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) ((w) >> SCHED_FIXEDPOINT_SHIFT)
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) (w)
125 # define scale_load_down(w) (w)
129 * Task weight (visible to users) and its load (invisible to users) have
130 * independent resolution, but they should be well calibrated. We use
131 * scale_load() and scale_load_down(w) to convert between them. The
132 * following must be true:
134 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
137 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
140 * Single value that decides SCHED_DEADLINE internal math precision.
141 * 10 -> just above 1us
142 * 9 -> just above 0.5us
147 * Single value that denotes runtime == period, ie unlimited time.
149 #define RUNTIME_INF ((u64)~0ULL)
151 static inline int idle_policy(int policy)
153 return policy == SCHED_IDLE;
155 static inline int fair_policy(int policy)
157 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
160 static inline int rt_policy(int policy)
162 return policy == SCHED_FIFO || policy == SCHED_RR;
165 static inline int dl_policy(int policy)
167 return policy == SCHED_DEADLINE;
169 static inline bool valid_policy(int policy)
171 return idle_policy(policy) || fair_policy(policy) ||
172 rt_policy(policy) || dl_policy(policy);
175 static inline int task_has_idle_policy(struct task_struct *p)
177 return idle_policy(p->policy);
180 static inline int task_has_rt_policy(struct task_struct *p)
182 return rt_policy(p->policy);
185 static inline int task_has_dl_policy(struct task_struct *p)
187 return dl_policy(p->policy);
190 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
193 * !! For sched_setattr_nocheck() (kernel) only !!
195 * This is actually gross. :(
197 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
198 * tasks, but still be able to sleep. We need this on platforms that cannot
199 * atomically change clock frequency. Remove once fast switching will be
200 * available on such platforms.
202 * SUGOV stands for SchedUtil GOVernor.
204 #define SCHED_FLAG_SUGOV 0x10000000
206 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
208 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
209 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
216 * Tells if entity @a should preempt entity @b.
219 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
221 return dl_entity_is_special(a) ||
222 dl_time_before(a->deadline, b->deadline);
226 * This is the priority-queue data structure of the RT scheduling class:
228 struct rt_prio_array {
229 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
230 struct list_head queue[MAX_RT_PRIO];
233 struct rt_bandwidth {
234 /* nests inside the rq lock: */
235 raw_spinlock_t rt_runtime_lock;
238 struct hrtimer rt_period_timer;
239 unsigned int rt_period_active;
242 void __dl_clear_params(struct task_struct *p);
245 * To keep the bandwidth of -deadline tasks and groups under control
246 * we need some place where:
247 * - store the maximum -deadline bandwidth of the system (the group);
248 * - cache the fraction of that bandwidth that is currently allocated.
250 * This is all done in the data structure below. It is similar to the
251 * one used for RT-throttling (rt_bandwidth), with the main difference
252 * that, since here we are only interested in admission control, we
253 * do not decrease any runtime while the group "executes", neither we
254 * need a timer to replenish it.
256 * With respect to SMP, the bandwidth is given on a per-CPU basis,
258 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
259 * - dl_total_bw array contains, in the i-eth element, the currently
260 * allocated bandwidth on the i-eth CPU.
261 * Moreover, groups consume bandwidth on each CPU, while tasks only
262 * consume bandwidth on the CPU they're running on.
263 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
264 * that will be shown the next time the proc or cgroup controls will
265 * be red. It on its turn can be changed by writing on its own
268 struct dl_bandwidth {
269 raw_spinlock_t dl_runtime_lock;
274 static inline int dl_bandwidth_enabled(void)
276 return sysctl_sched_rt_runtime >= 0;
285 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
288 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
290 dl_b->total_bw -= tsk_bw;
291 __dl_update(dl_b, (s32)tsk_bw / cpus);
295 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
297 dl_b->total_bw += tsk_bw;
298 __dl_update(dl_b, -((s32)tsk_bw / cpus));
302 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
304 return dl_b->bw != -1 &&
305 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
308 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
309 extern void init_dl_bw(struct dl_bw *dl_b);
310 extern int sched_dl_global_validate(void);
311 extern void sched_dl_do_global(void);
312 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
313 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
314 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
315 extern bool __checkparam_dl(const struct sched_attr *attr);
316 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
317 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
318 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
319 extern bool dl_cpu_busy(unsigned int cpu);
321 #ifdef CONFIG_CGROUP_SCHED
323 #include <linux/cgroup.h>
324 #include <linux/psi.h>
329 extern struct list_head task_groups;
331 struct cfs_bandwidth {
332 #ifdef CONFIG_CFS_BANDWIDTH
337 s64 hierarchical_quota;
341 u8 distribute_running;
343 struct hrtimer period_timer;
344 struct hrtimer slack_timer;
345 struct list_head throttled_cfs_rq;
354 /* Task group related information */
356 struct cgroup_subsys_state css;
358 #ifdef CONFIG_FAIR_GROUP_SCHED
359 /* schedulable entities of this group on each CPU */
360 struct sched_entity **se;
361 /* runqueue "owned" by this group on each CPU */
362 struct cfs_rq **cfs_rq;
363 unsigned long shares;
367 * load_avg can be heavily contended at clock tick time, so put
368 * it in its own cacheline separated from the fields above which
369 * will also be accessed at each tick.
371 atomic_long_t load_avg ____cacheline_aligned;
375 #ifdef CONFIG_RT_GROUP_SCHED
376 struct sched_rt_entity **rt_se;
377 struct rt_rq **rt_rq;
379 struct rt_bandwidth rt_bandwidth;
383 struct list_head list;
385 struct task_group *parent;
386 struct list_head siblings;
387 struct list_head children;
389 #ifdef CONFIG_SCHED_AUTOGROUP
390 struct autogroup *autogroup;
393 struct cfs_bandwidth cfs_bandwidth;
395 #ifdef CONFIG_UCLAMP_TASK_GROUP
396 /* The two decimal precision [%] value requested from user-space */
397 unsigned int uclamp_pct[UCLAMP_CNT];
398 /* Clamp values requested for a task group */
399 struct uclamp_se uclamp_req[UCLAMP_CNT];
400 /* Effective clamp values used for a task group */
401 struct uclamp_se uclamp[UCLAMP_CNT];
406 #ifdef CONFIG_FAIR_GROUP_SCHED
407 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
410 * A weight of 0 or 1 can cause arithmetics problems.
411 * A weight of a cfs_rq is the sum of weights of which entities
412 * are queued on this cfs_rq, so a weight of a entity should not be
413 * too large, so as the shares value of a task group.
