-          ZSTD_compressBlock_greedy_dedicatedDictSearch,
-          ZSTD_compressBlock_lazy_dedicatedDictSearch,
-          ZSTD_compressBlock_lazy2_dedicatedDictSearch,
+          ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH,
+          ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH,
+          ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH,
           NULL,
           NULL,
           NULL,
@@ -2681,18 +2996,26 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramS
     DEBUGLOG(4, "Selected block compressor: dictMode=%d strat=%d rowMatchfinder=%d", (int)dictMode, (int)strat, (int)useRowMatchFinder);
     if (ZSTD_rowMatchFinderUsed(strat, useRowMatchFinder)) {
         static const ZSTD_blockCompressor rowBasedBlockCompressors[4][3] = {
-            { ZSTD_compressBlock_greedy_row,
-            ZSTD_compressBlock_lazy_row,
-            ZSTD_compressBlock_lazy2_row },
-            { ZSTD_compressBlock_greedy_extDict_row,
-            ZSTD_compressBlock_lazy_extDict_row,
-            ZSTD_compressBlock_lazy2_extDict_row },
-            { ZSTD_compressBlock_greedy_dictMatchState_row,
-            ZSTD_compressBlock_lazy_dictMatchState_row,
-            ZSTD_compressBlock_lazy2_dictMatchState_row },
-            { ZSTD_compressBlock_greedy_dedicatedDictSearch_row,
-            ZSTD_compressBlock_lazy_dedicatedDictSearch_row,
-            ZSTD_compressBlock_lazy2_dedicatedDictSearch_row }
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW
+            }
         };
         DEBUGLOG(4, "Selecting a row-based matchfinder");
         assert(useRowMatchFinder != ZSTD_ps_auto);
@@ -2718,6 +3041,72 @@ void ZSTD_resetSeqStore(seqStore_t* ssPtr)
     ssPtr->longLengthType = ZSTD_llt_none;
 }
 
+/* ZSTD_postProcessSequenceProducerResult() :
+ * Validates and post-processes sequences obtained through the external matchfinder API:
+ *   - Checks whether nbExternalSeqs represents an error condition.
+ *   - Appends a block delimiter to outSeqs if one is not already present.
+ *     See zstd.h for context regarding block delimiters.
+ * Returns the number of sequences after post-processing, or an error code. */
+static size_t ZSTD_postProcessSequenceProducerResult(
+    ZSTD_Sequence* outSeqs, size_t nbExternalSeqs, size_t outSeqsCapacity, size_t srcSize
+) {
+    RETURN_ERROR_IF(
+        nbExternalSeqs > outSeqsCapacity,
+        sequenceProducer_failed,
+        "External sequence producer returned error code %lu",
+        (unsigned long)nbExternalSeqs
+    );
+
+    RETURN_ERROR_IF(
+        nbExternalSeqs == 0 && srcSize > 0,
+        sequenceProducer_failed,
+        "Got zero sequences from external sequence producer for a non-empty src buffer!"
+    );
+
+    if (srcSize == 0) {
+        ZSTD_memset(&outSeqs[0], 0, sizeof(ZSTD_Sequence));
+        return 1;
+    }
+
+    {
+        ZSTD_Sequence const lastSeq = outSeqs[nbExternalSeqs - 1];
+
+        /* We can return early if lastSeq is already a block delimiter. */
+        if (lastSeq.offset == 0 && lastSeq.matchLength == 0) {
+            return nbExternalSeqs;
+        }
+
+        /* This error condition is only possible if the external matchfinder
+         * produced an invalid parse, by definition of ZSTD_sequenceBound(). */
+        RETURN_ERROR_IF(
+            nbExternalSeqs == outSeqsCapacity,
+            sequenceProducer_failed,
+            "nbExternalSeqs == outSeqsCapacity but lastSeq is not a block delimiter!"
+        );
+
+        /* lastSeq is not a block delimiter, so we need to append one. */
+        ZSTD_memset(&outSeqs[nbExternalSeqs], 0, sizeof(ZSTD_Sequence));
+        return nbExternalSeqs + 1;
+    }
+}
+
+/* ZSTD_fastSequenceLengthSum() :
+ * Returns sum(litLen) + sum(matchLen) + lastLits for *seqBuf*.
+ * Similar to another function in zstd_compress.c (determine_blockSize),
+ * except it doesn't check for a block delimiter to end summation.
+ * Removing the early exit allows the compiler to auto-vectorize (https://godbolt.org/z/cY1cajz9P).
+ * This function can be deleted and replaced by determine_blockSize after we resolve issue #3456. */
+static size_t ZSTD_fastSequenceLengthSum(ZSTD_Sequence const* seqBuf, size_t seqBufSize) {
+    size_t matchLenSum, litLenSum, i;
+    matchLenSum = 0;
+    litLenSum = 0;
+    for (i = 0; i < seqBufSize; i++) {
+        litLenSum += seqBuf[i].litLength;
+        matchLenSum += seqBuf[i].matchLength;
+    }
+    return litLenSum + matchLenSum;
+}
+
 typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
 
 static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
@@ -2727,7 +3116,9 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
     assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
     /* Assert that we have correctly flushed the ctx params into the ms's copy */
     ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
-    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+    /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+     * additional 1. We need to revisit and change this logic to be more consistent */
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
         if (zc->appliedParams.cParams.strategy >= ZSTD_btopt) {
             ZSTD_ldm_skipRawSeqStoreBytes(&zc->externSeqStore, srcSize);
         } else {
@@ -2763,6 +3154,15 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
         }
         if (zc->externSeqStore.pos < zc->externSeqStore.size) {
             assert(zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_disable);
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                ZSTD_hasExtSeqProd(&zc->appliedParams),
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
             /* Updates ldmSeqStore.pos */
             lastLLSize =
                 ZSTD_ldm_blockCompress(&zc->externSeqStore,
@@ -2774,6 +3174,14 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
         } else if (zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
             rawSeqStore_t ldmSeqStore = kNullRawSeqStore;
 
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                ZSTD_hasExtSeqProd(&zc->appliedParams),
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
             ldmSeqStore.seq = zc->ldmSequences;
             ldmSeqStore.capacity = zc->maxNbLdmSequences;
             /* Updates ldmSeqStore.size */
@@ -2788,10 +3196,74 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                                        zc->appliedParams.useRowMatchFinder,
                                        src, srcSize);
             assert(ldmSeqStore.pos == ldmSeqStore.size);
-        } else {   /* not long range mode */
-            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy,
-                                                                                    zc->appliedParams.useRowMatchFinder,
-                                                                                    dictMode);
+        } else if (ZSTD_hasExtSeqProd(&zc->appliedParams)) {
+            assert(
+                zc->extSeqBufCapacity >= ZSTD_sequenceBound(srcSize)
+            );
+            assert(zc->appliedParams.extSeqProdFunc != NULL);
+
+            {   U32 const windowSize = (U32)1 << zc->appliedParams.cParams.windowLog;
+
+                size_t const nbExternalSeqs = (zc->appliedParams.extSeqProdFunc)(
+                    zc->appliedParams.extSeqProdState,
+                    zc->extSeqBuf,
+                    zc->extSeqBufCapacity,
+                    src, srcSize,
+                    NULL, 0,  /* dict and dictSize, currently not supported */
+                    zc->appliedParams.compressionLevel,
+                    windowSize
+                );
+
+                size_t const nbPostProcessedSeqs = ZSTD_postProcessSequenceProducerResult(
+                    zc->extSeqBuf,
+                    nbExternalSeqs,
+                    zc->extSeqBufCapacity,
+                    srcSize
+                );
+
+                /* Return early if there is no error, since we don't need to worry about last literals */
+                if (!ZSTD_isError(nbPostProcessedSeqs)) {
+                    ZSTD_sequencePosition seqPos = {0,0,0};
+                    size_t const seqLenSum = ZSTD_fastSequenceLengthSum(zc->extSeqBuf, nbPostProcessedSeqs);
+                    RETURN_ERROR_IF(seqLenSum > srcSize, externalSequences_invalid, "External sequences imply too large a block!");
+                    FORWARD_IF_ERROR(
+                        ZSTD_copySequencesToSeqStoreExplicitBlockDelim(
+                            zc, &seqPos,
+                            zc->extSeqBuf, nbPostProcessedSeqs,
+                            src, srcSize,
+                            zc->appliedParams.searchForExternalRepcodes
+                        ),
+                        "Failed to copy external sequences to seqStore!"
+                    );
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(5, "Copied %lu sequences from external sequence producer to internal seqStore.", (unsigned long)nbExternalSeqs);
+                    return ZSTDbss_compress;
+                }
+
+                /* Propagate the error if fallback is disabled */
+                if (!zc->appliedParams.enableMatchFinderFallback) {
+                    return nbPostProcessedSeqs;
+                }
+
+                /* Fallback to software matchfinder */
+                {   ZSTD_blockCompressor const blockCompressor =
+                        ZSTD_selectBlockCompressor(
+                            zc->appliedParams.cParams.strategy,
+                            zc->appliedParams.useRowMatchFinder,
+                            dictMode);
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(
+                        5,
+                        "External sequence producer returned error code %lu. Falling back to internal parser.",
+                        (unsigned long)nbExternalSeqs
+                    );
+                    lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+            }   }
+        } else {   /* not long range mode and no external matchfinder */
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(
+                    zc->appliedParams.cParams.strategy,
+                    zc->appliedParams.useRowMatchFinder,
+                    dictMode);
             ms->ldmSeqStore = NULL;
             lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
         }
@@ -2801,29 +3273,38 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
     return ZSTDbss_compress;
 }
 
-static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
+static size_t ZSTD_copyBlockSequences(SeqCollector* seqCollector, const seqStore_t* seqStore, const U32 prevRepcodes[ZSTD_REP_NUM])
 {
-    const seqStore_t* seqStore = ZSTD_getSeqStore(zc);
-    const seqDef* seqStoreSeqs = seqStore->sequencesStart;
-    size_t seqStoreSeqSize = seqStore->sequences - seqStoreSeqs;
-    size_t seqStoreLiteralsSize = (size_t)(seqStore->lit - seqStore->litStart);
-    size_t literalsRead = 0;
-    size_t lastLLSize;
+    const seqDef* inSeqs = seqStore->sequencesStart;
+    const size_t nbInSequences = seqStore->sequences - inSeqs;
+    const size_t nbInLiterals = (size_t)(seqStore->lit - seqStore->litStart);
 
-    ZSTD_Sequence* outSeqs = &zc->seqCollector.seqStart[zc->seqCollector.seqIndex];
+    ZSTD_Sequence* outSeqs = seqCollector->seqIndex == 0 ? seqCollector->seqStart : seqCollector->seqStart + seqCollector->seqIndex;
+    const size_t nbOutSequences = nbInSequences + 1;
+    size_t nbOutLiterals = 0;
+    repcodes_t repcodes;
     size_t i;
-    repcodes_t updatedRepcodes;
 
-    assert(zc->seqCollector.seqIndex + 1 < zc->seqCollector.maxSequences);
-    /* Ensure we have enough space for last literals "sequence" */
-    assert(zc->seqCollector.maxSequences >= seqStoreSeqSize + 1);
-    ZSTD_memcpy(updatedRepcodes.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
-    for (i = 0; i < seqStoreSeqSize; ++i) {
-        U32 rawOffset = seqStoreSeqs[i].offBase - ZSTD_REP_NUM;
-        outSeqs[i].litLength = seqStoreSeqs[i].litLength;
-        outSeqs[i].matchLength = seqStoreSeqs[i].mlBase + MINMATCH;
+    /* Bounds check that we have enough space for every input sequence
+     * and the block delimiter
+     */
+    assert(seqCollector->seqIndex <= seqCollector->maxSequences);
+    RETURN_ERROR_IF(
+        nbOutSequences > (size_t)(seqCollector->maxSequences - seqCollector->seqIndex),
+        dstSize_tooSmall,
+        "Not enough space to copy sequences");
+
+    ZSTD_memcpy(&repcodes, prevRepcodes, sizeof(repcodes));
+    for (i = 0; i < nbInSequences; ++i) {
+        U32 rawOffset;
+        outSeqs[i].litLength = inSeqs[i].litLength;
+        outSeqs[i].matchLength = inSeqs[i].mlBase + MINMATCH;
         outSeqs[i].rep = 0;
 
+        /* Handle the possible single length >= 64K
+         * There can only be one because we add MINMATCH to every match length,
+         * and blocks are at most 128K.
+         */
         if (i == seqStore->longLengthPos) {
             if (seqStore->longLengthType == ZSTD_llt_literalLength) {
                 outSeqs[i].litLength += 0x10000;
@@ -2832,37 +3313,55 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
             }
         }
 
-        if (seqStoreSeqs[i].offBase <= ZSTD_REP_NUM) {
-            /* Derive the correct offset corresponding to a repcode */
-            outSeqs[i].rep = seqStoreSeqs[i].offBase;
+        /* Determine the raw offset given the offBase, which may be a repcode. */
+        if (OFFBASE_IS_REPCODE(inSeqs[i].offBase)) {
+            const U32 repcode = OFFBASE_TO_REPCODE(inSeqs[i].offBase);
+            assert(repcode > 0);
+            outSeqs[i].rep = repcode;
             if (outSeqs[i].litLength != 0) {
-                rawOffset = updatedRepcodes.rep[outSeqs[i].rep - 1];
+                rawOffset = repcodes.rep[repcode - 1];
             } else {
-                if (outSeqs[i].rep == 3) {
-                    rawOffset = updatedRepcodes.rep[0] - 1;
+                if (repcode == 3) {
+                    assert(repcodes.rep[0] > 1);
+                    rawOffset = repcodes.rep[0] - 1;
                 } else {
-                    rawOffset = updatedRepcodes.rep[outSeqs[i].rep];
+                    rawOffset = repcodes.rep[repcode];
                 }
             }
+        } else {
+            rawOffset = OFFBASE_TO_OFFSET(inSeqs[i].offBase);
         }
         outSeqs[i].offset = rawOffset;
-        /* seqStoreSeqs[i].offset == offCode+1, and ZSTD_updateRep() expects offCode
-           so we provide seqStoreSeqs[i].offset - 1 */
-        ZSTD_updateRep(updatedRepcodes.rep,
-                       seqStoreSeqs[i].offBase - 1,
-                       seqStoreSeqs[i].litLength == 0);
-        literalsRead += outSeqs[i].litLength;
+
+        /* Update repcode history for the sequence */
+        ZSTD_updateRep(repcodes.rep,
+                       inSeqs[i].offBase,
+                       inSeqs[i].litLength == 0);
+
+        nbOutLiterals += outSeqs[i].litLength;
     }
     /* Insert last literals (if any exist) in the block as a sequence with ml == off == 0.
      * If there are no last literals, then we'll emit (of: 0, ml: 0, ll: 0), which is a marker
      * for the block boundary, according to the API.
      */
-    assert(seqStoreLiteralsSize >= literalsRead);
-    lastLLSize = seqStoreLiteralsSize - literalsRead;
-    outSeqs[i].litLength = (U32)lastLLSize;
-    outSeqs[i].matchLength = outSeqs[i].offset = outSeqs[i].rep = 0;
-    seqStoreSeqSize++;
-    zc->seqCollector.seqIndex += seqStoreSeqSize;
+    assert(nbInLiterals >= nbOutLiterals);
+    {
+        const size_t lastLLSize = nbInLiterals - nbOutLiterals;
+        outSeqs[nbInSequences].litLength = (U32)lastLLSize;
+        outSeqs[nbInSequences].matchLength = 0;
+        outSeqs[nbInSequences].offset = 0;
+        assert(nbOutSequences == nbInSequences + 1);
+    }
+    seqCollector->seqIndex += nbOutSequences;
+    assert(seqCollector->seqIndex <= seqCollector->maxSequences);
+
+    return 0;
+}
+
+size_t ZSTD_sequenceBound(size_t srcSize) {
+    const size_t maxNbSeq = (srcSize / ZSTD_MINMATCH_MIN) + 1;
+    const size_t maxNbDelims = (srcSize / ZSTD_BLOCKSIZE_MAX_MIN) + 1;
+    return maxNbSeq + maxNbDelims;
 }
 
 size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
@@ -2871,6 +3370,16 @@ size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
     const size_t dstCapacity = ZSTD_compressBound(srcSize);
     void* dst = ZSTD_customMalloc(dstCapacity, ZSTD_defaultCMem);
     SeqCollector seqCollector;
+    {
+        int targetCBlockSize;
+        FORWARD_IF_ERROR(ZSTD_CCtx_getParameter(zc, ZSTD_c_targetCBlockSize, &targetCBlockSize), "");
+        RETURN_ERROR_IF(targetCBlockSize != 0, parameter_unsupported, "targetCBlockSize != 0");
+    }
+    {
+        int nbWorkers;
+        FORWARD_IF_ERROR(ZSTD_CCtx_getParameter(zc, ZSTD_c_nbWorkers, &nbWorkers), "");
+        RETURN_ERROR_IF(nbWorkers != 0, parameter_unsupported, "nbWorkers != 0");
+    }
 
     RETURN_ERROR_IF(dst == NULL, memory_allocation, "NULL pointer!");
 
@@ -2880,8 +3389,12 @@ size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
     seqCollector.maxSequences = outSeqsSize;
     zc->seqCollector = seqCollector;
 
-    ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
-    ZSTD_customFree(dst, ZSTD_defaultCMem);
+    {
+        const size_t ret = ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
+        ZSTD_customFree(dst, ZSTD_defaultCMem);
+        FORWARD_IF_ERROR(ret, "ZSTD_compress2 failed");
+    }
+    assert(zc->seqCollector.seqIndex <= ZSTD_sequenceBound(srcSize));
     return zc->seqCollector.seqIndex;
 }
 
@@ -2910,19 +3423,17 @@ static int ZSTD_isRLE(const BYTE* src, size_t length) {
     const size_t unrollMask = unrollSize - 1;
     const size_t prefixLength = length & unrollMask;
     size_t i;
-    size_t u;
     if (length == 1) return 1;
     /* Check if prefix is RLE first before using unrolled loop */
     if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
         return 0;
     }
     for (i = prefixLength; i != length; i += unrollSize) {
+        size_t u;
         for (u = 0; u < unrollSize; u += sizeof(size_t)) {
             if (MEM_readST(ip + i + u) != valueST) {
                 return 0;
-            }
-        }
-    }
+    }   }   }
     return 1;
 }
 
@@ -2938,7 +3449,8 @@ static int ZSTD_maybeRLE(seqStore_t const* seqStore)
     return nbSeqs < 4 && nbLits < 10;
 }
 
-static void ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
+static void
+ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
 {
     ZSTD_compressedBlockState_t* const tmp = bs->prevCBlock;
     bs->prevCBlock = bs->nextCBlock;
@@ -2946,7 +3458,9 @@ static void ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* c
 }
 
 /* Writes the block header */
-static void writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock) {
+static void
+writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock)
+{
     U32 const cBlockHeader = cSize == 1 ?
                         lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
                         lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
@@ -2959,13 +3473,16 @@ static void writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastB
  *  Stores literals block type (raw, rle, compressed, repeat) and
  *  huffman description table to hufMetadata.
  *  Requires ENTROPY_WORKSPACE_SIZE workspace
- *  @return : size of huffman description table or error code */
-static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
-                                            const ZSTD_hufCTables_t* prevHuf,
-                                                  ZSTD_hufCTables_t* nextHuf,
-                                                  ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                                  const int literalsCompressionIsDisabled,
-                                                  void* workspace, size_t wkspSize)
+ * @return : size of huffman description table, or an error code
+ */
+static size_t
+ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
+                               const ZSTD_hufCTables_t* prevHuf,
+                                     ZSTD_hufCTables_t* nextHuf,
+                                     ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               const int literalsCompressionIsDisabled,
+                                     void* workspace, size_t wkspSize,
+                                     int hufFlags)
 {
     BYTE* const wkspStart = (BYTE*)workspace;
     BYTE* const wkspEnd = wkspStart + wkspSize;
@@ -2973,9 +3490,9 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
     unsigned* const countWksp = (unsigned*)workspace;
     const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
     BYTE* const nodeWksp = countWkspStart + countWkspSize;
-    const size_t nodeWkspSize = wkspEnd-nodeWksp;
+    const size_t nodeWkspSize = (size_t)(wkspEnd - nodeWksp);
     unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
-    unsigned huffLog = HUF_TABLELOG_DEFAULT;
+    unsigned huffLog = LitHufLog;
     HUF_repeat repeat = prevHuf->repeatMode;
     DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_literals (srcSize=%zu)", srcSize);
 
@@ -2990,73 +3507,77 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
 
     /* small ? don't even attempt compression (speed opt) */
 #ifndef COMPRESS_LITERALS_SIZE_MIN
-#define COMPRESS_LITERALS_SIZE_MIN 63
+# define COMPRESS_LITERALS_SIZE_MIN 63  /* heuristic */
 #endif
     {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
         if (srcSize <= minLitSize) {
             DEBUGLOG(5, "set_basic - too small");
             hufMetadata->hType = set_basic;
             return 0;
-        }
-    }
+    }   }
 
     /* Scan input and build symbol stats */
-    {   size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
+    {   size_t const largest =
+            HIST_count_wksp (countWksp, &maxSymbolValue,
+                            (const BYTE*)src, srcSize,
+                            workspace, wkspSize);
         FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
         if (largest == srcSize) {
+            /* only one literal symbol */
             DEBUGLOG(5, "set_rle");
             hufMetadata->hType = set_rle;
             return 0;
         }
         if (largest <= (srcSize >> 7)+4) {
+            /* heuristic: likely not compressible */
             DEBUGLOG(5, "set_basic - no gain");
             hufMetadata->hType = set_basic;
             return 0;
-        }
-    }
+    }   }
 
     /* Validate the previous Huffman table */
-    if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
+    if (repeat == HUF_repeat_check
+      && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
         repeat = HUF_repeat_none;
     }
 
     /* Build Huffman Tree */
     ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
-    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, nodeWksp, nodeWkspSize, nextHuf->CTable, countWksp, hufFlags);
+    assert(huffLog <= LitHufLog);
     {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
                                                     maxSymbolValue, huffLog,
                                                     nodeWksp, nodeWkspSize);
         FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
         huffLog = (U32)maxBits;
-        {   /* Build and write the CTable */
-            size_t const newCSize = HUF_estimateCompressedSize(
-                    (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
-            size_t const hSize = HUF_writeCTable_wksp(
-                    hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
-                    (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
-                    nodeWksp, nodeWkspSize);
-            /* Check against repeating the previous CTable */
-            if (repeat != HUF_repeat_none) {
-                size_t const oldCSize = HUF_estimateCompressedSize(
-                        (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
-                if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
-                    DEBUGLOG(5, "set_repeat - smaller");
-                    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-                    hufMetadata->hType = set_repeat;
-                    return 0;
-                }
-            }
-            if (newCSize + hSize >= srcSize) {
-                DEBUGLOG(5, "set_basic - no gains");
+    }
+    {   /* Build and write the CTable */
+        size_t const newCSize = HUF_estimateCompressedSize(
+                (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
+        size_t const hSize = HUF_writeCTable_wksp(
+                hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
+                (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
+                nodeWksp, nodeWkspSize);
+        /* Check against repeating the previous CTable */
+        if (repeat != HUF_repeat_none) {
+            size_t const oldCSize = HUF_estimateCompressedSize(
+                    (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
+            if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
+                DEBUGLOG(5, "set_repeat - smaller");
                 ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-                hufMetadata->hType = set_basic;
+                hufMetadata->hType = set_repeat;
                 return 0;
-            }
-            DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
-            hufMetadata->hType = set_compressed;
-            nextHuf->repeatMode = HUF_repeat_check;
-            return hSize;
+        }   }
+        if (newCSize + hSize >= srcSize) {
+            DEBUGLOG(5, "set_basic - no gains");
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            hufMetadata->hType = set_basic;
+            return 0;
         }
+        DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
+        hufMetadata->hType = set_compressed;
+        nextHuf->repeatMode = HUF_repeat_check;
+        return hSize;
     }
 }
 
@@ -3066,8 +3587,9 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
  * and updates nextEntropy to the appropriate repeatMode.
  */
 static ZSTD_symbolEncodingTypeStats_t
-ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy) {
-    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0};
+ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy)
+{
+    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0, 0};
     nextEntropy->litlength_repeatMode = FSE_repeat_none;
     nextEntropy->offcode_repeatMode = FSE_repeat_none;
     nextEntropy->matchlength_repeatMode = FSE_repeat_none;
@@ -3078,16 +3600,18 @@ ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy) {
  *  Builds entropy for the sequences.
  *  Stores symbol compression modes and fse table to fseMetadata.
  *  Requires ENTROPY_WORKSPACE_SIZE wksp.
- *  @return : size of fse tables or error code */
-static size_t ZSTD_buildBlockEntropyStats_sequences(seqStore_t* seqStorePtr,
-                                              const ZSTD_fseCTables_t* prevEntropy,
-                                                    ZSTD_fseCTables_t* nextEntropy,
-                                              const ZSTD_CCtx_params* cctxParams,
-                                                    ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                                    void* workspace, size_t wkspSize)
+ * @return : size of fse tables or error code */
+static size_t
+ZSTD_buildBlockEntropyStats_sequences(
+                const seqStore_t* seqStorePtr,
+                const ZSTD_fseCTables_t* prevEntropy,
+                      ZSTD_fseCTables_t* nextEntropy,
+                const ZSTD_CCtx_params* cctxParams,
+                      ZSTD_fseCTablesMetadata_t* fseMetadata,
+                      void* workspace, size_t wkspSize)
 {
     ZSTD_strategy const strategy = cctxParams->cParams.strategy;
-    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     BYTE* const ostart = fseMetadata->fseTablesBuffer;
     BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
     BYTE* op = ostart;
@@ -3114,23 +3638,28 @@ static size_t ZSTD_buildBlockEntropyStats_sequences(seqStore_t* seqStorePtr,
 /* ZSTD_buildBlockEntropyStats() :
  *  Builds entropy for the block.
  *  Requires workspace size ENTROPY_WORKSPACE_SIZE
- *
- *  @return : 0 on success or error code
+ * @return : 0 on success, or an error code
+ *  Note : also employed in superblock
  */
-size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
-                             const ZSTD_entropyCTables_t* prevEntropy,
-                                   ZSTD_entropyCTables_t* nextEntropy,
-                             const ZSTD_CCtx_params* cctxParams,
-                                   ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                   void* workspace, size_t wkspSize)
-{
-    size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
+size_t ZSTD_buildBlockEntropyStats(
+            const seqStore_t* seqStorePtr,
+            const ZSTD_entropyCTables_t* prevEntropy,
+                  ZSTD_entropyCTables_t* nextEntropy,
+            const ZSTD_CCtx_params* cctxParams,
+                  ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                  void* workspace, size_t wkspSize)
+{
+    size_t const litSize = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+    int const huf_useOptDepth = (cctxParams->cParams.strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD);
+    int const hufFlags = huf_useOptDepth ? HUF_flags_optimalDepth : 0;
+
     entropyMetadata->hufMetadata.hufDesSize =
         ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
                                             &prevEntropy->huf, &nextEntropy->huf,
                                             &entropyMetadata->hufMetadata,
                                             ZSTD_literalsCompressionIsDisabled(cctxParams),
-                                            workspace, wkspSize);
+                                            workspace, wkspSize, hufFlags);
+
     FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
     entropyMetadata->fseMetadata.fseTablesSize =
         ZSTD_buildBlockEntropyStats_sequences(seqStorePtr,
@@ -3143,11 +3672,12 @@ size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
 }
 
 /* Returns the size estimate for the literals section (header + content) of a block */
-static size_t ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
-                                                const ZSTD_hufCTables_t* huf,
-                                                const ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                                void* workspace, size_t wkspSize,
-                                                int writeEntropy)
+static size_t
+ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
+                               const ZSTD_hufCTables_t* huf,
+                               const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               void* workspace, size_t wkspSize,
+                               int writeEntropy)
 {
     unsigned* const countWksp = (unsigned*)workspace;
     unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
@@ -3169,12 +3699,13 @@ static size_t ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSiz
 }
 
 /* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
-static size_t ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
-                        const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
-                        const FSE_CTable* fseCTable,
-                        const U8* additionalBits,
-                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
-                        void* workspace, size_t wkspSize)
+static size_t
+ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
+                    const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
+                    const FSE_CTable* fseCTable,
+                    const U8* additionalBits,
+                    short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                    void* workspace, size_t wkspSize)
 {
     unsigned* const countWksp = (unsigned*)workspace;
     const BYTE* ctp = codeTable;
@@ -3206,99 +3737,107 @@ static size_t ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
 }
 
 /* Returns the size estimate for the sequences section (header + content) of a block */
-static size_t ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
-                                                  const BYTE* llCodeTable,
-                                                  const BYTE* mlCodeTable,
-                                                  size_t nbSeq,
-                                                  const ZSTD_fseCTables_t* fseTables,
-                                                  const ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                                  void* workspace, size_t wkspSize,
-                                                  int writeEntropy)
+static size_t
+ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
+                                 const BYTE* llCodeTable,
+                                 const BYTE* mlCodeTable,
+                                 size_t nbSeq,
+                                 const ZSTD_fseCTables_t* fseTables,
+                                 const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                 void* workspace, size_t wkspSize,
+                                 int writeEntropy)
 {
     size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
     size_t cSeqSizeEstimate = 0;
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
-                                         fseTables->offcodeCTable, NULL,
-                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
-                                         workspace, wkspSize);
+                                    fseTables->offcodeCTable, NULL,
+                                    OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                    workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
-                                         fseTables->litlengthCTable, LL_bits,
-                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
-                                         workspace, wkspSize);
+                                    fseTables->litlengthCTable, LL_bits,
+                                    LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                    workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
-                                         fseTables->matchlengthCTable, ML_bits,
-                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
-                                         workspace, wkspSize);
+                                    fseTables->matchlengthCTable, ML_bits,
+                                    ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                    workspace, wkspSize);
     if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
     return cSeqSizeEstimate + sequencesSectionHeaderSize;
 }
 
 /* Returns the size estimate for a given stream of literals, of, ll, ml */
-static size_t ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
-                                     const BYTE* ofCodeTable,
-                                     const BYTE* llCodeTable,
-                                     const BYTE* mlCodeTable,
-                                     size_t nbSeq,
-                                     const ZSTD_entropyCTables_t* entropy,
-                                     const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                     void* workspace, size_t wkspSize,
-                                     int writeLitEntropy, int writeSeqEntropy) {
+static size_t
+ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
+                       const BYTE* ofCodeTable,
+                       const BYTE* llCodeTable,
+                       const BYTE* mlCodeTable,
+                       size_t nbSeq,
+                       const ZSTD_entropyCTables_t* entropy,
+                       const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                       void* workspace, size_t wkspSize,
+                       int writeLitEntropy, int writeSeqEntropy)
+{
     size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
-                                                         &entropy->huf, &entropyMetadata->hufMetadata,
-                                                         workspace, wkspSize, writeLitEntropy);
+                                    &entropy->huf, &entropyMetadata->hufMetadata,
+                                    workspace, wkspSize, writeLitEntropy);
     size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
-                                                         nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
-                                                         workspace, wkspSize, writeSeqEntropy);
+                                    nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                    workspace, wkspSize, writeSeqEntropy);
     return seqSize + literalsSize + ZSTD_blockHeaderSize;
 }
 
 /* Builds entropy statistics and uses them for blocksize estimation.
  *
- * Returns the estimated compressed size of the seqStore, or a zstd error.
+ * @return: estimated compressed size of the seqStore, or a zstd error.
  */
-static size_t ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, ZSTD_CCtx* zc) {
-    ZSTD_entropyCTablesMetadata_t* entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
+static size_t
+ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, ZSTD_CCtx* zc)
+{
+    ZSTD_entropyCTablesMetadata_t* const entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
     DEBUGLOG(6, "ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize()");
     FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
                     &zc->blockState.prevCBlock->entropy,
                     &zc->blockState.nextCBlock->entropy,
                     &zc->appliedParams,
                     entropyMetadata,
-                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
-    return ZSTD_estimateBlockSize(seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE), "");
+    return ZSTD_estimateBlockSize(
+                    seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
                     seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
                     (size_t)(seqStore->sequences - seqStore->sequencesStart),
-                    &zc->blockState.nextCBlock->entropy, entropyMetadata, zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
+                    &zc->blockState.nextCBlock->entropy,
+                    entropyMetadata,
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
                     (int)(entropyMetadata->hufMetadata.hType == set_compressed), 1);
 }
 
 /* Returns literals bytes represented in a seqStore */
-static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore) {
+static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore)
+{
     size_t literalsBytes = 0;
-    size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
     size_t i;
     for (i = 0; i < nbSeqs; ++i) {
-        seqDef seq = seqStore->sequencesStart[i];
+        seqDef const seq = seqStore->sequencesStart[i];
         literalsBytes += seq.litLength;
         if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_literalLength) {
             literalsBytes += 0x10000;
-        }
-    }
+    }   }
     return literalsBytes;
 }
 
 /* Returns match bytes represented in a seqStore */
-static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore) {
+static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore)
+{
     size_t matchBytes = 0;
-    size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
     size_t i;
     for (i = 0; i < nbSeqs; ++i) {
         seqDef seq = seqStore->sequencesStart[i];
         matchBytes += seq.mlBase + MINMATCH;
         if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_matchLength) {
             matchBytes += 0x10000;
-        }
-    }
+    }   }
     return matchBytes;
 }
 
@@ -3307,15 +3846,12 @@ static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore) {
  */
 static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
                                const seqStore_t* originalSeqStore,
-                                     size_t startIdx, size_t endIdx) {
-    BYTE* const litEnd = originalSeqStore->lit;
-    size_t literalsBytes;
-    size_t literalsBytesPreceding = 0;
-
+                                     size_t startIdx, size_t endIdx)
+{
     *resultSeqStore = *originalSeqStore;
     if (startIdx > 0) {
         resultSeqStore->sequences = originalSeqStore->sequencesStart + startIdx;
-        literalsBytesPreceding = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->litStart += ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
     }
 
     /* Move longLengthPos into the correct position if necessary */
@@ -3328,13 +3864,12 @@ static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
     }
     resultSeqStore->sequencesStart = originalSeqStore->sequencesStart + startIdx;
     resultSeqStore->sequences = originalSeqStore->sequencesStart + endIdx;
-    literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
-    resultSeqStore->litStart += literalsBytesPreceding;
     if (endIdx == (size_t)(originalSeqStore->sequences - originalSeqStore->sequencesStart)) {
         /* This accounts for possible last literals if the derived chunk reaches the end of the block */
-        resultSeqStore->lit = litEnd;
+        assert(resultSeqStore->lit == originalSeqStore->lit);
     } else {
-        resultSeqStore->lit = resultSeqStore->litStart+literalsBytes;
+        size_t const literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->lit = resultSeqStore->litStart + literalsBytes;
     }
     resultSeqStore->llCode += startIdx;
     resultSeqStore->mlCode += startIdx;
@@ -3342,20 +3877,26 @@ static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
 }
 
 /*
- * Returns the raw offset represented by the combination of offCode, ll0, and repcode history.
- * offCode must represent a repcode in the numeric representation of ZSTD_storeSeq().
+ * Returns the raw offset represented by the combination of offBase, ll0, and repcode history.
+ * offBase must represent a repcode in the numeric representation of ZSTD_storeSeq().
  */
 static U32
-ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offCode, const U32 ll0)
-{
-    U32 const adjustedOffCode = STORED_REPCODE(offCode) - 1 + ll0;  /* [ 0 - 3 ] */
-    assert(STORED_IS_REPCODE(offCode));
-    if (adjustedOffCode == ZSTD_REP_NUM) {
-        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1 */
-        assert(rep[0] > 0);
+ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offBase, const U32 ll0)
+{
+    U32 const adjustedRepCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;  /* [ 0 - 3 ] */
+    assert(OFFBASE_IS_REPCODE(offBase));
+    if (adjustedRepCode == ZSTD_REP_NUM) {
+        assert(ll0);
+        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1
+         * This is only valid if it results in a valid offset value, aka > 0.
+         * Note : it may happen that `rep[0]==1` in exceptional circumstances.
+         * In which case this function will return 0, which is an invalid offset.
+         * It's not an issue though, since this value will be
+         * compared and discarded within ZSTD_seqStore_resolveOffCodes().
+         */
         return rep[0] - 1;
     }
-    return rep[adjustedOffCode];
+    return rep[adjustedRepCode];
 }
 
 /*
@@ -3371,30 +3912,33 @@ ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offCode, c
  *        1-3 : repcode 1-3
  *        4+ : real_offset+3
  */
-static void ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_t* const cRepcodes,
-                                          seqStore_t* const seqStore, U32 const nbSeq) {
+static void
+ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_t* const cRepcodes,
+                        const seqStore_t* const seqStore, U32 const nbSeq)
+{
     U32 idx = 0;
+    U32 const longLitLenIdx = seqStore->longLengthType == ZSTD_llt_literalLength ? seqStore->longLengthPos : nbSeq;
     for (; idx < nbSeq; ++idx) {
         seqDef* const seq = seqStore->sequencesStart + idx;
-        U32 const ll0 = (seq->litLength == 0);
-        U32 const offCode = OFFBASE_TO_STORED(seq->offBase);
-        assert(seq->offBase > 0);
-        if (STORED_IS_REPCODE(offCode)) {
-            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offCode, ll0);
-            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offCode, ll0);
+        U32 const ll0 = (seq->litLength == 0) && (idx != longLitLenIdx);
+        U32 const offBase = seq->offBase;
+        assert(offBase > 0);
+        if (OFFBASE_IS_REPCODE(offBase)) {
+            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offBase, ll0);
+            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offBase, ll0);
             /* Adjust simulated decompression repcode history if we come across a mismatch. Replace
              * the repcode with the offset it actually references, determined by the compression
              * repcode history.
              */
             if (dRawOffset != cRawOffset) {
-                seq->offBase = cRawOffset + ZSTD_REP_NUM;
+                seq->offBase = OFFSET_TO_OFFBASE(cRawOffset);
             }
         }
         /* Compression repcode history is always updated with values directly from the unmodified seqStore.
          * Decompression repcode history may use modified seq->offset value taken from compression repcode history.
          */
-        ZSTD_updateRep(dRepcodes->rep, OFFBASE_TO_STORED(seq->offBase), ll0);
-        ZSTD_updateRep(cRepcodes->rep, offCode, ll0);
+        ZSTD_updateRep(dRepcodes->rep, seq->offBase, ll0);
+        ZSTD_updateRep(cRepcodes->rep, offBase, ll0);
     }
 }
 
@@ -3404,10 +3948,11 @@ static void ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_
  * Returns the total size of that block (including header) or a ZSTD error code.
  */
 static size_t
-ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const seqStore,
+ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc,
+                            const seqStore_t* const seqStore,
                                   repcodes_t* const dRep, repcodes_t* const cRep,
                                   void* dst, size_t dstCapacity,
-                                  const void* src, size_t srcSize,
+                            const void* src, size_t srcSize,
                                   U32 lastBlock, U32 isPartition)
 {
     const U32 rleMaxLength = 25;
@@ -3442,8 +3987,9 @@ ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const seqStore,
         cSeqsSize = 1;
     }
 
+    /* Sequence collection not supported when block splitting */
     if (zc->seqCollector.collectSequences) {
-        ZSTD_copyBlockSequences(zc);
+        FORWARD_IF_ERROR(ZSTD_copyBlockSequences(&zc->seqCollector, seqStore, dRepOriginal.rep), "copyBlockSequences failed");
         ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
         return 0;
     }
@@ -3481,45 +4027,49 @@ typedef struct {
 
 /* Helper function to perform the recursive search for block splits.
  * Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
- * If advantageous to split, then we recurse down the two sub-blocks. If not, or if an error occurred in estimation, then
- * we do not recurse.
+ * If advantageous to split, then we recurse down the two sub-blocks.
+ * If not, or if an error occurred in estimation, then we do not recurse.
  *
- * Note: The recursion depth is capped by a heuristic minimum number of sequences, defined by MIN_SEQUENCES_BLOCK_SPLITTING.
+ * Note: The recursion depth is capped by a heuristic minimum number of sequences,
+ * defined by MIN_SEQUENCES_BLOCK_SPLITTING.
  * In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
  * In practice, recursion depth usually doesn't go beyond 4.
  *
- * Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS. At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
+ * Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS.
+ * At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
  * maximum of 128 KB, this value is actually impossible to reach.
  */
 static void
 ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
                              ZSTD_CCtx* zc, const seqStore_t* origSeqStore)
 {
-    seqStore_t* fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
-    seqStore_t* firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
-    seqStore_t* secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
+    seqStore_t* const fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
+    seqStore_t* const firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
+    seqStore_t* const secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
     size_t estimatedOriginalSize;
     size_t estimatedFirstHalfSize;
     size_t estimatedSecondHalfSize;
     size_t midIdx = (startIdx + endIdx)/2;
 
+    DEBUGLOG(5, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
+    assert(endIdx >= startIdx);
     if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= ZSTD_MAX_NB_BLOCK_SPLITS) {
-        DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences");
+        DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences (%zu)", endIdx - startIdx);
         return;
     }
-    DEBUGLOG(4, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
     ZSTD_deriveSeqStoreChunk(fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
     ZSTD_deriveSeqStoreChunk(firstHalfSeqStore, origSeqStore, startIdx, midIdx);
     ZSTD_deriveSeqStoreChunk(secondHalfSeqStore, origSeqStore, midIdx, endIdx);
     estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(fullSeqStoreChunk, zc);
     estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(firstHalfSeqStore, zc);
     estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(secondHalfSeqStore, zc);
-    DEBUGLOG(4, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
+    DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
              estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
     if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
         return;
     }
     if (estimatedFirstHalfSize + estimatedSecondHalfSize < estimatedOriginalSize) {
+        DEBUGLOG(5, "split decided at seqNb:%zu", midIdx);
         ZSTD_deriveBlockSplitsHelper(splits, startIdx, midIdx, zc, origSeqStore);
         splits->splitLocations[splits->idx] = (U32)midIdx;
         splits->idx++;
@@ -3527,14 +4077,18 @@ ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t end
     }
 }
 
-/* Base recursive function. Populates a table with intra-block partition indices that can improve compression ratio.
+/* Base recursive function.
+ * Populates a table with intra-block partition indices that can improve compression ratio.
  *
- * Returns the number of splits made (which equals the size of the partition table - 1).
+ * @return: number of splits made (which equals the size of the partition table - 1).
  */
-static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq) {
-    seqStoreSplits splits = {partitions, 0};
+static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
+{
+    seqStoreSplits splits;
+    splits.splitLocations = partitions;
+    splits.idx = 0;
     if (nbSeq <= 4) {
-        DEBUGLOG(4, "ZSTD_deriveBlockSplits: Too few sequences to split");
+        DEBUGLOG(5, "ZSTD_deriveBlockSplits: Too few sequences to split (%u <= 4)", nbSeq);
         /* Refuse to try and split anything with less than 4 sequences */
         return 0;
     }
@@ -3550,18 +4104,20 @@ static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
  * Returns combined size of all blocks (which includes headers), or a ZSTD error code.
  */
 static size_t
-ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity,
-                                       const void* src, size_t blockSize, U32 lastBlock, U32 nbSeq)
+ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t blockSize,
+                                    U32 lastBlock, U32 nbSeq)
 {
     size_t cSize = 0;
     const BYTE* ip = (const BYTE*)src;
     BYTE* op = (BYTE*)dst;
     size_t i = 0;
     size_t srcBytesTotal = 0;
-    U32* partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
-    seqStore_t* nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
-    seqStore_t* currSeqStore = &zc->blockSplitCtx.currSeqStore;
-    size_t numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
+    U32* const partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
+    seqStore_t* const nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
+    seqStore_t* const currSeqStore = &zc->blockSplitCtx.currSeqStore;
+    size_t const numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
 
     /* If a block is split and some partitions are emitted as RLE/uncompressed, then repcode history
      * may become invalid. In order to reconcile potentially invalid repcodes, we keep track of two
@@ -3583,30 +4139,31 @@ ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapac
     ZSTD_memcpy(cRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
     ZSTD_memset(nextSeqStore, 0, sizeof(seqStore_t));
 
-    DEBUGLOG(4, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+    DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
                 (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
                 (unsigned)zc->blockState.matchState.nextToUpdate);
 
     if (numSplits == 0) {
-        size_t cSizeSingleBlock = ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
-                                                                   &dRep, &cRep,
-                                                                    op, dstCapacity,
-                                                                    ip, blockSize,
-                                                                    lastBlock, 0 /* isPartition */);
+        size_t cSizeSingleBlock =
+            ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
+                                            &dRep, &cRep,
+                                            op, dstCapacity,
+                                            ip, blockSize,
+                                            lastBlock, 0 /* isPartition */);
         FORWARD_IF_ERROR(cSizeSingleBlock, "Compressing single block from splitBlock_internal() failed!");
         DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: No splits");
-        assert(cSizeSingleBlock <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
+        assert(zc->blockSize <= ZSTD_BLOCKSIZE_MAX);
+        assert(cSizeSingleBlock <= zc->blockSize + ZSTD_blockHeaderSize);
         return cSizeSingleBlock;
     }
 
     ZSTD_deriveSeqStoreChunk(currSeqStore, &zc->seqStore, 0, partitions[0]);
     for (i = 0; i <= numSplits; ++i) {
-        size_t srcBytes;
         size_t cSizeChunk;
         U32 const lastPartition = (i == numSplits);
         U32 lastBlockEntireSrc = 0;
 
-        srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
+        size_t srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
         srcBytesTotal += srcBytes;
         if (lastPartition) {
             /* This is the final partition, need to account for possible last literals */
@@ -3621,7 +4178,8 @@ ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapac
                                                        op, dstCapacity,
                                                        ip, srcBytes,
                                                        lastBlockEntireSrc, 1 /* isPartition */);
-        DEBUGLOG(5, "Estimated size: %zu actual size: %zu", ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
+        DEBUGLOG(5, "Estimated size: %zu vs %zu : actual size",
+                    ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
         FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
 
         ip += srcBytes;
@@ -3629,10 +4187,10 @@ ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapac
         dstCapacity -= cSizeChunk;
         cSize += cSizeChunk;
         *currSeqStore = *nextSeqStore;
-        assert(cSizeChunk <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
+        assert(cSizeChunk <= zc->blockSize + ZSTD_blockHeaderSize);
     }
-    /* cRep and dRep may have diverged during the compression. If so, we use the dRep repcodes
-     * for the next block.
+    /* cRep and dRep may have diverged during the compression.
+     * If so, we use the dRep repcodes for the next block.
      */
     ZSTD_memcpy(zc->blockState.prevCBlock->rep, dRep.rep, sizeof(repcodes_t));
     return cSize;
@@ -3643,8 +4201,6 @@ ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
                               void* dst, size_t dstCapacity,
                               const void* src, size_t srcSize, U32 lastBlock)
 {
-    const BYTE* ip = (const BYTE*)src;
-    BYTE* op = (BYTE*)dst;
     U32 nbSeq;
     size_t cSize;
     DEBUGLOG(4, "ZSTD_compressBlock_splitBlock");
@@ -3655,7 +4211,8 @@ ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
         if (bss == ZSTDbss_noCompress) {
             if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
                 zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
-            cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
+            RETURN_ERROR_IF(zc->seqCollector.collectSequences, sequenceProducer_failed, "Uncompressible block");
+            cSize = ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
             FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
             DEBUGLOG(4, "ZSTD_compressBlock_splitBlock: Nocompress block");
             return cSize;
@@ -3673,9 +4230,9 @@ ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
                             void* dst, size_t dstCapacity,
                             const void* src, size_t srcSize, U32 frame)
 {
-    /* This the upper bound for the length of an rle block.
-     * This isn't the actual upper bound. Finding the real threshold
-     * needs further investigation.
+    /* This is an estimated upper bound for the length of an rle block.
+     * This isn't the actual upper bound.
+     * Finding the real threshold needs further investigation.
      */
     const U32 rleMaxLength = 25;
     size_t cSize;
@@ -3687,11 +4244,15 @@ ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
 
     {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
         FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
-        if (bss == ZSTDbss_noCompress) { cSize = 0; goto out; }
+        if (bss == ZSTDbss_noCompress) {
+            RETURN_ERROR_IF(zc->seqCollector.collectSequences, sequenceProducer_failed, "Uncompressible block");
+            cSize = 0;
+            goto out;
+        }
     }
 
     if (zc->seqCollector.collectSequences) {
-        ZSTD_copyBlockSequences(zc);
+        FORWARD_IF_ERROR(ZSTD_copyBlockSequences(&zc->seqCollector, ZSTD_getSeqStore(zc), zc->blockState.prevCBlock->rep), "copyBlockSequences failed");
         ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
         return 0;
     }
@@ -3767,10 +4328,11 @@ static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
          *   * cSize >= blockBound(srcSize): We have expanded the block too much so
          *     emit an uncompressed block.
          */
-        {
-            size_t const cSize = ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
+        {   size_t const cSize =
+                ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
             if (cSize != ERROR(dstSize_tooSmall)) {
-                size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
+                size_t const maxCSize =
+                    srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
                 FORWARD_IF_ERROR(cSize, "ZSTD_compressSuperBlock failed");
                 if (cSize != 0 && cSize < maxCSize + ZSTD_blockHeaderSize) {
                     ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
@@ -3778,7 +4340,7 @@ static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
                 }
             }
         }
-    }
+    } /* if (bss == ZSTDbss_compress)*/
 
