1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
76 #include "cpudeadline.h"
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED 1
89 #define TASK_ON_RQ_MIGRATING 2
91 extern __read_mostly int scheduler_running;
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
100 * Helpers for converting nanosecond timing to jiffy resolution
102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
105 * Increase resolution of nice-level calculations for 64-bit architectures.
106 * The extra resolution improves shares distribution and load balancing of
107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108 * hierarchies, especially on larger systems. This is not a user-visible change
109 * and does not change the user-interface for setting shares/weights.
111 * We increase resolution only if we have enough bits to allow this increased
112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113 * are pretty high and the returns do not justify the increased costs.
115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116 * increase coverage and consistency always enable it on 64-bit platforms.
119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w) \
123 unsigned long __w = (w); \
125 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
130 # define scale_load(w) (w)
131 # define scale_load_down(w) (w)
135 * Task weight (visible to users) and its load (invisible to users) have
136 * independent resolution, but they should be well calibrated. We use
137 * scale_load() and scale_load_down(w) to convert between them. The
138 * following must be true:
140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
146 * Single value that decides SCHED_DEADLINE internal math precision.
147 * 10 -> just above 1us
148 * 9 -> just above 0.5us
153 * Single value that denotes runtime == period, ie unlimited time.
155 #define RUNTIME_INF ((u64)~0ULL)
157 static inline int idle_policy(int policy)
159 return policy == SCHED_IDLE;
161 static inline int fair_policy(int policy)
163 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
166 static inline int rt_policy(int policy)
168 return policy == SCHED_FIFO || policy == SCHED_RR;
171 static inline int dl_policy(int policy)
173 return policy == SCHED_DEADLINE;
175 static inline bool valid_policy(int policy)
177 return idle_policy(policy) || fair_policy(policy) ||
178 rt_policy(policy) || dl_policy(policy);
181 static inline int task_has_idle_policy(struct task_struct *p)
183 return idle_policy(p->policy);
186 static inline int task_has_rt_policy(struct task_struct *p)
188 return rt_policy(p->policy);
191 static inline int task_has_dl_policy(struct task_struct *p)
193 return dl_policy(p->policy);
196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
198 static inline void update_avg(u64 *avg, u64 sample)
200 s64 diff = sample - *avg;
205 * !! For sched_setattr_nocheck() (kernel) only !!
207 * This is actually gross. :(
209 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
210 * tasks, but still be able to sleep. We need this on platforms that cannot
211 * atomically change clock frequency. Remove once fast switching will be
212 * available on such platforms.
214 * SUGOV stands for SchedUtil GOVernor.
216 #define SCHED_FLAG_SUGOV 0x10000000
218 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
220 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
221 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
228 * Tells if entity @a should preempt entity @b.
231 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
233 return dl_entity_is_special(a) ||
234 dl_time_before(a->deadline, b->deadline);
238 * This is the priority-queue data structure of the RT scheduling class:
240 struct rt_prio_array {
241 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
242 struct list_head queue[MAX_RT_PRIO];
245 struct rt_bandwidth {
246 /* nests inside the rq lock: */
247 raw_spinlock_t rt_runtime_lock;
250 struct hrtimer rt_period_timer;
251 unsigned int rt_period_active;
254 void __dl_clear_params(struct task_struct *p);
257 * To keep the bandwidth of -deadline tasks and groups under control
258 * we need some place where:
259 * - store the maximum -deadline bandwidth of the system (the group);
260 * - cache the fraction of that bandwidth that is currently allocated.
262 * This is all done in the data structure below. It is similar to the
263 * one used for RT-throttling (rt_bandwidth), with the main difference
264 * that, since here we are only interested in admission control, we
265 * do not decrease any runtime while the group "executes", neither we
266 * need a timer to replenish it.
268 * With respect to SMP, the bandwidth is given on a per-CPU basis,
270 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
271 * - dl_total_bw array contains, in the i-eth element, the currently
272 * allocated bandwidth on the i-eth CPU.
273 * Moreover, groups consume bandwidth on each CPU, while tasks only
274 * consume bandwidth on the CPU they're running on.
275 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
276 * that will be shown the next time the proc or cgroup controls will
277 * be red. It on its turn can be changed by writing on its own
280 struct dl_bandwidth {
281 raw_spinlock_t dl_runtime_lock;
286 static inline int dl_bandwidth_enabled(void)
288 return sysctl_sched_rt_runtime >= 0;
297 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
300 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
302 dl_b->total_bw -= tsk_bw;
303 __dl_update(dl_b, (s32)tsk_bw / cpus);
307 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
309 dl_b->total_bw += tsk_bw;
310 __dl_update(dl_b, -((s32)tsk_bw / cpus));
314 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
316 return dl_b->bw != -1 &&
317 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
320 extern void init_dl_bw(struct dl_bw *dl_b);
321 extern int sched_dl_global_validate(void);
322 extern void sched_dl_do_global(void);
323 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
324 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
325 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
326 extern bool __checkparam_dl(const struct sched_attr *attr);
327 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
328 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
329 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
330 extern bool dl_cpu_busy(unsigned int cpu);
332 #ifdef CONFIG_CGROUP_SCHED
334 #include <linux/cgroup.h>
335 #include <linux/psi.h>
340 extern struct list_head task_groups;
342 struct cfs_bandwidth {
343 #ifdef CONFIG_CFS_BANDWIDTH
348 s64 hierarchical_quota;
353 struct hrtimer period_timer;
354 struct hrtimer slack_timer;
355 struct list_head throttled_cfs_rq;
364 /* Task group related information */
366 struct cgroup_subsys_state css;
368 #ifdef CONFIG_FAIR_GROUP_SCHED
369 /* schedulable entities of this group on each CPU */
370 struct sched_entity **se;
371 /* runqueue "owned" by this group on each CPU */
372 struct cfs_rq **cfs_rq;
373 unsigned long shares;
377 * load_avg can be heavily contended at clock tick time, so put
378 * it in its own cacheline separated from the fields above which
379 * will also be accessed at each tick.
381 atomic_long_t load_avg ____cacheline_aligned;
385 #ifdef CONFIG_RT_GROUP_SCHED
386 struct sched_rt_entity **rt_se;
387 struct rt_rq **rt_rq;
389 struct rt_bandwidth rt_bandwidth;
393 struct list_head list;
395 struct task_group *parent;
396 struct list_head siblings;
397 struct list_head children;
399 #ifdef CONFIG_SCHED_AUTOGROUP
400 struct autogroup *autogroup;
403 struct cfs_bandwidth cfs_bandwidth;
405 #ifdef CONFIG_UCLAMP_TASK_GROUP
406 /* The two decimal precision [%] value requested from user-space */
407 unsigned int uclamp_pct[UCLAMP_CNT];
408 /* Clamp values requested for a task group */
409 struct uclamp_se uclamp_req[UCLAMP_CNT];
410 /* Effective clamp values used for a task group */
411 struct uclamp_se uclamp[UCLAMP_CNT];
416 #ifdef CONFIG_FAIR_GROUP_SCHED
417 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
420 * A weight of 0 or 1 can cause arithmetics problems.
