2 #include <linux/sched.h>
3 #include <linux/sched/autogroup.h>
4 #include <linux/sched/sysctl.h>
5 #include <linux/sched/topology.h>
6 #include <linux/sched/rt.h>
7 #include <linux/sched/deadline.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/wake_q.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/numa_balancing.h>
12 #include <linux/sched/mm.h>
13 #include <linux/sched/cpufreq.h>
14 #include <linux/sched/stat.h>
15 #include <linux/sched/nohz.h>
16 #include <linux/sched/debug.h>
17 #include <linux/sched/hotplug.h>
18 #include <linux/sched/task.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/sched/cputime.h>
22 #include <linux/u64_stats_sync.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/binfmts.h>
25 #include <linux/mutex.h>
26 #include <linux/spinlock.h>
27 #include <linux/stop_machine.h>
28 #include <linux/irq_work.h>
29 #include <linux/tick.h>
30 #include <linux/slab.h>
32 #ifdef CONFIG_PARAVIRT
33 #include <asm/paravirt.h>
37 #include "cpudeadline.h"
40 #ifdef CONFIG_SCHED_DEBUG
41 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
43 #define SCHED_WARN_ON(x) ((void)(x))
49 /* task_struct::on_rq states: */
50 #define TASK_ON_RQ_QUEUED 1
51 #define TASK_ON_RQ_MIGRATING 2
53 extern __read_mostly int scheduler_running;
55 extern unsigned long calc_load_update;
56 extern atomic_long_t calc_load_tasks;
58 extern void calc_global_load_tick(struct rq *this_rq);
59 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
62 extern void cpu_load_update_active(struct rq *this_rq);
64 static inline void cpu_load_update_active(struct rq *this_rq) { }
68 * Helpers for converting nanosecond timing to jiffy resolution
70 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
73 * Increase resolution of nice-level calculations for 64-bit architectures.
74 * The extra resolution improves shares distribution and load balancing of
75 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
76 * hierarchies, especially on larger systems. This is not a user-visible change
77 * and does not change the user-interface for setting shares/weights.
79 * We increase resolution only if we have enough bits to allow this increased
80 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
81 * pretty high and the returns do not justify the increased costs.
83 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
84 * increase coverage and consistency always enable it on 64bit platforms.
87 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
88 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
89 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
91 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
92 # define scale_load(w) (w)
93 # define scale_load_down(w) (w)
97 * Task weight (visible to users) and its load (invisible to users) have
98 * independent resolution, but they should be well calibrated. We use
99 * scale_load() and scale_load_down(w) to convert between them. The
100 * following must be true:
102 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
105 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
108 * Single value that decides SCHED_DEADLINE internal math precision.
109 * 10 -> just above 1us
110 * 9 -> just above 0.5us
112 #define DL_SCALE (10)
115 * These are the 'tuning knobs' of the scheduler:
119 * single value that denotes runtime == period, ie unlimited time.
121 #define RUNTIME_INF ((u64)~0ULL)
123 static inline int idle_policy(int policy)
125 return policy == SCHED_IDLE;
127 static inline int fair_policy(int policy)
129 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
132 static inline int rt_policy(int policy)
134 return policy == SCHED_FIFO || policy == SCHED_RR;
137 static inline int dl_policy(int policy)
139 return policy == SCHED_DEADLINE;
141 static inline bool valid_policy(int policy)
143 return idle_policy(policy) || fair_policy(policy) ||
144 rt_policy(policy) || dl_policy(policy);
147 static inline int task_has_rt_policy(struct task_struct *p)
149 return rt_policy(p->policy);
152 static inline int task_has_dl_policy(struct task_struct *p)
154 return dl_policy(p->policy);
158 * Tells if entity @a should preempt entity @b.
161 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
163 return dl_time_before(a->deadline, b->deadline);
167 * This is the priority-queue data structure of the RT scheduling class:
169 struct rt_prio_array {
170 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
171 struct list_head queue[MAX_RT_PRIO];
174 struct rt_bandwidth {
175 /* nests inside the rq lock: */
176 raw_spinlock_t rt_runtime_lock;
179 struct hrtimer rt_period_timer;
180 unsigned int rt_period_active;
183 void __dl_clear_params(struct task_struct *p);
186 * To keep the bandwidth of -deadline tasks and groups under control
187 * we need some place where:
188 * - store the maximum -deadline bandwidth of the system (the group);
189 * - cache the fraction of that bandwidth that is currently allocated.
191 * This is all done in the data structure below. It is similar to the
192 * one used for RT-throttling (rt_bandwidth), with the main difference
193 * that, since here we are only interested in admission control, we
194 * do not decrease any runtime while the group "executes", neither we
195 * need a timer to replenish it.
197 * With respect to SMP, the bandwidth is given on a per-CPU basis,
199 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
200 * - dl_total_bw array contains, in the i-eth element, the currently
201 * allocated bandwidth on the i-eth CPU.
202 * Moreover, groups consume bandwidth on each CPU, while tasks only
203 * consume bandwidth on the CPU they're running on.
