1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
76 #include "cpudeadline.h"
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED 1
89 #define TASK_ON_RQ_MIGRATING 2
91 extern __read_mostly int scheduler_running;
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
100 * Helpers for converting nanosecond timing to jiffy resolution
102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
105 * Increase resolution of nice-level calculations for 64-bit architectures.
106 * The extra resolution improves shares distribution and load balancing of
107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108 * hierarchies, especially on larger systems. This is not a user-visible change
109 * and does not change the user-interface for setting shares/weights.
111 * We increase resolution only if we have enough bits to allow this increased
112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113 * are pretty high and the returns do not justify the increased costs.
115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116 * increase coverage and consistency always enable it on 64-bit platforms.
119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w) \
123 unsigned long __w = (w); \
125 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
130 # define scale_load(w) (w)
131 # define scale_load_down(w) (w)
135 * Task weight (visible to users) and its load (invisible to users) have
136 * independent resolution, but they should be well calibrated. We use
137 * scale_load() and scale_load_down(w) to convert between them. The
138 * following must be true:
140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
146 * Single value that decides SCHED_DEADLINE internal math precision.
147 * 10 -> just above 1us
148 * 9 -> just above 0.5us
153 * Single value that denotes runtime == period, ie unlimited time.
155 #define RUNTIME_INF ((u64)~0ULL)
157 static inline int idle_policy(int policy)
159 return policy == SCHED_IDLE;
161 static inline int fair_policy(int policy)
163 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
166 static inline int rt_policy(int policy)
168 return policy == SCHED_FIFO || policy == SCHED_RR;
171 static inline int dl_policy(int policy)
173 return policy == SCHED_DEADLINE;
175 static inline bool valid_policy(int policy)
177 return idle_policy(policy) || fair_policy(policy) ||
178 rt_policy(policy) || dl_policy(policy);
181 static inline int task_has_idle_policy(struct task_struct *p)
183 return idle_policy(p->policy);
186 static inline int task_has_rt_policy(struct task_struct *p)
188 return rt_policy(p->policy);
191 static inline int task_has_dl_policy(struct task_struct *p)
193 return dl_policy(p->policy);
196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
199 * !! For sched_setattr_nocheck() (kernel) only !!
201 * This is actually gross. :(
203 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
204 * tasks, but still be able to sleep. We need this on platforms that cannot
205 * atomically change clock frequency. Remove once fast switching will be
206 * available on such platforms.
208 * SUGOV stands for SchedUtil GOVernor.
210 #define SCHED_FLAG_SUGOV 0x10000000
212 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
214 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
216 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
217 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
224 * Tells if entity @a should preempt entity @b.
227 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
229 return dl_entity_is_special(a) ||
230 dl_time_before(a->deadline, b->deadline);
234 * This is the priority-queue data structure of the RT scheduling class:
236 struct rt_prio_array {
237 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
238 struct list_head queue[MAX_RT_PRIO];
241 struct rt_bandwidth {
242 /* nests inside the rq lock: */
243 raw_spinlock_t rt_runtime_lock;
246 struct hrtimer rt_period_timer;
247 unsigned int rt_period_active;
250 void __dl_clear_params(struct task_struct *p);
252 struct dl_bandwidth {
253 raw_spinlock_t dl_runtime_lock;
258 static inline int dl_bandwidth_enabled(void)
260 return sysctl_sched_rt_runtime >= 0;
264 * To keep the bandwidth of -deadline tasks under control
265 * we need some place where:
266 * - store the maximum -deadline bandwidth of each cpu;
267 * - cache the fraction of bandwidth that is currently allocated in
270 * This is all done in the data structure below. It is similar to the
271 * one used for RT-throttling (rt_bandwidth), with the main difference
272 * that, since here we are only interested in admission control, we
273 * do not decrease any runtime while the group "executes", neither we
274 * need a timer to replenish it.
276 * With respect to SMP, bandwidth is given on a per root domain basis,
278 * - bw (< 100%) is the deadline bandwidth of each CPU;
279 * - total_bw is the currently allocated bandwidth in each root domain;
287 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
290 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
292 dl_b->total_bw -= tsk_bw;
293 __dl_update(dl_b, (s32)tsk_bw / cpus);
297 void __dl_add(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 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
306 return dl_b->bw != -1 &&
307 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
310 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
311 extern void init_dl_bw(struct dl_bw *dl_b);
312 extern int sched_dl_global_validate(void);
313 extern void sched_dl_do_global(void);
314 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
315 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
316 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
317 extern bool __checkparam_dl(const struct sched_attr *attr);
318 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
319 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
320 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
321 extern bool dl_cpu_busy(unsigned int cpu);
323 #ifdef CONFIG_CGROUP_SCHED
325 #include <linux/cgroup.h>
326 #include <linux/psi.h>
331 extern struct list_head task_groups;
333 struct cfs_bandwidth {
334 #ifdef CONFIG_CFS_BANDWIDTH
339 s64 hierarchical_quota;
343 u8 distribute_running;
345 struct hrtimer period_timer;
346 struct hrtimer slack_timer;
347 struct list_head throttled_cfs_rq;
356 /* Task group related information */
358 struct cgroup_subsys_state css;
360 #ifdef CONFIG_FAIR_GROUP_SCHED
361 /* schedulable entities of this group on each CPU */
362 struct sched_entity **se;
363 /* runqueue "owned" by this group on each CPU */
364 struct cfs_rq **cfs_rq;
365 unsigned long shares;
369 * load_avg can be heavily contended at clock tick time, so put
370 * it in its own cacheline separated from the fields above which
371 * will also be accessed at each tick.
373 atomic_long_t load_avg ____cacheline_aligned;
377 #ifdef CONFIG_RT_GROUP_SCHED
378 struct sched_rt_entity **rt_se;
379 struct rt_rq **rt_rq;
381 struct rt_bandwidth rt_bandwidth;
385 struct list_head list;
387 struct task_group *parent;
388 struct list_head siblings;
389 struct list_head children;
391 #ifdef CONFIG_SCHED_AUTOGROUP
392 struct autogroup *autogroup;
395 struct cfs_bandwidth cfs_bandwidth;
397 #ifdef CONFIG_UCLAMP_TASK_GROUP
398 /* The two decimal precision [%] value requested from user-space */
399 unsigned int uclamp_pct[UCLAMP_CNT];
400 /* Clamp values requested for a task group */
401 struct uclamp_se uclamp_req[UCLAMP_CNT];
402 /* Effective clamp values used for a task group */
403 struct uclamp_se uclamp[UCLAMP_CNT];
408 #ifdef CONFIG_FAIR_GROUP_SCHED
409 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
412 * A weight of 0 or 1 can cause arithmetics problems.
413 * A weight of a cfs_rq is the sum of weights of which entities
414 * are queued on this cfs_rq, so a weight of a entity should not be
415 * too large, so as the shares value of a task group.
