1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/context_tracking.h>
31 #include <linux/cpufreq.h>
32 #include <linux/cpumask_api.h>
33 #include <linux/ctype.h>
34 #include <linux/file.h>
35 #include <linux/fs_api.h>
36 #include <linux/hrtimer_api.h>
37 #include <linux/interrupt.h>
38 #include <linux/irq_work.h>
39 #include <linux/jiffies.h>
40 #include <linux/kref_api.h>
41 #include <linux/kthread.h>
42 #include <linux/ktime_api.h>
43 #include <linux/lockdep_api.h>
44 #include <linux/lockdep.h>
45 #include <linux/minmax.h>
47 #include <linux/module.h>
48 #include <linux/mutex_api.h>
49 #include <linux/plist.h>
50 #include <linux/poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/psi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/seq_file.h>
56 #include <linux/seqlock.h>
57 #include <linux/softirq.h>
58 #include <linux/spinlock_api.h>
59 #include <linux/static_key.h>
60 #include <linux/stop_machine.h>
61 #include <linux/syscalls_api.h>
62 #include <linux/syscalls.h>
63 #include <linux/tick.h>
64 #include <linux/topology.h>
65 #include <linux/types.h>
66 #include <linux/u64_stats_sync_api.h>
67 #include <linux/uaccess.h>
68 #include <linux/wait_api.h>
69 #include <linux/wait_bit.h>
70 #include <linux/workqueue_api.h>
72 #include <trace/events/power.h>
73 #include <trace/events/sched.h>
75 #include "../workqueue_internal.h"
77 #ifdef CONFIG_CGROUP_SCHED
78 #include <linux/cgroup.h>
79 #include <linux/psi.h>
82 #ifdef CONFIG_SCHED_DEBUG
83 # include <linux/static_key.h>
86 #ifdef CONFIG_PARAVIRT
87 # include <asm/paravirt.h>
88 # include <asm/paravirt_api_clock.h>
92 #include "cpudeadline.h"
94 #ifdef CONFIG_SCHED_DEBUG
95 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
97 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
101 struct cpuidle_state;
103 /* task_struct::on_rq states: */
104 #define TASK_ON_RQ_QUEUED 1
105 #define TASK_ON_RQ_MIGRATING 2
107 extern __read_mostly int scheduler_running;
109 extern unsigned long calc_load_update;
110 extern atomic_long_t calc_load_tasks;
112 extern unsigned int sysctl_sched_child_runs_first;
114 extern void calc_global_load_tick(struct rq *this_rq);
115 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
117 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
119 extern unsigned int sysctl_sched_rt_period;
120 extern int sysctl_sched_rt_runtime;
121 extern int sched_rr_timeslice;
124 * Helpers for converting nanosecond timing to jiffy resolution
126 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
129 * Increase resolution of nice-level calculations for 64-bit architectures.
130 * The extra resolution improves shares distribution and load balancing of
131 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
132 * hierarchies, especially on larger systems. This is not a user-visible change
133 * and does not change the user-interface for setting shares/weights.
135 * We increase resolution only if we have enough bits to allow this increased
136 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
137 * are pretty high and the returns do not justify the increased costs.
139 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
140 * increase coverage and consistency always enable it on 64-bit platforms.
143 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
144 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
145 # define scale_load_down(w) \
147 unsigned long __w = (w); \
149 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
153 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
154 # define scale_load(w) (w)
155 # define scale_load_down(w) (w)
159 * Task weight (visible to users) and its load (invisible to users) have
160 * independent resolution, but they should be well calibrated. We use
161 * scale_load() and scale_load_down(w) to convert between them. The
162 * following must be true:
164 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
167 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
170 * Single value that decides SCHED_DEADLINE internal math precision.
171 * 10 -> just above 1us
172 * 9 -> just above 0.5us
177 * Single value that denotes runtime == period, ie unlimited time.
179 #define RUNTIME_INF ((u64)~0ULL)
181 static inline int idle_policy(int policy)
183 return policy == SCHED_IDLE;
185 static inline int fair_policy(int policy)
187 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
190 static inline int rt_policy(int policy)
192 return policy == SCHED_FIFO || policy == SCHED_RR;
195 static inline int dl_policy(int policy)
197 return policy == SCHED_DEADLINE;
199 static inline bool valid_policy(int policy)
201 return idle_policy(policy) || fair_policy(policy) ||
202 rt_policy(policy) || dl_policy(policy);
205 static inline int task_has_idle_policy(struct task_struct *p)
207 return idle_policy(p->policy);
210 static inline int task_has_rt_policy(struct task_struct *p)
212 return rt_policy(p->policy);
215 static inline int task_has_dl_policy(struct task_struct *p)
217 return dl_policy(p->policy);
220 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
222 static inline void update_avg(u64 *avg, u64 sample)
224 s64 diff = sample - *avg;
229 * Shifting a value by an exponent greater *or equal* to the size of said value
230 * is UB; cap at size-1.
232 #define shr_bound(val, shift) \
233 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
236 * !! For sched_setattr_nocheck() (kernel) only !!
238 * This is actually gross. :(
240 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
241 * tasks, but still be able to sleep. We need this on platforms that cannot
242 * atomically change clock frequency. Remove once fast switching will be
243 * available on such platforms.
245 * SUGOV stands for SchedUtil GOVernor.
247 #define SCHED_FLAG_SUGOV 0x10000000
249 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
251 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
253 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
254 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
261 * Tells if entity @a should preempt entity @b.
264 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
266 return dl_entity_is_special(a) ||
267 dl_time_before(a->deadline, b->deadline);
271 * This is the priority-queue data structure of the RT scheduling class:
273 struct rt_prio_array {
274 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
275 struct list_head queue[MAX_RT_PRIO];
278 struct rt_bandwidth {
279 /* nests inside the rq lock: */
280 raw_spinlock_t rt_runtime_lock;
283 struct hrtimer rt_period_timer;
284 unsigned int rt_period_active;
287 void __dl_clear_params(struct task_struct *p);
289 struct dl_bandwidth {
290 raw_spinlock_t dl_runtime_lock;
295 static inline int dl_bandwidth_enabled(void)
297 return sysctl_sched_rt_runtime >= 0;
301 * To keep the bandwidth of -deadline tasks under control
302 * we need some place where:
303 * - store the maximum -deadline bandwidth of each cpu;
304 * - cache the fraction of bandwidth that is currently allocated in
307 * This is all done in the data structure below. It is similar to the
308 * one used for RT-throttling (rt_bandwidth), with the main difference
309 * that, since here we are only interested in admission control, we
310 * do not decrease any runtime while the group "executes", neither we
311 * need a timer to replenish it.
313 * With respect to SMP, bandwidth is given on a per root domain basis,
315 * - bw (< 100%) is the deadline bandwidth of each CPU;
316 * - total_bw is the currently allocated bandwidth in each root domain;
325 * Verify the fitness of task @p to run on @cpu taking into account the
326 * CPU original capacity and the runtime/deadline ratio of the task.
328 * The function will return true if the CPU original capacity of the
329 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
330 * task and false otherwise.
332 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
334 unsigned long cap = arch_scale_cpu_capacity(cpu);
336 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
339 extern void init_dl_bw(struct dl_bw *dl_b);
340 extern int sched_dl_global_validate(void);
341 extern void sched_dl_do_global(void);
342 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
343 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
344 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
345 extern bool __checkparam_dl(const struct sched_attr *attr);
346 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
347 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
348 extern int dl_cpu_busy(int cpu, struct task_struct *p);
350 #ifdef CONFIG_CGROUP_SCHED
355 extern struct list_head task_groups;
357 struct cfs_bandwidth {
358 #ifdef CONFIG_CFS_BANDWIDTH
365 s64 hierarchical_quota;
370 struct hrtimer period_timer;
371 struct hrtimer slack_timer;
372 struct list_head throttled_cfs_rq;
383 /* Task group related information */
385 struct cgroup_subsys_state css;
387 #ifdef CONFIG_FAIR_GROUP_SCHED
388 /* schedulable entities of this group on each CPU */
389 struct sched_entity **se;
390 /* runqueue "owned" by this group on each CPU */
391 struct cfs_rq **cfs_rq;
392 unsigned long shares;
394 /* A positive value indicates that this is a SCHED_IDLE group. */
399 * load_avg can be heavily contended at clock tick time, so put
400 * it in its own cacheline separated from the fields above which
401 * will also be accessed at each tick.
