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/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/ratelimit.h>
61 #include <linux/rcupdate_wait.h>
62 #include <linux/security.h>
63 #include <linux/stop_machine.h>
64 #include <linux/suspend.h>
65 #include <linux/swait.h>
66 #include <linux/syscalls.h>
67 #include <linux/task_work.h>
68 #include <linux/tsacct_kern.h>
72 #ifdef CONFIG_PARAVIRT
73 # include <asm/paravirt.h>
77 #include "cpudeadline.h"
79 #include <trace/events/sched.h>
81 #ifdef CONFIG_SCHED_DEBUG
82 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
84 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
90 /* task_struct::on_rq states: */
91 #define TASK_ON_RQ_QUEUED 1
92 #define TASK_ON_RQ_MIGRATING 2
94 extern __read_mostly int scheduler_running;
96 extern unsigned long calc_load_update;
97 extern atomic_long_t calc_load_tasks;
99 extern void calc_global_load_tick(struct rq *this_rq);
100 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
102 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
104 * Helpers for converting nanosecond timing to jiffy resolution
106 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
109 * Increase resolution of nice-level calculations for 64-bit architectures.
110 * The extra resolution improves shares distribution and load balancing of
111 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
112 * hierarchies, especially on larger systems. This is not a user-visible change
113 * and does not change the user-interface for setting shares/weights.
115 * We increase resolution only if we have enough bits to allow this increased
116 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
117 * are pretty high and the returns do not justify the increased costs.
119 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
120 * increase coverage and consistency always enable it on 64-bit platforms.
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load_down(w) \
127 unsigned long __w = (w); \
129 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
133 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
134 # define scale_load(w) (w)
135 # define scale_load_down(w) (w)
139 * Task weight (visible to users) and its load (invisible to users) have
140 * independent resolution, but they should be well calibrated. We use
141 * scale_load() and scale_load_down(w) to convert between them. The
142 * following must be true:
144 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
147 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
150 * Single value that decides SCHED_DEADLINE internal math precision.
151 * 10 -> just above 1us
152 * 9 -> just above 0.5us
157 * Single value that denotes runtime == period, ie unlimited time.
159 #define RUNTIME_INF ((u64)~0ULL)
161 static inline int idle_policy(int policy)
163 return policy == SCHED_IDLE;
165 static inline int fair_policy(int policy)
167 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
170 static inline int rt_policy(int policy)
172 return policy == SCHED_FIFO || policy == SCHED_RR;
175 static inline int dl_policy(int policy)
177 return policy == SCHED_DEADLINE;
179 static inline bool valid_policy(int policy)
181 return idle_policy(policy) || fair_policy(policy) ||
182 rt_policy(policy) || dl_policy(policy);
185 static inline int task_has_idle_policy(struct task_struct *p)
187 return idle_policy(p->policy);
190 static inline int task_has_rt_policy(struct task_struct *p)
192 return rt_policy(p->policy);
195 static inline int task_has_dl_policy(struct task_struct *p)
197 return dl_policy(p->policy);
200 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
202 static inline void update_avg(u64 *avg, u64 sample)
204 s64 diff = sample - *avg;
209 * Shifting a value by an exponent greater *or equal* to the size of said value
210 * is UB; cap at size-1.
212 #define shr_bound(val, shift) \
213 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
216 * !! For sched_setattr_nocheck() (kernel) only !!
218 * This is actually gross. :(
220 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
221 * tasks, but still be able to sleep. We need this on platforms that cannot
222 * atomically change clock frequency. Remove once fast switching will be
223 * available on such platforms.
225 * SUGOV stands for SchedUtil GOVernor.
227 #define SCHED_FLAG_SUGOV 0x10000000
229 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
231 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
233 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
234 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
241 * Tells if entity @a should preempt entity @b.
244 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
246 return dl_entity_is_special(a) ||
247 dl_time_before(a->deadline, b->deadline);
251 * This is the priority-queue data structure of the RT scheduling class:
253 struct rt_prio_array {
254 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
255 struct list_head queue[MAX_RT_PRIO];
258 struct rt_bandwidth {
259 /* nests inside the rq lock: */
260 raw_spinlock_t rt_runtime_lock;
263 struct hrtimer rt_period_timer;
264 unsigned int rt_period_active;
267 void __dl_clear_params(struct task_struct *p);
269 struct dl_bandwidth {
270 raw_spinlock_t dl_runtime_lock;
275 static inline int dl_bandwidth_enabled(void)
277 return sysctl_sched_rt_runtime >= 0;
281 * To keep the bandwidth of -deadline tasks under control
282 * we need some place where:
283 * - store the maximum -deadline bandwidth of each cpu;
284 * - cache the fraction of bandwidth that is currently allocated in
287 * This is all done in the data structure below. It is similar to the
288 * one used for RT-throttling (rt_bandwidth), with the main difference
289 * that, since here we are only interested in admission control, we
290 * do not decrease any runtime while the group "executes", neither we
291 * need a timer to replenish it.
293 * With respect to SMP, bandwidth is given on a per root domain basis,
295 * - bw (< 100%) is the deadline bandwidth of each CPU;
296 * - total_bw is the currently allocated bandwidth in each root domain;
304 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
307 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
309 dl_b->total_bw -= tsk_bw;
310 __dl_update(dl_b, (s32)tsk_bw / cpus);
314 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
316 dl_b->total_bw += tsk_bw;
317 __dl_update(dl_b, -((s32)tsk_bw / cpus));
320 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
321 u64 old_bw, u64 new_bw)
323 return dl_b->bw != -1 &&
324 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
328 * Verify the fitness of task @p to run on @cpu taking into account the
329 * CPU original capacity and the runtime/deadline ratio of the task.
331 * The function will return true if the CPU original capacity of the
332 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
333 * task and false otherwise.
335 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
337 unsigned long cap = arch_scale_cpu_capacity(cpu);
339 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
342 extern void init_dl_bw(struct dl_bw *dl_b);
343 extern int sched_dl_global_validate(void);
344 extern void sched_dl_do_global(void);
345 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
346 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
347 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
348 extern bool __checkparam_dl(const struct sched_attr *attr);
349 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
350 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
351 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
352 extern bool dl_cpu_busy(unsigned int cpu);
354 #ifdef CONFIG_CGROUP_SCHED
356 #include <linux/cgroup.h>
357 #include <linux/psi.h>
362 extern struct list_head task_groups;
364 struct cfs_bandwidth {
365 #ifdef CONFIG_CFS_BANDWIDTH
372 s64 hierarchical_quota;
377 struct hrtimer period_timer;
378 struct hrtimer slack_timer;
379 struct list_head throttled_cfs_rq;
390 /* Task group related information */
392 struct cgroup_subsys_state css;
394 #ifdef CONFIG_FAIR_GROUP_SCHED
395 /* schedulable entities of this group on each CPU */
396 struct sched_entity **se;
397 /* runqueue "owned" by this group on each CPU */
398 struct cfs_rq **cfs_rq;
399 unsigned long shares;
401 /* A positive value indicates that this is a SCHED_IDLE group. */
406 * load_avg can be heavily contended at clock tick time, so put
407 * it in its own cacheline separated from the fields above which
408 * will also be accessed at each tick.
410 atomic_long_t load_avg ____cacheline_aligned;
414 #ifdef CONFIG_RT_GROUP_SCHED
415 struct sched_rt_entity **rt_se;
416 struct rt_rq **rt_rq;
418 struct rt_bandwidth rt_bandwidth;
422 struct list_head list;
424 struct task_group *parent;
425 struct list_head siblings;
426 struct list_head children;
428 #ifdef CONFIG_SCHED_AUTOGROUP
429 struct autogroup *autogroup;
432 struct cfs_bandwidth cfs_bandwidth;
434 #ifdef CONFIG_UCLAMP_TASK_GROUP
435 /* The two decimal precision [%] value requested from user-space */
436 unsigned int uclamp_pct[UCLAMP_CNT];
437 /* Clamp values requested for a task group */
438 struct uclamp_se uclamp_req[UCLAMP_CNT];
439 /* Effective clamp values used for a task group */
440 struct uclamp_se uclamp[UCLAMP_CNT];
445 #ifdef CONFIG_FAIR_GROUP_SCHED
446 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
449 * A weight of 0 or 1 can cause arithmetics problems.
