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/binfmts.h>
8 #include <linux/mutex.h>
9 #include <linux/spinlock.h>
10 #include <linux/stop_machine.h>
11 #include <linux/irq_work.h>
12 #include <linux/tick.h>
13 #include <linux/slab.h>
16 #include "cpudeadline.h"
19 #ifdef CONFIG_SCHED_DEBUG
20 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
22 #define SCHED_WARN_ON(x) ((void)(x))
28 /* task_struct::on_rq states: */
29 #define TASK_ON_RQ_QUEUED 1
30 #define TASK_ON_RQ_MIGRATING 2
32 extern __read_mostly int scheduler_running;
34 extern unsigned long calc_load_update;
35 extern atomic_long_t calc_load_tasks;
37 extern void calc_global_load_tick(struct rq *this_rq);
38 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
41 extern void cpu_load_update_active(struct rq *this_rq);
43 static inline void cpu_load_update_active(struct rq *this_rq) { }
47 * Helpers for converting nanosecond timing to jiffy resolution
49 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
52 * Increase resolution of nice-level calculations for 64-bit architectures.
53 * The extra resolution improves shares distribution and load balancing of
54 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
55 * hierarchies, especially on larger systems. This is not a user-visible change
56 * and does not change the user-interface for setting shares/weights.
58 * We increase resolution only if we have enough bits to allow this increased
59 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
60 * pretty high and the returns do not justify the increased costs.
62 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
63 * increase coverage and consistency always enable it on 64bit platforms.
66 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
67 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
68 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
70 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
71 # define scale_load(w) (w)
72 # define scale_load_down(w) (w)
76 * Task weight (visible to users) and its load (invisible to users) have
77 * independent resolution, but they should be well calibrated. We use
78 * scale_load() and scale_load_down(w) to convert between them. The
79 * following must be true:
81 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
84 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
87 * Single value that decides SCHED_DEADLINE internal math precision.
88 * 10 -> just above 1us
89 * 9 -> just above 0.5us
94 * These are the 'tuning knobs' of the scheduler:
98 * single value that denotes runtime == period, ie unlimited time.
100 #define RUNTIME_INF ((u64)~0ULL)
102 static inline int idle_policy(int policy)
104 return policy == SCHED_IDLE;
106 static inline int fair_policy(int policy)
108 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
111 static inline int rt_policy(int policy)
113 return policy == SCHED_FIFO || policy == SCHED_RR;
116 static inline int dl_policy(int policy)
118 return policy == SCHED_DEADLINE;
120 static inline bool valid_policy(int policy)
122 return idle_policy(policy) || fair_policy(policy) ||
123 rt_policy(policy) || dl_policy(policy);
126 static inline int task_has_rt_policy(struct task_struct *p)
128 return rt_policy(p->policy);
131 static inline int task_has_dl_policy(struct task_struct *p)
133 return dl_policy(p->policy);
137 * Tells if entity @a should preempt entity @b.
140 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
142 return dl_time_before(a->deadline, b->deadline);
146 * This is the priority-queue data structure of the RT scheduling class:
148 struct rt_prio_array {
149 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
150 struct list_head queue[MAX_RT_PRIO];
153 struct rt_bandwidth {
154 /* nests inside the rq lock: */
155 raw_spinlock_t rt_runtime_lock;
158 struct hrtimer rt_period_timer;
159 unsigned int rt_period_active;
162 void __dl_clear_params(struct task_struct *p);
165 * To keep the bandwidth of -deadline tasks and groups under control
166 * we need some place where:
167 * - store the maximum -deadline bandwidth of the system (the group);
168 * - cache the fraction of that bandwidth that is currently allocated.
170 * This is all done in the data structure below. It is similar to the
171 * one used for RT-throttling (rt_bandwidth), with the main difference
172 * that, since here we are only interested in admission control, we
173 * do not decrease any runtime while the group "executes", neither we
174 * need a timer to replenish it.
176 * With respect to SMP, the bandwidth is given on a per-CPU basis,
178 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
179 * - dl_total_bw array contains, in the i-eth element, the currently
180 * allocated bandwidth on the i-eth CPU.
181 * Moreover, groups consume bandwidth on each CPU, while tasks only
182 * consume bandwidth on the CPU they're running on.
183 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
184 * that will be shown the next time the proc or cgroup controls will
185 * be red. It on its turn can be changed by writing on its own
188 struct dl_bandwidth {
189 raw_spinlock_t dl_runtime_lock;
194 static inline int dl_bandwidth_enabled(void)
196 return sysctl_sched_rt_runtime >= 0;
199 extern struct dl_bw *dl_bw_of(int i);
207 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
209 dl_b->total_bw -= tsk_bw;
213 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
215 dl_b->total_bw += tsk_bw;
219 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
221 return dl_b->bw != -1 &&
222 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
225 extern struct mutex sched_domains_mutex;
227 #ifdef CONFIG_CGROUP_SCHED
229 #include <linux/cgroup.h>
234 extern struct list_head task_groups;
236 struct cfs_bandwidth {
237 #ifdef CONFIG_CFS_BANDWIDTH
241 s64 hierarchical_quota;
244 int idle, period_active;
245 struct hrtimer period_timer, slack_timer;
246 struct list_head throttled_cfs_rq;
249 int nr_periods, nr_throttled;
254 /* task group related information */
256 struct cgroup_subsys_state css;
258 #ifdef CONFIG_FAIR_GROUP_SCHED
259 /* schedulable entities of this group on each cpu */
260 struct sched_entity **se;
261 /* runqueue "owned" by this group on each cpu */
262 struct cfs_rq **cfs_rq;
263 unsigned long shares;
267 * load_avg can be heavily contended at clock tick time, so put
268 * it in its own cacheline separated from the fields above which
269 * will also be accessed at each tick.