204 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
205 * that will be shown the next time the proc or cgroup controls will
206 * be red. It on its turn can be changed by writing on its own
209 struct dl_bandwidth {
210 raw_spinlock_t dl_runtime_lock;
215 static inline int dl_bandwidth_enabled(void)
217 return sysctl_sched_rt_runtime >= 0;
220 extern struct dl_bw *dl_bw_of(int i);
228 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
230 dl_b->total_bw -= tsk_bw;
234 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
236 dl_b->total_bw += tsk_bw;
240 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
242 return dl_b->bw != -1 &&
243 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
246 extern void init_dl_bw(struct dl_bw *dl_b);
248 #ifdef CONFIG_CGROUP_SCHED
250 #include <linux/cgroup.h>
255 extern struct list_head task_groups;
257 struct cfs_bandwidth {
258 #ifdef CONFIG_CFS_BANDWIDTH
262 s64 hierarchical_quota;
265 int idle, period_active;
266 struct hrtimer period_timer, slack_timer;
267 struct list_head throttled_cfs_rq;
270 int nr_periods, nr_throttled;
275 /* task group related information */
277 struct cgroup_subsys_state css;
279 #ifdef CONFIG_FAIR_GROUP_SCHED
280 /* schedulable entities of this group on each cpu */
281 struct sched_entity **se;
282 /* runqueue "owned" by this group on each cpu */
283 struct cfs_rq **cfs_rq;
284 unsigned long shares;
288 * load_avg can be heavily contended at clock tick time, so put
289 * it in its own cacheline separated from the fields above which
290 * will also be accessed at each tick.
292 atomic_long_t load_avg ____cacheline_aligned;
296 #ifdef CONFIG_RT_GROUP_SCHED
297 struct sched_rt_entity **rt_se;
298 struct rt_rq **rt_rq;
300 struct rt_bandwidth rt_bandwidth;
304 struct list_head list;
306 struct task_group *parent;
307 struct list_head siblings;
308 struct list_head children;
310 #ifdef CONFIG_SCHED_AUTOGROUP
311 struct autogroup *autogroup;
314 struct cfs_bandwidth cfs_bandwidth;
317 #ifdef CONFIG_FAIR_GROUP_SCHED
318 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
321 * A weight of 0 or 1 can cause arithmetics problems.
322 * A weight of a cfs_rq is the sum of weights of which entities
323 * are queued on this cfs_rq, so a weight of a entity should not be
324 * too large, so as the shares value of a task group.
325 * (The default weight is 1024 - so there's no practical
326 * limitation from this.)
328 #define MIN_SHARES (1UL << 1)
329 #define MAX_SHARES (1UL << 18)
332 typedef int (*tg_visitor)(struct task_group *, void *);
334 extern int walk_tg_tree_from(struct task_group *from,
335 tg_visitor down, tg_visitor up, void *data);
338 * Iterate the full tree, calling @down when first entering a node and @up when
339 * leaving it for the final time.
341 * Caller must hold rcu_lock or sufficient equivalent.
343 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
345 return walk_tg_tree_from(&root_task_group, down, up, data);
348 extern int tg_nop(struct task_group *tg, void *data);
350 extern void free_fair_sched_group(struct task_group *tg);
351 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
352 extern void online_fair_sched_group(struct task_group *tg);
353 extern void unregister_fair_sched_group(struct task_group *tg);
354 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
355 struct sched_entity *se, int cpu,
356 struct sched_entity *parent);
357 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
359 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
360 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
361 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
363 extern void free_rt_sched_group(struct task_group *tg);
364 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
365 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
366 struct sched_rt_entity *rt_se, int cpu,
367 struct sched_rt_entity *parent);
369 extern struct task_group *sched_create_group(struct task_group *parent);
370 extern void sched_online_group(struct task_group *tg,
371 struct task_group *parent);
372 extern void sched_destroy_group(struct task_group *tg);
373 extern void sched_offline_group(struct task_group *tg);
375 extern void sched_move_task(struct task_struct *tsk);
377 #ifdef CONFIG_FAIR_GROUP_SCHED
378 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
381 extern void set_task_rq_fair(struct sched_entity *se,
382 struct cfs_rq *prev, struct cfs_rq *next);
383 #else /* !CONFIG_SMP */
384 static inline void set_task_rq_fair(struct sched_entity *se,
385 struct cfs_rq *prev, struct cfs_rq *next) { }
386 #endif /* CONFIG_SMP */
387 #endif /* CONFIG_FAIR_GROUP_SCHED */
389 #else /* CONFIG_CGROUP_SCHED */
391 struct cfs_bandwidth { };
393 #endif /* CONFIG_CGROUP_SCHED */
395 /* CFS-related fields in a runqueue */
397 struct load_weight load;
398 unsigned int nr_running, h_nr_running;
403 u64 min_vruntime_copy;
406 struct rb_root tasks_timeline;
407 struct rb_node *rb_leftmost;
410 * 'curr' points to currently running entity on this cfs_rq.
411 * It is set to NULL otherwise (i.e when none are currently running).
413 struct sched_entity *curr, *next, *last, *skip;
415 #ifdef CONFIG_SCHED_DEBUG
416 unsigned int nr_spread_over;
423 struct sched_avg avg;
424 u64 runnable_load_sum;
425 unsigned long runnable_load_avg;
426 #ifdef CONFIG_FAIR_GROUP_SCHED
427 unsigned long tg_load_avg_contrib;
428 unsigned long propagate_avg;
430 atomic_long_t removed_load_avg, removed_util_avg;
432 u64 load_last_update_time_copy;
435 #ifdef CONFIG_FAIR_GROUP_SCHED
437 * h_load = weight * f(tg)
439 * Where f(tg) is the recursive weight fraction assigned to
442 unsigned long h_load;
443 u64 last_h_load_update;
444 struct sched_entity *h_load_next;
445 #endif /* CONFIG_FAIR_GROUP_SCHED */
446 #endif /* CONFIG_SMP */
448 #ifdef CONFIG_FAIR_GROUP_SCHED
449 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
452 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
453 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
454 * (like users, containers etc.)
