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/stat.h>
27 #include <linux/sched/sysctl.h>
28 #include <linux/sched/task.h>
29 #include <linux/sched/task_stack.h>
30 #include <linux/sched/topology.h>
31 #include <linux/sched/user.h>
32 #include <linux/sched/wake_q.h>
33 #include <linux/sched/xacct.h>
35 #include <uapi/linux/sched/types.h>
37 #include <linux/binfmts.h>
38 #include <linux/blkdev.h>
39 #include <linux/compat.h>
40 #include <linux/context_tracking.h>
41 #include <linux/cpufreq.h>
42 #include <linux/cpuidle.h>
43 #include <linux/cpuset.h>
44 #include <linux/ctype.h>
45 #include <linux/debugfs.h>
46 #include <linux/delayacct.h>
47 #include <linux/init_task.h>
48 #include <linux/kprobes.h>
49 #include <linux/kthread.h>
50 #include <linux/membarrier.h>
51 #include <linux/migrate.h>
52 #include <linux/mmu_context.h>
53 #include <linux/nmi.h>
54 #include <linux/proc_fs.h>
55 #include <linux/prefetch.h>
56 #include <linux/profile.h>
57 #include <linux/rcupdate_wait.h>
58 #include <linux/security.h>
59 #include <linux/stackprotector.h>
60 #include <linux/stop_machine.h>
61 #include <linux/suspend.h>
62 #include <linux/swait.h>
63 #include <linux/syscalls.h>
64 #include <linux/task_work.h>
65 #include <linux/tsacct_kern.h>
69 #ifdef CONFIG_PARAVIRT
70 # include <asm/paravirt.h>
74 #include "cpudeadline.h"
76 #ifdef CONFIG_SCHED_DEBUG
77 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
79 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
85 /* task_struct::on_rq states: */
86 #define TASK_ON_RQ_QUEUED 1
87 #define TASK_ON_RQ_MIGRATING 2
89 extern __read_mostly int scheduler_running;
91 extern unsigned long calc_load_update;
92 extern atomic_long_t calc_load_tasks;
94 extern void calc_global_load_tick(struct rq *this_rq);
95 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
98 extern void cpu_load_update_active(struct rq *this_rq);
100 static inline void cpu_load_update_active(struct rq *this_rq) { }
104 * Helpers for converting nanosecond timing to jiffy resolution
106 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
109 * Increase resolution of nice-level calculations for 64-bit architectures.
110 * The extra resolution improves shares distribution and load balancing of
111 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
112 * hierarchies, especially on larger systems. This is not a user-visible change
113 * and does not change the user-interface for setting shares/weights.
115 * We increase resolution only if we have enough bits to allow this increased
116 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
117 * are pretty high and the returns do not justify the increased costs.
119 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
120 * increase coverage and consistency always enable it on 64-bit platforms.
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
127 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
128 # define scale_load(w) (w)
129 # define scale_load_down(w) (w)
133 * Task weight (visible to users) and its load (invisible to users) have
134 * independent resolution, but they should be well calibrated. We use
135 * scale_load() and scale_load_down(w) to convert between them. The
136 * following must be true:
138 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
141 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
144 * Single value that decides SCHED_DEADLINE internal math precision.
145 * 10 -> just above 1us
146 * 9 -> just above 0.5us
151 * Single value that denotes runtime == period, ie unlimited time.
153 #define RUNTIME_INF ((u64)~0ULL)
155 static inline int idle_policy(int policy)
157 return policy == SCHED_IDLE;
159 static inline int fair_policy(int policy)
161 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
164 static inline int rt_policy(int policy)
166 return policy == SCHED_FIFO || policy == SCHED_RR;
169 static inline int dl_policy(int policy)
171 return policy == SCHED_DEADLINE;
173 static inline bool valid_policy(int policy)
175 return idle_policy(policy) || fair_policy(policy) ||
176 rt_policy(policy) || dl_policy(policy);
179 static inline int task_has_rt_policy(struct task_struct *p)
181 return rt_policy(p->policy);
184 static inline int task_has_dl_policy(struct task_struct *p)
186 return dl_policy(p->policy);
189 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
192 * !! For sched_setattr_nocheck() (kernel) only !!
194 * This is actually gross. :(
196 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
197 * tasks, but still be able to sleep. We need this on platforms that cannot
198 * atomically change clock frequency. Remove once fast switching will be
199 * available on such platforms.
201 * SUGOV stands for SchedUtil GOVernor.
203 #define SCHED_FLAG_SUGOV 0x10000000
205 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
207 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
208 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
215 * Tells if entity @a should preempt entity @b.
218 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
220 return dl_entity_is_special(a) ||
221 dl_time_before(a->deadline, b->deadline);
225 * This is the priority-queue data structure of the RT scheduling class:
227 struct rt_prio_array {
228 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
229 struct list_head queue[MAX_RT_PRIO];
232 struct rt_bandwidth {
233 /* nests inside the rq lock: */
234 raw_spinlock_t rt_runtime_lock;
237 struct hrtimer rt_period_timer;
238 unsigned int rt_period_active;
241 void __dl_clear_params(struct task_struct *p);
244 * To keep the bandwidth of -deadline tasks and groups under control
245 * we need some place where:
246 * - store the maximum -deadline bandwidth of the system (the group);
247 * - cache the fraction of that bandwidth that is currently allocated.
249 * This is all done in the data structure below. It is similar to the
250 * one used for RT-throttling (rt_bandwidth), with the main difference
251 * that, since here we are only interested in admission control, we
252 * do not decrease any runtime while the group "executes", neither we
253 * need a timer to replenish it.
255 * With respect to SMP, the bandwidth is given on a per-CPU basis,
257 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
258 * - dl_total_bw array contains, in the i-eth element, the currently
259 * allocated bandwidth on the i-eth CPU.
260 * Moreover, groups consume bandwidth on each CPU, while tasks only
261 * consume bandwidth on the CPU they're running on.
