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 #include <trace/events/sched.h>
80 #ifdef CONFIG_SCHED_DEBUG
81 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
83 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
89 /* task_struct::on_rq states: */
90 #define TASK_ON_RQ_QUEUED 1
91 #define TASK_ON_RQ_MIGRATING 2
93 extern __read_mostly int scheduler_running;
95 extern unsigned long calc_load_update;
96 extern atomic_long_t calc_load_tasks;
98 extern void calc_global_load_tick(struct rq *this_rq);
99 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
101 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
103 * Helpers for converting nanosecond timing to jiffy resolution
105 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108 * Increase resolution of nice-level calculations for 64-bit architectures.
109 * The extra resolution improves shares distribution and load balancing of
110 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
111 * hierarchies, especially on larger systems. This is not a user-visible change
112 * and does not change the user-interface for setting shares/weights.
114 * We increase resolution only if we have enough bits to allow this increased
115 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
116 * are pretty high and the returns do not justify the increased costs.
118 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
119 * increase coverage and consistency always enable it on 64-bit platforms.
122 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
123 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load_down(w) \
126 unsigned long __w = (w); \
128 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
132 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
133 # define scale_load(w) (w)
134 # define scale_load_down(w) (w)
138 * Task weight (visible to users) and its load (invisible to users) have
139 * independent resolution, but they should be well calibrated. We use
140 * scale_load() and scale_load_down(w) to convert between them. The
141 * following must be true:
143 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
146 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
149 * Single value that decides SCHED_DEADLINE internal math precision.
150 * 10 -> just above 1us
151 * 9 -> just above 0.5us
156 * Single value that denotes runtime == period, ie unlimited time.
158 #define RUNTIME_INF ((u64)~0ULL)
160 static inline int idle_policy(int policy)
162 return policy == SCHED_IDLE;
164 static inline int fair_policy(int policy)
166 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
169 static inline int rt_policy(int policy)
171 return policy == SCHED_FIFO || policy == SCHED_RR;
174 static inline int dl_policy(int policy)
176 return policy == SCHED_DEADLINE;
178 static inline bool valid_policy(int policy)
180 return idle_policy(policy) || fair_policy(policy) ||
181 rt_policy(policy) || dl_policy(policy);
184 static inline int task_has_idle_policy(struct task_struct *p)
186 return idle_policy(p->policy);
189 static inline int task_has_rt_policy(struct task_struct *p)
191 return rt_policy(p->policy);
194 static inline int task_has_dl_policy(struct task_struct *p)
196 return dl_policy(p->policy);
199 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
201 static inline void update_avg(u64 *avg, u64 sample)
203 s64 diff = sample - *avg;
208 * !! For sched_setattr_nocheck() (kernel) only !!
210 * This is actually gross. :(
212 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
213 * tasks, but still be able to sleep. We need this on platforms that cannot
214 * atomically change clock frequency. Remove once fast switching will be
215 * available on such platforms.
217 * SUGOV stands for SchedUtil GOVernor.
219 #define SCHED_FLAG_SUGOV 0x10000000
221 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
223 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
224 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
231 * Tells if entity @a should preempt entity @b.
234 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
236 return dl_entity_is_special(a) ||
237 dl_time_before(a->deadline, b->deadline);
241 * This is the priority-queue data structure of the RT scheduling class:
243 struct rt_prio_array {
244 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
245 struct list_head queue[MAX_RT_PRIO];
248 struct rt_bandwidth {
249 /* nests inside the rq lock: */
250 raw_spinlock_t rt_runtime_lock;
253 struct hrtimer rt_period_timer;
254 unsigned int rt_period_active;
257 void __dl_clear_params(struct task_struct *p);
259 struct dl_bandwidth {
260 raw_spinlock_t dl_runtime_lock;
265 static inline int dl_bandwidth_enabled(void)
267 return sysctl_sched_rt_runtime >= 0;
271 * To keep the bandwidth of -deadline tasks under control
272 * we need some place where:
273 * - store the maximum -deadline bandwidth of each cpu;
274 * - cache the fraction of bandwidth that is currently allocated in
277 * This is all done in the data structure below. It is similar to the
278 * one used for RT-throttling (rt_bandwidth), with the main difference
279 * that, since here we are only interested in admission control, we
280 * do not decrease any runtime while the group "executes", neither we
281 * need a timer to replenish it.
283 * With respect to SMP, bandwidth is given on a per root domain basis,
285 * - bw (< 100%) is the deadline bandwidth of each CPU;
286 * - total_bw is the currently allocated bandwidth in each root domain;
294 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
297 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
299 dl_b->total_bw -= tsk_bw;
300 __dl_update(dl_b, (s32)tsk_bw / cpus);
304 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
306 dl_b->total_bw += tsk_bw;
307 __dl_update(dl_b, -((s32)tsk_bw / cpus));
310 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
311 u64 old_bw, u64 new_bw)
313 return dl_b->bw != -1 &&
314 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
318 * Verify the fitness of task @p to run on @cpu taking into account the
319 * CPU original capacity and the runtime/deadline ratio of the task.
321 * The function will return true if the CPU original capacity of the
322 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
323 * task and false otherwise.
325 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
327 unsigned long cap = arch_scale_cpu_capacity(cpu);
329 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
332 extern void init_dl_bw(struct dl_bw *dl_b);
333 extern int sched_dl_global_validate(void);
334 extern void sched_dl_do_global(void);
335 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
336 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
337 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
338 extern bool __checkparam_dl(const struct sched_attr *attr);
339 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
340 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
341 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
342 extern bool dl_cpu_busy(unsigned int cpu);
344 #ifdef CONFIG_CGROUP_SCHED
346 #include <linux/cgroup.h>
347 #include <linux/psi.h>
352 extern struct list_head task_groups;
354 struct cfs_bandwidth {
355 #ifdef CONFIG_CFS_BANDWIDTH
360 s64 hierarchical_quota;
365 struct hrtimer period_timer;
366 struct hrtimer slack_timer;
367 struct list_head throttled_cfs_rq;
376 /* Task group related information */
378 struct cgroup_subsys_state css;
380 #ifdef CONFIG_FAIR_GROUP_SCHED
381 /* schedulable entities of this group on each CPU */
382 struct sched_entity **se;
383 /* runqueue "owned" by this group on each CPU */
384 struct cfs_rq **cfs_rq;
385 unsigned long shares;
389 * load_avg can be heavily contended at clock tick time, so put
390 * it in its own cacheline separated from the fields above which
391 * will also be accessed at each tick.
393 atomic_long_t load_avg ____cacheline_aligned;
397 #ifdef CONFIG_RT_GROUP_SCHED
398 struct sched_rt_entity **rt_se;
399 struct rt_rq **rt_rq;
401 struct rt_bandwidth rt_bandwidth;
405 struct list_head list;
407 struct task_group *parent;
408 struct list_head siblings;
409 struct list_head children;
411 #ifdef CONFIG_SCHED_AUTOGROUP
412 struct autogroup *autogroup;
415 struct cfs_bandwidth cfs_bandwidth;
417 #ifdef CONFIG_UCLAMP_TASK_GROUP
418 /* The two decimal precision [%] value requested from user-space */
419 unsigned int uclamp_pct[UCLAMP_CNT];
420 /* Clamp values requested for a task group */
421 struct uclamp_se uclamp_req[UCLAMP_CNT];
422 /* Effective clamp values used for a task group */
423 struct uclamp_se uclamp[UCLAMP_CNT];
428 #ifdef CONFIG_FAIR_GROUP_SCHED
429 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
432 * A weight of 0 or 1 can cause arithmetics problems.