456 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
457 * list is used during load balance.
460 struct list_head leaf_cfs_rq_list;
461 struct task_group *tg; /* group that "owns" this runqueue */
463 #ifdef CONFIG_CFS_BANDWIDTH
466 s64 runtime_remaining;
468 u64 throttled_clock, throttled_clock_task;
469 u64 throttled_clock_task_time;
470 int throttled, throttle_count;
471 struct list_head throttled_list;
472 #endif /* CONFIG_CFS_BANDWIDTH */
473 #endif /* CONFIG_FAIR_GROUP_SCHED */
476 static inline int rt_bandwidth_enabled(void)
478 return sysctl_sched_rt_runtime >= 0;
481 /* RT IPI pull logic requires IRQ_WORK */
482 #ifdef CONFIG_IRQ_WORK
483 # define HAVE_RT_PUSH_IPI
486 /* Real-Time classes' related field in a runqueue: */
488 struct rt_prio_array active;
489 unsigned int rt_nr_running;
490 unsigned int rr_nr_running;
491 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
493 int curr; /* highest queued rt task prio */
495 int next; /* next highest */
500 unsigned long rt_nr_migratory;
501 unsigned long rt_nr_total;
503 struct plist_head pushable_tasks;
504 #ifdef HAVE_RT_PUSH_IPI
507 struct irq_work push_work;
508 raw_spinlock_t push_lock;
510 #endif /* CONFIG_SMP */
516 /* Nests inside the rq lock: */
517 raw_spinlock_t rt_runtime_lock;
519 #ifdef CONFIG_RT_GROUP_SCHED
520 unsigned long rt_nr_boosted;
523 struct task_group *tg;
527 /* Deadline class' related fields in a runqueue */
529 /* runqueue is an rbtree, ordered by deadline */
530 struct rb_root rb_root;
531 struct rb_node *rb_leftmost;
533 unsigned long dl_nr_running;
537 * Deadline values of the currently executing and the
538 * earliest ready task on this rq. Caching these facilitates
539 * the decision wether or not a ready but not running task
540 * should migrate somewhere else.
547 unsigned long dl_nr_migratory;
551 * Tasks on this rq that can be pushed away. They are kept in
552 * an rb-tree, ordered by tasks' deadlines, with caching
553 * of the leftmost (earliest deadline) element.
555 struct rb_root pushable_dl_tasks_root;
556 struct rb_node *pushable_dl_tasks_leftmost;
564 static inline bool sched_asym_prefer(int a, int b)
566 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
570 * We add the notion of a root-domain which will be used to define per-domain
571 * variables. Each exclusive cpuset essentially defines an island domain by
572 * fully partitioning the member cpus from any other cpuset. Whenever a new
573 * exclusive cpuset is created, we also create and attach a new root-domain
582 cpumask_var_t online;
584 /* Indicate more than one runnable task for any CPU */
588 * The bit corresponding to a CPU gets set here if such CPU has more
589 * than one runnable -deadline task (as it is below for RT tasks).
591 cpumask_var_t dlo_mask;
597 * The "RT overload" flag: it gets set if a CPU has more than
598 * one runnable RT task.
600 cpumask_var_t rto_mask;
601 struct cpupri cpupri;
603 unsigned long max_cpu_capacity;
606 extern struct root_domain def_root_domain;
607 extern struct mutex sched_domains_mutex;
608 extern cpumask_var_t fallback_doms;
609 extern cpumask_var_t sched_domains_tmpmask;
611 extern void init_defrootdomain(void);
612 extern int init_sched_domains(const struct cpumask *cpu_map);
613 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
615 #endif /* CONFIG_SMP */
618 * This is the main, per-CPU runqueue data structure.
620 * Locking rule: those places that want to lock multiple runqueues
621 * (such as the load balancing or the thread migration code), lock
622 * acquire operations must be ordered by ascending &runqueue.
629 * nr_running and cpu_load should be in the same cacheline because
630 * remote CPUs use both these fields when doing load calculation.
