2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/sched/deadline.h>
6 #include <linux/binfmts.h>
7 #include <linux/mutex.h>
8 #include <linux/spinlock.h>
9 #include <linux/stop_machine.h>
10 #include <linux/irq_work.h>
11 #include <linux/tick.h>
12 #include <linux/slab.h>
15 #include "cpudeadline.h"
21 /* task_struct::on_rq states: */
22 #define TASK_ON_RQ_QUEUED 1
23 #define TASK_ON_RQ_MIGRATING 2
25 extern __read_mostly int scheduler_running;
27 extern unsigned long calc_load_update;
28 extern atomic_long_t calc_load_tasks;
30 extern void calc_global_load_tick(struct rq *this_rq);
31 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
34 extern void cpu_load_update_active(struct rq *this_rq);
36 static inline void cpu_load_update_active(struct rq *this_rq) { }
39 #ifdef CONFIG_SCHED_SMT
40 extern void update_idle_core(struct rq *rq);
42 static inline void update_idle_core(struct rq *rq) { }
46 * Helpers for converting nanosecond timing to jiffy resolution
48 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
51 * Increase resolution of nice-level calculations for 64-bit architectures.
52 * The extra resolution improves shares distribution and load balancing of
53 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
54 * hierarchies, especially on larger systems. This is not a user-visible change
55 * and does not change the user-interface for setting shares/weights.
57 * We increase resolution only if we have enough bits to allow this increased
58 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
59 * pretty high and the returns do not justify the increased costs.
61 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
62 * increase coverage and consistency always enable it on 64bit platforms.
65 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
66 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
67 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
69 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
70 # define scale_load(w) (w)
71 # define scale_load_down(w) (w)
75 * Task weight (visible to users) and its load (invisible to users) have
76 * independent resolution, but they should be well calibrated. We use
77 * scale_load() and scale_load_down(w) to convert between them. The
78 * following must be true:
80 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
83 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
86 * Single value that decides SCHED_DEADLINE internal math precision.
87 * 10 -> just above 1us
88 * 9 -> just above 0.5us
93 * These are the 'tuning knobs' of the scheduler:
97 * single value that denotes runtime == period, ie unlimited time.
99 #define RUNTIME_INF ((u64)~0ULL)
101 static inline int idle_policy(int policy)
103 return policy == SCHED_IDLE;
105 static inline int fair_policy(int policy)
107 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
110 static inline int rt_policy(int policy)
112 return policy == SCHED_FIFO || policy == SCHED_RR;
115 static inline int dl_policy(int policy)
117 return policy == SCHED_DEADLINE;
119 static inline bool valid_policy(int policy)
121 return idle_policy(policy) || fair_policy(policy) ||
122 rt_policy(policy) || dl_policy(policy);
125 static inline int task_has_rt_policy(struct task_struct *p)
127 return rt_policy(p->policy);
130 static inline int task_has_dl_policy(struct task_struct *p)
132 return dl_policy(p->policy);
136 * Tells if entity @a should preempt entity @b.
139 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
141 return dl_time_before(a->deadline, b->deadline);
145 * This is the priority-queue data structure of the RT scheduling class:
147 struct rt_prio_array {
148 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
149 struct list_head queue[MAX_RT_PRIO];
152 struct rt_bandwidth {
153 /* nests inside the rq lock: */
154 raw_spinlock_t rt_runtime_lock;
157 struct hrtimer rt_period_timer;
158 unsigned int rt_period_active;
161 void __dl_clear_params(struct task_struct *p);
164 * To keep the bandwidth of -deadline tasks and groups under control
165 * we need some place where:
166 * - store the maximum -deadline bandwidth of the system (the group);
167 * - cache the fraction of that bandwidth that is currently allocated.
169 * This is all done in the data structure below. It is similar to the
170 * one used for RT-throttling (rt_bandwidth), with the main difference
171 * that, since here we are only interested in admission control, we
172 * do not decrease any runtime while the group "executes", neither we
173 * need a timer to replenish it.
175 * With respect to SMP, the bandwidth is given on a per-CPU basis,
177 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
178 * - dl_total_bw array contains, in the i-eth element, the currently
179 * allocated bandwidth on the i-eth CPU.
180 * Moreover, groups consume bandwidth on each CPU, while tasks only
181 * consume bandwidth on the CPU they're running on.
182 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
183 * that will be shown the next time the proc or cgroup controls will
184 * be red. It on its turn can be changed by writing on its own
187 struct dl_bandwidth {
188 raw_spinlock_t dl_runtime_lock;
193 static inline int dl_bandwidth_enabled(void)
195 return sysctl_sched_rt_runtime >= 0;
198 extern struct dl_bw *dl_bw_of(int i);
206 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
208 dl_b->total_bw -= tsk_bw;
212 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
214 dl_b->total_bw += tsk_bw;
218 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
220 return dl_b->bw != -1 &&
221 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
224 extern struct mutex sched_domains_mutex;
226 #ifdef CONFIG_CGROUP_SCHED
228 #include <linux/cgroup.h>
233 extern struct list_head task_groups;
235 struct cfs_bandwidth {
236 #ifdef CONFIG_CFS_BANDWIDTH
240 s64 hierarchical_quota;
243 int idle, period_active;
244 struct hrtimer period_timer, slack_timer;
245 struct list_head throttled_cfs_rq;
248 int nr_periods, nr_throttled;
253 /* task group related information */
255 struct cgroup_subsys_state css;
257 #ifdef CONFIG_FAIR_GROUP_SCHED
258 /* schedulable entities of this group on each cpu */
259 struct sched_entity **se;
260 /* runqueue "owned" by this group on each cpu */
261 struct cfs_rq **cfs_rq;
262 unsigned long shares;
266 * load_avg can be heavily contended at clock tick time, so put
267 * it in its own cacheline separated from the fields above which
268 * will also be accessed at each tick.
