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/init_task.h>
49 #include <linux/kprobes.h>
50 #include <linux/kthread.h>
51 #include <linux/membarrier.h>
52 #include <linux/migrate.h>
53 #include <linux/mmu_context.h>
54 #include <linux/nmi.h>
55 #include <linux/proc_fs.h>
56 #include <linux/prefetch.h>
57 #include <linux/profile.h>
58 #include <linux/psi.h>
59 #include <linux/rcupdate_wait.h>
60 #include <linux/security.h>
61 #include <linux/stop_machine.h>
62 #include <linux/suspend.h>
63 #include <linux/swait.h>
64 #include <linux/syscalls.h>
65 #include <linux/task_work.h>
66 #include <linux/tsacct_kern.h>
70 #ifdef CONFIG_PARAVIRT
71 # include <asm/paravirt.h>
75 #include "cpudeadline.h"
77 #ifdef CONFIG_SCHED_DEBUG
78 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
80 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
86 /* task_struct::on_rq states: */
87 #define TASK_ON_RQ_QUEUED 1
88 #define TASK_ON_RQ_MIGRATING 2
90 extern __read_mostly int scheduler_running;
92 extern unsigned long calc_load_update;
93 extern atomic_long_t calc_load_tasks;
95 extern void calc_global_load_tick(struct rq *this_rq);
96 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
99 extern void cpu_load_update_active(struct rq *this_rq);
101 static inline void cpu_load_update_active(struct rq *this_rq) { }
105 * Helpers for converting nanosecond timing to jiffy resolution
107 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
116 * We increase resolution only if we have enough bits to allow this increased
117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
124 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
128 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
129 # define scale_load(w) (w)
130 # define scale_load_down(w) (w)
134 * Task weight (visible to users) and its load (invisible to users) have
135 * independent resolution, but they should be well calibrated. We use
136 * scale_load() and scale_load_down(w) to convert between them. The
137 * following must be true:
139 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
142 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
145 * Single value that decides SCHED_DEADLINE internal math precision.
146 * 10 -> just above 1us
147 * 9 -> just above 0.5us
152 * Single value that denotes runtime == period, ie unlimited time.
154 #define RUNTIME_INF ((u64)~0ULL)
156 static inline int idle_policy(int policy)
158 return policy == SCHED_IDLE;
160 static inline int fair_policy(int policy)
162 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
165 static inline int rt_policy(int policy)
167 return policy == SCHED_FIFO || policy == SCHED_RR;
170 static inline int dl_policy(int policy)
172 return policy == SCHED_DEADLINE;
174 static inline bool valid_policy(int policy)
176 return idle_policy(policy) || fair_policy(policy) ||
177 rt_policy(policy) || dl_policy(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 struct hrtimer period_timer;
344 struct hrtimer slack_timer;
345 struct list_head throttled_cfs_rq;
352 bool distribute_running;
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;
491 u64 min_vruntime_copy;
494 struct rb_root_cached tasks_timeline;
497 * 'curr' points to currently running entity on this cfs_rq.
498 * It is set to NULL otherwise (i.e when none are currently running).
500 struct sched_entity *curr;
501 struct sched_entity *next;
502 struct sched_entity *last;
503 struct sched_entity *skip;
505 #ifdef CONFIG_SCHED_DEBUG
506 unsigned int nr_spread_over;
513 struct sched_avg avg;
515 u64 load_last_update_time_copy;
518 raw_spinlock_t lock ____cacheline_aligned;
520 unsigned long load_avg;
521 unsigned long util_avg;
522 unsigned long runnable_sum;
525 #ifdef CONFIG_FAIR_GROUP_SCHED
526 unsigned long tg_load_avg_contrib;
528 long prop_runnable_sum;
531 * h_load = weight * f(tg)
533 * Where f(tg) is the recursive weight fraction assigned to
536 unsigned long h_load;
537 u64 last_h_load_update;
538 struct sched_entity *h_load_next;
539 #endif /* CONFIG_FAIR_GROUP_SCHED */
540 #endif /* CONFIG_SMP */
542 #ifdef CONFIG_FAIR_GROUP_SCHED
543 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
546 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
547 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
548 * (like users, containers etc.)
550 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
551 * This list is used during load balance.
554 struct list_head leaf_cfs_rq_list;
555 struct task_group *tg; /* group that "owns" this runqueue */
557 #ifdef CONFIG_CFS_BANDWIDTH
561 s64 runtime_remaining;
564 u64 throttled_clock_task;
565 u64 throttled_clock_task_time;
568 struct list_head throttled_list;
569 #endif /* CONFIG_CFS_BANDWIDTH */
570 #endif /* CONFIG_FAIR_GROUP_SCHED */
573 static inline int rt_bandwidth_enabled(void)
575 return sysctl_sched_rt_runtime >= 0;
578 /* RT IPI pull logic requires IRQ_WORK */
579 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
580 # define HAVE_RT_PUSH_IPI
583 /* Real-Time classes' related field in a runqueue: */
585 struct rt_prio_array active;
586 unsigned int rt_nr_running;
587 unsigned int rr_nr_running;
588 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
590 int curr; /* highest queued rt task prio */
592 int next; /* next highest */
597 unsigned long rt_nr_migratory;
598 unsigned long rt_nr_total;
600 struct plist_head pushable_tasks;
602 #endif /* CONFIG_SMP */
608 /* Nests inside the rq lock: */
609 raw_spinlock_t rt_runtime_lock;
611 #ifdef CONFIG_RT_GROUP_SCHED
612 unsigned long rt_nr_boosted;
615 struct task_group *tg;
619 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
621 return rt_rq->rt_queued && rt_rq->rt_nr_running;
624 /* Deadline class' related fields in a runqueue */
626 /* runqueue is an rbtree, ordered by deadline */
627 struct rb_root_cached root;
629 unsigned long dl_nr_running;
633 * Deadline values of the currently executing and the
634 * earliest ready task on this rq. Caching these facilitates
635 * the decision wether or not a ready but not running task
636 * should migrate somewhere else.