262 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
263 * that will be shown the next time the proc or cgroup controls will
264 * be red. It on its turn can be changed by writing on its own
267 struct dl_bandwidth {
268 raw_spinlock_t dl_runtime_lock;
273 static inline int dl_bandwidth_enabled(void)
275 return sysctl_sched_rt_runtime >= 0;
284 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
287 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
289 dl_b->total_bw -= tsk_bw;
290 __dl_update(dl_b, (s32)tsk_bw / cpus);
294 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
296 dl_b->total_bw += tsk_bw;
297 __dl_update(dl_b, -((s32)tsk_bw / cpus));
301 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
303 return dl_b->bw != -1 &&
304 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
307 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
308 extern void init_dl_bw(struct dl_bw *dl_b);
309 extern int sched_dl_global_validate(void);
310 extern void sched_dl_do_global(void);
311 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
312 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
313 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
314 extern bool __checkparam_dl(const struct sched_attr *attr);
315 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
316 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
317 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
318 extern bool dl_cpu_busy(unsigned int cpu);
320 #ifdef CONFIG_CGROUP_SCHED
322 #include <linux/cgroup.h>
327 extern struct list_head task_groups;
329 struct cfs_bandwidth {
330 #ifdef CONFIG_CFS_BANDWIDTH
335 s64 hierarchical_quota;
341 struct hrtimer period_timer;
342 struct hrtimer slack_timer;
343 struct list_head throttled_cfs_rq;
352 /* Task group related information */
354 struct cgroup_subsys_state css;
356 #ifdef CONFIG_FAIR_GROUP_SCHED
357 /* schedulable entities of this group on each CPU */
358 struct sched_entity **se;
359 /* runqueue "owned" by this group on each CPU */
360 struct cfs_rq **cfs_rq;
361 unsigned long shares;
365 * load_avg can be heavily contended at clock tick time, so put
366 * it in its own cacheline separated from the fields above which
367 * will also be accessed at each tick.
369 atomic_long_t load_avg ____cacheline_aligned;
373 #ifdef CONFIG_RT_GROUP_SCHED
374 struct sched_rt_entity **rt_se;
375 struct rt_rq **rt_rq;
377 struct rt_bandwidth rt_bandwidth;
381 struct list_head list;
383 struct task_group *parent;
384 struct list_head siblings;
385 struct list_head children;
387 #ifdef CONFIG_SCHED_AUTOGROUP
388 struct autogroup *autogroup;
391 struct cfs_bandwidth cfs_bandwidth;
394 #ifdef CONFIG_FAIR_GROUP_SCHED
395 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
398 * A weight of 0 or 1 can cause arithmetics problems.
399 * A weight of a cfs_rq is the sum of weights of which entities
400 * are queued on this cfs_rq, so a weight of a entity should not be
401 * too large, so as the shares value of a task group.
402 * (The default weight is 1024 - so there's no practical
403 * limitation from this.)
405 #define MIN_SHARES (1UL << 1)
406 #define MAX_SHARES (1UL << 18)
409 typedef int (*tg_visitor)(struct task_group *, void *);
411 extern int walk_tg_tree_from(struct task_group *from,
412 tg_visitor down, tg_visitor up, void *data);
415 * Iterate the full tree, calling @down when first entering a node and @up when
416 * leaving it for the final time.
418 * Caller must hold rcu_lock or sufficient equivalent.
420 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
422 return walk_tg_tree_from(&root_task_group, down, up, data);
425 extern int tg_nop(struct task_group *tg, void *data);
427 extern void free_fair_sched_group(struct task_group *tg);
428 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
429 extern void online_fair_sched_group(struct task_group *tg);
430 extern void unregister_fair_sched_group(struct task_group *tg);
431 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
432 struct sched_entity *se, int cpu,
433 struct sched_entity *parent);
434 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
436 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
437 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
438 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
440 extern void free_rt_sched_group(struct task_group *tg);
441 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
442 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
443 struct sched_rt_entity *rt_se, int cpu,
444 struct sched_rt_entity *parent);
445 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
446 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
447 extern long sched_group_rt_runtime(struct task_group *tg);
448 extern long sched_group_rt_period(struct task_group *tg);
449 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
451 extern struct task_group *sched_create_group(struct task_group *parent);
452 extern void sched_online_group(struct task_group *tg,
453 struct task_group *parent);
454 extern void sched_destroy_group(struct task_group *tg);
455 extern void sched_offline_group(struct task_group *tg);
457 extern void sched_move_task(struct task_struct *tsk);
459 #ifdef CONFIG_FAIR_GROUP_SCHED
460 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
463 extern void set_task_rq_fair(struct sched_entity *se,
464 struct cfs_rq *prev, struct cfs_rq *next);
465 #else /* !CONFIG_SMP */
466 static inline void set_task_rq_fair(struct sched_entity *se,
467 struct cfs_rq *prev, struct cfs_rq *next) { }
468 #endif /* CONFIG_SMP */
469 #endif /* CONFIG_FAIR_GROUP_SCHED */
471 #else /* CONFIG_CGROUP_SCHED */
473 struct cfs_bandwidth { };
475 #endif /* CONFIG_CGROUP_SCHED */
477 /* CFS-related fields in a runqueue */
479 struct load_weight load;
480 unsigned long runnable_weight;
481 unsigned int nr_running;
482 unsigned int h_nr_running;
487 u64 min_vruntime_copy;
490 struct rb_root_cached tasks_timeline;
493 * 'curr' points to currently running entity on this cfs_rq.
494 * It is set to NULL otherwise (i.e when none are currently running).
496 struct sched_entity *curr;
497 struct sched_entity *next;
498 struct sched_entity *last;
499 struct sched_entity *skip;
501 #ifdef CONFIG_SCHED_DEBUG
502 unsigned int nr_spread_over;
509 struct sched_avg avg;
511 u64 load_last_update_time_copy;
514 raw_spinlock_t lock ____cacheline_aligned;
516 unsigned long load_avg;
517 unsigned long util_avg;
518 unsigned long runnable_sum;
521 #ifdef CONFIG_FAIR_GROUP_SCHED
522 unsigned long tg_load_avg_contrib;
524 long prop_runnable_sum;
527 * h_load = weight * f(tg)
529 * Where f(tg) is the recursive weight fraction assigned to
532 unsigned long h_load;
533 u64 last_h_load_update;
534 struct sched_entity *h_load_next;
535 #endif /* CONFIG_FAIR_GROUP_SCHED */
536 #endif /* CONFIG_SMP */
538 #ifdef CONFIG_FAIR_GROUP_SCHED
539 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
542 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
543 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
544 * (like users, containers etc.)
