1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/cpufreq.h>
31 #include <linux/cpumask_api.h>
32 #include <linux/ctype.h>
33 #include <linux/file.h>
34 #include <linux/fs_api.h>
35 #include <linux/hrtimer_api.h>
36 #include <linux/interrupt.h>
37 #include <linux/irq_work.h>
38 #include <linux/jiffies.h>
39 #include <linux/kref_api.h>
40 #include <linux/kthread.h>
41 #include <linux/ktime_api.h>
42 #include <linux/lockdep_api.h>
43 #include <linux/lockdep.h>
44 #include <linux/minmax.h>
46 #include <linux/module.h>
47 #include <linux/mutex_api.h>
48 #include <linux/plist.h>
49 #include <linux/poll.h>
50 #include <linux/proc_fs.h>
51 #include <linux/profile.h>
52 #include <linux/psi.h>
53 #include <linux/rcupdate.h>
54 #include <linux/seq_file.h>
55 #include <linux/seqlock.h>
56 #include <linux/softirq.h>
57 #include <linux/spinlock_api.h>
58 #include <linux/static_key.h>
59 #include <linux/stop_machine.h>
60 #include <linux/syscalls_api.h>
61 #include <linux/syscalls.h>
62 #include <linux/tick.h>
63 #include <linux/topology.h>
64 #include <linux/types.h>
65 #include <linux/u64_stats_sync_api.h>
66 #include <linux/uaccess.h>
67 #include <linux/wait_api.h>
68 #include <linux/wait_bit.h>
69 #include <linux/workqueue_api.h>
71 #include <trace/events/power.h>
72 #include <trace/events/sched.h>
74 #include "../workqueue_internal.h"
76 #ifdef CONFIG_CGROUP_SCHED
77 #include <linux/cgroup.h>
78 #include <linux/psi.h>
81 #ifdef CONFIG_SCHED_DEBUG
82 # include <linux/static_key.h>
85 #ifdef CONFIG_PARAVIRT
86 # include <asm/paravirt.h>
87 # include <asm/paravirt_api_clock.h>
91 #include "cpudeadline.h"
93 #ifdef CONFIG_SCHED_DEBUG
94 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
96 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
100 struct cpuidle_state;
102 /* task_struct::on_rq states: */
103 #define TASK_ON_RQ_QUEUED 1
104 #define TASK_ON_RQ_MIGRATING 2
106 extern __read_mostly int scheduler_running;
108 extern unsigned long calc_load_update;
109 extern atomic_long_t calc_load_tasks;
111 extern unsigned int sysctl_sched_child_runs_first;
113 extern void calc_global_load_tick(struct rq *this_rq);
114 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
116 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
118 extern unsigned int sysctl_sched_rt_period;
119 extern int sysctl_sched_rt_runtime;
120 extern int sched_rr_timeslice;
123 * Helpers for converting nanosecond timing to jiffy resolution
125 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
128 * Increase resolution of nice-level calculations for 64-bit architectures.
129 * The extra resolution improves shares distribution and load balancing of
130 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
131 * hierarchies, especially on larger systems. This is not a user-visible change
132 * and does not change the user-interface for setting shares/weights.
134 * We increase resolution only if we have enough bits to allow this increased
135 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
136 * are pretty high and the returns do not justify the increased costs.
138 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
139 * increase coverage and consistency always enable it on 64-bit platforms.
142 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
143 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
144 # define scale_load_down(w) \
146 unsigned long __w = (w); \
148 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
152 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
153 # define scale_load(w) (w)
154 # define scale_load_down(w) (w)
158 * Task weight (visible to users) and its load (invisible to users) have
159 * independent resolution, but they should be well calibrated. We use
160 * scale_load() and scale_load_down(w) to convert between them. The
161 * following must be true:
163 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
166 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
169 * Single value that decides SCHED_DEADLINE internal math precision.
170 * 10 -> just above 1us
171 * 9 -> just above 0.5us
176 * Single value that denotes runtime == period, ie unlimited time.
178 #define RUNTIME_INF ((u64)~0ULL)
180 static inline int idle_policy(int policy)
182 return policy == SCHED_IDLE;
184 static inline int fair_policy(int policy)
186 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
189 static inline int rt_policy(int policy)
191 return policy == SCHED_FIFO || policy == SCHED_RR;
194 static inline int dl_policy(int policy)
196 return policy == SCHED_DEADLINE;
198 static inline bool valid_policy(int policy)
200 return idle_policy(policy) || fair_policy(policy) ||
201 rt_policy(policy) || dl_policy(policy);
204 static inline int task_has_idle_policy(struct task_struct *p)
206 return idle_policy(p->policy);
209 static inline int task_has_rt_policy(struct task_struct *p)
211 return rt_policy(p->policy);
214 static inline int task_has_dl_policy(struct task_struct *p)
216 return dl_policy(p->policy);
219 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
221 static inline void update_avg(u64 *avg, u64 sample)
223 s64 diff = sample - *avg;
228 * Shifting a value by an exponent greater *or equal* to the size of said value
229 * is UB; cap at size-1.
231 #define shr_bound(val, shift) \
232 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
235 * !! For sched_setattr_nocheck() (kernel) only !!
237 * This is actually gross. :(
239 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
240 * tasks, but still be able to sleep. We need this on platforms that cannot
241 * atomically change clock frequency. Remove once fast switching will be
242 * available on such platforms.
244 * SUGOV stands for SchedUtil GOVernor.
246 #define SCHED_FLAG_SUGOV 0x10000000
248 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
250 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
252 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
253 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
260 * Tells if entity @a should preempt entity @b.
263 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
265 return dl_entity_is_special(a) ||
266 dl_time_before(a->deadline, b->deadline);
270 * This is the priority-queue data structure of the RT scheduling class:
272 struct rt_prio_array {
273 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
274 struct list_head queue[MAX_RT_PRIO];
277 struct rt_bandwidth {
278 /* nests inside the rq lock: */
279 raw_spinlock_t rt_runtime_lock;
282 struct hrtimer rt_period_timer;
283 unsigned int rt_period_active;
286 void __dl_clear_params(struct task_struct *p);
288 struct dl_bandwidth {
289 raw_spinlock_t dl_runtime_lock;
294 static inline int dl_bandwidth_enabled(void)
296 return sysctl_sched_rt_runtime >= 0;
300 * To keep the bandwidth of -deadline tasks under control
301 * we need some place where:
302 * - store the maximum -deadline bandwidth of each cpu;
303 * - cache the fraction of bandwidth that is currently allocated in
306 * This is all done in the data structure below. It is similar to the
307 * one used for RT-throttling (rt_bandwidth), with the main difference
308 * that, since here we are only interested in admission control, we
309 * do not decrease any runtime while the group "executes", neither we
310 * need a timer to replenish it.
312 * With respect to SMP, bandwidth is given on a per root domain basis,
314 * - bw (< 100%) is the deadline bandwidth of each CPU;
315 * - total_bw is the currently allocated bandwidth in each root domain;
324 * Verify the fitness of task @p to run on @cpu taking into account the
325 * CPU original capacity and the runtime/deadline ratio of the task.
327 * The function will return true if the CPU original capacity of the
328 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
329 * task and false otherwise.
