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;
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;
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
402 * A weight of 0 or 1 can cause arithmetics problems.
403 * A weight of a cfs_rq is the sum of weights of which entities
404 * are queued on this cfs_rq, so a weight of a entity should not be
405 * too large, so as the shares value of a task group.
406 * (The default weight is 1024 - so there's no practical
407 * limitation from this.)
409 #define MIN_SHARES (1UL << 1)
410 #define MAX_SHARES (1UL << 18)
413 typedef int (*tg_visitor)(struct task_group *, void *);
415 extern int walk_tg_tree_from(struct task_group *from,
416 tg_visitor down, tg_visitor up, void *data);
419 * Iterate the full tree, calling @down when first entering a node and @up when
420 * leaving it for the final time.
422 * Caller must hold rcu_lock or sufficient equivalent.
424 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
426 return walk_tg_tree_from(&root_task_group, down, up, data);
429 extern int tg_nop(struct task_group *tg, void *data);
431 extern void free_fair_sched_group(struct task_group *tg);
432 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
433 extern void online_fair_sched_group(struct task_group *tg);
434 extern void unregister_fair_sched_group(struct task_group *tg);
435 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
436 struct sched_entity *se, int cpu,
437 struct sched_entity *parent);
438 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
440 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
441 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
442 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
444 extern void free_rt_sched_group(struct task_group *tg);
445 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
446 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
447 struct sched_rt_entity *rt_se, int cpu,
448 struct sched_rt_entity *parent);
449 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
450 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
451 extern long sched_group_rt_runtime(struct task_group *tg);
452 extern long sched_group_rt_period(struct task_group *tg);
453 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
455 extern struct task_group *sched_create_group(struct task_group *parent);
456 extern void sched_online_group(struct task_group *tg,
457 struct task_group *parent);
458 extern void sched_destroy_group(struct task_group *tg);
459 extern void sched_offline_group(struct task_group *tg);
461 extern void sched_move_task(struct task_struct *tsk);
463 #ifdef CONFIG_FAIR_GROUP_SCHED
464 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
467 extern void set_task_rq_fair(struct sched_entity *se,
468 struct cfs_rq *prev, struct cfs_rq *next);
469 #else /* !CONFIG_SMP */
470 static inline void set_task_rq_fair(struct sched_entity *se,
471 struct cfs_rq *prev, struct cfs_rq *next) { }
472 #endif /* CONFIG_SMP */
473 #endif /* CONFIG_FAIR_GROUP_SCHED */
475 #else /* CONFIG_CGROUP_SCHED */
477 struct cfs_bandwidth { };
479 #endif /* CONFIG_CGROUP_SCHED */
481 /* CFS-related fields in a runqueue */
483 struct load_weight load;
484 unsigned long runnable_weight;
485 unsigned int nr_running;
486 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
487 unsigned int idle_h_nr_running; /* SCHED_IDLE */
492 u64 min_vruntime_copy;
495 struct rb_root_cached tasks_timeline;
498 * 'curr' points to currently running entity on this cfs_rq.
499 * It is set to NULL otherwise (i.e when none are currently running).
501 struct sched_entity *curr;
502 struct sched_entity *next;
503 struct sched_entity *last;
504 struct sched_entity *skip;
506 #ifdef CONFIG_SCHED_DEBUG
507 unsigned int nr_spread_over;
514 struct sched_avg avg;
516 u64 load_last_update_time_copy;
519 raw_spinlock_t lock ____cacheline_aligned;
521 unsigned long load_avg;
522 unsigned long util_avg;
523 unsigned long runnable_sum;
526 #ifdef CONFIG_FAIR_GROUP_SCHED
527 unsigned long tg_load_avg_contrib;
529 long prop_runnable_sum;
532 * h_load = weight * f(tg)
534 * Where f(tg) is the recursive weight fraction assigned to
537 unsigned long h_load;
538 u64 last_h_load_update;
539 struct sched_entity *h_load_next;
540 #endif /* CONFIG_FAIR_GROUP_SCHED */
541 #endif /* CONFIG_SMP */
543 #ifdef CONFIG_FAIR_GROUP_SCHED
544 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
547 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
548 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
549 * (like users, containers etc.)
551 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
552 * This list is used during load balance.
555 struct list_head leaf_cfs_rq_list;
556 struct task_group *tg; /* group that "owns" this runqueue */
558 #ifdef CONFIG_CFS_BANDWIDTH
562 s64 runtime_remaining;
565 u64 throttled_clock_task;
566 u64 throttled_clock_task_time;
569 struct list_head throttled_list;
570 #endif /* CONFIG_CFS_BANDWIDTH */
571 #endif /* CONFIG_FAIR_GROUP_SCHED */
574 static inline int rt_bandwidth_enabled(void)
576 return sysctl_sched_rt_runtime >= 0;
579 /* RT IPI pull logic requires IRQ_WORK */
580 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
581 # define HAVE_RT_PUSH_IPI
584 /* Real-Time classes' related field in a runqueue: */
586 struct rt_prio_array active;
587 unsigned int rt_nr_running;
588 unsigned int rr_nr_running;
589 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
591 int curr; /* highest queued rt task prio */
593 int next; /* next highest */
598 unsigned long rt_nr_migratory;
599 unsigned long rt_nr_total;
601 struct plist_head pushable_tasks;
603 #endif /* CONFIG_SMP */
609 /* Nests inside the rq lock: */
610 raw_spinlock_t rt_runtime_lock;
612 #ifdef CONFIG_RT_GROUP_SCHED
613 unsigned long rt_nr_boosted;
616 struct task_group *tg;
620 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
622 return rt_rq->rt_queued && rt_rq->rt_nr_running;
625 /* Deadline class' related fields in a runqueue */
627 /* runqueue is an rbtree, ordered by deadline */
628 struct rb_root_cached root;
630 unsigned long dl_nr_running;
634 * Deadline values of the currently executing and the
635 * earliest ready task on this rq. Caching these facilitates
636 * the decision whether or not a ready but not running task
637 * should migrate somewhere else.
