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);
34 extern void cpu_load_update_active(struct rq *this_rq);
36 static inline void cpu_load_update_active(struct rq *this_rq) { }
40 * Helpers for converting nanosecond timing to jiffy resolution
42 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
45 * Increase resolution of nice-level calculations for 64-bit architectures.
46 * The extra resolution improves shares distribution and load balancing of
47 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
48 * hierarchies, especially on larger systems. This is not a user-visible change
49 * and does not change the user-interface for setting shares/weights.
51 * We increase resolution only if we have enough bits to allow this increased
52 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
53 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
56 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
57 # define SCHED_LOAD_RESOLUTION 10
58 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
59 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
61 # define SCHED_LOAD_RESOLUTION 0
62 # define scale_load(w) (w)
63 # define scale_load_down(w) (w)
66 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
67 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
69 #define NICE_0_LOAD SCHED_LOAD_SCALE
70 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
73 * Single value that decides SCHED_DEADLINE internal math precision.
74 * 10 -> just above 1us
75 * 9 -> just above 0.5us
80 * These are the 'tuning knobs' of the scheduler:
84 * single value that denotes runtime == period, ie unlimited time.
86 #define RUNTIME_INF ((u64)~0ULL)
88 static inline int idle_policy(int policy)
90 return policy == SCHED_IDLE;
92 static inline int fair_policy(int policy)
94 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
97 static inline int rt_policy(int policy)
99 return policy == SCHED_FIFO || policy == SCHED_RR;
102 static inline int dl_policy(int policy)
104 return policy == SCHED_DEADLINE;
106 static inline bool valid_policy(int policy)
108 return idle_policy(policy) || fair_policy(policy) ||
109 rt_policy(policy) || dl_policy(policy);
112 static inline int task_has_rt_policy(struct task_struct *p)
114 return rt_policy(p->policy);
117 static inline int task_has_dl_policy(struct task_struct *p)
119 return dl_policy(p->policy);
123 * Tells if entity @a should preempt entity @b.
126 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
128 return dl_time_before(a->deadline, b->deadline);
132 * This is the priority-queue data structure of the RT scheduling class:
134 struct rt_prio_array {
135 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
136 struct list_head queue[MAX_RT_PRIO];
139 struct rt_bandwidth {
140 /* nests inside the rq lock: */
141 raw_spinlock_t rt_runtime_lock;
144 struct hrtimer rt_period_timer;
145 unsigned int rt_period_active;
148 void __dl_clear_params(struct task_struct *p);
151 * To keep the bandwidth of -deadline tasks and groups under control
152 * we need some place where:
153 * - store the maximum -deadline bandwidth of the system (the group);
154 * - cache the fraction of that bandwidth that is currently allocated.
156 * This is all done in the data structure below. It is similar to the
157 * one used for RT-throttling (rt_bandwidth), with the main difference
158 * that, since here we are only interested in admission control, we
159 * do not decrease any runtime while the group "executes", neither we
160 * need a timer to replenish it.
162 * With respect to SMP, the bandwidth is given on a per-CPU basis,
164 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
165 * - dl_total_bw array contains, in the i-eth element, the currently
166 * allocated bandwidth on the i-eth CPU.
167 * Moreover, groups consume bandwidth on each CPU, while tasks only
168 * consume bandwidth on the CPU they're running on.
169 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
170 * that will be shown the next time the proc or cgroup controls will
171 * be red. It on its turn can be changed by writing on its own
174 struct dl_bandwidth {
175 raw_spinlock_t dl_runtime_lock;
180 static inline int dl_bandwidth_enabled(void)
182 return sysctl_sched_rt_runtime >= 0;
185 extern struct dl_bw *dl_bw_of(int i);
193 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
195 dl_b->total_bw -= tsk_bw;
199 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
201 dl_b->total_bw += tsk_bw;
205 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
207 return dl_b->bw != -1 &&
208 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
211 extern struct mutex sched_domains_mutex;
213 #ifdef CONFIG_CGROUP_SCHED
215 #include <linux/cgroup.h>
220 extern struct list_head task_groups;
222 struct cfs_bandwidth {
223 #ifdef CONFIG_CFS_BANDWIDTH
227 s64 hierarchical_quota;
230 int idle, period_active;
231 struct hrtimer period_timer, slack_timer;
232 struct list_head throttled_cfs_rq;
235 int nr_periods, nr_throttled;
240 /* task group related information */
242 struct cgroup_subsys_state css;
244 #ifdef CONFIG_FAIR_GROUP_SCHED
245 /* schedulable entities of this group on each cpu */
246 struct sched_entity **se;
247 /* runqueue "owned" by this group on each cpu */
248 struct cfs_rq **cfs_rq;
249 unsigned long shares;
253 * load_avg can be heavily contended at clock tick time, so put
254 * it in its own cacheline separated from the fields above which
255 * will also be accessed at each tick.
