From 318c40e6ac298c062db3e34a9e94e75b81d3a653 Mon Sep 17 00:00:00 2001 From: Eric Naim Date: Mon, 28 Oct 2024 10:11:08 +0800 Subject: [PATCH] bore-cachy-ext Signed-off-by: Eric Naim --- include/linux/sched.h | 20 +- include/linux/sched/bore.h | 37 ++++ init/Kconfig | 17 ++ kernel/Kconfig.hz | 43 +++++ kernel/fork.c | 5 + kernel/sched/Makefile | 1 + kernel/sched/bore.c | 381 +++++++++++++++++++++++++++++++++++++ kernel/sched/core.c | 7 + kernel/sched/debug.c | 67 ++++++- kernel/sched/fair.c | 114 +++++++++-- kernel/sched/features.h | 4 + kernel/sched/sched.h | 16 ++ 12 files changed, 694 insertions(+), 18 deletions(-) create mode 100644 include/linux/sched/bore.h create mode 100644 kernel/sched/bore.c diff --git a/include/linux/sched.h b/include/linux/sched.h index c5a7901b2580..bab2d659b667 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -537,6 +537,14 @@ struct sched_statistics { #endif /* CONFIG_SCHEDSTATS */ } ____cacheline_aligned; +#ifdef CONFIG_SCHED_BORE +struct sched_burst_cache { + u8 score; + u32 count; + u64 timestamp; +}; +#endif // CONFIG_SCHED_BORE + struct sched_entity { /* For load-balancing: */ struct load_weight load; @@ -545,12 +553,22 @@ struct sched_entity { u64 min_vruntime; struct list_head group_node; - unsigned int on_rq; + unsigned char on_rq; + unsigned char rel_deadline; u64 exec_start; u64 sum_exec_runtime; u64 prev_sum_exec_runtime; u64 vruntime; +#ifdef CONFIG_SCHED_BORE + u64 burst_time; + u8 prev_burst_penalty; + u8 curr_burst_penalty; + u8 burst_penalty; + u8 burst_score; + struct sched_burst_cache child_burst; + struct sched_burst_cache group_burst; +#endif // CONFIG_SCHED_BORE s64 vlag; u64 slice; diff --git a/include/linux/sched/bore.h b/include/linux/sched/bore.h new file mode 100644 index 000000000000..12a613a94ff0 --- /dev/null +++ b/include/linux/sched/bore.h @@ -0,0 +1,37 @@ + +#include +#include + +#ifndef _LINUX_SCHED_BORE_H +#define _LINUX_SCHED_BORE_H + +#ifdef CONFIG_SCHED_BORE +extern u8 __read_mostly sched_bore; +extern u8 __read_mostly sched_burst_exclude_kthreads; +extern u8 __read_mostly sched_burst_smoothness_long; +extern u8 __read_mostly sched_burst_smoothness_short; +extern u8 __read_mostly sched_burst_fork_atavistic; +extern u8 __read_mostly sched_burst_parity_threshold; +extern u8 __read_mostly sched_burst_penalty_offset; +extern uint __read_mostly sched_burst_penalty_scale; +extern uint __read_mostly sched_burst_cache_lifetime; +extern uint __read_mostly sched_deadline_boost_mask; + +extern void update_burst_score(struct sched_entity *se); +extern void update_burst_penalty(struct sched_entity *se); + +extern void restart_burst(struct sched_entity *se); +extern void restart_burst_rescale_deadline(struct sched_entity *se); + +extern int sched_bore_update_handler(const struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos); + +extern void sched_clone_bore( + struct task_struct *p, struct task_struct *parent, u64 clone_flags); + +extern void init_task_bore(struct task_struct *p); + +extern void reweight_entity( + struct cfs_rq *cfs_rq, struct sched_entity *se, unsigned long weight); +#endif // CONFIG_SCHED_BORE +#endif // _LINUX_SCHED_BORE_H diff --git a/init/Kconfig b/init/Kconfig index e1a88d48d652..3aea8e43c360 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1327,6 +1327,23 @@ 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. + + If unsure, say Y here. + config SCHED_AUTOGROUP bool "Automatic process group scheduling" select CGROUPS diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz index 0f78364efd4f..4cf2d88916bd 100644 --- a/kernel/Kconfig.hz +++ b/kernel/Kconfig.hz @@ -79,5 +79,48 @@ config HZ default 750 if HZ_750 default 1000 if HZ_1000 +config