2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/u64_stats_sync.h>
6 #include <linux/sched/deadline.h>
7 #include <linux/kernel_stat.h>
8 #include <linux/binfmts.h>
9 #include <linux/mutex.h>
10 #include <linux/spinlock.h>
11 #include <linux/stop_machine.h>
12 #include <linux/irq_work.h>
13 #include <linux/tick.h>
14 #include <linux/slab.h>
17 #include "cpudeadline.h"
20 #ifdef CONFIG_SCHED_DEBUG
21 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
23 #define SCHED_WARN_ON(x) ((void)(x))
29 /* task_struct::on_rq states: */
30 #define TASK_ON_RQ_QUEUED 1
31 #define TASK_ON_RQ_MIGRATING 2
33 extern __read_mostly int scheduler_running;
35 extern unsigned long calc_load_update;
36 extern atomic_long_t calc_load_tasks;
38 extern void calc_global_load_tick(struct rq *this_rq);
39 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
42 extern void cpu_load_update_active(struct rq *this_rq);
44 static inline void cpu_load_update_active(struct rq *this_rq) { }
47 #ifdef CONFIG_SCHED_SMT
48 extern void update_idle_core(struct rq *rq);
50 static inline void update_idle_core(struct rq *rq) { }
54 * Helpers for converting nanosecond timing to jiffy resolution
56 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
59 * Increase resolution of nice-level calculations for 64-bit architectures.
60 * The extra resolution improves shares distribution and load balancing of
61 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
62 * hierarchies, especially on larger systems. This is not a user-visible change
63 * and does not change the user-interface for setting shares/weights.
65 * We increase resolution only if we have enough bits to allow this increased
66 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
67 * pretty high and the returns do not justify the increased costs.
69 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
70 * increase coverage and consistency always enable it on 64bit platforms.
73 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
74 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
75 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
77 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
78 # define scale_load(w) (w)
79 # define scale_load_down(w) (w)
83 * Task weight (visible to users) and its load (invisible to users) have
84 * independent resolution, but they should be well calibrated. We use
85 * scale_load() and scale_load_down(w) to convert between them. The
86 * following must be true:
88 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
91 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
94 * Single value that decides SCHED_DEADLINE internal math precision.
95 * 10 -> just above 1us
96 * 9 -> just above 0.5us
101 * These are the 'tuning knobs' of the scheduler:
105 * single value that denotes runtime == period, ie unlimited time.
107 #define RUNTIME_INF ((u64)~0ULL)
109 static inline int idle_policy(int policy)
111 return policy == SCHED_IDLE;
113 static inline int fair_policy(int policy)
115 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
118 static inline int rt_policy(int policy)
120 return policy == SCHED_FIFO || policy == SCHED_RR;
123 static inline int dl_policy(int policy)
125 return policy == SCHED_DEADLINE;
127 static inline bool valid_policy(int policy)
129 return idle_policy(policy) || fair_policy(policy) ||
130 rt_policy(policy) || dl_policy(policy);
133 static inline int task_has_rt_policy(struct task_struct *p)
135 return rt_policy(p->policy);
138 static inline int task_has_dl_policy(struct task_struct *p)
140 return dl_policy(p->policy);
144 * Tells if entity @a should preempt entity @b.
147 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
149 return dl_time_before(a->deadline, b->deadline);
153 * This is the priority-queue data structure of the RT scheduling class:
155 struct rt_prio_array {
156 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
157 struct list_head queue[MAX_RT_PRIO];
160 struct rt_bandwidth {
161 /* nests inside the rq lock: */
162 raw_spinlock_t rt_runtime_lock;
165 struct hrtimer rt_period_timer;
166 unsigned int rt_period_active;
169 void __dl_clear_params(struct task_struct *p);
172 * To keep the bandwidth of -deadline tasks and groups under control
173 * we need some place where:
174 * - store the maximum -deadline bandwidth of the system (the group);
175 * - cache the fraction of that bandwidth that is currently allocated.
177 * This is all done in the data structure below. It is similar to the
178 * one used for RT-throttling (rt_bandwidth), with the main difference
179 * that, since here we are only interested in admission control, we
180 * do not decrease any runtime while the group "executes", neither we
181 * need a timer to replenish it.
183 * With respect to SMP, the bandwidth is given on a per-CPU basis,
185 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
186 * - dl_total_bw array contains, in the i-eth element, the currently
187 * allocated bandwidth on the i-eth CPU.
188 * Moreover, groups consume bandwidth on each CPU, while tasks only
189 * consume bandwidth on the CPU they're running on.