433 * A weight of a cfs_rq is the sum of weights of which entities
434 * are queued on this cfs_rq, so a weight of a entity should not be
435 * too large, so as the shares value of a task group.
436 * (The default weight is 1024 - so there's no practical
437 * limitation from this.)
439 #define MIN_SHARES (1UL << 1)
440 #define MAX_SHARES (1UL << 18)
443 typedef int (*tg_visitor)(struct task_group *, void *);
445 extern int walk_tg_tree_from(struct task_group *from,
446 tg_visitor down, tg_visitor up, void *data);
449 * Iterate the full tree, calling @down when first entering a node and @up when
450 * leaving it for the final time.
452 * Caller must hold rcu_lock or sufficient equivalent.
454 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
456 return walk_tg_tree_from(&root_task_group, down, up, data);
459 extern int tg_nop(struct task_group *tg, void *data);
461 extern void free_fair_sched_group(struct task_group *tg);
462 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
463 extern void online_fair_sched_group(struct task_group *tg);
464 extern void unregister_fair_sched_group(struct task_group *tg);
465 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
466 struct sched_entity *se, int cpu,
467 struct sched_entity *parent);
468 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
470 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
471 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
472 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
474 extern void free_rt_sched_group(struct task_group *tg);
475 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
476 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
477 struct sched_rt_entity *rt_se, int cpu,
478 struct sched_rt_entity *parent);
479 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
480 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
481 extern long sched_group_rt_runtime(struct task_group *tg);
482 extern long sched_group_rt_period(struct task_group *tg);
483 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
485 extern struct task_group *sched_create_group(struct task_group *parent);
486 extern void sched_online_group(struct task_group *tg,
487 struct task_group *parent);
488 extern void sched_destroy_group(struct task_group *tg);
489 extern void sched_offline_group(struct task_group *tg);
491 extern void sched_move_task(struct task_struct *tsk);
493 #ifdef CONFIG_FAIR_GROUP_SCHED
494 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
497 extern void set_task_rq_fair(struct sched_entity *se,
498 struct cfs_rq *prev, struct cfs_rq *next);
499 #else /* !CONFIG_SMP */
500 static inline void set_task_rq_fair(struct sched_entity *se,
501 struct cfs_rq *prev, struct cfs_rq *next) { }
502 #endif /* CONFIG_SMP */
503 #endif /* CONFIG_FAIR_GROUP_SCHED */
505 #else /* CONFIG_CGROUP_SCHED */
507 struct cfs_bandwidth { };
509 #endif /* CONFIG_CGROUP_SCHED */
511 /* CFS-related fields in a runqueue */
513 struct load_weight load;
514 unsigned int nr_running;
515 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
516 unsigned int idle_h_nr_running; /* SCHED_IDLE */
521 u64 min_vruntime_copy;
524 struct rb_root_cached tasks_timeline;
527 * 'curr' points to currently running entity on this cfs_rq.
528 * It is set to NULL otherwise (i.e when none are currently running).
530 struct sched_entity *curr;
531 struct sched_entity *next;
532 struct sched_entity *last;
533 struct sched_entity *skip;
535 #ifdef CONFIG_SCHED_DEBUG
536 unsigned int nr_spread_over;
543 struct sched_avg avg;
545 u64 load_last_update_time_copy;
548 raw_spinlock_t lock ____cacheline_aligned;
550 unsigned long load_avg;
551 unsigned long util_avg;
552 unsigned long runnable_avg;
555 #ifdef CONFIG_FAIR_GROUP_SCHED
556 unsigned long tg_load_avg_contrib;
558 long prop_runnable_sum;
561 * h_load = weight * f(tg)
563 * Where f(tg) is the recursive weight fraction assigned to
566 unsigned long h_load;
567 u64 last_h_load_update;
568 struct sched_entity *h_load_next;
569 #endif /* CONFIG_FAIR_GROUP_SCHED */
570 #endif /* CONFIG_SMP */
572 #ifdef CONFIG_FAIR_GROUP_SCHED
573 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
576 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
577 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
578 * (like users, containers etc.)
580 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
581 * This list is used during load balance.
584 struct list_head leaf_cfs_rq_list;
585 struct task_group *tg; /* group that "owns" this runqueue */
587 #ifdef CONFIG_CFS_BANDWIDTH
589 s64 runtime_remaining;
592 u64 throttled_clock_task;
593 u64 throttled_clock_task_time;
596 struct list_head throttled_list;
597 #endif /* CONFIG_CFS_BANDWIDTH */
598 #endif /* CONFIG_FAIR_GROUP_SCHED */
601 static inline int rt_bandwidth_enabled(void)
603 return sysctl_sched_rt_runtime >= 0;
606 /* RT IPI pull logic requires IRQ_WORK */
607 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
608 # define HAVE_RT_PUSH_IPI
611 /* Real-Time classes' related field in a runqueue: */
613 struct rt_prio_array active;
614 unsigned int rt_nr_running;
615 unsigned int rr_nr_running;
616 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
618 int curr; /* highest queued rt task prio */
620 int next; /* next highest */
625 unsigned long rt_nr_migratory;
626 unsigned long rt_nr_total;
628 struct plist_head pushable_tasks;
630 #endif /* CONFIG_SMP */
636 /* Nests inside the rq lock: */
637 raw_spinlock_t rt_runtime_lock;
639 #ifdef CONFIG_RT_GROUP_SCHED
640 unsigned long rt_nr_boosted;
643 struct task_group *tg;
647 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
649 return rt_rq->rt_queued && rt_rq->rt_nr_running;
652 /* Deadline class' related fields in a runqueue */
654 /* runqueue is an rbtree, ordered by deadline */
655 struct rb_root_cached root;
657 unsigned long dl_nr_running;
661 * Deadline values of the currently executing and the
662 * earliest ready task on this rq. Caching these facilitates
663 * the decision whether or not a ready but not running task
664 * should migrate somewhere else.
671 unsigned long dl_nr_migratory;
675 * Tasks on this rq that can be pushed away. They are kept in
676 * an rb-tree, ordered by tasks' deadlines, with caching
677 * of the leftmost (earliest deadline) element.
679 struct rb_root_cached pushable_dl_tasks_root;
684 * "Active utilization" for this runqueue: increased when a
685 * task wakes up (becomes TASK_RUNNING) and decreased when a
691 * Utilization of the tasks "assigned" to this runqueue (including
692 * the tasks that are in runqueue and the tasks that executed on this
693 * CPU and blocked). Increased when a task moves to this runqueue, and
694 * decreased when the task moves away (migrates, changes scheduling
695 * policy, or terminates).
696 * This is needed to compute the "inactive utilization" for the
697 * runqueue (inactive utilization = this_bw - running_bw).
703 * Inverse of the fraction of CPU utilization that can be reclaimed
704 * by the GRUB algorithm.
709 #ifdef CONFIG_FAIR_GROUP_SCHED
710 /* An entity is a task if it doesn't "own" a runqueue */
711 #define entity_is_task(se) (!se->my_q)
713 static inline void se_update_runnable(struct sched_entity *se)
715 if (!entity_is_task(se))
716 se->runnable_weight = se->my_q->h_nr_running;
719 static inline long se_runnable(struct sched_entity *se)
721 if (entity_is_task(se))
724 return se->runnable_weight;
728 #define entity_is_task(se) 1
730 static inline void se_update_runnable(struct sched_entity *se) {}
732 static inline long se_runnable(struct sched_entity *se)
740 * XXX we want to get rid of these helpers and use the full load resolution.