414 * (The default weight is 1024 - so there's no practical
415 * limitation from this.)
417 #define MIN_SHARES (1UL << 1)
418 #define MAX_SHARES (1UL << 18)
421 typedef int (*tg_visitor)(struct task_group *, void *);
423 extern int walk_tg_tree_from(struct task_group *from,
424 tg_visitor down, tg_visitor up, void *data);
427 * Iterate the full tree, calling @down when first entering a node and @up when
428 * leaving it for the final time.
430 * Caller must hold rcu_lock or sufficient equivalent.
432 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
434 return walk_tg_tree_from(&root_task_group, down, up, data);
437 extern int tg_nop(struct task_group *tg, void *data);
439 extern void free_fair_sched_group(struct task_group *tg);
440 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
441 extern void online_fair_sched_group(struct task_group *tg);
442 extern void unregister_fair_sched_group(struct task_group *tg);
443 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
444 struct sched_entity *se, int cpu,
445 struct sched_entity *parent);
446 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
448 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
449 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
450 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
452 extern void free_rt_sched_group(struct task_group *tg);
453 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
454 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
455 struct sched_rt_entity *rt_se, int cpu,
456 struct sched_rt_entity *parent);
457 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
458 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
459 extern long sched_group_rt_runtime(struct task_group *tg);
460 extern long sched_group_rt_period(struct task_group *tg);
461 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
463 extern struct task_group *sched_create_group(struct task_group *parent);
464 extern void sched_online_group(struct task_group *tg,
465 struct task_group *parent);
466 extern void sched_destroy_group(struct task_group *tg);
467 extern void sched_offline_group(struct task_group *tg);
469 extern void sched_move_task(struct task_struct *tsk);
471 #ifdef CONFIG_FAIR_GROUP_SCHED
472 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
475 extern void set_task_rq_fair(struct sched_entity *se,
476 struct cfs_rq *prev, struct cfs_rq *next);
477 #else /* !CONFIG_SMP */
478 static inline void set_task_rq_fair(struct sched_entity *se,
479 struct cfs_rq *prev, struct cfs_rq *next) { }
480 #endif /* CONFIG_SMP */
481 #endif /* CONFIG_FAIR_GROUP_SCHED */
483 #else /* CONFIG_CGROUP_SCHED */
485 struct cfs_bandwidth { };
487 #endif /* CONFIG_CGROUP_SCHED */
489 /* CFS-related fields in a runqueue */
491 struct load_weight load;
492 unsigned long runnable_weight;
493 unsigned int nr_running;
494 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
495 unsigned int idle_h_nr_running; /* SCHED_IDLE */
500 u64 min_vruntime_copy;
503 struct rb_root_cached tasks_timeline;
506 * 'curr' points to currently running entity on this cfs_rq.
507 * It is set to NULL otherwise (i.e when none are currently running).
509 struct sched_entity *curr;
510 struct sched_entity *next;
511 struct sched_entity *last;
512 struct sched_entity *skip;
514 #ifdef CONFIG_SCHED_DEBUG
515 unsigned int nr_spread_over;
522 struct sched_avg avg;
524 u64 load_last_update_time_copy;
527 raw_spinlock_t lock ____cacheline_aligned;
529 unsigned long load_avg;
530 unsigned long util_avg;
531 unsigned long runnable_sum;
534 #ifdef CONFIG_FAIR_GROUP_SCHED
535 unsigned long tg_load_avg_contrib;
537 long prop_runnable_sum;
540 * h_load = weight * f(tg)
542 * Where f(tg) is the recursive weight fraction assigned to
545 unsigned long h_load;
546 u64 last_h_load_update;
547 struct sched_entity *h_load_next;
548 #endif /* CONFIG_FAIR_GROUP_SCHED */
549 #endif /* CONFIG_SMP */
551 #ifdef CONFIG_FAIR_GROUP_SCHED
552 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
555 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
556 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
557 * (like users, containers etc.)
559 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
560 * This list is used during load balance.
563 struct list_head leaf_cfs_rq_list;
564 struct task_group *tg; /* group that "owns" this runqueue */
566 #ifdef CONFIG_CFS_BANDWIDTH
568 s64 runtime_remaining;
571 u64 throttled_clock_task;
572 u64 throttled_clock_task_time;
575 struct list_head throttled_list;
576 #endif /* CONFIG_CFS_BANDWIDTH */
577 #endif /* CONFIG_FAIR_GROUP_SCHED */
580 static inline int rt_bandwidth_enabled(void)
582 return sysctl_sched_rt_runtime >= 0;
585 /* RT IPI pull logic requires IRQ_WORK */
586 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
587 # define HAVE_RT_PUSH_IPI
590 /* Real-Time classes' related field in a runqueue: */
592 struct rt_prio_array active;
593 unsigned int rt_nr_running;
594 unsigned int rr_nr_running;
595 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
597 int curr; /* highest queued rt task prio */
599 int next; /* next highest */
604 unsigned long rt_nr_migratory;
605 unsigned long rt_nr_total;
607 struct plist_head pushable_tasks;
609 #endif /* CONFIG_SMP */
615 /* Nests inside the rq lock: */
616 raw_spinlock_t rt_runtime_lock;
618 #ifdef CONFIG_RT_GROUP_SCHED
619 unsigned long rt_nr_boosted;
622 struct task_group *tg;
626 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
628 return rt_rq->rt_queued && rt_rq->rt_nr_running;
631 /* Deadline class' related fields in a runqueue */
633 /* runqueue is an rbtree, ordered by deadline */
634 struct rb_root_cached root;
636 unsigned long dl_nr_running;
640 * Deadline values of the currently executing and the
641 * earliest ready task on this rq. Caching these facilitates
642 * the decision whether or not a ready but not running task
643 * should migrate somewhere else.
650 unsigned long dl_nr_migratory;
654 * Tasks on this rq that can be pushed away. They are kept in
655 * an rb-tree, ordered by tasks' deadlines, with caching
656 * of the leftmost (earliest deadline) element.
658 struct rb_root_cached pushable_dl_tasks_root;
663 * "Active utilization" for this runqueue: increased when a
664 * task wakes up (becomes TASK_RUNNING) and decreased when a
670 * Utilization of the tasks "assigned" to this runqueue (including
671 * the tasks that are in runqueue and the tasks that executed on this
672 * CPU and blocked). Increased when a task moves to this runqueue, and
673 * decreased when the task moves away (migrates, changes scheduling
674 * policy, or terminates).