546 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
547 * This list is used during load balance.
550 struct list_head leaf_cfs_rq_list;
551 struct task_group *tg; /* group that "owns" this runqueue */
553 #ifdef CONFIG_CFS_BANDWIDTH
557 s64 runtime_remaining;
560 u64 throttled_clock_task;
561 u64 throttled_clock_task_time;
564 struct list_head throttled_list;
565 #endif /* CONFIG_CFS_BANDWIDTH */
566 #endif /* CONFIG_FAIR_GROUP_SCHED */
569 static inline int rt_bandwidth_enabled(void)
571 return sysctl_sched_rt_runtime >= 0;
574 /* RT IPI pull logic requires IRQ_WORK */
575 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
576 # define HAVE_RT_PUSH_IPI
579 /* Real-Time classes' related field in a runqueue: */
581 struct rt_prio_array active;
582 unsigned int rt_nr_running;
583 unsigned int rr_nr_running;
584 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
586 int curr; /* highest queued rt task prio */
588 int next; /* next highest */
593 unsigned long rt_nr_migratory;
594 unsigned long rt_nr_total;
596 struct plist_head pushable_tasks;
598 #endif /* CONFIG_SMP */
604 /* Nests inside the rq lock: */
605 raw_spinlock_t rt_runtime_lock;
607 #ifdef CONFIG_RT_GROUP_SCHED
608 unsigned long rt_nr_boosted;
611 struct task_group *tg;
615 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
617 return rt_rq->rt_queued && rt_rq->rt_nr_running;
620 /* Deadline class' related fields in a runqueue */
622 /* runqueue is an rbtree, ordered by deadline */
623 struct rb_root_cached root;
625 unsigned long dl_nr_running;
629 * Deadline values of the currently executing and the
630 * earliest ready task on this rq. Caching these facilitates
631 * the decision wether or not a ready but not running task
632 * should migrate somewhere else.
639 unsigned long dl_nr_migratory;
643 * Tasks on this rq that can be pushed away. They are kept in
644 * an rb-tree, ordered by tasks' deadlines, with caching
645 * of the leftmost (earliest deadline) element.
647 struct rb_root_cached pushable_dl_tasks_root;
652 * "Active utilization" for this runqueue: increased when a
653 * task wakes up (becomes TASK_RUNNING) and decreased when a
659 * Utilization of the tasks "assigned" to this runqueue (including
660 * the tasks that are in runqueue and the tasks that executed on this
661 * CPU and blocked). Increased when a task moves to this runqueue, and
662 * decreased when the task moves away (migrates, changes scheduling
663 * policy, or terminates).
664 * This is needed to compute the "inactive utilization" for the
665 * runqueue (inactive utilization = this_bw - running_bw).
671 * Inverse of the fraction of CPU utilization that can be reclaimed
672 * by the GRUB algorithm.
677 #ifdef CONFIG_FAIR_GROUP_SCHED
678 /* An entity is a task if it doesn't "own" a runqueue */
679 #define entity_is_task(se) (!se->my_q)
681 #define entity_is_task(se) 1
686 * XXX we want to get rid of these helpers and use the full load resolution.
688 static inline long se_weight(struct sched_entity *se)
690 return scale_load_down(se->load.weight);
693 static inline long se_runnable(struct sched_entity *se)
695 return scale_load_down(se->runnable_weight);
698 static inline bool sched_asym_prefer(int a, int b)
700 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
704 * We add the notion of a root-domain which will be used to define per-domain
705 * variables. Each exclusive cpuset essentially defines an island domain by
706 * fully partitioning the member CPUs from any other cpuset. Whenever a new
707 * exclusive cpuset is created, we also create and attach a new root-domain
716 cpumask_var_t online;
719 * Indicate pullable load on at least one CPU, e.g:
720 * - More than one runnable task
721 * - Running task is misfit
726 * The bit corresponding to a CPU gets set here if such CPU has more
727 * than one runnable -deadline task (as it is below for RT tasks).
729 cpumask_var_t dlo_mask;
734 #ifdef HAVE_RT_PUSH_IPI
736 * For IPI pull requests, loop across the rto_mask.
738 struct irq_work rto_push_work;
739 raw_spinlock_t rto_lock;
740 /* These are only updated and read within rto_lock */
743 /* These atomics are updated outside of a lock */
744 atomic_t rto_loop_next;
745 atomic_t rto_loop_start;
748 * The "RT overload" flag: it gets set if a CPU has more than
749 * one runnable RT task.
751 cpumask_var_t rto_mask;
752 struct cpupri cpupri;
754 unsigned long max_cpu_capacity;
757 extern struct root_domain def_root_domain;
758 extern struct mutex sched_domains_mutex;
760 extern void init_defrootdomain(void);
761 extern int sched_init_domains(const struct cpumask *cpu_map);
762 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
763 extern void sched_get_rd(struct root_domain *rd);
764 extern void sched_put_rd(struct root_domain *rd);
766 #ifdef HAVE_RT_PUSH_IPI
767 extern void rto_push_irq_work_func(struct irq_work *work);
769 #endif /* CONFIG_SMP */
772 * This is the main, per-CPU runqueue data structure.
774 * Locking rule: those places that want to lock multiple runqueues
775 * (such as the load balancing or the thread migration code), lock
776 * acquire operations must be ordered by ascending &runqueue.
783 * nr_running and cpu_load should be in the same cacheline because
784 * remote CPUs use both these fields when doing load calculation.
