From e4245907efed9fe7a978bf4a455893de86bbaf93 Mon Sep 17 00:00:00 2001 From: Piotr Gorski Date: Thu, 25 Jan 2024 15:56:35 +0100 Subject: [PATCH] bore-cachy-rt Signed-off-by: Piotr Gorski --- include/linux/sched.h | 11 ++ init/Kconfig | 19 +++ kernel/sched/core.c | 140 +++++++++++++++++++++ kernel/sched/debug.c | 57 ++++++++- kernel/sched/fair.c | 273 +++++++++++++++++++++++++++++++++++++--- kernel/sched/features.h | 4 + kernel/sched/sched.h | 18 +++ 7 files changed, 501 insertions(+), 21 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index c02fd12b4..04a3c636c 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -559,6 +559,17 @@ struct sched_entity { 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 slice_score; + u8 child_burst; + u16 child_burst_cnt; + u64 child_burst_last_cached; + u32 slice_load; +#endif // CONFIG_SCHED_BORE s64 vlag; u64 slice; diff --git a/init/Kconfig b/init/Kconfig index 9dee4c100..49d343e97 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1278,6 +1278,25 @@ 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. + + 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 7609a0a40..128a76df8 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4508,6 +4508,135 @@ int wake_up_state(struct task_struct *p, unsigned int state) return try_to_wake_up(p, state, 0); } +#ifdef CONFIG_SCHED_BORE +extern bool sched_bore; +extern u8 sched_burst_fork_atavistic; +extern uint sched_burst_cache_lifetime; + +void __init sched_init_bore(void) { + 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; + init_task.se.slice_score = 0; + init_task.se.child_burst_last_cached = 0; + init_task.se.slice_load = 0; +} + +void inline sched_fork_bore(struct task_struct *p) { + p->se.burst_time = 0; + p->se.curr_burst_penalty = 0; + p->se.slice_score = 0; + p->se.child_burst_last_cached = 0; + p->se.slice_load = 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->se.child_burst_last_cached + sched_burst_cache_lifetime < now); +} + +static void __update_child_burst_cache( + struct task_struct *p, u32 cnt, u32 sum, u64 now) { + u8 avg = 0; + if (cnt) avg = sum / cnt; + p->se.child_burst = max(avg, p->se.burst_penalty); + p->se.child_burst_cnt = cnt; + p->se.child_burst_last_cached = now; +} + +static inline void update_child_burst_direct(struct task_struct *p, u64 now) { + struct task_struct *child; + u32 cnt = 0; + u32 sum = 0; + + list_for_each_entry(child, &p->children, sibling) { + if (child->sched_class != &fair_sched_class) continue; + cnt++; + sum += child->se.burst_penalty; + } + + __update_child_burst_cache(p, cnt, sum, now); +} + +static inline u8 __inherit_burst_direct(struct task_struct *p, u64 now) { + struct task_struct *parent = p->real_parent; + if (child_burst_cache_expired(parent, now)) + update_child_burst_direct(parent, now); + + return parent->se.child_burst; +} + +static inline void update_child_burst_topological( + 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) { + if (dec->sched_class != &fair_sched_class) continue; + cnt++; + sum += dec->se.burst_penalty; + continue; + } + if (!child_burst_cache_expired(dec, now)) { + cnt += dec->se.child_burst_cnt; + sum += (u32)dec->se.child_burst * dec->se.child_burst_cnt; + continue; + } + update_child_burst_topological(dec, now, depth - 1, &cnt, &sum); + } + + __update_child_burst_cache(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->real_parent; + u32 cnt = 0, sum = 0; + + while (anc->real_parent != anc && count_child_tasks(anc) == 1) + anc = anc->real_parent; + + if (child_burst_cache_expired(anc, now)) + update_child_burst_topological( + anc, now, sched_burst_fork_atavistic - 1, &cnt, &sum); + + return anc->se.child_burst; +} + +static inline void inherit_burst(struct task_struct *p) { + u8 burst_cache; + u64 now = ktime_get_ns(); + + read_lock(&tasklist_lock); + burst_cache = likely(sched_burst_fork_atavistic)? + __inherit_burst_topological(p, now): + __inherit_burst_direct(p, now); + read_unlock(&tasklist_lock); + + p->se.prev_burst_penalty = max(p->se.prev_burst_penalty, burst_cache); +} + +static inline void sched_post_fork_bore(struct task_struct *p) { + if (p->sched_class == &fair_sched_class && likely(sched_bore)) + inherit_burst(p); + p->se.burst_penalty = p->se.prev_burst_penalty; +} +#endif // CONFIG_SCHED_BORE + /* * Perform scheduler related setup for a newly forked process p. * p is forked by current. @@ -4524,6 +4653,9 @@ 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; p->se.slice = sysctl_sched_base_slice; INIT_LIST_HEAD(&p->se.group_node); @@ -4843,6 +4975,9 @@ 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 + sched_post_fork_bore(p); +#endif // CONFIG_SCHED_BORE uclamp_post_fork(p); } @@ -9995,6 +10130,11 @@ 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 4.1.3 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 63e19b89c..7aa951e73 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -167,7 +167,52 @@ static const struct file_operations sched_feat_fops = { }; #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE +static ssize_t sched_min_base_slice_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; + if (!value) + return -EINVAL; + + sysctl_sched_min_base_slice = value; + sched_update_min_base_slice(); + + *ppos += cnt; + return cnt; +} + +static int sched_min_base_slice_show(struct seq_file *m, void *v) +{ + seq_printf(m, "%d\n", sysctl_sched_min_base_slice); + return 0; +} + +static int sched_min_base_slice_open(struct inode *inode, struct file *filp) +{ + return single_open(filp, sched_min_base_slice_show, NULL); +} + +static const struct file_operations sched_min_base_slice_fops = { + .open = sched_min_base_slice_open, + .write = sched_min_base_slice_write, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; +#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 +258,7 @@ static const struct file_operations sched_scaling_fops = { .llseek = seq_lseek, .release = single_release, }; - +#endif // CONFIG_SCHED_BORE #endif /* SMP */ #ifdef CONFIG_PREEMPT_DYNAMIC @@ -364,13 +409,20 @@ 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", 0400, 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 +#if !defined(CONFIG_SCHED_BORE) debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); +#endif // CONFIG_SCHED_BORE debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); @@ -614,6 +666,9 @@ 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.slice_score); +#endif // CONFIG_SCHED_BORE #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 988c68097..6210bd8e6 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) 2021-2024 Masahito Suzuki */ #include #include @@ -66,17 +69,29 @@ * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *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 default: max(1 sec / HZ, min_base_slice) constant, units: nanoseconds) + * (EEVDF default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds) */ +#ifdef CONFIG_SCHED_BORE +unsigned int sysctl_sched_base_slice = 1000000000ULL / HZ; +static unsigned int configured_sched_base_slice = 1000000000ULL / HZ; +unsigned int sysctl_sched_min_base_slice = 2000000ULL; +#else // CONFIG_SCHED_BORE unsigned int sysctl_sched_base_slice = 750000ULL; static unsigned int normalized_sysctl_sched_base_slice = 750000ULL; +#endif // CONFIG_SCHED_BORE /* * After fork, child runs first. If set to 0 (default) then @@ -86,6 +101,97 @@ unsigned int sysctl_sched_child_runs_first __read_mostly; const_debug unsigned int sysctl_sched_migration_cost = 500000UL; +#ifdef CONFIG_SCHED_BORE +bool __read_mostly sched_bore = 1; +bool __read_mostly sched_burst_score_rounding = 0; +bool __read_mostly sched_burst_smoothness_long = 1; +bool __read_mostly sched_burst_smoothness_short = 0; +u8 __read_mostly sched_burst_fork_atavistic = 2; +u8 __read_mostly sched_burst_penalty_offset = 22; +uint __read_mostly sched_burst_penalty_scale = 1280; +uint __read_mostly sched_burst_cache_lifetime = 60000000; +static u8 sixty_four = 64; +static uint maxval_12_bits = 4095; + +#define MAX_BURST_PENALTY (39U <<2) + +static inline u32 log2plus1_u64_u32f8(u64 v) { + u32 msb = fls64(v); + s32 excess_bits = msb - 9; + u8 fractional = (0 <= excess_bits)? v >> excess_bits: v << -excess_bits; + return msb << 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 - (s32)tolerance); + scaled_penalty = penalty * sched_burst_penalty_scale >> 16; + + return min(MAX_BURST_PENALTY, scaled_penalty); +} + +static inline 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 scale_slice(u64 delta, struct sched_entity *se) { + return mul_u64_u32_shr(delta, sched_prio_to_wmult[se->slice_score], 22); +} + +static inline u64 __unscale_slice(u64 delta, u8 score) { + return mul_u64_u32_shr(delta, sched_prio_to_weight[score], 10); +} + +static inline u64 unscale_slice(u64 delta, struct sched_entity *se) { + return __unscale_slice(delta, se->slice_score); +} + +static void avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se); +static void avg_vruntime_sub(struct cfs_rq *cfs_rq, struct sched_entity *se); + +static void update_slice_score(struct sched_entity *se) { + struct cfs_rq *cfs_rq = cfs_rq_of(se); + u8 prev_score = se->slice_score; + u32 penalty = se->burst_penalty; + if (sched_burst_score_rounding) penalty += 0x2U; + se->slice_score = penalty >> 2; + + if (se->slice_score != prev_score && se->slice_load) { + avg_vruntime_sub(cfs_rq, se); + avg_vruntime_add(cfs_rq, 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 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; + update_slice_score(se); +} + +static inline void restart_burst_rescale_deadline(struct sched_entity *se) { + u64 wremain, vremain = se->deadline - se->vruntime; + u8 prev_score = se->slice_score; + restart_burst(se); + if (prev_score > se->slice_score) { + wremain = __unscale_slice(vremain, prev_score); + se->deadline = se->vruntime + scale_slice(wremain, se); + } +} +#endif // CONFIG_SCHED_BORE + int sched_thermal_decay_shift; static int __init setup_sched_thermal_decay_shift(char *str) { @@ -145,6 +251,70 @@ 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(bool), + .mode = 0644, + .proc_handler = &proc_dobool, + }, + { + .procname = "sched_burst_cache_lifetime", + .data = &sched_burst_cache_lifetime, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, + { + .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_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_score_rounding", + .data = &sched_burst_score_rounding, + .maxlen = sizeof(bool), + .mode = 0644, + .proc_handler = &proc_dobool, + }, + { + .procname = "sched_burst_smoothness_long", + .data = &sched_burst_smoothness_long, + .maxlen = sizeof(bool), + .mode = 0644, + .proc_handler = &proc_dobool, + }, + { + .procname = "sched_burst_smoothness_short", + .data = &sched_burst_smoothness_short, + .maxlen = sizeof(bool), + .mode = 0644, + .proc_handler = &proc_dobool, + }, +#endif // CONFIG_SCHED_BORE { .procname = "sched_child_runs_first", .data = &sysctl_sched_child_runs_first, @@ -210,6 +380,13 @@ 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 update_sysctl(void) { + sysctl_sched_base_slice = + max(sysctl_sched_min_base_slice, configured_sched_base_slice); +} +void sched_update_min_base_slice(void) { update_sysctl(); } +#else // CONFIG_SCHED_BORE static unsigned int get_update_sysctl_factor(void) { unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8); @@ -240,6 +417,7 @@ static void update_sysctl(void) SET_SYSCTL(sched_base_slice); #undef SET_SYSCTL } +#endif // CONFIG_SCHED_BORE void __init sched_init_granularity(void) { @@ -313,6 +491,9 @@ static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se) 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 = scale_slice(delta, se); +#endif // CONFIG_SCHED_BORE return delta; } @@ -635,10 +816,27 @@ static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) * * As measured, the max (key * weight) value was ~44 bits for a kernel build. */ +#if !defined(CONFIG_SCHED_BORE) +#define entity_weight(se) scale_load_down(se->load.weight) +#else // CONFIG_SCHED_BORE +static unsigned long entity_weight(struct sched_entity *se) { + unsigned long weight = se->load.weight >> SCHED_AVG_LOAD_SHIFT; + if (likely(weight)) { + weight >>= SCHED_AVG_LOAD_SHIFT; + if (likely(sched_bore)) weight = unscale_slice(weight, se); + weight = max(2UL, weight); + } + return weight; +} +#endif // CONFIG_SCHED_BORE + static void avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se) { - unsigned long weight = scale_load_down(se->load.weight); + unsigned long weight = entity_weight(se); +#ifdef CONFIG_SCHED_BORE + se->slice_load = weight; +#endif // CONFIG_SCHED_BORE s64 key = entity_key(cfs_rq, se); cfs_rq->avg_vruntime += key * weight; @@ -648,7 +846,13 @@ avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se) static void avg_vruntime_sub(struct cfs_rq *cfs_rq, struct sched_entity *se) { - unsigned long weight = scale_load_down(se->load.weight); + unsigned long weight; +#if !defined(CONFIG_SCHED_BORE) + weight = scale_load_down(se->load.weight); +#else // CONFIG_SCHED_BORE + weight = se->slice_load; + se->slice_load = 0; +#endif // CONFIG_SCHED_BORE s64 key = entity_key(cfs_rq, se); cfs_rq->avg_vruntime -= key * weight; @@ -668,14 +872,14 @@ void avg_vruntime_update(struct cfs_rq *cfs_rq, s64 delta) * Specifically: avg_runtime() + 0 must result in entity_eligible() := true * For this to be so, the result of this function must have a left bias. */ -u64 avg_vruntime(struct cfs_rq *cfs_rq) +static u64 avg_key(struct cfs_rq *cfs_rq) { struct sched_entity *curr = cfs_rq->curr; s64 avg = cfs_rq->avg_vruntime; long load = cfs_rq->avg_load; if (curr && curr->on_rq) { - unsigned long weight = scale_load_down(curr->load.weight); + unsigned long weight = entity_weight(curr); avg += entity_key(cfs_rq, curr) * weight; load += weight; @@ -688,7 +892,11 @@ u64 avg_vruntime(struct cfs_rq *cfs_rq) avg = div_s64(avg, load); } - return cfs_rq->min_vruntime + avg; + return avg; +} + +inline u64 avg_vruntime(struct cfs_rq *cfs_rq) { + return cfs_rq->min_vruntime + avg_key(cfs_rq); } /* @@ -709,13 +917,8 @@ u64 avg_vruntime(struct cfs_rq *cfs_rq) */ static void update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se) { - s64 lag, limit; - SCHED_WARN_ON(!se->on_rq); - lag = avg_vruntime(cfs_rq) - se->vruntime; - - limit = calc_delta_fair(max_t(u64, 2*se->slice, TICK_NSEC), se); - se->vlag = clamp(lag, -limit, limit); + se->vlag = avg_vruntime(cfs_rq) - se->vruntime; } /* @@ -742,7 +945,7 @@ int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se) long load = cfs_rq->avg_load; if (curr && curr->on_rq) { - unsigned long weight = scale_load_down(curr->load.weight); + unsigned long weight = entity_weight(curr); avg += entity_key(cfs_rq, curr) * weight; load += weight; @@ -996,6 +1199,7 @@ 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(); @@ -1007,6 +1211,7 @@ int sched_update_scaling(void) return 0; } +#endif // CONFIG_SCHED_BORE #endif #endif @@ -1033,6 +1238,9 @@ static void update_deadline(struct cfs_rq *cfs_rq, struct sched_entity *se, bool /* * EEVDF: vd_i = ve_i + r_i / w_i */ +#ifdef CONFIG_SCHED_BORE + update_slice_score(se); +#endif // CONFIG_SCHED_BORE se->deadline = se->vruntime + calc_delta_fair(se->slice, se); /* @@ -1184,7 +1392,11 @@ static void __update_curr(struct cfs_rq *cfs_rq, bool