270 atomic_long_t load_avg ____cacheline_aligned;
274 #ifdef CONFIG_RT_GROUP_SCHED
275 struct sched_rt_entity **rt_se;
276 struct rt_rq **rt_rq;
278 struct rt_bandwidth rt_bandwidth;
282 struct list_head list;
284 struct task_group *parent;
285 struct list_head siblings;
286 struct list_head children;
288 #ifdef CONFIG_SCHED_AUTOGROUP
289 struct autogroup *autogroup;
292 struct cfs_bandwidth cfs_bandwidth;
295 #ifdef CONFIG_FAIR_GROUP_SCHED
296 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
299 * A weight of 0 or 1 can cause arithmetics problems.
300 * A weight of a cfs_rq is the sum of weights of which entities
301 * are queued on this cfs_rq, so a weight of a entity should not be
302 * too large, so as the shares value of a task group.
303 * (The default weight is 1024 - so there's no practical
304 * limitation from this.)
306 #define MIN_SHARES (1UL << 1)
307 #define MAX_SHARES (1UL << 18)
310 typedef int (*tg_visitor)(struct task_group *, void *);
312 extern int walk_tg_tree_from(struct task_group *from,
313 tg_visitor down, tg_visitor up, void *data);
316 * Iterate the full tree, calling @down when first entering a node and @up when
317 * leaving it for the final time.
319 * Caller must hold rcu_lock or sufficient equivalent.
321 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
323 return walk_tg_tree_from(&root_task_group, down, up, data);
326 extern int tg_nop(struct task_group *tg, void *data);
328 extern void free_fair_sched_group(struct task_group *tg);
329 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
330 extern void online_fair_sched_group(struct task_group *tg);
331 extern void unregister_fair_sched_group(struct task_group *tg);
332 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
333 struct sched_entity *se, int cpu,
334 struct sched_entity *parent);
335 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
337 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
338 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
339 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
341 extern void free_rt_sched_group(struct task_group *tg);
342 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
343 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
344 struct sched_rt_entity *rt_se, int cpu,
345 struct sched_rt_entity *parent);
347 extern struct task_group *sched_create_group(struct task_group *parent);
348 extern void sched_online_group(struct task_group *tg,
349 struct task_group *parent);
350 extern void sched_destroy_group(struct task_group *tg);
351 extern void sched_offline_group(struct task_group *tg);
353 extern void sched_move_task(struct task_struct *tsk);
355 #ifdef CONFIG_FAIR_GROUP_SCHED
356 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
359 extern void set_task_rq_fair(struct sched_entity *se,
360 struct cfs_rq *prev, struct cfs_rq *next);
361 #else /* !CONFIG_SMP */
362 static inline void set_task_rq_fair(struct sched_entity *se,
363 struct cfs_rq *prev, struct cfs_rq *next) { }
364 #endif /* CONFIG_SMP */
365 #endif /* CONFIG_FAIR_GROUP_SCHED */
367 #else /* CONFIG_CGROUP_SCHED */
369 struct cfs_bandwidth { };
371 #endif /* CONFIG_CGROUP_SCHED */
373 /* CFS-related fields in a runqueue */
375 struct load_weight load;
376 unsigned int nr_running, h_nr_running;
381 u64 min_vruntime_copy;
384 struct rb_root tasks_timeline;
385 struct rb_node *rb_leftmost;
388 * 'curr' points to currently running entity on this cfs_rq.
389 * It is set to NULL otherwise (i.e when none are currently running).
391 struct sched_entity *curr, *next, *last, *skip;
393 #ifdef CONFIG_SCHED_DEBUG
394 unsigned int nr_spread_over;
401 struct sched_avg avg;
402 u64 runnable_load_sum;
403 unsigned long runnable_load_avg;
404 #ifdef CONFIG_FAIR_GROUP_SCHED
405 unsigned long tg_load_avg_contrib;
407 atomic_long_t removed_load_avg, removed_util_avg;
409 u64 load_last_update_time_copy;
412 #ifdef CONFIG_FAIR_GROUP_SCHED
414 * h_load = weight * f(tg)
416 * Where f(tg) is the recursive weight fraction assigned to
419 unsigned long h_load;
420 u64 last_h_load_update;
421 struct sched_entity *h_load_next;
422 #endif /* CONFIG_FAIR_GROUP_SCHED */
423 #endif /* CONFIG_SMP */
425 #ifdef CONFIG_FAIR_GROUP_SCHED
426 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
429 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
430 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
431 * (like users, containers etc.)
