1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/init_task.h>
49 #include <linux/kprobes.h>
50 #include <linux/kthread.h>
51 #include <linux/membarrier.h>
52 #include <linux/migrate.h>
53 #include <linux/mmu_context.h>
54 #include <linux/nmi.h>
55 #include <linux/proc_fs.h>
56 #include <linux/prefetch.h>
57 #include <linux/profile.h>
58 #include <linux/psi.h>
59 #include <linux/rcupdate_wait.h>
60 #include <linux/security.h>
61 #include <linux/stop_machine.h>
62 #include <linux/suspend.h>
63 #include <linux/swait.h>
64 #include <linux/syscalls.h>
65 #include <linux/task_work.h>
66 #include <linux/tsacct_kern.h>
70 #ifdef CONFIG_PARAVIRT
71 # include <asm/paravirt.h>
75 #include "cpudeadline.h"
77 #ifdef CONFIG_SCHED_DEBUG
78 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
80 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
86 /* task_struct::on_rq states: */
87 #define TASK_ON_RQ_QUEUED 1
88 #define TASK_ON_RQ_MIGRATING 2
90 extern __read_mostly int scheduler_running;
92 extern unsigned long calc_load_update;
93 extern atomic_long_t calc_load_tasks;
95 extern void calc_global_load_tick(struct rq *this_rq);
96 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
99 extern void cpu_load_update_active(struct rq *this_rq);
101 static inline void cpu_load_update_active(struct rq *this_rq) { }
105 * Helpers for converting nanosecond timing to jiffy resolution
107 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
116 * We increase resolution only if we have enough bits to allow this increased
117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
124 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
128 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
129 # define scale_load(w) (w)
130 # define scale_load_down(w) (w)
134 * Task weight (visible to users) and its load (invisible to users) have
135 * independent resolution, but they should be well calibrated. We use
136 * scale_load() and scale_load_down(w) to convert between them. The
137 * following must be true:
139 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
142 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
145 * Single value that decides SCHED_DEADLINE internal math precision.
146 * 10 -> just above 1us
147 * 9 -> just above 0.5us
152 * Single value that denotes runtime == period, ie unlimited time.
154 #define RUNTIME_INF ((u64)~0ULL)
156 static inline int idle_policy(int policy)
158 return policy == SCHED_IDLE;
160 static inline int fair_policy(int policy)
162 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
165 static inline int rt_policy(int policy)
167 return policy == SCHED_FIFO || policy == SCHED_RR;
170 static inline int dl_policy(int policy)
172 return policy == SCHED_DEADLINE;
174 static inline bool valid_policy(int policy)
176 return idle_policy(policy) || fair_policy(policy) ||
177 rt_policy(policy) || dl_policy(policy);
180 static inline int task_has_idle_policy(struct task_struct *p)
182 return idle_policy(p->policy);
185 static inline int task_has_rt_policy(struct task_struct *p)
187 return rt_policy(p->policy);
190 static inline int task_has_dl_policy(struct task_struct *p)
192 return dl_policy(p->policy);
195 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
198 * !! For sched_setattr_nocheck() (kernel) only !!
200 * This is actually gross. :(
202 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
203 * tasks, but still be able to sleep. We need this on platforms that cannot
204 * atomically change clock frequency. Remove once fast switching will be
205 * available on such platforms.
207 * SUGOV stands for SchedUtil GOVernor.
209 #define SCHED_FLAG_SUGOV 0x10000000
211 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
213 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
214 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
221 * Tells if entity @a should preempt entity @b.
224 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
226 return dl_entity_is_special(a) ||
227 dl_time_before(a->deadline, b->deadline);
231 * This is the priority-queue data structure of the RT scheduling class:
233 struct rt_prio_array {
234 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
235 struct list_head queue[MAX_RT_PRIO];
238 struct rt_bandwidth {
239 /* nests inside the rq lock: */
240 raw_spinlock_t rt_runtime_lock;
243 struct hrtimer rt_period_timer;
244 unsigned int rt_period_active;
247 void __dl_clear_params(struct task_struct *p);
250 * To keep the bandwidth of -deadline tasks and groups under control
251 * we need some place where:
252 * - store the maximum -deadline bandwidth of the system (the group);
253 * - cache the fraction of that bandwidth that is currently allocated.
255 * This is all done in the data structure below. It is similar to the
256 * one used for RT-throttling (rt_bandwidth), with the main difference
257 * that, since here we are only interested in admission control, we
258 * do not decrease any runtime while the group "executes", neither we
259 * need a timer to replenish it.
261 * With respect to SMP, the bandwidth is given on a per-CPU basis,
263 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
264 * - dl_total_bw array contains, in the i-eth element, the currently
265 * allocated bandwidth on the i-eth CPU.
266 * Moreover, groups consume bandwidth on each CPU, while tasks only
267 * consume bandwidth on the CPU they're running on.
268 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
269 * that will be shown the next time the proc or cgroup controls will
270 * be red. It on its turn can be changed by writing on its own
273 struct dl_bandwidth {
274 raw_spinlock_t dl_runtime_lock;
279 static inline int dl_bandwidth_enabled(void)
281 return sysctl_sched_rt_runtime >= 0;
290 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
293 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
295 dl_b->total_bw -= tsk_bw;
296 __dl_update(dl_b, (s32)tsk_bw / cpus);
300 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
302 dl_b->total_bw += tsk_bw;
303 __dl_update(dl_b, -((s32)tsk_bw / cpus));
307 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
309 return dl_b->bw != -1 &&
310 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
313 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
314 extern void init_dl_bw(struct dl_bw *dl_b);
315 extern int sched_dl_global_validate(void);
316 extern void sched_dl_do_global(void);
317 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
318 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
319 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
320 extern bool __checkparam_dl(const struct sched_attr *attr);
321 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
322 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
323 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
324 extern bool dl_cpu_busy(unsigned int cpu);
326 #ifdef CONFIG_CGROUP_SCHED
328 #include <linux/cgroup.h>
329 #include <linux/psi.h>
334 extern struct list_head task_groups;
336 struct cfs_bandwidth {
337 #ifdef CONFIG_CFS_BANDWIDTH
342 s64 hierarchical_quota;
348 struct hrtimer period_timer;
349 struct hrtimer slack_timer;
350 struct list_head throttled_cfs_rq;
357 bool distribute_running;
361 /* Task group related information */
363 struct cgroup_subsys_state css;
365 #ifdef CONFIG_FAIR_GROUP_SCHED
366 /* schedulable entities of this group on each CPU */
367 struct sched_entity **se;
368 /* runqueue "owned" by this group on each CPU */
369 struct cfs_rq **cfs_rq;
370 unsigned long shares;
374 * load_avg can be heavily contended at clock tick time, so put
375 * it in its own cacheline separated from the fields above which
376 * will also be accessed at each tick.