450 * A weight of a cfs_rq is the sum of weights of which entities
451 * are queued on this cfs_rq, so a weight of a entity should not be
452 * too large, so as the shares value of a task group.
453 * (The default weight is 1024 - so there's no practical
454 * limitation from this.)
456 #define MIN_SHARES (1UL << 1)
457 #define MAX_SHARES (1UL << 18)
460 typedef int (*tg_visitor)(struct task_group *, void *);
462 extern int walk_tg_tree_from(struct task_group *from,
463 tg_visitor down, tg_visitor up, void *data);
466 * Iterate the full tree, calling @down when first entering a node and @up when
467 * leaving it for the final time.
469 * Caller must hold rcu_lock or sufficient equivalent.
471 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
473 return walk_tg_tree_from(&root_task_group, down, up, data);
476 extern int tg_nop(struct task_group *tg, void *data);
478 extern void free_fair_sched_group(struct task_group *tg);
479 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
480 extern void online_fair_sched_group(struct task_group *tg);
481 extern void unregister_fair_sched_group(struct task_group *tg);
482 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
483 struct sched_entity *se, int cpu,
484 struct sched_entity *parent);
485 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
487 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
488 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
489 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
491 extern void unregister_rt_sched_group(struct task_group *tg);
492 extern void free_rt_sched_group(struct task_group *tg);
493 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
494 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
495 struct sched_rt_entity *rt_se, int cpu,
496 struct sched_rt_entity *parent);
497 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
498 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
499 extern long sched_group_rt_runtime(struct task_group *tg);
500 extern long sched_group_rt_period(struct task_group *tg);
501 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
503 extern struct task_group *sched_create_group(struct task_group *parent);
504 extern void sched_online_group(struct task_group *tg,
505 struct task_group *parent);
506 extern void sched_destroy_group(struct task_group *tg);
507 extern void sched_release_group(struct task_group *tg);
509 extern void sched_move_task(struct task_struct *tsk);
511 #ifdef CONFIG_FAIR_GROUP_SCHED
512 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
514 extern int sched_group_set_idle(struct task_group *tg, long idle);
517 extern void set_task_rq_fair(struct sched_entity *se,
518 struct cfs_rq *prev, struct cfs_rq *next);
519 #else /* !CONFIG_SMP */
520 static inline void set_task_rq_fair(struct sched_entity *se,
521 struct cfs_rq *prev, struct cfs_rq *next) { }
522 #endif /* CONFIG_SMP */
523 #endif /* CONFIG_FAIR_GROUP_SCHED */
525 #else /* CONFIG_CGROUP_SCHED */
527 struct cfs_bandwidth { };
529 #endif /* CONFIG_CGROUP_SCHED */
531 /* CFS-related fields in a runqueue */
533 struct load_weight load;
534 unsigned int nr_running;
535 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
536 unsigned int idle_nr_running; /* SCHED_IDLE */
537 unsigned int idle_h_nr_running; /* SCHED_IDLE */
541 #ifdef CONFIG_SCHED_CORE
542 unsigned int forceidle_seq;
547 u64 min_vruntime_copy;
550 struct rb_root_cached tasks_timeline;
553 * 'curr' points to currently running entity on this cfs_rq.
554 * It is set to NULL otherwise (i.e when none are currently running).
556 struct sched_entity *curr;
557 struct sched_entity *next;
558 struct sched_entity *last;
559 struct sched_entity *skip;
561 #ifdef CONFIG_SCHED_DEBUG
562 unsigned int nr_spread_over;
569 struct sched_avg avg;
571 u64 load_last_update_time_copy;
574 raw_spinlock_t lock ____cacheline_aligned;
576 unsigned long load_avg;
577 unsigned long util_avg;
578 unsigned long runnable_avg;
581 #ifdef CONFIG_FAIR_GROUP_SCHED
582 unsigned long tg_load_avg_contrib;
584 long prop_runnable_sum;
587 * h_load = weight * f(tg)
589 * Where f(tg) is the recursive weight fraction assigned to
592 unsigned long h_load;
593 u64 last_h_load_update;
594 struct sched_entity *h_load_next;
595 #endif /* CONFIG_FAIR_GROUP_SCHED */
596 #endif /* CONFIG_SMP */
598 #ifdef CONFIG_FAIR_GROUP_SCHED
599 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
602 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
603 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
604 * (like users, containers etc.)
606 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
607 * This list is used during load balance.
610 struct list_head leaf_cfs_rq_list;
611 struct task_group *tg; /* group that "owns" this runqueue */
613 /* Locally cached copy of our task_group's idle value */
616 #ifdef CONFIG_CFS_BANDWIDTH
618 s64 runtime_remaining;
621 u64 throttled_clock_task;
622 u64 throttled_clock_task_time;
625 struct list_head throttled_list;
626 #endif /* CONFIG_CFS_BANDWIDTH */
627 #endif /* CONFIG_FAIR_GROUP_SCHED */
630 static inline int rt_bandwidth_enabled(void)
632 return sysctl_sched_rt_runtime >= 0;
635 /* RT IPI pull logic requires IRQ_WORK */
636 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
637 # define HAVE_RT_PUSH_IPI
640 /* Real-Time classes' related field in a runqueue: */
642 struct rt_prio_array active;
643 unsigned int rt_nr_running;
644 unsigned int rr_nr_running;
645 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
647 int curr; /* highest queued rt task prio */
649 int next; /* next highest */
654 unsigned int rt_nr_migratory;
655 unsigned int rt_nr_total;
657 struct plist_head pushable_tasks;
659 #endif /* CONFIG_SMP */
665 /* Nests inside the rq lock: */
666 raw_spinlock_t rt_runtime_lock;
668 #ifdef CONFIG_RT_GROUP_SCHED
669 unsigned int rt_nr_boosted;
672 struct task_group *tg;
676 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
678 return rt_rq->rt_queued && rt_rq->rt_nr_running;
681 /* Deadline class' related fields in a runqueue */
683 /* runqueue is an rbtree, ordered by deadline */
684 struct rb_root_cached root;
686 unsigned int dl_nr_running;
690 * Deadline values of the currently executing and the
691 * earliest ready task on this rq. Caching these facilitates
692 * the decision whether or not a ready but not running task
693 * should migrate somewhere else.
700 unsigned int dl_nr_migratory;
704 * Tasks on this rq that can be pushed away. They are kept in
705 * an rb-tree, ordered by tasks' deadlines, with caching
706 * of the leftmost (earliest deadline) element.
708 struct rb_root_cached pushable_dl_tasks_root;
713 * "Active utilization" for this runqueue: increased when a
714 * task wakes up (becomes TASK_RUNNING) and decreased when a
720 * Utilization of the tasks "assigned" to this runqueue (including
721 * the tasks that are in runqueue and the tasks that executed on this
722 * CPU and blocked). Increased when a task moves to this runqueue, and
723 * decreased when the task moves away (migrates, changes scheduling
724 * policy, or terminates).
725 * This is needed to compute the "inactive utilization" for the
726 * runqueue (inactive utilization = this_bw - running_bw).
732 * Inverse of the fraction of CPU utilization that can be reclaimed
733 * by the GRUB algorithm.
738 #ifdef CONFIG_FAIR_GROUP_SCHED
739 /* An entity is a task if it doesn't "own" a runqueue */
740 #define entity_is_task(se) (!se->my_q)
742 static inline void se_update_runnable(struct sched_entity *se)
744 if (!entity_is_task(se))
745 se->runnable_weight = se->my_q->h_nr_running;
748 static inline long se_runnable(struct sched_entity *se)
750 if (entity_is_task(se))
753 return se->runnable_weight;
757 #define entity_is_task(se) 1
759 static inline void se_update_runnable(struct sched_entity *se) {}
761 static inline long se_runnable(struct sched_entity *se)
769 * XXX we want to get rid of these helpers and use the full load resolution.
