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;
790 unsigned int numa_migrate_on;
792 #define CPU_LOAD_IDX_MAX 5
793 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
794 #ifdef CONFIG_NO_HZ_COMMON
796 unsigned long last_load_update_tick;
797 unsigned long last_blocked_load_update_tick;
798 unsigned int has_blocked_load;
799 #endif /* CONFIG_SMP */
800 unsigned int nohz_tick_stopped;
802 #endif /* CONFIG_NO_HZ_COMMON */
804 /* capture load from *all* tasks on this CPU: */
805 struct load_weight load;
806 unsigned long nr_load_updates;
813 #ifdef CONFIG_FAIR_GROUP_SCHED
814 /* list of leaf cfs_rq on this CPU: */
815 struct list_head leaf_cfs_rq_list;
816 struct list_head *tmp_alone_branch;
817 #endif /* CONFIG_FAIR_GROUP_SCHED */
820 * This is part of a global counter where only the total sum
821 * over all CPUs matters. A task can increase this counter on
822 * one CPU and if it got migrated afterwards it may decrease
823 * it on another CPU. Always updated under the runqueue lock:
825 unsigned long nr_uninterruptible;
827 struct task_struct *curr;
828 struct task_struct *idle;
829 struct task_struct *stop;
830 unsigned long next_balance;
831 struct mm_struct *prev_mm;
833 unsigned int clock_update_flags;
840 struct root_domain *rd;
841 struct sched_domain *sd;
843 unsigned long cpu_capacity;
844 unsigned long cpu_capacity_orig;
846 struct callback_head *balance_callback;
848 unsigned char idle_balance;
850 unsigned long misfit_task_load;
852 /* For active balancing */
855 struct cpu_stop_work active_balance_work;
857 /* CPU of this runqueue: */
861 struct list_head cfs_tasks;
863 struct sched_avg avg_rt;
864 struct sched_avg avg_dl;
865 #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
866 #define HAVE_SCHED_AVG_IRQ
867 struct sched_avg avg_irq;
872 /* This is used to determine avg_idle's max value */
873 u64 max_idle_balance_cost;
876 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
879 #ifdef CONFIG_PARAVIRT
882 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
883 u64 prev_steal_time_rq;
886 /* calc_load related fields */
887 unsigned long calc_load_update;
888 long calc_load_active;
890 #ifdef CONFIG_SCHED_HRTICK
892 int hrtick_csd_pending;
893 call_single_data_t hrtick_csd;
895 struct hrtimer hrtick_timer;
898 #ifdef CONFIG_SCHEDSTATS
900 struct sched_info rq_sched_info;
901 unsigned long long rq_cpu_time;
902 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
904 /* sys_sched_yield() stats */
905 unsigned int yld_count;
907 /* schedule() stats */
908 unsigned int sched_count;
909 unsigned int sched_goidle;
911 /* try_to_wake_up() stats */
912 unsigned int ttwu_count;
913 unsigned int ttwu_local;
917 struct llist_head wake_list;
920 #ifdef CONFIG_CPU_IDLE
921 /* Must be inspected within a rcu lock section */
922 struct cpuidle_state *idle_state;
926 static inline int cpu_of(struct rq *rq)
936 #ifdef CONFIG_SCHED_SMT
938 extern struct static_key_false sched_smt_present;
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 static inline u64 __rq_clock_broken(struct rq *rq)
962 return READ_ONCE(rq->clock);
966 * rq::clock_update_flags bits
968 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
969 * call to __schedule(). This is an optimisation to avoid
970 * neighbouring rq clock updates.
972 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
973 * in effect and calls to update_rq_clock() are being ignored.
975 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
976 * made to update_rq_clock() since the last time rq::lock was pinned.
