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/mutex.h>
7 #include <linux/spinlock.h>
8 #include <linux/stop_machine.h>
9 #include <linux/irq_work.h>
10 #include <linux/tick.h>
11 #include <linux/slab.h>
14 #include "cpudeadline.h"
20 /* task_struct::on_rq states: */
21 #define TASK_ON_RQ_QUEUED 1
22 #define TASK_ON_RQ_MIGRATING 2
24 extern __read_mostly int scheduler_running;
26 extern unsigned long calc_load_update;
27 extern atomic_long_t calc_load_tasks;
29 extern void calc_global_load_tick(struct rq *this_rq);
30 extern long calc_load_fold_active(struct rq *this_rq);
33 extern void update_cpu_load_active(struct rq *this_rq);
35 static inline void update_cpu_load_active(struct rq *this_rq) { }
39 * Helpers for converting nanosecond timing to jiffy resolution
41 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
44 * Increase resolution of nice-level calculations for 64-bit architectures.
45 * The extra resolution improves shares distribution and load balancing of
46 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
47 * hierarchies, especially on larger systems. This is not a user-visible change
48 * and does not change the user-interface for setting shares/weights.
50 * We increase resolution only if we have enough bits to allow this increased
51 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
52 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
55 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
56 # define SCHED_LOAD_RESOLUTION 10
57 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
58 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
60 # define SCHED_LOAD_RESOLUTION 0
61 # define scale_load(w) (w)
62 # define scale_load_down(w) (w)
65 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
66 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
68 #define NICE_0_LOAD SCHED_LOAD_SCALE
69 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
72 * Single value that decides SCHED_DEADLINE internal math precision.
73 * 10 -> just above 1us
74 * 9 -> just above 0.5us
79 * These are the 'tuning knobs' of the scheduler:
83 * single value that denotes runtime == period, ie unlimited time.
85 #define RUNTIME_INF ((u64)~0ULL)
87 static inline int idle_policy(int policy)
89 return policy == SCHED_IDLE;
91 static inline int fair_policy(int policy)
93 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
96 static inline int rt_policy(int policy)
98 return policy == SCHED_FIFO || policy == SCHED_RR;
101 static inline int dl_policy(int policy)
103 return policy == SCHED_DEADLINE;
105 static inline bool valid_policy(int policy)
107 return idle_policy(policy) || fair_policy(policy) ||
108 rt_policy(policy) || dl_policy(policy);
111 static inline int task_has_rt_policy(struct task_struct *p)
113 return rt_policy(p->policy);
116 static inline int task_has_dl_policy(struct task_struct *p)
118 return dl_policy(p->policy);
122 * Tells if entity @a should preempt entity @b.
125 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
127 return dl_time_before(a->deadline, b->deadline);
131 * This is the priority-queue data structure of the RT scheduling class:
133 struct rt_prio_array {
134 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
135 struct list_head queue[MAX_RT_PRIO];
138 struct rt_bandwidth {
139 /* nests inside the rq lock: */
140 raw_spinlock_t rt_runtime_lock;
143 struct hrtimer rt_period_timer;
144 unsigned int rt_period_active;
147 void __dl_clear_params(struct task_struct *p);
150 * To keep the bandwidth of -deadline tasks and groups under control
151 * we need some place where:
152 * - store the maximum -deadline bandwidth of the system (the group);
153 * - cache the fraction of that bandwidth that is currently allocated.
155 * This is all done in the data structure below. It is similar to the
156 * one used for RT-throttling (rt_bandwidth), with the main difference
157 * that, since here we are only interested in admission control, we
158 * do not decrease any runtime while the group "executes", neither we
159 * need a timer to replenish it.
161 * With respect to SMP, the bandwidth is given on a per-CPU basis,
163 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
164 * - dl_total_bw array contains, in the i-eth element, the currently
165 * allocated bandwidth on the i-eth CPU.
166 * Moreover, groups consume bandwidth on each CPU, while tasks only
167 * consume bandwidth on the CPU they're running on.
