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;
340 struct hrtimer period_timer;
341 struct hrtimer slack_timer;
342 struct list_head throttled_cfs_rq;
351 /* Task group related information */
353 struct cgroup_subsys_state css;
355 #ifdef CONFIG_FAIR_GROUP_SCHED
356 /* schedulable entities of this group on each CPU */
357 struct sched_entity **se;
358 /* runqueue "owned" by this group on each CPU */
359 struct cfs_rq **cfs_rq;
360 unsigned long shares;
364 * load_avg can be heavily contended at clock tick time, so put
365 * it in its own cacheline separated from the fields above which
366 * will also be accessed at each tick.
368 atomic_long_t load_avg ____cacheline_aligned;
372 #ifdef CONFIG_RT_GROUP_SCHED
373 struct sched_rt_entity **rt_se;
374 struct rt_rq **rt_rq;
376 struct rt_bandwidth rt_bandwidth;
380 struct list_head list;
382 struct task_group *parent;
383 struct list_head siblings;
384 struct list_head children;
386 #ifdef CONFIG_SCHED_AUTOGROUP
387 struct autogroup *autogroup;
390 struct cfs_bandwidth cfs_bandwidth;
393 #ifdef CONFIG_FAIR_GROUP_SCHED
394 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
397 * A weight of 0 or 1 can cause arithmetics problems.
398 * A weight of a cfs_rq is the sum of weights of which entities
399 * are queued on this cfs_rq, so a weight of a entity should not be
400 * too large, so as the shares value of a task group.
401 * (The default weight is 1024 - so there's no practical
402 * limitation from this.)
404 #define MIN_SHARES (1UL << 1)
405 #define MAX_SHARES (1UL << 18)
408 typedef int (*tg_visitor)(struct task_group *, void *);
410 extern int walk_tg_tree_from(struct task_group *from,
411 tg_visitor down, tg_visitor up, void *data);
414 * Iterate the full tree, calling @down when first entering a node and @up when
415 * leaving it for the final time.
417 * Caller must hold rcu_lock or sufficient equivalent.
419 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
421 return walk_tg_tree_from(&root_task_group, down, up, data);
424 extern int tg_nop(struct task_group *tg, void *data);
426 extern void free_fair_sched_group(struct task_group *tg);
427 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
428 extern void online_fair_sched_group(struct task_group *tg);
429 extern void unregister_fair_sched_group(struct task_group *tg);
430 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
431 struct sched_entity *se, int cpu,
432 struct sched_entity *parent);
433 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
435 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
436 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
437 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
439 extern void free_rt_sched_group(struct task_group *tg);
440 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
441 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
442 struct sched_rt_entity *rt_se, int cpu,
443 struct sched_rt_entity *parent);
444 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
445 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
446 extern long sched_group_rt_runtime(struct task_group *tg);
447 extern long sched_group_rt_period(struct task_group *tg);
448 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
450 extern struct task_group *sched_create_group(struct task_group *parent);
451 extern void sched_online_group(struct task_group *tg,
452 struct task_group *parent);
453 extern void sched_destroy_group(struct task_group *tg);
454 extern void sched_offline_group(struct task_group *tg);
456 extern void sched_move_task(struct task_struct *tsk);
458 #ifdef CONFIG_FAIR_GROUP_SCHED
459 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
462 extern void set_task_rq_fair(struct sched_entity *se,
463 struct cfs_rq *prev, struct cfs_rq *next);
464 #else /* !CONFIG_SMP */
465 static inline void set_task_rq_fair(struct sched_entity *se,
466 struct cfs_rq *prev, struct cfs_rq *next) { }
467 #endif /* CONFIG_SMP */
468 #endif /* CONFIG_FAIR_GROUP_SCHED */
470 #else /* CONFIG_CGROUP_SCHED */
472 struct cfs_bandwidth { };
474 #endif /* CONFIG_CGROUP_SCHED */
476 /* CFS-related fields in a runqueue */
478 struct load_weight load;
479 unsigned long runnable_weight;
480 unsigned int nr_running;
481 unsigned int h_nr_running;
486 u64 min_vruntime_copy;
489 struct rb_root_cached tasks_timeline;
492 * 'curr' points to currently running entity on this cfs_rq.
493 * It is set to NULL otherwise (i.e when none are currently running).
495 struct sched_entity *curr;
496 struct sched_entity *next;
497 struct sched_entity *last;
498 struct sched_entity *skip;
500 #ifdef CONFIG_SCHED_DEBUG
501 unsigned int nr_spread_over;
508 struct sched_avg avg;
510 u64 load_last_update_time_copy;
513 raw_spinlock_t lock ____cacheline_aligned;
515 unsigned long load_avg;
516 unsigned long util_avg;
517 unsigned long runnable_sum;
520 #ifdef CONFIG_FAIR_GROUP_SCHED
521 unsigned long tg_load_avg_contrib;
523 long prop_runnable_sum;
526 * h_load = weight * f(tg)
528 * Where f(tg) is the recursive weight fraction assigned to
531 unsigned long h_load;
532 u64 last_h_load_update;
533 struct sched_entity *h_load_next;
534 #endif /* CONFIG_FAIR_GROUP_SCHED */
535 #endif /* CONFIG_SMP */
537 #ifdef CONFIG_FAIR_GROUP_SCHED
538 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
541 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
542 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
543 * (like users, containers etc.)