433 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
434 * list is used during load balance.
437 struct list_head leaf_cfs_rq_list;
438 struct task_group *tg; /* group that "owns" this runqueue */
440 #ifdef CONFIG_CFS_BANDWIDTH
443 s64 runtime_remaining;
445 u64 throttled_clock, throttled_clock_task;
446 u64 throttled_clock_task_time;
447 int throttled, throttle_count;
448 struct list_head throttled_list;
449 #endif /* CONFIG_CFS_BANDWIDTH */
450 #endif /* CONFIG_FAIR_GROUP_SCHED */
453 static inline int rt_bandwidth_enabled(void)
455 return sysctl_sched_rt_runtime >= 0;
458 /* RT IPI pull logic requires IRQ_WORK */
459 #ifdef CONFIG_IRQ_WORK
460 # define HAVE_RT_PUSH_IPI
463 /* Real-Time classes' related field in a runqueue: */
465 struct rt_prio_array active;
466 unsigned int rt_nr_running;
467 unsigned int rr_nr_running;
468 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
470 int curr; /* highest queued rt task prio */
472 int next; /* next highest */
477 unsigned long rt_nr_migratory;
478 unsigned long rt_nr_total;
480 struct plist_head pushable_tasks;
481 #ifdef HAVE_RT_PUSH_IPI
484 struct irq_work push_work;
485 raw_spinlock_t push_lock;
487 #endif /* CONFIG_SMP */
493 /* Nests inside the rq lock: */
494 raw_spinlock_t rt_runtime_lock;
496 #ifdef CONFIG_RT_GROUP_SCHED
497 unsigned long rt_nr_boosted;
500 struct task_group *tg;
504 /* Deadline class' related fields in a runqueue */
506 /* runqueue is an rbtree, ordered by deadline */
507 struct rb_root rb_root;
508 struct rb_node *rb_leftmost;
510 unsigned long dl_nr_running;
514 * Deadline values of the currently executing and the
515 * earliest ready task on this rq. Caching these facilitates
516 * the decision wether or not a ready but not running task
517 * should migrate somewhere else.
524 unsigned long dl_nr_migratory;
528 * Tasks on this rq that can be pushed away. They are kept in
529 * an rb-tree, ordered by tasks' deadlines, with caching
530 * of the leftmost (earliest deadline) element.
532 struct rb_root pushable_dl_tasks_root;
533 struct rb_node *pushable_dl_tasks_leftmost;
542 * We add the notion of a root-domain which will be used to define per-domain
543 * variables. Each exclusive cpuset essentially defines an island domain by
544 * fully partitioning the member cpus from any other cpuset. Whenever a new
545 * exclusive cpuset is created, we also create and attach a new root-domain
554 cpumask_var_t online;
556 /* Indicate more than one runnable task for any CPU */
560 * The bit corresponding to a CPU gets set here if such CPU has more
561 * than one runnable -deadline task (as it is below for RT tasks).
563 cpumask_var_t dlo_mask;
569 * The "RT overload" flag: it gets set if a CPU has more than
570 * one runnable RT task.
572 cpumask_var_t rto_mask;
573 struct cpupri cpupri;
575 unsigned long max_cpu_capacity;
578 extern struct root_domain def_root_domain;
580 #endif /* CONFIG_SMP */
583 * This is the main, per-CPU runqueue data structure.
585 * Locking rule: those places that want to lock multiple runqueues
586 * (such as the load balancing or the thread migration code), lock
587 * acquire operations must be ordered by ascending &runqueue.
594 * nr_running and cpu_load should be in the same cacheline because
595 * remote CPUs use both these fields when doing load calculation.