675 * This is needed to compute the "inactive utilization" for the
676 * runqueue (inactive utilization = this_bw - running_bw).
682 * Inverse of the fraction of CPU utilization that can be reclaimed
683 * by the GRUB algorithm.
688 #ifdef CONFIG_FAIR_GROUP_SCHED
689 /* An entity is a task if it doesn't "own" a runqueue */
690 #define entity_is_task(se) (!se->my_q)
692 #define entity_is_task(se) 1
697 * XXX we want to get rid of these helpers and use the full load resolution.
699 static inline long se_weight(struct sched_entity *se)
701 return scale_load_down(se->load.weight);
704 static inline long se_runnable(struct sched_entity *se)
706 return scale_load_down(se->runnable_weight);
709 static inline bool sched_asym_prefer(int a, int b)
711 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
715 struct em_perf_domain *em_pd;
716 struct perf_domain *next;
720 /* Scheduling group status flags */
721 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
722 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
725 * We add the notion of a root-domain which will be used to define per-domain
726 * variables. Each exclusive cpuset essentially defines an island domain by
727 * fully partitioning the member CPUs from any other cpuset. Whenever a new
728 * exclusive cpuset is created, we also create and attach a new root-domain
737 cpumask_var_t online;
740 * Indicate pullable load on at least one CPU, e.g:
741 * - More than one runnable task
742 * - Running task is misfit
746 /* Indicate one or more cpus over-utilized (tipping point) */
750 * The bit corresponding to a CPU gets set here if such CPU has more
751 * than one runnable -deadline task (as it is below for RT tasks).
753 cpumask_var_t dlo_mask;
758 #ifdef HAVE_RT_PUSH_IPI
760 * For IPI pull requests, loop across the rto_mask.
762 struct irq_work rto_push_work;
763 raw_spinlock_t rto_lock;
764 /* These are only updated and read within rto_lock */
767 /* These atomics are updated outside of a lock */
768 atomic_t rto_loop_next;
769 atomic_t rto_loop_start;
772 * The "RT overload" flag: it gets set if a CPU has more than
773 * one runnable RT task.
775 cpumask_var_t rto_mask;
776 struct cpupri cpupri;
778 unsigned long max_cpu_capacity;
781 * NULL-terminated list of performance domains intersecting with the
782 * CPUs of the rd. Protected by RCU.
784 struct perf_domain __rcu *pd;
787 extern void init_defrootdomain(void);
788 extern int sched_init_domains(const struct cpumask *cpu_map);
789 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
790 extern void sched_get_rd(struct root_domain *rd);
791 extern void sched_put_rd(struct root_domain *rd);
793 #ifdef HAVE_RT_PUSH_IPI
794 extern void rto_push_irq_work_func(struct irq_work *work);
796 #endif /* CONFIG_SMP */
798 #ifdef CONFIG_UCLAMP_TASK
800 * struct uclamp_bucket - Utilization clamp bucket
801 * @value: utilization clamp value for tasks on this clamp bucket
802 * @tasks: number of RUNNABLE tasks on this clamp bucket
804 * Keep track of how many tasks are RUNNABLE for a given utilization
807 struct uclamp_bucket {
808 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
809 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
813 * struct uclamp_rq - rq's utilization clamp
814 * @value: currently active clamp values for a rq
815 * @bucket: utilization clamp buckets affecting a rq
817 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
818 * A clamp value is affecting a rq when there is at least one task RUNNABLE
819 * (or actually running) with that value.
821 * There are up to UCLAMP_CNT possible different clamp values, currently there
822 * are only two: minimum utilization and maximum utilization.
824 * All utilization clamping values are MAX aggregated, since:
825 * - for util_min: we want to run the CPU at least at the max of the minimum
826 * utilization required by its currently RUNNABLE tasks.
827 * - for util_max: we want to allow the CPU to run up to the max of the
828 * maximum utilization allowed by its currently RUNNABLE tasks.
830 * Since on each system we expect only a limited number of different
831 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
832 * the metrics required to compute all the per-rq utilization clamp values.
836 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
838 #endif /* CONFIG_UCLAMP_TASK */
841 * This is the main, per-CPU runqueue data structure.
843 * Locking rule: those places that want to lock multiple runqueues
844 * (such as the load balancing or the thread migration code), lock
845 * acquire operations must be ordered by ascending &runqueue.
852 * nr_running and cpu_load should be in the same cacheline because
853 * remote CPUs use both these fields when doing load calculation.
855 unsigned int nr_running;
856 #ifdef CONFIG_NUMA_BALANCING
857 unsigned int nr_numa_running;
858 unsigned int nr_preferred_running;
859 unsigned int numa_migrate_on;
861 #ifdef CONFIG_NO_HZ_COMMON
863 unsigned long last_load_update_tick;
864 unsigned long last_blocked_load_update_tick;
865 unsigned int has_blocked_load;
866 #endif /* CONFIG_SMP */
867 unsigned int nohz_tick_stopped;
869 #endif /* CONFIG_NO_HZ_COMMON */
871 unsigned long nr_load_updates;
874 #ifdef CONFIG_UCLAMP_TASK
875 /* Utilization clamp values based on CPU's RUNNABLE tasks */
876 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
877 unsigned int uclamp_flags;
878 #define UCLAMP_FLAG_IDLE 0x01
885 #ifdef CONFIG_FAIR_GROUP_SCHED
886 /* list of leaf cfs_rq on this CPU: */
887 struct list_head leaf_cfs_rq_list;
888 struct list_head *tmp_alone_branch;
889 #endif /* CONFIG_FAIR_GROUP_SCHED */
892 * This is part of a global counter where only the total sum
893 * over all CPUs matters. A task can increase this counter on
894 * one CPU and if it got migrated afterwards it may decrease