978 * If inside of __schedule(), clock_update_flags will have been
979 * shifted left (a left shift is a cheap operation for the fast path
980 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
982 * if (rq-clock_update_flags >= RQCF_UPDATED)
984 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
985 * one position though, because the next rq_unpin_lock() will shift it
988 #define RQCF_REQ_SKIP 0x01
989 #define RQCF_ACT_SKIP 0x02
990 #define RQCF_UPDATED 0x04
992 static inline void assert_clock_updated(struct rq *rq)
995 * The only reason for not seeing a clock update since the
996 * last rq_pin_lock() is if we're currently skipping updates.
998 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1001 static inline u64 rq_clock(struct rq *rq)
1003 lockdep_assert_held(&rq->lock);
1004 assert_clock_updated(rq);
1009 static inline u64 rq_clock_task(struct rq *rq)
1011 lockdep_assert_held(&rq->lock);
1012 assert_clock_updated(rq);
1014 return rq->clock_task;
1017 static inline void rq_clock_skip_update(struct rq *rq)
1019 lockdep_assert_held(&rq->lock);
1020 rq->clock_update_flags |= RQCF_REQ_SKIP;
1024 * See rt task throttling, which is the only time a skip
1025 * request is cancelled.
1027 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1029 lockdep_assert_held(&rq->lock);
1030 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1034 unsigned long flags;
1035 struct pin_cookie cookie;
1036 #ifdef CONFIG_SCHED_DEBUG
1038 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1039 * current pin context is stashed here in case it needs to be
1040 * restored in rq_repin_lock().
1042 unsigned int clock_update_flags;
1046 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1048 rf->cookie = lockdep_pin_lock(&rq->lock);
1050 #ifdef CONFIG_SCHED_DEBUG
1051 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1052 rf->clock_update_flags = 0;
1056 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1058 #ifdef CONFIG_SCHED_DEBUG
1059 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1060 rf->clock_update_flags = RQCF_UPDATED;
1063 lockdep_unpin_lock(&rq->lock, rf->cookie);
1066 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1068 lockdep_repin_lock(&rq->lock, rf->cookie);
1070 #ifdef CONFIG_SCHED_DEBUG
1072 * Restore the value we stashed in @rf for this pin context.
1074 rq->clock_update_flags |= rf->clock_update_flags;
1079 enum numa_topology_type {
1084 extern enum numa_topology_type sched_numa_topology_type;
1085 extern int sched_max_numa_distance;
1086 extern bool find_numa_distance(int distance);
1090 extern void sched_init_numa(void);
1091 extern void sched_domains_numa_masks_set(unsigned int cpu);
1092 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1094 static inline void sched_init_numa(void) { }
1095 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1096 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1099 #ifdef CONFIG_NUMA_BALANCING
1100 /* The regions in numa_faults array from task_struct */
1101 enum numa_faults_stats {
1107 extern void sched_setnuma(struct task_struct *p, int node);
1108 extern int migrate_task_to(struct task_struct *p, int cpu);
1109 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1111 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1114 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1117 #endif /* CONFIG_NUMA_BALANCING */
1122 queue_balance_callback(struct rq *rq,
1123 struct callback_head *head,
1124 void (*func)(struct rq *rq))
1126 lockdep_assert_held(&rq->lock);
1128 if (unlikely(head->next))
1131 head->func = (void (*)(struct callback_head *))func;
1132 head->next = rq->balance_callback;
1133 rq->balance_callback = head;
1136 extern void sched_ttwu_pending(void);
1138 #define rcu_dereference_check_sched_domain(p) \
1139 rcu_dereference_check((p), \
1140 lockdep_is_held(&sched_domains_mutex))
1143 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1144 * See detach_destroy_domains: synchronize_sched for details.
1146 * The domain tree of any CPU may only be accessed from within
1147 * preempt-disabled sections.
1149 #define for_each_domain(cpu, __sd) \
1150 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1151 __sd; __sd = __sd->parent)
1153 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1156 * highest_flag_domain - Return highest sched_domain containing flag.
1157 * @cpu: The CPU whose highest level of sched domain is to
1159 * @flag: The flag to check for the highest sched_domain
1160 * for the given CPU.