     DEBUGLOG(6, "Resorting to ZSTD_noCompressBlock()");
     /* Superblock compression failed, attempt to emit a single no compress block.
@@ -3836,7 +4398,7 @@ static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
 *   All blocks will be terminated, all input will be consumed.
 *   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
 *   Frame is supposed already started (header already produced)
-*   @return : compressed size, or an error code
+*  @return : compressed size, or an error code
 */
 static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
                                      void* dst, size_t dstCapacity,
@@ -3860,7 +4422,9 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
         ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
         U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
 
-        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE,
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE + 1,
                         dstSize_tooSmall,
                         "not enough space to store compressed block");
         if (remaining < blockSize) blockSize = remaining;
@@ -3899,7 +4463,7 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
                     MEM_writeLE24(op, cBlockHeader);
                     cSize += ZSTD_blockHeaderSize;
                 }
-            }
+            }  /* if (ZSTD_useTargetCBlockSize(&cctx->appliedParams))*/
 
 
             ip += blockSize;
@@ -4001,19 +4565,15 @@ size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
     }
 }
 
-size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+void ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
 {
-    RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong,
-                    "wrong cctx stage");
-    RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable,
-                    parameter_unsupported,
-                    "incompatible with ldm");
+    assert(cctx->stage == ZSTDcs_init);
+    assert(nbSeq == 0 || cctx->appliedParams.ldmParams.enableLdm != ZSTD_ps_enable);
     cctx->externSeqStore.seq = seq;
     cctx->externSeqStore.size = nbSeq;
     cctx->externSeqStore.capacity = nbSeq;
     cctx->externSeqStore.pos = 0;
     cctx->externSeqStore.posInSequence = 0;
-    return 0;
 }
 
 
@@ -4078,31 +4638,51 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
     }
 }
 
-size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
-                              void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize)
+size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (unsigned)srcSize);
     return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
 }
 
+/* NOTE: Must just wrap ZSTD_compressContinue_public() */
+size_t ZSTD_compressContinue(ZSTD_CCtx* cctx,
+                             void* dst, size_t dstCapacity,
+                       const void* src, size_t srcSize)
+{
+    return ZSTD_compressContinue_public(cctx, dst, dstCapacity, src, srcSize);
+}
 
-size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
+static size_t ZSTD_getBlockSize_deprecated(const ZSTD_CCtx* cctx)
 {
     ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
     assert(!ZSTD_checkCParams(cParams));
-    return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
+    return MIN(cctx->appliedParams.maxBlockSize, (size_t)1 << cParams.windowLog);
 }
 
-size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+/* NOTE: Must just wrap ZSTD_getBlockSize_deprecated() */
+size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
+{
+    return ZSTD_getBlockSize_deprecated(cctx);
+}
+
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_compressBlock: srcSize = %u", (unsigned)srcSize);
-    { size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
+    { size_t const blockSizeMax = ZSTD_getBlockSize_deprecated(cctx);
       RETURN_ERROR_IF(srcSize > blockSizeMax, srcSize_wrong, "input is larger than a block"); }
 
     return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
 }
 
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_deprecated(cctx, dst, dstCapacity, src, srcSize);
+}
+
 /*! ZSTD_loadDictionaryContent() :
  *  @return : 0, or an error code
  */
@@ -4111,25 +4691,36 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
                                          ZSTD_cwksp* ws,
                                          ZSTD_CCtx_params const* params,
                                          const void* src, size_t srcSize,
-                                         ZSTD_dictTableLoadMethod_e dtlm)
+                                         ZSTD_dictTableLoadMethod_e dtlm,
+                                         ZSTD_tableFillPurpose_e tfp)
 {
     const BYTE* ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
     int const loadLdmDict = params->ldmParams.enableLdm == ZSTD_ps_enable && ls != NULL;
 
-    /* Assert that we the ms params match the params we're being given */
+    /* Assert that the ms params match the params we're being given */
     ZSTD_assertEqualCParams(params->cParams, ms->cParams);
 
-    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+    {   /* Ensure large dictionaries can't cause index overflow */
+
         /* Allow the dictionary to set indices up to exactly ZSTD_CURRENT_MAX.
          * Dictionaries right at the edge will immediately trigger overflow
          * correction, but I don't want to insert extra constraints here.
          */
-        U32 const maxDictSize = ZSTD_CURRENT_MAX - 1;
-        /* We must have cleared our windows when our source is this large. */
-        assert(ZSTD_window_isEmpty(ms->window));
-        if (loadLdmDict)
-            assert(ZSTD_window_isEmpty(ls->window));
+        U32 maxDictSize = ZSTD_CURRENT_MAX - ZSTD_WINDOW_START_INDEX;
+
+        int const CDictTaggedIndices = ZSTD_CDictIndicesAreTagged(&params->cParams);
+        if (CDictTaggedIndices && tfp == ZSTD_tfp_forCDict) {
+            /* Some dictionary matchfinders in zstd use "short cache",
+             * which treats the lower ZSTD_SHORT_CACHE_TAG_BITS of each
+             * CDict hashtable entry as a tag rather than as part of an index.
+             * When short cache is used, we need to truncate the dictionary
+             * so that its indices don't overlap with the tag. */
+            U32 const shortCacheMaxDictSize = (1u << (32 - ZSTD_SHORT_CACHE_TAG_BITS)) - ZSTD_WINDOW_START_INDEX;
+            maxDictSize = MIN(maxDictSize, shortCacheMaxDictSize);
+            assert(!loadLdmDict);
+        }
+
         /* If the dictionary is too large, only load the suffix of the dictionary. */
         if (srcSize > maxDictSize) {
             ip = iend - maxDictSize;
@@ -4138,35 +4729,58 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
         }
     }
 
-    DEBUGLOG(4, "ZSTD_loadDictionaryContent(): useRowMatchFinder=%d", (int)params->useRowMatchFinder);
+    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+        /* We must have cleared our windows when our source is this large. */
+        assert(ZSTD_window_isEmpty(ms->window));
+        if (loadLdmDict) assert(ZSTD_window_isEmpty(ls->window));
+    }
     ZSTD_window_update(&ms->window, src, srcSize, /* forceNonContiguous */ 0);
-    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
-    ms->forceNonContiguous = params->deterministicRefPrefix;
 
-    if (loadLdmDict) {
+    DEBUGLOG(4, "ZSTD_loadDictionaryContent(): useRowMatchFinder=%d", (int)params->useRowMatchFinder);
+
+    if (loadLdmDict) { /* Load the entire dict into LDM matchfinders. */
         ZSTD_window_update(&ls->window, src, srcSize, /* forceNonContiguous */ 0);
         ls->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ls->window.base);
+        ZSTD_ldm_fillHashTable(ls, ip, iend, &params->ldmParams);
     }
 
+    /* If the dict is larger than we can reasonably index in our tables, only load the suffix. */
+    if (params->cParams.strategy < ZSTD_btultra) {
+        U32 maxDictSize = 8U << MIN(MAX(params->cParams.hashLog, params->cParams.chainLog), 28);
+        if (srcSize > maxDictSize) {
+            ip = iend - maxDictSize;
+            src = ip;
+            srcSize = maxDictSize;
+        }
+    }
+
+    ms->nextToUpdate = (U32)(ip - ms->window.base);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+    ms->forceNonContiguous = params->deterministicRefPrefix;
+
     if (srcSize <= HASH_READ_SIZE) return 0;
 
     ZSTD_overflowCorrectIfNeeded(ms, ws, params, ip, iend);
 
-    if (loadLdmDict)
-        ZSTD_ldm_fillHashTable(ls, ip, iend, &params->ldmParams);
-
     switch(params->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, iend, dtlm);
+        ZSTD_fillHashTable(ms, iend, dtlm, tfp);
         break;
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, iend, dtlm);
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+        ZSTD_fillDoubleHashTable(ms, iend, dtlm, tfp);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
         break;
 
     case ZSTD_greedy:
     case ZSTD_lazy:
     case ZSTD_lazy2:
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR)
         assert(srcSize >= HASH_READ_SIZE);
         if (ms->dedicatedDictSearch) {
             assert(ms->chainTable != NULL);
@@ -4174,7 +4788,7 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
         } else {
             assert(params->useRowMatchFinder != ZSTD_ps_auto);
             if (params->useRowMatchFinder == ZSTD_ps_enable) {
-                size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog) * sizeof(U16);
+                size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog);
                 ZSTD_memset(ms->tagTable, 0, tagTableSize);
                 ZSTD_row_update(ms, iend-HASH_READ_SIZE);
                 DEBUGLOG(4, "Using row-based hash table for lazy dict");
@@ -4183,14 +4797,23 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
                 DEBUGLOG(4, "Using chain-based hash table for lazy dict");
             }
         }
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
         break;
 
     case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
     case ZSTD_btopt:
     case ZSTD_btultra:
     case ZSTD_btultra2:
+#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
         assert(srcSize >= HASH_READ_SIZE);
         ZSTD_updateTree(ms, iend-HASH_READ_SIZE, iend);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
         break;
 
     default:
@@ -4237,11 +4860,10 @@ size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
 
         /* We only set the loaded table as valid if it contains all non-zero
          * weights. Otherwise, we set it to check */
-        if (!hasZeroWeights)
+        if (!hasZeroWeights && maxSymbolValue == 255)
             bs->entropy.huf.repeatMode = HUF_repeat_valid;
 
         RETURN_ERROR_IF(HUF_isError(hufHeaderSize), dictionary_corrupted, "");
-        RETURN_ERROR_IF(maxSymbolValue < 255, dictionary_corrupted, "");
         dictPtr += hufHeaderSize;
     }
 
@@ -4327,6 +4949,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
                                       ZSTD_CCtx_params const* params,
                                       const void* dict, size_t dictSize,
                                       ZSTD_dictTableLoadMethod_e dtlm,
+                                      ZSTD_tableFillPurpose_e tfp,
                                       void* workspace)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
@@ -4345,7 +4968,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
     {
         size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
         FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
-            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm), "");
+            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm, tfp), "");
     }
     return dictID;
 }
@@ -4361,6 +4984,7 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
                          const void* dict, size_t dictSize,
                                ZSTD_dictContentType_e dictContentType,
                                ZSTD_dictTableLoadMethod_e dtlm,
+                               ZSTD_tableFillPurpose_e tfp,
                                void* workspace)
 {
     DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
@@ -4373,13 +4997,13 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
 
     /* dict restricted modes */
     if (dictContentType == ZSTD_dct_rawContent)
-        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm);
+        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm, tfp);
 
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
         if (dictContentType == ZSTD_dct_auto) {
             DEBUGLOG(4, "raw content dictionary detected");
             return ZSTD_loadDictionaryContent(
-                ms, ls, ws, params, dict, dictSize, dtlm);
+                ms, ls, ws, params, dict, dictSize, dtlm, tfp);
         }
         RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
         assert(0);   /* impossible */
@@ -4387,13 +5011,14 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
 
     /* dict as full zstd dictionary */
     return ZSTD_loadZstdDictionary(
-        bs, ms, ws, params, dict, dictSize, dtlm, workspace);
+        bs, ms, ws, params, dict, dictSize, dtlm, tfp, workspace);
 }
 
 #define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
 #define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6ULL)
 
 /*! ZSTD_compressBegin_internal() :
+ * Assumption : either @dict OR @cdict (or none) is non-NULL, never both
  * @return : 0, or an error code */
 static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                                     const void* dict, size_t dictSize,
@@ -4426,11 +5051,11 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                         &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
                         cdict->dictContentSize, cdict->dictContentType, dtlm,
-                        cctx->entropyWorkspace)
+                        ZSTD_tfp_forCCtx, cctx->entropyWorkspace)
               : ZSTD_compress_insertDictionary(
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                         &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, dict, dictSize,
-                        dictContentType, dtlm, cctx->entropyWorkspace);
+                        dictContentType, dtlm, ZSTD_tfp_forCCtx, cctx->entropyWorkspace);
         FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
         assert(dictID <= UINT_MAX);
         cctx->dictID = (U32)dictID;
@@ -4471,11 +5096,11 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
                                             &cctxParams, pledgedSrcSize);
 }
 
-size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+static size_t
+ZSTD_compressBegin_usingDict_deprecated(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
 {
     ZSTD_CCtx_params cctxParams;
-    {
-        ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
+    {   ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
         ZSTD_CCtxParams_init_internal(&cctxParams, &params, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel);
     }
     DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
@@ -4483,9 +5108,15 @@ size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t di
                                        &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
 }
 
+size_t
+ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, dict, dictSize, compressionLevel);
+}
+
 size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
 {
-    return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, NULL, 0, compressionLevel);
 }
 
 
@@ -4496,14 +5127,13 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
 {
     BYTE* const ostart = (BYTE*)dst;
     BYTE* op = ostart;
-    size_t fhSize = 0;
 
     DEBUGLOG(4, "ZSTD_writeEpilogue");
     RETURN_ERROR_IF(cctx->stage == ZSTDcs_created, stage_wrong, "init missing");
 
     /* special case : empty frame */
     if (cctx->stage == ZSTDcs_init) {
-        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams, 0, 0);
+        size_t fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams, 0, 0);
         FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
         dstCapacity -= fhSize;
         op += fhSize;
@@ -4513,8 +5143,9 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
     if (cctx->stage != ZSTDcs_ending) {
         /* write one last empty block, make it the "last" block */
         U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
-        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for epilogue");
-        MEM_writeLE32(op, cBlockHeader24);
+        ZSTD_STATIC_ASSERT(ZSTD_BLOCKHEADERSIZE == 3);
+        RETURN_ERROR_IF(dstCapacity<3, dstSize_tooSmall, "no room for epilogue");
+        MEM_writeLE24(op, cBlockHeader24);
         op += ZSTD_blockHeaderSize;
         dstCapacity -= ZSTD_blockHeaderSize;
     }
@@ -4537,9 +5168,9 @@ void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize)
     (void)extraCSize;
 }
 
-size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
-                         void* dst, size_t dstCapacity,
-                   const void* src, size_t srcSize)
+size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize)
 {
     size_t endResult;
     size_t const cSize = ZSTD_compressContinue_internal(cctx,
@@ -4563,6 +5194,14 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
     return cSize + endResult;
 }
 
+/* NOTE: Must just wrap ZSTD_compressEnd_public() */
+size_t ZSTD_compressEnd(ZSTD_CCtx* cctx,
+                        void* dst, size_t dstCapacity,
+                  const void* src, size_t srcSize)
+{
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
 size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
                                void* dst, size_t dstCapacity,
                          const void* src, size_t srcSize,
@@ -4591,7 +5230,7 @@ size_t ZSTD_compress_advanced_internal(
     FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
                          dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
                          params, srcSize, ZSTDb_not_buffered) , "");
-    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
 }
 
 size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
@@ -4709,7 +5348,7 @@ static size_t ZSTD_initCDict_internal(
         {   size_t const dictID = ZSTD_compress_insertDictionary(
                     &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
                     &params, cdict->dictContent, cdict->dictContentSize,
-                    dictContentType, ZSTD_dtlm_full, cdict->entropyWorkspace);
+                    dictContentType, ZSTD_dtlm_full, ZSTD_tfp_forCDict, cdict->entropyWorkspace);
             FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
             assert(dictID <= (size_t)(U32)-1);
             cdict->dictID = (U32)dictID;
@@ -4811,7 +5450,7 @@ ZSTD_CDict* ZSTD_createCDict_advanced2(
                         cctxParams.useRowMatchFinder, cctxParams.enableDedicatedDictSearch,
                         customMem);
 
-    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+    if (!cdict || ZSTD_isError( ZSTD_initCDict_internal(cdict,
                                     dict, dictSize,
                                     dictLoadMethod, dictContentType,
                                     cctxParams) )) {
@@ -4906,6 +5545,7 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
     params.cParams = cParams;
     params.useRowMatchFinder = useRowMatchFinder;
     cdict->useRowMatchFinder = useRowMatchFinder;
+    cdict->compressionLevel = ZSTD_NO_CLEVEL;
 
     if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
                                               dict, dictSize,
@@ -4985,12 +5625,17 @@ size_t ZSTD_compressBegin_usingCDict_advanced(
 
 /* ZSTD_compressBegin_usingCDict() :
  * cdict must be != NULL */
-size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
 {
     ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
     return ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    return ZSTD_compressBegin_usingCDict_deprecated(cctx, cdict);
+}
+
 /*! ZSTD_compress_usingCDict_internal():
  * Implementation of various ZSTD_compress_usingCDict* functions.
  */
@@ -5000,7 +5645,7 @@ static size_t ZSTD_compress_usingCDict_internal(ZSTD_CCtx* cctx,
                                 const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
 {
     FORWARD_IF_ERROR(ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, srcSize), ""); /* will check if cdict != NULL */
-    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
 }
 
 /*! ZSTD_compress_usingCDict_advanced():
@@ -5197,30 +5842,41 @@ size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
 
 static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
 {
-    size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
-    if (hintInSize==0) hintInSize = cctx->blockSize;
-    return hintInSize;
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        return cctx->blockSize - cctx->stableIn_notConsumed;
+    }
+    assert(cctx->appliedParams.inBufferMode == ZSTD_bm_buffered);
+    {   size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
+        if (hintInSize==0) hintInSize = cctx->blockSize;
+        return hintInSize;
+    }
 }
 
 /* ZSTD_compressStream_generic():
  *  internal function for all *compressStream*() variants
- *  non-static, because can be called from zstdmt_compress.c
- * @return : hint size for next input */
+ * @return : hint size for next input to complete ongoing block */
 static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                           ZSTD_outBuffer* output,
                                           ZSTD_inBuffer* input,
                                           ZSTD_EndDirective const flushMode)
 {
-    const char* const istart = (const char*)input->src;
-    const char* const iend = input->size != 0 ? istart + input->size : istart;
-    const char* ip = input->pos != 0 ? istart + input->pos : istart;
-    char* const ostart = (char*)output->dst;
-    char* const oend = output->size != 0 ? ostart + output->size : ostart;
-    char* op = output->pos != 0 ? ostart + output->pos : ostart;
+    const char* const istart = (assert(input != NULL), (const char*)input->src);
+    const char* const iend = (istart != NULL) ? istart + input->size : istart;
+    const char* ip = (istart != NULL) ? istart + input->pos : istart;
+    char* const ostart = (assert(output != NULL), (char*)output->dst);
+    char* const oend = (ostart != NULL) ? ostart + output->size : ostart;
+    char* op = (ostart != NULL) ? ostart + output->pos : ostart;
     U32 someMoreWork = 1;
 
     /* check expectations */
-    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (unsigned)flushMode);
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%i, srcSize = %zu", (int)flushMode, input->size - input->pos);
+    assert(zcs != NULL);
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        assert(input->pos >= zcs->stableIn_notConsumed);
+        input->pos -= zcs->stableIn_notConsumed;
+        if (ip) ip -= zcs->stableIn_notConsumed;
+        zcs->stableIn_notConsumed = 0;
+    }
     if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
         assert(zcs->inBuff != NULL);
         assert(zcs->inBuffSize > 0);
@@ -5229,8 +5885,10 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
         assert(zcs->outBuff !=  NULL);
         assert(zcs->outBuffSize > 0);
     }
-    assert(output->pos <= output->size);
+    if (input->src == NULL) assert(input->size == 0);
     assert(input->pos <= input->size);
+    if (output->dst == NULL) assert(output->size == 0);
+    assert(output->pos <= output->size);
     assert((U32)flushMode <= (U32)ZSTD_e_end);
 
     while (someMoreWork) {
@@ -5245,7 +5903,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)  /* OR we are allowed to return dstSizeTooSmall */
               && (zcs->inBuffPos == 0) ) {
                 /* shortcut to compression pass directly into output buffer */
-                size_t const cSize = ZSTD_compressEnd(zcs,
+                size_t const cSize = ZSTD_compressEnd_public(zcs,
                                                 op, oend-op, ip, iend-ip);
                 DEBUGLOG(4, "ZSTD_compressEnd : cSize=%u", (unsigned)cSize);
                 FORWARD_IF_ERROR(cSize, "ZSTD_compressEnd failed");
@@ -5262,8 +5920,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                         zcs->inBuff + zcs->inBuffPos, toLoad,
                                         ip, iend-ip);
                 zcs->inBuffPos += loaded;
-                if (loaded != 0)
-                    ip += loaded;
+                if (ip) ip += loaded;
                 if ( (flushMode == ZSTD_e_continue)
                   && (zcs->inBuffPos < zcs->inBuffTarget) ) {
                     /* not enough input to fill full block : stop here */
@@ -5274,6 +5931,20 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     /* empty */
                     someMoreWork = 0; break;
                 }
+            } else {
+                assert(zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+                if ( (flushMode == ZSTD_e_continue)
+                  && ( (size_t)(iend - ip) < zcs->blockSize) ) {
+                    /* can't compress a full block : stop here */
+                    zcs->stableIn_notConsumed = (size_t)(iend - ip);
+                    ip = iend;  /* pretend to have consumed input */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (ip == iend) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
             }
             /* compress current block (note : this stage cannot be stopped in the middle) */
             DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
@@ -5281,9 +5952,8 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 void* cDst;
                 size_t cSize;
                 size_t oSize = oend-op;
-                size_t const iSize = inputBuffered
-                    ? zcs->inBuffPos - zcs->inToCompress
-                    : MIN((size_t)(iend - ip), zcs->blockSize);
+                size_t const iSize = inputBuffered ? zcs->inBuffPos - zcs->inToCompress
+                                                   : MIN((size_t)(iend - ip), zcs->blockSize);
                 if (oSize >= ZSTD_compressBound(iSize) || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)
                     cDst = op;   /* compress into output buffer, to skip flush stage */
                 else
@@ -5291,9 +5961,9 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 if (inputBuffered) {
                     unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
                     cSize = lastBlock ?
-                            ZSTD_compressEnd(zcs, cDst, oSize,
+                            ZSTD_compressEnd_public(zcs, cDst, oSize,
                                         zcs->inBuff + zcs->inToCompress, iSize) :
-                            ZSTD_compressContinue(zcs, cDst, oSize,
+                            ZSTD_compressContinue_public(zcs, cDst, oSize,
                                         zcs->inBuff + zcs->inToCompress, iSize);
                     FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
                     zcs->frameEnded = lastBlock;
@@ -5306,19 +5976,16 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     if (!lastBlock)
                         assert(zcs->inBuffTarget <= zcs->inBuffSize);
                     zcs->inToCompress = zcs->inBuffPos;
-                } else {
-                    unsigned const lastBlock = (ip + iSize == iend);
-                    assert(flushMode == ZSTD_e_end /* Already validated */);
+                } else { /* !inputBuffered, hence ZSTD_bm_stable */
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip + iSize == iend);
                     cSize = lastBlock ?
-                            ZSTD_compressEnd(zcs, cDst, oSize, ip, iSize) :
-                            ZSTD_compressContinue(zcs, cDst, oSize, ip, iSize);
+                            ZSTD_compressEnd_public(zcs, cDst, oSize, ip, iSize) :
+                            ZSTD_compressContinue_public(zcs, cDst, oSize, ip, iSize);
                     /* Consume the input prior to error checking to mirror buffered mode. */
-                    if (iSize > 0)
-                        ip += iSize;
+                    if (ip) ip += iSize;
                     FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
                     zcs->frameEnded = lastBlock;
-                    if (lastBlock)
-                        assert(ip == iend);
+                    if (lastBlock) assert(ip == iend);
                 }
                 if (cDst == op) {  /* no need to flush */
                     op += cSize;
@@ -5388,8 +6055,10 @@ size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuf
 /* After a compression call set the expected input/output buffer.
  * This is validated at the start of the next compression call.
  */
-static void ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, ZSTD_outBuffer const* output, ZSTD_inBuffer const* input)
+static void
+ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, const ZSTD_outBuffer* output, const ZSTD_inBuffer* input)
 {
+    DEBUGLOG(5, "ZSTD_setBufferExpectations (for advanced stable in/out modes)");
     if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
         cctx->expectedInBuffer = *input;
     }
@@ -5408,22 +6077,22 @@ static size_t ZSTD_checkBufferStability(ZSTD_CCtx const* cctx,
 {
     if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
         ZSTD_inBuffer const expect = cctx->expectedInBuffer;
-        if (expect.src != input->src || expect.pos != input->pos || expect.size != input->size)
-            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer enabled but input differs!");
-        if (endOp != ZSTD_e_end)
-            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer can only be used with ZSTD_e_end!");
+        if (expect.src != input->src || expect.pos != input->pos)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableInBuffer enabled but input differs!");
     }
+    (void)endOp;
     if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
         size_t const outBufferSize = output->size - output->pos;
         if (cctx->expectedOutBufferSize != outBufferSize)
-            RETURN_ERROR(dstBuffer_wrong, "ZSTD_c_stableOutBuffer enabled but output size differs!");
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableOutBuffer enabled but output size differs!");
     }
     return 0;
 }
 
 static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
                                              ZSTD_EndDirective endOp,
-                                             size_t inSize) {
+                                             size_t inSize)
+{
     ZSTD_CCtx_params params = cctx->requestedParams;
     ZSTD_prefixDict const prefixDict = cctx->prefixDict;
     FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
@@ -5437,9 +6106,9 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
         params.compressionLevel = cctx->cdict->compressionLevel;
     }
     DEBUGLOG(4, "ZSTD_compressStream2 : transparent init stage");
-    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-fix pledgedSrcSize */
-    {
-        size_t const dictSize = prefixDict.dict
+    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-determine pledgedSrcSize */
+
+    {   size_t const dictSize = prefixDict.dict
                 ? prefixDict.dictSize
                 : (cctx->cdict ? cctx->cdict->dictContentSize : 0);
         ZSTD_cParamMode_e const mode = ZSTD_getCParamMode(cctx->cdict, &params, cctx->pledgedSrcSizePlusOne - 1);
@@ -5451,6 +6120,9 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
     params.useBlockSplitter = ZSTD_resolveBlockSplitterMode(params.useBlockSplitter, &params.cParams);
     params.ldmParams.enableLdm = ZSTD_resolveEnableLdm(params.ldmParams.enableLdm, &params.cParams);
     params.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params.useRowMatchFinder, &params.cParams);
+    params.validateSequences = ZSTD_resolveExternalSequenceValidation(params.validateSequences);
+    params.maxBlockSize = ZSTD_resolveMaxBlockSize(params.maxBlockSize);
+    params.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(params.searchForExternalRepcodes, params.compressionLevel);
 
     {   U64 const pledgedSrcSize = cctx->pledgedSrcSizePlusOne - 1;
         assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
@@ -5477,6 +6149,8 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
     return 0;
 }
 
+/* @return provides a minimum amount of data remaining to be flushed from internal buffers
+ */
 size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
                              ZSTD_outBuffer* output,
                              ZSTD_inBuffer* input,
@@ -5491,8 +6165,27 @@ size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
 
     /* transparent initialization stage */
     if (cctx->streamStage == zcss_init) {
-        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, input->size), "CompressStream2 initialization failed");
-        ZSTD_setBufferExpectations(cctx, output, input);    /* Set initial buffer expectations now that we've initialized */
+        size_t const inputSize = input->size - input->pos;  /* no obligation to start from pos==0 */
+        size_t const totalInputSize = inputSize + cctx->stableIn_notConsumed;
+        if ( (cctx->requestedParams.inBufferMode == ZSTD_bm_stable) /* input is presumed stable, across invocations */
+          && (endOp == ZSTD_e_continue)                             /* no flush requested, more input to come */
+          && (totalInputSize < ZSTD_BLOCKSIZE_MAX) ) {              /* not even reached one block yet */
+            if (cctx->stableIn_notConsumed) {  /* not the first time */
+                /* check stable source guarantees */
+                RETURN_ERROR_IF(input->src != cctx->expectedInBuffer.src, stabilityCondition_notRespected, "stableInBuffer condition not respected: wrong src pointer");
+                RETURN_ERROR_IF(input->pos != cctx->expectedInBuffer.size, stabilityCondition_notRespected, "stableInBuffer condition not respected: externally modified pos");
+            }
+            /* pretend input was consumed, to give a sense forward progress */
+            input->pos = input->size;
+            /* save stable inBuffer, for later control, and flush/end */
+            cctx->expectedInBuffer = *input;
+            /* but actually input wasn't consumed, so keep track of position from where compression shall resume */
+            cctx->stableIn_notConsumed += inputSize;
+            /* don't initialize yet, wait for the first block of flush() order, for better parameters adaptation */
+            return ZSTD_FRAMEHEADERSIZE_MIN(cctx->requestedParams.format);  /* at least some header to produce */
+        }
+        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, totalInputSize), "compressStream2 initialization failed");
+        ZSTD_setBufferExpectations(cctx, output, input);   /* Set initial buffer expectations now that we've initialized */
     }
     /* end of transparent initialization stage */
 
@@ -5510,13 +6203,20 @@ size_t ZSTD_compressStream2_simpleArgs (
                       const void* src, size_t srcSize, size_t* srcPos,
                             ZSTD_EndDirective endOp)
 {
-    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
-    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
     /* ZSTD_compressStream2() will check validity of dstPos and srcPos */
-    size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
-    *dstPos = output.pos;
-    *srcPos = input.pos;
-    return cErr;
+    {   size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
 }
 
 size_t ZSTD_compress2(ZSTD_CCtx* cctx,
@@ -5539,6 +6239,7 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
         /* Reset to the original values. */
         cctx->requestedParams.inBufferMode = originalInBufferMode;
         cctx->requestedParams.outBufferMode = originalOutBufferMode;
+
         FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
         if (result != 0) {  /* compression not completed, due to lack of output space */
             assert(oPos == dstCapacity);
@@ -5549,64 +6250,61 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
     }
 }
 
-typedef struct {
-    U32 idx;             /* Index in array of ZSTD_Sequence */
-    U32 posInSequence;   /* Position within sequence at idx */
-    size_t posInSrc;        /* Number of bytes given by sequences provided so far */
-} ZSTD_sequencePosition;
-
 /* ZSTD_validateSequence() :
  * @offCode : is presumed to follow format required by ZSTD_storeSeq()
  * @returns a ZSTD error code if sequence is not valid
  */
 static size_t
-ZSTD_validateSequence(U32 offCode, U32 matchLength,
-                      size_t posInSrc, U32 windowLog, size_t dictSize)
+ZSTD_validateSequence(U32 offCode, U32 matchLength, U32 minMatch,
+                      size_t posInSrc, U32 windowLog, size_t dictSize, int useSequenceProducer)
 {
-    U32 const windowSize = 1 << windowLog;
+    U32 const windowSize = 1u << windowLog;
     /* posInSrc represents the amount of data the decoder would decode up to this point.
      * As long as the amount of data decoded is less than or equal to window size, offsets may be
      * larger than the total length of output decoded in order to reference the dict, even larger than
      * window size. After output surpasses windowSize, we're limited to windowSize offsets again.
      */
     size_t const offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
-    RETURN_ERROR_IF(offCode > STORE_OFFSET(offsetBound), corruption_detected, "Offset too large!");
-    RETURN_ERROR_IF(matchLength < MINMATCH, corruption_detected, "Matchlength too small");
+    size_t const matchLenLowerBound = (minMatch == 3 || useSequenceProducer) ? 3 : 4;
+    RETURN_ERROR_IF(offCode > OFFSET_TO_OFFBASE(offsetBound), externalSequences_invalid, "Offset too large!");
+    /* Validate maxNbSeq is large enough for the given matchLength and minMatch */
+    RETURN_ERROR_IF(matchLength < matchLenLowerBound, externalSequences_invalid, "Matchlength too small for the minMatch");
     return 0;
 }
 
 /* Returns an offset code, given a sequence's raw offset, the ongoing repcode array, and whether litLength == 0 */
-static U32 ZSTD_finalizeOffCode(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0)
+static U32 ZSTD_finalizeOffBase(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0)
 {
-    U32 offCode = STORE_OFFSET(rawOffset);
+    U32 offBase = OFFSET_TO_OFFBASE(rawOffset);
 
     if (!ll0 && rawOffset == rep[0]) {
-        offCode = STORE_REPCODE_1;
+        offBase = REPCODE1_TO_OFFBASE;
     } else if (rawOffset == rep[1]) {
-        offCode = STORE_REPCODE(2 - ll0);
+        offBase = REPCODE_TO_OFFBASE(2 - ll0);
     } else if (rawOffset == rep[2]) {
-        offCode = STORE_REPCODE(3 - ll0);
+        offBase = REPCODE_TO_OFFBASE(3 - ll0);
     } else if (ll0 && rawOffset == rep[0] - 1) {
-        offCode = STORE_REPCODE_3;
+        offBase = REPCODE3_TO_OFFBASE;
     }
-    return offCode;
+    return offBase;
 }
 
-/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
- * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
- */
-static size_t
+size_t
 ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
                                               ZSTD_sequencePosition* seqPos,
                                         const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                        const void* src, size_t blockSize)
+                                        const void* src, size_t blockSize,
+                                        ZSTD_paramSwitch_e externalRepSearch)
 {
     U32 idx = seqPos->idx;
+    U32 const startIdx = idx;
     BYTE const* ip = (BYTE const*)(src);
     const BYTE* const iend = ip + blockSize;
     repcodes_t updatedRepcodes;
     U32 dictSize;
 
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreExplicitBlockDelim (blockSize = %zu)", blockSize);
+
     if (cctx->cdict) {
         dictSize = (U32)cctx->cdict->dictContentSize;
     } else if (cctx->prefixDict.dict) {
@@ -5615,25 +6313,55 @@ ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
         dictSize = 0;
     }
     ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
-    for (; (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0) && idx < inSeqsSize; ++idx) {
+    for (; idx < inSeqsSize && (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0); ++idx) {
         U32 const litLength = inSeqs[idx].litLength;
-        U32 const ll0 = (litLength == 0);
         U32 const matchLength = inSeqs[idx].matchLength;
-        U32 const offCode = ZSTD_finalizeOffCode(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
-        ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
+        U32 offBase;
+
+        if (externalRepSearch == ZSTD_ps_disable) {
+            offBase = OFFSET_TO_OFFBASE(inSeqs[idx].offset);
+        } else {
+            U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
 
-        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
         if (cctx->appliedParams.validateSequences) {
             seqPos->posInSrc += litLength + matchLength;
-            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
-                                                cctx->appliedParams.cParams.windowLog, dictSize),
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                cctx->appliedParams.cParams.windowLog, dictSize, ZSTD_hasExtSeqProd(&cctx->appliedParams)),
                                                 "Sequence validation failed");
         }
-        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
                         "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
-        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength);
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
         ip += matchLength + litLength;
     }
+
+    /* If we skipped repcode search while parsing, we need to update repcodes now */
+    assert(externalRepSearch != ZSTD_ps_auto);
+    assert(idx >= startIdx);
+    if (externalRepSearch == ZSTD_ps_disable && idx != startIdx) {
+        U32* const rep = updatedRepcodes.rep;
+        U32 lastSeqIdx = idx - 1; /* index of last non-block-delimiter sequence */
+
+        if (lastSeqIdx >= startIdx + 2) {
+            rep[2] = inSeqs[lastSeqIdx - 2].offset;
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else if (lastSeqIdx == startIdx + 1) {
+            rep[2] = rep[0];
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else {
+            assert(lastSeqIdx == startIdx);
+            rep[2] = rep[1];
+            rep[1] = rep[0];
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        }
+    }
+
     ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
 
     if (inSeqs[idx].litLength) {
@@ -5642,26 +6370,15 @@ ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
         ip += inSeqs[idx].litLength;
         seqPos->posInSrc += inSeqs[idx].litLength;
     }
-    RETURN_ERROR_IF(ip != iend, corruption_detected, "Blocksize doesn't agree with block delimiter!");
+    RETURN_ERROR_IF(ip != iend, externalSequences_invalid, "Blocksize doesn't agree with block delimiter!");
     seqPos->idx = idx+1;
     return 0;
 }
 
-/* Returns the number of bytes to move the current read position back by. Only non-zero
- * if we ended up splitting a sequence. Otherwise, it may return a ZSTD error if something
- * went wrong.
- *
- * This function will attempt to scan through blockSize bytes represented by the sequences
- * in inSeqs, storing any (partial) sequences.
- *
- * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
- * avoid splitting a match, or to avoid splitting a match such that it would produce a match
- * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
- */
-static size_t
+size_t
 ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
                                    const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                   const void* src, size_t blockSize)
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch)
 {
     U32 idx = seqPos->idx;
     U32 startPosInSequence = seqPos->posInSequence;
@@ -5673,6 +6390,9 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
     U32 bytesAdjustment = 0;
     U32 finalMatchSplit = 0;
 
+    /* TODO(embg) support fast parsing mode in noBlockDelim mode */
+    (void)externalRepSearch;
+
     if (cctx->cdict) {
         dictSize = cctx->cdict->dictContentSize;
     } else if (cctx->prefixDict.dict) {
@@ -5680,7 +6400,7 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
     } else {
         dictSize = 0;
     }
-    DEBUGLOG(5, "ZSTD_copySequencesToSeqStore: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreNoBlockDelim: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
     DEBUGLOG(5, "Start seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
     ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
     while (endPosInSequence && idx < inSeqsSize && !finalMatchSplit) {
@@ -5688,7 +6408,7 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
         U32 litLength = currSeq.litLength;
         U32 matchLength = currSeq.matchLength;
         U32 const rawOffset = currSeq.offset;
-        U32 offCode;
+        U32 offBase;
 
         /* Modify the sequence depending on where endPosInSequence lies */
         if (endPosInSequence >= currSeq.litLength + currSeq.matchLength) {
@@ -5702,7 +6422,6 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
             /* Move to the next sequence */
             endPosInSequence -= currSeq.litLength + currSeq.matchLength;
             startPosInSequence = 0;
-            idx++;
         } else {
             /* This is the final (partial) sequence we're adding from inSeqs, and endPosInSequence
                does not reach the end of the match. So, we have to split the sequence */
@@ -5742,21 +6461,23 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
         }
         /* Check if this offset can be represented with a repcode */
         {   U32 const ll0 = (litLength == 0);
-            offCode = ZSTD_finalizeOffCode(rawOffset, updatedRepcodes.rep, ll0);
-            ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
+            offBase = ZSTD_finalizeOffBase(rawOffset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
         }
 
         if (cctx->appliedParams.validateSequences) {
             seqPos->posInSrc += litLength + matchLength;
-            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
-                                                   cctx->appliedParams.cParams.windowLog, dictSize),
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                   cctx->appliedParams.cParams.windowLog, dictSize, ZSTD_hasExtSeqProd(&cctx->appliedParams)),
                                                    "Sequence validation failed");
         }
-        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
-        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
                         "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
-        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength);
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
         ip += matchLength + litLength;
+        if (!finalMatchSplit)
+            idx++; /* Next Sequence */
     }
     DEBUGLOG(5, "Ending seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
     assert(idx == inSeqsSize || endPosInSequence <= inSeqs[idx].litLength + inSeqs[idx].matchLength);
@@ -5779,7 +6500,7 @@ ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition*
 
 typedef size_t (*ZSTD_sequenceCopier) (ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                       const void* src, size_t blockSize);
+                                       const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
 static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode)
 {
     ZSTD_sequenceCopier sequenceCopier = NULL;
@@ -5793,6 +6514,57 @@ static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode)
     return sequenceCopier;
 }
 
+/* Discover the size of next block by searching for the delimiter.
+ * Note that a block delimiter **must** exist in this mode,
+ * otherwise it's an input error.
+ * The block size retrieved will be later compared to ensure it remains within bounds */
+static size_t
+blockSize_explicitDelimiter(const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    int end = 0;
+    size_t blockSize = 0;
+    size_t spos = seqPos.idx;
+    DEBUGLOG(6, "blockSize_explicitDelimiter : seq %zu / %zu", spos, inSeqsSize);
+    assert(spos <= inSeqsSize);
+    while (spos < inSeqsSize) {
+        end = (inSeqs[spos].offset == 0);
+        blockSize += inSeqs[spos].litLength + inSeqs[spos].matchLength;
+        if (end) {
+            if (inSeqs[spos].matchLength != 0)
+                RETURN_ERROR(externalSequences_invalid, "delimiter format error : both matchlength and offset must be == 0");
+            break;
+        }
+        spos++;
+    }
+    if (!end)
+        RETURN_ERROR(externalSequences_invalid, "Reached end of sequences without finding a block delimiter");
+    return blockSize;
+}
+
+/* More a "target" block size */
+static size_t blockSize_noDelimiter(size_t blockSize, size_t remaining)
+{
+    int const lastBlock = (remaining <= blockSize);
+    return lastBlock ? remaining : blockSize;
+}
+
+static size_t determine_blockSize(ZSTD_sequenceFormat_e mode,
+                           size_t blockSize, size_t remaining,
+                     const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    DEBUGLOG(6, "determine_blockSize : remainingSize = %zu", remaining);
+    if (mode == ZSTD_sf_noBlockDelimiters)
+        return blockSize_noDelimiter(blockSize, remaining);
+    {   size_t const explicitBlockSize = blockSize_explicitDelimiter(inSeqs, inSeqsSize, seqPos);
+        FORWARD_IF_ERROR(explicitBlockSize, "Error while determining block size with explicit delimiters");
+        if (explicitBlockSize > blockSize)
+            RETURN_ERROR(externalSequences_invalid, "sequences incorrectly define a too large block");
+        if (explicitBlockSize > remaining)
+            RETURN_ERROR(externalSequences_invalid, "sequences define a frame longer than source");
+        return explicitBlockSize;
+    }
+}
+
 /* Compress, block-by-block, all of the sequences given.
  *
  * Returns the cumulative size of all compressed blocks (including their headers),
@@ -5805,9 +6577,6 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
                           const void* src, size_t srcSize)
 {
     size_t cSize = 0;
-    U32 lastBlock;
-    size_t blockSize;
-    size_t compressedSeqsSize;
     size_t remaining = srcSize;
     ZSTD_sequencePosition seqPos = {0, 0, 0};
 
@@ -5827,22 +6596,29 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
     }
 
     while (remaining) {
+        size_t compressedSeqsSize;
         size_t cBlockSize;
         size_t additionalByteAdjustment;
-        lastBlock = remaining <= cctx->blockSize;
-        blockSize = lastBlock ? (U32)remaining : (U32)cctx->blockSize;
+        size_t blockSize = determine_blockSize(cctx->appliedParams.blockDelimiters,
+                                        cctx->blockSize, remaining,
+                                        inSeqs, inSeqsSize, seqPos);
+        U32 const lastBlock = (blockSize == remaining);
+        FORWARD_IF_ERROR(blockSize, "Error while trying to determine block size");
+        assert(blockSize <= remaining);
         ZSTD_resetSeqStore(&cctx->seqStore);
-        DEBUGLOG(4, "Working on new block. Blocksize: %zu", blockSize);
+        DEBUGLOG(5, "Working on new block. Blocksize: %zu (total:%zu)", blockSize, (ip - (const BYTE*)src) + blockSize);
 
-        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize);
+        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize, cctx->appliedParams.searchForExternalRepcodes);
         FORWARD_IF_ERROR(additionalByteAdjustment, "Bad sequence copy");
         blockSize -= additionalByteAdjustment;
 
         /* If blocks are too small, emit as a nocompress block */
-        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
             cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
             FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
-            DEBUGLOG(4, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
+            DEBUGLOG(5, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
             cSize += cBlockSize;
             ip += blockSize;
             op += cBlockSize;
@@ -5851,6 +6627,7 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             continue;
         }
 
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "not enough dstCapacity to write a new compressed block");
         compressedSeqsSize = ZSTD_entropyCompressSeqStore(&cctx->seqStore,
                                 &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
                                 &cctx->appliedParams,
@@ -5859,11 +6636,11 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
                                 cctx->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
                                 cctx->bmi2);
         FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
-        DEBUGLOG(4, "Compressed sequences size: %zu", compressedSeqsSize);
+        DEBUGLOG(5, "Compressed sequences size: %zu", compressedSeqsSize);
 
         if (!cctx->isFirstBlock &&
             ZSTD_maybeRLE(&cctx->seqStore) &&
-            ZSTD_isRLE((BYTE const*)src, srcSize)) {
+            ZSTD_isRLE(ip, blockSize)) {
             /* We don't want to emit our first block as a RLE even if it qualifies because
             * doing so will cause the decoder (cli only) to throw a "should consume all input error."
             * This is only an issue for zstd <= v1.4.3
@@ -5874,12 +6651,12 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
         if (compressedSeqsSize == 0) {
             /* ZSTD_noCompressBlock writes the block header as well */
             cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
-            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
-            DEBUGLOG(4, "Writing out nocompress block, size: %zu", cBlockSize);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(5, "Writing out nocompress block, size: %zu", cBlockSize);
         } else if (compressedSeqsSize == 1) {
             cBlockSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, blockSize, lastBlock);
-            FORWARD_IF_ERROR(cBlockSize, "RLE compress block failed");
-            DEBUGLOG(4, "Writing out RLE block, size: %zu", cBlockSize);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_rleCompressBlock failed");
+            DEBUGLOG(5, "Writing out RLE block, size: %zu", cBlockSize);
         } else {
             U32 cBlockHeader;
             /* Error checking and repcodes update */
@@ -5891,11 +6668,10 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
             MEM_writeLE24(op, cBlockHeader);
             cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
-            DEBUGLOG(4, "Writing out compressed block, size: %zu", cBlockSize);
+            DEBUGLOG(5, "Writing out compressed block, size: %zu", cBlockSize);
         }
 
         cSize += cBlockSize;
-        DEBUGLOG(4, "cSize running total: %zu", cSize);
 
         if (lastBlock) {
             break;
@@ -5906,12 +6682,15 @@ ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             dstCapacity -= cBlockSize;
             cctx->isFirstBlock = 0;
         }
+        DEBUGLOG(5, "cSize running total: %zu (remaining dstCapacity=%zu)", cSize, dstCapacity);
     }
 
+    DEBUGLOG(4, "cSize final total: %zu", cSize);
     return cSize;
 }
 
-size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapacity,
+size_t ZSTD_compressSequences(ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
                               const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
                               const void* src, size_t srcSize)
 {
@@ -5921,7 +6700,7 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapaci
     size_t frameHeaderSize = 0;
 
     /* Transparent initialization stage, same as compressStream2() */
-    DEBUGLOG(3, "ZSTD_compressSequences()");
+    DEBUGLOG(4, "ZSTD_compressSequences (dstCapacity=%zu)", dstCapacity);
     assert(cctx != NULL);
     FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, srcSize), "CCtx initialization failed");
     /* Begin writing output, starting with frame header */
@@ -5949,26 +6728,34 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapaci
         cSize += 4;
     }
 
-    DEBUGLOG(3, "Final compressed size: %zu", cSize);
+    DEBUGLOG(4, "Final compressed size: %zu", cSize);
     return cSize;
 }
 
 /*======   Finalize   ======*/
 
+static ZSTD_inBuffer inBuffer_forEndFlush(const ZSTD_CStream* zcs)
+{
+    const ZSTD_inBuffer nullInput = { NULL, 0, 0 };
+    const int stableInput = (zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+    return stableInput ? zcs->expectedInBuffer : nullInput;
+}
+
 /*! ZSTD_flushStream() :
  * @return : amount of data remaining to flush */
 size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    ZSTD_inBuffer input = { NULL, 0, 0 };
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    input.size = input.pos; /* do not ingest more input during flush */
     return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_flush);
 }
 
 
 size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    ZSTD_inBuffer input = { NULL, 0, 0 };
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
     size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
-    FORWARD_IF_ERROR( remainingToFlush , "ZSTD_compressStream2 failed");
+    FORWARD_IF_ERROR(remainingToFlush , "ZSTD_compressStream2(,,ZSTD_e_end) failed");
     if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
     /* single thread mode : attempt to calculate remaining to flush more precisely */
     {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
@@ -6090,7 +6877,7 @@ static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel,
             cp.targetLength = (unsigned)(-clampedCompressionLevel);
         }
         /* refine parameters based on srcSize & dictSize */
-        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode);
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode, ZSTD_ps_auto);
     }
 }
 
