aboutsummaryrefslogtreecommitdiff
path: root/SOURCES/cachy-bore.patch
diff options
context:
space:
mode:
authorJan200101 <sentrycraft123@gmail.com>2024-10-09 20:08:15 +0200
committerJan200101 <sentrycraft123@gmail.com>2024-10-09 20:08:15 +0200
commitd4a54e18ca707d4bf2ab7732fce591d598a7c15e (patch)
tree8fb0ab49c0254f4629bd12f95b69b3408d6fc685 /SOURCES/cachy-bore.patch
parent2bc81be55c381d2bfd0543f36c4eafd8517c8972 (diff)
downloadkernel-fsync-d4a54e18ca707d4bf2ab7732fce591d598a7c15e.tar.gz
kernel-fsync-d4a54e18ca707d4bf2ab7732fce591d598a7c15e.zip
kernel 6.11.2
Diffstat (limited to 'SOURCES/cachy-bore.patch')
-rw-r--r--SOURCES/cachy-bore.patch1046
1 files changed, 554 insertions, 492 deletions
diff --git a/SOURCES/cachy-bore.patch b/SOURCES/cachy-bore.patch
index 7421807..f15d6da 100644
--- a/SOURCES/cachy-bore.patch
+++ b/SOURCES/cachy-bore.patch
@@ -1,24 +1,53 @@
-From 3816495f5635104fae1dda21b743f750c2914196 Mon Sep 17 00:00:00 2001
-From: Eric Naim <dnaim@proton.me>
-Date: Sat, 3 Aug 2024 15:23:30 +0700
-Subject: [PATCH] bore
+From 2328e7500823151cb2f119b847454e63e9da0f64 Mon Sep 17 00:00:00 2001
+From: Masahito S <firelzrd@gmail.com>
+Date: Thu, 3 Oct 2024 17:33:57 +0900
+Subject: [PATCH] linux6.11.y-bore5.6.0
---
- include/linux/sched.h | 10 ++
- init/Kconfig | 17 +++
- kernel/Kconfig.hz | 16 ++
- kernel/sched/core.c | 143 ++++++++++++++++++
- kernel/sched/debug.c | 60 +++++++-
- kernel/sched/fair.c | 322 ++++++++++++++++++++++++++++++++++++++--
- kernel/sched/features.h | 22 ++-
- kernel/sched/sched.h | 7 +
- 8 files changed, 583 insertions(+), 14 deletions(-)
+ include/linux/sched.h | 20 +-
+ include/linux/sched/bore.h | 37 ++++
+ init/Kconfig | 17 ++
+ kernel/Kconfig.hz | 17 ++
+ kernel/fork.c | 5 +
+ kernel/sched/Makefile | 1 +
+ kernel/sched/bore.c | 380 +++++++++++++++++++++++++++++++++++++
+ kernel/sched/core.c | 7 +
+ kernel/sched/debug.c | 60 +++++-
+ kernel/sched/fair.c | 89 ++++++++-
+ kernel/sched/features.h | 4 +
+ kernel/sched/sched.h | 7 +
+ 12 files changed, 639 insertions(+), 5 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 76214d7c8..9f65d367b 100644
+index f8d150343d..2481cf0125 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
-@@ -547,6 +547,16 @@ struct sched_entity {
+@@ -535,6 +535,14 @@ 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;
+@@ -543,12 +551,22 @@ 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;
@@ -28,18 +57,60 @@ index 76214d7c8..9f65d367b 100644
+ u8 curr_burst_penalty;
+ u8 burst_penalty;
+ u8 burst_score;
-+ u8 child_burst;
-+ u32 child_burst_cnt;
-+ u64 child_burst_last_cached;
++ 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 0000000000..12a613a94f
+--- /dev/null
++++ b/include/linux/sched/bore.h
+@@ -0,0 +1,37 @@
++
++#include <linux/sched.h>
++#include <linux/sched/cputime.h>
++
++#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 febdea2af..171b5d995 100644
+index 5783a0b875..b648ed538c 100644
--- a/init/Kconfig
+++ b/init/Kconfig
-@@ -1283,6 +1283,23 @@ config CHECKPOINT_RESTORE
+@@ -1297,6 +1297,23 @@ config CHECKPOINT_RESTORE
If unsure, say N here.