416 * (The default weight is 1024 - so there's no practical
417 * limitation from this.)
419 #define MIN_SHARES (1UL << 1)
420 #define MAX_SHARES (1UL << 18)
423 typedef int (*tg_visitor)(struct task_group *, void *);
425 extern int walk_tg_tree_from(struct task_group *from,
426 tg_visitor down, tg_visitor up, void *data);
429 * Iterate the full tree, calling @down when first entering a node and @up when
430 * leaving it for the final time.
432 * Caller must hold rcu_lock or sufficient equivalent.
434 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
436 return walk_tg_tree_from(&root_task_group, down, up, data);
439 extern int tg_nop(struct task_group *tg, void *data);
441 extern void free_fair_sched_group(struct task_group *tg);
442 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
443 extern void online_fair_sched_group(struct task_group *tg);
444 extern void unregister_fair_sched_group(struct task_group *tg);
445 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
446 struct sched_entity *se, int cpu,
447 struct sched_entity *parent);
448 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
450 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
451 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
452 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
454 extern void free_rt_sched_group(struct task_group *tg);
455 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
456 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
457 struct sched_rt_entity *rt_se, int cpu,
458 struct sched_rt_entity *parent);
459 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
460 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
461 extern long sched_group_rt_runtime(struct task_group *tg);
462 extern long sched_group_rt_period(struct task_group *tg);
463 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
465 extern struct task_group *sched_create_group(struct task_group *parent);
466 extern void sched_online_group(struct task_group *tg,
467 struct task_group *parent);
468 extern void sched_destroy_group(struct task_group *tg);
469 extern void sched_offline_group(struct task_group *tg);
471 extern void sched_move_task(struct task_struct *tsk);
473 #ifdef CONFIG_FAIR_GROUP_SCHED
474 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
477 extern void set_task_rq_fair(struct sched_entity *se,
478 struct cfs_rq *prev, struct cfs_rq *next);
479 #else /* !CONFIG_SMP */
480 static inline void set_task_rq_fair(struct sched_entity *se,
481 struct cfs_rq *prev, struct cfs_rq *next) { }
482 #endif /* CONFIG_SMP */
483 #endif /* CONFIG_FAIR_GROUP_SCHED */
485 #else /* CONFIG_CGROUP_SCHED */
487 struct cfs_bandwidth { };
489 #endif /* CONFIG_CGROUP_SCHED */
491 /* CFS-related fields in a runqueue */
493 struct load_weight load;
494 unsigned long runnable_weight;
495 unsigned int nr_running;
496 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
497 unsigned int idle_h_nr_running; /* SCHED_IDLE */
502 u64 min_vruntime_copy;
505 struct rb_root_cached tasks_timeline;
508 * 'curr' points to currently running entity on this cfs_rq.
509 * It is set to NULL otherwise (i.e when none are currently running).
511 struct sched_entity *curr;
512 struct sched_entity *next;
513 struct sched_entity *last;
514 struct sched_entity *skip;
516 #ifdef CONFIG_SCHED_DEBUG
517 unsigned int nr_spread_over;
524 struct sched_avg avg;
526 u64 load_last_update_time_copy;
529 raw_spinlock_t lock ____cacheline_aligned;
531 unsigned long load_avg;
532 unsigned long util_avg;
533 unsigned long runnable_sum;
536 #ifdef CONFIG_FAIR_GROUP_SCHED
537 unsigned long tg_load_avg_contrib;
539 long prop_runnable_sum;
542 * h_load = weight * f(tg)
544 * Where f(tg) is the recursive weight fraction assigned to
547 unsigned long h_load;
548 u64 last_h_load_update;
549 struct sched_entity *h_load_next;
550 #endif /* CONFIG_FAIR_GROUP_SCHED */
551 #endif /* CONFIG_SMP */
553 #ifdef CONFIG_FAIR_GROUP_SCHED
554 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
557 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
558 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
559 * (like users, containers etc.)
561 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
562 * This list is used during load balance.
565 struct list_head leaf_cfs_rq_list;
566 struct task_group *tg; /* group that "owns" this runqueue */
568 #ifdef CONFIG_CFS_BANDWIDTH
570 s64 runtime_remaining;
573 u64 throttled_clock_pelt;
574 u64 throttled_clock_pelt_time;
577 struct list_head throttled_list;
578 #endif /* CONFIG_CFS_BANDWIDTH */
579 #endif /* CONFIG_FAIR_GROUP_SCHED */
582 static inline int rt_bandwidth_enabled(void)
584 return sysctl_sched_rt_runtime >= 0;
587 /* RT IPI pull logic requires IRQ_WORK */
588 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
589 # define HAVE_RT_PUSH_IPI
592 /* Real-Time classes' related field in a runqueue: */
594 struct rt_prio_array active;
595 unsigned int rt_nr_running;
596 unsigned int rr_nr_running;
597 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
599 int curr; /* highest queued rt task prio */
601 int next; /* next highest */
606 unsigned long rt_nr_migratory;
607 unsigned long rt_nr_total;
609 struct plist_head pushable_tasks;
611 #endif /* CONFIG_SMP */
617 /* Nests inside the rq lock: */
618 raw_spinlock_t rt_runtime_lock;
620 #ifdef CONFIG_RT_GROUP_SCHED
621 unsigned long rt_nr_boosted;
624 struct task_group *tg;
628 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
630 return rt_rq->rt_queued && rt_rq->rt_nr_running;
633 /* Deadline class' related fields in a runqueue */
635 /* runqueue is an rbtree, ordered by deadline */
636 struct rb_root_cached root;
638 unsigned long dl_nr_running;
642 * Deadline values of the currently executing and the
643 * earliest ready task on this rq. Caching these facilitates
644 * the decision whether or not a ready but not running task
645 * should migrate somewhere else.
652 unsigned long dl_nr_migratory;
656 * Tasks on this rq that can be pushed away. They are kept in
657 * an rb-tree, ordered by tasks' deadlines, with caching
658 * of the leftmost (earliest deadline) element.
660 struct rb_root_cached pushable_dl_tasks_root;
665 * "Active utilization" for this runqueue: increased when a
666 * task wakes up (becomes TASK_RUNNING) and decreased when a
672 * Utilization of the tasks "assigned" to this runqueue (including
673 * the tasks that are in runqueue and the tasks that executed on this
674 * CPU and blocked). Increased when a task moves to this runqueue, and
675 * decreased when the task moves away (migrates, changes scheduling
676 * policy, or terminates).
677 * This is needed to compute the "inactive utilization" for the
678 * runqueue (inactive utilization = this_bw - running_bw).
684 * Inverse of the fraction of CPU utilization that can be reclaimed
685 * by the GRUB algorithm.