644 unsigned long dl_nr_migratory;
648 * Tasks on this rq that can be pushed away. They are kept in
649 * an rb-tree, ordered by tasks' deadlines, with caching
650 * of the leftmost (earliest deadline) element.
652 struct rb_root_cached pushable_dl_tasks_root;
657 * "Active utilization" for this runqueue: increased when a
658 * task wakes up (becomes TASK_RUNNING) and decreased when a
664 * Utilization of the tasks "assigned" to this runqueue (including
665 * the tasks that are in runqueue and the tasks that executed on this
666 * CPU and blocked). Increased when a task moves to this runqueue, and
667 * decreased when the task moves away (migrates, changes scheduling
668 * policy, or terminates).
669 * This is needed to compute the "inactive utilization" for the
670 * runqueue (inactive utilization = this_bw - running_bw).
676 * Inverse of the fraction of CPU utilization that can be reclaimed
677 * by the GRUB algorithm.
682 #ifdef CONFIG_FAIR_GROUP_SCHED
683 /* An entity is a task if it doesn't "own" a runqueue */
684 #define entity_is_task(se) (!se->my_q)
686 #define entity_is_task(se) 1
691 * XXX we want to get rid of these helpers and use the full load resolution.
693 static inline long se_weight(struct sched_entity *se)
695 return scale_load_down(se->load.weight);
698 static inline long se_runnable(struct sched_entity *se)
700 return scale_load_down(se->runnable_weight);
703 static inline bool sched_asym_prefer(int a, int b)
705 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
709 struct em_perf_domain *em_pd;
710 struct perf_domain *next;
714 /* Scheduling group status flags */
715 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
716 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
719 * We add the notion of a root-domain which will be used to define per-domain
720 * variables. Each exclusive cpuset essentially defines an island domain by
721 * fully partitioning the member CPUs from any other cpuset. Whenever a new
722 * exclusive cpuset is created, we also create and attach a new root-domain
731 cpumask_var_t online;
734 * Indicate pullable load on at least one CPU, e.g:
735 * - More than one runnable task
736 * - Running task is misfit
740 /* Indicate one or more cpus over-utilized (tipping point) */
744 * The bit corresponding to a CPU gets set here if such CPU has more
745 * than one runnable -deadline task (as it is below for RT tasks).
747 cpumask_var_t dlo_mask;
752 #ifdef HAVE_RT_PUSH_IPI
754 * For IPI pull requests, loop across the rto_mask.
756 struct irq_work rto_push_work;
757 raw_spinlock_t rto_lock;
758 /* These are only updated and read within rto_lock */
761 /* These atomics are updated outside of a lock */
762 atomic_t rto_loop_next;
763 atomic_t rto_loop_start;
766 * The "RT overload" flag: it gets set if a CPU has more than
767 * one runnable RT task.
769 cpumask_var_t rto_mask;
770 struct cpupri cpupri;
772 unsigned long max_cpu_capacity;
775 * NULL-terminated list of performance domains intersecting with the
776 * CPUs of the rd. Protected by RCU.
778 struct perf_domain __rcu *pd;
781 extern struct root_domain def_root_domain;
782 extern struct mutex sched_domains_mutex;
784 extern void init_defrootdomain(void);
785 extern int sched_init_domains(const struct cpumask *cpu_map);
786 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
787 extern void sched_get_rd(struct root_domain *rd);
788 extern void sched_put_rd(struct root_domain *rd);
790 #ifdef HAVE_RT_PUSH_IPI
791 extern void rto_push_irq_work_func(struct irq_work *work);
793 #endif /* CONFIG_SMP */
795 #ifdef CONFIG_UCLAMP_TASK
797 * struct uclamp_bucket - Utilization clamp bucket
798 * @value: utilization clamp value for tasks on this clamp bucket
799 * @tasks: number of RUNNABLE tasks on this clamp bucket
801 * Keep track of how many tasks are RUNNABLE for a given utilization
804 struct uclamp_bucket {
805 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
806 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
810 * struct uclamp_rq - rq's utilization clamp
811 * @value: currently active clamp values for a rq
812 * @bucket: utilization clamp buckets affecting a rq
814 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
815 * A clamp value is affecting a rq when there is at least one task RUNNABLE
816 * (or actually running) with that value.
818 * There are up to UCLAMP_CNT possible different clamp values, currently there
819 * are only two: minimum utilization and maximum utilization.
821 * All utilization clamping values are MAX aggregated, since:
822 * - for util_min: we want to run the CPU at least at the max of the minimum
823 * utilization required by its currently RUNNABLE tasks.
824 * - for util_max: we want to allow the CPU to run up to the max of the
825 * maximum utilization allowed by its currently RUNNABLE tasks.
827 * Since on each system we expect only a limited number of different
828 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
829 * the metrics required to compute all the per-rq utilization clamp values.
833 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
835 #endif /* CONFIG_UCLAMP_TASK */
838 * This is the main, per-CPU runqueue data structure.
840 * Locking rule: those places that want to lock multiple runqueues
841 * (such as the load balancing or the thread migration code), lock
842 * acquire operations must be ordered by ascending &runqueue.
849 * nr_running and cpu_load should be in the same cacheline because
850 * remote CPUs use both these fields when doing load calculation.