771 static inline long se_weight(struct sched_entity *se)
773 return scale_load_down(se->load.weight);
777 static inline bool sched_asym_prefer(int a, int b)
779 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
783 struct em_perf_domain *em_pd;
784 struct perf_domain *next;
788 /* Scheduling group status flags */
789 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
790 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
793 * We add the notion of a root-domain which will be used to define per-domain
794 * variables. Each exclusive cpuset essentially defines an island domain by
795 * fully partitioning the member CPUs from any other cpuset. Whenever a new
796 * exclusive cpuset is created, we also create and attach a new root-domain
805 cpumask_var_t online;
808 * Indicate pullable load on at least one CPU, e.g:
809 * - More than one runnable task
810 * - Running task is misfit
814 /* Indicate one or more cpus over-utilized (tipping point) */
818 * The bit corresponding to a CPU gets set here if such CPU has more
819 * than one runnable -deadline task (as it is below for RT tasks).
821 cpumask_var_t dlo_mask;
827 * Indicate whether a root_domain's dl_bw has been checked or
828 * updated. It's monotonously increasing value.
830 * Also, some corner cases, like 'wrap around' is dangerous, but given
831 * that u64 is 'big enough'. So that shouldn't be a concern.
835 #ifdef HAVE_RT_PUSH_IPI
837 * For IPI pull requests, loop across the rto_mask.
839 struct irq_work rto_push_work;
840 raw_spinlock_t rto_lock;
841 /* These are only updated and read within rto_lock */
844 /* These atomics are updated outside of a lock */
845 atomic_t rto_loop_next;
846 atomic_t rto_loop_start;
849 * The "RT overload" flag: it gets set if a CPU has more than
850 * one runnable RT task.
852 cpumask_var_t rto_mask;
853 struct cpupri cpupri;
855 unsigned long max_cpu_capacity;
858 * NULL-terminated list of performance domains intersecting with the
859 * CPUs of the rd. Protected by RCU.
861 struct perf_domain __rcu *pd;
864 extern void init_defrootdomain(void);
865 extern int sched_init_domains(const struct cpumask *cpu_map);
866 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
867 extern void sched_get_rd(struct root_domain *rd);
868 extern void sched_put_rd(struct root_domain *rd);
870 #ifdef HAVE_RT_PUSH_IPI
871 extern void rto_push_irq_work_func(struct irq_work *work);
873 #endif /* CONFIG_SMP */
875 #ifdef CONFIG_UCLAMP_TASK
877 * struct uclamp_bucket - Utilization clamp bucket
878 * @value: utilization clamp value for tasks on this clamp bucket
879 * @tasks: number of RUNNABLE tasks on this clamp bucket
881 * Keep track of how many tasks are RUNNABLE for a given utilization
884 struct uclamp_bucket {
885 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
886 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
890 * struct uclamp_rq - rq's utilization clamp
891 * @value: currently active clamp values for a rq
892 * @bucket: utilization clamp buckets affecting a rq
894 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
895 * A clamp value is affecting a rq when there is at least one task RUNNABLE
896 * (or actually running) with that value.
898 * There are up to UCLAMP_CNT possible different clamp values, currently there
899 * are only two: minimum utilization and maximum utilization.
901 * All utilization clamping values are MAX aggregated, since:
902 * - for util_min: we want to run the CPU at least at the max of the minimum
903 * utilization required by its currently RUNNABLE tasks.
904 * - for util_max: we want to allow the CPU to run up to the max of the
905 * maximum utilization allowed by its currently RUNNABLE tasks.
907 * Since on each system we expect only a limited number of different
908 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
909 * the metrics required to compute all the per-rq utilization clamp values.
913 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
916 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
917 #endif /* CONFIG_UCLAMP_TASK */
920 * This is the main, per-CPU runqueue data structure.
922 * Locking rule: those places that want to lock multiple runqueues
923 * (such as the load balancing or the thread migration code), lock
924 * acquire operations must be ordered by ascending &runqueue.
928 raw_spinlock_t __lock;
931 * nr_running and cpu_load should be in the same cacheline because
932 * remote CPUs use both these fields when doing load calculation.
934 unsigned int nr_running;
935 #ifdef CONFIG_NUMA_BALANCING
936 unsigned int nr_numa_running;
937 unsigned int nr_preferred_running;
938 unsigned int numa_migrate_on;
940 #ifdef CONFIG_NO_HZ_COMMON
942 unsigned long last_blocked_load_update_tick;
943 unsigned int has_blocked_load;
944 call_single_data_t nohz_csd;
945 #endif /* CONFIG_SMP */
946 unsigned int nohz_tick_stopped;
948 #endif /* CONFIG_NO_HZ_COMMON */
951 unsigned int ttwu_pending;
955 #ifdef CONFIG_UCLAMP_TASK
956 /* Utilization clamp values based on CPU's RUNNABLE tasks */
957 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
958 unsigned int uclamp_flags;
959 #define UCLAMP_FLAG_IDLE 0x01
966 #ifdef CONFIG_FAIR_GROUP_SCHED
967 /* list of leaf cfs_rq on this CPU: */
968 struct list_head leaf_cfs_rq_list;
969 struct list_head *tmp_alone_branch;
970 #endif /* CONFIG_FAIR_GROUP_SCHED */
973 * This is part of a global counter where only the total sum
974 * over all CPUs matters. A task can increase this counter on
975 * one CPU and if it got migrated afterwards it may decrease
976 * it on another CPU. Always updated under the runqueue lock:
978 unsigned int nr_uninterruptible;
980 struct task_struct __rcu *curr;
981 struct task_struct *idle;
982 struct task_struct *stop;
983 unsigned long next_balance;
984 struct mm_struct *prev_mm;
986 unsigned int clock_update_flags;
988 /* Ensure that all clocks are in the same cache line */
989 u64 clock_task ____cacheline_aligned;
991 unsigned long lost_idle_time;
995 #ifdef CONFIG_SCHED_DEBUG
996 u64 last_seen_need_resched_ns;
997 int ticks_without_resched;
1000 #ifdef CONFIG_MEMBARRIER
1001 int membarrier_state;
1005 struct root_domain *rd;
1006 struct sched_domain __rcu *sd;
1008 unsigned long cpu_capacity;
1009 unsigned long cpu_capacity_orig;
1011 struct callback_head *balance_callback;
1013 unsigned char nohz_idle_balance;
1014 unsigned char idle_balance;
1016 unsigned long misfit_task_load;
1018 /* For active balancing */
1021 struct cpu_stop_work active_balance_work;
1023 /* CPU of this runqueue: */
1027 struct list_head cfs_tasks;
1029 struct sched_avg avg_rt;
1030 struct sched_avg avg_dl;
1031 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1032 struct sched_avg avg_irq;
1034 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1035 struct sched_avg avg_thermal;
1040 unsigned long wake_stamp;
1043 /* This is used to determine avg_idle's max value */
1044 u64 max_idle_balance_cost;
1046 #ifdef CONFIG_HOTPLUG_CPU
1047 struct rcuwait hotplug_wait;
1049 #endif /* CONFIG_SMP */
1051 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1054 #ifdef CONFIG_PARAVIRT
1055 u64 prev_steal_time;
1057 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1058 u64 prev_steal_time_rq;
1061 /* calc_load related fields */
1062 unsigned long calc_load_update;
1063 long calc_load_active;
1065 #ifdef CONFIG_SCHED_HRTICK
1067 call_single_data_t hrtick_csd;
1069 struct hrtimer hrtick_timer;
1070 ktime_t hrtick_time;
1073 #ifdef CONFIG_SCHEDSTATS
1075 struct sched_info rq_sched_info;
1076 unsigned long long rq_cpu_time;
1077 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1079 /* sys_sched_yield() stats */
1080 unsigned int yld_count;
1082 /* schedule() stats */
1083 unsigned int sched_count;
1084 unsigned int sched_goidle;
1086 /* try_to_wake_up() stats */
1087 unsigned int ttwu_count;
1088 unsigned int ttwu_local;
1091 #ifdef CONFIG_CPU_IDLE
1092 /* Must be inspected within a rcu lock section */
1093 struct cpuidle_state *idle_state;
1097 unsigned int nr_pinned;
1099 unsigned int push_busy;
1100 struct cpu_stop_work push_work;
1102 #ifdef CONFIG_SCHED_CORE
1105 struct task_struct *core_pick;
1106 unsigned int core_enabled;
1107 unsigned int core_sched_seq;
1108 struct rb_root core_tree;
1110 /* shared state -- careful with sched_core_cpu_deactivate() */
1111 unsigned int core_task_seq;
1112 unsigned int core_pick_seq;
1113 unsigned long core_cookie;
1114 unsigned int core_forceidle_count;
1115 unsigned int core_forceidle_seq;
1116 unsigned int core_forceidle_occupation;
1117 u64 core_forceidle_start;
1121 #ifdef CONFIG_FAIR_GROUP_SCHED
1123 /* CPU runqueue to which this cfs_rq is attached */
1124 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1131 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1133 return container_of(cfs_rq, struct rq, cfs);
1137 static inline int cpu_of(struct rq *rq)
1146 #define MDF_PUSH 0x01
1148 static inline bool is_migration_disabled(struct task_struct *p)
1151 return p->migration_disabled;
1158 #ifdef CONFIG_SCHED_CORE
1159 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1161 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1163 static inline bool sched_core_enabled(struct rq *rq)
1165 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1168 static inline bool sched_core_disabled(void)
1170 return !static_branch_unlikely(&__sched_core_enabled);
1174 * Be careful with this function; not for general use. The return value isn't
1175 * stable unless you actually hold a relevant rq->__lock.