403 atomic_long_t load_avg ____cacheline_aligned;
407 #ifdef CONFIG_RT_GROUP_SCHED
408 struct sched_rt_entity **rt_se;
409 struct rt_rq **rt_rq;
411 struct rt_bandwidth rt_bandwidth;
415 struct list_head list;
417 struct task_group *parent;
418 struct list_head siblings;
419 struct list_head children;
421 #ifdef CONFIG_SCHED_AUTOGROUP
422 struct autogroup *autogroup;
425 struct cfs_bandwidth cfs_bandwidth;
427 #ifdef CONFIG_UCLAMP_TASK_GROUP
428 /* The two decimal precision [%] value requested from user-space */
429 unsigned int uclamp_pct[UCLAMP_CNT];
430 /* Clamp values requested for a task group */
431 struct uclamp_se uclamp_req[UCLAMP_CNT];
432 /* Effective clamp values used for a task group */
433 struct uclamp_se uclamp[UCLAMP_CNT];
438 #ifdef CONFIG_FAIR_GROUP_SCHED
439 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
442 * A weight of 0 or 1 can cause arithmetics problems.
443 * A weight of a cfs_rq is the sum of weights of which entities
444 * are queued on this cfs_rq, so a weight of a entity should not be
445 * too large, so as the shares value of a task group.
446 * (The default weight is 1024 - so there's no practical
447 * limitation from this.)
449 #define MIN_SHARES (1UL << 1)
450 #define MAX_SHARES (1UL << 18)
453 typedef int (*tg_visitor)(struct task_group *, void *);
455 extern int walk_tg_tree_from(struct task_group *from,
456 tg_visitor down, tg_visitor up, void *data);
459 * Iterate the full tree, calling @down when first entering a node and @up when
460 * leaving it for the final time.
462 * Caller must hold rcu_lock or sufficient equivalent.
464 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
466 return walk_tg_tree_from(&root_task_group, down, up, data);
469 extern int tg_nop(struct task_group *tg, void *data);
471 extern void free_fair_sched_group(struct task_group *tg);
472 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
473 extern void online_fair_sched_group(struct task_group *tg);
474 extern void unregister_fair_sched_group(struct task_group *tg);
475 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
476 struct sched_entity *se, int cpu,
477 struct sched_entity *parent);
478 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
480 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
481 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
482 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
484 extern void unregister_rt_sched_group(struct task_group *tg);
485 extern void free_rt_sched_group(struct task_group *tg);
486 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
487 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
488 struct sched_rt_entity *rt_se, int cpu,
489 struct sched_rt_entity *parent);
490 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
491 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
492 extern long sched_group_rt_runtime(struct task_group *tg);
493 extern long sched_group_rt_period(struct task_group *tg);
494 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
496 extern struct task_group *sched_create_group(struct task_group *parent);
497 extern void sched_online_group(struct task_group *tg,
498 struct task_group *parent);
499 extern void sched_destroy_group(struct task_group *tg);
500 extern void sched_release_group(struct task_group *tg);
502 extern void sched_move_task(struct task_struct *tsk);
504 #ifdef CONFIG_FAIR_GROUP_SCHED
505 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
507 extern int sched_group_set_idle(struct task_group *tg, long idle);
510 extern void set_task_rq_fair(struct sched_entity *se,
511 struct cfs_rq *prev, struct cfs_rq *next);
512 #else /* !CONFIG_SMP */
513 static inline void set_task_rq_fair(struct sched_entity *se,
514 struct cfs_rq *prev, struct cfs_rq *next) { }
515 #endif /* CONFIG_SMP */
516 #endif /* CONFIG_FAIR_GROUP_SCHED */
518 #else /* CONFIG_CGROUP_SCHED */
520 struct cfs_bandwidth { };
522 #endif /* CONFIG_CGROUP_SCHED */
524 /* CFS-related fields in a runqueue */
526 struct load_weight load;
527 unsigned int nr_running;
528 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
529 unsigned int idle_nr_running; /* SCHED_IDLE */
530 unsigned int idle_h_nr_running; /* SCHED_IDLE */
534 #ifdef CONFIG_SCHED_CORE
535 unsigned int forceidle_seq;
540 u64 min_vruntime_copy;
543 struct rb_root_cached tasks_timeline;
546 * 'curr' points to currently running entity on this cfs_rq.
547 * It is set to NULL otherwise (i.e when none are currently running).
549 struct sched_entity *curr;
550 struct sched_entity *next;
551 struct sched_entity *last;
552 struct sched_entity *skip;
554 #ifdef CONFIG_SCHED_DEBUG
555 unsigned int nr_spread_over;
562 struct sched_avg avg;
564 u64 load_last_update_time_copy;
567 raw_spinlock_t lock ____cacheline_aligned;
569 unsigned long load_avg;
570 unsigned long util_avg;
571 unsigned long runnable_avg;
574 #ifdef CONFIG_FAIR_GROUP_SCHED
575 unsigned long tg_load_avg_contrib;
577 long prop_runnable_sum;
580 * h_load = weight * f(tg)
582 * Where f(tg) is the recursive weight fraction assigned to
585 unsigned long h_load;
586 u64 last_h_load_update;
587 struct sched_entity *h_load_next;
588 #endif /* CONFIG_FAIR_GROUP_SCHED */
589 #endif /* CONFIG_SMP */
591 #ifdef CONFIG_FAIR_GROUP_SCHED
592 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
595 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
596 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
597 * (like users, containers etc.)
599 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
600 * This list is used during load balance.
603 struct list_head leaf_cfs_rq_list;
604 struct task_group *tg; /* group that "owns" this runqueue */
606 /* Locally cached copy of our task_group's idle value */
609 #ifdef CONFIG_CFS_BANDWIDTH
611 s64 runtime_remaining;
614 u64 throttled_clock_pelt;
615 u64 throttled_clock_pelt_time;
618 struct list_head throttled_list;
619 #endif /* CONFIG_CFS_BANDWIDTH */
620 #endif /* CONFIG_FAIR_GROUP_SCHED */
623 static inline int rt_bandwidth_enabled(void)
625 return sysctl_sched_rt_runtime >= 0;
628 /* RT IPI pull logic requires IRQ_WORK */
629 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
630 # define HAVE_RT_PUSH_IPI
633 /* Real-Time classes' related field in a runqueue: */
635 struct rt_prio_array active;
636 unsigned int rt_nr_running;
637 unsigned int rr_nr_running;
638 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
640 int curr; /* highest queued rt task prio */
642 int next; /* next highest */
647 unsigned int rt_nr_migratory;
648 unsigned int rt_nr_total;
650 struct plist_head pushable_tasks;
652 #endif /* CONFIG_SMP */
658 /* Nests inside the rq lock: */
659 raw_spinlock_t rt_runtime_lock;
661 #ifdef CONFIG_RT_GROUP_SCHED
662 unsigned int rt_nr_boosted;
665 struct task_group *tg;
669 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
671 return rt_rq->rt_queued && rt_rq->rt_nr_running;
674 /* Deadline class' related fields in a runqueue */
676 /* runqueue is an rbtree, ordered by deadline */
677 struct rb_root_cached root;
679 unsigned int dl_nr_running;
683 * Deadline values of the currently executing and the
684 * earliest ready task on this rq. Caching these facilitates
685 * the decision whether or not a ready but not running task
686 * should migrate somewhere else.
693 unsigned int dl_nr_migratory;
697 * Tasks on this rq that can be pushed away. They are kept in
698 * an rb-tree, ordered by tasks' deadlines, with caching
699 * of the leftmost (earliest deadline) element.
701 struct rb_root_cached pushable_dl_tasks_root;
706 * "Active utilization" for this runqueue: increased when a
707 * task wakes up (becomes TASK_RUNNING) and decreased when a
713 * Utilization of the tasks "assigned" to this runqueue (including
714 * the tasks that are in runqueue and the tasks that executed on this
715 * CPU and blocked). Increased when a task moves to this runqueue, and
716 * decreased when the task moves away (migrates, changes scheduling
717 * policy, or terminates).
718 * This is needed to compute the "inactive utilization" for the
719 * runqueue (inactive utilization = this_bw - running_bw).
725 * Inverse of the fraction of CPU utilization that can be reclaimed
726 * by the GRUB algorithm.
731 #ifdef CONFIG_FAIR_GROUP_SCHED
732 /* An entity is a task if it doesn't "own" a runqueue */
733 #define entity_is_task(se) (!se->my_q)
735 static inline void se_update_runnable(struct sched_entity *se)
737 if (!entity_is_task(se))
738 se->runnable_weight = se->my_q->h_nr_running;
741 static inline long se_runnable(struct sched_entity *se)
743 if (entity_is_task(se))
746 return se->runnable_weight;
750 #define entity_is_task(se) 1
752 static inline void se_update_runnable(struct sched_entity *se) {}
754 static inline long se_runnable(struct sched_entity *se)
762 * XXX we want to get rid of these helpers and use the full load resolution.