690 #ifdef CONFIG_FAIR_GROUP_SCHED
691 /* An entity is a task if it doesn't "own" a runqueue */
692 #define entity_is_task(se) (!se->my_q)
694 #define entity_is_task(se) 1
699 * XXX we want to get rid of these helpers and use the full load resolution.
701 static inline long se_weight(struct sched_entity *se)
703 return scale_load_down(se->load.weight);
706 static inline long se_runnable(struct sched_entity *se)
708 return scale_load_down(se->runnable_weight);
711 static inline bool sched_asym_prefer(int a, int b)
713 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
717 struct em_perf_domain *em_pd;
718 struct perf_domain *next;
722 /* Scheduling group status flags */
723 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
724 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
727 * We add the notion of a root-domain which will be used to define per-domain
728 * variables. Each exclusive cpuset essentially defines an island domain by
729 * fully partitioning the member CPUs from any other cpuset. Whenever a new
730 * exclusive cpuset is created, we also create and attach a new root-domain
739 cpumask_var_t online;
742 * Indicate pullable load on at least one CPU, e.g:
743 * - More than one runnable task
744 * - Running task is misfit
748 /* Indicate one or more cpus over-utilized (tipping point) */
752 * The bit corresponding to a CPU gets set here if such CPU has more
753 * than one runnable -deadline task (as it is below for RT tasks).
755 cpumask_var_t dlo_mask;
760 #ifdef HAVE_RT_PUSH_IPI
762 * For IPI pull requests, loop across the rto_mask.
764 struct irq_work rto_push_work;
765 raw_spinlock_t rto_lock;
766 /* These are only updated and read within rto_lock */
769 /* These atomics are updated outside of a lock */
770 atomic_t rto_loop_next;
771 atomic_t rto_loop_start;
774 * The "RT overload" flag: it gets set if a CPU has more than
775 * one runnable RT task.
777 cpumask_var_t rto_mask;
778 struct cpupri cpupri;
780 unsigned long max_cpu_capacity;
783 * NULL-terminated list of performance domains intersecting with the
784 * CPUs of the rd. Protected by RCU.
786 struct perf_domain __rcu *pd;
789 extern void init_defrootdomain(void);
790 extern int sched_init_domains(const struct cpumask *cpu_map);
791 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
792 extern void sched_get_rd(struct root_domain *rd);
793 extern void sched_put_rd(struct root_domain *rd);
795 #ifdef HAVE_RT_PUSH_IPI
796 extern void rto_push_irq_work_func(struct irq_work *work);
798 #endif /* CONFIG_SMP */
800 #ifdef CONFIG_UCLAMP_TASK
802 * struct uclamp_bucket - Utilization clamp bucket
803 * @value: utilization clamp value for tasks on this clamp bucket
804 * @tasks: number of RUNNABLE tasks on this clamp bucket
806 * Keep track of how many tasks are RUNNABLE for a given utilization
809 struct uclamp_bucket {
810 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
811 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
815 * struct uclamp_rq - rq's utilization clamp
816 * @value: currently active clamp values for a rq
817 * @bucket: utilization clamp buckets affecting a rq
819 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
820 * A clamp value is affecting a rq when there is at least one task RUNNABLE
821 * (or actually running) with that value.
823 * There are up to UCLAMP_CNT possible different clamp values, currently there
824 * are only two: minimum utilization and maximum utilization.
826 * All utilization clamping values are MAX aggregated, since:
827 * - for util_min: we want to run the CPU at least at the max of the minimum
828 * utilization required by its currently RUNNABLE tasks.
829 * - for util_max: we want to allow the CPU to run up to the max of the
830 * maximum utilization allowed by its currently RUNNABLE tasks.
832 * Since on each system we expect only a limited number of different
833 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
834 * the metrics required to compute all the per-rq utilization clamp values.
838 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
841 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
842 #endif /* CONFIG_UCLAMP_TASK */
845 * This is the main, per-CPU runqueue data structure.
847 * Locking rule: those places that want to lock multiple runqueues
848 * (such as the load balancing or the thread migration code), lock
849 * acquire operations must be ordered by ascending &runqueue.
856 * nr_running and cpu_load should be in the same cacheline because
857 * remote CPUs use both these fields when doing load calculation.
859 unsigned int nr_running;
860 #ifdef CONFIG_NUMA_BALANCING
861 unsigned int nr_numa_running;
862 unsigned int nr_preferred_running;
863 unsigned int numa_migrate_on;
865 #ifdef CONFIG_NO_HZ_COMMON
867 unsigned long last_load_update_tick;
868 unsigned long last_blocked_load_update_tick;
869 unsigned int has_blocked_load;
870 #endif /* CONFIG_SMP */
871 unsigned int nohz_tick_stopped;
873 #endif /* CONFIG_NO_HZ_COMMON */
875 unsigned long nr_load_updates;
878 #ifdef CONFIG_UCLAMP_TASK
879 /* Utilization clamp values based on CPU's RUNNABLE tasks */
880 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
881 unsigned int uclamp_flags;
882 #define UCLAMP_FLAG_IDLE 0x01
889 #ifdef CONFIG_FAIR_GROUP_SCHED
890 /* list of leaf cfs_rq on this CPU: */
891 struct list_head leaf_cfs_rq_list;
892 struct list_head *tmp_alone_branch;
893 #endif /* CONFIG_FAIR_GROUP_SCHED */
896 * This is part of a global counter where only the total sum
897 * over all CPUs matters. A task can increase this counter on
898 * one CPU and if it got migrated afterwards it may decrease
899 * it on another CPU. Always updated under the runqueue lock:
901 unsigned long nr_uninterruptible;
903 struct task_struct *curr;
904 struct task_struct *idle;
905 struct task_struct *stop;
906 unsigned long next_balance;
907 struct mm_struct *prev_mm;
909 unsigned int clock_update_flags;
911 /* Ensure that all clocks are in the same cache line */
912 u64 clock_task ____cacheline_aligned;
914 unsigned long lost_idle_time;
918 #ifdef CONFIG_MEMBARRIER
919 int membarrier_state;
923 struct root_domain *rd;
924 struct sched_domain __rcu *sd;
926 unsigned long cpu_capacity;
927 unsigned long cpu_capacity_orig;
929 struct callback_head *balance_callback;
931 unsigned char idle_balance;
933 unsigned long misfit_task_load;
935 /* For active balancing */
938 struct cpu_stop_work active_balance_work;
940 /* CPU of this runqueue: */
944 struct list_head cfs_tasks;
946 struct sched_avg avg_rt;
947 struct sched_avg avg_dl;
948 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
949 struct sched_avg avg_irq;
954 /* This is used to determine avg_idle's max value */
955 u64 max_idle_balance_cost;
958 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
961 #ifdef CONFIG_PARAVIRT
964 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
965 u64 prev_steal_time_rq;
968 /* calc_load related fields */
969 unsigned long calc_load_update;
970 long calc_load_active;
972 #ifdef CONFIG_SCHED_HRTICK
974 int hrtick_csd_pending;
975 call_single_data_t hrtick_csd;
977 struct hrtimer hrtick_timer;
981 #ifdef CONFIG_SCHEDSTATS
983 struct sched_info rq_sched_info;
984 unsigned long long rq_cpu_time;
985 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
987 /* sys_sched_yield() stats */
988 unsigned int yld_count;
990 /* schedule() stats */
991 unsigned int sched_count;
992 unsigned int sched_goidle;
994 /* try_to_wake_up() stats */
995 unsigned int ttwu_count;
996 unsigned int ttwu_local;
1000 struct llist_head wake_list;
1003 #ifdef CONFIG_CPU_IDLE
1004 /* Must be inspected within a rcu lock section */
1005 struct cpuidle_state *idle_state;
1009 #ifdef CONFIG_FAIR_GROUP_SCHED
1011 /* CPU runqueue to which this cfs_rq is attached */
1012 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1019 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1021 return container_of(cfs_rq, struct rq, cfs);
1025 static inline int cpu_of(struct rq *rq)
1035 #ifdef CONFIG_SCHED_SMT
1036 extern void __update_idle_core(struct rq *rq);
1038 static inline void update_idle_core(struct rq *rq)
1040 if (static_branch_unlikely(&sched_smt_present))
1041 __update_idle_core(rq);
1045 static inline void update_idle_core(struct rq *rq) { }
1048 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1050 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1051 #define this_rq() this_cpu_ptr(&runqueues)
1052 #define task_rq(p) cpu_rq(task_cpu(p))
1053 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1054 #define raw_rq() raw_cpu_ptr(&runqueues)
1056 extern void update_rq_clock(struct rq *rq);
1058 static inline u64 __rq_clock_broken(struct rq *rq)
1060 return READ_ONCE(rq->clock);
1064 * rq::clock_update_flags bits
1066 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1067 * call to __schedule(). This is an optimisation to avoid
1068 * neighbouring rq clock updates.