1177 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1179 if (sched_core_enabled(rq))
1180 return &rq->core->__lock;
1185 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1187 if (rq->core_enabled)
1188 return &rq->core->__lock;
1193 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1196 * Helpers to check if the CPU's core cookie matches with the task's cookie
1197 * when core scheduling is enabled.
1198 * A special case is that the task's cookie always matches with CPU's core
1199 * cookie if the CPU is in an idle core.
1201 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1203 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1204 if (!sched_core_enabled(rq))
1207 return rq->core->core_cookie == p->core_cookie;
1210 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1212 bool idle_core = true;
1215 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1216 if (!sched_core_enabled(rq))
1219 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1220 if (!available_idle_cpu(cpu)) {
1227 * A CPU in an idle core is always the best choice for tasks with
1230 return idle_core || rq->core->core_cookie == p->core_cookie;
1233 static inline bool sched_group_cookie_match(struct rq *rq,
1234 struct task_struct *p,
1235 struct sched_group *group)
1239 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1240 if (!sched_core_enabled(rq))
1243 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1244 if (sched_core_cookie_match(rq, p))
1250 extern void queue_core_balance(struct rq *rq);
1252 static inline bool sched_core_enqueued(struct task_struct *p)
1254 return !RB_EMPTY_NODE(&p->core_node);
1257 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1258 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1260 extern void sched_core_get(void);
1261 extern void sched_core_put(void);
1263 #else /* !CONFIG_SCHED_CORE */
1265 static inline bool sched_core_enabled(struct rq *rq)
1270 static inline bool sched_core_disabled(void)
1275 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1280 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1285 static inline void queue_core_balance(struct rq *rq)
1289 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1294 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1299 static inline bool sched_group_cookie_match(struct rq *rq,
1300 struct task_struct *p,
1301 struct sched_group *group)
1305 #endif /* CONFIG_SCHED_CORE */
1307 static inline void lockdep_assert_rq_held(struct rq *rq)
1309 lockdep_assert_held(__rq_lockp(rq));
1312 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1313 extern bool raw_spin_rq_trylock(struct rq *rq);
1314 extern void raw_spin_rq_unlock(struct rq *rq);
1316 static inline void raw_spin_rq_lock(struct rq *rq)
1318 raw_spin_rq_lock_nested(rq, 0);
1321 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1323 local_irq_disable();
1324 raw_spin_rq_lock(rq);
1327 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1329 raw_spin_rq_unlock(rq);
1333 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1335 unsigned long flags;
1336 local_irq_save(flags);
1337 raw_spin_rq_lock(rq);
1341 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1343 raw_spin_rq_unlock(rq);
1344 local_irq_restore(flags);
1347 #define raw_spin_rq_lock_irqsave(rq, flags) \
1349 flags = _raw_spin_rq_lock_irqsave(rq); \
1352 #ifdef CONFIG_SCHED_SMT
1353 extern void __update_idle_core(struct rq *rq);
1355 static inline void update_idle_core(struct rq *rq)
1357 if (static_branch_unlikely(&sched_smt_present))
1358 __update_idle_core(rq);
1362 static inline void update_idle_core(struct rq *rq) { }
1365 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1367 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1368 #define this_rq() this_cpu_ptr(&runqueues)
1369 #define task_rq(p) cpu_rq(task_cpu(p))
1370 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1371 #define raw_rq() raw_cpu_ptr(&runqueues)
1373 #ifdef CONFIG_FAIR_GROUP_SCHED
1374 static inline struct task_struct *task_of(struct sched_entity *se)
1376 SCHED_WARN_ON(!entity_is_task(se));
1377 return container_of(se, struct task_struct, se);
1380 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1382 return p->se.cfs_rq;
1385 /* runqueue on which this entity is (to be) queued */
1386 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1391 /* runqueue "owned" by this group */
1392 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1399 static inline struct task_struct *task_of(struct sched_entity *se)
1401 return container_of(se, struct task_struct, se);
1404 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1406 return &task_rq(p)->cfs;
1409 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1411 struct task_struct *p = task_of(se);
1412 struct rq *rq = task_rq(p);
1417 /* runqueue "owned" by this group */
1418 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1424 extern void update_rq_clock(struct rq *rq);
1427 * rq::clock_update_flags bits
1429 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1430 * call to __schedule(). This is an optimisation to avoid
1431 * neighbouring rq clock updates.
1433 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1434 * in effect and calls to update_rq_clock() are being ignored.
1436 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1437 * made to update_rq_clock() since the last time rq::lock was pinned.
1439 * If inside of __schedule(), clock_update_flags will have been
1440 * shifted left (a left shift is a cheap operation for the fast path
1441 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1443 * if (rq-clock_update_flags >= RQCF_UPDATED)
1445 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1446 * one position though, because the next rq_unpin_lock() will shift it
1449 #define RQCF_REQ_SKIP 0x01
1450 #define RQCF_ACT_SKIP 0x02
1451 #define RQCF_UPDATED 0x04
1453 static inline void assert_clock_updated(struct rq *rq)
1456 * The only reason for not seeing a clock update since the
1457 * last rq_pin_lock() is if we're currently skipping updates.
1459 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1462 static inline u64 rq_clock(struct rq *rq)
1464 lockdep_assert_rq_held(rq);
1465 assert_clock_updated(rq);
1470 static inline u64 rq_clock_task(struct rq *rq)
1472 lockdep_assert_rq_held(rq);
1473 assert_clock_updated(rq);
1475 return rq->clock_task;
1479 * By default the decay is the default pelt decay period.
1480 * The decay shift can change the decay period in
1482 * Decay shift Decay period(ms)
1489 extern int sched_thermal_decay_shift;
1491 static inline u64 rq_clock_thermal(struct rq *rq)
1493 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1496 static inline void rq_clock_skip_update(struct rq *rq)
1498 lockdep_assert_rq_held(rq);
1499 rq->clock_update_flags |= RQCF_REQ_SKIP;
1503 * See rt task throttling, which is the only time a skip
1504 * request is canceled.
1506 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1508 lockdep_assert_rq_held(rq);
1509 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1513 unsigned long flags;
1514 struct pin_cookie cookie;
1515 #ifdef CONFIG_SCHED_DEBUG
1517 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1518 * current pin context is stashed here in case it needs to be
1519 * restored in rq_repin_lock().
1521 unsigned int clock_update_flags;
1525 extern struct callback_head balance_push_callback;
1528 * Lockdep annotation that avoids accidental unlocks; it's like a
1529 * sticky/continuous lockdep_assert_held().
1531 * This avoids code that has access to 'struct rq *rq' (basically everything in
1532 * the scheduler) from accidentally unlocking the rq if they do not also have a
1533 * copy of the (on-stack) 'struct rq_flags rf'.
1535 * Also see Documentation/locking/lockdep-design.rst.
1537 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1539 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1541 #ifdef CONFIG_SCHED_DEBUG
1542 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1543 rf->clock_update_flags = 0;
1545 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1550 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1552 #ifdef CONFIG_SCHED_DEBUG
1553 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1554 rf->clock_update_flags = RQCF_UPDATED;
1557 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1560 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1562 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1564 #ifdef CONFIG_SCHED_DEBUG
1566 * Restore the value we stashed in @rf for this pin context.