1162 * Returns the highest sched_domain of a CPU which contains the given flag.
1164 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1166 struct sched_domain *sd, *hsd = NULL;
1168 for_each_domain(cpu, sd) {
1169 if (!(sd->flags & flag))
1177 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1179 struct sched_domain *sd;
1181 for_each_domain(cpu, sd) {
1182 if (sd->flags & flag)
1189 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1190 DECLARE_PER_CPU(int, sd_llc_size);
1191 DECLARE_PER_CPU(int, sd_llc_id);
1192 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1193 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1194 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1195 extern struct static_key_false sched_asym_cpucapacity;
1197 struct sched_group_capacity {
1200 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1203 unsigned long capacity;
1204 unsigned long min_capacity; /* Min per-CPU capacity in group */
1205 unsigned long max_capacity; /* Max per-CPU capacity in group */
1206 unsigned long next_update;
1207 int imbalance; /* XXX unrelated to capacity but shared group state */
1209 #ifdef CONFIG_SCHED_DEBUG
1213 unsigned long cpumask[0]; /* Balance mask */
1216 struct sched_group {
1217 struct sched_group *next; /* Must be a circular list */
1220 unsigned int group_weight;
1221 struct sched_group_capacity *sgc;
1222 int asym_prefer_cpu; /* CPU of highest priority in group */
1225 * The CPUs this group covers.
1227 * NOTE: this field is variable length. (Allocated dynamically
1228 * by attaching extra space to the end of the structure,
1229 * depending on how many CPUs the kernel has booted up with)
1231 unsigned long cpumask[0];
1234 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1236 return to_cpumask(sg->cpumask);
1240 * See build_balance_mask().
1242 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1244 return to_cpumask(sg->sgc->cpumask);
1248 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1249 * @group: The group whose first CPU is to be returned.
1251 static inline unsigned int group_first_cpu(struct sched_group *group)
1253 return cpumask_first(sched_group_span(group));
1256 extern int group_balance_cpu(struct sched_group *sg);
1258 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1259 void register_sched_domain_sysctl(void);
1260 void dirty_sched_domain_sysctl(int cpu);
1261 void unregister_sched_domain_sysctl(void);
1263 static inline void register_sched_domain_sysctl(void)
1266 static inline void dirty_sched_domain_sysctl(int cpu)
1269 static inline void unregister_sched_domain_sysctl(void)
1276 static inline void sched_ttwu_pending(void) { }
1278 #endif /* CONFIG_SMP */
1281 #include "autogroup.h"
1283 #ifdef CONFIG_CGROUP_SCHED
1286 * Return the group to which this tasks belongs.
1288 * We cannot use task_css() and friends because the cgroup subsystem
1289 * changes that value before the cgroup_subsys::attach() method is called,
1290 * therefore we cannot pin it and might observe the wrong value.
1292 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1293 * core changes this before calling sched_move_task().
1295 * Instead we use a 'copy' which is updated from sched_move_task() while
1296 * holding both task_struct::pi_lock and rq::lock.
1298 static inline struct task_group *task_group(struct task_struct *p)
1300 return p->sched_task_group;
1303 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1304 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1306 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1307 struct task_group *tg = task_group(p);
1310 #ifdef CONFIG_FAIR_GROUP_SCHED
1311 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1312 p->se.cfs_rq = tg->cfs_rq[cpu];
1313 p->se.parent = tg->se[cpu];
1316 #ifdef CONFIG_RT_GROUP_SCHED
1317 p->rt.rt_rq = tg->rt_rq[cpu];
1318 p->rt.parent = tg->rt_se[cpu];
1322 #else /* CONFIG_CGROUP_SCHED */
1324 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1325 static inline struct task_group *task_group(struct task_struct *p)
1330 #endif /* CONFIG_CGROUP_SCHED */
1332 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1334 set_task_rq(p, cpu);
1337 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1338 * successfuly executed on another CPU. We must ensure that updates of
1339 * per-task data have been completed by this moment.