643 unsigned long dl_nr_migratory;
647 * Tasks on this rq that can be pushed away. They are kept in
648 * an rb-tree, ordered by tasks' deadlines, with caching
649 * of the leftmost (earliest deadline) element.
651 struct rb_root_cached pushable_dl_tasks_root;
656 * "Active utilization" for this runqueue: increased when a
657 * task wakes up (becomes TASK_RUNNING) and decreased when a
663 * Utilization of the tasks "assigned" to this runqueue (including
664 * the tasks that are in runqueue and the tasks that executed on this
665 * CPU and blocked). Increased when a task moves to this runqueue, and
666 * decreased when the task moves away (migrates, changes scheduling
667 * policy, or terminates).
668 * This is needed to compute the "inactive utilization" for the
669 * runqueue (inactive utilization = this_bw - running_bw).
675 * Inverse of the fraction of CPU utilization that can be reclaimed
676 * by the GRUB algorithm.
681 #ifdef CONFIG_FAIR_GROUP_SCHED
682 /* An entity is a task if it doesn't "own" a runqueue */
683 #define entity_is_task(se) (!se->my_q)
685 #define entity_is_task(se) 1
690 * XXX we want to get rid of these helpers and use the full load resolution.
692 static inline long se_weight(struct sched_entity *se)
694 return scale_load_down(se->load.weight);
697 static inline long se_runnable(struct sched_entity *se)
699 return scale_load_down(se->runnable_weight);
702 static inline bool sched_asym_prefer(int a, int b)
704 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
708 * We add the notion of a root-domain which will be used to define per-domain
709 * variables. Each exclusive cpuset essentially defines an island domain by
710 * fully partitioning the member CPUs from any other cpuset. Whenever a new
711 * exclusive cpuset is created, we also create and attach a new root-domain
720 cpumask_var_t online;
723 * Indicate pullable load on at least one CPU, e.g:
724 * - More than one runnable task
725 * - Running task is misfit
730 * The bit corresponding to a CPU gets set here if such CPU has more
731 * than one runnable -deadline task (as it is below for RT tasks).
733 cpumask_var_t dlo_mask;
738 #ifdef HAVE_RT_PUSH_IPI
740 * For IPI pull requests, loop across the rto_mask.
742 struct irq_work rto_push_work;
743 raw_spinlock_t rto_lock;
744 /* These are only updated and read within rto_lock */
747 /* These atomics are updated outside of a lock */
748 atomic_t rto_loop_next;
749 atomic_t rto_loop_start;
752 * The "RT overload" flag: it gets set if a CPU has more than
753 * one runnable RT task.
755 cpumask_var_t rto_mask;
756 struct cpupri cpupri;
758 unsigned long max_cpu_capacity;
761 extern struct root_domain def_root_domain;
762 extern struct mutex sched_domains_mutex;
764 extern void init_defrootdomain(void);
765 extern int sched_init_domains(const struct cpumask *cpu_map);
766 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
767 extern void sched_get_rd(struct root_domain *rd);
768 extern void sched_put_rd(struct root_domain *rd);
770 #ifdef HAVE_RT_PUSH_IPI
771 extern void rto_push_irq_work_func(struct irq_work *work);
773 #endif /* CONFIG_SMP */
776 * This is the main, per-CPU runqueue data structure.
778 * Locking rule: those places that want to lock multiple runqueues
779 * (such as the load balancing or the thread migration code), lock
780 * acquire operations must be ordered by ascending &runqueue.
787 * nr_running and cpu_load should be in the same cacheline because
788 * remote CPUs use both these fields when doing load calculation.