1070 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1071 * in effect and calls to update_rq_clock() are being ignored.
1073 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1074 * made to update_rq_clock() since the last time rq::lock was pinned.
1076 * If inside of __schedule(), clock_update_flags will have been
1077 * shifted left (a left shift is a cheap operation for the fast path
1078 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1080 * if (rq-clock_update_flags >= RQCF_UPDATED)
1082 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1083 * one position though, because the next rq_unpin_lock() will shift it
1086 #define RQCF_REQ_SKIP 0x01
1087 #define RQCF_ACT_SKIP 0x02
1088 #define RQCF_UPDATED 0x04
1090 static inline void assert_clock_updated(struct rq *rq)
1093 * The only reason for not seeing a clock update since the
1094 * last rq_pin_lock() is if we're currently skipping updates.
1096 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1099 static inline u64 rq_clock(struct rq *rq)
1101 lockdep_assert_held(&rq->lock);
1102 assert_clock_updated(rq);
1107 static inline u64 rq_clock_task(struct rq *rq)
1109 lockdep_assert_held(&rq->lock);
1110 assert_clock_updated(rq);
1112 return rq->clock_task;
1115 static inline void rq_clock_skip_update(struct rq *rq)
1117 lockdep_assert_held(&rq->lock);
1118 rq->clock_update_flags |= RQCF_REQ_SKIP;
1122 * See rt task throttling, which is the only time a skip
1123 * request is cancelled.
1125 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1127 lockdep_assert_held(&rq->lock);
1128 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1132 unsigned long flags;
1133 struct pin_cookie cookie;
1134 #ifdef CONFIG_SCHED_DEBUG
1136 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1137 * current pin context is stashed here in case it needs to be
1138 * restored in rq_repin_lock().
1140 unsigned int clock_update_flags;
1144 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1146 rf->cookie = lockdep_pin_lock(&rq->lock);
1148 #ifdef CONFIG_SCHED_DEBUG
1149 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1150 rf->clock_update_flags = 0;
1154 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1156 #ifdef CONFIG_SCHED_DEBUG
1157 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1158 rf->clock_update_flags = RQCF_UPDATED;
1161 lockdep_unpin_lock(&rq->lock, rf->cookie);
1164 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1166 lockdep_repin_lock(&rq->lock, rf->cookie);
1168 #ifdef CONFIG_SCHED_DEBUG
1170 * Restore the value we stashed in @rf for this pin context.
1172 rq->clock_update_flags |= rf->clock_update_flags;
1176 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1177 __acquires(rq->lock);
1179 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1180 __acquires(p->pi_lock)
1181 __acquires(rq->lock);
1183 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1184 __releases(rq->lock)
1186 rq_unpin_lock(rq, rf);
1187 raw_spin_unlock(&rq->lock);
1191 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1192 __releases(rq->lock)
1193 __releases(p->pi_lock)
1195 rq_unpin_lock(rq, rf);
1196 raw_spin_unlock(&rq->lock);
1197 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1201 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1202 __acquires(rq->lock)
1204 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1205 rq_pin_lock(rq, rf);
1209 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1210 __acquires(rq->lock)
1212 raw_spin_lock_irq(&rq->lock);
1213 rq_pin_lock(rq, rf);
1217 rq_lock(struct rq *rq, struct rq_flags *rf)
1218 __acquires(rq->lock)
1220 raw_spin_lock(&rq->lock);
1221 rq_pin_lock(rq, rf);
1225 rq_relock(struct rq *rq, struct rq_flags *rf)
1226 __acquires(rq->lock)
1228 raw_spin_lock(&rq->lock);
1229 rq_repin_lock(rq, rf);
1233 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1234 __releases(rq->lock)
1236 rq_unpin_lock(rq, rf);
1237 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1241 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1242 __releases(rq->lock)
1244 rq_unpin_lock(rq, rf);
1245 raw_spin_unlock_irq(&rq->lock);
1249 rq_unlock(struct rq *rq, struct rq_flags *rf)
1250 __releases(rq->lock)
1252 rq_unpin_lock(rq, rf);
1253 raw_spin_unlock(&rq->lock);
1256 static inline struct rq *
1257 this_rq_lock_irq(struct rq_flags *rf)
1258 __acquires(rq->lock)
1262 local_irq_disable();
1269 enum numa_topology_type {
1274 extern enum numa_topology_type sched_numa_topology_type;
1275 extern int sched_max_numa_distance;
1276 extern bool find_numa_distance(int distance);
1277 extern void sched_init_numa(void);
1278 extern void sched_domains_numa_masks_set(unsigned int cpu);
1279 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1280 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1282 static inline void sched_init_numa(void) { }
1283 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1284 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1285 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1291 #ifdef CONFIG_NUMA_BALANCING
1292 /* The regions in numa_faults array from task_struct */
1293 enum numa_faults_stats {
1299 extern void sched_setnuma(struct task_struct *p, int node);
1300 extern int migrate_task_to(struct task_struct *p, int cpu);
1301 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1303 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1306 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1309 #endif /* CONFIG_NUMA_BALANCING */
1314 queue_balance_callback(struct rq *rq,
1315 struct callback_head *head,
1316 void (*func)(struct rq *rq))
1318 lockdep_assert_held(&rq->lock);
1320 if (unlikely(head->next))
1323 head->func = (void (*)(struct callback_head *))func;
1324 head->next = rq->balance_callback;
1325 rq->balance_callback = head;
1328 extern void sched_ttwu_pending(void);
1330 #define rcu_dereference_check_sched_domain(p) \
1331 rcu_dereference_check((p), \
1332 lockdep_is_held(&sched_domains_mutex))
1335 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1336 * See destroy_sched_domains: call_rcu for details.
1338 * The domain tree of any CPU may only be accessed from within
1339 * preempt-disabled sections.
1341 #define for_each_domain(cpu, __sd) \
1342 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1343 __sd; __sd = __sd->parent)
1345 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1348 * highest_flag_domain - Return highest sched_domain containing flag.
1349 * @cpu: The CPU whose highest level of sched domain is to
1351 * @flag: The flag to check for the highest sched_domain
1352 * for the given CPU.
1354 * Returns the highest sched_domain of a CPU which contains the given flag.
1356 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1358 struct sched_domain *sd, *hsd = NULL;
1360 for_each_domain(cpu, sd) {
1361 if (!(sd->flags & flag))
1369 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1371 struct sched_domain *sd;
1373 for_each_domain(cpu, sd) {
1374 if (sd->flags & flag)
1381 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1382 DECLARE_PER_CPU(int, sd_llc_size);
1383 DECLARE_PER_CPU(int, sd_llc_id);
1384 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1385 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1386 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1387 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1388 extern struct static_key_false sched_asym_cpucapacity;
1390 struct sched_group_capacity {
1393 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1396 unsigned long capacity;
1397 unsigned long min_capacity; /* Min per-CPU capacity in group */
1398 unsigned long max_capacity; /* Max per-CPU capacity in group */
1399 unsigned long next_update;
1400 int imbalance; /* XXX unrelated to capacity but shared group state */
1402 #ifdef CONFIG_SCHED_DEBUG
1406 unsigned long cpumask[0]; /* Balance mask */
1409 struct sched_group {
1410 struct sched_group *next; /* Must be a circular list */
1413 unsigned int group_weight;
1414 struct sched_group_capacity *sgc;
1415 int asym_prefer_cpu; /* CPU of highest priority in group */
1418 * The CPUs this group covers.
1420 * NOTE: this field is variable length. (Allocated dynamically
1421 * by attaching extra space to the end of the structure,
1422 * depending on how many CPUs the kernel has booted up with)
1424 unsigned long cpumask[0];
1427 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1429 return to_cpumask(sg->cpumask);
1433 * See build_balance_mask().
1435 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1437 return to_cpumask(sg->sgc->cpumask);
1441 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1442 * @group: The group whose first CPU is to be returned.
1444 static inline unsigned int group_first_cpu(struct sched_group *group)
1446 return cpumask_first(sched_group_span(group));
1449 extern int group_balance_cpu(struct sched_group *sg);
1451 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1452 void register_sched_domain_sysctl(void);
1453 void dirty_sched_domain_sysctl(int cpu);
1454 void unregister_sched_domain_sysctl(void);
1456 static inline void register_sched_domain_sysctl(void)
1459 static inline void dirty_sched_domain_sysctl(int cpu)
1462 static inline void unregister_sched_domain_sysctl(void)
1467 extern int newidle_balance(struct rq *this_rq, struct rq_flags *rf);
1471 static inline void sched_ttwu_pending(void) { }
1473 static inline int newidle_balance(struct rq *this_rq, struct rq_flags *rf) { return 0; }
1475 #endif /* CONFIG_SMP */
1478 #include "autogroup.h"
1480 #ifdef CONFIG_CGROUP_SCHED
1483 * Return the group to which this tasks belongs.
1485 * We cannot use task_css() and friends because the cgroup subsystem
1486 * changes that value before the cgroup_subsys::attach() method is called,
1487 * therefore we cannot pin it and might observe the wrong value.
1489 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1490 * core changes this before calling sched_move_task().
1492 * Instead we use a 'copy' which is updated from sched_move_task() while
1493 * holding both task_struct::pi_lock and rq::lock.
1495 static inline struct task_group *task_group(struct task_struct *p)
1497 return p->sched_task_group;
1500 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1501 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1503 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1504 struct task_group *tg = task_group(p);
1507 #ifdef CONFIG_FAIR_GROUP_SCHED
1508 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1509 p->se.cfs_rq = tg->cfs_rq[cpu];
1510 p->se.parent = tg->se[cpu];
1513 #ifdef CONFIG_RT_GROUP_SCHED
1514 p->rt.rt_rq = tg->rt_rq[cpu];
1515 p->rt.parent = tg->rt_se[cpu];
1519 #else /* CONFIG_CGROUP_SCHED */
1521 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1522 static inline struct task_group *task_group(struct task_struct *p)
1527 #endif /* CONFIG_CGROUP_SCHED */
1529 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1531 set_task_rq(p, cpu);
1534 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1535 * successfully executed on another CPU. We must ensure that updates of
1536 * per-task data have been completed by this moment.
1539 #ifdef CONFIG_THREAD_INFO_IN_TASK
1540 WRITE_ONCE(p->cpu, cpu);
1542 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1549 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1551 #ifdef CONFIG_SCHED_DEBUG
1552 # include <linux/static_key.h>
1553 # define const_debug __read_mostly
1555 # define const_debug const
1558 #define SCHED_FEAT(name, enabled) \
1559 __SCHED_FEAT_##name ,
1562 #include "features.h"
1568 #ifdef CONFIG_SCHED_DEBUG
1571 * To support run-time toggling of sched features, all the translation units
1572 * (but core.c) reference the sysctl_sched_features defined in core.c.