852 unsigned int nr_running;
853 #ifdef CONFIG_NUMA_BALANCING
854 unsigned int nr_numa_running;
855 unsigned int nr_preferred_running;
856 unsigned int numa_migrate_on;
858 #ifdef CONFIG_NO_HZ_COMMON
860 unsigned long last_load_update_tick;
861 unsigned long last_blocked_load_update_tick;
862 unsigned int has_blocked_load;
863 #endif /* CONFIG_SMP */
864 unsigned int nohz_tick_stopped;
866 #endif /* CONFIG_NO_HZ_COMMON */
868 unsigned long nr_load_updates;
871 #ifdef CONFIG_UCLAMP_TASK
872 /* Utilization clamp values based on CPU's RUNNABLE tasks */
873 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
874 unsigned int uclamp_flags;
875 #define UCLAMP_FLAG_IDLE 0x01
882 #ifdef CONFIG_FAIR_GROUP_SCHED
883 /* list of leaf cfs_rq on this CPU: */
884 struct list_head leaf_cfs_rq_list;
885 struct list_head *tmp_alone_branch;
886 #endif /* CONFIG_FAIR_GROUP_SCHED */
889 * This is part of a global counter where only the total sum
890 * over all CPUs matters. A task can increase this counter on
891 * one CPU and if it got migrated afterwards it may decrease
892 * it on another CPU. Always updated under the runqueue lock:
894 unsigned long nr_uninterruptible;
896 struct task_struct *curr;
897 struct task_struct *idle;
898 struct task_struct *stop;
899 unsigned long next_balance;
900 struct mm_struct *prev_mm;
902 unsigned int clock_update_flags;
904 /* Ensure that all clocks are in the same cache line */
905 u64 clock_task ____cacheline_aligned;
907 unsigned long lost_idle_time;
912 struct root_domain *rd;
913 struct sched_domain __rcu *sd;
915 unsigned long cpu_capacity;
916 unsigned long cpu_capacity_orig;
918 struct callback_head *balance_callback;
920 unsigned char idle_balance;
922 unsigned long misfit_task_load;
924 /* For active balancing */
927 struct cpu_stop_work active_balance_work;
929 /* CPU of this runqueue: */
933 struct list_head cfs_tasks;
935 struct sched_avg avg_rt;
936 struct sched_avg avg_dl;
937 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
938 struct sched_avg avg_irq;
943 /* This is used to determine avg_idle's max value */
944 u64 max_idle_balance_cost;
947 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
950 #ifdef CONFIG_PARAVIRT
953 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
954 u64 prev_steal_time_rq;
957 /* calc_load related fields */
958 unsigned long calc_load_update;
959 long calc_load_active;
961 #ifdef CONFIG_SCHED_HRTICK
963 int hrtick_csd_pending;
964 call_single_data_t hrtick_csd;
966 struct hrtimer hrtick_timer;
969 #ifdef CONFIG_SCHEDSTATS
971 struct sched_info rq_sched_info;
972 unsigned long long rq_cpu_time;
973 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
975 /* sys_sched_yield() stats */
976 unsigned int yld_count;
978 /* schedule() stats */
979 unsigned int sched_count;
980 unsigned int sched_goidle;
982 /* try_to_wake_up() stats */
983 unsigned int ttwu_count;
984 unsigned int ttwu_local;
988 struct llist_head wake_list;
991 #ifdef CONFIG_CPU_IDLE
992 /* Must be inspected within a rcu lock section */
993 struct cpuidle_state *idle_state;
997 #ifdef CONFIG_FAIR_GROUP_SCHED
999 /* CPU runqueue to which this cfs_rq is attached */
1000 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1007 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1009 return container_of(cfs_rq, struct rq, cfs);
1013 static inline int cpu_of(struct rq *rq)
1023 #ifdef CONFIG_SCHED_SMT
1024 extern void __update_idle_core(struct rq *rq);
1026 static inline void update_idle_core(struct rq *rq)
1028 if (static_branch_unlikely(&sched_smt_present))
1029 __update_idle_core(rq);
1033 static inline void update_idle_core(struct rq *rq) { }
1036 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1038 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1039 #define this_rq() this_cpu_ptr(&runqueues)
1040 #define task_rq(p) cpu_rq(task_cpu(p))
1041 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1042 #define raw_rq() raw_cpu_ptr(&runqueues)
1044 extern void update_rq_clock(struct rq *rq);
1046 static inline u64 __rq_clock_broken(struct rq *rq)
1048 return READ_ONCE(rq->clock);
1052 * rq::clock_update_flags bits
1054 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1055 * call to __schedule(). This is an optimisation to avoid
1056 * neighbouring rq clock updates.
1058 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1059 * in effect and calls to update_rq_clock() are being ignored.
1061 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1062 * made to update_rq_clock() since the last time rq::lock was pinned.
1064 * If inside of __schedule(), clock_update_flags will have been
1065 * shifted left (a left shift is a cheap operation for the fast path
1066 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1068 * if (rq-clock_update_flags >= RQCF_UPDATED)
1070 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1071 * one position though, because the next rq_unpin_lock() will shift it
1074 #define RQCF_REQ_SKIP 0x01
1075 #define RQCF_ACT_SKIP 0x02
1076 #define RQCF_UPDATED 0x04
1078 static inline void assert_clock_updated(struct rq *rq)
1081 * The only reason for not seeing a clock update since the
1082 * last rq_pin_lock() is if we're currently skipping updates.