790 unsigned int nr_running;
791 #ifdef CONFIG_NUMA_BALANCING
792 unsigned int nr_numa_running;
793 unsigned int nr_preferred_running;
794 unsigned int numa_migrate_on;
796 #define CPU_LOAD_IDX_MAX 5
797 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
798 #ifdef CONFIG_NO_HZ_COMMON
800 unsigned long last_load_update_tick;
801 unsigned long last_blocked_load_update_tick;
802 unsigned int has_blocked_load;
803 #endif /* CONFIG_SMP */
804 unsigned int nohz_tick_stopped;
806 #endif /* CONFIG_NO_HZ_COMMON */
808 /* capture load from *all* tasks on this CPU: */
809 struct load_weight load;
810 unsigned long nr_load_updates;
817 #ifdef CONFIG_FAIR_GROUP_SCHED
818 /* list of leaf cfs_rq on this CPU: */
819 struct list_head leaf_cfs_rq_list;
820 struct list_head *tmp_alone_branch;
821 #endif /* CONFIG_FAIR_GROUP_SCHED */
824 * This is part of a global counter where only the total sum
825 * over all CPUs matters. A task can increase this counter on
826 * one CPU and if it got migrated afterwards it may decrease
827 * it on another CPU. Always updated under the runqueue lock:
829 unsigned long nr_uninterruptible;
831 struct task_struct *curr;
832 struct task_struct *idle;
833 struct task_struct *stop;
834 unsigned long next_balance;
835 struct mm_struct *prev_mm;
837 unsigned int clock_update_flags;
844 struct root_domain *rd;
845 struct sched_domain *sd;
847 unsigned long cpu_capacity;
848 unsigned long cpu_capacity_orig;
850 struct callback_head *balance_callback;
852 unsigned char idle_balance;
854 unsigned long misfit_task_load;
856 /* For active balancing */
859 struct cpu_stop_work active_balance_work;
861 /* CPU of this runqueue: */
865 struct list_head cfs_tasks;
867 struct sched_avg avg_rt;
868 struct sched_avg avg_dl;
869 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
870 struct sched_avg avg_irq;
875 /* This is used to determine avg_idle's max value */
876 u64 max_idle_balance_cost;
879 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
882 #ifdef CONFIG_PARAVIRT
885 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
886 u64 prev_steal_time_rq;
889 /* calc_load related fields */
890 unsigned long calc_load_update;
891 long calc_load_active;
893 #ifdef CONFIG_SCHED_HRTICK
895 int hrtick_csd_pending;
896 call_single_data_t hrtick_csd;
898 struct hrtimer hrtick_timer;
901 #ifdef CONFIG_SCHEDSTATS
903 struct sched_info rq_sched_info;
904 unsigned long long rq_cpu_time;
905 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
907 /* sys_sched_yield() stats */
908 unsigned int yld_count;
910 /* schedule() stats */
911 unsigned int sched_count;
912 unsigned int sched_goidle;
914 /* try_to_wake_up() stats */
915 unsigned int ttwu_count;
916 unsigned int ttwu_local;
920 struct llist_head wake_list;
923 #ifdef CONFIG_CPU_IDLE
924 /* Must be inspected within a rcu lock section */
925 struct cpuidle_state *idle_state;
929 static inline int cpu_of(struct rq *rq)
939 #ifdef CONFIG_SCHED_SMT
940 extern void __update_idle_core(struct rq *rq);
942 static inline void update_idle_core(struct rq *rq)
944 if (static_branch_unlikely(&sched_smt_present))
945 __update_idle_core(rq);
949 static inline void update_idle_core(struct rq *rq) { }
952 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
954 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
955 #define this_rq() this_cpu_ptr(&runqueues)
956 #define task_rq(p) cpu_rq(task_cpu(p))
957 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
958 #define raw_rq() raw_cpu_ptr(&runqueues)
960 extern void update_rq_clock(struct rq *rq);
962 static inline u64 __rq_clock_broken(struct rq *rq)
964 return READ_ONCE(rq->clock);
968 * rq::clock_update_flags bits
970 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
971 * call to __schedule(). This is an optimisation to avoid
972 * neighbouring rq clock updates.
974 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
975 * in effect and calls to update_rq_clock() are being ignored.
977 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
978 * made to update_rq_clock() since the last time rq::lock was pinned.
980 * If inside of __schedule(), clock_update_flags will have been
981 * shifted left (a left shift is a cheap operation for the fast path
982 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
984 * if (rq-clock_update_flags >= RQCF_UPDATED)
986 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
987 * one position though, because the next rq_unpin_lock() will shift it
990 #define RQCF_REQ_SKIP 0x01
991 #define RQCF_ACT_SKIP 0x02
992 #define RQCF_UPDATED 0x04
994 static inline void assert_clock_updated(struct rq *rq)
997 * The only reason for not seeing a clock update since the
998 * last rq_pin_lock() is if we're currently skipping updates.
1000 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1003 static inline u64 rq_clock(struct rq *rq)
1005 lockdep_assert_held(&rq->lock);
1006 assert_clock_updated(rq);
1011 static inline u64 rq_clock_task(struct rq *rq)
1013 lockdep_assert_held(&rq->lock);
1014 assert_clock_updated(rq);
1016 return rq->clock_task;
1019 static inline void rq_clock_skip_update(struct rq *rq)
1021 lockdep_assert_held(&rq->lock);
1022 rq->clock_update_flags |= RQCF_REQ_SKIP;
1026 * See rt task throttling, which is the only time a skip
1027 * request is cancelled.
1029 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1031 lockdep_assert_held(&rq->lock);
1032 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1036 unsigned long flags;
1037 struct pin_cookie cookie;
1038 #ifdef CONFIG_SCHED_DEBUG
1040 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1041 * current pin context is stashed here in case it needs to be
1042 * restored in rq_repin_lock().
