1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
76 #include "cpudeadline.h"
78 #include <trace/events/sched.h>
80 #ifdef CONFIG_SCHED_DEBUG
81 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
83 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
89 /* task_struct::on_rq states: */
90 #define TASK_ON_RQ_QUEUED 1
91 #define TASK_ON_RQ_MIGRATING 2
93 extern __read_mostly int scheduler_running;
95 extern unsigned long calc_load_update;
96 extern atomic_long_t calc_load_tasks;
98 extern void calc_global_load_tick(struct rq *this_rq);
99 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
101 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
103 * Helpers for converting nanosecond timing to jiffy resolution
105 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108 * Increase resolution of nice-level calculations for 64-bit architectures.
109 * The extra resolution improves shares distribution and load balancing of
110 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
111 * hierarchies, especially on larger systems. This is not a user-visible change
112 * and does not change the user-interface for setting shares/weights.
114 * We increase resolution only if we have enough bits to allow this increased
115 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
116 * are pretty high and the returns do not justify the increased costs.
118 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
119 * increase coverage and consistency always enable it on 64-bit platforms.
122 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
123 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load_down(w) \
126 unsigned long __w = (w); \
128 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
132 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
133 # define scale_load(w) (w)
134 # define scale_load_down(w) (w)
138 * Task weight (visible to users) and its load (invisible to users) have
139 * independent resolution, but they should be well calibrated. We use
140 * scale_load() and scale_load_down(w) to convert between them. The
141 * following must be true:
143 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
146 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
149 * Single value that decides SCHED_DEADLINE internal math precision.
150 * 10 -> just above 1us
151 * 9 -> just above 0.5us
156 * Single value that denotes runtime == period, ie unlimited time.
158 #define RUNTIME_INF ((u64)~0ULL)
160 static inline int idle_policy(int policy)
162 return policy == SCHED_IDLE;
164 static inline int fair_policy(int policy)
166 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
169 static inline int rt_policy(int policy)
171 return policy == SCHED_FIFO || policy == SCHED_RR;
174 static inline int dl_policy(int policy)
176 return policy == SCHED_DEADLINE;
178 static inline bool valid_policy(int policy)
180 return idle_policy(policy) || fair_policy(policy) ||
181 rt_policy(policy) || dl_policy(policy);
184 static inline int task_has_idle_policy(struct task_struct *p)
186 return idle_policy(p->policy);
189 static inline int task_has_rt_policy(struct task_struct *p)
191 return rt_policy(p->policy);
194 static inline int task_has_dl_policy(struct task_struct *p)
196 return dl_policy(p->policy);
199 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
201 static inline void update_avg(u64 *avg, u64 sample)
203 s64 diff = sample - *avg;
208 * Shifting a value by an exponent greater *or equal* to the size of said value
209 * is UB; cap at size-1.
211 #define shr_bound(val, shift) \
212 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
215 * !! For sched_setattr_nocheck() (kernel) only !!
217 * This is actually gross. :(
219 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
220 * tasks, but still be able to sleep. We need this on platforms that cannot
221 * atomically change clock frequency. Remove once fast switching will be
222 * available on such platforms.
224 * SUGOV stands for SchedUtil GOVernor.
226 #define SCHED_FLAG_SUGOV 0x10000000
228 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
230 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
231 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
238 * Tells if entity @a should preempt entity @b.
241 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
243 return dl_entity_is_special(a) ||
244 dl_time_before(a->deadline, b->deadline);
248 * This is the priority-queue data structure of the RT scheduling class:
250 struct rt_prio_array {
251 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
252 struct list_head queue[MAX_RT_PRIO];
255 struct rt_bandwidth {
256 /* nests inside the rq lock: */
257 raw_spinlock_t rt_runtime_lock;
260 struct hrtimer rt_period_timer;
261 unsigned int rt_period_active;
264 void __dl_clear_params(struct task_struct *p);
266 struct dl_bandwidth {
267 raw_spinlock_t dl_runtime_lock;
272 static inline int dl_bandwidth_enabled(void)
274 return sysctl_sched_rt_runtime >= 0;
278 * To keep the bandwidth of -deadline tasks under control
279 * we need some place where:
280 * - store the maximum -deadline bandwidth of each cpu;
281 * - cache the fraction of bandwidth that is currently allocated in
284 * This is all done in the data structure below. It is similar to the
285 * one used for RT-throttling (rt_bandwidth), with the main difference
286 * that, since here we are only interested in admission control, we
287 * do not decrease any runtime while the group "executes", neither we
288 * need a timer to replenish it.
290 * With respect to SMP, bandwidth is given on a per root domain basis,
292 * - bw (< 100%) is the deadline bandwidth of each CPU;
293 * - total_bw is the currently allocated bandwidth in each root domain;
301 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
304 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
306 dl_b->total_bw -= tsk_bw;
307 __dl_update(dl_b, (s32)tsk_bw / cpus);
311 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
313 dl_b->total_bw += tsk_bw;
314 __dl_update(dl_b, -((s32)tsk_bw / cpus));
317 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
318 u64 old_bw, u64 new_bw)
320 return dl_b->bw != -1 &&
321 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
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_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
348 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
349 extern bool dl_cpu_busy(unsigned int cpu);
351 #ifdef CONFIG_CGROUP_SCHED
353 #include <linux/cgroup.h>
354 #include <linux/psi.h>
359 extern struct list_head task_groups;
361 struct cfs_bandwidth {
362 #ifdef CONFIG_CFS_BANDWIDTH
367 s64 hierarchical_quota;
372 struct hrtimer period_timer;
373 struct hrtimer slack_timer;
374 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;
396 * load_avg can be heavily contended at clock tick time, so put
397 * it in its own cacheline separated from the fields above which
398 * will also be accessed at each tick.
400 atomic_long_t load_avg ____cacheline_aligned;
404 #ifdef CONFIG_RT_GROUP_SCHED
405 struct sched_rt_entity **rt_se;
406 struct rt_rq **rt_rq;
408 struct rt_bandwidth rt_bandwidth;
412 struct list_head list;
414 struct task_group *parent;
415 struct list_head siblings;
416 struct list_head children;
418 #ifdef CONFIG_SCHED_AUTOGROUP
419 struct autogroup *autogroup;
422 struct cfs_bandwidth cfs_bandwidth;
424 #ifdef CONFIG_UCLAMP_TASK_GROUP
425 /* The two decimal precision [%] value requested from user-space */
426 unsigned int uclamp_pct[UCLAMP_CNT];
427 /* Clamp values requested for a task group */
428 struct uclamp_se uclamp_req[UCLAMP_CNT];
429 /* Effective clamp values used for a task group */
430 struct uclamp_se uclamp[UCLAMP_CNT];
435 #ifdef CONFIG_FAIR_GROUP_SCHED
436 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
439 * A weight of 0 or 1 can cause arithmetics problems.
440 * A weight of a cfs_rq is the sum of weights of which entities
441 * are queued on this cfs_rq, so a weight of a entity should not be
442 * too large, so as the shares value of a task group.
443 * (The default weight is 1024 - so there's no practical
444 * limitation from this.)
446 #define MIN_SHARES (1UL << 1)
447 #define MAX_SHARES (1UL << 18)
450 typedef int (*tg_visitor)(struct task_group *, void *);
452 extern int walk_tg_tree_from(struct task_group *from,
453 tg_visitor down, tg_visitor up, void *data);
456 * Iterate the full tree, calling @down when first entering a node and @up when
457 * leaving it for the final time.
459 * Caller must hold rcu_lock or sufficient equivalent.