421 * A weight of a cfs_rq is the sum of weights of which entities
422 * are queued on this cfs_rq, so a weight of a entity should not be
423 * too large, so as the shares value of a task group.
424 * (The default weight is 1024 - so there's no practical
425 * limitation from this.)
427 #define MIN_SHARES (1UL << 1)
428 #define MAX_SHARES (1UL << 18)
431 typedef int (*tg_visitor)(struct task_group *, void *);
433 extern int walk_tg_tree_from(struct task_group *from,
434 tg_visitor down, tg_visitor up, void *data);
437 * Iterate the full tree, calling @down when first entering a node and @up when
438 * leaving it for the final time.
440 * Caller must hold rcu_lock or sufficient equivalent.
442 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
444 return walk_tg_tree_from(&root_task_group, down, up, data);
447 extern int tg_nop(struct task_group *tg, void *data);
449 extern void free_fair_sched_group(struct task_group *tg);
450 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
451 extern void online_fair_sched_group(struct task_group *tg);
452 extern void unregister_fair_sched_group(struct task_group *tg);
453 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
454 struct sched_entity *se, int cpu,
455 struct sched_entity *parent);
456 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
458 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
459 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
460 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
462 extern void free_rt_sched_group(struct task_group *tg);
463 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
464 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
465 struct sched_rt_entity *rt_se, int cpu,
466 struct sched_rt_entity *parent);
467 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
468 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
469 extern long sched_group_rt_runtime(struct task_group *tg);
470 extern long sched_group_rt_period(struct task_group *tg);
471 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
473 extern struct task_group *sched_create_group(struct task_group *parent);
474 extern void sched_online_group(struct task_group *tg,
475 struct task_group *parent);
476 extern void sched_destroy_group(struct task_group *tg);
477 extern void sched_offline_group(struct task_group *tg);
479 extern void sched_move_task(struct task_struct *tsk);
481 #ifdef CONFIG_FAIR_GROUP_SCHED
482 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
485 extern void set_task_rq_fair(struct sched_entity *se,
486 struct cfs_rq *prev, struct cfs_rq *next);
487 #else /* !CONFIG_SMP */
488 static inline void set_task_rq_fair(struct sched_entity *se,
489 struct cfs_rq *prev, struct cfs_rq *next) { }
490 #endif /* CONFIG_SMP */
491 #endif /* CONFIG_FAIR_GROUP_SCHED */
493 #else /* CONFIG_CGROUP_SCHED */
495 struct cfs_bandwidth { };
497 #endif /* CONFIG_CGROUP_SCHED */
499 /* CFS-related fields in a runqueue */
501 struct load_weight load;
502 unsigned int nr_running;
503 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
504 unsigned int idle_h_nr_running; /* SCHED_IDLE */
509 u64 min_vruntime_copy;
512 struct rb_root_cached tasks_timeline;
515 * 'curr' points to currently running entity on this cfs_rq.
516 * It is set to NULL otherwise (i.e when none are currently running).
518 struct sched_entity *curr;
519 struct sched_entity *next;
520 struct sched_entity *last;
521 struct sched_entity *skip;
523 #ifdef CONFIG_SCHED_DEBUG
524 unsigned int nr_spread_over;
531 struct sched_avg avg;
533 u64 load_last_update_time_copy;
536 raw_spinlock_t lock ____cacheline_aligned;
538 unsigned long load_avg;
539 unsigned long util_avg;
540 unsigned long runnable_avg;
543 #ifdef CONFIG_FAIR_GROUP_SCHED
544 unsigned long tg_load_avg_contrib;
546 long prop_runnable_sum;
549 * h_load = weight * f(tg)
551 * Where f(tg) is the recursive weight fraction assigned to
554 unsigned long h_load;
555 u64 last_h_load_update;
556 struct sched_entity *h_load_next;
557 #endif /* CONFIG_FAIR_GROUP_SCHED */
558 #endif /* CONFIG_SMP */
560 #ifdef CONFIG_FAIR_GROUP_SCHED
561 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
564 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
565 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
566 * (like users, containers etc.)
568 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
569 * This list is used during load balance.
572 struct list_head leaf_cfs_rq_list;
573 struct task_group *tg; /* group that "owns" this runqueue */
575 #ifdef CONFIG_CFS_BANDWIDTH
577 s64 runtime_remaining;
580 u64 throttled_clock_task;
581 u64 throttled_clock_task_time;
584 struct list_head throttled_list;
585 #endif /* CONFIG_CFS_BANDWIDTH */
586 #endif /* CONFIG_FAIR_GROUP_SCHED */
589 static inline int rt_bandwidth_enabled(void)
591 return sysctl_sched_rt_runtime >= 0;
594 /* RT IPI pull logic requires IRQ_WORK */
595 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
596 # define HAVE_RT_PUSH_IPI
599 /* Real-Time classes' related field in a runqueue: */
601 struct rt_prio_array active;
602 unsigned int rt_nr_running;
603 unsigned int rr_nr_running;
604 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
606 int curr; /* highest queued rt task prio */
608 int next; /* next highest */
613 unsigned long rt_nr_migratory;
614 unsigned long rt_nr_total;
616 struct plist_head pushable_tasks;
618 #endif /* CONFIG_SMP */
624 /* Nests inside the rq lock: */
625 raw_spinlock_t rt_runtime_lock;
627 #ifdef CONFIG_RT_GROUP_SCHED
628 unsigned long rt_nr_boosted;
631 struct task_group *tg;
635 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
637 return rt_rq->rt_queued && rt_rq->rt_nr_running;
640 /* Deadline class' related fields in a runqueue */
642 /* runqueue is an rbtree, ordered by deadline */
643 struct rb_root_cached root;
645 unsigned long dl_nr_running;
649 * Deadline values of the currently executing and the
650 * earliest ready task on this rq. Caching these facilitates
651 * the decision whether or not a ready but not running task
652 * should migrate somewhere else.
659 unsigned long dl_nr_migratory;
663 * Tasks on this rq that can be pushed away. They are kept in
664 * an rb-tree, ordered by tasks' deadlines, with caching
665 * of the leftmost (earliest deadline) element.
667 struct rb_root_cached pushable_dl_tasks_root;
672 * "Active utilization" for this runqueue: increased when a
673 * task wakes up (becomes TASK_RUNNING) and decreased when a
679 * Utilization of the tasks "assigned" to this runqueue (including
680 * the tasks that are in runqueue and the tasks that executed on this
681 * CPU and blocked). Increased when a task moves to this runqueue, and
682 * decreased when the task moves away (migrates, changes scheduling
683 * policy, or terminates).