@@ -64,10 +135,10 @@ index febdea2af..171b5d995 100644
bool "Automatic process group scheduling"
select CGROUPS
diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz
-index 38ef6d068..5f6eecd1e 100644
+index 38ef6d0688..253c566b59 100644
--- a/kernel/Kconfig.hz
+++ b/kernel/Kconfig.hz
-@@ -55,5 +55,21 @@ config HZ
+@@ -55,5 +55,22 @@ config HZ
default 300 if HZ_300
default 1000 if HZ_1000
@@ -78,7 +149,8 @@ index 38ef6d068..5f6eecd1e 100644
+ The BORE Scheduler automatically calculates the optimal base
+ slice for the configured HZ using the following equation:
+
-+ base_slice_ns = max(min_base_slice_ns, 1000000000/HZ)
++ 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.
@@ -89,32 +161,188 @@ index 38ef6d068..5f6eecd1e 100644
+
config SCHED_HRTICK
def_bool HIGH_RES_TIMERS
-diff --git a/kernel/sched/core.c b/kernel/sched/core.c
-index ebf21373f..5d1c97612 100644
---- a/kernel/sched/core.c
-+++ b/kernel/sched/core.c
-@@ -4512,6 +4512,138 @@ int wake_up_state(struct task_struct *p, unsigned int state)
- return try_to_wake_up(p, state, 0);
- }
+diff --git a/kernel/fork.c b/kernel/fork.c
+index cc760491f2..179b884da3 100644
+--- a/kernel/fork.c
++++ b/kernel/fork.c
+@@ -111,6 +111,8 @@
+ #include <asm/cacheflush.h>
+ #include <asm/tlbflush.h>
+
++#include <linux/sched/bore.h>
++
+ #include <trace/events/sched.h>
+
+ #define CREATE_TRACE_POINTS
+@@ -2344,6 +2346,9 @@ __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 976092b7bd..293aad6754 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 0000000000..62a0191a32
+--- /dev/null
++++ b/kernel/sched/bore.c
+@@ -0,0 +1,380 @@
++/*
++ * Burst-Oriented Response Enhancer (BORE) CPU Scheduler
++ * Copyright (C) 2021-2024 Masahito Suzuki <firelzrd@gmail.com>
++ */
++#include <linux/cpuset.h>
++#include <linux/sched/bore.h>
++#include "sched.h"
++
+#ifdef CONFIG_SCHED_BORE
-+extern u8 sched_burst_fork_atavistic;
-+extern uint sched_burst_cache_lifetime;
-+
-+static 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.burst_score = 0;
-+ init_task.se.child_burst_last_cached = 0;
++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;
+}
+
-+inline void sched_fork_bore(struct task_struct *p) {
-+ p->se.burst_time = 0;
-+ p->se.curr_burst_penalty = 0;
-+ p->se.burst_score = 0;
-+ p->se.child_burst_last_cached = 0;
++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 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) {
++ 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) {
@@ -124,52 +352,45 @@ index ebf21373f..5d1c97612 100644
+ return cnt;
+}
+
-+static inline bool task_is_inheritable(struct task_struct *p) {
-+ return (p->sched_class == &fair_sched_class);
-+}
++static inline bool task_is_bore_eligible(struct task_struct *p)
++{return p->sched_class == &fair_sched_class;}
+
-+static inline bool child_burst_cache_expired(struct task_struct *p, u64 now) {
-+ u64 expiration_time =
-+ p->se.child_burst_last_cached + sched_burst_cache_lifetime;
-+ return ((s64)(expiration_time - now) < 0);
-+}
++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_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 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;
-+ u32 cnt = 0;
-+ u32 sum = 0;
+
+ list_for_each_entry(child, &p->children, sibling) {
-+ if (!task_is_inheritable(child)) continue;
++ if (!task_is_bore_eligible(child)) continue;
+ cnt++;
+ sum += child->se.burst_penalty;
+ }
+
-+ __update_child_burst_cache(p, cnt, sum, now);