632 unsigned int nr_running;
633 #ifdef CONFIG_NUMA_BALANCING
634 unsigned int nr_numa_running;
635 unsigned int nr_preferred_running;
637 #define CPU_LOAD_IDX_MAX 5
638 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
639 #ifdef CONFIG_NO_HZ_COMMON
641 unsigned long last_load_update_tick;
642 #endif /* CONFIG_SMP */
643 unsigned long nohz_flags;
644 #endif /* CONFIG_NO_HZ_COMMON */
645 #ifdef CONFIG_NO_HZ_FULL
646 unsigned long last_sched_tick;
648 /* capture load from *all* tasks on this cpu: */
649 struct load_weight load;
650 unsigned long nr_load_updates;
657 #ifdef CONFIG_FAIR_GROUP_SCHED
658 /* list of leaf cfs_rq on this cpu: */
659 struct list_head leaf_cfs_rq_list;
660 struct list_head *tmp_alone_branch;
661 #endif /* CONFIG_FAIR_GROUP_SCHED */
664 * This is part of a global counter where only the total sum
665 * over all CPUs matters. A task can increase this counter on
666 * one CPU and if it got migrated afterwards it may decrease
667 * it on another CPU. Always updated under the runqueue lock:
669 unsigned long nr_uninterruptible;
671 struct task_struct *curr, *idle, *stop;
672 unsigned long next_balance;
673 struct mm_struct *prev_mm;
675 unsigned int clock_update_flags;
682 struct root_domain *rd;
683 struct sched_domain *sd;
685 unsigned long cpu_capacity;
686 unsigned long cpu_capacity_orig;
688 struct callback_head *balance_callback;
690 unsigned char idle_balance;
691 /* For active balancing */
694 struct cpu_stop_work active_balance_work;
695 /* cpu of this runqueue: */
699 struct list_head cfs_tasks;
706 /* This is used to determine avg_idle's max value */
707 u64 max_idle_balance_cost;
710 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
713 #ifdef CONFIG_PARAVIRT
716 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
717 u64 prev_steal_time_rq;
720 /* calc_load related fields */
721 unsigned long calc_load_update;
722 long calc_load_active;
724 #ifdef CONFIG_SCHED_HRTICK
726 int hrtick_csd_pending;
727 struct call_single_data hrtick_csd;
729 struct hrtimer hrtick_timer;
732 #ifdef CONFIG_SCHEDSTATS
734 struct sched_info rq_sched_info;
735 unsigned long long rq_cpu_time;
736 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
738 /* sys_sched_yield() stats */
739 unsigned int yld_count;
741 /* schedule() stats */
742 unsigned int sched_count;
743 unsigned int sched_goidle;
745 /* try_to_wake_up() stats */
746 unsigned int ttwu_count;
747 unsigned int ttwu_local;
751 struct llist_head wake_list;
754 #ifdef CONFIG_CPU_IDLE
755 /* Must be inspected within a rcu lock section */
756 struct cpuidle_state *idle_state;
760 static inline int cpu_of(struct rq *rq)
770 #ifdef CONFIG_SCHED_SMT
772 extern struct static_key_false sched_smt_present;
774 extern void __update_idle_core(struct rq *rq);
776 static inline void update_idle_core(struct rq *rq)
778 if (static_branch_unlikely(&sched_smt_present))
779 __update_idle_core(rq);
783 static inline void update_idle_core(struct rq *rq) { }
786 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
788 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
789 #define this_rq() this_cpu_ptr(&runqueues)
790 #define task_rq(p) cpu_rq(task_cpu(p))
791 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
792 #define raw_rq() raw_cpu_ptr(&runqueues)
794 static inline u64 __rq_clock_broken(struct rq *rq)
796 return READ_ONCE(rq->clock);
800 * rq::clock_update_flags bits
802 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
803 * call to __schedule(). This is an optimisation to avoid
804 * neighbouring rq clock updates.
806 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
807 * in effect and calls to update_rq_clock() are being ignored.
809 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
810 * made to update_rq_clock() since the last time rq::lock was pinned.
812 * If inside of __schedule(), clock_update_flags will have been
813 * shifted left (a left shift is a cheap operation for the fast path
814 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
816 * if (rq-clock_update_flags >= RQCF_UPDATED)
818 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
819 * one position though, because the next rq_unpin_lock() will shift it
822 #define RQCF_REQ_SKIP 0x01
823 #define RQCF_ACT_SKIP 0x02
824 #define RQCF_UPDATED 0x04
826 static inline void assert_clock_updated(struct rq *rq)
829 * The only reason for not seeing a clock update since the
830 * last rq_pin_lock() is if we're currently skipping updates.
832 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
835 static inline u64 rq_clock(struct rq *rq)
837 lockdep_assert_held(&rq->lock);
838 assert_clock_updated(rq);
843 static inline u64 rq_clock_task(struct rq *rq)
845 lockdep_assert_held(&rq->lock);
846 assert_clock_updated(rq);
848 return rq->clock_task;
851 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
853 lockdep_assert_held(&rq->lock);
855 rq->clock_update_flags |= RQCF_REQ_SKIP;
857 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
862 struct pin_cookie cookie;
863 #ifdef CONFIG_SCHED_DEBUG
865 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
866 * current pin context is stashed here in case it needs to be
867 * restored in rq_repin_lock().
869 unsigned int clock_update_flags;
873 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
875 rf->cookie = lockdep_pin_lock(&rq->lock);
877 #ifdef CONFIG_SCHED_DEBUG
878 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
879 rf->clock_update_flags = 0;
883 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
885 #ifdef CONFIG_SCHED_DEBUG
886 if (rq->clock_update_flags > RQCF_ACT_SKIP)
887 rf->clock_update_flags = RQCF_UPDATED;
890 lockdep_unpin_lock(&rq->lock, rf->cookie);
893 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
895 lockdep_repin_lock(&rq->lock, rf->cookie);
897 #ifdef CONFIG_SCHED_DEBUG
899 * Restore the value we stashed in @rf for this pin context.
901 rq->clock_update_flags |= rf->clock_update_flags;
906 enum numa_topology_type {
911 extern enum numa_topology_type sched_numa_topology_type;
912 extern int sched_max_numa_distance;
913 extern bool find_numa_distance(int distance);
917 extern void sched_init_numa(void);
918 extern void sched_domains_numa_masks_set(unsigned int cpu);
919 extern void sched_domains_numa_masks_clear(unsigned int cpu);
921 static inline void sched_init_numa(void) { }
922 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
923 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
926 #ifdef CONFIG_NUMA_BALANCING
927 /* The regions in numa_faults array from task_struct */
928 enum numa_faults_stats {
934 extern void sched_setnuma(struct task_struct *p, int node);
935 extern int migrate_task_to(struct task_struct *p, int cpu);
936 extern int migrate_swap(struct task_struct *, struct task_struct *);
937 #endif /* CONFIG_NUMA_BALANCING */
942 queue_balance_callback(struct rq *rq,
943 struct callback_head *head,
944 void (*func)(struct rq *rq))
946 lockdep_assert_held(&rq->lock);
948 if (unlikely(head->next))
951 head->func = (void (*)(struct callback_head *))func;
952 head->next = rq->balance_callback;
953 rq->balance_callback = head;
956 extern void sched_ttwu_pending(void);
958 #define rcu_dereference_check_sched_domain(p) \
959 rcu_dereference_check((p), \
960 lockdep_is_held(&sched_domains_mutex))
963 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
964 * See detach_destroy_domains: synchronize_sched for details.