168 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
169 * that will be shown the next time the proc or cgroup controls will
170 * be red. It on its turn can be changed by writing on its own
173 struct dl_bandwidth {
174 raw_spinlock_t dl_runtime_lock;
179 static inline int dl_bandwidth_enabled(void)
181 return sysctl_sched_rt_runtime >= 0;
184 extern struct dl_bw *dl_bw_of(int i);
192 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
194 dl_b->total_bw -= tsk_bw;
198 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
200 dl_b->total_bw += tsk_bw;
204 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
206 return dl_b->bw != -1 &&
207 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
210 extern struct mutex sched_domains_mutex;
212 #ifdef CONFIG_CGROUP_SCHED
214 #include <linux/cgroup.h>
219 extern struct list_head task_groups;
221 struct cfs_bandwidth {
222 #ifdef CONFIG_CFS_BANDWIDTH
226 s64 hierarchical_quota;
229 int idle, period_active;
230 struct hrtimer period_timer, slack_timer;
231 struct list_head throttled_cfs_rq;
234 int nr_periods, nr_throttled;
239 /* task group related information */
241 struct cgroup_subsys_state css;
243 #ifdef CONFIG_FAIR_GROUP_SCHED
244 /* schedulable entities of this group on each cpu */
245 struct sched_entity **se;
246 /* runqueue "owned" by this group on each cpu */
247 struct cfs_rq **cfs_rq;
248 unsigned long shares;
251 atomic_long_t load_avg;
255 #ifdef CONFIG_RT_GROUP_SCHED
256 struct sched_rt_entity **rt_se;
257 struct rt_rq **rt_rq;
259 struct rt_bandwidth rt_bandwidth;
263 struct list_head list;
265 struct task_group *parent;
266 struct list_head siblings;
267 struct list_head children;
269 #ifdef CONFIG_SCHED_AUTOGROUP
270 struct autogroup *autogroup;
273 struct cfs_bandwidth cfs_bandwidth;
276 #ifdef CONFIG_FAIR_GROUP_SCHED
277 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
280 * A weight of 0 or 1 can cause arithmetics problems.
281 * A weight of a cfs_rq is the sum of weights of which entities
282 * are queued on this cfs_rq, so a weight of a entity should not be
283 * too large, so as the shares value of a task group.
284 * (The default weight is 1024 - so there's no practical
285 * limitation from this.)
287 #define MIN_SHARES (1UL << 1)
288 #define MAX_SHARES (1UL << 18)
291 typedef int (*tg_visitor)(struct task_group *, void *);
293 extern int walk_tg_tree_from(struct task_group *from,
294 tg_visitor down, tg_visitor up, void *data);
297 * Iterate the full tree, calling @down when first entering a node and @up when
298 * leaving it for the final time.
300 * Caller must hold rcu_lock or sufficient equivalent.
302 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
304 return walk_tg_tree_from(&root_task_group, down, up, data);
307 extern int tg_nop(struct task_group *tg, void *data);
309 extern void free_fair_sched_group(struct task_group *tg);
310 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
311 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
312 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
313 struct sched_entity *se, int cpu,
314 struct sched_entity *parent);
315 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
316 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
318 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
319 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
320 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
322 extern void free_rt_sched_group(struct task_group *tg);
323 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
324 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
325 struct sched_rt_entity *rt_se, int cpu,
326 struct sched_rt_entity *parent);
328 extern struct task_group *sched_create_group(struct task_group *parent);
329 extern void sched_online_group(struct task_group *tg,
330 struct task_group *parent);
331 extern void sched_destroy_group(struct task_group *tg);
332 extern void sched_offline_group(struct task_group *tg);
334 extern void sched_move_task(struct task_struct *tsk);
336 #ifdef CONFIG_FAIR_GROUP_SCHED
337 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
340 #else /* CONFIG_CGROUP_SCHED */
342 struct cfs_bandwidth { };
344 #endif /* CONFIG_CGROUP_SCHED */
346 /* CFS-related fields in a runqueue */
348 struct load_weight load;
349 unsigned int nr_running, h_nr_running;
354 u64 min_vruntime_copy;
357 struct rb_root tasks_timeline;
358 struct rb_node *rb_leftmost;
361 * 'curr' points to currently running entity on this cfs_rq.
362 * It is set to NULL otherwise (i.e when none are currently running).
364 struct sched_entity *curr, *next, *last, *skip;
366 #ifdef CONFIG_SCHED_DEBUG
367 unsigned int nr_spread_over;
374 struct sched_avg avg;
375 u64 runnable_load_sum;
376 unsigned long runnable_load_avg;
377 #ifdef CONFIG_FAIR_GROUP_SCHED
378 unsigned long tg_load_avg_contrib;
380 atomic_long_t removed_load_avg, removed_util_avg;
382 u64 load_last_update_time_copy;
385 #ifdef CONFIG_FAIR_GROUP_SCHED
387 * h_load = weight * f(tg)
389 * Where f(tg) is the recursive weight fraction assigned to
392 unsigned long h_load;
393 u64 last_h_load_update;
394 struct sched_entity *h_load_next;
395 #endif /* CONFIG_FAIR_GROUP_SCHED */
396 #endif /* CONFIG_SMP */
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
402 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
403 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
404 * (like users, containers etc.)
406 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
407 * list is used during load balance.