1574 extern const_debug unsigned int sysctl_sched_features;
1576 #ifdef CONFIG_JUMP_LABEL
1577 #define SCHED_FEAT(name, enabled) \
1578 static __always_inline bool static_branch_##name(struct static_key *key) \
1580 return static_key_##enabled(key); \
1583 #include "features.h"
1586 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1587 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1589 #else /* !CONFIG_JUMP_LABEL */
1591 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1593 #endif /* CONFIG_JUMP_LABEL */
1595 #else /* !SCHED_DEBUG */
1598 * Each translation unit has its own copy of sysctl_sched_features to allow
1599 * constants propagation at compile time and compiler optimization based on
1602 #define SCHED_FEAT(name, enabled) \
1603 (1UL << __SCHED_FEAT_##name) * enabled |
1604 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1605 #include "features.h"
1609 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1611 #endif /* SCHED_DEBUG */
1613 extern struct static_key_false sched_numa_balancing;
1614 extern struct static_key_false sched_schedstats;
1616 static inline u64 global_rt_period(void)
1618 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1621 static inline u64 global_rt_runtime(void)
1623 if (sysctl_sched_rt_runtime < 0)
1626 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1629 static inline int task_current(struct rq *rq, struct task_struct *p)
1631 return rq->curr == p;
1634 static inline int task_running(struct rq *rq, struct task_struct *p)
1639 return task_current(rq, p);
1643 static inline int task_on_rq_queued(struct task_struct *p)
1645 return p->on_rq == TASK_ON_RQ_QUEUED;
1648 static inline int task_on_rq_migrating(struct task_struct *p)
1650 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1656 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1657 #define WF_FORK 0x02 /* Child wakeup after fork */
1658 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1661 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1662 * of tasks with abnormal "nice" values across CPUs the contribution that
1663 * each task makes to its run queue's load is weighted according to its
1664 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1665 * scaled version of the new time slice allocation that they receive on time
1669 #define WEIGHT_IDLEPRIO 3
1670 #define WMULT_IDLEPRIO 1431655765
1672 extern const int sched_prio_to_weight[40];
1673 extern const u32 sched_prio_to_wmult[40];
1676 * {de,en}queue flags:
1678 * DEQUEUE_SLEEP - task is no longer runnable
1679 * ENQUEUE_WAKEUP - task just became runnable
1681 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1682 * are in a known state which allows modification. Such pairs
1683 * should preserve as much state as possible.
1685 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1688 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1689 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1690 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1694 #define DEQUEUE_SLEEP 0x01
1695 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1696 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1697 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1699 #define ENQUEUE_WAKEUP 0x01
1700 #define ENQUEUE_RESTORE 0x02
1701 #define ENQUEUE_MOVE 0x04
1702 #define ENQUEUE_NOCLOCK 0x08
1704 #define ENQUEUE_HEAD 0x10
1705 #define ENQUEUE_REPLENISH 0x20
1707 #define ENQUEUE_MIGRATED 0x40
1709 #define ENQUEUE_MIGRATED 0x00
1712 #define RETRY_TASK ((void *)-1UL)
1714 struct sched_class {
1715 const struct sched_class *next;
1717 #ifdef CONFIG_UCLAMP_TASK
1721 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1722 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1723 void (*yield_task) (struct rq *rq);
1724 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1726 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1729 * Both @prev and @rf are optional and may be NULL, in which case the
1730 * caller must already have invoked put_prev_task(rq, prev, rf).
1732 * Otherwise it is the responsibility of the pick_next_task() to call
1733 * put_prev_task() on the @prev task or something equivalent, IFF it
1734 * returns a next task.
1736 * In that case (@rf != NULL) it may return RETRY_TASK when it finds a
1737 * higher prio class has runnable tasks.
1739 struct task_struct * (*pick_next_task)(struct rq *rq,
1740 struct task_struct *prev,
1741 struct rq_flags *rf);
1742 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1743 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1746 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1747 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1748 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1750 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1752 void (*set_cpus_allowed)(struct task_struct *p,
1753 const struct cpumask *newmask);
1755 void (*rq_online)(struct rq *rq);
1756 void (*rq_offline)(struct rq *rq);
1759 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1760 void (*task_fork)(struct task_struct *p);
1761 void (*task_dead)(struct task_struct *p);
1764 * The switched_from() call is allowed to drop rq->lock, therefore we
1765 * cannot assume the switched_from/switched_to pair is serliazed by
1766 * rq->lock. They are however serialized by p->pi_lock.
1768 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1769 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1770 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1773 unsigned int (*get_rr_interval)(struct rq *rq,
1774 struct task_struct *task);
1776 void (*update_curr)(struct rq *rq);
1778 #define TASK_SET_GROUP 0
1779 #define TASK_MOVE_GROUP 1
1781 #ifdef CONFIG_FAIR_GROUP_SCHED
1782 void (*task_change_group)(struct task_struct *p, int type);
1786 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1788 WARN_ON_ONCE(rq->curr != prev);
1789 prev->sched_class->put_prev_task(rq, prev);
1792 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1794 WARN_ON_ONCE(rq->curr != next);
1795 next->sched_class->set_next_task(rq, next, false);
1799 #define sched_class_highest (&stop_sched_class)
1801 #define sched_class_highest (&dl_sched_class)
1804 #define for_class_range(class, _from, _to) \
1805 for (class = (_from); class != (_to); class = class->next)
1807 #define for_each_class(class) \
1808 for_class_range(class, sched_class_highest, NULL)
1810 extern const struct sched_class stop_sched_class;
1811 extern const struct sched_class dl_sched_class;
1812 extern const struct sched_class rt_sched_class;
1813 extern const struct sched_class fair_sched_class;
1814 extern const struct sched_class idle_sched_class;
1816 static inline bool sched_stop_runnable(struct rq *rq)
1818 return rq->stop && task_on_rq_queued(rq->stop);
1821 static inline bool sched_dl_runnable(struct rq *rq)
1823 return rq->dl.dl_nr_running > 0;
1826 static inline bool sched_rt_runnable(struct rq *rq)
1828 return rq->rt.rt_queued > 0;
1831 static inline bool sched_fair_runnable(struct rq *rq)
1833 return rq->cfs.nr_running > 0;
1838 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1840 extern void trigger_load_balance(struct rq *rq);
1842 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1846 #ifdef CONFIG_CPU_IDLE
1847 static inline void idle_set_state(struct rq *rq,
1848 struct cpuidle_state *idle_state)
1850 rq->idle_state = idle_state;
1853 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1855 SCHED_WARN_ON(!rcu_read_lock_held());
1857 return rq->idle_state;
1860 static inline void idle_set_state(struct rq *rq,
1861 struct cpuidle_state *idle_state)
1865 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1871 extern void schedule_idle(void);
1873 extern void sysrq_sched_debug_show(void);
1874 extern void sched_init_granularity(void);
1875 extern void update_max_interval(void);
1877 extern void init_sched_dl_class(void);
1878 extern void init_sched_rt_class(void);
1879 extern void init_sched_fair_class(void);
1881 extern void reweight_task(struct task_struct *p, int prio);
1883 extern void resched_curr(struct rq *rq);
1884 extern void resched_cpu(int cpu);
1886 extern struct rt_bandwidth def_rt_bandwidth;
1887 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1889 extern struct dl_bandwidth def_dl_bandwidth;
1890 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1891 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1892 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1893 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1896 #define BW_UNIT (1 << BW_SHIFT)
1897 #define RATIO_SHIFT 8
1898 #define MAX_BW_BITS (64 - BW_SHIFT)
1899 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
1900 unsigned long to_ratio(u64 period, u64 runtime);
1902 extern void init_entity_runnable_average(struct sched_entity *se);
1903 extern void post_init_entity_util_avg(struct task_struct *p);
1905 #ifdef CONFIG_NO_HZ_FULL
1906 extern bool sched_can_stop_tick(struct rq *rq);
1907 extern int __init sched_tick_offload_init(void);
1910 * Tick may be needed by tasks in the runqueue depending on their policy and
1911 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1912 * nohz mode if necessary.
