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 */
525 * u64_u32_load/u64_u32_store
527 * Use a copy of a u64 value to protect against data race. This is only
528 * applicable for 32-bits architectures.
531 # define u64_u32_load_copy(var, copy) var
532 # define u64_u32_store_copy(var, copy, val) (var = val)
534 # define u64_u32_load_copy(var, copy) \
536 u64 __val, __val_copy; \
540 * paired with u64_u32_store_copy(), ordering access \
545 } while (__val != __val_copy); \
548 # define u64_u32_store_copy(var, copy, val) \
550 typeof(val) __val = (val); \
553 * paired with u64_u32_load_copy(), ordering access to var and \
560 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
561 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
563 /* CFS-related fields in a runqueue */
565 struct load_weight load;
566 unsigned int nr_running;
567 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
568 unsigned int idle_nr_running; /* SCHED_IDLE */
569 unsigned int idle_h_nr_running; /* SCHED_IDLE */
573 #ifdef CONFIG_SCHED_CORE
574 unsigned int forceidle_seq;
579 u64 min_vruntime_copy;
582 struct rb_root_cached tasks_timeline;
585 * 'curr' points to currently running entity on this cfs_rq.
586 * It is set to NULL otherwise (i.e when none are currently running).
588 struct sched_entity *curr;
589 struct sched_entity *next;
590 struct sched_entity *last;
591 struct sched_entity *skip;
593 #ifdef CONFIG_SCHED_DEBUG
594 unsigned int nr_spread_over;
601 struct sched_avg avg;
603 u64 last_update_time_copy;
606 raw_spinlock_t lock ____cacheline_aligned;
608 unsigned long load_avg;
609 unsigned long util_avg;
610 unsigned long runnable_avg;
613 #ifdef CONFIG_FAIR_GROUP_SCHED
614 unsigned long tg_load_avg_contrib;
616 long prop_runnable_sum;
619 * h_load = weight * f(tg)
621 * Where f(tg) is the recursive weight fraction assigned to
624 unsigned long h_load;
625 u64 last_h_load_update;
626 struct sched_entity *h_load_next;
627 #endif /* CONFIG_FAIR_GROUP_SCHED */
628 #endif /* CONFIG_SMP */
630 #ifdef CONFIG_FAIR_GROUP_SCHED
631 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
634 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
635 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
636 * (like users, containers etc.)
638 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
639 * This list is used during load balance.
642 struct list_head leaf_cfs_rq_list;
643 struct task_group *tg; /* group that "owns" this runqueue */
645 /* Locally cached copy of our task_group's idle value */
648 #ifdef CONFIG_CFS_BANDWIDTH
650 s64 runtime_remaining;
652 u64 throttled_pelt_idle;
654 u64 throttled_pelt_idle_copy;
657 u64 throttled_clock_pelt;
658 u64 throttled_clock_pelt_time;
661 struct list_head throttled_list;
662 #endif /* CONFIG_CFS_BANDWIDTH */
663 #endif /* CONFIG_FAIR_GROUP_SCHED */
666 static inline int rt_bandwidth_enabled(void)
668 return sysctl_sched_rt_runtime >= 0;
671 /* RT IPI pull logic requires IRQ_WORK */
672 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
673 # define HAVE_RT_PUSH_IPI
676 /* Real-Time classes' related field in a runqueue: */
678 struct rt_prio_array active;
679 unsigned int rt_nr_running;
680 unsigned int rr_nr_running;
681 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
683 int curr; /* highest queued rt task prio */
685 int next; /* next highest */
690 unsigned int rt_nr_migratory;
691 unsigned int rt_nr_total;
693 struct plist_head pushable_tasks;
695 #endif /* CONFIG_SMP */
701 /* Nests inside the rq lock: */
702 raw_spinlock_t rt_runtime_lock;
704 #ifdef CONFIG_RT_GROUP_SCHED
705 unsigned int rt_nr_boosted;
708 struct task_group *tg;
712 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
714 return rt_rq->rt_queued && rt_rq->rt_nr_running;
717 /* Deadline class' related fields in a runqueue */
719 /* runqueue is an rbtree, ordered by deadline */
720 struct rb_root_cached root;
722 unsigned int dl_nr_running;
726 * Deadline values of the currently executing and the
727 * earliest ready task on this rq. Caching these facilitates
728 * the decision whether or not a ready but not running task
729 * should migrate somewhere else.
736 unsigned int dl_nr_migratory;
740 * Tasks on this rq that can be pushed away. They are kept in
741 * an rb-tree, ordered by tasks' deadlines, with caching
742 * of the leftmost (earliest deadline) element.
744 struct rb_root_cached pushable_dl_tasks_root;
749 * "Active utilization" for this runqueue: increased when a
750 * task wakes up (becomes TASK_RUNNING) and decreased when a
756 * Utilization of the tasks "assigned" to this runqueue (including
757 * the tasks that are in runqueue and the tasks that executed on this
758 * CPU and blocked). Increased when a task moves to this runqueue, and
759 * decreased when the task moves away (migrates, changes scheduling
760 * policy, or terminates).
761 * This is needed to compute the "inactive utilization" for the
762 * runqueue (inactive utilization = this_bw - running_bw).
768 * Inverse of the fraction of CPU utilization that can be reclaimed
769 * by the GRUB algorithm.
774 #ifdef CONFIG_FAIR_GROUP_SCHED
775 /* An entity is a task if it doesn't "own" a runqueue */
776 #define entity_is_task(se) (!se->my_q)
778 static inline void se_update_runnable(struct sched_entity *se)
780 if (!entity_is_task(se))
781 se->runnable_weight = se->my_q->h_nr_running;
784 static inline long se_runnable(struct sched_entity *se)
786 if (entity_is_task(se))
789 return se->runnable_weight;
793 #define entity_is_task(se) 1
795 static inline void se_update_runnable(struct sched_entity *se) {}
797 static inline long se_runnable(struct sched_entity *se)
805 * XXX we want to get rid of these helpers and use the full load resolution.
807 static inline long se_weight(struct sched_entity *se)
809 return scale_load_down(se->load.weight);
813 static inline bool sched_asym_prefer(int a, int b)
815 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
819 struct em_perf_domain *em_pd;
820 struct perf_domain *next;
824 /* Scheduling group status flags */
825 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
826 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
829 * We add the notion of a root-domain which will be used to define per-domain
830 * variables. Each exclusive cpuset essentially defines an island domain by
831 * fully partitioning the member CPUs from any other cpuset. Whenever a new
832 * exclusive cpuset is created, we also create and attach a new root-domain
841 cpumask_var_t online;
844 * Indicate pullable load on at least one CPU, e.g:
845 * - More than one runnable task
846 * - Running task is misfit
850 /* Indicate one or more cpus over-utilized (tipping point) */
854 * The bit corresponding to a CPU gets set here if such CPU has more
855 * than one runnable -deadline task (as it is below for RT tasks).
857 cpumask_var_t dlo_mask;
863 * Indicate whether a root_domain's dl_bw has been checked or
864 * updated. It's monotonously increasing value.
866 * Also, some corner cases, like 'wrap around' is dangerous, but given
867 * that u64 is 'big enough'. So that shouldn't be a concern.
871 #ifdef HAVE_RT_PUSH_IPI
873 * For IPI pull requests, loop across the rto_mask.
875 struct irq_work rto_push_work;
876 raw_spinlock_t rto_lock;
877 /* These are only updated and read within rto_lock */
880 /* These atomics are updated outside of a lock */
881 atomic_t rto_loop_next;
882 atomic_t rto_loop_start;
885 * The "RT overload" flag: it gets set if a CPU has more than
886 * one runnable RT task.
888 cpumask_var_t rto_mask;
889 struct cpupri cpupri;
891 unsigned long max_cpu_capacity;
894 * NULL-terminated list of performance domains intersecting with the
895 * CPUs of the rd. Protected by RCU.
897 struct perf_domain __rcu *pd;
900 extern void init_defrootdomain(void);
901 extern int sched_init_domains(const struct cpumask *cpu_map);
902 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
903 extern void sched_get_rd(struct root_domain *rd);
904 extern void sched_put_rd(struct root_domain *rd);
906 #ifdef HAVE_RT_PUSH_IPI
907 extern void rto_push_irq_work_func(struct irq_work *work);
909 #endif /* CONFIG_SMP */
911 #ifdef CONFIG_UCLAMP_TASK
913 * struct uclamp_bucket - Utilization clamp bucket
914 * @value: utilization clamp value for tasks on this clamp bucket
915 * @tasks: number of RUNNABLE tasks on this clamp bucket
917 * Keep track of how many tasks are RUNNABLE for a given utilization
920 struct uclamp_bucket {
921 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
922 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
926 * struct uclamp_rq - rq's utilization clamp
927 * @value: currently active clamp values for a rq
928 * @bucket: utilization clamp buckets affecting a rq
930 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
931 * A clamp value is affecting a rq when there is at least one task RUNNABLE
932 * (or actually running) with that value.
934 * There are up to UCLAMP_CNT possible different clamp values, currently there
935 * are only two: minimum utilization and maximum utilization.
937 * All utilization clamping values are MAX aggregated, since:
938 * - for util_min: we want to run the CPU at least at the max of the minimum
939 * utilization required by its currently RUNNABLE tasks.
940 * - for util_max: we want to allow the CPU to run up to the max of the
941 * maximum utilization allowed by its currently RUNNABLE tasks.
943 * Since on each system we expect only a limited number of different
944 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
945 * the metrics required to compute all the per-rq utilization clamp values.
949 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
952 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
953 #endif /* CONFIG_UCLAMP_TASK */
956 * This is the main, per-CPU runqueue data structure.
958 * Locking rule: those places that want to lock multiple runqueues
959 * (such as the load balancing or the thread migration code), lock
960 * acquire operations must be ordered by ascending &runqueue.
964 raw_spinlock_t __lock;
967 * nr_running and cpu_load should be in the same cacheline because
968 * remote CPUs use both these fields when doing load calculation.
970 unsigned int nr_running;
971 #ifdef CONFIG_NUMA_BALANCING
972 unsigned int nr_numa_running;
973 unsigned int nr_preferred_running;
974 unsigned int numa_migrate_on;
976 #ifdef CONFIG_NO_HZ_COMMON
978 unsigned long last_blocked_load_update_tick;
979 unsigned int has_blocked_load;
980 call_single_data_t nohz_csd;
981 #endif /* CONFIG_SMP */
982 unsigned int nohz_tick_stopped;
984 #endif /* CONFIG_NO_HZ_COMMON */
987 unsigned int ttwu_pending;
991 #ifdef CONFIG_UCLAMP_TASK
992 /* Utilization clamp values based on CPU's RUNNABLE tasks */
993 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
994 unsigned int uclamp_flags;
995 #define UCLAMP_FLAG_IDLE 0x01
1002 #ifdef CONFIG_FAIR_GROUP_SCHED
1003 /* list of leaf cfs_rq on this CPU: */
1004 struct list_head leaf_cfs_rq_list;
1005 struct list_head *tmp_alone_branch;
1006 #endif /* CONFIG_FAIR_GROUP_SCHED */
1009 * This is part of a global counter where only the total sum
1010 * over all CPUs matters. A task can increase this counter on
1011 * one CPU and if it got migrated afterwards it may decrease
1012 * it on another CPU. Always updated under the runqueue lock:
1014 unsigned int nr_uninterruptible;
1016 struct task_struct __rcu *curr;
1017 struct task_struct *idle;
1018 struct task_struct *stop;
1019 unsigned long next_balance;
1020 struct mm_struct *prev_mm;
1022 unsigned int clock_update_flags;
1024 /* Ensure that all clocks are in the same cache line */
1025 u64 clock_task ____cacheline_aligned;
1027 unsigned long lost_idle_time;
1028 u64 clock_pelt_idle;
1030 #ifndef CONFIG_64BIT
1031 u64 clock_pelt_idle_copy;
1032 u64 clock_idle_copy;
1037 #ifdef CONFIG_SCHED_DEBUG
1038 u64 last_seen_need_resched_ns;
1039 int ticks_without_resched;
1042 #ifdef CONFIG_MEMBARRIER
1043 int membarrier_state;
1047 struct root_domain *rd;
1048 struct sched_domain __rcu *sd;
1050 unsigned long cpu_capacity;
1051 unsigned long cpu_capacity_orig;
1053 struct callback_head *balance_callback;
1055 unsigned char nohz_idle_balance;
1056 unsigned char idle_balance;
1058 unsigned long misfit_task_load;
1060 /* For active balancing */
1063 struct cpu_stop_work active_balance_work;
1065 /* CPU of this runqueue: */
1069 struct list_head cfs_tasks;
1071 struct sched_avg avg_rt;
1072 struct sched_avg avg_dl;
1073 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1074 struct sched_avg avg_irq;
1076 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1077 struct sched_avg avg_thermal;
1082 unsigned long wake_stamp;
1085 /* This is used to determine avg_idle's max value */
1086 u64 max_idle_balance_cost;
1088 #ifdef CONFIG_HOTPLUG_CPU
1089 struct rcuwait hotplug_wait;
1091 #endif /* CONFIG_SMP */
1093 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1096 #ifdef CONFIG_PARAVIRT
1097 u64 prev_steal_time;
1099 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1100 u64 prev_steal_time_rq;
1103 /* calc_load related fields */
1104 unsigned long calc_load_update;
1105 long calc_load_active;
1107 #ifdef CONFIG_SCHED_HRTICK
1109 call_single_data_t hrtick_csd;
1111 struct hrtimer hrtick_timer;
1112 ktime_t hrtick_time;
1115 #ifdef CONFIG_SCHEDSTATS
1117 struct sched_info rq_sched_info;
1118 unsigned long long rq_cpu_time;
1119 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1121 /* sys_sched_yield() stats */
1122 unsigned int yld_count;
1124 /* schedule() stats */
1125 unsigned int sched_count;
1126 unsigned int sched_goidle;
1128 /* try_to_wake_up() stats */
1129 unsigned int ttwu_count;
1130 unsigned int ttwu_local;
1133 #ifdef CONFIG_CPU_IDLE
1134 /* Must be inspected within a rcu lock section */
1135 struct cpuidle_state *idle_state;
1139 unsigned int nr_pinned;
1141 unsigned int push_busy;
1142 struct cpu_stop_work push_work;
1144 #ifdef CONFIG_SCHED_CORE
1147 struct task_struct *core_pick;
1148 unsigned int core_enabled;
1149 unsigned int core_sched_seq;
1150 struct rb_root core_tree;
1152 /* shared state -- careful with sched_core_cpu_deactivate() */
1153 unsigned int core_task_seq;
1154 unsigned int core_pick_seq;
1155 unsigned long core_cookie;
1156 unsigned int core_forceidle_count;
1157 unsigned int core_forceidle_seq;
1158 unsigned int core_forceidle_occupation;
1159 u64 core_forceidle_start;
1163 #ifdef CONFIG_FAIR_GROUP_SCHED
1165 /* CPU runqueue to which this cfs_rq is attached */
1166 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1173 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1175 return container_of(cfs_rq, struct rq, cfs);
1179 static inline int cpu_of(struct rq *rq)
1188 #define MDF_PUSH 0x01
1190 static inline bool is_migration_disabled(struct task_struct *p)
1193 return p->migration_disabled;
1200 #ifdef CONFIG_SCHED_CORE
1201 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1203 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1205 static inline bool sched_core_enabled(struct rq *rq)
1207 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1210 static inline bool sched_core_disabled(void)
1212 return !static_branch_unlikely(&__sched_core_enabled);
1216 * Be careful with this function; not for general use. The return value isn't
1217 * stable unless you actually hold a relevant rq->__lock.
1219 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1221 if (sched_core_enabled(rq))
1222 return &rq->core->__lock;
1227 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1229 if (rq->core_enabled)
1230 return &rq->core->__lock;
1235 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1238 * Helpers to check if the CPU's core cookie matches with the task's cookie
1239 * when core scheduling is enabled.
1240 * A special case is that the task's cookie always matches with CPU's core
1241 * cookie if the CPU is in an idle core.
1243 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1245 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1246 if (!sched_core_enabled(rq))
1249 return rq->core->core_cookie == p->core_cookie;
1252 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1254 bool idle_core = true;
1257 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1258 if (!sched_core_enabled(rq))
1261 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1262 if (!available_idle_cpu(cpu)) {
1269 * A CPU in an idle core is always the best choice for tasks with
1272 return idle_core || rq->core->core_cookie == p->core_cookie;
1275 static inline bool sched_group_cookie_match(struct rq *rq,
1276 struct task_struct *p,
1277 struct sched_group *group)
1281 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1282 if (!sched_core_enabled(rq))
1285 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1286 if (sched_core_cookie_match(rq, p))
1292 static inline bool sched_core_enqueued(struct task_struct *p)
1294 return !RB_EMPTY_NODE(&p->core_node);
1297 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1298 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1300 extern void sched_core_get(void);
1301 extern void sched_core_put(void);
1303 #else /* !CONFIG_SCHED_CORE */
1305 static inline bool sched_core_enabled(struct rq *rq)
1310 static inline bool sched_core_disabled(void)
1315 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1320 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1325 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1330 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1335 static inline bool sched_group_cookie_match(struct rq *rq,
1336 struct task_struct *p,
1337 struct sched_group *group)
1341 #endif /* CONFIG_SCHED_CORE */
1343 static inline void lockdep_assert_rq_held(struct rq *rq)
1345 lockdep_assert_held(__rq_lockp(rq));
1348 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1349 extern bool raw_spin_rq_trylock(struct rq *rq);
1350 extern void raw_spin_rq_unlock(struct rq *rq);
1352 static inline void raw_spin_rq_lock(struct rq *rq)
1354 raw_spin_rq_lock_nested(rq, 0);
1357 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1359 local_irq_disable();
1360 raw_spin_rq_lock(rq);
1363 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1365 raw_spin_rq_unlock(rq);
1369 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1371 unsigned long flags;
1372 local_irq_save(flags);
1373 raw_spin_rq_lock(rq);
1377 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1379 raw_spin_rq_unlock(rq);
1380 local_irq_restore(flags);
1383 #define raw_spin_rq_lock_irqsave(rq, flags) \
1385 flags = _raw_spin_rq_lock_irqsave(rq); \
1388 #ifdef CONFIG_SCHED_SMT
1389 extern void __update_idle_core(struct rq *rq);
1391 static inline void update_idle_core(struct rq *rq)
1393 if (static_branch_unlikely(&sched_smt_present))
1394 __update_idle_core(rq);
1398 static inline void update_idle_core(struct rq *rq) { }
1401 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1403 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1404 #define this_rq() this_cpu_ptr(&runqueues)
1405 #define task_rq(p) cpu_rq(task_cpu(p))
1406 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1407 #define raw_rq() raw_cpu_ptr(&runqueues)
1409 #ifdef CONFIG_FAIR_GROUP_SCHED
1410 static inline struct task_struct *task_of(struct sched_entity *se)
1412 SCHED_WARN_ON(!entity_is_task(se));
1413 return container_of(se, struct task_struct, se);
1416 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1418 return p->se.cfs_rq;
1421 /* runqueue on which this entity is (to be) queued */
1422 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1427 /* runqueue "owned" by this group */
1428 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1435 static inline struct task_struct *task_of(struct sched_entity *se)
1437 return container_of(se, struct task_struct, se);
1440 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1442 return &task_rq(p)->cfs;
1445 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1447 struct task_struct *p = task_of(se);
1448 struct rq *rq = task_rq(p);
1453 /* runqueue "owned" by this group */
1454 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1460 extern void update_rq_clock(struct rq *rq);
1463 * rq::clock_update_flags bits
1465 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1466 * call to __schedule(). This is an optimisation to avoid
1467 * neighbouring rq clock updates.
1469 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1470 * in effect and calls to update_rq_clock() are being ignored.
1472 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1473 * made to update_rq_clock() since the last time rq::lock was pinned.
1475 * If inside of __schedule(), clock_update_flags will have been
1476 * shifted left (a left shift is a cheap operation for the fast path
1477 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1479 * if (rq-clock_update_flags >= RQCF_UPDATED)
1481 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1482 * one position though, because the next rq_unpin_lock() will shift it
1485 #define RQCF_REQ_SKIP 0x01
1486 #define RQCF_ACT_SKIP 0x02
1487 #define RQCF_UPDATED 0x04
1489 static inline void assert_clock_updated(struct rq *rq)
1492 * The only reason for not seeing a clock update since the
1493 * last rq_pin_lock() is if we're currently skipping updates.
1495 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1498 static inline u64 rq_clock(struct rq *rq)
1500 lockdep_assert_rq_held(rq);
1501 assert_clock_updated(rq);
1506 static inline u64 rq_clock_task(struct rq *rq)
1508 lockdep_assert_rq_held(rq);
1509 assert_clock_updated(rq);
1511 return rq->clock_task;
1515 * By default the decay is the default pelt decay period.
1516 * The decay shift can change the decay period in
1518 * Decay shift Decay period(ms)
1525 extern int sched_thermal_decay_shift;
1527 static inline u64 rq_clock_thermal(struct rq *rq)
1529 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1532 static inline void rq_clock_skip_update(struct rq *rq)
1534 lockdep_assert_rq_held(rq);
1535 rq->clock_update_flags |= RQCF_REQ_SKIP;
1539 * See rt task throttling, which is the only time a skip
1540 * request is canceled.
1542 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1544 lockdep_assert_rq_held(rq);
1545 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1549 unsigned long flags;
1550 struct pin_cookie cookie;
1551 #ifdef CONFIG_SCHED_DEBUG
1553 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1554 * current pin context is stashed here in case it needs to be
1555 * restored in rq_repin_lock().
1557 unsigned int clock_update_flags;
1561 extern struct callback_head balance_push_callback;
1564 * Lockdep annotation that avoids accidental unlocks; it's like a
1565 * sticky/continuous lockdep_assert_held().
1567 * This avoids code that has access to 'struct rq *rq' (basically everything in
1568 * the scheduler) from accidentally unlocking the rq if they do not also have a
1569 * copy of the (on-stack) 'struct rq_flags rf'.
1571 * Also see Documentation/locking/lockdep-design.rst.
1573 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1575 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1577 #ifdef CONFIG_SCHED_DEBUG
1578 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1579 rf->clock_update_flags = 0;
1581 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1586 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1588 #ifdef CONFIG_SCHED_DEBUG
1589 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1590 rf->clock_update_flags = RQCF_UPDATED;
1593 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1596 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1598 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1600 #ifdef CONFIG_SCHED_DEBUG
1602 * Restore the value we stashed in @rf for this pin context.
1604 rq->clock_update_flags |= rf->clock_update_flags;
1608 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1609 __acquires(rq->lock);
1611 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1612 __acquires(p->pi_lock)
1613 __acquires(rq->lock);
1615 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1616 __releases(rq->lock)
1618 rq_unpin_lock(rq, rf);
1619 raw_spin_rq_unlock(rq);
1623 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1624 __releases(rq->lock)
1625 __releases(p->pi_lock)
1627 rq_unpin_lock(rq, rf);
1628 raw_spin_rq_unlock(rq);
1629 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1633 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1634 __acquires(rq->lock)
1636 raw_spin_rq_lock_irqsave(rq, rf->flags);
1637 rq_pin_lock(rq, rf);
1641 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1642 __acquires(rq->lock)
1644 raw_spin_rq_lock_irq(rq);
1645 rq_pin_lock(rq, rf);
1649 rq_lock(struct rq *rq, struct rq_flags *rf)
1650 __acquires(rq->lock)
1652 raw_spin_rq_lock(rq);
1653 rq_pin_lock(rq, rf);
1657 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1658 __releases(rq->lock)
1660 rq_unpin_lock(rq, rf);
1661 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1665 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1666 __releases(rq->lock)
1668 rq_unpin_lock(rq, rf);
1669 raw_spin_rq_unlock_irq(rq);
1673 rq_unlock(struct rq *rq, struct rq_flags *rf)
1674 __releases(rq->lock)
1676 rq_unpin_lock(rq, rf);
1677 raw_spin_rq_unlock(rq);
1680 static inline struct rq *
1681 this_rq_lock_irq(struct rq_flags *rf)
1682 __acquires(rq->lock)
1686 local_irq_disable();
1693 enum numa_topology_type {
1698 extern enum numa_topology_type sched_numa_topology_type;
1699 extern int sched_max_numa_distance;
1700 extern bool find_numa_distance(int distance);
1701 extern void sched_init_numa(int offline_node);
1702 extern void sched_update_numa(int cpu, bool online);
1703 extern void sched_domains_numa_masks_set(unsigned int cpu);
1704 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1705 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1707 static inline void sched_init_numa(int offline_node) { }
1708 static inline void sched_update_numa(int cpu, bool online) { }
1709 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1710 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1711 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1717 #ifdef CONFIG_NUMA_BALANCING
1718 /* The regions in numa_faults array from task_struct */
1719 enum numa_faults_stats {
1725 extern void sched_setnuma(struct task_struct *p, int node);
1726 extern int migrate_task_to(struct task_struct *p, int cpu);
1727 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1729 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1732 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1735 #endif /* CONFIG_NUMA_BALANCING */
1740 queue_balance_callback(struct rq *rq,
1741 struct callback_head *head,
1742 void (*func)(struct rq *rq))
1744 lockdep_assert_rq_held(rq);
1747 * Don't (re)queue an already queued item; nor queue anything when
1748 * balance_push() is active, see the comment with
1749 * balance_push_callback.
1751 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1754 head->func = (void (*)(struct callback_head *))func;
1755 head->next = rq->balance_callback;
1756 rq->balance_callback = head;
1759 #define rcu_dereference_check_sched_domain(p) \
1760 rcu_dereference_check((p), \
1761 lockdep_is_held(&sched_domains_mutex))
1764 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1765 * See destroy_sched_domains: call_rcu for details.
1767 * The domain tree of any CPU may only be accessed from within
1768 * preempt-disabled sections.
1770 #define for_each_domain(cpu, __sd) \
1771 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1772 __sd; __sd = __sd->parent)
1775 * highest_flag_domain - Return highest sched_domain containing flag.
1776 * @cpu: The CPU whose highest level of sched domain is to
1778 * @flag: The flag to check for the highest sched_domain
1779 * for the given CPU.
1781 * Returns the highest sched_domain of a CPU which contains the given flag.
1783 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1785 struct sched_domain *sd, *hsd = NULL;
1787 for_each_domain(cpu, sd) {
1788 if (!(sd->flags & flag))
1796 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1798 struct sched_domain *sd;
1800 for_each_domain(cpu, sd) {
1801 if (sd->flags & flag)
1808 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1809 DECLARE_PER_CPU(int, sd_llc_size);
1810 DECLARE_PER_CPU(int, sd_llc_id);
1811 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1812 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1813 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1814 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1815 extern struct static_key_false sched_asym_cpucapacity;
1817 struct sched_group_capacity {
1820 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1823 unsigned long capacity;
1824 unsigned long min_capacity; /* Min per-CPU capacity in group */
1825 unsigned long max_capacity; /* Max per-CPU capacity in group */
1826 unsigned long next_update;
1827 int imbalance; /* XXX unrelated to capacity but shared group state */
1829 #ifdef CONFIG_SCHED_DEBUG
1833 unsigned long cpumask[]; /* Balance mask */
1836 struct sched_group {
1837 struct sched_group *next; /* Must be a circular list */
1840 unsigned int group_weight;
1841 struct sched_group_capacity *sgc;
1842 int asym_prefer_cpu; /* CPU of highest priority in group */
1846 * The CPUs this group covers.
1848 * NOTE: this field is variable length. (Allocated dynamically
1849 * by attaching extra space to the end of the structure,
1850 * depending on how many CPUs the kernel has booted up with)
1852 unsigned long cpumask[];
1855 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1857 return to_cpumask(sg->cpumask);
1861 * See build_balance_mask().
1863 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1865 return to_cpumask(sg->sgc->cpumask);
1868 extern int group_balance_cpu(struct sched_group *sg);
1870 #ifdef CONFIG_SCHED_DEBUG
1871 void update_sched_domain_debugfs(void);
1872 void dirty_sched_domain_sysctl(int cpu);
1874 static inline void update_sched_domain_debugfs(void)
1877 static inline void dirty_sched_domain_sysctl(int cpu)
1882 extern int sched_update_scaling(void);
1883 #endif /* CONFIG_SMP */
1887 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1889 extern void __sched_core_account_forceidle(struct rq *rq);
1891 static inline void sched_core_account_forceidle(struct rq *rq)
1893 if (schedstat_enabled())
1894 __sched_core_account_forceidle(rq);
1897 extern void __sched_core_tick(struct rq *rq);
1899 static inline void sched_core_tick(struct rq *rq)
1901 if (sched_core_enabled(rq) && schedstat_enabled())
1902 __sched_core_tick(rq);
1907 static inline void sched_core_account_forceidle(struct rq *rq) {}
1909 static inline void sched_core_tick(struct rq *rq) {}
1911 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1913 #ifdef CONFIG_CGROUP_SCHED
1916 * Return the group to which this tasks belongs.
1918 * We cannot use task_css() and friends because the cgroup subsystem
1919 * changes that value before the cgroup_subsys::attach() method is called,
1920 * therefore we cannot pin it and might observe the wrong value.
1922 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1923 * core changes this before calling sched_move_task().
1925 * Instead we use a 'copy' which is updated from sched_move_task() while
1926 * holding both task_struct::pi_lock and rq::lock.
1928 static inline struct task_group *task_group(struct task_struct *p)
1930 return p->sched_task_group;
1933 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1934 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1936 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1937 struct task_group *tg = task_group(p);
1940 #ifdef CONFIG_FAIR_GROUP_SCHED
1941 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1942 p->se.cfs_rq = tg->cfs_rq[cpu];
1943 p->se.parent = tg->se[cpu];
1946 #ifdef CONFIG_RT_GROUP_SCHED
1947 p->rt.rt_rq = tg->rt_rq[cpu];
1948 p->rt.parent = tg->rt_se[cpu];
1952 #else /* CONFIG_CGROUP_SCHED */
1954 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1955 static inline struct task_group *task_group(struct task_struct *p)
1960 #endif /* CONFIG_CGROUP_SCHED */
1962 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1964 set_task_rq(p, cpu);
1967 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1968 * successfully executed on another CPU. We must ensure that updates of
1969 * per-task data have been completed by this moment.
1972 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1978 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1980 #ifdef CONFIG_SCHED_DEBUG
1981 # define const_debug __read_mostly
1983 # define const_debug const
1986 #define SCHED_FEAT(name, enabled) \
1987 __SCHED_FEAT_##name ,
1990 #include "features.h"
1996 #ifdef CONFIG_SCHED_DEBUG
1999 * To support run-time toggling of sched features, all the translation units
2000 * (but core.c) reference the sysctl_sched_features defined in core.c.
2002 extern const_debug unsigned int sysctl_sched_features;
2004 #ifdef CONFIG_JUMP_LABEL
2005 #define SCHED_FEAT(name, enabled) \
2006 static __always_inline bool static_branch_##name(struct static_key *key) \
2008 return static_key_##enabled(key); \
2011 #include "features.h"
2014 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2015 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2017 #else /* !CONFIG_JUMP_LABEL */
2019 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2021 #endif /* CONFIG_JUMP_LABEL */
2023 #else /* !SCHED_DEBUG */
2026 * Each translation unit has its own copy of sysctl_sched_features to allow
2027 * constants propagation at compile time and compiler optimization based on
2030 #define SCHED_FEAT(name, enabled) \
2031 (1UL << __SCHED_FEAT_##name) * enabled |
2032 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2033 #include "features.h"
2037 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2039 #endif /* SCHED_DEBUG */
2041 extern struct static_key_false sched_numa_balancing;
2042 extern struct static_key_false sched_schedstats;
2044 static inline u64 global_rt_period(void)
2046 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2049 static inline u64 global_rt_runtime(void)
2051 if (sysctl_sched_rt_runtime < 0)
2054 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2057 static inline int task_current(struct rq *rq, struct task_struct *p)
2059 return rq->curr == p;
2062 static inline int task_running(struct rq *rq, struct task_struct *p)
2067 return task_current(rq, p);
2071 static inline int task_on_rq_queued(struct task_struct *p)
2073 return p->on_rq == TASK_ON_RQ_QUEUED;
2076 static inline int task_on_rq_migrating(struct task_struct *p)
2078 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2081 /* Wake flags. The first three directly map to some SD flag value */
2082 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2083 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2084 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2086 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2087 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2090 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2091 static_assert(WF_FORK == SD_BALANCE_FORK);
2092 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2096 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2097 * of tasks with abnormal "nice" values across CPUs the contribution that
2098 * each task makes to its run queue's load is weighted according to its
2099 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2100 * scaled version of the new time slice allocation that they receive on time
2104 #define WEIGHT_IDLEPRIO 3
2105 #define WMULT_IDLEPRIO 1431655765
2107 extern const int sched_prio_to_weight[40];
2108 extern const u32 sched_prio_to_wmult[40];
2111 * {de,en}queue flags:
2113 * DEQUEUE_SLEEP - task is no longer runnable
2114 * ENQUEUE_WAKEUP - task just became runnable
2116 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2117 * are in a known state which allows modification. Such pairs
2118 * should preserve as much state as possible.
2120 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2123 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2124 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2125 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2129 #define DEQUEUE_SLEEP 0x01
2130 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2131 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2132 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2134 #define ENQUEUE_WAKEUP 0x01
2135 #define ENQUEUE_RESTORE 0x02
2136 #define ENQUEUE_MOVE 0x04
2137 #define ENQUEUE_NOCLOCK 0x08
2139 #define ENQUEUE_HEAD 0x10
2140 #define ENQUEUE_REPLENISH 0x20
2142 #define ENQUEUE_MIGRATED 0x40
2144 #define ENQUEUE_MIGRATED 0x00
2147 #define RETRY_TASK ((void *)-1UL)
2149 struct sched_class {
2151 #ifdef CONFIG_UCLAMP_TASK
2155 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2156 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2157 void (*yield_task) (struct rq *rq);
2158 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2160 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2162 struct task_struct *(*pick_next_task)(struct rq *rq);
2164 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2165 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2168 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2169 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2171 struct task_struct * (*pick_task)(struct rq *rq);
2173 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2175 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2177 void (*set_cpus_allowed)(struct task_struct *p,
2178 const struct cpumask *newmask,
2181 void (*rq_online)(struct rq *rq);
2182 void (*rq_offline)(struct rq *rq);
2184 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2187 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2188 void (*task_fork)(struct task_struct *p);
2189 void (*task_dead)(struct task_struct *p);
2192 * The switched_from() call is allowed to drop rq->lock, therefore we
2193 * cannot assume the switched_from/switched_to pair is serialized by
2194 * rq->lock. They are however serialized by p->pi_lock.
2196 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2197 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2198 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2201 unsigned int (*get_rr_interval)(struct rq *rq,
2202 struct task_struct *task);
2204 void (*update_curr)(struct rq *rq);
2206 #define TASK_SET_GROUP 0
2207 #define TASK_MOVE_GROUP 1
2209 #ifdef CONFIG_FAIR_GROUP_SCHED
2210 void (*task_change_group)(struct task_struct *p, int type);
2214 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2216 WARN_ON_ONCE(rq->curr != prev);
2217 prev->sched_class->put_prev_task(rq, prev);
2220 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2222 next->sched_class->set_next_task(rq, next, false);
2227 * Helper to define a sched_class instance; each one is placed in a separate
2228 * section which is ordered by the linker script:
2230 * include/asm-generic/vmlinux.lds.h
2232 * *CAREFUL* they are laid out in *REVERSE* order!!!
2234 * Also enforce alignment on the instance, not the type, to guarantee layout.
2236 #define DEFINE_SCHED_CLASS(name) \
2237 const struct sched_class name##_sched_class \
2238 __aligned(__alignof__(struct sched_class)) \
2239 __section("__" #name "_sched_class")
2241 /* Defined in include/asm-generic/vmlinux.lds.h */
2242 extern struct sched_class __sched_class_highest[];
2243 extern struct sched_class __sched_class_lowest[];
2245 #define for_class_range(class, _from, _to) \
2246 for (class = (_from); class < (_to); class++)
2248 #define for_each_class(class) \
2249 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2251 #define sched_class_above(_a, _b) ((_a) < (_b))
2253 extern const struct sched_class stop_sched_class;
2254 extern const struct sched_class dl_sched_class;
2255 extern const struct sched_class rt_sched_class;
2256 extern const struct sched_class fair_sched_class;
2257 extern const struct sched_class idle_sched_class;
2259 static inline bool sched_stop_runnable(struct rq *rq)
2261 return rq->stop && task_on_rq_queued(rq->stop);
2264 static inline bool sched_dl_runnable(struct rq *rq)
2266 return rq->dl.dl_nr_running > 0;
2269 static inline bool sched_rt_runnable(struct rq *rq)
2271 return rq->rt.rt_queued > 0;
2274 static inline bool sched_fair_runnable(struct rq *rq)
2276 return rq->cfs.nr_running > 0;
2279 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2280 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2282 #define SCA_CHECK 0x01
2283 #define SCA_MIGRATE_DISABLE 0x02
2284 #define SCA_MIGRATE_ENABLE 0x04
2285 #define SCA_USER 0x08
2289 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2291 extern void trigger_load_balance(struct rq *rq);
2293 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2295 static inline struct task_struct *get_push_task(struct rq *rq)
2297 struct task_struct *p = rq->curr;
2299 lockdep_assert_rq_held(rq);
2304 if (p->nr_cpus_allowed == 1)
2307 if (p->migration_disabled)
2310 rq->push_busy = true;
2311 return get_task_struct(p);
2314 extern int push_cpu_stop(void *arg);
2318 #ifdef CONFIG_CPU_IDLE
2319 static inline void idle_set_state(struct rq *rq,
2320 struct cpuidle_state *idle_state)
2322 rq->idle_state = idle_state;
2325 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2327 SCHED_WARN_ON(!rcu_read_lock_held());
2329 return rq->idle_state;
2332 static inline void idle_set_state(struct rq *rq,
2333 struct cpuidle_state *idle_state)
2337 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2343 extern void schedule_idle(void);
2345 extern void sysrq_sched_debug_show(void);
2346 extern void sched_init_granularity(void);
2347 extern void update_max_interval(void);
2349 extern void init_sched_dl_class(void);
2350 extern void init_sched_rt_class(void);
2351 extern void init_sched_fair_class(void);
2353 extern void reweight_task(struct task_struct *p, int prio);
2355 extern void resched_curr(struct rq *rq);
2356 extern void resched_cpu(int cpu);
2358 extern struct rt_bandwidth def_rt_bandwidth;
2359 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2360 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2362 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2363 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2364 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2367 #define BW_UNIT (1 << BW_SHIFT)
2368 #define RATIO_SHIFT 8
2369 #define MAX_BW_BITS (64 - BW_SHIFT)
2370 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2371 unsigned long to_ratio(u64 period, u64 runtime);
2373 extern void init_entity_runnable_average(struct sched_entity *se);
2374 extern void post_init_entity_util_avg(struct task_struct *p);
2376 #ifdef CONFIG_NO_HZ_FULL
2377 extern bool sched_can_stop_tick(struct rq *rq);
2378 extern int __init sched_tick_offload_init(void);
2381 * Tick may be needed by tasks in the runqueue depending on their policy and
2382 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2383 * nohz mode if necessary.
2385 static inline void sched_update_tick_dependency(struct rq *rq)
2387 int cpu = cpu_of(rq);
2389 if (!tick_nohz_full_cpu(cpu))
2392 if (sched_can_stop_tick(rq))
2393 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2395 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2398 static inline int sched_tick_offload_init(void) { return 0; }
2399 static inline void sched_update_tick_dependency(struct rq *rq) { }
2402 static inline void add_nr_running(struct rq *rq, unsigned count)
2404 unsigned prev_nr = rq->nr_running;
2406 rq->nr_running = prev_nr + count;
2407 if (trace_sched_update_nr_running_tp_enabled()) {
2408 call_trace_sched_update_nr_running(rq, count);
2412 if (prev_nr < 2 && rq->nr_running >= 2) {
2413 if (!READ_ONCE(rq->rd->overload))
2414 WRITE_ONCE(rq->rd->overload, 1);
2418 sched_update_tick_dependency(rq);
2421 static inline void sub_nr_running(struct rq *rq, unsigned count)
2423 rq->nr_running -= count;
2424 if (trace_sched_update_nr_running_tp_enabled()) {
2425 call_trace_sched_update_nr_running(rq, -count);
2428 /* Check if we still need preemption */
2429 sched_update_tick_dependency(rq);
2432 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2433 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2435 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2437 extern const_debug unsigned int sysctl_sched_nr_migrate;
2438 extern const_debug unsigned int sysctl_sched_migration_cost;
2440 #ifdef CONFIG_SCHED_DEBUG
2441 extern unsigned int sysctl_sched_latency;
2442 extern unsigned int sysctl_sched_min_granularity;
2443 extern unsigned int sysctl_sched_idle_min_granularity;
2444 extern unsigned int sysctl_sched_wakeup_granularity;
2445 extern int sysctl_resched_latency_warn_ms;
2446 extern int sysctl_resched_latency_warn_once;
2448 extern unsigned int sysctl_sched_tunable_scaling;
2450 extern unsigned int sysctl_numa_balancing_scan_delay;
2451 extern unsigned int sysctl_numa_balancing_scan_period_min;
2452 extern unsigned int sysctl_numa_balancing_scan_period_max;
2453 extern unsigned int sysctl_numa_balancing_scan_size;
2456 #ifdef CONFIG_SCHED_HRTICK
2460 * - enabled by features
2461 * - hrtimer is actually high res
2463 static inline int hrtick_enabled(struct rq *rq)
2465 if (!cpu_active(cpu_of(rq)))
2467 return hrtimer_is_hres_active(&rq->hrtick_timer);
2470 static inline int hrtick_enabled_fair(struct rq *rq)
2472 if (!sched_feat(HRTICK))
2474 return hrtick_enabled(rq);
2477 static inline int hrtick_enabled_dl(struct rq *rq)
2479 if (!sched_feat(HRTICK_DL))
2481 return hrtick_enabled(rq);
2484 void hrtick_start(struct rq *rq, u64 delay);
2488 static inline int hrtick_enabled_fair(struct rq *rq)
2493 static inline int hrtick_enabled_dl(struct rq *rq)
2498 static inline int hrtick_enabled(struct rq *rq)
2503 #endif /* CONFIG_SCHED_HRTICK */
2505 #ifndef arch_scale_freq_tick
2506 static __always_inline
2507 void arch_scale_freq_tick(void)
2512 #ifndef arch_scale_freq_capacity
2514 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2515 * @cpu: the CPU in question.
2517 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2520 * ------ * SCHED_CAPACITY_SCALE
2523 static __always_inline
2524 unsigned long arch_scale_freq_capacity(int cpu)
2526 return SCHED_CAPACITY_SCALE;
2530 #ifdef CONFIG_SCHED_DEBUG
2532 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2533 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2534 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2535 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2537 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2539 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2540 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2542 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2546 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2551 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2553 #ifdef CONFIG_SCHED_CORE
2555 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2556 * order by core-id first and cpu-id second.
2560 * double_rq_lock(0,3); will take core-0, core-1 lock
2561 * double_rq_lock(1,2); will take core-1, core-0 lock
2563 * when only cpu-id is considered.
2565 if (rq1->core->cpu < rq2->core->cpu)
2567 if (rq1->core->cpu > rq2->core->cpu)
2571 * __sched_core_flip() relies on SMT having cpu-id lock order.
2574 return rq1->cpu < rq2->cpu;
2577 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2579 #ifdef CONFIG_PREEMPTION
2582 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2583 * way at the expense of forcing extra atomic operations in all
2584 * invocations. This assures that the double_lock is acquired using the
2585 * same underlying policy as the spinlock_t on this architecture, which
2586 * reduces latency compared to the unfair variant below. However, it
2587 * also adds more overhead and therefore may reduce throughput.
2589 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2590 __releases(this_rq->lock)
2591 __acquires(busiest->lock)
2592 __acquires(this_rq->lock)
2594 raw_spin_rq_unlock(this_rq);
2595 double_rq_lock(this_rq, busiest);
2602 * Unfair double_lock_balance: Optimizes throughput at the expense of
2603 * latency by eliminating extra atomic operations when the locks are
2604 * already in proper order on entry. This favors lower CPU-ids and will
2605 * grant the double lock to lower CPUs over higher ids under contention,
2606 * regardless of entry order into the function.
2608 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2609 __releases(this_rq->lock)
2610 __acquires(busiest->lock)
2611 __acquires(this_rq->lock)
2613 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2614 likely(raw_spin_rq_trylock(busiest))) {
2615 double_rq_clock_clear_update(this_rq, busiest);
2619 if (rq_order_less(this_rq, busiest)) {
2620 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2621 double_rq_clock_clear_update(this_rq, busiest);
2625 raw_spin_rq_unlock(this_rq);
2626 double_rq_lock(this_rq, busiest);
2631 #endif /* CONFIG_PREEMPTION */
2634 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2636 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2638 lockdep_assert_irqs_disabled();
2640 return _double_lock_balance(this_rq, busiest);
2643 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2644 __releases(busiest->lock)
2646 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2647 raw_spin_rq_unlock(busiest);
2648 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2651 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2657 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2660 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2666 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2669 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2675 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2679 * double_rq_unlock - safely unlock two runqueues
2681 * Note this does not restore interrupts like task_rq_unlock,
2682 * you need to do so manually after calling.
2684 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2685 __releases(rq1->lock)
2686 __releases(rq2->lock)
2688 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2689 raw_spin_rq_unlock(rq2);
2691 __release(rq2->lock);
2692 raw_spin_rq_unlock(rq1);
2695 extern void set_rq_online (struct rq *rq);
2696 extern void set_rq_offline(struct rq *rq);
2697 extern bool sched_smp_initialized;
2699 #else /* CONFIG_SMP */
2702 * double_rq_lock - safely lock two runqueues
2704 * Note this does not disable interrupts like task_rq_lock,
2705 * you need to do so manually before calling.
2707 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2708 __acquires(rq1->lock)
2709 __acquires(rq2->lock)
2711 BUG_ON(!irqs_disabled());
2713 raw_spin_rq_lock(rq1);
2714 __acquire(rq2->lock); /* Fake it out ;) */
2715 double_rq_clock_clear_update(rq1, rq2);
2719 * double_rq_unlock - safely unlock two runqueues
2721 * Note this does not restore interrupts like task_rq_unlock,
2722 * you need to do so manually after calling.
2724 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2725 __releases(rq1->lock)
2726 __releases(rq2->lock)
2729 raw_spin_rq_unlock(rq1);
2730 __release(rq2->lock);
2735 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2736 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2738 #ifdef CONFIG_SCHED_DEBUG
2739 extern bool sched_debug_verbose;
2741 extern void print_cfs_stats(struct seq_file *m, int cpu);
2742 extern void print_rt_stats(struct seq_file *m, int cpu);
2743 extern void print_dl_stats(struct seq_file *m, int cpu);
2744 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2745 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2746 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2748 extern void resched_latency_warn(int cpu, u64 latency);
2749 #ifdef CONFIG_NUMA_BALANCING
2751 show_numa_stats(struct task_struct *p, struct seq_file *m);
2753 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2754 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2755 #endif /* CONFIG_NUMA_BALANCING */
2757 static inline void resched_latency_warn(int cpu, u64 latency) {}
2758 #endif /* CONFIG_SCHED_DEBUG */
2760 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2761 extern void init_rt_rq(struct rt_rq *rt_rq);
2762 extern void init_dl_rq(struct dl_rq *dl_rq);
2764 extern void cfs_bandwidth_usage_inc(void);
2765 extern void cfs_bandwidth_usage_dec(void);
2767 #ifdef CONFIG_NO_HZ_COMMON
2768 #define NOHZ_BALANCE_KICK_BIT 0
2769 #define NOHZ_STATS_KICK_BIT 1
2770 #define NOHZ_NEWILB_KICK_BIT 2
2771 #define NOHZ_NEXT_KICK_BIT 3
2773 /* Run rebalance_domains() */
2774 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2775 /* Update blocked load */
2776 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2777 /* Update blocked load when entering idle */
2778 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2779 /* Update nohz.next_balance */
2780 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2782 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2784 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2786 extern void nohz_balance_exit_idle(struct rq *rq);
2788 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2791 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2792 extern void nohz_run_idle_balance(int cpu);
2794 static inline void nohz_run_idle_balance(int cpu) { }
2797 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2802 struct u64_stats_sync sync;
2805 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2808 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2809 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2810 * and never move forward.
2812 static inline u64 irq_time_read(int cpu)
2814 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2819 seq = __u64_stats_fetch_begin(&irqtime->sync);
2820 total = irqtime->total;
2821 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2825 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2827 #ifdef CONFIG_CPU_FREQ
2828 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2831 * cpufreq_update_util - Take a note about CPU utilization changes.
2832 * @rq: Runqueue to carry out the update for.
2833 * @flags: Update reason flags.
2835 * This function is called by the scheduler on the CPU whose utilization is
2838 * It can only be called from RCU-sched read-side critical sections.
2840 * The way cpufreq is currently arranged requires it to evaluate the CPU
2841 * performance state (frequency/voltage) on a regular basis to prevent it from
2842 * being stuck in a completely inadequate performance level for too long.
2843 * That is not guaranteed to happen if the updates are only triggered from CFS
2844 * and DL, though, because they may not be coming in if only RT tasks are
2845 * active all the time (or there are RT tasks only).
2847 * As a workaround for that issue, this function is called periodically by the
2848 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2849 * but that really is a band-aid. Going forward it should be replaced with
2850 * solutions targeted more specifically at RT tasks.
2852 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2854 struct update_util_data *data;
2856 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2859 data->func(data, rq_clock(rq), flags);
2862 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2863 #endif /* CONFIG_CPU_FREQ */
2865 #ifdef arch_scale_freq_capacity
2866 # ifndef arch_scale_freq_invariant
2867 # define arch_scale_freq_invariant() true
2870 # define arch_scale_freq_invariant() false
2874 static inline unsigned long capacity_orig_of(int cpu)
2876 return cpu_rq(cpu)->cpu_capacity_orig;
2880 * enum cpu_util_type - CPU utilization type
2881 * @FREQUENCY_UTIL: Utilization used to select frequency
2882 * @ENERGY_UTIL: Utilization used during energy calculation
2884 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2885 * need to be aggregated differently depending on the usage made of them. This
2886 * enum is used within effective_cpu_util() to differentiate the types of
2887 * utilization expected by the callers, and adjust the aggregation accordingly.
2889 enum cpu_util_type {
2894 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2895 enum cpu_util_type type,
2896 struct task_struct *p);
2898 static inline unsigned long cpu_bw_dl(struct rq *rq)
2900 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2903 static inline unsigned long cpu_util_dl(struct rq *rq)
2905 return READ_ONCE(rq->avg_dl.util_avg);
2909 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2910 * @cpu: the CPU to get the utilization for.
2912 * The unit of the return value must be the same as the one of CPU capacity
2913 * so that CPU utilization can be compared with CPU capacity.
2915 * CPU utilization is the sum of running time of runnable tasks plus the
2916 * recent utilization of currently non-runnable tasks on that CPU.
2917 * It represents the amount of CPU capacity currently used by CFS tasks in
2918 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2919 * capacity at f_max.
2921 * The estimated CPU utilization is defined as the maximum between CPU
2922 * utilization and sum of the estimated utilization of the currently
2923 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2924 * previously-executed tasks, which helps better deduce how busy a CPU will
2925 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2926 * of such a task would be significantly decayed at this point of time.
2928 * CPU utilization can be higher than the current CPU capacity
2929 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2930 * of rounding errors as well as task migrations or wakeups of new tasks.
2931 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2932 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2933 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2934 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2935 * though since this is useful for predicting the CPU capacity required
2936 * after task migrations (scheduler-driven DVFS).
2938 * Return: (Estimated) utilization for the specified CPU.
2940 static inline unsigned long cpu_util_cfs(int cpu)
2942 struct cfs_rq *cfs_rq;
2945 cfs_rq = &cpu_rq(cpu)->cfs;
2946 util = READ_ONCE(cfs_rq->avg.util_avg);
2948 if (sched_feat(UTIL_EST)) {
2949 util = max_t(unsigned long, util,
2950 READ_ONCE(cfs_rq->avg.util_est.enqueued));
2953 return min(util, capacity_orig_of(cpu));
2956 static inline unsigned long cpu_util_rt(struct rq *rq)
2958 return READ_ONCE(rq->avg_rt.util_avg);
2962 #ifdef CONFIG_UCLAMP_TASK
2963 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2966 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2967 * @rq: The rq to clamp against. Must not be NULL.
2968 * @util: The util value to clamp.
2969 * @p: The task to clamp against. Can be NULL if you want to clamp
2972 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2974 * If sched_uclamp_used static key is disabled, then just return the util
2975 * without any clamping since uclamp aggregation at the rq level in the fast
2976 * path is disabled, rendering this operation a NOP.
2978 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2979 * will return the correct effective uclamp value of the task even if the
2980 * static key is disabled.
2982 static __always_inline
2983 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2984 struct task_struct *p)
2986 unsigned long min_util = 0;
2987 unsigned long max_util = 0;
2989 if (!static_branch_likely(&sched_uclamp_used))
2993 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2994 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2997 * Ignore last runnable task's max clamp, as this task will
2998 * reset it. Similarly, no need to read the rq's min clamp.
3000 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
3004 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
3005 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
3008 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3009 * RUNNABLE tasks with _different_ clamps, we can end up with an
3010 * inversion. Fix it now when the clamps are applied.
3012 if (unlikely(min_util >= max_util))
3015 return clamp(util, min_util, max_util);
3018 /* Is the rq being capped/throttled by uclamp_max? */
3019 static inline bool uclamp_rq_is_capped(struct rq *rq)
3021 unsigned long rq_util;
3022 unsigned long max_util;
3024 if (!static_branch_likely(&sched_uclamp_used))
3027 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3028 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3030 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3034 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3035 * by default in the fast path and only gets turned on once userspace performs
3036 * an operation that requires it.
3038 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3041 static inline bool uclamp_is_used(void)
3043 return static_branch_likely(&sched_uclamp_used);
3045 #else /* CONFIG_UCLAMP_TASK */
3047 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3048 struct task_struct *p)
3053 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3055 static inline bool uclamp_is_used(void)
3059 #endif /* CONFIG_UCLAMP_TASK */
3061 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3062 static inline unsigned long cpu_util_irq(struct rq *rq)
3064 return rq->avg_irq.util_avg;
3068 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3070 util *= (max - irq);
3077 static inline unsigned long cpu_util_irq(struct rq *rq)
3083 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3089 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3091 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3093 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3095 static inline bool sched_energy_enabled(void)
3097 return static_branch_unlikely(&sched_energy_present);
3100 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3102 #define perf_domain_span(pd) NULL
3103 static inline bool sched_energy_enabled(void) { return false; }
3105 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3107 #ifdef CONFIG_MEMBARRIER
3109 * The scheduler provides memory barriers required by membarrier between:
3110 * - prior user-space memory accesses and store to rq->membarrier_state,
3111 * - store to rq->membarrier_state and following user-space memory accesses.
3112 * In the same way it provides those guarantees around store to rq->curr.
3114 static inline void membarrier_switch_mm(struct rq *rq,
3115 struct mm_struct *prev_mm,
3116 struct mm_struct *next_mm)
3118 int membarrier_state;
3120 if (prev_mm == next_mm)
3123 membarrier_state = atomic_read(&next_mm->membarrier_state);
3124 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3127 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3130 static inline void membarrier_switch_mm(struct rq *rq,
3131 struct mm_struct *prev_mm,
3132 struct mm_struct *next_mm)
3138 static inline bool is_per_cpu_kthread(struct task_struct *p)
3140 if (!(p->flags & PF_KTHREAD))
3143 if (p->nr_cpus_allowed != 1)
3150 extern void swake_up_all_locked(struct swait_queue_head *q);
3151 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3153 #ifdef CONFIG_PREEMPT_DYNAMIC
3154 extern int preempt_dynamic_mode;
3155 extern int sched_dynamic_mode(const char *str);
3156 extern void sched_dynamic_update(int mode);
3159 #endif /* _KERNEL_SCHED_SCHED_H */