190 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
191 * that will be shown the next time the proc or cgroup controls will
192 * be red. It on its turn can be changed by writing on its own
195 struct dl_bandwidth {
196 raw_spinlock_t dl_runtime_lock;
201 static inline int dl_bandwidth_enabled(void)
203 return sysctl_sched_rt_runtime >= 0;
206 extern struct dl_bw *dl_bw_of(int i);
214 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
216 dl_b->total_bw -= tsk_bw;
220 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
222 dl_b->total_bw += tsk_bw;
226 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
228 return dl_b->bw != -1 &&
229 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
232 extern struct mutex sched_domains_mutex;
234 #ifdef CONFIG_CGROUP_SCHED
236 #include <linux/cgroup.h>
241 extern struct list_head task_groups;
243 struct cfs_bandwidth {
244 #ifdef CONFIG_CFS_BANDWIDTH
248 s64 hierarchical_quota;
251 int idle, period_active;
252 struct hrtimer period_timer, slack_timer;
253 struct list_head throttled_cfs_rq;
256 int nr_periods, nr_throttled;
259 bool distribute_running;
263 /* task group related information */
265 struct cgroup_subsys_state css;
267 #ifdef CONFIG_FAIR_GROUP_SCHED
268 /* schedulable entities of this group on each cpu */
269 struct sched_entity **se;
270 /* runqueue "owned" by this group on each cpu */
271 struct cfs_rq **cfs_rq;
272 unsigned long shares;
276 * load_avg can be heavily contended at clock tick time, so put
277 * it in its own cacheline separated from the fields above which
278 * will also be accessed at each tick.
280 atomic_long_t load_avg ____cacheline_aligned;
284 #ifdef CONFIG_RT_GROUP_SCHED
285 struct sched_rt_entity **rt_se;
286 struct rt_rq **rt_rq;
288 struct rt_bandwidth rt_bandwidth;
292 struct list_head list;
294 struct task_group *parent;
295 struct list_head siblings;
296 struct list_head children;
298 #ifdef CONFIG_SCHED_AUTOGROUP
299 struct autogroup *autogroup;
302 struct cfs_bandwidth cfs_bandwidth;
305 #ifdef CONFIG_FAIR_GROUP_SCHED
306 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
309 * A weight of 0 or 1 can cause arithmetics problems.
310 * A weight of a cfs_rq is the sum of weights of which entities
311 * are queued on this cfs_rq, so a weight of a entity should not be
312 * too large, so as the shares value of a task group.
313 * (The default weight is 1024 - so there's no practical
314 * limitation from this.)
316 #define MIN_SHARES (1UL << 1)
317 #define MAX_SHARES (1UL << 18)
320 typedef int (*tg_visitor)(struct task_group *, void *);
322 extern int walk_tg_tree_from(struct task_group *from,
323 tg_visitor down, tg_visitor up, void *data);
326 * Iterate the full tree, calling @down when first entering a node and @up when
327 * leaving it for the final time.
329 * Caller must hold rcu_lock or sufficient equivalent.
331 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
333 return walk_tg_tree_from(&root_task_group, down, up, data);
336 extern int tg_nop(struct task_group *tg, void *data);
338 extern void free_fair_sched_group(struct task_group *tg);
339 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
340 extern void online_fair_sched_group(struct task_group *tg);
341 extern void unregister_fair_sched_group(struct task_group *tg);
342 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
343 struct sched_entity *se, int cpu,
344 struct sched_entity *parent);
345 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
347 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
348 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
349 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
351 extern void free_rt_sched_group(struct task_group *tg);
352 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
353 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
354 struct sched_rt_entity *rt_se, int cpu,
355 struct sched_rt_entity *parent);
357 extern struct task_group *sched_create_group(struct task_group *parent);
358 extern void sched_online_group(struct task_group *tg,
359 struct task_group *parent);
360 extern void sched_destroy_group(struct task_group *tg);
361 extern void sched_offline_group(struct task_group *tg);
363 extern void sched_move_task(struct task_struct *tsk);
365 #ifdef CONFIG_FAIR_GROUP_SCHED
366 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
369 extern void set_task_rq_fair(struct sched_entity *se,
370 struct cfs_rq *prev, struct cfs_rq *next);
371 #else /* !CONFIG_SMP */
372 static inline void set_task_rq_fair(struct sched_entity *se,
373 struct cfs_rq *prev, struct cfs_rq *next) { }
374 #endif /* CONFIG_SMP */
375 #endif /* CONFIG_FAIR_GROUP_SCHED */
377 #else /* CONFIG_CGROUP_SCHED */
379 struct cfs_bandwidth { };
381 #endif /* CONFIG_CGROUP_SCHED */
383 /* CFS-related fields in a runqueue */
385 struct load_weight load;
386 unsigned int nr_running, h_nr_running;
391 u64 min_vruntime_copy;
394 struct rb_root tasks_timeline;
395 struct rb_node *rb_leftmost;
398 * 'curr' points to currently running entity on this cfs_rq.
399 * It is set to NULL otherwise (i.e when none are currently running).
401 struct sched_entity *curr, *next, *last, *skip;
403 #ifdef CONFIG_SCHED_DEBUG
404 unsigned int nr_spread_over;
411 struct sched_avg avg;
412 u64 runnable_load_sum;
413 unsigned long runnable_load_avg;
414 #ifdef CONFIG_FAIR_GROUP_SCHED
415 unsigned long tg_load_avg_contrib;
417 atomic_long_t removed_load_avg, removed_util_avg;
419 u64 load_last_update_time_copy;
422 #ifdef CONFIG_FAIR_GROUP_SCHED
424 * h_load = weight * f(tg)
426 * Where f(tg) is the recursive weight fraction assigned to
429 unsigned long h_load;
430 u64 last_h_load_update;
431 struct sched_entity *h_load_next;
432 #endif /* CONFIG_FAIR_GROUP_SCHED */
433 #endif /* CONFIG_SMP */
435 #ifdef CONFIG_FAIR_GROUP_SCHED
436 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
439 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
440 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
441 * (like users, containers etc.)