271 atomic_long_t load_avg ____cacheline_aligned;
275 #ifdef CONFIG_RT_GROUP_SCHED
276 struct sched_rt_entity **rt_se;
277 struct rt_rq **rt_rq;
279 struct rt_bandwidth rt_bandwidth;
283 struct list_head list;
285 struct task_group *parent;
286 struct list_head siblings;
287 struct list_head children;
289 #ifdef CONFIG_SCHED_AUTOGROUP
290 struct autogroup *autogroup;
293 struct cfs_bandwidth cfs_bandwidth;
296 #ifdef CONFIG_FAIR_GROUP_SCHED
297 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
300 * A weight of 0 or 1 can cause arithmetics problems.
301 * A weight of a cfs_rq is the sum of weights of which entities
302 * are queued on this cfs_rq, so a weight of a entity should not be
303 * too large, so as the shares value of a task group.
304 * (The default weight is 1024 - so there's no practical
305 * limitation from this.)
307 #define MIN_SHARES (1UL << 1)
308 #define MAX_SHARES (1UL << 18)
311 typedef int (*tg_visitor)(struct task_group *, void *);
313 extern int walk_tg_tree_from(struct task_group *from,
314 tg_visitor down, tg_visitor up, void *data);
317 * Iterate the full tree, calling @down when first entering a node and @up when
318 * leaving it for the final time.
320 * Caller must hold rcu_lock or sufficient equivalent.
322 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
324 return walk_tg_tree_from(&root_task_group, down, up, data);
327 extern int tg_nop(struct task_group *tg, void *data);
329 extern void free_fair_sched_group(struct task_group *tg);
330 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
331 extern void online_fair_sched_group(struct task_group *tg);
332 extern void unregister_fair_sched_group(struct task_group *tg);
333 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
334 struct sched_entity *se, int cpu,
335 struct sched_entity *parent);
336 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
338 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
339 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
340 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
342 extern void free_rt_sched_group(struct task_group *tg);
343 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
344 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
345 struct sched_rt_entity *rt_se, int cpu,
346 struct sched_rt_entity *parent);
348 extern struct task_group *sched_create_group(struct task_group *parent);
349 extern void sched_online_group(struct task_group *tg,
350 struct task_group *parent);
351 extern void sched_destroy_group(struct task_group *tg);
352 extern void sched_offline_group(struct task_group *tg);
354 extern void sched_move_task(struct task_struct *tsk);
356 #ifdef CONFIG_FAIR_GROUP_SCHED
357 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
360 extern void set_task_rq_fair(struct sched_entity *se,
361 struct cfs_rq *prev, struct cfs_rq *next);
362 #else /* !CONFIG_SMP */
363 static inline void set_task_rq_fair(struct sched_entity *se,
364 struct cfs_rq *prev, struct cfs_rq *next) { }
365 #endif /* CONFIG_SMP */
366 #endif /* CONFIG_FAIR_GROUP_SCHED */
368 #else /* CONFIG_CGROUP_SCHED */
370 struct cfs_bandwidth { };
372 #endif /* CONFIG_CGROUP_SCHED */
374 /* CFS-related fields in a runqueue */
376 struct load_weight load;
377 unsigned int nr_running, h_nr_running;
382 u64 min_vruntime_copy;
385 struct rb_root tasks_timeline;
386 struct rb_node *rb_leftmost;
389 * 'curr' points to currently running entity on this cfs_rq.
390 * It is set to NULL otherwise (i.e when none are currently running).
392 struct sched_entity *curr, *next, *last, *skip;
394 #ifdef CONFIG_SCHED_DEBUG
395 unsigned int nr_spread_over;
402 struct sched_avg avg;
403 u64 runnable_load_sum;
404 unsigned long runnable_load_avg;
405 #ifdef CONFIG_FAIR_GROUP_SCHED
406 unsigned long tg_load_avg_contrib;
408 atomic_long_t removed_load_avg, removed_util_avg;
410 u64 load_last_update_time_copy;
413 #ifdef CONFIG_FAIR_GROUP_SCHED
415 * h_load = weight * f(tg)
417 * Where f(tg) is the recursive weight fraction assigned to
420 unsigned long h_load;
421 u64 last_h_load_update;
422 struct sched_entity *h_load_next;
423 #endif /* CONFIG_FAIR_GROUP_SCHED */
424 #endif /* CONFIG_SMP */
426 #ifdef CONFIG_FAIR_GROUP_SCHED
427 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
430 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
431 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
432 * (like users, containers etc.)