597 unsigned int nr_running;
598 #ifdef CONFIG_NUMA_BALANCING
599 unsigned int nr_numa_running;
600 unsigned int nr_preferred_running;
602 #define CPU_LOAD_IDX_MAX 5
603 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
604 #ifdef CONFIG_NO_HZ_COMMON
606 unsigned long last_load_update_tick;
607 #endif /* CONFIG_SMP */
608 unsigned long nohz_flags;
609 #endif /* CONFIG_NO_HZ_COMMON */
610 #ifdef CONFIG_NO_HZ_FULL
611 unsigned long last_sched_tick;
613 /* capture load from *all* tasks on this cpu: */
614 struct load_weight load;
615 unsigned long nr_load_updates;
622 #ifdef CONFIG_FAIR_GROUP_SCHED
623 /* list of leaf cfs_rq on this cpu: */
624 struct list_head leaf_cfs_rq_list;
625 #endif /* CONFIG_FAIR_GROUP_SCHED */
628 * This is part of a global counter where only the total sum
629 * over all CPUs matters. A task can increase this counter on
630 * one CPU and if it got migrated afterwards it may decrease
631 * it on another CPU. Always updated under the runqueue lock:
633 unsigned long nr_uninterruptible;
635 struct task_struct *curr, *idle, *stop;
636 unsigned long next_balance;
637 struct mm_struct *prev_mm;
639 unsigned int clock_skip_update;
646 struct root_domain *rd;
647 struct sched_domain *sd;
649 unsigned long cpu_capacity;
650 unsigned long cpu_capacity_orig;
652 struct callback_head *balance_callback;
654 unsigned char idle_balance;
655 /* For active balancing */
658 struct cpu_stop_work active_balance_work;
659 /* cpu of this runqueue: */
663 struct list_head cfs_tasks;
670 /* This is used to determine avg_idle's max value */
671 u64 max_idle_balance_cost;
674 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
677 #ifdef CONFIG_PARAVIRT
680 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
681 u64 prev_steal_time_rq;
684 /* calc_load related fields */
685 unsigned long calc_load_update;
686 long calc_load_active;
688 #ifdef CONFIG_SCHED_HRTICK
690 int hrtick_csd_pending;
691 struct call_single_data hrtick_csd;
693 struct hrtimer hrtick_timer;
696 #ifdef CONFIG_SCHEDSTATS
698 struct sched_info rq_sched_info;
699 unsigned long long rq_cpu_time;
700 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
702 /* sys_sched_yield() stats */
703 unsigned int yld_count;
705 /* schedule() stats */
706 unsigned int sched_count;
707 unsigned int sched_goidle;
709 /* try_to_wake_up() stats */
710 unsigned int ttwu_count;
711 unsigned int ttwu_local;
715 struct llist_head wake_list;
718 #ifdef CONFIG_CPU_IDLE
719 /* Must be inspected within a rcu lock section */
720 struct cpuidle_state *idle_state;
724 static inline int cpu_of(struct rq *rq)
733 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
735 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
736 #define this_rq() this_cpu_ptr(&runqueues)
737 #define task_rq(p) cpu_rq(task_cpu(p))
738 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
739 #define raw_rq() raw_cpu_ptr(&runqueues)
741 static inline u64 __rq_clock_broken(struct rq *rq)
743 return READ_ONCE(rq->clock);
746 static inline u64 rq_clock(struct rq *rq)
748 lockdep_assert_held(&rq->lock);
752 static inline u64 rq_clock_task(struct rq *rq)
754 lockdep_assert_held(&rq->lock);
755 return rq->clock_task;
758 #define RQCF_REQ_SKIP 0x01
759 #define RQCF_ACT_SKIP 0x02
761 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
763 lockdep_assert_held(&rq->lock);
765 rq->clock_skip_update |= RQCF_REQ_SKIP;
767 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
771 enum numa_topology_type {
776 extern enum numa_topology_type sched_numa_topology_type;
777 extern int sched_max_numa_distance;
778 extern bool find_numa_distance(int distance);
781 #ifdef CONFIG_NUMA_BALANCING
782 /* The regions in numa_faults array from task_struct */
783 enum numa_faults_stats {
789 extern void sched_setnuma(struct task_struct *p, int node);
790 extern int migrate_task_to(struct task_struct *p, int cpu);
791 extern int migrate_swap(struct task_struct *, struct task_struct *);
792 #endif /* CONFIG_NUMA_BALANCING */
797 queue_balance_callback(struct rq *rq,
798 struct callback_head *head,
799 void (*func)(struct rq *rq))
801 lockdep_assert_held(&rq->lock);
803 if (unlikely(head->next))
806 head->func = (void (*)(struct callback_head *))func;
807 head->next = rq->balance_callback;
808 rq->balance_callback = head;
811 extern void sched_ttwu_pending(void);
813 #define rcu_dereference_check_sched_domain(p) \
814 rcu_dereference_check((p), \
815 lockdep_is_held(&sched_domains_mutex))
818 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
819 * See detach_destroy_domains: synchronize_sched for details.
821 * The domain tree of any CPU may only be accessed from within
822 * preempt-disabled sections.
824 #define for_each_domain(cpu, __sd) \
825 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
826 __sd; __sd = __sd->parent)
828 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
831 * highest_flag_domain - Return highest sched_domain containing flag.
832 * @cpu: The cpu whose highest level of sched domain is to
834 * @flag: The flag to check for the highest sched_domain
837 * Returns the highest sched_domain of a cpu which contains the given flag.
839 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
841 struct sched_domain *sd, *hsd = NULL;
843 for_each_domain(cpu, sd) {
844 if (!(sd->flags & flag))
852 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
854 struct sched_domain *sd;
856 for_each_domain(cpu, sd) {
857 if (sd->flags & flag)
864 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
865 DECLARE_PER_CPU(int, sd_llc_size);
866 DECLARE_PER_CPU(int, sd_llc_id);
867 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
868 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
869 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
871 struct sched_group_capacity {
874 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
877 unsigned int capacity;
878 unsigned long next_update;
879 int imbalance; /* XXX unrelated to capacity but shared group state */
881 unsigned long cpumask[0]; /* iteration mask */
885 struct sched_group *next; /* Must be a circular list */
888 unsigned int group_weight;
889 struct sched_group_capacity *sgc;
892 * The CPUs this group covers.