966 * The domain tree of any CPU may only be accessed from within
967 * preempt-disabled sections.
969 #define for_each_domain(cpu, __sd) \
970 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
971 __sd; __sd = __sd->parent)
973 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
976 * highest_flag_domain - Return highest sched_domain containing flag.
977 * @cpu: The cpu whose highest level of sched domain is to
979 * @flag: The flag to check for the highest sched_domain
982 * Returns the highest sched_domain of a cpu which contains the given flag.
984 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
986 struct sched_domain *sd, *hsd = NULL;
988 for_each_domain(cpu, sd) {
989 if (!(sd->flags & flag))
997 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
999 struct sched_domain *sd;
1001 for_each_domain(cpu, sd) {
1002 if (sd->flags & flag)
1009 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1010 DECLARE_PER_CPU(int, sd_llc_size);
1011 DECLARE_PER_CPU(int, sd_llc_id);
1012 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1013 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1014 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1016 struct sched_group_capacity {
1019 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1022 unsigned long capacity;
1023 unsigned long min_capacity; /* Min per-CPU capacity in group */
1024 unsigned long next_update;
1025 int imbalance; /* XXX unrelated to capacity but shared group state */
1027 unsigned long cpumask[0]; /* iteration mask */
1030 struct sched_group {
1031 struct sched_group *next; /* Must be a circular list */
1034 unsigned int group_weight;
1035 struct sched_group_capacity *sgc;
1036 int asym_prefer_cpu; /* cpu of highest priority in group */
1039 * The CPUs this group covers.
1041 * NOTE: this field is variable length. (Allocated dynamically
1042 * by attaching extra space to the end of the structure,
1043 * depending on how many CPUs the kernel has booted up with)
1045 unsigned long cpumask[0];
1048 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
1050 return to_cpumask(sg->cpumask);
1054 * cpumask masking which cpus in the group are allowed to iterate up the domain
1057 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
1059 return to_cpumask(sg->sgc->cpumask);
1063 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1064 * @group: The group whose first cpu is to be returned.
1066 static inline unsigned int group_first_cpu(struct sched_group *group)
1068 return cpumask_first(sched_group_cpus(group));
1071 extern int group_balance_cpu(struct sched_group *sg);
1073 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1074 void register_sched_domain_sysctl(void);
1075 void unregister_sched_domain_sysctl(void);
1077 static inline void register_sched_domain_sysctl(void)
1080 static inline void unregister_sched_domain_sysctl(void)
1087 static inline void sched_ttwu_pending(void) { }
1089 #endif /* CONFIG_SMP */
1092 #include "autogroup.h"
1094 #ifdef CONFIG_CGROUP_SCHED
1097 * Return the group to which this tasks belongs.
1099 * We cannot use task_css() and friends because the cgroup subsystem
1100 * changes that value before the cgroup_subsys::attach() method is called,
1101 * therefore we cannot pin it and might observe the wrong value.
1103 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1104 * core changes this before calling sched_move_task().
1106 * Instead we use a 'copy' which is updated from sched_move_task() while
1107 * holding both task_struct::pi_lock and rq::lock.
1109 static inline struct task_group *task_group(struct task_struct *p)
1111 return p->sched_task_group;
1114 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1115 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1117 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1118 struct task_group *tg = task_group(p);
1121 #ifdef CONFIG_FAIR_GROUP_SCHED
1122 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1123 p->se.cfs_rq = tg->cfs_rq[cpu];
1124 p->se.parent = tg->se[cpu];
1127 #ifdef CONFIG_RT_GROUP_SCHED
1128 p->rt.rt_rq = tg->rt_rq[cpu];
1129 p->rt.parent = tg->rt_se[cpu];
1133 #else /* CONFIG_CGROUP_SCHED */
1135 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1136 static inline struct task_group *task_group(struct task_struct *p)
1141 #endif /* CONFIG_CGROUP_SCHED */
1143 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1145 set_task_rq(p, cpu);
1148 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1149 * successfuly executed on another CPU. We must ensure that updates of
1150 * per-task data have been completed by this moment.