410 struct list_head leaf_cfs_rq_list;
411 struct task_group *tg; /* group that "owns" this runqueue */
413 #ifdef CONFIG_CFS_BANDWIDTH
416 s64 runtime_remaining;
418 u64 throttled_clock, throttled_clock_task;
419 u64 throttled_clock_task_time;
420 int throttled, throttle_count;
421 struct list_head throttled_list;
422 #endif /* CONFIG_CFS_BANDWIDTH */
423 #endif /* CONFIG_FAIR_GROUP_SCHED */
426 static inline int rt_bandwidth_enabled(void)
428 return sysctl_sched_rt_runtime >= 0;
431 /* RT IPI pull logic requires IRQ_WORK */
432 #ifdef CONFIG_IRQ_WORK
433 # define HAVE_RT_PUSH_IPI
436 /* Real-Time classes' related field in a runqueue: */
438 struct rt_prio_array active;
439 unsigned int rt_nr_running;
440 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
442 int curr; /* highest queued rt task prio */
444 int next; /* next highest */
449 unsigned long rt_nr_migratory;
450 unsigned long rt_nr_total;
452 struct plist_head pushable_tasks;
453 #ifdef HAVE_RT_PUSH_IPI
456 struct irq_work push_work;
457 raw_spinlock_t push_lock;
459 #endif /* CONFIG_SMP */
465 /* Nests inside the rq lock: */
466 raw_spinlock_t rt_runtime_lock;
468 #ifdef CONFIG_RT_GROUP_SCHED
469 unsigned long rt_nr_boosted;
472 struct task_group *tg;
476 /* Deadline class' related fields in a runqueue */
478 /* runqueue is an rbtree, ordered by deadline */
479 struct rb_root rb_root;
480 struct rb_node *rb_leftmost;
482 unsigned long dl_nr_running;
486 * Deadline values of the currently executing and the
487 * earliest ready task on this rq. Caching these facilitates
488 * the decision wether or not a ready but not running task
489 * should migrate somewhere else.
496 unsigned long dl_nr_migratory;
500 * Tasks on this rq that can be pushed away. They are kept in
501 * an rb-tree, ordered by tasks' deadlines, with caching
502 * of the leftmost (earliest deadline) element.
504 struct rb_root pushable_dl_tasks_root;
505 struct rb_node *pushable_dl_tasks_leftmost;
514 * We add the notion of a root-domain which will be used to define per-domain
515 * variables. Each exclusive cpuset essentially defines an island domain by
516 * fully partitioning the member cpus from any other cpuset. Whenever a new
517 * exclusive cpuset is created, we also create and attach a new root-domain
526 cpumask_var_t online;
528 /* Indicate more than one runnable task for any CPU */
531 /* Indicate one or more cpus over-utilized (tipping point) */
535 * The bit corresponding to a CPU gets set here if such CPU has more
536 * than one runnable -deadline task (as it is below for RT tasks).
538 cpumask_var_t dlo_mask;
544 * The "RT overload" flag: it gets set if a CPU has more than
545 * one runnable RT task.
547 cpumask_var_t rto_mask;
548 struct cpupri cpupri;
550 /* Maximum cpu capacity in the system. */
551 unsigned long max_cpu_capacity;
554 extern struct root_domain def_root_domain;
556 #endif /* CONFIG_SMP */
559 * This is the main, per-CPU runqueue data structure.
561 * Locking rule: those places that want to lock multiple runqueues
562 * (such as the load balancing or the thread migration code), lock
563 * acquire operations must be ordered by ascending &runqueue.
570 * nr_running and cpu_load should be in the same cacheline because
571 * remote CPUs use both these fields when doing load calculation.
573 unsigned int nr_running;
574 #ifdef CONFIG_NUMA_BALANCING
575 unsigned int nr_numa_running;
576 unsigned int nr_preferred_running;
578 #define CPU_LOAD_IDX_MAX 5
579 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
580 unsigned long last_load_update_tick;
581 #ifdef CONFIG_NO_HZ_COMMON
583 unsigned long nohz_flags;
585 #ifdef CONFIG_NO_HZ_FULL
586 unsigned long last_sched_tick;
588 /* capture load from *all* tasks on this cpu: */
589 struct load_weight load;
590 unsigned long nr_load_updates;
597 #ifdef CONFIG_FAIR_GROUP_SCHED
598 /* list of leaf cfs_rq on this cpu: */
599 struct list_head leaf_cfs_rq_list;
600 #endif /* CONFIG_FAIR_GROUP_SCHED */
603 * This is part of a global counter where only the total sum
604 * over all CPUs matters. A task can increase this counter on
605 * one CPU and if it got migrated afterwards it may decrease
606 * it on another CPU. Always updated under the runqueue lock:
608 unsigned long nr_uninterruptible;
610 struct task_struct *curr, *idle, *stop;
611 unsigned long next_balance;
612 struct mm_struct *prev_mm;
614 unsigned int clock_skip_update;
621 struct root_domain *rd;
622 struct sched_domain *sd;
624 unsigned long cpu_capacity;
625 unsigned long cpu_capacity_orig;
627 struct callback_head *balance_callback;
629 unsigned char idle_balance;
630 /* For active balancing */
633 struct cpu_stop_work active_balance_work;
634 /* cpu of this runqueue: */
638 struct list_head cfs_tasks;
645 /* This is used to determine avg_idle's max value */
646 u64 max_idle_balance_cost;