786 unsigned int nr_running;
787 #ifdef CONFIG_NUMA_BALANCING
788 unsigned int nr_numa_running;
789 unsigned int nr_preferred_running;
791 #define CPU_LOAD_IDX_MAX 5
792 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
793 #ifdef CONFIG_NO_HZ_COMMON
795 unsigned long last_load_update_tick;
796 unsigned long last_blocked_load_update_tick;
797 unsigned int has_blocked_load;
798 #endif /* CONFIG_SMP */
799 unsigned int nohz_tick_stopped;
801 #endif /* CONFIG_NO_HZ_COMMON */
803 /* capture load from *all* tasks on this CPU: */
804 struct load_weight load;
805 unsigned long nr_load_updates;
812 #ifdef CONFIG_FAIR_GROUP_SCHED
813 /* list of leaf cfs_rq on this CPU: */
814 struct list_head leaf_cfs_rq_list;
815 struct list_head *tmp_alone_branch;
816 #endif /* CONFIG_FAIR_GROUP_SCHED */
819 * This is part of a global counter where only the total sum
820 * over all CPUs matters. A task can increase this counter on
821 * one CPU and if it got migrated afterwards it may decrease
822 * it on another CPU. Always updated under the runqueue lock:
824 unsigned long nr_uninterruptible;
826 struct task_struct *curr;
827 struct task_struct *idle;
828 struct task_struct *stop;
829 unsigned long next_balance;
830 struct mm_struct *prev_mm;
832 unsigned int clock_update_flags;
839 struct root_domain *rd;
840 struct sched_domain *sd;
842 unsigned long cpu_capacity;
843 unsigned long cpu_capacity_orig;
845 struct callback_head *balance_callback;
847 unsigned char idle_balance;
849 unsigned long misfit_task_load;
851 /* For active balancing */
854 struct cpu_stop_work active_balance_work;
856 /* CPU of this runqueue: */
860 struct list_head cfs_tasks;
862 struct sched_avg avg_rt;
863 struct sched_avg avg_dl;
864 #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
865 #define HAVE_SCHED_AVG_IRQ
866 struct sched_avg avg_irq;
871 /* This is used to determine avg_idle's max value */
872 u64 max_idle_balance_cost;
875 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
878 #ifdef CONFIG_PARAVIRT
881 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
882 u64 prev_steal_time_rq;
885 /* calc_load related fields */
886 unsigned long calc_load_update;
887 long calc_load_active;
889 #ifdef CONFIG_SCHED_HRTICK
891 int hrtick_csd_pending;
892 call_single_data_t hrtick_csd;
894 struct hrtimer hrtick_timer;
897 #ifdef CONFIG_SCHEDSTATS
899 struct sched_info rq_sched_info;
900 unsigned long long rq_cpu_time;
901 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
903 /* sys_sched_yield() stats */
904 unsigned int yld_count;
906 /* schedule() stats */
907 unsigned int sched_count;
908 unsigned int sched_goidle;
910 /* try_to_wake_up() stats */
911 unsigned int ttwu_count;
912 unsigned int ttwu_local;
916 struct llist_head wake_list;
919 #ifdef CONFIG_CPU_IDLE
920 /* Must be inspected within a rcu lock section */
921 struct cpuidle_state *idle_state;
925 static inline int cpu_of(struct rq *rq)
935 #ifdef CONFIG_SCHED_SMT
937 extern struct static_key_false sched_smt_present;
939 extern void __update_idle_core(struct rq *rq);
941 static inline void update_idle_core(struct rq *rq)
943 if (static_branch_unlikely(&sched_smt_present))
944 __update_idle_core(rq);
948 static inline void update_idle_core(struct rq *rq) { }
951 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
953 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
954 #define this_rq() this_cpu_ptr(&runqueues)
955 #define task_rq(p) cpu_rq(task_cpu(p))
956 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
957 #define raw_rq() raw_cpu_ptr(&runqueues)
959 static inline u64 __rq_clock_broken(struct rq *rq)
961 return READ_ONCE(rq->clock);
965 * rq::clock_update_flags bits
967 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
968 * call to __schedule(). This is an optimisation to avoid
969 * neighbouring rq clock updates.
971 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
972 * in effect and calls to update_rq_clock() are being ignored.
974 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
975 * made to update_rq_clock() since the last time rq::lock was pinned.
977 * If inside of __schedule(), clock_update_flags will have been
978 * shifted left (a left shift is a cheap operation for the fast path
979 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
981 * if (rq-clock_update_flags >= RQCF_UPDATED)
983 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
984 * one position though, because the next rq_unpin_lock() will shift it
987 #define RQCF_REQ_SKIP 0x01
988 #define RQCF_ACT_SKIP 0x02
989 #define RQCF_UPDATED 0x04
991 static inline void assert_clock_updated(struct rq *rq)
994 * The only reason for not seeing a clock update since the
995 * last rq_pin_lock() is if we're currently skipping updates.
997 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1000 static inline u64 rq_clock(struct rq *rq)
1002 lockdep_assert_held(&rq->lock);
1003 assert_clock_updated(rq);
1008 static inline u64 rq_clock_task(struct rq *rq)
1010 lockdep_assert_held(&rq->lock);
1011 assert_clock_updated(rq);
1013 return rq->clock_task;
1016 static inline void rq_clock_skip_update(struct rq *rq)
1018 lockdep_assert_held(&rq->lock);
1019 rq->clock_update_flags |= RQCF_REQ_SKIP;
1023 * See rt task throttling, which is the only time a skip
1024 * request is cancelled.
1026 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1028 lockdep_assert_held(&rq->lock);
1029 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1033 unsigned long flags;
1034 struct pin_cookie cookie;
1035 #ifdef CONFIG_SCHED_DEBUG
1037 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1038 * current pin context is stashed here in case it needs to be
1039 * restored in rq_repin_lock().
1041 unsigned int clock_update_flags;
1045 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1047 rf->cookie = lockdep_pin_lock(&rq->lock);
1049 #ifdef CONFIG_SCHED_DEBUG
1050 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1051 rf->clock_update_flags = 0;
1055 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1057 #ifdef CONFIG_SCHED_DEBUG
1058 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1059 rf->clock_update_flags = RQCF_UPDATED;
1062 lockdep_unpin_lock(&rq->lock, rf->cookie);
1065 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1067 lockdep_repin_lock(&rq->lock, rf->cookie);
1069 #ifdef CONFIG_SCHED_DEBUG
1071 * Restore the value we stashed in @rf for this pin context.