764 static inline long se_weight(struct sched_entity *se)
766 return scale_load_down(se->load.weight);
770 static inline bool sched_asym_prefer(int a, int b)
772 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
776 struct em_perf_domain *em_pd;
777 struct perf_domain *next;
781 /* Scheduling group status flags */
782 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
783 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
786 * We add the notion of a root-domain which will be used to define per-domain
787 * variables. Each exclusive cpuset essentially defines an island domain by
788 * fully partitioning the member CPUs from any other cpuset. Whenever a new
789 * exclusive cpuset is created, we also create and attach a new root-domain
798 cpumask_var_t online;
801 * Indicate pullable load on at least one CPU, e.g:
802 * - More than one runnable task
803 * - Running task is misfit
807 /* Indicate one or more cpus over-utilized (tipping point) */
811 * The bit corresponding to a CPU gets set here if such CPU has more
812 * than one runnable -deadline task (as it is below for RT tasks).
814 cpumask_var_t dlo_mask;
820 * Indicate whether a root_domain's dl_bw has been checked or
821 * updated. It's monotonously increasing value.
823 * Also, some corner cases, like 'wrap around' is dangerous, but given
824 * that u64 is 'big enough'. So that shouldn't be a concern.
828 #ifdef HAVE_RT_PUSH_IPI
830 * For IPI pull requests, loop across the rto_mask.
832 struct irq_work rto_push_work;
833 raw_spinlock_t rto_lock;
834 /* These are only updated and read within rto_lock */
837 /* These atomics are updated outside of a lock */
838 atomic_t rto_loop_next;
839 atomic_t rto_loop_start;
842 * The "RT overload" flag: it gets set if a CPU has more than
843 * one runnable RT task.
845 cpumask_var_t rto_mask;
846 struct cpupri cpupri;
848 unsigned long max_cpu_capacity;
851 * NULL-terminated list of performance domains intersecting with the
852 * CPUs of the rd. Protected by RCU.
854 struct perf_domain __rcu *pd;
857 extern void init_defrootdomain(void);
858 extern int sched_init_domains(const struct cpumask *cpu_map);
859 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
860 extern void sched_get_rd(struct root_domain *rd);
861 extern void sched_put_rd(struct root_domain *rd);
863 #ifdef HAVE_RT_PUSH_IPI
864 extern void rto_push_irq_work_func(struct irq_work *work);
866 #endif /* CONFIG_SMP */
868 #ifdef CONFIG_UCLAMP_TASK
870 * struct uclamp_bucket - Utilization clamp bucket
871 * @value: utilization clamp value for tasks on this clamp bucket
872 * @tasks: number of RUNNABLE tasks on this clamp bucket
874 * Keep track of how many tasks are RUNNABLE for a given utilization
877 struct uclamp_bucket {
878 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
879 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
883 * struct uclamp_rq - rq's utilization clamp
884 * @value: currently active clamp values for a rq
885 * @bucket: utilization clamp buckets affecting a rq
887 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
888 * A clamp value is affecting a rq when there is at least one task RUNNABLE
889 * (or actually running) with that value.
891 * There are up to UCLAMP_CNT possible different clamp values, currently there
892 * are only two: minimum utilization and maximum utilization.
894 * All utilization clamping values are MAX aggregated, since:
895 * - for util_min: we want to run the CPU at least at the max of the minimum
896 * utilization required by its currently RUNNABLE tasks.
897 * - for util_max: we want to allow the CPU to run up to the max of the
898 * maximum utilization allowed by its currently RUNNABLE tasks.
900 * Since on each system we expect only a limited number of different
901 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
902 * the metrics required to compute all the per-rq utilization clamp values.
906 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
909 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
910 #endif /* CONFIG_UCLAMP_TASK */
913 * This is the main, per-CPU runqueue data structure.
915 * Locking rule: those places that want to lock multiple runqueues
916 * (such as the load balancing or the thread migration code), lock
917 * acquire operations must be ordered by ascending &runqueue.
921 raw_spinlock_t __lock;
924 * nr_running and cpu_load should be in the same cacheline because
925 * remote CPUs use both these fields when doing load calculation.
927 unsigned int nr_running;
928 #ifdef CONFIG_NUMA_BALANCING
929 unsigned int nr_numa_running;
930 unsigned int nr_preferred_running;
931 unsigned int numa_migrate_on;
933 #ifdef CONFIG_NO_HZ_COMMON
935 unsigned long last_blocked_load_update_tick;
936 unsigned int has_blocked_load;
937 call_single_data_t nohz_csd;
938 #endif /* CONFIG_SMP */
939 unsigned int nohz_tick_stopped;
941 #endif /* CONFIG_NO_HZ_COMMON */
944 unsigned int ttwu_pending;
948 #ifdef CONFIG_UCLAMP_TASK
949 /* Utilization clamp values based on CPU's RUNNABLE tasks */
950 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
951 unsigned int uclamp_flags;
952 #define UCLAMP_FLAG_IDLE 0x01
959 #ifdef CONFIG_FAIR_GROUP_SCHED
960 /* list of leaf cfs_rq on this CPU: */
961 struct list_head leaf_cfs_rq_list;
962 struct list_head *tmp_alone_branch;
963 #endif /* CONFIG_FAIR_GROUP_SCHED */
966 * This is part of a global counter where only the total sum
967 * over all CPUs matters. A task can increase this counter on
968 * one CPU and if it got migrated afterwards it may decrease
969 * it on another CPU. Always updated under the runqueue lock:
971 unsigned int nr_uninterruptible;
973 struct task_struct __rcu *curr;
974 struct task_struct *idle;
975 struct task_struct *stop;
976 unsigned long next_balance;
977 struct mm_struct *prev_mm;
979 unsigned int clock_update_flags;
981 /* Ensure that all clocks are in the same cache line */
982 u64 clock_task ____cacheline_aligned;
984 unsigned long lost_idle_time;
988 #ifdef CONFIG_SCHED_DEBUG
989 u64 last_seen_need_resched_ns;
990 int ticks_without_resched;
993 #ifdef CONFIG_MEMBARRIER
994 int membarrier_state;
998 struct root_domain *rd;
999 struct sched_domain __rcu *sd;
1001 unsigned long cpu_capacity;
1002 unsigned long cpu_capacity_orig;
1004 struct callback_head *balance_callback;
1006 unsigned char nohz_idle_balance;
1007 unsigned char idle_balance;
1009 unsigned long misfit_task_load;
1011 /* For active balancing */
1014 struct cpu_stop_work active_balance_work;
1016 /* CPU of this runqueue: */
1020 struct list_head cfs_tasks;
1022 struct sched_avg avg_rt;
1023 struct sched_avg avg_dl;
1024 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1025 struct sched_avg avg_irq;
1027 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1028 struct sched_avg avg_thermal;
1033 unsigned long wake_stamp;
1036 /* This is used to determine avg_idle's max value */
1037 u64 max_idle_balance_cost;
1039 #ifdef CONFIG_HOTPLUG_CPU
1040 struct rcuwait hotplug_wait;
1042 #endif /* CONFIG_SMP */
1044 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1047 #ifdef CONFIG_PARAVIRT
1048 u64 prev_steal_time;
1050 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1051 u64 prev_steal_time_rq;
1054 /* calc_load related fields */
1055 unsigned long calc_load_update;
1056 long calc_load_active;
1058 #ifdef CONFIG_SCHED_HRTICK
1060 call_single_data_t hrtick_csd;
1062 struct hrtimer hrtick_timer;
1063 ktime_t hrtick_time;
1066 #ifdef CONFIG_SCHEDSTATS
1068 struct sched_info rq_sched_info;
1069 unsigned long long rq_cpu_time;
1070 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1072 /* sys_sched_yield() stats */
1073 unsigned int yld_count;
1075 /* schedule() stats */
1076 unsigned int sched_count;
1077 unsigned int sched_goidle;
1079 /* try_to_wake_up() stats */
1080 unsigned int ttwu_count;
1081 unsigned int ttwu_local;
1084 #ifdef CONFIG_CPU_IDLE
1085 /* Must be inspected within a rcu lock section */
1086 struct cpuidle_state *idle_state;
1090 unsigned int nr_pinned;
1092 unsigned int push_busy;
1093 struct cpu_stop_work push_work;
1095 #ifdef CONFIG_SCHED_CORE
1098 struct task_struct *core_pick;
1099 unsigned int core_enabled;
1100 unsigned int core_sched_seq;
1101 struct rb_root core_tree;
1103 /* shared state -- careful with sched_core_cpu_deactivate() */
1104 unsigned int core_task_seq;
1105 unsigned int core_pick_seq;
1106 unsigned long core_cookie;
1107 unsigned int core_forceidle_count;
1108 unsigned int core_forceidle_seq;
1109 unsigned int core_forceidle_occupation;
1110 u64 core_forceidle_start;
1114 #ifdef CONFIG_FAIR_GROUP_SCHED
1116 /* CPU runqueue to which this cfs_rq is attached */
1117 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1124 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1126 return container_of(cfs_rq, struct rq, cfs);
1130 static inline int cpu_of(struct rq *rq)
1139 #define MDF_PUSH 0x01
1141 static inline bool is_migration_disabled(struct task_struct *p)
1144 return p->migration_disabled;
1151 #ifdef CONFIG_SCHED_CORE
1152 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1154 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1156 static inline bool sched_core_enabled(struct rq *rq)
1158 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1161 static inline bool sched_core_disabled(void)
1163 return !static_branch_unlikely(&__sched_core_enabled);
1167 * Be careful with this function; not for general use. The return value isn't
1168 * stable unless you actually hold a relevant rq->__lock.