257 atomic_long_t load_avg ____cacheline_aligned;
261 #ifdef CONFIG_RT_GROUP_SCHED
262 struct sched_rt_entity **rt_se;
263 struct rt_rq **rt_rq;
265 struct rt_bandwidth rt_bandwidth;
269 struct list_head list;
271 struct task_group *parent;
272 struct list_head siblings;
273 struct list_head children;
275 #ifdef CONFIG_SCHED_AUTOGROUP
276 struct autogroup *autogroup;
279 struct cfs_bandwidth cfs_bandwidth;
282 #ifdef CONFIG_FAIR_GROUP_SCHED
283 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
286 * A weight of 0 or 1 can cause arithmetics problems.
287 * A weight of a cfs_rq is the sum of weights of which entities
288 * are queued on this cfs_rq, so a weight of a entity should not be
289 * too large, so as the shares value of a task group.
290 * (The default weight is 1024 - so there's no practical
291 * limitation from this.)
293 #define MIN_SHARES (1UL << 1)
294 #define MAX_SHARES (1UL << 18)
297 typedef int (*tg_visitor)(struct task_group *, void *);
299 extern int walk_tg_tree_from(struct task_group *from,
300 tg_visitor down, tg_visitor up, void *data);
303 * Iterate the full tree, calling @down when first entering a node and @up when
304 * leaving it for the final time.
306 * Caller must hold rcu_lock or sufficient equivalent.
308 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
310 return walk_tg_tree_from(&root_task_group, down, up, data);
313 extern int tg_nop(struct task_group *tg, void *data);
315 extern void free_fair_sched_group(struct task_group *tg);
316 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
317 extern void unregister_fair_sched_group(struct task_group *tg);
318 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
319 struct sched_entity *se, int cpu,
320 struct sched_entity *parent);
321 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
323 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
324 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
325 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
327 extern void free_rt_sched_group(struct task_group *tg);
328 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
329 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
330 struct sched_rt_entity *rt_se, int cpu,
331 struct sched_rt_entity *parent);
333 extern struct task_group *sched_create_group(struct task_group *parent);
334 extern void sched_online_group(struct task_group *tg,
335 struct task_group *parent);
336 extern void sched_destroy_group(struct task_group *tg);
337 extern void sched_offline_group(struct task_group *tg);
339 extern void sched_move_task(struct task_struct *tsk);
341 #ifdef CONFIG_FAIR_GROUP_SCHED
342 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
345 extern void set_task_rq_fair(struct sched_entity *se,
346 struct cfs_rq *prev, struct cfs_rq *next);
347 #else /* !CONFIG_SMP */
348 static inline void set_task_rq_fair(struct sched_entity *se,
349 struct cfs_rq *prev, struct cfs_rq *next) { }
350 #endif /* CONFIG_SMP */
351 #endif /* CONFIG_FAIR_GROUP_SCHED */
353 #else /* CONFIG_CGROUP_SCHED */
355 struct cfs_bandwidth { };
357 #endif /* CONFIG_CGROUP_SCHED */
359 /* CFS-related fields in a runqueue */
361 struct load_weight load;
362 unsigned int nr_running, h_nr_running;
367 u64 min_vruntime_copy;
370 struct rb_root tasks_timeline;
371 struct rb_node *rb_leftmost;
374 * 'curr' points to currently running entity on this cfs_rq.
375 * It is set to NULL otherwise (i.e when none are currently running).
377 struct sched_entity *curr, *next, *last, *skip;
379 #ifdef CONFIG_SCHED_DEBUG
380 unsigned int nr_spread_over;
387 struct sched_avg avg;
388 u64 runnable_load_sum;
389 unsigned long runnable_load_avg;
390 #ifdef CONFIG_FAIR_GROUP_SCHED
391 unsigned long tg_load_avg_contrib;
393 atomic_long_t removed_load_avg, removed_util_avg;
395 u64 load_last_update_time_copy;
398 #ifdef CONFIG_FAIR_GROUP_SCHED
400 * h_load = weight * f(tg)
402 * Where f(tg) is the recursive weight fraction assigned to
405 unsigned long h_load;
406 u64 last_h_load_update;
407 struct sched_entity *h_load_next;
408 #endif /* CONFIG_FAIR_GROUP_SCHED */
409 #endif /* CONFIG_SMP */
411 #ifdef CONFIG_FAIR_GROUP_SCHED
412 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
415 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
416 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
417 * (like users, containers etc.)
419 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
420 * list is used during load balance.