434 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
435 * list is used during load balance.
438 struct list_head leaf_cfs_rq_list;
439 struct task_group *tg; /* group that "owns" this runqueue */
441 #ifdef CONFIG_CFS_BANDWIDTH
444 s64 runtime_remaining;
446 u64 throttled_clock, throttled_clock_task;
447 u64 throttled_clock_task_time;
448 int throttled, throttle_count;
449 struct list_head throttled_list;
450 #endif /* CONFIG_CFS_BANDWIDTH */
451 #endif /* CONFIG_FAIR_GROUP_SCHED */
454 static inline int rt_bandwidth_enabled(void)
456 return sysctl_sched_rt_runtime >= 0;
459 /* RT IPI pull logic requires IRQ_WORK */
460 #ifdef CONFIG_IRQ_WORK
461 # define HAVE_RT_PUSH_IPI
464 /* Real-Time classes' related field in a runqueue: */
466 struct rt_prio_array active;
467 unsigned int rt_nr_running;
468 unsigned int rr_nr_running;
469 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
471 int curr; /* highest queued rt task prio */
473 int next; /* next highest */
478 unsigned long rt_nr_migratory;
479 unsigned long rt_nr_total;
481 struct plist_head pushable_tasks;
482 #ifdef HAVE_RT_PUSH_IPI
485 struct irq_work push_work;
486 raw_spinlock_t push_lock;
488 #endif /* CONFIG_SMP */
494 /* Nests inside the rq lock: */
495 raw_spinlock_t rt_runtime_lock;
497 #ifdef CONFIG_RT_GROUP_SCHED
498 unsigned long rt_nr_boosted;
501 struct task_group *tg;
505 /* Deadline class' related fields in a runqueue */
507 /* runqueue is an rbtree, ordered by deadline */
508 struct rb_root rb_root;
509 struct rb_node *rb_leftmost;
511 unsigned long dl_nr_running;
515 * Deadline values of the currently executing and the
516 * earliest ready task on this rq. Caching these facilitates
517 * the decision wether or not a ready but not running task
518 * should migrate somewhere else.
525 unsigned long dl_nr_migratory;
529 * Tasks on this rq that can be pushed away. They are kept in
530 * an rb-tree, ordered by tasks' deadlines, with caching
531 * of the leftmost (earliest deadline) element.
533 struct rb_root pushable_dl_tasks_root;
534 struct rb_node *pushable_dl_tasks_leftmost;
543 * We add the notion of a root-domain which will be used to define per-domain
544 * variables. Each exclusive cpuset essentially defines an island domain by
545 * fully partitioning the member cpus from any other cpuset. Whenever a new
546 * exclusive cpuset is created, we also create and attach a new root-domain
555 cpumask_var_t online;
557 /* Indicate more than one runnable task for any CPU */
561 * The bit corresponding to a CPU gets set here if such CPU has more
562 * than one runnable -deadline task (as it is below for RT tasks).
564 cpumask_var_t dlo_mask;
570 * The "RT overload" flag: it gets set if a CPU has more than
571 * one runnable RT task.
573 cpumask_var_t rto_mask;
574 struct cpupri cpupri;
576 unsigned long max_cpu_capacity;
579 extern struct root_domain def_root_domain;
581 #endif /* CONFIG_SMP */
584 * This is the main, per-CPU runqueue data structure.
586 * Locking rule: those places that want to lock multiple runqueues
587 * (such as the load balancing or the thread migration code), lock
588 * acquire operations must be ordered by ascending &runqueue.
595 * nr_running and cpu_load should be in the same cacheline because
596 * remote CPUs use both these fields when doing load calculation.