894 * NOTE: this field is variable length. (Allocated dynamically
895 * by attaching extra space to the end of the structure,
896 * depending on how many CPUs the kernel has booted up with)
898 unsigned long cpumask[0];
901 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
903 return to_cpumask(sg->cpumask);
907 * cpumask masking which cpus in the group are allowed to iterate up the domain
910 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
912 return to_cpumask(sg->sgc->cpumask);
916 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
917 * @group: The group whose first cpu is to be returned.
919 static inline unsigned int group_first_cpu(struct sched_group *group)
921 return cpumask_first(sched_group_cpus(group));
924 extern int group_balance_cpu(struct sched_group *sg);
926 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
927 void register_sched_domain_sysctl(void);
928 void unregister_sched_domain_sysctl(void);
930 static inline void register_sched_domain_sysctl(void)
933 static inline void unregister_sched_domain_sysctl(void)
940 static inline void sched_ttwu_pending(void) { }
942 #endif /* CONFIG_SMP */
945 #include "auto_group.h"
947 #ifdef CONFIG_CGROUP_SCHED
950 * Return the group to which this tasks belongs.
952 * We cannot use task_css() and friends because the cgroup subsystem
953 * changes that value before the cgroup_subsys::attach() method is called,
954 * therefore we cannot pin it and might observe the wrong value.
956 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
957 * core changes this before calling sched_move_task().
959 * Instead we use a 'copy' which is updated from sched_move_task() while
960 * holding both task_struct::pi_lock and rq::lock.
962 static inline struct task_group *task_group(struct task_struct *p)
964 return p->sched_task_group;
967 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
968 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
970 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
971 struct task_group *tg = task_group(p);
974 #ifdef CONFIG_FAIR_GROUP_SCHED
975 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
976 p->se.cfs_rq = tg->cfs_rq[cpu];
977 p->se.parent = tg->se[cpu];
980 #ifdef CONFIG_RT_GROUP_SCHED
981 p->rt.rt_rq = tg->rt_rq[cpu];
982 p->rt.parent = tg->rt_se[cpu];
986 #else /* CONFIG_CGROUP_SCHED */
988 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
989 static inline struct task_group *task_group(struct task_struct *p)
994 #endif /* CONFIG_CGROUP_SCHED */
996 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1001 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1002 * successfuly executed on another CPU. We must ensure that updates of
1003 * per-task data have been completed by this moment.
1006 task_thread_info(p)->cpu = cpu;
1012 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1014 #ifdef CONFIG_SCHED_DEBUG
1015 # include <linux/static_key.h>
1016 # define const_debug __read_mostly
1018 # define const_debug const
1021 extern const_debug unsigned int sysctl_sched_features;
1023 #define SCHED_FEAT(name, enabled) \
1024 __SCHED_FEAT_##name ,
1027 #include "features.h"
1033 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1034 #define SCHED_FEAT(name, enabled) \
1035 static __always_inline bool static_branch_##name(struct static_key *key) \
1037 return static_key_##enabled(key); \
1040 #include "features.h"
1044 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1045 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1046 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1047 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1048 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1050 extern struct static_key_false sched_numa_balancing;
1051 extern struct static_key_false sched_schedstats;
1053 static inline u64 global_rt_period(void)
1055 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1058 static inline u64 global_rt_runtime(void)
1060 if (sysctl_sched_rt_runtime < 0)
1063 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1066 static inline int task_current(struct rq *rq, struct task_struct *p)
1068 return rq->curr == p;
1071 static inline int task_running(struct rq *rq, struct task_struct *p)
1076 return task_current(rq, p);
1080 static inline int task_on_rq_queued(struct task_struct *p)
1082 return p->on_rq == TASK_ON_RQ_QUEUED;
1085 static inline int task_on_rq_migrating(struct task_struct *p)
1087 return p->on_rq == TASK_ON_RQ_MIGRATING;
1090 #ifndef prepare_arch_switch
1091 # define prepare_arch_switch(next) do { } while (0)
1093 #ifndef finish_arch_post_lock_switch
1094 # define finish_arch_post_lock_switch() do { } while (0)
1097 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1101 * We can optimise this out completely for !SMP, because the
1102 * SMP rebalancing from interrupt is the only thing that cares
1109 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1113 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1114 * We must ensure this doesn't happen until the switch is completely
1117 * In particular, the load of prev->state in finish_task_switch() must
1118 * happen before this.
1120 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1122 smp_store_release(&prev->on_cpu, 0);
1124 #ifdef CONFIG_DEBUG_SPINLOCK
1125 /* this is a valid case when another task releases the spinlock */
1126 rq->lock.owner = current;
1129 * If we are tracking spinlock dependencies then we have to
1130 * fix up the runqueue lock - which gets 'carried over' from
1131 * prev into current:
1133 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1135 raw_spin_unlock_irq(&rq->lock);
1141 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1142 #define WF_FORK 0x02 /* child wakeup after fork */
1143 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1146 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1147 * of tasks with abnormal "nice" values across CPUs the contribution that
1148 * each task makes to its run queue's load is weighted according to its
1149 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1150 * scaled version of the new time slice allocation that they receive on time
1154 #define WEIGHT_IDLEPRIO 3
1155 #define WMULT_IDLEPRIO 1431655765
1157 extern const int sched_prio_to_weight[40];
1158 extern const u32 sched_prio_to_wmult[40];
1161 * {de,en}queue flags:
1163 * DEQUEUE_SLEEP - task is no longer runnable
1164 * ENQUEUE_WAKEUP - task just became runnable
1166 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1167 * are in a known state which allows modification. Such pairs
1168 * should preserve as much state as possible.