1568 rq->clock_update_flags |= rf->clock_update_flags;
1572 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1573 __acquires(rq->lock);
1575 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1576 __acquires(p->pi_lock)
1577 __acquires(rq->lock);
1579 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1580 __releases(rq->lock)
1582 rq_unpin_lock(rq, rf);
1583 raw_spin_rq_unlock(rq);
1587 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1588 __releases(rq->lock)
1589 __releases(p->pi_lock)
1591 rq_unpin_lock(rq, rf);
1592 raw_spin_rq_unlock(rq);
1593 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1597 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1598 __acquires(rq->lock)
1600 raw_spin_rq_lock_irqsave(rq, rf->flags);
1601 rq_pin_lock(rq, rf);
1605 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1606 __acquires(rq->lock)
1608 raw_spin_rq_lock_irq(rq);
1609 rq_pin_lock(rq, rf);
1613 rq_lock(struct rq *rq, struct rq_flags *rf)
1614 __acquires(rq->lock)
1616 raw_spin_rq_lock(rq);
1617 rq_pin_lock(rq, rf);
1621 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1622 __releases(rq->lock)
1624 rq_unpin_lock(rq, rf);
1625 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1629 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1630 __releases(rq->lock)
1632 rq_unpin_lock(rq, rf);
1633 raw_spin_rq_unlock_irq(rq);
1637 rq_unlock(struct rq *rq, struct rq_flags *rf)
1638 __releases(rq->lock)
1640 rq_unpin_lock(rq, rf);
1641 raw_spin_rq_unlock(rq);
1644 static inline struct rq *
1645 this_rq_lock_irq(struct rq_flags *rf)
1646 __acquires(rq->lock)
1650 local_irq_disable();
1657 enum numa_topology_type {
1662 extern enum numa_topology_type sched_numa_topology_type;
1663 extern int sched_max_numa_distance;
1664 extern bool find_numa_distance(int distance);
1665 extern void sched_init_numa(void);
1666 extern void sched_domains_numa_masks_set(unsigned int cpu);
1667 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1668 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1670 static inline void sched_init_numa(void) { }
1671 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1672 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1673 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1679 #ifdef CONFIG_NUMA_BALANCING
1680 /* The regions in numa_faults array from task_struct */
1681 enum numa_faults_stats {
1687 extern void sched_setnuma(struct task_struct *p, int node);
1688 extern int migrate_task_to(struct task_struct *p, int cpu);
1689 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1691 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1694 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1697 #endif /* CONFIG_NUMA_BALANCING */
1702 queue_balance_callback(struct rq *rq,
1703 struct callback_head *head,
1704 void (*func)(struct rq *rq))
1706 lockdep_assert_rq_held(rq);
1708 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1711 head->func = (void (*)(struct callback_head *))func;
1712 head->next = rq->balance_callback;
1713 rq->balance_callback = head;
1716 #define rcu_dereference_check_sched_domain(p) \
1717 rcu_dereference_check((p), \
1718 lockdep_is_held(&sched_domains_mutex))
1721 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1722 * See destroy_sched_domains: call_rcu for details.
1724 * The domain tree of any CPU may only be accessed from within
1725 * preempt-disabled sections.
1727 #define for_each_domain(cpu, __sd) \
1728 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1729 __sd; __sd = __sd->parent)
1732 * highest_flag_domain - Return highest sched_domain containing flag.
1733 * @cpu: The CPU whose highest level of sched domain is to
1735 * @flag: The flag to check for the highest sched_domain
1736 * for the given CPU.
1738 * Returns the highest sched_domain of a CPU which contains the given flag.
1740 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1742 struct sched_domain *sd, *hsd = NULL;
1744 for_each_domain(cpu, sd) {
1745 if (!(sd->flags & flag))
1753 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1755 struct sched_domain *sd;
1757 for_each_domain(cpu, sd) {
1758 if (sd->flags & flag)
1765 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1766 DECLARE_PER_CPU(int, sd_llc_size);
1767 DECLARE_PER_CPU(int, sd_llc_id);
1768 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1769 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1770 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1771 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1772 extern struct static_key_false sched_asym_cpucapacity;
1774 struct sched_group_capacity {
1777 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1780 unsigned long capacity;
1781 unsigned long min_capacity; /* Min per-CPU capacity in group */
1782 unsigned long max_capacity; /* Max per-CPU capacity in group */
1783 unsigned long next_update;
1784 int imbalance; /* XXX unrelated to capacity but shared group state */
1786 #ifdef CONFIG_SCHED_DEBUG
1790 unsigned long cpumask[]; /* Balance mask */
1793 struct sched_group {
1794 struct sched_group *next; /* Must be a circular list */
1797 unsigned int group_weight;
1798 struct sched_group_capacity *sgc;
1799 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 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1861 extern void __sched_core_account_forceidle(struct rq *rq);
1863 static inline void sched_core_account_forceidle(struct rq *rq)
1865 if (schedstat_enabled())
1866 __sched_core_account_forceidle(rq);
1869 extern void __sched_core_tick(struct rq *rq);
1871 static inline void sched_core_tick(struct rq *rq)
1873 if (sched_core_enabled(rq) && schedstat_enabled())
1874 __sched_core_tick(rq);
1879 static inline void sched_core_account_forceidle(struct rq *rq) {}
1881 static inline void sched_core_tick(struct rq *rq) {}
1883 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1885 #ifdef CONFIG_CGROUP_SCHED
1888 * Return the group to which this tasks belongs.
1890 * We cannot use task_css() and friends because the cgroup subsystem
1891 * changes that value before the cgroup_subsys::attach() method is called,
1892 * therefore we cannot pin it and might observe the wrong value.
1894 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1895 * core changes this before calling sched_move_task().
1897 * Instead we use a 'copy' which is updated from sched_move_task() while
1898 * holding both task_struct::pi_lock and rq::lock.
1900 static inline struct task_group *task_group(struct task_struct *p)
1902 return p->sched_task_group;
1905 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1906 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1908 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1909 struct task_group *tg = task_group(p);
1912 #ifdef CONFIG_FAIR_GROUP_SCHED
1913 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1914 p->se.cfs_rq = tg->cfs_rq[cpu];
1915 p->se.parent = tg->se[cpu];
1918 #ifdef CONFIG_RT_GROUP_SCHED
1919 p->rt.rt_rq = tg->rt_rq[cpu];
1920 p->rt.parent = tg->rt_se[cpu];
1924 #else /* CONFIG_CGROUP_SCHED */
1926 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1927 static inline struct task_group *task_group(struct task_struct *p)
1932 #endif /* CONFIG_CGROUP_SCHED */
1934 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1936 set_task_rq(p, cpu);
1939 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1940 * successfully executed on another CPU. We must ensure that updates of
1941 * per-task data have been completed by this moment.
1944 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1950 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1952 #ifdef CONFIG_SCHED_DEBUG
1953 # include <linux/static_key.h>
1954 # define const_debug __read_mostly
1956 # define const_debug const
1959 #define SCHED_FEAT(name, enabled) \
1960 __SCHED_FEAT_##name ,
1963 #include "features.h"
1969 #ifdef CONFIG_SCHED_DEBUG
1972 * To support run-time toggling of sched features, all the translation units
1973 * (but core.c) reference the sysctl_sched_features defined in core.c.