598 unsigned int nr_running;
599 #ifdef CONFIG_NUMA_BALANCING
600 unsigned int nr_numa_running;
601 unsigned int nr_preferred_running;
603 #define CPU_LOAD_IDX_MAX 5
604 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
605 #ifdef CONFIG_NO_HZ_COMMON
607 unsigned long last_load_update_tick;
608 #endif /* CONFIG_SMP */
609 unsigned long nohz_flags;
610 #endif /* CONFIG_NO_HZ_COMMON */
611 #ifdef CONFIG_NO_HZ_FULL
612 unsigned long last_sched_tick;
614 /* capture load from *all* tasks on this cpu: */
615 struct load_weight load;
616 unsigned long nr_load_updates;
623 #ifdef CONFIG_FAIR_GROUP_SCHED
624 /* list of leaf cfs_rq on this cpu: */
625 struct list_head leaf_cfs_rq_list;
626 #endif /* CONFIG_FAIR_GROUP_SCHED */
629 * This is part of a global counter where only the total sum
630 * over all CPUs matters. A task can increase this counter on
631 * one CPU and if it got migrated afterwards it may decrease
632 * it on another CPU. Always updated under the runqueue lock:
634 unsigned long nr_uninterruptible;
636 struct task_struct *curr, *idle, *stop;
637 unsigned long next_balance;
638 struct mm_struct *prev_mm;
640 unsigned int clock_skip_update;
647 struct root_domain *rd;
648 struct sched_domain *sd;
650 unsigned long cpu_capacity;
651 unsigned long cpu_capacity_orig;
653 struct callback_head *balance_callback;
655 unsigned char idle_balance;
656 /* For active balancing */
659 struct cpu_stop_work active_balance_work;
660 /* cpu of this runqueue: */
664 struct list_head cfs_tasks;
671 /* This is used to determine avg_idle's max value */
672 u64 max_idle_balance_cost;
675 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
678 #ifdef CONFIG_PARAVIRT
681 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
682 u64 prev_steal_time_rq;
685 /* calc_load related fields */
686 unsigned long calc_load_update;
687 long calc_load_active;
689 #ifdef CONFIG_SCHED_HRTICK
691 int hrtick_csd_pending;
692 struct call_single_data hrtick_csd;
694 struct hrtimer hrtick_timer;
697 #ifdef CONFIG_SCHEDSTATS
699 struct sched_info rq_sched_info;
700 unsigned long long rq_cpu_time;
701 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
703 /* sys_sched_yield() stats */
704 unsigned int yld_count;
706 /* schedule() stats */
707 unsigned int sched_count;
708 unsigned int sched_goidle;
710 /* try_to_wake_up() stats */
711 unsigned int ttwu_count;
712 unsigned int ttwu_local;
716 struct llist_head wake_list;
719 #ifdef CONFIG_CPU_IDLE
720 /* Must be inspected within a rcu lock section */
721 struct cpuidle_state *idle_state;
725 static inline int cpu_of(struct rq *rq)
735 #ifdef CONFIG_SCHED_SMT
737 extern struct static_key_false sched_smt_present;
739 extern void __update_idle_core(struct rq *rq);
741 static inline void update_idle_core(struct rq *rq)
743 if (static_branch_unlikely(&sched_smt_present))
744 __update_idle_core(rq);
748 static inline void update_idle_core(struct rq *rq) { }
751 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
753 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
754 #define this_rq() this_cpu_ptr(&runqueues)
755 #define task_rq(p) cpu_rq(task_cpu(p))
756 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
757 #define raw_rq() raw_cpu_ptr(&runqueues)
759 static inline u64 __rq_clock_broken(struct rq *rq)
761 return READ_ONCE(rq->clock);
764 static inline u64 rq_clock(struct rq *rq)
766 lockdep_assert_held(&rq->lock);
770 static inline u64 rq_clock_task(struct rq *rq)
772 lockdep_assert_held(&rq->lock);
773 return rq->clock_task;
776 #define RQCF_REQ_SKIP 0x01
777 #define RQCF_ACT_SKIP 0x02
779 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
781 lockdep_assert_held(&rq->lock);
783 rq->clock_skip_update |= RQCF_REQ_SKIP;
785 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
789 enum numa_topology_type {
794 extern enum numa_topology_type sched_numa_topology_type;
795 extern int sched_max_numa_distance;
796 extern bool find_numa_distance(int distance);
799 #ifdef CONFIG_NUMA_BALANCING
800 /* The regions in numa_faults array from task_struct */
801 enum numa_faults_stats {
807 extern void sched_setnuma(struct task_struct *p, int node);
808 extern int migrate_task_to(struct task_struct *p, int cpu);
809 extern int migrate_swap(struct task_struct *, struct task_struct *);
810 #endif /* CONFIG_NUMA_BALANCING */
815 queue_balance_callback(struct rq *rq,
816 struct callback_head *head,
817 void (*func)(struct rq *rq))
819 lockdep_assert_held(&rq->lock);
821 if (unlikely(head->next))
824 head->func = (void (*)(struct callback_head *))func;
825 head->next = rq->balance_callback;
826 rq->balance_callback = head;
829 extern void sched_ttwu_pending(void);
831 #define rcu_dereference_check_sched_domain(p) \
832 rcu_dereference_check((p), \
833 lockdep_is_held(&sched_domains_mutex))
836 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
837 * See detach_destroy_domains: synchronize_sched for details.
839 * The domain tree of any CPU may only be accessed from within
840 * preempt-disabled sections.
842 #define for_each_domain(cpu, __sd) \
843 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
844 __sd; __sd = __sd->parent)
846 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
849 * highest_flag_domain - Return highest sched_domain containing flag.
850 * @cpu: The cpu whose highest level of sched domain is to
852 * @flag: The flag to check for the highest sched_domain
855 * Returns the highest sched_domain of a cpu which contains the given flag.
857 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
859 struct sched_domain *sd, *hsd = NULL;
861 for_each_domain(cpu, sd) {
862 if (!(sd->flags & flag))
870 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
872 struct sched_domain *sd;
874 for_each_domain(cpu, sd) {
875 if (sd->flags & flag)
882 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
883 DECLARE_PER_CPU(int, sd_llc_size);
884 DECLARE_PER_CPU(int, sd_llc_id);
885 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
886 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
887 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
889 struct sched_group_capacity {
892 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
895 unsigned int capacity;
896 unsigned long next_update;
897 int imbalance; /* XXX unrelated to capacity but shared group state */
899 unsigned long cpumask[0]; /* iteration mask */
903 struct sched_group *next; /* Must be a circular list */
906 unsigned int group_weight;
907 struct sched_group_capacity *sgc;
910 * The CPUs this group covers.