MIN_BASE_SLICE_NS + int "Default value for min_base_slice_ns" + default 2000000 + help + The BORE Scheduler automatically calculates the optimal base + slice for the configured HZ using the following equation: + + base_slice_ns = + 1000000000/HZ * DIV_ROUNDUP(min_base_slice_ns, 1000000000/HZ) + + This option sets the default lower bound limit of the base slice + to prevent the loss of task throughput due to overscheduling. + + Setting this value too high can cause the system to boot with + an unnecessarily large base slice, resulting in high scheduling + latency and poor system responsiveness. + +config MIGRATION_COST_BASE_NS + int "Default value for migration_cost_base_ns" + default 300000 + help + The BORE Scheduler automatically calculates the optimal + migration_cost_ns using the following equation: + + migration_cost_ns = + migration_cost_base_ns + ilog2(ncpus) * migration_cost_step_ns + + This option sets the default migration_cost_base_ns + to be used in the automatic calculation. + +config MIGRATION_COST_STEP_NS + int "Default value for migration_cost_step_ns" + default 50000 + help + The BORE Scheduler automatically calculates the optimal + migration_cost_ns using the following equation: + + migration_cost_ns = + migration_cost_base_ns + ilog2(ncpus) * migration_cost_step_ns + + This option sets the default migration_cost_step_ns + to be used in the automatic calculation. + config SCHED_HRTICK def_bool HIGH_RES_TIMERS diff --git a/kernel/fork.c b/kernel/fork.c index eb290420d926..8f060c73877b 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -116,6 +116,8 @@ #include #include +#include + #include #define CREATE_TRACE_POINTS @@ -2351,6 +2353,9 @@ __latent_entropy struct task_struct *copy_process( retval = sched_fork(clone_flags, p); if (retval) goto bad_fork_cleanup_policy; +#ifdef CONFIG_SCHED_BORE + sched_clone_bore(p, current, clone_flags); +#endif // CONFIG_SCHED_BORE retval = perf_event_init_task(p, clone_flags); if (retval) diff --git a/kernel/sched/Makefile b/kernel/sched/Makefile index 976092b7bd45..293aad675444 100644 --- a/kernel/sched/Makefile +++ b/kernel/sched/Makefile @@ -32,3 +32,4 @@ obj-y += core.o obj-y += fair.o obj-y += build_policy.o obj-y += build_utility.o +obj-y += bore.o diff --git a/kernel/sched/bore.c b/kernel/sched/bore.c new file mode 100644 index 000000000000..cd7e8a8d6075 --- /dev/null +++ b/kernel/sched/bore.c @@ -0,0 +1,381 @@ +/* + * Burst-Oriented Response Enhancer (BORE) CPU Scheduler + * Copyright (C) 2021-2024 Masahito Suzuki + */ +#include +#include +#include "sched.h" + +#ifdef CONFIG_SCHED_BORE +u8 __read_mostly sched_bore = 1; +u8 __read_mostly sched_burst_exclude_kthreads = 1; +u8 __read_mostly sched_burst_smoothness_long = 1; +u8 __read_mostly sched_burst_smoothness_short = 0; +u8 __read_mostly sched_burst_fork_atavistic = 2; +u8 __read_mostly sched_burst_parity_threshold = 2; +u8 __read_mostly sched_burst_penalty_offset = 24; +uint __read_mostly sched_burst_penalty_scale = 1280; +uint __read_mostly sched_burst_cache_lifetime = 60000000; +uint __read_mostly sched_deadline_boost_mask = ENQUEUE_INITIAL + | ENQUEUE_WAKEUP; +static int __maybe_unused sixty_four = 64; +static int __maybe_unused maxval_u8 = 255; +static int __maybe_unused maxval_12_bits = 4095; + +#define MAX_BURST_PENALTY (39U <<2) + +static inline u32 log2plus1_u64_u32f8(u64 v) { + u32 integral = fls64(v); + u8 fractional = v << (64 - integral) >> 55; + return integral << 8 | fractional; +} + +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 - tolerance)); + scaled_penalty = penalty * sched_burst_penalty_scale >> 16; + + return min(MAX_BURST_PENALTY, scaled_penalty); +} + +static inline u64 __scale_slice(u64 