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 */
530 u64 min_vruntime_copy;
533 struct rb_root_cached tasks_timeline;
536 * 'curr' points to currently running entity on this cfs_rq.
537 * It is set to NULL otherwise (i.e when none are currently running).
539 struct sched_entity *curr;
540 struct sched_entity *next;
541 struct sched_entity *last;
542 struct sched_entity *skip;
544 #ifdef CONFIG_SCHED_DEBUG
545 unsigned int nr_spread_over;
552 struct sched_avg avg;
554 u64 load_last_update_time_copy;
557 raw_spinlock_t lock ____cacheline_aligned;
559 unsigned long load_avg;
560 unsigned long util_avg;
561 unsigned long runnable_avg;
564 #ifdef CONFIG_FAIR_GROUP_SCHED
565 unsigned long tg_load_avg_contrib;
567 long prop_runnable_sum;
570 * h_load = weight * f(tg)
572 * Where f(tg) is the recursive weight fraction assigned to
575 unsigned long h_load;
576 u64 last_h_load_update;
577 struct sched_entity *h_load_next;
578 #endif /* CONFIG_FAIR_GROUP_SCHED */
579 #endif /* CONFIG_SMP */
581 #ifdef CONFIG_FAIR_GROUP_SCHED
582 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
585 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
586 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
587 * (like users, containers etc.)
589 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
590 * This list is used during load balance.
593 struct list_head leaf_cfs_rq_list;
594 struct task_group *tg; /* group that "owns" this runqueue */
596 #ifdef CONFIG_CFS_BANDWIDTH
598 s64 runtime_remaining;
601 u64 throttled_clock_task;
602 u64 throttled_clock_task_time;
605 struct list_head throttled_list;
606 #endif /* CONFIG_CFS_BANDWIDTH */
607 #endif /* CONFIG_FAIR_GROUP_SCHED */
610 static inline int rt_bandwidth_enabled(void)
612 return sysctl_sched_rt_runtime >= 0;
615 /* RT IPI pull logic requires IRQ_WORK */
616 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
617 # define HAVE_RT_PUSH_IPI
620 /* Real-Time classes' related field in a runqueue: */
622 struct rt_prio_array active;
623 unsigned int rt_nr_running;
624 unsigned int rr_nr_running;
625 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
627 int curr; /* highest queued rt task prio */
629 int next; /* next highest */
634 unsigned long rt_nr_migratory;
635 unsigned long rt_nr_total;
637 struct plist_head pushable_tasks;
639 #endif /* CONFIG_SMP */
645 /* Nests inside the rq lock: */
646 raw_spinlock_t rt_runtime_lock;
648 #ifdef CONFIG_RT_GROUP_SCHED
649 unsigned long rt_nr_boosted;
652 struct task_group *tg;
656 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
658 return rt_rq->rt_queued && rt_rq->rt_nr_running;
661 /* Deadline class' related fields in a runqueue */
663 /* runqueue is an rbtree, ordered by deadline */
664 struct rb_root_cached root;
666 unsigned long dl_nr_running;
670 * Deadline values of the currently executing and the
671 * earliest ready task on this rq. Caching these facilitates
672 * the decision whether or not a ready but not running task
673 * should migrate somewhere else.
680 unsigned long dl_nr_migratory;
684 * Tasks on this rq that can be pushed away. They are kept in
685 * an rb-tree, ordered by tasks' deadlines, with caching
686 * of the leftmost (earliest deadline) element.
688 struct rb_root_cached pushable_dl_tasks_root;
693 * "Active utilization" for this runqueue: increased when a
694 * task wakes up (becomes TASK_RUNNING) and decreased when a
700 * Utilization of the tasks "assigned" to this runqueue (including
701 * the tasks that are in runqueue and the tasks that executed on this
702 * CPU and blocked). Increased when a task moves to this runqueue, and
703 * decreased when the task moves away (migrates, changes scheduling
704 * policy, or terminates).
705 * This is needed to compute the "inactive utilization" for the
706 * runqueue (inactive utilization = this_bw - running_bw).
712 * Inverse of the fraction of CPU utilization that can be reclaimed
713 * by the GRUB algorithm.
718 #ifdef CONFIG_FAIR_GROUP_SCHED
719 /* An entity is a task if it doesn't "own" a runqueue */
720 #define entity_is_task(se) (!se->my_q)
722 static inline void se_update_runnable(struct sched_entity *se)
724 if (!entity_is_task(se))
725 se->runnable_weight = se->my_q->h_nr_running;
728 static inline long se_runnable(struct sched_entity *se)
730 if (entity_is_task(se))
733 return se->runnable_weight;
737 #define entity_is_task(se) 1
739 static inline void se_update_runnable(struct sched_entity *se) {}
741 static inline long se_runnable(struct sched_entity *se)
749 * XXX we want to get rid of these helpers and use the full load resolution.
751 static inline long se_weight(struct sched_entity *se)
753 return scale_load_down(se->load.weight);
757 static inline bool sched_asym_prefer(int a, int b)
759 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
763 struct em_perf_domain *em_pd;
764 struct perf_domain *next;
768 /* Scheduling group status flags */
769 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
770 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
773 * We add the notion of a root-domain which will be used to define per-domain
774 * variables. Each exclusive cpuset essentially defines an island domain by
775 * fully partitioning the member CPUs from any other cpuset. Whenever a new
776 * exclusive cpuset is created, we also create and attach a new root-domain
785 cpumask_var_t online;
788 * Indicate pullable load on at least one CPU, e.g:
789 * - More than one runnable task
790 * - Running task is misfit
794 /* Indicate one or more cpus over-utilized (tipping point) */
798 * The bit corresponding to a CPU gets set here if such CPU has more
799 * than one runnable -deadline task (as it is below for RT tasks).
801 cpumask_var_t dlo_mask;
807 * Indicate whether a root_domain's dl_bw has been checked or
808 * updated. It's monotonously increasing value.
810 * Also, some corner cases, like 'wrap around' is dangerous, but given
811 * that u64 is 'big enough'. So that shouldn't be a concern.
815 #ifdef HAVE_RT_PUSH_IPI
817 * For IPI pull requests, loop across the rto_mask.
819 struct irq_work rto_push_work;
820 raw_spinlock_t rto_lock;
821 /* These are only updated and read within rto_lock */
824 /* These atomics are updated outside of a lock */
825 atomic_t rto_loop_next;
826 atomic_t rto_loop_start;
829 * The "RT overload" flag: it gets set if a CPU has more than
830 * one runnable RT task.