1914 static inline void sched_update_tick_dependency(struct rq *rq)
1918 if (!tick_nohz_full_enabled())
1923 if (!tick_nohz_full_cpu(cpu))
1926 if (sched_can_stop_tick(rq))
1927 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1929 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1932 static inline int sched_tick_offload_init(void) { return 0; }
1933 static inline void sched_update_tick_dependency(struct rq *rq) { }
1936 static inline void add_nr_running(struct rq *rq, unsigned count)
1938 unsigned prev_nr = rq->nr_running;
1940 rq->nr_running = prev_nr + count;
1943 if (prev_nr < 2 && rq->nr_running >= 2) {
1944 if (!READ_ONCE(rq->rd->overload))
1945 WRITE_ONCE(rq->rd->overload, 1);
1949 sched_update_tick_dependency(rq);
1952 static inline void sub_nr_running(struct rq *rq, unsigned count)
1954 rq->nr_running -= count;
1955 /* Check if we still need preemption */
1956 sched_update_tick_dependency(rq);
1959 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1960 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1962 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1964 extern const_debug unsigned int sysctl_sched_nr_migrate;
1965 extern const_debug unsigned int sysctl_sched_migration_cost;
1967 #ifdef CONFIG_SCHED_HRTICK
1971 * - enabled by features
1972 * - hrtimer is actually high res
1974 static inline int hrtick_enabled(struct rq *rq)
1976 if (!sched_feat(HRTICK))
1978 if (!cpu_active(cpu_of(rq)))
1980 return hrtimer_is_hres_active(&rq->hrtick_timer);
1983 void hrtick_start(struct rq *rq, u64 delay);
1987 static inline int hrtick_enabled(struct rq *rq)
1992 #endif /* CONFIG_SCHED_HRTICK */
1994 #ifndef arch_scale_freq_capacity
1995 static __always_inline
1996 unsigned long arch_scale_freq_capacity(int cpu)
1998 return SCHED_CAPACITY_SCALE;
2003 #ifdef CONFIG_PREEMPTION
2005 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2008 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2009 * way at the expense of forcing extra atomic operations in all
2010 * invocations. This assures that the double_lock is acquired using the
2011 * same underlying policy as the spinlock_t on this architecture, which
2012 * reduces latency compared to the unfair variant below. However, it
2013 * also adds more overhead and therefore may reduce throughput.
2015 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2016 __releases(this_rq->lock)
2017 __acquires(busiest->lock)
2018 __acquires(this_rq->lock)
2020 raw_spin_unlock(&this_rq->lock);
2021 double_rq_lock(this_rq, busiest);
2028 * Unfair double_lock_balance: Optimizes throughput at the expense of
2029 * latency by eliminating extra atomic operations when the locks are
2030 * already in proper order on entry. This favors lower CPU-ids and will
2031 * grant the double lock to lower CPUs over higher ids under contention,
2032 * regardless of entry order into the function.
2034 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2035 __releases(this_rq->lock)
2036 __acquires(busiest->lock)
2037 __acquires(this_rq->lock)
2041 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2042 if (busiest < this_rq) {
2043 raw_spin_unlock(&this_rq->lock);
2044 raw_spin_lock(&busiest->lock);
2045 raw_spin_lock_nested(&this_rq->lock,
2046 SINGLE_DEPTH_NESTING);
2049 raw_spin_lock_nested(&busiest->lock,
2050 SINGLE_DEPTH_NESTING);
2055 #endif /* CONFIG_PREEMPTION */
2058 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2060 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2062 if (unlikely(!irqs_disabled())) {
2063 /* printk() doesn't work well under rq->lock */
2064 raw_spin_unlock(&this_rq->lock);
2068 return _double_lock_balance(this_rq, busiest);
2071 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2072 __releases(busiest->lock)
2074 raw_spin_unlock(&busiest->lock);
2075 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2078 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2084 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2087 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2093 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2096 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2102 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2106 * double_rq_lock - safely lock two runqueues
2108 * Note this does not disable interrupts like task_rq_lock,
2109 * you need to do so manually before calling.
2111 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2112 __acquires(rq1->lock)
2113 __acquires(rq2->lock)
2115 BUG_ON(!irqs_disabled());
2117 raw_spin_lock(&rq1->lock);
2118 __acquire(rq2->lock); /* Fake it out ;) */
2121 raw_spin_lock(&rq1->lock);
2122 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2124 raw_spin_lock(&rq2->lock);
2125 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2131 * double_rq_unlock - safely unlock two runqueues
2133 * Note this does not restore interrupts like task_rq_unlock,
2134 * you need to do so manually after calling.
2136 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2137 __releases(rq1->lock)
2138 __releases(rq2->lock)
2140 raw_spin_unlock(&rq1->lock);
2142 raw_spin_unlock(&rq2->lock);
2144 __release(rq2->lock);
2147 extern void set_rq_online (struct rq *rq);
2148 extern void set_rq_offline(struct rq *rq);
2149 extern bool sched_smp_initialized;
2151 #else /* CONFIG_SMP */
2154 * double_rq_lock - safely lock two runqueues
2156 * Note this does not disable interrupts like task_rq_lock,
2157 * you need to do so manually before calling.
2159 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2160 __acquires(rq1->lock)
2161 __acquires(rq2->lock)
2163 BUG_ON(!irqs_disabled());
2165 raw_spin_lock(&rq1->lock);
2166 __acquire(rq2->lock); /* Fake it out ;) */
2170 * double_rq_unlock - safely unlock two runqueues
2172 * Note this does not restore interrupts like task_rq_unlock,
2173 * you need to do so manually after calling.