1073 rq->clock_update_flags |= rf->clock_update_flags;
1078 enum numa_topology_type {
1083 extern enum numa_topology_type sched_numa_topology_type;
1084 extern int sched_max_numa_distance;
1085 extern bool find_numa_distance(int distance);
1089 extern void sched_init_numa(void);
1090 extern void sched_domains_numa_masks_set(unsigned int cpu);
1091 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1093 static inline void sched_init_numa(void) { }
1094 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1095 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1098 #ifdef CONFIG_NUMA_BALANCING
1099 /* The regions in numa_faults array from task_struct */
1100 enum numa_faults_stats {
1106 extern void sched_setnuma(struct task_struct *p, int node);
1107 extern int migrate_task_to(struct task_struct *p, int cpu);
1108 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1110 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1113 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1116 #endif /* CONFIG_NUMA_BALANCING */
1121 queue_balance_callback(struct rq *rq,
1122 struct callback_head *head,
1123 void (*func)(struct rq *rq))
1125 lockdep_assert_held(&rq->lock);
1127 if (unlikely(head->next))
1130 head->func = (void (*)(struct callback_head *))func;
1131 head->next = rq->balance_callback;
1132 rq->balance_callback = head;
1135 extern void sched_ttwu_pending(void);
1137 #define rcu_dereference_check_sched_domain(p) \
1138 rcu_dereference_check((p), \
1139 lockdep_is_held(&sched_domains_mutex))
1142 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1143 * See detach_destroy_domains: synchronize_sched for details.
1145 * The domain tree of any CPU may only be accessed from within
1146 * preempt-disabled sections.
1148 #define for_each_domain(cpu, __sd) \
1149 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1150 __sd; __sd = __sd->parent)
1152 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1155 * highest_flag_domain - Return highest sched_domain containing flag.
1156 * @cpu: The CPU whose highest level of sched domain is to
1158 * @flag: The flag to check for the highest sched_domain
1159 * for the given CPU.
1161 * Returns the highest sched_domain of a CPU which contains the given flag.
1163 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1165 struct sched_domain *sd, *hsd = NULL;
1167 for_each_domain(cpu, sd) {
1168 if (!(sd->flags & flag))
1176 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1178 struct sched_domain *sd;
1180 for_each_domain(cpu, sd) {
1181 if (sd->flags & flag)
1188 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1189 DECLARE_PER_CPU(int, sd_llc_size);
1190 DECLARE_PER_CPU(int, sd_llc_id);
1191 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1192 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1193 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1194 extern struct static_key_false sched_asym_cpucapacity;
1196 struct sched_group_capacity {
1199 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1202 unsigned long capacity;
1203 unsigned long min_capacity; /* Min per-CPU capacity in group */
1204 unsigned long max_capacity; /* Max per-CPU capacity in group */
1205 unsigned long next_update;
1206 int imbalance; /* XXX unrelated to capacity but shared group state */
1208 #ifdef CONFIG_SCHED_DEBUG
1212 unsigned long cpumask[0]; /* Balance mask */
1215 struct sched_group {
1216 struct sched_group *next; /* Must be a circular list */
1219 unsigned int group_weight;
1220 struct sched_group_capacity *sgc;
1221 int asym_prefer_cpu; /* CPU of highest priority in group */
1224 * The CPUs this group covers.
1226 * NOTE: this field is variable length. (Allocated dynamically
1227 * by attaching extra space to the end of the structure,
1228 * depending on how many CPUs the kernel has booted up with)
1230 unsigned long cpumask[0];
1233 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1235 return to_cpumask(sg->cpumask);
1239 * See build_balance_mask().
1241 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1243 return to_cpumask(sg->sgc->cpumask);
1247 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1248 * @group: The group whose first CPU is to be returned.
1250 static inline unsigned int group_first_cpu(struct sched_group *group)
1252 return cpumask_first(sched_group_span(group));
1255 extern int group_balance_cpu(struct sched_group *sg);
1257 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1258 void register_sched_domain_sysctl(void);
1259 void dirty_sched_domain_sysctl(int cpu);
1260 void unregister_sched_domain_sysctl(void);
1262 static inline void register_sched_domain_sysctl(void)
1265 static inline void dirty_sched_domain_sysctl(int cpu)
1268 static inline void unregister_sched_domain_sysctl(void)
1275 static inline void sched_ttwu_pending(void) { }
1277 #endif /* CONFIG_SMP */
1280 #include "autogroup.h"
1282 #ifdef CONFIG_CGROUP_SCHED
1285 * Return the group to which this tasks belongs.
1287 * We cannot use task_css() and friends because the cgroup subsystem
1288 * changes that value before the cgroup_subsys::attach() method is called,
1289 * therefore we cannot pin it and might observe the wrong value.
1291 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1292 * core changes this before calling sched_move_task().
1294 * Instead we use a 'copy' which is updated from sched_move_task() while
1295 * holding both task_struct::pi_lock and rq::lock.
1297 static inline struct task_group *task_group(struct task_struct *p)
1299 return p->sched_task_group;
1302 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1303 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1305 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1306 struct task_group *tg = task_group(p);
1309 #ifdef CONFIG_FAIR_GROUP_SCHED
1310 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1311 p->se.cfs_rq = tg->cfs_rq[cpu];
1312 p->se.parent = tg->se[cpu];
1315 #ifdef CONFIG_RT_GROUP_SCHED
1316 p->rt.rt_rq = tg->rt_rq[cpu];
1317 p->rt.parent = tg->rt_se[cpu];
1321 #else /* CONFIG_CGROUP_SCHED */
1323 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1324 static inline struct task_group *task_group(struct task_struct *p)
1329 #endif /* CONFIG_CGROUP_SCHED */
1331 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1333 set_task_rq(p, cpu);
1336 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1337 * successfuly executed on another CPU. We must ensure that updates of
1338 * per-task data have been completed by this moment.
1341 #ifdef CONFIG_THREAD_INFO_IN_TASK
1344 task_thread_info(p)->cpu = cpu;
1351 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1353 #ifdef CONFIG_SCHED_DEBUG
1354 # include <linux/static_key.h>
1355 # define const_debug __read_mostly
1357 # define const_debug const
1360 #define SCHED_FEAT(name, enabled) \
1361 __SCHED_FEAT_##name ,
1364 #include "features.h"
1370 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1373 * To support run-time toggling of sched features, all the translation units
1374 * (but core.c) reference the sysctl_sched_features defined in core.c.