895 * it on another CPU. Always updated under the runqueue lock:
897 unsigned long nr_uninterruptible;
899 struct task_struct *curr;
900 struct task_struct *idle;
901 struct task_struct *stop;
902 unsigned long next_balance;
903 struct mm_struct *prev_mm;
905 unsigned int clock_update_flags;
907 /* Ensure that all clocks are in the same cache line */
908 u64 clock_task ____cacheline_aligned;
910 unsigned long lost_idle_time;
915 struct root_domain *rd;
916 struct sched_domain __rcu *sd;
918 unsigned long cpu_capacity;
919 unsigned long cpu_capacity_orig;
921 struct callback_head *balance_callback;
923 unsigned char idle_balance;
925 unsigned long misfit_task_load;
927 /* For active balancing */
930 struct cpu_stop_work active_balance_work;
932 /* CPU of this runqueue: */
936 struct list_head cfs_tasks;
938 struct sched_avg avg_rt;
939 struct sched_avg avg_dl;
940 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
941 struct sched_avg avg_irq;
946 /* This is used to determine avg_idle's max value */
947 u64 max_idle_balance_cost;
950 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
953 #ifdef CONFIG_PARAVIRT
956 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
957 u64 prev_steal_time_rq;
960 /* calc_load related fields */
961 unsigned long calc_load_update;
962 long calc_load_active;
964 #ifdef CONFIG_SCHED_HRTICK
966 int hrtick_csd_pending;
967 call_single_data_t hrtick_csd;
969 struct hrtimer hrtick_timer;
972 #ifdef CONFIG_SCHEDSTATS
974 struct sched_info rq_sched_info;
975 unsigned long long rq_cpu_time;
976 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
978 /* sys_sched_yield() stats */
979 unsigned int yld_count;
981 /* schedule() stats */
982 unsigned int sched_count;
983 unsigned int sched_goidle;
985 /* try_to_wake_up() stats */
986 unsigned int ttwu_count;
987 unsigned int ttwu_local;
991 struct llist_head wake_list;
994 #ifdef CONFIG_CPU_IDLE
995 /* Must be inspected within a rcu lock section */
996 struct cpuidle_state *idle_state;
1000 #ifdef CONFIG_FAIR_GROUP_SCHED
1002 /* CPU runqueue to which this cfs_rq is attached */
1003 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1010 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1012 return container_of(cfs_rq, struct rq, cfs);
1016 static inline int cpu_of(struct rq *rq)
1026 #ifdef CONFIG_SCHED_SMT
1027 extern void __update_idle_core(struct rq *rq);
1029 static inline void update_idle_core(struct rq *rq)
1031 if (static_branch_unlikely(&sched_smt_present))
1032 __update_idle_core(rq);
1036 static inline void update_idle_core(struct rq *rq) { }
1039 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1041 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1042 #define this_rq() this_cpu_ptr(&runqueues)
1043 #define task_rq(p) cpu_rq(task_cpu(p))
1044 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1045 #define raw_rq() raw_cpu_ptr(&runqueues)
1047 extern void update_rq_clock(struct rq *rq);
1049 static inline u64 __rq_clock_broken(struct rq *rq)
1051 return READ_ONCE(rq->clock);
1055 * rq::clock_update_flags bits
1057 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1058 * call to __schedule(). This is an optimisation to avoid
1059 * neighbouring rq clock updates.
1061 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1062 * in effect and calls to update_rq_clock() are being ignored.
1064 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1065 * made to update_rq_clock() since the last time rq::lock was pinned.
1067 * If inside of __schedule(), clock_update_flags will have been
1068 * shifted left (a left shift is a cheap operation for the fast path
1069 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1071 * if (rq-clock_update_flags >= RQCF_UPDATED)
1073 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1074 * one position though, because the next rq_unpin_lock() will shift it
1077 #define RQCF_REQ_SKIP 0x01
1078 #define RQCF_ACT_SKIP 0x02
1079 #define RQCF_UPDATED 0x04
1081 static inline void assert_clock_updated(struct rq *rq)
1084 * The only reason for not seeing a clock update since the
1085 * last rq_pin_lock() is if we're currently skipping updates.
1087 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1090 static inline u64 rq_clock(struct rq *rq)
1092 lockdep_assert_held(&rq->lock);
1093 assert_clock_updated(rq);
1098 static inline u64 rq_clock_task(struct rq *rq)
1100 lockdep_assert_held(&rq->lock);
1101 assert_clock_updated(rq);
1103 return rq->clock_task;
1106 static inline void rq_clock_skip_update(struct rq *rq)
1108 lockdep_assert_held(&rq->lock);
1109 rq->clock_update_flags |= RQCF_REQ_SKIP;
1113 * See rt task throttling, which is the only time a skip
1114 * request is cancelled.
1116 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1118 lockdep_assert_held(&rq->lock);
1119 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1123 unsigned long flags;
1124 struct pin_cookie cookie;
1125 #ifdef CONFIG_SCHED_DEBUG
1127 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1128 * current pin context is stashed here in case it needs to be
1129 * restored in rq_repin_lock().
1131 unsigned int clock_update_flags;
1135 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1137 rf->cookie = lockdep_pin_lock(&rq->lock);
1139 #ifdef CONFIG_SCHED_DEBUG
1140 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1141 rf->clock_update_flags = 0;
1145 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1147 #ifdef CONFIG_SCHED_DEBUG
1148 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1149 rf->clock_update_flags = RQCF_UPDATED;
1152 lockdep_unpin_lock(&rq->lock, rf->cookie);
1155 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1157 lockdep_repin_lock(&rq->lock, rf->cookie);
1159 #ifdef CONFIG_SCHED_DEBUG
1161 * Restore the value we stashed in @rf for this pin context.