1342 #ifdef CONFIG_THREAD_INFO_IN_TASK
1345 task_thread_info(p)->cpu = cpu;
1352 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1354 #ifdef CONFIG_SCHED_DEBUG
1355 # include <linux/static_key.h>
1356 # define const_debug __read_mostly
1358 # define const_debug const
1361 #define SCHED_FEAT(name, enabled) \
1362 __SCHED_FEAT_##name ,
1365 #include "features.h"
1371 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1374 * To support run-time toggling of sched features, all the translation units
1375 * (but core.c) reference the sysctl_sched_features defined in core.c.
1377 extern const_debug unsigned int sysctl_sched_features;
1379 #define SCHED_FEAT(name, enabled) \
1380 static __always_inline bool static_branch_##name(struct static_key *key) \
1382 return static_key_##enabled(key); \
1385 #include "features.h"
1388 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1389 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1391 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1394 * Each translation unit has its own copy of sysctl_sched_features to allow
1395 * constants propagation at compile time and compiler optimization based on
1398 #define SCHED_FEAT(name, enabled) \
1399 (1UL << __SCHED_FEAT_##name) * enabled |
1400 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1401 #include "features.h"
1405 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1407 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1409 extern struct static_key_false sched_numa_balancing;
1410 extern struct static_key_false sched_schedstats;
1412 static inline u64 global_rt_period(void)
1414 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1417 static inline u64 global_rt_runtime(void)
1419 if (sysctl_sched_rt_runtime < 0)
1422 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1425 static inline int task_current(struct rq *rq, struct task_struct *p)
1427 return rq->curr == p;
1430 static inline int task_running(struct rq *rq, struct task_struct *p)
1435 return task_current(rq, p);
1439 static inline int task_on_rq_queued(struct task_struct *p)
1441 return p->on_rq == TASK_ON_RQ_QUEUED;
1444 static inline int task_on_rq_migrating(struct task_struct *p)
1446 return p->on_rq == TASK_ON_RQ_MIGRATING;
1452 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1453 #define WF_FORK 0x02 /* Child wakeup after fork */
1454 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1457 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1458 * of tasks with abnormal "nice" values across CPUs the contribution that
1459 * each task makes to its run queue's load is weighted according to its
1460 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1461 * scaled version of the new time slice allocation that they receive on time
1465 #define WEIGHT_IDLEPRIO 3
1466 #define WMULT_IDLEPRIO 1431655765
1468 extern const int sched_prio_to_weight[40];
1469 extern const u32 sched_prio_to_wmult[40];
1472 * {de,en}queue flags:
1474 * DEQUEUE_SLEEP - task is no longer runnable
1475 * ENQUEUE_WAKEUP - task just became runnable
1477 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1478 * are in a known state which allows modification. Such pairs
1479 * should preserve as much state as possible.
1481 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1484 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1485 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1486 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1490 #define DEQUEUE_SLEEP 0x01
1491 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1492 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1493 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1495 #define ENQUEUE_WAKEUP 0x01
1496 #define ENQUEUE_RESTORE 0x02
1497 #define ENQUEUE_MOVE 0x04
1498 #define ENQUEUE_NOCLOCK 0x08
1500 #define ENQUEUE_HEAD 0x10
1501 #define ENQUEUE_REPLENISH 0x20
1503 #define ENQUEUE_MIGRATED 0x40
1505 #define ENQUEUE_MIGRATED 0x00
1508 #define RETRY_TASK ((void *)-1UL)
1510 struct sched_class {
1511 const struct sched_class *next;
1513 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1514 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1515 void (*yield_task) (struct rq *rq);
1516 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1518 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1521 * It is the responsibility of the pick_next_task() method that will
1522 * return the next task to call put_prev_task() on the @prev task or
1523 * something equivalent.