1170 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1173 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1174 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1175 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1179 #define DEQUEUE_SLEEP 0x01
1180 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1181 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1183 #define ENQUEUE_WAKEUP 0x01
1184 #define ENQUEUE_RESTORE 0x02
1185 #define ENQUEUE_MOVE 0x04
1187 #define ENQUEUE_HEAD 0x08
1188 #define ENQUEUE_REPLENISH 0x10
1190 #define ENQUEUE_MIGRATED 0x20
1192 #define ENQUEUE_MIGRATED 0x00
1195 #define RETRY_TASK ((void *)-1UL)
1197 struct sched_class {
1198 const struct sched_class *next;
1200 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1201 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1202 void (*yield_task) (struct rq *rq);
1203 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1205 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1208 * It is the responsibility of the pick_next_task() method that will
1209 * return the next task to call put_prev_task() on the @prev task or
1210 * something equivalent.
1212 * May return RETRY_TASK when it finds a higher prio class has runnable
1215 struct task_struct * (*pick_next_task) (struct rq *rq,
1216 struct task_struct *prev,
1217 struct pin_cookie cookie);
1218 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1221 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1222 void (*migrate_task_rq)(struct task_struct *p);
1224 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1226 void (*set_cpus_allowed)(struct task_struct *p,
1227 const struct cpumask *newmask);
1229 void (*rq_online)(struct rq *rq);
1230 void (*rq_offline)(struct rq *rq);
1233 void (*set_curr_task) (struct rq *rq);
1234 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1235 void (*task_fork) (struct task_struct *p);
1236 void (*task_dead) (struct task_struct *p);
1239 * The switched_from() call is allowed to drop rq->lock, therefore we
1240 * cannot assume the switched_from/switched_to pair is serliazed by
1241 * rq->lock. They are however serialized by p->pi_lock.
1243 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1244 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1245 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1248 unsigned int (*get_rr_interval) (struct rq *rq,
1249 struct task_struct *task);
1251 void (*update_curr) (struct rq *rq);
1253 #define TASK_SET_GROUP 0
1254 #define TASK_MOVE_GROUP 1
1256 #ifdef CONFIG_FAIR_GROUP_SCHED
1257 void (*task_change_group) (struct task_struct *p, int type);
1261 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1263 prev->sched_class->put_prev_task(rq, prev);
1266 #define sched_class_highest (&stop_sched_class)
1267 #define for_each_class(class) \
1268 for (class = sched_class_highest; class; class = class->next)
1270 extern const struct sched_class stop_sched_class;
1271 extern const struct sched_class dl_sched_class;
1272 extern const struct sched_class rt_sched_class;
1273 extern const struct sched_class fair_sched_class;
1274 extern const struct sched_class idle_sched_class;
1279 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1281 extern void trigger_load_balance(struct rq *rq);
1283 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1287 #ifdef CONFIG_CPU_IDLE
1288 static inline void idle_set_state(struct rq *rq,
1289 struct cpuidle_state *idle_state)
1291 rq->idle_state = idle_state;
1294 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1296 WARN_ON(!rcu_read_lock_held());
1297 return rq->idle_state;
1300 static inline void idle_set_state(struct rq *rq,
1301 struct cpuidle_state *idle_state)
1305 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1311 extern void sysrq_sched_debug_show(void);
1312 extern void sched_init_granularity(void);
1313 extern void update_max_interval(void);
1315 extern void init_sched_dl_class(void);
1316 extern void init_sched_rt_class(void);
1317 extern void init_sched_fair_class(void);
1319 extern void resched_curr(struct rq *rq);
1320 extern void resched_cpu(int cpu);
1322 extern struct rt_bandwidth def_rt_bandwidth;
1323 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1325 extern struct dl_bandwidth def_dl_bandwidth;
1326 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1327 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1329 unsigned long to_ratio(u64 period, u64 runtime);
1331 extern void init_entity_runnable_average(struct sched_entity *se);
1332 extern void post_init_entity_util_avg(struct sched_entity *se);
1334 #ifdef CONFIG_NO_HZ_FULL
1335 extern bool sched_can_stop_tick(struct rq *rq);
1338 * Tick may be needed by tasks in the runqueue depending on their policy and
1339 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1340 * nohz mode if necessary.