1163 rq->clock_update_flags |= rf->clock_update_flags;
1167 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1168 __acquires(rq->lock);
1170 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1171 __acquires(p->pi_lock)
1172 __acquires(rq->lock);
1174 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1175 __releases(rq->lock)
1177 rq_unpin_lock(rq, rf);
1178 raw_spin_unlock(&rq->lock);
1182 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1183 __releases(rq->lock)
1184 __releases(p->pi_lock)
1186 rq_unpin_lock(rq, rf);
1187 raw_spin_unlock(&rq->lock);
1188 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1192 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1193 __acquires(rq->lock)
1195 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1196 rq_pin_lock(rq, rf);
1200 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1201 __acquires(rq->lock)
1203 raw_spin_lock_irq(&rq->lock);
1204 rq_pin_lock(rq, rf);
1208 rq_lock(struct rq *rq, struct rq_flags *rf)
1209 __acquires(rq->lock)
1211 raw_spin_lock(&rq->lock);
1212 rq_pin_lock(rq, rf);
1216 rq_relock(struct rq *rq, struct rq_flags *rf)
1217 __acquires(rq->lock)
1219 raw_spin_lock(&rq->lock);
1220 rq_repin_lock(rq, rf);
1224 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1225 __releases(rq->lock)
1227 rq_unpin_lock(rq, rf);
1228 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1232 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1233 __releases(rq->lock)
1235 rq_unpin_lock(rq, rf);
1236 raw_spin_unlock_irq(&rq->lock);
1240 rq_unlock(struct rq *rq, struct rq_flags *rf)
1241 __releases(rq->lock)
1243 rq_unpin_lock(rq, rf);
1244 raw_spin_unlock(&rq->lock);
1247 static inline struct rq *
1248 this_rq_lock_irq(struct rq_flags *rf)
1249 __acquires(rq->lock)
1253 local_irq_disable();
1260 enum numa_topology_type {
1265 extern enum numa_topology_type sched_numa_topology_type;
1266 extern int sched_max_numa_distance;
1267 extern bool find_numa_distance(int distance);
1268 extern void sched_init_numa(void);
1269 extern void sched_domains_numa_masks_set(unsigned int cpu);
1270 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1271 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1273 static inline void sched_init_numa(void) { }
1274 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1275 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1276 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1282 #ifdef CONFIG_NUMA_BALANCING
1283 /* The regions in numa_faults array from task_struct */
1284 enum numa_faults_stats {
1290 extern void sched_setnuma(struct task_struct *p, int node);
1291 extern int migrate_task_to(struct task_struct *p, int cpu);
1292 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1294 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1297 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1300 #endif /* CONFIG_NUMA_BALANCING */
1305 queue_balance_callback(struct rq *rq,
1306 struct callback_head *head,
1307 void (*func)(struct rq *rq))
1309 lockdep_assert_held(&rq->lock);
1311 if (unlikely(head->next))
1314 head->func = (void (*)(struct callback_head *))func;
1315 head->next = rq->balance_callback;
1316 rq->balance_callback = head;
1319 extern void sched_ttwu_pending(void);
1321 #define rcu_dereference_check_sched_domain(p) \
1322 rcu_dereference_check((p), \
1323 lockdep_is_held(&sched_domains_mutex))
1326 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1327 * See destroy_sched_domains: call_rcu for details.
1329 * The domain tree of any CPU may only be accessed from within
1330 * preempt-disabled sections.
1332 #define for_each_domain(cpu, __sd) \
1333 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1334 __sd; __sd = __sd->parent)
1336 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1339 * highest_flag_domain - Return highest sched_domain containing flag.
1340 * @cpu: The CPU whose highest level of sched domain is to
1342 * @flag: The flag to check for the highest sched_domain
1343 * for the given CPU.
1345 * Returns the highest sched_domain of a CPU which contains the given flag.
1347 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1349 struct sched_domain *sd, *hsd = NULL;
1351 for_each_domain(cpu, sd) {
1352 if (!(sd->flags & flag))
1360 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1362 struct sched_domain *sd;
1364 for_each_domain(cpu, sd) {
1365 if (sd->flags & flag)
1372 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1373 DECLARE_PER_CPU(int, sd_llc_size);
1374 DECLARE_PER_CPU(int, sd_llc_id);
1375 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1376 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1377 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1378 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1379 extern struct static_key_false sched_asym_cpucapacity;
1381 struct sched_group_capacity {
1384 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1387 unsigned long capacity;
1388 unsigned long min_capacity; /* Min per-CPU capacity in group */
1389 unsigned long max_capacity; /* Max per-CPU capacity in group */
1390 unsigned long next_update;
1391 int imbalance; /* XXX unrelated to capacity but shared group state */
1393 #ifdef CONFIG_SCHED_DEBUG
1397 unsigned long cpumask[0]; /* Balance mask */
1400 struct sched_group {
1401 struct sched_group *next; /* Must be a circular list */
1404 unsigned int group_weight;
1405 struct sched_group_capacity *sgc;
1406 int asym_prefer_cpu; /* CPU of highest priority in group */
1409 * The CPUs this group covers.
1411 * NOTE: this field is variable length. (Allocated dynamically
1412 * by attaching extra space to the end of the structure,
1413 * depending on how many CPUs the kernel has booted up with)
1415 unsigned long cpumask[0];
1418 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1420 return to_cpumask(sg->cpumask);
1424 * See build_balance_mask().
1426 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1428 return to_cpumask(sg->sgc->cpumask);
1432 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1433 * @group: The group whose first CPU is to be returned.
1435 static inline unsigned int group_first_cpu(struct sched_group *group)
1437 return cpumask_first(sched_group_span(group));
1440 extern int group_balance_cpu(struct sched_group *sg);
1442 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1443 void register_sched_domain_sysctl(void);
1444 void dirty_sched_domain_sysctl(int cpu);
1445 void unregister_sched_domain_sysctl(void);
1447 static inline void register_sched_domain_sysctl(void)
1450 static inline void dirty_sched_domain_sysctl(int cpu)
1453 static inline void unregister_sched_domain_sysctl(void)
1458 extern int newidle_balance(struct rq *this_rq, struct rq_flags *rf);
1462 static inline void sched_ttwu_pending(void) { }
1464 static inline int newidle_balance(struct rq *this_rq, struct rq_flags *rf) { return 0; }
1466 #endif /* CONFIG_SMP */
1469 #include "autogroup.h"
1471 #ifdef CONFIG_CGROUP_SCHED
1474 * Return the group to which this tasks belongs.
1476 * We cannot use task_css() and friends because the cgroup subsystem
1477 * changes that value before the cgroup_subsys::attach() method is called,
1478 * therefore we cannot pin it and might observe the wrong value.
1480 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1481 * core changes this before calling sched_move_task().
1483 * Instead we use a 'copy' which is updated from sched_move_task() while
1484 * holding both task_struct::pi_lock and rq::lock.
1486 static inline struct task_group *task_group(struct task_struct *p)
1488 return p->sched_task_group;
1491 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1492 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1494 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1495 struct task_group *tg = task_group(p);
1498 #ifdef CONFIG_FAIR_GROUP_SCHED
1499 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1500 p->se.cfs_rq = tg->cfs_rq[cpu];
1501 p->se.parent = tg->se[cpu];
1504 #ifdef CONFIG_RT_GROUP_SCHED
1505 p->rt.rt_rq = tg->rt_rq[cpu];
1506 p->rt.parent = tg->rt_se[cpu];
1510 #else /* CONFIG_CGROUP_SCHED */
1512 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1513 static inline struct task_group *task_group(struct task_struct *p)
1518 #endif /* CONFIG_CGROUP_SCHED */
1520 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1522 set_task_rq(p, cpu);
1525 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1526 * successfully executed on another CPU. We must ensure that updates of
1527 * per-task data have been completed by this moment.