@@ -6125,3 +6912,29 @@ ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeH
     if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
     return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
 }
+
+void ZSTD_registerSequenceProducer(
+    ZSTD_CCtx* zc,
+    void* extSeqProdState,
+    ZSTD_sequenceProducer_F extSeqProdFunc
+) {
+    assert(zc != NULL);
+    ZSTD_CCtxParams_registerSequenceProducer(
+        &zc->requestedParams, extSeqProdState, extSeqProdFunc
+    );
+}
+
+void ZSTD_CCtxParams_registerSequenceProducer(
+  ZSTD_CCtx_params* params,
+  void* extSeqProdState,
+  ZSTD_sequenceProducer_F extSeqProdFunc
+) {
+    assert(params != NULL);
+    if (extSeqProdFunc != NULL) {
+        params->extSeqProdFunc = extSeqProdFunc;
+        params->extSeqProdState = extSeqProdState;
+    } else {
+        params->extSeqProdFunc = NULL;
+        params->extSeqProdState = NULL;
+    }
+}
diff --git a/lib/zstd/compress/zstd_compress_internal.h b/lib/zstd/compress/zstd_compress_internal.h
index 71697a11ae30..53cb582a8d2b 100644
--- a/lib/zstd/compress/zstd_compress_internal.h
+++ b/lib/zstd/compress/zstd_compress_internal.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -20,6 +21,7 @@
 ***************************************/
 #include "../common/zstd_internal.h"
 #include "zstd_cwksp.h"
+#include "../common/bits.h" /* ZSTD_highbit32, ZSTD_NbCommonBytes */
 
 
 /*-*************************************
@@ -32,7 +34,7 @@
                                        It's not a big deal though : candidate will just be sorted again.
                                        Additionally, candidate position 1 will be lost.
                                        But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
-                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table reuse with a different strategy.
                                        This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
 
 
@@ -111,12 +113,13 @@ typedef struct {
 /* ZSTD_buildBlockEntropyStats() :
  *  Builds entropy for the block.
  *  @return : 0 on success or error code */
-size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
-                             const ZSTD_entropyCTables_t* prevEntropy,
-                                   ZSTD_entropyCTables_t* nextEntropy,
-                             const ZSTD_CCtx_params* cctxParams,
-                                   ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                   void* workspace, size_t wkspSize);
+size_t ZSTD_buildBlockEntropyStats(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                          void* workspace, size_t wkspSize);
 
 /* *******************************
 *  Compression internals structs *
@@ -142,26 +145,33 @@ typedef struct {
   size_t capacity;      /* The capacity starting from `seq` pointer */
 } rawSeqStore_t;
 
+typedef struct {
+    U32 idx;            /* Index in array of ZSTD_Sequence */
+    U32 posInSequence;  /* Position within sequence at idx */
+    size_t posInSrc;    /* Number of bytes given by sequences provided so far */
+} ZSTD_sequencePosition;
+
 UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
 
 typedef struct {
-    int price;
-    U32 off;
-    U32 mlen;
-    U32 litlen;
-    U32 rep[ZSTD_REP_NUM];
+    int price;  /* price from beginning of segment to this position */
+    U32 off;    /* offset of previous match */
+    U32 mlen;   /* length of previous match */
+    U32 litlen; /* nb of literals since previous match */
+    U32 rep[ZSTD_REP_NUM];  /* offset history after previous match */
 } ZSTD_optimal_t;
 
 typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
 
+#define ZSTD_OPT_SIZE (ZSTD_OPT_NUM+3)
 typedef struct {
     /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
     unsigned* litFreq;           /* table of literals statistics, of size 256 */
     unsigned* litLengthFreq;     /* table of litLength statistics, of size (MaxLL+1) */
     unsigned* matchLengthFreq;   /* table of matchLength statistics, of size (MaxML+1) */
     unsigned* offCodeFreq;       /* table of offCode statistics, of size (MaxOff+1) */
-    ZSTD_match_t* matchTable;    /* list of found matches, of size ZSTD_OPT_NUM+1 */
-    ZSTD_optimal_t* priceTable;  /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+    ZSTD_match_t* matchTable;    /* list of found matches, of size ZSTD_OPT_SIZE */
+    ZSTD_optimal_t* priceTable;  /* All positions tracked by optimal parser, of size ZSTD_OPT_SIZE */
 
     U32  litSum;                 /* nb of literals */
     U32  litLengthSum;           /* nb of litLength codes */
@@ -212,8 +222,10 @@ struct ZSTD_matchState_t {
     U32 hashLog3;           /* dispatch table for matches of len==3 : larger == faster, more memory */
 
     U32 rowHashLog;                          /* For row-based matchfinder: Hashlog based on nb of rows in the hashTable.*/
-    U16* tagTable;                           /* For row-based matchFinder: A row-based table containing the hashes and head index. */
+    BYTE* tagTable;                          /* For row-based matchFinder: A row-based table containing the hashes and head index. */
     U32 hashCache[ZSTD_ROW_HASH_CACHE_SIZE]; /* For row-based matchFinder: a cache of hashes to improve speed */
+    U64 hashSalt;                            /* For row-based matchFinder: salts the hash for reuse of tag table */
+    U32 hashSaltEntropy;                     /* For row-based matchFinder: collects entropy for salt generation */
 
     U32* hashTable;
     U32* hashTable3;
@@ -228,6 +240,18 @@ struct ZSTD_matchState_t {
     const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
     const rawSeqStore_t* ldmSeqStore;
+
+    /* Controls prefetching in some dictMatchState matchfinders.
+     * This behavior is controlled from the cctx ms.
+     * This parameter has no effect in the cdict ms. */
+    int prefetchCDictTables;
+
+    /* When == 0, lazy match finders insert every position.
+     * When != 0, lazy match finders only insert positions they search.
+     * This allows them to skip much faster over incompressible data,
+     * at a small cost to compression ratio.
+     */
+    int lazySkipping;
 };
 
 typedef struct {
@@ -324,6 +348,25 @@ struct ZSTD_CCtx_params_s {
 
     /* Internal use, for createCCtxParams() and freeCCtxParams() only */
     ZSTD_customMem customMem;
+
+    /* Controls prefetching in some dictMatchState matchfinders */
+    ZSTD_paramSwitch_e prefetchCDictTables;
+
+    /* Controls whether zstd will fall back to an internal matchfinder
+     * if the external matchfinder returns an error code. */
+    int enableMatchFinderFallback;
+
+    /* Parameters for the external sequence producer API.
+     * Users set these parameters through ZSTD_registerSequenceProducer().
+     * It is not possible to set these parameters individually through the public API. */
+    void* extSeqProdState;
+    ZSTD_sequenceProducer_F extSeqProdFunc;
+
+    /* Adjust the max block size*/
+    size_t maxBlockSize;
+
+    /* Controls repcode search in external sequence parsing */
+    ZSTD_paramSwitch_e searchForExternalRepcodes;
 };  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
 
 #define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
@@ -404,6 +447,7 @@ struct ZSTD_CCtx_s {
 
     /* Stable in/out buffer verification */
     ZSTD_inBuffer expectedInBuffer;
+    size_t stableIn_notConsumed; /* nb bytes within stable input buffer that are said to be consumed but are not */
     size_t expectedOutBufferSize;
 
     /* Dictionary */
@@ -417,9 +461,14 @@ struct ZSTD_CCtx_s {
 
     /* Workspace for block splitter */
     ZSTD_blockSplitCtx blockSplitCtx;
+
+    /* Buffer for output from external sequence producer */
+    ZSTD_Sequence* extSeqBuf;
+    size_t extSeqBufCapacity;
 };
 
 typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
+typedef enum { ZSTD_tfp_forCCtx, ZSTD_tfp_forCDict } ZSTD_tableFillPurpose_e;
 
 typedef enum {
     ZSTD_noDict = 0,
@@ -441,7 +490,7 @@ typedef enum {
                                  * In this mode we take both the source size and the dictionary size
                                  * into account when selecting and adjusting the parameters.
                                  */
-    ZSTD_cpm_unknown = 3,       /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
+    ZSTD_cpm_unknown = 3        /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
                                  * We don't know what these parameters are for. We default to the legacy
                                  * behavior of taking both the source size and the dict size into account
                                  * when selecting and adjusting parameters.
@@ -500,9 +549,11 @@ MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
 /* ZSTD_noCompressBlock() :
  * Writes uncompressed block to dst buffer from given src.
  * Returns the size of the block */
-MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
+MEM_STATIC size_t
+ZSTD_noCompressBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
 {
     U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
+    DEBUGLOG(5, "ZSTD_noCompressBlock (srcSize=%zu, dstCapacity=%zu)", srcSize, dstCapacity);
     RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
                     dstSize_tooSmall, "dst buf too small for uncompressed block");
     MEM_writeLE24(dst, cBlockHeader24);
@@ -510,7 +561,8 @@ MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const voi
     return ZSTD_blockHeaderSize + srcSize;
 }
 
-MEM_STATIC size_t ZSTD_rleCompressBlock (void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
+MEM_STATIC size_t
+ZSTD_rleCompressBlock(void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
 {
     BYTE* const op = (BYTE*)dst;
     U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
@@ -529,7 +581,7 @@ MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
 {
     U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
     ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
-    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
     return (srcSize >> minlog) + 2;
 }
 
@@ -565,29 +617,27 @@ ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE con
     while (ip < iend) *op++ = *ip++;
 }
 
-#define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
-#define STORE_REPCODE_1 STORE_REPCODE(1)
-#define STORE_REPCODE_2 STORE_REPCODE(2)
-#define STORE_REPCODE_3 STORE_REPCODE(3)
-#define STORE_REPCODE(r) (assert((r)>=1), assert((r)<=3), (r)-1)
-#define STORE_OFFSET(o)  (assert((o)>0), o + ZSTD_REP_MOVE)
-#define STORED_IS_OFFSET(o)  ((o) > ZSTD_REP_MOVE)
-#define STORED_IS_REPCODE(o) ((o) <= ZSTD_REP_MOVE)
-#define STORED_OFFSET(o)  (assert(STORED_IS_OFFSET(o)), (o)-ZSTD_REP_MOVE)
-#define STORED_REPCODE(o) (assert(STORED_IS_REPCODE(o)), (o)+1)  /* returns ID 1,2,3 */
-#define STORED_TO_OFFBASE(o) ((o)+1)
-#define OFFBASE_TO_STORED(o) ((o)-1)
+
+#define REPCODE1_TO_OFFBASE REPCODE_TO_OFFBASE(1)
+#define REPCODE2_TO_OFFBASE REPCODE_TO_OFFBASE(2)
+#define REPCODE3_TO_OFFBASE REPCODE_TO_OFFBASE(3)
+#define REPCODE_TO_OFFBASE(r) (assert((r)>=1), assert((r)<=ZSTD_REP_NUM), (r)) /* accepts IDs 1,2,3 */
+#define OFFSET_TO_OFFBASE(o)  (assert((o)>0), o + ZSTD_REP_NUM)
+#define OFFBASE_IS_OFFSET(o)  ((o) > ZSTD_REP_NUM)
+#define OFFBASE_IS_REPCODE(o) ( 1 <= (o) && (o) <= ZSTD_REP_NUM)
+#define OFFBASE_TO_OFFSET(o)  (assert(OFFBASE_IS_OFFSET(o)), (o) - ZSTD_REP_NUM)
+#define OFFBASE_TO_REPCODE(o) (assert(OFFBASE_IS_REPCODE(o)), (o))  /* returns ID 1,2,3 */
 
 /*! ZSTD_storeSeq() :
- *  Store a sequence (litlen, litPtr, offCode and matchLength) into seqStore_t.
- *  @offBase_minus1 : Users should use employ macros STORE_REPCODE_X and STORE_OFFSET().
+ *  Store a sequence (litlen, litPtr, offBase and matchLength) into seqStore_t.
+ *  @offBase : Users should employ macros REPCODE_TO_OFFBASE() and OFFSET_TO_OFFBASE().
  *  @matchLength : must be >= MINMATCH
- *  Allowed to overread literals up to litLimit.
+ *  Allowed to over-read literals up to litLimit.
 */
 HINT_INLINE UNUSED_ATTR void
 ZSTD_storeSeq(seqStore_t* seqStorePtr,
               size_t litLength, const BYTE* literals, const BYTE* litLimit,
-              U32 offBase_minus1,
+              U32 offBase,
               size_t matchLength)
 {
     BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
@@ -596,8 +646,8 @@ ZSTD_storeSeq(seqStore_t* seqStorePtr,
     static const BYTE* g_start = NULL;
     if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
     {   U32 const pos = (U32)((const BYTE*)literals - g_start);
-        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offCode%7u",
-               pos, (U32)litLength, (U32)matchLength, (U32)offBase_minus1);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offBase%7u",
+               pos, (U32)litLength, (U32)matchLength, (U32)offBase);
     }
 #endif
     assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
@@ -607,9 +657,9 @@ ZSTD_storeSeq(seqStore_t* seqStorePtr,
     assert(literals + litLength <= litLimit);
     if (litEnd <= litLimit_w) {
         /* Common case we can use wildcopy.
-	 * First copy 16 bytes, because literals are likely short.
-	 */
-        assert(WILDCOPY_OVERLENGTH >= 16);
+         * First copy 16 bytes, because literals are likely short.
+         */
+        ZSTD_STATIC_ASSERT(WILDCOPY_OVERLENGTH >= 16);
         ZSTD_copy16(seqStorePtr->lit, literals);
         if (litLength > 16) {
             ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
@@ -628,7 +678,7 @@ ZSTD_storeSeq(seqStore_t* seqStorePtr,
     seqStorePtr->sequences[0].litLength = (U16)litLength;
 
     /* match offset */
-    seqStorePtr->sequences[0].offBase = STORED_TO_OFFBASE(offBase_minus1);
+    seqStorePtr->sequences[0].offBase = offBase;
 
     /* match Length */
     assert(matchLength >= MINMATCH);
@@ -646,17 +696,17 @@ ZSTD_storeSeq(seqStore_t* seqStorePtr,
 
 /* ZSTD_updateRep() :
  * updates in-place @rep (array of repeat offsets)
- * @offBase_minus1 : sum-type, with same numeric representation as ZSTD_storeSeq()
+ * @offBase : sum-type, using numeric representation of ZSTD_storeSeq()
  */
 MEM_STATIC void
-ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase_minus1, U32 const ll0)
+ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
 {
-    if (STORED_IS_OFFSET(offBase_minus1)) {  /* full offset */
+    if (OFFBASE_IS_OFFSET(offBase)) {  /* full offset */
         rep[2] = rep[1];
         rep[1] = rep[0];
-        rep[0] = STORED_OFFSET(offBase_minus1);
+        rep[0] = OFFBASE_TO_OFFSET(offBase);
     } else {   /* repcode */
-        U32 const repCode = STORED_REPCODE(offBase_minus1) - 1 + ll0;
+        U32 const repCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;
         if (repCode > 0) {  /* note : if repCode==0, no change */
             U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
             rep[2] = (repCode >= 2) ? rep[1] : rep[2];
@@ -673,11 +723,11 @@ typedef struct repcodes_s {
 } repcodes_t;
 
 MEM_STATIC repcodes_t
-ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase_minus1, U32 const ll0)
+ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
 {
     repcodes_t newReps;
     ZSTD_memcpy(&newReps, rep, sizeof(newReps));
-    ZSTD_updateRep(newReps.rep, offBase_minus1, ll0);
+    ZSTD_updateRep(newReps.rep, offBase, ll0);
     return newReps;
 }
 
@@ -685,59 +735,6 @@ ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase_minus1, U32 const ll0
 /*-*************************************
 *  Match length counter
 ***************************************/
-static unsigned ZSTD_NbCommonBytes (size_t val)
-{
-    if (MEM_isLittleEndian()) {
-        if (MEM_64bits()) {
-#       if (__GNUC__ >= 4)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
-                                                     0, 3, 1, 3, 1, 4, 2, 7,
-                                                     0, 2, 3, 6, 1, 5, 3, 5,
-                                                     1, 3, 4, 4, 2, 5, 6, 7,
-                                                     7, 0, 1, 2, 3, 3, 4, 6,
-                                                     2, 6, 5, 5, 3, 4, 5, 6,
-                                                     7, 1, 2, 4, 6, 4, 4, 5,
-                                                     7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        } else { /* 32 bits */
-#       if (__GNUC__ >= 3)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
-                                                     3, 2, 2, 1, 3, 2, 0, 1,
-                                                     3, 3, 1, 2, 2, 2, 2, 0,
-                                                     3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-        }
-    } else {  /* Big Endian CPU */
-        if (MEM_64bits()) {
-#       if (__GNUC__ >= 4)
-            return (__builtin_clzll(val) >> 3);
-#       else
-            unsigned r;
-            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
-            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        } else { /* 32 bits */
-#       if (__GNUC__ >= 3)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-    }   }
-}
-
-
 MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
 {
     const BYTE* const pStart = pIn;
@@ -783,32 +780,43 @@ ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
  *  Hashes
  ***************************************/
 static const U32 prime3bytes = 506832829U;
-static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
-MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
+static U32    ZSTD_hash3(U32 u, U32 h, U32 s) { assert(h <= 32); return (((u << (32-24)) * prime3bytes) ^ s)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h, 0); } /* only in zstd_opt.h */
+MEM_STATIC size_t ZSTD_hash3PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash3(MEM_readLE32(ptr), h, s); }
 
 static const U32 prime4bytes = 2654435761U;
-static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
-static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+static U32    ZSTD_hash4(U32 u, U32 h, U32 s) { assert(h <= 32); return ((u * prime4bytes) ^ s) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_readLE32(ptr), h, 0); }
+static size_t ZSTD_hash4PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash4(MEM_readLE32(ptr), h, s); }
 
 static const U64 prime5bytes = 889523592379ULL;
-static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
-static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
+static size_t ZSTD_hash5(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-40)) * prime5bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash5PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash5(MEM_readLE64(p), h, s); }
 
 static const U64 prime6bytes = 227718039650203ULL;
-static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
-static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+static size_t ZSTD_hash6(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-48)) * prime6bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash6PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash6(MEM_readLE64(p), h, s); }
 
 static const U64 prime7bytes = 58295818150454627ULL;
-static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
-static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
+static size_t ZSTD_hash7(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-56)) * prime7bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash7PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash7(MEM_readLE64(p), h, s); }
 
 static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
-static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
-static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+static size_t ZSTD_hash8(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u) * prime8bytes)  ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash8PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash8(MEM_readLE64(p), h, s); }
+
 
 MEM_STATIC FORCE_INLINE_ATTR
 size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
 {
+    /* Although some of these hashes do support hBits up to 64, some do not.
+     * To be on the safe side, always avoid hBits > 32. */
+    assert(hBits <= 32);
+
     switch(mls)
     {
     default:
@@ -820,6 +828,24 @@ size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
     }
 }
 
+MEM_STATIC FORCE_INLINE_ATTR
+size_t ZSTD_hashPtrSalted(const void* p, U32 hBits, U32 mls, const U64 hashSalt) {
+    /* Although some of these hashes do support hBits up to 64, some do not.
+     * To be on the safe side, always avoid hBits > 32. */
+    assert(hBits <= 32);
+
+    switch(mls)
+    {
+        default:
+        case 4: return ZSTD_hash4PtrS(p, hBits, (U32)hashSalt);
+        case 5: return ZSTD_hash5PtrS(p, hBits, hashSalt);
+        case 6: return ZSTD_hash6PtrS(p, hBits, hashSalt);
+        case 7: return ZSTD_hash7PtrS(p, hBits, hashSalt);
+        case 8: return ZSTD_hash8PtrS(p, hBits, hashSalt);
+    }
+}
+
+
 /* ZSTD_ipow() :
  * Return base^exponent.
  */
@@ -1011,7 +1037,9 @@ MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
  * The least significant cycleLog bits of the indices must remain the same,
  * which may be 0. Every index up to maxDist in the past must be valid.
  */
-MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
+MEM_STATIC
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
                                            U32 maxDist, void const* src)
 {
     /* preemptive overflow correction:
@@ -1167,10 +1195,15 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
                     (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
         assert(blockEndIdx >= loadedDictEnd);
 
-        if (blockEndIdx > loadedDictEnd + maxDist) {
+        if (blockEndIdx > loadedDictEnd + maxDist || loadedDictEnd != window->dictLimit) {
             /* On reaching window size, dictionaries are invalidated.
              * For simplification, if window size is reached anywhere within next block,
              * the dictionary is invalidated for the full block.
+             *
+             * We also have to invalidate the dictionary if ZSTD_window_update() has detected
+             * non-contiguous segments, which means that loadedDictEnd != window->dictLimit.
+             * loadedDictEnd may be 0, if forceWindow is true, but in that case we never use
+             * dictMatchState, so setting it to NULL is not a problem.
              */
             DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
             *loadedDictEndPtr = 0;
@@ -1199,7 +1232,9 @@ MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
  * forget about the extDict. Handles overlap of the prefix and extDict.
  * Returns non-zero if the segment is contiguous.
  */
-MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
+MEM_STATIC
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_window_update(ZSTD_window_t* window,
                                   void const* src, size_t srcSize,
                                   int forceNonContiguous)
 {
@@ -1302,6 +1337,42 @@ MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
 
 #endif
 
+/* Short Cache */
+
+/* Normally, zstd matchfinders follow this flow:
+ *     1. Compute hash at ip
+ *     2. Load index from hashTable[hash]
+ *     3. Check if *ip == *(base + index)
+ * In dictionary compression, loading *(base + index) is often an L2 or even L3 miss.
+ *
+ * Short cache is an optimization which allows us to avoid step 3 most of the time
+ * when the data doesn't actually match. With short cache, the flow becomes:
+ *     1. Compute (hash, currentTag) at ip. currentTag is an 8-bit independent hash at ip.
+ *     2. Load (index, matchTag) from hashTable[hash]. See ZSTD_writeTaggedIndex to understand how this works.
+ *     3. Only if currentTag == matchTag, check *ip == *(base + index). Otherwise, continue.
+ *
+ * Currently, short cache is only implemented in CDict hashtables. Thus, its use is limited to
+ * dictMatchState matchfinders.
+ */
+#define ZSTD_SHORT_CACHE_TAG_BITS 8
+#define ZSTD_SHORT_CACHE_TAG_MASK ((1u << ZSTD_SHORT_CACHE_TAG_BITS) - 1)
+
+/* Helper function for ZSTD_fillHashTable and ZSTD_fillDoubleHashTable.
+ * Unpacks hashAndTag into (hash, tag), then packs (index, tag) into hashTable[hash]. */
+MEM_STATIC void ZSTD_writeTaggedIndex(U32* const hashTable, size_t hashAndTag, U32 index) {
+    size_t const hash = hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+    U32 const tag = (U32)(hashAndTag & ZSTD_SHORT_CACHE_TAG_MASK);
+    assert(index >> (32 - ZSTD_SHORT_CACHE_TAG_BITS) == 0);
+    hashTable[hash] = (index << ZSTD_SHORT_CACHE_TAG_BITS) | tag;
+}
+
+/* Helper function for short cache matchfinders.
+ * Unpacks tag1 and tag2 from lower bits of packedTag1 and packedTag2, then checks if the tags match. */
+MEM_STATIC int ZSTD_comparePackedTags(size_t packedTag1, size_t packedTag2) {
+    U32 const tag1 = packedTag1 & ZSTD_SHORT_CACHE_TAG_MASK;
+    U32 const tag2 = packedTag2 & ZSTD_SHORT_CACHE_TAG_MASK;
+    return tag1 == tag2;
+}
 
 
 /* ===============================================================
@@ -1381,11 +1452,10 @@ size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
  * This cannot be used when long range matching is enabled.
  * Zstd will use these sequences, and pass the literals to a secondary block
  * compressor.
- * @return : An error code on failure.
  * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
  * access and data corruption.
  */
-size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
+void ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
 
 /* ZSTD_cycleLog() :
  *  condition for correct operation : hashLog > 1 */
@@ -1396,4 +1466,55 @@ U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
  */
 void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize);
 
+/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
+ * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
+ * Note that the block delimiter must include the last literals of the block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
+                                              ZSTD_sequencePosition* seqPos,
+                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                        const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
+/* Returns the number of bytes to move the current read position back by.
+ * Only non-zero if we ended up splitting a sequence.
+ * Otherwise, it may return a ZSTD error if something went wrong.
+ *
+ * This function will attempt to scan through blockSize bytes
+ * represented by the sequences in @inSeqs,
+ * storing any (partial) sequences.
+ *
+ * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
+ * avoid splitting a match, or to avoid splitting a match such that it would produce a match
+ * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                   const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
+/* Returns 1 if an external sequence producer is registered, otherwise returns 0. */
+MEM_STATIC int ZSTD_hasExtSeqProd(const ZSTD_CCtx_params* params) {
+    return params->extSeqProdFunc != NULL;
+}
+
+/* ===============================================================
+ * Deprecated definitions that are still used internally to avoid
+ * deprecation warnings. These functions are exactly equivalent to
+ * their public variants, but avoid the deprecation warnings.
+ * =============================================================== */
+
+size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
+
+size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t srcSize);
+
+size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
 #endif /* ZSTD_COMPRESS_H */
diff --git a/lib/zstd/compress/zstd_compress_literals.c b/lib/zstd/compress/zstd_compress_literals.c
index 52b0a8059aba..3e9ea46a670a 100644
--- a/lib/zstd/compress/zstd_compress_literals.c
+++ b/lib/zstd/compress/zstd_compress_literals.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -13,11 +14,36 @@
  ***************************************/
 #include "zstd_compress_literals.h"
 
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+#if DEBUGLEVEL >= 2
+
+static size_t showHexa(const void* src, size_t srcSize)
+{
+    const BYTE* const ip = (const BYTE*)src;
+    size_t u;
+    for (u=0; u<srcSize; u++) {
+        RAWLOG(5, " %02X", ip[u]); (void)ip;
+    }
+    RAWLOG(5, " \n");
+    return srcSize;
+}
+
+#endif
+
+
+/* **************************************************************
+*  Literals compression - special cases
+****************************************************************/
 size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
     BYTE* const ostart = (BYTE*)dst;
     U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
 
+    DEBUGLOG(5, "ZSTD_noCompressLiterals: srcSize=%zu, dstCapacity=%zu", srcSize, dstCapacity);
+
     RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
 
     switch(flSize)
@@ -36,16 +62,30 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
     }
 
     ZSTD_memcpy(ostart + flSize, src, srcSize);
-    DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
+    DEBUGLOG(5, "Raw (uncompressed) literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
     return srcSize + flSize;
 }
 
+static int allBytesIdentical(const void* src, size_t srcSize)
+{
+    assert(srcSize >= 1);
+    assert(src != NULL);
+    {   const BYTE b = ((const BYTE*)src)[0];
+        size_t p;
+        for (p=1; p<srcSize; p++) {
+            if (((const BYTE*)src)[p] != b) return 0;
+        }
+        return 1;
+    }
+}
+
 size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
     BYTE* const ostart = (BYTE*)dst;
     U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
 
-    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
+    assert(dstCapacity >= 4); (void)dstCapacity;
+    assert(allBytesIdentical(src, srcSize));
 
     switch(flSize)
     {
@@ -63,28 +103,51 @@ size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void*
     }
 
     ostart[flSize] = *(const BYTE*)src;
-    DEBUGLOG(5, "RLE literals: %u -> %u", (U32)srcSize, (U32)flSize + 1);
+    DEBUGLOG(5, "RLE : Repeated Literal (%02X: %u times) -> %u bytes encoded", ((const BYTE*)src)[0], (U32)srcSize, (U32)flSize + 1);
     return flSize+1;
 }
 
-size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
-                              ZSTD_hufCTables_t* nextHuf,
-                              ZSTD_strategy strategy, int disableLiteralCompression,
-                              void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize,
-                              void* entropyWorkspace, size_t entropyWorkspaceSize,
-                        const int bmi2,
-                        unsigned suspectUncompressible)
+/* ZSTD_minLiteralsToCompress() :
+ * returns minimal amount of literals
+ * for literal compression to even be attempted.
+ * Minimum is made tighter as compression strategy increases.
+ */
+static size_t
+ZSTD_minLiteralsToCompress(ZSTD_strategy strategy, HUF_repeat huf_repeat)
+{
+    assert((int)strategy >= 0);
+    assert((int)strategy <= 9);
+    /* btultra2 : min 8 bytes;
+     * then 2x larger for each successive compression strategy
+     * max threshold 64 bytes */
+    {   int const shift = MIN(9-(int)strategy, 3);
+        size_t const mintc = (huf_repeat == HUF_repeat_valid) ? 6 : (size_t)8 << shift;
+        DEBUGLOG(7, "minLiteralsToCompress = %zu", mintc);
+        return mintc;
+    }
+}
+
+size_t ZSTD_compressLiterals (
+                  void* dst, size_t dstCapacity,
+            const void* src, size_t srcSize,
+                  void* entropyWorkspace, size_t entropyWorkspaceSize,
+            const ZSTD_hufCTables_t* prevHuf,
+                  ZSTD_hufCTables_t* nextHuf,
+                  ZSTD_strategy strategy,
+                  int disableLiteralCompression,
+                  int suspectUncompressible,
+                  int bmi2)
 {
-    size_t const minGain = ZSTD_minGain(srcSize, strategy);
     size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
     BYTE*  const ostart = (BYTE*)dst;
     U32 singleStream = srcSize < 256;
     symbolEncodingType_e hType = set_compressed;
     size_t cLitSize;
 
-    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i srcSize=%u)",
-                disableLiteralCompression, (U32)srcSize);
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i, srcSize=%u, dstCapacity=%zu)",
+                disableLiteralCompression, (U32)srcSize, dstCapacity);
+
+    DEBUGLOG(6, "Completed literals listing (%zu bytes)", showHexa(src, srcSize));
 
     /* Prepare nextEntropy assuming reusing the existing table */
     ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
@@ -92,40 +155,51 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     if (disableLiteralCompression)
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
-    /* small ? don't even attempt compression (speed opt) */
-#   define COMPRESS_LITERALS_SIZE_MIN 63
-    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
-        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
+    /* if too small, don't even attempt compression (speed opt) */
+    if (srcSize < ZSTD_minLiteralsToCompress(strategy, prevHuf->repeatMode))
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
     RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
     {   HUF_repeat repeat = prevHuf->repeatMode;
-        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        int const flags = 0
+            | (bmi2 ? HUF_flags_bmi2 : 0)
+            | (strategy < ZSTD_lazy && srcSize <= 1024 ? HUF_flags_preferRepeat : 0)
+            | (strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD ? HUF_flags_optimalDepth : 0)
+            | (suspectUncompressible ? HUF_flags_suspectUncompressible : 0);
+
+        typedef size_t (*huf_compress_f)(void*, size_t, const void*, size_t, unsigned, unsigned, void*, size_t, HUF_CElt*, HUF_repeat*, int);
+        huf_compress_f huf_compress;
         if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
-        cLitSize = singleStream ?
-            HUF_compress1X_repeat(
-                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
-                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
-                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2, suspectUncompressible) :
-            HUF_compress4X_repeat(
-                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
-                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
-                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2, suspectUncompressible);
+        huf_compress = singleStream ? HUF_compress1X_repeat : HUF_compress4X_repeat;
+        cLitSize = huf_compress(ostart+lhSize, dstCapacity-lhSize,
+                                src, srcSize,
+                                HUF_SYMBOLVALUE_MAX, LitHufLog,
+                                entropyWorkspace, entropyWorkspaceSize,
+                                (HUF_CElt*)nextHuf->CTable,
+                                &repeat, flags);
+        DEBUGLOG(5, "%zu literals compressed into %zu bytes (before header)", srcSize, cLitSize);
         if (repeat != HUF_repeat_none) {
             /* reused the existing table */
-            DEBUGLOG(5, "Reusing previous huffman table");
+            DEBUGLOG(5, "reusing statistics from previous huffman block");
             hType = set_repeat;
         }
     }
 
-    if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
-        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
+    {   size_t const minGain = ZSTD_minGain(srcSize, strategy);
+        if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }   }
     if (cLitSize==1) {
-        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
-    }
+        /* A return value of 1 signals that the alphabet consists of a single symbol.
+         * However, in some rare circumstances, it could be the compressed size (a single byte).
+         * For that outcome to have a chance to happen, it's necessary that `srcSize < 8`.
+         * (it's also necessary to not generate statistics).
+         * Therefore, in such a case, actively check that all bytes are identical. */
+        if ((srcSize >= 8) || allBytesIdentical(src, srcSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }   }
 
     if (hType == set_compressed) {
         /* using a newly constructed table */
@@ -136,16 +210,19 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     switch(lhSize)
     {
     case 3: /* 2 - 2 - 10 - 10 */
-        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+        if (!singleStream) assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
+        {   U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
             MEM_writeLE24(ostart, lhc);
             break;
         }
     case 4: /* 2 - 2 - 14 - 14 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
         {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
             MEM_writeLE32(ostart, lhc);
             break;
         }
     case 5: /* 2 - 2 - 18 - 18 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
         {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
             MEM_writeLE32(ostart, lhc);
             ostart[4] = (BYTE)(cLitSize >> 10);
diff --git a/lib/zstd/compress/zstd_compress_literals.h b/lib/zstd/compress/zstd_compress_literals.h
index 9775fb97cb70..a2a85d6b69e5 100644
--- a/lib/zstd/compress/zstd_compress_literals.h
+++ b/lib/zstd/compress/zstd_compress_literals.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -16,16 +17,24 @@
 
 size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
+/* ZSTD_compressRleLiteralsBlock() :
+ * Conditions :
+ * - All bytes in @src are identical
+ * - dstCapacity >= 4 */
 size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
-/* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
-size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
-                              ZSTD_hufCTables_t* nextHuf,
-                              ZSTD_strategy strategy, int disableLiteralCompression,
-                              void* dst, size_t dstCapacity,
+/* ZSTD_compressLiterals():
+ * @entropyWorkspace: must be aligned on 4-bytes boundaries
+ * @entropyWorkspaceSize : must be >= HUF_WORKSPACE_SIZE
+ * @suspectUncompressible: sampling checks, to potentially skip huffman coding
+ */
+size_t ZSTD_compressLiterals (void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize,
                               void* entropyWorkspace, size_t entropyWorkspaceSize,
-                        const int bmi2,
-                        unsigned suspectUncompressible);
+                        const ZSTD_hufCTables_t* prevHuf,
+                              ZSTD_hufCTables_t* nextHuf,
+                              ZSTD_strategy strategy, int disableLiteralCompression,
+                              int suspectUncompressible,
+                              int bmi2);
 
 #endif /* ZSTD_COMPRESS_LITERALS_H */
diff --git a/lib/zstd/compress/zstd_compress_sequences.c b/lib/zstd/compress/zstd_compress_sequences.c
index 21ddc1b37acf..5c028c78d889 100644
--- a/lib/zstd/compress/zstd_compress_sequences.c
+++ b/lib/zstd/compress/zstd_compress_sequences.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -58,7 +59,7 @@ static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
 {
     /* Heuristic: This should cover most blocks <= 16K and
      * start to fade out after 16K to about 32K depending on
-     * comprssibility.
+     * compressibility.
      */
     return nbSeq >= 2048;
 }
@@ -166,7 +167,7 @@ ZSTD_selectEncodingType(
     if (mostFrequent == nbSeq) {
         *repeatMode = FSE_repeat_none;
         if (isDefaultAllowed && nbSeq <= 2) {
-            /* Prefer set_basic over set_rle when there are 2 or less symbols,
+            /* Prefer set_basic over set_rle when there are 2 or fewer symbols,
              * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
              * If basic encoding isn't possible, always choose RLE.
              */
diff --git a/lib/zstd/compress/zstd_compress_sequences.h b/lib/zstd/compress/zstd_compress_sequences.h
index 7991364c2f71..7fe6f4ff5cf2 100644
--- a/lib/zstd/compress/zstd_compress_sequences.h
+++ b/lib/zstd/compress/zstd_compress_sequences.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/compress/zstd_compress_superblock.c b/lib/zstd/compress/zstd_compress_superblock.c
index 17d836cc84e8..41f6521b27cd 100644
--- a/lib/zstd/compress/zstd_compress_superblock.c
+++ b/lib/zstd/compress/zstd_compress_superblock.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -36,13 +37,14 @@
  *      If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
  *      and the following sub-blocks' literals sections will be Treeless_Literals_Block.
  *  @return : compressed size of literals section of a sub-block
- *            Or 0 if it unable to compress.
+ *            Or 0 if unable to compress.
  *            Or error code */
-static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
-                                    const ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                    const BYTE* literals, size_t litSize,
-                                    void* dst, size_t dstSize,
-                                    const int bmi2, int writeEntropy, int* entropyWritten)
+static size_t
+ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
+                              const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                              const BYTE* literals, size_t litSize,
+                              void* dst, size_t dstSize,
+                              const int bmi2, int writeEntropy, int* entropyWritten)
 {
     size_t const header = writeEntropy ? 200 : 0;
     size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
@@ -53,8 +55,6 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
     size_t cLitSize = 0;
 
-    (void)bmi2; /* TODO bmi2... */
-
     DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
 
     *entropyWritten = 0;
@@ -76,9 +76,9 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
         DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
     }
 
-    /* TODO bmi2 */
-    {   const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable)
-                                          : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable);
+    {   int const flags = bmi2 ? HUF_flags_bmi2 : 0;
+        const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags)
+                                          : HUF_compress4X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags);
         op += cSize;
         cLitSize += cSize;
         if (cSize == 0 || ERR_isError(cSize)) {
@@ -103,7 +103,7 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     switch(lhSize)
     {
     case 3: /* 2 - 2 - 10 - 10 */
-        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
+        {   U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
             MEM_writeLE24(ostart, lhc);
             break;
         }
@@ -123,26 +123,30 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     }
     *entropyWritten = 1;
     DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
-static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef* sequences, size_t nbSeq, size_t litSize, int lastSequence) {
-    const seqDef* const sstart = sequences;
-    const seqDef* const send = sequences + nbSeq;
-    const seqDef* sp = sstart;
+static size_t
+ZSTD_seqDecompressedSize(seqStore_t const* seqStore,
+                   const seqDef* sequences, size_t nbSeqs,
+                         size_t litSize, int lastSubBlock)
+{
     size_t matchLengthSum = 0;
     size_t litLengthSum = 0;
-    (void)(litLengthSum); /* suppress unused variable warning on some environments */
-    while (send-sp > 0) {
-        ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
+    size_t n;
+    for (n=0; n<nbSeqs; n++) {
+        const ZSTD_sequenceLength seqLen = ZSTD_getSequenceLength(seqStore, sequences+n);
         litLengthSum += seqLen.litLength;
         matchLengthSum += seqLen.matchLength;
-        sp++;
     }
-    assert(litLengthSum <= litSize);
-    if (!lastSequence) {
+    DEBUGLOG(5, "ZSTD_seqDecompressedSize: %u sequences from %p: %u literals + %u matchlength",
+                (unsigned)nbSeqs, (const void*)sequences,
+                (unsigned)litLengthSum, (unsigned)matchLengthSum);
+    if (!lastSubBlock)
         assert(litLengthSum == litSize);
-    }
+    else
+        assert(litLengthSum <= litSize);
+    (void)litLengthSum;
     return matchLengthSum + litSize;
 }
 
@@ -156,13 +160,14 @@ static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef*
  *  @return : compressed size of sequences section of a sub-block
  *            Or 0 if it is unable to compress
  *            Or error code. */
-static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
-                                              const ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                              const seqDef* sequences, size_t nbSeq,
-                                              const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
-                                              const ZSTD_CCtx_params* cctxParams,
-                                              void* dst, size_t dstCapacity,
-                                              const int bmi2, int writeEntropy, int* entropyWritten)
+static size_t
+ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
+                                const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                const seqDef* sequences, size_t nbSeq,
+                                const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                const ZSTD_CCtx_params* cctxParams,
+                                void* dst, size_t dstCapacity,
+                                const int bmi2, int writeEntropy, int* entropyWritten)
 {
     const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     BYTE* const ostart = (BYTE*)dst;
@@ -176,14 +181,14 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
     /* Sequences Header */
     RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
                     dstSize_tooSmall, "");
-    if (nbSeq < 0x7F)
+    if (nbSeq < 128)
         *op++ = (BYTE)nbSeq;
     else if (nbSeq < LONGNBSEQ)
         op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
     else
         op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
     if (nbSeq==0) {
-        return op - ostart;
+        return (size_t)(op - ostart);
     }
 
     /* seqHead : flags for FSE encoding type */
@@ -205,7 +210,7 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
     }
 
     {   size_t const bitstreamSize = ZSTD_encodeSequences(
-                                        op, oend - op,
+                                        op, (size_t)(oend - op),
                                         fseTables->matchlengthCTable, mlCode,
                                         fseTables->offcodeCTable, ofCode,
                                         fseTables->litlengthCTable, llCode,
@@ -249,7 +254,7 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
 #endif
 
     *entropyWritten = 1;
-    return op - ostart;
+    return (size_t)(op - ostart);
 }
 
 /* ZSTD_compressSubBlock() :
@@ -275,7 +280,8 @@ static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
                 litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
     {   size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
                                                         &entropyMetadata->hufMetadata, literals, litSize,
-                                                        op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
+                                                        op, (size_t)(oend-op),
+                                                        bmi2, writeLitEntropy, litEntropyWritten);
         FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
         if (cLitSize == 0) return 0;
         op += cLitSize;
@@ -285,18 +291,18 @@ static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
                                                   sequences, nbSeq,
                                                   llCode, mlCode, ofCode,
                                                   cctxParams,
-                                                  op, oend-op,
+                                                  op, (size_t)(oend-op),
                                                   bmi2, writeSeqEntropy, seqEntropyWritten);
         FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
         if (cSeqSize == 0) return 0;
         op += cSeqSize;
     }
     /* Write block header */
-    {   size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
+    {   size_t cSize = (size_t)(op-ostart) - ZSTD_blockHeaderSize;
         U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
         MEM_writeLE24(ostart, cBlockHeader24);
     }
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
@@ -385,7 +391,11 @@ static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
     return cSeqSizeEstimate + sequencesSectionHeaderSize;
 }
 
-static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
+typedef struct {
+    size_t estLitSize;
+    size_t estBlockSize;
+} EstimatedBlockSize;
+static EstimatedBlockSize ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
                                         const BYTE* ofCodeTable,
                                         const BYTE* llCodeTable,
                                         const BYTE* mlCodeTable,
@@ -393,15 +403,17 @@ static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
                                         const ZSTD_entropyCTables_t* entropy,
                                         const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
                                         void* workspace, size_t wkspSize,
-                                        int writeLitEntropy, int writeSeqEntropy) {
-    size_t cSizeEstimate = 0;
-    cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
-                                                         &entropy->huf, &entropyMetadata->hufMetadata,
-                                                         workspace, wkspSize, writeLitEntropy);
-    cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
+                                        int writeLitEntropy, int writeSeqEntropy)
+{
+    EstimatedBlockSize ebs;
+    ebs.estLitSize = ZSTD_estimateSubBlockSize_literal(literals, litSize,
+                                                        &entropy->huf, &entropyMetadata->hufMetadata,
+                                                        workspace, wkspSize, writeLitEntropy);
+    ebs.estBlockSize = ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
                                                          nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
                                                          workspace, wkspSize, writeSeqEntropy);
-    return cSizeEstimate + ZSTD_blockHeaderSize;
+    ebs.estBlockSize += ebs.estLitSize + ZSTD_blockHeaderSize;
+    return ebs;
 }
 
 static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
@@ -415,13 +427,56 @@ static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMe
     return 0;
 }
 