1170 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1172 if (sched_core_enabled(rq))
1173 return &rq->core->__lock;
1178 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1180 if (rq->core_enabled)
1181 return &rq->core->__lock;
1186 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1189 * Helpers to check if the CPU's core cookie matches with the task's cookie
1190 * when core scheduling is enabled.
1191 * A special case is that the task's cookie always matches with CPU's core
1192 * cookie if the CPU is in an idle core.
1194 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1196 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1197 if (!sched_core_enabled(rq))
1200 return rq->core->core_cookie == p->core_cookie;
1203 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1205 bool idle_core = true;
1208 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1209 if (!sched_core_enabled(rq))
1212 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1213 if (!available_idle_cpu(cpu)) {
1220 * A CPU in an idle core is always the best choice for tasks with
1223 return idle_core || rq->core->core_cookie == p->core_cookie;
1226 static inline bool sched_group_cookie_match(struct rq *rq,
1227 struct task_struct *p,
1228 struct sched_group *group)
1232 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1233 if (!sched_core_enabled(rq))
1236 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1237 if (sched_core_cookie_match(rq, p))
1243 static inline bool sched_core_enqueued(struct task_struct *p)
1245 return !RB_EMPTY_NODE(&p->core_node);
1248 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1249 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1251 extern void sched_core_get(void);
1252 extern void sched_core_put(void);
1254 #else /* !CONFIG_SCHED_CORE */
1256 static inline bool sched_core_enabled(struct rq *rq)
1261 static inline bool sched_core_disabled(void)
1266 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1271 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1276 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1281 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1286 static inline bool sched_group_cookie_match(struct rq *rq,
1287 struct task_struct *p,
1288 struct sched_group *group)
1292 #endif /* CONFIG_SCHED_CORE */
1294 static inline void lockdep_assert_rq_held(struct rq *rq)
1296 lockdep_assert_held(__rq_lockp(rq));
1299 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1300 extern bool raw_spin_rq_trylock(struct rq *rq);
1301 extern void raw_spin_rq_unlock(struct rq *rq);
1303 static inline void raw_spin_rq_lock(struct rq *rq)
1305 raw_spin_rq_lock_nested(rq, 0);
1308 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1310 local_irq_disable();
1311 raw_spin_rq_lock(rq);
1314 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1316 raw_spin_rq_unlock(rq);
1320 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1322 unsigned long flags;
1323 local_irq_save(flags);
1324 raw_spin_rq_lock(rq);
1328 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1330 raw_spin_rq_unlock(rq);
1331 local_irq_restore(flags);
1334 #define raw_spin_rq_lock_irqsave(rq, flags) \
1336 flags = _raw_spin_rq_lock_irqsave(rq); \
1339 #ifdef CONFIG_SCHED_SMT
1340 extern void __update_idle_core(struct rq *rq);
1342 static inline void update_idle_core(struct rq *rq)
1344 if (static_branch_unlikely(&sched_smt_present))
1345 __update_idle_core(rq);
1349 static inline void update_idle_core(struct rq *rq) { }
1352 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1354 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1355 #define this_rq() this_cpu_ptr(&runqueues)
1356 #define task_rq(p) cpu_rq(task_cpu(p))
1357 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1358 #define raw_rq() raw_cpu_ptr(&runqueues)
1360 #ifdef CONFIG_FAIR_GROUP_SCHED
1361 static inline struct task_struct *task_of(struct sched_entity *se)
1363 SCHED_WARN_ON(!entity_is_task(se));
1364 return container_of(se, struct task_struct, se);
1367 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1369 return p->se.cfs_rq;
1372 /* runqueue on which this entity is (to be) queued */
1373 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1378 /* runqueue "owned" by this group */
1379 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1386 static inline struct task_struct *task_of(struct sched_entity *se)
1388 return container_of(se, struct task_struct, se);
1391 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1393 return &task_rq(p)->cfs;
1396 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1398 struct task_struct *p = task_of(se);
1399 struct rq *rq = task_rq(p);
1404 /* runqueue "owned" by this group */
1405 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1411 extern void update_rq_clock(struct rq *rq);
1414 * rq::clock_update_flags bits
1416 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1417 * call to __schedule(). This is an optimisation to avoid
1418 * neighbouring rq clock updates.
1420 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1421 * in effect and calls to update_rq_clock() are being ignored.
1423 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1424 * made to update_rq_clock() since the last time rq::lock was pinned.
1426 * If inside of __schedule(), clock_update_flags will have been
1427 * shifted left (a left shift is a cheap operation for the fast path
1428 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1430 * if (rq-clock_update_flags >= RQCF_UPDATED)
1432 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1433 * one position though, because the next rq_unpin_lock() will shift it
1436 #define RQCF_REQ_SKIP 0x01
1437 #define RQCF_ACT_SKIP 0x02
1438 #define RQCF_UPDATED 0x04
1440 static inline void assert_clock_updated(struct rq *rq)
1443 * The only reason for not seeing a clock update since the
1444 * last rq_pin_lock() is if we're currently skipping updates.
1446 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1449 static inline u64 rq_clock(struct rq *rq)
1451 lockdep_assert_rq_held(rq);
1452 assert_clock_updated(rq);
1457 static inline u64 rq_clock_task(struct rq *rq)
1459 lockdep_assert_rq_held(rq);
1460 assert_clock_updated(rq);
1462 return rq->clock_task;
1466 * By default the decay is the default pelt decay period.
1467 * The decay shift can change the decay period in
1469 * Decay shift Decay period(ms)
1476 extern int sched_thermal_decay_shift;
1478 static inline u64 rq_clock_thermal(struct rq *rq)
1480 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1483 static inline void rq_clock_skip_update(struct rq *rq)
1485 lockdep_assert_rq_held(rq);
1486 rq->clock_update_flags |= RQCF_REQ_SKIP;
1490 * See rt task throttling, which is the only time a skip
1491 * request is canceled.
1493 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1495 lockdep_assert_rq_held(rq);
1496 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1500 unsigned long flags;
1501 struct pin_cookie cookie;
1502 #ifdef CONFIG_SCHED_DEBUG
1504 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1505 * current pin context is stashed here in case it needs to be
1506 * restored in rq_repin_lock().
1508 unsigned int clock_update_flags;
1512 extern struct callback_head balance_push_callback;
1515 * Lockdep annotation that avoids accidental unlocks; it's like a
1516 * sticky/continuous lockdep_assert_held().
1518 * This avoids code that has access to 'struct rq *rq' (basically everything in
1519 * the scheduler) from accidentally unlocking the rq if they do not also have a
1520 * copy of the (on-stack) 'struct rq_flags rf'.
1522 * Also see Documentation/locking/lockdep-design.rst.
1524 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1526 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1528 #ifdef CONFIG_SCHED_DEBUG
1529 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1530 rf->clock_update_flags = 0;
1532 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1537 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1539 #ifdef CONFIG_SCHED_DEBUG
1540 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1541 rf->clock_update_flags = RQCF_UPDATED;
1544 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1547 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1549 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1551 #ifdef CONFIG_SCHED_DEBUG
1553 * Restore the value we stashed in @rf for this pin context.