545 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
546 * This list is used during load balance.
549 struct list_head leaf_cfs_rq_list;
550 struct task_group *tg; /* group that "owns" this runqueue */
552 #ifdef CONFIG_CFS_BANDWIDTH
555 s64 runtime_remaining;
558 u64 throttled_clock_task;
559 u64 throttled_clock_task_time;
562 struct list_head throttled_list;
563 #endif /* CONFIG_CFS_BANDWIDTH */
564 #endif /* CONFIG_FAIR_GROUP_SCHED */
567 static inline int rt_bandwidth_enabled(void)
569 return sysctl_sched_rt_runtime >= 0;
572 /* RT IPI pull logic requires IRQ_WORK */
573 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
574 # define HAVE_RT_PUSH_IPI
577 /* Real-Time classes' related field in a runqueue: */
579 struct rt_prio_array active;
580 unsigned int rt_nr_running;
581 unsigned int rr_nr_running;
582 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
584 int curr; /* highest queued rt task prio */
586 int next; /* next highest */
591 unsigned long rt_nr_migratory;
592 unsigned long rt_nr_total;
594 struct plist_head pushable_tasks;
595 #endif /* CONFIG_SMP */
601 /* Nests inside the rq lock: */
602 raw_spinlock_t rt_runtime_lock;
604 #ifdef CONFIG_RT_GROUP_SCHED
605 unsigned long rt_nr_boosted;
608 struct task_group *tg;
612 /* Deadline class' related fields in a runqueue */
614 /* runqueue is an rbtree, ordered by deadline */
615 struct rb_root_cached root;
617 unsigned long dl_nr_running;
621 * Deadline values of the currently executing and the
622 * earliest ready task on this rq. Caching these facilitates
623 * the decision wether or not a ready but not running task
624 * should migrate somewhere else.
631 unsigned long dl_nr_migratory;
635 * Tasks on this rq that can be pushed away. They are kept in
636 * an rb-tree, ordered by tasks' deadlines, with caching
637 * of the leftmost (earliest deadline) element.
639 struct rb_root_cached pushable_dl_tasks_root;
644 * "Active utilization" for this runqueue: increased when a
645 * task wakes up (becomes TASK_RUNNING) and decreased when a
651 * Utilization of the tasks "assigned" to this runqueue (including
652 * the tasks that are in runqueue and the tasks that executed on this
653 * CPU and blocked). Increased when a task moves to this runqueue, and
654 * decreased when the task moves away (migrates, changes scheduling
655 * policy, or terminates).
656 * This is needed to compute the "inactive utilization" for the
657 * runqueue (inactive utilization = this_bw - running_bw).
663 * Inverse of the fraction of CPU utilization that can be reclaimed
664 * by the GRUB algorithm.
671 static inline bool sched_asym_prefer(int a, int b)
673 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
677 * We add the notion of a root-domain which will be used to define per-domain
678 * variables. Each exclusive cpuset essentially defines an island domain by
679 * fully partitioning the member CPUs from any other cpuset. Whenever a new
680 * exclusive cpuset is created, we also create and attach a new root-domain
689 cpumask_var_t online;
691 /* Indicate more than one runnable task for any CPU */
695 * The bit corresponding to a CPU gets set here if such CPU has more
696 * than one runnable -deadline task (as it is below for RT tasks).
698 cpumask_var_t dlo_mask;
703 #ifdef HAVE_RT_PUSH_IPI
705 * For IPI pull requests, loop across the rto_mask.
707 struct irq_work rto_push_work;
708 raw_spinlock_t rto_lock;
709 /* These are only updated and read within rto_lock */
712 /* These atomics are updated outside of a lock */
713 atomic_t rto_loop_next;
714 atomic_t rto_loop_start;
717 * The "RT overload" flag: it gets set if a CPU has more than
718 * one runnable RT task.
720 cpumask_var_t rto_mask;
721 struct cpupri cpupri;
723 unsigned long max_cpu_capacity;
726 extern struct root_domain def_root_domain;
727 extern struct mutex sched_domains_mutex;
729 extern void init_defrootdomain(void);
730 extern int sched_init_domains(const struct cpumask *cpu_map);
731 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
732 extern void sched_get_rd(struct root_domain *rd);
733 extern void sched_put_rd(struct root_domain *rd);
735 #ifdef HAVE_RT_PUSH_IPI
736 extern void rto_push_irq_work_func(struct irq_work *work);
738 #endif /* CONFIG_SMP */
741 * This is the main, per-CPU runqueue data structure.
743 * Locking rule: those places that want to lock multiple runqueues
744 * (such as the load balancing or the thread migration code), lock
745 * acquire operations must be ordered by ascending &runqueue.
752 * nr_running and cpu_load should be in the same cacheline because
753 * remote CPUs use both these fields when doing load calculation.