1975 extern const_debug unsigned int sysctl_sched_features;
1977 #ifdef CONFIG_JUMP_LABEL
1978 #define SCHED_FEAT(name, enabled) \
1979 static __always_inline bool static_branch_##name(struct static_key *key) \
1981 return static_key_##enabled(key); \
1984 #include "features.h"
1987 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1988 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1990 #else /* !CONFIG_JUMP_LABEL */
1992 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1994 #endif /* CONFIG_JUMP_LABEL */
1996 #else /* !SCHED_DEBUG */
1999 * Each translation unit has its own copy of sysctl_sched_features to allow
2000 * constants propagation at compile time and compiler optimization based on
2003 #define SCHED_FEAT(name, enabled) \
2004 (1UL << __SCHED_FEAT_##name) * enabled |
2005 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2006 #include "features.h"
2010 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2012 #endif /* SCHED_DEBUG */
2014 extern struct static_key_false sched_numa_balancing;
2015 extern struct static_key_false sched_schedstats;
2017 static inline u64 global_rt_period(void)
2019 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2022 static inline u64 global_rt_runtime(void)
2024 if (sysctl_sched_rt_runtime < 0)
2027 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2030 static inline int task_current(struct rq *rq, struct task_struct *p)
2032 return rq->curr == p;
2035 static inline int task_running(struct rq *rq, struct task_struct *p)
2040 return task_current(rq, p);
2044 static inline int task_on_rq_queued(struct task_struct *p)
2046 return p->on_rq == TASK_ON_RQ_QUEUED;
2049 static inline int task_on_rq_migrating(struct task_struct *p)
2051 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2054 /* Wake flags. The first three directly map to some SD flag value */
2055 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2056 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2057 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2059 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2060 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2061 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2064 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2065 static_assert(WF_FORK == SD_BALANCE_FORK);
2066 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2070 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2071 * of tasks with abnormal "nice" values across CPUs the contribution that
2072 * each task makes to its run queue's load is weighted according to its
2073 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2074 * scaled version of the new time slice allocation that they receive on time
2078 #define WEIGHT_IDLEPRIO 3
2079 #define WMULT_IDLEPRIO 1431655765
2081 extern const int sched_prio_to_weight[40];
2082 extern const u32 sched_prio_to_wmult[40];
2085 * {de,en}queue flags:
2087 * DEQUEUE_SLEEP - task is no longer runnable
2088 * ENQUEUE_WAKEUP - task just became runnable
2090 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2091 * are in a known state which allows modification. Such pairs
2092 * should preserve as much state as possible.
2094 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2097 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2098 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2099 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2103 #define DEQUEUE_SLEEP 0x01
2104 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2105 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2106 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2108 #define ENQUEUE_WAKEUP 0x01
2109 #define ENQUEUE_RESTORE 0x02
2110 #define ENQUEUE_MOVE 0x04
2111 #define ENQUEUE_NOCLOCK 0x08
2113 #define ENQUEUE_HEAD 0x10
2114 #define ENQUEUE_REPLENISH 0x20
2116 #define ENQUEUE_MIGRATED 0x40
2118 #define ENQUEUE_MIGRATED 0x00
2121 #define RETRY_TASK ((void *)-1UL)
2123 struct sched_class {
2125 #ifdef CONFIG_UCLAMP_TASK
2129 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2130 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2131 void (*yield_task) (struct rq *rq);
2132 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2134 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2136 struct task_struct *(*pick_next_task)(struct rq *rq);
2138 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2139 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2142 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2143 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2145 struct task_struct * (*pick_task)(struct rq *rq);
2147 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2149 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2151 void (*set_cpus_allowed)(struct task_struct *p,
2152 const struct cpumask *newmask,
2155 void (*rq_online)(struct rq *rq);
2156 void (*rq_offline)(struct rq *rq);
2158 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2161 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2162 void (*task_fork)(struct task_struct *p);
2163 void (*task_dead)(struct task_struct *p);
2166 * The switched_from() call is allowed to drop rq->lock, therefore we
2167 * cannot assume the switched_from/switched_to pair is serialized by
2168 * rq->lock. They are however serialized by p->pi_lock.
2170 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2171 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2172 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2175 unsigned int (*get_rr_interval)(struct rq *rq,
2176 struct task_struct *task);
2178 void (*update_curr)(struct rq *rq);
2180 #define TASK_SET_GROUP 0
2181 #define TASK_MOVE_GROUP 1
2183 #ifdef CONFIG_FAIR_GROUP_SCHED
2184 void (*task_change_group)(struct task_struct *p, int type);
2188 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2190 WARN_ON_ONCE(rq->curr != prev);
2191 prev->sched_class->put_prev_task(rq, prev);
2194 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2196 next->sched_class->set_next_task(rq, next, false);
2201 * Helper to define a sched_class instance; each one is placed in a separate
2202 * section which is ordered by the linker script:
2204 * include/asm-generic/vmlinux.lds.h
2206 * Also enforce alignment on the instance, not the type, to guarantee layout.
2208 #define DEFINE_SCHED_CLASS(name) \
2209 const struct sched_class name##_sched_class \
2210 __aligned(__alignof__(struct sched_class)) \
2211 __section("__" #name "_sched_class")
2213 /* Defined in include/asm-generic/vmlinux.lds.h */
2214 extern struct sched_class __begin_sched_classes[];
2215 extern struct sched_class __end_sched_classes[];
2217 #define sched_class_highest (__end_sched_classes - 1)
2218 #define sched_class_lowest (__begin_sched_classes - 1)
2220 #define for_class_range(class, _from, _to) \
2221 for (class = (_from); class != (_to); class--)
2223 #define for_each_class(class) \
2224 for_class_range(class, sched_class_highest, sched_class_lowest)
2226 extern const struct sched_class stop_sched_class;
2227 extern const struct sched_class dl_sched_class;
2228 extern const struct sched_class rt_sched_class;
2229 extern const struct sched_class fair_sched_class;
2230 extern const struct sched_class idle_sched_class;
2232 static inline bool sched_stop_runnable(struct rq *rq)
2234 return rq->stop && task_on_rq_queued(rq->stop);
2237 static inline bool sched_dl_runnable(struct rq *rq)
2239 return rq->dl.dl_nr_running > 0;
2242 static inline bool sched_rt_runnable(struct rq *rq)
2244 return rq->rt.rt_queued > 0;
2247 static inline bool sched_fair_runnable(struct rq *rq)
2249 return rq->cfs.nr_running > 0;
2252 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2253 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2255 #define SCA_CHECK 0x01
2256 #define SCA_MIGRATE_DISABLE 0x02
2257 #define SCA_MIGRATE_ENABLE 0x04
2258 #define SCA_USER 0x08
2262 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2264 extern void trigger_load_balance(struct rq *rq);
2266 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2268 static inline struct task_struct *get_push_task(struct rq *rq)
2270 struct task_struct *p = rq->curr;
2272 lockdep_assert_rq_held(rq);
2277 if (p->nr_cpus_allowed == 1)
2280 if (p->migration_disabled)
2283 rq->push_busy = true;
2284 return get_task_struct(p);
2287 extern int push_cpu_stop(void *arg);
2291 #ifdef CONFIG_CPU_IDLE
2292 static inline void idle_set_state(struct rq *rq,
2293 struct cpuidle_state *idle_state)
2295 rq->idle_state = idle_state;
2298 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2300 SCHED_WARN_ON(!rcu_read_lock_held());
2302 return rq->idle_state;
2305 static inline void idle_set_state(struct rq *rq,
2306 struct cpuidle_state *idle_state)
2310 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2316 extern void schedule_idle(void);
2318 extern void sysrq_sched_debug_show(void);
2319 extern void sched_init_granularity(void);
2320 extern void update_max_interval(void);
2322 extern void init_sched_dl_class(void);
2323 extern void init_sched_rt_class(void);
2324 extern void init_sched_fair_class(void);
2326 extern void reweight_task(struct task_struct *p, int prio);
2328 extern void resched_curr(struct rq *rq);
2329 extern void resched_cpu(int cpu);
2331 extern struct rt_bandwidth def_rt_bandwidth;
2332 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2334 extern struct dl_bandwidth def_dl_bandwidth;
2335 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2336 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2337 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2340 #define BW_UNIT (1 << BW_SHIFT)
2341 #define RATIO_SHIFT 8
2342 #define MAX_BW_BITS (64 - BW_SHIFT)
2343 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2344 unsigned long to_ratio(u64 period, u64 runtime);
2346 extern void init_entity_runnable_average(struct sched_entity *se);
2347 extern void post_init_entity_util_avg(struct task_struct *p);
2349 #ifdef CONFIG_NO_HZ_FULL
2350 extern bool sched_can_stop_tick(struct rq *rq);
2351 extern int __init sched_tick_offload_init(void);
2354 * Tick may be needed by tasks in the runqueue depending on their policy and
2355 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2356 * nohz mode if necessary.