423 struct list_head leaf_cfs_rq_list;
424 struct task_group *tg; /* group that "owns" this runqueue */
426 #ifdef CONFIG_CFS_BANDWIDTH
429 s64 runtime_remaining;
431 u64 throttled_clock, throttled_clock_task;
432 u64 throttled_clock_task_time;
433 int throttled, throttle_count;
434 struct list_head throttled_list;
435 #endif /* CONFIG_CFS_BANDWIDTH */
436 #endif /* CONFIG_FAIR_GROUP_SCHED */
439 static inline int rt_bandwidth_enabled(void)
441 return sysctl_sched_rt_runtime >= 0;
444 /* RT IPI pull logic requires IRQ_WORK */
445 #ifdef CONFIG_IRQ_WORK
446 # define HAVE_RT_PUSH_IPI
449 /* Real-Time classes' related field in a runqueue: */
451 struct rt_prio_array active;
452 unsigned int rt_nr_running;
453 unsigned int rr_nr_running;
454 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
456 int curr; /* highest queued rt task prio */
458 int next; /* next highest */
463 unsigned long rt_nr_migratory;
464 unsigned long rt_nr_total;
466 struct plist_head pushable_tasks;
467 #ifdef HAVE_RT_PUSH_IPI
470 struct irq_work push_work;
471 raw_spinlock_t push_lock;
473 #endif /* CONFIG_SMP */
479 /* Nests inside the rq lock: */
480 raw_spinlock_t rt_runtime_lock;
482 #ifdef CONFIG_RT_GROUP_SCHED
483 unsigned long rt_nr_boosted;
486 struct task_group *tg;
490 /* Deadline class' related fields in a runqueue */
492 /* runqueue is an rbtree, ordered by deadline */
493 struct rb_root rb_root;
494 struct rb_node *rb_leftmost;
496 unsigned long dl_nr_running;
500 * Deadline values of the currently executing and the
501 * earliest ready task on this rq. Caching these facilitates
502 * the decision wether or not a ready but not running task
503 * should migrate somewhere else.
510 unsigned long dl_nr_migratory;
514 * Tasks on this rq that can be pushed away. They are kept in
515 * an rb-tree, ordered by tasks' deadlines, with caching
516 * of the leftmost (earliest deadline) element.
518 struct rb_root pushable_dl_tasks_root;
519 struct rb_node *pushable_dl_tasks_leftmost;
528 * We add the notion of a root-domain which will be used to define per-domain
529 * variables. Each exclusive cpuset essentially defines an island domain by
530 * fully partitioning the member cpus from any other cpuset. Whenever a new
531 * exclusive cpuset is created, we also create and attach a new root-domain
540 cpumask_var_t online;
542 /* Indicate more than one runnable task for any CPU */
546 * The bit corresponding to a CPU gets set here if such CPU has more
547 * than one runnable -deadline task (as it is below for RT tasks).
549 cpumask_var_t dlo_mask;
555 * The "RT overload" flag: it gets set if a CPU has more than
556 * one runnable RT task.
558 cpumask_var_t rto_mask;
559 struct cpupri cpupri;
562 extern struct root_domain def_root_domain;
564 #endif /* CONFIG_SMP */
567 * This is the main, per-CPU runqueue data structure.
569 * Locking rule: those places that want to lock multiple runqueues
570 * (such as the load balancing or the thread migration code), lock
571 * acquire operations must be ordered by ascending &runqueue.
578 * nr_running and cpu_load should be in the same cacheline because
579 * remote CPUs use both these fields when doing load calculation.
581 unsigned int nr_running;
582 #ifdef CONFIG_NUMA_BALANCING
583 unsigned int nr_numa_running;
584 unsigned int nr_preferred_running;
586 #define CPU_LOAD_IDX_MAX 5
587 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
588 #ifdef CONFIG_NO_HZ_COMMON
590 unsigned long last_load_update_tick;
591 #endif /* CONFIG_SMP */
593 unsigned long nohz_flags;
594 #endif /* CONFIG_NO_HZ_COMMON */
595 #ifdef CONFIG_NO_HZ_FULL
596 unsigned long last_sched_tick;
598 /* capture load from *all* tasks on this cpu: */
599 struct load_weight load;
600 unsigned long nr_load_updates;
607 #ifdef CONFIG_FAIR_GROUP_SCHED
608 /* list of leaf cfs_rq on this cpu: */
609 struct list_head leaf_cfs_rq_list;
610 #endif /* CONFIG_FAIR_GROUP_SCHED */
613 * This is part of a global counter where only the total sum
614 * over all CPUs matters. A task can increase this counter on
615 * one CPU and if it got migrated afterwards it may decrease
616 * it on another CPU. Always updated under the runqueue lock:
618 unsigned long nr_uninterruptible;
620 struct task_struct *curr, *idle, *stop;
621 unsigned long next_balance;
622 struct mm_struct *prev_mm;
624 unsigned int clock_skip_update;
631 struct root_domain *rd;
632 struct sched_domain *sd;
634 unsigned long cpu_capacity;
635 unsigned long cpu_capacity_orig;
637 struct callback_head *balance_callback;
639 unsigned char idle_balance;
640 /* For active balancing */
643 struct cpu_stop_work active_balance_work;
644 /* cpu of this runqueue: */
648 struct list_head cfs_tasks;
655 /* This is used to determine avg_idle's max value */
656 u64 max_idle_balance_cost;
659 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
662 #ifdef CONFIG_PARAVIRT
665 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
666 u64 prev_steal_time_rq;
669 /* calc_load related fields */
670 unsigned long calc_load_update;
671 long calc_load_active;
673 #ifdef CONFIG_SCHED_HRTICK
675 int hrtick_csd_pending;
676 struct call_single_data hrtick_csd;
678 struct hrtimer hrtick_timer;
681 #ifdef CONFIG_SCHEDSTATS
683 struct sched_info rq_sched_info;
684 unsigned long long rq_cpu_time;
685 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