378 atomic_long_t load_avg ____cacheline_aligned;
382 #ifdef CONFIG_RT_GROUP_SCHED
383 struct sched_rt_entity **rt_se;
384 struct rt_rq **rt_rq;
386 struct rt_bandwidth rt_bandwidth;
390 struct list_head list;
392 struct task_group *parent;
393 struct list_head siblings;
394 struct list_head children;
396 #ifdef CONFIG_SCHED_AUTOGROUP
397 struct autogroup *autogroup;
400 struct cfs_bandwidth cfs_bandwidth;
403 #ifdef CONFIG_FAIR_GROUP_SCHED
404 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
407 * A weight of 0 or 1 can cause arithmetics problems.
408 * A weight of a cfs_rq is the sum of weights of which entities
409 * are queued on this cfs_rq, so a weight of a entity should not be
410 * too large, so as the shares value of a task group.
411 * (The default weight is 1024 - so there's no practical
412 * limitation from this.)
414 #define MIN_SHARES (1UL << 1)
415 #define MAX_SHARES (1UL << 18)
418 typedef int (*tg_visitor)(struct task_group *, void *);
420 extern int walk_tg_tree_from(struct task_group *from,
421 tg_visitor down, tg_visitor up, void *data);
424 * Iterate the full tree, calling @down when first entering a node and @up when
425 * leaving it for the final time.
427 * Caller must hold rcu_lock or sufficient equivalent.
429 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
431 return walk_tg_tree_from(&root_task_group, down, up, data);
434 extern int tg_nop(struct task_group *tg, void *data);
436 extern void free_fair_sched_group(struct task_group *tg);
437 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
438 extern void online_fair_sched_group(struct task_group *tg);
439 extern void unregister_fair_sched_group(struct task_group *tg);
440 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
441 struct sched_entity *se, int cpu,
442 struct sched_entity *parent);
443 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
445 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
446 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
447 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
449 extern void free_rt_sched_group(struct task_group *tg);
450 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
451 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
452 struct sched_rt_entity *rt_se, int cpu,
453 struct sched_rt_entity *parent);
454 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
455 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
456 extern long sched_group_rt_runtime(struct task_group *tg);
457 extern long sched_group_rt_period(struct task_group *tg);
458 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
460 extern struct task_group *sched_create_group(struct task_group *parent);
461 extern void sched_online_group(struct task_group *tg,
462 struct task_group *parent);
463 extern void sched_destroy_group(struct task_group *tg);
464 extern void sched_offline_group(struct task_group *tg);
466 extern void sched_move_task(struct task_struct *tsk);
468 #ifdef CONFIG_FAIR_GROUP_SCHED
469 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
472 extern void set_task_rq_fair(struct sched_entity *se,
473 struct cfs_rq *prev, struct cfs_rq *next);
474 #else /* !CONFIG_SMP */
475 static inline void set_task_rq_fair(struct sched_entity *se,
476 struct cfs_rq *prev, struct cfs_rq *next) { }
477 #endif /* CONFIG_SMP */
478 #endif /* CONFIG_FAIR_GROUP_SCHED */
480 #else /* CONFIG_CGROUP_SCHED */
482 struct cfs_bandwidth { };
484 #endif /* CONFIG_CGROUP_SCHED */
486 /* CFS-related fields in a runqueue */
488 struct load_weight load;
489 unsigned long runnable_weight;
490 unsigned int nr_running;
491 unsigned int h_nr_running;
496 u64 min_vruntime_copy;
499 struct rb_root_cached tasks_timeline;
502 * 'curr' points to currently running entity on this cfs_rq.
503 * It is set to NULL otherwise (i.e when none are currently running).
505 struct sched_entity *curr;
506 struct sched_entity *next;
507 struct sched_entity *last;
508 struct sched_entity *skip;
510 #ifdef CONFIG_SCHED_DEBUG
511 unsigned int nr_spread_over;
518 struct sched_avg avg;
520 u64 load_last_update_time_copy;
523 raw_spinlock_t lock ____cacheline_aligned;
525 unsigned long load_avg;
526 unsigned long util_avg;
527 unsigned long runnable_sum;
530 #ifdef CONFIG_FAIR_GROUP_SCHED
531 unsigned long tg_load_avg_contrib;
533 long prop_runnable_sum;
536 * h_load = weight * f(tg)
538 * Where f(tg) is the recursive weight fraction assigned to
541 unsigned long h_load;
542 u64 last_h_load_update;
543 struct sched_entity *h_load_next;
544 #endif /* CONFIG_FAIR_GROUP_SCHED */
545 #endif /* CONFIG_SMP */
547 #ifdef CONFIG_FAIR_GROUP_SCHED
548 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
551 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
552 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
553 * (like users, containers etc.)
555 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
556 * This list is used during load balance.