+static size_t countLiterals(seqStore_t const* seqStore, const seqDef* sp, size_t seqCount)
+{
+    size_t n, total = 0;
+    assert(sp != NULL);
+    for (n=0; n<seqCount; n++) {
+        total += ZSTD_getSequenceLength(seqStore, sp+n).litLength;
+    }
+    DEBUGLOG(6, "countLiterals for %zu sequences from %p => %zu bytes", seqCount, (const void*)sp, total);
+    return total;
+}
+
+#define BYTESCALE 256
+
+static size_t sizeBlockSequences(const seqDef* sp, size_t nbSeqs,
+                size_t targetBudget, size_t avgLitCost, size_t avgSeqCost,
+                int firstSubBlock)
+{
+    size_t n, budget = 0, inSize=0;
+    /* entropy headers */
+    size_t const headerSize = (size_t)firstSubBlock * 120 * BYTESCALE; /* generous estimate */
+    assert(firstSubBlock==0 || firstSubBlock==1);
+    budget += headerSize;
+
+    /* first sequence => at least one sequence*/
+    budget += sp[0].litLength * avgLitCost + avgSeqCost;
+    if (budget > targetBudget) return 1;
+    inSize = sp[0].litLength + (sp[0].mlBase+MINMATCH);
+
+    /* loop over sequences */
+    for (n=1; n<nbSeqs; n++) {
+        size_t currentCost = sp[n].litLength * avgLitCost + avgSeqCost;
+        budget += currentCost;
+        inSize += sp[n].litLength + (sp[n].mlBase+MINMATCH);
+        /* stop when sub-block budget is reached */
+        if ( (budget > targetBudget)
+            /* though continue to expand until the sub-block is deemed compressible */
+          && (budget < inSize * BYTESCALE) )
+            break;
+    }
+
+    return n;
+}
+
 /* ZSTD_compressSubBlock_multi() :
  *  Breaks super-block into multiple sub-blocks and compresses them.
- *  Entropy will be written to the first block.
- *  The following blocks will use repeat mode to compress.
- *  All sub-blocks are compressed blocks (no raw or rle blocks).
- *  @return : compressed size of the super block (which is multiple ZSTD blocks)
- *            Or 0 if it failed to compress. */
+ *  Entropy will be written into the first block.
+ *  The following blocks use repeat_mode to compress.
+ *  Sub-blocks are all compressed, except the last one when beneficial.
+ *  @return : compressed size of the super block (which features multiple ZSTD blocks)
+ *            or 0 if it failed to compress. */
 static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
                             const ZSTD_compressedBlockState_t* prevCBlock,
                             ZSTD_compressedBlockState_t* nextCBlock,
@@ -434,10 +489,12 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
 {
     const seqDef* const sstart = seqStorePtr->sequencesStart;
     const seqDef* const send = seqStorePtr->sequences;
-    const seqDef* sp = sstart;
+    const seqDef* sp = sstart; /* tracks progresses within seqStorePtr->sequences */
+    size_t const nbSeqs = (size_t)(send - sstart);
     const BYTE* const lstart = seqStorePtr->litStart;
     const BYTE* const lend = seqStorePtr->lit;
     const BYTE* lp = lstart;
+    size_t const nbLiterals = (size_t)(lend - lstart);
     BYTE const* ip = (BYTE const*)src;
     BYTE const* const iend = ip + srcSize;
     BYTE* const ostart = (BYTE*)dst;
@@ -446,112 +503,171 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
     const BYTE* llCodePtr = seqStorePtr->llCode;
     const BYTE* mlCodePtr = seqStorePtr->mlCode;
     const BYTE* ofCodePtr = seqStorePtr->ofCode;
-    size_t targetCBlockSize = cctxParams->targetCBlockSize;
-    size_t litSize, seqCount;
-    int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
+    size_t const minTarget = ZSTD_TARGETCBLOCKSIZE_MIN; /* enforce minimum size, to reduce undesirable side effects */
+    size_t const targetCBlockSize = MAX(minTarget, cctxParams->targetCBlockSize);
+    int writeLitEntropy = (entropyMetadata->hufMetadata.hType == set_compressed);
     int writeSeqEntropy = 1;
-    int lastSequence = 0;
-
-    DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
-                (unsigned)(lend-lp), (unsigned)(send-sstart));
-
-    litSize = 0;
-    seqCount = 0;
-    do {
-        size_t cBlockSizeEstimate = 0;
-        if (sstart == send) {
-            lastSequence = 1;
-        } else {
-            const seqDef* const sequence = sp + seqCount;
-            lastSequence = sequence == send - 1;
-            litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
-            seqCount++;
-        }
-        if (lastSequence) {
-            assert(lp <= lend);
-            assert(litSize <= (size_t)(lend - lp));
-            litSize = (size_t)(lend - lp);
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi (srcSize=%u, litSize=%u, nbSeq=%u)",
+               (unsigned)srcSize, (unsigned)(lend-lstart), (unsigned)(send-sstart));
+
+        /* let's start by a general estimation for the full block */
+    if (nbSeqs > 0) {
+        EstimatedBlockSize const ebs =
+                ZSTD_estimateSubBlockSize(lp, nbLiterals,
+                                        ofCodePtr, llCodePtr, mlCodePtr, nbSeqs,
+                                        &nextCBlock->entropy, entropyMetadata,
+                                        workspace, wkspSize,
+                                        writeLitEntropy, writeSeqEntropy);
+        /* quick estimation */
+        size_t const avgLitCost = nbLiterals ? (ebs.estLitSize * BYTESCALE) / nbLiterals : BYTESCALE;
+        size_t const avgSeqCost = ((ebs.estBlockSize - ebs.estLitSize) * BYTESCALE) / nbSeqs;
+        const size_t nbSubBlocks = MAX((ebs.estBlockSize + (targetCBlockSize/2)) / targetCBlockSize, 1);
+        size_t n, avgBlockBudget, blockBudgetSupp=0;
+        avgBlockBudget = (ebs.estBlockSize * BYTESCALE) / nbSubBlocks;
+        DEBUGLOG(5, "estimated fullblock size=%u bytes ; avgLitCost=%.2f ; avgSeqCost=%.2f ; targetCBlockSize=%u, nbSubBlocks=%u ; avgBlockBudget=%.0f bytes",
+                    (unsigned)ebs.estBlockSize, (double)avgLitCost/BYTESCALE, (double)avgSeqCost/BYTESCALE,
+                    (unsigned)targetCBlockSize, (unsigned)nbSubBlocks, (double)avgBlockBudget/BYTESCALE);
+        /* simplification: if estimates states that the full superblock doesn't compress, just bail out immediately
+         * this will result in the production of a single uncompressed block covering @srcSize.*/
+        if (ebs.estBlockSize > srcSize) return 0;
+
+        /* compress and write sub-blocks */
+        assert(nbSubBlocks>0);
+        for (n=0; n < nbSubBlocks-1; n++) {
+            /* determine nb of sequences for current sub-block + nbLiterals from next sequence */
+            size_t const seqCount = sizeBlockSequences(sp, (size_t)(send-sp),
+                                        avgBlockBudget + blockBudgetSupp, avgLitCost, avgSeqCost, n==0);
+            /* if reached last sequence : break to last sub-block (simplification) */
+            assert(seqCount <= (size_t)(send-sp));
+            if (sp + seqCount == send) break;
+            assert(seqCount > 0);
+            /* compress sub-block */
+            {   int litEntropyWritten = 0;
+                int seqEntropyWritten = 0;
+                size_t litSize = countLiterals(seqStorePtr, sp, seqCount);
+                const size_t decompressedSize =
+                        ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 0);
+                size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
+                                                sp, seqCount,
+                                                lp, litSize,
+                                                llCodePtr, mlCodePtr, ofCodePtr,
+                                                cctxParams,
+                                                op, (size_t)(oend-op),
+                                                bmi2, writeLitEntropy, writeSeqEntropy,
+                                                &litEntropyWritten, &seqEntropyWritten,
+                                                0);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
+
+                /* check compressibility, update state components */
+                if (cSize > 0 && cSize < decompressedSize) {
+                    DEBUGLOG(5, "Committed sub-block compressing %u bytes => %u bytes",
+                                (unsigned)decompressedSize, (unsigned)cSize);
+                    assert(ip + decompressedSize <= iend);
+                    ip += decompressedSize;
+                    lp += litSize;
+                    op += cSize;
+                    llCodePtr += seqCount;
+                    mlCodePtr += seqCount;
+                    ofCodePtr += seqCount;
+                    /* Entropy only needs to be written once */
+                    if (litEntropyWritten) {
+                        writeLitEntropy = 0;
+                    }
+                    if (seqEntropyWritten) {
+                        writeSeqEntropy = 0;
+                    }
+                    sp += seqCount;
+                    blockBudgetSupp = 0;
+            }   }
+            /* otherwise : do not compress yet, coalesce current sub-block with following one */
         }
-        /* I think there is an optimization opportunity here.
-         * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
-         * since it recalculates estimate from scratch.
-         * For example, it would recount literal distribution and symbol codes every time.
-         */
-        cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
-                                                       &nextCBlock->entropy, entropyMetadata,
-                                                       workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
-        if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
-            int litEntropyWritten = 0;
-            int seqEntropyWritten = 0;
-            const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
-            const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
-                                                       sp, seqCount,
-                                                       lp, litSize,
-                                                       llCodePtr, mlCodePtr, ofCodePtr,
-                                                       cctxParams,
-                                                       op, oend-op,
-                                                       bmi2, writeLitEntropy, writeSeqEntropy,
-                                                       &litEntropyWritten, &seqEntropyWritten,
-                                                       lastBlock && lastSequence);
-            FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
-            if (cSize > 0 && cSize < decompressedSize) {
-                DEBUGLOG(5, "Committed the sub-block");
-                assert(ip + decompressedSize <= iend);
-                ip += decompressedSize;
-                sp += seqCount;
-                lp += litSize;
-                op += cSize;
-                llCodePtr += seqCount;
-                mlCodePtr += seqCount;
-                ofCodePtr += seqCount;
-                litSize = 0;
-                seqCount = 0;
-                /* Entropy only needs to be written once */
-                if (litEntropyWritten) {
-                    writeLitEntropy = 0;
-                }
-                if (seqEntropyWritten) {
-                    writeSeqEntropy = 0;
-                }
+    } /* if (nbSeqs > 0) */
+
+    /* write last block */
+    DEBUGLOG(5, "Generate last sub-block: %u sequences remaining", (unsigned)(send - sp));
+    {   int litEntropyWritten = 0;
+        int seqEntropyWritten = 0;
+        size_t litSize = (size_t)(lend - lp);
+        size_t seqCount = (size_t)(send - sp);
+        const size_t decompressedSize =
+                ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 1);
+        size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
+                                            sp, seqCount,
+                                            lp, litSize,
+                                            llCodePtr, mlCodePtr, ofCodePtr,
+                                            cctxParams,
+                                            op, (size_t)(oend-op),
+                                            bmi2, writeLitEntropy, writeSeqEntropy,
+                                            &litEntropyWritten, &seqEntropyWritten,
+                                            lastBlock);
+        FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
+
+        /* update pointers, the nb of literals borrowed from next sequence must be preserved */
+        if (cSize > 0 && cSize < decompressedSize) {
+            DEBUGLOG(5, "Last sub-block compressed %u bytes => %u bytes",
+                        (unsigned)decompressedSize, (unsigned)cSize);
+            assert(ip + decompressedSize <= iend);
+            ip += decompressedSize;
+            lp += litSize;
+            op += cSize;
+            llCodePtr += seqCount;
+            mlCodePtr += seqCount;
+            ofCodePtr += seqCount;
+            /* Entropy only needs to be written once */
+            if (litEntropyWritten) {
+                writeLitEntropy = 0;
             }
+            if (seqEntropyWritten) {
+                writeSeqEntropy = 0;
+            }
+            sp += seqCount;
         }
-    } while (!lastSequence);
+    }
+
+
     if (writeLitEntropy) {
-        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
+        DEBUGLOG(5, "Literal entropy tables were never written");
         ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
     }
     if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
         /* If we haven't written our entropy tables, then we've violated our contract and
          * must emit an uncompressed block.
          */
-        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
+        DEBUGLOG(5, "Sequence entropy tables were never written => cancel, emit an uncompressed block");
         return 0;
     }
+
     if (ip < iend) {
-        size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
-        DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
+        /* some data left : last part of the block sent uncompressed */
+        size_t const rSize = (size_t)((iend - ip));
+        size_t const cSize = ZSTD_noCompressBlock(op, (size_t)(oend - op), ip, rSize, lastBlock);
+        DEBUGLOG(5, "Generate last uncompressed sub-block of %u bytes", (unsigned)(rSize));
         FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
         assert(cSize != 0);
         op += cSize;
         /* We have to regenerate the repcodes because we've skipped some sequences */
         if (sp < send) {
-            seqDef const* seq;
+            const seqDef* seq;
             repcodes_t rep;
             ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
             for (seq = sstart; seq < sp; ++seq) {
-                ZSTD_updateRep(rep.rep, seq->offBase - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
+                ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
             }
             ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
         }
     }
-    DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
-    return op-ostart;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed all subBlocks: total compressed size = %u",
+                (unsigned)(op-ostart));
+    return (size_t)(op-ostart);
 }
 
 size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
                                void* dst, size_t dstCapacity,
-                               void const* src, size_t srcSize,
-                               unsigned lastBlock) {
+                               const void* src, size_t srcSize,
+                               unsigned lastBlock)
+{
     ZSTD_entropyCTablesMetadata_t entropyMetadata;
 
     FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
diff --git a/lib/zstd/compress/zstd_compress_superblock.h b/lib/zstd/compress/zstd_compress_superblock.h
index 224ece79546e..826bbc9e029b 100644
--- a/lib/zstd/compress/zstd_compress_superblock.h
+++ b/lib/zstd/compress/zstd_compress_superblock.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/compress/zstd_cwksp.h b/lib/zstd/compress/zstd_cwksp.h
index 349fc923c355..86bc3c2c23c7 100644
--- a/lib/zstd/compress/zstd_cwksp.h
+++ b/lib/zstd/compress/zstd_cwksp.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,7 +15,9 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customFree */
 #include "../common/zstd_internal.h"
+#include "../common/portability_macros.h"
 
 
 /*-*************************************
@@ -41,8 +44,9 @@
 ***************************************/
 typedef enum {
     ZSTD_cwksp_alloc_objects,
-    ZSTD_cwksp_alloc_buffers,
-    ZSTD_cwksp_alloc_aligned
+    ZSTD_cwksp_alloc_aligned_init_once,
+    ZSTD_cwksp_alloc_aligned,
+    ZSTD_cwksp_alloc_buffers
 } ZSTD_cwksp_alloc_phase_e;
 
 /*
@@ -95,8 +99,8 @@ typedef enum {
  *
  * Workspace Layout:
  *
- * [                        ... workspace ...                         ]
- * [objects][tables ... ->] free space [<- ... aligned][<- ... buffers]
+ * [                        ... workspace ...                           ]
+ * [objects][tables ->] free space [<- buffers][<- aligned][<- init once]
  *
  * The various objects that live in the workspace are divided into the
  * following categories, and are allocated separately:
@@ -120,9 +124,18 @@ typedef enum {
  *   uint32_t arrays, all of whose values are between 0 and (nextSrc - base).
  *   Their sizes depend on the cparams. These tables are 64-byte aligned.
  *
- * - Aligned: these buffers are used for various purposes that require 4 byte
- *   alignment, but don't require any initialization before they're used. These
- *   buffers are each aligned to 64 bytes.
+ * - Init once: these buffers require to be initialized at least once before
+ *   use. They should be used when we want to skip memory initialization
+ *   while not triggering memory checkers (like Valgrind) when reading from
+ *   from this memory without writing to it first.
+ *   These buffers should be used carefully as they might contain data
+ *   from previous compressions.
+ *   Buffers are aligned to 64 bytes.
+ *
+ * - Aligned: these buffers don't require any initialization before they're
+ *   used. The user of the buffer should make sure they write into a buffer
+ *   location before reading from it.
+ *   Buffers are aligned to 64 bytes.
  *
  * - Buffers: these buffers are used for various purposes that don't require
  *   any alignment or initialization before they're used. This means they can
@@ -134,8 +147,9 @@ typedef enum {
  * correctly packed into the workspace buffer. That order is:
  *
  * 1. Objects
- * 2. Buffers
- * 3. Aligned/Tables
+ * 2. Init once / Tables
+ * 3. Aligned / Tables
+ * 4. Buffers / Tables
  *
  * Attempts to reserve objects of different types out of order will fail.
  */
@@ -147,6 +161,7 @@ typedef struct {
     void* tableEnd;
     void* tableValidEnd;
     void* allocStart;
+    void* initOnceStart;
 
     BYTE allocFailed;
     int workspaceOversizedDuration;
@@ -159,6 +174,7 @@ typedef struct {
 ***************************************/
 
 MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws);
+MEM_STATIC void*  ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws);
 
 MEM_STATIC void ZSTD_cwksp_assert_internal_consistency(ZSTD_cwksp* ws) {
     (void)ws;
@@ -168,6 +184,8 @@ MEM_STATIC void ZSTD_cwksp_assert_internal_consistency(ZSTD_cwksp* ws) {
     assert(ws->tableEnd <= ws->allocStart);
     assert(ws->tableValidEnd <= ws->allocStart);
     assert(ws->allocStart <= ws->workspaceEnd);
+    assert(ws->initOnceStart <= ZSTD_cwksp_initialAllocStart(ws));
+    assert(ws->workspace <= ws->initOnceStart);
 }
 
 /*
@@ -210,14 +228,10 @@ MEM_STATIC size_t ZSTD_cwksp_aligned_alloc_size(size_t size) {
  * for internal purposes (currently only alignment).
  */
 MEM_STATIC size_t ZSTD_cwksp_slack_space_required(void) {
-    /* For alignment, the wksp will always allocate an additional n_1=[1, 64] bytes
-     * to align the beginning of tables section, as well as another n_2=[0, 63] bytes
-     * to align the beginning of the aligned section.
-     *
-     * n_1 + n_2 == 64 bytes if the cwksp is freshly allocated, due to tables and
-     * aligneds being sized in multiples of 64 bytes.
+    /* For alignment, the wksp will always allocate an additional 2*ZSTD_CWKSP_ALIGNMENT_BYTES
+     * bytes to align the beginning of tables section and end of buffers;
      */
-    size_t const slackSpace = ZSTD_CWKSP_ALIGNMENT_BYTES;
+    size_t const slackSpace = ZSTD_CWKSP_ALIGNMENT_BYTES * 2;
     return slackSpace;
 }
 
@@ -230,10 +244,18 @@ MEM_STATIC size_t ZSTD_cwksp_bytes_to_align_ptr(void* ptr, const size_t alignByt
     size_t const alignBytesMask = alignBytes - 1;
     size_t const bytes = (alignBytes - ((size_t)ptr & (alignBytesMask))) & alignBytesMask;
     assert((alignBytes & alignBytesMask) == 0);
-    assert(bytes != ZSTD_CWKSP_ALIGNMENT_BYTES);
+    assert(bytes < alignBytes);
     return bytes;
 }
 
+/*
+ * Returns the initial value for allocStart which is used to determine the position from
+ * which we can allocate from the end of the workspace.
+ */
+MEM_STATIC void*  ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws) {
+    return (void*)((size_t)ws->workspaceEnd & ~(ZSTD_CWKSP_ALIGNMENT_BYTES-1));
+}
+
 /*
  * Internal function. Do not use directly.
  * Reserves the given number of bytes within the aligned/buffer segment of the wksp,
@@ -274,27 +296,16 @@ ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase
 {
     assert(phase >= ws->phase);
     if (phase > ws->phase) {
-        /* Going from allocating objects to allocating buffers */
-        if (ws->phase < ZSTD_cwksp_alloc_buffers &&
-                phase >= ZSTD_cwksp_alloc_buffers) {
+        /* Going from allocating objects to allocating initOnce / tables */
+        if (ws->phase < ZSTD_cwksp_alloc_aligned_init_once &&
+            phase >= ZSTD_cwksp_alloc_aligned_init_once) {
             ws->tableValidEnd = ws->objectEnd;
-        }
+            ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
 
-        /* Going from allocating buffers to allocating aligneds/tables */
-        if (ws->phase < ZSTD_cwksp_alloc_aligned &&
-                phase >= ZSTD_cwksp_alloc_aligned) {
-            {   /* Align the start of the "aligned" to 64 bytes. Use [1, 64] bytes. */
-                size_t const bytesToAlign =
-                    ZSTD_CWKSP_ALIGNMENT_BYTES - ZSTD_cwksp_bytes_to_align_ptr(ws->allocStart, ZSTD_CWKSP_ALIGNMENT_BYTES);
-                DEBUGLOG(5, "reserving aligned alignment addtl space: %zu", bytesToAlign);
-                ZSTD_STATIC_ASSERT((ZSTD_CWKSP_ALIGNMENT_BYTES & (ZSTD_CWKSP_ALIGNMENT_BYTES - 1)) == 0); /* power of 2 */
-                RETURN_ERROR_IF(!ZSTD_cwksp_reserve_internal_buffer_space(ws, bytesToAlign),
-                                memory_allocation, "aligned phase - alignment initial allocation failed!");
-            }
             {   /* Align the start of the tables to 64 bytes. Use [0, 63] bytes */
-                void* const alloc = ws->objectEnd;
+                void *const alloc = ws->objectEnd;
                 size_t const bytesToAlign = ZSTD_cwksp_bytes_to_align_ptr(alloc, ZSTD_CWKSP_ALIGNMENT_BYTES);
-                void* const objectEnd = (BYTE*)alloc + bytesToAlign;
+                void *const objectEnd = (BYTE *) alloc + bytesToAlign;
                 DEBUGLOG(5, "reserving table alignment addtl space: %zu", bytesToAlign);
                 RETURN_ERROR_IF(objectEnd > ws->workspaceEnd, memory_allocation,
                                 "table phase - alignment initial allocation failed!");
@@ -302,7 +313,9 @@ ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase
                 ws->tableEnd = objectEnd;  /* table area starts being empty */
                 if (ws->tableValidEnd < ws->tableEnd) {
                     ws->tableValidEnd = ws->tableEnd;
-        }   }   }
+                }
+            }
+        }
         ws->phase = phase;
         ZSTD_cwksp_assert_internal_consistency(ws);
     }
@@ -314,7 +327,7 @@ ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase
  */
 MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr)
 {
-    return (ptr != NULL) && (ws->workspace <= ptr) && (ptr <= ws->workspaceEnd);
+    return (ptr != NULL) && (ws->workspace <= ptr) && (ptr < ws->workspaceEnd);
 }
 
 /*
@@ -343,6 +356,33 @@ MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes)
     return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers);
 }
 
+/*
+ * Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
+ * This memory has been initialized at least once in the past.
+ * This doesn't mean it has been initialized this time, and it might contain data from previous
+ * operations.
+ * The main usage is for algorithms that might need read access into uninitialized memory.
+ * The algorithm must maintain safety under these conditions and must make sure it doesn't
+ * leak any of the past data (directly or in side channels).
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_aligned_init_once(ZSTD_cwksp* ws, size_t bytes)
+{
+    size_t const alignedBytes = ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES);
+    void* ptr = ZSTD_cwksp_reserve_internal(ws, alignedBytes, ZSTD_cwksp_alloc_aligned_init_once);
+    assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
+    if(ptr && ptr < ws->initOnceStart) {
+        /* We assume the memory following the current allocation is either:
+         * 1. Not usable as initOnce memory (end of workspace)
+         * 2. Another initOnce buffer that has been allocated before (and so was previously memset)
+         * 3. An ASAN redzone, in which case we don't want to write on it
+         * For these reasons it should be fine to not explicitly zero every byte up to ws->initOnceStart.
+         * Note that we assume here that MSAN and ASAN cannot run in the same time. */
+        ZSTD_memset(ptr, 0, MIN((size_t)((U8*)ws->initOnceStart - (U8*)ptr), alignedBytes));
+        ws->initOnceStart = ptr;
+    }
+    return ptr;
+}
+
 /*
  * Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
  */
@@ -356,18 +396,22 @@ MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes)
 
 /*
  * Aligned on 64 bytes. These buffers have the special property that
- * their values remain constrained, allowing us to re-use them without
+ * their values remain constrained, allowing us to reuse them without
  * memset()-ing them.
  */
 MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes)
 {
-    const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned;
+    const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned_init_once;
     void* alloc;
     void* end;
     void* top;
 
-    if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase))) {
-        return NULL;
+    /* We can only start allocating tables after we are done reserving space for objects at the
+     * start of the workspace */
+    if(ws->phase < phase) {
+        if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase))) {
+            return NULL;
+        }
     }
     alloc = ws->tableEnd;
     end = (BYTE *)alloc + bytes;
@@ -451,7 +495,7 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
     assert(ws->tableValidEnd >= ws->objectEnd);
     assert(ws->tableValidEnd <= ws->allocStart);
     if (ws->tableValidEnd < ws->tableEnd) {
-        ZSTD_memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
+        ZSTD_memset(ws->tableValidEnd, 0, (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd));
     }
     ZSTD_cwksp_mark_tables_clean(ws);
 }
@@ -478,14 +522,23 @@ MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
 
 
     ws->tableEnd = ws->objectEnd;
-    ws->allocStart = ws->workspaceEnd;
+    ws->allocStart = ZSTD_cwksp_initialAllocStart(ws);
     ws->allocFailed = 0;
-    if (ws->phase > ZSTD_cwksp_alloc_buffers) {
-        ws->phase = ZSTD_cwksp_alloc_buffers;
+    if (ws->phase > ZSTD_cwksp_alloc_aligned_init_once) {
+        ws->phase = ZSTD_cwksp_alloc_aligned_init_once;
     }
     ZSTD_cwksp_assert_internal_consistency(ws);
 }
 
+MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
+}
+
+MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
+         + (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
+}
+
 /*
  * The provided workspace takes ownership of the buffer [start, start+size).
  * Any existing values in the workspace are ignored (the previously managed
@@ -498,6 +551,7 @@ MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size, ZSTD_c
     ws->workspaceEnd = (BYTE*)start + size;
     ws->objectEnd = ws->workspace;
     ws->tableValidEnd = ws->objectEnd;
+    ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
     ws->phase = ZSTD_cwksp_alloc_objects;
     ws->isStatic = isStatic;
     ZSTD_cwksp_clear(ws);
@@ -529,15 +583,6 @@ MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) {
     ZSTD_memset(src, 0, sizeof(ZSTD_cwksp));
 }
 
-MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
-    return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
-}
-
-MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
-    return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
-         + (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
-}
-
 MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
     return ws->allocFailed;
 }
@@ -550,17 +595,11 @@ MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
  * Returns if the estimated space needed for a wksp is within an acceptable limit of the
  * actual amount of space used.
  */
-MEM_STATIC int ZSTD_cwksp_estimated_space_within_bounds(const ZSTD_cwksp* const ws,
-                                                        size_t const estimatedSpace, int resizedWorkspace) {
-    if (resizedWorkspace) {
-        /* Resized/newly allocated wksp should have exact bounds */
-        return ZSTD_cwksp_used(ws) == estimatedSpace;
-    } else {
-        /* Due to alignment, when reusing a workspace, we can actually consume 63 fewer or more bytes
-         * than estimatedSpace. See the comments in zstd_cwksp.h for details.
-         */
-        return (ZSTD_cwksp_used(ws) >= estimatedSpace - 63) && (ZSTD_cwksp_used(ws) <= estimatedSpace + 63);
-    }
+MEM_STATIC int ZSTD_cwksp_estimated_space_within_bounds(const ZSTD_cwksp *const ws, size_t const estimatedSpace) {
+    /* We have an alignment space between objects and tables between tables and buffers, so we can have up to twice
+     * the alignment bytes difference between estimation and actual usage */
+    return (estimatedSpace - ZSTD_cwksp_slack_space_required()) <= ZSTD_cwksp_used(ws) &&
+           ZSTD_cwksp_used(ws) <= estimatedSpace;
 }
 
 
diff --git a/lib/zstd/compress/zstd_double_fast.c b/lib/zstd/compress/zstd_double_fast.c
index 76933dea2624..5ff54f17d92f 100644
--- a/lib/zstd/compress/zstd_double_fast.c
+++ b/lib/zstd/compress/zstd_double_fast.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,8 +12,49 @@
 #include "zstd_compress_internal.h"
 #include "zstd_double_fast.h"
 
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
 
-void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillDoubleHashTableForCDict(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0) {
+                ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
+            }
+            if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
+                ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
+            }
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+    }   }
+}
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillDoubleHashTableForCCtx(ZSTD_matchState_t* ms,
                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -43,11 +85,24 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
             /* Only load extra positions for ZSTD_dtlm_full */
             if (dtlm == ZSTD_dtlm_fast)
                 break;
-    }   }
+        }   }
+}
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
+    }
 }
 
 
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_compressBlock_doubleFast_noDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize, U32 const mls /* template */)
@@ -67,7 +122,7 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
 
     size_t mLength;
     U32 offset;
@@ -100,8 +155,8 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
         U32 const current = (U32)(ip - base);
         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
         U32 const maxRep = current - windowLow;
-        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
     }
 
     /* Outer Loop: one iteration per match found and stored */
@@ -131,7 +186,7 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
             if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
                 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
                 ip++;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
                 goto _match_stored;
             }
 
@@ -175,9 +230,13 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
         } while (ip1 <= ilimit);
 
 _cleanup:
+        /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+         * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+        offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
         /* save reps for next block */
-        rep[0] = offset_1 ? offset_1 : offsetSaved;
-        rep[1] = offset_2 ? offset_2 : offsetSaved;
+        rep[0] = offset_1 ? offset_1 : offsetSaved1;
+        rep[1] = offset_2 ? offset_2 : offsetSaved2;
 
         /* Return the last literals size */
         return (size_t)(iend - anchor);
@@ -217,7 +276,7 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
             hashLong[hl1] = (U32)(ip1 - base);
         }
 
-        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
 _match_stored:
         /* match found */
@@ -243,7 +302,7 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
                 U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
                 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
                 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, rLength);
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
                 ip += rLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
@@ -254,6 +313,7 @@ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
 
 
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
@@ -275,7 +335,6 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
 
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
     const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
@@ -286,8 +345,8 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
     const BYTE* const dictStart    = dictBase + dictStartIndex;
     const BYTE* const dictEnd      = dms->window.nextSrc;
     const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
-    const U32 dictHBitsL           = dictCParams->hashLog;
-    const U32 dictHBitsS           = dictCParams->chainLog;
+    const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
     const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
 
     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
@@ -295,6 +354,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
     /* if a dictionary is attached, it must be within window range */
     assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
 
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashLong, hashTableBytes);
+        PREFETCH_AREA(dictHashSmall, chainTableBytes);
+    }
+
     /* init */
     ip += (dictAndPrefixLength == 0);
 
@@ -309,8 +375,12 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
         U32 offset;
         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
-        size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
-        size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
+        U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
+        int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
         U32 const curr = (U32)(ip-base);
         U32 const matchIndexL = hashLong[h2];
         U32 matchIndexS = hashSmall[h];
@@ -328,7 +398,7 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
             goto _match_stored;
         }
 
@@ -340,9 +410,9 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
                 while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
                 goto _match_found;
             }
-        } else {
+        } else if (dictTagsMatchL) {
             /* check dictMatchState long match */
-            U32 const dictMatchIndexL = dictHashLong[dictHL];
+            U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
             const BYTE* dictMatchL = dictBase + dictMatchIndexL;
             assert(dictMatchL < dictEnd);
 
@@ -358,9 +428,9 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
             if (MEM_read32(match) == MEM_read32(ip)) {
                 goto _search_next_long;
             }
-        } else {
+        } else if (dictTagsMatchS) {
             /* check dictMatchState short match */
-            U32 const dictMatchIndexS = dictHashSmall[dictHS];
+            U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
             match = dictBase + dictMatchIndexS;
             matchIndexS = dictMatchIndexS + dictIndexDelta;
 
@@ -375,10 +445,11 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
         continue;
 
 _search_next_long:
-
         {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-            size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
             U32 const matchIndexL3 = hashLong[hl3];
+            U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
             const BYTE* matchL3 = base + matchIndexL3;
             hashLong[hl3] = curr + 1;
 
@@ -391,9 +462,9 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
                     while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
                     goto _match_found;
                 }
-            } else {
+            } else if (dictTagsMatchL3) {
                 /* check dict long +1 match */
-                U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
+                U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
                 const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
                 assert(dictMatchL3 < dictEnd);
                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
@@ -419,7 +490,7 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
         offset_2 = offset_1;
         offset_1 = offset;
 
-        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
 _match_stored:
         /* match found */
@@ -448,7 +519,7 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
                     const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
                     ip += repLength2;
@@ -461,8 +532,8 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
     }   /* while (ip < ilimit) */
 
     /* save reps for next block */
-    rep[0] = offset_1 ? offset_1 : offsetSaved;
-    rep[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
@@ -527,7 +598,9 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState(
 }
 
 
-static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_doubleFast_extDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
         U32 const mls /* template */)
@@ -585,7 +658,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
         } else {
             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
@@ -596,7 +669,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
@@ -621,7 +694,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 }
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
             } else {
                 ip += ((ip-anchor) >> kSearchStrength) + 1;
@@ -653,7 +726,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                     U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
                     ip += repLength2;
@@ -694,3 +767,5 @@ size_t ZSTD_compressBlock_doubleFast_extDict(
         return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
     }
 }
+
+#endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
diff --git a/lib/zstd/compress/zstd_double_fast.h b/lib/zstd/compress/zstd_double_fast.h
index 6822bde65a1d..b7ddc714f13e 100644
--- a/lib/zstd/compress/zstd_double_fast.h
+++ b/lib/zstd/compress/zstd_double_fast.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,8 +16,12 @@
 #include "../common/mem.h"      /* U32 */
 #include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
 
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+
 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
-                              void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                              ZSTD_tableFillPurpose_e tfp);
+
 size_t ZSTD_compressBlock_doubleFast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
@@ -27,6 +32,14 @@ size_t ZSTD_compressBlock_doubleFast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST ZSTD_compressBlock_doubleFast
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE ZSTD_compressBlock_doubleFast_dictMatchState
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT ZSTD_compressBlock_doubleFast_extDict
+#else
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST NULL
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT NULL
+#endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
 
 
 #endif /* ZSTD_DOUBLE_FAST_H */
diff --git a/lib/zstd/compress/zstd_fast.c b/lib/zstd/compress/zstd_fast.c
index a752e6beab52..b7a63ba4ce56 100644
--- a/lib/zstd/compress/zstd_fast.c
+++ b/lib/zstd/compress/zstd_fast.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,8 +12,46 @@
 #include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
 #include "zstd_fast.h"
 
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
 
-void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+    /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_full);
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        {   size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
+            ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr);   }
+
+        if (dtlm == ZSTD_dtlm_fast) continue;
+        /* Only load extra positions for ZSTD_dtlm_full */
+        {   U32 p;
+            for (p = 1; p < fastHashFillStep; ++p) {
+                size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
+                if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {  /* not yet filled */
+                    ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
+                }   }   }   }
+}
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
                         const void* const end,
                         ZSTD_dictTableLoadMethod_e dtlm)
 {
@@ -25,6 +64,10 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
     const U32 fastHashFillStep = 3;
 
+    /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_fast);
+
     /* Always insert every fastHashFillStep position into the hash table.
      * Insert the other positions if their hash entry is empty.
      */
@@ -42,6 +85,18 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
     }   }   }   }
 }
 
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
 
 /*
  * If you squint hard enough (and ignore repcodes), the search operation at any
@@ -89,8 +144,9 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
  *
  * This is also the work we do at the beginning to enter the loop initially.
  */
-FORCE_INLINE_TEMPLATE size_t
-ZSTD_compressBlock_fast_noDict_generic(
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_fast_noDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
         U32 const mls, U32 const hasStep)
@@ -117,7 +173,7 @@ ZSTD_compressBlock_fast_noDict_generic(
 
     U32 rep_offset1 = rep[0];
     U32 rep_offset2 = rep[1];
-    U32 offsetSaved = 0;
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
 
     size_t hash0; /* hash for ip0 */
     size_t hash1; /* hash for ip1 */
@@ -141,8 +197,8 @@ ZSTD_compressBlock_fast_noDict_generic(
     {   U32 const curr = (U32)(ip0 - base);
         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
         U32 const maxRep = curr - windowLow;
-        if (rep_offset2 > maxRep) offsetSaved = rep_offset2, rep_offset2 = 0;
-        if (rep_offset1 > maxRep) offsetSaved = rep_offset1, rep_offset1 = 0;
+        if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
+        if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
     }
 
     /* start each op */
@@ -180,8 +236,14 @@ ZSTD_compressBlock_fast_noDict_generic(
             mLength = ip0[-1] == match0[-1];
             ip0 -= mLength;
             match0 -= mLength;
-            offcode = STORE_REPCODE_1;
+            offcode = REPCODE1_TO_OFFBASE;
             mLength += 4;
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 is before the
+             * repcode (ip2). */
+            hashTable[hash1] = (U32)(ip1 - base);
+
             goto _match;
         }
 
@@ -195,6 +257,12 @@ ZSTD_compressBlock_fast_noDict_generic(
         /* check match at ip[0] */
         if (MEM_read32(ip0) == mval) {
             /* found a match! */
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 == ip0 + 1, so
+             * we know we will resume searching after ip1 */
+            hashTable[hash1] = (U32)(ip1 - base);
+
             goto _offset;
         }
 
@@ -224,6 +292,21 @@ ZSTD_compressBlock_fast_noDict_generic(
         /* check match at ip[0] */
         if (MEM_read32(ip0) == mval) {
             /* found a match! */
+
+            /* first write next hash table entry; we've already calculated it */
+            if (step <= 4) {
+                /* We need to avoid writing an index into the hash table >= the
+                 * position at which we will pick up our searching after we've
+                 * taken this match.
+                 *
+                 * The minimum possible match has length 4, so the earliest ip0
+                 * can be after we take this match will be the current ip0 + 4.
+                 * ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
+                 * write this position.
+                 */
+                hashTable[hash1] = (U32)(ip1 - base);
+            }
+
             goto _offset;
         }
 
@@ -254,9 +337,24 @@ ZSTD_compressBlock_fast_noDict_generic(
      * However, it seems to be a meaningful performance hit to try to search
      * them. So let's not. */
 
+    /* When the repcodes are outside of the prefix, we set them to zero before the loop.
+     * When the offsets are still zero, we need to restore them after the block to have a correct
+     * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
+     * offsets were invalid. We need to figure out which offset to refill with.
+     *     - If both offsets are zero they are in the same order.
+     *     - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
+     *     - If only one is zero, we need to decide which offset to restore.
+     *         - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
+     *         - It is impossible for rep_offset2 to be non-zero.
+     *
+     * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
+     * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
+     */
+    offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;
+
     /* save reps for next block */
-    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved;
-    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved;
+    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
+    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
@@ -267,7 +365,7 @@ ZSTD_compressBlock_fast_noDict_generic(
     match0 = base + idx;
     rep_offset2 = rep_offset1;
     rep_offset1 = (U32)(ip0-match0);
-    offcode = STORE_OFFSET(rep_offset1);
+    offcode = OFFSET_TO_OFFBASE(rep_offset1);
     mLength = 4;
 
     /* Count the backwards match length. */
@@ -287,11 +385,6 @@ ZSTD_compressBlock_fast_noDict_generic(
     ip0 += mLength;
     anchor = ip0;
 
-    /* write next hash table entry */
-    if (ip1 < ip0) {
-        hashTable[hash1] = (U32)(ip1 - base);
-    }
-
     /* Fill table and check for immediate repcode. */
     if (ip0 <= ilimit) {
         /* Fill Table */
@@ -306,7 +399,7 @@ ZSTD_compressBlock_fast_noDict_generic(
                 { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
                 hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
                 ip0 += rLength;
-                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, STORE_REPCODE_1, rLength);
+                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
                 anchor = ip0;
                 continue;   /* faster when present (confirmed on gcc-8) ... (?) */
     }   }   }
@@ -369,6 +462,7 @@ size_t ZSTD_compressBlock_fast(
 }
 
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_compressBlock_fast_dictMatchState_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
@@ -380,14 +474,14 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
     const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
+    const BYTE* ip0 = istart;
+    const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
     const BYTE* anchor = istart;
     const U32   prefixStartIndex = ms->window.dictLimit;
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
 
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
     const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
@@ -397,13 +491,13 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     const BYTE* const dictStart    = dictBase + dictStartIndex;
     const BYTE* const dictEnd      = dms->window.nextSrc;
     const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
-    const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
-    const U32 dictHLog             = dictCParams->hashLog;
+    const U32 dictAndPrefixLength  = (U32)(istart - prefixStart + dictEnd - dictStart);
+    const U32 dictHBits            = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
 
     /* if a dictionary is still attached, it necessarily means that
      * it is within window size. So we just check it. */
     const U32 maxDistance = 1U << cParams->windowLog;
-    const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
     assert(endIndex - prefixStartIndex <= maxDistance);
     (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
 
@@ -413,106 +507,155 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
      * when translating a dict index into a local index */
     assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
 
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashTable, hashTableBytes);
+    }
+
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
-    ip += (dictAndPrefixLength == 0);
+    ip0 += (dictAndPrefixLength == 0);
     /* dictMatchState repCode checks don't currently handle repCode == 0
      * disabling. */
     assert(offset_1 <= dictAndPrefixLength);
     assert(offset_2 <= dictAndPrefixLength);
 
-    /* Main Search Loop */
-    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+    /* Outer search loop */
+    assert(stepSize >= 1);
+    while (ip1 <= ilimit) {   /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
         size_t mLength;
-        size_t const h = ZSTD_hashPtr(ip, hlog, mls);
-        U32 const curr = (U32)(ip-base);
-        U32 const matchIndex = hashTable[h];
-        const BYTE* match = base + matchIndex;
-        const U32 repIndex = curr + 1 - offset_1;
-        const BYTE* repMatch = (repIndex < prefixStartIndex) ?
-                               dictBase + (repIndex - dictIndexDelta) :
-                               base + repIndex;
-        hashTable[h] = curr;   /* update hash table */
-
-        if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
-          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
-        } else if ( (matchIndex <= prefixStartIndex) ) {
-            size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
-            U32 const dictMatchIndex = dictHashTable[dictHash];
-            const BYTE* dictMatch = dictBase + dictMatchIndex;
-            if (dictMatchIndex <= dictStartIndex ||
-                MEM_read32(dictMatch) != MEM_read32(ip)) {
-                assert(stepSize >= 1);
-                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-                continue;
-            } else {
-                /* found a dict match */
-                U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);
-                mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
-                while (((ip>anchor) & (dictMatch>dictStart))
-                     && (ip[-1] == dictMatch[-1])) {
-                    ip--; dictMatch--; mLength++;
+        size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+
+        size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
+        U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);
+
+        U32 matchIndex = hashTable[hash0];
+        U32 curr = (U32)(ip0 - base);
+        size_t step = stepSize;
+        const size_t kStepIncr = 1 << kSearchStrength;
+        const BYTE* nextStep = ip0 + kStepIncr;
+
+        /* Inner search loop */
+        while (1) {
+            const BYTE* match = base + matchIndex;
+            const U32 repIndex = curr + 1 - offset_1;
+            const BYTE* repMatch = (repIndex < prefixStartIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+            size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
+            hashTable[hash0] = curr;   /* update hash table */
+
+            if (((U32) ((prefixStartIndex - 1) - repIndex) >=
+                 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
+                && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
+                const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
+                ip0++;
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                break;
+            }
+
+            if (dictTagsMatch) {
+                /* Found a possible dict match */
+                const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+                const BYTE* dictMatch = dictBase + dictMatchIndex;
+                if (dictMatchIndex > dictStartIndex &&
+                    MEM_read32(dictMatch) == MEM_read32(ip0)) {
+                    /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
+                    if (matchIndex <= prefixStartIndex) {
+                        U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
+                        mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
+                        while (((ip0 > anchor) & (dictMatch > dictStart))
+                            && (ip0[-1] == dictMatch[-1])) {
+                            ip0--;
+                            dictMatch--;
+                            mLength++;
+                        } /* catch up */
+                        offset_2 = offset_1;
+                        offset_1 = offset;
+                        ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                        break;
+                    }
+                }
+            }
+
+            if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
+                /* found a regular match */
+                U32 const offset = (U32) (ip0 - match);
+                mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
+                while (((ip0 > anchor) & (match > prefixStart))
+                       && (ip0[-1] == match[-1])) {
+                    ip0--;
+                    match--;
+                    mLength++;
                 } /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                break;
             }
-        } else if (MEM_read32(match) != MEM_read32(ip)) {
-            /* it's not a match, and we're not going to check the dictionary */
-            assert(stepSize >= 1);
-            ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-            continue;
-        } else {
-            /* found a regular match */
-            U32 const offset = (U32)(ip-match);
-            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-            while (((ip>anchor) & (match>prefixStart))
-                 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
-            offset_2 = offset_1;
-            offset_1 = offset;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
-        }
+
+            /* Prepare for next iteration */
+            dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
+            matchIndex = hashTable[hash1];
+
+            if (ip1 >= nextStep) {
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip0 = ip1;
+            ip1 = ip1 + step;
+            if (ip1 > ilimit) goto _cleanup;
+
+            curr = (U32)(ip0 - base);
+            hash0 = hash1;
+        }   /* end inner search loop */
 
         /* match found */
-        ip += mLength;
-        anchor = ip;
+        assert(mLength);
+        ip0 += mLength;
+        anchor = ip0;
 
-        if (ip <= ilimit) {
+        if (ip0 <= ilimit) {
             /* Fill Table */
             assert(base+curr+2 > istart);  /* check base overflow */
             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
-            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
 
             /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
+            while (ip0 <= ilimit) {
+                U32 const current2 = (U32)(ip0-base);
                 U32 const repIndex2 = current2 - offset_2;
                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
                         dictBase - dictIndexDelta + repIndex2 :
                         base + repIndex2;
                 if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                   && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
-                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
+                    ip0 += repLength2;
+                    anchor = ip0;
                     continue;
                 }
                 break;
             }
         }
+
+        /* Prepare for next iteration */
+        assert(ip0 == anchor);
+        ip1 = ip0 + stepSize;
     }
 
+_cleanup:
     /* save reps for next block */
-    rep[0] = offset_1 ? offset_1 : offsetSaved;
-    rep[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
@@ -545,7 +688,9 @@ size_t ZSTD_compressBlock_fast_dictMatchState(
 }
 
 
-static size_t ZSTD_compressBlock_fast_extDict_generic(
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_fast_extDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
 {
@@ -553,11 +698,10 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     U32* const hashTable = ms->hashTable;
     U32 const hlog = cParams->hashLog;
     /* support stepSize of 0 */
-    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
+    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
     const BYTE* const base = ms->window.base;
     const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
@@ -570,6 +714,28 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
     U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
+    const BYTE* ip2;
+    const BYTE* ip3;
+    U32 current0;
+
+
+    size_t hash0; /* hash for ip0 */
+    size_t hash1; /* hash for ip1 */
+    U32 idx; /* match idx for ip0 */
+    const BYTE* idxBase; /* base pointer for idx */
+
+    U32 offcode;
+    const BYTE* match0;
+    size_t mLength;
+    const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */
+
+    size_t step;
+    const BYTE* nextStep;
+    const size_t kStepIncr = (1 << (kSearchStrength - 1));
 
     (void)hasStep; /* not currently specialized on whether it's accelerated */
 
@@ -579,75 +745,202 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     if (prefixStartIndex == dictStartIndex)
         return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
 
-    /* Search Loop */
-    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        const size_t h = ZSTD_hashPtr(ip, hlog, mls);
-        const U32    matchIndex = hashTable[h];
-        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
-        const BYTE*  match = matchBase + matchIndex;
-        const U32    curr = (U32)(ip-base);
-        const U32    repIndex = curr + 1 - offset_1;
-        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
-        const BYTE* const repMatch = repBase + repIndex;
-        hashTable[h] = curr;   /* update hash table */
-        DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
-
-        if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */
-             & (offset_1 <= curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
-            size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, rLength);
-            ip += rLength;
-            anchor = ip;
-        } else {
-            if ( (matchIndex < dictStartIndex) ||
-                 (MEM_read32(match) != MEM_read32(ip)) ) {
-                assert(stepSize >= 1);
-                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-                continue;
+    {   U32 const curr = (U32)(ip0 - base);
+        U32 const maxRep = curr - dictStartIndex;
+        if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* start each op */
+_start: /* Requires: ip0 */
+
+    step = stepSize;
+    nextStep = ip0 + kStepIncr;
+
+    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
+    ip1 = ip0 + 1;
+    ip2 = ip0 + step;
+    ip3 = ip2 + 1;
+
+    if (ip3 >= ilimit) {
+        goto _cleanup;
+    }
+
+    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+    hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+
+    idx = hashTable[hash0];
+    idxBase = idx < prefixStartIndex ? dictBase : base;
+
+    do {
+        {   /* load repcode match for ip[2] */
+            U32 const current2 = (U32)(ip2 - base);
+            U32 const repIndex = current2 - offset_1;
+            const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+            U32 rval;
+            if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
+                 & (offset_1 > 0) ) {
+                rval = MEM_read32(repBase + repIndex);
+            } else {
+                rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
             }
-            {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
-                const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
-                U32 const offset = curr - matchIndex;
-                size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
-                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
-                offset_2 = offset_1; offset_1 = offset;  /* update offset history */
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
-                ip += mLength;
-                anchor = ip;
+
+            /* write back hash table entry */
+            current0 = (U32)(ip0 - base);
+            hashTable[hash0] = current0;
+
+            /* check repcode at ip[2] */
+            if (MEM_read32(ip2) == rval) {
+                ip0 = ip2;
+                match0 = repBase + repIndex;
+                matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                assert((match0 != prefixStart) & (match0 != dictStart));
+                mLength = ip0[-1] == match0[-1];
+                ip0 -= mLength;
+                match0 -= mLength;
+                offcode = REPCODE1_TO_OFFBASE;
+                mLength += 4;
+                goto _match;
         }   }
 
-        if (ip <= ilimit) {
-            /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;
-            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
-            /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
-                U32 const repIndex2 = current2 - offset_2;
-                const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 <= curr - dictStartIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
-                    { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, STORE_REPCODE_1, repLength2);
-                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
-                    continue;
-                }
-                break;
-    }   }   }
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip3;
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip0 + step;
+        ip3 = ip1 + step;
+
+        /* calculate step */
+        if (ip2 >= nextStep) {
+            step++;
+            PREFETCH_L1(ip1 + 64);
+            PREFETCH_L1(ip1 + 128);
+            nextStep += kStepIncr;
+        }
+    } while (ip3 < ilimit);
+
+_cleanup:
+    /* Note that there are probably still a couple positions we could search.
+     * However, it seems to be a meaningful performance hit to try to search
+     * them. So let's not. */
+
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
 
     /* save reps for next block */
-    rep[0] = offset_1;
-    rep[1] = offset_2;
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
+
+_offset: /* Requires: ip0, idx, idxBase */
+
+    /* Compute the offset code. */
+    {   U32 const offset = current0 - idx;
+        const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
+        matchEnd = idx < prefixStartIndex ? dictEnd : iend;
+        match0 = idxBase + idx;
+        offset_2 = offset_1;
+        offset_1 = offset;
+        offcode = OFFSET_TO_OFFBASE(offset);
+        mLength = 4;
+
+        /* Count the backwards match length. */
+        while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
+            ip0--;
+            match0--;
+            mLength++;
+    }   }
+
+_match: /* Requires: ip0, match0, offcode, matchEnd */
+
+    /* Count the forward length. */
+    assert(matchEnd != 0);
+    mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);
+
+    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
+
+    ip0 += mLength;
+    anchor = ip0;
+
+    /* write next hash table entry */
+    if (ip1 < ip0) {
+        hashTable[hash1] = (U32)(ip1 - base);
+    }
+
+    /* Fill table and check for immediate repcode. */
+    if (ip0 <= ilimit) {
+        /* Fill Table */
+        assert(base+current0+2 > istart);  /* check base overflow */
+        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+        while (ip0 <= ilimit) {
+            U32 const repIndex2 = (U32)(ip0-base) - offset_2;
+            const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+            if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0))  /* intentional underflow */
+                 && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
+                const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
+                ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                ip0 += repLength2;
+                anchor = ip0;
+                continue;
+            }
+            break;
+    }   }
+
+    goto _start;
 }
 
 ZSTD_GEN_FAST_FN(extDict, 4, 0)
@@ -660,6 +953,7 @@ size_t ZSTD_compressBlock_fast_extDict(
         void const* src, size_t srcSize)
 {
     U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState == NULL);
     switch(mls)
     {
     default: /* includes case 3 */
diff --git a/lib/zstd/compress/zstd_fast.h b/lib/zstd/compress/zstd_fast.h
index fddc2f532d21..e64d9e1b2d39 100644
--- a/lib/zstd/compress/zstd_fast.h
+++ b/lib/zstd/compress/zstd_fast.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -16,7 +17,8 @@
 #include "zstd_compress_internal.h"
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
-                        void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp);
 size_t ZSTD_compressBlock_fast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
diff --git a/lib/zstd/compress/zstd_lazy.c b/lib/zstd/compress/zstd_lazy.c
index 0298a01a7504..3e88d8a1a136 100644
--- a/lib/zstd/compress/zstd_lazy.c
+++ b/lib/zstd/compress/zstd_lazy.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -10,14 +11,23 @@
 
 #include "zstd_compress_internal.h"
 #include "zstd_lazy.h"
+#include "../common/bits.h" /* ZSTD_countTrailingZeros64 */
+
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
+
+#define kLazySkippingStep 8
 
 
 /*-*************************************
 *  Binary Tree search
 ***************************************/
 
-static void
-ZSTD_updateDUBT(ZSTD_matchState_t* ms,
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_updateDUBT(ZSTD_matchState_t* ms,
                 const BYTE* ip, const BYTE* iend,
                 U32 mls)
 {
@@ -60,8 +70,9 @@ ZSTD_updateDUBT(ZSTD_matchState_t* ms,
  *  sort one already inserted but unsorted position
  *  assumption : curr >= btlow == (curr - btmask)
  *  doesn't fail */
-static void
-ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
                  U32 curr, const BYTE* inputEnd,
                  U32 nbCompares, U32 btLow,
                  const ZSTD_dictMode_e dictMode)
@@ -149,8 +160,9 @@ ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
 }
 
 
-static size_t
-ZSTD_DUBT_findBetterDictMatch (
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_DUBT_findBetterDictMatch (
         const ZSTD_matchState_t* ms,
         const BYTE* const ip, const BYTE* const iend,
         size_t* offsetPtr,
@@ -197,8 +209,8 @@ ZSTD_DUBT_findBetterDictMatch (
             U32 matchIndex = dictMatchIndex + dictIndexDelta;
             if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
                 DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
-                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, STORE_OFFSET(curr - matchIndex), dictMatchIndex, matchIndex);
-                bestLength = matchLength, *offsetPtr = STORE_OFFSET(curr - matchIndex);
+                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
             }
             if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
                 break;   /* drop, to guarantee consistency (miss a little bit of compression) */
@@ -218,7 +230,7 @@ ZSTD_DUBT_findBetterDictMatch (
     }
 
     if (bestLength >= MINMATCH) {
-        U32 const mIndex = curr - (U32)STORED_OFFSET(*offsetPtr); (void)mIndex;
+        U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex;
         DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
                     curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
     }
@@ -227,10 +239,11 @@ ZSTD_DUBT_findBetterDictMatch (
 }
 
 
-static size_t
-ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iend,
-                        size_t* offsetPtr,
+                        size_t* offBasePtr,
                         U32 const mls,
                         const ZSTD_dictMode_e dictMode)
 {
@@ -327,8 +340,8 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
             if (matchLength > bestLength) {
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
-                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
-                    bestLength = matchLength, *offsetPtr = STORE_OFFSET(curr - matchIndex);
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) )
+                    bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex);
                 if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
                     if (dictMode == ZSTD_dictMatchState) {
                         nbCompares = 0; /* in addition to avoiding checking any
@@ -361,16 +374,16 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
         if (dictMode == ZSTD_dictMatchState && nbCompares) {
             bestLength = ZSTD_DUBT_findBetterDictMatch(
                     ms, ip, iend,
-                    offsetPtr, bestLength, nbCompares,
+                    offBasePtr, bestLength, nbCompares,
                     mls, dictMode);
         }
 
         assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */
         ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
         if (bestLength >= MINMATCH) {
-            U32 const mIndex = curr - (U32)STORED_OFFSET(*offsetPtr); (void)mIndex;
+            U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex;
             DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
-                        curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
+                        curr, (U32)bestLength, (U32)*offBasePtr, mIndex);
         }
         return bestLength;
     }
@@ -378,17 +391,18 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
 
 
 /* ZSTD_BtFindBestMatch() : Tree updater, providing best match */
-FORCE_INLINE_TEMPLATE size_t
-ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iLimit,
-                      size_t* offsetPtr,
+                      size_t* offBasePtr,
                 const U32 mls /* template */,
                 const ZSTD_dictMode_e dictMode)
 {
     DEBUGLOG(7, "ZSTD_BtFindBestMatch");
     if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
     ZSTD_updateDUBT(ms, ip, iLimit, mls);
-    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offsetPtr, mls, dictMode);
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode);
 }
 
 /* *********************************
@@ -561,7 +575,7 @@ size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nb
         /* save best solution */
         if (currentMl > ml) {
             ml = currentMl;
-            *offsetPtr = STORE_OFFSET(curr - (matchIndex + ddsIndexDelta));
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
             if (ip+currentMl == iLimit) {
                 /* best possible, avoids read overflow on next attempt */
                 return ml;
@@ -598,7 +612,7 @@ size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nb
             /* save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = STORE_OFFSET(curr - (matchIndex + ddsIndexDelta));
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
         }
@@ -614,10 +628,12 @@ size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nb
 
 /* Update chains up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-FORCE_INLINE_TEMPLATE U32 ZSTD_insertAndFindFirstIndex_internal(
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertAndFindFirstIndex_internal(
                         ZSTD_matchState_t* ms,
                         const ZSTD_compressionParameters* const cParams,
-                        const BYTE* ip, U32 const mls)
+                        const BYTE* ip, U32 const mls, U32 const lazySkipping)
 {
     U32* const hashTable  = ms->hashTable;
     const U32 hashLog = cParams->hashLog;
@@ -632,6 +648,9 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_insertAndFindFirstIndex_internal(
         NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
         hashTable[h] = idx;
         idx++;
+        /* Stop inserting every position when in the lazy skipping mode. */
+        if (lazySkipping)
+            break;
     }
 
     ms->nextToUpdate = target;
@@ -640,11 +659,12 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_insertAndFindFirstIndex_internal(
 
 U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
-    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch);
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch, /* lazySkipping*/ 0);
 }
 
 /* inlining is important to hardwire a hot branch (template emulation) */
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_HcFindBestMatch(
                         ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
@@ -684,14 +704,15 @@ size_t ZSTD_HcFindBestMatch(
     }
 
     /* HC4 match finder */
-    matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
+    matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls, ms->lazySkipping);
 
     for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) {
         size_t currentMl=0;
         if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
             const BYTE* const match = base + matchIndex;
             assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
-            if (match[ml] == ip[ml])   /* potentially better */
+            /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+            if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
                 currentMl = ZSTD_count(ip, match, iLimit);
         } else {
             const BYTE* const match = dictBase + matchIndex;
@@ -703,7 +724,7 @@ size_t ZSTD_HcFindBestMatch(
         /* save best solution */
         if (currentMl > ml) {
             ml = currentMl;
-            *offsetPtr = STORE_OFFSET(curr - matchIndex);
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
             if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
         }
 
@@ -739,7 +760,7 @@ size_t ZSTD_HcFindBestMatch(
             if (currentMl > ml) {
                 ml = currentMl;
                 assert(curr > matchIndex + dmsIndexDelta);
-                *offsetPtr = STORE_OFFSET(curr - (matchIndex + dmsIndexDelta));
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
 
@@ -756,8 +777,6 @@ size_t ZSTD_HcFindBestMatch(
 * (SIMD) Row-based matchfinder
 ***********************************/
 /* Constants for row-based hash */
-#define ZSTD_ROW_HASH_TAG_OFFSET 16     /* byte offset of hashes in the match state's tagTable from the beginning of a row */
-#define ZSTD_ROW_HASH_TAG_BITS 8        /* nb bits to use for the tag */
 #define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
 #define ZSTD_ROW_HASH_MAX_ENTRIES 64    /* absolute maximum number of entries per row, for all configurations */
 
@@ -769,64 +788,19 @@ typedef U64 ZSTD_VecMask;   /* Clarifies when we are interacting with a U64 repr
  * Starting from the LSB, returns the idx of the next non-zero bit.
  * Basically counting the nb of trailing zeroes.
  */
-static U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
-    assert(val != 0);
-#   if (defined(__GNUC__) && ((__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))
-    if (sizeof(size_t) == 4) {
-        U32 mostSignificantWord = (U32)(val >> 32);
-        U32 leastSignificantWord = (U32)val;
-        if (leastSignificantWord == 0) {
-            return 32 + (U32)__builtin_ctz(mostSignificantWord);
-        } else {
-            return (U32)__builtin_ctz(leastSignificantWord);
-        }
-    } else {
-        return (U32)__builtin_ctzll(val);
-    }
-#   else
-    /* Software ctz version: http://aggregate.org/MAGIC/#Trailing%20Zero%20Count
-     * and: https://stackoverflow.com/questions/2709430/count-number-of-bits-in-a-64-bit-long-big-integer
-     */
-    val = ~val & (val - 1ULL); /* Lowest set bit mask */
-    val = val - ((val >> 1) & 0x5555555555555555);
-    val = (val & 0x3333333333333333ULL) + ((val >> 2) & 0x3333333333333333ULL);
-    return (U32)((((val + (val >> 4)) & 0xF0F0F0F0F0F0F0FULL) * 0x101010101010101ULL) >> 56);
-#   endif
-}
-
-/* ZSTD_rotateRight_*():
- * Rotates a bitfield to the right by "count" bits.
- * https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
- */
-FORCE_INLINE_TEMPLATE
-U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
-    assert(count < 64);
-    count &= 0x3F; /* for fickle pattern recognition */
-    return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
-}
-
-FORCE_INLINE_TEMPLATE
-U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
-    assert(count < 32);
-    count &= 0x1F; /* for fickle pattern recognition */
-    return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
-}
-
-FORCE_INLINE_TEMPLATE
-U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
-    assert(count < 16);
-    count &= 0x0F; /* for fickle pattern recognition */
-    return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
+MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
+    return ZSTD_countTrailingZeros64(val);
 }
 
 /* ZSTD_row_nextIndex():
  * Returns the next index to insert at within a tagTable row, and updates the "head"
- * value to reflect the update. Essentially cycles backwards from [0, {entries per row})
+ * value to reflect the update. Essentially cycles backwards from [1, {entries per row})
  */
 FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextIndex(BYTE* const tagRow, U32 const rowMask) {
-  U32 const next = (*tagRow - 1) & rowMask;
-  *tagRow = (BYTE)next;
-  return next;
+    U32 next = (*tagRow-1) & rowMask;
+    next += (next == 0) ? rowMask : 0; /* skip first position */
+    *tagRow = (BYTE)next;
+    return next;
 }
 
 /* ZSTD_isAligned():
@@ -840,7 +814,7 @@ MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
 /* ZSTD_row_prefetch():
  * Performs prefetching for the hashTable and tagTable at a given row.
  */
-FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, U16 const* tagTable, U32 const relRow, U32 const rowLog) {
+FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, BYTE const* tagTable, U32 const relRow, U32 const rowLog) {
     PREFETCH_L1(hashTable + relRow);
     if (rowLog >= 5) {
         PREFETCH_L1(hashTable + relRow + 16);
@@ -859,18 +833,20 @@ FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, U16 const* ta
  * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
  * but not beyond iLimit.
  */
-FORCE_INLINE_TEMPLATE void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
                                    U32 const rowLog, U32 const mls,
                                    U32 idx, const BYTE* const iLimit)
 {
     U32 const* const hashTable = ms->hashTable;
-    U16 const* const tagTable = ms->tagTable;
+    BYTE const* const tagTable = ms->tagTable;
     U32 const hashLog = ms->rowHashLog;
     U32 const maxElemsToPrefetch = (base + idx) > iLimit ? 0 : (U32)(iLimit - (base + idx) + 1);
     U32 const lim = idx + MIN(ZSTD_ROW_HASH_CACHE_SIZE, maxElemsToPrefetch);
 
     for (; idx < lim; ++idx) {
-        U32 const hash = (U32)ZSTD_hashPtr(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+        U32 const hash = (U32)ZSTD_hashPtrSalted(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
         U32 const row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
         ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
         ms->hashCache[idx & ZSTD_ROW_HASH_CACHE_MASK] = hash;
@@ -885,12 +861,15 @@ FORCE_INLINE_TEMPLATE void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const B
  * Returns the hash of base + idx, and replaces the hash in the hash cache with the byte at
  * base + idx + ZSTD_ROW_HASH_CACHE_SIZE. Also prefetches the appropriate rows from hashTable and tagTable.
  */
-FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable,
-                                                  U16 const* tagTable, BYTE const* base,
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable,
+                                                  BYTE const* tagTable, BYTE const* base,
                                                   U32 idx, U32 const hashLog,
-                                                  U32 const rowLog, U32 const mls)
+                                                  U32 const rowLog, U32 const mls,
+                                                  U64 const hashSalt)
 {
-    U32 const newHash = (U32)ZSTD_hashPtr(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+    U32 const newHash = (U32)ZSTD_hashPtrSalted(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
     U32 const row = (newHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
     ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
     {   U32 const hash = cache[idx & ZSTD_ROW_HASH_CACHE_MASK];
@@ -902,28 +881,29 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTab
 /* ZSTD_row_update_internalImpl():
  * Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
  */
-FORCE_INLINE_TEMPLATE void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
-                                                        U32 updateStartIdx, U32 const updateEndIdx,
-                                                        U32 const mls, U32 const rowLog,
-                                                        U32 const rowMask, U32 const useCache)
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
+                                  U32 updateStartIdx, U32 const updateEndIdx,
+                                  U32 const mls, U32 const rowLog,
+                                  U32 const rowMask, U32 const useCache)
 {
     U32* const hashTable = ms->hashTable;
-    U16* const tagTable = ms->tagTable;
+    BYTE* const tagTable = ms->tagTable;
     U32 const hashLog = ms->rowHashLog;
     const BYTE* const base = ms->window.base;
 
     DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
     for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
-        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls)
-                                  : (U32)ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls, ms->hashSalt)
+                                  : (U32)ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
         U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
         U32* const row = hashTable + relRow;
-        BYTE* tagRow = (BYTE*)(tagTable + relRow);  /* Though tagTable is laid out as a table of U16, each tag is only 1 byte.
-                                                       Explicit cast allows us to get exact desired position within each row */
+        BYTE* tagRow = tagTable + relRow;
         U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
 
-        assert(hash == ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls));
-        ((BYTE*)tagRow)[pos + ZSTD_ROW_HASH_TAG_OFFSET] = hash & ZSTD_ROW_HASH_TAG_MASK;
+        assert(hash == ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt));
+        tagRow[pos] = hash & ZSTD_ROW_HASH_TAG_MASK;
         row[pos] = updateStartIdx;
     }
 }
@@ -932,9 +912,11 @@ FORCE_INLINE_TEMPLATE void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
  * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
  * Skips sections of long matches as is necessary.
  */
-FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
-                                                    U32 const mls, U32 const rowLog,
-                                                    U32 const rowMask, U32 const useCache)
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
+                              U32 const mls, U32 const rowLog,
+                              U32 const rowMask, U32 const useCache)
 {
     U32 idx = ms->nextToUpdate;
     const BYTE* const base = ms->window.base;
@@ -971,7 +953,35 @@ void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip) {
     const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
 
     DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog);
-    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* dont use cache */);
+    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */);
+}
+
+/* Returns the mask width of bits group of which will be set to 1. Given not all
+ * architectures have easy movemask instruction, this helps to iterate over
+ * groups of bits easier and faster.
+ */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    (void)rowEntries;
+#if defined(ZSTD_ARCH_ARM_NEON)
+    /* NEON path only works for little endian */
+    if (!MEM_isLittleEndian()) {
+        return 1;
+    }
+    if (rowEntries == 16) {
+        return 4;
+    }
+    if (rowEntries == 32) {
+        return 2;
+    }
+    if (rowEntries == 64) {
+        return 1;
+    }
+#endif
+    return 1;
 }
 
 #if defined(ZSTD_ARCH_X86_SSE2)
@@ -994,71 +1004,82 @@ ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U
 }
 #endif
 
-/* Returns a ZSTD_VecMask (U32) that has the nth bit set to 1 if the newly-computed "tag" matches
- * the hash at the nth position in a row of the tagTable.
- * Each row is a circular buffer beginning at the value of "head". So we must rotate the "matches" bitfield
- * to match up with the actual layout of the entries within the hashTable */
+#if defined(ZSTD_ARCH_ARM_NEON)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getNEONMask(const U32 rowEntries, const BYTE* const src, const BYTE tag, const U32 headGrouped)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    if (rowEntries == 16) {
+        /* vshrn_n_u16 shifts by 4 every u16 and narrows to 8 lower bits.
+         * After that groups of 4 bits represent the equalMask. We lower
+         * all bits except the highest in these groups by doing AND with
+         * 0x88 = 0b10001000.
+         */
+        const uint8x16_t chunk = vld1q_u8(src);
+        const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
+        const uint8x8_t res = vshrn_n_u16(equalMask, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x8888888888888888ull;
+    } else if (rowEntries == 32) {
+        /* Same idea as with rowEntries == 16 but doing AND with
+         * 0x55 = 0b01010101.
+         */
+        const uint16x8x2_t chunk = vld2q_u16((const uint16_t*)(const void*)src);
+        const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
+        const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x8_t t0 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk0, dup)), 6);
+        const uint8x8_t t1 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk1, dup)), 6);
+        const uint8x8_t res = vsli_n_u8(t0, t1, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0) ;
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x5555555555555555ull;
+    } else { /* rowEntries == 64 */
+        const uint8x16x4_t chunk = vld4q_u8(src);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
+        const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
+        const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
+        const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
+
+        const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
+        const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
+        const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
+        const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
+        const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped);
+    }
+}
+#endif
+
+/* Returns a ZSTD_VecMask (U64) that has the nth group (determined by
+ * ZSTD_row_matchMaskGroupWidth) of bits set to 1 if the newly-computed "tag"
+ * matches the hash at the nth position in a row of the tagTable.
+ * Each row is a circular buffer beginning at the value of "headGrouped". So we
+ * must rotate the "matches" bitfield to match up with the actual layout of the
+ * entries within the hashTable */
 FORCE_INLINE_TEMPLATE ZSTD_VecMask
-ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 head, const U32 rowEntries)
+ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 headGrouped, const U32 rowEntries)
 {
-    const BYTE* const src = tagRow + ZSTD_ROW_HASH_TAG_OFFSET;
+    const BYTE* const src = tagRow;
     assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
     assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    assert(ZSTD_row_matchMaskGroupWidth(rowEntries) * rowEntries <= sizeof(ZSTD_VecMask) * 8);
 
 #if defined(ZSTD_ARCH_X86_SSE2)
 
-    return ZSTD_row_getSSEMask(rowEntries / 16, src, tag, head);
+    return ZSTD_row_getSSEMask(rowEntries / 16, src, tag, headGrouped);
 
 #else /* SW or NEON-LE */
 
 # if defined(ZSTD_ARCH_ARM_NEON)
   /* This NEON path only works for little endian - otherwise use SWAR below */
     if (MEM_isLittleEndian()) {
-        if (rowEntries == 16) {
-            const uint8x16_t chunk = vld1q_u8(src);
-            const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
-            const uint16x8_t t0 = vshlq_n_u16(equalMask, 7);
-            const uint32x4_t t1 = vreinterpretq_u32_u16(vsriq_n_u16(t0, t0, 14));
-            const uint64x2_t t2 = vreinterpretq_u64_u32(vshrq_n_u32(t1, 14));
-            const uint8x16_t t3 = vreinterpretq_u8_u64(vsraq_n_u64(t2, t2, 28));
-            const U16 hi = (U16)vgetq_lane_u8(t3, 8);
-            const U16 lo = (U16)vgetq_lane_u8(t3, 0);
-            return ZSTD_rotateRight_U16((hi << 8) | lo, head);
-        } else if (rowEntries == 32) {
-            const uint16x8x2_t chunk = vld2q_u16((const U16*)(const void*)src);
-            const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
-            const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
-            const uint8x16_t equalMask0 = vceqq_u8(chunk0, vdupq_n_u8(tag));
-            const uint8x16_t equalMask1 = vceqq_u8(chunk1, vdupq_n_u8(tag));
-            const int8x8_t pack0 = vqmovn_s16(vreinterpretq_s16_u8(equalMask0));
-            const int8x8_t pack1 = vqmovn_s16(vreinterpretq_s16_u8(equalMask1));
-            const uint8x8_t t0 = vreinterpret_u8_s8(pack0);
-            const uint8x8_t t1 = vreinterpret_u8_s8(pack1);
-            const uint8x8_t t2 = vsri_n_u8(t1, t0, 2);
-            const uint8x8x2_t t3 = vuzp_u8(t2, t0);
-            const uint8x8_t t4 = vsri_n_u8(t3.val[1], t3.val[0], 4);
-            const U32 matches = vget_lane_u32(vreinterpret_u32_u8(t4), 0);
-            return ZSTD_rotateRight_U32(matches, head);
-        } else { /* rowEntries == 64 */
-            const uint8x16x4_t chunk = vld4q_u8(src);
-            const uint8x16_t dup = vdupq_n_u8(tag);
-            const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
-            const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
-            const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
-            const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
-
-            const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
-            const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
-            const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
-            const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
-            const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
-            const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
-            return ZSTD_rotateRight_U64(matches, head);
-        }
+        return ZSTD_row_getNEONMask(rowEntries, src, tag, headGrouped);
     }
 # endif /* ZSTD_ARCH_ARM_NEON */
     /* SWAR */
-    {   const size_t chunkSize = sizeof(size_t);
+    {   const int chunkSize = sizeof(size_t);
         const size_t shiftAmount = ((chunkSize * 8) - chunkSize);
         const size_t xFF = ~((size_t)0);
         const size_t x01 = xFF / 0xFF;
@@ -1091,11 +1112,11 @@ ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 head,
         }
         matches = ~matches;
         if (rowEntries == 16) {
-            return ZSTD_rotateRight_U16((U16)matches, head);
+            return ZSTD_rotateRight_U16((U16)matches, headGrouped);
         } else if (rowEntries == 32) {
-            return ZSTD_rotateRight_U32((U32)matches, head);
+            return ZSTD_rotateRight_U32((U32)matches, headGrouped);
         } else {
-            return ZSTD_rotateRight_U64((U64)matches, head);
+            return ZSTD_rotateRight_U64((U64)matches, headGrouped);
         }
     }
 #endif
@@ -1103,20 +1124,21 @@ ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 head,
 
 /* The high-level approach of the SIMD row based match finder is as follows:
  * - Figure out where to insert the new entry:
- *      - Generate a hash from a byte along with an additional 1-byte "short hash". The additional byte is our "tag"
- *      - The hashTable is effectively split into groups or "rows" of 16 or 32 entries of U32, and the hash determines
+ *      - Generate a hash for current input posistion and split it into a one byte of tag and `rowHashLog` bits of index.
+ *           - The hash is salted by a value that changes on every contex reset, so when the same table is used
+ *             we will avoid collisions that would otherwise slow us down by intorducing phantom matches.
+ *      - The hashTable is effectively split into groups or "rows" of 15 or 31 entries of U32, and the index determines
  *        which row to insert into.
- *      - Determine the correct position within the row to insert the entry into. Each row of 16 or 32 can
- *        be considered as a circular buffer with a "head" index that resides in the tagTable.
- *      - Also insert the "tag" into the equivalent row and position in the tagTable.
- *          - Note: The tagTable has 17 or 33 1-byte entries per row, due to 16 or 32 tags, and 1 "head" entry.
- *                  The 17 or 33 entry rows are spaced out to occur every 32 or 64 bytes, respectively,
- *                  for alignment/performance reasons, leaving some bytes unused.
- * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte "short hash" and
+ *      - Determine the correct position within the row to insert the entry into. Each row of 15 or 31 can
+ *        be considered as a circular buffer with a "head" index that resides in the tagTable (overall 16 or 32 bytes
+ *        per row).
+ * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte tag calculated for the position and
  *   generate a bitfield that we can cycle through to check the collisions in the hash table.
  * - Pick the longest match.
+ * - Insert the tag into the equivalent row and position in the tagTable.
  */
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_RowFindBestMatch(
                         ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
@@ -1125,7 +1147,7 @@ size_t ZSTD_RowFindBestMatch(
                         const U32 rowLog)
 {
     U32* const hashTable = ms->hashTable;
-    U16* const tagTable = ms->tagTable;
+    BYTE* const tagTable = ms->tagTable;
     U32* const hashCache = ms->hashCache;
     const U32 hashLog = ms->rowHashLog;
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -1143,8 +1165,11 @@ size_t ZSTD_RowFindBestMatch(
     const U32 rowEntries = (1U << rowLog);
     const U32 rowMask = rowEntries - 1;
     const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */
+    const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries);
+    const U64 hashSalt = ms->hashSalt;
     U32 nbAttempts = 1U << cappedSearchLog;
     size_t ml=4-1;
+    U32 hash;
 
     /* DMS/DDS variables that may be referenced laster */
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
@@ -1168,7 +1193,7 @@ size_t ZSTD_RowFindBestMatch(
     if (dictMode == ZSTD_dictMatchState) {
         /* Prefetch DMS rows */
         U32* const dmsHashTable = dms->hashTable;
-        U16* const dmsTagTable = dms->tagTable;
+        BYTE* const dmsTagTable = dms->tagTable;
         U32 const dmsHash = (U32)ZSTD_hashPtr(ip, dms->rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
         U32 const dmsRelRow = (dmsHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
         dmsTag = dmsHash & ZSTD_ROW_HASH_TAG_MASK;
@@ -1178,23 +1203,34 @@ size_t ZSTD_RowFindBestMatch(
     }
 
     /* Update the hashTable and tagTable up to (but not including) ip */
-    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */);
+    if (!ms->lazySkipping) {
+        ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */);
+        hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls, hashSalt);
+    } else {
+        /* Stop inserting every position when in the lazy skipping mode.
+         * The hash cache is also not kept up to date in this mode.
+         */
+        hash = (U32)ZSTD_hashPtrSalted(ip, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
+        ms->nextToUpdate = curr;
+    }
+    ms->hashSaltEntropy += hash; /* collect salt entropy */
+
     {   /* Get the hash for ip, compute the appropriate row */
-        U32 const hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls);
         U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
         U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK;
         U32* const row = hashTable + relRow;
         BYTE* tagRow = (BYTE*)(tagTable + relRow);
-        U32 const head = *tagRow & rowMask;
+        U32 const headGrouped = (*tagRow & rowMask) * groupWidth;
         U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
         size_t numMatches = 0;
         size_t currMatch = 0;
-        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, head, rowEntries);
+        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries);
 
         /* Cycle through the matches and prefetch */
-        for (; (matches > 0) && (nbAttempts > 0); --nbAttempts, matches &= (matches - 1)) {
-            U32 const matchPos = (head + ZSTD_VecMask_next(matches)) & rowMask;
+        for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+            U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
             U32 const matchIndex = row[matchPos];
+            if(matchPos == 0) continue;
             assert(numMatches < rowEntries);
             if (matchIndex < lowLimit)
                 break;
@@ -1204,13 +1240,14 @@ size_t ZSTD_RowFindBestMatch(
                 PREFETCH_L1(dictBase + matchIndex);
             }
             matchBuffer[numMatches++] = matchIndex;
+            --nbAttempts;
         }
 
         /* Speed opt: insert current byte into hashtable too. This allows us to avoid one iteration of the loop
            in ZSTD_row_update_internal() at the next search. */
         {
             U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
-            tagRow[pos + ZSTD_ROW_HASH_TAG_OFFSET] = (BYTE)tag;
+            tagRow[pos] = (BYTE)tag;
             row[pos] = ms->nextToUpdate++;
         }
 
@@ -1224,7 +1261,8 @@ size_t ZSTD_RowFindBestMatch(
             if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
                 const BYTE* const match = base + matchIndex;
                 assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
-                if (match[ml] == ip[ml])   /* potentially better */
+                /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+                if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
                     currentMl = ZSTD_count(ip, match, iLimit);
             } else {
                 const BYTE* const match = dictBase + matchIndex;
@@ -1236,7 +1274,7 @@ size_t ZSTD_RowFindBestMatch(
             /* Save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = STORE_OFFSET(curr - matchIndex);
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
         }
@@ -1254,19 +1292,21 @@ size_t ZSTD_RowFindBestMatch(
         const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
         const U32 dmsIndexDelta        = dictLimit - dmsSize;
 
-        {   U32 const head = *dmsTagRow & rowMask;
+        {   U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth;
             U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
             size_t numMatches = 0;
             size_t currMatch = 0;
-            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, head, rowEntries);
+            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries);
 
-            for (; (matches > 0) && (nbAttempts > 0); --nbAttempts, matches &= (matches - 1)) {
-                U32 const matchPos = (head + ZSTD_VecMask_next(matches)) & rowMask;
+            for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+                U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
                 U32 const matchIndex = dmsRow[matchPos];
+                if(matchPos == 0) continue;
                 if (matchIndex < dmsLowestIndex)
                     break;
                 PREFETCH_L1(dmsBase + matchIndex);
                 matchBuffer[numMatches++] = matchIndex;
+                --nbAttempts;
             }
 
             /* Return the longest match */
@@ -1285,7 +1325,7 @@ size_t ZSTD_RowFindBestMatch(
                 if (currentMl > ml) {
                     ml = currentMl;
                     assert(curr > matchIndex + dmsIndexDelta);
-                    *offsetPtr = STORE_OFFSET(curr - (matchIndex + dmsIndexDelta));
+                    *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
                     if (ip+currentMl == iLimit) break;
                 }
             }
@@ -1472,8 +1512,9 @@ FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax(
 *  Common parser - lazy strategy
 *********************************/
 
-FORCE_INLINE_TEMPLATE size_t
-ZSTD_compressBlock_lazy_generic(
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_lazy_generic(
                         ZSTD_matchState_t* ms, seqStore_t* seqStore,
                         U32 rep[ZSTD_REP_NUM],
                         const void* src, size_t srcSize,
@@ -1491,7 +1532,8 @@ ZSTD_compressBlock_lazy_generic(
     const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
     const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
 
-    U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
 
     const int isDMS = dictMode == ZSTD_dictMatchState;
     const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
@@ -1512,8 +1554,8 @@ ZSTD_compressBlock_lazy_generic(
         U32 const curr = (U32)(ip - base);
         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
         U32 const maxRep = curr - windowLow;
-        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
     }
     if (isDxS) {
         /* dictMatchState repCode checks don't currently handle repCode == 0
@@ -1522,10 +1564,11 @@ ZSTD_compressBlock_lazy_generic(
         assert(offset_2 <= dictAndPrefixLength);
     }
 
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
     if (searchMethod == search_rowHash) {
-        ZSTD_row_fillHashCache(ms, base, rowLog,
-                            MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */),
-                            ms->nextToUpdate, ilimit);
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
     }
 
     /* Match Loop */
@@ -1537,7 +1580,7 @@ ZSTD_compressBlock_lazy_generic(
 #endif
     while (ip < ilimit) {
         size_t matchLength=0;
-        size_t offcode=STORE_REPCODE_1;
+        size_t offBase = REPCODE1_TO_OFFBASE;
         const BYTE* start=ip+1;
         DEBUGLOG(7, "search baseline (depth 0)");
 
@@ -1562,14 +1605,23 @@ ZSTD_compressBlock_lazy_generic(
         }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 999999999;
-            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offsetFound, mls, rowLog, searchMethod, dictMode);
+        {   size_t offbaseFound = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offcode=offsetFound;
+                matchLength = ml2, start = ip, offBase = offbaseFound;
         }
 
         if (matchLength < 4) {
-            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */;
+            ip += step;
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
             continue;
         }
 
@@ -1579,12 +1631,12 @@ ZSTD_compressBlock_lazy_generic(
             DEBUGLOG(7, "search depth 1");
             ip ++;
             if ( (dictMode == ZSTD_noDict)
-              && (offcode) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+              && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                 size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                 int const gain2 = (int)(mlRep * 3);
-                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                 if ((mlRep >= 4) && (gain2 > gain1))
-                    matchLength = mlRep, offcode = STORE_REPCODE_1, start = ip;
+                    matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
             }
             if (isDxS) {
                 const U32 repIndex = (U32)(ip - base) - offset_1;
@@ -1596,17 +1648,17 @@ ZSTD_compressBlock_lazy_generic(
                     const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
                     size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
                     int const gain2 = (int)(mlRep * 3);
-                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((mlRep >= 4) && (gain2 > gain1))
-                        matchLength = mlRep, offcode = STORE_REPCODE_1, start = ip;
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }
             }
-            {   size_t offset2=999999999;
-                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offset2, mls, rowLog, searchMethod, dictMode);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offset2)));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 4);
+            {   size_t ofbCandidate=999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
-                    matchLength = ml2, offcode = offset2, start = ip;
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
                     continue;   /* search a better one */
             }   }
 
@@ -1615,12 +1667,12 @@ ZSTD_compressBlock_lazy_generic(
                 DEBUGLOG(7, "search depth 2");
                 ip ++;
                 if ( (dictMode == ZSTD_noDict)
-                  && (offcode) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                  && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                     size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                     int const gain2 = (int)(mlRep * 4);
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((mlRep >= 4) && (gain2 > gain1))
-                        matchLength = mlRep, offcode = STORE_REPCODE_1, start = ip;
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }
                 if (isDxS) {
                     const U32 repIndex = (U32)(ip - base) - offset_1;
@@ -1632,17 +1684,17 @@ ZSTD_compressBlock_lazy_generic(
                         const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
                         size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
                         int const gain2 = (int)(mlRep * 4);
-                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                         if ((mlRep >= 4) && (gain2 > gain1))
-                            matchLength = mlRep, offcode = STORE_REPCODE_1, start = ip;
+                            matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                     }
                 }
-                {   size_t offset2=999999999;
-                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offset2, mls, rowLog, searchMethod, dictMode);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offset2)));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 7);
+                {   size_t ofbCandidate=999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
-                        matchLength = ml2, offcode = offset2, start = ip;
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
@@ -1653,26 +1705,33 @@ ZSTD_compressBlock_lazy_generic(
          * notably if `value` is unsigned, resulting in a large positive `-value`.
          */
         /* catch up */
-        if (STORED_IS_OFFSET(offcode)) {
+        if (OFFBASE_IS_OFFSET(offBase)) {
             if (dictMode == ZSTD_noDict) {
-                while ( ((start > anchor) & (start - STORED_OFFSET(offcode) > prefixLowest))
-                     && (start[-1] == (start-STORED_OFFSET(offcode))[-1]) )  /* only search for offset within prefix */
+                while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest))
+                     && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) )  /* only search for offset within prefix */
                     { start--; matchLength++; }
             }
             if (isDxS) {
-                U32 const matchIndex = (U32)((size_t)(start-base) - STORED_OFFSET(offcode));
+                U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
                 const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
                 const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
                 while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
             }
-            offset_2 = offset_1; offset_1 = (U32)STORED_OFFSET(offcode);
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
         }
         /* store sequence */
 _storeSequence:
         {   size_t const litLength = (size_t)(start - anchor);
-            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offcode, matchLength);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
             anchor = ip = start + matchLength;
         }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
 
         /* check immediate repcode */
         if (isDxS) {
@@ -1686,8 +1745,8 @@ ZSTD_compressBlock_lazy_generic(
                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
                     const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
                     matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
-                    offcode = offset_2; offset_2 = offset_1; offset_1 = (U32)offcode;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, matchLength);
+                    offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                     ip += matchLength;
                     anchor = ip;
                     continue;
@@ -1701,166 +1760,181 @@ ZSTD_compressBlock_lazy_generic(
                  && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
                 /* store sequence */
                 matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                offcode = offset_2; offset_2 = offset_1; offset_1 = (U32)offcode; /* swap repcodes */
-                ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, matchLength);
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                 ip += matchLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
     }   }   }
 
-    /* Save reps for next block */
-    rep[0] = offset_1 ? offset_1 : savedOffset;
-    rep[1] = offset_2 ? offset_2 : savedOffset;
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
 }
+#endif /* build exclusions */
 
 
-size_t ZSTD_compressBlock_btlazy2(
+#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_greedy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_lazy2(
+size_t ZSTD_compressBlock_greedy_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_lazy(
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
 }
 
-size_t ZSTD_compressBlock_greedy(
+size_t ZSTD_compressBlock_greedy_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+size_t ZSTD_compressBlock_greedy_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_lazy2_dictMatchState(
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch);
 }
+#endif
 
-size_t ZSTD_compressBlock_lazy_dictMatchState(
+#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_greedy_dictMatchState(
+size_t ZSTD_compressBlock_lazy_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
 }
 
-
-size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
 }
 
-size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+size_t ZSTD_compressBlock_lazy_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+size_t ZSTD_compressBlock_lazy_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState);
 }
 
-/* Row-based matchfinder */
-size_t ZSTD_compressBlock_lazy2_row(
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch);
 }
+#endif
 
-size_t ZSTD_compressBlock_lazy_row(
+#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_greedy_row(
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
 }
 
-size_t ZSTD_compressBlock_lazy_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy2_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_greedy_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState);
 }
 
-
 size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dedicatedDictSearch);
 }
+#endif
 
-size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
+#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btlazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
 }
+#endif
 
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_compressBlock_lazy_extDict_generic(
                         ZSTD_matchState_t* ms, seqStore_t* seqStore,
                         U32 rep[ZSTD_REP_NUM],
@@ -1886,12 +1960,13 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
 
     DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
 
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
     /* init */
     ip += (ip == prefixStart);
     if (searchMethod == search_rowHash) {
-        ZSTD_row_fillHashCache(ms, base, rowLog,
-                               MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */),
-                               ms->nextToUpdate, ilimit);
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
     }
 
     /* Match Loop */
@@ -1903,7 +1978,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
 #endif
     while (ip < ilimit) {
         size_t matchLength=0;
-        size_t offcode=STORE_REPCODE_1;
+        size_t offBase = REPCODE1_TO_OFFBASE;
         const BYTE* start=ip+1;
         U32 curr = (U32)(ip-base);
 
@@ -1922,14 +1997,23 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         }   }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 999999999;
-            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offsetFound, mls, rowLog, searchMethod, ZSTD_extDict);
+        {   size_t ofbCandidate = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offcode=offsetFound;
+                matchLength = ml2, start = ip, offBase = ofbCandidate;
         }
 
         if (matchLength < 4) {
-            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength);
+            ip += step + 1;   /* jump faster over incompressible sections */
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
             continue;
         }
 
@@ -1939,7 +2023,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             ip ++;
             curr++;
             /* check repCode */
-            if (offcode) {
+            if (offBase) {
                 const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
                 const U32 repIndex = (U32)(curr - offset_1);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
@@ -1951,18 +2035,18 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                     const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                     size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                     int const gain2 = (int)(repLength * 3);
-                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((repLength >= 4) && (gain2 > gain1))
-                        matchLength = repLength, offcode = STORE_REPCODE_1, start = ip;
+                        matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
             }   }
 
             /* search match, depth 1 */
-            {   size_t offset2=999999999;
-                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offset2, mls, rowLog, searchMethod, ZSTD_extDict);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offset2)));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 4);
+            {   size_t ofbCandidate = 999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
-                    matchLength = ml2, offcode = offset2, start = ip;
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
                     continue;   /* search a better one */
             }   }
 
@@ -1971,7 +2055,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                 ip ++;
                 curr++;
                 /* check repCode */
-                if (offcode) {
+                if (offBase) {
                     const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
                     const U32 repIndex = (U32)(curr - offset_1);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
@@ -1983,38 +2067,45 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                         size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                         int const gain2 = (int)(repLength * 4);
-                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 1);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                         if ((repLength >= 4) && (gain2 > gain1))
-                            matchLength = repLength, offcode = STORE_REPCODE_1, start = ip;
+                            matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }   }
 
                 /* search match, depth 2 */
-                {   size_t offset2=999999999;
-                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offset2, mls, rowLog, searchMethod, ZSTD_extDict);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offset2)));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)STORED_TO_OFFBASE(offcode)) + 7);
+                {   size_t ofbCandidate = 999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
-                        matchLength = ml2, offcode = offset2, start = ip;
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
         }
 
         /* catch up */
-        if (STORED_IS_OFFSET(offcode)) {
-            U32 const matchIndex = (U32)((size_t)(start-base) - STORED_OFFSET(offcode));
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
             const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
             const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
             while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
-            offset_2 = offset_1; offset_1 = (U32)STORED_OFFSET(offcode);
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
         }
 
         /* store sequence */
 _storeSequence:
         {   size_t const litLength = (size_t)(start - anchor);
-            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offcode, matchLength);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
             anchor = ip = start + matchLength;
         }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
 
         /* check immediate repcode */
         while (ip <= ilimit) {
@@ -2029,8 +2120,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                 matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
-                offcode = offset_2; offset_2 = offset_1; offset_1 = (U32)offcode;   /* swap offset history */
-                ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, matchLength);
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                 ip += matchLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
@@ -2045,8 +2136,9 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
     /* Return the last literals size */
     return (size_t)(iend - anchor);
 }
+#endif /* build exclusions */
 
-
+#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
 size_t ZSTD_compressBlock_greedy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
@@ -2054,49 +2146,55 @@ size_t ZSTD_compressBlock_greedy_extDict(
     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0);
 }
 
-size_t ZSTD_compressBlock_lazy_extDict(
+size_t ZSTD_compressBlock_greedy_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
-
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0);
 }
+#endif
 
-size_t ZSTD_compressBlock_lazy2_extDict(
+#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
 }
 
-size_t ZSTD_compressBlock_btlazy2_extDict(
+size_t ZSTD_compressBlock_lazy_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1);
 }
+#endif
 
-size_t ZSTD_compressBlock_greedy_extDict_row(
+#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
+
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
 }
 
-size_t ZSTD_compressBlock_lazy_extDict_row(
+size_t ZSTD_compressBlock_lazy2_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
-
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
 }
+#endif
 
-size_t ZSTD_compressBlock_lazy2_extDict_row(
+#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btlazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
 }
+#endif
diff --git a/lib/zstd/compress/zstd_lazy.h b/lib/zstd/compress/zstd_lazy.h
index e5bdf4df8dde..22c9201f4e63 100644
--- a/lib/zstd/compress/zstd_lazy.h
+++ b/lib/zstd/compress/zstd_lazy.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -22,98 +23,175 @@
  */
 #define ZSTD_LAZY_DDSS_BUCKET_LOG 2
 
+#define ZSTD_ROW_HASH_TAG_BITS 8        /* nb bits to use for the tag */
+
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
 U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
 void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip);
 
 void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip);
 
 void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). preemptively increase value of ZSTD_DUBT_UNSORTED_MARK */
+#endif
 
-size_t ZSTD_compressBlock_btlazy2(
+#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_greedy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2(
+size_t ZSTD_compressBlock_greedy_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy(
+size_t ZSTD_compressBlock_greedy_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy(
+size_t ZSTD_compressBlock_greedy_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_row(
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_row(
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_row(
+size_t ZSTD_compressBlock_greedy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-
-size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+size_t ZSTD_compressBlock_greedy_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_dictMatchState(
+
+#define ZSTD_COMPRESSBLOCK_GREEDY ZSTD_compressBlock_greedy
+#define ZSTD_COMPRESSBLOCK_GREEDY_ROW ZSTD_compressBlock_greedy_row
+#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE ZSTD_compressBlock_greedy_dictMatchState
+#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW ZSTD_compressBlock_greedy_dictMatchState_row
+#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH ZSTD_compressBlock_greedy_dedicatedDictSearch
+#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_greedy_dedicatedDictSearch_row
+#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT ZSTD_compressBlock_greedy_extDict
+#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW ZSTD_compressBlock_greedy_extDict_row
+#else
+#define ZSTD_COMPRESSBLOCK_GREEDY NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_ROW NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT NULL
+#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW NULL
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_dictMatchState(
+size_t ZSTD_compressBlock_lazy_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_dictMatchState(
+size_t ZSTD_compressBlock_lazy_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_dictMatchState_row(
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-
-size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+size_t ZSTD_compressBlock_lazy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+size_t ZSTD_compressBlock_lazy_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+
+#define ZSTD_COMPRESSBLOCK_LAZY ZSTD_compressBlock_lazy
+#define ZSTD_COMPRESSBLOCK_LAZY_ROW ZSTD_compressBlock_lazy_row
+#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE ZSTD_compressBlock_lazy_dictMatchState
+#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW ZSTD_compressBlock_lazy_dictMatchState_row
+#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH ZSTD_compressBlock_lazy_dedicatedDictSearch
+#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_lazy_dedicatedDictSearch_row
+#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT ZSTD_compressBlock_lazy_extDict
+#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW ZSTD_compressBlock_lazy_extDict_row
+#else
+#define ZSTD_COMPRESSBLOCK_LAZY NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT NULL
+#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW NULL
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
+size_t ZSTD_compressBlock_lazy2_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
+size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-
-size_t ZSTD_compressBlock_greedy_extDict(
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_extDict(
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_lazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy_extDict_row(
+size_t ZSTD_compressBlock_lazy2_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_extDict_row(
+
+#define ZSTD_COMPRESSBLOCK_LAZY2 ZSTD_compressBlock_lazy2
+#define ZSTD_COMPRESSBLOCK_LAZY2_ROW ZSTD_compressBlock_lazy2_row
+#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE ZSTD_compressBlock_lazy2_dictMatchState
+#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW ZSTD_compressBlock_lazy2_dictMatchState_row
+#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH ZSTD_compressBlock_lazy2_dedicatedDictSearch
+#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_lazy2_dedicatedDictSearch_row
+#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT ZSTD_compressBlock_lazy2_extDict
+#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW ZSTD_compressBlock_lazy2_extDict_row
+#else
+#define ZSTD_COMPRESSBLOCK_LAZY2 NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT NULL
+#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW NULL
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btlazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_extDict_row(
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btlazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-        
+
+#define ZSTD_COMPRESSBLOCK_BTLAZY2 ZSTD_compressBlock_btlazy2
+#define ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE ZSTD_compressBlock_btlazy2_dictMatchState
+#define ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT ZSTD_compressBlock_btlazy2_extDict
+#else
+#define ZSTD_COMPRESSBLOCK_BTLAZY2 NULL
+#define ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT NULL
+#endif
+
 
 
 #endif /* ZSTD_LAZY_H */
diff --git a/lib/zstd/compress/zstd_ldm.c b/lib/zstd/compress/zstd_ldm.c
index dd86fc83e7dd..07f3bc6437ce 100644
--- a/lib/zstd/compress/zstd_ldm.c
+++ b/lib/zstd/compress/zstd_ldm.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -242,11 +243,15 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
     switch(ms->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
         break;
 
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
         break;
 
     case ZSTD_greedy:
@@ -318,7 +323,9 @@ static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
     }
 }
 
-static size_t ZSTD_ldm_generateSequences_internal(
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_ldm_generateSequences_internal(
         ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
         ldmParams_t const* params, void const* src, size_t srcSize)
 {
@@ -549,7 +556,7 @@ size_t ZSTD_ldm_generateSequences(
          * the window through early invalidation.
          * TODO: * Test the chunk size.
          *       * Try invalidation after the sequence generation and test the
-         *         the offset against maxDist directly.
+         *         offset against maxDist directly.
          *
          * NOTE: Because of dictionaries + sequence splitting we MUST make sure
          * that any offset used is valid at the END of the sequence, since it may
@@ -689,7 +696,6 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
         /* maybeSplitSequence updates rawSeqStore->pos */
         rawSeq const sequence = maybeSplitSequence(rawSeqStore,
                                                    (U32)(iend - ip), minMatch);
-        int i;
         /* End signal */
         if (sequence.offset == 0)
             break;
@@ -702,6 +708,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
         /* Run the block compressor */
         DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
         {
+            int i;
             size_t const newLitLength =
                 blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
             ip += sequence.litLength;
@@ -711,7 +718,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
             rep[0] = sequence.offset;
             /* Store the sequence */
             ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
-                          STORE_OFFSET(sequence.offset),
+                          OFFSET_TO_OFFBASE(sequence.offset),
                           sequence.matchLength);
             ip += sequence.matchLength;
         }
diff --git a/lib/zstd/compress/zstd_ldm.h b/lib/zstd/compress/zstd_ldm.h
index fbc6a5e88fd7..c540731abde7 100644
--- a/lib/zstd/compress/zstd_ldm.h
+++ b/lib/zstd/compress/zstd_ldm.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/compress/zstd_ldm_geartab.h b/lib/zstd/compress/zstd_ldm_geartab.h
index 647f865be290..cfccfc46f6f7 100644
--- a/lib/zstd/compress/zstd_ldm_geartab.h
+++ b/lib/zstd/compress/zstd_ldm_geartab.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/compress/zstd_opt.c b/lib/zstd/compress/zstd_opt.c
index fd82acfda62f..a87b66ac8d24 100644
--- a/lib/zstd/compress/zstd_opt.c
+++ b/lib/zstd/compress/zstd_opt.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,11 +13,14 @@
 #include "hist.h"
 #include "zstd_opt.h"
 
+#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
 
 #define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
 #define ZSTD_MAX_PRICE     (1<<30)
 
-#define ZSTD_PREDEF_THRESHOLD 1024   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
+#define ZSTD_PREDEF_THRESHOLD 8   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
 
 
 /*-*************************************
@@ -26,27 +30,35 @@
 #if 0    /* approximation at bit level (for tests) */
 #  define BITCOST_ACCURACY 0
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat, opt) ((void)opt, ZSTD_bitWeight(stat))
+#  define WEIGHT(stat, opt) ((void)(opt), ZSTD_bitWeight(stat))
 #elif 0  /* fractional bit accuracy (for tests) */
 #  define BITCOST_ACCURACY 8
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))
+#  define WEIGHT(stat,opt) ((void)(opt), ZSTD_fracWeight(stat))
 #else    /* opt==approx, ultra==accurate */
 #  define BITCOST_ACCURACY 8
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat,opt) (opt ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#  define WEIGHT(stat,opt) ((opt) ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
 #endif
 
+/* ZSTD_bitWeight() :
+ * provide estimated "cost" of a stat in full bits only */
 MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
 {
     return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
 }
 
+/* ZSTD_fracWeight() :
+ * provide fractional-bit "cost" of a stat,
+ * using linear interpolation approximation */
 MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
 {
     U32 const stat = rawStat + 1;
     U32 const hb = ZSTD_highbit32(stat);
     U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    /* Fweight was meant for "Fractional weight"
+     * but it's effectively a value between 1 and 2
+     * using fixed point arithmetic */
     U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
     U32 const weight = BWeight + FWeight;
     assert(hb + BITCOST_ACCURACY < 31);
@@ -57,7 +69,7 @@ MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
 /* debugging function,
  * @return price in bytes as fractional value
  * for debug messages only */
-MEM_STATIC double ZSTD_fCost(U32 price)
+MEM_STATIC double ZSTD_fCost(int price)
 {
     return (double)price / (BITCOST_MULTIPLIER*8);
 }
@@ -88,20 +100,26 @@ static U32 sum_u32(const unsigned table[], size_t nbElts)
     return total;
 }
 
-static U32 ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift)
+typedef enum { base_0possible=0, base_1guaranteed=1 } base_directive_e;
+
+static U32
+ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift, base_directive_e base1)
 {
     U32 s, sum=0;
-    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)", (unsigned)lastEltIndex+1, (unsigned)shift);
+    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)",
+            (unsigned)lastEltIndex+1, (unsigned)shift );
     assert(shift < 30);
     for (s=0; s<lastEltIndex+1; s++) {
-        table[s] = 1 + (table[s] >> shift);
-        sum += table[s];
+        unsigned const base = base1 ? 1 : (table[s]>0);
+        unsigned const newStat = base + (table[s] >> shift);
+        sum += newStat;
+        table[s] = newStat;
     }
     return sum;
 }
 
 /* ZSTD_scaleStats() :
- * reduce all elements in table is sum too large
+ * reduce all elt frequencies in table if sum too large
  * return the resulting sum of elements */
 static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
 {
@@ -110,7 +128,7 @@ static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
     DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
     assert(logTarget < 30);
     if (factor <= 1) return prevsum;
-    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor));
+    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor), base_1guaranteed);
 }
 
 /* ZSTD_rescaleFreqs() :
@@ -129,18 +147,22 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
     DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
     optPtr->priceType = zop_dynamic;
 
-    if (optPtr->litLengthSum == 0) {  /* first block : init */
-        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {  /* heuristic */
-            DEBUGLOG(5, "(srcSize <= ZSTD_PREDEF_THRESHOLD) => zop_predef");
+    if (optPtr->litLengthSum == 0) {  /* no literals stats collected -> first block assumed -> init */
+
+        /* heuristic: use pre-defined stats for too small inputs */
+        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {
+            DEBUGLOG(5, "srcSize <= %i : use predefined stats", ZSTD_PREDEF_THRESHOLD);
             optPtr->priceType = zop_predef;
         }
 
         assert(optPtr->symbolCosts != NULL);
         if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
-            /* huffman table presumed generated by dictionary */
+
+            /* huffman stats covering the full value set : table presumed generated by dictionary */
             optPtr->priceType = zop_dynamic;
 
             if (compressedLiterals) {
+                /* generate literals statistics from huffman table */
                 unsigned lit;
                 assert(optPtr->litFreq != NULL);
                 optPtr->litSum = 0;
@@ -188,13 +210,14 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
                     optPtr->offCodeSum += optPtr->offCodeFreq[of];
             }   }
 
-        } else {  /* not a dictionary */
+        } else {  /* first block, no dictionary */
 
             assert(optPtr->litFreq != NULL);
             if (compressedLiterals) {
+                /* base initial cost of literals on direct frequency within src */
                 unsigned lit = MaxLit;
                 HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
-                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8);
+                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8, base_0possible);
             }
 
             {   unsigned const baseLLfreqs[MaxLL+1] = {
@@ -224,10 +247,9 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
                 optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
             }
 
-
         }
 
-    } else {   /* new block : re-use previous statistics, scaled down */
+    } else {   /* new block : scale down accumulated statistics */
 
         if (compressedLiterals)
             optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
@@ -246,6 +268,7 @@ static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
                                 const optState_t* const optPtr,
                                 int optLevel)
 {
+    DEBUGLOG(8, "ZSTD_rawLiteralsCost (%u literals)", litLength);
     if (litLength == 0) return 0;
 
     if (!ZSTD_compressedLiterals(optPtr))
@@ -255,11 +278,14 @@ static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
         return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
 
     /* dynamic statistics */
-    {   U32 price = litLength * optPtr->litSumBasePrice;
+    {   U32 price = optPtr->litSumBasePrice * litLength;
+        U32 const litPriceMax = optPtr->litSumBasePrice - BITCOST_MULTIPLIER;
         U32 u;
+        assert(optPtr->litSumBasePrice >= BITCOST_MULTIPLIER);
         for (u=0; u < litLength; u++) {
-            assert(WEIGHT(optPtr->litFreq[literals[u]], optLevel) <= optPtr->litSumBasePrice);   /* literal cost should never be negative */
-            price -= WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            U32 litPrice = WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            if (UNLIKELY(litPrice > litPriceMax)) litPrice = litPriceMax;
+            price -= litPrice;
         }
         return price;
     }
@@ -272,10 +298,11 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
     assert(litLength <= ZSTD_BLOCKSIZE_MAX);
     if (optPtr->priceType == zop_predef)
         return WEIGHT(litLength, optLevel);
-    /* We can't compute the litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
-     * because it isn't representable in the zstd format. So instead just
-     * call it 1 bit more than ZSTD_BLOCKSIZE_MAX - 1. In this case the block
-     * would be all literals.
+
+    /* ZSTD_LLcode() can't compute litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
+     * because it isn't representable in the zstd format.
+     * So instead just pretend it would cost 1 bit more than ZSTD_BLOCKSIZE_MAX - 1.
+     * In such a case, the block would be all literals.
      */
     if (litLength == ZSTD_BLOCKSIZE_MAX)
         return BITCOST_MULTIPLIER + ZSTD_litLengthPrice(ZSTD_BLOCKSIZE_MAX - 1, optPtr, optLevel);
@@ -289,24 +316,25 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
 }
 
 /* ZSTD_getMatchPrice() :
- * Provides the cost of the match part (offset + matchLength) of a sequence
+ * Provides the cost of the match part (offset + matchLength) of a sequence.
  * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
- * @offcode : expects a scale where 0,1,2 are repcodes 1-3, and 3+ are real_offsets+2
+ * @offBase : sumtype, representing an offset or a repcode, and using numeric representation of ZSTD_storeSeq()
  * @optLevel: when <2, favors small offset for decompression speed (improved cache efficiency)
  */
 FORCE_INLINE_TEMPLATE U32
-ZSTD_getMatchPrice(U32 const offcode,
+ZSTD_getMatchPrice(U32 const offBase,
                    U32 const matchLength,
              const optState_t* const optPtr,
                    int const optLevel)
 {
     U32 price;
-    U32 const offCode = ZSTD_highbit32(STORED_TO_OFFBASE(offcode));
+    U32 const offCode = ZSTD_highbit32(offBase);
     U32 const mlBase = matchLength - MINMATCH;
     assert(matchLength >= MINMATCH);
 
-    if (optPtr->priceType == zop_predef)  /* fixed scheme, do not use statistics */
-        return WEIGHT(mlBase, optLevel) + ((16 + offCode) * BITCOST_MULTIPLIER);
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, does not use statistics */
+        return WEIGHT(mlBase, optLevel)
+             + ((16 + offCode) * BITCOST_MULTIPLIER); /* emulated offset cost */
 
     /* dynamic statistics */
     price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
@@ -325,10 +353,10 @@ ZSTD_getMatchPrice(U32 const offcode,
 }
 
 /* ZSTD_updateStats() :
- * assumption : literals + litLengtn <= iend */
+ * assumption : literals + litLength <= iend */
 static void ZSTD_updateStats(optState_t* const optPtr,
                              U32 litLength, const BYTE* literals,
-                             U32 offsetCode, U32 matchLength)
+                             U32 offBase, U32 matchLength)
 {
     /* literals */
     if (ZSTD_compressedLiterals(optPtr)) {
@@ -344,8 +372,8 @@ static void ZSTD_updateStats(optState_t* const optPtr,
         optPtr->litLengthSum++;
     }
 
-    /* offset code : expected to follow storeSeq() numeric representation */
-    {   U32 const offCode = ZSTD_highbit32(STORED_TO_OFFBASE(offsetCode));
+    /* offset code : follows storeSeq() numeric representation */
+    {   U32 const offCode = ZSTD_highbit32(offBase);
         assert(offCode <= MaxOff);
         optPtr->offCodeFreq[offCode]++;
         optPtr->offCodeSum++;
@@ -379,9 +407,11 @@ MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 
 /* Update hashTable3 up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
-                                              U32* nextToUpdate3,
-                                              const BYTE* const ip)
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
+                                       U32* nextToUpdate3,
+                                       const BYTE* const ip)
 {
     U32* const hashTable3 = ms->hashTable3;
     U32 const hashLog3 = ms->hashLog3;
@@ -408,7 +438,9 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
  * @param ip assumed <= iend-8 .
  * @param target The target of ZSTD_updateTree_internal() - we are filling to this position
  * @return : nb of positions added */
-static U32 ZSTD_insertBt1(
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertBt1(
                 const ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iend,
                 U32 const target,
@@ -527,6 +559,7 @@ static U32 ZSTD_insertBt1(
 }
 
 FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 void ZSTD_updateTree_internal(
                 ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iend,
@@ -535,7 +568,7 @@ void ZSTD_updateTree_internal(
     const BYTE* const base = ms->window.base;
     U32 const target = (U32)(ip - base);
     U32 idx = ms->nextToUpdate;
-    DEBUGLOG(6, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
+    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
                 idx, target, dictMode);
 
     while(idx < target) {
@@ -553,15 +586,18 @@ void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
 }
 
 FORCE_INLINE_TEMPLATE
-U32 ZSTD_insertBtAndGetAllMatches (
-                    ZSTD_match_t* matches,   /* store result (found matches) in this table (presumed large enough) */
-                    ZSTD_matchState_t* ms,
-                    U32* nextToUpdate3,
-                    const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
-                    const U32 rep[ZSTD_REP_NUM],
-                    U32 const ll0,   /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
-                    const U32 lengthToBeat,
-                    U32 const mls /* template */)
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32
+ZSTD_insertBtAndGetAllMatches (
+                ZSTD_match_t* matches,  /* store result (found matches) in this table (presumed large enough) */
+                ZSTD_matchState_t* ms,
+                U32* nextToUpdate3,
+                const BYTE* const ip, const BYTE* const iLimit,
+                const ZSTD_dictMode_e dictMode,
+                const U32 rep[ZSTD_REP_NUM],
+                const U32 ll0,  /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
+                const U32 lengthToBeat,
+                const U32 mls /* template */)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
@@ -644,7 +680,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
                             repCode, ll0, repOffset, repLen);
                 bestLength = repLen;
-                matches[mnum].off = STORE_REPCODE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
+                matches[mnum].off = REPCODE_TO_OFFBASE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
                 matches[mnum].len = (U32)repLen;
                 mnum++;
                 if ( (repLen > sufficient_len)
@@ -673,7 +709,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 bestLength = mlen;
                 assert(curr > matchIndex3);
                 assert(mnum==0);  /* no prior solution */
-                matches[0].off = STORE_OFFSET(curr - matchIndex3);
+                matches[0].off = OFFSET_TO_OFFBASE(curr - matchIndex3);
                 matches[0].len = (U32)mlen;
                 mnum = 1;
                 if ( (mlen > sufficient_len) |
@@ -706,13 +742,13 @@ U32 ZSTD_insertBtAndGetAllMatches (
         }
 
         if (matchLength > bestLength) {
-            DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
-                    (U32)matchLength, curr - matchIndex, STORE_OFFSET(curr - matchIndex));
+            DEBUGLOG(8, "found match of length %u at distance %u (offBase=%u)",
+                    (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
             assert(matchEndIdx > matchIndex);
             if (matchLength > matchEndIdx - matchIndex)
                 matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = STORE_OFFSET(curr - matchIndex);
+            matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
             matches[mnum].len = (U32)matchLength;
             mnum++;
             if ( (matchLength > ZSTD_OPT_NUM)
@@ -754,12 +790,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
             if (matchLength > bestLength) {
                 matchIndex = dictMatchIndex + dmsIndexDelta;
-                DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
-                        (U32)matchLength, curr - matchIndex, STORE_OFFSET(curr - matchIndex));
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offBase=%u)",
+                        (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
                 bestLength = matchLength;
-                matches[mnum].off = STORE_OFFSET(curr - matchIndex);
+                matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
                 matches[mnum].len = (U32)matchLength;
                 mnum++;
                 if ( (matchLength > ZSTD_OPT_NUM)
@@ -792,7 +828,9 @@ typedef U32 (*ZSTD_getAllMatchesFn)(
     U32 const ll0,
     U32 const lengthToBeat);
 
-FORCE_INLINE_TEMPLATE U32 ZSTD_btGetAllMatches_internal(
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_btGetAllMatches_internal(
         ZSTD_match_t* matches,
         ZSTD_matchState_t* ms,
         U32* nextToUpdate3,
@@ -960,7 +998,7 @@ static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
                                       const ZSTD_optLdm_t* optLdm, U32 currPosInBlock)
 {
     U32 const posDiff = currPosInBlock - optLdm->startPosInBlock;
-    /* Note: ZSTD_match_t actually contains offCode and matchLength (before subtracting MINMATCH) */
+    /* Note: ZSTD_match_t actually contains offBase and matchLength (before subtracting MINMATCH) */
     U32 const candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
 
     /* Ensure that current block position is not outside of the match */
@@ -971,11 +1009,11 @@ static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
     }
 
     if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
-        U32 const candidateOffCode = STORE_OFFSET(optLdm->offset);
-        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offCode: %u matchLength %u) at block position=%u",
-                 candidateOffCode, candidateMatchLength, currPosInBlock);
+        U32 const candidateOffBase = OFFSET_TO_OFFBASE(optLdm->offset);
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offBase: %u matchLength %u) at block position=%u",
+                 candidateOffBase, candidateMatchLength, currPosInBlock);
         matches[*nbMatches].len = candidateMatchLength;
-        matches[*nbMatches].off = candidateOffCode;
+        matches[*nbMatches].off = candidateOffBase;
         (*nbMatches)++;
     }
 }
@@ -1011,11 +1049,6 @@ ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm,
 *  Optimal parser
 *********************************/
 
-static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
-{
-    return sol.litlen + sol.mlen;
-}
-
 #if 0 /* debug */
 
 static void
@@ -1033,7 +1066,13 @@ listStats(const U32* table, int lastEltID)
 
 #endif
 
-FORCE_INLINE_TEMPLATE size_t
+#define LIT_PRICE(_p) (int)ZSTD_rawLiteralsCost(_p, 1, optStatePtr, optLevel)
+#define LL_PRICE(_l) (int)ZSTD_litLengthPrice(_l, optStatePtr, optLevel)
+#define LL_INCPRICE(_l) (LL_PRICE(_l) - LL_PRICE(_l-1))
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t
 ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                                seqStore_t* seqStore,
                                U32 rep[ZSTD_REP_NUM],
@@ -1059,9 +1098,11 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
 
     ZSTD_optimal_t* const opt = optStatePtr->priceTable;
     ZSTD_match_t* const matches = optStatePtr->matchTable;
-    ZSTD_optimal_t lastSequence;
+    ZSTD_optimal_t lastStretch;
     ZSTD_optLdm_t optLdm;
 
+    ZSTD_memset(&lastStretch, 0, sizeof(ZSTD_optimal_t));
+
     optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
     optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
     ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
@@ -1082,103 +1123,139 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
             U32 const ll0 = !litlen;
             U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, ip, iend, rep, ll0, minMatch);
             ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
-                                              (U32)(ip-istart), (U32)(iend - ip));
-            if (!nbMatches) { ip++; continue; }
+                                              (U32)(ip-istart), (U32)(iend-ip));
+            if (!nbMatches) {
+                DEBUGLOG(8, "no match found at cPos %u", (unsigned)(ip-istart));
+                ip++;
+                continue;
+            }
+
+            /* Match found: let's store this solution, and eventually find more candidates.
+             * During this forward pass, @opt is used to store stretches,
+             * defined as "a match followed by N literals".
+             * Note how this is different from a Sequence, which is "N literals followed by a match".
+             * Storing stretches allows us to store different match predecessors
+             * for each literal position part of a literals run. */
 
             /* initialize opt[0] */
-            { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-            opt[0].mlen = 0;  /* means is_a_literal */
+            opt[0].mlen = 0;  /* there are only literals so far */
             opt[0].litlen = litlen;
-            /* We don't need to include the actual price of the literals because
-             * it is static for the duration of the forward pass, and is included
-             * in every price. We include the literal length to avoid negative
-             * prices when we subtract the previous literal length.
+            /* No need to include the actual price of the literals before the first match
+             * because it is static for the duration of the forward pass, and is included
+             * in every subsequent price. But, we include the literal length because
+             * the cost variation of litlen depends on the value of litlen.
              */
-            opt[0].price = (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
+            opt[0].price = LL_PRICE(litlen);
+            ZSTD_STATIC_ASSERT(sizeof(opt[0].rep[0]) == sizeof(rep[0]));
+            ZSTD_memcpy(&opt[0].rep, rep, sizeof(opt[0].rep));
 
             /* large match -> immediate encoding */
             {   U32 const maxML = matches[nbMatches-1].len;
-                U32 const maxOffcode = matches[nbMatches-1].off;
-                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffCode=%u at cPos=%u => start new series",
-                            nbMatches, maxML, maxOffcode, (U32)(ip-prefixStart));
+                U32 const maxOffBase = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffBase=%u at cPos=%u => start new series",
+                            nbMatches, maxML, maxOffBase, (U32)(ip-prefixStart));
 
                 if (maxML > sufficient_len) {
-                    lastSequence.litlen = litlen;
-                    lastSequence.mlen = maxML;
-                    lastSequence.off = maxOffcode;
-                    DEBUGLOG(6, "large match (%u>%u), immediate encoding",
+                    lastStretch.litlen = 0;
+                    lastStretch.mlen = maxML;
+                    lastStretch.off = maxOffBase;
+                    DEBUGLOG(6, "large match (%u>%u) => immediate encoding",
                                 maxML, sufficient_len);
                     cur = 0;
-                    last_pos = ZSTD_totalLen(lastSequence);
+                    last_pos = maxML;
                     goto _shortestPath;
             }   }
 
             /* set prices for first matches starting position == 0 */
             assert(opt[0].price >= 0);
-            {   U32 const literalsPrice = (U32)opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
-                U32 pos;
+            {   U32 pos;
                 U32 matchNb;
                 for (pos = 1; pos < minMatch; pos++) {
-                    opt[pos].price = ZSTD_MAX_PRICE;   /* mlen, litlen and price will be fixed during forward scanning */
+                    opt[pos].price = ZSTD_MAX_PRICE;
+                    opt[pos].mlen = 0;
+                    opt[pos].litlen = litlen + pos;
                 }
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
-                    U32 const offcode = matches[matchNb].off;
+                    U32 const offBase = matches[matchNb].off;
                     U32 const end = matches[matchNb].len;
                     for ( ; pos <= end ; pos++ ) {
-                        U32 const matchPrice = ZSTD_getMatchPrice(offcode, pos, optStatePtr, optLevel);
-                        U32 const sequencePrice = literalsPrice + matchPrice;
+                        int const matchPrice = (int)ZSTD_getMatchPrice(offBase, pos, optStatePtr, optLevel);
+                        int const sequencePrice = opt[0].price + matchPrice;
                         DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
                                     pos, ZSTD_fCost(sequencePrice));
                         opt[pos].mlen = pos;
-                        opt[pos].off = offcode;
-                        opt[pos].litlen = litlen;
-                        opt[pos].price = (int)sequencePrice;
-                }   }
+                        opt[pos].off = offBase;
+                        opt[pos].litlen = 0; /* end of match */
+                        opt[pos].price = sequencePrice + LL_PRICE(0);
+                    }
+                }
                 last_pos = pos-1;
+                opt[pos].price = ZSTD_MAX_PRICE;
             }
         }
 
         /* check further positions */
         for (cur = 1; cur <= last_pos; cur++) {
             const BYTE* const inr = ip + cur;
-            assert(cur < ZSTD_OPT_NUM);
-            DEBUGLOG(7, "cPos:%zi==rPos:%u", inr-istart, cur)
+            assert(cur <= ZSTD_OPT_NUM);
+            DEBUGLOG(7, "cPos:%zi==rPos:%u", inr-istart, cur);
 
             /* Fix current position with one literal if cheaper */
-            {   U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
+            {   U32 const litlen = opt[cur-1].litlen + 1;
                 int const price = opt[cur-1].price
-                                + (int)ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
-                                + (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
-                                - (int)ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
+                                + LIT_PRICE(ip+cur-1)
+                                + LL_INCPRICE(litlen);
                 assert(price < 1000000000); /* overflow check */
                 if (price <= opt[cur].price) {
+                    ZSTD_optimal_t const prevMatch = opt[cur];
                     DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
                                 inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
                                 opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
-                    opt[cur].mlen = 0;
-                    opt[cur].off = 0;
+                    opt[cur] = opt[cur-1];
                     opt[cur].litlen = litlen;
                     opt[cur].price = price;
+                    if ( (optLevel >= 1) /* additional check only for higher modes */
+                      && (prevMatch.litlen == 0) /* replace a match */
+                      && (LL_INCPRICE(1) < 0) /* ll1 is cheaper than ll0 */
+                      && LIKELY(ip + cur < iend)
+                    ) {
+                        /* check next position, in case it would be cheaper */
+                        int with1literal = prevMatch.price + LIT_PRICE(ip+cur) + LL_INCPRICE(1);
+                        int withMoreLiterals = price + LIT_PRICE(ip+cur) + LL_INCPRICE(litlen+1);
+                        DEBUGLOG(7, "then at next rPos %u : match+1lit %.2f vs %ulits %.2f",
+                                cur+1, ZSTD_fCost(with1literal), litlen+1, ZSTD_fCost(withMoreLiterals));
+                        if ( (with1literal < withMoreLiterals)
+                          && (with1literal < opt[cur+1].price) ) {
+                            /* update offset history - before it disappears */
+                            U32 const prev = cur - prevMatch.mlen;
+                            repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, prevMatch.off, opt[prev].litlen==0);
+                            assert(cur >= prevMatch.mlen);
+                            DEBUGLOG(7, "==> match+1lit is cheaper (%.2f < %.2f) (hist:%u,%u,%u) !",
+                                        ZSTD_fCost(with1literal), ZSTD_fCost(withMoreLiterals),
+                                        newReps.rep[0], newReps.rep[1], newReps.rep[2] );
+                            opt[cur+1] = prevMatch;  /* mlen & offbase */
+                            ZSTD_memcpy(opt[cur+1].rep, &newReps, sizeof(repcodes_t));
+                            opt[cur+1].litlen = 1;
+                            opt[cur+1].price = with1literal;
+                            if (last_pos < cur+1) last_pos = cur+1;
+                        }
+                    }
                 } else {
-                    DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f) (hist:%u,%u,%u)",
-                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price),
-                                opt[cur].rep[0], opt[cur].rep[1], opt[cur].rep[2]);
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price));
                 }
             }
 
-            /* Set the repcodes of the current position. We must do it here
-             * because we rely on the repcodes of the 2nd to last sequence being
-             * correct to set the next chunks repcodes during the backward
-             * traversal.
+            /* Offset history is not updated during match comparison.
+             * Do it here, now that the match is selected and confirmed.
              */
             ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(repcodes_t));
             assert(cur >= opt[cur].mlen);
-            if (opt[cur].mlen != 0) {
+            if (opt[cur].litlen == 0) {
+                /* just finished a match => alter offset history */
                 U32 const prev = cur - opt[cur].mlen;
-                repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
+                repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[prev].litlen==0);
                 ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
-            } else {
-                ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
             }
 
             /* last match must start at a minimum distance of 8 from oend */
@@ -1188,15 +1265,14 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
 
             if ( (optLevel==0) /*static_test*/
               && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
-                DEBUGLOG(7, "move to next rPos:%u : price is <=", cur+1);
+                DEBUGLOG(7, "skip current position : next rPos(%u) price is cheaper", cur+1);
                 continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
             }
 
             assert(opt[cur].price >= 0);
-            {   U32 const ll0 = (opt[cur].mlen != 0);
-                U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
-                U32 const previousPrice = (U32)opt[cur].price;
-                U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+            {   U32 const ll0 = (opt[cur].litlen == 0);
+                int const previousPrice = opt[cur].price;
+                int const basePrice = previousPrice + LL_PRICE(0);
                 U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, inr, iend, opt[cur].rep, ll0, minMatch);
                 U32 matchNb;
 
@@ -1208,18 +1284,17 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     continue;
                 }
 
-                {   U32 const maxML = matches[nbMatches-1].len;
-                    DEBUGLOG(7, "cPos:%zi==rPos:%u, found %u matches, of maxLength=%u",
-                                inr-istart, cur, nbMatches, maxML);
-
-                    if ( (maxML > sufficient_len)
-                      || (cur + maxML >= ZSTD_OPT_NUM) ) {
-                        lastSequence.mlen = maxML;
-                        lastSequence.off = matches[nbMatches-1].off;
-                        lastSequence.litlen = litlen;
-                        cur -= (opt[cur].mlen==0) ? opt[cur].litlen : 0;  /* last sequence is actually only literals, fix cur to last match - note : may underflow, in which case, it's first sequence, and it's okay */
-                        last_pos = cur + ZSTD_totalLen(lastSequence);
-                        if (cur > ZSTD_OPT_NUM) cur = 0;   /* underflow => first match */
+                {   U32 const longestML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u, found %u matches, of longest ML=%u",
+                                inr-istart, cur, nbMatches, longestML);
+
+                    if ( (longestML > sufficient_len)
+                      || (cur + longestML >= ZSTD_OPT_NUM)
+                      || (ip + cur + longestML >= iend) ) {
+                        lastStretch.mlen = longestML;
+                        lastStretch.off = matches[nbMatches-1].off;
+                        lastStretch.litlen = 0;
+                        last_pos = cur + longestML;
                         goto _shortestPath;
                 }   }
 
@@ -1230,20 +1305,25 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
                     U32 mlen;
 
-                    DEBUGLOG(7, "testing match %u => offCode=%4u, mlen=%2u, llen=%2u",
-                                matchNb, matches[matchNb].off, lastML, litlen);
+                    DEBUGLOG(7, "testing match %u => offBase=%4u, mlen=%2u, llen=%2u",
+                                matchNb, matches[matchNb].off, lastML, opt[cur].litlen);
 
                     for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
                         U32 const pos = cur + mlen;
-                        int const price = (int)basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+                        int const price = basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
 
                         if ((pos > last_pos) || (price < opt[pos].price)) {
                             DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
                                         pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
-                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }   /* fill empty positions */
+                            while (last_pos < pos) {
+                                /* fill empty positions, for future comparisons */
+                                last_pos++;
+                                opt[last_pos].price = ZSTD_MAX_PRICE;
+                                opt[last_pos].litlen = !0;  /* just needs to be != 0, to mean "not an end of match" */
+                            }
                             opt[pos].mlen = mlen;
                             opt[pos].off = offset;
-                            opt[pos].litlen = litlen;
+                            opt[pos].litlen = 0;
                             opt[pos].price = price;
                         } else {
                             DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
@@ -1251,52 +1331,86 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                             if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
                         }
             }   }   }
+            opt[last_pos+1].price = ZSTD_MAX_PRICE;
         }  /* for (cur = 1; cur <= last_pos; cur++) */
 
-        lastSequence = opt[last_pos];
-        cur = last_pos > ZSTD_totalLen(lastSequence) ? last_pos - ZSTD_totalLen(lastSequence) : 0;  /* single sequence, and it starts before `ip` */
-        assert(cur < ZSTD_OPT_NUM);  /* control overflow*/
+        lastStretch = opt[last_pos];
+        assert(cur >= lastStretch.mlen);
+        cur = last_pos - lastStretch.mlen;
 
 _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
         assert(opt[0].mlen == 0);
+        assert(last_pos >= lastStretch.mlen);
+        assert(cur == last_pos - lastStretch.mlen);
 
-        /* Set the next chunk's repcodes based on the repcodes of the beginning
-         * of the last match, and the last sequence. This avoids us having to
-         * update them while traversing the sequences.
-         */
-        if (lastSequence.mlen != 0) {
-            repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
-            ZSTD_memcpy(rep, &reps, sizeof(reps));
+        if (lastStretch.mlen==0) {
+            /* no solution : all matches have been converted into literals */
+            assert(lastStretch.litlen == (ip - anchor) + last_pos);
+            ip += last_pos;
+            continue;
+        }
+        assert(lastStretch.off > 0);
+
+        /* Update offset history */
+        if (lastStretch.litlen == 0) {
+            /* finishing on a match : update offset history */
+            repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastStretch.off, opt[cur].litlen==0);
+            ZSTD_memcpy(rep, &reps, sizeof(repcodes_t));
         } else {
-            ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
+            ZSTD_memcpy(rep, lastStretch.rep, sizeof(repcodes_t));
+            assert(cur >= lastStretch.litlen);
+            cur -= lastStretch.litlen;
         }
 
-        {   U32 const storeEnd = cur + 1;
+        /* Let's write the shortest path solution.
+         * It is stored in @opt in reverse order,
+         * starting from @storeEnd (==cur+2),
+         * effectively partially @opt overwriting.
+         * Content is changed too:
+         * - So far, @opt stored stretches, aka a match followed by literals
+         * - Now, it will store sequences, aka literals followed by a match
+         */
+        {   U32 const storeEnd = cur + 2;
             U32 storeStart = storeEnd;
-            U32 seqPos = cur;
+            U32 stretchPos = cur;
 
             DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
                         last_pos, cur); (void)last_pos;
-            assert(storeEnd < ZSTD_OPT_NUM);
-            DEBUGLOG(6, "last sequence copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
-                        storeEnd, lastSequence.litlen, lastSequence.mlen, lastSequence.off);
-            opt[storeEnd] = lastSequence;
-            while (seqPos > 0) {
-                U32 const backDist = ZSTD_totalLen(opt[seqPos]);
+            assert(storeEnd < ZSTD_OPT_SIZE);
+            DEBUGLOG(6, "last stretch copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                        storeEnd, lastStretch.litlen, lastStretch.mlen, lastStretch.off);
+            if (lastStretch.litlen > 0) {
+                /* last "sequence" is unfinished: just a bunch of literals */
+                opt[storeEnd].litlen = lastStretch.litlen;
+                opt[storeEnd].mlen = 0;
+                storeStart = storeEnd-1;
+                opt[storeStart] = lastStretch;
+            } {
+                opt[storeEnd] = lastStretch;  /* note: litlen will be fixed */
+                storeStart = storeEnd;
+            }
+            while (1) {
+                ZSTD_optimal_t nextStretch = opt[stretchPos];
+                opt[storeStart].litlen = nextStretch.litlen;
+                DEBUGLOG(6, "selected sequence (llen=%u,mlen=%u,ofc=%u)",
+                            opt[storeStart].litlen, opt[storeStart].mlen, opt[storeStart].off);
+                if (nextStretch.mlen == 0) {
+                    /* reaching beginning of segment */
+                    break;
+                }
                 storeStart--;
-                DEBUGLOG(6, "sequence from rPos=%u copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
-                            seqPos, storeStart, opt[seqPos].litlen, opt[seqPos].mlen, opt[seqPos].off);
-                opt[storeStart] = opt[seqPos];
-                seqPos = (seqPos > backDist) ? seqPos - backDist : 0;
+                opt[storeStart] = nextStretch; /* note: litlen will be fixed */
+                assert(nextStretch.litlen + nextStretch.mlen <= stretchPos);
+                stretchPos -= nextStretch.litlen + nextStretch.mlen;
             }
 
             /* save sequences */
-            DEBUGLOG(6, "sending selected sequences into seqStore")
+            DEBUGLOG(6, "sending selected sequences into seqStore");
             {   U32 storePos;
                 for (storePos=storeStart; storePos <= storeEnd; storePos++) {
                     U32 const llen = opt[storePos].litlen;
                     U32 const mlen = opt[storePos].mlen;
-                    U32 const offCode = opt[storePos].off;
+                    U32 const offBase = opt[storePos].off;
                     U32 const advance = llen + mlen;
                     DEBUGLOG(6, "considering seq starting at %zi, llen=%u, mlen=%u",
                                 anchor - istart, (unsigned)llen, (unsigned)mlen);
@@ -1308,11 +1422,14 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     }
 
                     assert(anchor + llen <= iend);
-                    ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
-                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offCode, mlen);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offBase, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offBase, mlen);
                     anchor += advance;
                     ip = anchor;
             }   }
+            DEBUGLOG(7, "new offset history : %u, %u, %u", rep[0], rep[1], rep[2]);
+
+            /* update all costs */
             ZSTD_setBasePrices(optStatePtr, optLevel);
         }
     }   /* while (ip < ilimit) */
@@ -1320,21 +1437,27 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
     /* Return the last literals size */
     return (size_t)(iend - anchor);
 }
+#endif /* build exclusions */
 
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
 static size_t ZSTD_compressBlock_opt0(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
 {
     return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /* optLevel */, dictMode);
 }
+#endif
 
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
 static size_t ZSTD_compressBlock_opt2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
 {
     return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /* optLevel */, dictMode);
 }
+#endif
 
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
 size_t ZSTD_compressBlock_btopt(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
@@ -1342,20 +1465,23 @@ size_t ZSTD_compressBlock_btopt(
     DEBUGLOG(5, "ZSTD_compressBlock_btopt");
     return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
 }
+#endif
 
 
 
 
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
 /* ZSTD_initStats_ultra():
  * make a first compression pass, just to seed stats with more accurate starting values.
  * only works on first block, with no dictionary and no ldm.
- * this function cannot error, hence its contract must be respected.
+ * this function cannot error out, its narrow contract must be respected.
  */
-static void
-ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
-                     seqStore_t* seqStore,
-                     U32 rep[ZSTD_REP_NUM],
-               const void* src, size_t srcSize)
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
+                          seqStore_t* seqStore,
+                          U32 rep[ZSTD_REP_NUM],
+                    const void* src, size_t srcSize)
 {
     U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
     ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
@@ -1368,7 +1494,7 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
 
     ZSTD_compressBlock_opt2(ms, seqStore, tmpRep, src, srcSize, ZSTD_noDict);   /* generate stats into ms->opt*/
 
-    /* invalidate first scan from history */
+    /* invalidate first scan from history, only keep entropy stats */
     ZSTD_resetSeqStore(seqStore);
     ms->window.base -= srcSize;
     ms->window.dictLimit += (U32)srcSize;
@@ -1392,10 +1518,10 @@ size_t ZSTD_compressBlock_btultra2(
     U32 const curr = (U32)((const BYTE*)src - ms->window.base);
     DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
 
-    /* 2-pass strategy:
+    /* 2-passes strategy:
      * this strategy makes a first pass over first block to collect statistics
-     * and seed next round's statistics with it.
-     * After 1st pass, function forgets everything, and starts a new block.
+     * in order to seed next round's statistics with it.
+     * After 1st pass, function forgets history, and starts a new block.
      * Consequently, this can only work if no data has been previously loaded in tables,
      * aka, no dictionary, no prefix, no ldm preprocessing.
      * The compression ratio gain is generally small (~0.5% on first block),
@@ -1404,15 +1530,17 @@ size_t ZSTD_compressBlock_btultra2(
     if ( (ms->opt.litLengthSum==0)   /* first block */
       && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
       && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
-      && (curr == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
-      && (srcSize > ZSTD_PREDEF_THRESHOLD)
+      && (curr == ms->window.dictLimit)    /* start of frame, nothing already loaded nor skipped */
+      && (srcSize > ZSTD_PREDEF_THRESHOLD) /* input large enough to not employ default stats */
       ) {
         ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
     }
 
     return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
 }
+#endif
 
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
 size_t ZSTD_compressBlock_btopt_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
@@ -1420,18 +1548,20 @@ size_t ZSTD_compressBlock_btopt_dictMatchState(
     return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_btultra_dictMatchState(
+size_t ZSTD_compressBlock_btopt_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
 }
+#endif
 
-size_t ZSTD_compressBlock_btopt_extDict(
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btultra_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
 }
 
 size_t ZSTD_compressBlock_btultra_extDict(
@@ -1440,6 +1570,7 @@ size_t ZSTD_compressBlock_btultra_extDict(
 {
     return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
 }
+#endif
 
 /* note : no btultra2 variant for extDict nor dictMatchState,
  * because btultra2 is not meant to work with dictionaries
diff --git a/lib/zstd/compress/zstd_opt.h b/lib/zstd/compress/zstd_opt.h
index 22b862858ba7..ac1b743d27cd 100644
--- a/lib/zstd/compress/zstd_opt.h
+++ b/lib/zstd/compress/zstd_opt.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,30 +15,40 @@
 
 #include "zstd_compress_internal.h"
 
+#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
 /* used in ZSTD_loadDictionaryContent() */
 void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend);
+#endif
 
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
 size_t ZSTD_compressBlock_btopt(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_btultra(
+size_t ZSTD_compressBlock_btopt_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-size_t ZSTD_compressBlock_btultra2(
+size_t ZSTD_compressBlock_btopt_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 
+#define ZSTD_COMPRESSBLOCK_BTOPT ZSTD_compressBlock_btopt
+#define ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE ZSTD_compressBlock_btopt_dictMatchState
+#define ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT ZSTD_compressBlock_btopt_extDict
+#else
+#define ZSTD_COMPRESSBLOCK_BTOPT NULL
+#define ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT NULL
+#endif
 
-size_t ZSTD_compressBlock_btopt_dictMatchState(
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btultra(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btultra_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-
-size_t ZSTD_compressBlock_btopt_extDict(
-        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btultra_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
@@ -45,6 +56,20 @@ size_t ZSTD_compressBlock_btultra_extDict(
         /* note : no btultra2 variant for extDict nor dictMatchState,
          * because btultra2 is not meant to work with dictionaries
          * and is only specific for the first block (no prefix) */
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+#define ZSTD_COMPRESSBLOCK_BTULTRA ZSTD_compressBlock_btultra
+#define ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE ZSTD_compressBlock_btultra_dictMatchState
+#define ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT ZSTD_compressBlock_btultra_extDict
+#define ZSTD_COMPRESSBLOCK_BTULTRA2 ZSTD_compressBlock_btultra2
+#else
+#define ZSTD_COMPRESSBLOCK_BTULTRA NULL
+#define ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE NULL
+#define ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT NULL
+#define ZSTD_COMPRESSBLOCK_BTULTRA2 NULL
+#endif
 
 
 #endif /* ZSTD_OPT_H */
diff --git a/lib/zstd/decompress/huf_decompress.c b/lib/zstd/decompress/huf_decompress.c
index 60958afebc41..ac8b87f48f84 100644
--- a/lib/zstd/decompress/huf_decompress.c
+++ b/lib/zstd/decompress/huf_decompress.c
@@ -1,7 +1,8 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /* ******************************************************************
  * huff0 huffman decoder,
  * part of Finite State Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -19,10 +20,10 @@
 #include "../common/compiler.h"
 #include "../common/bitstream.h"  /* BIT_* */
 #include "../common/fse.h"        /* to compress headers */
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/error_private.h"
 #include "../common/zstd_internal.h"
+#include "../common/bits.h"       /* ZSTD_highbit32, ZSTD_countTrailingZeros64 */
 
 /* **************************************************************
 *  Constants
@@ -34,6 +35,12 @@
 *  Macros
 ****************************************************************/
 
+#ifdef HUF_DISABLE_FAST_DECODE
+# define HUF_ENABLE_FAST_DECODE 0
+#else
+# define HUF_ENABLE_FAST_DECODE 1
+#endif
+
 /* These two optional macros force the use one way or another of the two
  * Huffman decompression implementations. You can't force in both directions
  * at the same time.
@@ -43,27 +50,25 @@
 #error "Cannot force the use of the X1 and X2 decoders at the same time!"
 #endif
 
-#if ZSTD_ENABLE_ASM_X86_64_BMI2 && DYNAMIC_BMI2
-# define HUF_ASM_X86_64_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
+/* When DYNAMIC_BMI2 is enabled, fast decoders are only called when bmi2 is
+ * supported at runtime, so we can add the BMI2 target attribute.
+ * When it is disabled, we will still get BMI2 if it is enabled statically.
+ */
+#if DYNAMIC_BMI2
+# define HUF_FAST_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
 #else
-# define HUF_ASM_X86_64_BMI2_ATTRS
+# define HUF_FAST_BMI2_ATTRS
 #endif
 
 #define HUF_EXTERN_C
 #define HUF_ASM_DECL HUF_EXTERN_C
 
-#if DYNAMIC_BMI2 || (ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__))
+#if DYNAMIC_BMI2
 # define HUF_NEED_BMI2_FUNCTION 1
 #else
 # define HUF_NEED_BMI2_FUNCTION 0
 #endif
 
-#if !(ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__))
-# define HUF_NEED_DEFAULT_FUNCTION 1
-#else
-# define HUF_NEED_DEFAULT_FUNCTION 0
-#endif
-
 /* **************************************************************
 *  Error Management
 ****************************************************************/
@@ -80,6 +85,11 @@
 /* **************************************************************
 *  BMI2 Variant Wrappers
 ****************************************************************/
+typedef size_t (*HUF_DecompressUsingDTableFn)(void *dst, size_t dstSize,
+                                              const void *cSrc,
+                                              size_t cSrcSize,
+                                              const HUF_DTable *DTable);
+
 #if DYNAMIC_BMI2
 
 #define HUF_DGEN(fn)                                                        \
@@ -101,9 +111,9 @@
     }                                                                       \
                                                                             \
     static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
-                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
     {                                                                       \
-        if (bmi2) {                                                         \
+        if (flags & HUF_flags_bmi2) {                                       \
             return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
         }                                                                   \
         return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
@@ -113,9 +123,9 @@
 
 #define HUF_DGEN(fn)                                                        \
     static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
-                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
     {                                                                       \
-        (void)bmi2;                                                         \
+        (void)flags;                                                        \
         return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
     }
 
@@ -134,43 +144,66 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
     return dtd;
 }
 
-#if ZSTD_ENABLE_ASM_X86_64_BMI2
-
-static size_t HUF_initDStream(BYTE const* ip) {
+static size_t HUF_initFastDStream(BYTE const* ip) {
     BYTE const lastByte = ip[7];
-    size_t const bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+    size_t const bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
     size_t const value = MEM_readLEST(ip) | 1;
     assert(bitsConsumed <= 8);
+    assert(sizeof(size_t) == 8);
     return value << bitsConsumed;
 }
+
+
+/*
+ * The input/output arguments to the Huffman fast decoding loop:
+ *
+ * ip [in/out] - The input pointers, must be updated to reflect what is consumed.
+ * op [in/out] - The output pointers, must be updated to reflect what is written.
+ * bits [in/out] - The bitstream containers, must be updated to reflect the current state.
+ * dt [in] - The decoding table.
+ * ilowest [in] - The beginning of the valid range of the input. Decoders may read
+ *                down to this pointer. It may be below iend[0].
+ * oend [in] - The end of the output stream. op[3] must not cross oend.
+ * iend [in] - The end of each input stream. ip[i] may cross iend[i],
+ *             as long as it is above ilowest, but that indicates corruption.
+ */
 typedef struct {
     BYTE const* ip[4];
     BYTE* op[4];
     U64 bits[4];
     void const* dt;
-    BYTE const* ilimit;
+    BYTE const* ilowest;
     BYTE* oend;
     BYTE const* iend[4];
-} HUF_DecompressAsmArgs;
+} HUF_DecompressFastArgs;
+
+typedef void (*HUF_DecompressFastLoopFn)(HUF_DecompressFastArgs*);
 
 /*
- * Initializes args for the asm decoding loop.
- * @returns 0 on success
- *          1 if the fallback implementation should be used.
+ * Initializes args for the fast decoding loop.
+ * @returns 1 on success
+ *          0 if the fallback implementation should be used.
  *          Or an error code on failure.
  */
-static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
+static size_t HUF_DecompressFastArgs_init(HUF_DecompressFastArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
 {
     void const* dt = DTable + 1;
     U32 const dtLog = HUF_getDTableDesc(DTable).tableLog;
 
-    const BYTE* const ilimit = (const BYTE*)src + 6 + 8;
+    const BYTE* const istart = (const BYTE*)src;
 
-    BYTE* const oend = (BYTE*)dst + dstSize;
+    BYTE* const oend = ZSTD_maybeNullPtrAdd((BYTE*)dst, dstSize);
 
-    /* The following condition is false on x32 platform,
-     * but HUF_asm is not compatible with this ABI */
-    if (!(MEM_isLittleEndian() && !MEM_32bits())) return 1;
+    /* The fast decoding loop assumes 64-bit little-endian.
+     * This condition is false on x32.
+     */
+    if (!MEM_isLittleEndian() || MEM_32bits())
+        return 0;
+
+    /* Avoid nullptr addition */
+    if (dstSize == 0)
+        return 0;
+    assert(dst != NULL);
 
     /* strict minimum : jump table + 1 byte per stream */
     if (srcSize < 10)
@@ -181,11 +214,10 @@ static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst,
      * On small inputs we don't have enough data to trigger the fast loop, so use the old decoder.
      */
     if (dtLog != HUF_DECODER_FAST_TABLELOG)
-        return 1;
+        return 0;
 
     /* Read the jump table. */
     {
-        const BYTE* const istart = (const BYTE*)src;
         size_t const length1 = MEM_readLE16(istart);
         size_t const length2 = MEM_readLE16(istart+2);
         size_t const length3 = MEM_readLE16(istart+4);
@@ -195,13 +227,11 @@ static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst,
         args->iend[2] = args->iend[1] + length2;
         args->iend[3] = args->iend[2] + length3;
 
-        /* HUF_initDStream() requires this, and this small of an input
+        /* HUF_initFastDStream() requires this, and this small of an input
          * won't benefit from the ASM loop anyways.
-         * length1 must be >= 16 so that ip[0] >= ilimit before the loop
-         * starts.
          */
-        if (length1 < 16 || length2 < 8 || length3 < 8 || length4 < 8)
-            return 1;
+        if (length1 < 8 || length2 < 8 || length3 < 8 || length4 < 8)
+            return 0;
         if (length4 > srcSize) return ERROR(corruption_detected);   /* overflow */
     }
     /* ip[] contains the position that is currently loaded into bits[]. */
@@ -218,7 +248,7 @@ static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst,
 
     /* No point to call the ASM loop for tiny outputs. */
     if (args->op[3] >= oend)
-        return 1;
+        return 0;
 
     /* bits[] is the bit container.
         * It is read from the MSB down to the LSB.
@@ -227,24 +257,25 @@ static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst,
         * set, so that CountTrailingZeros(bits[]) can be used
         * to count how many bits we've consumed.
         */
-    args->bits[0] = HUF_initDStream(args->ip[0]);
-    args->bits[1] = HUF_initDStream(args->ip[1]);
-    args->bits[2] = HUF_initDStream(args->ip[2]);
-    args->bits[3] = HUF_initDStream(args->ip[3]);
-
-    /* If ip[] >= ilimit, it is guaranteed to be safe to
-        * reload bits[]. It may be beyond its section, but is
-        * guaranteed to be valid (>= istart).
-        */
-    args->ilimit = ilimit;
+    args->bits[0] = HUF_initFastDStream(args->ip[0]);
+    args->bits[1] = HUF_initFastDStream(args->ip[1]);
+    args->bits[2] = HUF_initFastDStream(args->ip[2]);
+    args->bits[3] = HUF_initFastDStream(args->ip[3]);
+
+    /* The decoders must be sure to never read beyond ilowest.
+     * This is lower than iend[0], but allowing decoders to read
+     * down to ilowest can allow an extra iteration or two in the
+     * fast loop.
+     */
+    args->ilowest = istart;
 
     args->oend = oend;
     args->dt = dt;
 
-    return 0;
+    return 1;
 }
 
-static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressAsmArgs const* args, int stream, BYTE* segmentEnd)
+static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressFastArgs const* args, int stream, BYTE* segmentEnd)
 {
     /* Validate that we haven't overwritten. */
     if (args->op[stream] > segmentEnd)
@@ -258,15 +289,33 @@ static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressAsmArgs
         return ERROR(corruption_detected);
 
     /* Construct the BIT_DStream_t. */
-    bit->bitContainer = MEM_readLE64(args->ip[stream]);
-    bit->bitsConsumed = ZSTD_countTrailingZeros((size_t)args->bits[stream]);
-    bit->start = (const char*)args->iend[0];
+    assert(sizeof(size_t) == 8);
+    bit->bitContainer = MEM_readLEST(args->ip[stream]);
+    bit->bitsConsumed = ZSTD_countTrailingZeros64(args->bits[stream]);
+    bit->start = (const char*)args->ilowest;
     bit->limitPtr = bit->start + sizeof(size_t);
     bit->ptr = (const char*)args->ip[stream];
 
     return 0;
 }
-#endif
+
+/* Calls X(N) for each stream 0, 1, 2, 3. */
+#define HUF_4X_FOR_EACH_STREAM(X) \
+    do {                          \
+        X(0);                     \
+        X(1);                     \
+        X(2);                     \
+        X(3);                     \
+    } while (0)
+
+/* Calls X(N, var) for each stream 0, 1, 2, 3. */
+#define HUF_4X_FOR_EACH_STREAM_WITH_VAR(X, var) \
+    do {                                        \
+        X(0, (var));                            \
+        X(1, (var));                            \
+        X(2, (var));                            \
+        X(3, (var));                            \
+    } while (0)
 
 
 #ifndef HUF_FORCE_DECOMPRESS_X2
@@ -283,10 +332,11 @@ typedef struct { BYTE nbBits; BYTE byte; } HUF_DEltX1;   /* single-symbol decodi
 static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
     U64 D4;
     if (MEM_isLittleEndian()) {
-        D4 = (symbol << 8) + nbBits;
+        D4 = (U64)((symbol << 8) + nbBits);
     } else {
-        D4 = symbol + (nbBits << 8);
+        D4 = (U64)(symbol + (nbBits << 8));
     }
+    assert(D4 < (1U << 16));
     D4 *= 0x0001000100010001ULL;
     return D4;
 }
@@ -329,13 +379,7 @@ typedef struct {
         BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
 } HUF_ReadDTableX1_Workspace;
 
-
-size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
-{
-    return HUF_readDTableX1_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
-}
-
-size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags)
 {
     U32 tableLog = 0;
     U32 nbSymbols = 0;
@@ -350,7 +394,7 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
     DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), bmi2);
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), flags);
     if (HUF_isError(iSize)) return iSize;
 
 
@@ -377,9 +421,8 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
      * rankStart[0] is not filled because there are no entries in the table for
      * weight 0.
      */
-    {
-        int n;
-        int nextRankStart = 0;
+    {   int n;
+        U32 nextRankStart = 0;
         int const unroll = 4;
         int const nLimit = (int)nbSymbols - unroll + 1;
         for (n=0; n<(int)tableLog+1; n++) {
@@ -406,10 +449,9 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
      * We can switch based on the length to a different inner loop which is
      * optimized for that particular case.
      */
-    {
-        U32 w;
-        int symbol=wksp->rankVal[0];
-        int rankStart=0;
+    {   U32 w;
+        int symbol = wksp->rankVal[0];
+        int rankStart = 0;
         for (w=1; w<tableLog+1; ++w) {
             int const symbolCount = wksp->rankVal[w];
             int const length = (1 << w) >> 1;
@@ -483,15 +525,19 @@ HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog
 }
 
 #define HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr) \
-    *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog)
+    do { *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog); } while (0)
 
-#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)  \
-    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
+#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)      \
+    do {                                            \
+        if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+            HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr); \
+    } while (0)
 
-#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr) \
-    if (MEM_64bits()) \
-        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
+#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr)      \
+    do {                                            \
+        if (MEM_64bits())                           \
+            HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr); \
+    } while (0)
 
 HINT_INLINE size_t
 HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX1* const dt, const U32 dtLog)
@@ -519,7 +565,7 @@ HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, cons
     while (p < pEnd)
         HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
-    return pEnd-pStart;
+    return (size_t)(pEnd-pStart);
 }
 
 FORCE_INLINE_TEMPLATE size_t
@@ -529,7 +575,7 @@ HUF_decompress1X1_usingDTable_internal_body(
     const HUF_DTable* DTable)
 {
     BYTE* op = (BYTE*)dst;
-    BYTE* const oend = op + dstSize;
+    BYTE* const oend = ZSTD_maybeNullPtrAdd(op, dstSize);
     const void* dtPtr = DTable + 1;
     const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
     BIT_DStream_t bitD;
@@ -545,6 +591,10 @@ HUF_decompress1X1_usingDTable_internal_body(
     return dstSize;
 }
 
+/* HUF_decompress4X1_usingDTable_internal_body():
+ * Conditions :
+ * @dstSize >= 6
+ */
 FORCE_INLINE_TEMPLATE size_t
 HUF_decompress4X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
@@ -553,6 +603,7 @@ HUF_decompress4X1_usingDTable_internal_body(
 {
     /* Check */
     if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+    if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
 
     {   const BYTE* const istart = (const BYTE*) cSrc;
         BYTE* const ostart = (BYTE*) dst;
@@ -588,6 +639,7 @@ HUF_decompress4X1_usingDTable_internal_body(
 
         if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
         if (opStart4 > oend) return ERROR(corruption_detected);      /* overflow */
+        assert(dstSize >= 6); /* validated above */
         CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
         CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
         CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
@@ -650,52 +702,173 @@ size_t HUF_decompress4X1_usingDTable_internal_bmi2(void* dst, size_t dstSize, vo
 }
 #endif
 
-#if HUF_NEED_DEFAULT_FUNCTION
 static
 size_t HUF_decompress4X1_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
                     size_t cSrcSize, HUF_DTable const* DTable) {
     return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
 }
-#endif
 
 #if ZSTD_ENABLE_ASM_X86_64_BMI2
 
-HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop(HUF_DecompressAsmArgs* args) ZSTDLIB_HIDDEN;
+HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X1_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U16 const* const dtable = (U16 const*)args->dt;
+    BYTE* const oend = args->oend;
+    BYTE const* const ilowest = args->ilowest;
+
+    /* Copy the arguments to local variables */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= (stream == 3 ? oend : op[stream + 1]));
+            assert(ip[stream] >= ilowest);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each iteration produces 5 output symbols per stream */
+            size_t const oiters = (size_t)(oend - op[3]) / 5;
+            /* Each iteration consumes up to 11 bits * 5 = 55 bits < 7 bytes
+             * per stream.
+             */
+            size_t const iiters = (size_t)(ip[0] - ilowest) / 7;
+            /* We can safely run iters iterations before running bounds checks */
+            size_t const iters = MIN(oiters, iiters);
+            size_t const symbols = iters * 5;
+
+            /* We can simply check that op[3] < olimit, instead of checking all
+             * of our bounds, since we can't hit the other bounds until we've run
+             * iters iterations, which only happens when op[3] == olimit.
+             */
+            olimit = op[3] + symbols;
+
+            /* Exit fast decoding loop once we reach the end. */
+            if (op[3] == olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+#define HUF_4X1_DECODE_SYMBOL(_stream, _symbol)                 \
+    do {                                                        \
+        int const index = (int)(bits[(_stream)] >> 53);         \
+        int const entry = (int)dtable[index];                   \
+        bits[(_stream)] <<= (entry & 0x3F);                     \
+        op[(_stream)][(_symbol)] = (BYTE)((entry >> 8) & 0xFF); \
+    } while (0)
+
+#define HUF_4X1_RELOAD_STREAM(_stream)                              \
+    do {                                                            \
+        int const ctz = ZSTD_countTrailingZeros64(bits[(_stream)]); \
+        int const nbBits = ctz & 7;                                 \
+        int const nbBytes = ctz >> 3;                               \
+        op[(_stream)] += 5;                                         \
+        ip[(_stream)] -= nbBytes;                                   \
+        bits[(_stream)] = MEM_read64(ip[(_stream)]) | 1;            \
+        bits[(_stream)] <<= nbBits;                                 \
+    } while (0)
+
+        /* Manually unroll the loop because compilers don't consistently
+         * unroll the inner loops, which destroys performance.
+         */
+        do {
+            /* Decode 5 symbols in each of the 4 streams */
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 1);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 2);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 3);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 4);
+
+            /* Reload each of the 4 the bitstreams */
+            HUF_4X_FOR_EACH_STREAM(HUF_4X1_RELOAD_STREAM);
+        } while (op[3] < olimit);
+
+#undef HUF_4X1_DECODE_SYMBOL
+#undef HUF_4X1_RELOAD_STREAM
+    }
 
-static HUF_ASM_X86_64_BMI2_ATTRS
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+/*
+ * @returns @p dstSize on success (>= 6)
+ *          0 if the fallback implementation should be used
+ *          An error if an error occurred
+ */
+static HUF_FAST_BMI2_ATTRS
 size_t
-HUF_decompress4X1_usingDTable_internal_bmi2_asm(
+HUF_decompress4X1_usingDTable_internal_fast(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn)
 {
     void const* dt = DTable + 1;
-    const BYTE* const iend = (const BYTE*)cSrc + 6;
-    BYTE* const oend = (BYTE*)dst + dstSize;
-    HUF_DecompressAsmArgs args;
-    {
-        size_t const ret = HUF_DecompressAsmArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
-        FORWARD_IF_ERROR(ret, "Failed to init asm args");
-        if (ret != 0)
-            return HUF_decompress4X1_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
+    BYTE const* const ilowest = (BYTE const*)cSrc;
+    BYTE* const oend = ZSTD_maybeNullPtrAdd((BYTE*)dst, dstSize);
+    HUF_DecompressFastArgs args;
+    {   size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init fast loop args");
+        if (ret == 0)
+            return 0;
     }
 
-    assert(args.ip[0] >= args.ilimit);
-    HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop(&args);
+    assert(args.ip[0] >= args.ilowest);
+    loopFn(&args);
 
-    /* Our loop guarantees that ip[] >= ilimit and that we haven't
+    /* Our loop guarantees that ip[] >= ilowest and that we haven't
     * overwritten any op[].
     */
-    assert(args.ip[0] >= iend);
-    assert(args.ip[1] >= iend);
-    assert(args.ip[2] >= iend);
-    assert(args.ip[3] >= iend);
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[1] >= ilowest);
+    assert(args.ip[2] >= ilowest);
+    assert(args.ip[3] >= ilowest);
     assert(args.op[3] <= oend);
-    (void)iend;
+
+    assert(ilowest == args.ilowest);
+    assert(ilowest + 6 == args.iend[0]);
+    (void)ilowest;
 
     /* finish bit streams one by one. */
-    {
-        size_t const segmentSize = (dstSize+3) / 4;
+    {   size_t const segmentSize = (dstSize+3) / 4;
         BYTE* segmentEnd = (BYTE*)dst;
         int i;
         for (i = 0; i < 4; ++i) {
@@ -712,97 +885,59 @@ HUF_decompress4X1_usingDTable_internal_bmi2_asm(
     }
 
     /* decoded size */
+    assert(dstSize != 0);
     return dstSize;
 }
-#endif /* ZSTD_ENABLE_ASM_X86_64_BMI2 */
-
-typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
-                                               const void *cSrc,
-                                               size_t cSrcSize,
-                                               const HUF_DTable *DTable);
 
 HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
 
 static size_t HUF_decompress4X1_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
-                    size_t cSrcSize, HUF_DTable const* DTable, int bmi2)
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
 {
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X1_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X1_usingDTable_internal_fast_c_loop;
+
 #if DYNAMIC_BMI2
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X1_usingDTable_internal_bmi2;
 # if ZSTD_ENABLE_ASM_X86_64_BMI2
-        return HUF_decompress4X1_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
-# else
-        return HUF_decompress4X1_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+        }
 # endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
     }
-#else
-    (void)bmi2;
 #endif
 
 #if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
-    return HUF_decompress4X1_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
-#else
-    return HUF_decompress4X1_usingDTable_internal_default(dst, dstSize, cSrc, cSrcSize, DTable);
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+    }
 #endif
-}
-
-
-size_t HUF_decompress1X1_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-}
 
-size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    const BYTE* ip = (const BYTE*) cSrc;
-
-    size_t const hSize = HUF_readDTableX1_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
-
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
-}
-
-
-size_t HUF_decompress4X1_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    if (HUF_ENABLE_FAST_DECODE && !(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X1_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
 }
 
-static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize, int bmi2)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
-}
-
-size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 
-
 #endif /* HUF_FORCE_DECOMPRESS_X2 */
 
 
@@ -985,7 +1120,7 @@ static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 targetLog, const U32
 
 static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
                            const sortedSymbol_t* sortedList,
-                           const U32* rankStart, rankValCol_t *rankValOrigin, const U32 maxWeight,
+                           const U32* rankStart, rankValCol_t* rankValOrigin, const U32 maxWeight,
                            const U32 nbBitsBaseline)
 {
     U32* const rankVal = rankValOrigin[0];
@@ -1040,14 +1175,7 @@ typedef struct {
 
 size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
                        const void* src, size_t srcSize,
-                             void* workSpace, size_t wkspSize)
-{
-    return HUF_readDTableX2_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
-}
-
-size_t HUF_readDTableX2_wksp_bmi2(HUF_DTable* DTable,
-                       const void* src, size_t srcSize,
-                             void* workSpace, size_t wkspSize, int bmi2)
+                             void* workSpace, size_t wkspSize, int flags)
 {
     U32 tableLog, maxW, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
@@ -1069,7 +1197,7 @@ size_t HUF_readDTableX2_wksp_bmi2(HUF_DTable* DTable,
     if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* ZSTD_memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), bmi2);
+    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), flags);
     if (HUF_isError(iSize)) return iSize;
 
     /* check result */
@@ -1159,15 +1287,19 @@ HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, c
 }
 
 #define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
-    ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+    do { ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); } while (0)
 
-#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
-    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr)                     \
+    do {                                                           \
+        if (MEM_64bits() || (HUF_TABLELOG_MAX<=12))                \
+            ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); \
+    } while (0)
 
-#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
-    if (MEM_64bits()) \
-        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr)                     \
+    do {                                                           \
+        if (MEM_64bits())                                          \
+            ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); \
+    } while (0)
 
 HINT_INLINE size_t
 HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
@@ -1227,7 +1359,7 @@ HUF_decompress1X2_usingDTable_internal_body(
 
     /* decode */
     {   BYTE* const ostart = (BYTE*) dst;
-        BYTE* const oend = ostart + dstSize;
+        BYTE* const oend = ZSTD_maybeNullPtrAdd(ostart, dstSize);
         const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
         const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
@@ -1240,6 +1372,11 @@ HUF_decompress1X2_usingDTable_internal_body(
     /* decoded size */
     return dstSize;
 }
+
+/* HUF_decompress4X2_usingDTable_internal_body():
+ * Conditions:
+ * @dstSize >= 6
+ */
 FORCE_INLINE_TEMPLATE size_t
 HUF_decompress4X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
@@ -1247,6 +1384,7 @@ HUF_decompress4X2_usingDTable_internal_body(
     const HUF_DTable* DTable)
 {
     if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+    if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
 
     {   const BYTE* const istart = (const BYTE*) cSrc;
         BYTE* const ostart = (BYTE*) dst;
@@ -1280,8 +1418,9 @@ HUF_decompress4X2_usingDTable_internal_body(
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
         U32 const dtLog = dtd.tableLog;
 
-        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        if (opStart4 > oend) return ERROR(corruption_detected);      /* overflow */
+        if (length4 > cSrcSize) return ERROR(corruption_detected);  /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);     /* overflow */
+        assert(dstSize >= 6 /* validated above */);
         CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
         CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
         CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
@@ -1366,44 +1505,191 @@ size_t HUF_decompress4X2_usingDTable_internal_bmi2(void* dst, size_t dstSize, vo
 }
 #endif
 
-#if HUF_NEED_DEFAULT_FUNCTION
 static
 size_t HUF_decompress4X2_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
                     size_t cSrcSize, HUF_DTable const* DTable) {
     return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
 }
-#endif
 
 #if ZSTD_ENABLE_ASM_X86_64_BMI2
 
-HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop(HUF_DecompressAsmArgs* args) ZSTDLIB_HIDDEN;
+HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X2_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    BYTE* oend[4];
+    HUF_DEltX2 const* const dtable = (HUF_DEltX2 const*)args->dt;
+    BYTE const* const ilowest = args->ilowest;
+
+    /* Copy the arguments to local registers. */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    oend[0] = op[1];
+    oend[1] = op[2];
+    oend[2] = op[3];
+    oend[3] = args->oend;
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= oend[stream]);
+            assert(ip[stream] >= ilowest);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each loop does 5 table lookups for each of the 4 streams.
+             * Each table lookup consumes up to 11 bits of input, and produces
+             * up to 2 bytes of output.
+             */
+            /* We can consume up to 7 bytes of input per iteration per stream.
+             * We also know that each input pointer is >= ip[0]. So we can run
+             * iters loops before running out of input.
+             */
+            size_t iters = (size_t)(ip[0] - ilowest) / 7;
+            /* Each iteration can produce up to 10 bytes of output per stream.
+             * Each output stream my advance at different rates. So take the
+             * minimum number of safe iterations among all the output streams.
+             */
+            for (stream = 0; stream < 4; ++stream) {
+                size_t const oiters = (size_t)(oend[stream] - op[stream]) / 10;
+                iters = MIN(iters, oiters);
+            }
+
+            /* Each iteration produces at least 5 output symbols. So until
+             * op[3] crosses olimit, we know we haven't executed iters
+             * iterations yet. This saves us maintaining an iters counter,
+             * at the expense of computing the remaining # of iterations
+             * more frequently.
+             */
+            olimit = op[3] + (iters * 5);
+
+            /* Exit the fast decoding loop once we reach the end. */
+            if (op[3] == olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
 
-static HUF_ASM_X86_64_BMI2_ATTRS size_t
-HUF_decompress4X2_usingDTable_internal_bmi2_asm(
+#define HUF_4X2_DECODE_SYMBOL(_stream, _decode3)                      \
+    do {                                                              \
+        if ((_decode3) || (_stream) != 3) {                           \
+            int const index = (int)(bits[(_stream)] >> 53);           \
+            HUF_DEltX2 const entry = dtable[index];                   \
+            MEM_write16(op[(_stream)], entry.sequence); \
+            bits[(_stream)] <<= (entry.nbBits) & 0x3F;                \
+            op[(_stream)] += (entry.length);                          \
+        }                                                             \
+    } while (0)
+
+#define HUF_4X2_RELOAD_STREAM(_stream)                                  \
+    do {                                                                \
+        HUF_4X2_DECODE_SYMBOL(3, 1);                                    \
+        {                                                               \
+            int const ctz = ZSTD_countTrailingZeros64(bits[(_stream)]); \
+            int const nbBits = ctz & 7;                                 \
+            int const nbBytes = ctz >> 3;                               \
+            ip[(_stream)] -= nbBytes;                                   \
+            bits[(_stream)] = MEM_read64(ip[(_stream)]) | 1;            \
+            bits[(_stream)] <<= nbBits;                                 \
+        }                                                               \
+    } while (0)
+
+        /* Manually unroll the loop because compilers don't consistently
+         * unroll the inner loops, which destroys performance.
+         */
+        do {
+            /* Decode 5 symbols from each of the first 3 streams.
+             * The final stream will be decoded during the reload phase
+             * to reduce register pressure.
+             */
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+
+            /* Decode one symbol from the final stream */
+            HUF_4X2_DECODE_SYMBOL(3, 1);
+
+            /* Decode 4 symbols from the final stream & reload bitstreams.
+             * The final stream is reloaded last, meaning that all 5 symbols
+             * are decoded from the final stream before it is reloaded.
+             */
+            HUF_4X_FOR_EACH_STREAM(HUF_4X2_RELOAD_STREAM);
+        } while (op[3] < olimit);
+    }
+
+#undef HUF_4X2_DECODE_SYMBOL
+#undef HUF_4X2_RELOAD_STREAM
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+
+static HUF_FAST_BMI2_ATTRS size_t
+HUF_decompress4X2_usingDTable_internal_fast(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable) {
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn) {
     void const* dt = DTable + 1;
-    const BYTE* const iend = (const BYTE*)cSrc + 6;
-    BYTE* const oend = (BYTE*)dst + dstSize;
-    HUF_DecompressAsmArgs args;
+    const BYTE* const ilowest = (const BYTE*)cSrc;
+    BYTE* const oend = ZSTD_maybeNullPtrAdd((BYTE*)dst, dstSize);
+    HUF_DecompressFastArgs args;
     {
-        size_t const ret = HUF_DecompressAsmArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
         FORWARD_IF_ERROR(ret, "Failed to init asm args");
-        if (ret != 0)
-            return HUF_decompress4X2_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
+        if (ret == 0)
+            return 0;
     }
 
-    assert(args.ip[0] >= args.ilimit);
-    HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop(&args);
+    assert(args.ip[0] >= args.ilowest);
+    loopFn(&args);
 
     /* note : op4 already verified within main loop */
-    assert(args.ip[0] >= iend);
-    assert(args.ip[1] >= iend);
-    assert(args.ip[2] >= iend);
-    assert(args.ip[3] >= iend);
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[1] >= ilowest);
+    assert(args.ip[2] >= ilowest);
+    assert(args.ip[3] >= ilowest);
     assert(args.op[3] <= oend);
-    (void)iend;
+
+    assert(ilowest == args.ilowest);
+    assert(ilowest + 6 == args.iend[0]);
+    (void)ilowest;
 
     /* finish bitStreams one by one */
     {
@@ -1426,91 +1712,72 @@ HUF_decompress4X2_usingDTable_internal_bmi2_asm(
     /* decoded size */
     return dstSize;
 }
-#endif /* ZSTD_ENABLE_ASM_X86_64_BMI2 */
 
 static size_t HUF_decompress4X2_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
-                    size_t cSrcSize, HUF_DTable const* DTable, int bmi2)
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
 {
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X2_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X2_usingDTable_internal_fast_c_loop;
+
 #if DYNAMIC_BMI2
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X2_usingDTable_internal_bmi2;
 # if ZSTD_ENABLE_ASM_X86_64_BMI2
-        return HUF_decompress4X2_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
-# else
-        return HUF_decompress4X2_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+        }
 # endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
     }
-#else
-    (void)bmi2;
 #endif
 
 #if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
-    return HUF_decompress4X2_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
-#else
-    return HUF_decompress4X2_usingDTable_internal_default(dst, dstSize, cSrc, cSrcSize, DTable);
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+    }
 #endif
+
+    if (HUF_ENABLE_FAST_DECODE && !(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X2_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
 }
 
 HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
 
-size_t HUF_decompress1X2_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-}
-
 size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
     size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
-                                               workSpace, wkspSize);
+                                               workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, flags);
 }
 
-
-size_t HUF_decompress4X2_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-}
-
-static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize, int bmi2)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
     size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
-                                         workSpace, wkspSize);
+                                         workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
-}
-
-
 #endif /* HUF_FORCE_DECOMPRESS_X1 */
 
 
@@ -1518,44 +1785,6 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 /* Universal decompression selectors */
 /* ***********************************/
 
-size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize,
-                                    const void* cSrc, size_t cSrcSize,
-                                    const HUF_DTable* DTable)
-{
-    DTableDesc const dtd = HUF_getDTableDesc(DTable);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-    (void)dtd;
-    assert(dtd.tableType == 0);
-    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-    (void)dtd;
-    assert(dtd.tableType == 1);
-    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#else
-    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#endif
-}
-
-size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
-                                    const void* cSrc, size_t cSrcSize,
-                                    const HUF_DTable* DTable)
-{
-    DTableDesc const dtd = HUF_getDTableDesc(DTable);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-    (void)dtd;
-    assert(dtd.tableType == 0);
-    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-    (void)dtd;
-    assert(dtd.tableType == 1);
-    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#else
-    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#endif
-}
-
 
 #if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
 typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
@@ -1610,36 +1839,9 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
 #endif
 }
 
-
-size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
-                                     size_t dstSize, const void* cSrc,
-                                     size_t cSrcSize, void* workSpace,
-                                     size_t wkspSize)
-{
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize == 0) return ERROR(corruption_detected);
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#else
-        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                            cSrcSize, workSpace, wkspSize):
-                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#endif
-    }
-}
-
 size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                   const void* cSrc, size_t cSrcSize,
-                                  void* workSpace, size_t wkspSize)
+                                  void* workSpace, size_t wkspSize, int flags)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -1652,71 +1854,71 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
         (void)algoNb;
         assert(algoNb == 0);
         return HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
         (void)algoNb;
         assert(algoNb == 1);
         return HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #else
         return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize):
+                                cSrcSize, workSpace, wkspSize, flags):
                         HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #endif
     }
 }
 
 
-size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
 #if defined(HUF_FORCE_DECOMPRESS_X1)
     (void)dtd;
     assert(dtd.tableType == 0);
-    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
     (void)dtd;
     assert(dtd.tableType == 1);
-    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #else
-    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #endif
 }
 
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 #endif
 
-size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
 #if defined(HUF_FORCE_DECOMPRESS_X1)
     (void)dtd;
     assert(dtd.tableType == 0);
-    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
     (void)dtd;
     assert(dtd.tableType == 1);
-    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #else
-    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #endif
 }
 
-size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -1726,15 +1928,14 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
 #if defined(HUF_FORCE_DECOMPRESS_X1)
         (void)algoNb;
         assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
         (void)algoNb;
         assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #else
-        return algoNb ? HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
-                        HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags) :
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #endif
     }
 }
-
diff --git a/lib/zstd/decompress/zstd_ddict.c b/lib/zstd/decompress/zstd_ddict.c
index dbbc7919de53..30ef65e1ab5c 100644
--- a/lib/zstd/decompress/zstd_ddict.c
+++ b/lib/zstd/decompress/zstd_ddict.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,12 +15,12 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customFree */
 #include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
 #include "../common/cpu.h"         /* bmi2 */
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "zstd_decompress_internal.h"
 #include "zstd_ddict.h"
@@ -131,7 +132,7 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
         ZSTD_memcpy(internalBuffer, dict, dictSize);
     }
     ddict->dictSize = dictSize;
-    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    ddict->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
 
     /* parse dictionary content */
     FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
@@ -237,5 +238,5 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
 unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;
-    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+    return ddict->dictID;
 }
diff --git a/lib/zstd/decompress/zstd_ddict.h b/lib/zstd/decompress/zstd_ddict.h
index 8c1a79d666f8..de459a0dacd1 100644
--- a/lib/zstd/decompress/zstd_ddict.h
+++ b/lib/zstd/decompress/zstd_ddict.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/decompress/zstd_decompress.c b/lib/zstd/decompress/zstd_decompress.c
index 6b3177c94711..c9cbc45f6ed9 100644
--- a/lib/zstd/decompress/zstd_decompress.c
+++ b/lib/zstd/decompress/zstd_decompress.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -53,13 +54,15 @@
 *  Dependencies
 *********************************************************/
 #include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/error_private.h"
+#include "../common/zstd_internal.h"  /* blockProperties_t */
 #include "../common/mem.h"         /* low level memory routines */
+#include "../common/bits.h"  /* ZSTD_highbit32 */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include <linux/xxhash.h> /* xxh64_reset, xxh64_update, xxh64_digest, XXH64 */
-#include "../common/zstd_internal.h"  /* blockProperties_t */
 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
 #include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
@@ -72,11 +75,11 @@
  *************************************/
 
 #define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
-#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3   /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
-                                                     * Currently, that means a 0.75 load factor.
-                                                     * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
-                                                     * the load factor of the ddict hash set.
-                                                     */
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3  /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
+                                                    * Currently, that means a 0.75 load factor.
+                                                    * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
+                                                    * the load factor of the ddict hash set.
+                                                    */
 
 #define DDICT_HASHSET_TABLE_BASE_SIZE 64
 #define DDICT_HASHSET_RESIZE_FACTOR 2
@@ -237,6 +240,8 @@ static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
     dctx->outBufferMode = ZSTD_bm_buffered;
     dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
     dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
+    dctx->disableHufAsm = 0;
+    dctx->maxBlockSizeParam = 0;
 }
 
 static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
@@ -253,6 +258,7 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->streamStage = zdss_init;
     dctx->noForwardProgress = 0;
     dctx->oversizedDuration = 0;
+    dctx->isFrameDecompression = 1;
 #if DYNAMIC_BMI2
     dctx->bmi2 = ZSTD_cpuSupportsBmi2();
 #endif
@@ -421,16 +427,40 @@ size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
  *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
  * @return : 0, `zfhPtr` is correctly filled,
  *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
- *           or an error code, which can be tested using ZSTD_isError() */
+**           or an error code, which can be tested using ZSTD_isError() */
 size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
 {
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
-    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
-    if (srcSize < minInputSize) return minInputSize;
-    RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
+    DEBUGLOG(5, "ZSTD_getFrameHeader_advanced: minInputSize = %zu, srcSize = %zu", minInputSize, srcSize);
+
+    if (srcSize > 0) {
+        /* note : technically could be considered an assert(), since it's an invalid entry */
+        RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter : src==NULL, but srcSize>0");
+    }
+    if (srcSize < minInputSize) {
+        if (srcSize > 0 && format != ZSTD_f_zstd1_magicless) {
+            /* when receiving less than @minInputSize bytes,
+             * control these bytes at least correspond to a supported magic number
+             * in order to error out early if they don't.
+            **/
+            size_t const toCopy = MIN(4, srcSize);
+            unsigned char hbuf[4]; MEM_writeLE32(hbuf, ZSTD_MAGICNUMBER);
+            assert(src != NULL);
+            ZSTD_memcpy(hbuf, src, toCopy);
+            if ( MEM_readLE32(hbuf) != ZSTD_MAGICNUMBER ) {
+                /* not a zstd frame : let's check if it's a skippable frame */
+                MEM_writeLE32(hbuf, ZSTD_MAGIC_SKIPPABLE_START);
+                ZSTD_memcpy(hbuf, src, toCopy);
+                if ((MEM_readLE32(hbuf) & ZSTD_MAGIC_SKIPPABLE_MASK) != ZSTD_MAGIC_SKIPPABLE_START) {
+                    RETURN_ERROR(prefix_unknown,
+                                "first bytes don't correspond to any supported magic number");
+        }   }   }
+        return minInputSize;
+    }
 
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzers may not understand that zfhPtr will be read only if return value is zero, since they are 2 different signals */
     if ( (format != ZSTD_f_zstd1_magicless)
       && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
         if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
@@ -540,61 +570,62 @@ static size_t readSkippableFrameSize(void const* src, size_t srcSize)
     sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
     RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
                     frameParameter_unsupported, "");
-    {
-        size_t const skippableSize = skippableHeaderSize + sizeU32;
+    {   size_t const skippableSize = skippableHeaderSize + sizeU32;
         RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong, "");
         return skippableSize;
     }
 }
 
 /*! ZSTD_readSkippableFrame() :
- * Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
+ * Retrieves content of a skippable frame, and writes it to dst buffer.
  *
  * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
  * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
  * in the magicVariant.
  *
- * Returns an error if destination buffer is not large enough, or if the frame is not skippable.
+ * Returns an error if destination buffer is not large enough, or if this is not a valid skippable frame.
  *
  * @return : number of bytes written or a ZSTD error.
  */
-ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
-                                            const void* src, size_t srcSize)
+size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity,
+                               unsigned* magicVariant,  /* optional, can be NULL */
+                         const void* src, size_t srcSize)
 {
-    U32 const magicNumber = MEM_readLE32(src);
-    size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
-    size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
-
-    /* check input validity */
-    RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
-    RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
-    RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
 
-    /* deliver payload */
-    if (skippableContentSize > 0  && dst != NULL)
-        ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
-    if (magicVariant != NULL)
-        *magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
-    return skippableContentSize;
+    {   U32 const magicNumber = MEM_readLE32(src);
+        size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
+        size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
+
+        /* check input validity */
+        RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
+        RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
+
+        /* deliver payload */
+        if (skippableContentSize > 0  && dst != NULL)
+            ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
+        if (magicVariant != NULL)
+            *magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
+        return skippableContentSize;
+    }
 }
 
 /* ZSTD_findDecompressedSize() :
- *  compatible with legacy mode
  *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
  *      skippable frames
- *  @return : decompressed size of the frames contained */
+ *  note: compatible with legacy mode
+ * @return : decompressed size of the frames contained */
 unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 {
-    unsigned long long totalDstSize = 0;
+    U64 totalDstSize = 0;
 
     while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
         U32 const magicNumber = MEM_readLE32(src);
 
         if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
             size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-            if (ZSTD_isError(skippableSize)) {
-                return ZSTD_CONTENTSIZE_ERROR;
-            }
+            if (ZSTD_isError(skippableSize)) return ZSTD_CONTENTSIZE_ERROR;
             assert(skippableSize <= srcSize);
 
             src = (const BYTE *)src + skippableSize;
@@ -602,17 +633,17 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
             continue;
         }
 
-        {   unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
-            if (ret >= ZSTD_CONTENTSIZE_ERROR) return ret;
+        {   unsigned long long const fcs = ZSTD_getFrameContentSize(src, srcSize);
+            if (fcs >= ZSTD_CONTENTSIZE_ERROR) return fcs;
 
-            /* check for overflow */
-            if (totalDstSize + ret < totalDstSize) return ZSTD_CONTENTSIZE_ERROR;
-            totalDstSize += ret;
+            if (U64_MAX - totalDstSize < fcs)
+                return ZSTD_CONTENTSIZE_ERROR; /* check for overflow */
+            totalDstSize += fcs;
         }
+        /* skip to next frame */
         {   size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
-            if (ZSTD_isError(frameSrcSize)) {
-                return ZSTD_CONTENTSIZE_ERROR;
-            }
+            if (ZSTD_isError(frameSrcSize)) return ZSTD_CONTENTSIZE_ERROR;
+            assert(frameSrcSize <= srcSize);
 
             src = (const BYTE *)src + frameSrcSize;
             srcSize -= frameSrcSize;
@@ -676,13 +707,13 @@ static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
     return frameSizeInfo;
 }
 
-static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
+static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize, ZSTD_format_e format)
 {
     ZSTD_frameSizeInfo frameSizeInfo;
     ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
 
 
-    if ((srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
+    if (format == ZSTD_f_zstd1 && (srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
         && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
         frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
         assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
@@ -696,7 +727,7 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
         ZSTD_frameHeader zfh;
 
         /* Extract Frame Header */
-        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
+        {   size_t const ret = ZSTD_getFrameHeader_advanced(&zfh, src, srcSize, format);
             if (ZSTD_isError(ret))
                 return ZSTD_errorFrameSizeInfo(ret);
             if (ret > 0)
@@ -730,23 +761,26 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
             ip += 4;
         }
 
+        frameSizeInfo.nbBlocks = nbBlocks;
         frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
         frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
                                         ? zfh.frameContentSize
-                                        : nbBlocks * zfh.blockSizeMax;
+                                        : (unsigned long long)nbBlocks * zfh.blockSizeMax;
         return frameSizeInfo;
     }
 }
 
+static size_t ZSTD_findFrameCompressedSize_advanced(const void *src, size_t srcSize, ZSTD_format_e format) {
+    ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, format);
+    return frameSizeInfo.compressedSize;
+}
+
 /* ZSTD_findFrameCompressedSize() :
- *  compatible with legacy mode
- *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
- *  `srcSize` must be at least as large as the frame contained
- *  @return : the compressed size of the frame starting at `src` */
+ * See docs in zstd.h
+ * Note: compatible with legacy mode */
 size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
 {
-    ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
-    return frameSizeInfo.compressedSize;
+    return ZSTD_findFrameCompressedSize_advanced(src, srcSize, ZSTD_f_zstd1);
 }
 
 /* ZSTD_decompressBound() :
@@ -760,7 +794,7 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
     unsigned long long bound = 0;
     /* Iterate over each frame */
     while (srcSize > 0) {
-        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, ZSTD_f_zstd1);
         size_t const compressedSize = frameSizeInfo.compressedSize;
         unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
         if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
@@ -773,6 +807,48 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
     return bound;
 }
 
+size_t ZSTD_decompressionMargin(void const* src, size_t srcSize)
+{
+    size_t margin = 0;
+    unsigned maxBlockSize = 0;
+
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, ZSTD_f_zstd1);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        ZSTD_frameHeader zfh;
+
+        FORWARD_IF_ERROR(ZSTD_getFrameHeader(&zfh, src, srcSize), "");
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ERROR(corruption_detected);
+
+        if (zfh.frameType == ZSTD_frame) {
+            /* Add the frame header to our margin */
+            margin += zfh.headerSize;
+            /* Add the checksum to our margin */
+            margin += zfh.checksumFlag ? 4 : 0;
+            /* Add 3 bytes per block */
+            margin += 3 * frameSizeInfo.nbBlocks;
+
+            /* Compute the max block size */
+            maxBlockSize = MAX(maxBlockSize, zfh.blockSizeMax);
+        } else {
+            assert(zfh.frameType == ZSTD_skippableFrame);
+            /* Add the entire skippable frame size to our margin. */
+            margin += compressedSize;
+        }
+
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+    }
+
+    /* Add the max block size back to the margin. */
+    margin += maxBlockSize;
+
+    return margin;
+}
 
 /*-*************************************************************
  *   Frame decoding
@@ -856,6 +932,10 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         ip += frameHeaderSize; remainingSrcSize -= frameHeaderSize;
     }
 
+    /* Shrink the blockSizeMax if enabled */
+    if (dctx->maxBlockSizeParam != 0)
+        dctx->fParams.blockSizeMax = MIN(dctx->fParams.blockSizeMax, (unsigned)dctx->maxBlockSizeParam);
+
     /* Loop on each block */
     while (1) {
         BYTE* oBlockEnd = oend;
@@ -888,7 +968,8 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, /* frame */ 1, not_streaming);
+            assert(dctx->isFrameDecompression == 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, not_streaming);
             break;
         case bt_raw :
             /* Use oend instead of oBlockEnd because this function is safe to overlap. It uses memmove. */
@@ -901,12 +982,14 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         default:
             RETURN_ERROR(corruption_detected, "invalid block type");
         }
-
-        if (ZSTD_isError(decodedSize)) return decodedSize;
-        if (dctx->validateChecksum)
+        FORWARD_IF_ERROR(decodedSize, "Block decompression failure");
+        DEBUGLOG(5, "Decompressed block of dSize = %u", (unsigned)decodedSize);
+        if (dctx->validateChecksum) {
             xxh64_update(&dctx->xxhState, op, decodedSize);
-        if (decodedSize != 0)
+        }
+        if (decodedSize) /* support dst = NULL,0 */ {
             op += decodedSize;
+        }
         assert(ip != NULL);
         ip += cBlockSize;
         remainingSrcSize -= cBlockSize;
@@ -930,12 +1013,15 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
     }
     ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
     /* Allow caller to get size read */
+    DEBUGLOG(4, "ZSTD_decompressFrame: decompressed frame of size %zi, consuming %zi bytes of input", op-ostart, ip - (const BYTE*)*srcPtr);
     *srcPtr = ip;
     *srcSizePtr = remainingSrcSize;
     return (size_t)(op-ostart);
 }
 
-static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                         void* dst, size_t dstCapacity,
                                   const void* src, size_t srcSize,
                                   const void* dict, size_t dictSize,
@@ -955,17 +1041,18 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
     while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
 
 
-        {   U32 const magicNumber = MEM_readLE32(src);
-            DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
-                        (unsigned)magicNumber, ZSTD_MAGICNUMBER);
+        if (dctx->format == ZSTD_f_zstd1 && srcSize >= 4) {
+            U32 const magicNumber = MEM_readLE32(src);
+            DEBUGLOG(5, "reading magic number %08X", (unsigned)magicNumber);
             if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+                /* skippable frame detected : skip it */
                 size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-                FORWARD_IF_ERROR(skippableSize, "readSkippableFrameSize failed");
+                FORWARD_IF_ERROR(skippableSize, "invalid skippable frame");
                 assert(skippableSize <= srcSize);
 
                 src = (const BYTE *)src + skippableSize;
                 srcSize -= skippableSize;
-                continue;
+                continue; /* check next frame */
         }   }
 
         if (ddict) {
@@ -1061,8 +1148,8 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
 
 /*
- * Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed,
- * we allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed, we
+ * allow taking a partial block as the input. Currently only raw uncompressed blocks can
  * be streamed.
  *
  * For blocks that can be streamed, this allows us to reduce the latency until we produce
@@ -1181,7 +1268,8 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             {
             case bt_compressed:
                 DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
-                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1, is_streaming);
+                assert(dctx->isFrameDecompression == 1);
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, is_streaming);
                 dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_raw :
@@ -1250,6 +1338,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     case ZSTDds_decodeSkippableHeader:
         assert(src != NULL);
         assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
+        assert(dctx->format != ZSTD_f_zstd1_magicless);
         ZSTD_memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
         dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
         dctx->stage = ZSTDds_skipFrame;
@@ -1262,7 +1351,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 
     default:
         assert(0);   /* impossible */
-        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
     }
 }
 
@@ -1303,11 +1392,11 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
         /* in minimal huffman, we always use X1 variants */
         size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
                                                 dictPtr, dictEnd - dictPtr,
-                                                workspace, workspaceSize);
+                                                workspace, workspaceSize, /* flags */ 0);
 #else
         size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
                                                 dictPtr, (size_t)(dictEnd - dictPtr),
-                                                workspace, workspaceSize);
+                                                workspace, workspaceSize, /* flags */ 0);
 #endif
         RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
         dictPtr += hSize;
@@ -1403,10 +1492,11 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
     dctx->prefixStart = NULL;
     dctx->virtualStart = NULL;
     dctx->dictEnd = NULL;
-    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    dctx->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
     dctx->litEntropy = dctx->fseEntropy = 0;
     dctx->dictID = 0;
     dctx->bType = bt_reserved;
+    dctx->isFrameDecompression = 1;
     ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
     ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
     dctx->LLTptr = dctx->entropy.LLTable;
@@ -1465,7 +1555,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
  *  This could for one of the following reasons :
  *  - The frame does not require a dictionary (most common case).
  *  - The frame was built with dictID intentionally removed.
- *    Needed dictionary is a hidden information.
+ *    Needed dictionary is a hidden piece of information.
  *    Note : this use case also happens when using a non-conformant dictionary.
  *  - `srcSize` is too small, and as a result, frame header could not be decoded.
  *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
@@ -1474,7 +1564,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
  *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
 unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
 {
-    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0 };
+    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0, 0, 0 };
     size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
     if (ZSTD_isError(hError)) return 0;
     return zfp.dictID;
@@ -1581,7 +1671,9 @@ size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t di
 size_t ZSTD_initDStream(ZSTD_DStream* zds)
 {
     DEBUGLOG(4, "ZSTD_initDStream");
-    return ZSTD_initDStream_usingDDict(zds, NULL);
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(zds, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_DCtx_refDDict(zds, NULL), "");
+    return ZSTD_startingInputLength(zds->format);
 }
 
 /* ZSTD_initDStream_usingDDict() :
@@ -1589,6 +1681,7 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
  * this function cannot fail */
 size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
 {
+    DEBUGLOG(4, "ZSTD_initDStream_usingDDict");
     FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
     FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
     return ZSTD_startingInputLength(dctx->format);
@@ -1599,6 +1692,7 @@ size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
  * this function cannot fail */
 size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
+    DEBUGLOG(4, "ZSTD_resetDStream");
     FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
     return ZSTD_startingInputLength(dctx->format);
 }
@@ -1670,6 +1764,15 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
             bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
             bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
             return bounds;
+        case ZSTD_d_disableHuffmanAssembly:
+            bounds.lowerBound = 0;
+            bounds.upperBound = 1;
+            return bounds;
+        case ZSTD_d_maxBlockSize:
+            bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+            bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+            return bounds;
+
         default:;
     }
     bounds.error = ERROR(parameter_unsupported);
@@ -1710,6 +1813,12 @@ size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value
         case ZSTD_d_refMultipleDDicts:
             *value = (int)dctx->refMultipleDDicts;
             return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            *value = (int)dctx->disableHufAsm;
+            return 0;
+        case ZSTD_d_maxBlockSize:
+            *value = dctx->maxBlockSizeParam;
+            return 0;
         default:;
     }
     RETURN_ERROR(parameter_unsupported, "");
@@ -1743,6 +1852,14 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
             }
             dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
             return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            CHECK_DBOUNDS(ZSTD_d_disableHuffmanAssembly, value);
+            dctx->disableHufAsm = value != 0;
+            return 0;
+        case ZSTD_d_maxBlockSize:
+            if (value != 0) CHECK_DBOUNDS(ZSTD_d_maxBlockSize, value);
+            dctx->maxBlockSizeParam = value;
+            return 0;
         default:;
     }
     RETURN_ERROR(parameter_unsupported, "");
@@ -1754,6 +1871,7 @@ size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
       || (reset == ZSTD_reset_session_and_parameters) ) {
         dctx->streamStage = zdss_init;
         dctx->noForwardProgress = 0;
+        dctx->isFrameDecompression = 1;
     }
     if ( (reset == ZSTD_reset_parameters)
       || (reset == ZSTD_reset_session_and_parameters) ) {
@@ -1770,11 +1888,17 @@ size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
     return ZSTD_sizeof_DCtx(dctx);
 }
 
-size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
+static size_t ZSTD_decodingBufferSize_internal(unsigned long long windowSize, unsigned long long frameContentSize, size_t blockSizeMax)
 {
-    size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
-    /* space is needed to store the litbuffer after the output of a given block without stomping the extDict of a previous run, as well as to cover both windows against wildcopy*/
-    unsigned long long const neededRBSize = windowSize + blockSize + ZSTD_BLOCKSIZE_MAX + (WILDCOPY_OVERLENGTH * 2);
+    size_t const blockSize = MIN((size_t)MIN(windowSize, ZSTD_BLOCKSIZE_MAX), blockSizeMax);
+    /* We need blockSize + WILDCOPY_OVERLENGTH worth of buffer so that if a block
+     * ends at windowSize + WILDCOPY_OVERLENGTH + 1 bytes, we can start writing
+     * the block at the beginning of the output buffer, and maintain a full window.
+     *
+     * We need another blockSize worth of buffer so that we can store split
+     * literals at the end of the block without overwriting the extDict window.
+     */
+    unsigned long long const neededRBSize = windowSize + (blockSize * 2) + (WILDCOPY_OVERLENGTH * 2);
     unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
     size_t const minRBSize = (size_t) neededSize;
     RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
@@ -1782,6 +1906,11 @@ size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long
     return minRBSize;
 }
 
+size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
+{
+    return ZSTD_decodingBufferSize_internal(windowSize, frameContentSize, ZSTD_BLOCKSIZE_MAX);
+}
+
 size_t ZSTD_estimateDStreamSize(size_t windowSize)
 {
     size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
@@ -1918,7 +2047,6 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 if (zds->refMultipleDDicts && zds->ddictSet) {
                     ZSTD_DCtx_selectFrameDDict(zds);
                 }
-                DEBUGLOG(5, "header size : %u", (U32)hSize);
                 if (ZSTD_isError(hSize)) {
                     return hSize;   /* error */
                 }
@@ -1932,6 +2060,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                             zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
+                        /* check first few bytes */
+                        FORWARD_IF_ERROR(
+                            ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format),
+                            "First few bytes detected incorrect" );
+                        /* return hint input size */
                         return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     assert(ip != NULL);
@@ -1943,14 +2076,15 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
                 && zds->fParams.frameType != ZSTD_skippableFrame
                 && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
-                size_t const cSize = ZSTD_findFrameCompressedSize(istart, (size_t)(iend-istart));
+                size_t const cSize = ZSTD_findFrameCompressedSize_advanced(istart, (size_t)(iend-istart), zds->format);
                 if (cSize <= (size_t)(iend-istart)) {
                     /* shortcut : using single-pass mode */
                     size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
                     if (ZSTD_isError(decompressedSize)) return decompressedSize;
-                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()");
+                    assert(istart != NULL);
                     ip = istart + cSize;
-                    op += decompressedSize;
+                    op = op ? op + decompressedSize : op; /* can occur if frameContentSize = 0 (empty frame) */
                     zds->expected = 0;
                     zds->streamStage = zdss_init;
                     someMoreWork = 0;
@@ -1969,7 +2103,8 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             DEBUGLOG(4, "Consume header");
             FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)), "");
 
-            if ((MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+            if (zds->format == ZSTD_f_zstd1
+                && (MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
                 zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
                 zds->stage = ZSTDds_skipFrame;
             } else {
@@ -1985,11 +2120,13 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
             RETURN_ERROR_IF(zds->fParams.windowSize > zds->maxWindowSize,
                             frameParameter_windowTooLarge, "");
+            if (zds->maxBlockSizeParam != 0)
+                zds->fParams.blockSizeMax = MIN(zds->fParams.blockSizeMax, (unsigned)zds->maxBlockSizeParam);
 
             /* Adapt buffer sizes to frame header instructions */
             {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
                 size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
-                        ? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
+                        ? ZSTD_decodingBufferSize_internal(zds->fParams.windowSize, zds->fParams.frameContentSize, zds->fParams.blockSizeMax)
                         : 0;
 
                 ZSTD_DCtx_updateOversizedDuration(zds, neededInBuffSize, neededOutBuffSize);
@@ -2034,6 +2171,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
                     FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
+                    assert(ip != NULL);
                     ip += neededInSize;
                     /* Function modifies the stage so we must break */
                     break;
@@ -2048,7 +2186,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 int const isSkipFrame = ZSTD_isSkipFrame(zds);
                 size_t loadedSize;
                 /* At this point we shouldn't be decompressing a block that we can stream. */
-                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip));
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip)));
                 if (isSkipFrame) {
                     loadedSize = MIN(toLoad, (size_t)(iend-ip));
                 } else {
@@ -2057,8 +2195,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                                     "should never happen");
                     loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
                 }
-                ip += loadedSize;
-                zds->inPos += loadedSize;
+                if (loadedSize != 0) {
+                    /* ip may be NULL */
+                    ip += loadedSize;
+                    zds->inPos += loadedSize;
+                }
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
@@ -2068,14 +2209,17 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 break;
             }
         case zdss_flush:
-            {   size_t const toFlushSize = zds->outEnd - zds->outStart;
+            {
+                size_t const toFlushSize = zds->outEnd - zds->outStart;
                 size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
-                op += flushedSize;
+
+                op = op ? op + flushedSize : op;
+
                 zds->outStart += flushedSize;
                 if (flushedSize == toFlushSize) {  /* flush completed */
                     zds->streamStage = zdss_read;
                     if ( (zds->outBuffSize < zds->fParams.frameContentSize)
-                      && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
                         DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
                                 (int)(zds->outBuffSize - zds->outStart),
                                 (U32)zds->fParams.blockSizeMax);
@@ -2089,7 +2233,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
         default:
             assert(0);    /* impossible */
-            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
     }   }
 
     /* result */
@@ -2102,8 +2246,8 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
     if ((ip==istart) && (op==ostart)) {  /* no forward progress */
         zds->noForwardProgress ++;
         if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
-            RETURN_ERROR_IF(op==oend, dstSize_tooSmall, "");
-            RETURN_ERROR_IF(ip==iend, srcSize_wrong, "");
+            RETURN_ERROR_IF(op==oend, noForwardProgress_destFull, "");
+            RETURN_ERROR_IF(ip==iend, noForwardProgress_inputEmpty, "");
             assert(0);
         }
     } else {
@@ -2140,11 +2284,17 @@ size_t ZSTD_decompressStream_simpleArgs (
                             void* dst, size_t dstCapacity, size_t* dstPos,
                       const void* src, size_t srcSize, size_t* srcPos)
 {
-    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
-    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
-    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
-    size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
-    *dstPos = output.pos;
-    *srcPos = input.pos;
-    return cErr;
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    {   size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
 }
diff --git a/lib/zstd/decompress/zstd_decompress_block.c b/lib/zstd/decompress/zstd_decompress_block.c
index c1913b8e7c89..9fe9a12c8a2c 100644
--- a/lib/zstd/decompress/zstd_decompress_block.c
+++ b/lib/zstd/decompress/zstd_decompress_block.c
@@ -1,5 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -20,12 +21,12 @@
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/zstd_internal.h"
 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
 #include "zstd_decompress_block.h"
+#include "../common/bits.h"  /* ZSTD_highbit32 */
 
 /*_*******************************************************
 *  Macros
@@ -51,6 +52,13 @@ static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
  *   Block decoding
  ***************************************************************/
 
+static size_t ZSTD_blockSizeMax(ZSTD_DCtx const* dctx)
+{
+    size_t const blockSizeMax = dctx->isFrameDecompression ? dctx->fParams.blockSizeMax : ZSTD_BLOCKSIZE_MAX;
+    assert(blockSizeMax <= ZSTD_BLOCKSIZE_MAX);
+    return blockSizeMax;
+}
+
 /*! ZSTD_getcBlockSize() :
  *  Provides the size of compressed block from block header `src` */
 size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
@@ -73,41 +81,49 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
 static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
     const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
 {
-    if (streaming == not_streaming && dstCapacity > ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH)
-    {
-        /* room for litbuffer to fit without read faulting */
-        dctx->litBuffer = (BYTE*)dst + ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH;
+    size_t const blockSizeMax = ZSTD_blockSizeMax(dctx);
+    assert(litSize <= blockSizeMax);
+    assert(dctx->isFrameDecompression || streaming == not_streaming);
+    assert(expectedWriteSize <= blockSizeMax);
+    if (streaming == not_streaming && dstCapacity > blockSizeMax + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH) {
+        /* If we aren't streaming, we can just put the literals after the output
+         * of the current block. We don't need to worry about overwriting the
+         * extDict of our window, because it doesn't exist.
+         * So if we have space after the end of the block, just put it there.
+         */
+        dctx->litBuffer = (BYTE*)dst + blockSizeMax + WILDCOPY_OVERLENGTH;
         dctx->litBufferEnd = dctx->litBuffer + litSize;
         dctx->litBufferLocation = ZSTD_in_dst;
-    }
-    else if (litSize > ZSTD_LITBUFFEREXTRASIZE)
-    {
-        /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+    } else if (litSize <= ZSTD_LITBUFFEREXTRASIZE) {
+        /* Literals fit entirely within the extra buffer, put them there to avoid
+         * having to split the literals.
+         */
+        dctx->litBuffer = dctx->litExtraBuffer;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    } else {
+        assert(blockSizeMax > ZSTD_LITBUFFEREXTRASIZE);
+        /* Literals must be split between the output block and the extra lit
+         * buffer. We fill the extra lit buffer with the tail of the literals,
+         * and put the rest of the literals at the end of the block, with
+         * WILDCOPY_OVERLENGTH of buffer room to allow for overreads.
+         * This MUST not write more than our maxBlockSize beyond dst, because in
+         * streaming mode, that could overwrite part of our extDict window.
+         */
         if (splitImmediately) {
             /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
             dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
             dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
-        }
-        else {
-            /* initially this will be stored entirely in dst during huffman decoding, it will partially shifted to litExtraBuffer after */
+        } else {
+            /* initially this will be stored entirely in dst during huffman decoding, it will partially be shifted to litExtraBuffer after */
             dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
             dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
         }
         dctx->litBufferLocation = ZSTD_split;
-    }
-    else
-    {
-        /* fits entirely within litExtraBuffer, so no split is necessary */
-        dctx->litBuffer = dctx->litExtraBuffer;
-        dctx->litBufferEnd = dctx->litBuffer + litSize;
-        dctx->litBufferLocation = ZSTD_not_in_dst;
+        assert(dctx->litBufferEnd <= (BYTE*)dst + expectedWriteSize);
     }
 }
 
-/* Hidden declaration for fullbench */
-size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
-                          const void* src, size_t srcSize,
-                          void* dst, size_t dstCapacity, const streaming_operation streaming);
 /*! ZSTD_decodeLiteralsBlock() :
  * Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
  * in the dstBuffer.  If there is room to do so, it will be stored in full in the excess dst space after where the current
@@ -116,7 +132,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
  *
  * @return : nb of bytes read from src (< srcSize )
  *  note : symbol not declared but exposed for fullbench */
-size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+static size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                           const void* src, size_t srcSize,   /* note : srcSize < BLOCKSIZE */
                           void* dst, size_t dstCapacity, const streaming_operation streaming)
 {
@@ -125,6 +141,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 
     {   const BYTE* const istart = (const BYTE*) src;
         symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
+        size_t const blockSizeMax = ZSTD_blockSizeMax(dctx);
 
         switch(litEncType)
         {
@@ -134,13 +151,16 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
             ZSTD_FALLTHROUGH;
 
         case set_compressed:
-            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
+            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need up to 5 for case 3");
             {   size_t lhSize, litSize, litCSize;
                 U32 singleStream=0;
                 U32 const lhlCode = (istart[0] >> 2) & 3;
                 U32 const lhc = MEM_readLE32(istart);
                 size_t hufSuccess;
-                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
+                int const flags = 0
+                    | (ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0)
+                    | (dctx->disableHufAsm ? HUF_flags_disableAsm : 0);
                 switch(lhlCode)
                 {
                 case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -164,7 +184,11 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 }
                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
-                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
+                if (!singleStream)
+                    RETURN_ERROR_IF(litSize < MIN_LITERALS_FOR_4_STREAMS, literals_headerWrong,
+                        "Not enough literals (%zu) for the 4-streams mode (min %u)",
+                        litSize, MIN_LITERALS_FOR_4_STREAMS);
                 RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
                 RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
@@ -176,13 +200,14 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 
                 if (litEncType==set_repeat) {
                     if (singleStream) {
-                        hufSuccess = HUF_decompress1X_usingDTable_bmi2(
+                        hufSuccess = HUF_decompress1X_usingDTable(
                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                            dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
+                            dctx->HUFptr, flags);
                     } else {
-                        hufSuccess = HUF_decompress4X_usingDTable_bmi2(
+                        assert(litSize >= MIN_LITERALS_FOR_4_STREAMS);
+                        hufSuccess = HUF_decompress4X_usingDTable(
                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                            dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
+                            dctx->HUFptr, flags);
                     }
                 } else {
                     if (singleStream) {
@@ -190,26 +215,28 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                         hufSuccess = HUF_decompress1X_DCtx_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace));
+                            sizeof(dctx->workspace), flags);
 #else
-                        hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
+                        hufSuccess = HUF_decompress1X1_DCtx_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
+                            sizeof(dctx->workspace), flags);
 #endif
                     } else {
-                        hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
+                        hufSuccess = HUF_decompress4X_hufOnly_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
+                            sizeof(dctx->workspace), flags);
                     }
                 }
                 if (dctx->litBufferLocation == ZSTD_split)
                 {
+                    assert(litSize > ZSTD_LITBUFFEREXTRASIZE);
                     ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
                     ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
                     dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
                     dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
+                    assert(dctx->litBufferEnd <= (BYTE*)dst + blockSizeMax);
                 }
 
                 RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
@@ -224,7 +251,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         case set_basic:
             {   size_t litSize, lhSize;
                 U32 const lhlCode = ((istart[0]) >> 2) & 3;
-                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
                 switch(lhlCode)
                 {
                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -237,11 +264,13 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 case 3:
                     lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize = 3");
                     litSize = MEM_readLE24(istart) >> 4;
                     break;
                 }
 
                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
                 RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
                 if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
@@ -270,7 +299,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         case set_rle:
             {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
                 size_t litSize, lhSize;
-                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
                 switch(lhlCode)
                 {
                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -279,16 +308,17 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 case 1:
                     lhSize = 2;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 3");
                     litSize = MEM_readLE16(istart) >> 4;
                     break;
                 case 3:
                     lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 4");
                     litSize = MEM_readLE24(istart) >> 4;
-                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
                     break;
                 }
                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
-                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
                 RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
                 if (dctx->litBufferLocation == ZSTD_split)
@@ -310,6 +340,18 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
     }
 }
 
+/* Hidden declaration for fullbench */
+size_t ZSTD_decodeLiteralsBlock_wrapper(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity);
+size_t ZSTD_decodeLiteralsBlock_wrapper(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity)
+{
+    dctx->isFrameDecompression = 0;
+    return ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, not_streaming);
+}
+
 /* Default FSE distribution tables.
  * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
  * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#default-distributions
@@ -506,14 +548,15 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
                 for (i = 8; i < n; i += 8) {
                     MEM_write64(spread + pos + i, sv);
                 }
-                pos += n;
+                assert(n>=0);
+                pos += (size_t)n;
             }
         }
         /* Now we spread those positions across the table.
-         * The benefit of doing it in two stages is that we avoid the the
+         * The benefit of doing it in two stages is that we avoid the
          * variable size inner loop, which caused lots of branch misses.
          * Now we can run through all the positions without any branch misses.
-         * We unroll the loop twice, since that is what emperically worked best.
+         * We unroll the loop twice, since that is what empirically worked best.
          */
         {
             size_t position = 0;
@@ -540,7 +583,7 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
             for (i=0; i<n; i++) {
                 tableDecode[position].baseValue = s;
                 position = (position + step) & tableMask;
-                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+                while (UNLIKELY(position > highThreshold)) position = (position + step) & tableMask;   /* lowprob area */
         }   }
         assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
     }
@@ -551,7 +594,7 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
         for (u=0; u<tableSize; u++) {
             U32 const symbol = tableDecode[u].baseValue;
             U32 const nextState = symbolNext[symbol]++;
-            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
             tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
             assert(nbAdditionalBits[symbol] < 255);
             tableDecode[u].nbAdditionalBits = nbAdditionalBits[symbol];
@@ -664,11 +707,6 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
 
     /* SeqHead */
     nbSeq = *ip++;
-    if (!nbSeq) {
-        *nbSeqPtr=0;
-        RETURN_ERROR_IF(srcSize != 1, srcSize_wrong, "");
-        return 1;
-    }
     if (nbSeq > 0x7F) {
         if (nbSeq == 0xFF) {
             RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
@@ -681,8 +719,16 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
     }
     *nbSeqPtr = nbSeq;
 
+    if (nbSeq == 0) {
+        /* No sequence : section ends immediately */
+        RETURN_ERROR_IF(ip != iend, corruption_detected,
+            "extraneous data present in the Sequences section");
+        return (size_t)(ip - istart);
+    }
+
     /* FSE table descriptors */
     RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong, ""); /* minimum possible size: 1 byte for symbol encoding types */
+    RETURN_ERROR_IF(*ip & 3, corruption_detected, ""); /* The last field, Reserved, must be all-zeroes. */
     {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
         symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
         symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
@@ -829,7 +875,7 @@ static void ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, pt
 /* ZSTD_safecopyDstBeforeSrc():
  * This version allows overlap with dst before src, or handles the non-overlap case with dst after src
  * Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
-static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length) {
+static void ZSTD_safecopyDstBeforeSrc(BYTE* op, const BYTE* ip, ptrdiff_t length) {
     ptrdiff_t const diff = op - ip;
     BYTE* const oend = op + length;
 
@@ -858,6 +904,7 @@ static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length
  * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
  */
 FORCE_NOINLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_execSequenceEnd(BYTE* op,
     BYTE* const oend, seq_t sequence,
     const BYTE** litPtr, const BYTE* const litLimit,
@@ -905,6 +952,7 @@ size_t ZSTD_execSequenceEnd(BYTE* op,
  * This version is intended to be used during instances where the litBuffer is still split.  It is kept separate to avoid performance impact for the good case.
  */
 FORCE_NOINLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
     BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
     const BYTE** litPtr, const BYTE* const litLimit,
@@ -950,6 +998,7 @@ size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
 }
 
 HINT_INLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_execSequence(BYTE* op,
     BYTE* const oend, seq_t sequence,
     const BYTE** litPtr, const BYTE* const litLimit,
@@ -964,6 +1013,11 @@ size_t ZSTD_execSequence(BYTE* op,
 
     assert(op != NULL /* Precondition */);
     assert(oend_w < oend /* No underflow */);
+
+#if defined(__aarch64__)
+    /* prefetch sequence starting from match that will be used for copy later */
+    PREFETCH_L1(match);
+#endif
     /* Handle edge cases in a slow path:
      *   - Read beyond end of literals
      *   - Match end is within WILDCOPY_OVERLIMIT of oend
@@ -1043,6 +1097,7 @@ size_t ZSTD_execSequence(BYTE* op,
 }
 
 HINT_INLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
     BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
     const BYTE** litPtr, const BYTE* const litLimit,
@@ -1154,7 +1209,7 @@ ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16
 }
 
 /* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
- * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
+ * offset bits. But we can only read at most STREAM_ACCUMULATOR_MIN_32
  * bits before reloading. This value is the maximum number of bytes we read
  * after reloading when we are decoding long offsets.
  */
@@ -1165,13 +1220,37 @@ ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16
 
 typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
 
+/*
+ * ZSTD_decodeSequence():
+ * @p longOffsets : tells the decoder to reload more bit while decoding large offsets
+ *                  only used in 32-bit mode
+ * @return : Sequence (litL + matchL + offset)
+ */
 FORCE_INLINE_TEMPLATE seq_t
-ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, const int isLastSeq)
 {
     seq_t seq;
+    /*
+     * ZSTD_seqSymbol is a 64 bits wide structure.
+     * It can be loaded in one operation
+     * and its fields extracted by simply shifting or bit-extracting on aarch64.
+     * GCC doesn't recognize this and generates more unnecessary ldr/ldrb/ldrh
+     * operations that cause performance drop. This can be avoided by using this
+     * ZSTD_memcpy hack.
+     */
+#if defined(__aarch64__) && (defined(__GNUC__) && !defined(__clang__))
+    ZSTD_seqSymbol llDInfoS, mlDInfoS, ofDInfoS;
+    ZSTD_seqSymbol* const llDInfo = &llDInfoS;
+    ZSTD_seqSymbol* const mlDInfo = &mlDInfoS;
+    ZSTD_seqSymbol* const ofDInfo = &ofDInfoS;
+    ZSTD_memcpy(llDInfo, seqState->stateLL.table + seqState->stateLL.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(mlDInfo, seqState->stateML.table + seqState->stateML.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(ofDInfo, seqState->stateOffb.table + seqState->stateOffb.state, sizeof(ZSTD_seqSymbol));
+#else
     const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
     const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
     const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
+#endif
     seq.matchLength = mlDInfo->baseValue;
     seq.litLength = llDInfo->baseValue;
     {   U32 const ofBase = ofDInfo->baseValue;
@@ -1186,28 +1265,31 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
         U32 const llnbBits = llDInfo->nbBits;
         U32 const mlnbBits = mlDInfo->nbBits;
         U32 const ofnbBits = ofDInfo->nbBits;
+
+        assert(llBits <= MaxLLBits);
+        assert(mlBits <= MaxMLBits);
+        assert(ofBits <= MaxOff);
         /*
          * As gcc has better branch and block analyzers, sometimes it is only
-         * valuable to mark likelyness for clang, it gives around 3-4% of
+         * valuable to mark likeliness for clang, it gives around 3-4% of
          * performance.
          */
 
         /* sequence */
         {   size_t offset;
-    #if defined(__clang__)
-            if (LIKELY(ofBits > 1)) {
-    #else
             if (ofBits > 1) {
-    #endif
                 ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
                 ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
-                assert(ofBits <= MaxOff);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
                 if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
-                    U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
+                    /* Always read extra bits, this keeps the logic simple,
+                     * avoids branches, and avoids accidentally reading 0 bits.
+                     */
+                    U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
                     offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
                     BIT_reloadDStream(&seqState->DStream);
-                    if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
-                    assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
+                    offset += BIT_readBitsFast(&seqState->DStream, extraBits);
                 } else {
                     offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
                     if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
@@ -1224,7 +1306,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
                 } else {
                     offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
                     {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
-                        temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                        temp -= !temp; /* 0 is not valid: input corrupted => force offset to -1 => corruption detected at execSequence */
                         if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
                         seqState->prevOffset[1] = seqState->prevOffset[0];
                         seqState->prevOffset[0] = offset = temp;
@@ -1232,11 +1314,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
             seq.offset = offset;
         }
 
-    #if defined(__clang__)
-        if (UNLIKELY(mlBits > 0))
-    #else
         if (mlBits > 0)
-    #endif
             seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
 
         if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
@@ -1246,11 +1324,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
         /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
         ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
 
-    #if defined(__clang__)
-        if (UNLIKELY(llBits > 0))
-    #else
         if (llBits > 0)
-    #endif
             seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
 
         if (MEM_32bits())
@@ -1259,17 +1333,22 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
         DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
                     (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
 
-        ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
-        ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
-        if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-        ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
+        if (!isLastSeq) {
+            /* don't update FSE state for last Sequence */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
+            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
+            BIT_reloadDStream(&seqState->DStream);
+        }
     }
 
     return seq;
 }
 
-#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
-MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+#if DEBUGLEVEL >= 1
+static int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
 {
     size_t const windowSize = dctx->fParams.windowSize;
     /* No dictionary used. */
@@ -1283,30 +1362,33 @@ MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefix
     /* Dictionary is active. */
     return 1;
 }
+#endif
 
-MEM_STATIC void ZSTD_assertValidSequence(
+static void ZSTD_assertValidSequence(
         ZSTD_DCtx const* dctx,
         BYTE const* op, BYTE const* oend,
         seq_t const seq,
         BYTE const* prefixStart, BYTE const* virtualStart)
 {
 #if DEBUGLEVEL >= 1
-    size_t const windowSize = dctx->fParams.windowSize;
-    size_t const sequenceSize = seq.litLength + seq.matchLength;
-    BYTE const* const oLitEnd = op + seq.litLength;
-    DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
-            (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
-    assert(op <= oend);
-    assert((size_t)(oend - op) >= sequenceSize);
-    assert(sequenceSize <= ZSTD_BLOCKSIZE_MAX);
-    if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
-        size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
-        /* Offset must be within the dictionary. */
-        assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
-        assert(seq.offset <= windowSize + dictSize);
-    } else {
-        /* Offset must be within our window. */
-        assert(seq.offset <= windowSize);
+    if (dctx->isFrameDecompression) {
+        size_t const windowSize = dctx->fParams.windowSize;
+        size_t const sequenceSize = seq.litLength + seq.matchLength;
+        BYTE const* const oLitEnd = op + seq.litLength;
+        DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
+                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+        assert(op <= oend);
+        assert((size_t)(oend - op) >= sequenceSize);
+        assert(sequenceSize <= ZSTD_blockSizeMax(dctx));
+        if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
+            size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
+            /* Offset must be within the dictionary. */
+            assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
+            assert(seq.offset <= windowSize + dictSize);
+        } else {
+            /* Offset must be within our window. */
+            assert(seq.offset <= windowSize);
+        }
     }
 #else
     (void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
@@ -1322,23 +1404,21 @@ DONT_VECTORIZE
 ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
                          const void* seqStart, size_t seqSize, int nbSeq,
-                         const ZSTD_longOffset_e isLongOffset,
-                         const int frame)
+                         const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
     BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = ostart + maxDstSize;
+    BYTE* const oend = ZSTD_maybeNullPtrAdd(ostart, maxDstSize);
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* litBufferEnd = dctx->litBufferEnd;
     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
     const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer");
-    (void)frame;
+    DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer (%i seqs)", nbSeq);
 
-    /* Regen sequences */
+    /* Literals are split between internal buffer & output buffer */
     if (nbSeq) {
         seqState_t seqState;
         dctx->fseEntropy = 1;
@@ -1357,8 +1437,7 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
                 BIT_DStream_completed < BIT_DStream_overflow);
 
         /* decompress without overrunning litPtr begins */
-        {
-            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+        {   seq_t sequence = {0,0,0};  /* some static analyzer believe that @sequence is not initialized (it necessarily is, since for(;;) loop as at least one iteration) */
             /* Align the decompression loop to 32 + 16 bytes.
                 *
                 * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
@@ -1420,27 +1499,26 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
 #endif
 
             /* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
-            for (; litPtr + sequence.litLength <= dctx->litBufferEnd; ) {
-                size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+            for ( ; nbSeq; nbSeq--) {
+                sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
+                if (litPtr + sequence.litLength > dctx->litBufferEnd) break;
+                {   size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
-                assert(!ZSTD_isError(oneSeqSize));
-                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
 #endif
-                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
-                    return oneSeqSize;
-                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
-                op += oneSeqSize;
-                if (UNLIKELY(!--nbSeq))
-                    break;
-                BIT_reloadDStream(&(seqState.DStream));
-                sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
-            }
+                    if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                        return oneSeqSize;
+                    DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                    op += oneSeqSize;
+            }   }
+            DEBUGLOG(6, "reached: (litPtr + sequence.litLength > dctx->litBufferEnd)");
 
             /* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
             if (nbSeq > 0) {
                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
-                if (leftoverLit)
-                {
+                DEBUGLOG(6, "There are %i sequences left, and %zu/%zu literals left in buffer", nbSeq, leftoverLit, sequence.litLength);
+                if (leftoverLit) {
                     RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
                     ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
                     sequence.litLength -= leftoverLit;
@@ -1449,24 +1527,22 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
                 litPtr = dctx->litExtraBuffer;
                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
                 dctx->litBufferLocation = ZSTD_not_in_dst;
-                {
-                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
                     assert(!ZSTD_isError(oneSeqSize));
-                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+                    ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
 #endif
                     if (UNLIKELY(ZSTD_isError(oneSeqSize)))
                         return oneSeqSize;
                     DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
                     op += oneSeqSize;
-                    if (--nbSeq)
-                        BIT_reloadDStream(&(seqState.DStream));
                 }
+                nbSeq--;
             }
         }
 
-        if (nbSeq > 0) /* there is remaining lit from extra buffer */
-        {
+        if (nbSeq > 0) {
+            /* there is remaining lit from extra buffer */
 
 #if defined(__x86_64__)
             __asm__(".p2align 6");
@@ -1485,35 +1561,34 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
 #  endif
 #endif
 
-            for (; ; ) {
-                seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            for ( ; nbSeq ; nbSeq--) {
+                seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
                 size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
                 assert(!ZSTD_isError(oneSeqSize));
-                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+                ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
 #endif
                 if (UNLIKELY(ZSTD_isError(oneSeqSize)))
                     return oneSeqSize;
                 DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
                 op += oneSeqSize;
-                if (UNLIKELY(!--nbSeq))
-                    break;
-                BIT_reloadDStream(&(seqState.DStream));
             }
         }
 
         /* check if reached exact end */
         DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
         RETURN_ERROR_IF(nbSeq, corruption_detected, "");
-        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        DEBUGLOG(5, "bitStream : start=%p, ptr=%p, bitsConsumed=%u", seqState.DStream.start, seqState.DStream.ptr, seqState.DStream.bitsConsumed);
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
         /* save reps for next block */
         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
     }
 
     /* last literal segment */
-    if (dctx->litBufferLocation == ZSTD_split)  /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
-    {
-        size_t const lastLLSize = litBufferEnd - litPtr;
+    if (dctx->litBufferLocation == ZSTD_split) {
+        /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
+        size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        DEBUGLOG(6, "copy last literals from segment : %u", (U32)lastLLSize);
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
         if (op != NULL) {
             ZSTD_memmove(op, litPtr, lastLLSize);
@@ -1523,15 +1598,17 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
         litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
         dctx->litBufferLocation = ZSTD_not_in_dst;
     }
-    {   size_t const lastLLSize = litBufferEnd - litPtr;
+    /* copy last literals from internal buffer */
+    {   size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        DEBUGLOG(6, "copy last literals from internal buffer : %u", (U32)lastLLSize);
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
         if (op != NULL) {
             ZSTD_memcpy(op, litPtr, lastLLSize);
             op += lastLLSize;
-        }
-    }
+    }   }
 
-    return op-ostart;
+    DEBUGLOG(6, "decoded block of size %u bytes", (U32)(op - ostart));
+    return (size_t)(op - ostart);
 }
 
 FORCE_INLINE_TEMPLATE size_t
@@ -1539,21 +1616,19 @@ DONT_VECTORIZE
 ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
     void* dst, size_t maxDstSize,
     const void* seqStart, size_t seqSize, int nbSeq,
-    const ZSTD_longOffset_e isLongOffset,
-    const int frame)
+    const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
     BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ostart + maxDstSize : dctx->litBuffer;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ZSTD_maybeNullPtrAdd(ostart, maxDstSize) : dctx->litBuffer;
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* const litEnd = litPtr + dctx->litSize;
     const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
     const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
-    DEBUGLOG(5, "ZSTD_decompressSequences_body");
-    (void)frame;
+    DEBUGLOG(5, "ZSTD_decompressSequences_body: nbSeq = %d", nbSeq);
 
     /* Regen sequences */
     if (nbSeq) {
@@ -1568,11 +1643,6 @@ ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
         assert(dst != NULL);
 
-        ZSTD_STATIC_ASSERT(
-            BIT_DStream_unfinished < BIT_DStream_completed &&
-            BIT_DStream_endOfBuffer < BIT_DStream_completed &&
-            BIT_DStream_completed < BIT_DStream_overflow);
-
 #if defined(__x86_64__)
             __asm__(".p2align 6");
             __asm__("nop");
@@ -1587,73 +1657,70 @@ ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
 #  endif
 #endif
 
-        for ( ; ; ) {
-            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+        for ( ; nbSeq ; nbSeq--) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
             size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
             assert(!ZSTD_isError(oneSeqSize));
-            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+            ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
 #endif
             if (UNLIKELY(ZSTD_isError(oneSeqSize)))
                 return oneSeqSize;
             DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
             op += oneSeqSize;
-            if (UNLIKELY(!--nbSeq))
-                break;
-            BIT_reloadDStream(&(seqState.DStream));
         }
 
         /* check if reached exact end */
-        DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
-        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
-        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        assert(nbSeq == 0);
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
         /* save reps for next block */
         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
     }
 
     /* last literal segment */
-    {   size_t const lastLLSize = litEnd - litPtr;
+    {   size_t const lastLLSize = (size_t)(litEnd - litPtr);
+        DEBUGLOG(6, "copy last literals : %u", (U32)lastLLSize);
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
         if (op != NULL) {
             ZSTD_memcpy(op, litPtr, lastLLSize);
             op += lastLLSize;
-        }
-    }
+    }   }
 
-    return op-ostart;
+    DEBUGLOG(6, "decoded block of size %u bytes", (U32)(op - ostart));
+    return (size_t)(op - ostart);
 }
 
 static size_t
 ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
-                           const ZSTD_longOffset_e isLongOffset,
-                           const int frame)
+                           const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 
 static size_t
 ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
                                                void* dst, size_t maxDstSize,
                                          const void* seqStart, size_t seqSize, int nbSeq,
-                                         const ZSTD_longOffset_e isLongOffset,
-                                         const int frame)
+                                         const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
 
-FORCE_INLINE_TEMPLATE size_t
-ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
+FORCE_INLINE_TEMPLATE
+
+size_t ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
                    const BYTE* const prefixStart, const BYTE* const dictEnd)
 {
     prefetchPos += sequence.litLength;
     {   const BYTE* const matchBase = (sequence.offset > prefetchPos) ? dictEnd : prefixStart;
-        const BYTE* const match = matchBase + prefetchPos - sequence.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
-                                                                              * No consequence though : memory address is only used for prefetching, not for dereferencing */
+        /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+         * No consequence though : memory address is only used for prefetching, not for dereferencing */
+        const BYTE* const match = ZSTD_wrappedPtrSub(ZSTD_wrappedPtrAdd(matchBase, prefetchPos), sequence.offset);
         PREFETCH_L1(match); PREFETCH_L1(match+CACHELINE_SIZE);   /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
     }
     return prefetchPos + sequence.matchLength;
@@ -1668,20 +1735,18 @@ ZSTD_decompressSequencesLong_body(
                                ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
                          const void* seqStart, size_t seqSize, int nbSeq,
-                         const ZSTD_longOffset_e isLongOffset,
-                         const int frame)
+                         const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
     BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ostart + maxDstSize;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ZSTD_maybeNullPtrAdd(ostart, maxDstSize);
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* litBufferEnd = dctx->litBufferEnd;
     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
     const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    (void)frame;
 
     /* Regen sequences */
     if (nbSeq) {
@@ -1706,20 +1771,17 @@ ZSTD_decompressSequencesLong_body(
         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
 
         /* prepare in advance */
-        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
-            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+        for (seqNb=0; seqNb<seqAdvance; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, seqNb == nbSeq-1);
             prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
             sequences[seqNb] = sequence;
         }
-        RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
 
         /* decompress without stomping litBuffer */
-        for (; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb < nbSeq); seqNb++) {
-            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
-            size_t oneSeqSize;
+        for (; seqNb < nbSeq; seqNb++) {
+            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset, seqNb == nbSeq-1);
 
-            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd)
-            {
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd) {
                 /* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
                 if (leftoverLit)
@@ -1732,26 +1794,26 @@ ZSTD_decompressSequencesLong_body(
                 litPtr = dctx->litExtraBuffer;
                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
                 dctx->litBufferLocation = ZSTD_not_in_dst;
-                oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
-                assert(!ZSTD_isError(oneSeqSize));
-                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
-                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                    if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
 
-                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
-                sequences[seqNb & STORED_SEQS_MASK] = sequence;
-                op += oneSeqSize;
-            }
+                    prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                    sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                    op += oneSeqSize;
+            }   }
             else
             {
                 /* lit buffer is either wholly contained in first or second split, or not split at all*/
-                oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
                     ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
                     ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
                 assert(!ZSTD_isError(oneSeqSize));
-                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+                ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
 
@@ -1760,17 +1822,15 @@ ZSTD_decompressSequencesLong_body(
                 op += oneSeqSize;
             }
         }
-        RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected, "");
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
 
         /* finish queue */
         seqNb -= seqAdvance;
         for ( ; seqNb<nbSeq ; seqNb++) {
             seq_t *sequence = &(sequences[seqNb&STORED_SEQS_MASK]);
-            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd)
-            {
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd) {
                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
-                if (leftoverLit)
-                {
+                if (leftoverLit) {
                     RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
                     ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
                     sequence->litLength -= leftoverLit;
@@ -1779,11 +1839,10 @@ ZSTD_decompressSequencesLong_body(
                 litPtr = dctx->litExtraBuffer;
                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
                 dctx->litBufferLocation = ZSTD_not_in_dst;
-                {
-                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
                     assert(!ZSTD_isError(oneSeqSize));
-                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+                    ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
                     if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
                     op += oneSeqSize;
@@ -1796,7 +1855,7 @@ ZSTD_decompressSequencesLong_body(
                     ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
                 assert(!ZSTD_isError(oneSeqSize));
-                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+                ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
                 op += oneSeqSize;
@@ -1808,8 +1867,7 @@ ZSTD_decompressSequencesLong_body(
     }
 
     /* last literal segment */
-    if (dctx->litBufferLocation == ZSTD_split)  /* first deplete literal buffer in dst, then copy litExtraBuffer */
-    {
+    if (dctx->litBufferLocation == ZSTD_split) { /* first deplete literal buffer in dst, then copy litExtraBuffer */
         size_t const lastLLSize = litBufferEnd - litPtr;
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
         if (op != NULL) {
@@ -1827,17 +1885,16 @@ ZSTD_decompressSequencesLong_body(
         }
     }
 
-    return op-ostart;
+    return (size_t)(op - ostart);
 }
 
 static size_t
 ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
-                           const ZSTD_longOffset_e isLongOffset,
-                           const int frame)
+                           const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
 
@@ -1851,20 +1908,18 @@ DONT_VECTORIZE
 ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
-                           const ZSTD_longOffset_e isLongOffset,
-                           const int frame)
+                           const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 static BMI2_TARGET_ATTRIBUTE size_t
 DONT_VECTORIZE
 ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
-                           const ZSTD_longOffset_e isLongOffset,
-                           const int frame)
+                           const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
@@ -1873,10 +1928,9 @@ static BMI2_TARGET_ATTRIBUTE size_t
 ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
-                           const ZSTD_longOffset_e isLongOffset,
-                           const int frame)
+                           const ZSTD_longOffset_e isLongOffset)
 {
-    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
 
@@ -1886,37 +1940,34 @@ typedef size_t (*ZSTD_decompressSequences_t)(
                             ZSTD_DCtx* dctx,
                             void* dst, size_t maxDstSize,
                             const void* seqStart, size_t seqSize, int nbSeq,
-                            const ZSTD_longOffset_e isLongOffset,
-                            const int frame);
+                            const ZSTD_longOffset_e isLongOffset);
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
 static size_t
 ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
                    const void* seqStart, size_t seqSize, int nbSeq,
-                   const ZSTD_longOffset_e isLongOffset,
-                   const int frame)
+                   const ZSTD_longOffset_e isLongOffset)
 {
     DEBUGLOG(5, "ZSTD_decompressSequences");
 #if DYNAMIC_BMI2
     if (ZSTD_DCtx_get_bmi2(dctx)) {
-        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
     }
 #endif
-    return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 static size_t
 ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
                                  const void* seqStart, size_t seqSize, int nbSeq,
-                                 const ZSTD_longOffset_e isLongOffset,
-                                 const int frame)
+                                 const ZSTD_longOffset_e isLongOffset)
 {
     DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
 #if DYNAMIC_BMI2
     if (ZSTD_DCtx_get_bmi2(dctx)) {
-        return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+        return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
     }
 #endif
-    return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
@@ -1931,69 +1982,114 @@ static size_t
 ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
                              void* dst, size_t maxDstSize,
                              const void* seqStart, size_t seqSize, int nbSeq,
-                             const ZSTD_longOffset_e isLongOffset,
-                             const int frame)
+                             const ZSTD_longOffset_e isLongOffset)
 {
     DEBUGLOG(5, "ZSTD_decompressSequencesLong");
 #if DYNAMIC_BMI2
     if (ZSTD_DCtx_get_bmi2(dctx)) {
-        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
     }
 #endif
-  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
 
 
+/*
+ * @returns The total size of the history referenceable by zstd, including
+ * both the prefix and the extDict. At @p op any offset larger than this
+ * is invalid.
+ */
+static size_t ZSTD_totalHistorySize(BYTE* op, BYTE const* virtualStart)
+{
+    return (size_t)(op - virtualStart);
+}
+
+typedef struct {
+    unsigned longOffsetShare;
+    unsigned maxNbAdditionalBits;
+} ZSTD_OffsetInfo;
 
-#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
-    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-/* ZSTD_getLongOffsetsShare() :
+/* ZSTD_getOffsetInfo() :
  * condition : offTable must be valid
  * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
- *           compared to maximum possible of (1<<OffFSELog) */
-static unsigned
-ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
+ *           compared to maximum possible of (1<<OffFSELog),
+ *           as well as the maximum number additional bits required.
+ */
+static ZSTD_OffsetInfo
+ZSTD_getOffsetInfo(const ZSTD_seqSymbol* offTable, int nbSeq)
 {
-    const void* ptr = offTable;
-    U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
-    const ZSTD_seqSymbol* table = offTable + 1;
-    U32 const max = 1 << tableLog;
-    U32 u, total = 0;
-    DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
-
-    assert(max <= (1 << OffFSELog));  /* max not too large */
-    for (u=0; u<max; u++) {
-        if (table[u].nbAdditionalBits > 22) total += 1;
+    ZSTD_OffsetInfo info = {0, 0};
+    /* If nbSeq == 0, then the offTable is uninitialized, but we have
+     * no sequences, so both values should be 0.
+     */
+    if (nbSeq != 0) {
+        const void* ptr = offTable;
+        U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+        const ZSTD_seqSymbol* table = offTable + 1;
+        U32 const max = 1 << tableLog;
+        U32 u;
+        DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+        assert(max <= (1 << OffFSELog));  /* max not too large */
+        for (u=0; u<max; u++) {
+            info.maxNbAdditionalBits = MAX(info.maxNbAdditionalBits, table[u].nbAdditionalBits);
+            if (table[u].nbAdditionalBits > 22) info.longOffsetShare += 1;
+        }
+
+        assert(tableLog <= OffFSELog);
+        info.longOffsetShare <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
     }
 
-    assert(tableLog <= OffFSELog);
-    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+    return info;
+}
 
-    return total;
+/*
+ * @returns The maximum offset we can decode in one read of our bitstream, without
+ * reloading more bits in the middle of the offset bits read. Any offsets larger
+ * than this must use the long offset decoder.
+ */
+static size_t ZSTD_maxShortOffset(void)
+{
+    if (MEM_64bits()) {
+        /* We can decode any offset without reloading bits.
+         * This might change if the max window size grows.
+         */
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        return (size_t)-1;
+    } else {
+        /* The maximum offBase is (1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1.
+         * This offBase would require STREAM_ACCUMULATOR_MIN extra bits.
+         * Then we have to subtract ZSTD_REP_NUM to get the maximum possible offset.
+         */
+        size_t const maxOffbase = ((size_t)1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1;
+        size_t const maxOffset = maxOffbase - ZSTD_REP_NUM;
+        assert(ZSTD_highbit32((U32)maxOffbase) == STREAM_ACCUMULATOR_MIN);
+        return maxOffset;
+    }
 }
-#endif
 
 size_t
 ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                               void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
+                        const void* src, size_t srcSize, const streaming_operation streaming)
 {   /* blockType == blockCompressed */
     const BYTE* ip = (const BYTE*)src;
-    /* isLongOffset must be true if there are long offsets.
-     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
-     * We don't expect that to be the case in 64-bit mode.
-     * In block mode, window size is not known, so we have to be conservative.
-     * (note: but it could be evaluated from current-lowLimit)
-     */
-    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN))));
-    DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
-
-    RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (cSize : %u)", (unsigned)srcSize);
+
+    /* Note : the wording of the specification
+     * allows compressed block to be sized exactly ZSTD_blockSizeMax(dctx).
+     * This generally does not happen, as it makes little sense,
+     * since an uncompressed block would feature same size and have no decompression cost.
+     * Also, note that decoder from reference libzstd before < v1.5.4
+     * would consider this edge case as an error.
+     * As a consequence, avoid generating compressed blocks of size ZSTD_blockSizeMax(dctx)
+     * for broader compatibility with the deployed ecosystem of zstd decoders */
+    RETURN_ERROR_IF(srcSize > ZSTD_blockSizeMax(dctx), srcSize_wrong, "");
 
     /* Decode literals section */
     {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
-        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : cSize=%u, nbLiterals=%zu", (U32)litCSize, dctx->litSize);
         if (ZSTD_isError(litCSize)) return litCSize;
         ip += litCSize;
         srcSize -= litCSize;
@@ -2001,6 +2097,23 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 
     /* Build Decoding Tables */
     {
+        /* Compute the maximum block size, which must also work when !frame and fParams are unset.
+         * Additionally, take the min with dstCapacity to ensure that the totalHistorySize fits in a size_t.
+         */
+        size_t const blockSizeMax = MIN(dstCapacity, ZSTD_blockSizeMax(dctx));
+        size_t const totalHistorySize = ZSTD_totalHistorySize(ZSTD_maybeNullPtrAdd((BYTE*)dst, blockSizeMax), (BYTE const*)dctx->virtualStart);
+        /* isLongOffset must be true if there are long offsets.
+         * Offsets are long if they are larger than ZSTD_maxShortOffset().
+         * We don't expect that to be the case in 64-bit mode.
+         *
+         * We check here to see if our history is large enough to allow long offsets.
+         * If it isn't, then we can't possible have (valid) long offsets. If the offset
+         * is invalid, then it is okay to read it incorrectly.
+         *
+         * If isLongOffsets is true, then we will later check our decoding table to see
+         * if it is even possible to generate long offsets.
+         */
+        ZSTD_longOffset_e isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (totalHistorySize > ZSTD_maxShortOffset()));
         /* These macros control at build-time which decompressor implementation
          * we use. If neither is defined, we do some inspection and dispatch at
          * runtime.
@@ -2008,6 +2121,11 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
         int usePrefetchDecoder = dctx->ddictIsCold;
+#else
+        /* Set to 1 to avoid computing offset info if we don't need to.
+         * Otherwise this value is ignored.
+         */
+        int usePrefetchDecoder = 1;
 #endif
         int nbSeq;
         size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
@@ -2015,40 +2133,55 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
         ip += seqHSize;
         srcSize -= seqHSize;
 
-        RETURN_ERROR_IF(dst == NULL && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
+        RETURN_ERROR_IF((dst == NULL || dstCapacity == 0) && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
+        RETURN_ERROR_IF(MEM_64bits() && sizeof(size_t) == sizeof(void*) && (size_t)(-1) - (size_t)dst < (size_t)(1 << 20), dstSize_tooSmall,
+                "invalid dst");
 
-#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
-    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-        if ( !usePrefetchDecoder
-          && (!frame || (dctx->fParams.windowSize > (1<<24)))
-          && (nbSeq>ADVANCED_SEQS) ) {  /* could probably use a larger nbSeq limit */
-            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
-            U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
-            usePrefetchDecoder = (shareLongOffsets >= minShare);
+        /* If we could potentially have long offsets, or we might want to use the prefetch decoder,
+         * compute information about the share of long offsets, and the maximum nbAdditionalBits.
+         * NOTE: could probably use a larger nbSeq limit
+         */
+        if (isLongOffset || (!usePrefetchDecoder && (totalHistorySize > (1u << 24)) && (nbSeq > 8))) {
+            ZSTD_OffsetInfo const info = ZSTD_getOffsetInfo(dctx->OFTptr, nbSeq);
+            if (isLongOffset && info.maxNbAdditionalBits <= STREAM_ACCUMULATOR_MIN) {
+                /* If isLongOffset, but the maximum number of additional bits that we see in our table is small
+                 * enough, then we know it is impossible to have too long an offset in this block, so we can
+                 * use the regular offset decoder.
+                 */
+                isLongOffset = ZSTD_lo_isRegularOffset;
+            }
+            if (!usePrefetchDecoder) {
+                U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+                usePrefetchDecoder = (info.longOffsetShare >= minShare);
+            }
         }
-#endif
 
         dctx->ddictIsCold = 0;
 
 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-        if (usePrefetchDecoder)
+        if (usePrefetchDecoder) {
+#else
+        (void)usePrefetchDecoder;
+        {
 #endif
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
-            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
 #endif
+        }
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
         /* else */
         if (dctx->litBufferLocation == ZSTD_split)
-            return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+            return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
         else
-            return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+            return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
 #endif
     }
 }
 
 
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
 void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
 {
     if (dst != dctx->previousDstEnd && dstSize > 0) {   /* not contiguous */
@@ -2060,13 +2193,24 @@ void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
 }
 
 
-size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
-                            void* dst, size_t dstCapacity,
-                      const void* src, size_t srcSize)
+size_t ZSTD_decompressBlock_deprecated(ZSTD_DCtx* dctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
 {
     size_t dSize;
+    dctx->isFrameDecompression = 0;
     ZSTD_checkContinuity(dctx, dst, dstCapacity);
-    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0, not_streaming);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, not_streaming);
+    FORWARD_IF_ERROR(dSize, "");
     dctx->previousDstEnd = (char*)dst + dSize;
     return dSize;
 }
+
+
+/* NOTE: Must just wrap ZSTD_decompressBlock_deprecated() */
+size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    return ZSTD_decompressBlock_deprecated(dctx, dst, dstCapacity, src, srcSize);
+}
diff --git a/lib/zstd/decompress/zstd_decompress_block.h b/lib/zstd/decompress/zstd_decompress_block.h
index 3d2d57a5d25a..becffbd89364 100644
--- a/lib/zstd/decompress/zstd_decompress_block.h
+++ b/lib/zstd/decompress/zstd_decompress_block.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -47,7 +48,7 @@ typedef enum {
  */
 size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                                void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize, const int frame, const streaming_operation streaming);
+                         const void* src, size_t srcSize, const streaming_operation streaming);
 
 /* ZSTD_buildFSETable() :
  * generate FSE decoding table for one symbol (ll, ml or off)
@@ -64,5 +65,10 @@ void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
                    unsigned tableLog, void* wksp, size_t wkspSize,
                    int bmi2);
 
+/* Internal definition of ZSTD_decompressBlock() to avoid deprecation warnings. */
+size_t ZSTD_decompressBlock_deprecated(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize);
+
 
 #endif /* ZSTD_DEC_BLOCK_H */
diff --git a/lib/zstd/decompress/zstd_decompress_internal.h b/lib/zstd/decompress/zstd_decompress_internal.h
index 98102edb6a83..0f02526be774 100644
--- a/lib/zstd/decompress/zstd_decompress_internal.h
+++ b/lib/zstd/decompress/zstd_decompress_internal.h
@@ -1,5 +1,6 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -75,12 +76,13 @@ static UNUSED_ATTR const U32 ML_base[MaxML+1] = {
 
 #define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
 #define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32 ((ZSTD_BUILD_FSE_TABLE_WKSP_SIZE + sizeof(U32) - 1) / sizeof(U32))
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
 
 typedef struct {
     ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
     ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
     ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
-    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUF_decompress4X */
     U32 rep[ZSTD_REP_NUM];
     U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
 } ZSTD_entropyDTables_t;
@@ -152,6 +154,7 @@ struct ZSTD_DCtx_s
     size_t litSize;
     size_t rleSize;
     size_t staticSize;
+    int isFrameDecompression;
 #if DYNAMIC_BMI2 != 0
     int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
 #endif
@@ -164,6 +167,8 @@ struct ZSTD_DCtx_s
     ZSTD_dictUses_e dictUses;
     ZSTD_DDictHashSet* ddictSet;                    /* Hash set for multiple ddicts */
     ZSTD_refMultipleDDicts_e refMultipleDDicts;     /* User specified: if == 1, will allow references to multiple DDicts. Default == 0 (disabled) */
+    int disableHufAsm;
+    int maxBlockSizeParam;
 
     /* streaming */
     ZSTD_dStreamStage streamStage;
diff --git a/lib/zstd/decompress_sources.h b/lib/zstd/decompress_sources.h
index a06ca187aab5..8a47eb2a4514 100644
--- a/lib/zstd/decompress_sources.h
+++ b/lib/zstd/decompress_sources.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/zstd_common_module.c b/lib/zstd/zstd_common_module.c
index 22686e367e6f..466828e35752 100644
--- a/lib/zstd/zstd_common_module.c
+++ b/lib/zstd/zstd_common_module.c
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -24,9 +24,6 @@ EXPORT_SYMBOL_GPL(HUF_readStats_wksp);
 EXPORT_SYMBOL_GPL(ZSTD_isError);
 EXPORT_SYMBOL_GPL(ZSTD_getErrorName);
 EXPORT_SYMBOL_GPL(ZSTD_getErrorCode);
-EXPORT_SYMBOL_GPL(ZSTD_customMalloc);
-EXPORT_SYMBOL_GPL(ZSTD_customCalloc);
-EXPORT_SYMBOL_GPL(ZSTD_customFree);
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_DESCRIPTION("Zstd Common");
diff --git a/lib/zstd/zstd_compress_module.c b/lib/zstd/zstd_compress_module.c
index 04e1b5c01d9b..8ecf43226af2 100644
--- a/lib/zstd/zstd_compress_module.c
+++ b/lib/zstd/zstd_compress_module.c
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/lib/zstd/zstd_decompress_module.c b/lib/zstd/zstd_decompress_module.c
index f4ed952ed485..7d31518e9d5a 100644
--- a/lib/zstd/zstd_decompress_module.c
+++ b/lib/zstd/zstd_decompress_module.c
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -77,7 +77,7 @@ EXPORT_SYMBOL(zstd_init_dstream);
 
 size_t zstd_reset_dstream(zstd_dstream *dstream)
 {
-	return ZSTD_resetDStream(dstream);
+	return ZSTD_DCtx_reset(dstream, ZSTD_reset_session_only);
 }
 EXPORT_SYMBOL(zstd_reset_dstream);
 
-- 
2.44.0