1 /* SPDX-License-Identifier: GPL-2.0 */
3 #include <linux/sched.h>
4 #include <linux/sched/autogroup.h>
5 #include <linux/sched/sysctl.h>
6 #include <linux/sched/topology.h>
7 #include <linux/sched/rt.h>
8 #include <linux/sched/deadline.h>
9 #include <linux/sched/clock.h>
10 #include <linux/sched/wake_q.h>
11 #include <linux/sched/signal.h>
12 #include <linux/sched/numa_balancing.h>
13 #include <linux/sched/mm.h>
14 #include <linux/sched/cpufreq.h>
15 #include <linux/sched/stat.h>
16 #include <linux/sched/nohz.h>
17 #include <linux/sched/debug.h>
18 #include <linux/sched/hotplug.h>
19 #include <linux/sched/task.h>
20 #include <linux/sched/task_stack.h>
21 #include <linux/sched/cputime.h>
22 #include <linux/sched/init.h>
24 #include <linux/u64_stats_sync.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/binfmts.h>
27 #include <linux/mutex.h>
28 #include <linux/spinlock.h>
29 #include <linux/stop_machine.h>
30 #include <linux/irq_work.h>
31 #include <linux/tick.h>
32 #include <linux/slab.h>
33 #include <linux/cgroup.h>
35 #ifdef CONFIG_PARAVIRT
36 #include <asm/paravirt.h>
40 #include "cpudeadline.h"
42 #ifdef CONFIG_SCHED_DEBUG
43 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
45 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
51 /* task_struct::on_rq states: */
52 #define TASK_ON_RQ_QUEUED 1
53 #define TASK_ON_RQ_MIGRATING 2
55 extern __read_mostly int scheduler_running;
57 extern unsigned long calc_load_update;
58 extern atomic_long_t calc_load_tasks;
60 extern void calc_global_load_tick(struct rq *this_rq);
61 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
64 extern void cpu_load_update_active(struct rq *this_rq);
66 static inline void cpu_load_update_active(struct rq *this_rq) { }
70 * Helpers for converting nanosecond timing to jiffy resolution
72 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
75 * Increase resolution of nice-level calculations for 64-bit architectures.
76 * The extra resolution improves shares distribution and load balancing of
77 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
78 * hierarchies, especially on larger systems. This is not a user-visible change
79 * and does not change the user-interface for setting shares/weights.
81 * We increase resolution only if we have enough bits to allow this increased
82 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
83 * pretty high and the returns do not justify the increased costs.
85 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
86 * increase coverage and consistency always enable it on 64bit platforms.
89 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
90 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
91 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
93 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
94 # define scale_load(w) (w)
95 # define scale_load_down(w) (w)
99 * Task weight (visible to users) and its load (invisible to users) have
100 * independent resolution, but they should be well calibrated. We use
101 * scale_load() and scale_load_down(w) to convert between them. The
102 * following must be true:
104 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
107 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
110 * Single value that decides SCHED_DEADLINE internal math precision.
111 * 10 -> just above 1us
112 * 9 -> just above 0.5us
114 #define DL_SCALE (10)
117 * These are the 'tuning knobs' of the scheduler:
121 * single value that denotes runtime == period, ie unlimited time.
123 #define RUNTIME_INF ((u64)~0ULL)
125 static inline int idle_policy(int policy)
127 return policy == SCHED_IDLE;
129 static inline int fair_policy(int policy)
131 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
134 static inline int rt_policy(int policy)
136 return policy == SCHED_FIFO || policy == SCHED_RR;
139 static inline int dl_policy(int policy)
141 return policy == SCHED_DEADLINE;
143 static inline bool valid_policy(int policy)
145 return idle_policy(policy) || fair_policy(policy) ||
146 rt_policy(policy) || dl_policy(policy);
149 static inline int task_has_rt_policy(struct task_struct *p)
151 return rt_policy(p->policy);
154 static inline int task_has_dl_policy(struct task_struct *p)
156 return dl_policy(p->policy);
159 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
162 * !! For sched_setattr_nocheck() (kernel) only !!
164 * This is actually gross. :(
166 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
167 * tasks, but still be able to sleep. We need this on platforms that cannot
168 * atomically change clock frequency. Remove once fast switching will be
169 * available on such platforms.
171 * SUGOV stands for SchedUtil GOVernor.
173 #define SCHED_FLAG_SUGOV 0x10000000
175 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
177 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
178 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
185 * Tells if entity @a should preempt entity @b.
188 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
190 return dl_entity_is_special(a) ||
191 dl_time_before(a->deadline, b->deadline);
195 * This is the priority-queue data structure of the RT scheduling class:
197 struct rt_prio_array {
198 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
199 struct list_head queue[MAX_RT_PRIO];
202 struct rt_bandwidth {
203 /* nests inside the rq lock: */
204 raw_spinlock_t rt_runtime_lock;
207 struct hrtimer rt_period_timer;
208 unsigned int rt_period_active;
211 void __dl_clear_params(struct task_struct *p);
214 * To keep the bandwidth of -deadline tasks and groups under control
215 * we need some place where:
216 * - store the maximum -deadline bandwidth of the system (the group);
217 * - cache the fraction of that bandwidth that is currently allocated.
219 * This is all done in the data structure below. It is similar to the
220 * one used for RT-throttling (rt_bandwidth), with the main difference
221 * that, since here we are only interested in admission control, we
222 * do not decrease any runtime while the group "executes", neither we
223 * need a timer to replenish it.
225 * With respect to SMP, the bandwidth is given on a per-CPU basis,
227 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
228 * - dl_total_bw array contains, in the i-eth element, the currently
229 * allocated bandwidth on the i-eth CPU.
230 * Moreover, groups consume bandwidth on each CPU, while tasks only
231 * consume bandwidth on the CPU they're running on.