649 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
652 #ifdef CONFIG_PARAVIRT
655 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
656 u64 prev_steal_time_rq;
659 /* calc_load related fields */
660 unsigned long calc_load_update;
661 long calc_load_active;
663 #ifdef CONFIG_SCHED_HRTICK
665 int hrtick_csd_pending;
666 struct call_single_data hrtick_csd;
668 struct hrtimer hrtick_timer;
671 #ifdef CONFIG_SCHEDSTATS
673 struct sched_info rq_sched_info;
674 unsigned long long rq_cpu_time;
675 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
677 /* sys_sched_yield() stats */
678 unsigned int yld_count;
680 /* schedule() stats */
681 unsigned int sched_count;
682 unsigned int sched_goidle;
684 /* try_to_wake_up() stats */
685 unsigned int ttwu_count;
686 unsigned int ttwu_local;
690 struct llist_head wake_list;
693 #ifdef CONFIG_CPU_IDLE
694 /* Must be inspected within a rcu lock section */
695 struct cpuidle_state *idle_state;
700 static inline int cpu_of(struct rq *rq)
709 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
711 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
712 #define this_rq() this_cpu_ptr(&runqueues)
713 #define task_rq(p) cpu_rq(task_cpu(p))
714 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
715 #define raw_rq() raw_cpu_ptr(&runqueues)
717 static inline u64 __rq_clock_broken(struct rq *rq)
719 return READ_ONCE(rq->clock);
722 static inline u64 rq_clock(struct rq *rq)
724 lockdep_assert_held(&rq->lock);
728 static inline u64 rq_clock_task(struct rq *rq)
730 lockdep_assert_held(&rq->lock);
731 return rq->clock_task;
734 #define RQCF_REQ_SKIP 0x01
735 #define RQCF_ACT_SKIP 0x02
737 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
739 lockdep_assert_held(&rq->lock);
741 rq->clock_skip_update |= RQCF_REQ_SKIP;
743 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
747 enum numa_topology_type {
752 extern enum numa_topology_type sched_numa_topology_type;
753 extern int sched_max_numa_distance;
754 extern bool find_numa_distance(int distance);
757 #ifdef CONFIG_NUMA_BALANCING
758 /* The regions in numa_faults array from task_struct */
759 enum numa_faults_stats {
765 extern void sched_setnuma(struct task_struct *p, int node);
766 extern int migrate_task_to(struct task_struct *p, int cpu);
767 extern int migrate_swap(struct task_struct *, struct task_struct *);
768 #endif /* CONFIG_NUMA_BALANCING */
773 queue_balance_callback(struct rq *rq,
774 struct callback_head *head,
775 void (*func)(struct rq *rq))
777 lockdep_assert_held(&rq->lock);
779 if (unlikely(head->next))
782 head->func = (void (*)(struct callback_head *))func;
783 head->next = rq->balance_callback;
784 rq->balance_callback = head;
787 extern void sched_ttwu_pending(void);
789 #define rcu_dereference_check_sched_domain(p) \
790 rcu_dereference_check((p), \
791 lockdep_is_held(&sched_domains_mutex))
794 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
795 * See detach_destroy_domains: synchronize_sched for details.
797 * The domain tree of any CPU may only be accessed from within
798 * preempt-disabled sections.
800 #define for_each_domain(cpu, __sd) \
801 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
802 __sd; __sd = __sd->parent)
804 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
807 * highest_flag_domain - Return highest sched_domain containing flag.
808 * @cpu: The cpu whose highest level of sched domain is to
810 * @flag: The flag to check for the highest sched_domain
813 * Returns the highest sched_domain of a cpu which contains the given flag.
815 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
817 struct sched_domain *sd, *hsd = NULL;
819 for_each_domain(cpu, sd) {
820 if (!(sd->flags & flag))
828 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
830 struct sched_domain *sd;
832 for_each_domain(cpu, sd) {
833 if (sd->flags & flag)
840 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
841 DECLARE_PER_CPU(int, sd_llc_size);
842 DECLARE_PER_CPU(int, sd_llc_id);
843 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
844 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
845 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
846 DECLARE_PER_CPU(struct sched_domain *, sd_ea);
847 DECLARE_PER_CPU(struct sched_domain *, sd_scs);
849 struct sched_group_capacity {
852 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
855 unsigned int capacity;
856 unsigned long next_update;
857 int imbalance; /* XXX unrelated to capacity but shared group state */
859 * Number of busy cpus in this group.
861 atomic_t nr_busy_cpus;
863 unsigned long cpumask[0]; /* iteration mask */
867 struct sched_group *next; /* Must be a circular list */
870 unsigned int group_weight;
871 struct sched_group_capacity *sgc;
872 const struct sched_group_energy const *sge;
875 * The CPUs this group covers.