2358 static inline void sched_update_tick_dependency(struct rq *rq)
2360 int cpu = cpu_of(rq);
2362 if (!tick_nohz_full_cpu(cpu))
2365 if (sched_can_stop_tick(rq))
2366 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2368 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2371 static inline int sched_tick_offload_init(void) { return 0; }
2372 static inline void sched_update_tick_dependency(struct rq *rq) { }
2375 static inline void add_nr_running(struct rq *rq, unsigned count)
2377 unsigned prev_nr = rq->nr_running;
2379 rq->nr_running = prev_nr + count;
2380 if (trace_sched_update_nr_running_tp_enabled()) {
2381 call_trace_sched_update_nr_running(rq, count);
2385 if (prev_nr < 2 && rq->nr_running >= 2) {
2386 if (!READ_ONCE(rq->rd->overload))
2387 WRITE_ONCE(rq->rd->overload, 1);
2391 sched_update_tick_dependency(rq);
2394 static inline void sub_nr_running(struct rq *rq, unsigned count)
2396 rq->nr_running -= count;
2397 if (trace_sched_update_nr_running_tp_enabled()) {
2398 call_trace_sched_update_nr_running(rq, -count);
2401 /* Check if we still need preemption */
2402 sched_update_tick_dependency(rq);
2405 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2406 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2408 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2410 extern const_debug unsigned int sysctl_sched_nr_migrate;
2411 extern const_debug unsigned int sysctl_sched_migration_cost;
2413 #ifdef CONFIG_SCHED_DEBUG
2414 extern unsigned int sysctl_sched_latency;
2415 extern unsigned int sysctl_sched_min_granularity;
2416 extern unsigned int sysctl_sched_idle_min_granularity;
2417 extern unsigned int sysctl_sched_wakeup_granularity;
2418 extern int sysctl_resched_latency_warn_ms;
2419 extern int sysctl_resched_latency_warn_once;
2421 extern unsigned int sysctl_sched_tunable_scaling;
2423 extern unsigned int sysctl_numa_balancing_scan_delay;
2424 extern unsigned int sysctl_numa_balancing_scan_period_min;
2425 extern unsigned int sysctl_numa_balancing_scan_period_max;
2426 extern unsigned int sysctl_numa_balancing_scan_size;
2429 #ifdef CONFIG_SCHED_HRTICK
2433 * - enabled by features
2434 * - hrtimer is actually high res
2436 static inline int hrtick_enabled(struct rq *rq)
2438 if (!cpu_active(cpu_of(rq)))
2440 return hrtimer_is_hres_active(&rq->hrtick_timer);
2443 static inline int hrtick_enabled_fair(struct rq *rq)
2445 if (!sched_feat(HRTICK))
2447 return hrtick_enabled(rq);
2450 static inline int hrtick_enabled_dl(struct rq *rq)
2452 if (!sched_feat(HRTICK_DL))
2454 return hrtick_enabled(rq);
2457 void hrtick_start(struct rq *rq, u64 delay);
2461 static inline int hrtick_enabled_fair(struct rq *rq)
2466 static inline int hrtick_enabled_dl(struct rq *rq)
2471 static inline int hrtick_enabled(struct rq *rq)
2476 #endif /* CONFIG_SCHED_HRTICK */
2478 #ifndef arch_scale_freq_tick
2479 static __always_inline
2480 void arch_scale_freq_tick(void)
2485 #ifndef arch_scale_freq_capacity
2487 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2488 * @cpu: the CPU in question.
2490 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2493 * ------ * SCHED_CAPACITY_SCALE
2496 static __always_inline
2497 unsigned long arch_scale_freq_capacity(int cpu)
2499 return SCHED_CAPACITY_SCALE;
2506 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2508 #ifdef CONFIG_SCHED_CORE
2510 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2511 * order by core-id first and cpu-id second.
2515 * double_rq_lock(0,3); will take core-0, core-1 lock
2516 * double_rq_lock(1,2); will take core-1, core-0 lock
2518 * when only cpu-id is considered.
2520 if (rq1->core->cpu < rq2->core->cpu)
2522 if (rq1->core->cpu > rq2->core->cpu)
2526 * __sched_core_flip() relies on SMT having cpu-id lock order.
2529 return rq1->cpu < rq2->cpu;
2532 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2534 #ifdef CONFIG_PREEMPTION
2537 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2538 * way at the expense of forcing extra atomic operations in all
2539 * invocations. This assures that the double_lock is acquired using the
2540 * same underlying policy as the spinlock_t on this architecture, which
2541 * reduces latency compared to the unfair variant below. However, it
2542 * also adds more overhead and therefore may reduce throughput.
2544 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2545 __releases(this_rq->lock)
2546 __acquires(busiest->lock)
2547 __acquires(this_rq->lock)
2549 raw_spin_rq_unlock(this_rq);
2550 double_rq_lock(this_rq, busiest);
2557 * Unfair double_lock_balance: Optimizes throughput at the expense of
2558 * latency by eliminating extra atomic operations when the locks are
2559 * already in proper order on entry. This favors lower CPU-ids and will
2560 * grant the double lock to lower CPUs over higher ids under contention,
2561 * regardless of entry order into the function.
2563 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2564 __releases(this_rq->lock)
2565 __acquires(busiest->lock)
2566 __acquires(this_rq->lock)
2568 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2571 if (likely(raw_spin_rq_trylock(busiest)))
2574 if (rq_order_less(this_rq, busiest)) {
2575 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2579 raw_spin_rq_unlock(this_rq);
2580 double_rq_lock(this_rq, busiest);
2585 #endif /* CONFIG_PREEMPTION */
2588 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2590 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2592 lockdep_assert_irqs_disabled();
2594 return _double_lock_balance(this_rq, busiest);
2597 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2598 __releases(busiest->lock)
2600 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2601 raw_spin_rq_unlock(busiest);
2602 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2605 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2611 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2614 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2620 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2623 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2629 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2633 * double_rq_unlock - safely unlock two runqueues
2635 * Note this does not restore interrupts like task_rq_unlock,
2636 * you need to do so manually after calling.
2638 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2639 __releases(rq1->lock)
2640 __releases(rq2->lock)
2642 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2643 raw_spin_rq_unlock(rq2);
2645 __release(rq2->lock);
2646 raw_spin_rq_unlock(rq1);
2649 extern void set_rq_online (struct rq *rq);
2650 extern void set_rq_offline(struct rq *rq);
2651 extern bool sched_smp_initialized;
2653 #else /* CONFIG_SMP */
2656 * double_rq_lock - safely lock two runqueues
2658 * Note this does not disable interrupts like task_rq_lock,
2659 * you need to do so manually before calling.
2661 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2662 __acquires(rq1->lock)
2663 __acquires(rq2->lock)
2665 BUG_ON(!irqs_disabled());
2667 raw_spin_rq_lock(rq1);
2668 __acquire(rq2->lock); /* Fake it out ;) */
2672 * double_rq_unlock - safely unlock two runqueues
2674 * Note this does not restore interrupts like task_rq_unlock,
2675 * you need to do so manually after calling.
2677 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2678 __releases(rq1->lock)
2679 __releases(rq2->lock)
2682 raw_spin_rq_unlock(rq1);
2683 __release(rq2->lock);
2688 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2689 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2691 #ifdef CONFIG_SCHED_DEBUG
2692 extern bool sched_debug_verbose;
2694 extern void print_cfs_stats(struct seq_file *m, int cpu);
2695 extern void print_rt_stats(struct seq_file *m, int cpu);
2696 extern void print_dl_stats(struct seq_file *m, int cpu);
2697 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2698 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2699 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2701 extern void resched_latency_warn(int cpu, u64 latency);
2702 #ifdef CONFIG_NUMA_BALANCING
2704 show_numa_stats(struct task_struct *p, struct seq_file *m);
2706 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2707 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2708 #endif /* CONFIG_NUMA_BALANCING */
2710 static inline void resched_latency_warn(int cpu, u64 latency) {}
2711 #endif /* CONFIG_SCHED_DEBUG */
2713 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2714 extern void init_rt_rq(struct rt_rq *rt_rq);
2715 extern void init_dl_rq(struct dl_rq *dl_rq);
2717 extern void cfs_bandwidth_usage_inc(void);
2718 extern void cfs_bandwidth_usage_dec(void);
2720 #ifdef CONFIG_NO_HZ_COMMON
2721 #define NOHZ_BALANCE_KICK_BIT 0
2722 #define NOHZ_STATS_KICK_BIT 1
2723 #define NOHZ_NEWILB_KICK_BIT 2
2724 #define NOHZ_NEXT_KICK_BIT 3
2726 /* Run rebalance_domains() */
2727 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2728 /* Update blocked load */
2729 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2730 /* Update blocked load when entering idle */
2731 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2732 /* Update nohz.next_balance */
2733 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2735 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2737 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2739 extern void nohz_balance_exit_idle(struct rq *rq);
2741 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2744 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2745 extern void nohz_run_idle_balance(int cpu);
2747 static inline void nohz_run_idle_balance(int cpu) { }
2752 void __dl_update(struct dl_bw *dl_b, s64 bw)
2754 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2757 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2758 "sched RCU must be held");
2759 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2760 struct rq *rq = cpu_rq(i);
2762 rq->dl.extra_bw += bw;
2767 void __dl_update(struct dl_bw *dl_b, s64 bw)
2769 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2776 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2781 struct u64_stats_sync sync;
2784 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2787 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2788 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2789 * and never move forward.