1153 #ifdef CONFIG_THREAD_INFO_IN_TASK
1156 task_thread_info(p)->cpu = cpu;
1163 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1165 #ifdef CONFIG_SCHED_DEBUG
1166 # include <linux/static_key.h>
1167 # define const_debug __read_mostly
1169 # define const_debug const
1172 extern const_debug unsigned int sysctl_sched_features;
1174 #define SCHED_FEAT(name, enabled) \
1175 __SCHED_FEAT_##name ,
1178 #include "features.h"
1184 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1185 #define SCHED_FEAT(name, enabled) \
1186 static __always_inline bool static_branch_##name(struct static_key *key) \
1188 return static_key_##enabled(key); \
1191 #include "features.h"
1195 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1196 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1197 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1198 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1199 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1201 extern struct static_key_false sched_numa_balancing;
1202 extern struct static_key_false sched_schedstats;
1204 static inline u64 global_rt_period(void)
1206 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1209 static inline u64 global_rt_runtime(void)
1211 if (sysctl_sched_rt_runtime < 0)
1214 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1217 static inline int task_current(struct rq *rq, struct task_struct *p)
1219 return rq->curr == p;
1222 static inline int task_running(struct rq *rq, struct task_struct *p)
1227 return task_current(rq, p);
1231 static inline int task_on_rq_queued(struct task_struct *p)
1233 return p->on_rq == TASK_ON_RQ_QUEUED;
1236 static inline int task_on_rq_migrating(struct task_struct *p)
1238 return p->on_rq == TASK_ON_RQ_MIGRATING;
1241 #ifndef prepare_arch_switch
1242 # define prepare_arch_switch(next) do { } while (0)
1244 #ifndef finish_arch_post_lock_switch
1245 # define finish_arch_post_lock_switch() do { } while (0)
1248 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1252 * We can optimise this out completely for !SMP, because the
1253 * SMP rebalancing from interrupt is the only thing that cares
1260 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1264 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1265 * We must ensure this doesn't happen until the switch is completely
1268 * In particular, the load of prev->state in finish_task_switch() must
1269 * happen before this.
1271 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1273 smp_store_release(&prev->on_cpu, 0);
1275 #ifdef CONFIG_DEBUG_SPINLOCK
1276 /* this is a valid case when another task releases the spinlock */
1277 rq->lock.owner = current;
1280 * If we are tracking spinlock dependencies then we have to
1281 * fix up the runqueue lock - which gets 'carried over' from
1282 * prev into current:
1284 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1286 raw_spin_unlock_irq(&rq->lock);
1292 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1293 #define WF_FORK 0x02 /* child wakeup after fork */
1294 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1297 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1298 * of tasks with abnormal "nice" values across CPUs the contribution that
1299 * each task makes to its run queue's load is weighted according to its
1300 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1301 * scaled version of the new time slice allocation that they receive on time
1305 #define WEIGHT_IDLEPRIO 3
1306 #define WMULT_IDLEPRIO 1431655765
1308 extern const int sched_prio_to_weight[40];
1309 extern const u32 sched_prio_to_wmult[40];
1312 * {de,en}queue flags:
1314 * DEQUEUE_SLEEP - task is no longer runnable
1315 * ENQUEUE_WAKEUP - task just became runnable
1317 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1318 * are in a known state which allows modification. Such pairs
1319 * should preserve as much state as possible.
1321 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1324 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1325 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1326 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1330 #define DEQUEUE_SLEEP 0x01
1331 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1332 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1334 #define ENQUEUE_WAKEUP 0x01
1335 #define ENQUEUE_RESTORE 0x02
1336 #define ENQUEUE_MOVE 0x04
1338 #define ENQUEUE_HEAD 0x08
1339 #define ENQUEUE_REPLENISH 0x10
1341 #define ENQUEUE_MIGRATED 0x20
1343 #define ENQUEUE_MIGRATED 0x00
1346 #define RETRY_TASK ((void *)-1UL)
1348 struct sched_class {
1349 const struct sched_class *next;
1351 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1352 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1353 void (*yield_task) (struct rq *rq);
1354 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1356 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1359 * It is the responsibility of the pick_next_task() method that will
1360 * return the next task to call put_prev_task() on the @prev task or
1361 * something equivalent.
1363 * May return RETRY_TASK when it finds a higher prio class has runnable
1366 struct task_struct * (*pick_next_task) (struct rq *rq,
1367 struct task_struct *prev,
1368 struct rq_flags *rf);
1369 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1372 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1373 void (*migrate_task_rq)(struct task_struct *p);
1375 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1377 void (*set_cpus_allowed)(struct task_struct *p,
1378 const struct cpumask *newmask);
1380 void (*rq_online)(struct rq *rq);
1381 void (*rq_offline)(struct rq *rq);
1384 void (*set_curr_task) (struct rq *rq);
1385 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1386 void (*task_fork) (struct task_struct *p);
1387 void (*task_dead) (struct task_struct *p);
1390 * The switched_from() call is allowed to drop rq->lock, therefore we
1391 * cannot assume the switched_from/switched_to pair is serliazed by
1392 * rq->lock. They are however serialized by p->pi_lock.
1394 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1395 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1396 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1399 unsigned int (*get_rr_interval) (struct rq *rq,
1400 struct task_struct *task);
1402 void (*update_curr) (struct rq *rq);
1404 #define TASK_SET_GROUP 0
1405 #define TASK_MOVE_GROUP 1
1407 #ifdef CONFIG_FAIR_GROUP_SCHED
1408 void (*task_change_group) (struct task_struct *p, int type);
1412 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1414 prev->sched_class->put_prev_task(rq, prev);
1417 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1419 curr->sched_class->set_curr_task(rq);
1422 #define sched_class_highest (&stop_sched_class)
1423 #define for_each_class(class) \
1424 for (class = sched_class_highest; class; class = class->next)
1426 extern const struct sched_class stop_sched_class;
1427 extern const struct sched_class dl_sched_class;
1428 extern const struct sched_class rt_sched_class;
1429 extern const struct sched_class fair_sched_class;
1430 extern const struct sched_class idle_sched_class;
1435 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1437 extern void trigger_load_balance(struct rq *rq);
1439 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1443 #ifdef CONFIG_CPU_IDLE
1444 static inline void idle_set_state(struct rq *rq,
1445 struct cpuidle_state *idle_state)
1447 rq->idle_state = idle_state;
1450 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1452 SCHED_WARN_ON(!rcu_read_lock_held());
1453 return rq->idle_state;
1456 static inline void idle_set_state(struct rq *rq,
1457 struct cpuidle_state *idle_state)
1461 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1467 extern void sysrq_sched_debug_show(void);
1468 extern void sched_init_granularity(void);
1469 extern void update_max_interval(void);
1471 extern void init_sched_dl_class(void);
1472 extern void init_sched_rt_class(void);
1473 extern void init_sched_fair_class(void);
1475 extern void resched_curr(struct rq *rq);
1476 extern void resched_cpu(int cpu);
1478 extern struct rt_bandwidth def_rt_bandwidth;
1479 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1481 extern struct dl_bandwidth def_dl_bandwidth;
1482 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1483 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1485 unsigned long to_ratio(u64 period, u64 runtime);
1487 extern void init_entity_runnable_average(struct sched_entity *se);
1488 extern void post_init_entity_util_avg(struct sched_entity *se);
1490 #ifdef CONFIG_NO_HZ_FULL
1491 extern bool sched_can_stop_tick(struct rq *rq);
1494 * Tick may be needed by tasks in the runqueue depending on their policy and
1495 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1496 * nohz mode if necessary.