461 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
463 return walk_tg_tree_from(&root_task_group, down, up, data);
466 extern int tg_nop(struct task_group *tg, void *data);
468 extern void free_fair_sched_group(struct task_group *tg);
469 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
470 extern void online_fair_sched_group(struct task_group *tg);
471 extern void unregister_fair_sched_group(struct task_group *tg);
472 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
473 struct sched_entity *se, int cpu,
474 struct sched_entity *parent);
475 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
477 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
478 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
479 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
481 extern void free_rt_sched_group(struct task_group *tg);
482 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
483 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
484 struct sched_rt_entity *rt_se, int cpu,
485 struct sched_rt_entity *parent);
486 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
487 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
488 extern long sched_group_rt_runtime(struct task_group *tg);
489 extern long sched_group_rt_period(struct task_group *tg);
490 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
492 extern struct task_group *sched_create_group(struct task_group *parent);
493 extern void sched_online_group(struct task_group *tg,
494 struct task_group *parent);
495 extern void sched_destroy_group(struct task_group *tg);
496 extern void sched_offline_group(struct task_group *tg);
498 extern void sched_move_task(struct task_struct *tsk);
500 #ifdef CONFIG_FAIR_GROUP_SCHED
501 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
504 extern void set_task_rq_fair(struct sched_entity *se,
505 struct cfs_rq *prev, struct cfs_rq *next);
506 #else /* !CONFIG_SMP */
507 static inline void set_task_rq_fair(struct sched_entity *se,
508 struct cfs_rq *prev, struct cfs_rq *next) { }
509 #endif /* CONFIG_SMP */
510 #endif /* CONFIG_FAIR_GROUP_SCHED */
512 #else /* CONFIG_CGROUP_SCHED */
514 struct cfs_bandwidth { };
516 #endif /* CONFIG_CGROUP_SCHED */
518 /* CFS-related fields in a runqueue */
520 struct load_weight load;
521 unsigned int nr_running;
522 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
523 unsigned int idle_h_nr_running; /* SCHED_IDLE */
528 u64 min_vruntime_copy;
531 struct rb_root_cached tasks_timeline;
534 * 'curr' points to currently running entity on this cfs_rq.
535 * It is set to NULL otherwise (i.e when none are currently running).
537 struct sched_entity *curr;
538 struct sched_entity *next;
539 struct sched_entity *last;
540 struct sched_entity *skip;
542 #ifdef CONFIG_SCHED_DEBUG
543 unsigned int nr_spread_over;
550 struct sched_avg avg;
552 u64 load_last_update_time_copy;
555 raw_spinlock_t lock ____cacheline_aligned;
557 unsigned long load_avg;
558 unsigned long util_avg;
559 unsigned long runnable_avg;
562 #ifdef CONFIG_FAIR_GROUP_SCHED
563 unsigned long tg_load_avg_contrib;
565 long prop_runnable_sum;
568 * h_load = weight * f(tg)
570 * Where f(tg) is the recursive weight fraction assigned to
573 unsigned long h_load;
574 u64 last_h_load_update;
575 struct sched_entity *h_load_next;
576 #endif /* CONFIG_FAIR_GROUP_SCHED */
577 #endif /* CONFIG_SMP */
579 #ifdef CONFIG_FAIR_GROUP_SCHED
580 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
583 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
584 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
585 * (like users, containers etc.)
587 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
588 * This list is used during load balance.
591 struct list_head leaf_cfs_rq_list;
592 struct task_group *tg; /* group that "owns" this runqueue */
594 #ifdef CONFIG_CFS_BANDWIDTH
596 s64 runtime_remaining;
599 u64 throttled_clock_task;
600 u64 throttled_clock_task_time;
603 struct list_head throttled_list;
604 #endif /* CONFIG_CFS_BANDWIDTH */
605 #endif /* CONFIG_FAIR_GROUP_SCHED */
608 static inline int rt_bandwidth_enabled(void)
610 return sysctl_sched_rt_runtime >= 0;
613 /* RT IPI pull logic requires IRQ_WORK */
614 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
615 # define HAVE_RT_PUSH_IPI
618 /* Real-Time classes' related field in a runqueue: */
620 struct rt_prio_array active;
621 unsigned int rt_nr_running;
622 unsigned int rr_nr_running;
623 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
625 int curr; /* highest queued rt task prio */
627 int next; /* next highest */
632 unsigned long rt_nr_migratory;
633 unsigned long rt_nr_total;
635 struct plist_head pushable_tasks;
637 #endif /* CONFIG_SMP */
643 /* Nests inside the rq lock: */
644 raw_spinlock_t rt_runtime_lock;
646 #ifdef CONFIG_RT_GROUP_SCHED
647 unsigned long rt_nr_boosted;
650 struct task_group *tg;
654 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
656 return rt_rq->rt_queued && rt_rq->rt_nr_running;
659 /* Deadline class' related fields in a runqueue */
661 /* runqueue is an rbtree, ordered by deadline */
662 struct rb_root_cached root;
664 unsigned long dl_nr_running;
668 * Deadline values of the currently executing and the
669 * earliest ready task on this rq. Caching these facilitates
670 * the decision whether or not a ready but not running task
671 * should migrate somewhere else.