1084 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1087 static inline u64 rq_clock(struct rq *rq)
1089 lockdep_assert_held(&rq->lock);
1090 assert_clock_updated(rq);
1095 static inline u64 rq_clock_task(struct rq *rq)
1097 lockdep_assert_held(&rq->lock);
1098 assert_clock_updated(rq);
1100 return rq->clock_task;
1103 static inline void rq_clock_skip_update(struct rq *rq)
1105 lockdep_assert_held(&rq->lock);
1106 rq->clock_update_flags |= RQCF_REQ_SKIP;
1110 * See rt task throttling, which is the only time a skip
1111 * request is cancelled.
1113 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1115 lockdep_assert_held(&rq->lock);
1116 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1120 unsigned long flags;
1121 struct pin_cookie cookie;
1122 #ifdef CONFIG_SCHED_DEBUG
1124 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1125 * current pin context is stashed here in case it needs to be
1126 * restored in rq_repin_lock().
1128 unsigned int clock_update_flags;
1132 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1134 rf->cookie = lockdep_pin_lock(&rq->lock);
1136 #ifdef CONFIG_SCHED_DEBUG
1137 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1138 rf->clock_update_flags = 0;
1142 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1144 #ifdef CONFIG_SCHED_DEBUG
1145 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1146 rf->clock_update_flags = RQCF_UPDATED;
1149 lockdep_unpin_lock(&rq->lock, rf->cookie);
1152 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1154 lockdep_repin_lock(&rq->lock, rf->cookie);
1156 #ifdef CONFIG_SCHED_DEBUG
1158 * Restore the value we stashed in @rf for this pin context.
1160 rq->clock_update_flags |= rf->clock_update_flags;
1164 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1165 __acquires(rq->lock);
1167 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1168 __acquires(p->pi_lock)
1169 __acquires(rq->lock);
1171 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1172 __releases(rq->lock)
1174 rq_unpin_lock(rq, rf);
1175 raw_spin_unlock(&rq->lock);
1179 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1180 __releases(rq->lock)
1181 __releases(p->pi_lock)
1183 rq_unpin_lock(rq, rf);
1184 raw_spin_unlock(&rq->lock);
1185 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1189 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1190 __acquires(rq->lock)
1192 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1193 rq_pin_lock(rq, rf);
1197 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1198 __acquires(rq->lock)
1200 raw_spin_lock_irq(&rq->lock);
1201 rq_pin_lock(rq, rf);
1205 rq_lock(struct rq *rq, struct rq_flags *rf)
1206 __acquires(rq->lock)
1208 raw_spin_lock(&rq->lock);
1209 rq_pin_lock(rq, rf);
1213 rq_relock(struct rq *rq, struct rq_flags *rf)
1214 __acquires(rq->lock)
1216 raw_spin_lock(&rq->lock);
1217 rq_repin_lock(rq, rf);
1221 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1222 __releases(rq->lock)
1224 rq_unpin_lock(rq, rf);
1225 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1229 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1230 __releases(rq->lock)
1232 rq_unpin_lock(rq, rf);
1233 raw_spin_unlock_irq(&rq->lock);
1237 rq_unlock(struct rq *rq, struct rq_flags *rf)
1238 __releases(rq->lock)
1240 rq_unpin_lock(rq, rf);
1241 raw_spin_unlock(&rq->lock);
1244 static inline struct rq *
1245 this_rq_lock_irq(struct rq_flags *rf)
1246 __acquires(rq->lock)
1250 local_irq_disable();
1257 enum numa_topology_type {
1262 extern enum numa_topology_type sched_numa_topology_type;
1263 extern int sched_max_numa_distance;
1264 extern bool find_numa_distance(int distance);
1265 extern void sched_init_numa(void);
1266 extern void sched_domains_numa_masks_set(unsigned int cpu);
1267 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1268 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1270 static inline void sched_init_numa(void) { }
1271 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1272 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1273 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1279 #ifdef CONFIG_NUMA_BALANCING
1280 /* The regions in numa_faults array from task_struct */
1281 enum numa_faults_stats {
1287 extern void sched_setnuma(struct task_struct *p, int node);
1288 extern int migrate_task_to(struct task_struct *p, int cpu);
1289 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1291 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1294 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1297 #endif /* CONFIG_NUMA_BALANCING */
1302 queue_balance_callback(struct rq *rq,
1303 struct callback_head *head,
1304 void (*func)(struct rq *rq))
1306 lockdep_assert_held(&rq->lock);
1308 if (unlikely(head->next))
1311 head->func = (void (*)(struct callback_head *))func;
1312 head->next = rq->balance_callback;
1313 rq->balance_callback = head;
1316 extern void sched_ttwu_pending(void);
1318 #define rcu_dereference_check_sched_domain(p) \
1319 rcu_dereference_check((p), \
1320 lockdep_is_held(&sched_domains_mutex))
1323 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1324 * See destroy_sched_domains: call_rcu for details.
1326 * The domain tree of any CPU may only be accessed from within
1327 * preempt-disabled sections.
1329 #define for_each_domain(cpu, __sd) \
1330 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1331 __sd; __sd = __sd->parent)
1333 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1336 * highest_flag_domain - Return highest sched_domain containing flag.
1337 * @cpu: The CPU whose highest level of sched domain is to
1339 * @flag: The flag to check for the highest sched_domain
1340 * for the given CPU.
1342 * Returns the highest sched_domain of a CPU which contains the given flag.
1344 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1346 struct sched_domain *sd, *hsd = NULL;
1348 for_each_domain(cpu, sd) {
1349 if (!(sd->flags & flag))
1357 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1359 struct sched_domain *sd;
1361 for_each_domain(cpu, sd) {
1362 if (sd->flags & flag)
1369 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1370 DECLARE_PER_CPU(int, sd_llc_size);
1371 DECLARE_PER_CPU(int, sd_llc_id);
1372 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1373 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1374 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1375 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1376 extern struct static_key_false sched_asym_cpucapacity;
1378 struct sched_group_capacity {
1381 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1384 unsigned long capacity;
1385 unsigned long min_capacity; /* Min per-CPU capacity in group */
1386 unsigned long max_capacity; /* Max per-CPU capacity in group */
1387 unsigned long next_update;
1388 int imbalance; /* XXX unrelated to capacity but shared group state */
1390 #ifdef CONFIG_SCHED_DEBUG
1394 unsigned long cpumask[0]; /* Balance mask */
1397 struct sched_group {
1398 struct sched_group *next; /* Must be a circular list */
1401 unsigned int group_weight;
1402 struct sched_group_capacity *sgc;
1403 int asym_prefer_cpu; /* CPU of highest priority in group */
1406 * The CPUs this group covers.
1408 * NOTE: this field is variable length. (Allocated dynamically
1409 * by attaching extra space to the end of the structure,
1410 * depending on how many CPUs the kernel has booted up with)
1412 unsigned long cpumask[0];
1415 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1417 return to_cpumask(sg->cpumask);
1421 * See build_balance_mask().
1423 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1425 return to_cpumask(sg->sgc->cpumask);
1429 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1430 * @group: The group whose first CPU is to be returned.
1432 static inline unsigned int group_first_cpu(struct sched_group *group)
1434 return cpumask_first(sched_group_span(group));
1437 extern int group_balance_cpu(struct sched_group *sg);
1439 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1440 void register_sched_domain_sysctl(void);
1441 void dirty_sched_domain_sysctl(int cpu);
1442 void unregister_sched_domain_sysctl(void);
1444 static inline void register_sched_domain_sysctl(void)
1447 static inline void dirty_sched_domain_sysctl(int cpu)
1450 static inline void unregister_sched_domain_sysctl(void)
1457 static inline void sched_ttwu_pending(void) { }
1459 #endif /* CONFIG_SMP */
1462 #include "autogroup.h"
1464 #ifdef CONFIG_CGROUP_SCHED
1467 * Return the group to which this tasks belongs.
1469 * We cannot use task_css() and friends because the cgroup subsystem
1470 * changes that value before the cgroup_subsys::attach() method is called,
1471 * therefore we cannot pin it and might observe the wrong value.
1473 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1474 * core changes this before calling sched_move_task().
1476 * Instead we use a 'copy' which is updated from sched_move_task() while
1477 * holding both task_struct::pi_lock and rq::lock.
1479 static inline struct task_group *task_group(struct task_struct *p)
1481 return p->sched_task_group;
1484 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1485 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1487 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1488 struct task_group *tg = task_group(p);
1491 #ifdef CONFIG_FAIR_GROUP_SCHED
1492 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1493 p->se.cfs_rq = tg->cfs_rq[cpu];
1494 p->se.parent = tg->se[cpu];
1497 #ifdef CONFIG_RT_GROUP_SCHED
1498 p->rt.rt_rq = tg->rt_rq[cpu];
1499 p->rt.parent = tg->rt_se[cpu];
1503 #else /* CONFIG_CGROUP_SCHED */
1505 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1506 static inline struct task_group *task_group(struct task_struct *p)
1511 #endif /* CONFIG_CGROUP_SCHED */
1513 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1515 set_task_rq(p, cpu);
1518 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1519 * successfully executed on another CPU. We must ensure that updates of
1520 * per-task data have been completed by this moment.
1523 #ifdef CONFIG_THREAD_INFO_IN_TASK
1524 WRITE_ONCE(p->cpu, cpu);
1526 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1533 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1535 #ifdef CONFIG_SCHED_DEBUG
1536 # include <linux/static_key.h>
1537 # define const_debug __read_mostly
1539 # define const_debug const
1542 #define SCHED_FEAT(name, enabled) \
1543 __SCHED_FEAT_##name ,
1546 #include "features.h"
1552 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1555 * To support run-time toggling of sched features, all the translation units
1556 * (but core.c) reference the sysctl_sched_features defined in core.c.
1558 extern const_debug unsigned int sysctl_sched_features;
1560 #define SCHED_FEAT(name, enabled) \
1561 static __always_inline bool static_branch_##name(struct static_key *key) \
1563 return static_key_##enabled(key); \
1566 #include "features.h"
1569 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1570 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1572 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1575 * Each translation unit has its own copy of sysctl_sched_features to allow
1576 * constants propagation at compile time and compiler optimization based on
1579 #define SCHED_FEAT(name, enabled) \
1580 (1UL << __SCHED_FEAT_##name) * enabled |
1581 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1582 #include "features.h"
1586 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1588 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1590 extern struct static_key_false sched_numa_balancing;
1591 extern struct static_key_false sched_schedstats;
1593 static inline u64 global_rt_period(void)
1595 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1598 static inline u64 global_rt_runtime(void)
1600 if (sysctl_sched_rt_runtime < 0)
1603 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1606 static inline int task_current(struct rq *rq, struct task_struct *p)
1608 return rq->curr == p;
1611 static inline int task_running(struct rq *rq, struct task_struct *p)
1616 return task_current(rq, p);
1620 static inline int task_on_rq_queued(struct task_struct *p)
1622 return p->on_rq == TASK_ON_RQ_QUEUED;
1625 static inline int task_on_rq_migrating(struct task_struct *p)
1627 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1633 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1634 #define WF_FORK 0x02 /* Child wakeup after fork */
1635 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1638 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1639 * of tasks with abnormal "nice" values across CPUs the contribution that
1640 * each task makes to its run queue's load is weighted according to its
1641 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1642 * scaled version of the new time slice allocation that they receive on time
1646 #define WEIGHT_IDLEPRIO 3
1647 #define WMULT_IDLEPRIO 1431655765
1649 extern const int sched_prio_to_weight[40];
1650 extern const u32 sched_prio_to_wmult[40];
1653 * {de,en}queue flags:
1655 * DEQUEUE_SLEEP - task is no longer runnable
1656 * ENQUEUE_WAKEUP - task just became runnable
1658 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1659 * are in a known state which allows modification. Such pairs
1660 * should preserve as much state as possible.
1662 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1665 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1666 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1667 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1671 #define DEQUEUE_SLEEP 0x01
1672 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1673 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1674 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1676 #define ENQUEUE_WAKEUP 0x01
1677 #define ENQUEUE_RESTORE 0x02
1678 #define ENQUEUE_MOVE 0x04
1679 #define ENQUEUE_NOCLOCK 0x08
1681 #define ENQUEUE_HEAD 0x10
1682 #define ENQUEUE_REPLENISH 0x20
1684 #define ENQUEUE_MIGRATED 0x40
1686 #define ENQUEUE_MIGRATED 0x00
1689 #define RETRY_TASK ((void *)-1UL)
1691 struct sched_class {
1692 const struct sched_class *next;
1694 #ifdef CONFIG_UCLAMP_TASK
1698 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1699 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1700 void (*yield_task) (struct rq *rq);
1701 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1703 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1706 * It is the responsibility of the pick_next_task() method that will
1707 * return the next task to call put_prev_task() on the @prev task or
1708 * something equivalent.
1710 * May return RETRY_TASK when it finds a higher prio class has runnable
1713 struct task_struct * (*pick_next_task)(struct rq *rq,
1714 struct task_struct *prev,
1715 struct rq_flags *rf);
1716 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1719 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1720 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1722 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1724 void (*set_cpus_allowed)(struct task_struct *p,
1725 const struct cpumask *newmask);
1727 void (*rq_online)(struct rq *rq);
1728 void (*rq_offline)(struct rq *rq);
1731 void (*set_curr_task)(struct rq *rq);
1732 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1733 void (*task_fork)(struct task_struct *p);
1734 void (*task_dead)(struct task_struct *p);
1737 * The switched_from() call is allowed to drop rq->lock, therefore we
1738 * cannot assume the switched_from/switched_to pair is serliazed by
1739 * rq->lock. They are however serialized by p->pi_lock.
1741 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1742 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1743 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1746 unsigned int (*get_rr_interval)(struct rq *rq,
1747 struct task_struct *task);
1749 void (*update_curr)(struct rq *rq);
1751 #define TASK_SET_GROUP 0
1752 #define TASK_MOVE_GROUP 1
1754 #ifdef CONFIG_FAIR_GROUP_SCHED
1755 void (*task_change_group)(struct task_struct *p, int type);
1759 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1761 prev->sched_class->put_prev_task(rq, prev);
1764 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1766 curr->sched_class->set_curr_task(rq);
1770 #define sched_class_highest (&stop_sched_class)
1772 #define sched_class_highest (&dl_sched_class)
1774 #define for_each_class(class) \
1775 for (class = sched_class_highest; class; class = class->next)
1777 extern const struct sched_class stop_sched_class;
1778 extern const struct sched_class dl_sched_class;
1779 extern const struct sched_class rt_sched_class;
1780 extern const struct sched_class fair_sched_class;
1781 extern const struct sched_class idle_sched_class;
1786 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1788 extern void trigger_load_balance(struct rq *rq);
1790 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1794 #ifdef CONFIG_CPU_IDLE
1795 static inline void idle_set_state(struct rq *rq,
1796 struct cpuidle_state *idle_state)
1798 rq->idle_state = idle_state;
1801 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1803 SCHED_WARN_ON(!rcu_read_lock_held());
1805 return rq->idle_state;
1808 static inline void idle_set_state(struct rq *rq,
1809 struct cpuidle_state *idle_state)
1813 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1819 extern void schedule_idle(void);
1821 extern void sysrq_sched_debug_show(void);
1822 extern void sched_init_granularity(void);
1823 extern void update_max_interval(void);
1825 extern void init_sched_dl_class(void);
1826 extern void init_sched_rt_class(void);
1827 extern void init_sched_fair_class(void);
1829 extern void reweight_task(struct task_struct *p, int prio);
1831 extern void resched_curr(struct rq *rq);
1832 extern void resched_cpu(int cpu);
1834 extern struct rt_bandwidth def_rt_bandwidth;
1835 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1837 extern struct dl_bandwidth def_dl_bandwidth;
1838 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1839 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1840 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1841 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1844 #define BW_UNIT (1 << BW_SHIFT)
1845 #define RATIO_SHIFT 8
1846 unsigned long to_ratio(u64 period, u64 runtime);
1848 extern void init_entity_runnable_average(struct sched_entity *se);
1849 extern void post_init_entity_util_avg(struct task_struct *p);
1851 #ifdef CONFIG_NO_HZ_FULL
1852 extern bool sched_can_stop_tick(struct rq *rq);
1853 extern int __init sched_tick_offload_init(void);
1856 * Tick may be needed by tasks in the runqueue depending on their policy and
1857 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1858 * nohz mode if necessary.