912 * NOTE: this field is variable length. (Allocated dynamically
913 * by attaching extra space to the end of the structure,
914 * depending on how many CPUs the kernel has booted up with)
916 unsigned long cpumask[0];
919 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
921 return to_cpumask(sg->cpumask);
925 * cpumask masking which cpus in the group are allowed to iterate up the domain
928 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
930 return to_cpumask(sg->sgc->cpumask);
934 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
935 * @group: The group whose first cpu is to be returned.
937 static inline unsigned int group_first_cpu(struct sched_group *group)
939 return cpumask_first(sched_group_cpus(group));
942 extern int group_balance_cpu(struct sched_group *sg);
944 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
945 void register_sched_domain_sysctl(void);
946 void unregister_sched_domain_sysctl(void);
948 static inline void register_sched_domain_sysctl(void)
951 static inline void unregister_sched_domain_sysctl(void)
958 static inline void sched_ttwu_pending(void) { }
960 #endif /* CONFIG_SMP */
963 #include "auto_group.h"
965 #ifdef CONFIG_CGROUP_SCHED
968 * Return the group to which this tasks belongs.
970 * We cannot use task_css() and friends because the cgroup subsystem
971 * changes that value before the cgroup_subsys::attach() method is called,
972 * therefore we cannot pin it and might observe the wrong value.
974 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
975 * core changes this before calling sched_move_task().
977 * Instead we use a 'copy' which is updated from sched_move_task() while
978 * holding both task_struct::pi_lock and rq::lock.
980 static inline struct task_group *task_group(struct task_struct *p)
982 return p->sched_task_group;
985 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
986 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
988 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
989 struct task_group *tg = task_group(p);
992 #ifdef CONFIG_FAIR_GROUP_SCHED
993 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
994 p->se.cfs_rq = tg->cfs_rq[cpu];
995 p->se.parent = tg->se[cpu];
998 #ifdef CONFIG_RT_GROUP_SCHED
999 p->rt.rt_rq = tg->rt_rq[cpu];
1000 p->rt.parent = tg->rt_se[cpu];
1004 #else /* CONFIG_CGROUP_SCHED */
1006 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1007 static inline struct task_group *task_group(struct task_struct *p)
1012 #endif /* CONFIG_CGROUP_SCHED */
1014 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1016 set_task_rq(p, cpu);
1019 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1020 * successfuly executed on another CPU. We must ensure that updates of
1021 * per-task data have been completed by this moment.
1024 task_thread_info(p)->cpu = cpu;
1030 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1032 #ifdef CONFIG_SCHED_DEBUG
1033 # include <linux/static_key.h>
1034 # define const_debug __read_mostly
1036 # define const_debug const
1039 extern const_debug unsigned int sysctl_sched_features;
1041 #define SCHED_FEAT(name, enabled) \
1042 __SCHED_FEAT_##name ,
1045 #include "features.h"
1051 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1052 #define SCHED_FEAT(name, enabled) \
1053 static __always_inline bool static_branch_##name(struct static_key *key) \
1055 return static_key_##enabled(key); \
1058 #include "features.h"
1062 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1063 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1064 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1065 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1066 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1068 extern struct static_key_false sched_numa_balancing;
1069 extern struct static_key_false sched_schedstats;
1071 static inline u64 global_rt_period(void)
1073 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1076 static inline u64 global_rt_runtime(void)
1078 if (sysctl_sched_rt_runtime < 0)
1081 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1084 static inline int task_current(struct rq *rq, struct task_struct *p)
1086 return rq->curr == p;
1089 static inline int task_running(struct rq *rq, struct task_struct *p)
1094 return task_current(rq, p);
1098 static inline int task_on_rq_queued(struct task_struct *p)
1100 return p->on_rq == TASK_ON_RQ_QUEUED;
1103 static inline int task_on_rq_migrating(struct task_struct *p)
1105 return p->on_rq == TASK_ON_RQ_MIGRATING;
1108 #ifndef prepare_arch_switch
1109 # define prepare_arch_switch(next) do { } while (0)
1111 #ifndef finish_arch_post_lock_switch
1112 # define finish_arch_post_lock_switch() do { } while (0)
1115 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1119 * We can optimise this out completely for !SMP, because the
1120 * SMP rebalancing from interrupt is the only thing that cares
1127 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1131 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1132 * We must ensure this doesn't happen until the switch is completely
1135 * In particular, the load of prev->state in finish_task_switch() must
1136 * happen before this.