331 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
333 unsigned long cap = arch_scale_cpu_capacity(cpu);
335 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
338 extern void init_dl_bw(struct dl_bw *dl_b);
339 extern int sched_dl_global_validate(void);
340 extern void sched_dl_do_global(void);
341 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
342 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
343 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
344 extern bool __checkparam_dl(const struct sched_attr *attr);
345 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
346 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
347 extern int dl_cpu_busy(int cpu, struct task_struct *p);
349 #ifdef CONFIG_CGROUP_SCHED
354 extern struct list_head task_groups;
356 struct cfs_bandwidth {
357 #ifdef CONFIG_CFS_BANDWIDTH
364 s64 hierarchical_quota;
369 struct hrtimer period_timer;
370 struct hrtimer slack_timer;
371 struct list_head throttled_cfs_rq;
382 /* Task group related information */
384 struct cgroup_subsys_state css;
386 #ifdef CONFIG_FAIR_GROUP_SCHED
387 /* schedulable entities of this group on each CPU */
388 struct sched_entity **se;
389 /* runqueue "owned" by this group on each CPU */
390 struct cfs_rq **cfs_rq;
391 unsigned long shares;
393 /* A positive value indicates that this is a SCHED_IDLE group. */
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 unregister_rt_sched_group(struct task_group *tg);
484 extern void free_rt_sched_group(struct task_group *tg);
485 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
486 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
487 struct sched_rt_entity *rt_se, int cpu,
488 struct sched_rt_entity *parent);
489 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
490 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
491 extern long sched_group_rt_runtime(struct task_group *tg);
492 extern long sched_group_rt_period(struct task_group *tg);
493 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
495 extern struct task_group *sched_create_group(struct task_group *parent);
496 extern void sched_online_group(struct task_group *tg,
497 struct task_group *parent);
498 extern void sched_destroy_group(struct task_group *tg);
499 extern void sched_release_group(struct task_group *tg);
501 extern void sched_move_task(struct task_struct *tsk);
503 #ifdef CONFIG_FAIR_GROUP_SCHED
504 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
506 extern int sched_group_set_idle(struct task_group *tg, long idle);
509 extern void set_task_rq_fair(struct sched_entity *se,
510 struct cfs_rq *prev, struct cfs_rq *next);
511 #else /* !CONFIG_SMP */
512 static inline void set_task_rq_fair(struct sched_entity *se,
513 struct cfs_rq *prev, struct cfs_rq *next) { }
514 #endif /* CONFIG_SMP */
515 #endif /* CONFIG_FAIR_GROUP_SCHED */
517 #else /* CONFIG_CGROUP_SCHED */
519 struct cfs_bandwidth { };
521 #endif /* CONFIG_CGROUP_SCHED */
523 /* CFS-related fields in a runqueue */
525 struct load_weight load;
526 unsigned int nr_running;
527 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
528 unsigned int idle_nr_running; /* SCHED_IDLE */
529 unsigned int idle_h_nr_running; /* SCHED_IDLE */
533 #ifdef CONFIG_SCHED_CORE
534 unsigned int forceidle_seq;
539 u64 min_vruntime_copy;
542 struct rb_root_cached tasks_timeline;
545 * 'curr' points to currently running entity on this cfs_rq.
546 * It is set to NULL otherwise (i.e when none are currently running).
548 struct sched_entity *curr;
549 struct sched_entity *next;
550 struct sched_entity *last;
551 struct sched_entity *skip;
553 #ifdef CONFIG_SCHED_DEBUG
554 unsigned int nr_spread_over;
561 struct sched_avg avg;
563 u64 load_last_update_time_copy;
566 raw_spinlock_t lock ____cacheline_aligned;
568 unsigned long load_avg;
569 unsigned long util_avg;
570 unsigned long runnable_avg;
573 #ifdef CONFIG_FAIR_GROUP_SCHED
574 unsigned long tg_load_avg_contrib;
576 long prop_runnable_sum;
579 * h_load = weight * f(tg)
581 * Where f(tg) is the recursive weight fraction assigned to
584 unsigned long h_load;
585 u64 last_h_load_update;
586 struct sched_entity *h_load_next;
587 #endif /* CONFIG_FAIR_GROUP_SCHED */
588 #endif /* CONFIG_SMP */
590 #ifdef CONFIG_FAIR_GROUP_SCHED
591 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
594 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
595 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
596 * (like users, containers etc.)
598 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
599 * This list is used during load balance.
602 struct list_head leaf_cfs_rq_list;
603 struct task_group *tg; /* group that "owns" this runqueue */
605 /* Locally cached copy of our task_group's idle value */
608 #ifdef CONFIG_CFS_BANDWIDTH
610 s64 runtime_remaining;
613 u64 throttled_clock_pelt;
614 u64 throttled_clock_pelt_time;
617 struct list_head throttled_list;
618 #endif /* CONFIG_CFS_BANDWIDTH */
619 #endif /* CONFIG_FAIR_GROUP_SCHED */
622 static inline int rt_bandwidth_enabled(void)
624 return sysctl_sched_rt_runtime >= 0;
627 /* RT IPI pull logic requires IRQ_WORK */
628 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
629 # define HAVE_RT_PUSH_IPI
632 /* Real-Time classes' related field in a runqueue: */
634 struct rt_prio_array active;
635 unsigned int rt_nr_running;
636 unsigned int rr_nr_running;
637 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
639 int curr; /* highest queued rt task prio */
641 int next; /* next highest */
646 unsigned int rt_nr_migratory;
647 unsigned int rt_nr_total;
649 struct plist_head pushable_tasks;
651 #endif /* CONFIG_SMP */
657 /* Nests inside the rq lock: */
658 raw_spinlock_t rt_runtime_lock;
660 #ifdef CONFIG_RT_GROUP_SCHED
661 unsigned int rt_nr_boosted;
664 struct task_group *tg;
668 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
670 return rt_rq->rt_queued && rt_rq->rt_nr_running;
673 /* Deadline class' related fields in a runqueue */
675 /* runqueue is an rbtree, ordered by deadline */
676 struct rb_root_cached root;
678 unsigned int dl_nr_running;
682 * Deadline values of the currently executing and the
683 * earliest ready task on this rq. Caching these facilitates
684 * the decision whether or not a ready but not running task
685 * should migrate somewhere else.
692 unsigned int dl_nr_migratory;
696 * Tasks on this rq that can be pushed away. They are kept in
697 * an rb-tree, ordered by tasks' deadlines, with caching
698 * of the leftmost (earliest deadline) element.
700 struct rb_root_cached pushable_dl_tasks_root;
705 * "Active utilization" for this runqueue: increased when a
706 * task wakes up (becomes TASK_RUNNING) and decreased when a
712 * Utilization of the tasks "assigned" to this runqueue (including
713 * the tasks that are in runqueue and the tasks that executed on this
714 * CPU and blocked). Increased when a task moves to this runqueue, and
715 * decreased when the task moves away (migrates, changes scheduling
716 * policy, or terminates).
717 * This is needed to compute the "inactive utilization" for the
718 * runqueue (inactive utilization = this_bw - running_bw).
724 * Inverse of the fraction of CPU utilization that can be reclaimed
725 * by the GRUB algorithm.
730 #ifdef CONFIG_FAIR_GROUP_SCHED
731 /* An entity is a task if it doesn't "own" a runqueue */
732 #define entity_is_task(se) (!se->my_q)
734 static inline void se_update_runnable(struct sched_entity *se)
736 if (!entity_is_task(se))
737 se->runnable_weight = se->my_q->h_nr_running;
740 static inline long se_runnable(struct sched_entity *se)
742 if (entity_is_task(se))
745 return se->runnable_weight;
749 #define entity_is_task(se) 1
751 static inline void se_update_runnable(struct sched_entity *se) {}
753 static inline long se_runnable(struct sched_entity *se)
761 * XXX we want to get rid of these helpers and use the full load resolution.
763 static inline long se_weight(struct sched_entity *se)
765 return scale_load_down(se->load.weight);
769 static inline bool sched_asym_prefer(int a, int b)
771 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
775 struct em_perf_domain *em_pd;
776 struct perf_domain *next;
780 /* Scheduling group status flags */
781 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
782 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
785 * We add the notion of a root-domain which will be used to define per-domain
786 * variables. Each exclusive cpuset essentially defines an island domain by
787 * fully partitioning the member CPUs from any other cpuset. Whenever a new
788 * exclusive cpuset is created, we also create and attach a new root-domain
797 cpumask_var_t online;
800 * Indicate pullable load on at least one CPU, e.g:
801 * - More than one runnable task
802 * - Running task is misfit
806 /* Indicate one or more cpus over-utilized (tipping point) */
810 * The bit corresponding to a CPU gets set here if such CPU has more
811 * than one runnable -deadline task (as it is below for RT tasks).
813 cpumask_var_t dlo_mask;
819 * Indicate whether a root_domain's dl_bw has been checked or
820 * updated. It's monotonously increasing value.
822 * Also, some corner cases, like 'wrap around' is dangerous, but given
823 * that u64 is 'big enough'. So that shouldn't be a concern.
827 #ifdef HAVE_RT_PUSH_IPI
829 * For IPI pull requests, loop across the rto_mask.
831 struct irq_work rto_push_work;
832 raw_spinlock_t rto_lock;
833 /* These are only updated and read within rto_lock */
836 /* These atomics are updated outside of a lock */
837 atomic_t rto_loop_next;
838 atomic_t rto_loop_start;
841 * The "RT overload" flag: it gets set if a CPU has more than
842 * one runnable RT task.