232 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
233 * that will be shown the next time the proc or cgroup controls will
234 * be red. It on its turn can be changed by writing on its own
237 struct dl_bandwidth {
238 raw_spinlock_t dl_runtime_lock;
243 static inline int dl_bandwidth_enabled(void)
245 return sysctl_sched_rt_runtime >= 0;
253 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
256 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
258 dl_b->total_bw -= tsk_bw;
259 __dl_update(dl_b, (s32)tsk_bw / cpus);
263 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
265 dl_b->total_bw += tsk_bw;
266 __dl_update(dl_b, -((s32)tsk_bw / cpus));
270 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
272 return dl_b->bw != -1 &&
273 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
276 void dl_change_utilization(struct task_struct *p, u64 new_bw);
277 extern void init_dl_bw(struct dl_bw *dl_b);
278 extern int sched_dl_global_validate(void);
279 extern void sched_dl_do_global(void);
280 extern int sched_dl_overflow(struct task_struct *p, int policy,
281 const struct sched_attr *attr);
282 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
283 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
284 extern bool __checkparam_dl(const struct sched_attr *attr);
285 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
286 extern int dl_task_can_attach(struct task_struct *p,
287 const struct cpumask *cs_cpus_allowed);
288 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur,
289 const struct cpumask *trial);
290 extern bool dl_cpu_busy(unsigned int cpu);
292 #ifdef CONFIG_CGROUP_SCHED
294 #include <linux/cgroup.h>
299 extern struct list_head task_groups;
301 struct cfs_bandwidth {
302 #ifdef CONFIG_CFS_BANDWIDTH
306 s64 hierarchical_quota;
309 int idle, period_active;
310 struct hrtimer period_timer, slack_timer;
311 struct list_head throttled_cfs_rq;
314 int nr_periods, nr_throttled;
319 /* task group related information */
321 struct cgroup_subsys_state css;
323 #ifdef CONFIG_FAIR_GROUP_SCHED
324 /* schedulable entities of this group on each cpu */
325 struct sched_entity **se;
326 /* runqueue "owned" by this group on each cpu */
327 struct cfs_rq **cfs_rq;
328 unsigned long shares;
332 * load_avg can be heavily contended at clock tick time, so put
333 * it in its own cacheline separated from the fields above which
334 * will also be accessed at each tick.
336 atomic_long_t load_avg ____cacheline_aligned;
340 #ifdef CONFIG_RT_GROUP_SCHED
341 struct sched_rt_entity **rt_se;
342 struct rt_rq **rt_rq;
344 struct rt_bandwidth rt_bandwidth;
348 struct list_head list;
350 struct task_group *parent;
351 struct list_head siblings;
352 struct list_head children;
354 #ifdef CONFIG_SCHED_AUTOGROUP
355 struct autogroup *autogroup;
358 struct cfs_bandwidth cfs_bandwidth;
361 #ifdef CONFIG_FAIR_GROUP_SCHED
362 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
365 * A weight of 0 or 1 can cause arithmetics problems.
366 * A weight of a cfs_rq is the sum of weights of which entities
367 * are queued on this cfs_rq, so a weight of a entity should not be
368 * too large, so as the shares value of a task group.
369 * (The default weight is 1024 - so there's no practical
370 * limitation from this.)
372 #define MIN_SHARES (1UL << 1)
373 #define MAX_SHARES (1UL << 18)
376 typedef int (*tg_visitor)(struct task_group *, void *);
378 extern int walk_tg_tree_from(struct task_group *from,
379 tg_visitor down, tg_visitor up, void *data);
382 * Iterate the full tree, calling @down when first entering a node and @up when
383 * leaving it for the final time.
385 * Caller must hold rcu_lock or sufficient equivalent.
387 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
389 return walk_tg_tree_from(&root_task_group, down, up, data);
392 extern int tg_nop(struct task_group *tg, void *data);
394 extern void free_fair_sched_group(struct task_group *tg);
395 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
396 extern void online_fair_sched_group(struct task_group *tg);
397 extern void unregister_fair_sched_group(struct task_group *tg);
398 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
399 struct sched_entity *se, int cpu,
400 struct sched_entity *parent);
401 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
403 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
404 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
405 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
407 extern void free_rt_sched_group(struct task_group *tg);
408 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
409 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
410 struct sched_rt_entity *rt_se, int cpu,
411 struct sched_rt_entity *parent);
412 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
413 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
414 extern long sched_group_rt_runtime(struct task_group *tg);
415 extern long sched_group_rt_period(struct task_group *tg);
416 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
418 extern struct task_group *sched_create_group(struct task_group *parent);
419 extern void sched_online_group(struct task_group *tg,
420 struct task_group *parent);
421 extern void sched_destroy_group(struct task_group *tg);
422 extern void sched_offline_group(struct task_group *tg);
424 extern void sched_move_task(struct task_struct *tsk);
426 #ifdef CONFIG_FAIR_GROUP_SCHED
427 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
430 extern void set_task_rq_fair(struct sched_entity *se,
431 struct cfs_rq *prev, struct cfs_rq *next);
432 #else /* !CONFIG_SMP */
433 static inline void set_task_rq_fair(struct sched_entity *se,
434 struct cfs_rq *prev, struct cfs_rq *next) { }
435 #endif /* CONFIG_SMP */
436 #endif /* CONFIG_FAIR_GROUP_SCHED */
438 #else /* CONFIG_CGROUP_SCHED */
440 struct cfs_bandwidth { };
442 #endif /* CONFIG_CGROUP_SCHED */
444 /* CFS-related fields in a runqueue */
446 struct load_weight load;
447 unsigned long runnable_weight;
448 unsigned int nr_running, h_nr_running;
453 u64 min_vruntime_copy;
456 struct rb_root_cached tasks_timeline;
459 * 'curr' points to currently running entity on this cfs_rq.
460 * It is set to NULL otherwise (i.e when none are currently running).
462 struct sched_entity *curr, *next, *last, *skip;
464 #ifdef CONFIG_SCHED_DEBUG
465 unsigned int nr_spread_over;
472 struct sched_avg avg;
474 u64 load_last_update_time_copy;
477 raw_spinlock_t lock ____cacheline_aligned;
479 unsigned long load_avg;
480 unsigned long util_avg;
481 unsigned long runnable_sum;
484 #ifdef CONFIG_FAIR_GROUP_SCHED
485 unsigned long tg_load_avg_contrib;
487 long prop_runnable_sum;
490 * h_load = weight * f(tg)
492 * Where f(tg) is the recursive weight fraction assigned to
495 unsigned long h_load;
496 u64 last_h_load_update;
497 struct sched_entity *h_load_next;
498 #endif /* CONFIG_FAIR_GROUP_SCHED */
499 #endif /* CONFIG_SMP */
501 #ifdef CONFIG_FAIR_GROUP_SCHED
502 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
505 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
506 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
507 * (like users, containers etc.)
509 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
510 * list is used during load balance.
513 struct list_head leaf_cfs_rq_list;
514 struct task_group *tg; /* group that "owns" this runqueue */
516 #ifdef CONFIG_CFS_BANDWIDTH
519 s64 runtime_remaining;
521 u64 throttled_clock, throttled_clock_task;
522 u64 throttled_clock_task_time;
523 int throttled, throttle_count;
524 struct list_head throttled_list;
525 #endif /* CONFIG_CFS_BANDWIDTH */
526 #endif /* CONFIG_FAIR_GROUP_SCHED */
529 static inline int rt_bandwidth_enabled(void)
531 return sysctl_sched_rt_runtime >= 0;
534 /* RT IPI pull logic requires IRQ_WORK */
535 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
536 # define HAVE_RT_PUSH_IPI
539 /* Real-Time classes' related field in a runqueue: */
541 struct rt_prio_array active;
542 unsigned int rt_nr_running;
543 unsigned int rr_nr_running;
544 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
546 int curr; /* highest queued rt task prio */
548 int next; /* next highest */
553 unsigned long rt_nr_migratory;
554 unsigned long rt_nr_total;
556 struct plist_head pushable_tasks;
557 #endif /* CONFIG_SMP */
563 /* Nests inside the rq lock: */
564 raw_spinlock_t rt_runtime_lock;
566 #ifdef CONFIG_RT_GROUP_SCHED
567 unsigned long rt_nr_boosted;
570 struct task_group *tg;
574 /* Deadline class' related fields in a runqueue */
576 /* runqueue is an rbtree, ordered by deadline */
577 struct rb_root_cached root;
579 unsigned long dl_nr_running;
583 * Deadline values of the currently executing and the
584 * earliest ready task on this rq. Caching these facilitates
585 * the decision wether or not a ready but not running task
586 * should migrate somewhere else.
593 unsigned long dl_nr_migratory;
597 * Tasks on this rq that can be pushed away. They are kept in
598 * an rb-tree, ordered by tasks' deadlines, with caching
599 * of the leftmost (earliest deadline) element.
601 struct rb_root_cached pushable_dl_tasks_root;
606 * "Active utilization" for this runqueue: increased when a
607 * task wakes up (becomes TASK_RUNNING) and decreased when a
613 * Utilization of the tasks "assigned" to this runqueue (including
614 * the tasks that are in runqueue and the tasks that executed on this
615 * CPU and blocked). Increased when a task moves to this runqueue, and
616 * decreased when the task moves away (migrates, changes scheduling
617 * policy, or terminates).
618 * This is needed to compute the "inactive utilization" for the
619 * runqueue (inactive utilization = this_bw - running_bw).
625 * Inverse of the fraction of CPU utilization that can be reclaimed
626 * by the GRUB algorithm.
633 static inline bool sched_asym_prefer(int a, int b)
635 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
639 * We add the notion of a root-domain which will be used to define per-domain
640 * variables. Each exclusive cpuset essentially defines an island domain by
641 * fully partitioning the member cpus from any other cpuset. Whenever a new
642 * exclusive cpuset is created, we also create and attach a new root-domain
651 cpumask_var_t online;
653 /* Indicate more than one runnable task for any CPU */
657 * The bit corresponding to a CPU gets set here if such CPU has more
658 * than one runnable -deadline task (as it is below for RT tasks).
660 cpumask_var_t dlo_mask;
665 #ifdef HAVE_RT_PUSH_IPI
667 * For IPI pull requests, loop across the rto_mask.
669 struct irq_work rto_push_work;
670 raw_spinlock_t rto_lock;
671 /* These are only updated and read within rto_lock */
674 /* These atomics are updated outside of a lock */
675 atomic_t rto_loop_next;
676 atomic_t rto_loop_start;
679 * The "RT overload" flag: it gets set if a CPU has more than
680 * one runnable RT task.
682 cpumask_var_t rto_mask;
683 struct cpupri cpupri;
685 unsigned long max_cpu_capacity;
688 extern struct root_domain def_root_domain;
689 extern struct mutex sched_domains_mutex;
691 extern void init_defrootdomain(void);
692 extern int sched_init_domains(const struct cpumask *cpu_map);
693 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
695 #ifdef HAVE_RT_PUSH_IPI
696 extern void rto_push_irq_work_func(struct irq_work *work);
698 #endif /* CONFIG_SMP */
701 * This is the main, per-CPU runqueue data structure.
703 * Locking rule: those places that want to lock multiple runqueues
704 * (such as the load balancing or the thread migration code), lock
705 * acquire operations must be ordered by ascending &runqueue.
712 * nr_running and cpu_load should be in the same cacheline because
713 * remote CPUs use both these fields when doing load calculation.
715 unsigned int nr_running;
716 #ifdef CONFIG_NUMA_BALANCING
717 unsigned int nr_numa_running;
718 unsigned int nr_preferred_running;
720 #define CPU_LOAD_IDX_MAX 5
721 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
722 #ifdef CONFIG_NO_HZ_COMMON
724 unsigned long last_load_update_tick;
725 #endif /* CONFIG_SMP */
726 unsigned long nohz_flags;
727 #endif /* CONFIG_NO_HZ_COMMON */
728 #ifdef CONFIG_NO_HZ_FULL
729 unsigned long last_sched_tick;
731 /* capture load from *all* tasks on this cpu: */
732 struct load_weight load;
733 unsigned long nr_load_updates;
740 #ifdef CONFIG_FAIR_GROUP_SCHED
741 /* list of leaf cfs_rq on this cpu: */
742 struct list_head leaf_cfs_rq_list;
743 struct list_head *tmp_alone_branch;
744 #endif /* CONFIG_FAIR_GROUP_SCHED */
747 * This is part of a global counter where only the total sum
748 * over all CPUs matters. A task can increase this counter on
749 * one CPU and if it got migrated afterwards it may decrease
750 * it on another CPU. Always updated under the runqueue lock:
752 unsigned long nr_uninterruptible;
754 struct task_struct *curr, *idle, *stop;
755 unsigned long next_balance;
756 struct mm_struct *prev_mm;
758 unsigned int clock_update_flags;
765 struct root_domain *rd;
766 struct sched_domain *sd;
768 unsigned long cpu_capacity;
769 unsigned long cpu_capacity_orig;
771 struct callback_head *balance_callback;
773 unsigned char idle_balance;
774 /* For active balancing */
777 struct cpu_stop_work active_balance_work;
778 /* cpu of this runqueue: */
782 struct list_head cfs_tasks;
789 /* This is used to determine avg_idle's max value */
790 u64 max_idle_balance_cost;
793 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
796 #ifdef CONFIG_PARAVIRT
799 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
800 u64 prev_steal_time_rq;
803 /* calc_load related fields */
804 unsigned long calc_load_update;
805 long calc_load_active;
807 #ifdef CONFIG_SCHED_HRTICK
809 int hrtick_csd_pending;
810 call_single_data_t hrtick_csd;
812 struct hrtimer hrtick_timer;
815 #ifdef CONFIG_SCHEDSTATS
817 struct sched_info rq_sched_info;
818 unsigned long long rq_cpu_time;
819 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
821 /* sys_sched_yield() stats */
822 unsigned int yld_count;
824 /* schedule() stats */
825 unsigned int sched_count;
826 unsigned int sched_goidle;
828 /* try_to_wake_up() stats */
829 unsigned int ttwu_count;
830 unsigned int ttwu_local;
834 struct llist_head wake_list;
837 #ifdef CONFIG_CPU_IDLE
838 /* Must be inspected within a rcu lock section */
839 struct cpuidle_state *idle_state;
843 static inline int cpu_of(struct rq *rq)
853 #ifdef CONFIG_SCHED_SMT
855 extern struct static_key_false sched_smt_present;
857 extern void __update_idle_core(struct rq *rq);
859 static inline void update_idle_core(struct rq *rq)
861 if (static_branch_unlikely(&sched_smt_present))
862 __update_idle_core(rq);
866 static inline void update_idle_core(struct rq *rq) { }
869 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
871 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
872 #define this_rq() this_cpu_ptr(&runqueues)
873 #define task_rq(p) cpu_rq(task_cpu(p))
874 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
875 #define raw_rq() raw_cpu_ptr(&runqueues)
877 static inline u64 __rq_clock_broken(struct rq *rq)
879 return READ_ONCE(rq->clock);
883 * rq::clock_update_flags bits
885 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
886 * call to __schedule(). This is an optimisation to avoid
887 * neighbouring rq clock updates.
889 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
890 * in effect and calls to update_rq_clock() are being ignored.
892 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
893 * made to update_rq_clock() since the last time rq::lock was pinned.
895 * If inside of __schedule(), clock_update_flags will have been
896 * shifted left (a left shift is a cheap operation for the fast path
897 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
899 * if (rq-clock_update_flags >= RQCF_UPDATED)
901 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
902 * one position though, because the next rq_unpin_lock() will shift it
905 #define RQCF_REQ_SKIP 0x01
906 #define RQCF_ACT_SKIP 0x02
907 #define RQCF_UPDATED 0x04
909 static inline void assert_clock_updated(struct rq *rq)
912 * The only reason for not seeing a clock update since the
913 * last rq_pin_lock() is if we're currently skipping updates.
915 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
918 static inline u64 rq_clock(struct rq *rq)
920 lockdep_assert_held(&rq->lock);
921 assert_clock_updated(rq);
926 static inline u64 rq_clock_task(struct rq *rq)
928 lockdep_assert_held(&rq->lock);
929 assert_clock_updated(rq);
931 return rq->clock_task;
934 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
936 lockdep_assert_held(&rq->lock);
938 rq->clock_update_flags |= RQCF_REQ_SKIP;
940 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
945 struct pin_cookie cookie;
946 #ifdef CONFIG_SCHED_DEBUG
948 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
949 * current pin context is stashed here in case it needs to be
950 * restored in rq_repin_lock().
952 unsigned int clock_update_flags;
956 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
958 rf->cookie = lockdep_pin_lock(&rq->lock);
960 #ifdef CONFIG_SCHED_DEBUG
961 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
962 rf->clock_update_flags = 0;
966 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
968 #ifdef CONFIG_SCHED_DEBUG
969 if (rq->clock_update_flags > RQCF_ACT_SKIP)
970 rf->clock_update_flags = RQCF_UPDATED;
973 lockdep_unpin_lock(&rq->lock, rf->cookie);
976 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
978 lockdep_repin_lock(&rq->lock, rf->cookie);
980 #ifdef CONFIG_SCHED_DEBUG
982 * Restore the value we stashed in @rf for this pin context.
984 rq->clock_update_flags |= rf->clock_update_flags;
989 enum numa_topology_type {
994 extern enum numa_topology_type sched_numa_topology_type;
995 extern int sched_max_numa_distance;
996 extern bool find_numa_distance(int distance);
1000 extern void sched_init_numa(void);
1001 extern void sched_domains_numa_masks_set(unsigned int cpu);
1002 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1004 static inline void sched_init_numa(void) { }
1005 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1006 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1009 #ifdef CONFIG_NUMA_BALANCING
1010 /* The regions in numa_faults array from task_struct */
1011 enum numa_faults_stats {
1017 extern void sched_setnuma(struct task_struct *p, int node);
1018 extern int migrate_task_to(struct task_struct *p, int cpu);
1019 extern int migrate_swap(struct task_struct *, struct task_struct *);
1020 #endif /* CONFIG_NUMA_BALANCING */
1025 queue_balance_callback(struct rq *rq,
1026 struct callback_head *head,
1027 void (*func)(struct rq *rq))
1029 lockdep_assert_held(&rq->lock);
1031 if (unlikely(head->next))
1034 head->func = (void (*)(struct callback_head *))func;
1035 head->next = rq->balance_callback;
1036 rq->balance_callback = head;
1039 extern void sched_ttwu_pending(void);
1041 #define rcu_dereference_check_sched_domain(p) \
1042 rcu_dereference_check((p), \
1043 lockdep_is_held(&sched_domains_mutex))
1046 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1047 * See detach_destroy_domains: synchronize_sched for details.
1049 * The domain tree of any CPU may only be accessed from within
1050 * preempt-disabled sections.
1052 #define for_each_domain(cpu, __sd) \
1053 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1054 __sd; __sd = __sd->parent)
1056 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1059 * highest_flag_domain - Return highest sched_domain containing flag.
1060 * @cpu: The cpu whose highest level of sched domain is to
1062 * @flag: The flag to check for the highest sched_domain
1063 * for the given cpu.
1065 * Returns the highest sched_domain of a cpu which contains the given flag.
1067 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1069 struct sched_domain *sd, *hsd = NULL;
1071 for_each_domain(cpu, sd) {
1072 if (!(sd->flags & flag))
1080 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1082 struct sched_domain *sd;
1084 for_each_domain(cpu, sd) {
1085 if (sd->flags & flag)
1092 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1093 DECLARE_PER_CPU(int, sd_llc_size);
1094 DECLARE_PER_CPU(int, sd_llc_id);
1095 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1096 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1097 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1099 struct sched_group_capacity {
1102 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1105 unsigned long capacity;
1106 unsigned long min_capacity; /* Min per-CPU capacity in group */
1107 unsigned long next_update;
1108 int imbalance; /* XXX unrelated to capacity but shared group state */
1110 #ifdef CONFIG_SCHED_DEBUG
1114 unsigned long cpumask[0]; /* balance mask */
1117 struct sched_group {
1118 struct sched_group *next; /* Must be a circular list */
1121 unsigned int group_weight;
1122 struct sched_group_capacity *sgc;
1123 int asym_prefer_cpu; /* cpu of highest priority in group */
1126 * The CPUs this group covers.
1128 * NOTE: this field is variable length. (Allocated dynamically
1129 * by attaching extra space to the end of the structure,
1130 * depending on how many CPUs the kernel has booted up with)
1132 unsigned long cpumask[0];
1135 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1137 return to_cpumask(sg->cpumask);
1141 * See build_balance_mask().
1143 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1145 return to_cpumask(sg->sgc->cpumask);
1149 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1150 * @group: The group whose first cpu is to be returned.
1152 static inline unsigned int group_first_cpu(struct sched_group *group)
1154 return cpumask_first(sched_group_span(group));
1157 extern int group_balance_cpu(struct sched_group *sg);
1159 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1160 void register_sched_domain_sysctl(void);
1161 void dirty_sched_domain_sysctl(int cpu);
1162 void unregister_sched_domain_sysctl(void);
1164 static inline void register_sched_domain_sysctl(void)
1167 static inline void dirty_sched_domain_sysctl(int cpu)
1170 static inline void unregister_sched_domain_sysctl(void)
1177 static inline void sched_ttwu_pending(void) { }
1179 #endif /* CONFIG_SMP */
1182 #include "autogroup.h"
1184 #ifdef CONFIG_CGROUP_SCHED
1187 * Return the group to which this tasks belongs.
1189 * We cannot use task_css() and friends because the cgroup subsystem
1190 * changes that value before the cgroup_subsys::attach() method is called,
1191 * therefore we cannot pin it and might observe the wrong value.
1193 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1194 * core changes this before calling sched_move_task().
1196 * Instead we use a 'copy' which is updated from sched_move_task() while
1197 * holding both task_struct::pi_lock and rq::lock.
1199 static inline struct task_group *task_group(struct task_struct *p)
1201 return p->sched_task_group;
1204 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1205 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1207 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1208 struct task_group *tg = task_group(p);
1211 #ifdef CONFIG_FAIR_GROUP_SCHED
1212 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1213 p->se.cfs_rq = tg->cfs_rq[cpu];
1214 p->se.parent = tg->se[cpu];
1217 #ifdef CONFIG_RT_GROUP_SCHED
1218 p->rt.rt_rq = tg->rt_rq[cpu];
1219 p->rt.parent = tg->rt_se[cpu];
1223 #else /* CONFIG_CGROUP_SCHED */
1225 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1226 static inline struct task_group *task_group(struct task_struct *p)
1231 #endif /* CONFIG_CGROUP_SCHED */
1233 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1235 set_task_rq(p, cpu);
1238 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1239 * successfuly executed on another CPU. We must ensure that updates of
1240 * per-task data have been completed by this moment.
1243 #ifdef CONFIG_THREAD_INFO_IN_TASK
1246 task_thread_info(p)->cpu = cpu;
1253 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1255 #ifdef CONFIG_SCHED_DEBUG
1256 # include <linux/static_key.h>
1257 # define const_debug __read_mostly
1259 # define const_debug const
1262 #define SCHED_FEAT(name, enabled) \
1263 __SCHED_FEAT_##name ,
1266 #include "features.h"
1272 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1275 * To support run-time toggling of sched features, all the translation units
1276 * (but core.c) reference the sysctl_sched_features defined in core.c.
1278 extern const_debug unsigned int sysctl_sched_features;
1280 #define SCHED_FEAT(name, enabled) \
1281 static __always_inline bool static_branch_##name(struct static_key *key) \
1283 return static_key_##enabled(key); \
1286 #include "features.h"
1289 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1290 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1292 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1295 * Each translation unit has its own copy of sysctl_sched_features to allow
1296 * constants propagation at compile time and compiler optimization based on
1299 #define SCHED_FEAT(name, enabled) \
1300 (1UL << __SCHED_FEAT_##name) * enabled |
1301 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1302 #include "features.h"
1306 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1308 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1310 extern struct static_key_false sched_numa_balancing;
1311 extern struct static_key_false sched_schedstats;
1313 static inline u64 global_rt_period(void)
1315 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1318 static inline u64 global_rt_runtime(void)
1320 if (sysctl_sched_rt_runtime < 0)
1323 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1326 static inline int task_current(struct rq *rq, struct task_struct *p)
1328 return rq->curr == p;
1331 static inline int task_running(struct rq *rq, struct task_struct *p)
1336 return task_current(rq, p);
1340 static inline int task_on_rq_queued(struct task_struct *p)
1342 return p->on_rq == TASK_ON_RQ_QUEUED;
1345 static inline int task_on_rq_migrating(struct task_struct *p)
1347 return p->on_rq == TASK_ON_RQ_MIGRATING;
1350 #ifndef prepare_arch_switch
1351 # define prepare_arch_switch(next) do { } while (0)
1353 #ifndef finish_arch_post_lock_switch
1354 # define finish_arch_post_lock_switch() do { } while (0)
1360 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1361 #define WF_FORK 0x02 /* child wakeup after fork */
1362 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1365 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1366 * of tasks with abnormal "nice" values across CPUs the contribution that
1367 * each task makes to its run queue's load is weighted according to its
1368 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1369 * scaled version of the new time slice allocation that they receive on time
1373 #define WEIGHT_IDLEPRIO 3
1374 #define WMULT_IDLEPRIO 1431655765
1376 extern const int sched_prio_to_weight[40];
1377 extern const u32 sched_prio_to_wmult[40];
1380 * {de,en}queue flags:
1382 * DEQUEUE_SLEEP - task is no longer runnable
1383 * ENQUEUE_WAKEUP - task just became runnable
1385 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1386 * are in a known state which allows modification. Such pairs
1387 * should preserve as much state as possible.
1389 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1392 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1393 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1394 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1398 #define DEQUEUE_SLEEP 0x01
1399 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1400 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1401 #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */
1403 #define ENQUEUE_WAKEUP 0x01
1404 #define ENQUEUE_RESTORE 0x02
1405 #define ENQUEUE_MOVE 0x04
1406 #define ENQUEUE_NOCLOCK 0x08
1408 #define ENQUEUE_HEAD 0x10
1409 #define ENQUEUE_REPLENISH 0x20
1411 #define ENQUEUE_MIGRATED 0x40
1413 #define ENQUEUE_MIGRATED 0x00
1416 #define RETRY_TASK ((void *)-1UL)
1418 struct sched_class {
1419 const struct sched_class *next;
1421 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1422 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1423 void (*yield_task) (struct rq *rq);
1424 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1426 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1429 * It is the responsibility of the pick_next_task() method that will
1430 * return the next task to call put_prev_task() on the @prev task or
1431 * something equivalent.
1433 * May return RETRY_TASK when it finds a higher prio class has runnable
1436 struct task_struct * (*pick_next_task) (struct rq *rq,
1437 struct task_struct *prev,
1438 struct rq_flags *rf);
1439 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1442 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1443 void (*migrate_task_rq)(struct task_struct *p);
1445 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1447 void (*set_cpus_allowed)(struct task_struct *p,
1448 const struct cpumask *newmask);
1450 void (*rq_online)(struct rq *rq);
1451 void (*rq_offline)(struct rq *rq);
1454 void (*set_curr_task) (struct rq *rq);
1455 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1456 void (*task_fork) (struct task_struct *p);
1457 void (*task_dead) (struct task_struct *p);
1460 * The switched_from() call is allowed to drop rq->lock, therefore we
1461 * cannot assume the switched_from/switched_to pair is serliazed by
1462 * rq->lock. They are however serialized by p->pi_lock.
1464 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1465 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1466 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1469 unsigned int (*get_rr_interval) (struct rq *rq,
1470 struct task_struct *task);
1472 void (*update_curr) (struct rq *rq);
1474 #define TASK_SET_GROUP 0
1475 #define TASK_MOVE_GROUP 1
1477 #ifdef CONFIG_FAIR_GROUP_SCHED
1478 void (*task_change_group) (struct task_struct *p, int type);
1482 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1484 prev->sched_class->put_prev_task(rq, prev);
1487 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1489 curr->sched_class->set_curr_task(rq);
1493 #define sched_class_highest (&stop_sched_class)
1495 #define sched_class_highest (&dl_sched_class)
1497 #define for_each_class(class) \
1498 for (class = sched_class_highest; class; class = class->next)
1500 extern const struct sched_class stop_sched_class;
1501 extern const struct sched_class dl_sched_class;
1502 extern const struct sched_class rt_sched_class;
1503 extern const struct sched_class fair_sched_class;
1504 extern const struct sched_class idle_sched_class;
1509 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1511 extern void trigger_load_balance(struct rq *rq);
1513 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1517 #ifdef CONFIG_CPU_IDLE
1518 static inline void idle_set_state(struct rq *rq,
1519 struct cpuidle_state *idle_state)
1521 rq->idle_state = idle_state;
1524 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1526 SCHED_WARN_ON(!rcu_read_lock_held());
1527 return rq->idle_state;
1530 static inline void idle_set_state(struct rq *rq,
1531 struct cpuidle_state *idle_state)
1535 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1541 extern void schedule_idle(void);
1543 extern void sysrq_sched_debug_show(void);
1544 extern void sched_init_granularity(void);
1545 extern void update_max_interval(void);
1547 extern void init_sched_dl_class(void);
1548 extern void init_sched_rt_class(void);
1549 extern void init_sched_fair_class(void);
1551 extern void reweight_task(struct task_struct *p, int prio);
1553 extern void resched_curr(struct rq *rq);
1554 extern void resched_cpu(int cpu);
1556 extern struct rt_bandwidth def_rt_bandwidth;
1557 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1559 extern struct dl_bandwidth def_dl_bandwidth;
1560 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1561 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1562 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1563 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1566 #define BW_UNIT (1 << BW_SHIFT)
1567 #define RATIO_SHIFT 8
1568 unsigned long to_ratio(u64 period, u64 runtime);
1570 extern void init_entity_runnable_average(struct sched_entity *se);
1571 extern void post_init_entity_util_avg(struct sched_entity *se);
1573 #ifdef CONFIG_NO_HZ_FULL
1574 extern bool sched_can_stop_tick(struct rq *rq);
1577 * Tick may be needed by tasks in the runqueue depending on their policy and
1578 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1579 * nohz mode if necessary.
1581 static inline void sched_update_tick_dependency(struct rq *rq)
1585 if (!tick_nohz_full_enabled())
1590 if (!tick_nohz_full_cpu(cpu))
1593 if (sched_can_stop_tick(rq))
1594 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1596 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1599 static inline void sched_update_tick_dependency(struct rq *rq) { }
1602 static inline void add_nr_running(struct rq *rq, unsigned count)
1604 unsigned prev_nr = rq->nr_running;
1606 rq->nr_running = prev_nr + count;
1608 if (prev_nr < 2 && rq->nr_running >= 2) {
1610 if (!rq->rd->overload)
1611 rq->rd->overload = true;
1615 sched_update_tick_dependency(rq);
1618 static inline void sub_nr_running(struct rq *rq, unsigned count)
1620 rq->nr_running -= count;
1621 /* Check if we still need preemption */
1622 sched_update_tick_dependency(rq);
1625 static inline void rq_last_tick_reset(struct rq *rq)
1627 #ifdef CONFIG_NO_HZ_FULL
1628 rq->last_sched_tick = jiffies;
1632 extern void update_rq_clock(struct rq *rq);
1634 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1635 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1637 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1639 extern const_debug unsigned int sysctl_sched_time_avg;
1640 extern const_debug unsigned int sysctl_sched_nr_migrate;
1641 extern const_debug unsigned int sysctl_sched_migration_cost;
1643 static inline u64 sched_avg_period(void)
1645 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1648 #ifdef CONFIG_SCHED_HRTICK
1652 * - enabled by features
1653 * - hrtimer is actually high res
1655 static inline int hrtick_enabled(struct rq *rq)
1657 if (!sched_feat(HRTICK))
1659 if (!cpu_active(cpu_of(rq)))
1661 return hrtimer_is_hres_active(&rq->hrtick_timer);
1664 void hrtick_start(struct rq *rq, u64 delay);
1668 static inline int hrtick_enabled(struct rq *rq)
1673 #endif /* CONFIG_SCHED_HRTICK */
1675 #ifndef arch_scale_freq_capacity
1676 static __always_inline
1677 unsigned long arch_scale_freq_capacity(int cpu)
1679 return SCHED_CAPACITY_SCALE;
1684 extern void sched_avg_update(struct rq *rq);
1686 #ifndef arch_scale_cpu_capacity
1687 static __always_inline
1688 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1690 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1691 return sd->smt_gain / sd->span_weight;
1693 return SCHED_CAPACITY_SCALE;
1697 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1699 rq->rt_avg += rt_delta * arch_scale_freq_capacity(cpu_of(rq));
1700 sched_avg_update(rq);
1703 #ifndef arch_scale_cpu_capacity
1704 static __always_inline
1705 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1707 return SCHED_CAPACITY_SCALE;
1710 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1711 static inline void sched_avg_update(struct rq *rq) { }
1714 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1715 __acquires(rq->lock);
1717 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1718 __acquires(p->pi_lock)
1719 __acquires(rq->lock);
1721 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1722 __releases(rq->lock)
1724 rq_unpin_lock(rq, rf);
1725 raw_spin_unlock(&rq->lock);
1729 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1730 __releases(rq->lock)
1731 __releases(p->pi_lock)
1733 rq_unpin_lock(rq, rf);
1734 raw_spin_unlock(&rq->lock);
1735 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1739 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1740 __acquires(rq->lock)
1742 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1743 rq_pin_lock(rq, rf);
1747 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1748 __acquires(rq->lock)
1750 raw_spin_lock_irq(&rq->lock);
1751 rq_pin_lock(rq, rf);
1755 rq_lock(struct rq *rq, struct rq_flags *rf)
1756 __acquires(rq->lock)
1758 raw_spin_lock(&rq->lock);
1759 rq_pin_lock(rq, rf);
1763 rq_relock(struct rq *rq, struct rq_flags *rf)
1764 __acquires(rq->lock)
1766 raw_spin_lock(&rq->lock);
1767 rq_repin_lock(rq, rf);
1771 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1772 __releases(rq->lock)
1774 rq_unpin_lock(rq, rf);
1775 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1779 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1780 __releases(rq->lock)
1782 rq_unpin_lock(rq, rf);
1783 raw_spin_unlock_irq(&rq->lock);
1787 rq_unlock(struct rq *rq, struct rq_flags *rf)
1788 __releases(rq->lock)
1790 rq_unpin_lock(rq, rf);
1791 raw_spin_unlock(&rq->lock);
1795 #ifdef CONFIG_PREEMPT
1797 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1800 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1801 * way at the expense of forcing extra atomic operations in all
1802 * invocations. This assures that the double_lock is acquired using the
1803 * same underlying policy as the spinlock_t on this architecture, which
1804 * reduces latency compared to the unfair variant below. However, it
1805 * also adds more overhead and therefore may reduce throughput.
1807 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1808 __releases(this_rq->lock)
1809 __acquires(busiest->lock)
1810 __acquires(this_rq->lock)
1812 raw_spin_unlock(&this_rq->lock);
1813 double_rq_lock(this_rq, busiest);
1820 * Unfair double_lock_balance: Optimizes throughput at the expense of
1821 * latency by eliminating extra atomic operations when the locks are
1822 * already in proper order on entry. This favors lower cpu-ids and will
1823 * grant the double lock to lower cpus over higher ids under contention,
1824 * regardless of entry order into the function.
1826 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1827 __releases(this_rq->lock)
1828 __acquires(busiest->lock)
1829 __acquires(this_rq->lock)
1833 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1834 if (busiest < this_rq) {
1835 raw_spin_unlock(&this_rq->lock);
1836 raw_spin_lock(&busiest->lock);
1837 raw_spin_lock_nested(&this_rq->lock,
1838 SINGLE_DEPTH_NESTING);
1841 raw_spin_lock_nested(&busiest->lock,
1842 SINGLE_DEPTH_NESTING);
1847 #endif /* CONFIG_PREEMPT */
1850 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1852 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1854 if (unlikely(!irqs_disabled())) {
1855 /* printk() doesn't work good under rq->lock */
1856 raw_spin_unlock(&this_rq->lock);
1860 return _double_lock_balance(this_rq, busiest);
1863 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1864 __releases(busiest->lock)
1866 raw_spin_unlock(&busiest->lock);
1867 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1870 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1876 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1879 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1885 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1888 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1894 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1898 * double_rq_lock - safely lock two runqueues
1900 * Note this does not disable interrupts like task_rq_lock,
1901 * you need to do so manually before calling.
1903 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1904 __acquires(rq1->lock)
1905 __acquires(rq2->lock)
1907 BUG_ON(!irqs_disabled());
1909 raw_spin_lock(&rq1->lock);
1910 __acquire(rq2->lock); /* Fake it out ;) */
1913 raw_spin_lock(&rq1->lock);
1914 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1916 raw_spin_lock(&rq2->lock);
1917 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1923 * double_rq_unlock - safely unlock two runqueues
1925 * Note this does not restore interrupts like task_rq_unlock,
1926 * you need to do so manually after calling.
1928 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1929 __releases(rq1->lock)
1930 __releases(rq2->lock)
1932 raw_spin_unlock(&rq1->lock);
1934 raw_spin_unlock(&rq2->lock);
1936 __release(rq2->lock);
1939 extern void set_rq_online (struct rq *rq);
1940 extern void set_rq_offline(struct rq *rq);
1941 extern bool sched_smp_initialized;
1943 #else /* CONFIG_SMP */
1946 * double_rq_lock - safely lock two runqueues
1948 * Note this does not disable interrupts like task_rq_lock,
1949 * you need to do so manually before calling.
1951 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1952 __acquires(rq1->lock)
1953 __acquires(rq2->lock)
1955 BUG_ON(!irqs_disabled());
1957 raw_spin_lock(&rq1->lock);
1958 __acquire(rq2->lock); /* Fake it out ;) */
1962 * double_rq_unlock - safely unlock two runqueues
1964 * Note this does not restore interrupts like task_rq_unlock,
1965 * you need to do so manually after calling.
1967 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1968 __releases(rq1->lock)
1969 __releases(rq2->lock)
1972 raw_spin_unlock(&rq1->lock);
1973 __release(rq2->lock);
1978 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1979 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1981 #ifdef CONFIG_SCHED_DEBUG
1982 extern bool sched_debug_enabled;
1984 extern void print_cfs_stats(struct seq_file *m, int cpu);
1985 extern void print_rt_stats(struct seq_file *m, int cpu);
1986 extern void print_dl_stats(struct seq_file *m, int cpu);
1988 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1989 #ifdef CONFIG_NUMA_BALANCING
1991 show_numa_stats(struct task_struct *p, struct seq_file *m);
1993 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1994 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1995 #endif /* CONFIG_NUMA_BALANCING */
1996 #endif /* CONFIG_SCHED_DEBUG */
1998 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1999 extern void init_rt_rq(struct rt_rq *rt_rq);
2000 extern void init_dl_rq(struct dl_rq *dl_rq);
2002 extern void cfs_bandwidth_usage_inc(void);
2003 extern void cfs_bandwidth_usage_dec(void);
2005 #ifdef CONFIG_NO_HZ_COMMON
2006 enum rq_nohz_flag_bits {
2011 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2013 extern void nohz_balance_exit_idle(unsigned int cpu);
2015 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
2021 void __dl_update(struct dl_bw *dl_b, s64 bw)
2023 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2026 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2027 "sched RCU must be held");
2028 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2029 struct rq *rq = cpu_rq(i);
2031 rq->dl.extra_bw += bw;
2036 void __dl_update(struct dl_bw *dl_b, s64 bw)
2038 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2045 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2050 struct u64_stats_sync sync;
2053 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2056 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2057 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2058 * and never move forward.
2060 static inline u64 irq_time_read(int cpu)
2062 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2067 seq = __u64_stats_fetch_begin(&irqtime->sync);
2068 total = irqtime->total;
2069 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2073 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2075 #ifdef CONFIG_CPU_FREQ
2076 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2079 * cpufreq_update_util - Take a note about CPU utilization changes.
2080 * @rq: Runqueue to carry out the update for.
2081 * @flags: Update reason flags.
2083 * This function is called by the scheduler on the CPU whose utilization is
2086 * It can only be called from RCU-sched read-side critical sections.
2088 * The way cpufreq is currently arranged requires it to evaluate the CPU
2089 * performance state (frequency/voltage) on a regular basis to prevent it from
2090 * being stuck in a completely inadequate performance level for too long.
2091 * That is not guaranteed to happen if the updates are only triggered from CFS
2092 * and DL, though, because they may not be coming in if only RT tasks are
2093 * active all the time (or there are RT tasks only).
2095 * As a workaround for that issue, this function is called periodically by the
2096 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2097 * but that really is a band-aid. Going forward it should be replaced with
2098 * solutions targeted more specifically at RT tasks.
2100 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2102 struct update_util_data *data;
2104 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2107 data->func(data, rq_clock(rq), flags);
2110 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2111 #endif /* CONFIG_CPU_FREQ */
2113 #ifdef arch_scale_freq_capacity
2114 #ifndef arch_scale_freq_invariant
2115 #define arch_scale_freq_invariant() (true)
2117 #else /* arch_scale_freq_capacity */
2118 #define arch_scale_freq_invariant() (false)
2121 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2123 static inline unsigned long cpu_util_dl(struct rq *rq)
2125 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2128 static inline unsigned long cpu_util_cfs(struct rq *rq)
2130 return rq->cfs.avg.util_avg;