diff --git a/patches/0002-bore-cachy.patch b/patches/0002-bore-cachy.patch index a911e12..2785cb7 100644 --- a/patches/0002-bore-cachy.patch +++ b/patches/0002-bore-cachy.patch @@ -1,6 +1,6 @@ -From 81c203e0e6bc5c0c9e53e2c81122c6387afe309e Mon Sep 17 00:00:00 2001 +From 22d040c06d6becbc23319ded3bb6471596f19ffd Mon Sep 17 00:00:00 2001 From: Eric Naim -Date: Sun, 3 Nov 2024 10:16:41 +0800 +Date: Thu, 7 Nov 2024 22:52:43 +0800 Subject: [PATCH] bore-cachy Signed-off-by: Eric Naim @@ -8,15 +8,15 @@ Signed-off-by: Eric Naim include/linux/sched.h | 17 ++ include/linux/sched/bore.h | 37 ++++ init/Kconfig | 17 ++ - kernel/Kconfig.hz | 43 +++++ + kernel/Kconfig.hz | 17 ++ 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 | 104 ++++++++-- - kernel/sched/sched.h | 16 ++ - 11 files changed, 677 insertions(+), 18 deletions(-) + kernel/sched/debug.c | 61 +++++- + kernel/sched/fair.c | 88 +++++++-- + kernel/sched/sched.h | 9 + + 11 files changed, 622 insertions(+), 18 deletions(-) create mode 100644 include/linux/sched/bore.h create mode 100644 kernel/sched/bore.c @@ -127,10 +127,10 @@ index 504e8a7c4e2a..43daffa3ee03 100644 bool "Automatic process group scheduling" select CGROUPS diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz -index 0f78364efd4f..b412ea29d0b3 100644 +index 0f78364efd4f..83a6b919ab29 100644 --- a/kernel/Kconfig.hz +++ b/kernel/Kconfig.hz -@@ -79,5 +79,48 @@ config HZ +@@ -79,5 +79,22 @@ config HZ default 750 if HZ_750 default 1000 if HZ_1000 @@ -150,40 +150,14 @@ index 0f78364efd4f..b412ea29d0b3 100644 + 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 500000 -+ 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 0 -+ 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 671b85ce818b..370cfd9886d3 100644 +index 8287afdd01d2..e4274ce3ac32 100644 --- a/kernel/fork.c +++ b/kernel/fork.c -@@ -116,6 +116,8 @@ +@@ -117,6 +117,8 @@ #include #include @@ -192,7 +166,7 @@ index 671b85ce818b..370cfd9886d3 100644 #include #define CREATE_TRACE_POINTS -@@ -2350,6 +2352,9 @@ __latent_entropy struct task_struct *copy_process( +@@ -2352,6 +2354,9 @@ __latent_entropy struct task_struct *copy_process( retval = sched_fork(clone_flags, p); if (retval) goto bad_fork_cleanup_policy; @@ -599,7 +573,7 @@ index 000000000000..cd7e8a8d6075 +#endif // CONFIG_SYSCTL +#endif // CONFIG_SCHED_BORE diff --git a/kernel/sched/core.c b/kernel/sched/core.c -index dbfb5717d6af..9aa89fa98c90 100644 +index 719e0ed1e976..02f8b495738b 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -97,6 +97,8 @@ @@ -616,7 +590,7 @@ index dbfb5717d6af..9aa89fa98c90 100644 #endif +#ifdef CONFIG_SCHED_BORE -+ printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 5.7.2 by Masahito Suzuki"); ++ printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 5.7.3 by Masahito Suzuki"); + init_task_bore(&init_task); +#endif // CONFIG_SCHED_BORE + @@ -624,10 +598,10 @@ index dbfb5717d6af..9aa89fa98c90 100644 #ifdef CONFIG_FAIR_GROUP_SCHED diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c -index f4035c7a0fa1..ddd03cd517c7 100644 +index f4035c7a0fa1..0edb3a216f5d 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c -@@ -167,7 +167,55 @@ static const struct file_operations sched_feat_fops = { +@@ -167,7 +167,53 @@ static const struct file_operations sched_feat_fops = { }; #ifdef CONFIG_SMP @@ -675,15 +649,13 @@ index f4035c7a0fa1..ddd03cd517c7 100644 +}; + +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 = { +@@ -213,7 +259,7 @@ static const struct file_operations sched_scaling_fops = { .llseek = seq_lseek, .release = single_release, }; @@ -692,7 +664,7 @@ index f4035c7a0fa1..ddd03cd517c7 100644 #endif /* SMP */ #ifdef CONFIG_PREEMPT_DYNAMIC -@@ -504,14 +552,25 @@ static __init int sched_init_debug(void) +@@ -504,13 +550,20 @@ static __init int sched_init_debug(void) debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); #endif @@ -707,18 +679,13 @@ index f4035c7a0fa1..ddd03cd517c7 100644 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 ++#if !defined(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("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); - mutex_lock(&sched_domains_mutex); -@@ -755,6 +814,9 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) +@@ -755,6 +808,9 @@ 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))); @@ -728,7 +695,7 @@ index f4035c7a0fa1..ddd03cd517c7 100644 #ifdef CONFIG_NUMA_BALANCING SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); #endif -@@ -1243,6 +1305,9 @@ void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, +@@ -1243,6 +1299,9 @@ void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, P(se.load.weight); #ifdef CONFIG_SMP @@ -739,7 +706,7 @@ index f4035c7a0fa1..ddd03cd517c7 100644 P(se.avg.runnable_sum); P(se.avg.util_sum); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c -index b55fa314df37..b8c4649c72e8 100644 +index 54e7c4c3e2c5..94c54154a175 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -55,6 +55,8 @@ @@ -751,7 +718,7 @@ index b55fa314df37..b8c4649c72e8 100644 /* * The initial- and re-scaling of tunables is configurable * -@@ -64,28 +66,38 @@ +@@ -64,28 +66,32 @@ * SCHED_TUNABLESCALING_LOG - scaled logarithmically, *1+ilog(ncpus) * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus * @@ -778,7 +745,7 @@ index b55fa314df37..b8c4649c72e8 100644 -static unsigned int normalized_sysctl_sched_base_slice = 350000ULL; -#else +#ifdef CONFIG_SCHED_BORE -+const static unsigned int nsecs_per_tick = 1000000000ULL / HZ; ++static const unsigned int nsecs_per_tick = 1000000000ULL / HZ; +unsigned int sysctl_sched_min_base_slice = CONFIG_MIN_BASE_SLICE_NS; +__read_mostly uint sysctl_sched_base_slice = nsecs_per_tick; +#else // !CONFIG_SCHED_BORE @@ -790,18 +757,12 @@ index b55fa314df37..b8c4649c72e8 100644 -#ifdef CONFIG_CACHY -const_debug unsigned int sysctl_sched_migration_cost = 300000UL; -#else -+#ifdef CONFIG_SCHED_BORE -+unsigned int sysctl_sched_migration_cost_base = CONFIG_MIGRATION_COST_BASE_NS; -+unsigned int 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) +@@ -130,12 +136,8 @@ int __weak arch_asym_cpu_priority(int cpu) * * (default: 5 msec, units: microseconds) */ @@ -814,48 +775,29 @@ index b55fa314df37..b8c4649c72e8 100644 #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) +@@ -201,6 +203,13 @@ 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 auto_calculate_base_slice(void) { ++static void update_sysctl(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(); } ++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); -@@ -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) +@@ -231,6 +240,7 @@ static void update_sysctl(void) SET_SYSCTL(sched_base_slice); #undef SET_SYSCTL -+#endif // CONFIG_SCHED_BORE } ++#endif // CONFIG_SCHED_BORE void __init sched_init_granularity(void) -@@ -708,6 +734,9 @@ static s64 entity_lag(u64 avruntime, struct sched_entity *se) + { +@@ -708,6 +718,9 @@ 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); @@ -865,7 +807,7 @@ index b55fa314df37..b8c4649c72e8 100644 return clamp(vlag, -limit, limit); } -@@ -939,6 +968,10 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) +@@ -939,6 +952,10 @@ 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) @@ -876,7 +818,7 @@ index b55fa314df37..b8c4649c72e8 100644 return curr; /* Pick the leftmost entity if it's eligible */ -@@ -997,6 +1030,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) +@@ -997,6 +1014,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) * Scheduling class statistics methods: */ #ifdef CONFIG_SMP @@ -884,7 +826,7 @@ index b55fa314df37..b8c4649c72e8 100644 int sched_update_scaling(void) { unsigned int factor = get_update_sysctl_factor(); -@@ -1008,6 +1042,7 @@ int sched_update_scaling(void) +@@ -1008,6 +1026,7 @@ int sched_update_scaling(void) return 0; } @@ -892,7 +834,7 @@ index b55fa314df37..b8c4649c72e8 100644 #endif #endif -@@ -1240,6 +1275,10 @@ static void update_curr(struct cfs_rq *cfs_rq) +@@ -1240,6 +1259,10 @@ static void update_curr(struct cfs_rq *cfs_rq) if (unlikely(delta_exec <= 0)) return; @@ -903,7 +845,7 @@ index b55fa314df37..b8c4649c72e8 100644 curr->vruntime += calc_delta_fair(delta_exec, curr); resched = update_deadline(cfs_rq, curr); update_min_vruntime(cfs_rq); -@@ -3884,7 +3923,7 @@ static void reweight_eevdf(struct sched_entity *se, u64 avruntime, +@@ -3884,7 +3907,7 @@ static void reweight_eevdf(struct sched_entity *se, u64 avruntime, se->deadline = avruntime + vslice; } @@ -912,7 +854,7 @@ index b55fa314df37..b8c4649c72e8 100644 unsigned long weight) { bool curr = cfs_rq->curr == se; -@@ -5293,6 +5332,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) +@@ -5293,6 +5316,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) * * EEVDF: placement strategy #1 / #2 */ @@ -922,7 +864,7 @@ index b55fa314df37..b8c4649c72e8 100644 if (sched_feat(PLACE_LAG) && cfs_rq->nr_running) { struct sched_entity *curr = cfs_rq->curr; unsigned long load; -@@ -5368,7 +5410,11 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) +@@ -5368,7 +5394,11 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) se->rel_deadline = 0; return; } @@ -935,7 +877,7 @@ index b55fa314df37..b8c4649c72e8 100644 /* * When joining the competition; the existing tasks will be, * on average, halfway through their slice, as such start tasks -@@ -7109,6 +7155,14 @@ static int dequeue_entities(struct rq *rq, struct sched_entity *se, int flags) +@@ -7110,6 +7140,14 @@ static int dequeue_entities(struct rq *rq, struct sched_entity *se, int flags) p = task_of(se); h_nr_running = 1; idle_h_nr_running = task_has_idle_policy(p); @@ -950,7 +892,7 @@ index b55fa314df37..b8c4649c72e8 100644 } else { cfs_rq = group_cfs_rq(se); slice = cfs_rq_min_slice(cfs_rq); -@@ -9018,16 +9072,25 @@ static void yield_task_fair(struct rq *rq) +@@ -9025,16 +9063,25 @@ static void yield_task_fair(struct rq *rq) /* * Are we the only task in the tree? */ @@ -976,7 +918,7 @@ index b55fa314df37..b8c4649c72e8 100644 /* * Tell update_rq_clock() that we've just updated, * so we don't do microscopic update in schedule() -@@ -13079,6 +13142,9 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) +@@ -13086,6 +13133,9 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) static void task_fork_fair(struct task_struct *p) { set_task_max_allowed_capacity(p); @@ -986,7 +928,7 @@ index b55fa314df37..b8c4649c72e8 100644 } /* -@@ -13189,6 +13255,10 @@ static void attach_task_cfs_rq(struct task_struct *p) +@@ -13196,6 +13246,10 @@ static void attach_task_cfs_rq(struct task_struct *p) static void switched_from_fair(struct rq *rq, struct task_struct *p) { @@ -998,33 +940,24 @@ index b55fa314df37..b8c4649c72e8 100644 } diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h -index 991226ab5250..043ee801bb19 100644 +index f610df2e0811..996c86741e28 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h -@@ -2074,7 +2074,12 @@ static inline void update_sched_domain_debugfs(void) { } +@@ -2074,7 +2074,11 @@ 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) { -@@ -2792,9 +2797,20 @@ extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); - #endif - +@@ -2824,7 +2828,12 @@ extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); extern const_debug unsigned int sysctl_sched_nr_migrate; -+#ifdef CONFIG_SCHED_BORE -+extern unsigned int sysctl_sched_migration_cost_base; -+extern unsigned int sysctl_sched_migration_cost_step; -+extern __read_mostly uint 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 unsigned int sysctl_sched_min_base_slice; @@ -1037,4 +970,3 @@ index 991226ab5250..043ee801bb19 100644 extern int sysctl_resched_latency_warn_ms; -- 2.47.0 -