1044 unsigned int clock_update_flags;
1048 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1050 rf->cookie = lockdep_pin_lock(&rq->lock);
1052 #ifdef CONFIG_SCHED_DEBUG
1053 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1054 rf->clock_update_flags = 0;
1058 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1060 #ifdef CONFIG_SCHED_DEBUG
1061 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1062 rf->clock_update_flags = RQCF_UPDATED;
1065 lockdep_unpin_lock(&rq->lock, rf->cookie);
1068 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1070 lockdep_repin_lock(&rq->lock, rf->cookie);
1072 #ifdef CONFIG_SCHED_DEBUG
1074 * Restore the value we stashed in @rf for this pin context.
1076 rq->clock_update_flags |= rf->clock_update_flags;
1080 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1081 __acquires(rq->lock);
1083 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1084 __acquires(p->pi_lock)
1085 __acquires(rq->lock);
1087 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1088 __releases(rq->lock)
1090 rq_unpin_lock(rq, rf);
1091 raw_spin_unlock(&rq->lock);
1095 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1096 __releases(rq->lock)
1097 __releases(p->pi_lock)
1099 rq_unpin_lock(rq, rf);
1100 raw_spin_unlock(&rq->lock);
1101 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1105 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1106 __acquires(rq->lock)
1108 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1109 rq_pin_lock(rq, rf);
1113 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1114 __acquires(rq->lock)
1116 raw_spin_lock_irq(&rq->lock);
1117 rq_pin_lock(rq, rf);
1121 rq_lock(struct rq *rq, struct rq_flags *rf)
1122 __acquires(rq->lock)
1124 raw_spin_lock(&rq->lock);
1125 rq_pin_lock(rq, rf);
1129 rq_relock(struct rq *rq, struct rq_flags *rf)
1130 __acquires(rq->lock)
1132 raw_spin_lock(&rq->lock);
1133 rq_repin_lock(rq, rf);
1137 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1138 __releases(rq->lock)
1140 rq_unpin_lock(rq, rf);
1141 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1145 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1146 __releases(rq->lock)
1148 rq_unpin_lock(rq, rf);
1149 raw_spin_unlock_irq(&rq->lock);
1153 rq_unlock(struct rq *rq, struct rq_flags *rf)
1154 __releases(rq->lock)
1156 rq_unpin_lock(rq, rf);
1157 raw_spin_unlock(&rq->lock);
1160 static inline struct rq *
1161 this_rq_lock_irq(struct rq_flags *rf)
1162 __acquires(rq->lock)
1166 local_irq_disable();
1173 enum numa_topology_type {
1178 extern enum numa_topology_type sched_numa_topology_type;
1179 extern int sched_max_numa_distance;
1180 extern bool find_numa_distance(int distance);
1184 extern void sched_init_numa(void);
1185 extern void sched_domains_numa_masks_set(unsigned int cpu);
1186 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1188 static inline void sched_init_numa(void) { }
1189 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1190 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1193 #ifdef CONFIG_NUMA_BALANCING
1194 /* The regions in numa_faults array from task_struct */
1195 enum numa_faults_stats {
1201 extern void sched_setnuma(struct task_struct *p, int node);
1202 extern int migrate_task_to(struct task_struct *p, int cpu);
1203 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1205 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1208 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1211 #endif /* CONFIG_NUMA_BALANCING */
1216 queue_balance_callback(struct rq *rq,
1217 struct callback_head *head,
1218 void (*func)(struct rq *rq))
1220 lockdep_assert_held(&rq->lock);
1222 if (unlikely(head->next))
1225 head->func = (void (*)(struct callback_head *))func;
1226 head->next = rq->balance_callback;
1227 rq->balance_callback = head;
1230 extern void sched_ttwu_pending(void);
1232 #define rcu_dereference_check_sched_domain(p) \
1233 rcu_dereference_check((p), \
1234 lockdep_is_held(&sched_domains_mutex))
1237 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1238 * See detach_destroy_domains: synchronize_sched for details.
1240 * The domain tree of any CPU may only be accessed from within
1241 * preempt-disabled sections.
1243 #define for_each_domain(cpu, __sd) \
1244 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1245 __sd; __sd = __sd->parent)
1247 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1250 * highest_flag_domain - Return highest sched_domain containing flag.
1251 * @cpu: The CPU whose highest level of sched domain is to
1253 * @flag: The flag to check for the highest sched_domain
1254 * for the given CPU.
1256 * Returns the highest sched_domain of a CPU which contains the given flag.
1258 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1260 struct sched_domain *sd, *hsd = NULL;
1262 for_each_domain(cpu, sd) {
1263 if (!(sd->flags & flag))
1271 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1273 struct sched_domain *sd;
1275 for_each_domain(cpu, sd) {
1276 if (sd->flags & flag)
1283 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1284 DECLARE_PER_CPU(int, sd_llc_size);
1285 DECLARE_PER_CPU(int, sd_llc_id);
1286 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1287 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1288 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1289 extern struct static_key_false sched_asym_cpucapacity;
1291 struct sched_group_capacity {
1294 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1297 unsigned long capacity;
1298 unsigned long min_capacity; /* Min per-CPU capacity in group */
1299 unsigned long max_capacity; /* Max per-CPU capacity in group */
1300 unsigned long next_update;
1301 int imbalance; /* XXX unrelated to capacity but shared group state */
1303 #ifdef CONFIG_SCHED_DEBUG
1307 unsigned long cpumask[0]; /* Balance mask */
1310 struct sched_group {
1311 struct sched_group *next; /* Must be a circular list */
1314 unsigned int group_weight;
1315 struct sched_group_capacity *sgc;
1316 int asym_prefer_cpu; /* CPU of highest priority in group */
1319 * The CPUs this group covers.