delta, u8 score) +{return mul_u64_u32_shr(delta, sched_prio_to_wmult[score], 22);} + +static inline u64 __unscale_slice(u64 delta, u8 score) +{return mul_u64_u32_shr(delta, sched_prio_to_weight[score], 10);} + +static void reweight_task_by_prio(struct task_struct *p, int prio) { + struct sched_entity *se = &p->se; + unsigned long weight = scale_load(sched_prio_to_weight[prio]); + + reweight_entity(cfs_rq_of(se), se, weight); + se->load.inv_weight = sched_prio_to_wmult[prio]; +} + +static inline u8 effective_prio(struct task_struct *p) { + u8 prio = p->static_prio - MAX_RT_PRIO; + if (likely(sched_bore)) + prio += p->se.burst_score; + return min(39, prio); +} + +void update_burst_score(struct sched_entity *se) { + if (!entity_is_task(se)) return; + struct task_struct *p = task_of(se); + u8 prev_prio = effective_prio(p); + + u8 burst_score = 0; + if (!((p->flags & PF_KTHREAD) && likely(sched_burst_exclude_kthreads))) + burst_score = se->burst_penalty >> 2; + se->burst_score = burst_score; + + u8 new_prio = effective_prio(p); + if (new_prio != prev_prio) + reweight_task_by_prio(p, new_prio); +} + +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); + update_burst_score(se); +} + +static inline u32 binary_smooth(u32 new, u32 old) { + int increment = new - old; + return (0 <= increment)? + old + ( increment >> (int)sched_burst_smoothness_long): + old - (-increment >> (int)sched_burst_smoothness_short); +} + +static void revolve_burst_penalty(struct sched_entity *se) { + se->prev_burst_penalty = + binary_smooth(se->curr_burst_penalty, se->prev_burst_penalty); + se->burst_time = 0; + se->curr_burst_penalty = 0; +} + +inline void restart_burst(struct sched_entity *se) { + revolve_burst_penalty(se); + se->burst_penalty = se->prev_burst_penalty; + update_burst_score(se); +} + +void restart_burst_rescale_deadline(struct sched_entity *se) { + s64 vscaled, wremain, vremain = se->deadline - se->vruntime; + struct task_struct *p = task_of(se); + u8 prev_prio = effective_prio(p); + restart_burst(se); + u8 new_prio = effective_prio(p); + if (prev_prio > new_prio) { + wremain = __unscale_slice(abs(vremain), prev_prio); + vscaled = __scale_slice(wremain, new_prio); + if (unlikely(vremain < 0)) + vscaled = -vscaled; + se->deadline = se->vruntime + vscaled; + } +} + +static inline bool task_is_bore_eligible(struct task_struct *p) +{return p->sched_class == &fair_sched_class;} + +static void reset_task_weights_bore(void) { + struct task_struct *task; + struct rq *rq; + struct rq_flags rf; + + write_lock_irq(&tasklist_lock); + for_each_process(task) { + if (!task_is_bore_eligible(task)) continue; + rq = task_rq(task); + rq_lock_irqsave(rq, &rf); + reweight_task_by_prio(task, effective_prio(task)); + rq_unlock_irqrestore(rq, &rf); + } + write_unlock_irq(&tasklist_lock); +} + +int sched_bore_update_handler(const 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; + + reset_task_weights_bore(); + + return 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 burst_cache_expired(struct sched_burst_cache *bc, u64 now) +{return (s64)(bc->timestamp + sched_burst_cache_lifetime - now) < 0;} + +static void update_burst_cache(struct sched_burst_cache *bc, + struct task_struct *p, u32 cnt, u32 sum, u64 now) { + u8 avg = cnt ? sum / cnt : 0; + bc->score = max(avg, p->se.burst_penalty); + bc->count = cnt; + bc->timestamp = now; +} + +static inline void update_child_burst_direct(struct task_struct *p, u64 now) { + u32 cnt = 0, sum = 0; + struct task_struct *child; + + list_for_each_entry(child, &p->children, sibling) { + if (!task_is_bore_eligible(child)) continue; + cnt++; + sum += child->se.burst_penalty; + } + + update_burst_cache(&p->se.child_burst, p, cnt, sum, now); +} + +static