443 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
444 * list is used during load balance.
447 struct list_head leaf_cfs_rq_list;
448 struct task_group *tg; /* group that "owns" this runqueue */
450 #ifdef CONFIG_CFS_BANDWIDTH
453 s64 runtime_remaining;
455 u64 throttled_clock, throttled_clock_task;
456 u64 throttled_clock_task_time;
457 int throttled, throttle_count;
458 struct list_head throttled_list;
459 #endif /* CONFIG_CFS_BANDWIDTH */
460 #endif /* CONFIG_FAIR_GROUP_SCHED */
463 static inline int rt_bandwidth_enabled(void)
465 return sysctl_sched_rt_runtime >= 0;
468 /* RT IPI pull logic requires IRQ_WORK */
469 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
470 # define HAVE_RT_PUSH_IPI
473 /* Real-Time classes' related field in a runqueue: */
475 struct rt_prio_array active;
476 unsigned int rt_nr_running;
477 unsigned int rr_nr_running;
478 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
480 int curr; /* highest queued rt task prio */
482 int next; /* next highest */
487 unsigned long rt_nr_migratory;
488 unsigned long rt_nr_total;
490 struct plist_head pushable_tasks;
491 #endif /* CONFIG_SMP */
497 /* Nests inside the rq lock: */
498 raw_spinlock_t rt_runtime_lock;
500 #ifdef CONFIG_RT_GROUP_SCHED
501 unsigned long rt_nr_boosted;
504 struct task_group *tg;
508 /* Deadline class' related fields in a runqueue */
510 /* runqueue is an rbtree, ordered by deadline */
511 struct rb_root rb_root;
512 struct rb_node *rb_leftmost;
514 unsigned long dl_nr_running;
518 * Deadline values of the currently executing and the
519 * earliest ready task on this rq. Caching these facilitates
520 * the decision wether or not a ready but not running task
521 * should migrate somewhere else.
528 unsigned long dl_nr_migratory;
532 * Tasks on this rq that can be pushed away. They are kept in
533 * an rb-tree, ordered by tasks' deadlines, with caching
534 * of the leftmost (earliest deadline) element.
536 struct rb_root pushable_dl_tasks_root;
537 struct rb_node *pushable_dl_tasks_leftmost;
546 * We add the notion of a root-domain which will be used to define per-domain
547 * variables. Each exclusive cpuset essentially defines an island domain by
548 * fully partitioning the member cpus from any other cpuset. Whenever a new
549 * exclusive cpuset is created, we also create and attach a new root-domain
558 cpumask_var_t online;
560 /* Indicate more than one runnable task for any CPU */
564 * The bit corresponding to a CPU gets set here if such CPU has more
565 * than one runnable -deadline task (as it is below for RT tasks).
567 cpumask_var_t dlo_mask;
572 #ifdef HAVE_RT_PUSH_IPI
574 * For IPI pull requests, loop across the rto_mask.
576 struct irq_work rto_push_work;
577 raw_spinlock_t rto_lock;
578 /* These are only updated and read within rto_lock */
581 /* These atomics are updated outside of a lock */
582 atomic_t rto_loop_next;
583 atomic_t rto_loop_start;
586 * The "RT overload" flag: it gets set if a CPU has more than
587 * one runnable RT task.
589 cpumask_var_t rto_mask;
590 struct cpupri cpupri;
592 unsigned long max_cpu_capacity;
595 extern struct root_domain def_root_domain;
596 extern void sched_get_rd(struct root_domain *rd);
597 extern void sched_put_rd(struct root_domain *rd);
599 #ifdef HAVE_RT_PUSH_IPI
600 extern void rto_push_irq_work_func(struct irq_work *work);
602 #endif /* CONFIG_SMP */
605 * This is the main, per-CPU runqueue data structure.
607 * Locking rule: those places that want to lock multiple runqueues
608 * (such as the load balancing or the thread migration code), lock
609 * acquire operations must be ordered by ascending &runqueue.
616 * nr_running and cpu_load should be in the same cacheline because
617 * remote CPUs use both these fields when doing load calculation.
619 unsigned int nr_running;
620 #ifdef CONFIG_NUMA_BALANCING
621 unsigned int nr_numa_running;
622 unsigned int nr_preferred_running;
624 #define CPU_LOAD_IDX_MAX 5
625 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
626 #ifdef CONFIG_NO_HZ_COMMON
628 unsigned long last_load_update_tick;
629 #endif /* CONFIG_SMP */
630 unsigned long nohz_flags;
631 #endif /* CONFIG_NO_HZ_COMMON */
632 #ifdef CONFIG_NO_HZ_FULL
633 unsigned long last_sched_tick;
635 /* capture load from *all* tasks on this cpu: */
636 struct load_weight load;
637 unsigned long nr_load_updates;
644 #ifdef CONFIG_FAIR_GROUP_SCHED
645 /* list of leaf cfs_rq on this cpu: */
646 struct list_head leaf_cfs_rq_list;
647 #endif /* CONFIG_FAIR_GROUP_SCHED */
650 * This is part of a global counter where only the total sum
651 * over all CPUs matters. A task can increase this counter on
652 * one CPU and if it got migrated afterwards it may decrease
653 * it on another CPU. Always updated under the runqueue lock:
655 unsigned long nr_uninterruptible;
657 struct task_struct *curr, *idle, *stop;
658 unsigned long next_balance;
659 struct mm_struct *prev_mm;
661 unsigned int clock_skip_update;
668 struct root_domain *rd;
669 struct sched_domain *sd;
671 unsigned long cpu_capacity;
672 unsigned long cpu_capacity_orig;
674 struct callback_head *balance_callback;
676 unsigned char idle_balance;
677 /* For active balancing */
680 struct cpu_stop_work active_balance_work;
681 /* cpu of this runqueue: */
685 struct list_head cfs_tasks;
692 /* This is used to determine avg_idle's max value */
693 u64 max_idle_balance_cost;
696 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
699 #ifdef CONFIG_PARAVIRT
702 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
703 u64 prev_steal_time_rq;
706 /* calc_load related fields */
707 unsigned long calc_load_update;
708 long calc_load_active;
710 #ifdef CONFIG_SCHED_HRTICK
712 int hrtick_csd_pending;
713 struct call_single_data hrtick_csd;
715 struct hrtimer hrtick_timer;
718 #ifdef CONFIG_SCHEDSTATS
720 struct sched_info rq_sched_info;
721 unsigned long long rq_cpu_time;
722 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
724 /* sys_sched_yield() stats */
725 unsigned int yld_count;
727 /* schedule() stats */
728 unsigned int sched_count;
729 unsigned int sched_goidle;
731 /* try_to_wake_up() stats */
732 unsigned int ttwu_count;
733 unsigned int ttwu_local;
737 struct llist_head wake_list;
740 #ifdef CONFIG_CPU_IDLE
741 /* Must be inspected within a rcu lock section */
742 struct cpuidle_state *idle_state;
746 static inline int cpu_of(struct rq *rq)
755 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
757 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
758 #define this_rq() this_cpu_ptr(&runqueues)
759 #define task_rq(p) cpu_rq(task_cpu(p))
760 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
761 #define raw_rq() raw_cpu_ptr(&runqueues)
763 static inline u64 __rq_clock_broken(struct rq *rq)
765 return READ_ONCE(rq->clock);
768 static inline u64 rq_clock(struct rq *rq)
770 lockdep_assert_held(&rq->lock);
774 static inline u64 rq_clock_task(struct rq *rq)
776 lockdep_assert_held(&rq->lock);
777 return rq->clock_task;
780 #define RQCF_REQ_SKIP 0x01
781 #define RQCF_ACT_SKIP 0x02
783 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
785 lockdep_assert_held(&rq->lock);
787 rq->clock_skip_update |= RQCF_REQ_SKIP;
789 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
793 enum numa_topology_type {
798 extern enum numa_topology_type sched_numa_topology_type;
799 extern int sched_max_numa_distance;
800 extern bool find_numa_distance(int distance);
803 #ifdef CONFIG_NUMA_BALANCING
804 /* The regions in numa_faults array from task_struct */
805 enum numa_faults_stats {
811 extern void sched_setnuma(struct task_struct *p, int node);
812 extern int migrate_task_to(struct task_struct *p, int cpu);
813 extern int migrate_swap(struct task_struct *, struct task_struct *);
814 #endif /* CONFIG_NUMA_BALANCING */
819 queue_balance_callback(struct rq *rq,
820 struct callback_head *head,
821 void (*func)(struct rq *rq))
823 lockdep_assert_held(&rq->lock);
825 if (unlikely(head->next))
828 head->func = (void (*)(struct callback_head *))func;
829 head->next = rq->balance_callback;
830 rq->balance_callback = head;
833 extern void sched_ttwu_pending(void);
835 #define rcu_dereference_check_sched_domain(p) \
836 rcu_dereference_check((p), \
837 lockdep_is_held(&sched_domains_mutex))
840 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
841 * See detach_destroy_domains: synchronize_sched for details.
843 * The domain tree of any CPU may only be accessed from within
844 * preempt-disabled sections.
846 #define for_each_domain(cpu, __sd) \
847 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
848 __sd; __sd = __sd->parent)
850 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
853 * highest_flag_domain - Return highest sched_domain containing flag.
854 * @cpu: The cpu whose highest level of sched domain is to
856 * @flag: The flag to check for the highest sched_domain
859 * Returns the highest sched_domain of a cpu which contains the given flag.
861 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
863 struct sched_domain *sd, *hsd = NULL;
865 for_each_domain(cpu, sd) {
866 if (!(sd->flags & flag))
874 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
876 struct sched_domain *sd;
878 for_each_domain(cpu, sd) {
879 if (sd->flags & flag)
886 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
887 DECLARE_PER_CPU(int, sd_llc_size);
888 DECLARE_PER_CPU(int, sd_llc_id);
889 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
890 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
891 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
893 struct sched_group_capacity {
896 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
899 unsigned int capacity;
900 unsigned long next_update;
901 int imbalance; /* XXX unrelated to capacity but shared group state */
903 unsigned long cpumask[0]; /* iteration mask */
907 struct sched_group *next; /* Must be a circular list */
910 unsigned int group_weight;
911 struct sched_group_capacity *sgc;
914 * The CPUs this group covers.
916 * NOTE: this field is variable length. (Allocated dynamically
917 * by attaching extra space to the end of the structure,
918 * depending on how many CPUs the kernel has booted up with)
920 unsigned long cpumask[0];
923 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
925 return to_cpumask(sg->cpumask);
929 * cpumask masking which cpus in the group are allowed to iterate up the domain
932 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
934 return to_cpumask(sg->sgc->cpumask);
938 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
939 * @group: The group whose first cpu is to be returned.
941 static inline unsigned int group_first_cpu(struct sched_group *group)
943 return cpumask_first(sched_group_cpus(group));
946 extern int group_balance_cpu(struct sched_group *sg);
948 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
949 void register_sched_domain_sysctl(void);
950 void unregister_sched_domain_sysctl(void);
952 static inline void register_sched_domain_sysctl(void)
955 static inline void unregister_sched_domain_sysctl(void)
962 static inline void sched_ttwu_pending(void) { }
964 #endif /* CONFIG_SMP */
967 #include "auto_group.h"
969 #ifdef CONFIG_CGROUP_SCHED
972 * Return the group to which this tasks belongs.
974 * We cannot use task_css() and friends because the cgroup subsystem
975 * changes that value before the cgroup_subsys::attach() method is called,
976 * therefore we cannot pin it and might observe the wrong value.
978 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
979 * core changes this before calling sched_move_task().
981 * Instead we use a 'copy' which is updated from sched_move_task() while
982 * holding both task_struct::pi_lock and rq::lock.
984 static inline struct task_group *task_group(struct task_struct *p)
986 return p->sched_task_group;
989 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
990 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
992 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
993 struct task_group *tg = task_group(p);
996 #ifdef CONFIG_FAIR_GROUP_SCHED
997 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
998 p->se.cfs_rq = tg->cfs_rq[cpu];
999 p->se.parent = tg->se[cpu];
1002 #ifdef CONFIG_RT_GROUP_SCHED
1003 p->rt.rt_rq = tg->rt_rq[cpu];
1004 p->rt.parent = tg->rt_se[cpu];
1008 #else /* CONFIG_CGROUP_SCHED */
1010 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1011 static inline struct task_group *task_group(struct task_struct *p)
1016 #endif /* CONFIG_CGROUP_SCHED */
1018 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1020 set_task_rq(p, cpu);
1023 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1024 * successfuly executed on another CPU. We must ensure that updates of
1025 * per-task data have been completed by this moment.
1028 #ifdef CONFIG_THREAD_INFO_IN_TASK
1031 task_thread_info(p)->cpu = cpu;
1038 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1040 #ifdef CONFIG_SCHED_DEBUG
1041 # include <linux/static_key.h>
1042 # define const_debug __read_mostly
1044 # define const_debug const
1047 extern const_debug unsigned int sysctl_sched_features;
1049 #define SCHED_FEAT(name, enabled) \
1050 __SCHED_FEAT_##name ,
1053 #include "features.h"
1059 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1060 #define SCHED_FEAT(name, enabled) \
1061 static __always_inline bool static_branch_##name(struct static_key *key) \
1063 return static_key_##enabled(key); \
1066 #include "features.h"
1070 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1071 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1072 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1073 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1074 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1076 extern struct static_key_false sched_numa_balancing;
1077 extern struct static_key_false sched_schedstats;
1079 static inline u64 global_rt_period(void)
1081 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1084 static inline u64 global_rt_runtime(void)
1086 if (sysctl_sched_rt_runtime < 0)
1089 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1092 static inline int task_current(struct rq *rq, struct task_struct *p)
1094 return rq->curr == p;
1097 static inline int task_running(struct rq *rq, struct task_struct *p)
1102 return task_current(rq, p);
1106 static inline int task_on_rq_queued(struct task_struct *p)
1108 return p->on_rq == TASK_ON_RQ_QUEUED;
1111 static inline int task_on_rq_migrating(struct task_struct *p)
1113 return p->on_rq == TASK_ON_RQ_MIGRATING;
1116 #ifndef prepare_arch_switch
1117 # define prepare_arch_switch(next) do { } while (0)
1119 #ifndef finish_arch_post_lock_switch
1120 # define finish_arch_post_lock_switch() do { } while (0)
1123 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1127 * We can optimise this out completely for !SMP, because the
1128 * SMP rebalancing from interrupt is the only thing that cares
1135 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1139 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1140 * We must ensure this doesn't happen until the switch is completely
1143 * In particular, the load of prev->state in finish_task_switch() must
1144 * happen before this.