684 * This is needed to compute the "inactive utilization" for the
685 * runqueue (inactive utilization = this_bw - running_bw).
691 * Inverse of the fraction of CPU utilization that can be reclaimed
692 * by the GRUB algorithm.
697 #ifdef CONFIG_FAIR_GROUP_SCHED
698 /* An entity is a task if it doesn't "own" a runqueue */
699 #define entity_is_task(se) (!se->my_q)
701 static inline void se_update_runnable(struct sched_entity *se)
703 if (!entity_is_task(se))
704 se->runnable_weight = se->my_q->h_nr_running;
707 static inline long se_runnable(struct sched_entity *se)
709 if (entity_is_task(se))
712 return se->runnable_weight;
716 #define entity_is_task(se) 1
718 static inline void se_update_runnable(struct sched_entity *se) {}
720 static inline long se_runnable(struct sched_entity *se)
728 * XXX we want to get rid of these helpers and use the full load resolution.
730 static inline long se_weight(struct sched_entity *se)
732 return scale_load_down(se->load.weight);
736 static inline bool sched_asym_prefer(int a, int b)
738 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
742 struct em_perf_domain *em_pd;
743 struct perf_domain *next;
747 /* Scheduling group status flags */
748 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
749 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
752 * We add the notion of a root-domain which will be used to define per-domain
753 * variables. Each exclusive cpuset essentially defines an island domain by
754 * fully partitioning the member CPUs from any other cpuset. Whenever a new
755 * exclusive cpuset is created, we also create and attach a new root-domain
764 cpumask_var_t online;
767 * Indicate pullable load on at least one CPU, e.g:
768 * - More than one runnable task
769 * - Running task is misfit
773 /* Indicate one or more cpus over-utilized (tipping point) */
777 * The bit corresponding to a CPU gets set here if such CPU has more
778 * than one runnable -deadline task (as it is below for RT tasks).
780 cpumask_var_t dlo_mask;
785 #ifdef HAVE_RT_PUSH_IPI
787 * For IPI pull requests, loop across the rto_mask.
789 struct irq_work rto_push_work;
790 raw_spinlock_t rto_lock;
791 /* These are only updated and read within rto_lock */
794 /* These atomics are updated outside of a lock */
795 atomic_t rto_loop_next;
796 atomic_t rto_loop_start;
799 * The "RT overload" flag: it gets set if a CPU has more than
800 * one runnable RT task.
802 cpumask_var_t rto_mask;
803 struct cpupri cpupri;
805 unsigned long max_cpu_capacity;
808 * NULL-terminated list of performance domains intersecting with the
809 * CPUs of the rd. Protected by RCU.
811 struct perf_domain __rcu *pd;
814 extern void init_defrootdomain(void);
815 extern int sched_init_domains(const struct cpumask *cpu_map);
816 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
817 extern void sched_get_rd(struct root_domain *rd);
818 extern void sched_put_rd(struct root_domain *rd);
820 #ifdef HAVE_RT_PUSH_IPI
821 extern void rto_push_irq_work_func(struct irq_work *work);
823 #endif /* CONFIG_SMP */
825 #ifdef CONFIG_UCLAMP_TASK
827 * struct uclamp_bucket - Utilization clamp bucket
828 * @value: utilization clamp value for tasks on this clamp bucket
829 * @tasks: number of RUNNABLE tasks on this clamp bucket
831 * Keep track of how many tasks are RUNNABLE for a given utilization
834 struct uclamp_bucket {
835 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
836 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
840 * struct uclamp_rq - rq's utilization clamp
841 * @value: currently active clamp values for a rq
842 * @bucket: utilization clamp buckets affecting a rq
844 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
845 * A clamp value is affecting a rq when there is at least one task RUNNABLE
846 * (or actually running) with that value.
848 * There are up to UCLAMP_CNT possible different clamp values, currently there
849 * are only two: minimum utilization and maximum utilization.
851 * All utilization clamping values are MAX aggregated, since:
852 * - for util_min: we want to run the CPU at least at the max of the minimum
853 * utilization required by its currently RUNNABLE tasks.
854 * - for util_max: we want to allow the CPU to run up to the max of the
855 * maximum utilization allowed by its currently RUNNABLE tasks.
857 * Since on each system we expect only a limited number of different
858 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
859 * the metrics required to compute all the per-rq utilization clamp values.
863 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
865 #endif /* CONFIG_UCLAMP_TASK */
868 * This is the main, per-CPU runqueue data structure.
870 * Locking rule: those places that want to lock multiple runqueues
871 * (such as the load balancing or the thread migration code), lock
872 * acquire operations must be ordered by ascending &runqueue.
879 * nr_running and cpu_load should be in the same cacheline because
880 * remote CPUs use both these fields when doing load calculation.