1530 #ifdef CONFIG_THREAD_INFO_IN_TASK
1531 WRITE_ONCE(p->cpu, cpu);
1533 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1540 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1542 #ifdef CONFIG_SCHED_DEBUG
1543 # include <linux/static_key.h>
1544 # define const_debug __read_mostly
1546 # define const_debug const
1549 #define SCHED_FEAT(name, enabled) \
1550 __SCHED_FEAT_##name ,
1553 #include "features.h"
1559 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1562 * To support run-time toggling of sched features, all the translation units
1563 * (but core.c) reference the sysctl_sched_features defined in core.c.
1565 extern const_debug unsigned int sysctl_sched_features;
1567 #define SCHED_FEAT(name, enabled) \
1568 static __always_inline bool static_branch_##name(struct static_key *key) \
1570 return static_key_##enabled(key); \
1573 #include "features.h"
1576 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1577 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1579 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1582 * Each translation unit has its own copy of sysctl_sched_features to allow
1583 * constants propagation at compile time and compiler optimization based on
1586 #define SCHED_FEAT(name, enabled) \
1587 (1UL << __SCHED_FEAT_##name) * enabled |
1588 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1589 #include "features.h"
1593 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1595 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1597 extern struct static_key_false sched_numa_balancing;
1598 extern struct static_key_false sched_schedstats;
1600 static inline u64 global_rt_period(void)
1602 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1605 static inline u64 global_rt_runtime(void)
1607 if (sysctl_sched_rt_runtime < 0)
1610 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1613 static inline int task_current(struct rq *rq, struct task_struct *p)
1615 return rq->curr == p;
1618 static inline int task_running(struct rq *rq, struct task_struct *p)
1623 return task_current(rq, p);
1627 static inline int task_on_rq_queued(struct task_struct *p)
1629 return p->on_rq == TASK_ON_RQ_QUEUED;
1632 static inline int task_on_rq_migrating(struct task_struct *p)
1634 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1640 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1641 #define WF_FORK 0x02 /* Child wakeup after fork */
1642 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1645 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1646 * of tasks with abnormal "nice" values across CPUs the contribution that
1647 * each task makes to its run queue's load is weighted according to its
1648 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1649 * scaled version of the new time slice allocation that they receive on time
1653 #define WEIGHT_IDLEPRIO 3
1654 #define WMULT_IDLEPRIO 1431655765
1656 extern const int sched_prio_to_weight[40];
1657 extern const u32 sched_prio_to_wmult[40];
1660 * {de,en}queue flags:
1662 * DEQUEUE_SLEEP - task is no longer runnable
1663 * ENQUEUE_WAKEUP - task just became runnable
1665 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1666 * are in a known state which allows modification. Such pairs
1667 * should preserve as much state as possible.
1669 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1672 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1673 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1674 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1678 #define DEQUEUE_SLEEP 0x01
1679 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1680 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1681 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1683 #define ENQUEUE_WAKEUP 0x01
1684 #define ENQUEUE_RESTORE 0x02
1685 #define ENQUEUE_MOVE 0x04
1686 #define ENQUEUE_NOCLOCK 0x08
1688 #define ENQUEUE_HEAD 0x10
1689 #define ENQUEUE_REPLENISH 0x20
1691 #define ENQUEUE_MIGRATED 0x40
1693 #define ENQUEUE_MIGRATED 0x00
1696 #define RETRY_TASK ((void *)-1UL)
1698 struct sched_class {
1699 const struct sched_class *next;
1701 #ifdef CONFIG_UCLAMP_TASK
1705 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1706 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1707 void (*yield_task) (struct rq *rq);
1708 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1710 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1713 * Both @prev and @rf are optional and may be NULL, in which case the
1714 * caller must already have invoked put_prev_task(rq, prev, rf).
1716 * Otherwise it is the responsibility of the pick_next_task() to call
1717 * put_prev_task() on the @prev task or something equivalent, IFF it
1718 * returns a next task.
1720 * In that case (@rf != NULL) it may return RETRY_TASK when it finds a
1721 * higher prio class has runnable tasks.
1723 struct task_struct * (*pick_next_task)(struct rq *rq,
1724 struct task_struct *prev,
1725 struct rq_flags *rf);
1726 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct rq_flags *rf);
1727 void (*set_next_task)(struct rq *rq, struct task_struct *p);
1730 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1731 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1733 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1735 void (*set_cpus_allowed)(struct task_struct *p,
1736 const struct cpumask *newmask);
1738 void (*rq_online)(struct rq *rq);
1739 void (*rq_offline)(struct rq *rq);
1742 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1743 void (*task_fork)(struct task_struct *p);
1744 void (*task_dead)(struct task_struct *p);
1747 * The switched_from() call is allowed to drop rq->lock, therefore we
1748 * cannot assume the switched_from/switched_to pair is serliazed by
1749 * rq->lock. They are however serialized by p->pi_lock.
1751 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1752 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1753 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1756 unsigned int (*get_rr_interval)(struct rq *rq,
1757 struct task_struct *task);
1759 void (*update_curr)(struct rq *rq);
1761 #define TASK_SET_GROUP 0
1762 #define TASK_MOVE_GROUP 1
1764 #ifdef CONFIG_FAIR_GROUP_SCHED
1765 void (*task_change_group)(struct task_struct *p, int type);
1769 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1771 WARN_ON_ONCE(rq->curr != prev);
1772 prev->sched_class->put_prev_task(rq, prev, NULL);
1775 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1777 WARN_ON_ONCE(rq->curr != next);
1778 next->sched_class->set_next_task(rq, next);
1782 #define sched_class_highest (&stop_sched_class)
1784 #define sched_class_highest (&dl_sched_class)
1786 #define for_each_class(class) \
1787 for (class = sched_class_highest; class; class = class->next)
1789 extern const struct sched_class stop_sched_class;
1790 extern const struct sched_class dl_sched_class;
1791 extern const struct sched_class rt_sched_class;
1792 extern const struct sched_class fair_sched_class;
1793 extern const struct sched_class idle_sched_class;
1798 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1800 extern void trigger_load_balance(struct rq *rq);
1802 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1806 #ifdef CONFIG_CPU_IDLE
1807 static inline void idle_set_state(struct rq *rq,
1808 struct cpuidle_state *idle_state)
1810 rq->idle_state = idle_state;
1813 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1815 SCHED_WARN_ON(!rcu_read_lock_held());
1817 return rq->idle_state;
1820 static inline void idle_set_state(struct rq *rq,
1821 struct cpuidle_state *idle_state)
1825 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1831 extern void schedule_idle(void);
1833 extern void sysrq_sched_debug_show(void);
1834 extern void sched_init_granularity(void);
1835 extern void update_max_interval(void);
1837 extern void init_sched_dl_class(void);
1838 extern void init_sched_rt_class(void);
1839 extern void init_sched_fair_class(void);
1841 extern void reweight_task(struct task_struct *p, int prio);
1843 extern void resched_curr(struct rq *rq);
1844 extern void resched_cpu(int cpu);
1846 extern struct rt_bandwidth def_rt_bandwidth;
1847 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1849 extern struct dl_bandwidth def_dl_bandwidth;
1850 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1851 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1852 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1853 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1856 #define BW_UNIT (1 << BW_SHIFT)
1857 #define RATIO_SHIFT 8
1858 unsigned long to_ratio(u64 period, u64 runtime);
1860 extern void init_entity_runnable_average(struct sched_entity *se);
1861 extern void post_init_entity_util_avg(struct task_struct *p);
1863 #ifdef CONFIG_NO_HZ_FULL
1864 extern bool sched_can_stop_tick(struct rq *rq);
1865 extern int __init sched_tick_offload_init(void);
1868 * Tick may be needed by tasks in the runqueue depending on their policy and
1869 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1870 * nohz mode if necessary.