832 cpumask_var_t rto_mask;
833 struct cpupri cpupri;
835 unsigned long max_cpu_capacity;
838 * NULL-terminated list of performance domains intersecting with the
839 * CPUs of the rd. Protected by RCU.
841 struct perf_domain __rcu *pd;
844 extern void init_defrootdomain(void);
845 extern int sched_init_domains(const struct cpumask *cpu_map);
846 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
847 extern void sched_get_rd(struct root_domain *rd);
848 extern void sched_put_rd(struct root_domain *rd);
850 #ifdef HAVE_RT_PUSH_IPI
851 extern void rto_push_irq_work_func(struct irq_work *work);
853 #endif /* CONFIG_SMP */
855 #ifdef CONFIG_UCLAMP_TASK
857 * struct uclamp_bucket - Utilization clamp bucket
858 * @value: utilization clamp value for tasks on this clamp bucket
859 * @tasks: number of RUNNABLE tasks on this clamp bucket
861 * Keep track of how many tasks are RUNNABLE for a given utilization
864 struct uclamp_bucket {
865 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
866 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
870 * struct uclamp_rq - rq's utilization clamp
871 * @value: currently active clamp values for a rq
872 * @bucket: utilization clamp buckets affecting a rq
874 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
875 * A clamp value is affecting a rq when there is at least one task RUNNABLE
876 * (or actually running) with that value.
878 * There are up to UCLAMP_CNT possible different clamp values, currently there
879 * are only two: minimum utilization and maximum utilization.
881 * All utilization clamping values are MAX aggregated, since:
882 * - for util_min: we want to run the CPU at least at the max of the minimum
883 * utilization required by its currently RUNNABLE tasks.
884 * - for util_max: we want to allow the CPU to run up to the max of the
885 * maximum utilization allowed by its currently RUNNABLE tasks.
887 * Since on each system we expect only a limited number of different
888 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
889 * the metrics required to compute all the per-rq utilization clamp values.
893 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
896 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
897 #endif /* CONFIG_UCLAMP_TASK */
900 * This is the main, per-CPU runqueue data structure.
902 * Locking rule: those places that want to lock multiple runqueues
903 * (such as the load balancing or the thread migration code), lock
904 * acquire operations must be ordered by ascending &runqueue.
908 raw_spinlock_t __lock;
911 * nr_running and cpu_load should be in the same cacheline because
912 * remote CPUs use both these fields when doing load calculation.
914 unsigned int nr_running;
915 #ifdef CONFIG_NUMA_BALANCING
916 unsigned int nr_numa_running;
917 unsigned int nr_preferred_running;
918 unsigned int numa_migrate_on;
920 #ifdef CONFIG_NO_HZ_COMMON
922 unsigned long last_blocked_load_update_tick;
923 unsigned int has_blocked_load;
924 call_single_data_t nohz_csd;
925 #endif /* CONFIG_SMP */
926 unsigned int nohz_tick_stopped;
928 #endif /* CONFIG_NO_HZ_COMMON */
931 unsigned int ttwu_pending;
935 #ifdef CONFIG_UCLAMP_TASK
936 /* Utilization clamp values based on CPU's RUNNABLE tasks */
937 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
938 unsigned int uclamp_flags;
939 #define UCLAMP_FLAG_IDLE 0x01
946 #ifdef CONFIG_FAIR_GROUP_SCHED
947 /* list of leaf cfs_rq on this CPU: */
948 struct list_head leaf_cfs_rq_list;
949 struct list_head *tmp_alone_branch;
950 #endif /* CONFIG_FAIR_GROUP_SCHED */
953 * This is part of a global counter where only the total sum
954 * over all CPUs matters. A task can increase this counter on
955 * one CPU and if it got migrated afterwards it may decrease
956 * it on another CPU. Always updated under the runqueue lock:
958 unsigned long nr_uninterruptible;
960 struct task_struct __rcu *curr;
961 struct task_struct *idle;
962 struct task_struct *stop;
963 unsigned long next_balance;
964 struct mm_struct *prev_mm;
966 unsigned int clock_update_flags;
968 /* Ensure that all clocks are in the same cache line */
969 u64 clock_task ____cacheline_aligned;
971 unsigned long lost_idle_time;
975 #ifdef CONFIG_SCHED_DEBUG
976 u64 last_seen_need_resched_ns;
977 int ticks_without_resched;
980 #ifdef CONFIG_MEMBARRIER
981 int membarrier_state;
985 struct root_domain *rd;
986 struct sched_domain __rcu *sd;
988 unsigned long cpu_capacity;
989 unsigned long cpu_capacity_orig;
991 struct callback_head *balance_callback;
993 unsigned char nohz_idle_balance;
994 unsigned char idle_balance;
996 unsigned long misfit_task_load;
998 /* For active balancing */
1001 struct cpu_stop_work active_balance_work;
1003 /* CPU of this runqueue: */
1007 struct list_head cfs_tasks;
1009 struct sched_avg avg_rt;
1010 struct sched_avg avg_dl;
1011 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1012 struct sched_avg avg_irq;
1014 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1015 struct sched_avg avg_thermal;
1020 /* This is used to determine avg_idle's max value */
1021 u64 max_idle_balance_cost;
1023 #ifdef CONFIG_HOTPLUG_CPU
1024 struct rcuwait hotplug_wait;
1026 #endif /* CONFIG_SMP */
1028 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1031 #ifdef CONFIG_PARAVIRT
1032 u64 prev_steal_time;
1034 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1035 u64 prev_steal_time_rq;
1038 /* calc_load related fields */
1039 unsigned long calc_load_update;
1040 long calc_load_active;
1042 #ifdef CONFIG_SCHED_HRTICK
1044 call_single_data_t hrtick_csd;
1046 struct hrtimer hrtick_timer;
1047 ktime_t hrtick_time;
1050 #ifdef CONFIG_SCHEDSTATS
1052 struct sched_info rq_sched_info;
1053 unsigned long long rq_cpu_time;
1054 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1056 /* sys_sched_yield() stats */
1057 unsigned int yld_count;
1059 /* schedule() stats */
1060 unsigned int sched_count;
1061 unsigned int sched_goidle;
1063 /* try_to_wake_up() stats */
1064 unsigned int ttwu_count;
1065 unsigned int ttwu_local;
1068 #ifdef CONFIG_CPU_IDLE
1069 /* Must be inspected within a rcu lock section */
1070 struct cpuidle_state *idle_state;
1074 unsigned int nr_pinned;
1076 unsigned int push_busy;
1077 struct cpu_stop_work push_work;
1079 #ifdef CONFIG_SCHED_CORE
1082 unsigned int core_enabled;
1086 #ifdef CONFIG_FAIR_GROUP_SCHED
1088 /* CPU runqueue to which this cfs_rq is attached */
1089 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1096 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1098 return container_of(cfs_rq, struct rq, cfs);
1102 static inline int cpu_of(struct rq *rq)
1111 #define MDF_PUSH 0x01
1113 static inline bool is_migration_disabled(struct task_struct *p)
1116 return p->migration_disabled;
1122 #ifdef CONFIG_SCHED_CORE
1124 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1126 static inline bool sched_core_enabled(struct rq *rq)
1128 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1131 static inline bool sched_core_disabled(void)
1133 return !static_branch_unlikely(&__sched_core_enabled);
1137 * Be careful with this function; not for general use. The return value isn't
1138 * stable unless you actually hold a relevant rq->__lock.
1140 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1142 if (sched_core_enabled(rq))
1143 return &rq->core->__lock;
1148 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1150 if (rq->core_enabled)
1151 return &rq->core->__lock;
1156 #else /* !CONFIG_SCHED_CORE */
1158 static inline bool sched_core_enabled(struct rq *rq)
1163 static inline bool sched_core_disabled(void)
1168 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1173 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1178 #endif /* CONFIG_SCHED_CORE */
1180 static inline void lockdep_assert_rq_held(struct rq *rq)
1182 lockdep_assert_held(__rq_lockp(rq));
1185 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1186 extern bool raw_spin_rq_trylock(struct rq *rq);
1187 extern void raw_spin_rq_unlock(struct rq *rq);
1189 static inline void raw_spin_rq_lock(struct rq *rq)
1191 raw_spin_rq_lock_nested(rq, 0);
1194 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1196 local_irq_disable();
1197 raw_spin_rq_lock(rq);
1200 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1202 raw_spin_rq_unlock(rq);
1206 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1208 unsigned long flags;
1209 local_irq_save(flags);
1210 raw_spin_rq_lock(rq);
1214 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1216 raw_spin_rq_unlock(rq);
1217 local_irq_restore(flags);
1220 #define raw_spin_rq_lock_irqsave(rq, flags) \
1222 flags = _raw_spin_rq_lock_irqsave(rq); \
1225 #ifdef CONFIG_SCHED_SMT
1226 extern void __update_idle_core(struct rq *rq);
1228 static inline void update_idle_core(struct rq *rq)
1230 if (static_branch_unlikely(&sched_smt_present))
1231 __update_idle_core(rq);
1235 static inline void update_idle_core(struct rq *rq) { }
1238 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1240 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1241 #define this_rq() this_cpu_ptr(&runqueues)
1242 #define task_rq(p) cpu_rq(task_cpu(p))
1243 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1244 #define raw_rq() raw_cpu_ptr(&runqueues)
1246 extern void update_rq_clock(struct rq *rq);
1248 static inline u64 __rq_clock_broken(struct rq *rq)
1250 return READ_ONCE(rq->clock);
1254 * rq::clock_update_flags bits
1256 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1257 * call to __schedule(). This is an optimisation to avoid
1258 * neighbouring rq clock updates.
1260 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1261 * in effect and calls to update_rq_clock() are being ignored.
1263 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1264 * made to update_rq_clock() since the last time rq::lock was pinned.
1266 * If inside of __schedule(), clock_update_flags will have been
1267 * shifted left (a left shift is a cheap operation for the fast path
1268 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1270 * if (rq-clock_update_flags >= RQCF_UPDATED)
1272 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1273 * one position though, because the next rq_unpin_lock() will shift it
1276 #define RQCF_REQ_SKIP 0x01
1277 #define RQCF_ACT_SKIP 0x02
1278 #define RQCF_UPDATED 0x04
1280 static inline void assert_clock_updated(struct rq *rq)
1283 * The only reason for not seeing a clock update since the
1284 * last rq_pin_lock() is if we're currently skipping updates.
1286 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1289 static inline u64 rq_clock(struct rq *rq)
1291 lockdep_assert_rq_held(rq);
1292 assert_clock_updated(rq);
1297 static inline u64 rq_clock_task(struct rq *rq)
1299 lockdep_assert_rq_held(rq);
1300 assert_clock_updated(rq);
1302 return rq->clock_task;
1306 * By default the decay is the default pelt decay period.
1307 * The decay shift can change the decay period in
1309 * Decay shift Decay period(ms)
1316 extern int sched_thermal_decay_shift;
1318 static inline u64 rq_clock_thermal(struct rq *rq)
1320 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1323 static inline void rq_clock_skip_update(struct rq *rq)
1325 lockdep_assert_rq_held(rq);
1326 rq->clock_update_flags |= RQCF_REQ_SKIP;
1330 * See rt task throttling, which is the only time a skip
1331 * request is canceled.
1333 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1335 lockdep_assert_rq_held(rq);
1336 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1340 unsigned long flags;
1341 struct pin_cookie cookie;
1342 #ifdef CONFIG_SCHED_DEBUG
1344 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1345 * current pin context is stashed here in case it needs to be
1346 * restored in rq_repin_lock().
1348 unsigned int clock_update_flags;
1352 extern struct callback_head balance_push_callback;
1355 * Lockdep annotation that avoids accidental unlocks; it's like a
1356 * sticky/continuous lockdep_assert_held().
1358 * This avoids code that has access to 'struct rq *rq' (basically everything in
1359 * the scheduler) from accidentally unlocking the rq if they do not also have a
1360 * copy of the (on-stack) 'struct rq_flags rf'.
1362 * Also see Documentation/locking/lockdep-design.rst.
1364 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1366 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1368 #ifdef CONFIG_SCHED_DEBUG
1369 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1370 rf->clock_update_flags = 0;
1372 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1377 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1379 #ifdef CONFIG_SCHED_DEBUG
1380 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1381 rf->clock_update_flags = RQCF_UPDATED;
1384 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1387 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1389 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1391 #ifdef CONFIG_SCHED_DEBUG
1393 * Restore the value we stashed in @rf for this pin context.
1395 rq->clock_update_flags |= rf->clock_update_flags;
1399 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1400 __acquires(rq->lock);
1402 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1403 __acquires(p->pi_lock)
1404 __acquires(rq->lock);
1406 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1407 __releases(rq->lock)
1409 rq_unpin_lock(rq, rf);
1410 raw_spin_rq_unlock(rq);
1414 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1415 __releases(rq->lock)
1416 __releases(p->pi_lock)
1418 rq_unpin_lock(rq, rf);
1419 raw_spin_rq_unlock(rq);
1420 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1424 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1425 __acquires(rq->lock)
1427 raw_spin_rq_lock_irqsave(rq, rf->flags);
1428 rq_pin_lock(rq, rf);
1432 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1433 __acquires(rq->lock)
1435 raw_spin_rq_lock_irq(rq);
1436 rq_pin_lock(rq, rf);
1440 rq_lock(struct rq *rq, struct rq_flags *rf)
1441 __acquires(rq->lock)
1443 raw_spin_rq_lock(rq);
1444 rq_pin_lock(rq, rf);
1448 rq_relock(struct rq *rq, struct rq_flags *rf)
1449 __acquires(rq->lock)
1451 raw_spin_rq_lock(rq);
1452 rq_repin_lock(rq, rf);
1456 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1457 __releases(rq->lock)
1459 rq_unpin_lock(rq, rf);
1460 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1464 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1465 __releases(rq->lock)
1467 rq_unpin_lock(rq, rf);
1468 raw_spin_rq_unlock_irq(rq);
1472 rq_unlock(struct rq *rq, struct rq_flags *rf)
1473 __releases(rq->lock)
1475 rq_unpin_lock(rq, rf);
1476 raw_spin_rq_unlock(rq);
1479 static inline struct rq *
1480 this_rq_lock_irq(struct rq_flags *rf)
1481 __acquires(rq->lock)
1485 local_irq_disable();
1492 enum numa_topology_type {
1497 extern enum numa_topology_type sched_numa_topology_type;
1498 extern int sched_max_numa_distance;
1499 extern bool find_numa_distance(int distance);
1500 extern void sched_init_numa(void);
1501 extern void sched_domains_numa_masks_set(unsigned int cpu);
1502 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1503 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1505 static inline void sched_init_numa(void) { }
1506 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1507 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1508 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1514 #ifdef CONFIG_NUMA_BALANCING
1515 /* The regions in numa_faults array from task_struct */
1516 enum numa_faults_stats {
1522 extern void sched_setnuma(struct task_struct *p, int node);
1523 extern int migrate_task_to(struct task_struct *p, int cpu);
1524 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1526 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1529 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1532 #endif /* CONFIG_NUMA_BALANCING */
1537 queue_balance_callback(struct rq *rq,
1538 struct callback_head *head,
1539 void (*func)(struct rq *rq))
1541 lockdep_assert_rq_held(rq);
1543 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1546 head->func = (void (*)(struct callback_head *))func;
1547 head->next = rq->balance_callback;
1548 rq->balance_callback = head;
1551 #define rcu_dereference_check_sched_domain(p) \
1552 rcu_dereference_check((p), \
1553 lockdep_is_held(&sched_domains_mutex))
1556 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1557 * See destroy_sched_domains: call_rcu for details.
1559 * The domain tree of any CPU may only be accessed from within
1560 * preempt-disabled sections.
1562 #define for_each_domain(cpu, __sd) \
1563 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1564 __sd; __sd = __sd->parent)
1567 * highest_flag_domain - Return highest sched_domain containing flag.
1568 * @cpu: The CPU whose highest level of sched domain is to
1570 * @flag: The flag to check for the highest sched_domain
1571 * for the given CPU.
1573 * Returns the highest sched_domain of a CPU which contains the given flag.
1575 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1577 struct sched_domain *sd, *hsd = NULL;
1579 for_each_domain(cpu, sd) {
1580 if (!(sd->flags & flag))
1588 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1590 struct sched_domain *sd;
1592 for_each_domain(cpu, sd) {
1593 if (sd->flags & flag)
1600 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1601 DECLARE_PER_CPU(int, sd_llc_size);
1602 DECLARE_PER_CPU(int, sd_llc_id);
1603 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1604 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1605 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1606 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1607 extern struct static_key_false sched_asym_cpucapacity;
1609 struct sched_group_capacity {
1612 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1615 unsigned long capacity;
1616 unsigned long min_capacity; /* Min per-CPU capacity in group */
1617 unsigned long max_capacity; /* Max per-CPU capacity in group */
1618 unsigned long next_update;
1619 int imbalance; /* XXX unrelated to capacity but shared group state */
1621 #ifdef CONFIG_SCHED_DEBUG
1625 unsigned long cpumask[]; /* Balance mask */
1628 struct sched_group {
1629 struct sched_group *next; /* Must be a circular list */
1632 unsigned int group_weight;
1633 struct sched_group_capacity *sgc;
1634 int asym_prefer_cpu; /* CPU of highest priority in group */
1637 * The CPUs this group covers.
1639 * NOTE: this field is variable length. (Allocated dynamically
1640 * by attaching extra space to the end of the structure,
1641 * depending on how many CPUs the kernel has booted up with)
1643 unsigned long cpumask[];
1646 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1648 return to_cpumask(sg->cpumask);
1652 * See build_balance_mask().
1654 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1656 return to_cpumask(sg->sgc->cpumask);
1660 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1661 * @group: The group whose first CPU is to be returned.
1663 static inline unsigned int group_first_cpu(struct sched_group *group)
1665 return cpumask_first(sched_group_span(group));
1668 extern int group_balance_cpu(struct sched_group *sg);
1670 #ifdef CONFIG_SCHED_DEBUG
1671 void update_sched_domain_debugfs(void);
1672 void dirty_sched_domain_sysctl(int cpu);
1674 static inline void update_sched_domain_debugfs(void)
1677 static inline void dirty_sched_domain_sysctl(int cpu)
1682 extern int sched_update_scaling(void);
1684 extern void flush_smp_call_function_from_idle(void);
1686 #else /* !CONFIG_SMP: */
1687 static inline void flush_smp_call_function_from_idle(void) { }
1691 #include "autogroup.h"
1693 #ifdef CONFIG_CGROUP_SCHED
1696 * Return the group to which this tasks belongs.
1698 * We cannot use task_css() and friends because the cgroup subsystem
1699 * changes that value before the cgroup_subsys::attach() method is called,
1700 * therefore we cannot pin it and might observe the wrong value.
1702 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1703 * core changes this before calling sched_move_task().
1705 * Instead we use a 'copy' which is updated from sched_move_task() while
1706 * holding both task_struct::pi_lock and rq::lock.
1708 static inline struct task_group *task_group(struct task_struct *p)
1710 return p->sched_task_group;
1713 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1714 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1716 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1717 struct task_group *tg = task_group(p);
1720 #ifdef CONFIG_FAIR_GROUP_SCHED
1721 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1722 p->se.cfs_rq = tg->cfs_rq[cpu];
1723 p->se.parent = tg->se[cpu];
1726 #ifdef CONFIG_RT_GROUP_SCHED
1727 p->rt.rt_rq = tg->rt_rq[cpu];
1728 p->rt.parent = tg->rt_se[cpu];
1732 #else /* CONFIG_CGROUP_SCHED */
1734 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1735 static inline struct task_group *task_group(struct task_struct *p)
1740 #endif /* CONFIG_CGROUP_SCHED */
1742 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1744 set_task_rq(p, cpu);
1747 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1748 * successfully executed on another CPU. We must ensure that updates of
1749 * per-task data have been completed by this moment.
1752 #ifdef CONFIG_THREAD_INFO_IN_TASK
1753 WRITE_ONCE(p->cpu, cpu);
1755 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1762 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1764 #ifdef CONFIG_SCHED_DEBUG
1765 # include <linux/static_key.h>
1766 # define const_debug __read_mostly
1768 # define const_debug const
1771 #define SCHED_FEAT(name, enabled) \
1772 __SCHED_FEAT_##name ,
1775 #include "features.h"
1781 #ifdef CONFIG_SCHED_DEBUG
1784 * To support run-time toggling of sched features, all the translation units
1785 * (but core.c) reference the sysctl_sched_features defined in core.c.
1787 extern const_debug unsigned int sysctl_sched_features;
1789 #ifdef CONFIG_JUMP_LABEL
1790 #define SCHED_FEAT(name, enabled) \
1791 static __always_inline bool static_branch_##name(struct static_key *key) \
1793 return static_key_##enabled(key); \
1796 #include "features.h"
1799 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1800 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1802 #else /* !CONFIG_JUMP_LABEL */
1804 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1806 #endif /* CONFIG_JUMP_LABEL */
1808 #else /* !SCHED_DEBUG */
1811 * Each translation unit has its own copy of sysctl_sched_features to allow
1812 * constants propagation at compile time and compiler optimization based on
1815 #define SCHED_FEAT(name, enabled) \
1816 (1UL << __SCHED_FEAT_##name) * enabled |
1817 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1818 #include "features.h"
1822 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1824 #endif /* SCHED_DEBUG */
1826 extern struct static_key_false sched_numa_balancing;
1827 extern struct static_key_false sched_schedstats;
1829 static inline u64 global_rt_period(void)
1831 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1834 static inline u64 global_rt_runtime(void)
1836 if (sysctl_sched_rt_runtime < 0)
1839 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1842 static inline int task_current(struct rq *rq, struct task_struct *p)
1844 return rq->curr == p;
1847 static inline int task_running(struct rq *rq, struct task_struct *p)
1852 return task_current(rq, p);
1856 static inline int task_on_rq_queued(struct task_struct *p)
1858 return p->on_rq == TASK_ON_RQ_QUEUED;
1861 static inline int task_on_rq_migrating(struct task_struct *p)
1863 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1866 /* Wake flags. The first three directly map to some SD flag value */
1867 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
1868 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
1869 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
1871 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
1872 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
1873 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
1876 static_assert(WF_EXEC == SD_BALANCE_EXEC);
1877 static_assert(WF_FORK == SD_BALANCE_FORK);
1878 static_assert(WF_TTWU == SD_BALANCE_WAKE);
1882 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1883 * of tasks with abnormal "nice" values across CPUs the contribution that
1884 * each task makes to its run queue's load is weighted according to its
1885 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1886 * scaled version of the new time slice allocation that they receive on time
1890 #define WEIGHT_IDLEPRIO 3
1891 #define WMULT_IDLEPRIO 1431655765
1893 extern const int sched_prio_to_weight[40];
1894 extern const u32 sched_prio_to_wmult[40];
1897 * {de,en}queue flags:
1899 * DEQUEUE_SLEEP - task is no longer runnable
1900 * ENQUEUE_WAKEUP - task just became runnable
1902 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1903 * are in a known state which allows modification. Such pairs
1904 * should preserve as much state as possible.
1906 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1909 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1910 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1911 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1915 #define DEQUEUE_SLEEP 0x01
1916 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1917 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1918 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1920 #define ENQUEUE_WAKEUP 0x01
1921 #define ENQUEUE_RESTORE 0x02
1922 #define ENQUEUE_MOVE 0x04
1923 #define ENQUEUE_NOCLOCK 0x08
1925 #define ENQUEUE_HEAD 0x10
1926 #define ENQUEUE_REPLENISH 0x20
1928 #define ENQUEUE_MIGRATED 0x40
1930 #define ENQUEUE_MIGRATED 0x00
1933 #define RETRY_TASK ((void *)-1UL)
1935 struct sched_class {
1937 #ifdef CONFIG_UCLAMP_TASK
1941 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1942 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1943 void (*yield_task) (struct rq *rq);
1944 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1946 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1948 struct task_struct *(*pick_next_task)(struct rq *rq);
1950 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1951 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1954 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1955 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
1957 struct task_struct * (*pick_task)(struct rq *rq);
1959 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1961 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1963 void (*set_cpus_allowed)(struct task_struct *p,
1964 const struct cpumask *newmask,
1967 void (*rq_online)(struct rq *rq);
1968 void (*rq_offline)(struct rq *rq);
1970 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
1973 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1974 void (*task_fork)(struct task_struct *p);
1975 void (*task_dead)(struct task_struct *p);
1978 * The switched_from() call is allowed to drop rq->lock, therefore we
1979 * cannot assume the switched_from/switched_to pair is serialized by
1980 * rq->lock. They are however serialized by p->pi_lock.
1982 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1983 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1984 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1987 unsigned int (*get_rr_interval)(struct rq *rq,
1988 struct task_struct *task);
1990 void (*update_curr)(struct rq *rq);
1992 #define TASK_SET_GROUP 0
1993 #define TASK_MOVE_GROUP 1
1995 #ifdef CONFIG_FAIR_GROUP_SCHED
1996 void (*task_change_group)(struct task_struct *p, int type);
2000 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2002 WARN_ON_ONCE(rq->curr != prev);
2003 prev->sched_class->put_prev_task(rq, prev);
2006 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2008 WARN_ON_ONCE(rq->curr != next);
2009 next->sched_class->set_next_task(rq, next, false);
2014 * Helper to define a sched_class instance; each one is placed in a separate
2015 * section which is ordered by the linker script:
2017 * include/asm-generic/vmlinux.lds.h
2019 * Also enforce alignment on the instance, not the type, to guarantee layout.
2021 #define DEFINE_SCHED_CLASS(name) \
2022 const struct sched_class name##_sched_class \
2023 __aligned(__alignof__(struct sched_class)) \
2024 __section("__" #name "_sched_class")
2026 /* Defined in include/asm-generic/vmlinux.lds.h */
2027 extern struct sched_class __begin_sched_classes[];
2028 extern struct sched_class __end_sched_classes[];
2030 #define sched_class_highest (__end_sched_classes - 1)
2031 #define sched_class_lowest (__begin_sched_classes - 1)
2033 #define for_class_range(class, _from, _to) \
2034 for (class = (_from); class != (_to); class--)
2036 #define for_each_class(class) \
2037 for_class_range(class, sched_class_highest, sched_class_lowest)
2039 extern const struct sched_class stop_sched_class;
2040 extern const struct sched_class dl_sched_class;
2041 extern const struct sched_class rt_sched_class;
2042 extern const struct sched_class fair_sched_class;
2043 extern const struct sched_class idle_sched_class;
2045 static inline bool sched_stop_runnable(struct rq *rq)
2047 return rq->stop && task_on_rq_queued(rq->stop);
2050 static inline bool sched_dl_runnable(struct rq *rq)
2052 return rq->dl.dl_nr_running > 0;
2055 static inline bool sched_rt_runnable(struct rq *rq)
2057 return rq->rt.rt_queued > 0;
2060 static inline bool sched_fair_runnable(struct rq *rq)
2062 return rq->cfs.nr_running > 0;
2065 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2066 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2068 #define SCA_CHECK 0x01
2069 #define SCA_MIGRATE_DISABLE 0x02
2070 #define SCA_MIGRATE_ENABLE 0x04
2074 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2076 extern void trigger_load_balance(struct rq *rq);
2078 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2080 static inline struct task_struct *get_push_task(struct rq *rq)
2082 struct task_struct *p = rq->curr;
2084 lockdep_assert_rq_held(rq);
2089 if (p->nr_cpus_allowed == 1)
2092 rq->push_busy = true;
2093 return get_task_struct(p);
2096 extern int push_cpu_stop(void *arg);
2100 #ifdef CONFIG_CPU_IDLE
2101 static inline void idle_set_state(struct rq *rq,
2102 struct cpuidle_state *idle_state)
2104 rq->idle_state = idle_state;
2107 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2109 SCHED_WARN_ON(!rcu_read_lock_held());
2111 return rq->idle_state;
2114 static inline void idle_set_state(struct rq *rq,
2115 struct cpuidle_state *idle_state)
2119 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2125 extern void schedule_idle(void);
2127 extern void sysrq_sched_debug_show(void);
2128 extern void sched_init_granularity(void);
2129 extern void update_max_interval(void);
2131 extern void init_sched_dl_class(void);
2132 extern void init_sched_rt_class(void);
2133 extern void init_sched_fair_class(void);
2135 extern void reweight_task(struct task_struct *p, int prio);
2137 extern void resched_curr(struct rq *rq);
2138 extern void resched_cpu(int cpu);
2140 extern struct rt_bandwidth def_rt_bandwidth;
2141 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2143 extern struct dl_bandwidth def_dl_bandwidth;
2144 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2145 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2146 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2149 #define BW_UNIT (1 << BW_SHIFT)
2150 #define RATIO_SHIFT 8
2151 #define MAX_BW_BITS (64 - BW_SHIFT)
2152 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2153 unsigned long to_ratio(u64 period, u64 runtime);
2155 extern void init_entity_runnable_average(struct sched_entity *se);
2156 extern void post_init_entity_util_avg(struct task_struct *p);
2158 #ifdef CONFIG_NO_HZ_FULL
2159 extern bool sched_can_stop_tick(struct rq *rq);
2160 extern int __init sched_tick_offload_init(void);
2163 * Tick may be needed by tasks in the runqueue depending on their policy and
2164 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2165 * nohz mode if necessary.
2167 static inline void sched_update_tick_dependency(struct rq *rq)
2169 int cpu = cpu_of(rq);
2171 if (!tick_nohz_full_cpu(cpu))
2174 if (sched_can_stop_tick(rq))
2175 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2177 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2180 static inline int sched_tick_offload_init(void) { return 0; }
2181 static inline void sched_update_tick_dependency(struct rq *rq) { }
2184 static inline void add_nr_running(struct rq *rq, unsigned count)
2186 unsigned prev_nr = rq->nr_running;
2188 rq->nr_running = prev_nr + count;
2189 if (trace_sched_update_nr_running_tp_enabled()) {
2190 call_trace_sched_update_nr_running(rq, count);
2194 if (prev_nr < 2 && rq->nr_running >= 2) {
2195 if (!READ_ONCE(rq->rd->overload))
2196 WRITE_ONCE(rq->rd->overload, 1);
2200 sched_update_tick_dependency(rq);
2203 static inline void sub_nr_running(struct rq *rq, unsigned count)
2205 rq->nr_running -= count;
2206 if (trace_sched_update_nr_running_tp_enabled()) {
2207 call_trace_sched_update_nr_running(rq, -count);
2210 /* Check if we still need preemption */
2211 sched_update_tick_dependency(rq);
2214 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2215 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2217 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2219 extern const_debug unsigned int sysctl_sched_nr_migrate;
2220 extern const_debug unsigned int sysctl_sched_migration_cost;
2222 #ifdef CONFIG_SCHED_HRTICK
2226 * - enabled by features
2227 * - hrtimer is actually high res
2229 static inline int hrtick_enabled(struct rq *rq)
2231 if (!cpu_active(cpu_of(rq)))
2233 return hrtimer_is_hres_active(&rq->hrtick_timer);
2236 static inline int hrtick_enabled_fair(struct rq *rq)
2238 if (!sched_feat(HRTICK))
2240 return hrtick_enabled(rq);
2243 static inline int hrtick_enabled_dl(struct rq *rq)
2245 if (!sched_feat(HRTICK_DL))
2247 return hrtick_enabled(rq);
2250 void hrtick_start(struct rq *rq, u64 delay);
2254 static inline int hrtick_enabled_fair(struct rq *rq)
2259 static inline int hrtick_enabled_dl(struct rq *rq)
2264 static inline int hrtick_enabled(struct rq *rq)
2269 #endif /* CONFIG_SCHED_HRTICK */
2271 #ifndef arch_scale_freq_tick
2272 static __always_inline
2273 void arch_scale_freq_tick(void)
2278 #ifndef arch_scale_freq_capacity
2280 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2281 * @cpu: the CPU in question.
2283 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2286 * ------ * SCHED_CAPACITY_SCALE
2289 static __always_inline
2290 unsigned long arch_scale_freq_capacity(int cpu)
2292 return SCHED_CAPACITY_SCALE;
2299 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2301 #ifdef CONFIG_SCHED_CORE
2303 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2304 * order by core-id first and cpu-id second.
2308 * double_rq_lock(0,3); will take core-0, core-1 lock
2309 * double_rq_lock(1,2); will take core-1, core-0 lock
2311 * when only cpu-id is considered.
2313 if (rq1->core->cpu < rq2->core->cpu)
2315 if (rq1->core->cpu > rq2->core->cpu)
2319 * __sched_core_flip() relies on SMT having cpu-id lock order.
2322 return rq1->cpu < rq2->cpu;
2325 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2327 #ifdef CONFIG_PREEMPTION
2330 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2331 * way at the expense of forcing extra atomic operations in all
2332 * invocations. This assures that the double_lock is acquired using the
2333 * same underlying policy as the spinlock_t on this architecture, which
2334 * reduces latency compared to the unfair variant below. However, it
2335 * also adds more overhead and therefore may reduce throughput.
2337 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2338 __releases(this_rq->lock)
2339 __acquires(busiest->lock)
2340 __acquires(this_rq->lock)
2342 raw_spin_rq_unlock(this_rq);
2343 double_rq_lock(this_rq, busiest);
2350 * Unfair double_lock_balance: Optimizes throughput at the expense of
2351 * latency by eliminating extra atomic operations when the locks are
2352 * already in proper order on entry. This favors lower CPU-ids and will
2353 * grant the double lock to lower CPUs over higher ids under contention,
2354 * regardless of entry order into the function.
2356 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2357 __releases(this_rq->lock)
2358 __acquires(busiest->lock)
2359 __acquires(this_rq->lock)
2361 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2364 if (likely(raw_spin_rq_trylock(busiest)))
2367 if (rq_order_less(this_rq, busiest)) {
2368 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2372 raw_spin_rq_unlock(this_rq);
2373 double_rq_lock(this_rq, busiest);
2378 #endif /* CONFIG_PREEMPTION */
2381 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2383 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2385 lockdep_assert_irqs_disabled();
2387 return _double_lock_balance(this_rq, busiest);
2390 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2391 __releases(busiest->lock)
2393 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2394 raw_spin_rq_unlock(busiest);
2395 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2398 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2404 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2407 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2413 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2416 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2422 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2426 * double_rq_unlock - safely unlock two runqueues
2428 * Note this does not restore interrupts like task_rq_unlock,
2429 * you need to do so manually after calling.
2431 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2432 __releases(rq1->lock)
2433 __releases(rq2->lock)
2435 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2436 raw_spin_rq_unlock(rq2);
2438 __release(rq2->lock);
2439 raw_spin_rq_unlock(rq1);
2442 extern void set_rq_online (struct rq *rq);
2443 extern void set_rq_offline(struct rq *rq);
2444 extern bool sched_smp_initialized;
2446 #else /* CONFIG_SMP */
2449 * double_rq_lock - safely lock two runqueues
2451 * Note this does not disable interrupts like task_rq_lock,
2452 * you need to do so manually before calling.
2454 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2455 __acquires(rq1->lock)
2456 __acquires(rq2->lock)
2458 BUG_ON(!irqs_disabled());
2460 raw_spin_rq_lock(rq1);
2461 __acquire(rq2->lock); /* Fake it out ;) */
2465 * double_rq_unlock - safely unlock two runqueues
2467 * Note this does not restore interrupts like task_rq_unlock,
2468 * you need to do so manually after calling.
2470 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2471 __releases(rq1->lock)
2472 __releases(rq2->lock)
2475 raw_spin_rq_unlock(rq1);
2476 __release(rq2->lock);
2481 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2482 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2484 #ifdef CONFIG_SCHED_DEBUG
2485 extern bool sched_debug_verbose;
2487 extern void print_cfs_stats(struct seq_file *m, int cpu);
2488 extern void print_rt_stats(struct seq_file *m, int cpu);
2489 extern void print_dl_stats(struct seq_file *m, int cpu);
2490 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2491 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2492 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2494 extern void resched_latency_warn(int cpu, u64 latency);
2495 #ifdef CONFIG_NUMA_BALANCING
2497 show_numa_stats(struct task_struct *p, struct seq_file *m);
2499 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2500 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2501 #endif /* CONFIG_NUMA_BALANCING */
2503 static inline void resched_latency_warn(int cpu, u64 latency) {}
2504 #endif /* CONFIG_SCHED_DEBUG */
2506 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2507 extern void init_rt_rq(struct rt_rq *rt_rq);
2508 extern void init_dl_rq(struct dl_rq *dl_rq);
2510 extern void cfs_bandwidth_usage_inc(void);
2511 extern void cfs_bandwidth_usage_dec(void);
2513 #ifdef CONFIG_NO_HZ_COMMON
2514 #define NOHZ_BALANCE_KICK_BIT 0
2515 #define NOHZ_STATS_KICK_BIT 1
2516 #define NOHZ_NEWILB_KICK_BIT 2
2518 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2519 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2520 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2522 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2524 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2526 extern void nohz_balance_exit_idle(struct rq *rq);
2528 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2531 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2532 extern void nohz_run_idle_balance(int cpu);
2534 static inline void nohz_run_idle_balance(int cpu) { }
2539 void __dl_update(struct dl_bw *dl_b, s64 bw)
2541 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2544 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2545 "sched RCU must be held");
2546 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2547 struct rq *rq = cpu_rq(i);
2549 rq->dl.extra_bw += bw;
2554 void __dl_update(struct dl_bw *dl_b, s64 bw)
2556 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2563 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2568 struct u64_stats_sync sync;
2571 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2574 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2575 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2576 * and never move forward.
2578 static inline u64 irq_time_read(int cpu)
2580 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2585 seq = __u64_stats_fetch_begin(&irqtime->sync);
2586 total = irqtime->total;
2587 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2591 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2593 #ifdef CONFIG_CPU_FREQ
2594 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2597 * cpufreq_update_util - Take a note about CPU utilization changes.
2598 * @rq: Runqueue to carry out the update for.
2599 * @flags: Update reason flags.
2601 * This function is called by the scheduler on the CPU whose utilization is
2604 * It can only be called from RCU-sched read-side critical sections.
2606 * The way cpufreq is currently arranged requires it to evaluate the CPU
2607 * performance state (frequency/voltage) on a regular basis to prevent it from
2608 * being stuck in a completely inadequate performance level for too long.
2609 * That is not guaranteed to happen if the updates are only triggered from CFS
2610 * and DL, though, because they may not be coming in if only RT tasks are
2611 * active all the time (or there are RT tasks only).
2613 * As a workaround for that issue, this function is called periodically by the
2614 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2615 * but that really is a band-aid. Going forward it should be replaced with
2616 * solutions targeted more specifically at RT tasks.
2618 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2620 struct update_util_data *data;
2622 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2625 data->func(data, rq_clock(rq), flags);
2628 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2629 #endif /* CONFIG_CPU_FREQ */
2631 #ifdef CONFIG_UCLAMP_TASK
2632 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2635 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2636 * @rq: The rq to clamp against. Must not be NULL.
2637 * @util: The util value to clamp.
2638 * @p: The task to clamp against. Can be NULL if you want to clamp
2641 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2643 * If sched_uclamp_used static key is disabled, then just return the util
2644 * without any clamping since uclamp aggregation at the rq level in the fast
2645 * path is disabled, rendering this operation a NOP.
2647 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2648 * will return the correct effective uclamp value of the task even if the
2649 * static key is disabled.
2651 static __always_inline
2652 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2653 struct task_struct *p)
2655 unsigned long min_util;
2656 unsigned long max_util;
2658 if (!static_branch_likely(&sched_uclamp_used))
2661 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2662 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2665 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2666 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2670 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2671 * RUNNABLE tasks with _different_ clamps, we can end up with an
2672 * inversion. Fix it now when the clamps are applied.
2674 if (unlikely(min_util >= max_util))
2677 return clamp(util, min_util, max_util);
2681 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2682 * by default in the fast path and only gets turned on once userspace performs
2683 * an operation that requires it.
2685 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2688 static inline bool uclamp_is_used(void)
2690 return static_branch_likely(&sched_uclamp_used);
2692 #else /* CONFIG_UCLAMP_TASK */
2694 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2695 struct task_struct *p)
2700 static inline bool uclamp_is_used(void)
2704 #endif /* CONFIG_UCLAMP_TASK */
2706 #ifdef arch_scale_freq_capacity
2707 # ifndef arch_scale_freq_invariant
2708 # define arch_scale_freq_invariant() true
2711 # define arch_scale_freq_invariant() false
2715 static inline unsigned long capacity_orig_of(int cpu)
2717 return cpu_rq(cpu)->cpu_capacity_orig;
2721 * enum cpu_util_type - CPU utilization type
2722 * @FREQUENCY_UTIL: Utilization used to select frequency
2723 * @ENERGY_UTIL: Utilization used during energy calculation
2725 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2726 * need to be aggregated differently depending on the usage made of them. This
2727 * enum is used within effective_cpu_util() to differentiate the types of
2728 * utilization expected by the callers, and adjust the aggregation accordingly.
2730 enum cpu_util_type {
2735 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2736 unsigned long max, enum cpu_util_type type,
2737 struct task_struct *p);
2739 static inline unsigned long cpu_bw_dl(struct rq *rq)
2741 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2744 static inline unsigned long cpu_util_dl(struct rq *rq)
2746 return READ_ONCE(rq->avg_dl.util_avg);
2749 static inline unsigned long cpu_util_cfs(struct rq *rq)
2751 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2753 if (sched_feat(UTIL_EST)) {
2754 util = max_t(unsigned long, util,
2755 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2761 static inline unsigned long cpu_util_rt(struct rq *rq)
2763 return READ_ONCE(rq->avg_rt.util_avg);
2767 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2768 static inline unsigned long cpu_util_irq(struct rq *rq)
2770 return rq->avg_irq.util_avg;
2774 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2776 util *= (max - irq);
2783 static inline unsigned long cpu_util_irq(struct rq *rq)
2789 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2795 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2797 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2799 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2801 static inline bool sched_energy_enabled(void)
2803 return static_branch_unlikely(&sched_energy_present);
2806 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2808 #define perf_domain_span(pd) NULL
2809 static inline bool sched_energy_enabled(void) { return false; }
2811 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2813 #ifdef CONFIG_MEMBARRIER
2815 * The scheduler provides memory barriers required by membarrier between:
2816 * - prior user-space memory accesses and store to rq->membarrier_state,
2817 * - store to rq->membarrier_state and following user-space memory accesses.
2818 * In the same way it provides those guarantees around store to rq->curr.
2820 static inline void membarrier_switch_mm(struct rq *rq,
2821 struct mm_struct *prev_mm,
2822 struct mm_struct *next_mm)
2824 int membarrier_state;
2826 if (prev_mm == next_mm)
2829 membarrier_state = atomic_read(&next_mm->membarrier_state);
2830 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2833 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2836 static inline void membarrier_switch_mm(struct rq *rq,
2837 struct mm_struct *prev_mm,
2838 struct mm_struct *next_mm)
2844 static inline bool is_per_cpu_kthread(struct task_struct *p)
2846 if (!(p->flags & PF_KTHREAD))
2849 if (p->nr_cpus_allowed != 1)
2856 extern void swake_up_all_locked(struct swait_queue_head *q);
2857 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
2859 #ifdef CONFIG_PREEMPT_DYNAMIC
2860 extern int preempt_dynamic_mode;
2861 extern int sched_dynamic_mode(const char *str);
2862 extern void sched_dynamic_update(int mode);