844 cpumask_var_t rto_mask;
845 struct cpupri cpupri;
847 unsigned long max_cpu_capacity;
850 * NULL-terminated list of performance domains intersecting with the
851 * CPUs of the rd. Protected by RCU.
853 struct perf_domain __rcu *pd;
856 extern void init_defrootdomain(void);
857 extern int sched_init_domains(const struct cpumask *cpu_map);
858 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
859 extern void sched_get_rd(struct root_domain *rd);
860 extern void sched_put_rd(struct root_domain *rd);
862 #ifdef HAVE_RT_PUSH_IPI
863 extern void rto_push_irq_work_func(struct irq_work *work);
865 #endif /* CONFIG_SMP */
867 #ifdef CONFIG_UCLAMP_TASK
869 * struct uclamp_bucket - Utilization clamp bucket
870 * @value: utilization clamp value for tasks on this clamp bucket
871 * @tasks: number of RUNNABLE tasks on this clamp bucket
873 * Keep track of how many tasks are RUNNABLE for a given utilization
876 struct uclamp_bucket {
877 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
878 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
882 * struct uclamp_rq - rq's utilization clamp
883 * @value: currently active clamp values for a rq
884 * @bucket: utilization clamp buckets affecting a rq
886 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
887 * A clamp value is affecting a rq when there is at least one task RUNNABLE
888 * (or actually running) with that value.
890 * There are up to UCLAMP_CNT possible different clamp values, currently there
891 * are only two: minimum utilization and maximum utilization.
893 * All utilization clamping values are MAX aggregated, since:
894 * - for util_min: we want to run the CPU at least at the max of the minimum
895 * utilization required by its currently RUNNABLE tasks.
896 * - for util_max: we want to allow the CPU to run up to the max of the
897 * maximum utilization allowed by its currently RUNNABLE tasks.
899 * Since on each system we expect only a limited number of different
900 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
901 * the metrics required to compute all the per-rq utilization clamp values.
905 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
908 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
909 #endif /* CONFIG_UCLAMP_TASK */
912 * This is the main, per-CPU runqueue data structure.
914 * Locking rule: those places that want to lock multiple runqueues
915 * (such as the load balancing or the thread migration code), lock
916 * acquire operations must be ordered by ascending &runqueue.
920 raw_spinlock_t __lock;
923 * nr_running and cpu_load should be in the same cacheline because
924 * remote CPUs use both these fields when doing load calculation.
926 unsigned int nr_running;
927 #ifdef CONFIG_NUMA_BALANCING
928 unsigned int nr_numa_running;
929 unsigned int nr_preferred_running;
930 unsigned int numa_migrate_on;
932 #ifdef CONFIG_NO_HZ_COMMON
934 unsigned long last_blocked_load_update_tick;
935 unsigned int has_blocked_load;
936 call_single_data_t nohz_csd;
937 #endif /* CONFIG_SMP */
938 unsigned int nohz_tick_stopped;
940 #endif /* CONFIG_NO_HZ_COMMON */
943 unsigned int ttwu_pending;
947 #ifdef CONFIG_UCLAMP_TASK
948 /* Utilization clamp values based on CPU's RUNNABLE tasks */
949 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
950 unsigned int uclamp_flags;
951 #define UCLAMP_FLAG_IDLE 0x01
958 #ifdef CONFIG_FAIR_GROUP_SCHED
959 /* list of leaf cfs_rq on this CPU: */
960 struct list_head leaf_cfs_rq_list;
961 struct list_head *tmp_alone_branch;
962 #endif /* CONFIG_FAIR_GROUP_SCHED */
965 * This is part of a global counter where only the total sum
966 * over all CPUs matters. A task can increase this counter on
967 * one CPU and if it got migrated afterwards it may decrease
968 * it on another CPU. Always updated under the runqueue lock:
970 unsigned int nr_uninterruptible;
972 struct task_struct __rcu *curr;
973 struct task_struct *idle;
974 struct task_struct *stop;
975 unsigned long next_balance;
976 struct mm_struct *prev_mm;
978 unsigned int clock_update_flags;
980 /* Ensure that all clocks are in the same cache line */
981 u64 clock_task ____cacheline_aligned;
983 unsigned long lost_idle_time;
987 #ifdef CONFIG_SCHED_DEBUG
988 u64 last_seen_need_resched_ns;
989 int ticks_without_resched;
992 #ifdef CONFIG_MEMBARRIER
993 int membarrier_state;
997 struct root_domain *rd;
998 struct sched_domain __rcu *sd;
1000 unsigned long cpu_capacity;
1001 unsigned long cpu_capacity_orig;
1003 struct callback_head *balance_callback;
1005 unsigned char nohz_idle_balance;
1006 unsigned char idle_balance;
1008 unsigned long misfit_task_load;
1010 /* For active balancing */
1013 struct cpu_stop_work active_balance_work;
1015 /* CPU of this runqueue: */
1019 struct list_head cfs_tasks;
1021 struct sched_avg avg_rt;
1022 struct sched_avg avg_dl;
1023 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1024 struct sched_avg avg_irq;
1026 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1027 struct sched_avg avg_thermal;
1032 unsigned long wake_stamp;
1035 /* This is used to determine avg_idle's max value */
1036 u64 max_idle_balance_cost;
1038 #ifdef CONFIG_HOTPLUG_CPU
1039 struct rcuwait hotplug_wait;
1041 #endif /* CONFIG_SMP */
1043 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1046 #ifdef CONFIG_PARAVIRT
1047 u64 prev_steal_time;
1049 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1050 u64 prev_steal_time_rq;
1053 /* calc_load related fields */
1054 unsigned long calc_load_update;
1055 long calc_load_active;
1057 #ifdef CONFIG_SCHED_HRTICK
1059 call_single_data_t hrtick_csd;
1061 struct hrtimer hrtick_timer;
1062 ktime_t hrtick_time;
1065 #ifdef CONFIG_SCHEDSTATS
1067 struct sched_info rq_sched_info;
1068 unsigned long long rq_cpu_time;
1069 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1071 /* sys_sched_yield() stats */
1072 unsigned int yld_count;
1074 /* schedule() stats */
1075 unsigned int sched_count;
1076 unsigned int sched_goidle;
1078 /* try_to_wake_up() stats */
1079 unsigned int ttwu_count;
1080 unsigned int ttwu_local;
1083 #ifdef CONFIG_CPU_IDLE
1084 /* Must be inspected within a rcu lock section */
1085 struct cpuidle_state *idle_state;
1089 unsigned int nr_pinned;
1091 unsigned int push_busy;
1092 struct cpu_stop_work push_work;
1094 #ifdef CONFIG_SCHED_CORE
1097 struct task_struct *core_pick;
1098 unsigned int core_enabled;
1099 unsigned int core_sched_seq;
1100 struct rb_root core_tree;
1102 /* shared state -- careful with sched_core_cpu_deactivate() */
1103 unsigned int core_task_seq;
1104 unsigned int core_pick_seq;
1105 unsigned long core_cookie;
1106 unsigned int core_forceidle_count;
1107 unsigned int core_forceidle_seq;
1108 unsigned int core_forceidle_occupation;
1109 u64 core_forceidle_start;
1113 #ifdef CONFIG_FAIR_GROUP_SCHED
1115 /* CPU runqueue to which this cfs_rq is attached */
1116 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1123 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1125 return container_of(cfs_rq, struct rq, cfs);
1129 static inline int cpu_of(struct rq *rq)
1138 #define MDF_PUSH 0x01
1140 static inline bool is_migration_disabled(struct task_struct *p)
1143 return p->migration_disabled;
1150 #ifdef CONFIG_SCHED_CORE
1151 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1153 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1155 static inline bool sched_core_enabled(struct rq *rq)
1157 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1160 static inline bool sched_core_disabled(void)
1162 return !static_branch_unlikely(&__sched_core_enabled);
1166 * Be careful with this function; not for general use. The return value isn't
1167 * stable unless you actually hold a relevant rq->__lock.
1169 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1171 if (sched_core_enabled(rq))
1172 return &rq->core->__lock;
1177 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1179 if (rq->core_enabled)
1180 return &rq->core->__lock;
1185 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1188 * Helpers to check if the CPU's core cookie matches with the task's cookie
1189 * when core scheduling is enabled.