1525 * May return RETRY_TASK when it finds a higher prio class has runnable
1528 struct task_struct * (*pick_next_task)(struct rq *rq,
1529 struct task_struct *prev,
1530 struct rq_flags *rf);
1531 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1534 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1535 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1537 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1539 void (*set_cpus_allowed)(struct task_struct *p,
1540 const struct cpumask *newmask);
1542 void (*rq_online)(struct rq *rq);
1543 void (*rq_offline)(struct rq *rq);
1546 void (*set_curr_task)(struct rq *rq);
1547 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1548 void (*task_fork)(struct task_struct *p);
1549 void (*task_dead)(struct task_struct *p);
1552 * The switched_from() call is allowed to drop rq->lock, therefore we
1553 * cannot assume the switched_from/switched_to pair is serliazed by
1554 * rq->lock. They are however serialized by p->pi_lock.
1556 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1557 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1558 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1561 unsigned int (*get_rr_interval)(struct rq *rq,
1562 struct task_struct *task);
1564 void (*update_curr)(struct rq *rq);
1566 #define TASK_SET_GROUP 0
1567 #define TASK_MOVE_GROUP 1
1569 #ifdef CONFIG_FAIR_GROUP_SCHED
1570 void (*task_change_group)(struct task_struct *p, int type);
1574 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1576 prev->sched_class->put_prev_task(rq, prev);
1579 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1581 curr->sched_class->set_curr_task(rq);
1585 #define sched_class_highest (&stop_sched_class)
1587 #define sched_class_highest (&dl_sched_class)
1589 #define for_each_class(class) \
1590 for (class = sched_class_highest; class; class = class->next)
1592 extern const struct sched_class stop_sched_class;
1593 extern const struct sched_class dl_sched_class;
1594 extern const struct sched_class rt_sched_class;
1595 extern const struct sched_class fair_sched_class;
1596 extern const struct sched_class idle_sched_class;
1601 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1603 extern void trigger_load_balance(struct rq *rq);
1605 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1609 #ifdef CONFIG_CPU_IDLE
1610 static inline void idle_set_state(struct rq *rq,
1611 struct cpuidle_state *idle_state)
1613 rq->idle_state = idle_state;
1616 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1618 SCHED_WARN_ON(!rcu_read_lock_held());
1620 return rq->idle_state;
1623 static inline void idle_set_state(struct rq *rq,
1624 struct cpuidle_state *idle_state)
1628 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1634 extern void schedule_idle(void);
1636 extern void sysrq_sched_debug_show(void);
1637 extern void sched_init_granularity(void);
1638 extern void update_max_interval(void);
1640 extern void init_sched_dl_class(void);
1641 extern void init_sched_rt_class(void);
1642 extern void init_sched_fair_class(void);
1644 extern void reweight_task(struct task_struct *p, int prio);
1646 extern void resched_curr(struct rq *rq);
1647 extern void resched_cpu(int cpu);
1649 extern struct rt_bandwidth def_rt_bandwidth;
1650 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1652 extern struct dl_bandwidth def_dl_bandwidth;
1653 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1654 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1655 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1656 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1659 #define BW_UNIT (1 << BW_SHIFT)
1660 #define RATIO_SHIFT 8
1661 unsigned long to_ratio(u64 period, u64 runtime);
1663 extern void init_entity_runnable_average(struct sched_entity *se);
1664 extern void post_init_entity_util_avg(struct sched_entity *se);
1666 #ifdef CONFIG_NO_HZ_FULL
1667 extern bool sched_can_stop_tick(struct rq *rq);
1668 extern int __init sched_tick_offload_init(void);
1671 * Tick may be needed by tasks in the runqueue depending on their policy and
1672 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1673 * nohz mode if necessary.