1860 static inline void sched_update_tick_dependency(struct rq *rq)
1864 if (!tick_nohz_full_enabled())
1869 if (!tick_nohz_full_cpu(cpu))
1872 if (sched_can_stop_tick(rq))
1873 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1875 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1878 static inline int sched_tick_offload_init(void) { return 0; }
1879 static inline void sched_update_tick_dependency(struct rq *rq) { }
1882 static inline void add_nr_running(struct rq *rq, unsigned count)
1884 unsigned prev_nr = rq->nr_running;
1886 rq->nr_running = prev_nr + count;
1889 if (prev_nr < 2 && rq->nr_running >= 2) {
1890 if (!READ_ONCE(rq->rd->overload))
1891 WRITE_ONCE(rq->rd->overload, 1);
1895 sched_update_tick_dependency(rq);
1898 static inline void sub_nr_running(struct rq *rq, unsigned count)
1900 rq->nr_running -= count;
1901 /* Check if we still need preemption */
1902 sched_update_tick_dependency(rq);
1905 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1906 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1908 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1910 extern const_debug unsigned int sysctl_sched_nr_migrate;
1911 extern const_debug unsigned int sysctl_sched_migration_cost;
1913 #ifdef CONFIG_SCHED_HRTICK
1917 * - enabled by features
1918 * - hrtimer is actually high res
1920 static inline int hrtick_enabled(struct rq *rq)
1922 if (!sched_feat(HRTICK))
1924 if (!cpu_active(cpu_of(rq)))
1926 return hrtimer_is_hres_active(&rq->hrtick_timer);
1929 void hrtick_start(struct rq *rq, u64 delay);
1933 static inline int hrtick_enabled(struct rq *rq)
1938 #endif /* CONFIG_SCHED_HRTICK */
1940 #ifndef arch_scale_freq_capacity
1941 static __always_inline
1942 unsigned long arch_scale_freq_capacity(int cpu)
1944 return SCHED_CAPACITY_SCALE;
1949 #ifdef CONFIG_PREEMPT
1951 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1954 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1955 * way at the expense of forcing extra atomic operations in all
1956 * invocations. This assures that the double_lock is acquired using the
1957 * same underlying policy as the spinlock_t on this architecture, which
1958 * reduces latency compared to the unfair variant below. However, it
1959 * also adds more overhead and therefore may reduce throughput.
1961 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1962 __releases(this_rq->lock)
1963 __acquires(busiest->lock)
1964 __acquires(this_rq->lock)
1966 raw_spin_unlock(&this_rq->lock);
1967 double_rq_lock(this_rq, busiest);
1974 * Unfair double_lock_balance: Optimizes throughput at the expense of
1975 * latency by eliminating extra atomic operations when the locks are
1976 * already in proper order on entry. This favors lower CPU-ids and will
1977 * grant the double lock to lower CPUs over higher ids under contention,
1978 * regardless of entry order into the function.
1980 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1981 __releases(this_rq->lock)
1982 __acquires(busiest->lock)
1983 __acquires(this_rq->lock)
1987 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1988 if (busiest < this_rq) {
1989 raw_spin_unlock(&this_rq->lock);
1990 raw_spin_lock(&busiest->lock);
1991 raw_spin_lock_nested(&this_rq->lock,
1992 SINGLE_DEPTH_NESTING);
1995 raw_spin_lock_nested(&busiest->lock,
1996 SINGLE_DEPTH_NESTING);
2001 #endif /* CONFIG_PREEMPT */
2004 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2006 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2008 if (unlikely(!irqs_disabled())) {
2009 /* printk() doesn't work well under rq->lock */
2010 raw_spin_unlock(&this_rq->lock);
2014 return _double_lock_balance(this_rq, busiest);
2017 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2018 __releases(busiest->lock)
2020 raw_spin_unlock(&busiest->lock);
2021 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2024 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2030 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2033 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2039 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2042 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2048 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2052 * double_rq_lock - safely lock two runqueues
2054 * Note this does not disable interrupts like task_rq_lock,
2055 * you need to do so manually before calling.
2057 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2058 __acquires(rq1->lock)
2059 __acquires(rq2->lock)
2061 BUG_ON(!irqs_disabled());
2063 raw_spin_lock(&rq1->lock);
2064 __acquire(rq2->lock); /* Fake it out ;) */
2067 raw_spin_lock(&rq1->lock);
2068 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2070 raw_spin_lock(&rq2->lock);
2071 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2077 * double_rq_unlock - safely unlock two runqueues
2079 * Note this does not restore interrupts like task_rq_unlock,
2080 * you need to do so manually after calling.
2082 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2083 __releases(rq1->lock)
2084 __releases(rq2->lock)
2086 raw_spin_unlock(&rq1->lock);
2088 raw_spin_unlock(&rq2->lock);
2090 __release(rq2->lock);
2093 extern void set_rq_online (struct rq *rq);
2094 extern void set_rq_offline(struct rq *rq);
2095 extern bool sched_smp_initialized;
2097 #else /* CONFIG_SMP */
2100 * double_rq_lock - safely lock two runqueues
2102 * Note this does not disable interrupts like task_rq_lock,
2103 * you need to do so manually before calling.
2105 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2106 __acquires(rq1->lock)
2107 __acquires(rq2->lock)
2109 BUG_ON(!irqs_disabled());
2111 raw_spin_lock(&rq1->lock);
2112 __acquire(rq2->lock); /* Fake it out ;) */
2116 * double_rq_unlock - safely unlock two runqueues
2118 * Note this does not restore interrupts like task_rq_unlock,
2119 * you need to do so manually after calling.