1872 static inline void sched_update_tick_dependency(struct rq *rq)
1876 if (!tick_nohz_full_enabled())
1881 if (!tick_nohz_full_cpu(cpu))
1884 if (sched_can_stop_tick(rq))
1885 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1887 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1890 static inline int sched_tick_offload_init(void) { return 0; }
1891 static inline void sched_update_tick_dependency(struct rq *rq) { }
1894 static inline void add_nr_running(struct rq *rq, unsigned count)
1896 unsigned prev_nr = rq->nr_running;
1898 rq->nr_running = prev_nr + count;
1901 if (prev_nr < 2 && rq->nr_running >= 2) {
1902 if (!READ_ONCE(rq->rd->overload))
1903 WRITE_ONCE(rq->rd->overload, 1);
1907 sched_update_tick_dependency(rq);
1910 static inline void sub_nr_running(struct rq *rq, unsigned count)
1912 rq->nr_running -= count;
1913 /* Check if we still need preemption */
1914 sched_update_tick_dependency(rq);
1917 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1918 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1920 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1922 extern const_debug unsigned int sysctl_sched_nr_migrate;
1923 extern const_debug unsigned int sysctl_sched_migration_cost;
1925 #ifdef CONFIG_SCHED_HRTICK
1929 * - enabled by features
1930 * - hrtimer is actually high res
1932 static inline int hrtick_enabled(struct rq *rq)
1934 if (!sched_feat(HRTICK))
1936 if (!cpu_active(cpu_of(rq)))
1938 return hrtimer_is_hres_active(&rq->hrtick_timer);
1941 void hrtick_start(struct rq *rq, u64 delay);
1945 static inline int hrtick_enabled(struct rq *rq)
1950 #endif /* CONFIG_SCHED_HRTICK */
1952 #ifndef arch_scale_freq_capacity
1953 static __always_inline
1954 unsigned long arch_scale_freq_capacity(int cpu)
1956 return SCHED_CAPACITY_SCALE;
1961 #ifdef CONFIG_PREEMPTION
1963 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1966 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1967 * way at the expense of forcing extra atomic operations in all
1968 * invocations. This assures that the double_lock is acquired using the
1969 * same underlying policy as the spinlock_t on this architecture, which
1970 * reduces latency compared to the unfair variant below. However, it
1971 * also adds more overhead and therefore may reduce throughput.
1973 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1974 __releases(this_rq->lock)
1975 __acquires(busiest->lock)
1976 __acquires(this_rq->lock)
1978 raw_spin_unlock(&this_rq->lock);
1979 double_rq_lock(this_rq, busiest);
1986 * Unfair double_lock_balance: Optimizes throughput at the expense of
1987 * latency by eliminating extra atomic operations when the locks are
1988 * already in proper order on entry. This favors lower CPU-ids and will
1989 * grant the double lock to lower CPUs over higher ids under contention,
1990 * regardless of entry order into the function.
1992 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1993 __releases(this_rq->lock)
1994 __acquires(busiest->lock)
1995 __acquires(this_rq->lock)
1999 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2000 if (busiest < this_rq) {
2001 raw_spin_unlock(&this_rq->lock);
2002 raw_spin_lock(&busiest->lock);
2003 raw_spin_lock_nested(&this_rq->lock,
2004 SINGLE_DEPTH_NESTING);
2007 raw_spin_lock_nested(&busiest->lock,
2008 SINGLE_DEPTH_NESTING);
2013 #endif /* CONFIG_PREEMPTION */
2016 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2018 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2020 if (unlikely(!irqs_disabled())) {
2021 /* printk() doesn't work well under rq->lock */
2022 raw_spin_unlock(&this_rq->lock);
2026 return _double_lock_balance(this_rq, busiest);
2029 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2030 __releases(busiest->lock)
2032 raw_spin_unlock(&busiest->lock);
2033 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2036 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2042 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2045 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2051 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2054 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2060 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2064 * double_rq_lock - safely lock two runqueues
2066 * Note this does not disable interrupts like task_rq_lock,
2067 * you need to do so manually before calling.
2069 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2070 __acquires(rq1->lock)
2071 __acquires(rq2->lock)
2073 BUG_ON(!irqs_disabled());
2075 raw_spin_lock(&rq1->lock);
2076 __acquire(rq2->lock); /* Fake it out ;) */
2079 raw_spin_lock(&rq1->lock);
2080 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2082 raw_spin_lock(&rq2->lock);
2083 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2089 * double_rq_unlock - safely unlock two runqueues
2091 * Note this does not restore interrupts like task_rq_unlock,
2092 * you need to do so manually after calling.
2094 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2095 __releases(rq1->lock)
2096 __releases(rq2->lock)
2098 raw_spin_unlock(&rq1->lock);
2100 raw_spin_unlock(&rq2->lock);
2102 __release(rq2->lock);
2105 extern void set_rq_online (struct rq *rq);
2106 extern void set_rq_offline(struct rq *rq);
2107 extern bool sched_smp_initialized;
2109 #else /* CONFIG_SMP */
2112 * double_rq_lock - safely lock two runqueues
2114 * Note this does not disable interrupts like task_rq_lock,
2115 * you need to do so manually before calling.