1146 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1148 smp_store_release(&prev->on_cpu, 0);
1150 #ifdef CONFIG_DEBUG_SPINLOCK
1151 /* this is a valid case when another task releases the spinlock */
1152 rq->lock.owner = current;
1155 * If we are tracking spinlock dependencies then we have to
1156 * fix up the runqueue lock - which gets 'carried over' from
1157 * prev into current:
1159 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1161 raw_spin_unlock_irq(&rq->lock);
1167 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1168 #define WF_FORK 0x02 /* child wakeup after fork */
1169 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1172 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1173 * of tasks with abnormal "nice" values across CPUs the contribution that
1174 * each task makes to its run queue's load is weighted according to its
1175 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1176 * scaled version of the new time slice allocation that they receive on time
1180 #define WEIGHT_IDLEPRIO 3
1181 #define WMULT_IDLEPRIO 1431655765
1183 extern const int sched_prio_to_weight[40];
1184 extern const u32 sched_prio_to_wmult[40];
1187 * {de,en}queue flags:
1189 * DEQUEUE_SLEEP - task is no longer runnable
1190 * ENQUEUE_WAKEUP - task just became runnable
1192 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1193 * are in a known state which allows modification. Such pairs
1194 * should preserve as much state as possible.
1196 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1199 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1200 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1201 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1205 #define DEQUEUE_SLEEP 0x01
1206 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1207 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1209 #define ENQUEUE_WAKEUP 0x01
1210 #define ENQUEUE_RESTORE 0x02
1211 #define ENQUEUE_MOVE 0x04
1213 #define ENQUEUE_HEAD 0x08
1214 #define ENQUEUE_REPLENISH 0x10
1216 #define ENQUEUE_MIGRATED 0x20
1218 #define ENQUEUE_MIGRATED 0x00
1221 #define RETRY_TASK ((void *)-1UL)
1223 struct sched_class {
1224 const struct sched_class *next;
1226 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1227 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1228 void (*yield_task) (struct rq *rq);
1229 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1231 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1234 * It is the responsibility of the pick_next_task() method that will
1235 * return the next task to call put_prev_task() on the @prev task or
1236 * something equivalent.
1238 * May return RETRY_TASK when it finds a higher prio class has runnable
1241 struct task_struct * (*pick_next_task) (struct rq *rq,
1242 struct task_struct *prev,
1243 struct pin_cookie cookie);
1244 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1247 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1248 void (*migrate_task_rq)(struct task_struct *p);
1250 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1252 void (*set_cpus_allowed)(struct task_struct *p,
1253 const struct cpumask *newmask);
1255 void (*rq_online)(struct rq *rq);
1256 void (*rq_offline)(struct rq *rq);
1259 void (*set_curr_task) (struct rq *rq);
1260 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1261 void (*task_fork) (struct task_struct *p);
1262 void (*task_dead) (struct task_struct *p);
1265 * The switched_from() call is allowed to drop rq->lock, therefore we
1266 * cannot assume the switched_from/switched_to pair is serliazed by
1267 * rq->lock. They are however serialized by p->pi_lock.
1269 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1270 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1271 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1274 unsigned int (*get_rr_interval) (struct rq *rq,
1275 struct task_struct *task);
1277 void (*update_curr) (struct rq *rq);
1279 #define TASK_SET_GROUP 0
1280 #define TASK_MOVE_GROUP 1
1282 #ifdef CONFIG_FAIR_GROUP_SCHED
1283 void (*task_change_group) (struct task_struct *p, int type);
1287 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1289 prev->sched_class->put_prev_task(rq, prev);
1292 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1294 curr->sched_class->set_curr_task(rq);
1297 #define sched_class_highest (&stop_sched_class)
1298 #define for_each_class(class) \
1299 for (class = sched_class_highest; class; class = class->next)
1301 extern const struct sched_class stop_sched_class;
1302 extern const struct sched_class dl_sched_class;
1303 extern const struct sched_class rt_sched_class;
1304 extern const struct sched_class fair_sched_class;
1305 extern const struct sched_class idle_sched_class;
1310 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1312 extern void trigger_load_balance(struct rq *rq);
1314 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1318 #ifdef CONFIG_CPU_IDLE
1319 static inline void idle_set_state(struct rq *rq,
1320 struct cpuidle_state *idle_state)
1322 rq->idle_state = idle_state;
1325 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1327 SCHED_WARN_ON(!rcu_read_lock_held());
1328 return rq->idle_state;
1331 static inline void idle_set_state(struct rq *rq,
1332 struct cpuidle_state *idle_state)
1336 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1342 extern void sysrq_sched_debug_show(void);
1343 extern void sched_init_granularity(void);
1344 extern void update_max_interval(void);
1346 extern void init_sched_dl_class(void);
1347 extern void init_sched_rt_class(void);
1348 extern void init_sched_fair_class(void);
1350 extern void resched_curr(struct rq *rq);
1351 extern void resched_cpu(int cpu);
1353 extern struct rt_bandwidth def_rt_bandwidth;
1354 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1356 extern struct dl_bandwidth def_dl_bandwidth;
1357 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1358 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1360 unsigned long to_ratio(u64 period, u64 runtime);
1362 extern void init_entity_runnable_average(struct sched_entity *se);
1363 extern void post_init_entity_util_avg(struct sched_entity *se);
1365 #ifdef CONFIG_NO_HZ_FULL
1366 extern bool sched_can_stop_tick(struct rq *rq);
1369 * Tick may be needed by tasks in the runqueue depending on their policy and
1370 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1371 * nohz mode if necessary.