882 unsigned int nr_running;
883 #ifdef CONFIG_NUMA_BALANCING
884 unsigned int nr_numa_running;
885 unsigned int nr_preferred_running;
886 unsigned int numa_migrate_on;
888 #ifdef CONFIG_NO_HZ_COMMON
890 unsigned long last_blocked_load_update_tick;
891 unsigned int has_blocked_load;
892 call_single_data_t nohz_csd;
893 #endif /* CONFIG_SMP */
894 unsigned int nohz_tick_stopped;
896 #endif /* CONFIG_NO_HZ_COMMON */
898 unsigned long nr_load_updates;
901 #ifdef CONFIG_UCLAMP_TASK
902 /* Utilization clamp values based on CPU's RUNNABLE tasks */
903 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
904 unsigned int uclamp_flags;
905 #define UCLAMP_FLAG_IDLE 0x01
912 #ifdef CONFIG_FAIR_GROUP_SCHED
913 /* list of leaf cfs_rq on this CPU: */
914 struct list_head leaf_cfs_rq_list;
915 struct list_head *tmp_alone_branch;
916 #endif /* CONFIG_FAIR_GROUP_SCHED */
919 * This is part of a global counter where only the total sum
920 * over all CPUs matters. A task can increase this counter on
921 * one CPU and if it got migrated afterwards it may decrease
922 * it on another CPU. Always updated under the runqueue lock:
924 unsigned long nr_uninterruptible;
926 struct task_struct __rcu *curr;
927 struct task_struct *idle;
928 struct task_struct *stop;
929 unsigned long next_balance;
930 struct mm_struct *prev_mm;
932 unsigned int clock_update_flags;
934 /* Ensure that all clocks are in the same cache line */
935 u64 clock_task ____cacheline_aligned;
937 unsigned long lost_idle_time;
941 #ifdef CONFIG_MEMBARRIER
942 int membarrier_state;
946 struct root_domain *rd;
947 struct sched_domain __rcu *sd;
949 unsigned long cpu_capacity;
950 unsigned long cpu_capacity_orig;
952 struct callback_head *balance_callback;
954 unsigned char nohz_idle_balance;
955 unsigned char idle_balance;
957 unsigned long misfit_task_load;
959 /* For active balancing */
962 struct cpu_stop_work active_balance_work;
964 /* CPU of this runqueue: */
968 struct list_head cfs_tasks;
970 struct sched_avg avg_rt;
971 struct sched_avg avg_dl;
972 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
973 struct sched_avg avg_irq;
975 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
976 struct sched_avg avg_thermal;
981 /* This is used to determine avg_idle's max value */
982 u64 max_idle_balance_cost;
983 #endif /* CONFIG_SMP */
985 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
988 #ifdef CONFIG_PARAVIRT
991 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
992 u64 prev_steal_time_rq;
995 /* calc_load related fields */
996 unsigned long calc_load_update;
997 long calc_load_active;
999 #ifdef CONFIG_SCHED_HRTICK
1001 call_single_data_t hrtick_csd;
1003 struct hrtimer hrtick_timer;
1006 #ifdef CONFIG_SCHEDSTATS
1008 struct sched_info rq_sched_info;
1009 unsigned long long rq_cpu_time;
1010 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1012 /* sys_sched_yield() stats */
1013 unsigned int yld_count;
1015 /* schedule() stats */
1016 unsigned int sched_count;
1017 unsigned int sched_goidle;
1019 /* try_to_wake_up() stats */
1020 unsigned int ttwu_count;
1021 unsigned int ttwu_local;
1025 call_single_data_t wake_csd;
1026 struct llist_head wake_list;
1029 #ifdef CONFIG_CPU_IDLE
1030 /* Must be inspected within a rcu lock section */
1031 struct cpuidle_state *idle_state;
1035 #ifdef CONFIG_FAIR_GROUP_SCHED
1037 /* CPU runqueue to which this cfs_rq is attached */
1038 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1045 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1047 return container_of(cfs_rq, struct rq, cfs);
1051 static inline int cpu_of(struct rq *rq)
1061 #ifdef CONFIG_SCHED_SMT
1062 extern void __update_idle_core(struct rq *rq);
1064 static inline void update_idle_core(struct rq *rq)
1066 if (static_branch_unlikely(&sched_smt_present))
1067 __update_idle_core(rq);
1071 static inline void update_idle_core(struct rq *rq) { }
1074 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1076 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1077 #define this_rq() this_cpu_ptr(&runqueues)
1078 #define task_rq(p) cpu_rq(task_cpu(p))
1079 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1080 #define raw_rq() raw_cpu_ptr(&runqueues)
1082 extern void update_rq_clock(struct rq *rq);
1084 static inline u64 __rq_clock_broken(struct rq *rq)
1086 return READ_ONCE(rq->clock);
1090 * rq::clock_update_flags bits
1092 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1093 * call to __schedule(). This is an optimisation to avoid
1094 * neighbouring rq clock updates.
1096 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1097 * in effect and calls to update_rq_clock() are being ignored.
1099 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1100 * made to update_rq_clock() since the last time rq::lock was pinned.
1102 * If inside of __schedule(), clock_update_flags will have been
1103 * shifted left (a left shift is a cheap operation for the fast path
1104 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1106 * if (rq-clock_update_flags >= RQCF_UPDATED)
1108 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1109 * one position though, because the next rq_unpin_lock() will shift it
1112 #define RQCF_REQ_SKIP 0x01
1113 #define RQCF_ACT_SKIP 0x02
1114 #define RQCF_UPDATED 0x04
1116 static inline void assert_clock_updated(struct rq *rq)
1119 * The only reason for not seeing a clock update since the
1120 * last rq_pin_lock() is if we're currently skipping updates.
1122 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1125 static inline u64 rq_clock(struct rq *rq)
1127 lockdep_assert_held(&rq->lock);
1128 assert_clock_updated(rq);
1133 static inline u64 rq_clock_task(struct rq *rq)
1135 lockdep_assert_held(&rq->lock);
1136 assert_clock_updated(rq);
1138 return rq->clock_task;
1142 * By default the decay is the default pelt decay period.
1143 * The decay shift can change the decay period in
1145 * Decay shift Decay period(ms)
1152 extern int sched_thermal_decay_shift;
1154 static inline u64 rq_clock_thermal(struct rq *rq)
1156 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1159 static inline void rq_clock_skip_update(struct rq *rq)
1161 lockdep_assert_held(&rq->lock);
1162 rq->clock_update_flags |= RQCF_REQ_SKIP;
1166 * See rt task throttling, which is the only time a skip
1167 * request is cancelled.
1169 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1171 lockdep_assert_held(&rq->lock);
1172 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1176 unsigned long flags;
1177 struct pin_cookie cookie;
1178 #ifdef CONFIG_SCHED_DEBUG
1180 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1181 * current pin context is stashed here in case it needs to be
1182 * restored in rq_repin_lock().
1184 unsigned int clock_update_flags;
1188 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1190 rf->cookie = lockdep_pin_lock(&rq->lock);
1192 #ifdef CONFIG_SCHED_DEBUG
1193 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1194 rf->clock_update_flags = 0;
1198 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1200 #ifdef CONFIG_SCHED_DEBUG
1201 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1202 rf->clock_update_flags = RQCF_UPDATED;
1205 lockdep_unpin_lock(&rq->lock, rf->cookie);
1208 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1210 lockdep_repin_lock(&rq->lock, rf->cookie);
1212 #ifdef CONFIG_SCHED_DEBUG
1214 * Restore the value we stashed in @rf for this pin context.
1216 rq->clock_update_flags |= rf->clock_update_flags;
1220 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1221 __acquires(rq->lock);
1223 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1224 __acquires(p->pi_lock)
1225 __acquires(rq->lock);
1227 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1228 __releases(rq->lock)
1230 rq_unpin_lock(rq, rf);
1231 raw_spin_unlock(&rq->lock);
1235 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1236 __releases(rq->lock)
1237 __releases(p->pi_lock)
1239 rq_unpin_lock(rq, rf);
1240 raw_spin_unlock(&rq->lock);
1241 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1245 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1246 __acquires(rq->lock)
1248 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1249 rq_pin_lock(rq, rf);
1253 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1254 __acquires(rq->lock)
1256 raw_spin_lock_irq(&rq->lock);
1257 rq_pin_lock(rq, rf);
1261 rq_lock(struct rq *rq, struct rq_flags *rf)
1262 __acquires(rq->lock)
1264 raw_spin_lock(&rq->lock);
1265 rq_pin_lock(rq, rf);
1269 rq_relock(struct rq *rq, struct rq_flags *rf)
1270 __acquires(rq->lock)
1272 raw_spin_lock(&rq->lock);
1273 rq_repin_lock(rq, rf);
1277 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1278 __releases(rq->lock)
1280 rq_unpin_lock(rq, rf);
1281 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1285 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1286 __releases(rq->lock)
1288 rq_unpin_lock(rq, rf);
1289 raw_spin_unlock_irq(&rq->lock);
1293 rq_unlock(struct rq *rq, struct rq_flags *rf)
1294 __releases(rq->lock)
1296 rq_unpin_lock(rq, rf);
1297 raw_spin_unlock(&rq->lock);
1300 static inline struct rq *
1301 this_rq_lock_irq(struct rq_flags *rf)
1302 __acquires(rq->lock)
1306 local_irq_disable();
1313 enum numa_topology_type {
1318 extern enum numa_topology_type sched_numa_topology_type;
1319 extern int sched_max_numa_distance;
1320 extern bool find_numa_distance(int distance);
1321 extern void sched_init_numa(void);
1322 extern void sched_domains_numa_masks_set(unsigned int cpu);
1323 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1324 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1326 static inline void sched_init_numa(void) { }
1327 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1328 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1329 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1335 #ifdef CONFIG_NUMA_BALANCING
1336 /* The regions in numa_faults array from task_struct */
1337 enum numa_faults_stats {
1343 extern void sched_setnuma(struct task_struct *p, int node);
1344 extern int migrate_task_to(struct task_struct *p, int cpu);
1345 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1347 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1350 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1353 #endif /* CONFIG_NUMA_BALANCING */
1358 queue_balance_callback(struct rq *rq,
1359 struct callback_head *head,
1360 void (*func)(struct rq *rq))
1362 lockdep_assert_held(&rq->lock);
1364 if (unlikely(head->next))
1367 head->func = (void (*)(struct callback_head *))func;
1368 head->next = rq->balance_callback;
1369 rq->balance_callback = head;
1372 extern void sched_ttwu_pending(void);
1374 #define rcu_dereference_check_sched_domain(p) \
1375 rcu_dereference_check((p), \
1376 lockdep_is_held(&sched_domains_mutex))
1379 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1380 * See destroy_sched_domains: call_rcu for details.
1382 * The domain tree of any CPU may only be accessed from within
1383 * preempt-disabled sections.
1385 #define for_each_domain(cpu, __sd) \
1386 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1387 __sd; __sd = __sd->parent)
1390 * highest_flag_domain - Return highest sched_domain containing flag.
1391 * @cpu: The CPU whose highest level of sched domain is to
1393 * @flag: The flag to check for the highest sched_domain
1394 * for the given CPU.
1396 * Returns the highest sched_domain of a CPU which contains the given flag.
1398 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1400 struct sched_domain *sd, *hsd = NULL;
1402 for_each_domain(cpu, sd) {
1403 if (!(sd->flags & flag))
1411 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1413 struct sched_domain *sd;
1415 for_each_domain(cpu, sd) {
1416 if (sd->flags & flag)
1423 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1424 DECLARE_PER_CPU(int, sd_llc_size);
1425 DECLARE_PER_CPU(int, sd_llc_id);
1426 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1427 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1428 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1429 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1430 extern struct static_key_false sched_asym_cpucapacity;
1432 struct sched_group_capacity {
1435 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1438 unsigned long capacity;
1439 unsigned long min_capacity; /* Min per-CPU capacity in group */
1440 unsigned long max_capacity; /* Max per-CPU capacity in group */
1441 unsigned long next_update;
1442 int imbalance; /* XXX unrelated to capacity but shared group state */
1444 #ifdef CONFIG_SCHED_DEBUG
1448 unsigned long cpumask[0]; /* Balance mask */
1451 struct sched_group {
1452 struct sched_group *next; /* Must be a circular list */
1455 unsigned int group_weight;
1456 struct sched_group_capacity *sgc;
1457 int asym_prefer_cpu; /* CPU of highest priority in group */
1460 * The CPUs this group covers.
1462 * NOTE: this field is variable length. (Allocated dynamically
1463 * by attaching extra space to the end of the structure,
1464 * depending on how many CPUs the kernel has booted up with)
1466 unsigned long cpumask[];
1469 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1471 return to_cpumask(sg->cpumask);
1475 * See build_balance_mask().
1477 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1479 return to_cpumask(sg->sgc->cpumask);
1483 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1484 * @group: The group whose first CPU is to be returned.
1486 static inline unsigned int group_first_cpu(struct sched_group *group)
1488 return cpumask_first(sched_group_span(group));
1491 extern int group_balance_cpu(struct sched_group *sg);
1493 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1494 void register_sched_domain_sysctl(void);
1495 void dirty_sched_domain_sysctl(int cpu);
1496 void unregister_sched_domain_sysctl(void);
1498 static inline void register_sched_domain_sysctl(void)
1501 static inline void dirty_sched_domain_sysctl(int cpu)
1504 static inline void unregister_sched_domain_sysctl(void)
1509 extern void flush_smp_call_function_from_idle(void);
1511 #else /* !CONFIG_SMP: */
1512 static inline void flush_smp_call_function_from_idle(void) { }
1513 static inline void sched_ttwu_pending(void) { }
1517 #include "autogroup.h"
1519 #ifdef CONFIG_CGROUP_SCHED
1522 * Return the group to which this tasks belongs.
1524 * We cannot use task_css() and friends because the cgroup subsystem
1525 * changes that value before the cgroup_subsys::attach() method is called,
1526 * therefore we cannot pin it and might observe the wrong value.
1528 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1529 * core changes this before calling sched_move_task().
1531 * Instead we use a 'copy' which is updated from sched_move_task() while
1532 * holding both task_struct::pi_lock and rq::lock.
1534 static inline struct task_group *task_group(struct task_struct *p)
1536 return p->sched_task_group;
1539 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1540 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1542 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1543 struct task_group *tg = task_group(p);
1546 #ifdef CONFIG_FAIR_GROUP_SCHED
1547 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1548 p->se.cfs_rq = tg->cfs_rq[cpu];
1549 p->se.parent = tg->se[cpu];
1552 #ifdef CONFIG_RT_GROUP_SCHED
1553 p->rt.rt_rq = tg->rt_rq[cpu];
1554 p->rt.parent = tg->rt_se[cpu];
1558 #else /* CONFIG_CGROUP_SCHED */
1560 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1561 static inline struct task_group *task_group(struct task_struct *p)
1566 #endif /* CONFIG_CGROUP_SCHED */
1568 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1570 set_task_rq(p, cpu);
1573 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1574 * successfully executed on another CPU. We must ensure that updates of
1575 * per-task data have been completed by this moment.
1578 #ifdef CONFIG_THREAD_INFO_IN_TASK
1579 WRITE_ONCE(p->cpu, cpu);
1581 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1588 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1590 #ifdef CONFIG_SCHED_DEBUG
1591 # include <linux/static_key.h>
1592 # define const_debug __read_mostly
1594 # define const_debug const
1597 #define SCHED_FEAT(name, enabled) \
1598 __SCHED_FEAT_##name ,
1601 #include "features.h"
1607 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1610 * To support run-time toggling of sched features, all the translation units
1611 * (but core.c) reference the sysctl_sched_features defined in core.c.
1613 extern const_debug unsigned int sysctl_sched_features;
1615 #define SCHED_FEAT(name, enabled) \
1616 static __always_inline bool static_branch_##name(struct static_key *key) \
1618 return static_key_##enabled(key); \
1621 #include "features.h"
1624 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1625 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1627 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1630 * Each translation unit has its own copy of sysctl_sched_features to allow
1631 * constants propagation at compile time and compiler optimization based on
1634 #define SCHED_FEAT(name, enabled) \
1635 (1UL << __SCHED_FEAT_##name) * enabled |
1636 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1637 #include "features.h"
1641 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1643 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1645 extern struct static_key_false sched_numa_balancing;
1646 extern struct static_key_false sched_schedstats;
1648 static inline u64 global_rt_period(void)
1650 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1653 static inline u64 global_rt_runtime(void)
1655 if (sysctl_sched_rt_runtime < 0)
1658 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1661 static inline int task_current(struct rq *rq, struct task_struct *p)
1663 return rq->curr == p;
1666 static inline int task_running(struct rq *rq, struct task_struct *p)
1671 return task_current(rq, p);
1675 static inline int task_on_rq_queued(struct task_struct *p)
1677 return p->on_rq == TASK_ON_RQ_QUEUED;
1680 static inline int task_on_rq_migrating(struct task_struct *p)
1682 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1688 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1689 #define WF_FORK 0x02 /* Child wakeup after fork */
1690 #define WF_MIGRATED 0x04 /* Internal use, task got migrated */
1691 #define WF_ON_RQ 0x08 /* Wakee is on_rq */
1694 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1695 * of tasks with abnormal "nice" values across CPUs the contribution that
1696 * each task makes to its run queue's load is weighted according to its
1697 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1698 * scaled version of the new time slice allocation that they receive on time
1702 #define WEIGHT_IDLEPRIO 3
1703 #define WMULT_IDLEPRIO 1431655765
1705 extern const int sched_prio_to_weight[40];
1706 extern const u32 sched_prio_to_wmult[40];
1709 * {de,en}queue flags:
1711 * DEQUEUE_SLEEP - task is no longer runnable
1712 * ENQUEUE_WAKEUP - task just became runnable
1714 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1715 * are in a known state which allows modification. Such pairs
1716 * should preserve as much state as possible.
1718 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1721 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1722 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1723 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1727 #define DEQUEUE_SLEEP 0x01
1728 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1729 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1730 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1732 #define ENQUEUE_WAKEUP 0x01
1733 #define ENQUEUE_RESTORE 0x02
1734 #define ENQUEUE_MOVE 0x04
1735 #define ENQUEUE_NOCLOCK 0x08
1737 #define ENQUEUE_HEAD 0x10
1738 #define ENQUEUE_REPLENISH 0x20
1740 #define ENQUEUE_MIGRATED 0x40
1742 #define ENQUEUE_MIGRATED 0x00
1745 #define RETRY_TASK ((void *)-1UL)
1747 struct sched_class {
1748 const struct sched_class *next;
1750 #ifdef CONFIG_UCLAMP_TASK
1754 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1755 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1756 void (*yield_task) (struct rq *rq);
1757 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1759 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1761 struct task_struct *(*pick_next_task)(struct rq *rq);
1763 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1764 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1767 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1768 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1769 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1771 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1773 void (*set_cpus_allowed)(struct task_struct *p,
1774 const struct cpumask *newmask);
1776 void (*rq_online)(struct rq *rq);
1777 void (*rq_offline)(struct rq *rq);
1780 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1781 void (*task_fork)(struct task_struct *p);
1782 void (*task_dead)(struct task_struct *p);
1785 * The switched_from() call is allowed to drop rq->lock, therefore we
1786 * cannot assume the switched_from/switched_to pair is serliazed by
1787 * rq->lock. They are however serialized by p->pi_lock.
1789 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1790 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1791 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1794 unsigned int (*get_rr_interval)(struct rq *rq,
1795 struct task_struct *task);
1797 void (*update_curr)(struct rq *rq);
1799 #define TASK_SET_GROUP 0
1800 #define TASK_MOVE_GROUP 1
1802 #ifdef CONFIG_FAIR_GROUP_SCHED
1803 void (*task_change_group)(struct task_struct *p, int type);
1807 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1809 WARN_ON_ONCE(rq->curr != prev);
1810 prev->sched_class->put_prev_task(rq, prev);
1813 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1815 WARN_ON_ONCE(rq->curr != next);
1816 next->sched_class->set_next_task(rq, next, false);
1820 #define sched_class_highest (&stop_sched_class)
1822 #define sched_class_highest (&dl_sched_class)
1825 #define for_class_range(class, _from, _to) \
1826 for (class = (_from); class != (_to); class = class->next)
1828 #define for_each_class(class) \
1829 for_class_range(class, sched_class_highest, NULL)
1831 extern const struct sched_class stop_sched_class;
1832 extern const struct sched_class dl_sched_class;
1833 extern const struct sched_class rt_sched_class;
1834 extern const struct sched_class fair_sched_class;
1835 extern const struct sched_class idle_sched_class;
1837 static inline bool sched_stop_runnable(struct rq *rq)
1839 return rq->stop && task_on_rq_queued(rq->stop);
1842 static inline bool sched_dl_runnable(struct rq *rq)
1844 return rq->dl.dl_nr_running > 0;
1847 static inline bool sched_rt_runnable(struct rq *rq)
1849 return rq->rt.rt_queued > 0;
1852 static inline bool sched_fair_runnable(struct rq *rq)
1854 return rq->cfs.nr_running > 0;
1857 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1858 extern struct task_struct *pick_next_task_idle(struct rq *rq);
1862 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1864 extern void trigger_load_balance(struct rq *rq);
1866 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1870 #ifdef CONFIG_CPU_IDLE
1871 static inline void idle_set_state(struct rq *rq,
1872 struct cpuidle_state *idle_state)
1874 rq->idle_state = idle_state;
1877 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1879 SCHED_WARN_ON(!rcu_read_lock_held());
1881 return rq->idle_state;
1884 static inline void idle_set_state(struct rq *rq,
1885 struct cpuidle_state *idle_state)
1889 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1895 extern void schedule_idle(void);
1897 extern void sysrq_sched_debug_show(void);
1898 extern void sched_init_granularity(void);
1899 extern void update_max_interval(void);
1901 extern void init_sched_dl_class(void);
1902 extern void init_sched_rt_class(void);
1903 extern void init_sched_fair_class(void);
1905 extern void reweight_task(struct task_struct *p, int prio);
1907 extern void resched_curr(struct rq *rq);
1908 extern void resched_cpu(int cpu);
1910 extern struct rt_bandwidth def_rt_bandwidth;
1911 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1913 extern struct dl_bandwidth def_dl_bandwidth;
1914 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1915 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1916 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1919 #define BW_UNIT (1 << BW_SHIFT)
1920 #define RATIO_SHIFT 8
1921 #define MAX_BW_BITS (64 - BW_SHIFT)
1922 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
1923 unsigned long to_ratio(u64 period, u64 runtime);
1925 extern void init_entity_runnable_average(struct sched_entity *se);
1926 extern void post_init_entity_util_avg(struct task_struct *p);
1928 #ifdef CONFIG_NO_HZ_FULL
1929 extern bool sched_can_stop_tick(struct rq *rq);
1930 extern int __init sched_tick_offload_init(void);
1933 * Tick may be needed by tasks in the runqueue depending on their policy and
1934 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1935 * nohz mode if necessary.