742 static inline long se_weight(struct sched_entity *se)
744 return scale_load_down(se->load.weight);
748 static inline bool sched_asym_prefer(int a, int b)
750 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
754 struct em_perf_domain *em_pd;
755 struct perf_domain *next;
759 /* Scheduling group status flags */
760 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
761 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
764 * We add the notion of a root-domain which will be used to define per-domain
765 * variables. Each exclusive cpuset essentially defines an island domain by
766 * fully partitioning the member CPUs from any other cpuset. Whenever a new
767 * exclusive cpuset is created, we also create and attach a new root-domain
776 cpumask_var_t online;
779 * Indicate pullable load on at least one CPU, e.g:
780 * - More than one runnable task
781 * - Running task is misfit
785 /* Indicate one or more cpus over-utilized (tipping point) */
789 * The bit corresponding to a CPU gets set here if such CPU has more
790 * than one runnable -deadline task (as it is below for RT tasks).
792 cpumask_var_t dlo_mask;
798 * Indicate whether a root_domain's dl_bw has been checked or
799 * updated. It's monotonously increasing value.
801 * Also, some corner cases, like 'wrap around' is dangerous, but given
802 * that u64 is 'big enough'. So that shouldn't be a concern.
806 #ifdef HAVE_RT_PUSH_IPI
808 * For IPI pull requests, loop across the rto_mask.
810 struct irq_work rto_push_work;
811 raw_spinlock_t rto_lock;
812 /* These are only updated and read within rto_lock */
815 /* These atomics are updated outside of a lock */
816 atomic_t rto_loop_next;
817 atomic_t rto_loop_start;
820 * The "RT overload" flag: it gets set if a CPU has more than
821 * one runnable RT task.
823 cpumask_var_t rto_mask;
824 struct cpupri cpupri;
826 unsigned long max_cpu_capacity;
829 * NULL-terminated list of performance domains intersecting with the
830 * CPUs of the rd. Protected by RCU.
832 struct perf_domain __rcu *pd;
835 extern void init_defrootdomain(void);
836 extern int sched_init_domains(const struct cpumask *cpu_map);
837 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
838 extern void sched_get_rd(struct root_domain *rd);
839 extern void sched_put_rd(struct root_domain *rd);
841 #ifdef HAVE_RT_PUSH_IPI
842 extern void rto_push_irq_work_func(struct irq_work *work);
844 #endif /* CONFIG_SMP */
846 #ifdef CONFIG_UCLAMP_TASK
848 * struct uclamp_bucket - Utilization clamp bucket
849 * @value: utilization clamp value for tasks on this clamp bucket
850 * @tasks: number of RUNNABLE tasks on this clamp bucket
852 * Keep track of how many tasks are RUNNABLE for a given utilization
855 struct uclamp_bucket {
856 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
857 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
861 * struct uclamp_rq - rq's utilization clamp
862 * @value: currently active clamp values for a rq
863 * @bucket: utilization clamp buckets affecting a rq
865 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
866 * A clamp value is affecting a rq when there is at least one task RUNNABLE
867 * (or actually running) with that value.
869 * There are up to UCLAMP_CNT possible different clamp values, currently there
870 * are only two: minimum utilization and maximum utilization.
872 * All utilization clamping values are MAX aggregated, since:
873 * - for util_min: we want to run the CPU at least at the max of the minimum
874 * utilization required by its currently RUNNABLE tasks.
875 * - for util_max: we want to allow the CPU to run up to the max of the
876 * maximum utilization allowed by its currently RUNNABLE tasks.
878 * Since on each system we expect only a limited number of different
879 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
880 * the metrics required to compute all the per-rq utilization clamp values.
884 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
887 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
888 #endif /* CONFIG_UCLAMP_TASK */
891 * This is the main, per-CPU runqueue data structure.
893 * Locking rule: those places that want to lock multiple runqueues
894 * (such as the load balancing or the thread migration code), lock
895 * acquire operations must be ordered by ascending &runqueue.
902 * nr_running and cpu_load should be in the same cacheline because
903 * remote CPUs use both these fields when doing load calculation.
905 unsigned int nr_running;
906 #ifdef CONFIG_NUMA_BALANCING
907 unsigned int nr_numa_running;
908 unsigned int nr_preferred_running;
909 unsigned int numa_migrate_on;
911 #ifdef CONFIG_NO_HZ_COMMON
913 unsigned long last_blocked_load_update_tick;
914 unsigned int has_blocked_load;
915 call_single_data_t nohz_csd;
916 #endif /* CONFIG_SMP */
917 unsigned int nohz_tick_stopped;
919 #endif /* CONFIG_NO_HZ_COMMON */
922 unsigned int ttwu_pending;
926 #ifdef CONFIG_UCLAMP_TASK
927 /* Utilization clamp values based on CPU's RUNNABLE tasks */
928 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
929 unsigned int uclamp_flags;
930 #define UCLAMP_FLAG_IDLE 0x01
937 #ifdef CONFIG_FAIR_GROUP_SCHED
938 /* list of leaf cfs_rq on this CPU: */
939 struct list_head leaf_cfs_rq_list;
940 struct list_head *tmp_alone_branch;
941 #endif /* CONFIG_FAIR_GROUP_SCHED */
944 * This is part of a global counter where only the total sum
945 * over all CPUs matters. A task can increase this counter on
946 * one CPU and if it got migrated afterwards it may decrease
947 * it on another CPU. Always updated under the runqueue lock:
949 unsigned long nr_uninterruptible;
951 struct task_struct __rcu *curr;
952 struct task_struct *idle;
953 struct task_struct *stop;
954 unsigned long next_balance;
955 struct mm_struct *prev_mm;
957 unsigned int clock_update_flags;
959 /* Ensure that all clocks are in the same cache line */
960 u64 clock_task ____cacheline_aligned;
962 unsigned long lost_idle_time;
966 #ifdef CONFIG_MEMBARRIER
967 int membarrier_state;
971 struct root_domain *rd;
972 struct sched_domain __rcu *sd;
974 unsigned long cpu_capacity;
975 unsigned long cpu_capacity_orig;
977 struct callback_head *balance_callback;
978 unsigned char balance_push;
980 unsigned char nohz_idle_balance;
981 unsigned char idle_balance;
983 unsigned long misfit_task_load;
985 /* For active balancing */
988 struct cpu_stop_work active_balance_work;
990 /* CPU of this runqueue: */
994 struct list_head cfs_tasks;
996 struct sched_avg avg_rt;
997 struct sched_avg avg_dl;
998 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
999 struct sched_avg avg_irq;
1001 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1002 struct sched_avg avg_thermal;
1007 /* This is used to determine avg_idle's max value */
1008 u64 max_idle_balance_cost;
1010 #ifdef CONFIG_HOTPLUG_CPU
1011 struct rcuwait hotplug_wait;
1013 #endif /* CONFIG_SMP */
1015 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1018 #ifdef CONFIG_PARAVIRT
1019 u64 prev_steal_time;
1021 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1022 u64 prev_steal_time_rq;
1025 /* calc_load related fields */
1026 unsigned long calc_load_update;
1027 long calc_load_active;
1029 #ifdef CONFIG_SCHED_HRTICK
1031 call_single_data_t hrtick_csd;
1033 struct hrtimer hrtick_timer;
1034 ktime_t hrtick_time;
1037 #ifdef CONFIG_SCHEDSTATS
1039 struct sched_info rq_sched_info;
1040 unsigned long long rq_cpu_time;
1041 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1043 /* sys_sched_yield() stats */
1044 unsigned int yld_count;
1046 /* schedule() stats */
1047 unsigned int sched_count;
1048 unsigned int sched_goidle;
1050 /* try_to_wake_up() stats */
1051 unsigned int ttwu_count;
1052 unsigned int ttwu_local;
1055 #ifdef CONFIG_CPU_IDLE
1056 /* Must be inspected within a rcu lock section */
1057 struct cpuidle_state *idle_state;
1061 unsigned int nr_pinned;
1063 unsigned int push_busy;
1064 struct cpu_stop_work push_work;
1067 #ifdef CONFIG_FAIR_GROUP_SCHED
1069 /* CPU runqueue to which this cfs_rq is attached */