1555 rq->clock_update_flags |= rf->clock_update_flags;
1559 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1560 __acquires(rq->lock);
1562 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1563 __acquires(p->pi_lock)
1564 __acquires(rq->lock);
1566 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1567 __releases(rq->lock)
1569 rq_unpin_lock(rq, rf);
1570 raw_spin_rq_unlock(rq);
1574 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1575 __releases(rq->lock)
1576 __releases(p->pi_lock)
1578 rq_unpin_lock(rq, rf);
1579 raw_spin_rq_unlock(rq);
1580 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1584 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1585 __acquires(rq->lock)
1587 raw_spin_rq_lock_irqsave(rq, rf->flags);
1588 rq_pin_lock(rq, rf);
1592 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1593 __acquires(rq->lock)
1595 raw_spin_rq_lock_irq(rq);
1596 rq_pin_lock(rq, rf);
1600 rq_lock(struct rq *rq, struct rq_flags *rf)
1601 __acquires(rq->lock)
1603 raw_spin_rq_lock(rq);
1604 rq_pin_lock(rq, rf);
1608 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1609 __releases(rq->lock)
1611 rq_unpin_lock(rq, rf);
1612 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1616 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1617 __releases(rq->lock)
1619 rq_unpin_lock(rq, rf);
1620 raw_spin_rq_unlock_irq(rq);
1624 rq_unlock(struct rq *rq, struct rq_flags *rf)
1625 __releases(rq->lock)
1627 rq_unpin_lock(rq, rf);
1628 raw_spin_rq_unlock(rq);
1631 static inline struct rq *
1632 this_rq_lock_irq(struct rq_flags *rf)
1633 __acquires(rq->lock)
1637 local_irq_disable();
1644 enum numa_topology_type {
1649 extern enum numa_topology_type sched_numa_topology_type;
1650 extern int sched_max_numa_distance;
1651 extern bool find_numa_distance(int distance);
1652 extern void sched_init_numa(int offline_node);
1653 extern void sched_update_numa(int cpu, bool online);
1654 extern void sched_domains_numa_masks_set(unsigned int cpu);
1655 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1656 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1658 static inline void sched_init_numa(int offline_node) { }
1659 static inline void sched_update_numa(int cpu, bool online) { }
1660 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1661 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1662 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1668 #ifdef CONFIG_NUMA_BALANCING
1669 /* The regions in numa_faults array from task_struct */
1670 enum numa_faults_stats {
1676 extern void sched_setnuma(struct task_struct *p, int node);
1677 extern int migrate_task_to(struct task_struct *p, int cpu);
1678 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1680 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1683 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1686 #endif /* CONFIG_NUMA_BALANCING */
1691 queue_balance_callback(struct rq *rq,
1692 struct callback_head *head,
1693 void (*func)(struct rq *rq))
1695 lockdep_assert_rq_held(rq);
1698 * Don't (re)queue an already queued item; nor queue anything when
1699 * balance_push() is active, see the comment with
1700 * balance_push_callback.
1702 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1705 head->func = (void (*)(struct callback_head *))func;
1706 head->next = rq->balance_callback;
1707 rq->balance_callback = head;
1710 #define rcu_dereference_check_sched_domain(p) \
1711 rcu_dereference_check((p), \
1712 lockdep_is_held(&sched_domains_mutex))
1715 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1716 * See destroy_sched_domains: call_rcu for details.
1718 * The domain tree of any CPU may only be accessed from within
1719 * preempt-disabled sections.
1721 #define for_each_domain(cpu, __sd) \
1722 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1723 __sd; __sd = __sd->parent)
1726 * highest_flag_domain - Return highest sched_domain containing flag.
1727 * @cpu: The CPU whose highest level of sched domain is to
1729 * @flag: The flag to check for the highest sched_domain
1730 * for the given CPU.
1732 * Returns the highest sched_domain of a CPU which contains the given flag.
1734 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1736 struct sched_domain *sd, *hsd = NULL;
1738 for_each_domain(cpu, sd) {
1739 if (!(sd->flags & flag))
1747 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1749 struct sched_domain *sd;
1751 for_each_domain(cpu, sd) {
1752 if (sd->flags & flag)
1759 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1760 DECLARE_PER_CPU(int, sd_llc_size);
1761 DECLARE_PER_CPU(int, sd_llc_id);
1762 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1763 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1764 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1765 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1766 extern struct static_key_false sched_asym_cpucapacity;
1768 struct sched_group_capacity {
1771 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1774 unsigned long capacity;
1775 unsigned long min_capacity; /* Min per-CPU capacity in group */
1776 unsigned long max_capacity; /* Max per-CPU capacity in group */
1777 unsigned long next_update;
1778 int imbalance; /* XXX unrelated to capacity but shared group state */
1780 #ifdef CONFIG_SCHED_DEBUG
1784 unsigned long cpumask[]; /* Balance mask */
1787 struct sched_group {
1788 struct sched_group *next; /* Must be a circular list */
1791 unsigned int group_weight;
1792 struct sched_group_capacity *sgc;
1793 int asym_prefer_cpu; /* CPU of highest priority in group */
1797 * The CPUs this group covers.
1799 * NOTE: this field is variable length. (Allocated dynamically
1800 * by attaching extra space to the end of the structure,
1801 * depending on how many CPUs the kernel has booted up with)
1803 unsigned long cpumask[];
1806 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1808 return to_cpumask(sg->cpumask);
1812 * See build_balance_mask().
1814 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1816 return to_cpumask(sg->sgc->cpumask);
1820 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1821 * @group: The group whose first CPU is to be returned.
1823 static inline unsigned int group_first_cpu(struct sched_group *group)
1825 return cpumask_first(sched_group_span(group));
1828 extern int group_balance_cpu(struct sched_group *sg);
1830 #ifdef CONFIG_SCHED_DEBUG
1831 void update_sched_domain_debugfs(void);
1832 void dirty_sched_domain_sysctl(int cpu);
1834 static inline void update_sched_domain_debugfs(void)
1837 static inline void dirty_sched_domain_sysctl(int cpu)
1842 extern int sched_update_scaling(void);
1843 #endif /* CONFIG_SMP */
1847 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1849 extern void __sched_core_account_forceidle(struct rq *rq);
1851 static inline void sched_core_account_forceidle(struct rq *rq)
1853 if (schedstat_enabled())
1854 __sched_core_account_forceidle(rq);
1857 extern void __sched_core_tick(struct rq *rq);
1859 static inline void sched_core_tick(struct rq *rq)
1861 if (sched_core_enabled(rq) && schedstat_enabled())
1862 __sched_core_tick(rq);
1867 static inline void sched_core_account_forceidle(struct rq *rq) {}
1869 static inline void sched_core_tick(struct rq *rq) {}
1871 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1873 #ifdef CONFIG_CGROUP_SCHED
1876 * Return the group to which this tasks belongs.
1878 * We cannot use task_css() and friends because the cgroup subsystem
1879 * changes that value before the cgroup_subsys::attach() method is called,
1880 * therefore we cannot pin it and might observe the wrong value.
1882 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1883 * core changes this before calling sched_move_task().
1885 * Instead we use a 'copy' which is updated from sched_move_task() while
1886 * holding both task_struct::pi_lock and rq::lock.
1888 static inline struct task_group *task_group(struct task_struct *p)
1890 return p->sched_task_group;
1893 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1894 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1896 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1897 struct task_group *tg = task_group(p);
1900 #ifdef CONFIG_FAIR_GROUP_SCHED
1901 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1902 p->se.cfs_rq = tg->cfs_rq[cpu];
1903 p->se.parent = tg->se[cpu];
1906 #ifdef CONFIG_RT_GROUP_SCHED
1907 p->rt.rt_rq = tg->rt_rq[cpu];
1908 p->rt.parent = tg->rt_se[cpu];
1912 #else /* CONFIG_CGROUP_SCHED */
1914 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1915 static inline struct task_group *task_group(struct task_struct *p)
1920 #endif /* CONFIG_CGROUP_SCHED */
1922 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1924 set_task_rq(p, cpu);
1927 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1928 * successfully executed on another CPU. We must ensure that updates of
1929 * per-task data have been completed by this moment.
1932 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1938 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1940 #ifdef CONFIG_SCHED_DEBUG
1941 # define const_debug __read_mostly
1943 # define const_debug const
1946 #define SCHED_FEAT(name, enabled) \
1947 __SCHED_FEAT_##name ,
1950 #include "features.h"
1956 #ifdef CONFIG_SCHED_DEBUG
1959 * To support run-time toggling of sched features, all the translation units
1960 * (but core.c) reference the sysctl_sched_features defined in core.c.