559 struct list_head leaf_cfs_rq_list;
560 struct task_group *tg; /* group that "owns" this runqueue */
562 #ifdef CONFIG_CFS_BANDWIDTH
566 s64 runtime_remaining;
569 u64 throttled_clock_task;
570 u64 throttled_clock_task_time;
573 struct list_head throttled_list;
574 #endif /* CONFIG_CFS_BANDWIDTH */
575 #endif /* CONFIG_FAIR_GROUP_SCHED */
578 static inline int rt_bandwidth_enabled(void)
580 return sysctl_sched_rt_runtime >= 0;
583 /* RT IPI pull logic requires IRQ_WORK */
584 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
585 # define HAVE_RT_PUSH_IPI
588 /* Real-Time classes' related field in a runqueue: */
590 struct rt_prio_array active;
591 unsigned int rt_nr_running;
592 unsigned int rr_nr_running;
593 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
595 int curr; /* highest queued rt task prio */
597 int next; /* next highest */
602 unsigned long rt_nr_migratory;
603 unsigned long rt_nr_total;
605 struct plist_head pushable_tasks;
607 #endif /* CONFIG_SMP */
613 /* Nests inside the rq lock: */
614 raw_spinlock_t rt_runtime_lock;
616 #ifdef CONFIG_RT_GROUP_SCHED
617 unsigned long rt_nr_boosted;
620 struct task_group *tg;
624 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
626 return rt_rq->rt_queued && rt_rq->rt_nr_running;
629 /* Deadline class' related fields in a runqueue */
631 /* runqueue is an rbtree, ordered by deadline */
632 struct rb_root_cached root;
634 unsigned long dl_nr_running;
638 * Deadline values of the currently executing and the
639 * earliest ready task on this rq. Caching these facilitates
640 * the decision whether or not a ready but not running task
641 * should migrate somewhere else.
648 unsigned long dl_nr_migratory;
652 * Tasks on this rq that can be pushed away. They are kept in
653 * an rb-tree, ordered by tasks' deadlines, with caching
654 * of the leftmost (earliest deadline) element.
656 struct rb_root_cached pushable_dl_tasks_root;
661 * "Active utilization" for this runqueue: increased when a
662 * task wakes up (becomes TASK_RUNNING) and decreased when a
668 * Utilization of the tasks "assigned" to this runqueue (including
669 * the tasks that are in runqueue and the tasks that executed on this
670 * CPU and blocked). Increased when a task moves to this runqueue, and
671 * decreased when the task moves away (migrates, changes scheduling
672 * policy, or terminates).
673 * This is needed to compute the "inactive utilization" for the
674 * runqueue (inactive utilization = this_bw - running_bw).
680 * Inverse of the fraction of CPU utilization that can be reclaimed
681 * by the GRUB algorithm.
686 #ifdef CONFIG_FAIR_GROUP_SCHED
687 /* An entity is a task if it doesn't "own" a runqueue */
688 #define entity_is_task(se) (!se->my_q)
690 #define entity_is_task(se) 1
695 * XXX we want to get rid of these helpers and use the full load resolution.
697 static inline long se_weight(struct sched_entity *se)
699 return scale_load_down(se->load.weight);
702 static inline long se_runnable(struct sched_entity *se)
704 return scale_load_down(se->runnable_weight);
707 static inline bool sched_asym_prefer(int a, int b)
709 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
713 * We add the notion of a root-domain which will be used to define per-domain
714 * variables. Each exclusive cpuset essentially defines an island domain by
715 * fully partitioning the member CPUs from any other cpuset. Whenever a new
716 * exclusive cpuset is created, we also create and attach a new root-domain
725 cpumask_var_t online;
728 * Indicate pullable load on at least one CPU, e.g:
729 * - More than one runnable task
730 * - Running task is misfit
735 * The bit corresponding to a CPU gets set here if such CPU has more
736 * than one runnable -deadline task (as it is below for RT tasks).
738 cpumask_var_t dlo_mask;
743 #ifdef HAVE_RT_PUSH_IPI
745 * For IPI pull requests, loop across the rto_mask.
747 struct irq_work rto_push_work;
748 raw_spinlock_t rto_lock;
749 /* These are only updated and read within rto_lock */
752 /* These atomics are updated outside of a lock */
753 atomic_t rto_loop_next;
754 atomic_t rto_loop_start;
757 * The "RT overload" flag: it gets set if a CPU has more than
758 * one runnable RT task.
760 cpumask_var_t rto_mask;
761 struct cpupri cpupri;
763 unsigned long max_cpu_capacity;
766 extern struct root_domain def_root_domain;
767 extern struct mutex sched_domains_mutex;
769 extern void init_defrootdomain(void);
770 extern int sched_init_domains(const struct cpumask *cpu_map);
771 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
772 extern void sched_get_rd(struct root_domain *rd);
773 extern void sched_put_rd(struct root_domain *rd);
775 #ifdef HAVE_RT_PUSH_IPI
776 extern void rto_push_irq_work_func(struct irq_work *work);
778 #endif /* CONFIG_SMP */
781 * This is the main, per-CPU runqueue data structure.
783 * Locking rule: those places that want to lock multiple runqueues
784 * (such as the load balancing or the thread migration code), lock
785 * acquire operations must be ordered by ascending &runqueue.
792 * nr_running and cpu_load should be in the same cacheline because
793 * remote CPUs use both these fields when doing load calculation.