1373 static inline void sched_update_tick_dependency(struct rq *rq)
1377 if (!tick_nohz_full_enabled())
1382 if (!tick_nohz_full_cpu(cpu))
1385 if (sched_can_stop_tick(rq))
1386 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1388 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1391 static inline void sched_update_tick_dependency(struct rq *rq) { }
1394 static inline void add_nr_running(struct rq *rq, unsigned count)
1396 unsigned prev_nr = rq->nr_running;
1398 rq->nr_running = prev_nr + count;
1400 if (prev_nr < 2 && rq->nr_running >= 2) {
1402 if (!rq->rd->overload)
1403 rq->rd->overload = true;
1407 sched_update_tick_dependency(rq);
1410 static inline void sub_nr_running(struct rq *rq, unsigned count)
1412 rq->nr_running -= count;
1413 /* Check if we still need preemption */
1414 sched_update_tick_dependency(rq);
1417 static inline void rq_last_tick_reset(struct rq *rq)
1419 #ifdef CONFIG_NO_HZ_FULL
1420 rq->last_sched_tick = jiffies;
1424 extern void update_rq_clock(struct rq *rq);
1426 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1427 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1429 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1431 extern const_debug unsigned int sysctl_sched_time_avg;
1432 extern const_debug unsigned int sysctl_sched_nr_migrate;
1433 extern const_debug unsigned int sysctl_sched_migration_cost;
1435 static inline u64 sched_avg_period(void)
1437 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1440 #ifdef CONFIG_SCHED_HRTICK
1444 * - enabled by features
1445 * - hrtimer is actually high res
1447 static inline int hrtick_enabled(struct rq *rq)
1449 if (!sched_feat(HRTICK))
1451 if (!cpu_active(cpu_of(rq)))
1453 return hrtimer_is_hres_active(&rq->hrtick_timer);
1456 void hrtick_start(struct rq *rq, u64 delay);
1460 static inline int hrtick_enabled(struct rq *rq)
1465 #endif /* CONFIG_SCHED_HRTICK */
1468 extern void sched_avg_update(struct rq *rq);
1470 #ifndef arch_scale_freq_capacity
1471 static __always_inline
1472 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1474 return SCHED_CAPACITY_SCALE;
1478 #ifndef arch_scale_cpu_capacity
1479 static __always_inline
1480 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1482 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1483 return sd->smt_gain / sd->span_weight;
1485 return SCHED_CAPACITY_SCALE;
1489 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1491 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1492 sched_avg_update(rq);
1495 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1496 static inline void sched_avg_update(struct rq *rq) { }
1500 unsigned long flags;
1501 struct pin_cookie cookie;
1504 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1505 __acquires(rq->lock);
1506 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1507 __acquires(p->pi_lock)
1508 __acquires(rq->lock);
1510 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1511 __releases(rq->lock)
1513 lockdep_unpin_lock(&rq->lock, rf->cookie);
1514 raw_spin_unlock(&rq->lock);
1518 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1519 __releases(rq->lock)
1520 __releases(p->pi_lock)
1522 lockdep_unpin_lock(&rq->lock, rf->cookie);
1523 raw_spin_unlock(&rq->lock);
1524 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1528 #ifdef CONFIG_PREEMPT
1530 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1533 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1534 * way at the expense of forcing extra atomic operations in all
1535 * invocations. This assures that the double_lock is acquired using the
1536 * same underlying policy as the spinlock_t on this architecture, which
1537 * reduces latency compared to the unfair variant below. However, it
1538 * also adds more overhead and therefore may reduce throughput.
1540 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1541 __releases(this_rq->lock)
1542 __acquires(busiest->lock)
1543 __acquires(this_rq->lock)
1545 raw_spin_unlock(&this_rq->lock);
1546 double_rq_lock(this_rq, busiest);
1553 * Unfair double_lock_balance: Optimizes throughput at the expense of
1554 * latency by eliminating extra atomic operations when the locks are
1555 * already in proper order on entry. This favors lower cpu-ids and will
1556 * grant the double lock to lower cpus over higher ids under contention,
1557 * regardless of entry order into the function.
1559 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1560 __releases(this_rq->lock)
1561 __acquires(busiest->lock)
1562 __acquires(this_rq->lock)
1566 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1567 if (busiest < this_rq) {
1568 raw_spin_unlock(&this_rq->lock);
1569 raw_spin_lock(&busiest->lock);
1570 raw_spin_lock_nested(&this_rq->lock,
1571 SINGLE_DEPTH_NESTING);
1574 raw_spin_lock_nested(&busiest->lock,
1575 SINGLE_DEPTH_NESTING);
1580 #endif /* CONFIG_PREEMPT */
1583 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1585 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1587 if (unlikely(!irqs_disabled())) {
1588 /* printk() doesn't work good under rq->lock */
1589 raw_spin_unlock(&this_rq->lock);
1593 return _double_lock_balance(this_rq, busiest);
1596 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1597 __releases(busiest->lock)
1599 raw_spin_unlock(&busiest->lock);
1600 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1603 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1609 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1612 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1618 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1621 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1627 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1631 * double_rq_lock - safely lock two runqueues
1633 * Note this does not disable interrupts like task_rq_lock,
1634 * you need to do so manually before calling.