1321 * NOTE: this field is variable length. (Allocated dynamically
1322 * by attaching extra space to the end of the structure,
1323 * depending on how many CPUs the kernel has booted up with)
1325 unsigned long cpumask[0];
1328 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1330 return to_cpumask(sg->cpumask);
1334 * See build_balance_mask().
1336 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1338 return to_cpumask(sg->sgc->cpumask);
1342 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1343 * @group: The group whose first CPU is to be returned.
1345 static inline unsigned int group_first_cpu(struct sched_group *group)
1347 return cpumask_first(sched_group_span(group));
1350 extern int group_balance_cpu(struct sched_group *sg);
1352 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1353 void register_sched_domain_sysctl(void);
1354 void dirty_sched_domain_sysctl(int cpu);
1355 void unregister_sched_domain_sysctl(void);
1357 static inline void register_sched_domain_sysctl(void)
1360 static inline void dirty_sched_domain_sysctl(int cpu)
1363 static inline void unregister_sched_domain_sysctl(void)
1370 static inline void sched_ttwu_pending(void) { }
1372 #endif /* CONFIG_SMP */
1375 #include "autogroup.h"
1377 #ifdef CONFIG_CGROUP_SCHED
1380 * Return the group to which this tasks belongs.
1382 * We cannot use task_css() and friends because the cgroup subsystem
1383 * changes that value before the cgroup_subsys::attach() method is called,
1384 * therefore we cannot pin it and might observe the wrong value.
1386 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1387 * core changes this before calling sched_move_task().
1389 * Instead we use a 'copy' which is updated from sched_move_task() while
1390 * holding both task_struct::pi_lock and rq::lock.
1392 static inline struct task_group *task_group(struct task_struct *p)
1394 return p->sched_task_group;
1397 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1398 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1400 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1401 struct task_group *tg = task_group(p);
1404 #ifdef CONFIG_FAIR_GROUP_SCHED
1405 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1406 p->se.cfs_rq = tg->cfs_rq[cpu];
1407 p->se.parent = tg->se[cpu];
1410 #ifdef CONFIG_RT_GROUP_SCHED
1411 p->rt.rt_rq = tg->rt_rq[cpu];
1412 p->rt.parent = tg->rt_se[cpu];
1416 #else /* CONFIG_CGROUP_SCHED */
1418 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1419 static inline struct task_group *task_group(struct task_struct *p)
1424 #endif /* CONFIG_CGROUP_SCHED */
1426 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1428 set_task_rq(p, cpu);
1431 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1432 * successfuly executed on another CPU. We must ensure that updates of
1433 * per-task data have been completed by this moment.
1436 #ifdef CONFIG_THREAD_INFO_IN_TASK
1439 task_thread_info(p)->cpu = cpu;
1446 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1448 #ifdef CONFIG_SCHED_DEBUG
1449 # include <linux/static_key.h>
1450 # define const_debug __read_mostly
1452 # define const_debug const
1455 #define SCHED_FEAT(name, enabled) \
1456 __SCHED_FEAT_##name ,
1459 #include "features.h"
1465 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1468 * To support run-time toggling of sched features, all the translation units
1469 * (but core.c) reference the sysctl_sched_features defined in core.c.
1471 extern const_debug unsigned int sysctl_sched_features;
1473 #define SCHED_FEAT(name, enabled) \
1474 static __always_inline bool static_branch_##name(struct static_key *key) \
1476 return static_key_##enabled(key); \
1479 #include "features.h"
1482 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1483 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1485 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1488 * Each translation unit has its own copy of sysctl_sched_features to allow
1489 * constants propagation at compile time and compiler optimization based on
1492 #define SCHED_FEAT(name, enabled) \
1493 (1UL << __SCHED_FEAT_##name) * enabled |
1494 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1495 #include "features.h"
1499 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1501 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1503 extern struct static_key_false sched_numa_balancing;
1504 extern struct static_key_false sched_schedstats;
1506 static inline u64 global_rt_period(void)
1508 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1511 static inline u64 global_rt_runtime(void)
1513 if (sysctl_sched_rt_runtime < 0)
1516 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1519 static inline int task_current(struct rq *rq, struct task_struct *p)
1521 return rq->curr == p;
1524 static inline int task_running(struct rq *rq, struct task_struct *p)
1529 return task_current(rq, p);
1533 static inline int task_on_rq_queued(struct task_struct *p)
1535 return p->on_rq == TASK_ON_RQ_QUEUED;
1538 static inline int task_on_rq_migrating(struct task_struct *p)
1540 return p->on_rq == TASK_ON_RQ_MIGRATING;
1546 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1547 #define WF_FORK 0x02 /* Child wakeup after fork */
1548 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1551 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1552 * of tasks with abnormal "nice" values across CPUs the contribution that
1553 * each task makes to its run queue's load is weighted according to its
1554 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1555 * scaled version of the new time slice allocation that they receive on time
1559 #define WEIGHT_IDLEPRIO 3
1560 #define WMULT_IDLEPRIO 1431655765
1562 extern const int sched_prio_to_weight[40];
1563 extern const u32 sched_prio_to_wmult[40];
1566 * {de,en}queue flags:
1568 * DEQUEUE_SLEEP - task is no longer runnable
1569 * ENQUEUE_WAKEUP - task just became runnable
1571 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1572 * are in a known state which allows modification. Such pairs
1573 * should preserve as much state as possible.