1675 static inline void sched_update_tick_dependency(struct rq *rq)
1679 if (!tick_nohz_full_enabled())
1684 if (!tick_nohz_full_cpu(cpu))
1687 if (sched_can_stop_tick(rq))
1688 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1690 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1693 static inline int sched_tick_offload_init(void) { return 0; }
1694 static inline void sched_update_tick_dependency(struct rq *rq) { }
1697 static inline void add_nr_running(struct rq *rq, unsigned count)
1699 unsigned prev_nr = rq->nr_running;
1701 rq->nr_running = prev_nr + count;
1703 if (prev_nr < 2 && rq->nr_running >= 2) {
1705 if (!READ_ONCE(rq->rd->overload))
1706 WRITE_ONCE(rq->rd->overload, 1);
1710 sched_update_tick_dependency(rq);
1713 static inline void sub_nr_running(struct rq *rq, unsigned count)
1715 rq->nr_running -= count;
1716 /* Check if we still need preemption */
1717 sched_update_tick_dependency(rq);
1720 extern void update_rq_clock(struct rq *rq);
1722 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1723 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1725 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1727 extern const_debug unsigned int sysctl_sched_nr_migrate;
1728 extern const_debug unsigned int sysctl_sched_migration_cost;
1730 #ifdef CONFIG_SCHED_HRTICK
1734 * - enabled by features
1735 * - hrtimer is actually high res
1737 static inline int hrtick_enabled(struct rq *rq)
1739 if (!sched_feat(HRTICK))
1741 if (!cpu_active(cpu_of(rq)))
1743 return hrtimer_is_hres_active(&rq->hrtick_timer);
1746 void hrtick_start(struct rq *rq, u64 delay);
1750 static inline int hrtick_enabled(struct rq *rq)
1755 #endif /* CONFIG_SCHED_HRTICK */
1757 #ifndef arch_scale_freq_capacity
1758 static __always_inline
1759 unsigned long arch_scale_freq_capacity(int cpu)
1761 return SCHED_CAPACITY_SCALE;
1766 #ifndef arch_scale_cpu_capacity
1767 static __always_inline
1768 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1770 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1771 return sd->smt_gain / sd->span_weight;
1773 return SCHED_CAPACITY_SCALE;
1777 #ifndef arch_scale_cpu_capacity
1778 static __always_inline
1779 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1781 return SCHED_CAPACITY_SCALE;
1786 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1787 __acquires(rq->lock);
1789 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1790 __acquires(p->pi_lock)
1791 __acquires(rq->lock);
1793 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1794 __releases(rq->lock)
1796 rq_unpin_lock(rq, rf);
1797 raw_spin_unlock(&rq->lock);
1801 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1802 __releases(rq->lock)
1803 __releases(p->pi_lock)
1805 rq_unpin_lock(rq, rf);
1806 raw_spin_unlock(&rq->lock);
1807 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1811 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1812 __acquires(rq->lock)
1814 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1815 rq_pin_lock(rq, rf);
1819 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1820 __acquires(rq->lock)
1822 raw_spin_lock_irq(&rq->lock);
1823 rq_pin_lock(rq, rf);
1827 rq_lock(struct rq *rq, struct rq_flags *rf)
1828 __acquires(rq->lock)
1830 raw_spin_lock(&rq->lock);
1831 rq_pin_lock(rq, rf);
1835 rq_relock(struct rq *rq, struct rq_flags *rf)
1836 __acquires(rq->lock)
1838 raw_spin_lock(&rq->lock);
1839 rq_repin_lock(rq, rf);
1843 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1844 __releases(rq->lock)
1846 rq_unpin_lock(rq, rf);
1847 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1851 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1852 __releases(rq->lock)
1854 rq_unpin_lock(rq, rf);
1855 raw_spin_unlock_irq(&rq->lock);
1859 rq_unlock(struct rq *rq, struct rq_flags *rf)
1860 __releases(rq->lock)
1862 rq_unpin_lock(rq, rf);
1863 raw_spin_unlock(&rq->lock);
1867 #ifdef CONFIG_PREEMPT
1869 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1872 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1873 * way at the expense of forcing extra atomic operations in all
1874 * invocations. This assures that the double_lock is acquired using the
1875 * same underlying policy as the spinlock_t on this architecture, which
1876 * reduces latency compared to the unfair variant below. However, it
1877 * also adds more overhead and therefore may reduce throughput.