877 * NOTE: this field is variable length. (Allocated dynamically
878 * by attaching extra space to the end of the structure,
879 * depending on how many CPUs the kernel has booted up with)
881 unsigned long cpumask[0];
884 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
886 return to_cpumask(sg->cpumask);
890 * cpumask masking which cpus in the group are allowed to iterate up the domain
893 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
895 return to_cpumask(sg->sgc->cpumask);
899 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
900 * @group: The group whose first cpu is to be returned.
902 static inline unsigned int group_first_cpu(struct sched_group *group)
904 return cpumask_first(sched_group_cpus(group));
907 extern int group_balance_cpu(struct sched_group *sg);
911 static inline void sched_ttwu_pending(void) { }
913 #endif /* CONFIG_SMP */
916 #include "auto_group.h"
918 #ifdef CONFIG_CGROUP_SCHED
921 * Return the group to which this tasks belongs.
923 * We cannot use task_css() and friends because the cgroup subsystem
924 * changes that value before the cgroup_subsys::attach() method is called,
925 * therefore we cannot pin it and might observe the wrong value.
927 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
928 * core changes this before calling sched_move_task().
930 * Instead we use a 'copy' which is updated from sched_move_task() while
931 * holding both task_struct::pi_lock and rq::lock.
933 static inline struct task_group *task_group(struct task_struct *p)
935 return p->sched_task_group;
938 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
939 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
941 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
942 struct task_group *tg = task_group(p);
945 #ifdef CONFIG_FAIR_GROUP_SCHED
946 p->se.cfs_rq = tg->cfs_rq[cpu];
947 p->se.parent = tg->se[cpu];
950 #ifdef CONFIG_RT_GROUP_SCHED
951 p->rt.rt_rq = tg->rt_rq[cpu];
952 p->rt.parent = tg->rt_se[cpu];
956 #else /* CONFIG_CGROUP_SCHED */
958 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
959 static inline struct task_group *task_group(struct task_struct *p)
964 #endif /* CONFIG_CGROUP_SCHED */
966 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
971 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
972 * successfuly executed on another CPU. We must ensure that updates of
973 * per-task data have been completed by this moment.
976 task_thread_info(p)->cpu = cpu;
982 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
984 #ifdef CONFIG_SCHED_DEBUG
985 # include <linux/static_key.h>
986 # define const_debug __read_mostly
988 # define const_debug const
991 extern const_debug unsigned int sysctl_sched_features;
993 #define SCHED_FEAT(name, enabled) \
994 __SCHED_FEAT_##name ,
997 #include "features.h"
1003 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1004 #define SCHED_FEAT(name, enabled) \
1005 static __always_inline bool static_branch_##name(struct static_key *key) \
1007 return static_key_##enabled(key); \
1010 #include "features.h"
1014 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1015 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1016 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1017 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1018 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1020 extern struct static_key_false sched_numa_balancing;
1022 static inline u64 global_rt_period(void)
1024 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1027 static inline u64 global_rt_runtime(void)
1029 if (sysctl_sched_rt_runtime < 0)
1032 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1035 static inline int task_current(struct rq *rq, struct task_struct *p)
1037 return rq->curr == p;
1040 static inline int task_running(struct rq *rq, struct task_struct *p)
1045 return task_current(rq, p);
1049 static inline int task_on_rq_queued(struct task_struct *p)
1051 return p->on_rq == TASK_ON_RQ_QUEUED;
1054 static inline int task_on_rq_migrating(struct task_struct *p)
1056 return p->on_rq == TASK_ON_RQ_MIGRATING;
1059 #ifndef prepare_arch_switch
1060 # define prepare_arch_switch(next) do { } while (0)
1062 #ifndef finish_arch_post_lock_switch
1063 # define finish_arch_post_lock_switch() do { } while (0)
1066 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1070 * We can optimise this out completely for !SMP, because the
1071 * SMP rebalancing from interrupt is the only thing that cares
1078 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1082 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1083 * We must ensure this doesn't happen until the switch is completely
1086 * In particular, the load of prev->state in finish_task_switch() must
1087 * happen before this.
1089 * Pairs with the control dependency and rmb in try_to_wake_up().
1091 smp_store_release(&prev->on_cpu, 0);
1093 #ifdef CONFIG_DEBUG_SPINLOCK
1094 /* this is a valid case when another task releases the spinlock */
1095 rq->lock.owner = current;
1098 * If we are tracking spinlock dependencies then we have to
1099 * fix up the runqueue lock - which gets 'carried over' from
1100 * prev into current:
1102 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1104 raw_spin_unlock_irq(&rq->lock);
1110 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1111 #define WF_FORK 0x02 /* child wakeup after fork */
1112 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1115 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1116 * of tasks with abnormal "nice" values across CPUs the contribution that
1117 * each task makes to its run queue's load is weighted according to its
1118 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1119 * scaled version of the new time slice allocation that they receive on time
1123 #define WEIGHT_IDLEPRIO 3
1124 #define WMULT_IDLEPRIO 1431655765
1127 * Nice levels are multiplicative, with a gentle 10% change for every
1128 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1129 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1130 * that remained on nice 0.
1132 * The "10% effect" is relative and cumulative: from _any_ nice level,
1133 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1134 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1135 * If a task goes up by ~10% and another task goes down by ~10% then
1136 * the relative distance between them is ~25%.)
1138 static const int prio_to_weight[40] = {
1139 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1140 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1141 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1142 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1143 /* 0 */ 1024, 820, 655, 526, 423,
1144 /* 5 */ 335, 272, 215, 172, 137,
1145 /* 10 */ 110, 87, 70, 56, 45,
1146 /* 15 */ 36, 29, 23, 18, 15,
1150 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1152 * In cases where the weight does not change often, we can use the
1153 * precalculated inverse to speed up arithmetics by turning divisions
1154 * into multiplications:
1156 static const u32 prio_to_wmult[40] = {
1157 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1158 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1159 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1160 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1161 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1162 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1163 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1164 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1167 #define ENQUEUE_WAKEUP 0x01
1168 #define ENQUEUE_HEAD 0x02
1170 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1172 #define ENQUEUE_WAKING 0x00
1174 #define ENQUEUE_REPLENISH 0x08
1175 #define ENQUEUE_RESTORE 0x10
1177 #define DEQUEUE_SLEEP 0x01
1178 #define DEQUEUE_SAVE 0x02
1180 #define RETRY_TASK ((void *)-1UL)
1182 struct sched_class {
1183 const struct sched_class *next;
1185 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1186 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1187 void (*yield_task) (struct rq *rq);
1188 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1190 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1193 * It is the responsibility of the pick_next_task() method that will
1194 * return the next task to call put_prev_task() on the @prev task or
1195 * something equivalent.
1197 * May return RETRY_TASK when it finds a higher prio class has runnable
1200 struct task_struct * (*pick_next_task) (struct rq *rq,
1201 struct task_struct *prev);
1202 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1205 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1206 void (*migrate_task_rq)(struct task_struct *p);
1208 void (*task_waking) (struct task_struct *task);
1209 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1211 void (*set_cpus_allowed)(struct task_struct *p,
1212 const struct cpumask *newmask);
1214 void (*rq_online)(struct rq *rq);
1215 void (*rq_offline)(struct rq *rq);
1218 void (*set_curr_task) (struct rq *rq);
1219 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1220 void (*task_fork) (struct task_struct *p);
1221 void (*task_dead) (struct task_struct *p);
1224 * The switched_from() call is allowed to drop rq->lock, therefore we
1225 * cannot assume the switched_from/switched_to pair is serliazed by
1226 * rq->lock. They are however serialized by p->pi_lock.
1228 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1229 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1230 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1233 unsigned int (*get_rr_interval) (struct rq *rq,
1234 struct task_struct *task);
1236 void (*update_curr) (struct rq *rq);
1238 #ifdef CONFIG_FAIR_GROUP_SCHED
1239 void (*task_move_group) (struct task_struct *p);
1243 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1245 prev->sched_class->put_prev_task(rq, prev);
1248 #define sched_class_highest (&stop_sched_class)
1249 #define for_each_class(class) \
1250 for (class = sched_class_highest; class; class = class->next)
1252 extern const struct sched_class stop_sched_class;
1253 extern const struct sched_class dl_sched_class;
1254 extern const struct sched_class rt_sched_class;
1255 extern const struct sched_class fair_sched_class;
1256 extern const struct sched_class idle_sched_class;
1261 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1263 extern void trigger_load_balance(struct rq *rq);
1265 extern void idle_enter_fair(struct rq *this_rq);
1266 extern void idle_exit_fair(struct rq *this_rq);
1268 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1272 static inline void idle_enter_fair(struct rq *rq) { }
1273 static inline void idle_exit_fair(struct rq *rq) { }
1277 #ifdef CONFIG_CPU_IDLE
1278 static inline void idle_set_state(struct rq *rq,
1279 struct cpuidle_state *idle_state)
1281 rq->idle_state = idle_state;
1284 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1286 WARN_ON(!rcu_read_lock_held());
1287 return rq->idle_state;
1290 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1292 rq->idle_state_idx = idle_state_idx;
1295 static inline int idle_get_state_idx(struct rq *rq)
1297 WARN_ON(!rcu_read_lock_held());
1298 return rq->idle_state_idx;
1301 static inline void idle_set_state(struct rq *rq,
1302 struct cpuidle_state *idle_state)
1306 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1311 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1315 static inline int idle_get_state_idx(struct rq *rq)
1321 extern void sysrq_sched_debug_show(void);
1322 extern void sched_init_granularity(void);
1323 extern void update_max_interval(void);
1325 extern void init_sched_dl_class(void);
1326 extern void init_sched_rt_class(void);
1327 extern void init_sched_fair_class(void);
1329 extern void resched_curr(struct rq *rq);
1330 extern void resched_cpu(int cpu);
1332 extern struct rt_bandwidth def_rt_bandwidth;
1333 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1335 extern struct dl_bandwidth def_dl_bandwidth;
1336 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1337 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1339 unsigned long to_ratio(u64 period, u64 runtime);
1341 extern void init_entity_runnable_average(struct sched_entity *se);
1343 static inline void add_nr_running(struct rq *rq, unsigned count)
1345 unsigned prev_nr = rq->nr_running;
1347 rq->nr_running = prev_nr + count;
1349 if (prev_nr < 2 && rq->nr_running >= 2) {
1351 if (!rq->rd->overload)
1352 rq->rd->overload = true;
1355 #ifdef CONFIG_NO_HZ_FULL
1356 if (tick_nohz_full_cpu(rq->cpu)) {
1358 * Tick is needed if more than one task runs on a CPU.