1498 static inline void sched_update_tick_dependency(struct rq *rq)
1502 if (!tick_nohz_full_enabled())
1507 if (!tick_nohz_full_cpu(cpu))
1510 if (sched_can_stop_tick(rq))
1511 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1513 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1516 static inline void sched_update_tick_dependency(struct rq *rq) { }
1519 static inline void add_nr_running(struct rq *rq, unsigned count)
1521 unsigned prev_nr = rq->nr_running;
1523 rq->nr_running = prev_nr + count;
1525 if (prev_nr < 2 && rq->nr_running >= 2) {
1527 if (!rq->rd->overload)
1528 rq->rd->overload = true;
1532 sched_update_tick_dependency(rq);
1535 static inline void sub_nr_running(struct rq *rq, unsigned count)
1537 rq->nr_running -= count;
1538 /* Check if we still need preemption */
1539 sched_update_tick_dependency(rq);
1542 static inline void rq_last_tick_reset(struct rq *rq)
1544 #ifdef CONFIG_NO_HZ_FULL
1545 rq->last_sched_tick = jiffies;
1549 extern void update_rq_clock(struct rq *rq);
1551 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1552 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1554 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1556 extern const_debug unsigned int sysctl_sched_time_avg;
1557 extern const_debug unsigned int sysctl_sched_nr_migrate;
1558 extern const_debug unsigned int sysctl_sched_migration_cost;
1560 static inline u64 sched_avg_period(void)
1562 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1565 #ifdef CONFIG_SCHED_HRTICK
1569 * - enabled by features
1570 * - hrtimer is actually high res
1572 static inline int hrtick_enabled(struct rq *rq)
1574 if (!sched_feat(HRTICK))
1576 if (!cpu_active(cpu_of(rq)))
1578 return hrtimer_is_hres_active(&rq->hrtick_timer);
1581 void hrtick_start(struct rq *rq, u64 delay);
1585 static inline int hrtick_enabled(struct rq *rq)
1590 #endif /* CONFIG_SCHED_HRTICK */
1593 extern void sched_avg_update(struct rq *rq);
1595 #ifndef arch_scale_freq_capacity
1596 static __always_inline
1597 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1599 return SCHED_CAPACITY_SCALE;
1603 #ifndef arch_scale_cpu_capacity
1604 static __always_inline
1605 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1607 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1608 return sd->smt_gain / sd->span_weight;
1610 return SCHED_CAPACITY_SCALE;
1614 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1616 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1617 sched_avg_update(rq);
1620 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1621 static inline void sched_avg_update(struct rq *rq) { }
1624 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1625 __acquires(rq->lock);
1626 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1627 __acquires(p->pi_lock)
1628 __acquires(rq->lock);
1630 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1631 __releases(rq->lock)
1633 rq_unpin_lock(rq, rf);
1634 raw_spin_unlock(&rq->lock);
1638 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1639 __releases(rq->lock)
1640 __releases(p->pi_lock)
1642 rq_unpin_lock(rq, rf);
1643 raw_spin_unlock(&rq->lock);
1644 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1648 #ifdef CONFIG_PREEMPT
1650 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1653 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1654 * way at the expense of forcing extra atomic operations in all
1655 * invocations. This assures that the double_lock is acquired using the
1656 * same underlying policy as the spinlock_t on this architecture, which
1657 * reduces latency compared to the unfair variant below. However, it
1658 * also adds more overhead and therefore may reduce throughput.
1660 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1661 __releases(this_rq->lock)
1662 __acquires(busiest->lock)
1663 __acquires(this_rq->lock)
1665 raw_spin_unlock(&this_rq->lock);
1666 double_rq_lock(this_rq, busiest);
1673 * Unfair double_lock_balance: Optimizes throughput at the expense of
1674 * latency by eliminating extra atomic operations when the locks are
1675 * already in proper order on entry. This favors lower cpu-ids and will
1676 * grant the double lock to lower cpus over higher ids under contention,
1677 * regardless of entry order into the function.