1879 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1880 __releases(this_rq->lock)
1881 __acquires(busiest->lock)
1882 __acquires(this_rq->lock)
1884 raw_spin_unlock(&this_rq->lock);
1885 double_rq_lock(this_rq, busiest);
1892 * Unfair double_lock_balance: Optimizes throughput at the expense of
1893 * latency by eliminating extra atomic operations when the locks are
1894 * already in proper order on entry. This favors lower CPU-ids and will
1895 * grant the double lock to lower CPUs over higher ids under contention,
1896 * regardless of entry order into the function.
1898 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1899 __releases(this_rq->lock)
1900 __acquires(busiest->lock)
1901 __acquires(this_rq->lock)
1905 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1906 if (busiest < this_rq) {
1907 raw_spin_unlock(&this_rq->lock);
1908 raw_spin_lock(&busiest->lock);
1909 raw_spin_lock_nested(&this_rq->lock,
1910 SINGLE_DEPTH_NESTING);
1913 raw_spin_lock_nested(&busiest->lock,
1914 SINGLE_DEPTH_NESTING);
1919 #endif /* CONFIG_PREEMPT */
1922 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1924 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1926 if (unlikely(!irqs_disabled())) {
1927 /* printk() doesn't work well under rq->lock */
1928 raw_spin_unlock(&this_rq->lock);
1932 return _double_lock_balance(this_rq, busiest);
1935 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1936 __releases(busiest->lock)
1938 raw_spin_unlock(&busiest->lock);
1939 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1942 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1948 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1951 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1957 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1960 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1966 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1970 * double_rq_lock - safely lock two runqueues
1972 * Note this does not disable interrupts like task_rq_lock,
1973 * you need to do so manually before calling.
1975 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1976 __acquires(rq1->lock)
1977 __acquires(rq2->lock)
1979 BUG_ON(!irqs_disabled());
1981 raw_spin_lock(&rq1->lock);
1982 __acquire(rq2->lock); /* Fake it out ;) */
1985 raw_spin_lock(&rq1->lock);
1986 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1988 raw_spin_lock(&rq2->lock);
1989 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1995 * double_rq_unlock - safely unlock two runqueues
1997 * Note this does not restore interrupts like task_rq_unlock,
1998 * you need to do so manually after calling.
2000 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2001 __releases(rq1->lock)
2002 __releases(rq2->lock)
2004 raw_spin_unlock(&rq1->lock);
2006 raw_spin_unlock(&rq2->lock);
2008 __release(rq2->lock);
2011 extern void set_rq_online (struct rq *rq);
2012 extern void set_rq_offline(struct rq *rq);
2013 extern bool sched_smp_initialized;
2015 #else /* CONFIG_SMP */
2018 * double_rq_lock - safely lock two runqueues
2020 * Note this does not disable interrupts like task_rq_lock,
2021 * you need to do so manually before calling.
2023 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2024 __acquires(rq1->lock)
2025 __acquires(rq2->lock)
2027 BUG_ON(!irqs_disabled());
2029 raw_spin_lock(&rq1->lock);
2030 __acquire(rq2->lock); /* Fake it out ;) */
2034 * double_rq_unlock - safely unlock two runqueues
2036 * Note this does not restore interrupts like task_rq_unlock,
2037 * you need to do so manually after calling.