1190 * A special case is that the task's cookie always matches with CPU's core
1191 * cookie if the CPU is in an idle core.
1193 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1195 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1196 if (!sched_core_enabled(rq))
1199 return rq->core->core_cookie == p->core_cookie;
1202 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1204 bool idle_core = true;
1207 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1208 if (!sched_core_enabled(rq))
1211 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1212 if (!available_idle_cpu(cpu)) {
1219 * A CPU in an idle core is always the best choice for tasks with
1222 return idle_core || rq->core->core_cookie == p->core_cookie;
1225 static inline bool sched_group_cookie_match(struct rq *rq,
1226 struct task_struct *p,
1227 struct sched_group *group)
1231 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1232 if (!sched_core_enabled(rq))
1235 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1236 if (sched_core_cookie_match(rq, p))
1242 static inline bool sched_core_enqueued(struct task_struct *p)
1244 return !RB_EMPTY_NODE(&p->core_node);
1247 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1248 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1250 extern void sched_core_get(void);
1251 extern void sched_core_put(void);
1253 #else /* !CONFIG_SCHED_CORE */
1255 static inline bool sched_core_enabled(struct rq *rq)
1260 static inline bool sched_core_disabled(void)
1265 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1270 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1275 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1280 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1285 static inline bool sched_group_cookie_match(struct rq *rq,
1286 struct task_struct *p,
1287 struct sched_group *group)
1291 #endif /* CONFIG_SCHED_CORE */
1293 static inline void lockdep_assert_rq_held(struct rq *rq)
1295 lockdep_assert_held(__rq_lockp(rq));
1298 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1299 extern bool raw_spin_rq_trylock(struct rq *rq);
1300 extern void raw_spin_rq_unlock(struct rq *rq);
1302 static inline void raw_spin_rq_lock(struct rq *rq)
1304 raw_spin_rq_lock_nested(rq, 0);
1307 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1309 local_irq_disable();
1310 raw_spin_rq_lock(rq);
1313 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1315 raw_spin_rq_unlock(rq);
1319 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1321 unsigned long flags;
1322 local_irq_save(flags);
1323 raw_spin_rq_lock(rq);
1327 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1329 raw_spin_rq_unlock(rq);
1330 local_irq_restore(flags);
1333 #define raw_spin_rq_lock_irqsave(rq, flags) \
1335 flags = _raw_spin_rq_lock_irqsave(rq); \
1338 #ifdef CONFIG_SCHED_SMT
1339 extern void __update_idle_core(struct rq *rq);
1341 static inline void update_idle_core(struct rq *rq)
1343 if (static_branch_unlikely(&sched_smt_present))
1344 __update_idle_core(rq);
1348 static inline void update_idle_core(struct rq *rq) { }
1351 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1353 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1354 #define this_rq() this_cpu_ptr(&runqueues)
1355 #define task_rq(p) cpu_rq(task_cpu(p))
1356 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1357 #define raw_rq() raw_cpu_ptr(&runqueues)
1359 #ifdef CONFIG_FAIR_GROUP_SCHED
1360 static inline struct task_struct *task_of(struct sched_entity *se)
1362 SCHED_WARN_ON(!entity_is_task(se));
1363 return container_of(se, struct task_struct, se);
1366 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1368 return p->se.cfs_rq;
1371 /* runqueue on which this entity is (to be) queued */
1372 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1377 /* runqueue "owned" by this group */
1378 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1385 static inline struct task_struct *task_of(struct sched_entity *se)
1387 return container_of(se, struct task_struct, se);
1390 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1392 return &task_rq(p)->cfs;
1395 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1397 struct task_struct *p = task_of(se);
1398 struct rq *rq = task_rq(p);
1403 /* runqueue "owned" by this group */
1404 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1410 extern void update_rq_clock(struct rq *rq);
1413 * rq::clock_update_flags bits
1415 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1416 * call to __schedule(). This is an optimisation to avoid
1417 * neighbouring rq clock updates.
1419 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1420 * in effect and calls to update_rq_clock() are being ignored.
1422 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1423 * made to update_rq_clock() since the last time rq::lock was pinned.
1425 * If inside of __schedule(), clock_update_flags will have been
1426 * shifted left (a left shift is a cheap operation for the fast path
1427 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1429 * if (rq-clock_update_flags >= RQCF_UPDATED)
1431 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1432 * one position though, because the next rq_unpin_lock() will shift it
1435 #define RQCF_REQ_SKIP 0x01
1436 #define RQCF_ACT_SKIP 0x02
1437 #define RQCF_UPDATED 0x04
1439 static inline void assert_clock_updated(struct rq *rq)
1442 * The only reason for not seeing a clock update since the
1443 * last rq_pin_lock() is if we're currently skipping updates.
1445 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1448 static inline u64 rq_clock(struct rq *rq)
1450 lockdep_assert_rq_held(rq);
1451 assert_clock_updated(rq);
1456 static inline u64 rq_clock_task(struct rq *rq)
1458 lockdep_assert_rq_held(rq);
1459 assert_clock_updated(rq);
1461 return rq->clock_task;
1465 * By default the decay is the default pelt decay period.
1466 * The decay shift can change the decay period in
1468 * Decay shift Decay period(ms)
1475 extern int sched_thermal_decay_shift;
1477 static inline u64 rq_clock_thermal(struct rq *rq)
1479 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1482 static inline void rq_clock_skip_update(struct rq *rq)
1484 lockdep_assert_rq_held(rq);
1485 rq->clock_update_flags |= RQCF_REQ_SKIP;
1489 * See rt task throttling, which is the only time a skip
1490 * request is canceled.
1492 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1494 lockdep_assert_rq_held(rq);
1495 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1499 unsigned long flags;
1500 struct pin_cookie cookie;
1501 #ifdef CONFIG_SCHED_DEBUG
1503 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1504 * current pin context is stashed here in case it needs to be
1505 * restored in rq_repin_lock().
1507 unsigned int clock_update_flags;
1511 extern struct callback_head balance_push_callback;
1514 * Lockdep annotation that avoids accidental unlocks; it's like a
1515 * sticky/continuous lockdep_assert_held().
1517 * This avoids code that has access to 'struct rq *rq' (basically everything in
1518 * the scheduler) from accidentally unlocking the rq if they do not also have a
1519 * copy of the (on-stack) 'struct rq_flags rf'.
1521 * Also see Documentation/locking/lockdep-design.rst.
1523 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1525 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1527 #ifdef CONFIG_SCHED_DEBUG
1528 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1529 rf->clock_update_flags = 0;
1531 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1536 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1538 #ifdef CONFIG_SCHED_DEBUG
1539 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1540 rf->clock_update_flags = RQCF_UPDATED;
1543 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1546 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1548 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1550 #ifdef CONFIG_SCHED_DEBUG
1552 * Restore the value we stashed in @rf for this pin context.
1554 rq->clock_update_flags |= rf->clock_update_flags;
1558 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1559 __acquires(rq->lock);
1561 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1562 __acquires(p->pi_lock)
1563 __acquires(rq->lock);
1565 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1566 __releases(rq->lock)
1568 rq_unpin_lock(rq, rf);
1569 raw_spin_rq_unlock(rq);
1573 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1574 __releases(rq->lock)
1575 __releases(p->pi_lock)
1577 rq_unpin_lock(rq, rf);
1578 raw_spin_rq_unlock(rq);
1579 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1583 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1584 __acquires(rq->lock)
1586 raw_spin_rq_lock_irqsave(rq, rf->flags);
1587 rq_pin_lock(rq, rf);
1591 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1592 __acquires(rq->lock)
1594 raw_spin_rq_lock_irq(rq);
1595 rq_pin_lock(rq, rf);
1599 rq_lock(struct rq *rq, struct rq_flags *rf)
1600 __acquires(rq->lock)
1602 raw_spin_rq_lock(rq);
1603 rq_pin_lock(rq, rf);
1607 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1608 __releases(rq->lock)
1610 rq_unpin_lock(rq, rf);
1611 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1615 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1616 __releases(rq->lock)
1618 rq_unpin_lock(rq, rf);
1619 raw_spin_rq_unlock_irq(rq);
1623 rq_unlock(struct rq *rq, struct rq_flags *rf)
1624 __releases(rq->lock)
1626 rq_unpin_lock(rq, rf);
1627 raw_spin_rq_unlock(rq);
1630 static inline struct rq *
1631 this_rq_lock_irq(struct rq_flags *rf)
1632 __acquires(rq->lock)
1636 local_irq_disable();
1643 enum numa_topology_type {
1648 extern enum numa_topology_type sched_numa_topology_type;
1649 extern int sched_max_numa_distance;
1650 extern bool find_numa_distance(int distance);
1651 extern void sched_init_numa(int offline_node);
1652 extern void sched_update_numa(int cpu, bool online);
1653 extern void sched_domains_numa_masks_set(unsigned int cpu);
1654 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1655 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1657 static inline void sched_init_numa(int offline_node) { }
1658 static inline void sched_update_numa(int cpu, bool online) { }
1659 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1660 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1661 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1667 #ifdef CONFIG_NUMA_BALANCING
1668 /* The regions in numa_faults array from task_struct */
1669 enum numa_faults_stats {
1675 extern void sched_setnuma(struct task_struct *p, int node);
1676 extern int migrate_task_to(struct task_struct *p, int cpu);
1677 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1679 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1682 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1685 #endif /* CONFIG_NUMA_BALANCING */
1690 queue_balance_callback(struct rq *rq,
1691 struct callback_head *head,
1692 void (*func)(struct rq *rq))
1694 lockdep_assert_rq_held(rq);
1697 * Don't (re)queue an already queued item; nor queue anything when
1698 * balance_push() is active, see the comment with
1699 * balance_push_callback.