1070 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1077 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1079 return container_of(cfs_rq, struct rq, cfs);
1083 static inline int cpu_of(struct rq *rq)
1092 #define MDF_PUSH 0x01
1094 static inline bool is_migration_disabled(struct task_struct *p)
1097 return p->migration_disabled;
1103 #ifdef CONFIG_SCHED_SMT
1104 extern void __update_idle_core(struct rq *rq);
1106 static inline void update_idle_core(struct rq *rq)
1108 if (static_branch_unlikely(&sched_smt_present))
1109 __update_idle_core(rq);
1113 static inline void update_idle_core(struct rq *rq) { }
1116 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1118 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1119 #define this_rq() this_cpu_ptr(&runqueues)
1120 #define task_rq(p) cpu_rq(task_cpu(p))
1121 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1122 #define raw_rq() raw_cpu_ptr(&runqueues)
1124 extern void update_rq_clock(struct rq *rq);
1126 static inline u64 __rq_clock_broken(struct rq *rq)
1128 return READ_ONCE(rq->clock);
1132 * rq::clock_update_flags bits
1134 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1135 * call to __schedule(). This is an optimisation to avoid
1136 * neighbouring rq clock updates.
1138 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1139 * in effect and calls to update_rq_clock() are being ignored.
1141 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1142 * made to update_rq_clock() since the last time rq::lock was pinned.
1144 * If inside of __schedule(), clock_update_flags will have been
1145 * shifted left (a left shift is a cheap operation for the fast path
1146 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1148 * if (rq-clock_update_flags >= RQCF_UPDATED)
1150 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1151 * one position though, because the next rq_unpin_lock() will shift it
1154 #define RQCF_REQ_SKIP 0x01
1155 #define RQCF_ACT_SKIP 0x02
1156 #define RQCF_UPDATED 0x04
1158 static inline void assert_clock_updated(struct rq *rq)
1161 * The only reason for not seeing a clock update since the
1162 * last rq_pin_lock() is if we're currently skipping updates.
1164 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1167 static inline u64 rq_clock(struct rq *rq)
1169 lockdep_assert_held(&rq->lock);
1170 assert_clock_updated(rq);
1175 static inline u64 rq_clock_task(struct rq *rq)
1177 lockdep_assert_held(&rq->lock);
1178 assert_clock_updated(rq);
1180 return rq->clock_task;
1184 * By default the decay is the default pelt decay period.
1185 * The decay shift can change the decay period in
1187 * Decay shift Decay period(ms)
1194 extern int sched_thermal_decay_shift;
1196 static inline u64 rq_clock_thermal(struct rq *rq)
1198 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1201 static inline void rq_clock_skip_update(struct rq *rq)
1203 lockdep_assert_held(&rq->lock);
1204 rq->clock_update_flags |= RQCF_REQ_SKIP;
1208 * See rt task throttling, which is the only time a skip
1209 * request is cancelled.
1211 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1213 lockdep_assert_held(&rq->lock);
1214 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1218 unsigned long flags;
1219 struct pin_cookie cookie;
1220 #ifdef CONFIG_SCHED_DEBUG
1222 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1223 * current pin context is stashed here in case it needs to be
1224 * restored in rq_repin_lock().
1226 unsigned int clock_update_flags;
1230 extern struct callback_head balance_push_callback;
1233 * Lockdep annotation that avoids accidental unlocks; it's like a
1234 * sticky/continuous lockdep_assert_held().
1236 * This avoids code that has access to 'struct rq *rq' (basically everything in
1237 * the scheduler) from accidentally unlocking the rq if they do not also have a
1238 * copy of the (on-stack) 'struct rq_flags rf'.
1240 * Also see Documentation/locking/lockdep-design.rst.
1242 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1244 rf->cookie = lockdep_pin_lock(&rq->lock);
1246 #ifdef CONFIG_SCHED_DEBUG
1247 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1248 rf->clock_update_flags = 0;
1250 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1255 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1257 #ifdef CONFIG_SCHED_DEBUG
1258 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1259 rf->clock_update_flags = RQCF_UPDATED;
1262 lockdep_unpin_lock(&rq->lock, rf->cookie);
1265 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1267 lockdep_repin_lock(&rq->lock, rf->cookie);
1269 #ifdef CONFIG_SCHED_DEBUG
1271 * Restore the value we stashed in @rf for this pin context.
1273 rq->clock_update_flags |= rf->clock_update_flags;
1277 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1278 __acquires(rq->lock);
1280 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1281 __acquires(p->pi_lock)
1282 __acquires(rq->lock);
1284 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1285 __releases(rq->lock)
1287 rq_unpin_lock(rq, rf);
1288 raw_spin_unlock(&rq->lock);
1292 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1293 __releases(rq->lock)
1294 __releases(p->pi_lock)
1296 rq_unpin_lock(rq, rf);
1297 raw_spin_unlock(&rq->lock);
1298 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1302 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1303 __acquires(rq->lock)
1305 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1306 rq_pin_lock(rq, rf);
1310 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1311 __acquires(rq->lock)
1313 raw_spin_lock_irq(&rq->lock);
1314 rq_pin_lock(rq, rf);
1318 rq_lock(struct rq *rq, struct rq_flags *rf)
1319 __acquires(rq->lock)
1321 raw_spin_lock(&rq->lock);
1322 rq_pin_lock(rq, rf);
1326 rq_relock(struct rq *rq, struct rq_flags *rf)
1327 __acquires(rq->lock)
1329 raw_spin_lock(&rq->lock);
1330 rq_repin_lock(rq, rf);
1334 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1335 __releases(rq->lock)
1337 rq_unpin_lock(rq, rf);
1338 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1342 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1343 __releases(rq->lock)
1345 rq_unpin_lock(rq, rf);
1346 raw_spin_unlock_irq(&rq->lock);
1350 rq_unlock(struct rq *rq, struct rq_flags *rf)
1351 __releases(rq->lock)
1353 rq_unpin_lock(rq, rf);
1354 raw_spin_unlock(&rq->lock);
1357 static inline struct rq *
1358 this_rq_lock_irq(struct rq_flags *rf)
1359 __acquires(rq->lock)
1363 local_irq_disable();
1370 enum numa_topology_type {
1375 extern enum numa_topology_type sched_numa_topology_type;
1376 extern int sched_max_numa_distance;
1377 extern bool find_numa_distance(int distance);
1378 extern void sched_init_numa(void);
1379 extern void sched_domains_numa_masks_set(unsigned int cpu);
1380 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1381 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1383 static inline void sched_init_numa(void) { }
1384 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1385 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1386 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1392 #ifdef CONFIG_NUMA_BALANCING
1393 /* The regions in numa_faults array from task_struct */
1394 enum numa_faults_stats {
1400 extern void sched_setnuma(struct task_struct *p, int node);
1401 extern int migrate_task_to(struct task_struct *p, int cpu);
1402 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1404 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1407 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1410 #endif /* CONFIG_NUMA_BALANCING */
1415 queue_balance_callback(struct rq *rq,
1416 struct callback_head *head,
1417 void (*func)(struct rq *rq))
1419 lockdep_assert_held(&rq->lock);
1421 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1424 head->func = (void (*)(struct callback_head *))func;
1425 head->next = rq->balance_callback;
1426 rq->balance_callback = head;
1429 #define rcu_dereference_check_sched_domain(p) \
1430 rcu_dereference_check((p), \
1431 lockdep_is_held(&sched_domains_mutex))
1434 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1435 * See destroy_sched_domains: call_rcu for details.