1575 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1578 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1579 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1580 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1584 #define DEQUEUE_SLEEP 0x01
1585 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1586 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1587 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1589 #define ENQUEUE_WAKEUP 0x01
1590 #define ENQUEUE_RESTORE 0x02
1591 #define ENQUEUE_MOVE 0x04
1592 #define ENQUEUE_NOCLOCK 0x08
1594 #define ENQUEUE_HEAD 0x10
1595 #define ENQUEUE_REPLENISH 0x20
1597 #define ENQUEUE_MIGRATED 0x40
1599 #define ENQUEUE_MIGRATED 0x00
1602 #define RETRY_TASK ((void *)-1UL)
1604 struct sched_class {
1605 const struct sched_class *next;
1607 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1608 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1609 void (*yield_task) (struct rq *rq);
1610 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1612 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1615 * It is the responsibility of the pick_next_task() method that will
1616 * return the next task to call put_prev_task() on the @prev task or
1617 * something equivalent.
1619 * May return RETRY_TASK when it finds a higher prio class has runnable
1622 struct task_struct * (*pick_next_task)(struct rq *rq,
1623 struct task_struct *prev,
1624 struct rq_flags *rf);
1625 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1628 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1629 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1631 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1633 void (*set_cpus_allowed)(struct task_struct *p,
1634 const struct cpumask *newmask);
1636 void (*rq_online)(struct rq *rq);
1637 void (*rq_offline)(struct rq *rq);
1640 void (*set_curr_task)(struct rq *rq);
1641 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1642 void (*task_fork)(struct task_struct *p);
1643 void (*task_dead)(struct task_struct *p);
1646 * The switched_from() call is allowed to drop rq->lock, therefore we
1647 * cannot assume the switched_from/switched_to pair is serliazed by
1648 * rq->lock. They are however serialized by p->pi_lock.
1650 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1651 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1652 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1655 unsigned int (*get_rr_interval)(struct rq *rq,
1656 struct task_struct *task);
1658 void (*update_curr)(struct rq *rq);
1660 #define TASK_SET_GROUP 0
1661 #define TASK_MOVE_GROUP 1
1663 #ifdef CONFIG_FAIR_GROUP_SCHED
1664 void (*task_change_group)(struct task_struct *p, int type);
1668 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1670 prev->sched_class->put_prev_task(rq, prev);
1673 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1675 curr->sched_class->set_curr_task(rq);
1679 #define sched_class_highest (&stop_sched_class)
1681 #define sched_class_highest (&dl_sched_class)
1683 #define for_each_class(class) \
1684 for (class = sched_class_highest; class; class = class->next)
1686 extern const struct sched_class stop_sched_class;
1687 extern const struct sched_class dl_sched_class;
1688 extern const struct sched_class rt_sched_class;
1689 extern const struct sched_class fair_sched_class;
1690 extern const struct sched_class idle_sched_class;
1695 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1697 extern void trigger_load_balance(struct rq *rq);
1699 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1703 #ifdef CONFIG_CPU_IDLE
1704 static inline void idle_set_state(struct rq *rq,
1705 struct cpuidle_state *idle_state)
1707 rq->idle_state = idle_state;
1710 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1712 SCHED_WARN_ON(!rcu_read_lock_held());
1714 return rq->idle_state;
1717 static inline void idle_set_state(struct rq *rq,
1718 struct cpuidle_state *idle_state)
1722 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1728 extern void schedule_idle(void);
1730 extern void sysrq_sched_debug_show(void);
1731 extern void sched_init_granularity(void);
1732 extern void update_max_interval(void);
1734 extern void init_sched_dl_class(void);
1735 extern void init_sched_rt_class(void);
1736 extern void init_sched_fair_class(void);
1738 extern void reweight_task(struct task_struct *p, int prio);
1740 extern void resched_curr(struct rq *rq);
1741 extern void resched_cpu(int cpu);
1743 extern struct rt_bandwidth def_rt_bandwidth;
1744 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1746 extern struct dl_bandwidth def_dl_bandwidth;
1747 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1748 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1749 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1750 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1753 #define BW_UNIT (1 << BW_SHIFT)
1754 #define RATIO_SHIFT 8
1755 unsigned long to_ratio(u64 period, u64 runtime);
1757 extern void init_entity_runnable_average(struct sched_entity *se);
1758 extern void post_init_entity_util_avg(struct sched_entity *se);
1760 #ifdef CONFIG_NO_HZ_FULL
1761 extern bool sched_can_stop_tick(struct rq *rq);
1762 extern int __init sched_tick_offload_init(void);
1765 * Tick may be needed by tasks in the runqueue depending on their policy and
1766 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1767 * nohz mode if necessary.