1701 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1704 head->func = (void (*)(struct callback_head *))func;
1705 head->next = rq->balance_callback;
1706 rq->balance_callback = head;
1709 #define rcu_dereference_check_sched_domain(p) \
1710 rcu_dereference_check((p), \
1711 lockdep_is_held(&sched_domains_mutex))
1714 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1715 * See destroy_sched_domains: call_rcu for details.
1717 * The domain tree of any CPU may only be accessed from within
1718 * preempt-disabled sections.
1720 #define for_each_domain(cpu, __sd) \
1721 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1722 __sd; __sd = __sd->parent)
1725 * highest_flag_domain - Return highest sched_domain containing flag.
1726 * @cpu: The CPU whose highest level of sched domain is to
1728 * @flag: The flag to check for the highest sched_domain
1729 * for the given CPU.
1731 * Returns the highest sched_domain of a CPU which contains the given flag.
1733 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1735 struct sched_domain *sd, *hsd = NULL;
1737 for_each_domain(cpu, sd) {
1738 if (!(sd->flags & flag))
1746 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1748 struct sched_domain *sd;
1750 for_each_domain(cpu, sd) {
1751 if (sd->flags & flag)
1758 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1759 DECLARE_PER_CPU(int, sd_llc_size);
1760 DECLARE_PER_CPU(int, sd_llc_id);
1761 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1762 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1763 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1764 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1765 extern struct static_key_false sched_asym_cpucapacity;
1767 struct sched_group_capacity {
1770 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1773 unsigned long capacity;
1774 unsigned long min_capacity; /* Min per-CPU capacity in group */
1775 unsigned long max_capacity; /* Max per-CPU capacity in group */
1776 unsigned long next_update;
1777 int imbalance; /* XXX unrelated to capacity but shared group state */
1779 #ifdef CONFIG_SCHED_DEBUG
1783 unsigned long cpumask[]; /* Balance mask */
1786 struct sched_group {
1787 struct sched_group *next; /* Must be a circular list */
1790 unsigned int group_weight;
1791 struct sched_group_capacity *sgc;
1792 int asym_prefer_cpu; /* CPU of highest priority in group */
1796 * The CPUs this group covers.
1798 * NOTE: this field is variable length. (Allocated dynamically
1799 * by attaching extra space to the end of the structure,
1800 * depending on how many CPUs the kernel has booted up with)
1802 unsigned long cpumask[];
1805 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1807 return to_cpumask(sg->cpumask);
1811 * See build_balance_mask().
1813 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1815 return to_cpumask(sg->sgc->cpumask);
1819 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1820 * @group: The group whose first CPU is to be returned.
1822 static inline unsigned int group_first_cpu(struct sched_group *group)
1824 return cpumask_first(sched_group_span(group));
1827 extern int group_balance_cpu(struct sched_group *sg);
1829 #ifdef CONFIG_SCHED_DEBUG
1830 void update_sched_domain_debugfs(void);
1831 void dirty_sched_domain_sysctl(int cpu);
1833 static inline void update_sched_domain_debugfs(void)
1836 static inline void dirty_sched_domain_sysctl(int cpu)
1841 extern int sched_update_scaling(void);
1842 #endif /* CONFIG_SMP */
1846 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1848 extern void __sched_core_account_forceidle(struct rq *rq);
1850 static inline void sched_core_account_forceidle(struct rq *rq)
1852 if (schedstat_enabled())
1853 __sched_core_account_forceidle(rq);
1856 extern void __sched_core_tick(struct rq *rq);
1858 static inline void sched_core_tick(struct rq *rq)
1860 if (sched_core_enabled(rq) && schedstat_enabled())
1861 __sched_core_tick(rq);
1866 static inline void sched_core_account_forceidle(struct rq *rq) {}
1868 static inline void sched_core_tick(struct rq *rq) {}
1870 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1872 #ifdef CONFIG_CGROUP_SCHED
1875 * Return the group to which this tasks belongs.
1877 * We cannot use task_css() and friends because the cgroup subsystem
1878 * changes that value before the cgroup_subsys::attach() method is called,
1879 * therefore we cannot pin it and might observe the wrong value.
1881 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1882 * core changes this before calling sched_move_task().
1884 * Instead we use a 'copy' which is updated from sched_move_task() while
1885 * holding both task_struct::pi_lock and rq::lock.
1887 static inline struct task_group *task_group(struct task_struct *p)
1889 return p->sched_task_group;
1892 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1893 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1895 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1896 struct task_group *tg = task_group(p);
1899 #ifdef CONFIG_FAIR_GROUP_SCHED
1900 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1901 p->se.cfs_rq = tg->cfs_rq[cpu];
1902 p->se.parent = tg->se[cpu];
1905 #ifdef CONFIG_RT_GROUP_SCHED
1906 p->rt.rt_rq = tg->rt_rq[cpu];
1907 p->rt.parent = tg->rt_se[cpu];
1911 #else /* CONFIG_CGROUP_SCHED */
1913 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1914 static inline struct task_group *task_group(struct task_struct *p)
1919 #endif /* CONFIG_CGROUP_SCHED */
1921 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1923 set_task_rq(p, cpu);
1926 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1927 * successfully executed on another CPU. We must ensure that updates of
1928 * per-task data have been completed by this moment.
1931 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1937 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1939 #ifdef CONFIG_SCHED_DEBUG
1940 # define const_debug __read_mostly
1942 # define const_debug const
1945 #define SCHED_FEAT(name, enabled) \
1946 __SCHED_FEAT_##name ,
1949 #include "features.h"
1955 #ifdef CONFIG_SCHED_DEBUG
1958 * To support run-time toggling of sched features, all the translation units
1959 * (but core.c) reference the sysctl_sched_features defined in core.c.