1437 * The domain tree of any CPU may only be accessed from within
1438 * preempt-disabled sections.
1440 #define for_each_domain(cpu, __sd) \
1441 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1442 __sd; __sd = __sd->parent)
1445 * highest_flag_domain - Return highest sched_domain containing flag.
1446 * @cpu: The CPU whose highest level of sched domain is to
1448 * @flag: The flag to check for the highest sched_domain
1449 * for the given CPU.
1451 * Returns the highest sched_domain of a CPU which contains the given flag.
1453 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1455 struct sched_domain *sd, *hsd = NULL;
1457 for_each_domain(cpu, sd) {
1458 if (!(sd->flags & flag))
1466 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1468 struct sched_domain *sd;
1470 for_each_domain(cpu, sd) {
1471 if (sd->flags & flag)
1478 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1479 DECLARE_PER_CPU(int, sd_llc_size);
1480 DECLARE_PER_CPU(int, sd_llc_id);
1481 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1482 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1483 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1484 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1485 extern struct static_key_false sched_asym_cpucapacity;
1487 struct sched_group_capacity {
1490 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1493 unsigned long capacity;
1494 unsigned long min_capacity; /* Min per-CPU capacity in group */
1495 unsigned long max_capacity; /* Max per-CPU capacity in group */
1496 unsigned long next_update;
1497 int imbalance; /* XXX unrelated to capacity but shared group state */
1499 #ifdef CONFIG_SCHED_DEBUG
1503 unsigned long cpumask[]; /* Balance mask */
1506 struct sched_group {
1507 struct sched_group *next; /* Must be a circular list */
1510 unsigned int group_weight;
1511 struct sched_group_capacity *sgc;
1512 int asym_prefer_cpu; /* CPU of highest priority in group */
1515 * The CPUs this group covers.
1517 * NOTE: this field is variable length. (Allocated dynamically
1518 * by attaching extra space to the end of the structure,
1519 * depending on how many CPUs the kernel has booted up with)
1521 unsigned long cpumask[];
1524 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1526 return to_cpumask(sg->cpumask);
1530 * See build_balance_mask().
1532 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1534 return to_cpumask(sg->sgc->cpumask);
1538 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1539 * @group: The group whose first CPU is to be returned.
1541 static inline unsigned int group_first_cpu(struct sched_group *group)
1543 return cpumask_first(sched_group_span(group));
1546 extern int group_balance_cpu(struct sched_group *sg);
1548 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1549 void register_sched_domain_sysctl(void);
1550 void dirty_sched_domain_sysctl(int cpu);
1551 void unregister_sched_domain_sysctl(void);
1553 static inline void register_sched_domain_sysctl(void)
1556 static inline void dirty_sched_domain_sysctl(int cpu)
1559 static inline void unregister_sched_domain_sysctl(void)
1564 extern void flush_smp_call_function_from_idle(void);
1566 #else /* !CONFIG_SMP: */
1567 static inline void flush_smp_call_function_from_idle(void) { }
1571 #include "autogroup.h"
1573 #ifdef CONFIG_CGROUP_SCHED
1576 * Return the group to which this tasks belongs.
1578 * We cannot use task_css() and friends because the cgroup subsystem
1579 * changes that value before the cgroup_subsys::attach() method is called,
1580 * therefore we cannot pin it and might observe the wrong value.
1582 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1583 * core changes this before calling sched_move_task().
1585 * Instead we use a 'copy' which is updated from sched_move_task() while
1586 * holding both task_struct::pi_lock and rq::lock.
1588 static inline struct task_group *task_group(struct task_struct *p)
1590 return p->sched_task_group;
1593 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1594 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1596 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1597 struct task_group *tg = task_group(p);
1600 #ifdef CONFIG_FAIR_GROUP_SCHED
1601 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1602 p->se.cfs_rq = tg->cfs_rq[cpu];
1603 p->se.parent = tg->se[cpu];
1606 #ifdef CONFIG_RT_GROUP_SCHED
1607 p->rt.rt_rq = tg->rt_rq[cpu];
1608 p->rt.parent = tg->rt_se[cpu];
1612 #else /* CONFIG_CGROUP_SCHED */
1614 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1615 static inline struct task_group *task_group(struct task_struct *p)
1620 #endif /* CONFIG_CGROUP_SCHED */
1622 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1624 set_task_rq(p, cpu);
1627 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1628 * successfully executed on another CPU. We must ensure that updates of
1629 * per-task data have been completed by this moment.
1632 #ifdef CONFIG_THREAD_INFO_IN_TASK
1633 WRITE_ONCE(p->cpu, cpu);
1635 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1642 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1644 #ifdef CONFIG_SCHED_DEBUG
1645 # include <linux/static_key.h>
1646 # define const_debug __read_mostly
1648 # define const_debug const
1651 #define SCHED_FEAT(name, enabled) \
1652 __SCHED_FEAT_##name ,
1655 #include "features.h"
1661 #ifdef CONFIG_SCHED_DEBUG
1664 * To support run-time toggling of sched features, all the translation units
1665 * (but core.c) reference the sysctl_sched_features defined in core.c.