755 unsigned int nr_running;
756 #ifdef CONFIG_NUMA_BALANCING
757 unsigned int nr_numa_running;
758 unsigned int nr_preferred_running;
760 #define CPU_LOAD_IDX_MAX 5
761 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
762 #ifdef CONFIG_NO_HZ_COMMON
764 unsigned long last_load_update_tick;
765 unsigned long last_blocked_load_update_tick;
766 unsigned int has_blocked_load;
767 #endif /* CONFIG_SMP */
768 unsigned int nohz_tick_stopped;
770 #endif /* CONFIG_NO_HZ_COMMON */
772 /* capture load from *all* tasks on this CPU: */
773 struct load_weight load;
774 unsigned long nr_load_updates;
781 #ifdef CONFIG_FAIR_GROUP_SCHED
782 /* list of leaf cfs_rq on this CPU: */
783 struct list_head leaf_cfs_rq_list;
784 struct list_head *tmp_alone_branch;
785 #endif /* CONFIG_FAIR_GROUP_SCHED */
788 * This is part of a global counter where only the total sum
789 * over all CPUs matters. A task can increase this counter on
790 * one CPU and if it got migrated afterwards it may decrease
791 * it on another CPU. Always updated under the runqueue lock:
793 unsigned long nr_uninterruptible;
795 struct task_struct *curr;
796 struct task_struct *idle;
797 struct task_struct *stop;
798 unsigned long next_balance;
799 struct mm_struct *prev_mm;
801 unsigned int clock_update_flags;
808 struct root_domain *rd;
809 struct sched_domain *sd;
811 unsigned long cpu_capacity;
812 unsigned long cpu_capacity_orig;
814 struct callback_head *balance_callback;
816 unsigned char idle_balance;
818 /* For active balancing */
821 struct cpu_stop_work active_balance_work;
823 /* CPU of this runqueue: */
827 struct list_head cfs_tasks;
834 /* This is used to determine avg_idle's max value */
835 u64 max_idle_balance_cost;
838 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
841 #ifdef CONFIG_PARAVIRT
844 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
845 u64 prev_steal_time_rq;
848 /* calc_load related fields */
849 unsigned long calc_load_update;
850 long calc_load_active;
852 #ifdef CONFIG_SCHED_HRTICK
854 int hrtick_csd_pending;
855 call_single_data_t hrtick_csd;
857 struct hrtimer hrtick_timer;
860 #ifdef CONFIG_SCHEDSTATS
862 struct sched_info rq_sched_info;
863 unsigned long long rq_cpu_time;
864 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
866 /* sys_sched_yield() stats */
867 unsigned int yld_count;
869 /* schedule() stats */
870 unsigned int sched_count;
871 unsigned int sched_goidle;
873 /* try_to_wake_up() stats */
874 unsigned int ttwu_count;
875 unsigned int ttwu_local;
879 struct llist_head wake_list;
882 #ifdef CONFIG_CPU_IDLE
883 /* Must be inspected within a rcu lock section */
884 struct cpuidle_state *idle_state;
888 static inline int cpu_of(struct rq *rq)
898 #ifdef CONFIG_SCHED_SMT
900 extern struct static_key_false sched_smt_present;
902 extern void __update_idle_core(struct rq *rq);
904 static inline void update_idle_core(struct rq *rq)
906 if (static_branch_unlikely(&sched_smt_present))
907 __update_idle_core(rq);
911 static inline void update_idle_core(struct rq *rq) { }
914 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
916 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
917 #define this_rq() this_cpu_ptr(&runqueues)
918 #define task_rq(p) cpu_rq(task_cpu(p))
919 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
920 #define raw_rq() raw_cpu_ptr(&runqueues)
922 static inline u64 __rq_clock_broken(struct rq *rq)
924 return READ_ONCE(rq->clock);
928 * rq::clock_update_flags bits
930 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
931 * call to __schedule(). This is an optimisation to avoid
932 * neighbouring rq clock updates.
934 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
935 * in effect and calls to update_rq_clock() are being ignored.
937 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
938 * made to update_rq_clock() since the last time rq::lock was pinned.
940 * If inside of __schedule(), clock_update_flags will have been
941 * shifted left (a left shift is a cheap operation for the fast path
942 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
944 * if (rq-clock_update_flags >= RQCF_UPDATED)
946 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
947 * one position though, because the next rq_unpin_lock() will shift it
950 #define RQCF_REQ_SKIP 0x01
951 #define RQCF_ACT_SKIP 0x02
952 #define RQCF_UPDATED 0x04
954 static inline void assert_clock_updated(struct rq *rq)
957 * The only reason for not seeing a clock update since the
958 * last rq_pin_lock() is if we're currently skipping updates.
960 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
963 static inline u64 rq_clock(struct rq *rq)
965 lockdep_assert_held(&rq->lock);
966 assert_clock_updated(rq);
971 static inline u64 rq_clock_task(struct rq *rq)
973 lockdep_assert_held(&rq->lock);
974 assert_clock_updated(rq);
976 return rq->clock_task;
979 static inline void rq_clock_skip_update(struct rq *rq)
981 lockdep_assert_held(&rq->lock);
982 rq->clock_update_flags |= RQCF_REQ_SKIP;
986 * See rt task throttling, which is the only time a skip
987 * request is cancelled.