687 /* sys_sched_yield() stats */
688 unsigned int yld_count;
690 /* schedule() stats */
691 unsigned int sched_count;
692 unsigned int sched_goidle;
694 /* try_to_wake_up() stats */
695 unsigned int ttwu_count;
696 unsigned int ttwu_local;
700 struct llist_head wake_list;
703 #ifdef CONFIG_CPU_IDLE
704 /* Must be inspected within a rcu lock section */
705 struct cpuidle_state *idle_state;
709 static inline int cpu_of(struct rq *rq)
718 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
720 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
721 #define this_rq() this_cpu_ptr(&runqueues)
722 #define task_rq(p) cpu_rq(task_cpu(p))
723 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
724 #define raw_rq() raw_cpu_ptr(&runqueues)
726 static inline u64 __rq_clock_broken(struct rq *rq)
728 return READ_ONCE(rq->clock);
731 static inline u64 rq_clock(struct rq *rq)
733 lockdep_assert_held(&rq->lock);
737 static inline u64 rq_clock_task(struct rq *rq)
739 lockdep_assert_held(&rq->lock);
740 return rq->clock_task;
743 #define RQCF_REQ_SKIP 0x01
744 #define RQCF_ACT_SKIP 0x02
746 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
748 lockdep_assert_held(&rq->lock);
750 rq->clock_skip_update |= RQCF_REQ_SKIP;
752 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
756 enum numa_topology_type {
761 extern enum numa_topology_type sched_numa_topology_type;
762 extern int sched_max_numa_distance;
763 extern bool find_numa_distance(int distance);
766 #ifdef CONFIG_NUMA_BALANCING
767 /* The regions in numa_faults array from task_struct */
768 enum numa_faults_stats {
774 extern void sched_setnuma(struct task_struct *p, int node);
775 extern int migrate_task_to(struct task_struct *p, int cpu);
776 extern int migrate_swap(struct task_struct *, struct task_struct *);
777 #endif /* CONFIG_NUMA_BALANCING */
782 queue_balance_callback(struct rq *rq,
783 struct callback_head *head,
784 void (*func)(struct rq *rq))
786 lockdep_assert_held(&rq->lock);
788 if (unlikely(head->next))
791 head->func = (void (*)(struct callback_head *))func;
792 head->next = rq->balance_callback;
793 rq->balance_callback = head;
796 extern void sched_ttwu_pending(void);
798 #define rcu_dereference_check_sched_domain(p) \
799 rcu_dereference_check((p), \
800 lockdep_is_held(&sched_domains_mutex))
803 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
804 * See detach_destroy_domains: synchronize_sched for details.
806 * The domain tree of any CPU may only be accessed from within
807 * preempt-disabled sections.
809 #define for_each_domain(cpu, __sd) \
810 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
811 __sd; __sd = __sd->parent)
813 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
816 * highest_flag_domain - Return highest sched_domain containing flag.
817 * @cpu: The cpu whose highest level of sched domain is to
819 * @flag: The flag to check for the highest sched_domain
822 * Returns the highest sched_domain of a cpu which contains the given flag.
824 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
826 struct sched_domain *sd, *hsd = NULL;
828 for_each_domain(cpu, sd) {
829 if (!(sd->flags & flag))
837 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
839 struct sched_domain *sd;
841 for_each_domain(cpu, sd) {
842 if (sd->flags & flag)
849 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
850 DECLARE_PER_CPU(int, sd_llc_size);
851 DECLARE_PER_CPU(int, sd_llc_id);
852 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
853 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
854 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
856 struct sched_group_capacity {
859 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
862 unsigned int capacity;
863 unsigned long next_update;
864 int imbalance; /* XXX unrelated to capacity but shared group state */
866 * Number of busy cpus in this group.
868 atomic_t nr_busy_cpus;
870 unsigned long cpumask[0]; /* iteration mask */
874 struct sched_group *next; /* Must be a circular list */
877 unsigned int group_weight;
878 struct sched_group_capacity *sgc;
881 * The CPUs this group covers.
883 * NOTE: this field is variable length. (Allocated dynamically
884 * by attaching extra space to the end of the structure,
885 * depending on how many CPUs the kernel has booted up with)
887 unsigned long cpumask[0];
890 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
892 return to_cpumask(sg->cpumask);
896 * cpumask masking which cpus in the group are allowed to iterate up the domain
899 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
901 return to_cpumask(sg->sgc->cpumask);
905 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
906 * @group: The group whose first cpu is to be returned.
908 static inline unsigned int group_first_cpu(struct sched_group *group)
910 return cpumask_first(sched_group_cpus(group));
913 extern int group_balance_cpu(struct sched_group *sg);
915 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
916 void register_sched_domain_sysctl(void);
917 void unregister_sched_domain_sysctl(void);
919 static inline void register_sched_domain_sysctl(void)
922 static inline void unregister_sched_domain_sysctl(void)
929 static inline void sched_ttwu_pending(void) { }
931 #endif /* CONFIG_SMP */
934 #include "auto_group.h"
936 #ifdef CONFIG_CGROUP_SCHED
939 * Return the group to which this tasks belongs.