++ 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->real_parent;
-+ if (child_burst_cache_expired(parent, 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;
++ 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;
-+ u32 cnt = 0, dcnt = 0;
-+ u32 sum = 0;
+
+ list_for_each_entry(child, &p->children, sibling) {
+ dec = child;
@@ -177,95 +398,214 @@ index ebf21373f..5d1c97612 100644
+ dec = list_first_entry(&dec->children, struct task_struct, sibling);
+
+ if (!dcnt || !depth) {
-+ if (!task_is_inheritable(dec)) continue;
++ if (!task_is_bore_eligible(dec)) 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;
++ 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_child_burst_cache(p, cnt, sum, now);
++ 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->real_parent;
++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 (child_burst_cache_expired(anc, now))
++ 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;
++ return anc->se.child_burst.score;
+}
+
-+static inline void inherit_burst(struct task_struct *p) {
-+ u8 burst_cache;
-+ u64 now = ktime_get_ns();
++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);
-+ burst_cache = likely(sched_burst_fork_atavistic)?
-+ __inherit_burst_topological(p, now):
-+ __inherit_burst_direct(p, now);
++ 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);
+
-+ p->se.prev_burst_penalty = max(p->se.prev_burst_penalty, burst_cache);
++ 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;
+}
+
-+static void sched_post_fork_bore(struct task_struct *p) {
-+ if (p->sched_class == &fair_sched_class)
-+ inherit_burst(p);
-+ p->se.burst_penalty = p->se.prev_burst_penalty;
++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));
+}
-+#endif // CONFIG_SCHED_BORE
+
- /*
- * Perform scheduler related setup for a newly forked process p.
- * p is forked by current.
-@@ -4528,6 +4660,9 @@ 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);
++#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
- p->se.vlag = 0;
- p->se.slice = sysctl_sched_base_slice;
- INIT_LIST_HEAD(&p->se.group_node);
-@@ -4843,6 +4978,9 @@ void sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs)
+diff --git a/kernel/sched/core.c b/kernel/sched/core.c
+index f3951e4a55..9264e542c9 100644
+--- a/kernel/sched/core.c
++++ b/kernel/sched/core.c
+@@ -97,6 +97,8 @@
+ #include "../../io_uring/io-wq.h"
+ #include "../smpboot.h"
- 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);
- }
++#include <linux/sched/bore.h>
++
+ EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpu);
+ EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpumask);
-@@ -9930,6 +10068,11 @@ void __init sched_init(void)
+@@ -8197,6 +8199,11 @@ 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 5.2.8 by Masahito Suzuki");
++ printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 5.6.0 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 c1eb9a1af..e2da8d773 100644
+index c1eb9a1afd..e2da8d7738 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -167,7 +167,52 @@ static const struct file_operations sched_feat_fops = {
@@ -372,21 +712,20 @@ index c1eb9a1af..e2da8d773 100644
P(se.avg.runnable_sum);
P(se.avg.util_sum);
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
-index 483c137b9..4c8d7fbd5 100644
+index 9057584ec0..465d2626ee 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
-@@ -19,6 +19,9 @@
+@@ -55,6 +55,8 @@
+ #include "stats.h"
+ #include "autogroup.h"
+
++#include <linux/sched/bore.h>
++
+ /*
+ * The initial- and re-scaling of tunables is configurable
*
- * 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 <firelzrd@gmail.com>
- */
- #include <linux/energy_model.h>
- #include <linux/mmap_lock.h>
-@@ -64,20 +67,146 @@
- * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
+@@ -64,17 +66,29 @@
+ * SCHED_TUNABLESCALING_LOG - scaled logarithmically, *1+ilog(ncpus)
* SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
*
- * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
@@ -417,231 +756,26 @@ index 483c137b9..4c8d7fbd5 100644
const_debug unsigned int sysctl_sched_migration_cost = 500000UL;
-+#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_penalty_offset = 22;
-+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;
-+uint __read_mostly sched_deadline_preserve_mask = ENQUEUE_RESTORE
-+ | ENQUEUE_MIGRATED;
-+static int __maybe_unused sixty_four = 64;
-+static int __maybe_unused 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 u64 scale_slice(u64 delta, struct sched_entity *se) {
-+ return mul_u64_u32_shr(delta, sched_prio_to_wmult[se->burst_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->burst_score);
-+}
-+
-+static void reweight_entity(
-+ struct cfs_rq *cfs_rq, struct sched_entity *se, unsigned long weight);
-+
-+static void renice_task(struct task_struct *p, int prio)
-+{
-+ struct sched_entity *se = &p->se;
-+ struct cfs_rq *cfs_rq = cfs_rq_of(se);
-+ struct load_weight *load = &se->load;
-+ unsigned long weight = scale_load(sched_prio_to_weight[prio]);
-+
-+ reweight_entity(cfs_rq, se, weight);
-+ load->inv_weight = sched_prio_to_wmult[prio];
-+}
-+
-+static void update_burst_score(struct sched_entity *se) {
-+ if (!entity_is_task(se)) return;
-+ struct task_struct *p = task_of(se);
-+ u8 prio = p->static_prio - MAX_RT_PRIO;
-+ u8 prev_prio = min(39, prio + se->burst_score);
-+
-+ u8 burst_score = 0;
-+ if (!(sched_burst_exclude_kthreads && (p->flags & PF_KTHREAD)))
-+ burst_score = se->burst_penalty >> 2;
-+
-+ se->burst_score = burst_score;
-+
-+ u8 new_prio = min(39, prio + se->burst_score);
-+ if (new_prio != prev_prio)
-+ renice_task(p, new_prio);
-+}
-+
-+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);
-+ 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 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_burst_score(se);
-+}
-+
-+static void restart_burst_rescale_deadline(struct sched_entity *se) {
-+ s64 vscaled, wremain, vremain = se->deadline - se->vruntime;
-+ u8 prev_score = se->burst_score;
-+ restart_burst(se);
-+ if (prev_score > se->burst_score) {
-+ wremain = __unscale_slice(abs(vremain), prev_score);
-+ vscaled = scale_slice(wremain, se);
-+ if (unlikely(vremain < 0))
-+ vscaled = -vscaled;
-+ se->deadline = se->vruntime + vscaled;
-+ }
-+}
-+#endif // CONFIG_SCHED_BORE
-+
- static int __init setup_sched_thermal_decay_shift(char *str)
- {
- pr_warn("Ignoring the deprecated sched_thermal_decay_shift= option\n");
-@@ -131,6 +260,92 @@ 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(u8),
-+ .mode = 0644,
-+ .proc_handler = proc_dou8vec_minmax,
-+ .extra1 = SYSCTL_ONE,
-+ .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_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,
-+ },
-+ {
-+ .procname = "sched_deadline_preserve_mask",
-+ .data = &sched_deadline_preserve_mask,
-+ .maxlen = sizeof(uint),
-+ .mode = 0644,
-+ .proc_handler = proc_douintvec,
-+ },
-+#endif // CONFIG_SCHED_BORE
- #ifdef CONFIG_CFS_BANDWIDTH
- {
- .procname = "sched_cfs_bandwidth_slice_us",
-@@ -188,6 +403,13 @@ static inline void update_load_set(struct load_weight *lw, unsigned long w)
+@@ -188,6 +202,18 @@ 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 update_sysctl(void) {
-+ sysctl_sched_base_slice =
-+ max(sysctl_sched_min_base_slice, configured_sched_base_slice);
++ unsigned int base_slice = configured_sched_base_slice;
++ unsigned int min_base_slice = sysctl_sched_min_base_slice;
++
++ if (min_base_slice)
++ base_slice *= DIV_ROUND_UP(min_base_slice, base_slice);
++
++ sysctl_sched_base_slice = 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);
-@@ -218,6 +440,7 @@ static void update_sysctl(void)
+@@ -218,6 +244,7 @@ static void update_sysctl(void)
SET_SYSCTL(sched_base_slice);
#undef SET_SYSCTL
}
@@ -649,93 +783,28 @@ index 483c137b9..4c8d7fbd5 100644
void __init sched_init_granularity(void)
{
-@@ -695,6 +918,9 @@ static s64 entity_lag(u64 avruntime, struct sched_entity *se)
+@@ -695,6 +722,9 @@ 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 >>= 1;
++ limit >>= !!sched_bore;
+#endif // CONFIG_SCHED_BORE
return clamp(vlag, -limit, limit);
}
-@@ -855,6 +1081,39 @@ struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
- return __node_2_se(left);
- }
-
-+static inline bool pick_curr(struct cfs_rq *cfs_rq,
-+ struct sched_entity *curr, struct sched_entity *wakee)
-+{
-+ /*
-+ * Nothing to preserve...
-+ */
-+ if (!curr || !sched_feat(RESPECT_SLICE))
-+ return false;
-+
-+ /*
-+ * Allow preemption at the 0-lag point -- even if not all of the slice
-+ * is consumed. Note: placement of positive lag can push V left and render
-+ * @curr instantly ineligible irrespective the time on-cpu.
-+ */
-+ if (sched_feat(RUN_TO_PARITY) && !entity_eligible(cfs_rq, curr))
-+ return false;
-+
-+ /*
-+ * Don't preserve @curr when the @wakee has a shorter slice and earlier
-+ * deadline. IOW, explicitly allow preemption.
-+ */
-+ if (sched_feat(PREEMPT_SHORT) && wakee &&
-+ wakee->slice < curr->slice &&
-+ (s64)(wakee->deadline - curr->deadline) < 0)
-+ return false;
-+
-+ /*
-+ * Preserve @curr to allow it to finish its first slice.
-+ * See the HACK in set_next_entity().
-+ */
-+ return curr->vlag == curr->deadline;
-+}
-+
- /*
- * Earliest Eligible Virtual Deadline First
- *
-@@ -874,28 +1133,27 @@ struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
- *
- * Which allows tree pruning through eligibility.
- */
--static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq)
-+static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq, struct sched_entity *wakee)
- {
- struct rb_node *node = cfs_rq->tasks_timeline.rb_root.rb_node;
- struct sched_entity *se = __pick_first_entity(cfs_rq);
- struct sched_entity *curr = cfs_rq->curr;
- struct sched_entity *best = NULL;
-
-+ if (curr && !curr->on_rq)
-+ curr = NULL;
-+
- /*
- * We can safely skip eligibility check if there is only one entity
- * in this cfs_rq, saving some cycles.
- */
- if (cfs_rq->nr_running == 1)
-- return curr && curr->on_rq ? curr : se;
--
-- if (curr && (!curr->on_rq || !entity_eligible(cfs_rq, curr)))
-- curr = NULL;
-+ return curr ?: se;
-
- /*
-- * Once selected, run a task until it either becomes non-eligible or
-- * until it gets a new slice. See the HACK in set_next_entity().
-+ * Preserve @curr to let it finish its slice.
+@@ -896,6 +926,10 @@ 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)
-+ if (pick_curr(cfs_rq, curr, wakee))
+ 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 */
-@@ -954,6 +1212,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
+@@ -954,6 +988,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
* Scheduling class statistics methods:
*/
#ifdef CONFIG_SMP
@@ -743,7 +812,7 @@ index 483c137b9..4c8d7fbd5 100644
int sched_update_scaling(void)
{
unsigned int factor = get_update_sysctl_factor();
-@@ -965,6 +1224,7 @@ int sched_update_scaling(void)
+@@ -965,6 +1000,7 @@ int sched_update_scaling(void)
return 0;
}
@@ -751,33 +820,27 @@ index 483c137b9..4c8d7fbd5 100644
#endif
#endif
-@@ -1165,7 +1425,13 @@ static void update_curr(struct cfs_rq *cfs_rq)
+@@ -1165,6 +1201,10 @@ 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);
-+ curr->vruntime += max(1ULL, calc_delta_fair(delta_exec, curr));
-+#else // !CONFIG_SCHED_BORE
- curr->vruntime += calc_delta_fair(delta_exec, curr);
+#endif // CONFIG_SCHED_BORE
+ curr->vruntime += calc_delta_fair(delta_exec, curr);
update_deadline(cfs_rq, curr);
update_min_vruntime(cfs_rq);
+@@ -3782,7 +3822,7 @@ static void reweight_eevdf(struct sched_entity *se, u64 avruntime,
+ se->deadline = avruntime + vslice;
+ }
-@@ -5179,6 +5445,11 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
- s64 lag = 0;
-
- se->slice = sysctl_sched_base_slice;
-+#ifdef CONFIG_SCHED_BORE
-+ if (flags & ~sched_deadline_boost_mask & sched_deadline_preserve_mask)
-+ vslice = se->deadline - se->vruntime;
-+ else
-+#endif // CONFIG_SCHED_BORE
- vslice = calc_delta_fair(se->slice, se);
-
- /*
-@@ -5189,6 +5460,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
+-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;
+@@ -5189,6 +5229,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
*
* EEVDF: placement strategy #1 / #2
*/
@@ -787,28 +850,44 @@ index 483c137b9..4c8d7fbd5 100644
if (sched_feat(PLACE_LAG) && cfs_rq->nr_running) {
struct sched_entity *curr = cfs_rq->curr;
unsigned long load;
-@@ -5264,7 +5538,11 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
- * on average, halfway through their slice, as such start tasks
- * off with half a slice to ease into the competition.
- */
-+#if !defined(CONFIG_SCHED_BORE)
- if (sched_feat(PLACE_DEADLINE_INITIAL) && (flags & ENQUEUE_INITIAL))
-+#else // CONFIG_SCHED_BORE
-+ if (flags & sched_deadline_boost_mask)
-+#endif // CONFIG_SCHED_BORE
- vslice /= 2;
+@@ -5259,6 +5302,16 @@ 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
/*
-@@ -5478,7 +5756,7 @@ pick_next_entity(struct cfs_rq *cfs_rq)
- cfs_rq->next && entity_eligible(cfs_rq, cfs_rq->next))
- return cfs_rq->next;
+ * When joining the competition; the existing tasks will be,
+ * on average, halfway through their slice, as such start tasks
+@@ -5368,6 +5421,7 @@ 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;
-- return pick_eevdf(cfs_rq);
-+ return pick_eevdf(cfs_rq, NULL);
- }
+ if (entity_is_task(se) && task_on_rq_migrating(task_of(se)))
+@@ -5395,6 +5449,11 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
+ clear_buddies(cfs_rq, se);
- static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
-@@ -6846,6 +7124,14 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
+ 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;
+@@ -6846,6 +6905,14 @@ 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);
@@ -823,16 +902,7 @@ index 483c137b9..4c8d7fbd5 100644
for_each_sched_entity(se) {
cfs_rq = cfs_rq_of(se);
-@@ -8414,7 +8700,7 @@ static void check_preempt_wakeup_fair(struct rq *rq, struct task_struct *p, int
- /*
- * XXX pick_eevdf(cfs_rq) != se ?