tick) 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, tick); update_min_vruntime(cfs_rq); @@ -5096,7 +5308,12 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) struct sched_entity *curr = cfs_rq->curr; unsigned long load; - lag = se->vlag; + u64 limit = calc_delta_fair(max_t(u64, se->slice*2, TICK_NSEC), se); + s64 overmet = limit, undermet = limit; +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) overmet = div_s64(overmet, 2); +#endif // CONFIG_SCHED_BORE + lag = clamp(se->vlag, -overmet, undermet); /* * If we want to place a task and preserve lag, we have to @@ -5152,9 +5369,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) */ load = cfs_rq->avg_load; if (curr && curr->on_rq) - load += scale_load_down(curr->load.weight); + load += entity_weight(curr); - lag *= load + scale_load_down(se->load.weight); + lag *= load + entity_weight(se); if (WARN_ON_ONCE(!load)) load = 1; lag = div_s64(lag, load); @@ -6714,6 +6931,12 @@ 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) { + update_curr(cfs_rq_of(se)); + restart_burst(se); + } +#endif // CONFIG_SCHED_BORE for_each_sched_entity(se) { cfs_rq = cfs_rq_of(se); @@ -8445,16 +8668,21 @@ static void yield_task_fair(struct rq *rq) /* * Are we the only task in the tree? */ +#ifdef CONFIG_SCHED_BORE + if (unlikely(!sched_bore)) +#endif // CONFIG_SCHED_BORE if (unlikely(rq->nr_running == 1)) return; - clear_buddies(cfs_rq, se); - 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; +#endif // CONFIG_SCHED_BORE /* * Tell update_rq_clock() that we've just updated, * so we don't do microscopic update in schedule() @@ -8462,6 +8690,8 @@ static void yield_task_fair(struct rq *rq) */ rq_clock_skip_update(rq); + clear_buddies(cfs_rq, se); + se->deadline += calc_delta_fair(se->slice, se); } @@ -12531,6 +12761,9 @@ static void task_fork_fair(struct task_struct *p) curr = cfs_rq->curr; if (curr) update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + update_slice_score(se); +#endif // CONFIG_SCHED_BORE place_entity(cfs_rq, se, ENQUEUE_INITIAL); rq_unlock(rq, &rf); } diff --git a/kernel/sched/features.h b/kernel/sched/features.h index dd8b35f67..dbac533df 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -6,7 +6,11 @@ */ SCHED_FEAT(PLACE_LAG, true) SCHED_FEAT(PLACE_DEADLINE_INITIAL, true) +#ifdef CONFIG_SCHED_BORE +SCHED_FEAT(RUN_TO_PARITY, false) +#else // CONFIG_SCHED_BORE SCHED_FEAT(RUN_TO_PARITY, true) +#endif // CONFIG_SCHED_BORE /* * Prefer to schedule the task we woke last (assuming it failed diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index c4307934a..1f5b9b96d 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -155,6 +155,17 @@ extern int sched_rr_timeslice; # define scale_load_down(w) (w) #endif +#ifdef CONFIG_SCHED_BORE +# ifdef CONFIG_64BIT +# define SCHED_AVG_LOAD_EXTRA_RESOLUTION 5 +# define SCHED_AVG_LOAD_SHIFT \ + (SCHED_FIXEDPOINT_SHIFT - SCHED_AVG_LOAD_EXTRA_RESOLUTION) +# else // CONFIG_64BIT +# define SCHED_AVG_LOAD_EXTRA_RESOLUTION 0 +# define SCHED_AVG_LOAD_SHIFT 0 +# endif // CONFIG_64BIT +#endif // CONFIG_SCHED_BORE + /* * Task weight (visible to users) and its load (invisible to users) have * independent resolution, but they should be well calibrated. We use @@ -1945,7 +1956,11 @@ static inline void dirty_sched_domain_sysctl(int cpu) } #endif +#ifdef CONFIG_SCHED_BORE +extern void sched_update_min_base_slice(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) { @@ -2526,6 +2541,9 @@ extern const_debug unsigned int sysctl_sched_nr_migrate; extern const_debug unsigned int sysctl_sched_migration_cost; extern unsigned int sysctl_sched_base_slice; +#ifdef CONFIG_SCHED_BORE +extern unsigned int sysctl_sched_min_base_slice; +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_SCHED_DEBUG extern int sysctl_resched_latency_warn_ms; -- 2.43.0.232.ge79552d197