2175 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2176 __releases(rq1->lock)
2177 __releases(rq2->lock)
2180 raw_spin_unlock(&rq1->lock);
2181 __release(rq2->lock);
2186 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2187 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2189 #ifdef CONFIG_SCHED_DEBUG
2190 extern bool sched_debug_enabled;
2192 extern void print_cfs_stats(struct seq_file *m, int cpu);
2193 extern void print_rt_stats(struct seq_file *m, int cpu);
2194 extern void print_dl_stats(struct seq_file *m, int cpu);
2195 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2196 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2197 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2198 #ifdef CONFIG_NUMA_BALANCING
2200 show_numa_stats(struct task_struct *p, struct seq_file *m);
2202 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2203 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2204 #endif /* CONFIG_NUMA_BALANCING */
2205 #endif /* CONFIG_SCHED_DEBUG */
2207 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2208 extern void init_rt_rq(struct rt_rq *rt_rq);
2209 extern void init_dl_rq(struct dl_rq *dl_rq);
2211 extern void cfs_bandwidth_usage_inc(void);
2212 extern void cfs_bandwidth_usage_dec(void);
2214 #ifdef CONFIG_NO_HZ_COMMON
2215 #define NOHZ_BALANCE_KICK_BIT 0
2216 #define NOHZ_STATS_KICK_BIT 1
2218 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2219 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2221 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2223 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2225 extern void nohz_balance_exit_idle(struct rq *rq);
2227 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2233 void __dl_update(struct dl_bw *dl_b, s64 bw)
2235 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2238 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2239 "sched RCU must be held");
2240 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2241 struct rq *rq = cpu_rq(i);
2243 rq->dl.extra_bw += bw;
2248 void __dl_update(struct dl_bw *dl_b, s64 bw)
2250 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2257 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2262 struct u64_stats_sync sync;
2265 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2268 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2269 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2270 * and never move forward.
2272 static inline u64 irq_time_read(int cpu)
2274 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2279 seq = __u64_stats_fetch_begin(&irqtime->sync);
2280 total = irqtime->total;
2281 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2285 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2287 #ifdef CONFIG_CPU_FREQ
2288 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2291 * cpufreq_update_util - Take a note about CPU utilization changes.
2292 * @rq: Runqueue to carry out the update for.
2293 * @flags: Update reason flags.
2295 * This function is called by the scheduler on the CPU whose utilization is
2298 * It can only be called from RCU-sched read-side critical sections.
2300 * The way cpufreq is currently arranged requires it to evaluate the CPU
2301 * performance state (frequency/voltage) on a regular basis to prevent it from
2302 * being stuck in a completely inadequate performance level for too long.
2303 * That is not guaranteed to happen if the updates are only triggered from CFS
2304 * and DL, though, because they may not be coming in if only RT tasks are
2305 * active all the time (or there are RT tasks only).
2307 * As a workaround for that issue, this function is called periodically by the
2308 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2309 * but that really is a band-aid. Going forward it should be replaced with
2310 * solutions targeted more specifically at RT tasks.
2312 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2314 struct update_util_data *data;
2316 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2319 data->func(data, rq_clock(rq), flags);
2322 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2323 #endif /* CONFIG_CPU_FREQ */
2325 #ifdef CONFIG_UCLAMP_TASK
2326 unsigned int uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2329 * uclamp_util_with - clamp @util with @rq and @p effective uclamp values.
2330 * @rq: The rq to clamp against. Must not be NULL.
2331 * @util: The util value to clamp.
2332 * @p: The task to clamp against. Can be NULL if you want to clamp
2335 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2337 * If sched_uclamp_used static key is disabled, then just return the util
2338 * without any clamping since uclamp aggregation at the rq level in the fast
2339 * path is disabled, rendering this operation a NOP.
2341 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2342 * will return the correct effective uclamp value of the task even if the
2343 * static key is disabled.
2345 static __always_inline
2346 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2347 struct task_struct *p)
2349 unsigned int min_util;
2350 unsigned int max_util;
2352 if (!static_branch_likely(&sched_uclamp_used))
2355 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2356 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2359 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2360 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2364 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2365 * RUNNABLE tasks with _different_ clamps, we can end up with an
2366 * inversion. Fix it now when the clamps are applied.
2368 if (unlikely(min_util >= max_util))
2371 return clamp(util, min_util, max_util);
2374 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2376 return uclamp_util_with(rq, util, NULL);
2380 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2381 * by default in the fast path and only gets turned on once userspace performs
2382 * an operation that requires it.
2384 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2387 static inline bool uclamp_is_used(void)
2389 return static_branch_likely(&sched_uclamp_used);
2391 #else /* CONFIG_UCLAMP_TASK */
2392 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2393 struct task_struct *p)
2397 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2402 static inline bool uclamp_is_used(void)
2406 #endif /* CONFIG_UCLAMP_TASK */
2408 #ifdef arch_scale_freq_capacity
2409 # ifndef arch_scale_freq_invariant
2410 # define arch_scale_freq_invariant() true
2413 # define arch_scale_freq_invariant() false
2417 static inline unsigned long capacity_orig_of(int cpu)
2419 return cpu_rq(cpu)->cpu_capacity_orig;
2424 * enum schedutil_type - CPU utilization type
2425 * @FREQUENCY_UTIL: Utilization used to select frequency
2426 * @ENERGY_UTIL: Utilization used during energy calculation
2428 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2429 * need to be aggregated differently depending on the usage made of them. This
2430 * enum is used within schedutil_freq_util() to differentiate the types of
2431 * utilization expected by the callers, and adjust the aggregation accordingly.
2433 enum schedutil_type {
2438 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2440 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2441 unsigned long max, enum schedutil_type type,
2442 struct task_struct *p);
2444 static inline unsigned long cpu_bw_dl(struct rq *rq)
2446 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2449 static inline unsigned long cpu_util_dl(struct rq *rq)
2451 return READ_ONCE(rq->avg_dl.util_avg);
2454 static inline unsigned long cpu_util_cfs(struct rq *rq)
2456 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2458 if (sched_feat(UTIL_EST)) {
2459 util = max_t(unsigned long, util,
2460 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2466 static inline unsigned long cpu_util_rt(struct rq *rq)
2468 return READ_ONCE(rq->avg_rt.util_avg);
2470 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2471 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2472 unsigned long max, enum schedutil_type type,
2473 struct task_struct *p)
2477 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2479 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2480 static inline unsigned long cpu_util_irq(struct rq *rq)
2482 return rq->avg_irq.util_avg;
2486 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2488 util *= (max - irq);
2495 static inline unsigned long cpu_util_irq(struct rq *rq)
2501 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2507 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2509 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2511 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2513 static inline bool sched_energy_enabled(void)
2515 return static_branch_unlikely(&sched_energy_present);
2518 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2520 #define perf_domain_span(pd) NULL
2521 static inline bool sched_energy_enabled(void) { return false; }
2523 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2525 #ifdef CONFIG_MEMBARRIER
2527 * The scheduler provides memory barriers required by membarrier between:
2528 * - prior user-space memory accesses and store to rq->membarrier_state,
2529 * - store to rq->membarrier_state and following user-space memory accesses.
2530 * In the same way it provides those guarantees around store to rq->curr.
2532 static inline void membarrier_switch_mm(struct rq *rq,
2533 struct mm_struct *prev_mm,
2534 struct mm_struct *next_mm)
2536 int membarrier_state;
2538 if (prev_mm == next_mm)
2541 membarrier_state = atomic_read(&next_mm->membarrier_state);
2542 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2545 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2548 static inline void membarrier_switch_mm(struct rq *rq,
2549 struct mm_struct *prev_mm,
2550 struct mm_struct *next_mm)