941 * We cannot use task_css() and friends because the cgroup subsystem
942 * changes that value before the cgroup_subsys::attach() method is called,
943 * therefore we cannot pin it and might observe the wrong value.
945 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
946 * core changes this before calling sched_move_task().
948 * Instead we use a 'copy' which is updated from sched_move_task() while
949 * holding both task_struct::pi_lock and rq::lock.
951 static inline struct task_group *task_group(struct task_struct *p)
953 return p->sched_task_group;
956 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
957 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
959 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
960 struct task_group *tg = task_group(p);
963 #ifdef CONFIG_FAIR_GROUP_SCHED
964 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
965 p->se.cfs_rq = tg->cfs_rq[cpu];
966 p->se.parent = tg->se[cpu];
969 #ifdef CONFIG_RT_GROUP_SCHED
970 p->rt.rt_rq = tg->rt_rq[cpu];
971 p->rt.parent = tg->rt_se[cpu];
975 #else /* CONFIG_CGROUP_SCHED */
977 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
978 static inline struct task_group *task_group(struct task_struct *p)
983 #endif /* CONFIG_CGROUP_SCHED */
985 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
990 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
991 * successfuly executed on another CPU. We must ensure that updates of
992 * per-task data have been completed by this moment.
995 task_thread_info(p)->cpu = cpu;
1001 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1003 #ifdef CONFIG_SCHED_DEBUG
1004 # include <linux/static_key.h>
1005 # define const_debug __read_mostly
1007 # define const_debug const
1010 extern const_debug unsigned int sysctl_sched_features;
1012 #define SCHED_FEAT(name, enabled) \
1013 __SCHED_FEAT_##name ,
1016 #include "features.h"
1022 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1023 #define SCHED_FEAT(name, enabled) \
1024 static __always_inline bool static_branch_##name(struct static_key *key) \
1026 return static_key_##enabled(key); \
1029 #include "features.h"
1033 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1034 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1035 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1036 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1037 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1039 extern struct static_key_false sched_numa_balancing;
1040 extern struct static_key_false sched_schedstats;
1042 static inline u64 global_rt_period(void)
1044 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1047 static inline u64 global_rt_runtime(void)
1049 if (sysctl_sched_rt_runtime < 0)
1052 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1055 static inline int task_current(struct rq *rq, struct task_struct *p)
1057 return rq->curr == p;
1060 static inline int task_running(struct rq *rq, struct task_struct *p)
1065 return task_current(rq, p);
1069 static inline int task_on_rq_queued(struct task_struct *p)
1071 return p->on_rq == TASK_ON_RQ_QUEUED;
1074 static inline int task_on_rq_migrating(struct task_struct *p)
1076 return p->on_rq == TASK_ON_RQ_MIGRATING;
1079 #ifndef prepare_arch_switch
1080 # define prepare_arch_switch(next) do { } while (0)
1082 #ifndef finish_arch_post_lock_switch
1083 # define finish_arch_post_lock_switch() do { } while (0)
1086 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1090 * We can optimise this out completely for !SMP, because the
1091 * SMP rebalancing from interrupt is the only thing that cares
1098 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1102 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1103 * We must ensure this doesn't happen until the switch is completely
1106 * In particular, the load of prev->state in finish_task_switch() must
1107 * happen before this.
1109 * Pairs with the smp_cond_acquire() in try_to_wake_up().
1111 smp_store_release(&prev->on_cpu, 0);
1113 #ifdef CONFIG_DEBUG_SPINLOCK
1114 /* this is a valid case when another task releases the spinlock */
1115 rq->lock.owner = current;
1118 * If we are tracking spinlock dependencies then we have to
1119 * fix up the runqueue lock - which gets 'carried over' from
1120 * prev into current:
1122 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1124 raw_spin_unlock_irq(&rq->lock);
1130 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1131 #define WF_FORK 0x02 /* child wakeup after fork */
1132 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1135 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1136 * of tasks with abnormal "nice" values across CPUs the contribution that
1137 * each task makes to its run queue's load is weighted according to its
1138 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1139 * scaled version of the new time slice allocation that they receive on time
1143 #define WEIGHT_IDLEPRIO 3
1144 #define WMULT_IDLEPRIO 1431655765
1146 extern const int sched_prio_to_weight[40];
1147 extern const u32 sched_prio_to_wmult[40];
1150 * {de,en}queue flags:
1152 * DEQUEUE_SLEEP - task is no longer runnable
1153 * ENQUEUE_WAKEUP - task just became runnable
1155 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1156 * are in a known state which allows modification. Such pairs
1157 * should preserve as much state as possible.
1159 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1162 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1163 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1164 * ENQUEUE_WAKING - sched_class::task_waking was called
1168 #define DEQUEUE_SLEEP 0x01
1169 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1170 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1172 #define ENQUEUE_WAKEUP 0x01
1173 #define ENQUEUE_RESTORE 0x02
1174 #define ENQUEUE_MOVE 0x04
1176 #define ENQUEUE_HEAD 0x08
1177 #define ENQUEUE_REPLENISH 0x10
1179 #define ENQUEUE_WAKING 0x20
1181 #define ENQUEUE_WAKING 0x00
1184 #define RETRY_TASK ((void *)-1UL)
1186 struct sched_class {
1187 const struct sched_class *next;
1189 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1190 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1191 void (*yield_task) (struct rq *rq);
1192 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1194 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1197 * It is the responsibility of the pick_next_task() method that will
1198 * return the next task to call put_prev_task() on the @prev task or
1199 * something equivalent.
1201 * May return RETRY_TASK when it finds a higher prio class has runnable
1204 struct task_struct * (*pick_next_task) (struct rq *rq,
1205 struct task_struct *prev);
1206 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1209 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1210 void (*migrate_task_rq)(struct task_struct *p);
1212 void (*task_waking) (struct task_struct *task);
1213 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1215 void (*set_cpus_allowed)(struct task_struct *p,
1216 const struct cpumask *newmask);
1218 void (*rq_online)(struct rq *rq);
1219 void (*rq_offline)(struct rq *rq);
1222 void (*set_curr_task) (struct rq *rq);
1223 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1224 void (*task_fork) (struct task_struct *p);
1225 void (*task_dead) (struct task_struct *p);
1228 * The switched_from() call is allowed to drop rq->lock, therefore we
1229 * cannot assume the switched_from/switched_to pair is serliazed by
1230 * rq->lock. They are however serialized by p->pi_lock.
1232 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1233 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1234 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1237 unsigned int (*get_rr_interval) (struct rq *rq,
1238 struct task_struct *task);
1240 void (*update_curr) (struct rq *rq);
1242 #ifdef CONFIG_FAIR_GROUP_SCHED
1243 void (*task_move_group) (struct task_struct *p);
1247 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1249 prev->sched_class->put_prev_task(rq, prev);
1252 #define sched_class_highest (&stop_sched_class)
1253 #define for_each_class(class) \
1254 for (class = sched_class_highest; class; class = class->next)
1256 extern const struct sched_class stop_sched_class;
1257 extern const struct sched_class dl_sched_class;
1258 extern const struct sched_class rt_sched_class;
1259 extern const struct sched_class fair_sched_class;
1260 extern const struct sched_class idle_sched_class;
1265 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1267 extern void trigger_load_balance(struct rq *rq);
1269 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1273 #ifdef CONFIG_CPU_IDLE
1274 static inline void idle_set_state(struct rq *rq,
1275 struct cpuidle_state *idle_state)
1277 rq->idle_state = idle_state;
1280 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1282 WARN_ON(!rcu_read_lock_held());
1283 return rq->idle_state;
1286 static inline void idle_set_state(struct rq *rq,
1287 struct cpuidle_state *idle_state)
1291 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1297 extern void sysrq_sched_debug_show(void);
1298 extern void sched_init_granularity(void);
1299 extern void update_max_interval(void);
1301 extern void init_sched_dl_class(void);
1302 extern void init_sched_rt_class(void);
1303 extern void init_sched_fair_class(void);
1305 extern void resched_curr(struct rq *rq);
1306 extern void resched_cpu(int cpu);
1308 extern struct rt_bandwidth def_rt_bandwidth;
1309 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1311 extern struct dl_bandwidth def_dl_bandwidth;
1312 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1313 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1315 unsigned long to_ratio(u64 period, u64 runtime);
1317 extern void init_entity_runnable_average(struct sched_entity *se);
1318 extern void post_init_entity_util_avg(struct sched_entity *se);
1320 #ifdef CONFIG_NO_HZ_FULL
1321 extern bool sched_can_stop_tick(struct rq *rq);
1324 * Tick may be needed by tasks in the runqueue depending on their policy and
1325 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1326 * nohz mode if necessary.
1328 static inline void sched_update_tick_dependency(struct rq *rq)
1332 if (!tick_nohz_full_enabled())
1337 if (!tick_nohz_full_cpu(cpu))
1340 if (sched_can_stop_tick(rq))
1341 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1343 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1346 static inline void sched_update_tick_dependency(struct rq *rq) { }
1349 static inline void add_nr_running(struct rq *rq, unsigned count)
1351 unsigned prev_nr = rq->nr_running;
1353 rq->nr_running = prev_nr + count;
1355 if (prev_nr < 2 && rq->nr_running >= 2) {
1357 if (!rq->rd->overload)
1358 rq->rd->overload = true;
1362 sched_update_tick_dependency(rq);
1365 static inline void sub_nr_running(struct rq *rq, unsigned count)
1367 rq->nr_running -= count;
1368 /* Check if we still need preemption */
1369 sched_update_tick_dependency(rq);
1372 static inline void rq_last_tick_reset(struct rq *rq)
1374 #ifdef CONFIG_NO_HZ_FULL
1375 rq->last_sched_tick = jiffies;
1379 extern void update_rq_clock(struct rq *rq);
1381 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1382 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1384 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1386 extern const_debug unsigned int sysctl_sched_time_avg;
1387 extern const_debug unsigned int sysctl_sched_nr_migrate;
1388 extern const_debug unsigned int sysctl_sched_migration_cost;
1390 static inline u64 sched_avg_period(void)
1392 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1395 #ifdef CONFIG_SCHED_HRTICK
1399 * - enabled by features
1400 * - hrtimer is actually high res
1402 static inline int hrtick_enabled(struct rq *rq)
1404 if (!sched_feat(HRTICK))
1406 if (!cpu_active(cpu_of(rq)))
1408 return hrtimer_is_hres_active(&rq->hrtick_timer);
1411 void hrtick_start(struct rq *rq, u64 delay);
1415 static inline int hrtick_enabled(struct rq *rq)
1420 #endif /* CONFIG_SCHED_HRTICK */
1423 extern void sched_avg_update(struct rq *rq);
1425 #ifndef arch_scale_freq_capacity
1426 static __always_inline
1427 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1429 return SCHED_CAPACITY_SCALE;
1433 #ifndef arch_scale_cpu_capacity
1434 static __always_inline
1435 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1437 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1438 return sd->smt_gain / sd->span_weight;
1440 return SCHED_CAPACITY_SCALE;
1444 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1446 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1447 sched_avg_update(rq);
1450 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1451 static inline void sched_avg_update(struct rq *rq) { }
1455 * __task_rq_lock - lock the rq @p resides on.