1667 extern const_debug unsigned int sysctl_sched_features;
1669 #ifdef CONFIG_JUMP_LABEL
1670 #define SCHED_FEAT(name, enabled) \
1671 static __always_inline bool static_branch_##name(struct static_key *key) \
1673 return static_key_##enabled(key); \
1676 #include "features.h"
1679 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1680 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1682 #else /* !CONFIG_JUMP_LABEL */
1684 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1686 #endif /* CONFIG_JUMP_LABEL */
1688 #else /* !SCHED_DEBUG */
1691 * Each translation unit has its own copy of sysctl_sched_features to allow
1692 * constants propagation at compile time and compiler optimization based on
1695 #define SCHED_FEAT(name, enabled) \
1696 (1UL << __SCHED_FEAT_##name) * enabled |
1697 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1698 #include "features.h"
1702 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1704 #endif /* SCHED_DEBUG */
1706 extern struct static_key_false sched_numa_balancing;
1707 extern struct static_key_false sched_schedstats;
1709 static inline u64 global_rt_period(void)
1711 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1714 static inline u64 global_rt_runtime(void)
1716 if (sysctl_sched_rt_runtime < 0)
1719 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1722 static inline int task_current(struct rq *rq, struct task_struct *p)
1724 return rq->curr == p;
1727 static inline int task_running(struct rq *rq, struct task_struct *p)
1732 return task_current(rq, p);
1736 static inline int task_on_rq_queued(struct task_struct *p)
1738 return p->on_rq == TASK_ON_RQ_QUEUED;
1741 static inline int task_on_rq_migrating(struct task_struct *p)
1743 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1746 /* Wake flags. The first three directly map to some SD flag value */
1747 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
1748 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
1749 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
1751 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
1752 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
1753 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
1756 static_assert(WF_EXEC == SD_BALANCE_EXEC);
1757 static_assert(WF_FORK == SD_BALANCE_FORK);
1758 static_assert(WF_TTWU == SD_BALANCE_WAKE);
1762 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1763 * of tasks with abnormal "nice" values across CPUs the contribution that
1764 * each task makes to its run queue's load is weighted according to its
1765 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1766 * scaled version of the new time slice allocation that they receive on time
1770 #define WEIGHT_IDLEPRIO 3
1771 #define WMULT_IDLEPRIO 1431655765
1773 extern const int sched_prio_to_weight[40];
1774 extern const u32 sched_prio_to_wmult[40];
1777 * {de,en}queue flags:
1779 * DEQUEUE_SLEEP - task is no longer runnable
1780 * ENQUEUE_WAKEUP - task just became runnable
1782 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1783 * are in a known state which allows modification. Such pairs
1784 * should preserve as much state as possible.
1786 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1789 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1790 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1791 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1795 #define DEQUEUE_SLEEP 0x01
1796 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1797 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1798 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1800 #define ENQUEUE_WAKEUP 0x01
1801 #define ENQUEUE_RESTORE 0x02
1802 #define ENQUEUE_MOVE 0x04
1803 #define ENQUEUE_NOCLOCK 0x08
1805 #define ENQUEUE_HEAD 0x10
1806 #define ENQUEUE_REPLENISH 0x20
1808 #define ENQUEUE_MIGRATED 0x40
1810 #define ENQUEUE_MIGRATED 0x00
1813 #define RETRY_TASK ((void *)-1UL)
1815 struct sched_class {
1817 #ifdef CONFIG_UCLAMP_TASK
1821 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1822 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1823 void (*yield_task) (struct rq *rq);
1824 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1826 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1828 struct task_struct *(*pick_next_task)(struct rq *rq);
1830 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1831 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1834 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1835 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
1836 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1838 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1840 void (*set_cpus_allowed)(struct task_struct *p,
1841 const struct cpumask *newmask,
1844 void (*rq_online)(struct rq *rq);
1845 void (*rq_offline)(struct rq *rq);
1847 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
1850 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1851 void (*task_fork)(struct task_struct *p);
1852 void (*task_dead)(struct task_struct *p);
1855 * The switched_from() call is allowed to drop rq->lock, therefore we
1856 * cannot assume the switched_from/switched_to pair is serliazed by
1857 * rq->lock. They are however serialized by p->pi_lock.
1859 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1860 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1861 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1864 unsigned int (*get_rr_interval)(struct rq *rq,
1865 struct task_struct *task);
1867 void (*update_curr)(struct rq *rq);
1869 #define TASK_SET_GROUP 0
1870 #define TASK_MOVE_GROUP 1
1872 #ifdef CONFIG_FAIR_GROUP_SCHED
1873 void (*task_change_group)(struct task_struct *p, int type);
1877 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1879 WARN_ON_ONCE(rq->curr != prev);
1880 prev->sched_class->put_prev_task(rq, prev);
1883 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1885 WARN_ON_ONCE(rq->curr != next);
1886 next->sched_class->set_next_task(rq, next, false);
1891 * Helper to define a sched_class instance; each one is placed in a separate
1892 * section which is ordered by the linker script:
1894 * include/asm-generic/vmlinux.lds.h
1896 * Also enforce alignment on the instance, not the type, to guarantee layout.
1898 #define DEFINE_SCHED_CLASS(name) \
1899 const struct sched_class name##_sched_class \
1900 __aligned(__alignof__(struct sched_class)) \
1901 __section("__" #name "_sched_class")
1903 /* Defined in include/asm-generic/vmlinux.lds.h */
1904 extern struct sched_class __begin_sched_classes[];
1905 extern struct sched_class __end_sched_classes[];
1907 #define sched_class_highest (__end_sched_classes - 1)
1908 #define sched_class_lowest (__begin_sched_classes - 1)
1910 #define for_class_range(class, _from, _to) \
1911 for (class = (_from); class != (_to); class--)
1913 #define for_each_class(class) \
1914 for_class_range(class, sched_class_highest, sched_class_lowest)
1916 extern const struct sched_class stop_sched_class;
1917 extern const struct sched_class dl_sched_class;
1918 extern const struct sched_class rt_sched_class;
1919 extern const struct sched_class fair_sched_class;
1920 extern const struct sched_class idle_sched_class;
1922 static inline bool sched_stop_runnable(struct rq *rq)
1924 return rq->stop && task_on_rq_queued(rq->stop);
1927 static inline bool sched_dl_runnable(struct rq *rq)
1929 return rq->dl.dl_nr_running > 0;
1932 static inline bool sched_rt_runnable(struct rq *rq)
1934 return rq->rt.rt_queued > 0;
1937 static inline bool sched_fair_runnable(struct rq *rq)
1939 return rq->cfs.nr_running > 0;
1942 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1943 extern struct task_struct *pick_next_task_idle(struct rq *rq);
1945 #define SCA_CHECK 0x01
1946 #define SCA_MIGRATE_DISABLE 0x02
1947 #define SCA_MIGRATE_ENABLE 0x04
1951 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1953 extern void trigger_load_balance(struct rq *rq);
1955 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
1957 static inline struct task_struct *get_push_task(struct rq *rq)
1959 struct task_struct *p = rq->curr;
1961 lockdep_assert_held(&rq->lock);
1966 if (p->nr_cpus_allowed == 1)
1969 rq->push_busy = true;
1970 return get_task_struct(p);
1973 extern int push_cpu_stop(void *arg);
1977 #ifdef CONFIG_CPU_IDLE
1978 static inline void idle_set_state(struct rq *rq,
1979 struct cpuidle_state *idle_state)
1981 rq->idle_state = idle_state;
1984 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1986 SCHED_WARN_ON(!rcu_read_lock_held());
1988 return rq->idle_state;
1991 static inline void idle_set_state(struct rq *rq,
1992 struct cpuidle_state *idle_state)
1996 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2002 extern void schedule_idle(void);
2004 extern void sysrq_sched_debug_show(void);
2005 extern void sched_init_granularity(void);
2006 extern void update_max_interval(void);
2008 extern void init_sched_dl_class(void);
2009 extern void init_sched_rt_class(void);
2010 extern void init_sched_fair_class(void);
2012 extern void reweight_task(struct task_struct *p, int prio);
2014 extern void resched_curr(struct rq *rq);
2015 extern void resched_cpu(int cpu);
2017 extern struct rt_bandwidth def_rt_bandwidth;
2018 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2020 extern struct dl_bandwidth def_dl_bandwidth;
2021 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2022 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2023 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2026 #define BW_UNIT (1 << BW_SHIFT)
2027 #define RATIO_SHIFT 8
2028 #define MAX_BW_BITS (64 - BW_SHIFT)
2029 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2030 unsigned long to_ratio(u64 period, u64 runtime);
2032 extern void init_entity_runnable_average(struct sched_entity *se);
2033 extern void post_init_entity_util_avg(struct task_struct *p);
2035 #ifdef CONFIG_NO_HZ_FULL
2036 extern bool sched_can_stop_tick(struct rq *rq);
2037 extern int __init sched_tick_offload_init(void);
2040 * Tick may be needed by tasks in the runqueue depending on their policy and
2041 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2042 * nohz mode if necessary.