1376 extern const_debug unsigned int sysctl_sched_features;
1378 #define SCHED_FEAT(name, enabled) \
1379 static __always_inline bool static_branch_##name(struct static_key *key) \
1381 return static_key_##enabled(key); \
1384 #include "features.h"
1387 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1388 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1390 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1393 * Each translation unit has its own copy of sysctl_sched_features to allow
1394 * constants propagation at compile time and compiler optimization based on
1397 #define SCHED_FEAT(name, enabled) \
1398 (1UL << __SCHED_FEAT_##name) * enabled |
1399 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1400 #include "features.h"
1404 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1406 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1408 extern struct static_key_false sched_numa_balancing;
1409 extern struct static_key_false sched_schedstats;
1411 static inline u64 global_rt_period(void)
1413 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1416 static inline u64 global_rt_runtime(void)
1418 if (sysctl_sched_rt_runtime < 0)
1421 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1424 static inline int task_current(struct rq *rq, struct task_struct *p)
1426 return rq->curr == p;
1429 static inline int task_running(struct rq *rq, struct task_struct *p)
1434 return task_current(rq, p);
1438 static inline int task_on_rq_queued(struct task_struct *p)
1440 return p->on_rq == TASK_ON_RQ_QUEUED;
1443 static inline int task_on_rq_migrating(struct task_struct *p)
1445 return p->on_rq == TASK_ON_RQ_MIGRATING;
1451 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1452 #define WF_FORK 0x02 /* Child wakeup after fork */
1453 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1456 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1457 * of tasks with abnormal "nice" values across CPUs the contribution that
1458 * each task makes to its run queue's load is weighted according to its
1459 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1460 * scaled version of the new time slice allocation that they receive on time
1464 #define WEIGHT_IDLEPRIO 3
1465 #define WMULT_IDLEPRIO 1431655765
1467 extern const int sched_prio_to_weight[40];
1468 extern const u32 sched_prio_to_wmult[40];
1471 * {de,en}queue flags:
1473 * DEQUEUE_SLEEP - task is no longer runnable
1474 * ENQUEUE_WAKEUP - task just became runnable
1476 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1477 * are in a known state which allows modification. Such pairs
1478 * should preserve as much state as possible.
1480 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1483 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1484 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1485 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1489 #define DEQUEUE_SLEEP 0x01
1490 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1491 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1492 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1494 #define ENQUEUE_WAKEUP 0x01
1495 #define ENQUEUE_RESTORE 0x02
1496 #define ENQUEUE_MOVE 0x04
1497 #define ENQUEUE_NOCLOCK 0x08
1499 #define ENQUEUE_HEAD 0x10
1500 #define ENQUEUE_REPLENISH 0x20
1502 #define ENQUEUE_MIGRATED 0x40
1504 #define ENQUEUE_MIGRATED 0x00
1507 #define RETRY_TASK ((void *)-1UL)
1509 struct sched_class {
1510 const struct sched_class *next;
1512 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1513 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1514 void (*yield_task) (struct rq *rq);
1515 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1517 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1520 * It is the responsibility of the pick_next_task() method that will
1521 * return the next task to call put_prev_task() on the @prev task or
1522 * something equivalent.
1524 * May return RETRY_TASK when it finds a higher prio class has runnable
1527 struct task_struct * (*pick_next_task)(struct rq *rq,
1528 struct task_struct *prev,
1529 struct rq_flags *rf);
1530 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1533 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1534 void (*migrate_task_rq)(struct task_struct *p);
1536 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1538 void (*set_cpus_allowed)(struct task_struct *p,
1539 const struct cpumask *newmask);
1541 void (*rq_online)(struct rq *rq);
1542 void (*rq_offline)(struct rq *rq);
1545 void (*set_curr_task)(struct rq *rq);
1546 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1547 void (*task_fork)(struct task_struct *p);
1548 void (*task_dead)(struct task_struct *p);
1551 * The switched_from() call is allowed to drop rq->lock, therefore we
1552 * cannot assume the switched_from/switched_to pair is serliazed by
1553 * rq->lock. They are however serialized by p->pi_lock.
1555 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1556 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1557 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1560 unsigned int (*get_rr_interval)(struct rq *rq,
1561 struct task_struct *task);
1563 void (*update_curr)(struct rq *rq);
1565 #define TASK_SET_GROUP 0
1566 #define TASK_MOVE_GROUP 1
1568 #ifdef CONFIG_FAIR_GROUP_SCHED
1569 void (*task_change_group)(struct task_struct *p, int type);
1573 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1575 prev->sched_class->put_prev_task(rq, prev);
1578 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1580 curr->sched_class->set_curr_task(rq);
1584 #define sched_class_highest (&stop_sched_class)
1586 #define sched_class_highest (&dl_sched_class)
1588 #define for_each_class(class) \
1589 for (class = sched_class_highest; class; class = class->next)
1591 extern const struct sched_class stop_sched_class;
1592 extern const struct sched_class dl_sched_class;
1593 extern const struct sched_class rt_sched_class;
1594 extern const struct sched_class fair_sched_class;
1595 extern const struct sched_class idle_sched_class;
1600 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1602 extern void trigger_load_balance(struct rq *rq);
1604 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1608 #ifdef CONFIG_CPU_IDLE
1609 static inline void idle_set_state(struct rq *rq,
1610 struct cpuidle_state *idle_state)
1612 rq->idle_state = idle_state;
1615 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1617 SCHED_WARN_ON(!rcu_read_lock_held());
1619 return rq->idle_state;
1622 static inline void idle_set_state(struct rq *rq,
1623 struct cpuidle_state *idle_state)
1627 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1633 extern void schedule_idle(void);
1635 extern void sysrq_sched_debug_show(void);
1636 extern void sched_init_granularity(void);
1637 extern void update_max_interval(void);
1639 extern void init_sched_dl_class(void);
1640 extern void init_sched_rt_class(void);
1641 extern void init_sched_fair_class(void);
1643 extern void reweight_task(struct task_struct *p, int prio);
1645 extern void resched_curr(struct rq *rq);
1646 extern void resched_cpu(int cpu);
1648 extern struct rt_bandwidth def_rt_bandwidth;
1649 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1651 extern struct dl_bandwidth def_dl_bandwidth;
1652 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1653 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1654 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1655 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1658 #define BW_UNIT (1 << BW_SHIFT)
1659 #define RATIO_SHIFT 8
1660 unsigned long to_ratio(u64 period, u64 runtime);
1662 extern void init_entity_runnable_average(struct sched_entity *se);
1663 extern void post_init_entity_util_avg(struct sched_entity *se);
1665 #ifdef CONFIG_NO_HZ_FULL
1666 extern bool sched_can_stop_tick(struct rq *rq);
1667 extern int __init sched_tick_offload_init(void);
1670 * Tick may be needed by tasks in the runqueue depending on their policy and
1671 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1672 * nohz mode if necessary.