1962 extern const_debug unsigned int sysctl_sched_features;
1964 #ifdef CONFIG_JUMP_LABEL
1965 #define SCHED_FEAT(name, enabled) \
1966 static __always_inline bool static_branch_##name(struct static_key *key) \
1968 return static_key_##enabled(key); \
1971 #include "features.h"
1974 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1975 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1977 #else /* !CONFIG_JUMP_LABEL */
1979 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1981 #endif /* CONFIG_JUMP_LABEL */
1983 #else /* !SCHED_DEBUG */
1986 * Each translation unit has its own copy of sysctl_sched_features to allow
1987 * constants propagation at compile time and compiler optimization based on
1990 #define SCHED_FEAT(name, enabled) \
1991 (1UL << __SCHED_FEAT_##name) * enabled |
1992 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1993 #include "features.h"
1997 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1999 #endif /* SCHED_DEBUG */
2001 extern struct static_key_false sched_numa_balancing;
2002 extern struct static_key_false sched_schedstats;
2004 static inline u64 global_rt_period(void)
2006 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2009 static inline u64 global_rt_runtime(void)
2011 if (sysctl_sched_rt_runtime < 0)
2014 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2017 static inline int task_current(struct rq *rq, struct task_struct *p)
2019 return rq->curr == p;
2022 static inline int task_running(struct rq *rq, struct task_struct *p)
2027 return task_current(rq, p);
2031 static inline int task_on_rq_queued(struct task_struct *p)
2033 return p->on_rq == TASK_ON_RQ_QUEUED;
2036 static inline int task_on_rq_migrating(struct task_struct *p)
2038 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2041 /* Wake flags. The first three directly map to some SD flag value */
2042 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2043 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2044 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2046 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2047 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2048 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2051 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2052 static_assert(WF_FORK == SD_BALANCE_FORK);
2053 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2057 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2058 * of tasks with abnormal "nice" values across CPUs the contribution that
2059 * each task makes to its run queue's load is weighted according to its
2060 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2061 * scaled version of the new time slice allocation that they receive on time
2065 #define WEIGHT_IDLEPRIO 3
2066 #define WMULT_IDLEPRIO 1431655765
2068 extern const int sched_prio_to_weight[40];
2069 extern const u32 sched_prio_to_wmult[40];
2072 * {de,en}queue flags:
2074 * DEQUEUE_SLEEP - task is no longer runnable
2075 * ENQUEUE_WAKEUP - task just became runnable
2077 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2078 * are in a known state which allows modification. Such pairs
2079 * should preserve as much state as possible.
2081 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2084 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2085 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2086 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2090 #define DEQUEUE_SLEEP 0x01
2091 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2092 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2093 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2095 #define ENQUEUE_WAKEUP 0x01
2096 #define ENQUEUE_RESTORE 0x02
2097 #define ENQUEUE_MOVE 0x04
2098 #define ENQUEUE_NOCLOCK 0x08
2100 #define ENQUEUE_HEAD 0x10
2101 #define ENQUEUE_REPLENISH 0x20
2103 #define ENQUEUE_MIGRATED 0x40
2105 #define ENQUEUE_MIGRATED 0x00
2108 #define RETRY_TASK ((void *)-1UL)
2110 struct sched_class {
2112 #ifdef CONFIG_UCLAMP_TASK
2116 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2117 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2118 void (*yield_task) (struct rq *rq);
2119 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2121 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2123 struct task_struct *(*pick_next_task)(struct rq *rq);
2125 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2126 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2129 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2130 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2132 struct task_struct * (*pick_task)(struct rq *rq);
2134 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2136 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2138 void (*set_cpus_allowed)(struct task_struct *p,
2139 const struct cpumask *newmask,
2142 void (*rq_online)(struct rq *rq);
2143 void (*rq_offline)(struct rq *rq);
2145 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2148 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2149 void (*task_fork)(struct task_struct *p);
2150 void (*task_dead)(struct task_struct *p);
2153 * The switched_from() call is allowed to drop rq->lock, therefore we
2154 * cannot assume the switched_from/switched_to pair is serialized by
2155 * rq->lock. They are however serialized by p->pi_lock.
2157 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2158 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2159 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2162 unsigned int (*get_rr_interval)(struct rq *rq,
2163 struct task_struct *task);
2165 void (*update_curr)(struct rq *rq);
2167 #define TASK_SET_GROUP 0
2168 #define TASK_MOVE_GROUP 1
2170 #ifdef CONFIG_FAIR_GROUP_SCHED
2171 void (*task_change_group)(struct task_struct *p, int type);
2175 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2177 WARN_ON_ONCE(rq->curr != prev);
2178 prev->sched_class->put_prev_task(rq, prev);
2181 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2183 next->sched_class->set_next_task(rq, next, false);
2188 * Helper to define a sched_class instance; each one is placed in a separate
2189 * section which is ordered by the linker script:
2191 * include/asm-generic/vmlinux.lds.h
2193 * *CAREFUL* they are laid out in *REVERSE* order!!!
2195 * Also enforce alignment on the instance, not the type, to guarantee layout.
2197 #define DEFINE_SCHED_CLASS(name) \
2198 const struct sched_class name##_sched_class \
2199 __aligned(__alignof__(struct sched_class)) \
2200 __section("__" #name "_sched_class")
2202 /* Defined in include/asm-generic/vmlinux.lds.h */
2203 extern struct sched_class __sched_class_highest[];
2204 extern struct sched_class __sched_class_lowest[];
2206 #define for_class_range(class, _from, _to) \
2207 for (class = (_from); class < (_to); class++)
2209 #define for_each_class(class) \
2210 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2212 #define sched_class_above(_a, _b) ((_a) < (_b))
2214 extern const struct sched_class stop_sched_class;
2215 extern const struct sched_class dl_sched_class;
2216 extern const struct sched_class rt_sched_class;
2217 extern const struct sched_class fair_sched_class;
2218 extern const struct sched_class idle_sched_class;
2220 static inline bool sched_stop_runnable(struct rq *rq)
2222 return rq->stop && task_on_rq_queued(rq->stop);
2225 static inline bool sched_dl_runnable(struct rq *rq)
2227 return rq->dl.dl_nr_running > 0;
2230 static inline bool sched_rt_runnable(struct rq *rq)
2232 return rq->rt.rt_queued > 0;
2235 static inline bool sched_fair_runnable(struct rq *rq)
2237 return rq->cfs.nr_running > 0;
2240 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2241 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2243 #define SCA_CHECK 0x01
2244 #define SCA_MIGRATE_DISABLE 0x02
2245 #define SCA_MIGRATE_ENABLE 0x04
2246 #define SCA_USER 0x08
2250 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2252 extern void trigger_load_balance(struct rq *rq);
2254 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2256 static inline struct task_struct *get_push_task(struct rq *rq)
2258 struct task_struct *p = rq->curr;
2260 lockdep_assert_rq_held(rq);
2265 if (p->nr_cpus_allowed == 1)
2268 if (p->migration_disabled)
2271 rq->push_busy = true;
2272 return get_task_struct(p);
2275 extern int push_cpu_stop(void *arg);
2279 #ifdef CONFIG_CPU_IDLE
2280 static inline void idle_set_state(struct rq *rq,
2281 struct cpuidle_state *idle_state)
2283 rq->idle_state = idle_state;
2286 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2288 SCHED_WARN_ON(!rcu_read_lock_held());
2290 return rq->idle_state;
2293 static inline void idle_set_state(struct rq *rq,
2294 struct cpuidle_state *idle_state)
2298 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2304 extern void schedule_idle(void);
2306 extern void sysrq_sched_debug_show(void);
2307 extern void sched_init_granularity(void);
2308 extern void update_max_interval(void);
2310 extern void init_sched_dl_class(void);
2311 extern void init_sched_rt_class(void);
2312 extern void init_sched_fair_class(void);
2314 extern void reweight_task(struct task_struct *p, int prio);
2316 extern void resched_curr(struct rq *rq);
2317 extern void resched_cpu(int cpu);
2319 extern struct rt_bandwidth def_rt_bandwidth;
2320 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2321 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2323 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2324 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2325 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2328 #define BW_UNIT (1 << BW_SHIFT)
2329 #define RATIO_SHIFT 8
2330 #define MAX_BW_BITS (64 - BW_SHIFT)
2331 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2332 unsigned long to_ratio(u64 period, u64 runtime);
2334 extern void init_entity_runnable_average(struct sched_entity *se);
2335 extern void post_init_entity_util_avg(struct task_struct *p);
2337 #ifdef CONFIG_NO_HZ_FULL
2338 extern bool sched_can_stop_tick(struct rq *rq);
2339 extern int __init sched_tick_offload_init(void);
2342 * Tick may be needed by tasks in the runqueue depending on their policy and
2343 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2344 * nohz mode if necessary.