1961 extern const_debug unsigned int sysctl_sched_features;
1963 #ifdef CONFIG_JUMP_LABEL
1964 #define SCHED_FEAT(name, enabled) \
1965 static __always_inline bool static_branch_##name(struct static_key *key) \
1967 return static_key_##enabled(key); \
1970 #include "features.h"
1973 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1974 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1976 #else /* !CONFIG_JUMP_LABEL */
1978 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1980 #endif /* CONFIG_JUMP_LABEL */
1982 #else /* !SCHED_DEBUG */
1985 * Each translation unit has its own copy of sysctl_sched_features to allow
1986 * constants propagation at compile time and compiler optimization based on
1989 #define SCHED_FEAT(name, enabled) \
1990 (1UL << __SCHED_FEAT_##name) * enabled |
1991 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1992 #include "features.h"
1996 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1998 #endif /* SCHED_DEBUG */
2000 extern struct static_key_false sched_numa_balancing;
2001 extern struct static_key_false sched_schedstats;
2003 static inline u64 global_rt_period(void)
2005 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2008 static inline u64 global_rt_runtime(void)
2010 if (sysctl_sched_rt_runtime < 0)
2013 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2016 static inline int task_current(struct rq *rq, struct task_struct *p)
2018 return rq->curr == p;
2021 static inline int task_running(struct rq *rq, struct task_struct *p)
2026 return task_current(rq, p);
2030 static inline int task_on_rq_queued(struct task_struct *p)
2032 return p->on_rq == TASK_ON_RQ_QUEUED;
2035 static inline int task_on_rq_migrating(struct task_struct *p)
2037 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2040 /* Wake flags. The first three directly map to some SD flag value */
2041 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2042 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2043 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2045 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2046 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2049 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2050 static_assert(WF_FORK == SD_BALANCE_FORK);
2051 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2055 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2056 * of tasks with abnormal "nice" values across CPUs the contribution that
2057 * each task makes to its run queue's load is weighted according to its
2058 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2059 * scaled version of the new time slice allocation that they receive on time
2063 #define WEIGHT_IDLEPRIO 3
2064 #define WMULT_IDLEPRIO 1431655765
2066 extern const int sched_prio_to_weight[40];
2067 extern const u32 sched_prio_to_wmult[40];
2070 * {de,en}queue flags:
2072 * DEQUEUE_SLEEP - task is no longer runnable
2073 * ENQUEUE_WAKEUP - task just became runnable
2075 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2076 * are in a known state which allows modification. Such pairs
2077 * should preserve as much state as possible.
2079 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2082 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2083 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2084 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2088 #define DEQUEUE_SLEEP 0x01
2089 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2090 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2091 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2093 #define ENQUEUE_WAKEUP 0x01
2094 #define ENQUEUE_RESTORE 0x02
2095 #define ENQUEUE_MOVE 0x04
2096 #define ENQUEUE_NOCLOCK 0x08
2098 #define ENQUEUE_HEAD 0x10
2099 #define ENQUEUE_REPLENISH 0x20
2101 #define ENQUEUE_MIGRATED 0x40
2103 #define ENQUEUE_MIGRATED 0x00
2106 #define RETRY_TASK ((void *)-1UL)
2108 struct sched_class {
2110 #ifdef CONFIG_UCLAMP_TASK
2114 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2115 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2116 void (*yield_task) (struct rq *rq);
2117 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2119 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2121 struct task_struct *(*pick_next_task)(struct rq *rq);
2123 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2124 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2127 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2128 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2130 struct task_struct * (*pick_task)(struct rq *rq);
2132 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2134 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2136 void (*set_cpus_allowed)(struct task_struct *p,
2137 const struct cpumask *newmask,
2140 void (*rq_online)(struct rq *rq);
2141 void (*rq_offline)(struct rq *rq);
2143 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2146 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2147 void (*task_fork)(struct task_struct *p);
2148 void (*task_dead)(struct task_struct *p);
2151 * The switched_from() call is allowed to drop rq->lock, therefore we
2152 * cannot assume the switched_from/switched_to pair is serialized by
2153 * rq->lock. They are however serialized by p->pi_lock.
2155 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2156 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2157 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2160 unsigned int (*get_rr_interval)(struct rq *rq,
2161 struct task_struct *task);
2163 void (*update_curr)(struct rq *rq);
2165 #define TASK_SET_GROUP 0
2166 #define TASK_MOVE_GROUP 1
2168 #ifdef CONFIG_FAIR_GROUP_SCHED
2169 void (*task_change_group)(struct task_struct *p, int type);
2173 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2175 WARN_ON_ONCE(rq->curr != prev);
2176 prev->sched_class->put_prev_task(rq, prev);
2179 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2181 next->sched_class->set_next_task(rq, next, false);
2186 * Helper to define a sched_class instance; each one is placed in a separate
2187 * section which is ordered by the linker script:
2189 * include/asm-generic/vmlinux.lds.h
2191 * *CAREFUL* they are laid out in *REVERSE* order!!!
2193 * Also enforce alignment on the instance, not the type, to guarantee layout.
2195 #define DEFINE_SCHED_CLASS(name) \
2196 const struct sched_class name##_sched_class \
2197 __aligned(__alignof__(struct sched_class)) \
2198 __section("__" #name "_sched_class")
2200 /* Defined in include/asm-generic/vmlinux.lds.h */
2201 extern struct sched_class __sched_class_highest[];
2202 extern struct sched_class __sched_class_lowest[];
2204 #define for_class_range(class, _from, _to) \
2205 for (class = (_from); class < (_to); class++)
2207 #define for_each_class(class) \
2208 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2210 #define sched_class_above(_a, _b) ((_a) < (_b))
2212 extern const struct sched_class stop_sched_class;
2213 extern const struct sched_class dl_sched_class;
2214 extern const struct sched_class rt_sched_class;
2215 extern const struct sched_class fair_sched_class;
2216 extern const struct sched_class idle_sched_class;
2218 static inline bool sched_stop_runnable(struct rq *rq)
2220 return rq->stop && task_on_rq_queued(rq->stop);
2223 static inline bool sched_dl_runnable(struct rq *rq)
2225 return rq->dl.dl_nr_running > 0;
2228 static inline bool sched_rt_runnable(struct rq *rq)
2230 return rq->rt.rt_queued > 0;
2233 static inline bool sched_fair_runnable(struct rq *rq)
2235 return rq->cfs.nr_running > 0;
2238 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2239 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2241 #define SCA_CHECK 0x01
2242 #define SCA_MIGRATE_DISABLE 0x02
2243 #define SCA_MIGRATE_ENABLE 0x04
2244 #define SCA_USER 0x08
2248 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2250 extern void trigger_load_balance(struct rq *rq);
2252 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2254 static inline struct task_struct *get_push_task(struct rq *rq)
2256 struct task_struct *p = rq->curr;
2258 lockdep_assert_rq_held(rq);
2263 if (p->nr_cpus_allowed == 1)
2266 if (p->migration_disabled)
2269 rq->push_busy = true;
2270 return get_task_struct(p);
2273 extern int push_cpu_stop(void *arg);
2277 #ifdef CONFIG_CPU_IDLE
2278 static inline void idle_set_state(struct rq *rq,
2279 struct cpuidle_state *idle_state)
2281 rq->idle_state = idle_state;
2284 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2286 SCHED_WARN_ON(!rcu_read_lock_held());
2288 return rq->idle_state;
2291 static inline void idle_set_state(struct rq *rq,
2292 struct cpuidle_state *idle_state)
2296 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2302 extern void schedule_idle(void);
2304 extern void sysrq_sched_debug_show(void);
2305 extern void sched_init_granularity(void);
2306 extern void update_max_interval(void);
2308 extern void init_sched_dl_class(void);
2309 extern void init_sched_rt_class(void);
2310 extern void init_sched_fair_class(void);
2312 extern void reweight_task(struct task_struct *p, int prio);
2314 extern void resched_curr(struct rq *rq);
2315 extern void resched_cpu(int cpu);
2317 extern struct rt_bandwidth def_rt_bandwidth;
2318 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2319 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2321 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2322 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2323 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2326 #define BW_UNIT (1 << BW_SHIFT)
2327 #define RATIO_SHIFT 8
2328 #define MAX_BW_BITS (64 - BW_SHIFT)
2329 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2330 unsigned long to_ratio(u64 period, u64 runtime);
2332 extern void init_entity_runnable_average(struct sched_entity *se);
2333 extern void post_init_entity_util_avg(struct task_struct *p);
2335 #ifdef CONFIG_NO_HZ_FULL
2336 extern bool sched_can_stop_tick(struct rq *rq);
2337 extern int __init sched_tick_offload_init(void);
2340 * Tick may be needed by tasks in the runqueue depending on their policy and
2341 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2342 * nohz mode if necessary.