1457 static inline struct rq *__task_rq_lock(struct task_struct *p)
1458 __acquires(rq->lock)
1462 lockdep_assert_held(&p->pi_lock);
1466 raw_spin_lock(&rq->lock);
1467 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1468 lockdep_pin_lock(&rq->lock);
1471 raw_spin_unlock(&rq->lock);
1473 while (unlikely(task_on_rq_migrating(p)))
1479 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1481 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1482 __acquires(p->pi_lock)
1483 __acquires(rq->lock)
1488 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1490 raw_spin_lock(&rq->lock);
1492 * move_queued_task() task_rq_lock()
1494 * ACQUIRE (rq->lock)
1495 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1496 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1497 * [S] ->cpu = new_cpu [L] task_rq()
1499 * RELEASE (rq->lock)
1501 * If we observe the old cpu in task_rq_lock, the acquire of
1502 * the old rq->lock will fully serialize against the stores.
1504 * If we observe the new cpu in task_rq_lock, the acquire will
1505 * pair with the WMB to ensure we must then also see migrating.
1507 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1508 lockdep_pin_lock(&rq->lock);
1511 raw_spin_unlock(&rq->lock);
1512 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1514 while (unlikely(task_on_rq_migrating(p)))
1519 static inline void __task_rq_unlock(struct rq *rq)
1520 __releases(rq->lock)
1522 lockdep_unpin_lock(&rq->lock);
1523 raw_spin_unlock(&rq->lock);
1527 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1528 __releases(rq->lock)
1529 __releases(p->pi_lock)
1531 lockdep_unpin_lock(&rq->lock);
1532 raw_spin_unlock(&rq->lock);
1533 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1537 #ifdef CONFIG_PREEMPT
1539 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1542 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1543 * way at the expense of forcing extra atomic operations in all
1544 * invocations. This assures that the double_lock is acquired using the
1545 * same underlying policy as the spinlock_t on this architecture, which
1546 * reduces latency compared to the unfair variant below. However, it
1547 * also adds more overhead and therefore may reduce throughput.
1549 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1550 __releases(this_rq->lock)
1551 __acquires(busiest->lock)
1552 __acquires(this_rq->lock)
1554 raw_spin_unlock(&this_rq->lock);
1555 double_rq_lock(this_rq, busiest);
1562 * Unfair double_lock_balance: Optimizes throughput at the expense of
1563 * latency by eliminating extra atomic operations when the locks are
1564 * already in proper order on entry. This favors lower cpu-ids and will
1565 * grant the double lock to lower cpus over higher ids under contention,
1566 * regardless of entry order into the function.
1568 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1569 __releases(this_rq->lock)
1570 __acquires(busiest->lock)
1571 __acquires(this_rq->lock)
1575 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1576 if (busiest < this_rq) {
1577 raw_spin_unlock(&this_rq->lock);
1578 raw_spin_lock(&busiest->lock);
1579 raw_spin_lock_nested(&this_rq->lock,
1580 SINGLE_DEPTH_NESTING);
1583 raw_spin_lock_nested(&busiest->lock,
1584 SINGLE_DEPTH_NESTING);
1589 #endif /* CONFIG_PREEMPT */
1592 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1594 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1596 if (unlikely(!irqs_disabled())) {
1597 /* printk() doesn't work good under rq->lock */
1598 raw_spin_unlock(&this_rq->lock);
1602 return _double_lock_balance(this_rq, busiest);
1605 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1606 __releases(busiest->lock)
1608 raw_spin_unlock(&busiest->lock);
1609 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1612 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1618 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1621 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1627 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1630 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1636 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1640 * double_rq_lock - safely lock two runqueues
1642 * Note this does not disable interrupts like task_rq_lock,
1643 * you need to do so manually before calling.