1359 * Send the target an IPI to kick it out of nohz mode.
1361 * We assume that IPI implies full memory barrier and the
1362 * new value of rq->nr_running is visible on reception
1365 tick_nohz_full_kick_cpu(rq->cpu);
1371 static inline void sub_nr_running(struct rq *rq, unsigned count)
1373 rq->nr_running -= count;
1376 static inline void rq_last_tick_reset(struct rq *rq)
1378 #ifdef CONFIG_NO_HZ_FULL
1379 rq->last_sched_tick = jiffies;
1383 extern void update_rq_clock(struct rq *rq);
1385 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1386 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1388 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1390 extern const_debug unsigned int sysctl_sched_time_avg;
1391 extern const_debug unsigned int sysctl_sched_nr_migrate;
1392 extern const_debug unsigned int sysctl_sched_migration_cost;
1394 static inline u64 sched_avg_period(void)
1396 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1399 #ifdef CONFIG_SCHED_HRTICK
1403 * - enabled by features
1404 * - hrtimer is actually high res
1406 static inline int hrtick_enabled(struct rq *rq)
1408 if (!sched_feat(HRTICK))
1410 if (!cpu_active(cpu_of(rq)))
1412 return hrtimer_is_hres_active(&rq->hrtick_timer);
1415 void hrtick_start(struct rq *rq, u64 delay);
1419 static inline int hrtick_enabled(struct rq *rq)
1424 #endif /* CONFIG_SCHED_HRTICK */
1427 extern void sched_avg_update(struct rq *rq);
1429 #ifndef arch_scale_freq_capacity
1430 static __always_inline
1431 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1433 return SCHED_CAPACITY_SCALE;
1437 #ifndef arch_scale_cpu_capacity
1438 static __always_inline
1439 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1441 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1442 return sd->smt_gain / sd->span_weight;
1444 return SCHED_CAPACITY_SCALE;
1448 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1450 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1451 sched_avg_update(rq);
1454 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1455 static inline void sched_avg_update(struct rq *rq) { }
1459 * __task_rq_lock - lock the rq @p resides on.
1461 static inline struct rq *__task_rq_lock(struct task_struct *p)
1462 __acquires(rq->lock)
1466 lockdep_assert_held(&p->pi_lock);
1470 raw_spin_lock(&rq->lock);
1471 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1472 lockdep_pin_lock(&rq->lock);
1475 raw_spin_unlock(&rq->lock);
1477 while (unlikely(task_on_rq_migrating(p)))
1483 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1485 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1486 __acquires(p->pi_lock)
1487 __acquires(rq->lock)
1492 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1494 raw_spin_lock(&rq->lock);
1496 * move_queued_task() task_rq_lock()
1498 * ACQUIRE (rq->lock)
1499 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1500 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1501 * [S] ->cpu = new_cpu [L] task_rq()
1503 * RELEASE (rq->lock)
1505 * If we observe the old cpu in task_rq_lock, the acquire of
1506 * the old rq->lock will fully serialize against the stores.
1508 * If we observe the new cpu in task_rq_lock, the acquire will
1509 * pair with the WMB to ensure we must then also see migrating.
1511 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1512 lockdep_pin_lock(&rq->lock);
1515 raw_spin_unlock(&rq->lock);
1516 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1518 while (unlikely(task_on_rq_migrating(p)))
1523 static inline void __task_rq_unlock(struct rq *rq)
1524 __releases(rq->lock)
1526 lockdep_unpin_lock(&rq->lock);
1527 raw_spin_unlock(&rq->lock);
1531 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1532 __releases(rq->lock)
1533 __releases(p->pi_lock)
1535 lockdep_unpin_lock(&rq->lock);
1536 raw_spin_unlock(&rq->lock);
1537 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1541 #ifdef CONFIG_PREEMPT
1543 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1546 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1547 * way at the expense of forcing extra atomic operations in all
1548 * invocations. This assures that the double_lock is acquired using the
1549 * same underlying policy as the spinlock_t on this architecture, which
1550 * reduces latency compared to the unfair variant below. However, it
1551 * also adds more overhead and therefore may reduce throughput.