1937 static inline void sched_update_tick_dependency(struct rq *rq)
1941 if (!tick_nohz_full_enabled())
1946 if (!tick_nohz_full_cpu(cpu))
1949 if (sched_can_stop_tick(rq))
1950 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1952 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1955 static inline int sched_tick_offload_init(void) { return 0; }
1956 static inline void sched_update_tick_dependency(struct rq *rq) { }
1959 static inline void add_nr_running(struct rq *rq, unsigned count)
1961 unsigned prev_nr = rq->nr_running;
1963 rq->nr_running = prev_nr + count;
1966 if (prev_nr < 2 && rq->nr_running >= 2) {
1967 if (!READ_ONCE(rq->rd->overload))
1968 WRITE_ONCE(rq->rd->overload, 1);
1972 sched_update_tick_dependency(rq);
1975 static inline void sub_nr_running(struct rq *rq, unsigned count)
1977 rq->nr_running -= count;
1978 /* Check if we still need preemption */
1979 sched_update_tick_dependency(rq);
1982 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1983 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1985 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1987 extern const_debug unsigned int sysctl_sched_nr_migrate;
1988 extern const_debug unsigned int sysctl_sched_migration_cost;
1990 #ifdef CONFIG_SCHED_HRTICK
1994 * - enabled by features
1995 * - hrtimer is actually high res
1997 static inline int hrtick_enabled(struct rq *rq)
1999 if (!sched_feat(HRTICK))
2001 if (!cpu_active(cpu_of(rq)))
2003 return hrtimer_is_hres_active(&rq->hrtick_timer);
2006 void hrtick_start(struct rq *rq, u64 delay);
2010 static inline int hrtick_enabled(struct rq *rq)
2015 #endif /* CONFIG_SCHED_HRTICK */
2017 #ifndef arch_scale_freq_tick
2018 static __always_inline
2019 void arch_scale_freq_tick(void)
2024 #ifndef arch_scale_freq_capacity
2025 static __always_inline
2026 unsigned long arch_scale_freq_capacity(int cpu)
2028 return SCHED_CAPACITY_SCALE;
2033 #ifdef CONFIG_PREEMPTION
2035 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2038 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2039 * way at the expense of forcing extra atomic operations in all
2040 * invocations. This assures that the double_lock is acquired using the
2041 * same underlying policy as the spinlock_t on this architecture, which
2042 * reduces latency compared to the unfair variant below. However, it
2043 * also adds more overhead and therefore may reduce throughput.
2045 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2046 __releases(this_rq->lock)
2047 __acquires(busiest->lock)
2048 __acquires(this_rq->lock)
2050 raw_spin_unlock(&this_rq->lock);
2051 double_rq_lock(this_rq, busiest);
2058 * Unfair double_lock_balance: Optimizes throughput at the expense of
2059 * latency by eliminating extra atomic operations when the locks are
2060 * already in proper order on entry. This favors lower CPU-ids and will
2061 * grant the double lock to lower CPUs over higher ids under contention,
2062 * regardless of entry order into the function.
2064 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2065 __releases(this_rq->lock)
2066 __acquires(busiest->lock)
2067 __acquires(this_rq->lock)
2071 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2072 if (busiest < this_rq) {
2073 raw_spin_unlock(&this_rq->lock);
2074 raw_spin_lock(&busiest->lock);
2075 raw_spin_lock_nested(&this_rq->lock,
2076 SINGLE_DEPTH_NESTING);
2079 raw_spin_lock_nested(&busiest->lock,
2080 SINGLE_DEPTH_NESTING);
2085 #endif /* CONFIG_PREEMPTION */
2088 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2090 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2092 if (unlikely(!irqs_disabled())) {
2093 /* printk() doesn't work well under rq->lock */
2094 raw_spin_unlock(&this_rq->lock);
2098 return _double_lock_balance(this_rq, busiest);
2101 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2102 __releases(busiest->lock)
2104 raw_spin_unlock(&busiest->lock);
2105 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2108 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2114 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2117 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2123 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2126 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2132 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2136 * double_rq_lock - safely lock two runqueues
2138 * Note this does not disable interrupts like task_rq_lock,
2139 * you need to do so manually before calling.
2141 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2142 __acquires(rq1->lock)
2143 __acquires(rq2->lock)
2145 BUG_ON(!irqs_disabled());
2147 raw_spin_lock(&rq1->lock);
2148 __acquire(rq2->lock); /* Fake it out ;) */
2151 raw_spin_lock(&rq1->lock);
2152 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2154 raw_spin_lock(&rq2->lock);
2155 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2161 * double_rq_unlock - safely unlock two runqueues
2163 * Note this does not restore interrupts like task_rq_unlock,
2164 * you need to do so manually after calling.
2166 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2167 __releases(rq1->lock)
2168 __releases(rq2->lock)
2170 raw_spin_unlock(&rq1->lock);
2172 raw_spin_unlock(&rq2->lock);
2174 __release(rq2->lock);
2177 extern void set_rq_online (struct rq *rq);
2178 extern void set_rq_offline(struct rq *rq);
2179 extern bool sched_smp_initialized;
2181 #else /* CONFIG_SMP */
2184 * double_rq_lock - safely lock two runqueues
2186 * Note this does not disable interrupts like task_rq_lock,
2187 * you need to do so manually before calling.
2189 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2190 __acquires(rq1->lock)
2191 __acquires(rq2->lock)
2193 BUG_ON(!irqs_disabled());
2195 raw_spin_lock(&rq1->lock);
2196 __acquire(rq2->lock); /* Fake it out ;) */
2200 * double_rq_unlock - safely unlock two runqueues
2202 * Note this does not restore interrupts like task_rq_unlock,
2203 * you need to do so manually after calling.