- */
-- if (pick_eevdf(cfs_rq) == pse)
-+ if (pick_eevdf(cfs_rq, pse) == pse)
- goto preempt;
-
- return;
-@@ -8632,16 +8918,25 @@ static void yield_task_fair(struct rq *rq)
+@@ -8632,16 +8699,25 @@ static void yield_task_fair(struct rq *rq)
/*
* Are we the only task in the tree?
*/
@@ -858,7 +928,7 @@ index 483c137b9..4c8d7fbd5 100644
/*
* Tell update_rq_clock() that we've just updated,
* so we don't do microscopic update in schedule()
-@@ -12709,6 +13004,9 @@ static void task_fork_fair(struct task_struct *p)
+@@ -12716,6 +12792,9 @@ static void task_fork_fair(struct task_struct *p)
curr = cfs_rq->curr;
if (curr)
update_curr(cfs_rq);
@@ -868,46 +938,38 @@ index 483c137b9..4c8d7fbd5 100644
place_entity(cfs_rq, se, ENQUEUE_INITIAL);
rq_unlock(rq, &rf);
}
+@@ -12828,6 +12907,10 @@ 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 143f55df8..3aad8900c 100644
+index 143f55df89..e97b7b68bd 100644
--- a/kernel/sched/features.h
+++ b/kernel/sched/features.h
-@@ -5,8 +5,28 @@
- * sleep+wake cycles. EEVDF placement strategy #1, #2 if disabled.
+@@ -6,6 +6,10 @@
*/
SCHED_FEAT(PLACE_LAG, true)
-+/*
-+ * Give new tasks half a slice to ease into the competition.
-+ */
-+#if !defined(CONFIG_SCHED_BORE)
SCHED_FEAT(PLACE_DEADLINE_INITIAL, true)
--SCHED_FEAT(RUN_TO_PARITY, true)
-+#endif // CONFIG_SCHED_BORE
-+/*
-+ * Inhibit (wakeup) preemption until the current task has exhausted its slice.
-+ */
-+#ifdef CONFIG_SCHED_BORE
-+SCHED_FEAT(RESPECT_SLICE, false)
-+#else // !CONFIG_SCHED_BORE
-+SCHED_FEAT(RESPECT_SLICE, true)
-+#endif // CONFIG_SCHED_BORE
+/*
-+ * Relax RESPECT_SLICE to allow preemption once current has reached 0-lag.
++ * Preserve relative virtual deadline on 'migration'.
+ */
-+SCHED_FEAT(RUN_TO_PARITY, false)
-+/*
-+ * Allow tasks with a shorter slice to disregard RESPECT_SLICE
-+ */
-+SCHED_FEAT(PREEMPT_SHORT, true)
++SCHED_FEAT(PLACE_REL_DEADLINE, true)
+ SCHED_FEAT(RUN_TO_PARITY, true)
/*
- * Prefer to schedule the task we woke last (assuming it failed
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
-index 38aeedd8a..aa0ae3fb9 100644
+index 4c36cc6803..cc18ad228e 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
-@@ -1969,7 +1969,11 @@ static inline void dirty_sched_domain_sysctl(int cpu)
- }
+@@ -1984,7 +1984,11 @@ static inline void update_sched_domain_debugfs(void) { }
+ static inline void dirty_sched_domain_sysctl(int cpu) { }
#endif
+#ifdef CONFIG_SCHED_BORE
@@ -918,7 +980,7 @@ index 38aeedd8a..aa0ae3fb9 100644
static inline const struct cpumask *task_user_cpus(struct task_struct *p)
{
-@@ -2554,6 +2558,9 @@ extern const_debug unsigned int sysctl_sched_nr_migrate;
+@@ -2601,6 +2605,9 @@ 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;
@@ -929,5 +991,5 @@ index 38aeedd8a..aa0ae3fb9 100644
#ifdef CONFIG_SCHED_DEBUG
extern int sysctl_resched_latency_warn_ms;
--
-2.46.0
+2.34.1