inline u8 inherit_burst_direct(struct task_struct *p, u64 now) { + struct task_struct *parent = p; + if (burst_cache_expired(&parent->se.child_burst, now)) + update_child_burst_direct(parent, now); + + return parent->se.child_burst.score; +} + +static void update_child_burst_topological( + struct task_struct *p, u64 now, u32 depth, u32 *acnt, u32 *asum) { + u32 cnt = 0, dcnt = 0, sum = 0; + struct task_struct *child, *dec; + + 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) { + if (!task_is_bore_eligible(dec)) continue; + cnt++; + sum += dec->se.burst_penalty; + continue; + } + if (!burst_cache_expired(&dec->se.child_burst, now)) { + cnt += dec->se.child_burst.count; + sum += (u32)dec->se.child_burst.score * dec->se.child_burst.count; + continue; + } + update_child_burst_topological(dec, now, depth - 1, &cnt, &sum); + } + + update_burst_cache(&p->se.child_burst, p, cnt, sum, now); + *acnt += cnt; + *asum += sum; +} + +static inline u8 inherit_burst_topological(struct task_struct *p, u64 now) { + struct task_struct *anc = p; + u32 cnt = 0, sum = 0; + + while (anc->real_parent != anc && count_child_tasks(anc) == 1) + anc = anc->real_parent; + + if (burst_cache_expired(&anc->se.child_burst, now)) + update_child_burst_topological( + anc, now, sched_burst_fork_atavistic - 1, &cnt, &sum); + + return anc->se.child_burst.score; +} + +static inline void update_tg_burst(struct task_struct *p, u64 now) { + struct task_struct *task; + u32 cnt = 0, sum = 0; + + for_each_thread(p, task) { + if (!task_is_bore_eligible(task)) continue; + cnt++; + sum += task->se.burst_penalty; + } + + update_burst_cache(&p->se.group_burst, p, cnt, sum, now); +} + +static inline u8 inherit_burst_tg(struct task_struct *p, u64 now) { + struct task_struct *parent = p->group_leader; + if (burst_cache_expired(&parent->se.group_burst, now)) + update_tg_burst(parent, now); + + return parent->se.group_burst.score; +} + +void sched_clone_bore( + struct task_struct *p, struct task_struct *parent, u64 clone_flags) { + if (!task_is_bore_eligible(p)) return; + + u64 now = ktime_get_ns(); + read_lock(&tasklist_lock); + u8 penalty = (clone_flags & CLONE_THREAD) ? + inherit_burst_tg(parent, now) : + likely(sched_burst_fork_atavistic) ? + inherit_burst_topological(parent, now): + inherit_burst_direct(parent, now); + read_unlock(&tasklist_lock); + + struct sched_entity *se = &p->se; + revolve_burst_penalty(se); + se->burst_penalty = se->prev_burst_penalty = + max(se->prev_burst_penalty, penalty); + se->child_burst.timestamp = 0; + se->group_burst.timestamp = 0; +} + +void init_task_bore(struct task_struct *p) { + p->se.burst_time = 0; + p->se.prev_burst_penalty = 0; + p->se.curr_burst_penalty = 0; + p->se.burst_penalty = 0; + p->se.burst_score = 0; + memset(&p->se.child_burst, 0, sizeof(struct sched_burst_cache)); + memset(&p->se.group_burst, 0, sizeof(struct sched_burst_cache)); +} + +#ifdef CONFIG_SYSCTL +static struct ctl_table sched_bore_sysctls[] = { + { + .procname = "sched_bore", + .data = &sched_bore, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = sched_bore_update_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_exclude_kthreads", + .data = &sched_burst_exclude_kthreads, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_smoothness_long", + .data = &sched_burst_smoothness_long, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_smoothness_short", + .data = &sched_burst_smoothness_short, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_fork_atavistic", + .data = &sched_burst_fork_atavistic, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_THREE, + }, + { + .procname = "sched_burst_parity_threshold", + .data = &sched_burst_parity_threshold, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &maxval_u8, + }, + { + .procname = "sched_burst_penalty_offset", + .data = &sched_burst_penalty_offset, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sixty_four, + }, + { + .procname = "sched_burst_penalty_scale", + .data = &sched_burst_penalty_scale, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &maxval_12_bits, + }, + { + .procname = "sched_burst_cache_lifetime", + .data = &sched_burst_cache_lifetime, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, + { + .procname = "sched_deadline_boost_mask", + .data = &sched_deadline_boost_mask, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, +}; + +static int __init sched_bore_sysctl_init(void) { + register_sysctl_init("kernel", sched_bore_sysctls); + return 0; +} +late_initcall(sched_bore_sysctl_init); +#endif // CONFIG_SYSCTL +#endif // CONFIG_SCHED_BORE diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 8ae04bd4a5a4..896db098c4c5 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -97,6 +97,8 @@ #include "../../io_uring/io-wq.h" #include "../smpboot.h" +#include + EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpu); EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpumask); @@ -8290,6 +8292,11 @@ void __init sched_init(void) BUG_ON(!sched_class_above(&ext_sched_class, &idle_sched_class)); #endif +#ifdef CONFIG_SCHED_BORE + printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 5.7.1 by Masahito Suzuki"); + init_task_bore(&init_task); +#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 c057ef46c5f8..b71ce5182500 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -167,7 +167,55 @@ static const struct file_operations sched_feat_fops = { }; #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE +#define DEFINE_SYSCTL_SCHED_FUNC(name, update_func) \ +static ssize_t sched_##name##_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos) \ +{ \ + char buf[16]; \ + unsigned int value; \ +\ + if (cnt > 15) \ + cnt = 15; \ +\ + if (copy_from_user(&buf, ubuf, cnt)) \ + return -EFAULT; \ + buf[cnt] = '\0'; \ +\ + if (kstrtouint(buf, 10, &value)) \ + return -EINVAL; \ +\ + sysctl_sched_##name = value; \ + sched_update_##update_func(); \ +\ + *ppos += cnt; \ + return cnt; \ +} \ +\ +static int sched_##name##_show(struct seq_file *m, void *v) \ +{ \ + seq_printf(m, "%d\n", sysctl_sched_##name); \ + return 0; \ +} \ +\ +static int sched_##name##_open(struct inode *inode, struct file *filp) \ +{ \ + return single_open(filp, sched_##name##_show, NULL); \ +} \ +\ +static const struct file_operations sched_##name##_fops = { \ + .open = sched_##name##_open, \ + .write = sched_##name##_write, \ + .read = seq_read, \ + .llseek = seq_lseek, \ + .release = single_release, \ +}; + +DEFINE_SYSCTL_SCHED_FUNC(min_base_slice, min_base_slice) +DEFINE_SYSCTL_SCHED_FUNC(migration_cost_base, migration_cost) +DEFINE_SYSCTL_SCHED_FUNC(migration_cost_step, migration_cost) +#undef DEFINE_SYSCTL_SCHED_FUNC +#else // !CONFIG_SCHED_BORE static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos) { @@ -213,7 +261,7 @@ static const struct file_operations sched_scaling_fops = { .llseek = seq_lseek, .release = single_release, }; - +#endif // CONFIG_SCHED_BORE #endif /* SMP */ #ifdef CONFIG_PREEMPT_DYNAMIC @@ -347,14 +395,25 @@ static __init int sched_init_debug(void) debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); #endif +#ifdef CONFIG_SCHED_BORE + debugfs_create_file("min_base_slice_ns", 0644, debugfs_sched, NULL, &sched_min_base_slice_fops); + debugfs_create_u32("base_slice_ns", 0444, debugfs_sched, &sysctl_sched_base_slice); +#else // !CONFIG_SCHED_BORE debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); +#endif // CONFIG_SCHED_BORE debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE + debugfs_create_file("migration_cost_base_ns", 0644, debugfs_sched, NULL, &sched_migration_cost_base_fops); + debugfs_create_file("migration_cost_step_ns", 0644, debugfs_sched, NULL, &sched_migration_cost_step_fops); + debugfs_create_u32("migration_cost_ns", 0444, debugfs_sched, &sysctl_sched_migration_cost); +#else // !CONFIG_SCHED_BORE debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); +#endif // CONFIG_SCHED_BORE debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); mutex_lock(&sched_domains_mutex); @@ -596,6 +655,9 @@ 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", p->se.burst_score); +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_NUMA_BALANCING SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); #endif @@ -1069,6 +1131,9 @@ void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, P(se.load.weight); #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE + P(se.burst_score); +#endif // CONFIG_SCHED_BORE P(se.avg.load_sum); P(se.avg.runnable_sum); P(se.avg.util_sum); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a36e37a674e8..bdd7366db711 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -55,6 +55,8 @@ #include "stats.h" #include "autogroup.h" +#include + /* * The initial- and re-scaling of tunables is configurable * @@ -64,28 +66,38 @@ * SCHED_TUNABLESCALING_LOG - scaled logarithmically, *1+ilog(ncpus) * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus * - * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus)) + * BORE : default SCHED_TUNABLESCALING_NONE = *1 constant + * EEVDF: default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus)) */ +#ifdef CONFIG_SCHED_BORE +unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_NONE; +#else // !CONFIG_SCHED_BORE unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG; +#endif // CONFIG_SCHED_BORE /* * Minimal preemption granularity for CPU-bound tasks: * - * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds) + * BORE : base_slice = minimum multiple of nsecs_per_tick >= min_base_slice + * (default min_base_slice = 2000000 constant, units: nanoseconds) + * EEVDF: 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 +#ifdef CONFIG_SCHED_BORE +const static uint nsecs_per_tick = 1000000000ULL / HZ; +const_debug uint sysctl_sched_min_base_slice = CONFIG_MIN_BASE_SLICE_NS; +__read_mostly uint sysctl_sched_base_slice = nsecs_per_tick; +#else // !CONFIG_SCHED_BORE unsigned int sysctl_sched_base_slice = 750000ULL; static unsigned int normalized_sysctl_sched_base_slice = 750000ULL; -#endif +#endif // CONFIG_SCHED_BORE -#ifdef CONFIG_CACHY -const_debug unsigned int sysctl_sched_migration_cost = 300000UL; -#else +#ifdef CONFIG_SCHED_BORE +const_debug uint sysctl_sched_migration_cost_base = CONFIG_MIGRATION_COST_BASE_NS; +const_debug uint sysctl_sched_migration_cost_step = CONFIG_MIGRATION_COST_STEP_NS; +__read_mostly uint sysctl_sched_migration_cost = CONFIG_MIGRATION_COST_BASE_NS; +#else // !CONFIG_SCHED_BORE const_debug unsigned int sysctl_sched_migration_cost = 500000UL; -#endif +#endif // CONFIG_SCHED_BORE static int __init setup_sched_thermal_decay_shift(char *str) { @@ -130,12 +142,8 @@ 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. */ @@ -201,6 +209,18 @@ static inline void update_load_set(struct load_weight *lw, unsigned long w) * * This idea comes from the SD scheduler of Con Kolivas: */ +#ifdef CONFIG_SCHED_BORE +static void auto_calculate_base_slice(void) { + sysctl_sched_base_slice = nsecs_per_tick * + max(1UL, DIV_ROUND_UP(sysctl_sched_min_base_slice, nsecs_per_tick)); +} +static void auto_calculate_migration_cost(void) { + sysctl_sched_migration_cost = sysctl_sched_migration_cost_base + + ilog2(num_online_cpus()) * sysctl_sched_migration_cost_step; +} +void sched_update_min_base_slice(void) { auto_calculate_base_slice(); } +void sched_update_migration_cost(void) { auto_calculate_migration_cost(); } +#else // !CONFIG_SCHED_BORE static unsigned int get_update_sysctl_factor(void) { unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8); @@ -221,15 +241,21 @@ static unsigned int get_update_sysctl_factor(void) return factor; } +#endif // CONFIG_SCHED_BORE static void update_sysctl(void) { +#ifdef CONFIG_SCHED_BORE + auto_calculate_base_slice(); + auto_calculate_migration_cost(); +#else // !CONFIG_SCHED_BORE unsigned int factor = get_update_sysctl_factor(); #define SET_SYSCTL(name) \ (sysctl_##name = (factor) * normalized_sysctl_##name) SET_SYSCTL(sched_base_slice); #undef SET_SYSCTL +#endif // CONFIG_SCHED_BORE } void __init sched_init_granularity(void) @@ -708,6 +734,9 @@ static s64 entity_lag(u64 avruntime, struct sched_entity *se) vlag = avruntime - se->vruntime; limit = calc_delta_fair(max_t(u64, 2*se->slice, TICK_NSEC), se); +#ifdef CONFIG_SCHED_BORE + limit >>= !!sched_bore; +#endif // CONFIG_SCHED_BORE return clamp(vlag, -limit, limit); } @@ -909,6 +938,10 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) * until it gets a new slice. See the HACK in set_next_entity(). */ if (sched_feat(RUN_TO_PARITY) && curr && curr->vlag == curr->deadline) +#ifdef CONFIG_SCHED_BORE + if (!(likely(sched_bore) && likely(sched_burst_parity_threshold) && + sched_burst_parity_threshold < cfs_rq->nr_running)) +#endif // CONFIG_SCHED_BORE return curr; /* Pick the leftmost entity if it's eligible */ @@ -967,6 +1000,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) * Scheduling class statistics methods: */ #ifdef CONFIG_SMP +#if !defined(CONFIG_SCHED_BORE) int sched_update_scaling(void) { unsigned int factor = get_update_sysctl_factor(); @@ -978,6 +1012,7 @@ int sched_update_scaling(void) return 0; } +#endif // CONFIG_SCHED_BORE #endif #endif @@ -1178,6 +1213,10 @@ static void update_curr(struct cfs_rq *cfs_rq) if (unlikely(delta_exec <= 0)) return; +#ifdef CONFIG_SCHED_BORE + curr->burst_time += delta_exec; + update_burst_penalty(curr); +#endif // CONFIG_SCHED_BORE curr->vruntime += calc_delta_fair(delta_exec, curr); update_deadline(cfs_rq, curr); update_min_vruntime(cfs_rq); @@ -3804,7 +3843,7 @@ static void reweight_eevdf(struct sched_entity *se, u64 avruntime, se->deadline = avruntime + vslice; } -static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, +void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, unsigned long weight) { bool curr = cfs_rq->curr == se; @@ -5212,6 +5251,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) * * EEVDF: placement strategy #1 / #2 */ +#ifdef CONFIG_SCHED_BORE + if (se->vlag) +#endif // CONFIG_SCHED_BORE if (sched_feat(PLACE_LAG) && cfs_rq->nr_running) { struct sched_entity *curr = cfs_rq->curr; unsigned long load; @@ -5282,6 +5324,16 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) se->vruntime = vruntime - lag; + if (sched_feat(PLACE_REL_DEADLINE) && se->rel_deadline) { + se->deadline += se->vruntime; + se->rel_deadline = 0; + return; + } +#ifdef CONFIG_SCHED_BORE + else if (likely(sched_bore)) + vslice >>= !!