1769 static inline void sched_update_tick_dependency(struct rq *rq)
1773 if (!tick_nohz_full_enabled())
1778 if (!tick_nohz_full_cpu(cpu))
1781 if (sched_can_stop_tick(rq))
1782 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1784 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1787 static inline int sched_tick_offload_init(void) { return 0; }
1788 static inline void sched_update_tick_dependency(struct rq *rq) { }
1791 static inline void add_nr_running(struct rq *rq, unsigned count)
1793 unsigned prev_nr = rq->nr_running;
1795 rq->nr_running = prev_nr + count;
1797 if (prev_nr < 2 && rq->nr_running >= 2) {
1799 if (!READ_ONCE(rq->rd->overload))
1800 WRITE_ONCE(rq->rd->overload, 1);
1804 sched_update_tick_dependency(rq);
1807 static inline void sub_nr_running(struct rq *rq, unsigned count)
1809 rq->nr_running -= count;
1810 /* Check if we still need preemption */
1811 sched_update_tick_dependency(rq);
1814 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1815 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1817 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1819 extern const_debug unsigned int sysctl_sched_nr_migrate;
1820 extern const_debug unsigned int sysctl_sched_migration_cost;
1822 #ifdef CONFIG_SCHED_HRTICK
1826 * - enabled by features
1827 * - hrtimer is actually high res
1829 static inline int hrtick_enabled(struct rq *rq)
1831 if (!sched_feat(HRTICK))
1833 if (!cpu_active(cpu_of(rq)))
1835 return hrtimer_is_hres_active(&rq->hrtick_timer);
1838 void hrtick_start(struct rq *rq, u64 delay);
1842 static inline int hrtick_enabled(struct rq *rq)
1847 #endif /* CONFIG_SCHED_HRTICK */
1849 #ifndef arch_scale_freq_capacity
1850 static __always_inline
1851 unsigned long arch_scale_freq_capacity(int cpu)
1853 return SCHED_CAPACITY_SCALE;
1858 #ifndef arch_scale_cpu_capacity
1859 static __always_inline
1860 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1862 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1863 return sd->smt_gain / sd->span_weight;
1865 return SCHED_CAPACITY_SCALE;
1869 #ifndef arch_scale_cpu_capacity
1870 static __always_inline
1871 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1873 return SCHED_CAPACITY_SCALE;
1879 #ifdef CONFIG_PREEMPT
1881 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1884 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1885 * way at the expense of forcing extra atomic operations in all
1886 * invocations. This assures that the double_lock is acquired using the
1887 * same underlying policy as the spinlock_t on this architecture, which
1888 * reduces latency compared to the unfair variant below. However, it
1889 * also adds more overhead and therefore may reduce throughput.
1891 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1892 __releases(this_rq->lock)
1893 __acquires(busiest->lock)
1894 __acquires(this_rq->lock)
1896 raw_spin_unlock(&this_rq->lock);
1897 double_rq_lock(this_rq, busiest);
1904 * Unfair double_lock_balance: Optimizes throughput at the expense of
1905 * latency by eliminating extra atomic operations when the locks are
1906 * already in proper order on entry. This favors lower CPU-ids and will
1907 * grant the double lock to lower CPUs over higher ids under contention,
1908 * regardless of entry order into the function.
1910 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1911 __releases(this_rq->lock)
1912 __acquires(busiest->lock)
1913 __acquires(this_rq->lock)
1917 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1918 if (busiest < this_rq) {
1919 raw_spin_unlock(&this_rq->lock);
1920 raw_spin_lock(&busiest->lock);
1921 raw_spin_lock_nested(&this_rq->lock,
1922 SINGLE_DEPTH_NESTING);
1925 raw_spin_lock_nested(&busiest->lock,
1926 SINGLE_DEPTH_NESTING);
1931 #endif /* CONFIG_PREEMPT */
1934 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1936 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1938 if (unlikely(!irqs_disabled())) {
1939 /* printk() doesn't work well under rq->lock */
1940 raw_spin_unlock(&this_rq->lock);
1944 return _double_lock_balance(this_rq, busiest);
1947 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1948 __releases(busiest->lock)
1950 raw_spin_unlock(&busiest->lock);
1951 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1954 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1960 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1963 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1969 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1972 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1978 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1982 * double_rq_lock - safely lock two runqueues
1984 * Note this does not disable interrupts like task_rq_lock,
1985 * you need to do so manually before calling.
1987 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1988 __acquires(rq1->lock)
1989 __acquires(rq2->lock)
1991 BUG_ON(!irqs_disabled());
1993 raw_spin_lock(&rq1->lock);
1994 __acquire(rq2->lock); /* Fake it out ;) */
1997 raw_spin_lock(&rq1->lock);
1998 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2000 raw_spin_lock(&rq2->lock);
2001 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2007 * double_rq_unlock - safely unlock two runqueues
2009 * Note this does not restore interrupts like task_rq_unlock,
2010 * you need to do so manually after calling.
