linux-pikaos-template/patches/0002-eevdfbore.patch

From c92de794ee60ef526ca33cfee59e96f0d95b0697 Mon Sep 17 00:00:00 2001
From: Piotr Gorski <lucjan.lucjanov@gmail.com>
Date: Sun, 17 Sep 2023 22:51:20 +0200
Subject: [PATCH] bore-eevdf

Signed-off-by: Piotr Gorski <lucjan.lucjanov@gmail.com>
---
 include/linux/sched.h |  31 ++++++++
 init/Kconfig          |  20 +++++
 kernel/sched/core.c   | 160 +++++++++++++++++++++++++++++++++++++++
 kernel/sched/debug.c  |   3 +
 kernel/sched/fair.c   | 169 +++++++++++++++++++++++++++++++++++++++++-
 5 files changed, 381 insertions(+), 2 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index c940c4dc8..e88251927 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -545,6 +545,24 @@ struct sched_statistics {
 #endif /* CONFIG_SCHEDSTATS */
 } ____cacheline_aligned;
 
+#ifdef CONFIG_SCHED_BORE
+typedef union {
+	u16	u16;
+	s16	s16;
+	u8	u8[2];
+	s8	s8[2];
+} x16;
+
+typedef union {
+	u32	u32;
+	s32	s32;
+	u16	u16[2];
+	s16	s16[2];
+	u8	u8[4];
+	s8	s8[4];
+} x32;
+#endif // CONFIG_SCHED_BORE
+
 struct sched_entity {
 	/* For load-balancing: */
 	struct load_weight		load;
@@ -559,6 +577,12 @@ struct sched_entity {
 	u64				sum_exec_runtime;
 	u64				prev_sum_exec_runtime;
 	u64				vruntime;
+#ifdef CONFIG_SCHED_BORE
+	u64				burst_time;
+	u16				prev_burst_penalty;
+	u16				curr_burst_penalty;
+	u16				burst_penalty;
+#endif // CONFIG_SCHED_BORE
 	s64				vlag;
 	u64				slice;
 
@@ -990,6 +1014,13 @@ struct task_struct {
 	struct list_head		children;
 	struct list_head		sibling;
 	struct task_struct		*group_leader;
+#ifdef CONFIG_SCHED_BORE
+	u16	child_burst_cache;
+	u16	child_burst_count_cache;
+	u64	child_burst_last_cached;
+	u16	group_burst_cache;
+	u64	group_burst_last_cached;
+#endif // CONFIG_SCHED_BORE
 
 	/*
 	 * 'ptraced' is the list of tasks this task is using ptrace() on.
diff --git a/init/Kconfig b/init/Kconfig
index 25193a9d5..acc02af18 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1278,6 +1278,26 @@ config CHECKPOINT_RESTORE
 
 	  If unsure, say N here.
 
+config SCHED_BORE
+	bool "Burst-Oriented Response Enhancer"
+	default y
+	help
+	  In Desktop and Mobile computing, one might prefer interactive
+	  tasks to keep responsive no matter what they run in the background.
+
+	  Enabling this kernel feature modifies the scheduler to discriminate
+	  tasks by their burst time (runtime since it last went sleeping or
+	  yielding state) and prioritize those that run less bursty.
+	  Such tasks usually include window compositor, widgets backend,
+	  terminal emulator, video playback, games and so on.
+	  With a little impact to scheduling fairness, it may improve
+	  responsiveness especially under heavy background workload.
+
+	  You can turn it off by setting the sysctl kernel.sched_bore = 0.
+	  Enabling this feature implies NO_GENTLE_FAIR_SLEEPERS by default.
+
+	  If unsure say Y here.
+
 config SCHED_AUTOGROUP
 	bool "Automatic process group scheduling"
 	select CGROUPS
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index aff81e124..2cc47b723 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -4491,6 +4491,155 @@ int wake_up_state(struct task_struct *p, unsigned int state)
 	return try_to_wake_up(p, state, 0);
 }
 