1553 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1554 __releases(this_rq->lock)
1555 __acquires(busiest->lock)
1556 __acquires(this_rq->lock)
1558 raw_spin_unlock(&this_rq->lock);
1559 double_rq_lock(this_rq, busiest);
1566 * Unfair double_lock_balance: Optimizes throughput at the expense of
1567 * latency by eliminating extra atomic operations when the locks are
1568 * already in proper order on entry. This favors lower cpu-ids and will
1569 * grant the double lock to lower cpus over higher ids under contention,
1570 * regardless of entry order into the function.
1572 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1573 __releases(this_rq->lock)
1574 __acquires(busiest->lock)
1575 __acquires(this_rq->lock)
1579 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1580 if (busiest < this_rq) {
1581 raw_spin_unlock(&this_rq->lock);
1582 raw_spin_lock(&busiest->lock);
1583 raw_spin_lock_nested(&this_rq->lock,
1584 SINGLE_DEPTH_NESTING);
1587 raw_spin_lock_nested(&busiest->lock,
1588 SINGLE_DEPTH_NESTING);
1593 #endif /* CONFIG_PREEMPT */
1596 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1598 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1600 if (unlikely(!irqs_disabled())) {
1601 /* printk() doesn't work good under rq->lock */
1602 raw_spin_unlock(&this_rq->lock);
1606 return _double_lock_balance(this_rq, busiest);
1609 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1610 __releases(busiest->lock)
1612 raw_spin_unlock(&busiest->lock);
1613 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1616 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1622 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1625 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1631 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1634 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1640 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
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 ;) */
1659 raw_spin_lock(&rq1->lock);
1660 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1662 raw_spin_lock(&rq2->lock);
1663 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1669 * double_rq_unlock - safely unlock two runqueues
1671 * Note this does not restore interrupts like task_rq_unlock,
1672 * you need to do so manually after calling.
1674 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1675 __releases(rq1->lock)
1676 __releases(rq2->lock)
1678 raw_spin_unlock(&rq1->lock);
1680 raw_spin_unlock(&rq2->lock);
1682 __release(rq2->lock);
1685 #else /* CONFIG_SMP */
1688 * double_rq_lock - safely lock two runqueues
1690 * Note this does not disable interrupts like task_rq_lock,
1691 * you need to do so manually before calling.
1693 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1694 __acquires(rq1->lock)
1695 __acquires(rq2->lock)
1697 BUG_ON(!irqs_disabled());
1699 raw_spin_lock(&rq1->lock);
1700 __acquire(rq2->lock); /* Fake it out ;) */
1704 * double_rq_unlock - safely unlock two runqueues
1706 * Note this does not restore interrupts like task_rq_unlock,
1707 * you need to do so manually after calling.
1709 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1710 __releases(rq1->lock)
1711 __releases(rq2->lock)
1714 raw_spin_unlock(&rq1->lock);
1715 __release(rq2->lock);
1720 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1721 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1723 #ifdef CONFIG_SCHED_DEBUG
1724 extern void print_cfs_stats(struct seq_file *m, int cpu);
1725 extern void print_rt_stats(struct seq_file *m, int cpu);
1726 extern void print_dl_stats(struct seq_file *m, int cpu);
1728 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1730 #ifdef CONFIG_NUMA_BALANCING
1732 show_numa_stats(struct task_struct *p, struct seq_file *m);
1734 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1735 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1736 #endif /* CONFIG_NUMA_BALANCING */
1737 #endif /* CONFIG_SCHED_DEBUG */
1739 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1740 extern void init_rt_rq(struct rt_rq *rt_rq);
1741 extern void init_dl_rq(struct dl_rq *dl_rq);
1743 extern void cfs_bandwidth_usage_inc(void);
1744 extern void cfs_bandwidth_usage_dec(void);
1746 #ifdef CONFIG_NO_HZ_COMMON
1747 enum rq_nohz_flag_bits {
1752 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1755 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1757 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1758 DECLARE_PER_CPU(u64, cpu_softirq_time);
1760 #ifndef CONFIG_64BIT
1761 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1763 static inline void irq_time_write_begin(void)
1765 __this_cpu_inc(irq_time_seq.sequence);
1769 static inline void irq_time_write_end(void)
1772 __this_cpu_inc(irq_time_seq.sequence);
1775 static inline u64 irq_time_read(int cpu)
1781 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1782 irq_time = per_cpu(cpu_softirq_time, cpu) +
1783 per_cpu(cpu_hardirq_time, cpu);
1784 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1788 #else /* CONFIG_64BIT */
1789 static inline void irq_time_write_begin(void)
1793 static inline void irq_time_write_end(void)
1797 static inline u64 irq_time_read(int cpu)
1799 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1801 #endif /* CONFIG_64BIT */
1802 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1804 static inline void account_reset_rq(struct rq *rq)
1806 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1807 rq->prev_irq_time = 0;
1809 #ifdef CONFIG_PARAVIRT
1810 rq->prev_steal_time = 0;
1812 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1813 rq->prev_steal_time_rq = 0;