989 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
991 lockdep_assert_held(&rq->lock);
992 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
997 struct pin_cookie cookie;
998 #ifdef CONFIG_SCHED_DEBUG
1000 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1001 * current pin context is stashed here in case it needs to be
1002 * restored in rq_repin_lock().
1004 unsigned int clock_update_flags;
1008 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1010 rf->cookie = lockdep_pin_lock(&rq->lock);
1012 #ifdef CONFIG_SCHED_DEBUG
1013 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1014 rf->clock_update_flags = 0;
1018 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1020 #ifdef CONFIG_SCHED_DEBUG
1021 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1022 rf->clock_update_flags = RQCF_UPDATED;
1025 lockdep_unpin_lock(&rq->lock, rf->cookie);
1028 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1030 lockdep_repin_lock(&rq->lock, rf->cookie);
1032 #ifdef CONFIG_SCHED_DEBUG
1034 * Restore the value we stashed in @rf for this pin context.
1036 rq->clock_update_flags |= rf->clock_update_flags;
1041 enum numa_topology_type {
1046 extern enum numa_topology_type sched_numa_topology_type;
1047 extern int sched_max_numa_distance;
1048 extern bool find_numa_distance(int distance);
1052 extern void sched_init_numa(void);
1053 extern void sched_domains_numa_masks_set(unsigned int cpu);
1054 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1056 static inline void sched_init_numa(void) { }
1057 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1058 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1061 #ifdef CONFIG_NUMA_BALANCING
1062 /* The regions in numa_faults array from task_struct */
1063 enum numa_faults_stats {
1069 extern void sched_setnuma(struct task_struct *p, int node);
1070 extern int migrate_task_to(struct task_struct *p, int cpu);
1071 extern int migrate_swap(struct task_struct *, struct task_struct *);
1072 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1075 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1078 #endif /* CONFIG_NUMA_BALANCING */
1083 queue_balance_callback(struct rq *rq,
1084 struct callback_head *head,
1085 void (*func)(struct rq *rq))
1087 lockdep_assert_held(&rq->lock);
1089 if (unlikely(head->next))
1092 head->func = (void (*)(struct callback_head *))func;
1093 head->next = rq->balance_callback;
1094 rq->balance_callback = head;
1097 extern void sched_ttwu_pending(void);
1099 #define rcu_dereference_check_sched_domain(p) \
1100 rcu_dereference_check((p), \
1101 lockdep_is_held(&sched_domains_mutex))
1104 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1105 * See detach_destroy_domains: synchronize_sched for details.
1107 * The domain tree of any CPU may only be accessed from within
1108 * preempt-disabled sections.
1110 #define for_each_domain(cpu, __sd) \
1111 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1112 __sd; __sd = __sd->parent)
1114 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1117 * highest_flag_domain - Return highest sched_domain containing flag.
1118 * @cpu: The CPU whose highest level of sched domain is to
1120 * @flag: The flag to check for the highest sched_domain
1121 * for the given CPU.
1123 * Returns the highest sched_domain of a CPU which contains the given flag.
1125 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1127 struct sched_domain *sd, *hsd = NULL;
1129 for_each_domain(cpu, sd) {
1130 if (!(sd->flags & flag))
1138 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1140 struct sched_domain *sd;
1142 for_each_domain(cpu, sd) {
1143 if (sd->flags & flag)
1150 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1151 DECLARE_PER_CPU(int, sd_llc_size);
1152 DECLARE_PER_CPU(int, sd_llc_id);
1153 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1154 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1155 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1157 struct sched_group_capacity {
1160 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1163 unsigned long capacity;
1164 unsigned long min_capacity; /* Min per-CPU capacity in group */
1165 unsigned long next_update;
1166 int imbalance; /* XXX unrelated to capacity but shared group state */
1168 #ifdef CONFIG_SCHED_DEBUG
1172 unsigned long cpumask[0]; /* Balance mask */
1175 struct sched_group {
1176 struct sched_group *next; /* Must be a circular list */
1179 unsigned int group_weight;
1180 struct sched_group_capacity *sgc;
1181 int asym_prefer_cpu; /* CPU of highest priority in group */
1184 * The CPUs this group covers.
1186 * NOTE: this field is variable length. (Allocated dynamically
1187 * by attaching extra space to the end of the structure,
1188 * depending on how many CPUs the kernel has booted up with)
1190 unsigned long cpumask[0];
1193 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1195 return to_cpumask(sg->cpumask);
1199 * See build_balance_mask().
1201 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1203 return to_cpumask(sg->sgc->cpumask);
1207 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1208 * @group: The group whose first CPU is to be returned.
1210 static inline unsigned int group_first_cpu(struct sched_group *group)
1212 return cpumask_first(sched_group_span(group));
1215 extern int group_balance_cpu(struct sched_group *sg);
1217 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1218 void register_sched_domain_sysctl(void);
1219 void dirty_sched_domain_sysctl(int cpu);
1220 void unregister_sched_domain_sysctl(void);
1222 static inline void register_sched_domain_sysctl(void)
1225 static inline void dirty_sched_domain_sysctl(int cpu)
1228 static inline void unregister_sched_domain_sysctl(void)
1235 static inline void sched_ttwu_pending(void) { }
1237 #endif /* CONFIG_SMP */
1240 #include "autogroup.h"
1242 #ifdef CONFIG_CGROUP_SCHED
1245 * Return the group to which this tasks belongs.