2044 static inline void sched_update_tick_dependency(struct rq *rq)
2046 int cpu = cpu_of(rq);
2048 if (!tick_nohz_full_cpu(cpu))
2051 if (sched_can_stop_tick(rq))
2052 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2054 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2057 static inline int sched_tick_offload_init(void) { return 0; }
2058 static inline void sched_update_tick_dependency(struct rq *rq) { }
2061 static inline void add_nr_running(struct rq *rq, unsigned count)
2063 unsigned prev_nr = rq->nr_running;
2065 rq->nr_running = prev_nr + count;
2066 if (trace_sched_update_nr_running_tp_enabled()) {
2067 call_trace_sched_update_nr_running(rq, count);
2071 if (prev_nr < 2 && rq->nr_running >= 2) {
2072 if (!READ_ONCE(rq->rd->overload))
2073 WRITE_ONCE(rq->rd->overload, 1);
2077 sched_update_tick_dependency(rq);
2080 static inline void sub_nr_running(struct rq *rq, unsigned count)
2082 rq->nr_running -= count;
2083 if (trace_sched_update_nr_running_tp_enabled()) {
2084 call_trace_sched_update_nr_running(rq, -count);
2087 /* Check if we still need preemption */
2088 sched_update_tick_dependency(rq);
2091 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2092 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2094 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2096 extern const_debug unsigned int sysctl_sched_nr_migrate;
2097 extern const_debug unsigned int sysctl_sched_migration_cost;
2099 #ifdef CONFIG_SCHED_HRTICK
2103 * - enabled by features
2104 * - hrtimer is actually high res
2106 static inline int hrtick_enabled(struct rq *rq)
2108 if (!cpu_active(cpu_of(rq)))
2110 return hrtimer_is_hres_active(&rq->hrtick_timer);
2113 static inline int hrtick_enabled_fair(struct rq *rq)
2115 if (!sched_feat(HRTICK))
2117 return hrtick_enabled(rq);
2120 static inline int hrtick_enabled_dl(struct rq *rq)
2122 if (!sched_feat(HRTICK_DL))
2124 return hrtick_enabled(rq);
2127 void hrtick_start(struct rq *rq, u64 delay);
2131 static inline int hrtick_enabled_fair(struct rq *rq)
2136 static inline int hrtick_enabled_dl(struct rq *rq)
2141 static inline int hrtick_enabled(struct rq *rq)
2146 #endif /* CONFIG_SCHED_HRTICK */
2148 #ifndef arch_scale_freq_tick
2149 static __always_inline
2150 void arch_scale_freq_tick(void)
2155 #ifndef arch_scale_freq_capacity
2157 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2158 * @cpu: the CPU in question.
2160 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2163 * ------ * SCHED_CAPACITY_SCALE
2166 static __always_inline
2167 unsigned long arch_scale_freq_capacity(int cpu)
2169 return SCHED_CAPACITY_SCALE;
2174 #ifdef CONFIG_PREEMPTION
2176 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2179 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2180 * way at the expense of forcing extra atomic operations in all
2181 * invocations. This assures that the double_lock is acquired using the
2182 * same underlying policy as the spinlock_t on this architecture, which
2183 * reduces latency compared to the unfair variant below. However, it
2184 * also adds more overhead and therefore may reduce throughput.
2186 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2187 __releases(this_rq->lock)
2188 __acquires(busiest->lock)
2189 __acquires(this_rq->lock)
2191 raw_spin_unlock(&this_rq->lock);
2192 double_rq_lock(this_rq, busiest);
2199 * Unfair double_lock_balance: Optimizes throughput at the expense of
2200 * latency by eliminating extra atomic operations when the locks are
2201 * already in proper order on entry. This favors lower CPU-ids and will
2202 * grant the double lock to lower CPUs over higher ids under contention,
2203 * regardless of entry order into the function.
2205 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2206 __releases(this_rq->lock)
2207 __acquires(busiest->lock)
2208 __acquires(this_rq->lock)
2212 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2213 if (busiest < this_rq) {
2214 raw_spin_unlock(&this_rq->lock);
2215 raw_spin_lock(&busiest->lock);
2216 raw_spin_lock_nested(&this_rq->lock,
2217 SINGLE_DEPTH_NESTING);
2220 raw_spin_lock_nested(&busiest->lock,
2221 SINGLE_DEPTH_NESTING);
2226 #endif /* CONFIG_PREEMPTION */
2229 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2231 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2233 if (unlikely(!irqs_disabled())) {
2234 /* printk() doesn't work well under rq->lock */
2235 raw_spin_unlock(&this_rq->lock);
2239 return _double_lock_balance(this_rq, busiest);
2242 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2243 __releases(busiest->lock)
2245 raw_spin_unlock(&busiest->lock);
2246 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2249 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2255 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2258 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2264 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2267 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2273 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2277 * double_rq_lock - safely lock two runqueues
2279 * Note this does not disable interrupts like task_rq_lock,
2280 * you need to do so manually before calling.
2282 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2283 __acquires(rq1->lock)
2284 __acquires(rq2->lock)
2286 BUG_ON(!irqs_disabled());
2288 raw_spin_lock(&rq1->lock);
2289 __acquire(rq2->lock); /* Fake it out ;) */
2292 raw_spin_lock(&rq1->lock);
2293 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2295 raw_spin_lock(&rq2->lock);
2296 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2302 * double_rq_unlock - safely unlock two runqueues
2304 * Note this does not restore interrupts like task_rq_unlock,
2305 * you need to do so manually after calling.
2307 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2308 __releases(rq1->lock)
2309 __releases(rq2->lock)
2311 raw_spin_unlock(&rq1->lock);
2313 raw_spin_unlock(&rq2->lock);
2315 __release(rq2->lock);
2318 extern void set_rq_online (struct rq *rq);
2319 extern void set_rq_offline(struct rq *rq);
2320 extern bool sched_smp_initialized;
2322 #else /* CONFIG_SMP */
2325 * double_rq_lock - safely lock two runqueues
2327 * Note this does not disable interrupts like task_rq_lock,
2328 * you need to do so manually before calling.
2330 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2331 __acquires(rq1->lock)
2332 __acquires(rq2->lock)
2334 BUG_ON(!irqs_disabled());
2336 raw_spin_lock(&rq1->lock);
2337 __acquire(rq2->lock); /* Fake it out ;) */
2341 * double_rq_unlock - safely unlock two runqueues
2343 * Note this does not restore interrupts like task_rq_unlock,
2344 * you need to do so manually after calling.