(flags & sched_deadline_boost_mask); + else +#endif // CONFIG_SCHED_BORE /* * When joining the competition; the existing tasks will be, * on average, halfway through their slice, as such start tasks @@ -5391,6 +5443,7 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq); static void dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) { + bool sleep = flags & DEQUEUE_SLEEP; int action = UPDATE_TG; if (entity_is_task(se) && task_on_rq_migrating(task_of(se))) @@ -5418,6 +5471,11 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) clear_buddies(cfs_rq, se); update_entity_lag(cfs_rq, se); + if (sched_feat(PLACE_REL_DEADLINE) && !sleep) { + se->deadline -= se->vruntime; + se->rel_deadline = 1; + } + if (se != cfs_rq->curr) __dequeue_entity(cfs_rq, se); se->on_rq = 0; @@ -6869,6 +6927,14 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) bool was_sched_idle = sched_idle_rq(rq); util_est_dequeue(&rq->cfs, p); +#ifdef CONFIG_SCHED_BORE + if (task_sleep) { + cfs_rq = cfs_rq_of(se); + if (cfs_rq->curr == se) + update_curr(cfs_rq); + restart_burst(se); + } +#endif // CONFIG_SCHED_BORE for_each_sched_entity(se) { cfs_rq = cfs_rq_of(se); @@ -8651,16 +8717,25 @@ static void yield_task_fair(struct rq *rq) /* * Are we the only task in the tree? */ +#if !defined(CONFIG_SCHED_BORE) if (unlikely(rq->nr_running == 1)) return; clear_buddies(cfs_rq, se); +#endif // CONFIG_SCHED_BORE update_rq_clock(rq); /* * Update run-time statistics of the 'current'. */ update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + restart_burst_rescale_deadline(se); + if (unlikely(rq->nr_running == 1)) + return; + + clear_buddies(cfs_rq, se); +#endif // CONFIG_SCHED_BORE /* * Tell update_rq_clock() that we've just updated, * so we don't do microscopic update in schedule() @@ -12725,6 +12800,9 @@ static void task_fork_fair(struct task_struct *p) curr = cfs_rq->curr; if (curr) update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + update_burst_score(se); +#endif // CONFIG_SCHED_BORE place_entity(cfs_rq, se, ENQUEUE_INITIAL); rq_unlock(rq, &rf); } @@ -12837,6 +12915,10 @@ static void attach_task_cfs_rq(struct task_struct *p) static void switched_from_fair(struct rq *rq, struct task_struct *p) { + p->se.rel_deadline = 0; +#ifdef CONFIG_SCHED_BORE + init_task_bore(p); +#endif // CONFIG_SCHED_BORE detach_task_cfs_rq(p); } diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 143f55df890b..e97b7b68bdd3 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -6,6 +6,10 @@ */ SCHED_FEAT(PLACE_LAG, true) SCHED_FEAT(PLACE_DEADLINE_INITIAL, true) +/* + * Preserve relative virtual deadline on 'migration'. + */ +SCHED_FEAT(PLACE_REL_DEADLINE, true) SCHED_FEAT(RUN_TO_PARITY, true) /* diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 48d893de632b..8c3fa2ffa177 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2065,7 +2065,12 @@ static inline void update_sched_domain_debugfs(void) { } static inline void dirty_sched_domain_sysctl(int cpu) { } #endif +#ifdef CONFIG_SCHED_BORE +extern void sched_update_min_base_slice(void); +extern void sched_update_migration_cost(void); +#else // !CONFIG_SCHED_BORE extern int sched_update_scaling(void); +#endif // CONFIG_SCHED_BORE static inline const struct cpumask *task_user_cpus(struct task_struct *p) { @@ -2735,9 +2740,20 @@ extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); #endif extern const_debug unsigned int sysctl_sched_nr_migrate; +#ifdef CONFIG_SCHED_BORE +extern const_debug unsigned int sysctl_sched_migration_cost_base; +extern const_debug unsigned int sysctl_sched_migration_cost_step; +extern __read_mostly unsigned int sysctl_sched_migration_cost; +#else // !CONFIG_SCHED_BORE extern const_debug unsigned int sysctl_sched_migration_cost; +#endif // CONFIG_SCHED_BORE +#ifdef CONFIG_SCHED_BORE +extern const_debug unsigned int sysctl_sched_min_base_slice; +extern __read_mostly unsigned int sysctl_sched_base_slice; +#else // !CONFIG_SCHED_BORE extern unsigned int sysctl_sched_base_slice; +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_SCHED_DEBUG extern int sysctl_resched_latency_warn_ms; -- 2.47.0