1138 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1140 smp_store_release(&prev->on_cpu, 0);
1142 #ifdef CONFIG_DEBUG_SPINLOCK
1143 /* this is a valid case when another task releases the spinlock */
1144 rq->lock.owner = current;
1147 * If we are tracking spinlock dependencies then we have to
1148 * fix up the runqueue lock - which gets 'carried over' from
1149 * prev into current:
1151 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1153 raw_spin_unlock_irq(&rq->lock);
1159 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1160 #define WF_FORK 0x02 /* child wakeup after fork */
1161 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1164 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1165 * of tasks with abnormal "nice" values across CPUs the contribution that
1166 * each task makes to its run queue's load is weighted according to its
1167 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1168 * scaled version of the new time slice allocation that they receive on time
1172 #define WEIGHT_IDLEPRIO 3
1173 #define WMULT_IDLEPRIO 1431655765
1175 extern const int sched_prio_to_weight[40];
1176 extern const u32 sched_prio_to_wmult[40];
1179 * {de,en}queue flags:
1181 * DEQUEUE_SLEEP - task is no longer runnable
1182 * ENQUEUE_WAKEUP - task just became runnable
1184 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1185 * are in a known state which allows modification. Such pairs
1186 * should preserve as much state as possible.
1188 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1191 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1192 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1193 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1197 #define DEQUEUE_SLEEP 0x01
1198 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1199 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1201 #define ENQUEUE_WAKEUP 0x01
1202 #define ENQUEUE_RESTORE 0x02
1203 #define ENQUEUE_MOVE 0x04
1205 #define ENQUEUE_HEAD 0x08
1206 #define ENQUEUE_REPLENISH 0x10
1208 #define ENQUEUE_MIGRATED 0x20
1210 #define ENQUEUE_MIGRATED 0x00
1213 #define RETRY_TASK ((void *)-1UL)
1215 struct sched_class {
1216 const struct sched_class *next;
1218 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1219 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1220 void (*yield_task) (struct rq *rq);
1221 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1223 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1226 * It is the responsibility of the pick_next_task() method that will
1227 * return the next task to call put_prev_task() on the @prev task or
1228 * something equivalent.
1230 * May return RETRY_TASK when it finds a higher prio class has runnable
1233 struct task_struct * (*pick_next_task) (struct rq *rq,
1234 struct task_struct *prev,
1235 struct pin_cookie cookie);
1236 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1239 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1240 void (*migrate_task_rq)(struct task_struct *p);
1242 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1244 void (*set_cpus_allowed)(struct task_struct *p,
1245 const struct cpumask *newmask);
1247 void (*rq_online)(struct rq *rq);
1248 void (*rq_offline)(struct rq *rq);
1251 void (*set_curr_task) (struct rq *rq);
1252 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1253 void (*task_fork) (struct task_struct *p);
1254 void (*task_dead) (struct task_struct *p);
1257 * The switched_from() call is allowed to drop rq->lock, therefore we
1258 * cannot assume the switched_from/switched_to pair is serliazed by
1259 * rq->lock. They are however serialized by p->pi_lock.
1261 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1262 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1263 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1266 unsigned int (*get_rr_interval) (struct rq *rq,
1267 struct task_struct *task);
1269 void (*update_curr) (struct rq *rq);
1271 #define TASK_SET_GROUP 0
1272 #define TASK_MOVE_GROUP 1
1274 #ifdef CONFIG_FAIR_GROUP_SCHED
1275 void (*task_change_group) (struct task_struct *p, int type);
1279 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1281 prev->sched_class->put_prev_task(rq, prev);
1284 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1286 curr->sched_class->set_curr_task(rq);
1289 #define sched_class_highest (&stop_sched_class)
1290 #define for_each_class(class) \
1291 for (class = sched_class_highest; class; class = class->next)
1293 extern const struct sched_class stop_sched_class;
1294 extern const struct sched_class dl_sched_class;
1295 extern const struct sched_class rt_sched_class;
1296 extern const struct sched_class fair_sched_class;
1297 extern const struct sched_class idle_sched_class;
1302 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1304 extern void trigger_load_balance(struct rq *rq);
1306 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1310 #ifdef CONFIG_CPU_IDLE
1311 static inline void idle_set_state(struct rq *rq,
1312 struct cpuidle_state *idle_state)
1314 rq->idle_state = idle_state;
1317 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1319 SCHED_WARN_ON(!rcu_read_lock_held());
1320 return rq->idle_state;
1323 static inline void idle_set_state(struct rq *rq,
1324 struct cpuidle_state *idle_state)
1328 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1334 extern void sysrq_sched_debug_show(void);
1335 extern void sched_init_granularity(void);
1336 extern void update_max_interval(void);
1338 extern void init_sched_dl_class(void);
1339 extern void init_sched_rt_class(void);
1340 extern void init_sched_fair_class(void);
1342 extern void resched_curr(struct rq *rq);
1343 extern void resched_cpu(int cpu);
1345 extern struct rt_bandwidth def_rt_bandwidth;
1346 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1348 extern struct dl_bandwidth def_dl_bandwidth;
1349 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1350 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1352 unsigned long to_ratio(u64 period, u64 runtime);
1354 extern void init_entity_runnable_average(struct sched_entity *se);
1355 extern void post_init_entity_util_avg(struct sched_entity *se);
1357 #ifdef CONFIG_NO_HZ_FULL
1358 extern bool sched_can_stop_tick(struct rq *rq);
1361 * Tick may be needed by tasks in the runqueue depending on their policy and
1362 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1363 * nohz mode if necessary.