1674 static inline void sched_update_tick_dependency(struct rq *rq)
1678 if (!tick_nohz_full_enabled())
1683 if (!tick_nohz_full_cpu(cpu))
1686 if (sched_can_stop_tick(rq))
1687 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1689 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1692 static inline int sched_tick_offload_init(void) { return 0; }
1693 static inline void sched_update_tick_dependency(struct rq *rq) { }
1696 static inline void add_nr_running(struct rq *rq, unsigned count)
1698 unsigned prev_nr = rq->nr_running;
1700 rq->nr_running = prev_nr + count;
1702 if (prev_nr < 2 && rq->nr_running >= 2) {
1704 if (!READ_ONCE(rq->rd->overload))
1705 WRITE_ONCE(rq->rd->overload, 1);
1709 sched_update_tick_dependency(rq);
1712 static inline void sub_nr_running(struct rq *rq, unsigned count)
1714 rq->nr_running -= count;
1715 /* Check if we still need preemption */
1716 sched_update_tick_dependency(rq);
1719 extern void update_rq_clock(struct rq *rq);
1721 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1722 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1724 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1726 extern const_debug unsigned int sysctl_sched_nr_migrate;
1727 extern const_debug unsigned int sysctl_sched_migration_cost;
1729 #ifdef CONFIG_SCHED_HRTICK
1733 * - enabled by features
1734 * - hrtimer is actually high res
1736 static inline int hrtick_enabled(struct rq *rq)
1738 if (!sched_feat(HRTICK))
1740 if (!cpu_active(cpu_of(rq)))
1742 return hrtimer_is_hres_active(&rq->hrtick_timer);
1745 void hrtick_start(struct rq *rq, u64 delay);
1749 static inline int hrtick_enabled(struct rq *rq)
1754 #endif /* CONFIG_SCHED_HRTICK */
1756 #ifndef arch_scale_freq_capacity
1757 static __always_inline
1758 unsigned long arch_scale_freq_capacity(int cpu)
1760 return SCHED_CAPACITY_SCALE;
1765 #ifndef arch_scale_cpu_capacity
1766 static __always_inline
1767 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1769 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1770 return sd->smt_gain / sd->span_weight;
1772 return SCHED_CAPACITY_SCALE;
1776 #ifndef arch_scale_cpu_capacity
1777 static __always_inline
1778 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1780 return SCHED_CAPACITY_SCALE;
1785 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1786 __acquires(rq->lock);
1788 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1789 __acquires(p->pi_lock)
1790 __acquires(rq->lock);
1792 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1793 __releases(rq->lock)
1795 rq_unpin_lock(rq, rf);
1796 raw_spin_unlock(&rq->lock);
1800 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1801 __releases(rq->lock)
1802 __releases(p->pi_lock)
1804 rq_unpin_lock(rq, rf);
1805 raw_spin_unlock(&rq->lock);
1806 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1810 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1811 __acquires(rq->lock)
1813 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1814 rq_pin_lock(rq, rf);
1818 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1819 __acquires(rq->lock)
1821 raw_spin_lock_irq(&rq->lock);
1822 rq_pin_lock(rq, rf);
1826 rq_lock(struct rq *rq, struct rq_flags *rf)
1827 __acquires(rq->lock)
1829 raw_spin_lock(&rq->lock);
1830 rq_pin_lock(rq, rf);
1834 rq_relock(struct rq *rq, struct rq_flags *rf)
1835 __acquires(rq->lock)
1837 raw_spin_lock(&rq->lock);
1838 rq_repin_lock(rq, rf);
1842 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1843 __releases(rq->lock)
1845 rq_unpin_lock(rq, rf);
1846 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1850 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1851 __releases(rq->lock)
1853 rq_unpin_lock(rq, rf);
1854 raw_spin_unlock_irq(&rq->lock);
1858 rq_unlock(struct rq *rq, struct rq_flags *rf)
1859 __releases(rq->lock)
1861 rq_unpin_lock(rq, rf);
1862 raw_spin_unlock(&rq->lock);
1866 #ifdef CONFIG_PREEMPT
1868 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1871 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1872 * way at the expense of forcing extra atomic operations in all
1873 * invocations. This assures that the double_lock is acquired using the
1874 * same underlying policy as the spinlock_t on this architecture, which
1875 * reduces latency compared to the unfair variant below. However, it
1876 * also adds more overhead and therefore may reduce throughput.
1878 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1879 __releases(this_rq->lock)
1880 __acquires(busiest->lock)
1881 __acquires(this_rq->lock)
1883 raw_spin_unlock(&this_rq->lock);
1884 double_rq_lock(this_rq, busiest);
1891 * Unfair double_lock_balance: Optimizes throughput at the expense of
1892 * latency by eliminating extra atomic operations when the locks are
1893 * already in proper order on entry. This favors lower CPU-ids and will
1894 * grant the double lock to lower CPUs over higher ids under contention,
1895 * regardless of entry order into the function.
1897 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1898 __releases(this_rq->lock)
1899 __acquires(busiest->lock)
1900 __acquires(this_rq->lock)
1904 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1905 if (busiest < this_rq) {
1906 raw_spin_unlock(&this_rq->lock);
1907 raw_spin_lock(&busiest->lock);
1908 raw_spin_lock_nested(&this_rq->lock,
1909 SINGLE_DEPTH_NESTING);
1912 raw_spin_lock_nested(&busiest->lock,
1913 SINGLE_DEPTH_NESTING);
1918 #endif /* CONFIG_PREEMPT */
1921 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1923 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1925 if (unlikely(!irqs_disabled())) {
1926 /* printk() doesn't work well under rq->lock */
1927 raw_spin_unlock(&this_rq->lock);
1931 return _double_lock_balance(this_rq, busiest);
1934 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1935 __releases(busiest->lock)
1937 raw_spin_unlock(&busiest->lock);
1938 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1941 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1947 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1950 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1956 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1959 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1965 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1969 * double_rq_lock - safely lock two runqueues
1971 * Note this does not disable interrupts like task_rq_lock,
1972 * you need to do so manually before calling.