678 unsigned long dl_nr_migratory;
682 * Tasks on this rq that can be pushed away. They are kept in
683 * an rb-tree, ordered by tasks' deadlines, with caching
684 * of the leftmost (earliest deadline) element.
686 struct rb_root_cached pushable_dl_tasks_root;
691 * "Active utilization" for this runqueue: increased when a
692 * task wakes up (becomes TASK_RUNNING) and decreased when a
698 * Utilization of the tasks "assigned" to this runqueue (including
699 * the tasks that are in runqueue and the tasks that executed on this
700 * CPU and blocked). Increased when a task moves to this runqueue, and
701 * decreased when the task moves away (migrates, changes scheduling
702 * policy, or terminates).
703 * This is needed to compute the "inactive utilization" for the
704 * runqueue (inactive utilization = this_bw - running_bw).
710 * Inverse of the fraction of CPU utilization that can be reclaimed
711 * by the GRUB algorithm.
716 #ifdef CONFIG_FAIR_GROUP_SCHED
717 /* An entity is a task if it doesn't "own" a runqueue */
718 #define entity_is_task(se) (!se->my_q)
720 static inline void se_update_runnable(struct sched_entity *se)
722 if (!entity_is_task(se))
723 se->runnable_weight = se->my_q->h_nr_running;
726 static inline long se_runnable(struct sched_entity *se)
728 if (entity_is_task(se))
731 return se->runnable_weight;
735 #define entity_is_task(se) 1
737 static inline void se_update_runnable(struct sched_entity *se) {}
739 static inline long se_runnable(struct sched_entity *se)
747 * XXX we want to get rid of these helpers and use the full load resolution.
749 static inline long se_weight(struct sched_entity *se)
751 return scale_load_down(se->load.weight);
755 static inline bool sched_asym_prefer(int a, int b)
757 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
761 struct em_perf_domain *em_pd;
762 struct perf_domain *next;
766 /* Scheduling group status flags */
767 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
768 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
771 * We add the notion of a root-domain which will be used to define per-domain
772 * variables. Each exclusive cpuset essentially defines an island domain by
773 * fully partitioning the member CPUs from any other cpuset. Whenever a new
774 * exclusive cpuset is created, we also create and attach a new root-domain
783 cpumask_var_t online;
786 * Indicate pullable load on at least one CPU, e.g:
787 * - More than one runnable task
788 * - Running task is misfit
792 /* Indicate one or more cpus over-utilized (tipping point) */
796 * The bit corresponding to a CPU gets set here if such CPU has more
797 * than one runnable -deadline task (as it is below for RT tasks).
799 cpumask_var_t dlo_mask;
805 * Indicate whether a root_domain's dl_bw has been checked or
806 * updated. It's monotonously increasing value.
808 * Also, some corner cases, like 'wrap around' is dangerous, but given
809 * that u64 is 'big enough'. So that shouldn't be a concern.
813 #ifdef HAVE_RT_PUSH_IPI
815 * For IPI pull requests, loop across the rto_mask.
817 struct irq_work rto_push_work;
818 raw_spinlock_t rto_lock;
819 /* These are only updated and read within rto_lock */
822 /* These atomics are updated outside of a lock */
823 atomic_t rto_loop_next;
824 atomic_t rto_loop_start;
827 * The "RT overload" flag: it gets set if a CPU has more than
828 * one runnable RT task.
830 cpumask_var_t rto_mask;
831 struct cpupri cpupri;
833 unsigned long max_cpu_capacity;
836 * NULL-terminated list of performance domains intersecting with the
837 * CPUs of the rd. Protected by RCU.