1342 static inline void sched_update_tick_dependency(struct rq *rq)
1346 if (!tick_nohz_full_enabled())
1351 if (!tick_nohz_full_cpu(cpu))
1354 if (sched_can_stop_tick(rq))
1355 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1357 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1360 static inline void sched_update_tick_dependency(struct rq *rq) { }
1363 static inline void add_nr_running(struct rq *rq, unsigned count)
1365 unsigned prev_nr = rq->nr_running;
1367 rq->nr_running = prev_nr + count;
1369 if (prev_nr < 2 && rq->nr_running >= 2) {
1371 if (!rq->rd->overload)
1372 rq->rd->overload = true;
1376 sched_update_tick_dependency(rq);
1379 static inline void sub_nr_running(struct rq *rq, unsigned count)
1381 rq->nr_running -= count;
1382 /* Check if we still need preemption */
1383 sched_update_tick_dependency(rq);
1386 static inline void rq_last_tick_reset(struct rq *rq)
1388 #ifdef CONFIG_NO_HZ_FULL
1389 rq->last_sched_tick = jiffies;
1393 extern void update_rq_clock(struct rq *rq);
1395 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1396 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1398 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1400 extern const_debug unsigned int sysctl_sched_time_avg;
1401 extern const_debug unsigned int sysctl_sched_nr_migrate;
1402 extern const_debug unsigned int sysctl_sched_migration_cost;
1404 static inline u64 sched_avg_period(void)
1406 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1409 #ifdef CONFIG_SCHED_HRTICK
1413 * - enabled by features
1414 * - hrtimer is actually high res
1416 static inline int hrtick_enabled(struct rq *rq)
1418 if (!sched_feat(HRTICK))
1420 if (!cpu_active(cpu_of(rq)))
1422 return hrtimer_is_hres_active(&rq->hrtick_timer);
1425 void hrtick_start(struct rq *rq, u64 delay);
1429 static inline int hrtick_enabled(struct rq *rq)
1434 #endif /* CONFIG_SCHED_HRTICK */
1437 extern void sched_avg_update(struct rq *rq);
1439 #ifndef arch_scale_freq_capacity
1440 static __always_inline
1441 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1443 return SCHED_CAPACITY_SCALE;
1447 #ifndef arch_scale_cpu_capacity
1448 static __always_inline
1449 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1451 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1452 return sd->smt_gain / sd->span_weight;
1454 return SCHED_CAPACITY_SCALE;
1458 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1460 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1461 sched_avg_update(rq);
1464 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1465 static inline void sched_avg_update(struct rq *rq) { }
1469 unsigned long flags;
1470 struct pin_cookie cookie;
1473 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1474 __acquires(rq->lock);
1475 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1476 __acquires(p->pi_lock)
1477 __acquires(rq->lock);
1479 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1480 __releases(rq->lock)
1482 lockdep_unpin_lock(&rq->lock, rf->cookie);
1483 raw_spin_unlock(&rq->lock);
1487 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1488 __releases(rq->lock)
1489 __releases(p->pi_lock)
1491 lockdep_unpin_lock(&rq->lock, rf->cookie);
1492 raw_spin_unlock(&rq->lock);
1493 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1497 #ifdef CONFIG_PREEMPT
1499 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1502 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1503 * way at the expense of forcing extra atomic operations in all
1504 * invocations. This assures that the double_lock is acquired using the
1505 * same underlying policy as the spinlock_t on this architecture, which
1506 * reduces latency compared to the unfair variant below. However, it
1507 * also adds more overhead and therefore may reduce throughput.
1509 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1510 __releases(this_rq->lock)
1511 __acquires(busiest->lock)
1512 __acquires(this_rq->lock)
1514 raw_spin_unlock(&this_rq->lock);
1515 double_rq_lock(this_rq, busiest);
1522 * Unfair double_lock_balance: Optimizes throughput at the expense of
1523 * latency by eliminating extra atomic operations when the locks are
1524 * already in proper order on entry. This favors lower cpu-ids and will
1525 * grant the double lock to lower cpus over higher ids under contention,
1526 * regardless of entry order into the function.
1528 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1529 __releases(this_rq->lock)
1530 __acquires(busiest->lock)
1531 __acquires(this_rq->lock)
1535 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1536 if (busiest < this_rq) {
1537 raw_spin_unlock(&this_rq->lock);
1538 raw_spin_lock(&busiest->lock);
1539 raw_spin_lock_nested(&this_rq->lock,
1540 SINGLE_DEPTH_NESTING);
1543 raw_spin_lock_nested(&busiest->lock,
1544 SINGLE_DEPTH_NESTING);
1549 #endif /* CONFIG_PREEMPT */
1552 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1554 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1556 if (unlikely(!irqs_disabled())) {
1557 /* printk() doesn't work good under rq->lock */
1558 raw_spin_unlock(&this_rq->lock);
1562 return _double_lock_balance(this_rq, busiest);
1565 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1566 __releases(busiest->lock)
1568 raw_spin_unlock(&busiest->lock);
1569 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1572 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1578 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1581 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1587 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1590 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1596 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1600 * double_rq_lock - safely lock two runqueues
1602 * Note this does not disable interrupts like task_rq_lock,
1603 * you need to do so manually before calling.