2012 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2013 __releases(rq1->lock)
2014 __releases(rq2->lock)
2016 raw_spin_unlock(&rq1->lock);
2018 raw_spin_unlock(&rq2->lock);
2020 __release(rq2->lock);
2023 extern void set_rq_online (struct rq *rq);
2024 extern void set_rq_offline(struct rq *rq);
2025 extern bool sched_smp_initialized;
2027 #else /* CONFIG_SMP */
2030 * double_rq_lock - safely lock two runqueues
2032 * Note this does not disable interrupts like task_rq_lock,
2033 * you need to do so manually before calling.
2035 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2036 __acquires(rq1->lock)
2037 __acquires(rq2->lock)
2039 BUG_ON(!irqs_disabled());
2041 raw_spin_lock(&rq1->lock);
2042 __acquire(rq2->lock); /* Fake it out ;) */
2046 * double_rq_unlock - safely unlock two runqueues
2048 * Note this does not restore interrupts like task_rq_unlock,
2049 * you need to do so manually after calling.
2051 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2052 __releases(rq1->lock)
2053 __releases(rq2->lock)
2056 raw_spin_unlock(&rq1->lock);
2057 __release(rq2->lock);
2062 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2063 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2065 #ifdef CONFIG_SCHED_DEBUG
2066 extern bool sched_debug_enabled;
2068 extern void print_cfs_stats(struct seq_file *m, int cpu);
2069 extern void print_rt_stats(struct seq_file *m, int cpu);
2070 extern void print_dl_stats(struct seq_file *m, int cpu);
2071 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2072 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2073 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2074 #ifdef CONFIG_NUMA_BALANCING
2076 show_numa_stats(struct task_struct *p, struct seq_file *m);
2078 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2079 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2080 #endif /* CONFIG_NUMA_BALANCING */
2081 #endif /* CONFIG_SCHED_DEBUG */
2083 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2084 extern void init_rt_rq(struct rt_rq *rt_rq);
2085 extern void init_dl_rq(struct dl_rq *dl_rq);
2087 extern void cfs_bandwidth_usage_inc(void);
2088 extern void cfs_bandwidth_usage_dec(void);
2090 #ifdef CONFIG_NO_HZ_COMMON
2091 #define NOHZ_BALANCE_KICK_BIT 0
2092 #define NOHZ_STATS_KICK_BIT 1
2094 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2095 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2097 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2099 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2101 extern void nohz_balance_exit_idle(struct rq *rq);
2103 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2109 void __dl_update(struct dl_bw *dl_b, s64 bw)
2111 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2114 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2115 "sched RCU must be held");
2116 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2117 struct rq *rq = cpu_rq(i);
2119 rq->dl.extra_bw += bw;
2124 void __dl_update(struct dl_bw *dl_b, s64 bw)
2126 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2133 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2138 struct u64_stats_sync sync;
2141 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2144 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2145 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2146 * and never move forward.
2148 static inline u64 irq_time_read(int cpu)
2150 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2155 seq = __u64_stats_fetch_begin(&irqtime->sync);
2156 total = irqtime->total;
2157 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2161 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2163 #ifdef CONFIG_CPU_FREQ
2164 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2167 * cpufreq_update_util - Take a note about CPU utilization changes.
2168 * @rq: Runqueue to carry out the update for.
2169 * @flags: Update reason flags.
2171 * This function is called by the scheduler on the CPU whose utilization is
2174 * It can only be called from RCU-sched read-side critical sections.
2176 * The way cpufreq is currently arranged requires it to evaluate the CPU
2177 * performance state (frequency/voltage) on a regular basis to prevent it from
2178 * being stuck in a completely inadequate performance level for too long.
2179 * That is not guaranteed to happen if the updates are only triggered from CFS
2180 * and DL, though, because they may not be coming in if only RT tasks are
2181 * active all the time (or there are RT tasks only).
2183 * As a workaround for that issue, this function is called periodically by the
2184 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2185 * but that really is a band-aid. Going forward it should be replaced with
2186 * solutions targeted more specifically at RT tasks.
2188 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2190 struct update_util_data *data;
2192 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2195 data->func(data, rq_clock(rq), flags);
2198 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2199 #endif /* CONFIG_CPU_FREQ */
2201 #ifdef arch_scale_freq_capacity
2202 # ifndef arch_scale_freq_invariant
2203 # define arch_scale_freq_invariant() true
2206 # define arch_scale_freq_invariant() false
2209 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2210 static inline unsigned long cpu_bw_dl(struct rq *rq)
2212 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2215 static inline unsigned long cpu_util_dl(struct rq *rq)
2217 return READ_ONCE(rq->avg_dl.util_avg);
2220 static inline unsigned long cpu_util_cfs(struct rq *rq)
2222 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2224 if (sched_feat(UTIL_EST)) {
2225 util = max_t(unsigned long, util,
2226 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2232 static inline unsigned long cpu_util_rt(struct rq *rq)
2234 return READ_ONCE(rq->avg_rt.util_avg);
2238 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2239 static inline unsigned long cpu_util_irq(struct rq *rq)
2241 return rq->avg_irq.util_avg;
2245 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2247 util *= (max - irq);
2254 static inline unsigned long cpu_util_irq(struct rq *rq)
2260 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)