839 struct perf_domain __rcu *pd;
842 extern void init_defrootdomain(void);
843 extern int sched_init_domains(const struct cpumask *cpu_map);
844 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
845 extern void sched_get_rd(struct root_domain *rd);
846 extern void sched_put_rd(struct root_domain *rd);
848 #ifdef HAVE_RT_PUSH_IPI
849 extern void rto_push_irq_work_func(struct irq_work *work);
851 #endif /* CONFIG_SMP */
853 #ifdef CONFIG_UCLAMP_TASK
855 * struct uclamp_bucket - Utilization clamp bucket
856 * @value: utilization clamp value for tasks on this clamp bucket
857 * @tasks: number of RUNNABLE tasks on this clamp bucket
859 * Keep track of how many tasks are RUNNABLE for a given utilization
862 struct uclamp_bucket {
863 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
864 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
868 * struct uclamp_rq - rq's utilization clamp
869 * @value: currently active clamp values for a rq
870 * @bucket: utilization clamp buckets affecting a rq
872 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
873 * A clamp value is affecting a rq when there is at least one task RUNNABLE
874 * (or actually running) with that value.
876 * There are up to UCLAMP_CNT possible different clamp values, currently there
877 * are only two: minimum utilization and maximum utilization.
879 * All utilization clamping values are MAX aggregated, since:
880 * - for util_min: we want to run the CPU at least at the max of the minimum
881 * utilization required by its currently RUNNABLE tasks.
882 * - for util_max: we want to allow the CPU to run up to the max of the
883 * maximum utilization allowed by its currently RUNNABLE tasks.
885 * Since on each system we expect only a limited number of different
886 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
887 * the metrics required to compute all the per-rq utilization clamp values.
891 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
894 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
895 #endif /* CONFIG_UCLAMP_TASK */
898 * This is the main, per-CPU runqueue data structure.
900 * Locking rule: those places that want to lock multiple runqueues
901 * (such as the load balancing or the thread migration code), lock
902 * acquire operations must be ordered by ascending &runqueue.
909 * nr_running and cpu_load should be in the same cacheline because
910 * remote CPUs use both these fields when doing load calculation.
912 unsigned int nr_running;
913 #ifdef CONFIG_NUMA_BALANCING
914 unsigned int nr_numa_running;
915 unsigned int nr_preferred_running;
916 unsigned int numa_migrate_on;
918 #ifdef CONFIG_NO_HZ_COMMON
920 unsigned long last_blocked_load_update_tick;
921 unsigned int has_blocked_load;
922 call_single_data_t nohz_csd;
923 #endif /* CONFIG_SMP */
924 unsigned int nohz_tick_stopped;
926 #endif /* CONFIG_NO_HZ_COMMON */
929 unsigned int ttwu_pending;
933 #ifdef CONFIG_UCLAMP_TASK
934 /* Utilization clamp values based on CPU's RUNNABLE tasks */
935 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
936 unsigned int uclamp_flags;
937 #define UCLAMP_FLAG_IDLE 0x01
944 #ifdef CONFIG_FAIR_GROUP_SCHED
945 /* list of leaf cfs_rq on this CPU: */
946 struct list_head leaf_cfs_rq_list;
947 struct list_head *tmp_alone_branch;
948 #endif /* CONFIG_FAIR_GROUP_SCHED */
951 * This is part of a global counter where only the total sum
952 * over all CPUs matters. A task can increase this counter on
953 * one CPU and if it got migrated afterwards it may decrease
954 * it on another CPU. Always updated under the runqueue lock:
956 unsigned long nr_uninterruptible;
958 struct task_struct __rcu *curr;
959 struct task_struct *idle;
960 struct task_struct *stop;
961 unsigned long next_balance;
962 struct mm_struct *prev_mm;
964 unsigned int clock_update_flags;
966 /* Ensure that all clocks are in the same cache line */
967 u64 clock_task ____cacheline_aligned;
969 unsigned long lost_idle_time;
973 #ifdef CONFIG_MEMBARRIER
974 int membarrier_state;
978 struct root_domain *rd;
979 struct sched_domain __rcu *sd;
981 unsigned long cpu_capacity;
982 unsigned long cpu_capacity_orig;
984 struct callback_head *balance_callback;