1365 static inline void sched_update_tick_dependency(struct rq *rq)
1369 if (!tick_nohz_full_enabled())
1374 if (!tick_nohz_full_cpu(cpu))
1377 if (sched_can_stop_tick(rq))
1378 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1380 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1383 static inline void sched_update_tick_dependency(struct rq *rq) { }
1386 static inline void add_nr_running(struct rq *rq, unsigned count)
1388 unsigned prev_nr = rq->nr_running;
1390 rq->nr_running = prev_nr + count;
1392 if (prev_nr < 2 && rq->nr_running >= 2) {
1394 if (!rq->rd->overload)
1395 rq->rd->overload = true;
1399 sched_update_tick_dependency(rq);
1402 static inline void sub_nr_running(struct rq *rq, unsigned count)
1404 rq->nr_running -= count;
1405 /* Check if we still need preemption */
1406 sched_update_tick_dependency(rq);
1409 static inline void rq_last_tick_reset(struct rq *rq)
1411 #ifdef CONFIG_NO_HZ_FULL
1412 rq->last_sched_tick = jiffies;
1416 extern void update_rq_clock(struct rq *rq);
1418 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1419 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1421 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1423 extern const_debug unsigned int sysctl_sched_time_avg;
1424 extern const_debug unsigned int sysctl_sched_nr_migrate;
1425 extern const_debug unsigned int sysctl_sched_migration_cost;
1427 static inline u64 sched_avg_period(void)
1429 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1432 #ifdef CONFIG_SCHED_HRTICK
1436 * - enabled by features
1437 * - hrtimer is actually high res
1439 static inline int hrtick_enabled(struct rq *rq)
1441 if (!sched_feat(HRTICK))
1443 if (!cpu_active(cpu_of(rq)))
1445 return hrtimer_is_hres_active(&rq->hrtick_timer);
1448 void hrtick_start(struct rq *rq, u64 delay);
1452 static inline int hrtick_enabled(struct rq *rq)
1457 #endif /* CONFIG_SCHED_HRTICK */
1460 extern void sched_avg_update(struct rq *rq);
1462 #ifndef arch_scale_freq_capacity
1463 static __always_inline
1464 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1466 return SCHED_CAPACITY_SCALE;
1470 #ifndef arch_scale_cpu_capacity
1471 static __always_inline
1472 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1474 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1475 return sd->smt_gain / sd->span_weight;
1477 return SCHED_CAPACITY_SCALE;
1481 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1483 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1484 sched_avg_update(rq);
1487 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1488 static inline void sched_avg_update(struct rq *rq) { }
1492 unsigned long flags;
1493 struct pin_cookie cookie;
1496 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1497 __acquires(rq->lock);
1498 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1499 __acquires(p->pi_lock)
1500 __acquires(rq->lock);
1502 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1503 __releases(rq->lock)
1505 lockdep_unpin_lock(&rq->lock, rf->cookie);
1506 raw_spin_unlock(&rq->lock);
1510 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1511 __releases(rq->lock)
1512 __releases(p->pi_lock)
1514 lockdep_unpin_lock(&rq->lock, rf->cookie);
1515 raw_spin_unlock(&rq->lock);
1516 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1520 #ifdef CONFIG_PREEMPT
1522 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1525 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1526 * way at the expense of forcing extra atomic operations in all
1527 * invocations. This assures that the double_lock is acquired using the
1528 * same underlying policy as the spinlock_t on this architecture, which
1529 * reduces latency compared to the unfair variant below. However, it
1530 * also adds more overhead and therefore may reduce throughput.
1532 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1533 __releases(this_rq->lock)
1534 __acquires(busiest->lock)
1535 __acquires(this_rq->lock)
1537 raw_spin_unlock(&this_rq->lock);
1538 double_rq_lock(this_rq, busiest);
1545 * Unfair double_lock_balance: Optimizes throughput at the expense of
1546 * latency by eliminating extra atomic operations when the locks are
1547 * already in proper order on entry. This favors lower cpu-ids and will
1548 * grant the double lock to lower cpus over higher ids under contention,
1549 * regardless of entry order into the function.
1551 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1552 __releases(this_rq->lock)
1553 __acquires(busiest->lock)
1554 __acquires(this_rq->lock)
1558 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1559 if (busiest < this_rq) {
1560 raw_spin_unlock(&this_rq->lock);
1561 raw_spin_lock(&busiest->lock);
1562 raw_spin_lock_nested(&this_rq->lock,
1563 SINGLE_DEPTH_NESTING);
1566 raw_spin_lock_nested(&busiest->lock,
1567 SINGLE_DEPTH_NESTING);
1572 #endif /* CONFIG_PREEMPT */
1575 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1577 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1579 if (unlikely(!irqs_disabled())) {
1580 /* printk() doesn't work good under rq->lock */
1581 raw_spin_unlock(&this_rq->lock);
1585 return _double_lock_balance(this_rq, busiest);
1588 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1589 __releases(busiest->lock)
1591 raw_spin_unlock(&busiest->lock);
1592 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1595 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1601 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1604 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1610 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1613 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1619 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1623 * double_rq_lock - safely lock two runqueues
1625 * Note this does not disable interrupts like task_rq_lock,
1626 * you need to do so manually before calling.