2344 static inline void sched_update_tick_dependency(struct rq *rq)
2346 int cpu = cpu_of(rq);
2348 if (!tick_nohz_full_cpu(cpu))
2351 if (sched_can_stop_tick(rq))
2352 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2354 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2357 static inline int sched_tick_offload_init(void) { return 0; }
2358 static inline void sched_update_tick_dependency(struct rq *rq) { }
2361 static inline void add_nr_running(struct rq *rq, unsigned count)
2363 unsigned prev_nr = rq->nr_running;
2365 rq->nr_running = prev_nr + count;
2366 if (trace_sched_update_nr_running_tp_enabled()) {
2367 call_trace_sched_update_nr_running(rq, count);
2371 if (prev_nr < 2 && rq->nr_running >= 2) {
2372 if (!READ_ONCE(rq->rd->overload))
2373 WRITE_ONCE(rq->rd->overload, 1);
2377 sched_update_tick_dependency(rq);
2380 static inline void sub_nr_running(struct rq *rq, unsigned count)
2382 rq->nr_running -= count;
2383 if (trace_sched_update_nr_running_tp_enabled()) {
2384 call_trace_sched_update_nr_running(rq, -count);
2387 /* Check if we still need preemption */
2388 sched_update_tick_dependency(rq);
2391 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2392 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2394 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2396 extern const_debug unsigned int sysctl_sched_nr_migrate;
2397 extern const_debug unsigned int sysctl_sched_migration_cost;
2399 #ifdef CONFIG_SCHED_DEBUG
2400 extern unsigned int sysctl_sched_latency;
2401 extern unsigned int sysctl_sched_min_granularity;
2402 extern unsigned int sysctl_sched_idle_min_granularity;
2403 extern unsigned int sysctl_sched_wakeup_granularity;
2404 extern int sysctl_resched_latency_warn_ms;
2405 extern int sysctl_resched_latency_warn_once;
2407 extern unsigned int sysctl_sched_tunable_scaling;
2409 extern unsigned int sysctl_numa_balancing_scan_delay;
2410 extern unsigned int sysctl_numa_balancing_scan_period_min;
2411 extern unsigned int sysctl_numa_balancing_scan_period_max;
2412 extern unsigned int sysctl_numa_balancing_scan_size;
2415 #ifdef CONFIG_SCHED_HRTICK
2419 * - enabled by features
2420 * - hrtimer is actually high res
2422 static inline int hrtick_enabled(struct rq *rq)
2424 if (!cpu_active(cpu_of(rq)))
2426 return hrtimer_is_hres_active(&rq->hrtick_timer);
2429 static inline int hrtick_enabled_fair(struct rq *rq)
2431 if (!sched_feat(HRTICK))
2433 return hrtick_enabled(rq);
2436 static inline int hrtick_enabled_dl(struct rq *rq)
2438 if (!sched_feat(HRTICK_DL))
2440 return hrtick_enabled(rq);
2443 void hrtick_start(struct rq *rq, u64 delay);
2447 static inline int hrtick_enabled_fair(struct rq *rq)
2452 static inline int hrtick_enabled_dl(struct rq *rq)
2457 static inline int hrtick_enabled(struct rq *rq)
2462 #endif /* CONFIG_SCHED_HRTICK */
2464 #ifndef arch_scale_freq_tick
2465 static __always_inline
2466 void arch_scale_freq_tick(void)
2471 #ifndef arch_scale_freq_capacity
2473 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2474 * @cpu: the CPU in question.
2476 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2479 * ------ * SCHED_CAPACITY_SCALE
2482 static __always_inline
2483 unsigned long arch_scale_freq_capacity(int cpu)
2485 return SCHED_CAPACITY_SCALE;
2489 #ifdef CONFIG_SCHED_DEBUG
2491 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2492 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2493 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2494 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2496 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2498 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2499 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2501 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2505 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2510 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2512 #ifdef CONFIG_SCHED_CORE
2514 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2515 * order by core-id first and cpu-id second.
2519 * double_rq_lock(0,3); will take core-0, core-1 lock
2520 * double_rq_lock(1,2); will take core-1, core-0 lock
2522 * when only cpu-id is considered.
2524 if (rq1->core->cpu < rq2->core->cpu)
2526 if (rq1->core->cpu > rq2->core->cpu)
2530 * __sched_core_flip() relies on SMT having cpu-id lock order.
2533 return rq1->cpu < rq2->cpu;
2536 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2538 #ifdef CONFIG_PREEMPTION
2541 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2542 * way at the expense of forcing extra atomic operations in all
2543 * invocations. This assures that the double_lock is acquired using the
2544 * same underlying policy as the spinlock_t on this architecture, which
2545 * reduces latency compared to the unfair variant below. However, it
2546 * also adds more overhead and therefore may reduce throughput.
2548 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2549 __releases(this_rq->lock)
2550 __acquires(busiest->lock)
2551 __acquires(this_rq->lock)
2553 raw_spin_rq_unlock(this_rq);
2554 double_rq_lock(this_rq, busiest);
2561 * Unfair double_lock_balance: Optimizes throughput at the expense of
2562 * latency by eliminating extra atomic operations when the locks are
2563 * already in proper order on entry. This favors lower CPU-ids and will
2564 * grant the double lock to lower CPUs over higher ids under contention,
2565 * regardless of entry order into the function.
2567 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2568 __releases(this_rq->lock)
2569 __acquires(busiest->lock)
2570 __acquires(this_rq->lock)
2572 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2573 likely(raw_spin_rq_trylock(busiest))) {
2574 double_rq_clock_clear_update(this_rq, busiest);
2578 if (rq_order_less(this_rq, busiest)) {
2579 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2580 double_rq_clock_clear_update(this_rq, busiest);
2584 raw_spin_rq_unlock(this_rq);
2585 double_rq_lock(this_rq, busiest);
2590 #endif /* CONFIG_PREEMPTION */
2593 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2595 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2597 lockdep_assert_irqs_disabled();
2599 return _double_lock_balance(this_rq, busiest);
2602 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2603 __releases(busiest->lock)
2605 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2606 raw_spin_rq_unlock(busiest);
2607 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2610 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2616 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2619 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2625 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2628 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2634 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2638 * double_rq_unlock - safely unlock two runqueues
2640 * Note this does not restore interrupts like task_rq_unlock,
2641 * you need to do so manually after calling.
2643 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2644 __releases(rq1->lock)
2645 __releases(rq2->lock)
2647 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2648 raw_spin_rq_unlock(rq2);
2650 __release(rq2->lock);
2651 raw_spin_rq_unlock(rq1);
2654 extern void set_rq_online (struct rq *rq);
2655 extern void set_rq_offline(struct rq *rq);
2656 extern bool sched_smp_initialized;
2658 #else /* CONFIG_SMP */
2661 * double_rq_lock - safely lock two runqueues
2663 * Note this does not disable interrupts like task_rq_lock,
2664 * you need to do so manually before calling.
2666 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2667 __acquires(rq1->lock)
2668 __acquires(rq2->lock)
2670 BUG_ON(!irqs_disabled());
2672 raw_spin_rq_lock(rq1);
2673 __acquire(rq2->lock); /* Fake it out ;) */
2674 double_rq_clock_clear_update(rq1, rq2);
2678 * double_rq_unlock - safely unlock two runqueues
2680 * Note this does not restore interrupts like task_rq_unlock,
2681 * you need to do so manually after calling.
2683 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2684 __releases(rq1->lock)
2685 __releases(rq2->lock)
2688 raw_spin_rq_unlock(rq1);
2689 __release(rq2->lock);
2694 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2695 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2697 #ifdef CONFIG_SCHED_DEBUG
2698 extern bool sched_debug_verbose;
2700 extern void print_cfs_stats(struct seq_file *m, int cpu);
2701 extern void print_rt_stats(struct seq_file *m, int cpu);
2702 extern void print_dl_stats(struct seq_file *m, int cpu);
2703 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2704 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2705 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2707 extern void resched_latency_warn(int cpu, u64 latency);
2708 #ifdef CONFIG_NUMA_BALANCING
2710 show_numa_stats(struct task_struct *p, struct seq_file *m);
2712 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2713 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2714 #endif /* CONFIG_NUMA_BALANCING */
2716 static inline void resched_latency_warn(int cpu, u64 latency) {}
2717 #endif /* CONFIG_SCHED_DEBUG */
2719 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2720 extern void init_rt_rq(struct rt_rq *rt_rq);
2721 extern void init_dl_rq(struct dl_rq *dl_rq);
2723 extern void cfs_bandwidth_usage_inc(void);
2724 extern void cfs_bandwidth_usage_dec(void);
2726 #ifdef CONFIG_NO_HZ_COMMON
2727 #define NOHZ_BALANCE_KICK_BIT 0
2728 #define NOHZ_STATS_KICK_BIT 1
2729 #define NOHZ_NEWILB_KICK_BIT 2
2730 #define NOHZ_NEXT_KICK_BIT 3
2732 /* Run rebalance_domains() */
2733 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2734 /* Update blocked load */
2735 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2736 /* Update blocked load when entering idle */
2737 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2738 /* Update nohz.next_balance */
2739 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2741 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2743 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2745 extern void nohz_balance_exit_idle(struct rq *rq);
2747 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2750 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2751 extern void nohz_run_idle_balance(int cpu);
2753 static inline void nohz_run_idle_balance(int cpu) { }
2756 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2761 struct u64_stats_sync sync;
2764 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2767 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2768 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2769 * and never move forward.