1636 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1637 __acquires(rq1->lock)
1638 __acquires(rq2->lock)
1640 BUG_ON(!irqs_disabled());
1642 raw_spin_lock(&rq1->lock);
1643 __acquire(rq2->lock); /* Fake it out ;) */
1646 raw_spin_lock(&rq1->lock);
1647 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1649 raw_spin_lock(&rq2->lock);
1650 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1656 * double_rq_unlock - safely unlock two runqueues
1658 * Note this does not restore interrupts like task_rq_unlock,
1659 * you need to do so manually after calling.
1661 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1662 __releases(rq1->lock)
1663 __releases(rq2->lock)
1665 raw_spin_unlock(&rq1->lock);
1667 raw_spin_unlock(&rq2->lock);
1669 __release(rq2->lock);
1672 #else /* CONFIG_SMP */
1675 * double_rq_lock - safely lock two runqueues
1677 * Note this does not disable interrupts like task_rq_lock,
1678 * you need to do so manually before calling.
1680 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1681 __acquires(rq1->lock)
1682 __acquires(rq2->lock)
1684 BUG_ON(!irqs_disabled());
1686 raw_spin_lock(&rq1->lock);
1687 __acquire(rq2->lock); /* Fake it out ;) */
1691 * double_rq_unlock - safely unlock two runqueues
1693 * Note this does not restore interrupts like task_rq_unlock,
1694 * you need to do so manually after calling.
1696 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1697 __releases(rq1->lock)
1698 __releases(rq2->lock)
1701 raw_spin_unlock(&rq1->lock);
1702 __release(rq2->lock);
1707 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1708 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1710 #ifdef CONFIG_SCHED_DEBUG
1711 extern void print_cfs_stats(struct seq_file *m, int cpu);
1712 extern void print_rt_stats(struct seq_file *m, int cpu);
1713 extern void print_dl_stats(struct seq_file *m, int cpu);
1715 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1717 #ifdef CONFIG_NUMA_BALANCING
1719 show_numa_stats(struct task_struct *p, struct seq_file *m);
1721 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1722 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1723 #endif /* CONFIG_NUMA_BALANCING */
1724 #endif /* CONFIG_SCHED_DEBUG */
1726 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1727 extern void init_rt_rq(struct rt_rq *rt_rq);
1728 extern void init_dl_rq(struct dl_rq *dl_rq);
1730 extern void cfs_bandwidth_usage_inc(void);
1731 extern void cfs_bandwidth_usage_dec(void);
1733 #ifdef CONFIG_NO_HZ_COMMON
1734 enum rq_nohz_flag_bits {
1739 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1741 extern void nohz_balance_exit_idle(unsigned int cpu);
1743 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1746 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1751 struct u64_stats_sync sync;
1754 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
1757 * Returns the irqtime minus the softirq time computed by ksoftirqd.
1758 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
1759 * and never move forward.
1761 static inline u64 irq_time_read(int cpu)
1763 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
1768 seq = __u64_stats_fetch_begin(&irqtime->sync);
1769 total = irqtime->total;
1770 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
1774 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1776 #ifdef CONFIG_CPU_FREQ
1777 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1780 * cpufreq_update_util - Take a note about CPU utilization changes.
1781 * @rq: Runqueue to carry out the update for.
1782 * @flags: Update reason flags.
1784 * This function is called by the scheduler on the CPU whose utilization is
1787 * It can only be called from RCU-sched read-side critical sections.
1789 * The way cpufreq is currently arranged requires it to evaluate the CPU
1790 * performance state (frequency/voltage) on a regular basis to prevent it from
1791 * being stuck in a completely inadequate performance level for too long.
1792 * That is not guaranteed to happen if the updates are only triggered from CFS,
1793 * though, because they may not be coming in if RT or deadline tasks are active
1794 * all the time (or there are RT and DL tasks only).
1796 * As a workaround for that issue, this function is called by the RT and DL
1797 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1798 * but that really is a band-aid. Going forward it should be replaced with
1799 * solutions targeted more specifically at RT and DL tasks.
1801 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
1803 struct update_util_data *data;
1805 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1807 data->func(data, rq_clock(rq), flags);
1810 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags)
1812 if (cpu_of(rq) == smp_processor_id())
1813 cpufreq_update_util(rq, flags);
1816 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
1817 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {}
1818 #endif /* CONFIG_CPU_FREQ */
1820 #ifdef arch_scale_freq_capacity
1821 #ifndef arch_scale_freq_invariant
1822 #define arch_scale_freq_invariant() (true)
1824 #else /* arch_scale_freq_capacity */
1825 #define arch_scale_freq_invariant() (false)