1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/context_tracking.h>
31 #include <linux/cpufreq.h>
32 #include <linux/cpumask_api.h>
33 #include <linux/ctype.h>
34 #include <linux/file.h>
35 #include <linux/fs_api.h>
36 #include <linux/hrtimer_api.h>
37 #include <linux/interrupt.h>
38 #include <linux/irq_work.h>
39 #include <linux/jiffies.h>
40 #include <linux/kref_api.h>
41 #include <linux/kthread.h>
42 #include <linux/ktime_api.h>
43 #include <linux/lockdep_api.h>
44 #include <linux/lockdep.h>
45 #include <linux/minmax.h>
47 #include <linux/module.h>
48 #include <linux/mutex_api.h>
49 #include <linux/plist.h>
50 #include <linux/poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/psi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/seq_file.h>
56 #include <linux/seqlock.h>
57 #include <linux/softirq.h>
58 #include <linux/spinlock_api.h>
59 #include <linux/static_key.h>
60 #include <linux/stop_machine.h>
61 #include <linux/syscalls_api.h>
62 #include <linux/syscalls.h>
63 #include <linux/tick.h>
64 #include <linux/topology.h>
65 #include <linux/types.h>
66 #include <linux/u64_stats_sync_api.h>
67 #include <linux/uaccess.h>
68 #include <linux/wait_api.h>
69 #include <linux/wait_bit.h>
70 #include <linux/workqueue_api.h>
72 #include <trace/events/power.h>
73 #include <trace/events/sched.h>
75 #include "../workqueue_internal.h"
77 #ifdef CONFIG_CGROUP_SCHED
78 #include <linux/cgroup.h>
79 #include <linux/psi.h>
82 #ifdef CONFIG_SCHED_DEBUG
83 # include <linux/static_key.h>
86 #ifdef CONFIG_PARAVIRT
87 # include <asm/paravirt.h>
88 # include <asm/paravirt_api_clock.h>
92 #include "cpudeadline.h"
94 #ifdef CONFIG_SCHED_DEBUG
95 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
97 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
101 struct cpuidle_state;
103 /* task_struct::on_rq states: */
104 #define TASK_ON_RQ_QUEUED 1
105 #define TASK_ON_RQ_MIGRATING 2
107 extern __read_mostly int scheduler_running;
109 extern unsigned long calc_load_update;
110 extern atomic_long_t calc_load_tasks;
112 extern unsigned int sysctl_sched_child_runs_first;
114 extern void calc_global_load_tick(struct rq *this_rq);
115 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
117 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
119 extern unsigned int sysctl_sched_rt_period;
120 extern int sysctl_sched_rt_runtime;
121 extern int sched_rr_timeslice;
124 * Helpers for converting nanosecond timing to jiffy resolution
126 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
129 * Increase resolution of nice-level calculations for 64-bit architectures.
130 * The extra resolution improves shares distribution and load balancing of
131 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
132 * hierarchies, especially on larger systems. This is not a user-visible change
133 * and does not change the user-interface for setting shares/weights.
135 * We increase resolution only if we have enough bits to allow this increased
136 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
137 * are pretty high and the returns do not justify the increased costs.
139 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
140 * increase coverage and consistency always enable it on 64-bit platforms.
143 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
144 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
145 # define scale_load_down(w) \
147 unsigned long __w = (w); \
149 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
153 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
154 # define scale_load(w) (w)
155 # define scale_load_down(w) (w)
159 * Task weight (visible to users) and its load (invisible to users) have
160 * independent resolution, but they should be well calibrated. We use
161 * scale_load() and scale_load_down(w) to convert between them. The
162 * following must be true:
164 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
167 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
170 * Single value that decides SCHED_DEADLINE internal math precision.
171 * 10 -> just above 1us
172 * 9 -> just above 0.5us
177 * Single value that denotes runtime == period, ie unlimited time.
179 #define RUNTIME_INF ((u64)~0ULL)
181 static inline int idle_policy(int policy)
183 return policy == SCHED_IDLE;
185 static inline int fair_policy(int policy)
187 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
190 static inline int rt_policy(int policy)
192 return policy == SCHED_FIFO || policy == SCHED_RR;
195 static inline int dl_policy(int policy)
197 return policy == SCHED_DEADLINE;
199 static inline bool valid_policy(int policy)
201 return idle_policy(policy) || fair_policy(policy) ||
202 rt_policy(policy) || dl_policy(policy);
205 static inline int task_has_idle_policy(struct task_struct *p)
207 return idle_policy(p->policy);
210 static inline int task_has_rt_policy(struct task_struct *p)
212 return rt_policy(p->policy);
215 static inline int task_has_dl_policy(struct task_struct *p)
217 return dl_policy(p->policy);
220 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
222 static inline void update_avg(u64 *avg, u64 sample)
224 s64 diff = sample - *avg;
229 * Shifting a value by an exponent greater *or equal* to the size of said value
230 * is UB; cap at size-1.
232 #define shr_bound(val, shift) \
233 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
236 * !! For sched_setattr_nocheck() (kernel) only !!
238 * This is actually gross. :(
240 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
241 * tasks, but still be able to sleep. We need this on platforms that cannot
242 * atomically change clock frequency. Remove once fast switching will be
243 * available on such platforms.
245 * SUGOV stands for SchedUtil GOVernor.
247 #define SCHED_FLAG_SUGOV 0x10000000
249 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
251 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
253 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
254 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
261 * Tells if entity @a should preempt entity @b.
264 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
266 return dl_entity_is_special(a) ||
267 dl_time_before(a->deadline, b->deadline);
271 * This is the priority-queue data structure of the RT scheduling class:
273 struct rt_prio_array {
274 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
275 struct list_head queue[MAX_RT_PRIO];
278 struct rt_bandwidth {
279 /* nests inside the rq lock: */
280 raw_spinlock_t rt_runtime_lock;
283 struct hrtimer rt_period_timer;
284 unsigned int rt_period_active;
287 void __dl_clear_params(struct task_struct *p);
289 struct dl_bandwidth {
290 raw_spinlock_t dl_runtime_lock;
295 static inline int dl_bandwidth_enabled(void)
297 return sysctl_sched_rt_runtime >= 0;
301 * To keep the bandwidth of -deadline tasks under control
302 * we need some place where:
303 * - store the maximum -deadline bandwidth of each cpu;
304 * - cache the fraction of bandwidth that is currently allocated in
307 * This is all done in the data structure below. It is similar to the
308 * one used for RT-throttling (rt_bandwidth), with the main difference
309 * that, since here we are only interested in admission control, we
310 * do not decrease any runtime while the group "executes", neither we
311 * need a timer to replenish it.
313 * With respect to SMP, bandwidth is given on a per root domain basis,
315 * - bw (< 100%) is the deadline bandwidth of each CPU;
316 * - total_bw is the currently allocated bandwidth in each root domain;
324 extern void init_dl_bw(struct dl_bw *dl_b);
325 extern int sched_dl_global_validate(void);
326 extern void sched_dl_do_global(void);
327 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
328 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
329 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
330 extern bool __checkparam_dl(const struct sched_attr *attr);
331 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
332 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
333 extern int dl_cpu_busy(int cpu, struct task_struct *p);
335 #ifdef CONFIG_CGROUP_SCHED
340 extern struct list_head task_groups;
342 struct cfs_bandwidth {
343 #ifdef CONFIG_CFS_BANDWIDTH
350 s64 hierarchical_quota;
355 struct hrtimer period_timer;
356 struct hrtimer slack_timer;
357 struct list_head throttled_cfs_rq;
368 /* Task group related information */
370 struct cgroup_subsys_state css;
372 #ifdef CONFIG_FAIR_GROUP_SCHED
373 /* schedulable entities of this group on each CPU */
374 struct sched_entity **se;
375 /* runqueue "owned" by this group on each CPU */
376 struct cfs_rq **cfs_rq;
377 unsigned long shares;
379 /* A positive value indicates that this is a SCHED_IDLE group. */
384 * load_avg can be heavily contended at clock tick time, so put
385 * it in its own cacheline separated from the fields above which
386 * will also be accessed at each tick.
388 atomic_long_t load_avg ____cacheline_aligned;
392 #ifdef CONFIG_RT_GROUP_SCHED
393 struct sched_rt_entity **rt_se;
394 struct rt_rq **rt_rq;
396 struct rt_bandwidth rt_bandwidth;
400 struct list_head list;
402 struct task_group *parent;
403 struct list_head siblings;
404 struct list_head children;
406 #ifdef CONFIG_SCHED_AUTOGROUP
407 struct autogroup *autogroup;
410 struct cfs_bandwidth cfs_bandwidth;
412 #ifdef CONFIG_UCLAMP_TASK_GROUP
413 /* The two decimal precision [%] value requested from user-space */
414 unsigned int uclamp_pct[UCLAMP_CNT];
415 /* Clamp values requested for a task group */
416 struct uclamp_se uclamp_req[UCLAMP_CNT];
417 /* Effective clamp values used for a task group */
418 struct uclamp_se uclamp[UCLAMP_CNT];
423 #ifdef CONFIG_FAIR_GROUP_SCHED
424 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
427 * A weight of 0 or 1 can cause arithmetics problems.
428 * A weight of a cfs_rq is the sum of weights of which entities
429 * are queued on this cfs_rq, so a weight of a entity should not be
430 * too large, so as the shares value of a task group.
431 * (The default weight is 1024 - so there's no practical
432 * limitation from this.)
434 #define MIN_SHARES (1UL << 1)
435 #define MAX_SHARES (1UL << 18)
438 typedef int (*tg_visitor)(struct task_group *, void *);
440 extern int walk_tg_tree_from(struct task_group *from,
441 tg_visitor down, tg_visitor up, void *data);
444 * Iterate the full tree, calling @down when first entering a node and @up when
445 * leaving it for the final time.
447 * Caller must hold rcu_lock or sufficient equivalent.
449 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
451 return walk_tg_tree_from(&root_task_group, down, up, data);
454 extern int tg_nop(struct task_group *tg, void *data);
456 extern void free_fair_sched_group(struct task_group *tg);
457 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
458 extern void online_fair_sched_group(struct task_group *tg);
459 extern void unregister_fair_sched_group(struct task_group *tg);
460 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
461 struct sched_entity *se, int cpu,
462 struct sched_entity *parent);
463 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
465 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
466 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
467 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
469 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
470 struct sched_rt_entity *rt_se, int cpu,
471 struct sched_rt_entity *parent);
472 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
473 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
474 extern long sched_group_rt_runtime(struct task_group *tg);
475 extern long sched_group_rt_period(struct task_group *tg);
476 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
478 extern struct task_group *sched_create_group(struct task_group *parent);
479 extern void sched_online_group(struct task_group *tg,
480 struct task_group *parent);
481 extern void sched_destroy_group(struct task_group *tg);
482 extern void sched_release_group(struct task_group *tg);
484 extern void sched_move_task(struct task_struct *tsk);
486 #ifdef CONFIG_FAIR_GROUP_SCHED
487 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
489 extern int sched_group_set_idle(struct task_group *tg, long idle);
492 extern void set_task_rq_fair(struct sched_entity *se,
493 struct cfs_rq *prev, struct cfs_rq *next);
494 #else /* !CONFIG_SMP */
495 static inline void set_task_rq_fair(struct sched_entity *se,
496 struct cfs_rq *prev, struct cfs_rq *next) { }
497 #endif /* CONFIG_SMP */
498 #endif /* CONFIG_FAIR_GROUP_SCHED */
500 #else /* CONFIG_CGROUP_SCHED */
502 struct cfs_bandwidth { };
504 #endif /* CONFIG_CGROUP_SCHED */
506 extern void unregister_rt_sched_group(struct task_group *tg);
507 extern void free_rt_sched_group(struct task_group *tg);
508 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
511 * u64_u32_load/u64_u32_store
513 * Use a copy of a u64 value to protect against data race. This is only
514 * applicable for 32-bits architectures.
517 # define u64_u32_load_copy(var, copy) var
518 # define u64_u32_store_copy(var, copy, val) (var = val)
520 # define u64_u32_load_copy(var, copy) \
522 u64 __val, __val_copy; \
526 * paired with u64_u32_store_copy(), ordering access \
531 } while (__val != __val_copy); \
534 # define u64_u32_store_copy(var, copy, val) \
536 typeof(val) __val = (val); \
539 * paired with u64_u32_load_copy(), ordering access to var and \
546 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
547 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
549 /* CFS-related fields in a runqueue */
551 struct load_weight load;
552 unsigned int nr_running;
553 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
554 unsigned int idle_nr_running; /* SCHED_IDLE */
555 unsigned int idle_h_nr_running; /* SCHED_IDLE */
559 #ifdef CONFIG_SCHED_CORE
560 unsigned int forceidle_seq;
565 u64 min_vruntime_copy;
568 struct rb_root_cached tasks_timeline;
571 * 'curr' points to currently running entity on this cfs_rq.
572 * It is set to NULL otherwise (i.e when none are currently running).
574 struct sched_entity *curr;
575 struct sched_entity *next;
576 struct sched_entity *last;
577 struct sched_entity *skip;
579 #ifdef CONFIG_SCHED_DEBUG
580 unsigned int nr_spread_over;
587 struct sched_avg avg;
589 u64 last_update_time_copy;
592 raw_spinlock_t lock ____cacheline_aligned;
594 unsigned long load_avg;
595 unsigned long util_avg;
596 unsigned long runnable_avg;
599 #ifdef CONFIG_FAIR_GROUP_SCHED
600 unsigned long tg_load_avg_contrib;
602 long prop_runnable_sum;
605 * h_load = weight * f(tg)
607 * Where f(tg) is the recursive weight fraction assigned to
610 unsigned long h_load;
611 u64 last_h_load_update;
612 struct sched_entity *h_load_next;
613 #endif /* CONFIG_FAIR_GROUP_SCHED */
614 #endif /* CONFIG_SMP */
616 #ifdef CONFIG_FAIR_GROUP_SCHED
617 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
620 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
621 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
622 * (like users, containers etc.)
624 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
625 * This list is used during load balance.
628 struct list_head leaf_cfs_rq_list;
629 struct task_group *tg; /* group that "owns" this runqueue */
631 /* Locally cached copy of our task_group's idle value */
634 #ifdef CONFIG_CFS_BANDWIDTH
636 s64 runtime_remaining;
638 u64 throttled_pelt_idle;
640 u64 throttled_pelt_idle_copy;
643 u64 throttled_clock_pelt;
644 u64 throttled_clock_pelt_time;
647 struct list_head throttled_list;
648 #endif /* CONFIG_CFS_BANDWIDTH */
649 #endif /* CONFIG_FAIR_GROUP_SCHED */
652 static inline int rt_bandwidth_enabled(void)
654 return sysctl_sched_rt_runtime >= 0;
657 /* RT IPI pull logic requires IRQ_WORK */
658 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
659 # define HAVE_RT_PUSH_IPI
662 /* Real-Time classes' related field in a runqueue: */
664 struct rt_prio_array active;
665 unsigned int rt_nr_running;
666 unsigned int rr_nr_running;
667 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
669 int curr; /* highest queued rt task prio */
671 int next; /* next highest */
676 unsigned int rt_nr_migratory;
677 unsigned int rt_nr_total;
679 struct plist_head pushable_tasks;
681 #endif /* CONFIG_SMP */
687 /* Nests inside the rq lock: */
688 raw_spinlock_t rt_runtime_lock;
690 #ifdef CONFIG_RT_GROUP_SCHED
691 unsigned int rt_nr_boosted;
694 struct task_group *tg;
698 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
700 return rt_rq->rt_queued && rt_rq->rt_nr_running;
703 /* Deadline class' related fields in a runqueue */
705 /* runqueue is an rbtree, ordered by deadline */
706 struct rb_root_cached root;
708 unsigned int dl_nr_running;
712 * Deadline values of the currently executing and the
713 * earliest ready task on this rq. Caching these facilitates
714 * the decision whether or not a ready but not running task
715 * should migrate somewhere else.
722 unsigned int dl_nr_migratory;
726 * Tasks on this rq that can be pushed away. They are kept in
727 * an rb-tree, ordered by tasks' deadlines, with caching
728 * of the leftmost (earliest deadline) element.
730 struct rb_root_cached pushable_dl_tasks_root;
735 * "Active utilization" for this runqueue: increased when a
736 * task wakes up (becomes TASK_RUNNING) and decreased when a
742 * Utilization of the tasks "assigned" to this runqueue (including
743 * the tasks that are in runqueue and the tasks that executed on this
744 * CPU and blocked). Increased when a task moves to this runqueue, and
745 * decreased when the task moves away (migrates, changes scheduling
746 * policy, or terminates).
747 * This is needed to compute the "inactive utilization" for the
748 * runqueue (inactive utilization = this_bw - running_bw).
754 * Inverse of the fraction of CPU utilization that can be reclaimed
755 * by the GRUB algorithm.
760 #ifdef CONFIG_FAIR_GROUP_SCHED
761 /* An entity is a task if it doesn't "own" a runqueue */
762 #define entity_is_task(se) (!se->my_q)
764 static inline void se_update_runnable(struct sched_entity *se)
766 if (!entity_is_task(se))
767 se->runnable_weight = se->my_q->h_nr_running;
770 static inline long se_runnable(struct sched_entity *se)
772 if (entity_is_task(se))
775 return se->runnable_weight;
779 #define entity_is_task(se) 1
781 static inline void se_update_runnable(struct sched_entity *se) {}
783 static inline long se_runnable(struct sched_entity *se)
791 * XXX we want to get rid of these helpers and use the full load resolution.
793 static inline long se_weight(struct sched_entity *se)
795 return scale_load_down(se->load.weight);
799 static inline bool sched_asym_prefer(int a, int b)
801 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
805 struct em_perf_domain *em_pd;
806 struct perf_domain *next;
810 /* Scheduling group status flags */
811 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
812 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
815 * We add the notion of a root-domain which will be used to define per-domain
816 * variables. Each exclusive cpuset essentially defines an island domain by
817 * fully partitioning the member CPUs from any other cpuset. Whenever a new
818 * exclusive cpuset is created, we also create and attach a new root-domain
827 cpumask_var_t online;
830 * Indicate pullable load on at least one CPU, e.g:
831 * - More than one runnable task
832 * - Running task is misfit
836 /* Indicate one or more cpus over-utilized (tipping point) */
840 * The bit corresponding to a CPU gets set here if such CPU has more
841 * than one runnable -deadline task (as it is below for RT tasks).
843 cpumask_var_t dlo_mask;
849 * Indicate whether a root_domain's dl_bw has been checked or
850 * updated. It's monotonously increasing value.
852 * Also, some corner cases, like 'wrap around' is dangerous, but given
853 * that u64 is 'big enough'. So that shouldn't be a concern.
857 #ifdef HAVE_RT_PUSH_IPI
859 * For IPI pull requests, loop across the rto_mask.
861 struct irq_work rto_push_work;
862 raw_spinlock_t rto_lock;
863 /* These are only updated and read within rto_lock */
866 /* These atomics are updated outside of a lock */
867 atomic_t rto_loop_next;
868 atomic_t rto_loop_start;
871 * The "RT overload" flag: it gets set if a CPU has more than
872 * one runnable RT task.
874 cpumask_var_t rto_mask;
875 struct cpupri cpupri;
877 unsigned long max_cpu_capacity;
880 * NULL-terminated list of performance domains intersecting with the
881 * CPUs of the rd. Protected by RCU.
883 struct perf_domain __rcu *pd;
886 extern void init_defrootdomain(void);
887 extern int sched_init_domains(const struct cpumask *cpu_map);
888 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
889 extern void sched_get_rd(struct root_domain *rd);
890 extern void sched_put_rd(struct root_domain *rd);
892 #ifdef HAVE_RT_PUSH_IPI
893 extern void rto_push_irq_work_func(struct irq_work *work);
895 #endif /* CONFIG_SMP */
897 #ifdef CONFIG_UCLAMP_TASK
899 * struct uclamp_bucket - Utilization clamp bucket
900 * @value: utilization clamp value for tasks on this clamp bucket
901 * @tasks: number of RUNNABLE tasks on this clamp bucket
903 * Keep track of how many tasks are RUNNABLE for a given utilization
906 struct uclamp_bucket {
907 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
908 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
912 * struct uclamp_rq - rq's utilization clamp
913 * @value: currently active clamp values for a rq
914 * @bucket: utilization clamp buckets affecting a rq
916 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
917 * A clamp value is affecting a rq when there is at least one task RUNNABLE
918 * (or actually running) with that value.
920 * There are up to UCLAMP_CNT possible different clamp values, currently there
921 * are only two: minimum utilization and maximum utilization.
923 * All utilization clamping values are MAX aggregated, since:
924 * - for util_min: we want to run the CPU at least at the max of the minimum
925 * utilization required by its currently RUNNABLE tasks.
926 * - for util_max: we want to allow the CPU to run up to the max of the
927 * maximum utilization allowed by its currently RUNNABLE tasks.
929 * Since on each system we expect only a limited number of different
930 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
931 * the metrics required to compute all the per-rq utilization clamp values.
935 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
938 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
939 #endif /* CONFIG_UCLAMP_TASK */
942 * This is the main, per-CPU runqueue data structure.
944 * Locking rule: those places that want to lock multiple runqueues
945 * (such as the load balancing or the thread migration code), lock
946 * acquire operations must be ordered by ascending &runqueue.
950 raw_spinlock_t __lock;
953 * nr_running and cpu_load should be in the same cacheline because
954 * remote CPUs use both these fields when doing load calculation.
956 unsigned int nr_running;
957 #ifdef CONFIG_NUMA_BALANCING
958 unsigned int nr_numa_running;
959 unsigned int nr_preferred_running;
960 unsigned int numa_migrate_on;
962 #ifdef CONFIG_NO_HZ_COMMON
964 unsigned long last_blocked_load_update_tick;
965 unsigned int has_blocked_load;
966 call_single_data_t nohz_csd;
967 #endif /* CONFIG_SMP */
968 unsigned int nohz_tick_stopped;
970 #endif /* CONFIG_NO_HZ_COMMON */
973 unsigned int ttwu_pending;
977 #ifdef CONFIG_UCLAMP_TASK
978 /* Utilization clamp values based on CPU's RUNNABLE tasks */
979 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
980 unsigned int uclamp_flags;
981 #define UCLAMP_FLAG_IDLE 0x01
988 #ifdef CONFIG_FAIR_GROUP_SCHED
989 /* list of leaf cfs_rq on this CPU: */
990 struct list_head leaf_cfs_rq_list;
991 struct list_head *tmp_alone_branch;
992 #endif /* CONFIG_FAIR_GROUP_SCHED */
995 * This is part of a global counter where only the total sum
996 * over all CPUs matters. A task can increase this counter on
997 * one CPU and if it got migrated afterwards it may decrease
998 * it on another CPU. Always updated under the runqueue lock:
1000 unsigned int nr_uninterruptible;
1002 struct task_struct __rcu *curr;
1003 struct task_struct *idle;
1004 struct task_struct *stop;
1005 unsigned long next_balance;
1006 struct mm_struct *prev_mm;
1008 unsigned int clock_update_flags;
1010 /* Ensure that all clocks are in the same cache line */
1011 u64 clock_task ____cacheline_aligned;
1013 unsigned long lost_idle_time;
1014 u64 clock_pelt_idle;
1016 #ifndef CONFIG_64BIT
1017 u64 clock_pelt_idle_copy;
1018 u64 clock_idle_copy;
1023 #ifdef CONFIG_SCHED_DEBUG
1024 u64 last_seen_need_resched_ns;
1025 int ticks_without_resched;
1028 #ifdef CONFIG_MEMBARRIER
1029 int membarrier_state;
1033 struct root_domain *rd;
1034 struct sched_domain __rcu *sd;
1036 unsigned long cpu_capacity;
1037 unsigned long cpu_capacity_orig;
1039 struct callback_head *balance_callback;
1041 unsigned char nohz_idle_balance;
1042 unsigned char idle_balance;
1044 unsigned long misfit_task_load;
1046 /* For active balancing */
1049 struct cpu_stop_work active_balance_work;
1051 /* CPU of this runqueue: */
1055 struct list_head cfs_tasks;
1057 struct sched_avg avg_rt;
1058 struct sched_avg avg_dl;
1059 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1060 struct sched_avg avg_irq;
1062 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1063 struct sched_avg avg_thermal;
1068 unsigned long wake_stamp;
1071 /* This is used to determine avg_idle's max value */
1072 u64 max_idle_balance_cost;
1074 #ifdef CONFIG_HOTPLUG_CPU
1075 struct rcuwait hotplug_wait;
1077 #endif /* CONFIG_SMP */
1079 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1082 #ifdef CONFIG_PARAVIRT
1083 u64 prev_steal_time;
1085 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1086 u64 prev_steal_time_rq;
1089 /* calc_load related fields */
1090 unsigned long calc_load_update;
1091 long calc_load_active;
1093 #ifdef CONFIG_SCHED_HRTICK
1095 call_single_data_t hrtick_csd;
1097 struct hrtimer hrtick_timer;
1098 ktime_t hrtick_time;
1101 #ifdef CONFIG_SCHEDSTATS
1103 struct sched_info rq_sched_info;
1104 unsigned long long rq_cpu_time;
1105 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1107 /* sys_sched_yield() stats */
1108 unsigned int yld_count;
1110 /* schedule() stats */
1111 unsigned int sched_count;
1112 unsigned int sched_goidle;
1114 /* try_to_wake_up() stats */
1115 unsigned int ttwu_count;
1116 unsigned int ttwu_local;
1119 #ifdef CONFIG_CPU_IDLE
1120 /* Must be inspected within a rcu lock section */
1121 struct cpuidle_state *idle_state;
1125 unsigned int nr_pinned;
1127 unsigned int push_busy;
1128 struct cpu_stop_work push_work;
1130 #ifdef CONFIG_SCHED_CORE
1133 struct task_struct *core_pick;
1134 unsigned int core_enabled;
1135 unsigned int core_sched_seq;
1136 struct rb_root core_tree;
1138 /* shared state -- careful with sched_core_cpu_deactivate() */
1139 unsigned int core_task_seq;
1140 unsigned int core_pick_seq;
1141 unsigned long core_cookie;
1142 unsigned int core_forceidle_count;
1143 unsigned int core_forceidle_seq;
1144 unsigned int core_forceidle_occupation;
1145 u64 core_forceidle_start;
1149 #ifdef CONFIG_FAIR_GROUP_SCHED
1151 /* CPU runqueue to which this cfs_rq is attached */
1152 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1159 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1161 return container_of(cfs_rq, struct rq, cfs);
1165 static inline int cpu_of(struct rq *rq)
1174 #define MDF_PUSH 0x01
1176 static inline bool is_migration_disabled(struct task_struct *p)
1179 return p->migration_disabled;
1186 #ifdef CONFIG_SCHED_CORE
1187 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1189 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1191 static inline bool sched_core_enabled(struct rq *rq)
1193 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1196 static inline bool sched_core_disabled(void)
1198 return !static_branch_unlikely(&__sched_core_enabled);
1202 * Be careful with this function; not for general use. The return value isn't
1203 * stable unless you actually hold a relevant rq->__lock.
1205 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1207 if (sched_core_enabled(rq))
1208 return &rq->core->__lock;
1213 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1215 if (rq->core_enabled)
1216 return &rq->core->__lock;
1221 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1224 * Helpers to check if the CPU's core cookie matches with the task's cookie
1225 * when core scheduling is enabled.
1226 * A special case is that the task's cookie always matches with CPU's core
1227 * cookie if the CPU is in an idle core.
1229 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1231 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1232 if (!sched_core_enabled(rq))
1235 return rq->core->core_cookie == p->core_cookie;
1238 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1240 bool idle_core = true;
1243 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1244 if (!sched_core_enabled(rq))
1247 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1248 if (!available_idle_cpu(cpu)) {
1255 * A CPU in an idle core is always the best choice for tasks with
1258 return idle_core || rq->core->core_cookie == p->core_cookie;
1261 static inline bool sched_group_cookie_match(struct rq *rq,
1262 struct task_struct *p,
1263 struct sched_group *group)
1267 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1268 if (!sched_core_enabled(rq))
1271 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1272 if (sched_core_cookie_match(rq, p))
1278 static inline bool sched_core_enqueued(struct task_struct *p)
1280 return !RB_EMPTY_NODE(&p->core_node);
1283 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1284 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1286 extern void sched_core_get(void);
1287 extern void sched_core_put(void);
1289 #else /* !CONFIG_SCHED_CORE */
1291 static inline bool sched_core_enabled(struct rq *rq)
1296 static inline bool sched_core_disabled(void)
1301 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1306 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1311 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1316 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1321 static inline bool sched_group_cookie_match(struct rq *rq,
1322 struct task_struct *p,
1323 struct sched_group *group)
1327 #endif /* CONFIG_SCHED_CORE */
1329 static inline void lockdep_assert_rq_held(struct rq *rq)
1331 lockdep_assert_held(__rq_lockp(rq));
1334 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1335 extern bool raw_spin_rq_trylock(struct rq *rq);
1336 extern void raw_spin_rq_unlock(struct rq *rq);
1338 static inline void raw_spin_rq_lock(struct rq *rq)
1340 raw_spin_rq_lock_nested(rq, 0);
1343 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1345 local_irq_disable();
1346 raw_spin_rq_lock(rq);
1349 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1351 raw_spin_rq_unlock(rq);
1355 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1357 unsigned long flags;
1358 local_irq_save(flags);
1359 raw_spin_rq_lock(rq);
1363 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1365 raw_spin_rq_unlock(rq);
1366 local_irq_restore(flags);
1369 #define raw_spin_rq_lock_irqsave(rq, flags) \
1371 flags = _raw_spin_rq_lock_irqsave(rq); \
1374 #ifdef CONFIG_SCHED_SMT
1375 extern void __update_idle_core(struct rq *rq);
1377 static inline void update_idle_core(struct rq *rq)
1379 if (static_branch_unlikely(&sched_smt_present))
1380 __update_idle_core(rq);
1384 static inline void update_idle_core(struct rq *rq) { }
1387 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1389 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1390 #define this_rq() this_cpu_ptr(&runqueues)
1391 #define task_rq(p) cpu_rq(task_cpu(p))
1392 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1393 #define raw_rq() raw_cpu_ptr(&runqueues)
1395 #ifdef CONFIG_FAIR_GROUP_SCHED
1396 static inline struct task_struct *task_of(struct sched_entity *se)
1398 SCHED_WARN_ON(!entity_is_task(se));
1399 return container_of(se, struct task_struct, se);
1402 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1404 return p->se.cfs_rq;
1407 /* runqueue on which this entity is (to be) queued */
1408 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1413 /* runqueue "owned" by this group */
1414 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1421 static inline struct task_struct *task_of(struct sched_entity *se)
1423 return container_of(se, struct task_struct, se);
1426 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1428 return &task_rq(p)->cfs;
1431 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1433 struct task_struct *p = task_of(se);
1434 struct rq *rq = task_rq(p);
1439 /* runqueue "owned" by this group */
1440 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1446 extern void update_rq_clock(struct rq *rq);
1449 * rq::clock_update_flags bits
1451 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1452 * call to __schedule(). This is an optimisation to avoid
1453 * neighbouring rq clock updates.
1455 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1456 * in effect and calls to update_rq_clock() are being ignored.
1458 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1459 * made to update_rq_clock() since the last time rq::lock was pinned.
1461 * If inside of __schedule(), clock_update_flags will have been
1462 * shifted left (a left shift is a cheap operation for the fast path
1463 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1465 * if (rq-clock_update_flags >= RQCF_UPDATED)
1467 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1468 * one position though, because the next rq_unpin_lock() will shift it
1471 #define RQCF_REQ_SKIP 0x01
1472 #define RQCF_ACT_SKIP 0x02
1473 #define RQCF_UPDATED 0x04
1475 static inline void assert_clock_updated(struct rq *rq)
1478 * The only reason for not seeing a clock update since the
1479 * last rq_pin_lock() is if we're currently skipping updates.
1481 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1484 static inline u64 rq_clock(struct rq *rq)
1486 lockdep_assert_rq_held(rq);
1487 assert_clock_updated(rq);
1492 static inline u64 rq_clock_task(struct rq *rq)
1494 lockdep_assert_rq_held(rq);
1495 assert_clock_updated(rq);
1497 return rq->clock_task;
1501 * By default the decay is the default pelt decay period.
1502 * The decay shift can change the decay period in
1504 * Decay shift Decay period(ms)
1511 extern int sched_thermal_decay_shift;
1513 static inline u64 rq_clock_thermal(struct rq *rq)
1515 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1518 static inline void rq_clock_skip_update(struct rq *rq)
1520 lockdep_assert_rq_held(rq);
1521 rq->clock_update_flags |= RQCF_REQ_SKIP;
1525 * See rt task throttling, which is the only time a skip
1526 * request is canceled.
1528 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1530 lockdep_assert_rq_held(rq);
1531 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1535 unsigned long flags;
1536 struct pin_cookie cookie;
1537 #ifdef CONFIG_SCHED_DEBUG
1539 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1540 * current pin context is stashed here in case it needs to be
1541 * restored in rq_repin_lock().
1543 unsigned int clock_update_flags;
1547 extern struct callback_head balance_push_callback;
1550 * Lockdep annotation that avoids accidental unlocks; it's like a
1551 * sticky/continuous lockdep_assert_held().
1553 * This avoids code that has access to 'struct rq *rq' (basically everything in
1554 * the scheduler) from accidentally unlocking the rq if they do not also have a
1555 * copy of the (on-stack) 'struct rq_flags rf'.
1557 * Also see Documentation/locking/lockdep-design.rst.
1559 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1561 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1563 #ifdef CONFIG_SCHED_DEBUG
1564 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1565 rf->clock_update_flags = 0;
1567 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1572 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1574 #ifdef CONFIG_SCHED_DEBUG
1575 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1576 rf->clock_update_flags = RQCF_UPDATED;
1579 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1582 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1584 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1586 #ifdef CONFIG_SCHED_DEBUG
1588 * Restore the value we stashed in @rf for this pin context.
1590 rq->clock_update_flags |= rf->clock_update_flags;
1594 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1595 __acquires(rq->lock);
1597 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1598 __acquires(p->pi_lock)
1599 __acquires(rq->lock);
1601 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1602 __releases(rq->lock)
1604 rq_unpin_lock(rq, rf);
1605 raw_spin_rq_unlock(rq);
1609 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1610 __releases(rq->lock)
1611 __releases(p->pi_lock)
1613 rq_unpin_lock(rq, rf);
1614 raw_spin_rq_unlock(rq);
1615 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1619 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1620 __acquires(rq->lock)
1622 raw_spin_rq_lock_irqsave(rq, rf->flags);
1623 rq_pin_lock(rq, rf);
1627 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1628 __acquires(rq->lock)
1630 raw_spin_rq_lock_irq(rq);
1631 rq_pin_lock(rq, rf);
1635 rq_lock(struct rq *rq, struct rq_flags *rf)
1636 __acquires(rq->lock)
1638 raw_spin_rq_lock(rq);
1639 rq_pin_lock(rq, rf);
1643 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1644 __releases(rq->lock)
1646 rq_unpin_lock(rq, rf);
1647 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1651 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1652 __releases(rq->lock)
1654 rq_unpin_lock(rq, rf);
1655 raw_spin_rq_unlock_irq(rq);
1659 rq_unlock(struct rq *rq, struct rq_flags *rf)
1660 __releases(rq->lock)
1662 rq_unpin_lock(rq, rf);
1663 raw_spin_rq_unlock(rq);
1666 static inline struct rq *
1667 this_rq_lock_irq(struct rq_flags *rf)
1668 __acquires(rq->lock)
1672 local_irq_disable();
1679 enum numa_topology_type {
1684 extern enum numa_topology_type sched_numa_topology_type;
1685 extern int sched_max_numa_distance;
1686 extern bool find_numa_distance(int distance);
1687 extern void sched_init_numa(int offline_node);
1688 extern void sched_update_numa(int cpu, bool online);
1689 extern void sched_domains_numa_masks_set(unsigned int cpu);
1690 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1691 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1693 static inline void sched_init_numa(int offline_node) { }
1694 static inline void sched_update_numa(int cpu, bool online) { }
1695 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1696 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1697 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1703 #ifdef CONFIG_NUMA_BALANCING
1704 /* The regions in numa_faults array from task_struct */
1705 enum numa_faults_stats {
1711 extern void sched_setnuma(struct task_struct *p, int node);
1712 extern int migrate_task_to(struct task_struct *p, int cpu);
1713 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1715 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1718 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1721 #endif /* CONFIG_NUMA_BALANCING */
1726 queue_balance_callback(struct rq *rq,
1727 struct callback_head *head,
1728 void (*func)(struct rq *rq))
1730 lockdep_assert_rq_held(rq);
1733 * Don't (re)queue an already queued item; nor queue anything when
1734 * balance_push() is active, see the comment with
1735 * balance_push_callback.
1737 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1740 head->func = (void (*)(struct callback_head *))func;
1741 head->next = rq->balance_callback;
1742 rq->balance_callback = head;
1745 #define rcu_dereference_check_sched_domain(p) \
1746 rcu_dereference_check((p), \
1747 lockdep_is_held(&sched_domains_mutex))
1750 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1751 * See destroy_sched_domains: call_rcu for details.
1753 * The domain tree of any CPU may only be accessed from within
1754 * preempt-disabled sections.
1756 #define for_each_domain(cpu, __sd) \
1757 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1758 __sd; __sd = __sd->parent)
1761 * highest_flag_domain - Return highest sched_domain containing flag.
1762 * @cpu: The CPU whose highest level of sched domain is to
1764 * @flag: The flag to check for the highest sched_domain
1765 * for the given CPU.
1767 * Returns the highest sched_domain of a CPU which contains the given flag.
1769 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1771 struct sched_domain *sd, *hsd = NULL;
1773 for_each_domain(cpu, sd) {
1774 if (!(sd->flags & flag))
1782 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1784 struct sched_domain *sd;
1786 for_each_domain(cpu, sd) {
1787 if (sd->flags & flag)
1794 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1795 DECLARE_PER_CPU(int, sd_llc_size);
1796 DECLARE_PER_CPU(int, sd_llc_id);
1797 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1798 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1799 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1800 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1801 extern struct static_key_false sched_asym_cpucapacity;
1803 static __always_inline bool sched_asym_cpucap_active(void)
1805 return static_branch_unlikely(&sched_asym_cpucapacity);
1808 struct sched_group_capacity {
1811 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1814 unsigned long capacity;
1815 unsigned long min_capacity; /* Min per-CPU capacity in group */
1816 unsigned long max_capacity; /* Max per-CPU capacity in group */
1817 unsigned long next_update;
1818 int imbalance; /* XXX unrelated to capacity but shared group state */
1820 #ifdef CONFIG_SCHED_DEBUG
1824 unsigned long cpumask[]; /* Balance mask */
1827 struct sched_group {
1828 struct sched_group *next; /* Must be a circular list */
1831 unsigned int group_weight;
1832 struct sched_group_capacity *sgc;
1833 int asym_prefer_cpu; /* CPU of highest priority in group */
1837 * The CPUs this group covers.
1839 * NOTE: this field is variable length. (Allocated dynamically
1840 * by attaching extra space to the end of the structure,
1841 * depending on how many CPUs the kernel has booted up with)
1843 unsigned long cpumask[];
1846 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1848 return to_cpumask(sg->cpumask);
1852 * See build_balance_mask().
1854 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1856 return to_cpumask(sg->sgc->cpumask);
1859 extern int group_balance_cpu(struct sched_group *sg);
1861 #ifdef CONFIG_SCHED_DEBUG
1862 void update_sched_domain_debugfs(void);
1863 void dirty_sched_domain_sysctl(int cpu);
1865 static inline void update_sched_domain_debugfs(void)
1868 static inline void dirty_sched_domain_sysctl(int cpu)
1873 extern int sched_update_scaling(void);
1874 #endif /* CONFIG_SMP */
1878 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1880 extern void __sched_core_account_forceidle(struct rq *rq);
1882 static inline void sched_core_account_forceidle(struct rq *rq)
1884 if (schedstat_enabled())
1885 __sched_core_account_forceidle(rq);
1888 extern void __sched_core_tick(struct rq *rq);
1890 static inline void sched_core_tick(struct rq *rq)
1892 if (sched_core_enabled(rq) && schedstat_enabled())
1893 __sched_core_tick(rq);
1898 static inline void sched_core_account_forceidle(struct rq *rq) {}
1900 static inline void sched_core_tick(struct rq *rq) {}
1902 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1904 #ifdef CONFIG_CGROUP_SCHED
1907 * Return the group to which this tasks belongs.
1909 * We cannot use task_css() and friends because the cgroup subsystem
1910 * changes that value before the cgroup_subsys::attach() method is called,
1911 * therefore we cannot pin it and might observe the wrong value.
1913 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1914 * core changes this before calling sched_move_task().
1916 * Instead we use a 'copy' which is updated from sched_move_task() while
1917 * holding both task_struct::pi_lock and rq::lock.
1919 static inline struct task_group *task_group(struct task_struct *p)
1921 return p->sched_task_group;
1924 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1925 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1927 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1928 struct task_group *tg = task_group(p);
1931 #ifdef CONFIG_FAIR_GROUP_SCHED
1932 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1933 p->se.cfs_rq = tg->cfs_rq[cpu];
1934 p->se.parent = tg->se[cpu];
1935 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
1938 #ifdef CONFIG_RT_GROUP_SCHED
1939 p->rt.rt_rq = tg->rt_rq[cpu];
1940 p->rt.parent = tg->rt_se[cpu];
1944 #else /* CONFIG_CGROUP_SCHED */
1946 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1947 static inline struct task_group *task_group(struct task_struct *p)
1952 #endif /* CONFIG_CGROUP_SCHED */
1954 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1956 set_task_rq(p, cpu);
1959 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1960 * successfully executed on another CPU. We must ensure that updates of
1961 * per-task data have been completed by this moment.
1964 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1970 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1972 #ifdef CONFIG_SCHED_DEBUG
1973 # define const_debug __read_mostly
1975 # define const_debug const
1978 #define SCHED_FEAT(name, enabled) \
1979 __SCHED_FEAT_##name ,
1982 #include "features.h"
1988 #ifdef CONFIG_SCHED_DEBUG
1991 * To support run-time toggling of sched features, all the translation units
1992 * (but core.c) reference the sysctl_sched_features defined in core.c.
1994 extern const_debug unsigned int sysctl_sched_features;
1996 #ifdef CONFIG_JUMP_LABEL
1997 #define SCHED_FEAT(name, enabled) \
1998 static __always_inline bool static_branch_##name(struct static_key *key) \
2000 return static_key_##enabled(key); \
2003 #include "features.h"
2006 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2007 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2009 #else /* !CONFIG_JUMP_LABEL */
2011 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2013 #endif /* CONFIG_JUMP_LABEL */
2015 #else /* !SCHED_DEBUG */
2018 * Each translation unit has its own copy of sysctl_sched_features to allow
2019 * constants propagation at compile time and compiler optimization based on
2022 #define SCHED_FEAT(name, enabled) \
2023 (1UL << __SCHED_FEAT_##name) * enabled |
2024 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2025 #include "features.h"
2029 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2031 #endif /* SCHED_DEBUG */
2033 extern struct static_key_false sched_numa_balancing;
2034 extern struct static_key_false sched_schedstats;
2036 static inline u64 global_rt_period(void)
2038 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2041 static inline u64 global_rt_runtime(void)
2043 if (sysctl_sched_rt_runtime < 0)
2046 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2049 static inline int task_current(struct rq *rq, struct task_struct *p)
2051 return rq->curr == p;
2054 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2059 return task_current(rq, p);
2063 static inline int task_on_rq_queued(struct task_struct *p)
2065 return p->on_rq == TASK_ON_RQ_QUEUED;
2068 static inline int task_on_rq_migrating(struct task_struct *p)
2070 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2073 /* Wake flags. The first three directly map to some SD flag value */
2074 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2075 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2076 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2078 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2079 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2082 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2083 static_assert(WF_FORK == SD_BALANCE_FORK);
2084 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2088 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2089 * of tasks with abnormal "nice" values across CPUs the contribution that
2090 * each task makes to its run queue's load is weighted according to its
2091 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2092 * scaled version of the new time slice allocation that they receive on time
2096 #define WEIGHT_IDLEPRIO 3
2097 #define WMULT_IDLEPRIO 1431655765
2099 extern const int sched_prio_to_weight[40];
2100 extern const u32 sched_prio_to_wmult[40];
2103 * {de,en}queue flags:
2105 * DEQUEUE_SLEEP - task is no longer runnable
2106 * ENQUEUE_WAKEUP - task just became runnable
2108 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2109 * are in a known state which allows modification. Such pairs
2110 * should preserve as much state as possible.
2112 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2115 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2116 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2117 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2121 #define DEQUEUE_SLEEP 0x01
2122 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2123 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2124 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2126 #define ENQUEUE_WAKEUP 0x01
2127 #define ENQUEUE_RESTORE 0x02
2128 #define ENQUEUE_MOVE 0x04
2129 #define ENQUEUE_NOCLOCK 0x08
2131 #define ENQUEUE_HEAD 0x10
2132 #define ENQUEUE_REPLENISH 0x20
2134 #define ENQUEUE_MIGRATED 0x40
2136 #define ENQUEUE_MIGRATED 0x00
2139 #define RETRY_TASK ((void *)-1UL)
2141 struct sched_class {
2143 #ifdef CONFIG_UCLAMP_TASK
2147 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2148 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2149 void (*yield_task) (struct rq *rq);
2150 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2152 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2154 struct task_struct *(*pick_next_task)(struct rq *rq);
2156 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2157 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2160 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2161 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2163 struct task_struct * (*pick_task)(struct rq *rq);
2165 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2167 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2169 void (*set_cpus_allowed)(struct task_struct *p,
2170 const struct cpumask *newmask,
2173 void (*rq_online)(struct rq *rq);
2174 void (*rq_offline)(struct rq *rq);
2176 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2179 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2180 void (*task_fork)(struct task_struct *p);
2181 void (*task_dead)(struct task_struct *p);
2184 * The switched_from() call is allowed to drop rq->lock, therefore we
2185 * cannot assume the switched_from/switched_to pair is serialized by
2186 * rq->lock. They are however serialized by p->pi_lock.
2188 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2189 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2190 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2193 unsigned int (*get_rr_interval)(struct rq *rq,
2194 struct task_struct *task);
2196 void (*update_curr)(struct rq *rq);
2198 #ifdef CONFIG_FAIR_GROUP_SCHED
2199 void (*task_change_group)(struct task_struct *p);
2203 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2205 WARN_ON_ONCE(rq->curr != prev);
2206 prev->sched_class->put_prev_task(rq, prev);
2209 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2211 next->sched_class->set_next_task(rq, next, false);
2216 * Helper to define a sched_class instance; each one is placed in a separate
2217 * section which is ordered by the linker script:
2219 * include/asm-generic/vmlinux.lds.h
2221 * *CAREFUL* they are laid out in *REVERSE* order!!!
2223 * Also enforce alignment on the instance, not the type, to guarantee layout.
2225 #define DEFINE_SCHED_CLASS(name) \
2226 const struct sched_class name##_sched_class \
2227 __aligned(__alignof__(struct sched_class)) \
2228 __section("__" #name "_sched_class")
2230 /* Defined in include/asm-generic/vmlinux.lds.h */
2231 extern struct sched_class __sched_class_highest[];
2232 extern struct sched_class __sched_class_lowest[];
2234 #define for_class_range(class, _from, _to) \
2235 for (class = (_from); class < (_to); class++)
2237 #define for_each_class(class) \
2238 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2240 #define sched_class_above(_a, _b) ((_a) < (_b))
2242 extern const struct sched_class stop_sched_class;
2243 extern const struct sched_class dl_sched_class;
2244 extern const struct sched_class rt_sched_class;
2245 extern const struct sched_class fair_sched_class;
2246 extern const struct sched_class idle_sched_class;
2248 static inline bool sched_stop_runnable(struct rq *rq)
2250 return rq->stop && task_on_rq_queued(rq->stop);
2253 static inline bool sched_dl_runnable(struct rq *rq)
2255 return rq->dl.dl_nr_running > 0;
2258 static inline bool sched_rt_runnable(struct rq *rq)
2260 return rq->rt.rt_queued > 0;
2263 static inline bool sched_fair_runnable(struct rq *rq)
2265 return rq->cfs.nr_running > 0;
2268 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2269 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2271 #define SCA_CHECK 0x01
2272 #define SCA_MIGRATE_DISABLE 0x02
2273 #define SCA_MIGRATE_ENABLE 0x04
2274 #define SCA_USER 0x08
2278 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2280 extern void trigger_load_balance(struct rq *rq);
2282 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2284 static inline struct task_struct *get_push_task(struct rq *rq)
2286 struct task_struct *p = rq->curr;
2288 lockdep_assert_rq_held(rq);
2293 if (p->nr_cpus_allowed == 1)
2296 if (p->migration_disabled)
2299 rq->push_busy = true;
2300 return get_task_struct(p);
2303 extern int push_cpu_stop(void *arg);
2307 #ifdef CONFIG_CPU_IDLE
2308 static inline void idle_set_state(struct rq *rq,
2309 struct cpuidle_state *idle_state)
2311 rq->idle_state = idle_state;
2314 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2316 SCHED_WARN_ON(!rcu_read_lock_held());
2318 return rq->idle_state;
2321 static inline void idle_set_state(struct rq *rq,
2322 struct cpuidle_state *idle_state)
2326 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2332 extern void schedule_idle(void);
2334 extern void sysrq_sched_debug_show(void);
2335 extern void sched_init_granularity(void);
2336 extern void update_max_interval(void);
2338 extern void init_sched_dl_class(void);
2339 extern void init_sched_rt_class(void);
2340 extern void init_sched_fair_class(void);
2342 extern void reweight_task(struct task_struct *p, int prio);
2344 extern void resched_curr(struct rq *rq);
2345 extern void resched_cpu(int cpu);
2347 extern struct rt_bandwidth def_rt_bandwidth;
2348 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2349 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2351 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2352 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2353 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2356 #define BW_UNIT (1 << BW_SHIFT)
2357 #define RATIO_SHIFT 8
2358 #define MAX_BW_BITS (64 - BW_SHIFT)
2359 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2360 unsigned long to_ratio(u64 period, u64 runtime);
2362 extern void init_entity_runnable_average(struct sched_entity *se);
2363 extern void post_init_entity_util_avg(struct task_struct *p);
2365 #ifdef CONFIG_NO_HZ_FULL
2366 extern bool sched_can_stop_tick(struct rq *rq);
2367 extern int __init sched_tick_offload_init(void);
2370 * Tick may be needed by tasks in the runqueue depending on their policy and
2371 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2372 * nohz mode if necessary.
2374 static inline void sched_update_tick_dependency(struct rq *rq)
2376 int cpu = cpu_of(rq);
2378 if (!tick_nohz_full_cpu(cpu))
2381 if (sched_can_stop_tick(rq))
2382 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2384 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2387 static inline int sched_tick_offload_init(void) { return 0; }
2388 static inline void sched_update_tick_dependency(struct rq *rq) { }
2391 static inline void add_nr_running(struct rq *rq, unsigned count)
2393 unsigned prev_nr = rq->nr_running;
2395 rq->nr_running = prev_nr + count;
2396 if (trace_sched_update_nr_running_tp_enabled()) {
2397 call_trace_sched_update_nr_running(rq, count);
2401 if (prev_nr < 2 && rq->nr_running >= 2) {
2402 if (!READ_ONCE(rq->rd->overload))
2403 WRITE_ONCE(rq->rd->overload, 1);
2407 sched_update_tick_dependency(rq);
2410 static inline void sub_nr_running(struct rq *rq, unsigned count)
2412 rq->nr_running -= count;
2413 if (trace_sched_update_nr_running_tp_enabled()) {
2414 call_trace_sched_update_nr_running(rq, -count);
2417 /* Check if we still need preemption */
2418 sched_update_tick_dependency(rq);
2421 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2422 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2424 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2426 #ifdef CONFIG_PREEMPT_RT
2427 #define SCHED_NR_MIGRATE_BREAK 8
2429 #define SCHED_NR_MIGRATE_BREAK 32
2432 extern const_debug unsigned int sysctl_sched_nr_migrate;
2433 extern const_debug unsigned int sysctl_sched_migration_cost;
2435 #ifdef CONFIG_SCHED_DEBUG
2436 extern unsigned int sysctl_sched_latency;
2437 extern unsigned int sysctl_sched_min_granularity;
2438 extern unsigned int sysctl_sched_idle_min_granularity;
2439 extern unsigned int sysctl_sched_wakeup_granularity;
2440 extern int sysctl_resched_latency_warn_ms;
2441 extern int sysctl_resched_latency_warn_once;
2443 extern unsigned int sysctl_sched_tunable_scaling;
2445 extern unsigned int sysctl_numa_balancing_scan_delay;
2446 extern unsigned int sysctl_numa_balancing_scan_period_min;
2447 extern unsigned int sysctl_numa_balancing_scan_period_max;
2448 extern unsigned int sysctl_numa_balancing_scan_size;
2449 extern unsigned int sysctl_numa_balancing_hot_threshold;
2452 #ifdef CONFIG_SCHED_HRTICK
2456 * - enabled by features
2457 * - hrtimer is actually high res
2459 static inline int hrtick_enabled(struct rq *rq)
2461 if (!cpu_active(cpu_of(rq)))
2463 return hrtimer_is_hres_active(&rq->hrtick_timer);
2466 static inline int hrtick_enabled_fair(struct rq *rq)
2468 if (!sched_feat(HRTICK))
2470 return hrtick_enabled(rq);
2473 static inline int hrtick_enabled_dl(struct rq *rq)
2475 if (!sched_feat(HRTICK_DL))
2477 return hrtick_enabled(rq);
2480 void hrtick_start(struct rq *rq, u64 delay);
2484 static inline int hrtick_enabled_fair(struct rq *rq)
2489 static inline int hrtick_enabled_dl(struct rq *rq)
2494 static inline int hrtick_enabled(struct rq *rq)
2499 #endif /* CONFIG_SCHED_HRTICK */
2501 #ifndef arch_scale_freq_tick
2502 static __always_inline
2503 void arch_scale_freq_tick(void)
2508 #ifndef arch_scale_freq_capacity
2510 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2511 * @cpu: the CPU in question.
2513 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2516 * ------ * SCHED_CAPACITY_SCALE
2519 static __always_inline
2520 unsigned long arch_scale_freq_capacity(int cpu)
2522 return SCHED_CAPACITY_SCALE;
2526 #ifdef CONFIG_SCHED_DEBUG
2528 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2529 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2530 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2531 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2533 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2535 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2536 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2538 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2542 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2547 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2549 #ifdef CONFIG_SCHED_CORE
2551 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2552 * order by core-id first and cpu-id second.
2556 * double_rq_lock(0,3); will take core-0, core-1 lock
2557 * double_rq_lock(1,2); will take core-1, core-0 lock
2559 * when only cpu-id is considered.
2561 if (rq1->core->cpu < rq2->core->cpu)
2563 if (rq1->core->cpu > rq2->core->cpu)
2567 * __sched_core_flip() relies on SMT having cpu-id lock order.
2570 return rq1->cpu < rq2->cpu;
2573 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2575 #ifdef CONFIG_PREEMPTION
2578 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2579 * way at the expense of forcing extra atomic operations in all
2580 * invocations. This assures that the double_lock is acquired using the
2581 * same underlying policy as the spinlock_t on this architecture, which
2582 * reduces latency compared to the unfair variant below. However, it
2583 * also adds more overhead and therefore may reduce throughput.
2585 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2586 __releases(this_rq->lock)
2587 __acquires(busiest->lock)
2588 __acquires(this_rq->lock)
2590 raw_spin_rq_unlock(this_rq);
2591 double_rq_lock(this_rq, busiest);
2598 * Unfair double_lock_balance: Optimizes throughput at the expense of
2599 * latency by eliminating extra atomic operations when the locks are
2600 * already in proper order on entry. This favors lower CPU-ids and will
2601 * grant the double lock to lower CPUs over higher ids under contention,
2602 * regardless of entry order into the function.
2604 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2605 __releases(this_rq->lock)
2606 __acquires(busiest->lock)
2607 __acquires(this_rq->lock)
2609 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2610 likely(raw_spin_rq_trylock(busiest))) {
2611 double_rq_clock_clear_update(this_rq, busiest);
2615 if (rq_order_less(this_rq, busiest)) {
2616 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2617 double_rq_clock_clear_update(this_rq, busiest);
2621 raw_spin_rq_unlock(this_rq);
2622 double_rq_lock(this_rq, busiest);
2627 #endif /* CONFIG_PREEMPTION */
2630 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2632 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2634 lockdep_assert_irqs_disabled();
2636 return _double_lock_balance(this_rq, busiest);
2639 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2640 __releases(busiest->lock)
2642 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2643 raw_spin_rq_unlock(busiest);
2644 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2647 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2653 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2656 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2662 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2665 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2671 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2675 * double_rq_unlock - safely unlock two runqueues
2677 * Note this does not restore interrupts like task_rq_unlock,
2678 * you need to do so manually after calling.
2680 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2681 __releases(rq1->lock)
2682 __releases(rq2->lock)
2684 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2685 raw_spin_rq_unlock(rq2);
2687 __release(rq2->lock);
2688 raw_spin_rq_unlock(rq1);
2691 extern void set_rq_online (struct rq *rq);
2692 extern void set_rq_offline(struct rq *rq);
2693 extern bool sched_smp_initialized;
2695 #else /* CONFIG_SMP */
2698 * double_rq_lock - safely lock two runqueues
2700 * Note this does not disable interrupts like task_rq_lock,
2701 * you need to do so manually before calling.
2703 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2704 __acquires(rq1->lock)
2705 __acquires(rq2->lock)
2707 WARN_ON_ONCE(!irqs_disabled());
2708 WARN_ON_ONCE(rq1 != rq2);
2709 raw_spin_rq_lock(rq1);
2710 __acquire(rq2->lock); /* Fake it out ;) */
2711 double_rq_clock_clear_update(rq1, rq2);
2715 * double_rq_unlock - safely unlock two runqueues
2717 * Note this does not restore interrupts like task_rq_unlock,
2718 * you need to do so manually after calling.
2720 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2721 __releases(rq1->lock)
2722 __releases(rq2->lock)
2724 WARN_ON_ONCE(rq1 != rq2);
2725 raw_spin_rq_unlock(rq1);
2726 __release(rq2->lock);
2731 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2732 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2734 #ifdef CONFIG_SCHED_DEBUG
2735 extern bool sched_debug_verbose;
2737 extern void print_cfs_stats(struct seq_file *m, int cpu);
2738 extern void print_rt_stats(struct seq_file *m, int cpu);
2739 extern void print_dl_stats(struct seq_file *m, int cpu);
2740 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2741 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2742 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2744 extern void resched_latency_warn(int cpu, u64 latency);
2745 #ifdef CONFIG_NUMA_BALANCING
2747 show_numa_stats(struct task_struct *p, struct seq_file *m);
2749 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2750 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2751 #endif /* CONFIG_NUMA_BALANCING */
2753 static inline void resched_latency_warn(int cpu, u64 latency) {}
2754 #endif /* CONFIG_SCHED_DEBUG */
2756 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2757 extern void init_rt_rq(struct rt_rq *rt_rq);
2758 extern void init_dl_rq(struct dl_rq *dl_rq);
2760 extern void cfs_bandwidth_usage_inc(void);
2761 extern void cfs_bandwidth_usage_dec(void);
2763 #ifdef CONFIG_NO_HZ_COMMON
2764 #define NOHZ_BALANCE_KICK_BIT 0
2765 #define NOHZ_STATS_KICK_BIT 1
2766 #define NOHZ_NEWILB_KICK_BIT 2
2767 #define NOHZ_NEXT_KICK_BIT 3
2769 /* Run rebalance_domains() */
2770 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2771 /* Update blocked load */
2772 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2773 /* Update blocked load when entering idle */
2774 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2775 /* Update nohz.next_balance */
2776 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2778 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2780 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2782 extern void nohz_balance_exit_idle(struct rq *rq);
2784 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2787 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2788 extern void nohz_run_idle_balance(int cpu);
2790 static inline void nohz_run_idle_balance(int cpu) { }
2793 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2798 struct u64_stats_sync sync;
2801 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2804 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2805 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2806 * and never move forward.
2808 static inline u64 irq_time_read(int cpu)
2810 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2815 seq = __u64_stats_fetch_begin(&irqtime->sync);
2816 total = irqtime->total;
2817 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2821 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2823 #ifdef CONFIG_CPU_FREQ
2824 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2827 * cpufreq_update_util - Take a note about CPU utilization changes.
2828 * @rq: Runqueue to carry out the update for.
2829 * @flags: Update reason flags.
2831 * This function is called by the scheduler on the CPU whose utilization is
2834 * It can only be called from RCU-sched read-side critical sections.
2836 * The way cpufreq is currently arranged requires it to evaluate the CPU
2837 * performance state (frequency/voltage) on a regular basis to prevent it from
2838 * being stuck in a completely inadequate performance level for too long.
2839 * That is not guaranteed to happen if the updates are only triggered from CFS
2840 * and DL, though, because they may not be coming in if only RT tasks are
2841 * active all the time (or there are RT tasks only).
2843 * As a workaround for that issue, this function is called periodically by the
2844 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2845 * but that really is a band-aid. Going forward it should be replaced with
2846 * solutions targeted more specifically at RT tasks.
2848 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2850 struct update_util_data *data;
2852 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2855 data->func(data, rq_clock(rq), flags);
2858 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2859 #endif /* CONFIG_CPU_FREQ */
2861 #ifdef arch_scale_freq_capacity
2862 # ifndef arch_scale_freq_invariant
2863 # define arch_scale_freq_invariant() true
2866 # define arch_scale_freq_invariant() false
2870 static inline unsigned long capacity_orig_of(int cpu)
2872 return cpu_rq(cpu)->cpu_capacity_orig;
2876 * enum cpu_util_type - CPU utilization type
2877 * @FREQUENCY_UTIL: Utilization used to select frequency
2878 * @ENERGY_UTIL: Utilization used during energy calculation
2880 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2881 * need to be aggregated differently depending on the usage made of them. This
2882 * enum is used within effective_cpu_util() to differentiate the types of
2883 * utilization expected by the callers, and adjust the aggregation accordingly.
2885 enum cpu_util_type {
2890 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2891 enum cpu_util_type type,
2892 struct task_struct *p);
2895 * Verify the fitness of task @p to run on @cpu taking into account the
2896 * CPU original capacity and the runtime/deadline ratio of the task.
2898 * The function will return true if the original capacity of @cpu is
2899 * greater than or equal to task's deadline density right shifted by
2900 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2902 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2904 unsigned long cap = arch_scale_cpu_capacity(cpu);
2906 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2909 static inline unsigned long cpu_bw_dl(struct rq *rq)
2911 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2914 static inline unsigned long cpu_util_dl(struct rq *rq)
2916 return READ_ONCE(rq->avg_dl.util_avg);
2920 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2921 * @cpu: the CPU to get the utilization for.
2923 * The unit of the return value must be the same as the one of CPU capacity
2924 * so that CPU utilization can be compared with CPU capacity.
2926 * CPU utilization is the sum of running time of runnable tasks plus the
2927 * recent utilization of currently non-runnable tasks on that CPU.
2928 * It represents the amount of CPU capacity currently used by CFS tasks in
2929 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2930 * capacity at f_max.
2932 * The estimated CPU utilization is defined as the maximum between CPU
2933 * utilization and sum of the estimated utilization of the currently
2934 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2935 * previously-executed tasks, which helps better deduce how busy a CPU will
2936 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2937 * of such a task would be significantly decayed at this point of time.
2939 * CPU utilization can be higher than the current CPU capacity
2940 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2941 * of rounding errors as well as task migrations or wakeups of new tasks.
2942 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2943 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2944 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2945 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2946 * though since this is useful for predicting the CPU capacity required
2947 * after task migrations (scheduler-driven DVFS).
2949 * Return: (Estimated) utilization for the specified CPU.
2951 static inline unsigned long cpu_util_cfs(int cpu)
2953 struct cfs_rq *cfs_rq;
2956 cfs_rq = &cpu_rq(cpu)->cfs;
2957 util = READ_ONCE(cfs_rq->avg.util_avg);
2959 if (sched_feat(UTIL_EST)) {
2960 util = max_t(unsigned long, util,
2961 READ_ONCE(cfs_rq->avg.util_est.enqueued));
2964 return min(util, capacity_orig_of(cpu));
2967 static inline unsigned long cpu_util_rt(struct rq *rq)
2969 return READ_ONCE(rq->avg_rt.util_avg);
2973 #ifdef CONFIG_UCLAMP_TASK
2974 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2977 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2978 * @rq: The rq to clamp against. Must not be NULL.
2979 * @util: The util value to clamp.
2980 * @p: The task to clamp against. Can be NULL if you want to clamp
2983 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2985 * If sched_uclamp_used static key is disabled, then just return the util
2986 * without any clamping since uclamp aggregation at the rq level in the fast
2987 * path is disabled, rendering this operation a NOP.
2989 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2990 * will return the correct effective uclamp value of the task even if the
2991 * static key is disabled.
2993 static __always_inline
2994 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2995 struct task_struct *p)
2997 unsigned long min_util = 0;
2998 unsigned long max_util = 0;
3000 if (!static_branch_likely(&sched_uclamp_used))
3004 min_util = uclamp_eff_value(p, UCLAMP_MIN);
3005 max_util = uclamp_eff_value(p, UCLAMP_MAX);
3008 * Ignore last runnable task's max clamp, as this task will
3009 * reset it. Similarly, no need to read the rq's min clamp.
3011 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
3015 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
3016 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
3019 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3020 * RUNNABLE tasks with _different_ clamps, we can end up with an
3021 * inversion. Fix it now when the clamps are applied.
3023 if (unlikely(min_util >= max_util))
3026 return clamp(util, min_util, max_util);
3029 /* Is the rq being capped/throttled by uclamp_max? */
3030 static inline bool uclamp_rq_is_capped(struct rq *rq)
3032 unsigned long rq_util;
3033 unsigned long max_util;
3035 if (!static_branch_likely(&sched_uclamp_used))
3038 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3039 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3041 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3045 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3046 * by default in the fast path and only gets turned on once userspace performs
3047 * an operation that requires it.
3049 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3052 static inline bool uclamp_is_used(void)
3054 return static_branch_likely(&sched_uclamp_used);
3056 #else /* CONFIG_UCLAMP_TASK */
3058 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3059 struct task_struct *p)
3064 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3066 static inline bool uclamp_is_used(void)
3070 #endif /* CONFIG_UCLAMP_TASK */
3072 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3073 static inline unsigned long cpu_util_irq(struct rq *rq)
3075 return rq->avg_irq.util_avg;
3079 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3081 util *= (max - irq);
3088 static inline unsigned long cpu_util_irq(struct rq *rq)
3094 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3100 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3102 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3104 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3106 static inline bool sched_energy_enabled(void)
3108 return static_branch_unlikely(&sched_energy_present);
3111 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3113 #define perf_domain_span(pd) NULL
3114 static inline bool sched_energy_enabled(void) { return false; }
3116 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3118 #ifdef CONFIG_MEMBARRIER
3120 * The scheduler provides memory barriers required by membarrier between:
3121 * - prior user-space memory accesses and store to rq->membarrier_state,
3122 * - store to rq->membarrier_state and following user-space memory accesses.
3123 * In the same way it provides those guarantees around store to rq->curr.
3125 static inline void membarrier_switch_mm(struct rq *rq,
3126 struct mm_struct *prev_mm,
3127 struct mm_struct *next_mm)
3129 int membarrier_state;
3131 if (prev_mm == next_mm)
3134 membarrier_state = atomic_read(&next_mm->membarrier_state);
3135 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3138 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3141 static inline void membarrier_switch_mm(struct rq *rq,
3142 struct mm_struct *prev_mm,
3143 struct mm_struct *next_mm)
3149 static inline bool is_per_cpu_kthread(struct task_struct *p)
3151 if (!(p->flags & PF_KTHREAD))
3154 if (p->nr_cpus_allowed != 1)
3161 extern void swake_up_all_locked(struct swait_queue_head *q);
3162 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3164 #ifdef CONFIG_PREEMPT_DYNAMIC
3165 extern int preempt_dynamic_mode;
3166 extern int sched_dynamic_mode(const char *str);
3167 extern void sched_dynamic_update(int mode);
3170 static inline void update_current_exec_runtime(struct task_struct *curr,
3171 u64 now, u64 delta_exec)
3173 curr->se.sum_exec_runtime += delta_exec;
3174 account_group_exec_runtime(curr, delta_exec);
3176 curr->se.exec_start = now;
3177 cgroup_account_cputime(curr, delta_exec);
3180 #endif /* _KERNEL_SCHED_SCHED_H */