2346 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2347 __releases(rq1->lock)
2348 __releases(rq2->lock)
2351 raw_spin_unlock(&rq1->lock);
2352 __release(rq2->lock);
2357 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2358 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2360 #ifdef CONFIG_SCHED_DEBUG
2361 extern bool sched_debug_enabled;
2363 extern void print_cfs_stats(struct seq_file *m, int cpu);
2364 extern void print_rt_stats(struct seq_file *m, int cpu);
2365 extern void print_dl_stats(struct seq_file *m, int cpu);
2366 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2367 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2368 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2369 #ifdef CONFIG_NUMA_BALANCING
2371 show_numa_stats(struct task_struct *p, struct seq_file *m);
2373 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2374 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2375 #endif /* CONFIG_NUMA_BALANCING */
2376 #endif /* CONFIG_SCHED_DEBUG */
2378 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2379 extern void init_rt_rq(struct rt_rq *rt_rq);
2380 extern void init_dl_rq(struct dl_rq *dl_rq);
2382 extern void cfs_bandwidth_usage_inc(void);
2383 extern void cfs_bandwidth_usage_dec(void);
2385 #ifdef CONFIG_NO_HZ_COMMON
2386 #define NOHZ_BALANCE_KICK_BIT 0
2387 #define NOHZ_STATS_KICK_BIT 1
2389 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2390 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2392 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2394 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2396 extern void nohz_balance_exit_idle(struct rq *rq);
2398 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2404 void __dl_update(struct dl_bw *dl_b, s64 bw)
2406 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2409 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2410 "sched RCU must be held");
2411 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2412 struct rq *rq = cpu_rq(i);
2414 rq->dl.extra_bw += bw;
2419 void __dl_update(struct dl_bw *dl_b, s64 bw)
2421 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2428 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2433 struct u64_stats_sync sync;
2436 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2439 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2440 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2441 * and never move forward.
2443 static inline u64 irq_time_read(int cpu)
2445 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2450 seq = __u64_stats_fetch_begin(&irqtime->sync);
2451 total = irqtime->total;
2452 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2456 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2458 #ifdef CONFIG_CPU_FREQ
2459 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2462 * cpufreq_update_util - Take a note about CPU utilization changes.
2463 * @rq: Runqueue to carry out the update for.
2464 * @flags: Update reason flags.
2466 * This function is called by the scheduler on the CPU whose utilization is
2469 * It can only be called from RCU-sched read-side critical sections.
2471 * The way cpufreq is currently arranged requires it to evaluate the CPU
2472 * performance state (frequency/voltage) on a regular basis to prevent it from
2473 * being stuck in a completely inadequate performance level for too long.
2474 * That is not guaranteed to happen if the updates are only triggered from CFS
2475 * and DL, though, because they may not be coming in if only RT tasks are
2476 * active all the time (or there are RT tasks only).
2478 * As a workaround for that issue, this function is called periodically by the
2479 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2480 * but that really is a band-aid. Going forward it should be replaced with
2481 * solutions targeted more specifically at RT tasks.
2483 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2485 struct update_util_data *data;
2487 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2490 data->func(data, rq_clock(rq), flags);
2493 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2494 #endif /* CONFIG_CPU_FREQ */
2496 #ifdef CONFIG_UCLAMP_TASK
2497 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2500 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2501 * @rq: The rq to clamp against. Must not be NULL.
2502 * @util: The util value to clamp.
2503 * @p: The task to clamp against. Can be NULL if you want to clamp
2506 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2508 * If sched_uclamp_used static key is disabled, then just return the util
2509 * without any clamping since uclamp aggregation at the rq level in the fast
2510 * path is disabled, rendering this operation a NOP.
2512 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2513 * will return the correct effective uclamp value of the task even if the
2514 * static key is disabled.
2516 static __always_inline
2517 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2518 struct task_struct *p)
2520 unsigned long min_util;
2521 unsigned long max_util;
2523 if (!static_branch_likely(&sched_uclamp_used))
2526 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2527 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2530 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2531 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2535 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2536 * RUNNABLE tasks with _different_ clamps, we can end up with an
2537 * inversion. Fix it now when the clamps are applied.
2539 if (unlikely(min_util >= max_util))
2542 return clamp(util, min_util, max_util);
2546 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2547 * by default in the fast path and only gets turned on once userspace performs
2548 * an operation that requires it.
2550 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2553 static inline bool uclamp_is_used(void)
2555 return static_branch_likely(&sched_uclamp_used);
2557 #else /* CONFIG_UCLAMP_TASK */
2559 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2560 struct task_struct *p)
2565 static inline bool uclamp_is_used(void)
2569 #endif /* CONFIG_UCLAMP_TASK */
2571 #ifdef arch_scale_freq_capacity
2572 # ifndef arch_scale_freq_invariant
2573 # define arch_scale_freq_invariant() true
2576 # define arch_scale_freq_invariant() false
2580 static inline unsigned long capacity_orig_of(int cpu)
2582 return cpu_rq(cpu)->cpu_capacity_orig;
2586 * enum cpu_util_type - CPU utilization type
2587 * @FREQUENCY_UTIL: Utilization used to select frequency
2588 * @ENERGY_UTIL: Utilization used during energy calculation
2590 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2591 * need to be aggregated differently depending on the usage made of them. This
2592 * enum is used within effective_cpu_util() to differentiate the types of
2593 * utilization expected by the callers, and adjust the aggregation accordingly.
2595 enum cpu_util_type {
2600 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2601 unsigned long max, enum cpu_util_type type,
2602 struct task_struct *p);
2604 static inline unsigned long cpu_bw_dl(struct rq *rq)
2606 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2609 static inline unsigned long cpu_util_dl(struct rq *rq)
2611 return READ_ONCE(rq->avg_dl.util_avg);
2614 static inline unsigned long cpu_util_cfs(struct rq *rq)
2616 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2618 if (sched_feat(UTIL_EST)) {
2619 util = max_t(unsigned long, util,
2620 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2626 static inline unsigned long cpu_util_rt(struct rq *rq)
2628 return READ_ONCE(rq->avg_rt.util_avg);
2632 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2633 static inline unsigned long cpu_util_irq(struct rq *rq)
2635 return rq->avg_irq.util_avg;
2639 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2641 util *= (max - irq);
2648 static inline unsigned long cpu_util_irq(struct rq *rq)
2654 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2660 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2662 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2664 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2666 static inline bool sched_energy_enabled(void)
2668 return static_branch_unlikely(&sched_energy_present);
2671 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2673 #define perf_domain_span(pd) NULL
2674 static inline bool sched_energy_enabled(void) { return false; }
2676 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2678 #ifdef CONFIG_MEMBARRIER
2680 * The scheduler provides memory barriers required by membarrier between:
2681 * - prior user-space memory accesses and store to rq->membarrier_state,
2682 * - store to rq->membarrier_state and following user-space memory accesses.
2683 * In the same way it provides those guarantees around store to rq->curr.
2685 static inline void membarrier_switch_mm(struct rq *rq,
2686 struct mm_struct *prev_mm,
2687 struct mm_struct *next_mm)
2689 int membarrier_state;
2691 if (prev_mm == next_mm)
2694 membarrier_state = atomic_read(&next_mm->membarrier_state);
2695 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2698 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2701 static inline void membarrier_switch_mm(struct rq *rq,
2702 struct mm_struct *prev_mm,
2703 struct mm_struct *next_mm)
2709 static inline bool is_per_cpu_kthread(struct task_struct *p)
2711 if (!(p->flags & PF_KTHREAD))
2714 if (p->nr_cpus_allowed != 1)
2721 void swake_up_all_locked(struct swait_queue_head *q);
2722 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);