2791 static inline u64 irq_time_read(int cpu)
2793 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2798 seq = __u64_stats_fetch_begin(&irqtime->sync);
2799 total = irqtime->total;
2800 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2804 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2806 #ifdef CONFIG_CPU_FREQ
2807 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2810 * cpufreq_update_util - Take a note about CPU utilization changes.
2811 * @rq: Runqueue to carry out the update for.
2812 * @flags: Update reason flags.
2814 * This function is called by the scheduler on the CPU whose utilization is
2817 * It can only be called from RCU-sched read-side critical sections.
2819 * The way cpufreq is currently arranged requires it to evaluate the CPU
2820 * performance state (frequency/voltage) on a regular basis to prevent it from
2821 * being stuck in a completely inadequate performance level for too long.
2822 * That is not guaranteed to happen if the updates are only triggered from CFS
2823 * and DL, though, because they may not be coming in if only RT tasks are
2824 * active all the time (or there are RT tasks only).
2826 * As a workaround for that issue, this function is called periodically by the
2827 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2828 * but that really is a band-aid. Going forward it should be replaced with
2829 * solutions targeted more specifically at RT tasks.
2831 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2833 struct update_util_data *data;
2835 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2838 data->func(data, rq_clock(rq), flags);
2841 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2842 #endif /* CONFIG_CPU_FREQ */
2844 #ifdef CONFIG_UCLAMP_TASK
2845 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2848 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2849 * @rq: The rq to clamp against. Must not be NULL.
2850 * @util: The util value to clamp.
2851 * @p: The task to clamp against. Can be NULL if you want to clamp
2854 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2856 * If sched_uclamp_used static key is disabled, then just return the util
2857 * without any clamping since uclamp aggregation at the rq level in the fast
2858 * path is disabled, rendering this operation a NOP.
2860 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2861 * will return the correct effective uclamp value of the task even if the
2862 * static key is disabled.
2864 static __always_inline
2865 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2866 struct task_struct *p)
2868 unsigned long min_util = 0;
2869 unsigned long max_util = 0;
2871 if (!static_branch_likely(&sched_uclamp_used))
2875 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2876 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2879 * Ignore last runnable task's max clamp, as this task will
2880 * reset it. Similarly, no need to read the rq's min clamp.
2882 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2886 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2887 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2890 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2891 * RUNNABLE tasks with _different_ clamps, we can end up with an
2892 * inversion. Fix it now when the clamps are applied.
2894 if (unlikely(min_util >= max_util))
2897 return clamp(util, min_util, max_util);
2901 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2902 * by default in the fast path and only gets turned on once userspace performs
2903 * an operation that requires it.
2905 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2908 static inline bool uclamp_is_used(void)
2910 return static_branch_likely(&sched_uclamp_used);
2912 #else /* CONFIG_UCLAMP_TASK */
2914 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2915 struct task_struct *p)
2920 static inline bool uclamp_is_used(void)
2924 #endif /* CONFIG_UCLAMP_TASK */
2926 #ifdef arch_scale_freq_capacity
2927 # ifndef arch_scale_freq_invariant
2928 # define arch_scale_freq_invariant() true
2931 # define arch_scale_freq_invariant() false
2935 static inline unsigned long capacity_orig_of(int cpu)
2937 return cpu_rq(cpu)->cpu_capacity_orig;
2941 * enum cpu_util_type - CPU utilization type
2942 * @FREQUENCY_UTIL: Utilization used to select frequency
2943 * @ENERGY_UTIL: Utilization used during energy calculation
2945 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2946 * need to be aggregated differently depending on the usage made of them. This
2947 * enum is used within effective_cpu_util() to differentiate the types of
2948 * utilization expected by the callers, and adjust the aggregation accordingly.
2950 enum cpu_util_type {
2955 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2956 unsigned long max, enum cpu_util_type type,
2957 struct task_struct *p);
2959 static inline unsigned long cpu_bw_dl(struct rq *rq)
2961 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2964 static inline unsigned long cpu_util_dl(struct rq *rq)
2966 return READ_ONCE(rq->avg_dl.util_avg);
2970 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2971 * @cpu: the CPU to get the utilization for.
2973 * The unit of the return value must be the same as the one of CPU capacity
2974 * so that CPU utilization can be compared with CPU capacity.
2976 * CPU utilization is the sum of running time of runnable tasks plus the
2977 * recent utilization of currently non-runnable tasks on that CPU.
2978 * It represents the amount of CPU capacity currently used by CFS tasks in
2979 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2980 * capacity at f_max.
2982 * The estimated CPU utilization is defined as the maximum between CPU
2983 * utilization and sum of the estimated utilization of the currently
2984 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2985 * previously-executed tasks, which helps better deduce how busy a CPU will
2986 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2987 * of such a task would be significantly decayed at this point of time.
2989 * CPU utilization can be higher than the current CPU capacity
2990 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2991 * of rounding errors as well as task migrations or wakeups of new tasks.
2992 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2993 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2994 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2995 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2996 * though since this is useful for predicting the CPU capacity required
2997 * after task migrations (scheduler-driven DVFS).
2999 * Return: (Estimated) utilization for the specified CPU.
3001 static inline unsigned long cpu_util_cfs(int cpu)
3003 struct cfs_rq *cfs_rq;
3006 cfs_rq = &cpu_rq(cpu)->cfs;
3007 util = READ_ONCE(cfs_rq->avg.util_avg);
3009 if (sched_feat(UTIL_EST)) {
3010 util = max_t(unsigned long, util,
3011 READ_ONCE(cfs_rq->avg.util_est.enqueued));
3014 return min(util, capacity_orig_of(cpu));
3017 static inline unsigned long cpu_util_rt(struct rq *rq)
3019 return READ_ONCE(rq->avg_rt.util_avg);
3023 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3024 static inline unsigned long cpu_util_irq(struct rq *rq)
3026 return rq->avg_irq.util_avg;
3030 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3032 util *= (max - irq);
3039 static inline unsigned long cpu_util_irq(struct rq *rq)
3045 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3051 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3053 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3055 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3057 static inline bool sched_energy_enabled(void)
3059 return static_branch_unlikely(&sched_energy_present);
3062 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3064 #define perf_domain_span(pd) NULL
3065 static inline bool sched_energy_enabled(void) { return false; }
3067 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3069 #ifdef CONFIG_MEMBARRIER
3071 * The scheduler provides memory barriers required by membarrier between:
3072 * - prior user-space memory accesses and store to rq->membarrier_state,
3073 * - store to rq->membarrier_state and following user-space memory accesses.
3074 * In the same way it provides those guarantees around store to rq->curr.
3076 static inline void membarrier_switch_mm(struct rq *rq,
3077 struct mm_struct *prev_mm,
3078 struct mm_struct *next_mm)
3080 int membarrier_state;
3082 if (prev_mm == next_mm)
3085 membarrier_state = atomic_read(&next_mm->membarrier_state);
3086 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3089 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3092 static inline void membarrier_switch_mm(struct rq *rq,
3093 struct mm_struct *prev_mm,
3094 struct mm_struct *next_mm)
3100 static inline bool is_per_cpu_kthread(struct task_struct *p)
3102 if (!(p->flags & PF_KTHREAD))
3105 if (p->nr_cpus_allowed != 1)
3112 extern void swake_up_all_locked(struct swait_queue_head *q);
3113 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3115 #ifdef CONFIG_PREEMPT_DYNAMIC
3116 extern int preempt_dynamic_mode;
3117 extern int sched_dynamic_mode(const char *str);
3118 extern void sched_dynamic_update(int mode);