1605 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1606 __acquires(rq1->lock)
1607 __acquires(rq2->lock)
1609 BUG_ON(!irqs_disabled());
1611 raw_spin_lock(&rq1->lock);
1612 __acquire(rq2->lock); /* Fake it out ;) */
1615 raw_spin_lock(&rq1->lock);
1616 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1618 raw_spin_lock(&rq2->lock);
1619 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1625 * double_rq_unlock - safely unlock two runqueues
1627 * Note this does not restore interrupts like task_rq_unlock,
1628 * you need to do so manually after calling.
1630 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1631 __releases(rq1->lock)
1632 __releases(rq2->lock)
1634 raw_spin_unlock(&rq1->lock);
1636 raw_spin_unlock(&rq2->lock);
1638 __release(rq2->lock);
1641 #else /* CONFIG_SMP */
1644 * double_rq_lock - safely lock two runqueues
1646 * Note this does not disable interrupts like task_rq_lock,
1647 * you need to do so manually before calling.
1649 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1650 __acquires(rq1->lock)
1651 __acquires(rq2->lock)
1653 BUG_ON(!irqs_disabled());
1655 raw_spin_lock(&rq1->lock);
1656 __acquire(rq2->lock); /* Fake it out ;) */
1660 * double_rq_unlock - safely unlock two runqueues
1662 * Note this does not restore interrupts like task_rq_unlock,
1663 * you need to do so manually after calling.
1665 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1666 __releases(rq1->lock)
1667 __releases(rq2->lock)
1670 raw_spin_unlock(&rq1->lock);
1671 __release(rq2->lock);
1676 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1677 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1679 #ifdef CONFIG_SCHED_DEBUG
1680 extern void print_cfs_stats(struct seq_file *m, int cpu);
1681 extern void print_rt_stats(struct seq_file *m, int cpu);
1682 extern void print_dl_stats(struct seq_file *m, int cpu);
1684 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1686 #ifdef CONFIG_NUMA_BALANCING
1688 show_numa_stats(struct task_struct *p, struct seq_file *m);
1690 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1691 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1692 #endif /* CONFIG_NUMA_BALANCING */
1693 #endif /* CONFIG_SCHED_DEBUG */
1695 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1696 extern void init_rt_rq(struct rt_rq *rt_rq);
1697 extern void init_dl_rq(struct dl_rq *dl_rq);
1699 extern void cfs_bandwidth_usage_inc(void);
1700 extern void cfs_bandwidth_usage_dec(void);
1702 #ifdef CONFIG_NO_HZ_COMMON
1703 enum rq_nohz_flag_bits {
1708 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1710 extern void nohz_balance_exit_idle(unsigned int cpu);
1712 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1715 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1717 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1718 DECLARE_PER_CPU(u64, cpu_softirq_time);
1720 #ifndef CONFIG_64BIT
1721 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1723 static inline void irq_time_write_begin(void)
1725 __this_cpu_inc(irq_time_seq.sequence);
1729 static inline void irq_time_write_end(void)
1732 __this_cpu_inc(irq_time_seq.sequence);
1735 static inline u64 irq_time_read(int cpu)
1741 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1742 irq_time = per_cpu(cpu_softirq_time, cpu) +
1743 per_cpu(cpu_hardirq_time, cpu);
1744 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1748 #else /* CONFIG_64BIT */
1749 static inline void irq_time_write_begin(void)
1753 static inline void irq_time_write_end(void)
1757 static inline u64 irq_time_read(int cpu)
1759 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1761 #endif /* CONFIG_64BIT */
1762 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1764 #ifdef CONFIG_CPU_FREQ
1765 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1768 * cpufreq_update_util - Take a note about CPU utilization changes.
1769 * @time: Current time.
1770 * @util: Current utilization.
1771 * @max: Utilization ceiling.
1773 * This function is called by the scheduler on every invocation of
1774 * update_load_avg() on the CPU whose utilization is being updated.
1776 * It can only be called from RCU-sched read-side critical sections.
1778 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max)
1780 struct update_util_data *data;
1782 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1784 data->func(data, time, util, max);
1788 * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed.
1789 * @time: Current time.
1791 * The way cpufreq is currently arranged requires it to evaluate the CPU
1792 * performance state (frequency/voltage) on a regular basis to prevent it from
1793 * being stuck in a completely inadequate performance level for too long.
1794 * That is not guaranteed to happen if the updates are only triggered from CFS,
1795 * though, because they may not be coming in if RT or deadline tasks are active
1796 * all the time (or there are RT and DL tasks only).
1798 * As a workaround for that issue, this function is called by the RT and DL
1799 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1800 * but that really is a band-aid. Going forward it should be replaced with
1801 * solutions targeted more specifically at RT and DL tasks.
1803 static inline void cpufreq_trigger_update(u64 time)
1805 cpufreq_update_util(time, ULONG_MAX, 0);
1808 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {}
1809 static inline void cpufreq_trigger_update(u64 time) {}
1810 #endif /* CONFIG_CPU_FREQ */
1812 #ifdef arch_scale_freq_capacity
1813 #ifndef arch_scale_freq_invariant
1814 #define arch_scale_freq_invariant() (true)
1816 #else /* arch_scale_freq_capacity */
1817 #define arch_scale_freq_invariant() (false)