795 unsigned int nr_running;
796 #ifdef CONFIG_NUMA_BALANCING
797 unsigned int nr_numa_running;
798 unsigned int nr_preferred_running;
799 unsigned int numa_migrate_on;
801 #define CPU_LOAD_IDX_MAX 5
802 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
803 #ifdef CONFIG_NO_HZ_COMMON
805 unsigned long last_load_update_tick;
806 unsigned long last_blocked_load_update_tick;
807 unsigned int has_blocked_load;
808 #endif /* CONFIG_SMP */
809 unsigned int nohz_tick_stopped;
811 #endif /* CONFIG_NO_HZ_COMMON */
813 /* capture load from *all* tasks on this CPU: */
814 struct load_weight load;
815 unsigned long nr_load_updates;
822 #ifdef CONFIG_FAIR_GROUP_SCHED
823 /* list of leaf cfs_rq on this CPU: */
824 struct list_head leaf_cfs_rq_list;
825 struct list_head *tmp_alone_branch;
826 #endif /* CONFIG_FAIR_GROUP_SCHED */
829 * This is part of a global counter where only the total sum
830 * over all CPUs matters. A task can increase this counter on
831 * one CPU and if it got migrated afterwards it may decrease
832 * it on another CPU. Always updated under the runqueue lock:
834 unsigned long nr_uninterruptible;
836 struct task_struct *curr;
837 struct task_struct *idle;
838 struct task_struct *stop;
839 unsigned long next_balance;
840 struct mm_struct *prev_mm;
842 unsigned int clock_update_flags;
849 struct root_domain *rd;
850 struct sched_domain *sd;
852 unsigned long cpu_capacity;
853 unsigned long cpu_capacity_orig;
855 struct callback_head *balance_callback;
857 unsigned char idle_balance;
859 unsigned long misfit_task_load;
861 /* For active balancing */
864 struct cpu_stop_work active_balance_work;
866 /* CPU of this runqueue: */
870 struct list_head cfs_tasks;
872 struct sched_avg avg_rt;
873 struct sched_avg avg_dl;
874 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
875 struct sched_avg avg_irq;
880 /* This is used to determine avg_idle's max value */
881 u64 max_idle_balance_cost;
884 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
887 #ifdef CONFIG_PARAVIRT
890 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
891 u64 prev_steal_time_rq;
894 /* calc_load related fields */
895 unsigned long calc_load_update;
896 long calc_load_active;
898 #ifdef CONFIG_SCHED_HRTICK
900 int hrtick_csd_pending;
901 call_single_data_t hrtick_csd;
903 struct hrtimer hrtick_timer;
906 #ifdef CONFIG_SCHEDSTATS
908 struct sched_info rq_sched_info;
909 unsigned long long rq_cpu_time;
910 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
912 /* sys_sched_yield() stats */
913 unsigned int yld_count;
915 /* schedule() stats */
916 unsigned int sched_count;
917 unsigned int sched_goidle;
919 /* try_to_wake_up() stats */
920 unsigned int ttwu_count;
921 unsigned int ttwu_local;
925 struct llist_head wake_list;
928 #ifdef CONFIG_CPU_IDLE
929 /* Must be inspected within a rcu lock section */
930 struct cpuidle_state *idle_state;
934 static inline int cpu_of(struct rq *rq)
944 #ifdef CONFIG_SCHED_SMT
945 extern void __update_idle_core(struct rq *rq);
947 static inline void update_idle_core(struct rq *rq)
949 if (static_branch_unlikely(&sched_smt_present))
950 __update_idle_core(rq);
954 static inline void update_idle_core(struct rq *rq) { }
957 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
959 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
960 #define this_rq() this_cpu_ptr(&runqueues)
961 #define task_rq(p) cpu_rq(task_cpu(p))
962 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
963 #define raw_rq() raw_cpu_ptr(&runqueues)
965 extern void update_rq_clock(struct rq *rq);
967 static inline u64 __rq_clock_broken(struct rq *rq)
969 return READ_ONCE(rq->clock);
973 * rq::clock_update_flags bits
975 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
976 * call to __schedule(). This is an optimisation to avoid
977 * neighbouring rq clock updates.
979 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
980 * in effect and calls to update_rq_clock() are being ignored.
982 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
983 * made to update_rq_clock() since the last time rq::lock was pinned.
985 * If inside of __schedule(), clock_update_flags will have been
986 * shifted left (a left shift is a cheap operation for the fast path
987 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
989 * if (rq-clock_update_flags >= RQCF_UPDATED)
991 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
992 * one position though, because the next rq_unpin_lock() will shift it
995 #define RQCF_REQ_SKIP 0x01
996 #define RQCF_ACT_SKIP 0x02
997 #define RQCF_UPDATED 0x04
999 static inline void assert_clock_updated(struct rq *rq)
1002 * The only reason for not seeing a clock update since the
1003 * last rq_pin_lock() is if we're currently skipping updates.
1005 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1008 static inline u64 rq_clock(struct rq *rq)
1010 lockdep_assert_held(&rq->lock);
1011 assert_clock_updated(rq);
1016 static inline u64 rq_clock_task(struct rq *rq)
1018 lockdep_assert_held(&rq->lock);
1019 assert_clock_updated(rq);
1021 return rq->clock_task;
1024 static inline void rq_clock_skip_update(struct rq *rq)
1026 lockdep_assert_held(&rq->lock);
1027 rq->clock_update_flags |= RQCF_REQ_SKIP;
1031 * See rt task throttling, which is the only time a skip
1032 * request is cancelled.
1034 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1036 lockdep_assert_held(&rq->lock);
1037 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1041 unsigned long flags;
1042 struct pin_cookie cookie;
1043 #ifdef CONFIG_SCHED_DEBUG
1045 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1046 * current pin context is stashed here in case it needs to be
1047 * restored in rq_repin_lock().
1049 unsigned int clock_update_flags;
1053 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1055 rf->cookie = lockdep_pin_lock(&rq->lock);
1057 #ifdef CONFIG_SCHED_DEBUG
1058 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1059 rf->clock_update_flags = 0;
1063 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1065 #ifdef CONFIG_SCHED_DEBUG
1066 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1067 rf->clock_update_flags = RQCF_UPDATED;
1070 lockdep_unpin_lock(&rq->lock, rf->cookie);
1073 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1075 lockdep_repin_lock(&rq->lock, rf->cookie);
1077 #ifdef CONFIG_SCHED_DEBUG
1079 * Restore the value we stashed in @rf for this pin context.