2121 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2122 __releases(rq1->lock)
2123 __releases(rq2->lock)
2126 raw_spin_unlock(&rq1->lock);
2127 __release(rq2->lock);
2132 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2133 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2135 #ifdef CONFIG_SCHED_DEBUG
2136 extern bool sched_debug_enabled;
2138 extern void print_cfs_stats(struct seq_file *m, int cpu);
2139 extern void print_rt_stats(struct seq_file *m, int cpu);
2140 extern void print_dl_stats(struct seq_file *m, int cpu);
2141 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2142 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2143 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2144 #ifdef CONFIG_NUMA_BALANCING
2146 show_numa_stats(struct task_struct *p, struct seq_file *m);
2148 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2149 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2150 #endif /* CONFIG_NUMA_BALANCING */
2151 #endif /* CONFIG_SCHED_DEBUG */
2153 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2154 extern void init_rt_rq(struct rt_rq *rt_rq);
2155 extern void init_dl_rq(struct dl_rq *dl_rq);
2157 extern void cfs_bandwidth_usage_inc(void);
2158 extern void cfs_bandwidth_usage_dec(void);
2160 #ifdef CONFIG_NO_HZ_COMMON
2161 #define NOHZ_BALANCE_KICK_BIT 0
2162 #define NOHZ_STATS_KICK_BIT 1
2164 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2165 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2167 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2169 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2171 extern void nohz_balance_exit_idle(struct rq *rq);
2173 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2179 void __dl_update(struct dl_bw *dl_b, s64 bw)
2181 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2184 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2185 "sched RCU must be held");
2186 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2187 struct rq *rq = cpu_rq(i);
2189 rq->dl.extra_bw += bw;
2194 void __dl_update(struct dl_bw *dl_b, s64 bw)
2196 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2203 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2208 struct u64_stats_sync sync;
2211 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2214 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2215 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2216 * and never move forward.
2218 static inline u64 irq_time_read(int cpu)
2220 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2225 seq = __u64_stats_fetch_begin(&irqtime->sync);
2226 total = irqtime->total;
2227 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2231 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2233 #ifdef CONFIG_CPU_FREQ
2234 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2237 * cpufreq_update_util - Take a note about CPU utilization changes.
2238 * @rq: Runqueue to carry out the update for.
2239 * @flags: Update reason flags.
2241 * This function is called by the scheduler on the CPU whose utilization is
2244 * It can only be called from RCU-sched read-side critical sections.
2246 * The way cpufreq is currently arranged requires it to evaluate the CPU
2247 * performance state (frequency/voltage) on a regular basis to prevent it from
2248 * being stuck in a completely inadequate performance level for too long.
2249 * That is not guaranteed to happen if the updates are only triggered from CFS
2250 * and DL, though, because they may not be coming in if only RT tasks are
2251 * active all the time (or there are RT tasks only).
2253 * As a workaround for that issue, this function is called periodically by the
2254 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2255 * but that really is a band-aid. Going forward it should be replaced with
2256 * solutions targeted more specifically at RT tasks.
2258 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2260 struct update_util_data *data;
2262 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2265 data->func(data, rq_clock(rq), flags);
2268 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2269 #endif /* CONFIG_CPU_FREQ */
2271 #ifdef CONFIG_UCLAMP_TASK
2272 unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id);
2274 static __always_inline
2275 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2276 struct task_struct *p)
2278 unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2279 unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2282 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2283 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2287 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2288 * RUNNABLE tasks with _different_ clamps, we can end up with an
2289 * inversion. Fix it now when the clamps are applied.
2291 if (unlikely(min_util >= max_util))
2294 return clamp(util, min_util, max_util);
2297 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2299 return uclamp_util_with(rq, util, NULL);
2301 #else /* CONFIG_UCLAMP_TASK */
2302 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2303 struct task_struct *p)
2307 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2311 #endif /* CONFIG_UCLAMP_TASK */
2313 #ifdef arch_scale_freq_capacity
2314 # ifndef arch_scale_freq_invariant
2315 # define arch_scale_freq_invariant() true
2318 # define arch_scale_freq_invariant() false
2322 static inline unsigned long capacity_orig_of(int cpu)
2324 return cpu_rq(cpu)->cpu_capacity_orig;
2329 * enum schedutil_type - CPU utilization type
2330 * @FREQUENCY_UTIL: Utilization used to select frequency
2331 * @ENERGY_UTIL: Utilization used during energy calculation
2333 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2334 * need to be aggregated differently depending on the usage made of them. This
2335 * enum is used within schedutil_freq_util() to differentiate the types of
2336 * utilization expected by the callers, and adjust the aggregation accordingly.
2338 enum schedutil_type {
2343 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2345 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2346 unsigned long max, enum schedutil_type type,
2347 struct task_struct *p);
2349 static inline unsigned long cpu_bw_dl(struct rq *rq)
2351 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2354 static inline unsigned long cpu_util_dl(struct rq *rq)
2356 return READ_ONCE(rq->avg_dl.util_avg);
2359 static inline unsigned long cpu_util_cfs(struct rq *rq)
2361 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2363 if (sched_feat(UTIL_EST)) {
2364 util = max_t(unsigned long, util,
2365 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2371 static inline unsigned long cpu_util_rt(struct rq *rq)
2373 return READ_ONCE(rq->avg_rt.util_avg);
2375 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2376 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2377 unsigned long max, enum schedutil_type type,
2378 struct task_struct *p)
2382 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2384 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2385 static inline unsigned long cpu_util_irq(struct rq *rq)
2387 return rq->avg_irq.util_avg;
2391 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2393 util *= (max - irq);
2400 static inline unsigned long cpu_util_irq(struct rq *rq)
2406 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2412 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2414 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2416 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2418 static inline bool sched_energy_enabled(void)
2420 return static_branch_unlikely(&sched_energy_present);
2423 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2425 #define perf_domain_span(pd) NULL
2426 static inline bool sched_energy_enabled(void) { return false; }
2428 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */