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/psi.h>
58 #include <linux/rcupdate_wait.h>
59 #include <linux/security.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_idle_policy(struct task_struct *p)
181 return idle_policy(p->policy);
184 static inline int task_has_rt_policy(struct task_struct *p)
186 return rt_policy(p->policy);
189 static inline int task_has_dl_policy(struct task_struct *p)
191 return dl_policy(p->policy);
194 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
197 * !! For sched_setattr_nocheck() (kernel) only !!
199 * This is actually gross. :(
201 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
202 * tasks, but still be able to sleep. We need this on platforms that cannot
203 * atomically change clock frequency. Remove once fast switching will be
204 * available on such platforms.
206 * SUGOV stands for SchedUtil GOVernor.
208 #define SCHED_FLAG_SUGOV 0x10000000
210 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
212 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
213 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
220 * Tells if entity @a should preempt entity @b.
223 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
225 return dl_entity_is_special(a) ||
226 dl_time_before(a->deadline, b->deadline);
230 * This is the priority-queue data structure of the RT scheduling class:
232 struct rt_prio_array {
233 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
234 struct list_head queue[MAX_RT_PRIO];
237 struct rt_bandwidth {
238 /* nests inside the rq lock: */
239 raw_spinlock_t rt_runtime_lock;
242 struct hrtimer rt_period_timer;
243 unsigned int rt_period_active;
246 void __dl_clear_params(struct task_struct *p);
249 * To keep the bandwidth of -deadline tasks and groups under control
250 * we need some place where:
251 * - store the maximum -deadline bandwidth of the system (the group);
252 * - cache the fraction of that bandwidth that is currently allocated.
254 * This is all done in the data structure below. It is similar to the
255 * one used for RT-throttling (rt_bandwidth), with the main difference
256 * that, since here we are only interested in admission control, we
257 * do not decrease any runtime while the group "executes", neither we
258 * need a timer to replenish it.
260 * With respect to SMP, the bandwidth is given on a per-CPU basis,
262 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
263 * - dl_total_bw array contains, in the i-eth element, the currently
264 * allocated bandwidth on the i-eth CPU.
265 * Moreover, groups consume bandwidth on each CPU, while tasks only
266 * consume bandwidth on the CPU they're running on.
267 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
268 * that will be shown the next time the proc or cgroup controls will
269 * be red. It on its turn can be changed by writing on its own
272 struct dl_bandwidth {
273 raw_spinlock_t dl_runtime_lock;
278 static inline int dl_bandwidth_enabled(void)
280 return sysctl_sched_rt_runtime >= 0;
289 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
292 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
294 dl_b->total_bw -= tsk_bw;
295 __dl_update(dl_b, (s32)tsk_bw / cpus);
299 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
301 dl_b->total_bw += tsk_bw;
302 __dl_update(dl_b, -((s32)tsk_bw / cpus));
306 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
308 return dl_b->bw != -1 &&
309 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
312 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
313 extern void init_dl_bw(struct dl_bw *dl_b);
314 extern int sched_dl_global_validate(void);
315 extern void sched_dl_do_global(void);
316 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
317 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
318 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
319 extern bool __checkparam_dl(const struct sched_attr *attr);
320 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
321 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
322 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
323 extern bool dl_cpu_busy(unsigned int cpu);
325 #ifdef CONFIG_CGROUP_SCHED
327 #include <linux/cgroup.h>
328 #include <linux/psi.h>
333 extern struct list_head task_groups;
335 struct cfs_bandwidth {
336 #ifdef CONFIG_CFS_BANDWIDTH
341 s64 hierarchical_quota;
347 struct hrtimer period_timer;
348 struct hrtimer slack_timer;
349 struct list_head throttled_cfs_rq;
356 bool distribute_running;
360 /* Task group related information */
362 struct cgroup_subsys_state css;
364 #ifdef CONFIG_FAIR_GROUP_SCHED
365 /* schedulable entities of this group on each CPU */
366 struct sched_entity **se;
367 /* runqueue "owned" by this group on each CPU */
368 struct cfs_rq **cfs_rq;
369 unsigned long shares;
373 * load_avg can be heavily contended at clock tick time, so put
374 * it in its own cacheline separated from the fields above which
375 * will also be accessed at each tick.
377 atomic_long_t load_avg ____cacheline_aligned;
381 #ifdef CONFIG_RT_GROUP_SCHED
382 struct sched_rt_entity **rt_se;
383 struct rt_rq **rt_rq;
385 struct rt_bandwidth rt_bandwidth;
389 struct list_head list;
391 struct task_group *parent;
392 struct list_head siblings;
393 struct list_head children;
395 #ifdef CONFIG_SCHED_AUTOGROUP
396 struct autogroup *autogroup;
399 struct cfs_bandwidth cfs_bandwidth;
402 #ifdef CONFIG_FAIR_GROUP_SCHED
403 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
406 * A weight of 0 or 1 can cause arithmetics problems.
407 * A weight of a cfs_rq is the sum of weights of which entities
408 * are queued on this cfs_rq, so a weight of a entity should not be
409 * too large, so as the shares value of a task group.
410 * (The default weight is 1024 - so there's no practical
411 * limitation from this.)
413 #define MIN_SHARES (1UL << 1)
414 #define MAX_SHARES (1UL << 18)
417 typedef int (*tg_visitor)(struct task_group *, void *);
419 extern int walk_tg_tree_from(struct task_group *from,
420 tg_visitor down, tg_visitor up, void *data);
423 * Iterate the full tree, calling @down when first entering a node and @up when
424 * leaving it for the final time.
426 * Caller must hold rcu_lock or sufficient equivalent.
428 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
430 return walk_tg_tree_from(&root_task_group, down, up, data);
433 extern int tg_nop(struct task_group *tg, void *data);
435 extern void free_fair_sched_group(struct task_group *tg);
436 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
437 extern void online_fair_sched_group(struct task_group *tg);
438 extern void unregister_fair_sched_group(struct task_group *tg);
439 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
440 struct sched_entity *se, int cpu,
441 struct sched_entity *parent);
442 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
444 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
445 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
446 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
448 extern void free_rt_sched_group(struct task_group *tg);
449 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
450 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
451 struct sched_rt_entity *rt_se, int cpu,
452 struct sched_rt_entity *parent);
453 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
454 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
455 extern long sched_group_rt_runtime(struct task_group *tg);
456 extern long sched_group_rt_period(struct task_group *tg);
457 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
459 extern struct task_group *sched_create_group(struct task_group *parent);
460 extern void sched_online_group(struct task_group *tg,
461 struct task_group *parent);
462 extern void sched_destroy_group(struct task_group *tg);
463 extern void sched_offline_group(struct task_group *tg);
465 extern void sched_move_task(struct task_struct *tsk);
467 #ifdef CONFIG_FAIR_GROUP_SCHED
468 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
471 extern void set_task_rq_fair(struct sched_entity *se,
472 struct cfs_rq *prev, struct cfs_rq *next);
473 #else /* !CONFIG_SMP */
474 static inline void set_task_rq_fair(struct sched_entity *se,
475 struct cfs_rq *prev, struct cfs_rq *next) { }
476 #endif /* CONFIG_SMP */
477 #endif /* CONFIG_FAIR_GROUP_SCHED */
479 #else /* CONFIG_CGROUP_SCHED */
481 struct cfs_bandwidth { };
483 #endif /* CONFIG_CGROUP_SCHED */
485 /* CFS-related fields in a runqueue */
487 struct load_weight load;
488 unsigned long runnable_weight;
489 unsigned int nr_running;
490 unsigned int h_nr_running;
495 u64 min_vruntime_copy;
498 struct rb_root_cached tasks_timeline;
501 * 'curr' points to currently running entity on this cfs_rq.
502 * It is set to NULL otherwise (i.e when none are currently running).
504 struct sched_entity *curr;
505 struct sched_entity *next;
506 struct sched_entity *last;
507 struct sched_entity *skip;
509 #ifdef CONFIG_SCHED_DEBUG
510 unsigned int nr_spread_over;
517 struct sched_avg avg;
519 u64 load_last_update_time_copy;
522 raw_spinlock_t lock ____cacheline_aligned;
524 unsigned long load_avg;
525 unsigned long util_avg;
526 unsigned long runnable_sum;
529 #ifdef CONFIG_FAIR_GROUP_SCHED
530 unsigned long tg_load_avg_contrib;
532 long prop_runnable_sum;
535 * h_load = weight * f(tg)
537 * Where f(tg) is the recursive weight fraction assigned to
540 unsigned long h_load;
541 u64 last_h_load_update;
542 struct sched_entity *h_load_next;
543 #endif /* CONFIG_FAIR_GROUP_SCHED */
544 #endif /* CONFIG_SMP */
546 #ifdef CONFIG_FAIR_GROUP_SCHED
547 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
550 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
551 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
552 * (like users, containers etc.)
554 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
555 * This list is used during load balance.
558 struct list_head leaf_cfs_rq_list;
559 struct task_group *tg; /* group that "owns" this runqueue */
561 #ifdef CONFIG_CFS_BANDWIDTH
565 s64 runtime_remaining;
568 u64 throttled_clock_task;
569 u64 throttled_clock_task_time;
572 struct list_head throttled_list;
573 #endif /* CONFIG_CFS_BANDWIDTH */
574 #endif /* CONFIG_FAIR_GROUP_SCHED */
577 static inline int rt_bandwidth_enabled(void)
579 return sysctl_sched_rt_runtime >= 0;
582 /* RT IPI pull logic requires IRQ_WORK */
583 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
584 # define HAVE_RT_PUSH_IPI
587 /* Real-Time classes' related field in a runqueue: */
589 struct rt_prio_array active;
590 unsigned int rt_nr_running;
591 unsigned int rr_nr_running;
592 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
594 int curr; /* highest queued rt task prio */
596 int next; /* next highest */
601 unsigned long rt_nr_migratory;
602 unsigned long rt_nr_total;
604 struct plist_head pushable_tasks;
606 #endif /* CONFIG_SMP */
612 /* Nests inside the rq lock: */
613 raw_spinlock_t rt_runtime_lock;
615 #ifdef CONFIG_RT_GROUP_SCHED
616 unsigned long rt_nr_boosted;
619 struct task_group *tg;
623 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
625 return rt_rq->rt_queued && rt_rq->rt_nr_running;
628 /* Deadline class' related fields in a runqueue */
630 /* runqueue is an rbtree, ordered by deadline */
631 struct rb_root_cached root;
633 unsigned long dl_nr_running;
637 * Deadline values of the currently executing and the
638 * earliest ready task on this rq. Caching these facilitates
639 * the decision wether or not a ready but not running task
640 * should migrate somewhere else.
647 unsigned long dl_nr_migratory;
651 * Tasks on this rq that can be pushed away. They are kept in
652 * an rb-tree, ordered by tasks' deadlines, with caching
653 * of the leftmost (earliest deadline) element.
655 struct rb_root_cached pushable_dl_tasks_root;
660 * "Active utilization" for this runqueue: increased when a
661 * task wakes up (becomes TASK_RUNNING) and decreased when a
667 * Utilization of the tasks "assigned" to this runqueue (including
668 * the tasks that are in runqueue and the tasks that executed on this
669 * CPU and blocked). Increased when a task moves to this runqueue, and
670 * decreased when the task moves away (migrates, changes scheduling
671 * policy, or terminates).
672 * This is needed to compute the "inactive utilization" for the
673 * runqueue (inactive utilization = this_bw - running_bw).
679 * Inverse of the fraction of CPU utilization that can be reclaimed
680 * by the GRUB algorithm.
685 #ifdef CONFIG_FAIR_GROUP_SCHED
686 /* An entity is a task if it doesn't "own" a runqueue */
687 #define entity_is_task(se) (!se->my_q)
689 #define entity_is_task(se) 1
694 * XXX we want to get rid of these helpers and use the full load resolution.
696 static inline long se_weight(struct sched_entity *se)
698 return scale_load_down(se->load.weight);
701 static inline long se_runnable(struct sched_entity *se)
703 return scale_load_down(se->runnable_weight);
706 static inline bool sched_asym_prefer(int a, int b)
708 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
712 * We add the notion of a root-domain which will be used to define per-domain
713 * variables. Each exclusive cpuset essentially defines an island domain by
714 * fully partitioning the member CPUs from any other cpuset. Whenever a new
715 * exclusive cpuset is created, we also create and attach a new root-domain
724 cpumask_var_t online;
727 * Indicate pullable load on at least one CPU, e.g:
728 * - More than one runnable task
729 * - Running task is misfit
734 * The bit corresponding to a CPU gets set here if such CPU has more
735 * than one runnable -deadline task (as it is below for RT tasks).
737 cpumask_var_t dlo_mask;
742 #ifdef HAVE_RT_PUSH_IPI
744 * For IPI pull requests, loop across the rto_mask.
746 struct irq_work rto_push_work;
747 raw_spinlock_t rto_lock;
748 /* These are only updated and read within rto_lock */
751 /* These atomics are updated outside of a lock */
752 atomic_t rto_loop_next;
753 atomic_t rto_loop_start;
756 * The "RT overload" flag: it gets set if a CPU has more than
757 * one runnable RT task.
759 cpumask_var_t rto_mask;
760 struct cpupri cpupri;
762 unsigned long max_cpu_capacity;
765 extern struct root_domain def_root_domain;
766 extern struct mutex sched_domains_mutex;
768 extern void init_defrootdomain(void);
769 extern int sched_init_domains(const struct cpumask *cpu_map);
770 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
771 extern void sched_get_rd(struct root_domain *rd);
772 extern void sched_put_rd(struct root_domain *rd);
774 #ifdef HAVE_RT_PUSH_IPI
775 extern void rto_push_irq_work_func(struct irq_work *work);
777 #endif /* CONFIG_SMP */
780 * This is the main, per-CPU runqueue data structure.
782 * Locking rule: those places that want to lock multiple runqueues
783 * (such as the load balancing or the thread migration code), lock
784 * acquire operations must be ordered by ascending &runqueue.
791 * nr_running and cpu_load should be in the same cacheline because
792 * remote CPUs use both these fields when doing load calculation.
794 unsigned int nr_running;
795 #ifdef CONFIG_NUMA_BALANCING
796 unsigned int nr_numa_running;
797 unsigned int nr_preferred_running;
798 unsigned int numa_migrate_on;
800 #define CPU_LOAD_IDX_MAX 5
801 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
802 #ifdef CONFIG_NO_HZ_COMMON
804 unsigned long last_load_update_tick;
805 unsigned long last_blocked_load_update_tick;
806 unsigned int has_blocked_load;
807 #endif /* CONFIG_SMP */
808 unsigned int nohz_tick_stopped;
810 #endif /* CONFIG_NO_HZ_COMMON */
812 /* capture load from *all* tasks on this CPU: */
813 struct load_weight load;
814 unsigned long nr_load_updates;
821 #ifdef CONFIG_FAIR_GROUP_SCHED
822 /* list of leaf cfs_rq on this CPU: */
823 struct list_head leaf_cfs_rq_list;
824 struct list_head *tmp_alone_branch;
825 #endif /* CONFIG_FAIR_GROUP_SCHED */
828 * This is part of a global counter where only the total sum
829 * over all CPUs matters. A task can increase this counter on
830 * one CPU and if it got migrated afterwards it may decrease
831 * it on another CPU. Always updated under the runqueue lock:
833 unsigned long nr_uninterruptible;
835 struct task_struct *curr;
836 struct task_struct *idle;
837 struct task_struct *stop;
838 unsigned long next_balance;
839 struct mm_struct *prev_mm;
841 unsigned int clock_update_flags;
848 struct root_domain *rd;
849 struct sched_domain *sd;
851 unsigned long cpu_capacity;
852 unsigned long cpu_capacity_orig;
854 struct callback_head *balance_callback;
856 unsigned char idle_balance;
858 unsigned long misfit_task_load;
860 /* For active balancing */
863 struct cpu_stop_work active_balance_work;
865 /* CPU of this runqueue: */
869 struct list_head cfs_tasks;
871 struct sched_avg avg_rt;
872 struct sched_avg avg_dl;
873 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
874 struct sched_avg avg_irq;
879 /* This is used to determine avg_idle's max value */
880 u64 max_idle_balance_cost;
883 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
886 #ifdef CONFIG_PARAVIRT
889 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
890 u64 prev_steal_time_rq;
893 /* calc_load related fields */
894 unsigned long calc_load_update;
895 long calc_load_active;
897 #ifdef CONFIG_SCHED_HRTICK
899 int hrtick_csd_pending;
900 call_single_data_t hrtick_csd;
902 struct hrtimer hrtick_timer;
905 #ifdef CONFIG_SCHEDSTATS
907 struct sched_info rq_sched_info;
908 unsigned long long rq_cpu_time;
909 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
911 /* sys_sched_yield() stats */
912 unsigned int yld_count;
914 /* schedule() stats */
915 unsigned int sched_count;
916 unsigned int sched_goidle;
918 /* try_to_wake_up() stats */
919 unsigned int ttwu_count;
920 unsigned int ttwu_local;
924 struct llist_head wake_list;
927 #ifdef CONFIG_CPU_IDLE
928 /* Must be inspected within a rcu lock section */
929 struct cpuidle_state *idle_state;
933 static inline int cpu_of(struct rq *rq)
943 #ifdef CONFIG_SCHED_SMT
945 extern struct static_key_false sched_smt_present;
947 extern void __update_idle_core(struct rq *rq);
949 static inline void update_idle_core(struct rq *rq)
951 if (static_branch_unlikely(&sched_smt_present))
952 __update_idle_core(rq);
956 static inline void update_idle_core(struct rq *rq) { }
959 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
961 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
962 #define this_rq() this_cpu_ptr(&runqueues)
963 #define task_rq(p) cpu_rq(task_cpu(p))
964 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
965 #define raw_rq() raw_cpu_ptr(&runqueues)
967 extern void update_rq_clock(struct rq *rq);
969 static inline u64 __rq_clock_broken(struct rq *rq)
971 return READ_ONCE(rq->clock);
975 * rq::clock_update_flags bits
977 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
978 * call to __schedule(). This is an optimisation to avoid
979 * neighbouring rq clock updates.
981 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
982 * in effect and calls to update_rq_clock() are being ignored.
984 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
985 * made to update_rq_clock() since the last time rq::lock was pinned.
987 * If inside of __schedule(), clock_update_flags will have been
988 * shifted left (a left shift is a cheap operation for the fast path
989 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
991 * if (rq-clock_update_flags >= RQCF_UPDATED)
993 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
994 * one position though, because the next rq_unpin_lock() will shift it
997 #define RQCF_REQ_SKIP 0x01
998 #define RQCF_ACT_SKIP 0x02
999 #define RQCF_UPDATED 0x04
1001 static inline void assert_clock_updated(struct rq *rq)
1004 * The only reason for not seeing a clock update since the
1005 * last rq_pin_lock() is if we're currently skipping updates.
1007 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1010 static inline u64 rq_clock(struct rq *rq)
1012 lockdep_assert_held(&rq->lock);
1013 assert_clock_updated(rq);
1018 static inline u64 rq_clock_task(struct rq *rq)
1020 lockdep_assert_held(&rq->lock);
1021 assert_clock_updated(rq);
1023 return rq->clock_task;
1026 static inline void rq_clock_skip_update(struct rq *rq)
1028 lockdep_assert_held(&rq->lock);
1029 rq->clock_update_flags |= RQCF_REQ_SKIP;
1033 * See rt task throttling, which is the only time a skip
1034 * request is cancelled.
1036 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1038 lockdep_assert_held(&rq->lock);
1039 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1043 unsigned long flags;
1044 struct pin_cookie cookie;
1045 #ifdef CONFIG_SCHED_DEBUG
1047 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1048 * current pin context is stashed here in case it needs to be
1049 * restored in rq_repin_lock().
1051 unsigned int clock_update_flags;
1055 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1057 rf->cookie = lockdep_pin_lock(&rq->lock);
1059 #ifdef CONFIG_SCHED_DEBUG
1060 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1061 rf->clock_update_flags = 0;
1065 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1067 #ifdef CONFIG_SCHED_DEBUG
1068 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1069 rf->clock_update_flags = RQCF_UPDATED;
1072 lockdep_unpin_lock(&rq->lock, rf->cookie);
1075 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1077 lockdep_repin_lock(&rq->lock, rf->cookie);
1079 #ifdef CONFIG_SCHED_DEBUG
1081 * Restore the value we stashed in @rf for this pin context.
1083 rq->clock_update_flags |= rf->clock_update_flags;
1087 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1088 __acquires(rq->lock);
1090 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1091 __acquires(p->pi_lock)
1092 __acquires(rq->lock);
1094 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1095 __releases(rq->lock)
1097 rq_unpin_lock(rq, rf);
1098 raw_spin_unlock(&rq->lock);
1102 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1103 __releases(rq->lock)
1104 __releases(p->pi_lock)
1106 rq_unpin_lock(rq, rf);
1107 raw_spin_unlock(&rq->lock);
1108 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1112 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1113 __acquires(rq->lock)
1115 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1116 rq_pin_lock(rq, rf);
1120 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1121 __acquires(rq->lock)
1123 raw_spin_lock_irq(&rq->lock);
1124 rq_pin_lock(rq, rf);
1128 rq_lock(struct rq *rq, struct rq_flags *rf)
1129 __acquires(rq->lock)
1131 raw_spin_lock(&rq->lock);
1132 rq_pin_lock(rq, rf);
1136 rq_relock(struct rq *rq, struct rq_flags *rf)
1137 __acquires(rq->lock)
1139 raw_spin_lock(&rq->lock);
1140 rq_repin_lock(rq, rf);
1144 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1145 __releases(rq->lock)
1147 rq_unpin_lock(rq, rf);
1148 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1152 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1153 __releases(rq->lock)
1155 rq_unpin_lock(rq, rf);
1156 raw_spin_unlock_irq(&rq->lock);
1160 rq_unlock(struct rq *rq, struct rq_flags *rf)
1161 __releases(rq->lock)
1163 rq_unpin_lock(rq, rf);
1164 raw_spin_unlock(&rq->lock);
1167 static inline struct rq *
1168 this_rq_lock_irq(struct rq_flags *rf)
1169 __acquires(rq->lock)
1173 local_irq_disable();
1180 enum numa_topology_type {
1185 extern enum numa_topology_type sched_numa_topology_type;
1186 extern int sched_max_numa_distance;
1187 extern bool find_numa_distance(int distance);
1191 extern void sched_init_numa(void);
1192 extern void sched_domains_numa_masks_set(unsigned int cpu);
1193 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1195 static inline void sched_init_numa(void) { }
1196 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1197 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1200 #ifdef CONFIG_NUMA_BALANCING
1201 /* The regions in numa_faults array from task_struct */
1202 enum numa_faults_stats {
1208 extern void sched_setnuma(struct task_struct *p, int node);
1209 extern int migrate_task_to(struct task_struct *p, int cpu);
1210 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1212 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1215 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1218 #endif /* CONFIG_NUMA_BALANCING */
1223 queue_balance_callback(struct rq *rq,
1224 struct callback_head *head,
1225 void (*func)(struct rq *rq))
1227 lockdep_assert_held(&rq->lock);
1229 if (unlikely(head->next))
1232 head->func = (void (*)(struct callback_head *))func;
1233 head->next = rq->balance_callback;
1234 rq->balance_callback = head;
1237 extern void sched_ttwu_pending(void);
1239 #define rcu_dereference_check_sched_domain(p) \
1240 rcu_dereference_check((p), \
1241 lockdep_is_held(&sched_domains_mutex))
1244 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1245 * See detach_destroy_domains: synchronize_sched for details.
1247 * The domain tree of any CPU may only be accessed from within
1248 * preempt-disabled sections.
1250 #define for_each_domain(cpu, __sd) \
1251 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1252 __sd; __sd = __sd->parent)
1254 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1257 * highest_flag_domain - Return highest sched_domain containing flag.
1258 * @cpu: The CPU whose highest level of sched domain is to
1260 * @flag: The flag to check for the highest sched_domain
1261 * for the given CPU.
1263 * Returns the highest sched_domain of a CPU which contains the given flag.
1265 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1267 struct sched_domain *sd, *hsd = NULL;
1269 for_each_domain(cpu, sd) {
1270 if (!(sd->flags & flag))
1278 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1280 struct sched_domain *sd;
1282 for_each_domain(cpu, sd) {
1283 if (sd->flags & flag)
1290 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1291 DECLARE_PER_CPU(int, sd_llc_size);
1292 DECLARE_PER_CPU(int, sd_llc_id);
1293 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1294 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1295 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1296 extern struct static_key_false sched_asym_cpucapacity;
1298 struct sched_group_capacity {
1301 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1304 unsigned long capacity;
1305 unsigned long min_capacity; /* Min per-CPU capacity in group */
1306 unsigned long max_capacity; /* Max per-CPU capacity in group */
1307 unsigned long next_update;
1308 int imbalance; /* XXX unrelated to capacity but shared group state */
1310 #ifdef CONFIG_SCHED_DEBUG
1314 unsigned long cpumask[0]; /* Balance mask */
1317 struct sched_group {
1318 struct sched_group *next; /* Must be a circular list */
1321 unsigned int group_weight;
1322 struct sched_group_capacity *sgc;
1323 int asym_prefer_cpu; /* CPU of highest priority in group */
1326 * The CPUs this group covers.
1328 * NOTE: this field is variable length. (Allocated dynamically
1329 * by attaching extra space to the end of the structure,
1330 * depending on how many CPUs the kernel has booted up with)
1332 unsigned long cpumask[0];
1335 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1337 return to_cpumask(sg->cpumask);
1341 * See build_balance_mask().
1343 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1345 return to_cpumask(sg->sgc->cpumask);
1349 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1350 * @group: The group whose first CPU is to be returned.
1352 static inline unsigned int group_first_cpu(struct sched_group *group)
1354 return cpumask_first(sched_group_span(group));
1357 extern int group_balance_cpu(struct sched_group *sg);
1359 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1360 void register_sched_domain_sysctl(void);
1361 void dirty_sched_domain_sysctl(int cpu);
1362 void unregister_sched_domain_sysctl(void);
1364 static inline void register_sched_domain_sysctl(void)
1367 static inline void dirty_sched_domain_sysctl(int cpu)
1370 static inline void unregister_sched_domain_sysctl(void)
1377 static inline void sched_ttwu_pending(void) { }
1379 #endif /* CONFIG_SMP */
1382 #include "autogroup.h"
1384 #ifdef CONFIG_CGROUP_SCHED
1387 * Return the group to which this tasks belongs.
1389 * We cannot use task_css() and friends because the cgroup subsystem
1390 * changes that value before the cgroup_subsys::attach() method is called,
1391 * therefore we cannot pin it and might observe the wrong value.
1393 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1394 * core changes this before calling sched_move_task().
1396 * Instead we use a 'copy' which is updated from sched_move_task() while
1397 * holding both task_struct::pi_lock and rq::lock.
1399 static inline struct task_group *task_group(struct task_struct *p)
1401 return p->sched_task_group;
1404 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1405 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1407 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1408 struct task_group *tg = task_group(p);
1411 #ifdef CONFIG_FAIR_GROUP_SCHED
1412 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1413 p->se.cfs_rq = tg->cfs_rq[cpu];
1414 p->se.parent = tg->se[cpu];
1417 #ifdef CONFIG_RT_GROUP_SCHED
1418 p->rt.rt_rq = tg->rt_rq[cpu];
1419 p->rt.parent = tg->rt_se[cpu];
1423 #else /* CONFIG_CGROUP_SCHED */
1425 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1426 static inline struct task_group *task_group(struct task_struct *p)
1431 #endif /* CONFIG_CGROUP_SCHED */
1433 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1435 set_task_rq(p, cpu);
1438 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1439 * successfuly executed on another CPU. We must ensure that updates of
1440 * per-task data have been completed by this moment.
1443 #ifdef CONFIG_THREAD_INFO_IN_TASK
1446 task_thread_info(p)->cpu = cpu;
1453 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1455 #ifdef CONFIG_SCHED_DEBUG
1456 # include <linux/static_key.h>
1457 # define const_debug __read_mostly
1459 # define const_debug const
1462 #define SCHED_FEAT(name, enabled) \
1463 __SCHED_FEAT_##name ,
1466 #include "features.h"
1472 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1475 * To support run-time toggling of sched features, all the translation units
1476 * (but core.c) reference the sysctl_sched_features defined in core.c.
1478 extern const_debug unsigned int sysctl_sched_features;
1480 #define SCHED_FEAT(name, enabled) \
1481 static __always_inline bool static_branch_##name(struct static_key *key) \
1483 return static_key_##enabled(key); \
1486 #include "features.h"
1489 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1490 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1492 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1495 * Each translation unit has its own copy of sysctl_sched_features to allow
1496 * constants propagation at compile time and compiler optimization based on
1499 #define SCHED_FEAT(name, enabled) \
1500 (1UL << __SCHED_FEAT_##name) * enabled |
1501 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1502 #include "features.h"
1506 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1508 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1510 extern struct static_key_false sched_numa_balancing;
1511 extern struct static_key_false sched_schedstats;
1513 static inline u64 global_rt_period(void)
1515 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1518 static inline u64 global_rt_runtime(void)
1520 if (sysctl_sched_rt_runtime < 0)
1523 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1526 static inline int task_current(struct rq *rq, struct task_struct *p)
1528 return rq->curr == p;
1531 static inline int task_running(struct rq *rq, struct task_struct *p)
1536 return task_current(rq, p);
1540 static inline int task_on_rq_queued(struct task_struct *p)
1542 return p->on_rq == TASK_ON_RQ_QUEUED;
1545 static inline int task_on_rq_migrating(struct task_struct *p)
1547 return p->on_rq == TASK_ON_RQ_MIGRATING;
1553 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1554 #define WF_FORK 0x02 /* Child wakeup after fork */
1555 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1558 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1559 * of tasks with abnormal "nice" values across CPUs the contribution that
1560 * each task makes to its run queue's load is weighted according to its
1561 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1562 * scaled version of the new time slice allocation that they receive on time
1566 #define WEIGHT_IDLEPRIO 3
1567 #define WMULT_IDLEPRIO 1431655765
1569 extern const int sched_prio_to_weight[40];
1570 extern const u32 sched_prio_to_wmult[40];
1573 * {de,en}queue flags:
1575 * DEQUEUE_SLEEP - task is no longer runnable
1576 * ENQUEUE_WAKEUP - task just became runnable
1578 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1579 * are in a known state which allows modification. Such pairs
1580 * should preserve as much state as possible.
1582 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1585 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1586 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1587 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1591 #define DEQUEUE_SLEEP 0x01
1592 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1593 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1594 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1596 #define ENQUEUE_WAKEUP 0x01
1597 #define ENQUEUE_RESTORE 0x02
1598 #define ENQUEUE_MOVE 0x04
1599 #define ENQUEUE_NOCLOCK 0x08
1601 #define ENQUEUE_HEAD 0x10
1602 #define ENQUEUE_REPLENISH 0x20
1604 #define ENQUEUE_MIGRATED 0x40
1606 #define ENQUEUE_MIGRATED 0x00
1609 #define RETRY_TASK ((void *)-1UL)
1611 struct sched_class {
1612 const struct sched_class *next;
1614 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1615 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1616 void (*yield_task) (struct rq *rq);
1617 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1619 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1622 * It is the responsibility of the pick_next_task() method that will
1623 * return the next task to call put_prev_task() on the @prev task or
1624 * something equivalent.
1626 * May return RETRY_TASK when it finds a higher prio class has runnable
1629 struct task_struct * (*pick_next_task)(struct rq *rq,
1630 struct task_struct *prev,
1631 struct rq_flags *rf);
1632 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1635 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1636 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1638 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1640 void (*set_cpus_allowed)(struct task_struct *p,
1641 const struct cpumask *newmask);
1643 void (*rq_online)(struct rq *rq);
1644 void (*rq_offline)(struct rq *rq);
1647 void (*set_curr_task)(struct rq *rq);
1648 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1649 void (*task_fork)(struct task_struct *p);
1650 void (*task_dead)(struct task_struct *p);
1653 * The switched_from() call is allowed to drop rq->lock, therefore we
1654 * cannot assume the switched_from/switched_to pair is serliazed by
1655 * rq->lock. They are however serialized by p->pi_lock.
1657 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1658 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1659 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1662 unsigned int (*get_rr_interval)(struct rq *rq,
1663 struct task_struct *task);
1665 void (*update_curr)(struct rq *rq);
1667 #define TASK_SET_GROUP 0
1668 #define TASK_MOVE_GROUP 1
1670 #ifdef CONFIG_FAIR_GROUP_SCHED
1671 void (*task_change_group)(struct task_struct *p, int type);
1675 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1677 prev->sched_class->put_prev_task(rq, prev);
1680 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1682 curr->sched_class->set_curr_task(rq);
1686 #define sched_class_highest (&stop_sched_class)
1688 #define sched_class_highest (&dl_sched_class)
1690 #define for_each_class(class) \
1691 for (class = sched_class_highest; class; class = class->next)
1693 extern const struct sched_class stop_sched_class;
1694 extern const struct sched_class dl_sched_class;
1695 extern const struct sched_class rt_sched_class;
1696 extern const struct sched_class fair_sched_class;
1697 extern const struct sched_class idle_sched_class;
1702 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1704 extern void trigger_load_balance(struct rq *rq);
1706 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1710 #ifdef CONFIG_CPU_IDLE
1711 static inline void idle_set_state(struct rq *rq,
1712 struct cpuidle_state *idle_state)
1714 rq->idle_state = idle_state;
1717 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1719 SCHED_WARN_ON(!rcu_read_lock_held());
1721 return rq->idle_state;
1724 static inline void idle_set_state(struct rq *rq,
1725 struct cpuidle_state *idle_state)
1729 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1735 extern void schedule_idle(void);
1737 extern void sysrq_sched_debug_show(void);
1738 extern void sched_init_granularity(void);
1739 extern void update_max_interval(void);
1741 extern void init_sched_dl_class(void);
1742 extern void init_sched_rt_class(void);
1743 extern void init_sched_fair_class(void);
1745 extern void reweight_task(struct task_struct *p, int prio);
1747 extern void resched_curr(struct rq *rq);
1748 extern void resched_cpu(int cpu);
1750 extern struct rt_bandwidth def_rt_bandwidth;
1751 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1753 extern struct dl_bandwidth def_dl_bandwidth;
1754 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1755 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1756 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1757 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1760 #define BW_UNIT (1 << BW_SHIFT)
1761 #define RATIO_SHIFT 8
1762 unsigned long to_ratio(u64 period, u64 runtime);
1764 extern void init_entity_runnable_average(struct sched_entity *se);
1765 extern void post_init_entity_util_avg(struct sched_entity *se);
1767 #ifdef CONFIG_NO_HZ_FULL
1768 extern bool sched_can_stop_tick(struct rq *rq);
1769 extern int __init sched_tick_offload_init(void);
1772 * Tick may be needed by tasks in the runqueue depending on their policy and
1773 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1774 * nohz mode if necessary.
1776 static inline void sched_update_tick_dependency(struct rq *rq)
1780 if (!tick_nohz_full_enabled())
1785 if (!tick_nohz_full_cpu(cpu))
1788 if (sched_can_stop_tick(rq))
1789 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1791 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1794 static inline int sched_tick_offload_init(void) { return 0; }
1795 static inline void sched_update_tick_dependency(struct rq *rq) { }
1798 static inline void add_nr_running(struct rq *rq, unsigned count)
1800 unsigned prev_nr = rq->nr_running;
1802 rq->nr_running = prev_nr + count;
1805 if (prev_nr < 2 && rq->nr_running >= 2) {
1806 if (!READ_ONCE(rq->rd->overload))
1807 WRITE_ONCE(rq->rd->overload, 1);
1811 sched_update_tick_dependency(rq);
1814 static inline void sub_nr_running(struct rq *rq, unsigned count)
1816 rq->nr_running -= count;
1817 /* Check if we still need preemption */
1818 sched_update_tick_dependency(rq);
1821 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1822 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1824 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1826 extern const_debug unsigned int sysctl_sched_nr_migrate;
1827 extern const_debug unsigned int sysctl_sched_migration_cost;
1829 #ifdef CONFIG_SCHED_HRTICK
1833 * - enabled by features
1834 * - hrtimer is actually high res
1836 static inline int hrtick_enabled(struct rq *rq)
1838 if (!sched_feat(HRTICK))
1840 if (!cpu_active(cpu_of(rq)))
1842 return hrtimer_is_hres_active(&rq->hrtick_timer);
1845 void hrtick_start(struct rq *rq, u64 delay);
1849 static inline int hrtick_enabled(struct rq *rq)
1854 #endif /* CONFIG_SCHED_HRTICK */
1856 #ifndef arch_scale_freq_capacity
1857 static __always_inline
1858 unsigned long arch_scale_freq_capacity(int cpu)
1860 return SCHED_CAPACITY_SCALE;
1865 #ifndef arch_scale_cpu_capacity
1866 static __always_inline
1867 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1869 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1870 return sd->smt_gain / sd->span_weight;
1872 return SCHED_CAPACITY_SCALE;
1876 #ifndef arch_scale_cpu_capacity
1877 static __always_inline
1878 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1880 return SCHED_CAPACITY_SCALE;
1886 #ifdef CONFIG_PREEMPT
1888 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1891 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1892 * way at the expense of forcing extra atomic operations in all
1893 * invocations. This assures that the double_lock is acquired using the
1894 * same underlying policy as the spinlock_t on this architecture, which
1895 * reduces latency compared to the unfair variant below. However, it
1896 * also adds more overhead and therefore may reduce throughput.
1898 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1899 __releases(this_rq->lock)
1900 __acquires(busiest->lock)
1901 __acquires(this_rq->lock)
1903 raw_spin_unlock(&this_rq->lock);
1904 double_rq_lock(this_rq, busiest);
1911 * Unfair double_lock_balance: Optimizes throughput at the expense of
1912 * latency by eliminating extra atomic operations when the locks are
1913 * already in proper order on entry. This favors lower CPU-ids and will
1914 * grant the double lock to lower CPUs over higher ids under contention,
1915 * regardless of entry order into the function.
1917 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1918 __releases(this_rq->lock)
1919 __acquires(busiest->lock)
1920 __acquires(this_rq->lock)
1924 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1925 if (busiest < this_rq) {
1926 raw_spin_unlock(&this_rq->lock);
1927 raw_spin_lock(&busiest->lock);
1928 raw_spin_lock_nested(&this_rq->lock,
1929 SINGLE_DEPTH_NESTING);
1932 raw_spin_lock_nested(&busiest->lock,
1933 SINGLE_DEPTH_NESTING);
1938 #endif /* CONFIG_PREEMPT */
1941 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1943 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1945 if (unlikely(!irqs_disabled())) {
1946 /* printk() doesn't work well under rq->lock */
1947 raw_spin_unlock(&this_rq->lock);
1951 return _double_lock_balance(this_rq, busiest);
1954 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1955 __releases(busiest->lock)
1957 raw_spin_unlock(&busiest->lock);
1958 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1961 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1967 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1970 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1976 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1979 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1985 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1989 * double_rq_lock - safely lock two runqueues
1991 * Note this does not disable interrupts like task_rq_lock,
1992 * you need to do so manually before calling.
1994 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1995 __acquires(rq1->lock)
1996 __acquires(rq2->lock)
1998 BUG_ON(!irqs_disabled());
2000 raw_spin_lock(&rq1->lock);
2001 __acquire(rq2->lock); /* Fake it out ;) */
2004 raw_spin_lock(&rq1->lock);
2005 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2007 raw_spin_lock(&rq2->lock);
2008 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2014 * double_rq_unlock - safely unlock two runqueues
2016 * Note this does not restore interrupts like task_rq_unlock,
2017 * you need to do so manually after calling.
2019 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2020 __releases(rq1->lock)
2021 __releases(rq2->lock)
2023 raw_spin_unlock(&rq1->lock);
2025 raw_spin_unlock(&rq2->lock);
2027 __release(rq2->lock);
2030 extern void set_rq_online (struct rq *rq);
2031 extern void set_rq_offline(struct rq *rq);
2032 extern bool sched_smp_initialized;
2034 #else /* CONFIG_SMP */
2037 * double_rq_lock - safely lock two runqueues
2039 * Note this does not disable interrupts like task_rq_lock,
2040 * you need to do so manually before calling.
2042 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2043 __acquires(rq1->lock)
2044 __acquires(rq2->lock)
2046 BUG_ON(!irqs_disabled());
2048 raw_spin_lock(&rq1->lock);
2049 __acquire(rq2->lock); /* Fake it out ;) */
2053 * double_rq_unlock - safely unlock two runqueues
2055 * Note this does not restore interrupts like task_rq_unlock,
2056 * you need to do so manually after calling.
2058 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2059 __releases(rq1->lock)
2060 __releases(rq2->lock)
2063 raw_spin_unlock(&rq1->lock);
2064 __release(rq2->lock);
2069 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2070 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2072 #ifdef CONFIG_SCHED_DEBUG
2073 extern bool sched_debug_enabled;
2075 extern void print_cfs_stats(struct seq_file *m, int cpu);
2076 extern void print_rt_stats(struct seq_file *m, int cpu);
2077 extern void print_dl_stats(struct seq_file *m, int cpu);
2078 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2079 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2080 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2081 #ifdef CONFIG_NUMA_BALANCING
2083 show_numa_stats(struct task_struct *p, struct seq_file *m);
2085 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2086 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2087 #endif /* CONFIG_NUMA_BALANCING */
2088 #endif /* CONFIG_SCHED_DEBUG */
2090 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2091 extern void init_rt_rq(struct rt_rq *rt_rq);
2092 extern void init_dl_rq(struct dl_rq *dl_rq);
2094 extern void cfs_bandwidth_usage_inc(void);
2095 extern void cfs_bandwidth_usage_dec(void);
2097 #ifdef CONFIG_NO_HZ_COMMON
2098 #define NOHZ_BALANCE_KICK_BIT 0
2099 #define NOHZ_STATS_KICK_BIT 1
2101 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2102 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2104 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2106 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2108 extern void nohz_balance_exit_idle(struct rq *rq);
2110 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2116 void __dl_update(struct dl_bw *dl_b, s64 bw)
2118 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2121 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2122 "sched RCU must be held");
2123 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2124 struct rq *rq = cpu_rq(i);
2126 rq->dl.extra_bw += bw;
2131 void __dl_update(struct dl_bw *dl_b, s64 bw)
2133 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2140 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2145 struct u64_stats_sync sync;
2148 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2151 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2152 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2153 * and never move forward.
2155 static inline u64 irq_time_read(int cpu)
2157 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2162 seq = __u64_stats_fetch_begin(&irqtime->sync);
2163 total = irqtime->total;
2164 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2168 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2170 #ifdef CONFIG_CPU_FREQ
2171 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2174 * cpufreq_update_util - Take a note about CPU utilization changes.
2175 * @rq: Runqueue to carry out the update for.
2176 * @flags: Update reason flags.
2178 * This function is called by the scheduler on the CPU whose utilization is
2181 * It can only be called from RCU-sched read-side critical sections.
2183 * The way cpufreq is currently arranged requires it to evaluate the CPU
2184 * performance state (frequency/voltage) on a regular basis to prevent it from
2185 * being stuck in a completely inadequate performance level for too long.
2186 * That is not guaranteed to happen if the updates are only triggered from CFS
2187 * and DL, though, because they may not be coming in if only RT tasks are
2188 * active all the time (or there are RT tasks only).
2190 * As a workaround for that issue, this function is called periodically by the
2191 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2192 * but that really is a band-aid. Going forward it should be replaced with
2193 * solutions targeted more specifically at RT tasks.
2195 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2197 struct update_util_data *data;
2199 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2202 data->func(data, rq_clock(rq), flags);
2205 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2206 #endif /* CONFIG_CPU_FREQ */
2208 #ifdef arch_scale_freq_capacity
2209 # ifndef arch_scale_freq_invariant
2210 # define arch_scale_freq_invariant() true
2213 # define arch_scale_freq_invariant() false
2216 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2217 static inline unsigned long cpu_bw_dl(struct rq *rq)
2219 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2222 static inline unsigned long cpu_util_dl(struct rq *rq)
2224 return READ_ONCE(rq->avg_dl.util_avg);
2227 static inline unsigned long cpu_util_cfs(struct rq *rq)
2229 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2231 if (sched_feat(UTIL_EST)) {
2232 util = max_t(unsigned long, util,
2233 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2239 static inline unsigned long cpu_util_rt(struct rq *rq)
2241 return READ_ONCE(rq->avg_rt.util_avg);
2245 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2246 static inline unsigned long cpu_util_irq(struct rq *rq)
2248 return rq->avg_irq.util_avg;
2252 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2254 util *= (max - irq);
2261 static inline unsigned long cpu_util_irq(struct rq *rq)
2267 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)