1628 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1629 __acquires(rq1->lock)
1630 __acquires(rq2->lock)
1632 BUG_ON(!irqs_disabled());
1634 raw_spin_lock(&rq1->lock);
1635 __acquire(rq2->lock); /* Fake it out ;) */
1638 raw_spin_lock(&rq1->lock);
1639 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1641 raw_spin_lock(&rq2->lock);
1642 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1648 * double_rq_unlock - safely unlock two runqueues
1650 * Note this does not restore interrupts like task_rq_unlock,
1651 * you need to do so manually after calling.
1653 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1654 __releases(rq1->lock)
1655 __releases(rq2->lock)
1657 raw_spin_unlock(&rq1->lock);
1659 raw_spin_unlock(&rq2->lock);
1661 __release(rq2->lock);
1664 #else /* CONFIG_SMP */
1667 * double_rq_lock - safely lock two runqueues
1669 * Note this does not disable interrupts like task_rq_lock,
1670 * you need to do so manually before calling.
1672 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1673 __acquires(rq1->lock)
1674 __acquires(rq2->lock)
1676 BUG_ON(!irqs_disabled());
1678 raw_spin_lock(&rq1->lock);
1679 __acquire(rq2->lock); /* Fake it out ;) */
1683 * double_rq_unlock - safely unlock two runqueues
1685 * Note this does not restore interrupts like task_rq_unlock,
1686 * you need to do so manually after calling.
1688 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1689 __releases(rq1->lock)
1690 __releases(rq2->lock)
1693 raw_spin_unlock(&rq1->lock);
1694 __release(rq2->lock);
1699 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1700 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1702 #ifdef CONFIG_SCHED_DEBUG
1703 extern void print_cfs_stats(struct seq_file *m, int cpu);
1704 extern void print_rt_stats(struct seq_file *m, int cpu);
1705 extern void print_dl_stats(struct seq_file *m, int cpu);
1707 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1709 #ifdef CONFIG_NUMA_BALANCING
1711 show_numa_stats(struct task_struct *p, struct seq_file *m);
1713 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1714 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1715 #endif /* CONFIG_NUMA_BALANCING */
1716 #endif /* CONFIG_SCHED_DEBUG */
1718 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1719 extern void init_rt_rq(struct rt_rq *rt_rq);
1720 extern void init_dl_rq(struct dl_rq *dl_rq);
1722 extern void cfs_bandwidth_usage_inc(void);
1723 extern void cfs_bandwidth_usage_dec(void);
1725 #ifdef CONFIG_NO_HZ_COMMON
1726 enum rq_nohz_flag_bits {
1731 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1733 extern void nohz_balance_exit_idle(unsigned int cpu);
1735 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1738 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1743 struct u64_stats_sync sync;
1746 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
1748 static inline u64 irq_time_read(int cpu)
1750 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
1755 seq = __u64_stats_fetch_begin(&irqtime->sync);
1756 total = irqtime->softirq_time + irqtime->hardirq_time;
1757 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
1761 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1763 #ifdef CONFIG_CPU_FREQ
1764 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1767 * cpufreq_update_util - Take a note about CPU utilization changes.
1768 * @time: Current time.
1769 * @util: Current utilization.
1770 * @max: Utilization ceiling.
1772 * This function is called by the scheduler on every invocation of
1773 * update_load_avg() on the CPU whose utilization is being updated.
1775 * It can only be called from RCU-sched read-side critical sections.
1777 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max)
1779 struct update_util_data *data;
1781 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1783 data->func(data, time, util, max);
1787 * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed.
1788 * @time: Current time.
1790 * The way cpufreq is currently arranged requires it to evaluate the CPU
1791 * performance state (frequency/voltage) on a regular basis to prevent it from
1792 * being stuck in a completely inadequate performance level for too long.
1793 * That is not guaranteed to happen if the updates are only triggered from CFS,
1794 * though, because they may not be coming in if RT or deadline tasks are active
1795 * all the time (or there are RT and DL tasks only).
1797 * As a workaround for that issue, this function is called by the RT and DL
1798 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1799 * but that really is a band-aid. Going forward it should be replaced with
1800 * solutions targeted more specifically at RT and DL tasks.
1802 static inline void cpufreq_trigger_update(u64 time)
1804 cpufreq_update_util(time, ULONG_MAX, 0);
1807 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {}
1808 static inline void cpufreq_trigger_update(u64 time) {}
1809 #endif /* CONFIG_CPU_FREQ */
1811 #ifdef arch_scale_freq_capacity
1812 #ifndef arch_scale_freq_invariant
1813 #define arch_scale_freq_invariant() (true)
1815 #else /* arch_scale_freq_capacity */
1816 #define arch_scale_freq_invariant() (false)