2205 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2206 __releases(rq1->lock)
2207 __releases(rq2->lock)
2210 raw_spin_unlock(&rq1->lock);
2211 __release(rq2->lock);
2216 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2217 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2219 #ifdef CONFIG_SCHED_DEBUG
2220 extern bool sched_debug_enabled;
2222 extern void print_cfs_stats(struct seq_file *m, int cpu);
2223 extern void print_rt_stats(struct seq_file *m, int cpu);
2224 extern void print_dl_stats(struct seq_file *m, int cpu);
2225 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2226 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2227 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2228 #ifdef CONFIG_NUMA_BALANCING
2230 show_numa_stats(struct task_struct *p, struct seq_file *m);
2232 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2233 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2234 #endif /* CONFIG_NUMA_BALANCING */
2235 #endif /* CONFIG_SCHED_DEBUG */
2237 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2238 extern void init_rt_rq(struct rt_rq *rt_rq);
2239 extern void init_dl_rq(struct dl_rq *dl_rq);
2241 extern void cfs_bandwidth_usage_inc(void);
2242 extern void cfs_bandwidth_usage_dec(void);
2244 #ifdef CONFIG_NO_HZ_COMMON
2245 #define NOHZ_BALANCE_KICK_BIT 0
2246 #define NOHZ_STATS_KICK_BIT 1
2248 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2249 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2251 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2253 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2255 extern void nohz_balance_exit_idle(struct rq *rq);
2257 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2263 void __dl_update(struct dl_bw *dl_b, s64 bw)
2265 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2268 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2269 "sched RCU must be held");
2270 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2271 struct rq *rq = cpu_rq(i);
2273 rq->dl.extra_bw += bw;
2278 void __dl_update(struct dl_bw *dl_b, s64 bw)
2280 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2287 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2292 struct u64_stats_sync sync;
2295 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2298 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2299 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2300 * and never move forward.
2302 static inline u64 irq_time_read(int cpu)
2304 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2309 seq = __u64_stats_fetch_begin(&irqtime->sync);
2310 total = irqtime->total;
2311 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2315 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2317 #ifdef CONFIG_CPU_FREQ
2318 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2321 * cpufreq_update_util - Take a note about CPU utilization changes.
2322 * @rq: Runqueue to carry out the update for.
2323 * @flags: Update reason flags.
2325 * This function is called by the scheduler on the CPU whose utilization is
2328 * It can only be called from RCU-sched read-side critical sections.
2330 * The way cpufreq is currently arranged requires it to evaluate the CPU
2331 * performance state (frequency/voltage) on a regular basis to prevent it from
2332 * being stuck in a completely inadequate performance level for too long.
2333 * That is not guaranteed to happen if the updates are only triggered from CFS
2334 * and DL, though, because they may not be coming in if only RT tasks are
2335 * active all the time (or there are RT tasks only).
2337 * As a workaround for that issue, this function is called periodically by the
2338 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2339 * but that really is a band-aid. Going forward it should be replaced with
2340 * solutions targeted more specifically at RT tasks.
2342 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2344 struct update_util_data *data;
2346 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2349 data->func(data, rq_clock(rq), flags);
2352 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2353 #endif /* CONFIG_CPU_FREQ */
2355 #ifdef CONFIG_UCLAMP_TASK
2356 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2358 static __always_inline
2359 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2360 struct task_struct *p)
2362 unsigned long min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2363 unsigned long max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2366 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2367 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2371 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2372 * RUNNABLE tasks with _different_ clamps, we can end up with an
2373 * inversion. Fix it now when the clamps are applied.
2375 if (unlikely(min_util >= max_util))
2378 return clamp(util, min_util, max_util);
2380 #else /* CONFIG_UCLAMP_TASK */
2382 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2383 struct task_struct *p)
2387 #endif /* CONFIG_UCLAMP_TASK */
2389 #ifdef arch_scale_freq_capacity
2390 # ifndef arch_scale_freq_invariant
2391 # define arch_scale_freq_invariant() true
2394 # define arch_scale_freq_invariant() false
2398 static inline unsigned long capacity_orig_of(int cpu)
2400 return cpu_rq(cpu)->cpu_capacity_orig;
2405 * enum schedutil_type - CPU utilization type
2406 * @FREQUENCY_UTIL: Utilization used to select frequency
2407 * @ENERGY_UTIL: Utilization used during energy calculation
2409 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2410 * need to be aggregated differently depending on the usage made of them. This
2411 * enum is used within schedutil_freq_util() to differentiate the types of
2412 * utilization expected by the callers, and adjust the aggregation accordingly.
2414 enum schedutil_type {
2419 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2421 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2422 unsigned long max, enum schedutil_type type,
2423 struct task_struct *p);
2425 static inline unsigned long cpu_bw_dl(struct rq *rq)
2427 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2430 static inline unsigned long cpu_util_dl(struct rq *rq)
2432 return READ_ONCE(rq->avg_dl.util_avg);
2435 static inline unsigned long cpu_util_cfs(struct rq *rq)
2437 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2439 if (sched_feat(UTIL_EST)) {
2440 util = max_t(unsigned long, util,
2441 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2447 static inline unsigned long cpu_util_rt(struct rq *rq)
2449 return READ_ONCE(rq->avg_rt.util_avg);
2451 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2452 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2453 unsigned long max, enum schedutil_type type,
2454 struct task_struct *p)
2458 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2460 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2461 static inline unsigned long cpu_util_irq(struct rq *rq)
2463 return rq->avg_irq.util_avg;
2467 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2469 util *= (max - irq);
2476 static inline unsigned long cpu_util_irq(struct rq *rq)
2482 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2488 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2490 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2492 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2494 static inline bool sched_energy_enabled(void)
2496 return static_branch_unlikely(&sched_energy_present);
2499 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2501 #define perf_domain_span(pd) NULL
2502 static inline bool sched_energy_enabled(void) { return false; }
2504 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2506 #ifdef CONFIG_MEMBARRIER
2508 * The scheduler provides memory barriers required by membarrier between:
2509 * - prior user-space memory accesses and store to rq->membarrier_state,
2510 * - store to rq->membarrier_state and following user-space memory accesses.
2511 * In the same way it provides those guarantees around store to rq->curr.
2513 static inline void membarrier_switch_mm(struct rq *rq,
2514 struct mm_struct *prev_mm,
2515 struct mm_struct *next_mm)
2517 int membarrier_state;
2519 if (prev_mm == next_mm)
2522 membarrier_state = atomic_read(&next_mm->membarrier_state);
2523 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2526 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2529 static inline void membarrier_switch_mm(struct rq *rq,
2530 struct mm_struct *prev_mm,
2531 struct mm_struct *next_mm)
2537 static inline bool is_per_cpu_kthread(struct task_struct *p)
2539 if (!(p->flags & PF_KTHREAD))
2542 if (p->nr_cpus_allowed != 1)
2549 void swake_up_all_locked(struct swait_queue_head *q);
2550 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);