1679 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1680 __releases(this_rq->lock)
1681 __acquires(busiest->lock)
1682 __acquires(this_rq->lock)
1686 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1687 if (busiest < this_rq) {
1688 raw_spin_unlock(&this_rq->lock);
1689 raw_spin_lock(&busiest->lock);
1690 raw_spin_lock_nested(&this_rq->lock,
1691 SINGLE_DEPTH_NESTING);
1694 raw_spin_lock_nested(&busiest->lock,
1695 SINGLE_DEPTH_NESTING);
1700 #endif /* CONFIG_PREEMPT */
1703 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1705 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1707 if (unlikely(!irqs_disabled())) {
1708 /* printk() doesn't work good under rq->lock */
1709 raw_spin_unlock(&this_rq->lock);
1713 return _double_lock_balance(this_rq, busiest);
1716 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1717 __releases(busiest->lock)
1719 raw_spin_unlock(&busiest->lock);
1720 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1723 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1729 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1732 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1738 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1741 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1747 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1751 * double_rq_lock - safely lock two runqueues
1753 * Note this does not disable interrupts like task_rq_lock,
1754 * you need to do so manually before calling.
1756 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1757 __acquires(rq1->lock)
1758 __acquires(rq2->lock)
1760 BUG_ON(!irqs_disabled());
1762 raw_spin_lock(&rq1->lock);
1763 __acquire(rq2->lock); /* Fake it out ;) */
1766 raw_spin_lock(&rq1->lock);
1767 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1769 raw_spin_lock(&rq2->lock);
1770 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1776 * double_rq_unlock - safely unlock two runqueues
1778 * Note this does not restore interrupts like task_rq_unlock,
1779 * you need to do so manually after calling.
1781 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1782 __releases(rq1->lock)
1783 __releases(rq2->lock)
1785 raw_spin_unlock(&rq1->lock);
1787 raw_spin_unlock(&rq2->lock);
1789 __release(rq2->lock);
1792 extern void set_rq_online (struct rq *rq);
1793 extern void set_rq_offline(struct rq *rq);
1794 extern bool sched_smp_initialized;
1796 #else /* CONFIG_SMP */
1799 * double_rq_lock - safely lock two runqueues
1801 * Note this does not disable interrupts like task_rq_lock,
1802 * you need to do so manually before calling.
1804 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1805 __acquires(rq1->lock)
1806 __acquires(rq2->lock)
1808 BUG_ON(!irqs_disabled());
1810 raw_spin_lock(&rq1->lock);
1811 __acquire(rq2->lock); /* Fake it out ;) */
1815 * double_rq_unlock - safely unlock two runqueues
1817 * Note this does not restore interrupts like task_rq_unlock,
1818 * you need to do so manually after calling.
1820 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1821 __releases(rq1->lock)
1822 __releases(rq2->lock)
1825 raw_spin_unlock(&rq1->lock);
1826 __release(rq2->lock);
1831 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1832 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1834 #ifdef CONFIG_SCHED_DEBUG
1835 extern void print_cfs_stats(struct seq_file *m, int cpu);
1836 extern void print_rt_stats(struct seq_file *m, int cpu);
1837 extern void print_dl_stats(struct seq_file *m, int cpu);
1839 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1840 #ifdef CONFIG_NUMA_BALANCING
1842 show_numa_stats(struct task_struct *p, struct seq_file *m);
1844 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1845 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1846 #endif /* CONFIG_NUMA_BALANCING */
1847 #endif /* CONFIG_SCHED_DEBUG */
1849 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1850 extern void init_rt_rq(struct rt_rq *rt_rq);
1851 extern void init_dl_rq(struct dl_rq *dl_rq);
1853 extern void cfs_bandwidth_usage_inc(void);
1854 extern void cfs_bandwidth_usage_dec(void);
1856 #ifdef CONFIG_NO_HZ_COMMON
1857 enum rq_nohz_flag_bits {
1862 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1864 extern void nohz_balance_exit_idle(unsigned int cpu);
1866 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1869 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1873 struct u64_stats_sync sync;
1876 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
1878 static inline u64 irq_time_read(int cpu)
1880 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
1881 u64 *cpustat = kcpustat_cpu(cpu).cpustat;
1886 seq = __u64_stats_fetch_begin(&irqtime->sync);
1887 total = cpustat[CPUTIME_SOFTIRQ] + cpustat[CPUTIME_IRQ];
1888 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
1892 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1894 #ifdef CONFIG_CPU_FREQ
1895 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1898 * cpufreq_update_util - Take a note about CPU utilization changes.
1899 * @rq: Runqueue to carry out the update for.
1900 * @flags: Update reason flags.
1902 * This function is called by the scheduler on the CPU whose utilization is
1905 * It can only be called from RCU-sched read-side critical sections.
1907 * The way cpufreq is currently arranged requires it to evaluate the CPU
1908 * performance state (frequency/voltage) on a regular basis to prevent it from
1909 * being stuck in a completely inadequate performance level for too long.
1910 * That is not guaranteed to happen if the updates are only triggered from CFS,
1911 * though, because they may not be coming in if RT or deadline tasks are active
1912 * all the time (or there are RT and DL tasks only).
1914 * As a workaround for that issue, this function is called by the RT and DL
1915 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1916 * but that really is a band-aid. Going forward it should be replaced with
1917 * solutions targeted more specifically at RT and DL tasks.
1919 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
1921 struct update_util_data *data;
1923 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1925 data->func(data, rq_clock(rq), flags);
1928 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags)
1930 if (cpu_of(rq) == smp_processor_id())
1931 cpufreq_update_util(rq, flags);
1934 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
1935 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {}
1936 #endif /* CONFIG_CPU_FREQ */
1938 #ifdef arch_scale_freq_capacity
1939 #ifndef arch_scale_freq_invariant
1940 #define arch_scale_freq_invariant() (true)
1942 #else /* arch_scale_freq_capacity */
1943 #define arch_scale_freq_invariant() (false)