2771 static inline u64 irq_time_read(int cpu)
2773 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2778 seq = __u64_stats_fetch_begin(&irqtime->sync);
2779 total = irqtime->total;
2780 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2784 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2786 #ifdef CONFIG_CPU_FREQ
2787 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2790 * cpufreq_update_util - Take a note about CPU utilization changes.
2791 * @rq: Runqueue to carry out the update for.
2792 * @flags: Update reason flags.
2794 * This function is called by the scheduler on the CPU whose utilization is
2797 * It can only be called from RCU-sched read-side critical sections.
2799 * The way cpufreq is currently arranged requires it to evaluate the CPU
2800 * performance state (frequency/voltage) on a regular basis to prevent it from
2801 * being stuck in a completely inadequate performance level for too long.
2802 * That is not guaranteed to happen if the updates are only triggered from CFS
2803 * and DL, though, because they may not be coming in if only RT tasks are
2804 * active all the time (or there are RT tasks only).
2806 * As a workaround for that issue, this function is called periodically by the
2807 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2808 * but that really is a band-aid. Going forward it should be replaced with
2809 * solutions targeted more specifically at RT tasks.
2811 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2813 struct update_util_data *data;
2815 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2818 data->func(data, rq_clock(rq), flags);
2821 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2822 #endif /* CONFIG_CPU_FREQ */
2824 #ifdef arch_scale_freq_capacity
2825 # ifndef arch_scale_freq_invariant
2826 # define arch_scale_freq_invariant() true
2829 # define arch_scale_freq_invariant() false
2833 static inline unsigned long capacity_orig_of(int cpu)
2835 return cpu_rq(cpu)->cpu_capacity_orig;
2839 * enum cpu_util_type - CPU utilization type
2840 * @FREQUENCY_UTIL: Utilization used to select frequency
2841 * @ENERGY_UTIL: Utilization used during energy calculation
2843 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2844 * need to be aggregated differently depending on the usage made of them. This
2845 * enum is used within effective_cpu_util() to differentiate the types of
2846 * utilization expected by the callers, and adjust the aggregation accordingly.
2848 enum cpu_util_type {
2853 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2854 unsigned long max, enum cpu_util_type type,
2855 struct task_struct *p);
2857 static inline unsigned long cpu_bw_dl(struct rq *rq)
2859 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2862 static inline unsigned long cpu_util_dl(struct rq *rq)
2864 return READ_ONCE(rq->avg_dl.util_avg);
2868 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2869 * @cpu: the CPU to get the utilization for.
2871 * The unit of the return value must be the same as the one of CPU capacity
2872 * so that CPU utilization can be compared with CPU capacity.
2874 * CPU utilization is the sum of running time of runnable tasks plus the
2875 * recent utilization of currently non-runnable tasks on that CPU.
2876 * It represents the amount of CPU capacity currently used by CFS tasks in
2877 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2878 * capacity at f_max.
2880 * The estimated CPU utilization is defined as the maximum between CPU
2881 * utilization and sum of the estimated utilization of the currently
2882 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2883 * previously-executed tasks, which helps better deduce how busy a CPU will
2884 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2885 * of such a task would be significantly decayed at this point of time.
2887 * CPU utilization can be higher than the current CPU capacity
2888 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2889 * of rounding errors as well as task migrations or wakeups of new tasks.
2890 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2891 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2892 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2893 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2894 * though since this is useful for predicting the CPU capacity required
2895 * after task migrations (scheduler-driven DVFS).
2897 * Return: (Estimated) utilization for the specified CPU.
2899 static inline unsigned long cpu_util_cfs(int cpu)
2901 struct cfs_rq *cfs_rq;
2904 cfs_rq = &cpu_rq(cpu)->cfs;
2905 util = READ_ONCE(cfs_rq->avg.util_avg);
2907 if (sched_feat(UTIL_EST)) {
2908 util = max_t(unsigned long, util,
2909 READ_ONCE(cfs_rq->avg.util_est.enqueued));
2912 return min(util, capacity_orig_of(cpu));
2915 static inline unsigned long cpu_util_rt(struct rq *rq)
2917 return READ_ONCE(rq->avg_rt.util_avg);
2921 #ifdef CONFIG_UCLAMP_TASK
2922 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2925 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2926 * @rq: The rq to clamp against. Must not be NULL.
2927 * @util: The util value to clamp.
2928 * @p: The task to clamp against. Can be NULL if you want to clamp
2931 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2933 * If sched_uclamp_used static key is disabled, then just return the util
2934 * without any clamping since uclamp aggregation at the rq level in the fast
2935 * path is disabled, rendering this operation a NOP.
2937 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2938 * will return the correct effective uclamp value of the task even if the
2939 * static key is disabled.
2941 static __always_inline
2942 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2943 struct task_struct *p)
2945 unsigned long min_util = 0;
2946 unsigned long max_util = 0;
2948 if (!static_branch_likely(&sched_uclamp_used))
2952 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2953 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2956 * Ignore last runnable task's max clamp, as this task will
2957 * reset it. Similarly, no need to read the rq's min clamp.
2959 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2963 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2964 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2967 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2968 * RUNNABLE tasks with _different_ clamps, we can end up with an
2969 * inversion. Fix it now when the clamps are applied.
2971 if (unlikely(min_util >= max_util))
2974 return clamp(util, min_util, max_util);
2977 /* Is the rq being capped/throttled by uclamp_max? */
2978 static inline bool uclamp_rq_is_capped(struct rq *rq)
2980 unsigned long rq_util;
2981 unsigned long max_util;
2983 if (!static_branch_likely(&sched_uclamp_used))
2986 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
2987 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2989 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
2993 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2994 * by default in the fast path and only gets turned on once userspace performs
2995 * an operation that requires it.
2997 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3000 static inline bool uclamp_is_used(void)
3002 return static_branch_likely(&sched_uclamp_used);
3004 #else /* CONFIG_UCLAMP_TASK */
3006 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3007 struct task_struct *p)
3012 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3014 static inline bool uclamp_is_used(void)
3018 #endif /* CONFIG_UCLAMP_TASK */
3020 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3021 static inline unsigned long cpu_util_irq(struct rq *rq)
3023 return rq->avg_irq.util_avg;
3027 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3029 util *= (max - irq);
3036 static inline unsigned long cpu_util_irq(struct rq *rq)
3042 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3048 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3050 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3052 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3054 static inline bool sched_energy_enabled(void)
3056 return static_branch_unlikely(&sched_energy_present);
3059 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3061 #define perf_domain_span(pd) NULL
3062 static inline bool sched_energy_enabled(void) { return false; }
3064 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3066 #ifdef CONFIG_MEMBARRIER
3068 * The scheduler provides memory barriers required by membarrier between:
3069 * - prior user-space memory accesses and store to rq->membarrier_state,
3070 * - store to rq->membarrier_state and following user-space memory accesses.
3071 * In the same way it provides those guarantees around store to rq->curr.
3073 static inline void membarrier_switch_mm(struct rq *rq,
3074 struct mm_struct *prev_mm,
3075 struct mm_struct *next_mm)
3077 int membarrier_state;
3079 if (prev_mm == next_mm)
3082 membarrier_state = atomic_read(&next_mm->membarrier_state);
3083 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3086 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3089 static inline void membarrier_switch_mm(struct rq *rq,
3090 struct mm_struct *prev_mm,
3091 struct mm_struct *next_mm)
3097 static inline bool is_per_cpu_kthread(struct task_struct *p)
3099 if (!(p->flags & PF_KTHREAD))
3102 if (p->nr_cpus_allowed != 1)
3109 extern void swake_up_all_locked(struct swait_queue_head *q);
3110 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3112 #ifdef CONFIG_PREEMPT_DYNAMIC
3113 extern int preempt_dynamic_mode;
3114 extern int sched_dynamic_mode(const char *str);
3115 extern void sched_dynamic_update(int mode);
3118 #endif /* _KERNEL_SCHED_SCHED_H */