1974 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1975 __acquires(rq1->lock)
1976 __acquires(rq2->lock)
1978 BUG_ON(!irqs_disabled());
1980 raw_spin_lock(&rq1->lock);
1981 __acquire(rq2->lock); /* Fake it out ;) */
1984 raw_spin_lock(&rq1->lock);
1985 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1987 raw_spin_lock(&rq2->lock);
1988 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1994 * double_rq_unlock - safely unlock two runqueues
1996 * Note this does not restore interrupts like task_rq_unlock,
1997 * you need to do so manually after calling.
1999 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2000 __releases(rq1->lock)
2001 __releases(rq2->lock)
2003 raw_spin_unlock(&rq1->lock);
2005 raw_spin_unlock(&rq2->lock);
2007 __release(rq2->lock);
2010 extern void set_rq_online (struct rq *rq);
2011 extern void set_rq_offline(struct rq *rq);
2012 extern bool sched_smp_initialized;
2014 #else /* CONFIG_SMP */
2017 * double_rq_lock - safely lock two runqueues
2019 * Note this does not disable interrupts like task_rq_lock,
2020 * you need to do so manually before calling.
2022 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2023 __acquires(rq1->lock)
2024 __acquires(rq2->lock)
2026 BUG_ON(!irqs_disabled());
2028 raw_spin_lock(&rq1->lock);
2029 __acquire(rq2->lock); /* Fake it out ;) */
2033 * double_rq_unlock - safely unlock two runqueues
2035 * Note this does not restore interrupts like task_rq_unlock,
2036 * you need to do so manually after calling.
2038 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2039 __releases(rq1->lock)
2040 __releases(rq2->lock)
2043 raw_spin_unlock(&rq1->lock);
2044 __release(rq2->lock);
2049 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2050 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2052 #ifdef CONFIG_SCHED_DEBUG
2053 extern bool sched_debug_enabled;
2055 extern void print_cfs_stats(struct seq_file *m, int cpu);
2056 extern void print_rt_stats(struct seq_file *m, int cpu);
2057 extern void print_dl_stats(struct seq_file *m, int cpu);
2058 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2059 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2060 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2061 #ifdef CONFIG_NUMA_BALANCING
2063 show_numa_stats(struct task_struct *p, struct seq_file *m);
2065 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2066 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2067 #endif /* CONFIG_NUMA_BALANCING */
2068 #endif /* CONFIG_SCHED_DEBUG */
2070 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2071 extern void init_rt_rq(struct rt_rq *rt_rq);
2072 extern void init_dl_rq(struct dl_rq *dl_rq);
2074 extern void cfs_bandwidth_usage_inc(void);
2075 extern void cfs_bandwidth_usage_dec(void);
2077 #ifdef CONFIG_NO_HZ_COMMON
2078 #define NOHZ_BALANCE_KICK_BIT 0
2079 #define NOHZ_STATS_KICK_BIT 1
2081 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2082 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2084 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2086 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2088 extern void nohz_balance_exit_idle(struct rq *rq);
2090 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2096 void __dl_update(struct dl_bw *dl_b, s64 bw)
2098 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2101 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2102 "sched RCU must be held");
2103 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2104 struct rq *rq = cpu_rq(i);
2106 rq->dl.extra_bw += bw;
2111 void __dl_update(struct dl_bw *dl_b, s64 bw)
2113 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2120 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2125 struct u64_stats_sync sync;
2128 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2131 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2132 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2133 * and never move forward.
2135 static inline u64 irq_time_read(int cpu)
2137 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2142 seq = __u64_stats_fetch_begin(&irqtime->sync);
2143 total = irqtime->total;
2144 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2148 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2150 #ifdef CONFIG_CPU_FREQ
2151 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2154 * cpufreq_update_util - Take a note about CPU utilization changes.
2155 * @rq: Runqueue to carry out the update for.
2156 * @flags: Update reason flags.
2158 * This function is called by the scheduler on the CPU whose utilization is
2161 * It can only be called from RCU-sched read-side critical sections.
2163 * The way cpufreq is currently arranged requires it to evaluate the CPU
2164 * performance state (frequency/voltage) on a regular basis to prevent it from
2165 * being stuck in a completely inadequate performance level for too long.
2166 * That is not guaranteed to happen if the updates are only triggered from CFS
2167 * and DL, though, because they may not be coming in if only RT tasks are
2168 * active all the time (or there are RT tasks only).
2170 * As a workaround for that issue, this function is called periodically by the
2171 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2172 * but that really is a band-aid. Going forward it should be replaced with
2173 * solutions targeted more specifically at RT tasks.
2175 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2177 struct update_util_data *data;
2179 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2182 data->func(data, rq_clock(rq), flags);
2185 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2186 #endif /* CONFIG_CPU_FREQ */
2188 #ifdef arch_scale_freq_capacity
2189 # ifndef arch_scale_freq_invariant
2190 # define arch_scale_freq_invariant() true
2193 # define arch_scale_freq_invariant() false
2196 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2197 static inline unsigned long cpu_bw_dl(struct rq *rq)
2199 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2202 static inline unsigned long cpu_util_dl(struct rq *rq)
2204 return READ_ONCE(rq->avg_dl.util_avg);
2207 static inline unsigned long cpu_util_cfs(struct rq *rq)
2209 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2211 if (sched_feat(UTIL_EST)) {
2212 util = max_t(unsigned long, util,
2213 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2219 static inline unsigned long cpu_util_rt(struct rq *rq)
2221 return READ_ONCE(rq->avg_rt.util_avg);
2225 #ifdef HAVE_SCHED_AVG_IRQ
2226 static inline unsigned long cpu_util_irq(struct rq *rq)
2228 return rq->avg_irq.util_avg;
2232 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2234 util *= (max - irq);
2241 static inline unsigned long cpu_util_irq(struct rq *rq)
2247 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)