+#ifdef CONFIG_SCHED_BORE
+extern unsigned int sched_burst_cache_lifetime;
+extern unsigned int sched_bore;
+extern unsigned int sched_burst_fork_atavistic;
+
+void __init sched_init_bore(void) {
+	init_task.child_burst_cache = 0;
+	init_task.child_burst_count_cache = 0;
+	init_task.child_burst_last_cached = 0;
+	init_task.group_burst_cache = 0;
+	init_task.group_burst_last_cached = 0;
+	init_task.se.burst_time = 0;
+	init_task.se.prev_burst_penalty = 0;
+	init_task.se.curr_burst_penalty = 0;
+	init_task.se.burst_penalty = 0;
+}
+
+void inline sched_fork_bore(struct task_struct *p) {
+	p->child_burst_cache = 0;
+	p->child_burst_count_cache = 0;
+	p->child_burst_last_cached = 0;
+	p->group_burst_cache = 0;
+	p->group_burst_last_cached = 0;
+	p->se.burst_time = 0;
+	p->se.curr_burst_penalty = 0;
+}
+
+static u32 count_child_tasks(struct task_struct *p) {
+	struct task_struct *child;
+	u32 cnt = 0;
+	list_for_each_entry(child, &p->children, sibling) {cnt++;}
+	return cnt;
+}
+
+static inline bool child_burst_cache_expired(struct task_struct *p, u64 now) {
+	return (p->child_burst_last_cached + sched_burst_cache_lifetime < now);
+}
+
+static inline bool group_burst_cache_expired(struct task_struct *p, u64 now) {
+	return (p->group_burst_last_cached + sched_burst_cache_lifetime < now);
+}
+
+static void __update_child_burst_cache(
+	struct task_struct *p, u32 cnt, u32 sum, u64 now) {
+	u16 avg = 0;
+	if (cnt) avg = DIV_ROUND_CLOSEST(sum, cnt);
+	p->child_burst_cache = max(avg, p->se.burst_penalty);
+	p->child_burst_count_cache = cnt;
+	p->child_burst_last_cached = now;
+}
+
+static void update_child_burst_cache(struct task_struct *p, u64 now) {
+	struct task_struct *child;
+	u32 cnt = 0;
+	u32 sum = 0;
+
+	list_for_each_entry(child, &p->children, sibling) {
+		cnt++;
+		sum += child->se.burst_penalty;
+	}
+
+	__update_child_burst_cache(p, cnt, sum, now);
+}
+
+static void update_child_burst_cache_atavistic(
+	struct task_struct *p, u64 now, u32 depth, u32 *acnt, u32 *asum) {
+	struct task_struct *child, *dec;
+	u32 cnt = 0, dcnt = 0;
+	u32 sum = 0;
+
+	list_for_each_entry(child, &p->children, sibling) {
+		dec = child;
+		while ((dcnt = count_child_tasks(dec)) == 1)
+			dec = list_first_entry(&dec->children, struct task_struct, sibling);
+		
+		if (!dcnt || !depth) {
+			cnt++;
+			sum += dec->se.burst_penalty;
+		} else {
+			if (child_burst_cache_expired(dec, now))
+				update_child_burst_cache_atavistic(dec, now, depth - 1, &cnt, &sum);
+			else {
+				cnt += dec->child_burst_count_cache;
+				sum += (u32)dec->child_burst_cache * dec->child_burst_count_cache;
+			}
+		}
+	}
+
+	__update_child_burst_cache(p, cnt, sum, now);
+	*acnt += cnt;
+	*asum += sum;
+}
+
+static void update_group_burst_cache(struct task_struct *p, u64 now) {
+	struct task_struct *member;
+	u32 cnt = 0, sum = 0;
+	u16 avg = 0;
+
+	for_each_thread(p, member) {
+		cnt++;
+		sum += member->se.burst_penalty;
+	}
+
+	if (cnt) avg = DIV_ROUND_CLOSEST(sum, cnt);
+	p->group_burst_cache = max(avg, p->se.burst_penalty);
+	p->group_burst_last_cached = now;
+}
+
+#define forked_task_is_process(p) (p->pid == p->tgid)
+
+static void fork_burst_penalty(struct task_struct *p) {
+	struct sched_entity *se = &p->se;
+	struct task_struct *anc;
+	u64 now = ktime_get_ns();
+	u32 cnt = 0, sum = 0;
+	u16 burst_cache;
+
+	if (likely(sched_bore)) {
+		read_lock(&tasklist_lock);
+
+		if (forked_task_is_process(p)) {
+			anc = p->real_parent;
+			if (likely(sched_burst_fork_atavistic)) {
+				while ((anc->real_parent != anc) &&
+				       (count_child_tasks(anc) == 1))
+					anc = anc->real_parent;
+				if (child_burst_cache_expired(anc, now))
+					update_child_burst_cache_atavistic(
+						anc, now, sched_burst_fork_atavistic - 1, &cnt, &sum);
+			} else
+				if (child_burst_cache_expired(anc, now))
+					update_child_burst_cache(anc, now);
+
+			burst_cache = anc->child_burst_cache;
+		} else {
+			anc = p->group_leader;
+			if (group_burst_cache_expired(anc, now))
+				update_group_burst_cache(anc, now);
+			
+			burst_cache = anc->group_burst_cache;
+		}
+
+		read_unlock(&tasklist_lock);
+		se->prev_burst_penalty = max(se->prev_burst_penalty, burst_cache);
+	}
+	se->burst_penalty = se->prev_burst_penalty;
+}
+#endif // CONFIG_SCHED_BORE
+
 /*
  * Perform scheduler related setup for a newly forked process p.
  * p is forked by current.
@@ -4507,6 +4656,9 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
 	p->se.prev_sum_exec_runtime	= 0;
 	p->se.nr_migrations		= 0;
 	p->se.vruntime			= 0;
+#ifdef CONFIG_SCHED_BORE
+	sched_fork_bore(p);
+#endif // CONFIG_SCHED_BORE
 	p->se.vlag			= 0;
 	INIT_LIST_HEAD(&p->se.group_node);
 
@@ -4828,6 +4980,9 @@ void sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs)
 
 void sched_post_fork(struct task_struct *p)
 {
+#ifdef CONFIG_SCHED_BORE
+	fork_burst_penalty(p);
+#endif // CONFIG_SCHED_BORE
 	uclamp_post_fork(p);
 }
 
@@ -9954,6 +10109,11 @@ void __init sched_init(void)
 	BUG_ON(&dl_sched_class != &stop_sched_class + 1);
 #endif
 
+#ifdef CONFIG_SCHED_BORE
+	sched_init_bore();
+	printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 3.1.4 by Masahito Suzuki");
+#endif // CONFIG_SCHED_BORE
+
 	wait_bit_init();
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index e7e83181f..6ebd52247 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -594,6 +594,9 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
 
+#ifdef CONFIG_SCHED_BORE
+	SEQ_printf(m, " %2d", ((x16*)&p->se.burst_penalty)->u8[1]);
+#endif
 #ifdef CONFIG_NUMA_BALANCING
 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 #endif
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d49c0f979..dffee0766 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -19,6 +19,9 @@
  *
  *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
  *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
+ *
+ *  Burst-Oriented Response Enhancer (BORE) CPU Scheduler
+ *  Copyright (C) 2021-2023 Masahito Suzuki <firelzrd@gmail.com>
  */
 #include <linux/energy_model.h>
 #include <linux/mmap_lock.h>
@@ -86,6 +89,67 @@ unsigned int sysctl_sched_child_runs_first __read_mostly;
 
 const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
 
+#ifdef CONFIG_SCHED_BORE
+unsigned int __read_mostly sched_bore                  = 1;
+unsigned int __read_mostly sched_bore_extra_flags      = 0;
+unsigned int __read_mostly sched_burst_cache_lifetime  = 60000000;
+unsigned int __read_mostly sched_burst_penalty_offset  = 22;
+unsigned int __read_mostly sched_burst_penalty_scale   = 1366;
+unsigned int __read_mostly sched_burst_smoothness_up   = 1;
+unsigned int __read_mostly sched_burst_smoothness_down = 0;
+unsigned int __read_mostly sched_burst_fork_atavistic  = 2;
+static int three          = 3;
+static int sixty_four     = 64;
+static int maxval_12_bits = 4095;
+
+#define MAX_BURST_PENALTY ((40U << 8) - 1)
+
+static inline u32 log2plus1_u64_u32f8(u64 v) {
+	x32 result;
+	int msb = fls64(v);
+	int excess_bits = msb - 9;
+	result.u8[0] = (0 <= excess_bits)? v >> excess_bits: v << -excess_bits;
+	result.u8[1] = msb;
+	return result.u32;
+}
+
+static inline u32 calc_burst_penalty(u64 burst_time) {
+	u32 greed, tolerance, penalty, scaled_penalty;
+	
+	greed = log2plus1_u64_u32f8(burst_time);
+	tolerance = sched_burst_penalty_offset << 8;
+	penalty = max(0, (s32)greed - (s32)tolerance);
+	scaled_penalty = penalty * sched_burst_penalty_scale >> 10;
+
+	return min(MAX_BURST_PENALTY, scaled_penalty);
+}
+
+static void update_burst_penalty(struct sched_entity *se) {
+	se->curr_burst_penalty = calc_burst_penalty(se->burst_time);
+	se->burst_penalty = max(se->prev_burst_penalty, se->curr_burst_penalty);
+}
+
+static inline u64 penalty_scale(u64 delta, struct sched_entity *se, bool half) {
+	u32 score = ((x16*)&se->burst_penalty)->u8[1];
+	if (half) score >>= 1;
+	return mul_u64_u32_shr(delta, sched_prio_to_wmult[score], 22);
+}
+
+static inline u32 binary_smooth(u32 new, u32 old) {
+  int increment = new - old;
+  return (0 <= increment)?
+    old + ( increment >> sched_burst_smoothness_up):
+    old - (-increment >> sched_burst_smoothness_down);
+}
+
+static void restart_burst(struct sched_entity *se) {
+	se->burst_penalty = se->prev_burst_penalty =
+		binary_smooth(se->curr_burst_penalty, se->prev_burst_penalty);
+	se->curr_burst_penalty = 0;
+	se->burst_time = 0;
+}
+#endif // CONFIG_SCHED_BORE
+
 int sched_thermal_decay_shift;
 static int __init setup_sched_thermal_decay_shift(char *str)
 {
@@ -145,6 +209,78 @@ static unsigned int sysctl_numa_balancing_promote_rate_limit = 65536;
 
 #ifdef CONFIG_SYSCTL
 static struct ctl_table sched_fair_sysctls[] = {
+#ifdef CONFIG_SCHED_BORE
+	{
+		.procname	= "sched_bore",
+		.data		= &sched_bore,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
+	{
+		.procname	= "sched_bore_extra_flags",
+		.data		= &sched_bore_extra_flags,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
+	{
+		.procname	= "sched_burst_cache_lifetime",
+		.data		= &sched_burst_cache_lifetime,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler = proc_dointvec,
+	},
+	{
+		.procname	= "sched_burst_fork_atavistic",
+		.data		= &sched_burst_fork_atavistic,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &three,
+	},
+	{
+		.procname	= "sched_burst_penalty_offset",
+		.data		= &sched_burst_penalty_offset,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &sixty_four,
+	},
+	{
+		.procname	= "sched_burst_penalty_scale",
+		.data		= &sched_burst_penalty_scale,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &maxval_12_bits,
+	},
+	{
+		.procname	= "sched_burst_smoothness_down",
+		.data		= &sched_burst_smoothness_down,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &three,
+	},
+	{
+		.procname	= "sched_burst_smoothness_up",
+		.data		= &sched_burst_smoothness_up,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &three,
+	},
+#endif // CONFIG_SCHED_BORE
 	{
 		.procname       = "sched_child_runs_first",
 		.data           = &sysctl_sched_child_runs_first,
@@ -308,11 +444,22 @@ static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight
 /*
  * delta /= w
  */
+#ifdef CONFIG_SCHED_BORE
+#define bore_start_debit_full_penalty (sched_bore_extra_flags)
+#define calc_delta_fair_debit(delta, se) \
+        __calc_delta_fair(delta, se, !bore_start_debit_full_penalty)
+#define calc_delta_fair(delta, se) __calc_delta_fair(delta, se, false)
+static inline u64 __calc_delta_fair(u64 delta, struct sched_entity *se, bool half)
+#else // CONFIG_SCHED_BORE
 static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
+#endif // CONFIG_SCHED_BORE
 {
 	if (unlikely(se->load.weight != NICE_0_LOAD))
 		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
 
+#ifdef CONFIG_SCHED_BORE
+	if (likely(sched_bore)) delta = penalty_scale(delta, se, half);
+#endif // CONFIG_SCHED_BORE
 	return delta;
 }
 
@@ -1128,7 +1275,11 @@ static void update_curr(struct cfs_rq *cfs_rq)
 	curr->sum_exec_runtime += delta_exec;
 	schedstat_add(cfs_rq->exec_clock, delta_exec);
 
-	curr->vruntime += calc_delta_fair(delta_exec, curr);
+#ifdef CONFIG_SCHED_BORE
+	curr->burst_time += delta_exec;
+	update_burst_penalty(curr);
+#endif // CONFIG_SCHED_BORE
+	curr->vruntime += max(1ULL, calc_delta_fair(delta_exec, curr));
 	update_deadline(cfs_rq, curr);
 	update_min_vruntime(cfs_rq);
 
@@ -4926,7 +5077,11 @@ static inline void update_misfit_status(struct task_struct *p, struct rq *rq) {}
 static void
 place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 {
+#ifdef CONFIG_SCHED_BORE
+	u64 vslice = calc_delta_fair_debit(se->slice, se);
+#else // CONFIG_SCHED_BORE
 	u64 vslice = calc_delta_fair(se->slice, se);
+#endif // CONFIG_SCHED_BORE
 	u64 vruntime = avg_vruntime(cfs_rq);
 	s64 lag = 0;
 
@@ -6482,6 +6637,9 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 	util_est_dequeue(&rq->cfs, p);
 
 	for_each_sched_entity(se) {
+#ifdef CONFIG_SCHED_BORE
+		if (task_sleep) restart_burst(se);
+#endif // CONFIG_SCHED_BORE
 		cfs_rq = cfs_rq_of(se);
 		dequeue_entity(cfs_rq, se, flags);
 
@@ -8206,8 +8364,12 @@ static void yield_task_fair(struct rq *rq)
 	/*
 	 * Are we the only task in the tree?
 	 */
-	if (unlikely(rq->nr_running == 1))
+	if (unlikely(rq->nr_running == 1)) {
+#ifdef CONFIG_SCHED_BORE
+		restart_burst(se);
+#endif // CONFIG_SCHED_BORE
 		return;
+	}
 
 	clear_buddies(cfs_rq, se);
 
@@ -8216,6 +8378,9 @@ static void yield_task_fair(struct rq *rq)
 	 * Update run-time statistics of the 'current'.
 	 */
 	update_curr(cfs_rq);
+#ifdef CONFIG_SCHED_BORE
+	restart_burst(se);
+#endif // CONFIG_SCHED_BORE
 	/*
 	 * Tell update_rq_clock() that we've just updated,
 	 * so we don't do microscopic update in schedule()
-- 
2.42.0