2346 static inline void sched_update_tick_dependency(struct rq *rq)
2348 int cpu = cpu_of(rq);
2350 if (!tick_nohz_full_cpu(cpu))
2353 if (sched_can_stop_tick(rq))
2354 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2356 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2359 static inline int sched_tick_offload_init(void) { return 0; }
2360 static inline void sched_update_tick_dependency(struct rq *rq) { }
2363 static inline void add_nr_running(struct rq *rq, unsigned count)
2365 unsigned prev_nr = rq->nr_running;
2367 rq->nr_running = prev_nr + count;
2368 if (trace_sched_update_nr_running_tp_enabled()) {
2369 call_trace_sched_update_nr_running(rq, count);
2373 if (prev_nr < 2 && rq->nr_running >= 2) {
2374 if (!READ_ONCE(rq->rd->overload))
2375 WRITE_ONCE(rq->rd->overload, 1);
2379 sched_update_tick_dependency(rq);
2382 static inline void sub_nr_running(struct rq *rq, unsigned count)
2384 rq->nr_running -= count;
2385 if (trace_sched_update_nr_running_tp_enabled()) {
2386 call_trace_sched_update_nr_running(rq, -count);
2389 /* Check if we still need preemption */
2390 sched_update_tick_dependency(rq);
2393 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2394 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2396 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2398 extern const_debug unsigned int sysctl_sched_nr_migrate;
2399 extern const_debug unsigned int sysctl_sched_migration_cost;
2401 #ifdef CONFIG_SCHED_DEBUG
2402 extern unsigned int sysctl_sched_latency;
2403 extern unsigned int sysctl_sched_min_granularity;
2404 extern unsigned int sysctl_sched_idle_min_granularity;
2405 extern unsigned int sysctl_sched_wakeup_granularity;
2406 extern int sysctl_resched_latency_warn_ms;
2407 extern int sysctl_resched_latency_warn_once;
2409 extern unsigned int sysctl_sched_tunable_scaling;
2411 extern unsigned int sysctl_numa_balancing_scan_delay;
2412 extern unsigned int sysctl_numa_balancing_scan_period_min;
2413 extern unsigned int sysctl_numa_balancing_scan_period_max;
2414 extern unsigned int sysctl_numa_balancing_scan_size;
2417 #ifdef CONFIG_SCHED_HRTICK
2421 * - enabled by features
2422 * - hrtimer is actually high res
2424 static inline int hrtick_enabled(struct rq *rq)
2426 if (!cpu_active(cpu_of(rq)))
2428 return hrtimer_is_hres_active(&rq->hrtick_timer);
2431 static inline int hrtick_enabled_fair(struct rq *rq)
2433 if (!sched_feat(HRTICK))
2435 return hrtick_enabled(rq);
2438 static inline int hrtick_enabled_dl(struct rq *rq)
2440 if (!sched_feat(HRTICK_DL))
2442 return hrtick_enabled(rq);
2445 void hrtick_start(struct rq *rq, u64 delay);
2449 static inline int hrtick_enabled_fair(struct rq *rq)
2454 static inline int hrtick_enabled_dl(struct rq *rq)
2459 static inline int hrtick_enabled(struct rq *rq)
2464 #endif /* CONFIG_SCHED_HRTICK */
2466 #ifndef arch_scale_freq_tick
2467 static __always_inline
2468 void arch_scale_freq_tick(void)
2473 #ifndef arch_scale_freq_capacity
2475 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2476 * @cpu: the CPU in question.
2478 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2481 * ------ * SCHED_CAPACITY_SCALE
2484 static __always_inline
2485 unsigned long arch_scale_freq_capacity(int cpu)
2487 return SCHED_CAPACITY_SCALE;
2491 #ifdef CONFIG_SCHED_DEBUG
2493 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2494 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2495 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2496 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2498 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2500 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2501 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2503 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2507 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2512 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2514 #ifdef CONFIG_SCHED_CORE
2516 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2517 * order by core-id first and cpu-id second.
2521 * double_rq_lock(0,3); will take core-0, core-1 lock
2522 * double_rq_lock(1,2); will take core-1, core-0 lock
2524 * when only cpu-id is considered.
2526 if (rq1->core->cpu < rq2->core->cpu)
2528 if (rq1->core->cpu > rq2->core->cpu)
2532 * __sched_core_flip() relies on SMT having cpu-id lock order.
2535 return rq1->cpu < rq2->cpu;
2538 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2540 #ifdef CONFIG_PREEMPTION
2543 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2544 * way at the expense of forcing extra atomic operations in all
2545 * invocations. This assures that the double_lock is acquired using the
2546 * same underlying policy as the spinlock_t on this architecture, which
2547 * reduces latency compared to the unfair variant below. However, it
2548 * also adds more overhead and therefore may reduce throughput.
2550 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2551 __releases(this_rq->lock)
2552 __acquires(busiest->lock)
2553 __acquires(this_rq->lock)
2555 raw_spin_rq_unlock(this_rq);
2556 double_rq_lock(this_rq, busiest);
2563 * Unfair double_lock_balance: Optimizes throughput at the expense of
2564 * latency by eliminating extra atomic operations when the locks are
2565 * already in proper order on entry. This favors lower CPU-ids and will
2566 * grant the double lock to lower CPUs over higher ids under contention,
2567 * regardless of entry order into the function.
2569 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2570 __releases(this_rq->lock)
2571 __acquires(busiest->lock)
2572 __acquires(this_rq->lock)
2574 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2575 likely(raw_spin_rq_trylock(busiest))) {
2576 double_rq_clock_clear_update(this_rq, busiest);
2580 if (rq_order_less(this_rq, busiest)) {
2581 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2582 double_rq_clock_clear_update(this_rq, busiest);
2586 raw_spin_rq_unlock(this_rq);
2587 double_rq_lock(this_rq, busiest);
2592 #endif /* CONFIG_PREEMPTION */
2595 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2597 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2599 lockdep_assert_irqs_disabled();
2601 return _double_lock_balance(this_rq, busiest);
2604 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2605 __releases(busiest->lock)
2607 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2608 raw_spin_rq_unlock(busiest);
2609 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2612 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2618 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2621 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2627 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2630 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2636 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2640 * double_rq_unlock - safely unlock two runqueues
2642 * Note this does not restore interrupts like task_rq_unlock,
2643 * you need to do so manually after calling.
2645 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2646 __releases(rq1->lock)
2647 __releases(rq2->lock)
2649 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2650 raw_spin_rq_unlock(rq2);
2652 __release(rq2->lock);
2653 raw_spin_rq_unlock(rq1);
2656 extern void set_rq_online (struct rq *rq);
2657 extern void set_rq_offline(struct rq *rq);
2658 extern bool sched_smp_initialized;
2660 #else /* CONFIG_SMP */
2663 * double_rq_lock - safely lock two runqueues
2665 * Note this does not disable interrupts like task_rq_lock,
2666 * you need to do so manually before calling.
2668 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2669 __acquires(rq1->lock)
2670 __acquires(rq2->lock)
2672 BUG_ON(!irqs_disabled());
2674 raw_spin_rq_lock(rq1);
2675 __acquire(rq2->lock); /* Fake it out ;) */
2676 double_rq_clock_clear_update(rq1, rq2);
2680 * double_rq_unlock - safely unlock two runqueues
2682 * Note this does not restore interrupts like task_rq_unlock,
2683 * you need to do so manually after calling.
2685 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2686 __releases(rq1->lock)
2687 __releases(rq2->lock)
2690 raw_spin_rq_unlock(rq1);
2691 __release(rq2->lock);
2696 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2697 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2699 #ifdef CONFIG_SCHED_DEBUG
2700 extern bool sched_debug_verbose;
2702 extern void print_cfs_stats(struct seq_file *m, int cpu);
2703 extern void print_rt_stats(struct seq_file *m, int cpu);
2704 extern void print_dl_stats(struct seq_file *m, int cpu);
2705 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2706 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2707 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2709 extern void resched_latency_warn(int cpu, u64 latency);
2710 #ifdef CONFIG_NUMA_BALANCING
2712 show_numa_stats(struct task_struct *p, struct seq_file *m);
2714 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2715 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2716 #endif /* CONFIG_NUMA_BALANCING */
2718 static inline void resched_latency_warn(int cpu, u64 latency) {}
2719 #endif /* CONFIG_SCHED_DEBUG */
2721 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2722 extern void init_rt_rq(struct rt_rq *rt_rq);
2723 extern void init_dl_rq(struct dl_rq *dl_rq);
2725 extern void cfs_bandwidth_usage_inc(void);
2726 extern void cfs_bandwidth_usage_dec(void);
2728 #ifdef CONFIG_NO_HZ_COMMON
2729 #define NOHZ_BALANCE_KICK_BIT 0
2730 #define NOHZ_STATS_KICK_BIT 1
2731 #define NOHZ_NEWILB_KICK_BIT 2
2732 #define NOHZ_NEXT_KICK_BIT 3
2734 /* Run rebalance_domains() */
2735 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2736 /* Update blocked load */
2737 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2738 /* Update blocked load when entering idle */
2739 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2740 /* Update nohz.next_balance */
2741 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2743 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2745 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2747 extern void nohz_balance_exit_idle(struct rq *rq);
2749 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2752 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2753 extern void nohz_run_idle_balance(int cpu);
2755 static inline void nohz_run_idle_balance(int cpu) { }
2758 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2763 struct u64_stats_sync sync;
2766 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2769 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2770 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2771 * and never move forward.