1081 rq->clock_update_flags |= rf->clock_update_flags;
1085 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1086 __acquires(rq->lock);
1088 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1089 __acquires(p->pi_lock)
1090 __acquires(rq->lock);
1092 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1093 __releases(rq->lock)
1095 rq_unpin_lock(rq, rf);
1096 raw_spin_unlock(&rq->lock);
1100 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1101 __releases(rq->lock)
1102 __releases(p->pi_lock)
1104 rq_unpin_lock(rq, rf);
1105 raw_spin_unlock(&rq->lock);
1106 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1110 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1111 __acquires(rq->lock)
1113 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1114 rq_pin_lock(rq, rf);
1118 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1119 __acquires(rq->lock)
1121 raw_spin_lock_irq(&rq->lock);
1122 rq_pin_lock(rq, rf);
1126 rq_lock(struct rq *rq, struct rq_flags *rf)
1127 __acquires(rq->lock)
1129 raw_spin_lock(&rq->lock);
1130 rq_pin_lock(rq, rf);
1134 rq_relock(struct rq *rq, struct rq_flags *rf)
1135 __acquires(rq->lock)
1137 raw_spin_lock(&rq->lock);
1138 rq_repin_lock(rq, rf);
1142 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1143 __releases(rq->lock)
1145 rq_unpin_lock(rq, rf);
1146 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1150 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1151 __releases(rq->lock)
1153 rq_unpin_lock(rq, rf);
1154 raw_spin_unlock_irq(&rq->lock);
1158 rq_unlock(struct rq *rq, struct rq_flags *rf)
1159 __releases(rq->lock)
1161 rq_unpin_lock(rq, rf);
1162 raw_spin_unlock(&rq->lock);
1165 static inline struct rq *
1166 this_rq_lock_irq(struct rq_flags *rf)
1167 __acquires(rq->lock)
1171 local_irq_disable();
1178 enum numa_topology_type {
1183 extern enum numa_topology_type sched_numa_topology_type;
1184 extern int sched_max_numa_distance;
1185 extern bool find_numa_distance(int distance);
1189 extern void sched_init_numa(void);
1190 extern void sched_domains_numa_masks_set(unsigned int cpu);
1191 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1193 static inline void sched_init_numa(void) { }
1194 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1195 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1198 #ifdef CONFIG_NUMA_BALANCING
1199 /* The regions in numa_faults array from task_struct */
1200 enum numa_faults_stats {
1206 extern void sched_setnuma(struct task_struct *p, int node);
1207 extern int migrate_task_to(struct task_struct *p, int cpu);
1208 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1210 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1213 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1216 #endif /* CONFIG_NUMA_BALANCING */
1221 queue_balance_callback(struct rq *rq,
1222 struct callback_head *head,
1223 void (*func)(struct rq *rq))
1225 lockdep_assert_held(&rq->lock);
1227 if (unlikely(head->next))
1230 head->func = (void (*)(struct callback_head *))func;
1231 head->next = rq->balance_callback;
1232 rq->balance_callback = head;
1235 extern void sched_ttwu_pending(void);
1237 #define rcu_dereference_check_sched_domain(p) \
1238 rcu_dereference_check((p), \
1239 lockdep_is_held(&sched_domains_mutex))
1242 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1243 * See detach_destroy_domains: synchronize_sched for details.
1245 * The domain tree of any CPU may only be accessed from within
1246 * preempt-disabled sections.
1248 #define for_each_domain(cpu, __sd) \
1249 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1250 __sd; __sd = __sd->parent)
1252 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1255 * highest_flag_domain - Return highest sched_domain containing flag.
1256 * @cpu: The CPU whose highest level of sched domain is to
1258 * @flag: The flag to check for the highest sched_domain
1259 * for the given CPU.
1261 * Returns the highest sched_domain of a CPU which contains the given flag.
1263 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1265 struct sched_domain *sd, *hsd = NULL;
1267 for_each_domain(cpu, sd) {
1268 if (!(sd->flags & flag))
1276 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1278 struct sched_domain *sd;
1280 for_each_domain(cpu, sd) {
1281 if (sd->flags & flag)
1288 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1289 DECLARE_PER_CPU(int, sd_llc_size);
1290 DECLARE_PER_CPU(int, sd_llc_id);
1291 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1292 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1293 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1294 extern struct static_key_false sched_asym_cpucapacity;
1296 struct sched_group_capacity {
1299 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1302 unsigned long capacity;
1303 unsigned long min_capacity; /* Min per-CPU capacity in group */
1304 unsigned long max_capacity; /* Max per-CPU capacity in group */
1305 unsigned long next_update;
1306 int imbalance; /* XXX unrelated to capacity but shared group state */
1308 #ifdef CONFIG_SCHED_DEBUG
1312 unsigned long cpumask[0]; /* Balance mask */
1315 struct sched_group {
1316 struct sched_group *next; /* Must be a circular list */
1319 unsigned int group_weight;
1320 struct sched_group_capacity *sgc;
1321 int asym_prefer_cpu; /* CPU of highest priority in group */
1324 * The CPUs this group covers.
1326 * NOTE: this field is variable length. (Allocated dynamically
1327 * by attaching extra space to the end of the structure,
1328 * depending on how many CPUs the kernel has booted up with)
1330 unsigned long cpumask[0];
1333 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1335 return to_cpumask(sg->cpumask);
1339 * See build_balance_mask().
1341 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1343 return to_cpumask(sg->sgc->cpumask);
1347 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1348 * @group: The group whose first CPU is to be returned.
1350 static inline unsigned int group_first_cpu(struct sched_group *group)
1352 return cpumask_first(sched_group_span(group));
1355 extern int group_balance_cpu(struct sched_group *sg);
1357 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1358 void register_sched_domain_sysctl(void);
1359 void dirty_sched_domain_sysctl(int cpu);
1360 void unregister_sched_domain_sysctl(void);
1362 static inline void register_sched_domain_sysctl(void)
1365 static inline void dirty_sched_domain_sysctl(int cpu)
1368 static inline void unregister_sched_domain_sysctl(void)
1375 static inline void sched_ttwu_pending(void) { }
1377 #endif /* CONFIG_SMP */
1380 #include "autogroup.h"
1382 #ifdef CONFIG_CGROUP_SCHED
1385 * Return the group to which this tasks belongs.
1387 * We cannot use task_css() and friends because the cgroup subsystem
1388 * changes that value before the cgroup_subsys::attach() method is called,
1389 * therefore we cannot pin it and might observe the wrong value.
1391 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1392 * core changes this before calling sched_move_task().
1394 * Instead we use a 'copy' which is updated from sched_move_task() while
1395 * holding both task_struct::pi_lock and rq::lock.
1397 static inline struct task_group *task_group(struct task_struct *p)
1399 return p->sched_task_group;
1402 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1403 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1405 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1406 struct task_group *tg = task_group(p);
1409 #ifdef CONFIG_FAIR_GROUP_SCHED
1410 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1411 p->se.cfs_rq = tg->cfs_rq[cpu];
1412 p->se.parent = tg->se[cpu];
1415 #ifdef CONFIG_RT_GROUP_SCHED
1416 p->rt.rt_rq = tg->rt_rq[cpu];
1417 p->rt.parent = tg->rt_se[cpu];
1421 #else /* CONFIG_CGROUP_SCHED */
1423 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1424 static inline struct task_group *task_group(struct task_struct *p)
1429 #endif /* CONFIG_CGROUP_SCHED */
1431 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1433 set_task_rq(p, cpu);
1436 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1437 * successfully executed on another CPU. We must ensure that updates of
1438 * per-task data have been completed by this moment.