1247 * We cannot use task_css() and friends because the cgroup subsystem
1248 * changes that value before the cgroup_subsys::attach() method is called,
1249 * therefore we cannot pin it and might observe the wrong value.
1251 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1252 * core changes this before calling sched_move_task().
1254 * Instead we use a 'copy' which is updated from sched_move_task() while
1255 * holding both task_struct::pi_lock and rq::lock.
1257 static inline struct task_group *task_group(struct task_struct *p)
1259 return p->sched_task_group;
1262 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1263 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1265 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1266 struct task_group *tg = task_group(p);
1269 #ifdef CONFIG_FAIR_GROUP_SCHED
1270 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1271 p->se.cfs_rq = tg->cfs_rq[cpu];
1272 p->se.parent = tg->se[cpu];
1275 #ifdef CONFIG_RT_GROUP_SCHED
1276 p->rt.rt_rq = tg->rt_rq[cpu];
1277 p->rt.parent = tg->rt_se[cpu];
1281 #else /* CONFIG_CGROUP_SCHED */
1283 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1284 static inline struct task_group *task_group(struct task_struct *p)
1289 #endif /* CONFIG_CGROUP_SCHED */
1291 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1293 set_task_rq(p, cpu);
1296 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1297 * successfuly executed on another CPU. We must ensure that updates of
1298 * per-task data have been completed by this moment.
1301 #ifdef CONFIG_THREAD_INFO_IN_TASK
1304 task_thread_info(p)->cpu = cpu;
1311 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1313 #ifdef CONFIG_SCHED_DEBUG
1314 # include <linux/static_key.h>
1315 # define const_debug __read_mostly
1317 # define const_debug const
1320 #define SCHED_FEAT(name, enabled) \
1321 __SCHED_FEAT_##name ,
1324 #include "features.h"
1330 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1333 * To support run-time toggling of sched features, all the translation units
1334 * (but core.c) reference the sysctl_sched_features defined in core.c.
1336 extern const_debug unsigned int sysctl_sched_features;
1338 #define SCHED_FEAT(name, enabled) \
1339 static __always_inline bool static_branch_##name(struct static_key *key) \
1341 return static_key_##enabled(key); \
1344 #include "features.h"
1347 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1348 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1350 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1353 * Each translation unit has its own copy of sysctl_sched_features to allow
1354 * constants propagation at compile time and compiler optimization based on
1357 #define SCHED_FEAT(name, enabled) \
1358 (1UL << __SCHED_FEAT_##name) * enabled |
1359 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1360 #include "features.h"
1364 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1366 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1368 extern struct static_key_false sched_numa_balancing;
1369 extern struct static_key_false sched_schedstats;
1371 static inline u64 global_rt_period(void)
1373 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1376 static inline u64 global_rt_runtime(void)
1378 if (sysctl_sched_rt_runtime < 0)
1381 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1384 static inline int task_current(struct rq *rq, struct task_struct *p)
1386 return rq->curr == p;
1389 static inline int task_running(struct rq *rq, struct task_struct *p)
1394 return task_current(rq, p);
1398 static inline int task_on_rq_queued(struct task_struct *p)
1400 return p->on_rq == TASK_ON_RQ_QUEUED;
1403 static inline int task_on_rq_migrating(struct task_struct *p)
1405 return p->on_rq == TASK_ON_RQ_MIGRATING;
1411 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1412 #define WF_FORK 0x02 /* Child wakeup after fork */
1413 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1416 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1417 * of tasks with abnormal "nice" values across CPUs the contribution that
1418 * each task makes to its run queue's load is weighted according to its
1419 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1420 * scaled version of the new time slice allocation that they receive on time
1424 #define WEIGHT_IDLEPRIO 3
1425 #define WMULT_IDLEPRIO 1431655765
1427 extern const int sched_prio_to_weight[40];
1428 extern const u32 sched_prio_to_wmult[40];
1431 * {de,en}queue flags:
1433 * DEQUEUE_SLEEP - task is no longer runnable
1434 * ENQUEUE_WAKEUP - task just became runnable
1436 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1437 * are in a known state which allows modification. Such pairs
1438 * should preserve as much state as possible.
1440 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1443 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1444 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1445 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1449 #define DEQUEUE_SLEEP 0x01
1450 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1451 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1452 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1454 #define ENQUEUE_WAKEUP 0x01
1455 #define ENQUEUE_RESTORE 0x02
1456 #define ENQUEUE_MOVE 0x04
1457 #define ENQUEUE_NOCLOCK 0x08
1459 #define ENQUEUE_HEAD 0x10
1460 #define ENQUEUE_REPLENISH 0x20
1462 #define ENQUEUE_MIGRATED 0x40
1464 #define ENQUEUE_MIGRATED 0x00
1467 #define RETRY_TASK ((void *)-1UL)
1469 struct sched_class {
1470 const struct sched_class *next;
1472 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1473 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1474 void (*yield_task) (struct rq *rq);
1475 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1477 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1480 * It is the responsibility of the pick_next_task() method that will
1481 * return the next task to call put_prev_task() on the @prev task or
1482 * something equivalent.
1484 * May return RETRY_TASK when it finds a higher prio class has runnable
1487 struct task_struct * (*pick_next_task)(struct rq *rq,
1488 struct task_struct *prev,
1489 struct rq_flags *rf);
1490 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1493 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1494 void (*migrate_task_rq)(struct task_struct *p);
1496 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1498 void (*set_cpus_allowed)(struct task_struct *p,
1499 const struct cpumask *newmask);
1501 void (*rq_online)(struct rq *rq);
1502 void (*rq_offline)(struct rq *rq);
1505 void (*set_curr_task)(struct rq *rq);
1506 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1507 void (*task_fork)(struct task_struct *p);
1508 void (*task_dead)(struct task_struct *p);
1511 * The switched_from() call is allowed to drop rq->lock, therefore we
1512 * cannot assume the switched_from/switched_to pair is serliazed by
1513 * rq->lock. They are however serialized by p->pi_lock.
1515 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1516 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1517 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1520 unsigned int (*get_rr_interval)(struct rq *rq,
1521 struct task_struct *task);
1523 void (*update_curr)(struct rq *rq);
1525 #define TASK_SET_GROUP 0
1526 #define TASK_MOVE_GROUP 1
1528 #ifdef CONFIG_FAIR_GROUP_SCHED
1529 void (*task_change_group)(struct task_struct *p, int type);
1533 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1535 prev->sched_class->put_prev_task(rq, prev);
1538 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1540 curr->sched_class->set_curr_task(rq);
1544 #define sched_class_highest (&stop_sched_class)
1546 #define sched_class_highest (&dl_sched_class)
1548 #define for_each_class(class) \
1549 for (class = sched_class_highest; class; class = class->next)
1551 extern const struct sched_class stop_sched_class;
1552 extern const struct sched_class dl_sched_class;
1553 extern const struct sched_class rt_sched_class;
1554 extern const struct sched_class fair_sched_class;
1555 extern const struct sched_class idle_sched_class;
1560 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1562 extern void trigger_load_balance(struct rq *rq);
1564 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1568 #ifdef CONFIG_CPU_IDLE
1569 static inline void idle_set_state(struct rq *rq,
1570 struct cpuidle_state *idle_state)
1572 rq->idle_state = idle_state;
1575 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1577 SCHED_WARN_ON(!rcu_read_lock_held());
1579 return rq->idle_state;
1582 static inline void idle_set_state(struct rq *rq,
1583 struct cpuidle_state *idle_state)
1587 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1593 extern void schedule_idle(void);
1595 extern void sysrq_sched_debug_show(void);
1596 extern void sched_init_granularity(void);
1597 extern void update_max_interval(void);
1599 extern void init_sched_dl_class(void);
1600 extern void init_sched_rt_class(void);
1601 extern void init_sched_fair_class(void);
1603 extern void reweight_task(struct task_struct *p, int prio);
1605 extern void resched_curr(struct rq *rq);
1606 extern void resched_cpu(int cpu);
1608 extern struct rt_bandwidth def_rt_bandwidth;
1609 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1611 extern struct dl_bandwidth def_dl_bandwidth;
1612 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1613 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1614 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1615 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1618 #define BW_UNIT (1 << BW_SHIFT)
1619 #define RATIO_SHIFT 8
1620 unsigned long to_ratio(u64 period, u64 runtime);
1622 extern void init_entity_runnable_average(struct sched_entity *se);
1623 extern void post_init_entity_util_avg(struct sched_entity *se);
1625 #ifdef CONFIG_NO_HZ_FULL
1626 extern bool sched_can_stop_tick(struct rq *rq);
1627 extern int __init sched_tick_offload_init(void);
1630 * Tick may be needed by tasks in the runqueue depending on their policy and
1631 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1632 * nohz mode if necessary.
1634 static inline void sched_update_tick_dependency(struct rq *rq)
1638 if (!tick_nohz_full_enabled())
1643 if (!tick_nohz_full_cpu(cpu))
1646 if (sched_can_stop_tick(rq))
1647 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1649 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1652 static inline int sched_tick_offload_init(void) { return 0; }
1653 static inline void sched_update_tick_dependency(struct rq *rq) { }
1656 static inline void add_nr_running(struct rq *rq, unsigned count)
1658 unsigned prev_nr = rq->nr_running;
1660 rq->nr_running = prev_nr + count;
1662 if (prev_nr < 2 && rq->nr_running >= 2) {
1664 if (!rq->rd->overload)
1665 rq->rd->overload = true;
1669 sched_update_tick_dependency(rq);
1672 static inline void sub_nr_running(struct rq *rq, unsigned count)
1674 rq->nr_running -= count;
1675 /* Check if we still need preemption */
1676 sched_update_tick_dependency(rq);
1679 extern void update_rq_clock(struct rq *rq);
1681 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1682 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1684 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1686 extern const_debug unsigned int sysctl_sched_time_avg;
1687 extern const_debug unsigned int sysctl_sched_nr_migrate;
1688 extern const_debug unsigned int sysctl_sched_migration_cost;
1690 static inline u64 sched_avg_period(void)
1692 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1695 #ifdef CONFIG_SCHED_HRTICK
1699 * - enabled by features
1700 * - hrtimer is actually high res
1702 static inline int hrtick_enabled(struct rq *rq)
1704 if (!sched_feat(HRTICK))
1706 if (!cpu_active(cpu_of(rq)))
1708 return hrtimer_is_hres_active(&rq->hrtick_timer);
1711 void hrtick_start(struct rq *rq, u64 delay);
1715 static inline int hrtick_enabled(struct rq *rq)
1720 #endif /* CONFIG_SCHED_HRTICK */
1722 #ifndef arch_scale_freq_capacity
1723 static __always_inline
1724 unsigned long arch_scale_freq_capacity(int cpu)
1726 return SCHED_CAPACITY_SCALE;
1731 extern void sched_avg_update(struct rq *rq);
1733 #ifndef arch_scale_cpu_capacity
1734 static __always_inline
1735 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1737 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1738 return sd->smt_gain / sd->span_weight;
1740 return SCHED_CAPACITY_SCALE;
1744 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1746 rq->rt_avg += rt_delta * arch_scale_freq_capacity(cpu_of(rq));
1747 sched_avg_update(rq);
1750 #ifndef arch_scale_cpu_capacity
1751 static __always_inline
1752 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1754 return SCHED_CAPACITY_SCALE;
1757 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1758 static inline void sched_avg_update(struct rq *rq) { }
1761 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1762 __acquires(rq->lock);
1764 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1765 __acquires(p->pi_lock)
1766 __acquires(rq->lock);
1768 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1769 __releases(rq->lock)
1771 rq_unpin_lock(rq, rf);
1772 raw_spin_unlock(&rq->lock);
1776 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1777 __releases(rq->lock)
1778 __releases(p->pi_lock)
1780 rq_unpin_lock(rq, rf);
1781 raw_spin_unlock(&rq->lock);
1782 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1786 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1787 __acquires(rq->lock)
1789 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1790 rq_pin_lock(rq, rf);
1794 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1795 __acquires(rq->lock)
1797 raw_spin_lock_irq(&rq->lock);
1798 rq_pin_lock(rq, rf);
1802 rq_lock(struct rq *rq, struct rq_flags *rf)
1803 __acquires(rq->lock)
1805 raw_spin_lock(&rq->lock);
1806 rq_pin_lock(rq, rf);
1810 rq_relock(struct rq *rq, struct rq_flags *rf)
1811 __acquires(rq->lock)
1813 raw_spin_lock(&rq->lock);
1814 rq_repin_lock(rq, rf);
1818 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1819 __releases(rq->lock)
1821 rq_unpin_lock(rq, rf);
1822 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1826 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1827 __releases(rq->lock)
1829 rq_unpin_lock(rq, rf);
1830 raw_spin_unlock_irq(&rq->lock);
1834 rq_unlock(struct rq *rq, struct rq_flags *rf)
1835 __releases(rq->lock)
1837 rq_unpin_lock(rq, rf);
1838 raw_spin_unlock(&rq->lock);
1842 #ifdef CONFIG_PREEMPT
1844 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1847 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1848 * way at the expense of forcing extra atomic operations in all
1849 * invocations. This assures that the double_lock is acquired using the
1850 * same underlying policy as the spinlock_t on this architecture, which
1851 * reduces latency compared to the unfair variant below. However, it
1852 * also adds more overhead and therefore may reduce throughput.
1854 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1855 __releases(this_rq->lock)
1856 __acquires(busiest->lock)
1857 __acquires(this_rq->lock)
1859 raw_spin_unlock(&this_rq->lock);
1860 double_rq_lock(this_rq, busiest);
1867 * Unfair double_lock_balance: Optimizes throughput at the expense of
1868 * latency by eliminating extra atomic operations when the locks are
1869 * already in proper order on entry. This favors lower CPU-ids and will
1870 * grant the double lock to lower CPUs over higher ids under contention,
1871 * regardless of entry order into the function.
1873 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1874 __releases(this_rq->lock)
1875 __acquires(busiest->lock)
1876 __acquires(this_rq->lock)
1880 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1881 if (busiest < this_rq) {
1882 raw_spin_unlock(&this_rq->lock);
1883 raw_spin_lock(&busiest->lock);
1884 raw_spin_lock_nested(&this_rq->lock,
1885 SINGLE_DEPTH_NESTING);
1888 raw_spin_lock_nested(&busiest->lock,
1889 SINGLE_DEPTH_NESTING);
1894 #endif /* CONFIG_PREEMPT */
1897 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1899 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1901 if (unlikely(!irqs_disabled())) {
1902 /* printk() doesn't work well under rq->lock */
1903 raw_spin_unlock(&this_rq->lock);
1907 return _double_lock_balance(this_rq, busiest);
1910 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1911 __releases(busiest->lock)
1913 raw_spin_unlock(&busiest->lock);
1914 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1917 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1923 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1926 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1932 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1935 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1941 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1945 * double_rq_lock - safely lock two runqueues
1947 * Note this does not disable interrupts like task_rq_lock,
1948 * you need to do so manually before calling.
1950 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1951 __acquires(rq1->lock)
1952 __acquires(rq2->lock)
1954 BUG_ON(!irqs_disabled());
1956 raw_spin_lock(&rq1->lock);
1957 __acquire(rq2->lock); /* Fake it out ;) */
1960 raw_spin_lock(&rq1->lock);
1961 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1963 raw_spin_lock(&rq2->lock);
1964 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1970 * double_rq_unlock - safely unlock two runqueues
1972 * Note this does not restore interrupts like task_rq_unlock,
1973 * you need to do so manually after calling.
1975 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1976 __releases(rq1->lock)
1977 __releases(rq2->lock)
1979 raw_spin_unlock(&rq1->lock);
1981 raw_spin_unlock(&rq2->lock);
1983 __release(rq2->lock);
1986 extern void set_rq_online (struct rq *rq);
1987 extern void set_rq_offline(struct rq *rq);
1988 extern bool sched_smp_initialized;
1990 #else /* CONFIG_SMP */
1993 * double_rq_lock - safely lock two runqueues
1995 * Note this does not disable interrupts like task_rq_lock,
1996 * you need to do so manually before calling.
1998 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1999 __acquires(rq1->lock)
2000 __acquires(rq2->lock)
2002 BUG_ON(!irqs_disabled());
2004 raw_spin_lock(&rq1->lock);
2005 __acquire(rq2->lock); /* Fake it out ;) */
2009 * double_rq_unlock - safely unlock two runqueues
2011 * Note this does not restore interrupts like task_rq_unlock,
2012 * you need to do so manually after calling.
2014 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2015 __releases(rq1->lock)
2016 __releases(rq2->lock)
2019 raw_spin_unlock(&rq1->lock);
2020 __release(rq2->lock);
2025 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2026 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2028 #ifdef CONFIG_SCHED_DEBUG
2029 extern bool sched_debug_enabled;
2031 extern void print_cfs_stats(struct seq_file *m, int cpu);
2032 extern void print_rt_stats(struct seq_file *m, int cpu);
2033 extern void print_dl_stats(struct seq_file *m, int cpu);
2034 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2035 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2036 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2037 #ifdef CONFIG_NUMA_BALANCING
2039 show_numa_stats(struct task_struct *p, struct seq_file *m);
2041 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2042 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2043 #endif /* CONFIG_NUMA_BALANCING */
2044 #endif /* CONFIG_SCHED_DEBUG */
2046 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2047 extern void init_rt_rq(struct rt_rq *rt_rq);
2048 extern void init_dl_rq(struct dl_rq *dl_rq);
2050 extern void cfs_bandwidth_usage_inc(void);
2051 extern void cfs_bandwidth_usage_dec(void);
2053 #ifdef CONFIG_NO_HZ_COMMON
2054 #define NOHZ_BALANCE_KICK_BIT 0
2055 #define NOHZ_STATS_KICK_BIT 1
2057 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2058 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2060 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2062 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2064 extern void nohz_balance_exit_idle(struct rq *rq);
2066 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2072 void __dl_update(struct dl_bw *dl_b, s64 bw)
2074 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2077 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2078 "sched RCU must be held");
2079 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2080 struct rq *rq = cpu_rq(i);
2082 rq->dl.extra_bw += bw;
2087 void __dl_update(struct dl_bw *dl_b, s64 bw)
2089 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2096 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2101 struct u64_stats_sync sync;
2104 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2107 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2108 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2109 * and never move forward.
2111 static inline u64 irq_time_read(int cpu)
2113 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2118 seq = __u64_stats_fetch_begin(&irqtime->sync);
2119 total = irqtime->total;
2120 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2124 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2126 #ifdef CONFIG_CPU_FREQ
2127 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2130 * cpufreq_update_util - Take a note about CPU utilization changes.
2131 * @rq: Runqueue to carry out the update for.
2132 * @flags: Update reason flags.
2134 * This function is called by the scheduler on the CPU whose utilization is
2137 * It can only be called from RCU-sched read-side critical sections.
2139 * The way cpufreq is currently arranged requires it to evaluate the CPU
2140 * performance state (frequency/voltage) on a regular basis to prevent it from
2141 * being stuck in a completely inadequate performance level for too long.
2142 * That is not guaranteed to happen if the updates are only triggered from CFS
2143 * and DL, though, because they may not be coming in if only RT tasks are
2144 * active all the time (or there are RT tasks only).
2146 * As a workaround for that issue, this function is called periodically by the
2147 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2148 * but that really is a band-aid. Going forward it should be replaced with
2149 * solutions targeted more specifically at RT tasks.
2151 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2153 struct update_util_data *data;
2155 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2158 data->func(data, rq_clock(rq), flags);
2161 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2162 #endif /* CONFIG_CPU_FREQ */
2164 #ifdef arch_scale_freq_capacity
2165 # ifndef arch_scale_freq_invariant
2166 # define arch_scale_freq_invariant() true
2169 # define arch_scale_freq_invariant() false
2172 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2173 static inline unsigned long cpu_util_dl(struct rq *rq)
2175 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2178 static inline unsigned long cpu_util_cfs(struct rq *rq)
2180 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2182 if (sched_feat(UTIL_EST)) {
2183 util = max_t(unsigned long, util,
2184 READ_ONCE(rq->cfs.avg.util_est.enqueued));