2773 static inline u64 irq_time_read(int cpu)
2775 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2780 seq = __u64_stats_fetch_begin(&irqtime->sync);
2781 total = irqtime->total;
2782 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2786 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2788 #ifdef CONFIG_CPU_FREQ
2789 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2792 * cpufreq_update_util - Take a note about CPU utilization changes.
2793 * @rq: Runqueue to carry out the update for.
2794 * @flags: Update reason flags.
2796 * This function is called by the scheduler on the CPU whose utilization is
2799 * It can only be called from RCU-sched read-side critical sections.
2801 * The way cpufreq is currently arranged requires it to evaluate the CPU
2802 * performance state (frequency/voltage) on a regular basis to prevent it from
2803 * being stuck in a completely inadequate performance level for too long.
2804 * That is not guaranteed to happen if the updates are only triggered from CFS
2805 * and DL, though, because they may not be coming in if only RT tasks are
2806 * active all the time (or there are RT tasks only).
2808 * As a workaround for that issue, this function is called periodically by the
2809 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2810 * but that really is a band-aid. Going forward it should be replaced with
2811 * solutions targeted more specifically at RT tasks.
2813 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2815 struct update_util_data *data;
2817 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2820 data->func(data, rq_clock(rq), flags);
2823 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2824 #endif /* CONFIG_CPU_FREQ */
2826 #ifdef arch_scale_freq_capacity
2827 # ifndef arch_scale_freq_invariant
2828 # define arch_scale_freq_invariant() true
2831 # define arch_scale_freq_invariant() false
2835 static inline unsigned long capacity_orig_of(int cpu)
2837 return cpu_rq(cpu)->cpu_capacity_orig;
2841 * enum cpu_util_type - CPU utilization type
2842 * @FREQUENCY_UTIL: Utilization used to select frequency
2843 * @ENERGY_UTIL: Utilization used during energy calculation
2845 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2846 * need to be aggregated differently depending on the usage made of them. This
2847 * enum is used within effective_cpu_util() to differentiate the types of
2848 * utilization expected by the callers, and adjust the aggregation accordingly.
2850 enum cpu_util_type {
2855 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2856 unsigned long max, enum cpu_util_type type,
2857 struct task_struct *p);
2859 static inline unsigned long cpu_bw_dl(struct rq *rq)
2861 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2864 static inline unsigned long cpu_util_dl(struct rq *rq)
2866 return READ_ONCE(rq->avg_dl.util_avg);
2870 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2871 * @cpu: the CPU to get the utilization for.
2873 * The unit of the return value must be the same as the one of CPU capacity
2874 * so that CPU utilization can be compared with CPU capacity.
2876 * CPU utilization is the sum of running time of runnable tasks plus the
2877 * recent utilization of currently non-runnable tasks on that CPU.
2878 * It represents the amount of CPU capacity currently used by CFS tasks in
2879 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2880 * capacity at f_max.
2882 * The estimated CPU utilization is defined as the maximum between CPU
2883 * utilization and sum of the estimated utilization of the currently
2884 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2885 * previously-executed tasks, which helps better deduce how busy a CPU will
2886 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2887 * of such a task would be significantly decayed at this point of time.
2889 * CPU utilization can be higher than the current CPU capacity
2890 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2891 * of rounding errors as well as task migrations or wakeups of new tasks.
2892 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2893 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2894 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2895 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2896 * though since this is useful for predicting the CPU capacity required
2897 * after task migrations (scheduler-driven DVFS).
2899 * Return: (Estimated) utilization for the specified CPU.
2901 static inline unsigned long cpu_util_cfs(int cpu)
2903 struct cfs_rq *cfs_rq;
2906 cfs_rq = &cpu_rq(cpu)->cfs;
2907 util = READ_ONCE(cfs_rq->avg.util_avg);
2909 if (sched_feat(UTIL_EST)) {
2910 util = max_t(unsigned long, util,
2911 READ_ONCE(cfs_rq->avg.util_est.enqueued));
2914 return min(util, capacity_orig_of(cpu));
2917 static inline unsigned long cpu_util_rt(struct rq *rq)
2919 return READ_ONCE(rq->avg_rt.util_avg);
2923 #ifdef CONFIG_UCLAMP_TASK
2924 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2927 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2928 * @rq: The rq to clamp against. Must not be NULL.
2929 * @util: The util value to clamp.
2930 * @p: The task to clamp against. Can be NULL if you want to clamp
2933 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2935 * If sched_uclamp_used static key is disabled, then just return the util
2936 * without any clamping since uclamp aggregation at the rq level in the fast
2937 * path is disabled, rendering this operation a NOP.
2939 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2940 * will return the correct effective uclamp value of the task even if the
2941 * static key is disabled.
2943 static __always_inline
2944 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2945 struct task_struct *p)
2947 unsigned long min_util = 0;
2948 unsigned long max_util = 0;
2950 if (!static_branch_likely(&sched_uclamp_used))
2954 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2955 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2958 * Ignore last runnable task's max clamp, as this task will
2959 * reset it. Similarly, no need to read the rq's min clamp.
2961 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2965 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2966 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2969 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2970 * RUNNABLE tasks with _different_ clamps, we can end up with an
2971 * inversion. Fix it now when the clamps are applied.
2973 if (unlikely(min_util >= max_util))
2976 return clamp(util, min_util, max_util);
2979 /* Is the rq being capped/throttled by uclamp_max? */
2980 static inline bool uclamp_rq_is_capped(struct rq *rq)
2982 unsigned long rq_util;
2983 unsigned long max_util;
2985 if (!static_branch_likely(&sched_uclamp_used))
2988 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
2989 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2991 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
2995 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2996 * by default in the fast path and only gets turned on once userspace performs
2997 * an operation that requires it.
2999 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3002 static inline bool uclamp_is_used(void)
3004 return static_branch_likely(&sched_uclamp_used);
3006 #else /* CONFIG_UCLAMP_TASK */
3008 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3009 struct task_struct *p)
3014 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3016 static inline bool uclamp_is_used(void)
3020 #endif /* CONFIG_UCLAMP_TASK */
3022 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3023 static inline unsigned long cpu_util_irq(struct rq *rq)
3025 return rq->avg_irq.util_avg;
3029 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3031 util *= (max - irq);
3038 static inline unsigned long cpu_util_irq(struct rq *rq)
3044 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3050 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3052 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3054 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3056 static inline bool sched_energy_enabled(void)
3058 return static_branch_unlikely(&sched_energy_present);
3061 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3063 #define perf_domain_span(pd) NULL
3064 static inline bool sched_energy_enabled(void) { return false; }
3066 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3068 #ifdef CONFIG_MEMBARRIER
3070 * The scheduler provides memory barriers required by membarrier between:
3071 * - prior user-space memory accesses and store to rq->membarrier_state,
3072 * - store to rq->membarrier_state and following user-space memory accesses.
3073 * In the same way it provides those guarantees around store to rq->curr.
3075 static inline void membarrier_switch_mm(struct rq *rq,
3076 struct mm_struct *prev_mm,
3077 struct mm_struct *next_mm)
3079 int membarrier_state;
3081 if (prev_mm == next_mm)
3084 membarrier_state = atomic_read(&next_mm->membarrier_state);
3085 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3088 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3091 static inline void membarrier_switch_mm(struct rq *rq,
3092 struct mm_struct *prev_mm,
3093 struct mm_struct *next_mm)
3099 static inline bool is_per_cpu_kthread(struct task_struct *p)
3101 if (!(p->flags & PF_KTHREAD))
3104 if (p->nr_cpus_allowed != 1)
3111 extern void swake_up_all_locked(struct swait_queue_head *q);
3112 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3114 #ifdef CONFIG_PREEMPT_DYNAMIC
3115 extern int preempt_dynamic_mode;
3116 extern int sched_dynamic_mode(const char *str);
3117 extern void sched_dynamic_update(int mode);
3120 #endif /* _KERNEL_SCHED_SCHED_H */