2117 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2118 __acquires(rq1->lock)
2119 __acquires(rq2->lock)
2121 BUG_ON(!irqs_disabled());
2123 raw_spin_lock(&rq1->lock);
2124 __acquire(rq2->lock); /* Fake it out ;) */
2128 * double_rq_unlock - safely unlock two runqueues
2130 * Note this does not restore interrupts like task_rq_unlock,
2131 * you need to do so manually after calling.
2133 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2134 __releases(rq1->lock)
2135 __releases(rq2->lock)
2138 raw_spin_unlock(&rq1->lock);
2139 __release(rq2->lock);
2144 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2145 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2147 #ifdef CONFIG_SCHED_DEBUG
2148 extern bool sched_debug_enabled;
2150 extern void print_cfs_stats(struct seq_file *m, int cpu);
2151 extern void print_rt_stats(struct seq_file *m, int cpu);
2152 extern void print_dl_stats(struct seq_file *m, int cpu);
2153 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2154 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2155 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2156 #ifdef CONFIG_NUMA_BALANCING
2158 show_numa_stats(struct task_struct *p, struct seq_file *m);
2160 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2161 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2162 #endif /* CONFIG_NUMA_BALANCING */
2163 #endif /* CONFIG_SCHED_DEBUG */
2165 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2166 extern void init_rt_rq(struct rt_rq *rt_rq);
2167 extern void init_dl_rq(struct dl_rq *dl_rq);
2169 extern void cfs_bandwidth_usage_inc(void);
2170 extern void cfs_bandwidth_usage_dec(void);
2172 #ifdef CONFIG_NO_HZ_COMMON
2173 #define NOHZ_BALANCE_KICK_BIT 0
2174 #define NOHZ_STATS_KICK_BIT 1
2176 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2177 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2179 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2181 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2183 extern void nohz_balance_exit_idle(struct rq *rq);
2185 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2191 void __dl_update(struct dl_bw *dl_b, s64 bw)
2193 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2196 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2197 "sched RCU must be held");
2198 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2199 struct rq *rq = cpu_rq(i);
2201 rq->dl.extra_bw += bw;
2206 void __dl_update(struct dl_bw *dl_b, s64 bw)
2208 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2215 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2220 struct u64_stats_sync sync;
2223 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2226 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2227 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2228 * and never move forward.
2230 static inline u64 irq_time_read(int cpu)
2232 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2237 seq = __u64_stats_fetch_begin(&irqtime->sync);
2238 total = irqtime->total;
2239 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2243 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2245 #ifdef CONFIG_CPU_FREQ
2246 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2249 * cpufreq_update_util - Take a note about CPU utilization changes.
2250 * @rq: Runqueue to carry out the update for.
2251 * @flags: Update reason flags.
2253 * This function is called by the scheduler on the CPU whose utilization is
2256 * It can only be called from RCU-sched read-side critical sections.
2258 * The way cpufreq is currently arranged requires it to evaluate the CPU
2259 * performance state (frequency/voltage) on a regular basis to prevent it from
2260 * being stuck in a completely inadequate performance level for too long.
2261 * That is not guaranteed to happen if the updates are only triggered from CFS
2262 * and DL, though, because they may not be coming in if only RT tasks are
2263 * active all the time (or there are RT tasks only).
2265 * As a workaround for that issue, this function is called periodically by the
2266 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2267 * but that really is a band-aid. Going forward it should be replaced with
2268 * solutions targeted more specifically at RT tasks.
2270 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2272 struct update_util_data *data;
2274 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2277 data->func(data, rq_clock(rq), flags);
2280 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2281 #endif /* CONFIG_CPU_FREQ */
2283 #ifdef CONFIG_UCLAMP_TASK
2284 enum uclamp_id uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2286 static __always_inline
2287 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2288 struct task_struct *p)
2290 unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2291 unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2294 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2295 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2299 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2300 * RUNNABLE tasks with _different_ clamps, we can end up with an
2301 * inversion. Fix it now when the clamps are applied.
2303 if (unlikely(min_util >= max_util))
2306 return clamp(util, min_util, max_util);
2309 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2311 return uclamp_util_with(rq, util, NULL);
2313 #else /* CONFIG_UCLAMP_TASK */
2314 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2315 struct task_struct *p)
2319 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2323 #endif /* CONFIG_UCLAMP_TASK */
2325 #ifdef arch_scale_freq_capacity
2326 # ifndef arch_scale_freq_invariant
2327 # define arch_scale_freq_invariant() true
2330 # define arch_scale_freq_invariant() false
2334 static inline unsigned long capacity_orig_of(int cpu)
2336 return cpu_rq(cpu)->cpu_capacity_orig;
2341 * enum schedutil_type - CPU utilization type
2342 * @FREQUENCY_UTIL: Utilization used to select frequency
2343 * @ENERGY_UTIL: Utilization used during energy calculation
2345 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2346 * need to be aggregated differently depending on the usage made of them. This
2347 * enum is used within schedutil_freq_util() to differentiate the types of
2348 * utilization expected by the callers, and adjust the aggregation accordingly.
2350 enum schedutil_type {
2355 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2357 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2358 unsigned long max, enum schedutil_type type,
2359 struct task_struct *p);
2361 static inline unsigned long cpu_bw_dl(struct rq *rq)
2363 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2366 static inline unsigned long cpu_util_dl(struct rq *rq)
2368 return READ_ONCE(rq->avg_dl.util_avg);
2371 static inline unsigned long cpu_util_cfs(struct rq *rq)
2373 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2375 if (sched_feat(UTIL_EST)) {
2376 util = max_t(unsigned long, util,
2377 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2383 static inline unsigned long cpu_util_rt(struct rq *rq)
2385 return READ_ONCE(rq->avg_rt.util_avg);
2387 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2388 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2389 unsigned long max, enum schedutil_type type,
2390 struct task_struct *p)
2394 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2396 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2397 static inline unsigned long cpu_util_irq(struct rq *rq)
2399 return rq->avg_irq.util_avg;
2403 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2405 util *= (max - irq);
2412 static inline unsigned long cpu_util_irq(struct rq *rq)
2418 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2424 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2426 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2428 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2430 static inline bool sched_energy_enabled(void)
2432 return static_branch_unlikely(&sched_energy_present);
2435 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2437 #define perf_domain_span(pd) NULL
2438 static inline bool sched_energy_enabled(void) { return false; }
2440 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */