1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/bitops.h>
40 #include <linux/blkdev.h>
41 #include <linux/compat.h>
42 #include <linux/context_tracking.h>
43 #include <linux/cpufreq.h>
44 #include <linux/cpuidle.h>
45 #include <linux/cpuset.h>
46 #include <linux/ctype.h>
47 #include <linux/debugfs.h>
48 #include <linux/delayacct.h>
49 #include <linux/energy_model.h>
50 #include <linux/init_task.h>
51 #include <linux/kprobes.h>
52 #include <linux/kthread.h>
53 #include <linux/membarrier.h>
54 #include <linux/migrate.h>
55 #include <linux/mmu_context.h>
56 #include <linux/nmi.h>
57 #include <linux/proc_fs.h>
58 #include <linux/prefetch.h>
59 #include <linux/profile.h>
60 #include <linux/psi.h>
61 #include <linux/ratelimit.h>
62 #include <linux/rcupdate_wait.h>
63 #include <linux/security.h>
64 #include <linux/stop_machine.h>
65 #include <linux/suspend.h>
66 #include <linux/swait.h>
67 #include <linux/syscalls.h>
68 #include <linux/task_work.h>
69 #include <linux/tsacct_kern.h>
73 #ifdef CONFIG_PARAVIRT
74 # include <asm/paravirt.h>
78 #include "cpudeadline.h"
80 #include <trace/events/sched.h>
82 #ifdef CONFIG_SCHED_DEBUG
83 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
85 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
91 /* task_struct::on_rq states: */
92 #define TASK_ON_RQ_QUEUED 1
93 #define TASK_ON_RQ_MIGRATING 2
95 extern __read_mostly int scheduler_running;
97 extern unsigned long calc_load_update;
98 extern atomic_long_t calc_load_tasks;
100 extern void calc_global_load_tick(struct rq *this_rq);
101 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
103 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
105 * Helpers for converting nanosecond timing to jiffy resolution
107 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
116 * We increase resolution only if we have enough bits to allow this increased
117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
124 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load_down(w) \
128 unsigned long __w = (w); \
130 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
134 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
135 # define scale_load(w) (w)
136 # define scale_load_down(w) (w)
140 * Task weight (visible to users) and its load (invisible to users) have
141 * independent resolution, but they should be well calibrated. We use
142 * scale_load() and scale_load_down(w) to convert between them. The
143 * following must be true:
145 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
148 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
151 * Single value that decides SCHED_DEADLINE internal math precision.
152 * 10 -> just above 1us
153 * 9 -> just above 0.5us
158 * Single value that denotes runtime == period, ie unlimited time.
160 #define RUNTIME_INF ((u64)~0ULL)
162 static inline int idle_policy(int policy)
164 return policy == SCHED_IDLE;
166 static inline int fair_policy(int policy)
168 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
171 static inline int rt_policy(int policy)
173 return policy == SCHED_FIFO || policy == SCHED_RR;
176 static inline int dl_policy(int policy)
178 return policy == SCHED_DEADLINE;
180 static inline bool valid_policy(int policy)
182 return idle_policy(policy) || fair_policy(policy) ||
183 rt_policy(policy) || dl_policy(policy);
186 static inline int task_has_idle_policy(struct task_struct *p)
188 return idle_policy(p->policy);
191 static inline int task_has_rt_policy(struct task_struct *p)
193 return rt_policy(p->policy);
196 static inline int task_has_dl_policy(struct task_struct *p)
198 return dl_policy(p->policy);
201 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
203 static inline void update_avg(u64 *avg, u64 sample)
205 s64 diff = sample - *avg;
210 * Shifting a value by an exponent greater *or equal* to the size of said value
211 * is UB; cap at size-1.
213 #define shr_bound(val, shift) \
214 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
217 * !! For sched_setattr_nocheck() (kernel) only !!
219 * This is actually gross. :(
221 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
222 * tasks, but still be able to sleep. We need this on platforms that cannot
223 * atomically change clock frequency. Remove once fast switching will be
224 * available on such platforms.
226 * SUGOV stands for SchedUtil GOVernor.
228 #define SCHED_FLAG_SUGOV 0x10000000
230 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
232 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
233 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
240 * Tells if entity @a should preempt entity @b.
243 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
245 return dl_entity_is_special(a) ||
246 dl_time_before(a->deadline, b->deadline);
250 * This is the priority-queue data structure of the RT scheduling class:
252 struct rt_prio_array {
253 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
254 struct list_head queue[MAX_RT_PRIO];
257 struct rt_bandwidth {
258 /* nests inside the rq lock: */
259 raw_spinlock_t rt_runtime_lock;
262 struct hrtimer rt_period_timer;
263 unsigned int rt_period_active;
266 void __dl_clear_params(struct task_struct *p);
268 struct dl_bandwidth {
269 raw_spinlock_t dl_runtime_lock;
274 static inline int dl_bandwidth_enabled(void)
276 return sysctl_sched_rt_runtime >= 0;
280 * To keep the bandwidth of -deadline tasks under control
281 * we need some place where:
282 * - store the maximum -deadline bandwidth of each cpu;
283 * - cache the fraction of bandwidth that is currently allocated in
286 * This is all done in the data structure below. It is similar to the
287 * one used for RT-throttling (rt_bandwidth), with the main difference
288 * that, since here we are only interested in admission control, we
289 * do not decrease any runtime while the group "executes", neither we
290 * need a timer to replenish it.
292 * With respect to SMP, bandwidth is given on a per root domain basis,
294 * - bw (< 100%) is the deadline bandwidth of each CPU;
295 * - total_bw is the currently allocated bandwidth in each root domain;
303 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
306 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
308 dl_b->total_bw -= tsk_bw;
309 __dl_update(dl_b, (s32)tsk_bw / cpus);
313 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
315 dl_b->total_bw += tsk_bw;
316 __dl_update(dl_b, -((s32)tsk_bw / cpus));
319 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
320 u64 old_bw, u64 new_bw)
322 return dl_b->bw != -1 &&
323 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
327 * Verify the fitness of task @p to run on @cpu taking into account the
328 * CPU original capacity and the runtime/deadline ratio of the task.
330 * The function will return true if the CPU original capacity of the
331 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
332 * task and false otherwise.
334 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
336 unsigned long cap = arch_scale_cpu_capacity(cpu);
338 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
341 extern void init_dl_bw(struct dl_bw *dl_b);
342 extern int sched_dl_global_validate(void);
343 extern void sched_dl_do_global(void);
344 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
345 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
346 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
347 extern bool __checkparam_dl(const struct sched_attr *attr);
348 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
349 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
350 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
351 extern bool dl_cpu_busy(unsigned int cpu);
353 #ifdef CONFIG_CGROUP_SCHED
355 #include <linux/cgroup.h>
356 #include <linux/psi.h>
361 extern struct list_head task_groups;
363 struct cfs_bandwidth {
364 #ifdef CONFIG_CFS_BANDWIDTH
369 s64 hierarchical_quota;
374 struct hrtimer period_timer;
375 struct hrtimer slack_timer;
376 struct list_head throttled_cfs_rq;
385 /* Task group related information */
387 struct cgroup_subsys_state css;
389 #ifdef CONFIG_FAIR_GROUP_SCHED
390 /* schedulable entities of this group on each CPU */
391 struct sched_entity **se;
392 /* runqueue "owned" by this group on each CPU */
393 struct cfs_rq **cfs_rq;
394 unsigned long shares;
398 * load_avg can be heavily contended at clock tick time, so put
399 * it in its own cacheline separated from the fields above which
400 * will also be accessed at each tick.
402 atomic_long_t load_avg ____cacheline_aligned;
406 #ifdef CONFIG_RT_GROUP_SCHED
407 struct sched_rt_entity **rt_se;
408 struct rt_rq **rt_rq;
410 struct rt_bandwidth rt_bandwidth;
414 struct list_head list;
416 struct task_group *parent;
417 struct list_head siblings;
418 struct list_head children;
420 #ifdef CONFIG_SCHED_AUTOGROUP
421 struct autogroup *autogroup;
424 struct cfs_bandwidth cfs_bandwidth;
426 #ifdef CONFIG_UCLAMP_TASK_GROUP
427 /* The two decimal precision [%] value requested from user-space */
428 unsigned int uclamp_pct[UCLAMP_CNT];
429 /* Clamp values requested for a task group */
430 struct uclamp_se uclamp_req[UCLAMP_CNT];
431 /* Effective clamp values used for a task group */
432 struct uclamp_se uclamp[UCLAMP_CNT];
437 #ifdef CONFIG_FAIR_GROUP_SCHED
438 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
441 * A weight of 0 or 1 can cause arithmetics problems.
442 * A weight of a cfs_rq is the sum of weights of which entities
443 * are queued on this cfs_rq, so a weight of a entity should not be
444 * too large, so as the shares value of a task group.
445 * (The default weight is 1024 - so there's no practical
446 * limitation from this.)
448 #define MIN_SHARES (1UL << 1)
449 #define MAX_SHARES (1UL << 18)
452 typedef int (*tg_visitor)(struct task_group *, void *);
454 extern int walk_tg_tree_from(struct task_group *from,
455 tg_visitor down, tg_visitor up, void *data);
458 * Iterate the full tree, calling @down when first entering a node and @up when
459 * leaving it for the final time.
461 * Caller must hold rcu_lock or sufficient equivalent.
463 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
465 return walk_tg_tree_from(&root_task_group, down, up, data);
468 extern int tg_nop(struct task_group *tg, void *data);
470 extern void free_fair_sched_group(struct task_group *tg);
471 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
472 extern void online_fair_sched_group(struct task_group *tg);
473 extern void unregister_fair_sched_group(struct task_group *tg);
474 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
475 struct sched_entity *se, int cpu,
476 struct sched_entity *parent);
477 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
479 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
480 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
481 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
483 extern void free_rt_sched_group(struct task_group *tg);
484 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
485 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
486 struct sched_rt_entity *rt_se, int cpu,
487 struct sched_rt_entity *parent);
488 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
489 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
490 extern long sched_group_rt_runtime(struct task_group *tg);
491 extern long sched_group_rt_period(struct task_group *tg);
492 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
494 extern struct task_group *sched_create_group(struct task_group *parent);
495 extern void sched_online_group(struct task_group *tg,
496 struct task_group *parent);
497 extern void sched_destroy_group(struct task_group *tg);
498 extern void sched_offline_group(struct task_group *tg);
500 extern void sched_move_task(struct task_struct *tsk);
502 #ifdef CONFIG_FAIR_GROUP_SCHED
503 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
506 extern void set_task_rq_fair(struct sched_entity *se,
507 struct cfs_rq *prev, struct cfs_rq *next);
508 #else /* !CONFIG_SMP */
509 static inline void set_task_rq_fair(struct sched_entity *se,
510 struct cfs_rq *prev, struct cfs_rq *next) { }
511 #endif /* CONFIG_SMP */
512 #endif /* CONFIG_FAIR_GROUP_SCHED */
514 #else /* CONFIG_CGROUP_SCHED */
516 struct cfs_bandwidth { };
518 #endif /* CONFIG_CGROUP_SCHED */
520 /* CFS-related fields in a runqueue */
522 struct load_weight load;
523 unsigned int nr_running;
524 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
525 unsigned int idle_h_nr_running; /* SCHED_IDLE */
529 #ifdef CONFIG_SCHED_CORE
530 unsigned int forceidle_seq;
535 u64 min_vruntime_copy;
538 struct rb_root_cached tasks_timeline;
541 * 'curr' points to currently running entity on this cfs_rq.
542 * It is set to NULL otherwise (i.e when none are currently running).
544 struct sched_entity *curr;
545 struct sched_entity *next;
546 struct sched_entity *last;
547 struct sched_entity *skip;
549 #ifdef CONFIG_SCHED_DEBUG
550 unsigned int nr_spread_over;
557 struct sched_avg avg;
559 u64 load_last_update_time_copy;
562 raw_spinlock_t lock ____cacheline_aligned;
564 unsigned long load_avg;
565 unsigned long util_avg;
566 unsigned long runnable_avg;
569 #ifdef CONFIG_FAIR_GROUP_SCHED
570 unsigned long tg_load_avg_contrib;
572 long prop_runnable_sum;
575 * h_load = weight * f(tg)
577 * Where f(tg) is the recursive weight fraction assigned to
580 unsigned long h_load;
581 u64 last_h_load_update;
582 struct sched_entity *h_load_next;
583 #endif /* CONFIG_FAIR_GROUP_SCHED */
584 #endif /* CONFIG_SMP */
586 #ifdef CONFIG_FAIR_GROUP_SCHED
587 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
590 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
591 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
592 * (like users, containers etc.)
594 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
595 * This list is used during load balance.
598 struct list_head leaf_cfs_rq_list;
599 struct task_group *tg; /* group that "owns" this runqueue */
601 #ifdef CONFIG_CFS_BANDWIDTH
603 s64 runtime_remaining;
606 u64 throttled_clock_task;
607 u64 throttled_clock_task_time;
610 struct list_head throttled_list;
611 #endif /* CONFIG_CFS_BANDWIDTH */
612 #endif /* CONFIG_FAIR_GROUP_SCHED */
615 static inline int rt_bandwidth_enabled(void)
617 return sysctl_sched_rt_runtime >= 0;
620 /* RT IPI pull logic requires IRQ_WORK */
621 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
622 # define HAVE_RT_PUSH_IPI
625 /* Real-Time classes' related field in a runqueue: */
627 struct rt_prio_array active;
628 unsigned int rt_nr_running;
629 unsigned int rr_nr_running;
630 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
632 int curr; /* highest queued rt task prio */
634 int next; /* next highest */
639 unsigned int rt_nr_migratory;
640 unsigned int rt_nr_total;
642 struct plist_head pushable_tasks;
644 #endif /* CONFIG_SMP */
650 /* Nests inside the rq lock: */
651 raw_spinlock_t rt_runtime_lock;
653 #ifdef CONFIG_RT_GROUP_SCHED
654 unsigned int rt_nr_boosted;
657 struct task_group *tg;
661 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
663 return rt_rq->rt_queued && rt_rq->rt_nr_running;
666 /* Deadline class' related fields in a runqueue */
668 /* runqueue is an rbtree, ordered by deadline */
669 struct rb_root_cached root;
671 unsigned int dl_nr_running;
675 * Deadline values of the currently executing and the
676 * earliest ready task on this rq. Caching these facilitates
677 * the decision whether or not a ready but not running task
678 * should migrate somewhere else.
685 unsigned int dl_nr_migratory;
689 * Tasks on this rq that can be pushed away. They are kept in
690 * an rb-tree, ordered by tasks' deadlines, with caching
691 * of the leftmost (earliest deadline) element.
693 struct rb_root_cached pushable_dl_tasks_root;
698 * "Active utilization" for this runqueue: increased when a
699 * task wakes up (becomes TASK_RUNNING) and decreased when a
705 * Utilization of the tasks "assigned" to this runqueue (including
706 * the tasks that are in runqueue and the tasks that executed on this
707 * CPU and blocked). Increased when a task moves to this runqueue, and
708 * decreased when the task moves away (migrates, changes scheduling
709 * policy, or terminates).
710 * This is needed to compute the "inactive utilization" for the
711 * runqueue (inactive utilization = this_bw - running_bw).
717 * Inverse of the fraction of CPU utilization that can be reclaimed
718 * by the GRUB algorithm.
723 #ifdef CONFIG_FAIR_GROUP_SCHED
724 /* An entity is a task if it doesn't "own" a runqueue */
725 #define entity_is_task(se) (!se->my_q)
727 static inline void se_update_runnable(struct sched_entity *se)
729 if (!entity_is_task(se))
730 se->runnable_weight = se->my_q->h_nr_running;
733 static inline long se_runnable(struct sched_entity *se)
735 if (entity_is_task(se))
738 return se->runnable_weight;
742 #define entity_is_task(se) 1
744 static inline void se_update_runnable(struct sched_entity *se) {}
746 static inline long se_runnable(struct sched_entity *se)
754 * XXX we want to get rid of these helpers and use the full load resolution.
756 static inline long se_weight(struct sched_entity *se)
758 return scale_load_down(se->load.weight);
762 static inline bool sched_asym_prefer(int a, int b)
764 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
768 struct em_perf_domain *em_pd;
769 struct perf_domain *next;
773 /* Scheduling group status flags */
774 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
775 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
778 * We add the notion of a root-domain which will be used to define per-domain
779 * variables. Each exclusive cpuset essentially defines an island domain by
780 * fully partitioning the member CPUs from any other cpuset. Whenever a new
781 * exclusive cpuset is created, we also create and attach a new root-domain
790 cpumask_var_t online;
793 * Indicate pullable load on at least one CPU, e.g:
794 * - More than one runnable task
795 * - Running task is misfit
799 /* Indicate one or more cpus over-utilized (tipping point) */
803 * The bit corresponding to a CPU gets set here if such CPU has more
804 * than one runnable -deadline task (as it is below for RT tasks).
806 cpumask_var_t dlo_mask;
812 * Indicate whether a root_domain's dl_bw has been checked or
813 * updated. It's monotonously increasing value.
815 * Also, some corner cases, like 'wrap around' is dangerous, but given
816 * that u64 is 'big enough'. So that shouldn't be a concern.
820 #ifdef HAVE_RT_PUSH_IPI
822 * For IPI pull requests, loop across the rto_mask.
824 struct irq_work rto_push_work;
825 raw_spinlock_t rto_lock;
826 /* These are only updated and read within rto_lock */
829 /* These atomics are updated outside of a lock */
830 atomic_t rto_loop_next;
831 atomic_t rto_loop_start;
834 * The "RT overload" flag: it gets set if a CPU has more than
835 * one runnable RT task.
837 cpumask_var_t rto_mask;
838 struct cpupri cpupri;
840 unsigned long max_cpu_capacity;
843 * NULL-terminated list of performance domains intersecting with the
844 * CPUs of the rd. Protected by RCU.
846 struct perf_domain __rcu *pd;
849 extern void init_defrootdomain(void);
850 extern int sched_init_domains(const struct cpumask *cpu_map);
851 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
852 extern void sched_get_rd(struct root_domain *rd);
853 extern void sched_put_rd(struct root_domain *rd);
855 #ifdef HAVE_RT_PUSH_IPI
856 extern void rto_push_irq_work_func(struct irq_work *work);
858 #endif /* CONFIG_SMP */
860 #ifdef CONFIG_UCLAMP_TASK
862 * struct uclamp_bucket - Utilization clamp bucket
863 * @value: utilization clamp value for tasks on this clamp bucket
864 * @tasks: number of RUNNABLE tasks on this clamp bucket
866 * Keep track of how many tasks are RUNNABLE for a given utilization
869 struct uclamp_bucket {
870 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
871 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
875 * struct uclamp_rq - rq's utilization clamp
876 * @value: currently active clamp values for a rq
877 * @bucket: utilization clamp buckets affecting a rq
879 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
880 * A clamp value is affecting a rq when there is at least one task RUNNABLE
881 * (or actually running) with that value.
883 * There are up to UCLAMP_CNT possible different clamp values, currently there
884 * are only two: minimum utilization and maximum utilization.
886 * All utilization clamping values are MAX aggregated, since:
887 * - for util_min: we want to run the CPU at least at the max of the minimum
888 * utilization required by its currently RUNNABLE tasks.
889 * - for util_max: we want to allow the CPU to run up to the max of the
890 * maximum utilization allowed by its currently RUNNABLE tasks.
892 * Since on each system we expect only a limited number of different
893 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
894 * the metrics required to compute all the per-rq utilization clamp values.
898 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
901 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
902 #endif /* CONFIG_UCLAMP_TASK */
905 * This is the main, per-CPU runqueue data structure.
907 * Locking rule: those places that want to lock multiple runqueues
908 * (such as the load balancing or the thread migration code), lock
909 * acquire operations must be ordered by ascending &runqueue.
913 raw_spinlock_t __lock;
916 * nr_running and cpu_load should be in the same cacheline because
917 * remote CPUs use both these fields when doing load calculation.
919 unsigned int nr_running;
920 #ifdef CONFIG_NUMA_BALANCING
921 unsigned int nr_numa_running;
922 unsigned int nr_preferred_running;
923 unsigned int numa_migrate_on;
925 #ifdef CONFIG_NO_HZ_COMMON
927 unsigned long last_blocked_load_update_tick;
928 unsigned int has_blocked_load;
929 call_single_data_t nohz_csd;
930 #endif /* CONFIG_SMP */
931 unsigned int nohz_tick_stopped;
933 #endif /* CONFIG_NO_HZ_COMMON */
936 unsigned int ttwu_pending;
940 #ifdef CONFIG_UCLAMP_TASK
941 /* Utilization clamp values based on CPU's RUNNABLE tasks */
942 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
943 unsigned int uclamp_flags;
944 #define UCLAMP_FLAG_IDLE 0x01
951 #ifdef CONFIG_FAIR_GROUP_SCHED
952 /* list of leaf cfs_rq on this CPU: */
953 struct list_head leaf_cfs_rq_list;
954 struct list_head *tmp_alone_branch;
955 #endif /* CONFIG_FAIR_GROUP_SCHED */
958 * This is part of a global counter where only the total sum
959 * over all CPUs matters. A task can increase this counter on
960 * one CPU and if it got migrated afterwards it may decrease
961 * it on another CPU. Always updated under the runqueue lock:
963 unsigned int nr_uninterruptible;
965 struct task_struct __rcu *curr;
966 struct task_struct *idle;
967 struct task_struct *stop;
968 unsigned long next_balance;
969 struct mm_struct *prev_mm;
971 unsigned int clock_update_flags;
973 /* Ensure that all clocks are in the same cache line */
974 u64 clock_task ____cacheline_aligned;
976 unsigned long lost_idle_time;
980 #ifdef CONFIG_SCHED_DEBUG
981 u64 last_seen_need_resched_ns;
982 int ticks_without_resched;
985 #ifdef CONFIG_MEMBARRIER
986 int membarrier_state;
990 struct root_domain *rd;
991 struct sched_domain __rcu *sd;
993 unsigned long cpu_capacity;
994 unsigned long cpu_capacity_orig;
996 struct callback_head *balance_callback;
998 unsigned char nohz_idle_balance;
999 unsigned char idle_balance;
1001 unsigned long misfit_task_load;
1003 /* For active balancing */
1006 struct cpu_stop_work active_balance_work;
1008 /* CPU of this runqueue: */
1012 struct list_head cfs_tasks;
1014 struct sched_avg avg_rt;
1015 struct sched_avg avg_dl;
1016 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1017 struct sched_avg avg_irq;
1019 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1020 struct sched_avg avg_thermal;
1025 unsigned long wake_stamp;
1028 /* This is used to determine avg_idle's max value */
1029 u64 max_idle_balance_cost;
1031 #ifdef CONFIG_HOTPLUG_CPU
1032 struct rcuwait hotplug_wait;
1034 #endif /* CONFIG_SMP */
1036 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1039 #ifdef CONFIG_PARAVIRT
1040 u64 prev_steal_time;
1042 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1043 u64 prev_steal_time_rq;
1046 /* calc_load related fields */
1047 unsigned long calc_load_update;
1048 long calc_load_active;
1050 #ifdef CONFIG_SCHED_HRTICK
1052 call_single_data_t hrtick_csd;
1054 struct hrtimer hrtick_timer;
1055 ktime_t hrtick_time;
1058 #ifdef CONFIG_SCHEDSTATS
1060 struct sched_info rq_sched_info;
1061 unsigned long long rq_cpu_time;
1062 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1064 /* sys_sched_yield() stats */
1065 unsigned int yld_count;
1067 /* schedule() stats */
1068 unsigned int sched_count;
1069 unsigned int sched_goidle;
1071 /* try_to_wake_up() stats */
1072 unsigned int ttwu_count;
1073 unsigned int ttwu_local;
1076 #ifdef CONFIG_CPU_IDLE
1077 /* Must be inspected within a rcu lock section */
1078 struct cpuidle_state *idle_state;
1082 unsigned int nr_pinned;
1084 unsigned int push_busy;
1085 struct cpu_stop_work push_work;
1087 #ifdef CONFIG_SCHED_CORE
1090 struct task_struct *core_pick;
1091 unsigned int core_enabled;
1092 unsigned int core_sched_seq;
1093 struct rb_root core_tree;
1096 unsigned int core_task_seq;
1097 unsigned int core_pick_seq;
1098 unsigned long core_cookie;
1099 unsigned char core_forceidle;
1100 unsigned int core_forceidle_seq;
1104 #ifdef CONFIG_FAIR_GROUP_SCHED
1106 /* CPU runqueue to which this cfs_rq is attached */
1107 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1114 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1116 return container_of(cfs_rq, struct rq, cfs);
1120 static inline int cpu_of(struct rq *rq)
1129 #define MDF_PUSH 0x01
1131 static inline bool is_migration_disabled(struct task_struct *p)
1134 return p->migration_disabled;
1141 #ifdef CONFIG_SCHED_CORE
1142 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1144 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1146 static inline bool sched_core_enabled(struct rq *rq)
1148 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1151 static inline bool sched_core_disabled(void)
1153 return !static_branch_unlikely(&__sched_core_enabled);
1157 * Be careful with this function; not for general use. The return value isn't
1158 * stable unless you actually hold a relevant rq->__lock.
1160 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1162 if (sched_core_enabled(rq))
1163 return &rq->core->__lock;
1168 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1170 if (rq->core_enabled)
1171 return &rq->core->__lock;
1176 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1179 * Helpers to check if the CPU's core cookie matches with the task's cookie
1180 * when core scheduling is enabled.
1181 * A special case is that the task's cookie always matches with CPU's core
1182 * cookie if the CPU is in an idle core.
1184 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1186 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1187 if (!sched_core_enabled(rq))
1190 return rq->core->core_cookie == p->core_cookie;
1193 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1195 bool idle_core = true;
1198 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1199 if (!sched_core_enabled(rq))
1202 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1203 if (!available_idle_cpu(cpu)) {
1210 * A CPU in an idle core is always the best choice for tasks with
1213 return idle_core || rq->core->core_cookie == p->core_cookie;
1216 static inline bool sched_group_cookie_match(struct rq *rq,
1217 struct task_struct *p,
1218 struct sched_group *group)
1222 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1223 if (!sched_core_enabled(rq))
1226 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1227 if (sched_core_cookie_match(rq, p))
1233 extern void queue_core_balance(struct rq *rq);
1235 static inline bool sched_core_enqueued(struct task_struct *p)
1237 return !RB_EMPTY_NODE(&p->core_node);
1240 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1241 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1243 extern void sched_core_get(void);
1244 extern void sched_core_put(void);
1246 extern unsigned long sched_core_alloc_cookie(void);
1247 extern void sched_core_put_cookie(unsigned long cookie);
1248 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1249 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1251 #else /* !CONFIG_SCHED_CORE */
1253 static inline bool sched_core_enabled(struct rq *rq)
1258 static inline bool sched_core_disabled(void)
1263 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1268 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1273 static inline void queue_core_balance(struct rq *rq)
1277 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1282 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1287 static inline bool sched_group_cookie_match(struct rq *rq,
1288 struct task_struct *p,
1289 struct sched_group *group)
1293 #endif /* CONFIG_SCHED_CORE */
1295 static inline void lockdep_assert_rq_held(struct rq *rq)
1297 lockdep_assert_held(__rq_lockp(rq));
1300 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1301 extern bool raw_spin_rq_trylock(struct rq *rq);
1302 extern void raw_spin_rq_unlock(struct rq *rq);
1304 static inline void raw_spin_rq_lock(struct rq *rq)
1306 raw_spin_rq_lock_nested(rq, 0);
1309 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1311 local_irq_disable();
1312 raw_spin_rq_lock(rq);
1315 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1317 raw_spin_rq_unlock(rq);
1321 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1323 unsigned long flags;
1324 local_irq_save(flags);
1325 raw_spin_rq_lock(rq);
1329 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1331 raw_spin_rq_unlock(rq);
1332 local_irq_restore(flags);
1335 #define raw_spin_rq_lock_irqsave(rq, flags) \
1337 flags = _raw_spin_rq_lock_irqsave(rq); \
1340 #ifdef CONFIG_SCHED_SMT
1341 extern void __update_idle_core(struct rq *rq);
1343 static inline void update_idle_core(struct rq *rq)
1345 if (static_branch_unlikely(&sched_smt_present))
1346 __update_idle_core(rq);
1350 static inline void update_idle_core(struct rq *rq) { }
1353 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1355 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1356 #define this_rq() this_cpu_ptr(&runqueues)
1357 #define task_rq(p) cpu_rq(task_cpu(p))
1358 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1359 #define raw_rq() raw_cpu_ptr(&runqueues)
1361 #ifdef CONFIG_FAIR_GROUP_SCHED
1362 static inline struct task_struct *task_of(struct sched_entity *se)
1364 SCHED_WARN_ON(!entity_is_task(se));
1365 return container_of(se, struct task_struct, se);
1368 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1370 return p->se.cfs_rq;
1373 /* runqueue on which this entity is (to be) queued */
1374 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1379 /* runqueue "owned" by this group */
1380 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1387 static inline struct task_struct *task_of(struct sched_entity *se)
1389 return container_of(se, struct task_struct, se);
1392 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1394 return &task_rq(p)->cfs;
1397 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1399 struct task_struct *p = task_of(se);
1400 struct rq *rq = task_rq(p);
1405 /* runqueue "owned" by this group */
1406 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1412 extern void update_rq_clock(struct rq *rq);
1414 static inline u64 __rq_clock_broken(struct rq *rq)
1416 return READ_ONCE(rq->clock);
1420 * rq::clock_update_flags bits
1422 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1423 * call to __schedule(). This is an optimisation to avoid
1424 * neighbouring rq clock updates.
1426 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1427 * in effect and calls to update_rq_clock() are being ignored.
1429 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1430 * made to update_rq_clock() since the last time rq::lock was pinned.
1432 * If inside of __schedule(), clock_update_flags will have been
1433 * shifted left (a left shift is a cheap operation for the fast path
1434 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1436 * if (rq-clock_update_flags >= RQCF_UPDATED)
1438 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1439 * one position though, because the next rq_unpin_lock() will shift it
1442 #define RQCF_REQ_SKIP 0x01
1443 #define RQCF_ACT_SKIP 0x02
1444 #define RQCF_UPDATED 0x04
1446 static inline void assert_clock_updated(struct rq *rq)
1449 * The only reason for not seeing a clock update since the
1450 * last rq_pin_lock() is if we're currently skipping updates.
1452 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1455 static inline u64 rq_clock(struct rq *rq)
1457 lockdep_assert_rq_held(rq);
1458 assert_clock_updated(rq);
1463 static inline u64 rq_clock_task(struct rq *rq)
1465 lockdep_assert_rq_held(rq);
1466 assert_clock_updated(rq);
1468 return rq->clock_task;
1472 * By default the decay is the default pelt decay period.
1473 * The decay shift can change the decay period in
1475 * Decay shift Decay period(ms)
1482 extern int sched_thermal_decay_shift;
1484 static inline u64 rq_clock_thermal(struct rq *rq)
1486 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1489 static inline void rq_clock_skip_update(struct rq *rq)
1491 lockdep_assert_rq_held(rq);
1492 rq->clock_update_flags |= RQCF_REQ_SKIP;
1496 * See rt task throttling, which is the only time a skip
1497 * request is canceled.
1499 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1501 lockdep_assert_rq_held(rq);
1502 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1506 unsigned long flags;
1507 struct pin_cookie cookie;
1508 #ifdef CONFIG_SCHED_DEBUG
1510 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1511 * current pin context is stashed here in case it needs to be
1512 * restored in rq_repin_lock().
1514 unsigned int clock_update_flags;
1518 extern struct callback_head balance_push_callback;
1521 * Lockdep annotation that avoids accidental unlocks; it's like a
1522 * sticky/continuous lockdep_assert_held().
1524 * This avoids code that has access to 'struct rq *rq' (basically everything in
1525 * the scheduler) from accidentally unlocking the rq if they do not also have a
1526 * copy of the (on-stack) 'struct rq_flags rf'.
1528 * Also see Documentation/locking/lockdep-design.rst.
1530 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1532 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1534 #ifdef CONFIG_SCHED_DEBUG
1535 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1536 rf->clock_update_flags = 0;
1538 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1543 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1545 #ifdef CONFIG_SCHED_DEBUG
1546 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1547 rf->clock_update_flags = RQCF_UPDATED;
1550 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1553 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1555 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1557 #ifdef CONFIG_SCHED_DEBUG
1559 * Restore the value we stashed in @rf for this pin context.
1561 rq->clock_update_flags |= rf->clock_update_flags;
1565 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1566 __acquires(rq->lock);
1568 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1569 __acquires(p->pi_lock)
1570 __acquires(rq->lock);
1572 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1573 __releases(rq->lock)
1575 rq_unpin_lock(rq, rf);
1576 raw_spin_rq_unlock(rq);
1580 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1581 __releases(rq->lock)
1582 __releases(p->pi_lock)
1584 rq_unpin_lock(rq, rf);
1585 raw_spin_rq_unlock(rq);
1586 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1590 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1591 __acquires(rq->lock)
1593 raw_spin_rq_lock_irqsave(rq, rf->flags);
1594 rq_pin_lock(rq, rf);
1598 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1599 __acquires(rq->lock)
1601 raw_spin_rq_lock_irq(rq);
1602 rq_pin_lock(rq, rf);
1606 rq_lock(struct rq *rq, struct rq_flags *rf)
1607 __acquires(rq->lock)
1609 raw_spin_rq_lock(rq);
1610 rq_pin_lock(rq, rf);
1614 rq_relock(struct rq *rq, struct rq_flags *rf)
1615 __acquires(rq->lock)
1617 raw_spin_rq_lock(rq);
1618 rq_repin_lock(rq, rf);
1622 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1623 __releases(rq->lock)
1625 rq_unpin_lock(rq, rf);
1626 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1630 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1631 __releases(rq->lock)
1633 rq_unpin_lock(rq, rf);
1634 raw_spin_rq_unlock_irq(rq);
1638 rq_unlock(struct rq *rq, struct rq_flags *rf)
1639 __releases(rq->lock)
1641 rq_unpin_lock(rq, rf);
1642 raw_spin_rq_unlock(rq);
1645 static inline struct rq *
1646 this_rq_lock_irq(struct rq_flags *rf)
1647 __acquires(rq->lock)
1651 local_irq_disable();
1658 enum numa_topology_type {
1663 extern enum numa_topology_type sched_numa_topology_type;
1664 extern int sched_max_numa_distance;
1665 extern bool find_numa_distance(int distance);
1666 extern void sched_init_numa(void);
1667 extern void sched_domains_numa_masks_set(unsigned int cpu);
1668 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1669 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1671 static inline void sched_init_numa(void) { }
1672 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1673 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1674 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1680 #ifdef CONFIG_NUMA_BALANCING
1681 /* The regions in numa_faults array from task_struct */
1682 enum numa_faults_stats {
1688 extern void sched_setnuma(struct task_struct *p, int node);
1689 extern int migrate_task_to(struct task_struct *p, int cpu);
1690 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1692 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1695 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1698 #endif /* CONFIG_NUMA_BALANCING */
1703 queue_balance_callback(struct rq *rq,
1704 struct callback_head *head,
1705 void (*func)(struct rq *rq))
1707 lockdep_assert_rq_held(rq);
1709 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1712 head->func = (void (*)(struct callback_head *))func;
1713 head->next = rq->balance_callback;
1714 rq->balance_callback = head;
1717 #define rcu_dereference_check_sched_domain(p) \
1718 rcu_dereference_check((p), \
1719 lockdep_is_held(&sched_domains_mutex))
1722 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1723 * See destroy_sched_domains: call_rcu for details.
1725 * The domain tree of any CPU may only be accessed from within
1726 * preempt-disabled sections.
1728 #define for_each_domain(cpu, __sd) \
1729 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1730 __sd; __sd = __sd->parent)
1733 * highest_flag_domain - Return highest sched_domain containing flag.
1734 * @cpu: The CPU whose highest level of sched domain is to
1736 * @flag: The flag to check for the highest sched_domain
1737 * for the given CPU.
1739 * Returns the highest sched_domain of a CPU which contains the given flag.
1741 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1743 struct sched_domain *sd, *hsd = NULL;
1745 for_each_domain(cpu, sd) {
1746 if (!(sd->flags & flag))
1754 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1756 struct sched_domain *sd;
1758 for_each_domain(cpu, sd) {
1759 if (sd->flags & flag)
1766 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1767 DECLARE_PER_CPU(int, sd_llc_size);
1768 DECLARE_PER_CPU(int, sd_llc_id);
1769 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1770 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1771 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1772 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1773 extern struct static_key_false sched_asym_cpucapacity;
1775 struct sched_group_capacity {
1778 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1781 unsigned long capacity;
1782 unsigned long min_capacity; /* Min per-CPU capacity in group */
1783 unsigned long max_capacity; /* Max per-CPU capacity in group */
1784 unsigned long next_update;
1785 int imbalance; /* XXX unrelated to capacity but shared group state */
1787 #ifdef CONFIG_SCHED_DEBUG
1791 unsigned long cpumask[]; /* Balance mask */
1794 struct sched_group {
1795 struct sched_group *next; /* Must be a circular list */
1798 unsigned int group_weight;
1799 struct sched_group_capacity *sgc;
1800 int asym_prefer_cpu; /* CPU of highest priority in group */
1803 * The CPUs this group covers.
1805 * NOTE: this field is variable length. (Allocated dynamically
1806 * by attaching extra space to the end of the structure,
1807 * depending on how many CPUs the kernel has booted up with)
1809 unsigned long cpumask[];
1812 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1814 return to_cpumask(sg->cpumask);
1818 * See build_balance_mask().
1820 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1822 return to_cpumask(sg->sgc->cpumask);
1826 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1827 * @group: The group whose first CPU is to be returned.
1829 static inline unsigned int group_first_cpu(struct sched_group *group)
1831 return cpumask_first(sched_group_span(group));
1834 extern int group_balance_cpu(struct sched_group *sg);
1836 #ifdef CONFIG_SCHED_DEBUG
1837 void update_sched_domain_debugfs(void);
1838 void dirty_sched_domain_sysctl(int cpu);
1840 static inline void update_sched_domain_debugfs(void)
1843 static inline void dirty_sched_domain_sysctl(int cpu)
1848 extern int sched_update_scaling(void);
1850 extern void flush_smp_call_function_from_idle(void);
1852 #else /* !CONFIG_SMP: */
1853 static inline void flush_smp_call_function_from_idle(void) { }
1857 #include "autogroup.h"
1859 #ifdef CONFIG_CGROUP_SCHED
1862 * Return the group to which this tasks belongs.
1864 * We cannot use task_css() and friends because the cgroup subsystem
1865 * changes that value before the cgroup_subsys::attach() method is called,
1866 * therefore we cannot pin it and might observe the wrong value.
1868 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1869 * core changes this before calling sched_move_task().
1871 * Instead we use a 'copy' which is updated from sched_move_task() while
1872 * holding both task_struct::pi_lock and rq::lock.
1874 static inline struct task_group *task_group(struct task_struct *p)
1876 return p->sched_task_group;
1879 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1880 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1882 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1883 struct task_group *tg = task_group(p);
1886 #ifdef CONFIG_FAIR_GROUP_SCHED
1887 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1888 p->se.cfs_rq = tg->cfs_rq[cpu];
1889 p->se.parent = tg->se[cpu];
1892 #ifdef CONFIG_RT_GROUP_SCHED
1893 p->rt.rt_rq = tg->rt_rq[cpu];
1894 p->rt.parent = tg->rt_se[cpu];
1898 #else /* CONFIG_CGROUP_SCHED */
1900 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1901 static inline struct task_group *task_group(struct task_struct *p)
1906 #endif /* CONFIG_CGROUP_SCHED */
1908 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1910 set_task_rq(p, cpu);
1913 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1914 * successfully executed on another CPU. We must ensure that updates of
1915 * per-task data have been completed by this moment.
1918 #ifdef CONFIG_THREAD_INFO_IN_TASK
1919 WRITE_ONCE(p->cpu, cpu);
1921 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1928 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1930 #ifdef CONFIG_SCHED_DEBUG
1931 # include <linux/static_key.h>
1932 # define const_debug __read_mostly
1934 # define const_debug const
1937 #define SCHED_FEAT(name, enabled) \
1938 __SCHED_FEAT_##name ,
1941 #include "features.h"
1947 #ifdef CONFIG_SCHED_DEBUG
1950 * To support run-time toggling of sched features, all the translation units
1951 * (but core.c) reference the sysctl_sched_features defined in core.c.
1953 extern const_debug unsigned int sysctl_sched_features;
1955 #ifdef CONFIG_JUMP_LABEL
1956 #define SCHED_FEAT(name, enabled) \
1957 static __always_inline bool static_branch_##name(struct static_key *key) \
1959 return static_key_##enabled(key); \
1962 #include "features.h"
1965 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1966 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1968 #else /* !CONFIG_JUMP_LABEL */
1970 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1972 #endif /* CONFIG_JUMP_LABEL */
1974 #else /* !SCHED_DEBUG */
1977 * Each translation unit has its own copy of sysctl_sched_features to allow
1978 * constants propagation at compile time and compiler optimization based on
1981 #define SCHED_FEAT(name, enabled) \
1982 (1UL << __SCHED_FEAT_##name) * enabled |
1983 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1984 #include "features.h"
1988 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1990 #endif /* SCHED_DEBUG */
1992 extern struct static_key_false sched_numa_balancing;
1993 extern struct static_key_false sched_schedstats;
1995 static inline u64 global_rt_period(void)
1997 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2000 static inline u64 global_rt_runtime(void)
2002 if (sysctl_sched_rt_runtime < 0)
2005 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2008 static inline int task_current(struct rq *rq, struct task_struct *p)
2010 return rq->curr == p;
2013 static inline int task_running(struct rq *rq, struct task_struct *p)
2018 return task_current(rq, p);
2022 static inline int task_on_rq_queued(struct task_struct *p)
2024 return p->on_rq == TASK_ON_RQ_QUEUED;
2027 static inline int task_on_rq_migrating(struct task_struct *p)
2029 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2032 /* Wake flags. The first three directly map to some SD flag value */
2033 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2034 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2035 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2037 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2038 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2039 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2042 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2043 static_assert(WF_FORK == SD_BALANCE_FORK);
2044 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2048 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2049 * of tasks with abnormal "nice" values across CPUs the contribution that
2050 * each task makes to its run queue's load is weighted according to its
2051 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2052 * scaled version of the new time slice allocation that they receive on time
2056 #define WEIGHT_IDLEPRIO 3
2057 #define WMULT_IDLEPRIO 1431655765
2059 extern const int sched_prio_to_weight[40];
2060 extern const u32 sched_prio_to_wmult[40];
2063 * {de,en}queue flags:
2065 * DEQUEUE_SLEEP - task is no longer runnable
2066 * ENQUEUE_WAKEUP - task just became runnable
2068 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2069 * are in a known state which allows modification. Such pairs
2070 * should preserve as much state as possible.
2072 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2075 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2076 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2077 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2081 #define DEQUEUE_SLEEP 0x01
2082 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2083 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2084 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2086 #define ENQUEUE_WAKEUP 0x01
2087 #define ENQUEUE_RESTORE 0x02
2088 #define ENQUEUE_MOVE 0x04
2089 #define ENQUEUE_NOCLOCK 0x08
2091 #define ENQUEUE_HEAD 0x10
2092 #define ENQUEUE_REPLENISH 0x20
2094 #define ENQUEUE_MIGRATED 0x40
2096 #define ENQUEUE_MIGRATED 0x00
2099 #define RETRY_TASK ((void *)-1UL)
2101 struct sched_class {
2103 #ifdef CONFIG_UCLAMP_TASK
2107 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2108 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2109 void (*yield_task) (struct rq *rq);
2110 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2112 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2114 struct task_struct *(*pick_next_task)(struct rq *rq);
2116 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2117 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2120 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2121 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2123 struct task_struct * (*pick_task)(struct rq *rq);
2125 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2127 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2129 void (*set_cpus_allowed)(struct task_struct *p,
2130 const struct cpumask *newmask,
2133 void (*rq_online)(struct rq *rq);
2134 void (*rq_offline)(struct rq *rq);
2136 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2139 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2140 void (*task_fork)(struct task_struct *p);
2141 void (*task_dead)(struct task_struct *p);
2144 * The switched_from() call is allowed to drop rq->lock, therefore we
2145 * cannot assume the switched_from/switched_to pair is serialized by
2146 * rq->lock. They are however serialized by p->pi_lock.
2148 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2149 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2150 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2153 unsigned int (*get_rr_interval)(struct rq *rq,
2154 struct task_struct *task);
2156 void (*update_curr)(struct rq *rq);
2158 #define TASK_SET_GROUP 0
2159 #define TASK_MOVE_GROUP 1
2161 #ifdef CONFIG_FAIR_GROUP_SCHED
2162 void (*task_change_group)(struct task_struct *p, int type);
2166 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2168 WARN_ON_ONCE(rq->curr != prev);
2169 prev->sched_class->put_prev_task(rq, prev);
2172 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2174 next->sched_class->set_next_task(rq, next, false);
2179 * Helper to define a sched_class instance; each one is placed in a separate
2180 * section which is ordered by the linker script:
2182 * include/asm-generic/vmlinux.lds.h
2184 * Also enforce alignment on the instance, not the type, to guarantee layout.
2186 #define DEFINE_SCHED_CLASS(name) \
2187 const struct sched_class name##_sched_class \
2188 __aligned(__alignof__(struct sched_class)) \
2189 __section("__" #name "_sched_class")
2191 /* Defined in include/asm-generic/vmlinux.lds.h */
2192 extern struct sched_class __begin_sched_classes[];
2193 extern struct sched_class __end_sched_classes[];
2195 #define sched_class_highest (__end_sched_classes - 1)
2196 #define sched_class_lowest (__begin_sched_classes - 1)
2198 #define for_class_range(class, _from, _to) \
2199 for (class = (_from); class != (_to); class--)
2201 #define for_each_class(class) \
2202 for_class_range(class, sched_class_highest, sched_class_lowest)
2204 extern const struct sched_class stop_sched_class;
2205 extern const struct sched_class dl_sched_class;
2206 extern const struct sched_class rt_sched_class;
2207 extern const struct sched_class fair_sched_class;
2208 extern const struct sched_class idle_sched_class;
2210 static inline bool sched_stop_runnable(struct rq *rq)
2212 return rq->stop && task_on_rq_queued(rq->stop);
2215 static inline bool sched_dl_runnable(struct rq *rq)
2217 return rq->dl.dl_nr_running > 0;
2220 static inline bool sched_rt_runnable(struct rq *rq)
2222 return rq->rt.rt_queued > 0;
2225 static inline bool sched_fair_runnable(struct rq *rq)
2227 return rq->cfs.nr_running > 0;
2230 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2231 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2233 #define SCA_CHECK 0x01
2234 #define SCA_MIGRATE_DISABLE 0x02
2235 #define SCA_MIGRATE_ENABLE 0x04
2239 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2241 extern void trigger_load_balance(struct rq *rq);
2243 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2245 static inline struct task_struct *get_push_task(struct rq *rq)
2247 struct task_struct *p = rq->curr;
2249 lockdep_assert_rq_held(rq);
2254 if (p->nr_cpus_allowed == 1)
2257 rq->push_busy = true;
2258 return get_task_struct(p);
2261 extern int push_cpu_stop(void *arg);
2265 #ifdef CONFIG_CPU_IDLE
2266 static inline void idle_set_state(struct rq *rq,
2267 struct cpuidle_state *idle_state)
2269 rq->idle_state = idle_state;
2272 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2274 SCHED_WARN_ON(!rcu_read_lock_held());
2276 return rq->idle_state;
2279 static inline void idle_set_state(struct rq *rq,
2280 struct cpuidle_state *idle_state)
2284 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2290 extern void schedule_idle(void);
2292 extern void sysrq_sched_debug_show(void);
2293 extern void sched_init_granularity(void);
2294 extern void update_max_interval(void);
2296 extern void init_sched_dl_class(void);
2297 extern void init_sched_rt_class(void);
2298 extern void init_sched_fair_class(void);
2300 extern void reweight_task(struct task_struct *p, int prio);
2302 extern void resched_curr(struct rq *rq);
2303 extern void resched_cpu(int cpu);
2305 extern struct rt_bandwidth def_rt_bandwidth;
2306 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2308 extern struct dl_bandwidth def_dl_bandwidth;
2309 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2310 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2311 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2314 #define BW_UNIT (1 << BW_SHIFT)
2315 #define RATIO_SHIFT 8
2316 #define MAX_BW_BITS (64 - BW_SHIFT)
2317 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2318 unsigned long to_ratio(u64 period, u64 runtime);
2320 extern void init_entity_runnable_average(struct sched_entity *se);
2321 extern void post_init_entity_util_avg(struct task_struct *p);
2323 #ifdef CONFIG_NO_HZ_FULL
2324 extern bool sched_can_stop_tick(struct rq *rq);
2325 extern int __init sched_tick_offload_init(void);
2328 * Tick may be needed by tasks in the runqueue depending on their policy and
2329 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2330 * nohz mode if necessary.
2332 static inline void sched_update_tick_dependency(struct rq *rq)
2334 int cpu = cpu_of(rq);
2336 if (!tick_nohz_full_cpu(cpu))
2339 if (sched_can_stop_tick(rq))
2340 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2342 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2345 static inline int sched_tick_offload_init(void) { return 0; }
2346 static inline void sched_update_tick_dependency(struct rq *rq) { }
2349 static inline void add_nr_running(struct rq *rq, unsigned count)
2351 unsigned prev_nr = rq->nr_running;
2353 rq->nr_running = prev_nr + count;
2354 if (trace_sched_update_nr_running_tp_enabled()) {
2355 call_trace_sched_update_nr_running(rq, count);
2359 if (prev_nr < 2 && rq->nr_running >= 2) {
2360 if (!READ_ONCE(rq->rd->overload))
2361 WRITE_ONCE(rq->rd->overload, 1);
2365 sched_update_tick_dependency(rq);
2368 static inline void sub_nr_running(struct rq *rq, unsigned count)
2370 rq->nr_running -= count;
2371 if (trace_sched_update_nr_running_tp_enabled()) {
2372 call_trace_sched_update_nr_running(rq, -count);
2375 /* Check if we still need preemption */
2376 sched_update_tick_dependency(rq);
2379 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2380 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2382 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2384 extern const_debug unsigned int sysctl_sched_nr_migrate;
2385 extern const_debug unsigned int sysctl_sched_migration_cost;
2387 #ifdef CONFIG_SCHED_HRTICK
2391 * - enabled by features
2392 * - hrtimer is actually high res
2394 static inline int hrtick_enabled(struct rq *rq)
2396 if (!cpu_active(cpu_of(rq)))
2398 return hrtimer_is_hres_active(&rq->hrtick_timer);
2401 static inline int hrtick_enabled_fair(struct rq *rq)
2403 if (!sched_feat(HRTICK))
2405 return hrtick_enabled(rq);
2408 static inline int hrtick_enabled_dl(struct rq *rq)
2410 if (!sched_feat(HRTICK_DL))
2412 return hrtick_enabled(rq);
2415 void hrtick_start(struct rq *rq, u64 delay);
2419 static inline int hrtick_enabled_fair(struct rq *rq)
2424 static inline int hrtick_enabled_dl(struct rq *rq)
2429 static inline int hrtick_enabled(struct rq *rq)
2434 #endif /* CONFIG_SCHED_HRTICK */
2436 #ifndef arch_scale_freq_tick
2437 static __always_inline
2438 void arch_scale_freq_tick(void)
2443 #ifndef arch_scale_freq_capacity
2445 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2446 * @cpu: the CPU in question.
2448 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2451 * ------ * SCHED_CAPACITY_SCALE
2454 static __always_inline
2455 unsigned long arch_scale_freq_capacity(int cpu)
2457 return SCHED_CAPACITY_SCALE;
2464 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2466 #ifdef CONFIG_SCHED_CORE
2468 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2469 * order by core-id first and cpu-id second.
2473 * double_rq_lock(0,3); will take core-0, core-1 lock
2474 * double_rq_lock(1,2); will take core-1, core-0 lock
2476 * when only cpu-id is considered.
2478 if (rq1->core->cpu < rq2->core->cpu)
2480 if (rq1->core->cpu > rq2->core->cpu)
2484 * __sched_core_flip() relies on SMT having cpu-id lock order.
2487 return rq1->cpu < rq2->cpu;
2490 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2492 #ifdef CONFIG_PREEMPTION
2495 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2496 * way at the expense of forcing extra atomic operations in all
2497 * invocations. This assures that the double_lock is acquired using the
2498 * same underlying policy as the spinlock_t on this architecture, which
2499 * reduces latency compared to the unfair variant below. However, it
2500 * also adds more overhead and therefore may reduce throughput.
2502 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2503 __releases(this_rq->lock)
2504 __acquires(busiest->lock)
2505 __acquires(this_rq->lock)
2507 raw_spin_rq_unlock(this_rq);
2508 double_rq_lock(this_rq, busiest);
2515 * Unfair double_lock_balance: Optimizes throughput at the expense of
2516 * latency by eliminating extra atomic operations when the locks are
2517 * already in proper order on entry. This favors lower CPU-ids and will
2518 * grant the double lock to lower CPUs over higher ids under contention,
2519 * regardless of entry order into the function.
2521 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2522 __releases(this_rq->lock)
2523 __acquires(busiest->lock)
2524 __acquires(this_rq->lock)
2526 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2529 if (likely(raw_spin_rq_trylock(busiest)))
2532 if (rq_order_less(this_rq, busiest)) {
2533 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2537 raw_spin_rq_unlock(this_rq);
2538 double_rq_lock(this_rq, busiest);
2543 #endif /* CONFIG_PREEMPTION */
2546 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2548 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2550 lockdep_assert_irqs_disabled();
2552 return _double_lock_balance(this_rq, busiest);
2555 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2556 __releases(busiest->lock)
2558 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2559 raw_spin_rq_unlock(busiest);
2560 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2563 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2569 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2572 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2578 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2581 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2587 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2591 * double_rq_unlock - safely unlock two runqueues
2593 * Note this does not restore interrupts like task_rq_unlock,
2594 * you need to do so manually after calling.
2596 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2597 __releases(rq1->lock)
2598 __releases(rq2->lock)
2600 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2601 raw_spin_rq_unlock(rq2);
2603 __release(rq2->lock);
2604 raw_spin_rq_unlock(rq1);
2607 extern void set_rq_online (struct rq *rq);
2608 extern void set_rq_offline(struct rq *rq);
2609 extern bool sched_smp_initialized;
2611 #else /* CONFIG_SMP */
2614 * double_rq_lock - safely lock two runqueues
2616 * Note this does not disable interrupts like task_rq_lock,
2617 * you need to do so manually before calling.
2619 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2620 __acquires(rq1->lock)
2621 __acquires(rq2->lock)
2623 BUG_ON(!irqs_disabled());
2625 raw_spin_rq_lock(rq1);
2626 __acquire(rq2->lock); /* Fake it out ;) */
2630 * double_rq_unlock - safely unlock two runqueues
2632 * Note this does not restore interrupts like task_rq_unlock,
2633 * you need to do so manually after calling.
2635 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2636 __releases(rq1->lock)
2637 __releases(rq2->lock)
2640 raw_spin_rq_unlock(rq1);
2641 __release(rq2->lock);
2646 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2647 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2649 #ifdef CONFIG_SCHED_DEBUG
2650 extern bool sched_debug_verbose;
2652 extern void print_cfs_stats(struct seq_file *m, int cpu);
2653 extern void print_rt_stats(struct seq_file *m, int cpu);
2654 extern void print_dl_stats(struct seq_file *m, int cpu);
2655 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2656 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2657 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2659 extern void resched_latency_warn(int cpu, u64 latency);
2660 #ifdef CONFIG_NUMA_BALANCING
2662 show_numa_stats(struct task_struct *p, struct seq_file *m);
2664 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2665 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2666 #endif /* CONFIG_NUMA_BALANCING */
2668 static inline void resched_latency_warn(int cpu, u64 latency) {}
2669 #endif /* CONFIG_SCHED_DEBUG */
2671 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2672 extern void init_rt_rq(struct rt_rq *rt_rq);
2673 extern void init_dl_rq(struct dl_rq *dl_rq);
2675 extern void cfs_bandwidth_usage_inc(void);
2676 extern void cfs_bandwidth_usage_dec(void);
2678 #ifdef CONFIG_NO_HZ_COMMON
2679 #define NOHZ_BALANCE_KICK_BIT 0
2680 #define NOHZ_STATS_KICK_BIT 1
2681 #define NOHZ_NEWILB_KICK_BIT 2
2683 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2684 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2685 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2687 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2689 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2691 extern void nohz_balance_exit_idle(struct rq *rq);
2693 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2696 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2697 extern void nohz_run_idle_balance(int cpu);
2699 static inline void nohz_run_idle_balance(int cpu) { }
2704 void __dl_update(struct dl_bw *dl_b, s64 bw)
2706 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2709 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2710 "sched RCU must be held");
2711 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2712 struct rq *rq = cpu_rq(i);
2714 rq->dl.extra_bw += bw;
2719 void __dl_update(struct dl_bw *dl_b, s64 bw)
2721 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2728 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2733 struct u64_stats_sync sync;
2736 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2739 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2740 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2741 * and never move forward.
2743 static inline u64 irq_time_read(int cpu)
2745 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2750 seq = __u64_stats_fetch_begin(&irqtime->sync);
2751 total = irqtime->total;
2752 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2756 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2758 #ifdef CONFIG_CPU_FREQ
2759 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2762 * cpufreq_update_util - Take a note about CPU utilization changes.
2763 * @rq: Runqueue to carry out the update for.
2764 * @flags: Update reason flags.
2766 * This function is called by the scheduler on the CPU whose utilization is
2769 * It can only be called from RCU-sched read-side critical sections.
2771 * The way cpufreq is currently arranged requires it to evaluate the CPU
2772 * performance state (frequency/voltage) on a regular basis to prevent it from
2773 * being stuck in a completely inadequate performance level for too long.
2774 * That is not guaranteed to happen if the updates are only triggered from CFS
2775 * and DL, though, because they may not be coming in if only RT tasks are
2776 * active all the time (or there are RT tasks only).
2778 * As a workaround for that issue, this function is called periodically by the
2779 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2780 * but that really is a band-aid. Going forward it should be replaced with
2781 * solutions targeted more specifically at RT tasks.
2783 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2785 struct update_util_data *data;
2787 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2790 data->func(data, rq_clock(rq), flags);
2793 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2794 #endif /* CONFIG_CPU_FREQ */
2796 #ifdef CONFIG_UCLAMP_TASK
2797 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2800 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2801 * @rq: The rq to clamp against. Must not be NULL.
2802 * @util: The util value to clamp.
2803 * @p: The task to clamp against. Can be NULL if you want to clamp
2806 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2808 * If sched_uclamp_used static key is disabled, then just return the util
2809 * without any clamping since uclamp aggregation at the rq level in the fast
2810 * path is disabled, rendering this operation a NOP.
2812 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2813 * will return the correct effective uclamp value of the task even if the
2814 * static key is disabled.
2816 static __always_inline
2817 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2818 struct task_struct *p)
2820 unsigned long min_util;
2821 unsigned long max_util;
2823 if (!static_branch_likely(&sched_uclamp_used))
2826 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2827 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2830 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2831 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2835 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2836 * RUNNABLE tasks with _different_ clamps, we can end up with an
2837 * inversion. Fix it now when the clamps are applied.
2839 if (unlikely(min_util >= max_util))
2842 return clamp(util, min_util, max_util);
2846 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2847 * by default in the fast path and only gets turned on once userspace performs
2848 * an operation that requires it.
2850 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2853 static inline bool uclamp_is_used(void)
2855 return static_branch_likely(&sched_uclamp_used);
2857 #else /* CONFIG_UCLAMP_TASK */
2859 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2860 struct task_struct *p)
2865 static inline bool uclamp_is_used(void)
2869 #endif /* CONFIG_UCLAMP_TASK */
2871 #ifdef arch_scale_freq_capacity
2872 # ifndef arch_scale_freq_invariant
2873 # define arch_scale_freq_invariant() true
2876 # define arch_scale_freq_invariant() false
2880 static inline unsigned long capacity_orig_of(int cpu)
2882 return cpu_rq(cpu)->cpu_capacity_orig;
2886 * enum cpu_util_type - CPU utilization type
2887 * @FREQUENCY_UTIL: Utilization used to select frequency
2888 * @ENERGY_UTIL: Utilization used during energy calculation
2890 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2891 * need to be aggregated differently depending on the usage made of them. This
2892 * enum is used within effective_cpu_util() to differentiate the types of
2893 * utilization expected by the callers, and adjust the aggregation accordingly.
2895 enum cpu_util_type {
2900 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2901 unsigned long max, enum cpu_util_type type,
2902 struct task_struct *p);
2904 static inline unsigned long cpu_bw_dl(struct rq *rq)
2906 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2909 static inline unsigned long cpu_util_dl(struct rq *rq)
2911 return READ_ONCE(rq->avg_dl.util_avg);
2914 static inline unsigned long cpu_util_cfs(struct rq *rq)
2916 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2918 if (sched_feat(UTIL_EST)) {
2919 util = max_t(unsigned long, util,
2920 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2926 static inline unsigned long cpu_util_rt(struct rq *rq)
2928 return READ_ONCE(rq->avg_rt.util_avg);
2932 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2933 static inline unsigned long cpu_util_irq(struct rq *rq)
2935 return rq->avg_irq.util_avg;
2939 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2941 util *= (max - irq);
2948 static inline unsigned long cpu_util_irq(struct rq *rq)
2954 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2960 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2962 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2964 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2966 static inline bool sched_energy_enabled(void)
2968 return static_branch_unlikely(&sched_energy_present);
2971 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2973 #define perf_domain_span(pd) NULL
2974 static inline bool sched_energy_enabled(void) { return false; }
2976 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2978 #ifdef CONFIG_MEMBARRIER
2980 * The scheduler provides memory barriers required by membarrier between:
2981 * - prior user-space memory accesses and store to rq->membarrier_state,
2982 * - store to rq->membarrier_state and following user-space memory accesses.
2983 * In the same way it provides those guarantees around store to rq->curr.
2985 static inline void membarrier_switch_mm(struct rq *rq,
2986 struct mm_struct *prev_mm,
2987 struct mm_struct *next_mm)
2989 int membarrier_state;
2991 if (prev_mm == next_mm)
2994 membarrier_state = atomic_read(&next_mm->membarrier_state);
2995 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2998 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3001 static inline void membarrier_switch_mm(struct rq *rq,
3002 struct mm_struct *prev_mm,
3003 struct mm_struct *next_mm)
3009 static inline bool is_per_cpu_kthread(struct task_struct *p)
3011 if (!(p->flags & PF_KTHREAD))
3014 if (p->nr_cpus_allowed != 1)
3021 extern void swake_up_all_locked(struct swait_queue_head *q);
3022 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3024 #ifdef CONFIG_PREEMPT_DYNAMIC
3025 extern int preempt_dynamic_mode;
3026 extern int sched_dynamic_mode(const char *str);
3027 extern void sched_dynamic_update(int mode);