1441 #ifdef CONFIG_THREAD_INFO_IN_TASK
1444 task_thread_info(p)->cpu = cpu;
1451 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1453 #ifdef CONFIG_SCHED_DEBUG
1454 # include <linux/static_key.h>
1455 # define const_debug __read_mostly
1457 # define const_debug const
1460 #define SCHED_FEAT(name, enabled) \
1461 __SCHED_FEAT_##name ,
1464 #include "features.h"
1470 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1473 * To support run-time toggling of sched features, all the translation units
1474 * (but core.c) reference the sysctl_sched_features defined in core.c.
1476 extern const_debug unsigned int sysctl_sched_features;
1478 #define SCHED_FEAT(name, enabled) \
1479 static __always_inline bool static_branch_##name(struct static_key *key) \
1481 return static_key_##enabled(key); \
1484 #include "features.h"
1487 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1488 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1490 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1493 * Each translation unit has its own copy of sysctl_sched_features to allow
1494 * constants propagation at compile time and compiler optimization based on
1497 #define SCHED_FEAT(name, enabled) \
1498 (1UL << __SCHED_FEAT_##name) * enabled |
1499 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1500 #include "features.h"
1504 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1506 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1508 extern struct static_key_false sched_numa_balancing;
1509 extern struct static_key_false sched_schedstats;
1511 static inline u64 global_rt_period(void)
1513 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1516 static inline u64 global_rt_runtime(void)
1518 if (sysctl_sched_rt_runtime < 0)
1521 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1524 static inline int task_current(struct rq *rq, struct task_struct *p)
1526 return rq->curr == p;
1529 static inline int task_running(struct rq *rq, struct task_struct *p)
1534 return task_current(rq, p);
1538 static inline int task_on_rq_queued(struct task_struct *p)
1540 return p->on_rq == TASK_ON_RQ_QUEUED;
1543 static inline int task_on_rq_migrating(struct task_struct *p)
1545 return p->on_rq == TASK_ON_RQ_MIGRATING;
1551 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1552 #define WF_FORK 0x02 /* Child wakeup after fork */
1553 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1556 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1557 * of tasks with abnormal "nice" values across CPUs the contribution that
1558 * each task makes to its run queue's load is weighted according to its
1559 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1560 * scaled version of the new time slice allocation that they receive on time
1564 #define WEIGHT_IDLEPRIO 3
1565 #define WMULT_IDLEPRIO 1431655765
1567 extern const int sched_prio_to_weight[40];
1568 extern const u32 sched_prio_to_wmult[40];
1571 * {de,en}queue flags:
1573 * DEQUEUE_SLEEP - task is no longer runnable
1574 * ENQUEUE_WAKEUP - task just became runnable
1576 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1577 * are in a known state which allows modification. Such pairs
1578 * should preserve as much state as possible.
1580 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1583 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1584 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1585 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1589 #define DEQUEUE_SLEEP 0x01
1590 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1591 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1592 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1594 #define ENQUEUE_WAKEUP 0x01
1595 #define ENQUEUE_RESTORE 0x02
1596 #define ENQUEUE_MOVE 0x04
1597 #define ENQUEUE_NOCLOCK 0x08
1599 #define ENQUEUE_HEAD 0x10
1600 #define ENQUEUE_REPLENISH 0x20
1602 #define ENQUEUE_MIGRATED 0x40
1604 #define ENQUEUE_MIGRATED 0x00
1607 #define RETRY_TASK ((void *)-1UL)
1609 struct sched_class {
1610 const struct sched_class *next;
1612 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1613 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1614 void (*yield_task) (struct rq *rq);
1615 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1617 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1620 * It is the responsibility of the pick_next_task() method that will
1621 * return the next task to call put_prev_task() on the @prev task or
1622 * something equivalent.
1624 * May return RETRY_TASK when it finds a higher prio class has runnable
1627 struct task_struct * (*pick_next_task)(struct rq *rq,
1628 struct task_struct *prev,
1629 struct rq_flags *rf);
1630 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1633 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1634 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1636 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1638 void (*set_cpus_allowed)(struct task_struct *p,
1639 const struct cpumask *newmask);
1641 void (*rq_online)(struct rq *rq);
1642 void (*rq_offline)(struct rq *rq);
1645 void (*set_curr_task)(struct rq *rq);
1646 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1647 void (*task_fork)(struct task_struct *p);
1648 void (*task_dead)(struct task_struct *p);
1651 * The switched_from() call is allowed to drop rq->lock, therefore we
1652 * cannot assume the switched_from/switched_to pair is serliazed by
1653 * rq->lock. They are however serialized by p->pi_lock.
1655 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1656 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1657 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1660 unsigned int (*get_rr_interval)(struct rq *rq,
1661 struct task_struct *task);
1663 void (*update_curr)(struct rq *rq);
1665 #define TASK_SET_GROUP 0
1666 #define TASK_MOVE_GROUP 1
1668 #ifdef CONFIG_FAIR_GROUP_SCHED
1669 void (*task_change_group)(struct task_struct *p, int type);
1673 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1675 prev->sched_class->put_prev_task(rq, prev);
1678 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1680 curr->sched_class->set_curr_task(rq);
1684 #define sched_class_highest (&stop_sched_class)
1686 #define sched_class_highest (&dl_sched_class)
1688 #define for_each_class(class) \
1689 for (class = sched_class_highest; class; class = class->next)
1691 extern const struct sched_class stop_sched_class;
1692 extern const struct sched_class dl_sched_class;
1693 extern const struct sched_class rt_sched_class;
1694 extern const struct sched_class fair_sched_class;
1695 extern const struct sched_class idle_sched_class;
1700 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1702 extern void trigger_load_balance(struct rq *rq);
1704 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1708 #ifdef CONFIG_CPU_IDLE
1709 static inline void idle_set_state(struct rq *rq,
1710 struct cpuidle_state *idle_state)
1712 rq->idle_state = idle_state;
1715 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1717 SCHED_WARN_ON(!rcu_read_lock_held());
1719 return rq->idle_state;
1722 static inline void idle_set_state(struct rq *rq,
1723 struct cpuidle_state *idle_state)
1727 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1733 extern void schedule_idle(void);
1735 extern void sysrq_sched_debug_show(void);
1736 extern void sched_init_granularity(void);
1737 extern void update_max_interval(void);
1739 extern void init_sched_dl_class(void);
1740 extern void init_sched_rt_class(void);
1741 extern void init_sched_fair_class(void);
1743 extern void reweight_task(struct task_struct *p, int prio);
1745 extern void resched_curr(struct rq *rq);
1746 extern void resched_cpu(int cpu);
1748 extern struct rt_bandwidth def_rt_bandwidth;
1749 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1751 extern struct dl_bandwidth def_dl_bandwidth;
1752 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1753 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1754 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1755 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1758 #define BW_UNIT (1 << BW_SHIFT)
1759 #define RATIO_SHIFT 8
1760 unsigned long to_ratio(u64 period, u64 runtime);
1762 extern void init_entity_runnable_average(struct sched_entity *se);
1763 extern void post_init_entity_util_avg(struct sched_entity *se);
1765 #ifdef CONFIG_NO_HZ_FULL
1766 extern bool sched_can_stop_tick(struct rq *rq);
1767 extern int __init sched_tick_offload_init(void);
1770 * Tick may be needed by tasks in the runqueue depending on their policy and
1771 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1772 * nohz mode if necessary.
1774 static inline void sched_update_tick_dependency(struct rq *rq)
1778 if (!tick_nohz_full_enabled())
1783 if (!tick_nohz_full_cpu(cpu))
1786 if (sched_can_stop_tick(rq))
1787 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1789 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1792 static inline int sched_tick_offload_init(void) { return 0; }
1793 static inline void sched_update_tick_dependency(struct rq *rq) { }
1796 static inline void add_nr_running(struct rq *rq, unsigned count)
1798 unsigned prev_nr = rq->nr_running;
1800 rq->nr_running = prev_nr + count;
1803 if (prev_nr < 2 && rq->nr_running >= 2) {
1804 if (!READ_ONCE(rq->rd->overload))
1805 WRITE_ONCE(rq->rd->overload, 1);
1809 sched_update_tick_dependency(rq);
1812 static inline void sub_nr_running(struct rq *rq, unsigned count)
1814 rq->nr_running -= count;
1815 /* Check if we still need preemption */
1816 sched_update_tick_dependency(rq);
1819 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1820 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1822 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1824 extern const_debug unsigned int sysctl_sched_nr_migrate;
1825 extern const_debug unsigned int sysctl_sched_migration_cost;
1827 #ifdef CONFIG_SCHED_HRTICK
1831 * - enabled by features
1832 * - hrtimer is actually high res
1834 static inline int hrtick_enabled(struct rq *rq)
1836 if (!sched_feat(HRTICK))
1838 if (!cpu_active(cpu_of(rq)))
1840 return hrtimer_is_hres_active(&rq->hrtick_timer);
1843 void hrtick_start(struct rq *rq, u64 delay);
1847 static inline int hrtick_enabled(struct rq *rq)
1852 #endif /* CONFIG_SCHED_HRTICK */
1854 #ifndef arch_scale_freq_capacity
1855 static __always_inline
1856 unsigned long arch_scale_freq_capacity(int cpu)
1858 return SCHED_CAPACITY_SCALE;
1863 #ifdef CONFIG_PREEMPT
1865 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1868 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1869 * way at the expense of forcing extra atomic operations in all
1870 * invocations. This assures that the double_lock is acquired using the
1871 * same underlying policy as the spinlock_t on this architecture, which
1872 * reduces latency compared to the unfair variant below. However, it
1873 * also adds more overhead and therefore may reduce throughput.
1875 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1876 __releases(this_rq->lock)
1877 __acquires(busiest->lock)
1878 __acquires(this_rq->lock)
1880 raw_spin_unlock(&this_rq->lock);
1881 double_rq_lock(this_rq, busiest);
1888 * Unfair double_lock_balance: Optimizes throughput at the expense of
1889 * latency by eliminating extra atomic operations when the locks are
1890 * already in proper order on entry. This favors lower CPU-ids and will
1891 * grant the double lock to lower CPUs over higher ids under contention,
1892 * regardless of entry order into the function.
1894 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1895 __releases(this_rq->lock)
1896 __acquires(busiest->lock)
1897 __acquires(this_rq->lock)
1901 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1902 if (busiest < this_rq) {
1903 raw_spin_unlock(&this_rq->lock);
1904 raw_spin_lock(&busiest->lock);
1905 raw_spin_lock_nested(&this_rq->lock,
1906 SINGLE_DEPTH_NESTING);
1909 raw_spin_lock_nested(&busiest->lock,
1910 SINGLE_DEPTH_NESTING);
1915 #endif /* CONFIG_PREEMPT */
1918 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1920 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1922 if (unlikely(!irqs_disabled())) {
1923 /* printk() doesn't work well under rq->lock */
1924 raw_spin_unlock(&this_rq->lock);
1928 return _double_lock_balance(this_rq, busiest);
1931 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1932 __releases(busiest->lock)
1934 raw_spin_unlock(&busiest->lock);
1935 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1938 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1944 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1947 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1953 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1956 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1962 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1966 * double_rq_lock - safely lock two runqueues
1968 * Note this does not disable interrupts like task_rq_lock,
1969 * you need to do so manually before calling.
1971 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1972 __acquires(rq1->lock)
1973 __acquires(rq2->lock)
1975 BUG_ON(!irqs_disabled());
1977 raw_spin_lock(&rq1->lock);
1978 __acquire(rq2->lock); /* Fake it out ;) */
1981 raw_spin_lock(&rq1->lock);
1982 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1984 raw_spin_lock(&rq2->lock);
1985 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1991 * double_rq_unlock - safely unlock two runqueues
1993 * Note this does not restore interrupts like task_rq_unlock,
1994 * you need to do so manually after calling.
1996 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1997 __releases(rq1->lock)
1998 __releases(rq2->lock)
2000 raw_spin_unlock(&rq1->lock);
2002 raw_spin_unlock(&rq2->lock);
2004 __release(rq2->lock);
2007 extern void set_rq_online (struct rq *rq);
2008 extern void set_rq_offline(struct rq *rq);
2009 extern bool sched_smp_initialized;
2011 #else /* CONFIG_SMP */
2014 * double_rq_lock - safely lock two runqueues
2016 * Note this does not disable interrupts like task_rq_lock,
2017 * you need to do so manually before calling.
2019 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2020 __acquires(rq1->lock)
2021 __acquires(rq2->lock)
2023 BUG_ON(!irqs_disabled());
2025 raw_spin_lock(&rq1->lock);
2026 __acquire(rq2->lock); /* Fake it out ;) */
2030 * double_rq_unlock - safely unlock two runqueues
2032 * Note this does not restore interrupts like task_rq_unlock,
2033 * you need to do so manually after calling.
2035 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2036 __releases(rq1->lock)
2037 __releases(rq2->lock)
2040 raw_spin_unlock(&rq1->lock);
2041 __release(rq2->lock);
2046 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2047 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2049 #ifdef CONFIG_SCHED_DEBUG
2050 extern bool sched_debug_enabled;
2052 extern void print_cfs_stats(struct seq_file *m, int cpu);
2053 extern void print_rt_stats(struct seq_file *m, int cpu);
2054 extern void print_dl_stats(struct seq_file *m, int cpu);
2055 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2056 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2057 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2058 #ifdef CONFIG_NUMA_BALANCING
2060 show_numa_stats(struct task_struct *p, struct seq_file *m);
2062 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2063 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2064 #endif /* CONFIG_NUMA_BALANCING */
2065 #endif /* CONFIG_SCHED_DEBUG */
2067 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2068 extern void init_rt_rq(struct rt_rq *rt_rq);
2069 extern void init_dl_rq(struct dl_rq *dl_rq);
2071 extern void cfs_bandwidth_usage_inc(void);
2072 extern void cfs_bandwidth_usage_dec(void);
2074 #ifdef CONFIG_NO_HZ_COMMON
2075 #define NOHZ_BALANCE_KICK_BIT 0
2076 #define NOHZ_STATS_KICK_BIT 1
2078 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2079 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2081 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2083 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2085 extern void nohz_balance_exit_idle(struct rq *rq);
2087 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2093 void __dl_update(struct dl_bw *dl_b, s64 bw)
2095 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2098 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2099 "sched RCU must be held");
2100 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2101 struct rq *rq = cpu_rq(i);
2103 rq->dl.extra_bw += bw;
2108 void __dl_update(struct dl_bw *dl_b, s64 bw)
2110 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2117 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2122 struct u64_stats_sync sync;
2125 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2128 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2129 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2130 * and never move forward.
2132 static inline u64 irq_time_read(int cpu)
2134 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2139 seq = __u64_stats_fetch_begin(&irqtime->sync);
2140 total = irqtime->total;
2141 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2145 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2147 #ifdef CONFIG_CPU_FREQ
2148 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2151 * cpufreq_update_util - Take a note about CPU utilization changes.
2152 * @rq: Runqueue to carry out the update for.
2153 * @flags: Update reason flags.
2155 * This function is called by the scheduler on the CPU whose utilization is
2158 * It can only be called from RCU-sched read-side critical sections.
2160 * The way cpufreq is currently arranged requires it to evaluate the CPU
2161 * performance state (frequency/voltage) on a regular basis to prevent it from
2162 * being stuck in a completely inadequate performance level for too long.
2163 * That is not guaranteed to happen if the updates are only triggered from CFS
2164 * and DL, though, because they may not be coming in if only RT tasks are
2165 * active all the time (or there are RT tasks only).
2167 * As a workaround for that issue, this function is called periodically by the
2168 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2169 * but that really is a band-aid. Going forward it should be replaced with
2170 * solutions targeted more specifically at RT tasks.
2172 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2174 struct update_util_data *data;
2176 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2179 data->func(data, rq_clock(rq), flags);
2182 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2183 #endif /* CONFIG_CPU_FREQ */
2185 #ifdef arch_scale_freq_capacity
2186 # ifndef arch_scale_freq_invariant
2187 # define arch_scale_freq_invariant() true
2190 # define arch_scale_freq_invariant() false
2193 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2195 * enum schedutil_type - CPU utilization type
2196 * @FREQUENCY_UTIL: Utilization used to select frequency
2197 * @ENERGY_UTIL: Utilization used during energy calculation
2199 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2200 * need to be aggregated differently depending on the usage made of them. This
2201 * enum is used within schedutil_freq_util() to differentiate the types of
2202 * utilization expected by the callers, and adjust the aggregation accordingly.
2204 enum schedutil_type {
2209 unsigned long schedutil_freq_util(int cpu, unsigned long util_cfs,
2210 unsigned long max, enum schedutil_type type);
2212 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2214 unsigned long max = arch_scale_cpu_capacity(NULL, cpu);
2216 return schedutil_freq_util(cpu, cfs, max, ENERGY_UTIL);
2219 static inline unsigned long cpu_bw_dl(struct rq *rq)
2221 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2224 static inline unsigned long cpu_util_dl(struct rq *rq)
2226 return READ_ONCE(rq->avg_dl.util_avg);
2229 static inline unsigned long cpu_util_cfs(struct rq *rq)
2231 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2233 if (sched_feat(UTIL_EST)) {
2234 util = max_t(unsigned long, util,
2235 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2241 static inline unsigned long cpu_util_rt(struct rq *rq)
2243 return READ_ONCE(rq->avg_rt.util_avg);
2245 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2246 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2252 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2253 static inline unsigned long cpu_util_irq(struct rq *rq)
2255 return rq->avg_irq.util_avg;
2259 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2261 util *= (max - irq);
2268 static inline unsigned long cpu_util_irq(struct rq *rq)
2274 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)