2039 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2040 __releases(rq1->lock)
2041 __releases(rq2->lock)
2044 raw_spin_unlock(&rq1->lock);
2045 __release(rq2->lock);
2050 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2051 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2053 #ifdef CONFIG_SCHED_DEBUG
2054 extern bool sched_debug_enabled;
2056 extern void print_cfs_stats(struct seq_file *m, int cpu);
2057 extern void print_rt_stats(struct seq_file *m, int cpu);
2058 extern void print_dl_stats(struct seq_file *m, int cpu);
2059 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2060 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2061 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2062 #ifdef CONFIG_NUMA_BALANCING
2064 show_numa_stats(struct task_struct *p, struct seq_file *m);
2066 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2067 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2068 #endif /* CONFIG_NUMA_BALANCING */
2069 #endif /* CONFIG_SCHED_DEBUG */
2071 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2072 extern void init_rt_rq(struct rt_rq *rt_rq);
2073 extern void init_dl_rq(struct dl_rq *dl_rq);
2075 extern void cfs_bandwidth_usage_inc(void);
2076 extern void cfs_bandwidth_usage_dec(void);
2078 #ifdef CONFIG_NO_HZ_COMMON
2079 #define NOHZ_BALANCE_KICK_BIT 0
2080 #define NOHZ_STATS_KICK_BIT 1
2082 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2083 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2085 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2087 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2089 extern void nohz_balance_exit_idle(struct rq *rq);
2091 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2097 void __dl_update(struct dl_bw *dl_b, s64 bw)
2099 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2102 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2103 "sched RCU must be held");
2104 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2105 struct rq *rq = cpu_rq(i);
2107 rq->dl.extra_bw += bw;
2112 void __dl_update(struct dl_bw *dl_b, s64 bw)
2114 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2121 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2126 struct u64_stats_sync sync;
2129 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2132 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2133 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2134 * and never move forward.
2136 static inline u64 irq_time_read(int cpu)
2138 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2143 seq = __u64_stats_fetch_begin(&irqtime->sync);
2144 total = irqtime->total;
2145 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2149 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2151 #ifdef CONFIG_CPU_FREQ
2152 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2155 * cpufreq_update_util - Take a note about CPU utilization changes.
2156 * @rq: Runqueue to carry out the update for.
2157 * @flags: Update reason flags.
2159 * This function is called by the scheduler on the CPU whose utilization is
2162 * It can only be called from RCU-sched read-side critical sections.
2164 * The way cpufreq is currently arranged requires it to evaluate the CPU
2165 * performance state (frequency/voltage) on a regular basis to prevent it from
2166 * being stuck in a completely inadequate performance level for too long.
2167 * That is not guaranteed to happen if the updates are only triggered from CFS
2168 * and DL, though, because they may not be coming in if only RT tasks are
2169 * active all the time (or there are RT tasks only).
2171 * As a workaround for that issue, this function is called periodically by the
2172 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2173 * but that really is a band-aid. Going forward it should be replaced with
2174 * solutions targeted more specifically at RT tasks.
2176 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2178 struct update_util_data *data;
2180 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2183 data->func(data, rq_clock(rq), flags);
2186 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2187 #endif /* CONFIG_CPU_FREQ */
2189 #ifdef arch_scale_freq_capacity
2190 # ifndef arch_scale_freq_invariant
2191 # define arch_scale_freq_invariant() true
2194 # define arch_scale_freq_invariant() false
2197 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2198 static inline unsigned long cpu_bw_dl(struct rq *rq)
2200 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2203 static inline unsigned long cpu_util_dl(struct rq *rq)
2205 return READ_ONCE(rq->avg_dl.util_avg);
2208 static inline unsigned long cpu_util_cfs(struct rq *rq)
2210 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2212 if (sched_feat(UTIL_EST)) {
2213 util = max_t(unsigned long, util,
2214 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2220 static inline unsigned long cpu_util_rt(struct rq *rq)
2222 return READ_ONCE(rq->avg_rt.util_avg);
2226 #ifdef HAVE_SCHED_AVG_IRQ
2227 static inline unsigned long cpu_util_irq(struct rq *rq)
2229 return rq->avg_irq.util_avg;
2233 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2235 util *= (max - irq);
2242 static inline unsigned long cpu_util_irq(struct rq *rq)
2248 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)