1645 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1646 __acquires(rq1->lock)
1647 __acquires(rq2->lock)
1649 BUG_ON(!irqs_disabled());
1651 raw_spin_lock(&rq1->lock);
1652 __acquire(rq2->lock); /* Fake it out ;) */
1655 raw_spin_lock(&rq1->lock);
1656 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1658 raw_spin_lock(&rq2->lock);
1659 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1665 * double_rq_unlock - safely unlock two runqueues
1667 * Note this does not restore interrupts like task_rq_unlock,
1668 * you need to do so manually after calling.
1670 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1671 __releases(rq1->lock)
1672 __releases(rq2->lock)
1674 raw_spin_unlock(&rq1->lock);
1676 raw_spin_unlock(&rq2->lock);
1678 __release(rq2->lock);
1681 #else /* CONFIG_SMP */
1684 * double_rq_lock - safely lock two runqueues
1686 * Note this does not disable interrupts like task_rq_lock,
1687 * you need to do so manually before calling.
1689 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1690 __acquires(rq1->lock)
1691 __acquires(rq2->lock)
1693 BUG_ON(!irqs_disabled());
1695 raw_spin_lock(&rq1->lock);
1696 __acquire(rq2->lock); /* Fake it out ;) */
1700 * double_rq_unlock - safely unlock two runqueues
1702 * Note this does not restore interrupts like task_rq_unlock,
1703 * you need to do so manually after calling.
1705 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1706 __releases(rq1->lock)
1707 __releases(rq2->lock)
1710 raw_spin_unlock(&rq1->lock);
1711 __release(rq2->lock);
1716 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1717 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1719 #ifdef CONFIG_SCHED_DEBUG
1720 extern void print_cfs_stats(struct seq_file *m, int cpu);
1721 extern void print_rt_stats(struct seq_file *m, int cpu);
1722 extern void print_dl_stats(struct seq_file *m, int cpu);
1724 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1726 #ifdef CONFIG_NUMA_BALANCING
1728 show_numa_stats(struct task_struct *p, struct seq_file *m);
1730 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1731 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1732 #endif /* CONFIG_NUMA_BALANCING */
1733 #endif /* CONFIG_SCHED_DEBUG */
1735 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1736 extern void init_rt_rq(struct rt_rq *rt_rq);
1737 extern void init_dl_rq(struct dl_rq *dl_rq);
1739 extern void cfs_bandwidth_usage_inc(void);
1740 extern void cfs_bandwidth_usage_dec(void);
1742 #ifdef CONFIG_NO_HZ_COMMON
1743 enum rq_nohz_flag_bits {
1748 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1751 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1753 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1754 DECLARE_PER_CPU(u64, cpu_softirq_time);
1756 #ifndef CONFIG_64BIT
1757 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1759 static inline void irq_time_write_begin(void)
1761 __this_cpu_inc(irq_time_seq.sequence);
1765 static inline void irq_time_write_end(void)
1768 __this_cpu_inc(irq_time_seq.sequence);
1771 static inline u64 irq_time_read(int cpu)
1777 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1778 irq_time = per_cpu(cpu_softirq_time, cpu) +
1779 per_cpu(cpu_hardirq_time, cpu);
1780 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1784 #else /* CONFIG_64BIT */
1785 static inline void irq_time_write_begin(void)
1789 static inline void irq_time_write_end(void)
1793 static inline u64 irq_time_read(int cpu)
1795 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1797 #endif /* CONFIG_64BIT */
1798 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1800 #ifdef CONFIG_CPU_FREQ
1801 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1804 * cpufreq_update_util - Take a note about CPU utilization changes.
1805 * @time: Current time.
1806 * @util: Current utilization.
1807 * @max: Utilization ceiling.
1809 * This function is called by the scheduler on every invocation of
1810 * update_load_avg() on the CPU whose utilization is being updated.
1812 * It can only be called from RCU-sched read-side critical sections.
1814 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max)
1816 struct update_util_data *data;
1818 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1820 data->func(data, time, util, max);
1824 * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed.
1825 * @time: Current time.
1827 * The way cpufreq is currently arranged requires it to evaluate the CPU
1828 * performance state (frequency/voltage) on a regular basis to prevent it from
1829 * being stuck in a completely inadequate performance level for too long.
1830 * That is not guaranteed to happen if the updates are only triggered from CFS,
1831 * though, because they may not be coming in if RT or deadline tasks are active
1832 * all the time (or there are RT and DL tasks only).
1834 * As a workaround for that issue, this function is called by the RT and DL
1835 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1836 * but that really is a band-aid. Going forward it should be replaced with
1837 * solutions targeted more specifically at RT and DL tasks.
1839 static inline void cpufreq_trigger_update(u64 time)
1841 cpufreq_update_util(time, ULONG_MAX, 0);
1844 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {}
1845 static inline void cpufreq_trigger_update(u64 time) {}
1846 #endif /* CONFIG_CPU_FREQ */
1848 static inline void account_reset_rq(struct rq *rq)
1850 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1851 rq->prev_irq_time = 0;
1853 #ifdef CONFIG_PARAVIRT
1854 rq->prev_steal_time = 0;
1856 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1857 rq->prev_steal_time_rq = 0;