985 unsigned char balance_push;
987 unsigned char nohz_idle_balance;
988 unsigned char idle_balance;
990 unsigned long misfit_task_load;
992 /* For active balancing */
995 struct cpu_stop_work active_balance_work;
997 /* CPU of this runqueue: */
1001 struct list_head cfs_tasks;
1003 struct sched_avg avg_rt;
1004 struct sched_avg avg_dl;
1005 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1006 struct sched_avg avg_irq;
1008 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1009 struct sched_avg avg_thermal;
1014 /* This is used to determine avg_idle's max value */
1015 u64 max_idle_balance_cost;
1017 #ifdef CONFIG_HOTPLUG_CPU
1018 struct rcuwait hotplug_wait;
1020 #endif /* CONFIG_SMP */
1022 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1025 #ifdef CONFIG_PARAVIRT
1026 u64 prev_steal_time;
1028 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1029 u64 prev_steal_time_rq;
1032 /* calc_load related fields */
1033 unsigned long calc_load_update;
1034 long calc_load_active;
1036 #ifdef CONFIG_SCHED_HRTICK
1038 call_single_data_t hrtick_csd;
1040 struct hrtimer hrtick_timer;
1041 ktime_t hrtick_time;
1044 #ifdef CONFIG_SCHEDSTATS
1046 struct sched_info rq_sched_info;
1047 unsigned long long rq_cpu_time;
1048 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1050 /* sys_sched_yield() stats */
1051 unsigned int yld_count;
1053 /* schedule() stats */
1054 unsigned int sched_count;
1055 unsigned int sched_goidle;
1057 /* try_to_wake_up() stats */
1058 unsigned int ttwu_count;
1059 unsigned int ttwu_local;
1062 #ifdef CONFIG_CPU_IDLE
1063 /* Must be inspected within a rcu lock section */
1064 struct cpuidle_state *idle_state;
1068 unsigned int nr_pinned;
1070 unsigned int push_busy;
1071 struct cpu_stop_work push_work;
1074 #ifdef CONFIG_FAIR_GROUP_SCHED
1076 /* CPU runqueue to which this cfs_rq is attached */
1077 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1084 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1086 return container_of(cfs_rq, struct rq, cfs);
1090 static inline int cpu_of(struct rq *rq)
1099 #define MDF_PUSH 0x01
1101 static inline bool is_migration_disabled(struct task_struct *p)
1104 return p->migration_disabled;
1110 #ifdef CONFIG_SCHED_SMT
1111 extern void __update_idle_core(struct rq *rq);
1113 static inline void update_idle_core(struct rq *rq)
1115 if (static_branch_unlikely(&sched_smt_present))
1116 __update_idle_core(rq);
1120 static inline void update_idle_core(struct rq *rq) { }
1123 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1125 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1126 #define this_rq() this_cpu_ptr(&runqueues)
1127 #define task_rq(p) cpu_rq(task_cpu(p))
1128 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1129 #define raw_rq() raw_cpu_ptr(&runqueues)
1131 extern void update_rq_clock(struct rq *rq);
1133 static inline u64 __rq_clock_broken(struct rq *rq)
1135 return READ_ONCE(rq->clock);
1139 * rq::clock_update_flags bits
1141 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1142 * call to __schedule(). This is an optimisation to avoid
1143 * neighbouring rq clock updates.
1145 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1146 * in effect and calls to update_rq_clock() are being ignored.
1148 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1149 * made to update_rq_clock() since the last time rq::lock was pinned.
1151 * If inside of __schedule(), clock_update_flags will have been
1152 * shifted left (a left shift is a cheap operation for the fast path
1153 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1155 * if (rq-clock_update_flags >= RQCF_UPDATED)
1157 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1158 * one position though, because the next rq_unpin_lock() will shift it
1161 #define RQCF_REQ_SKIP 0x01
1162 #define RQCF_ACT_SKIP 0x02
1163 #define RQCF_UPDATED 0x04
1165 static inline void assert_clock_updated(struct rq *rq)
1168 * The only reason for not seeing a clock update since the
1169 * last rq_pin_lock() is if we're currently skipping updates.
1171 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1174 static inline u64 rq_clock(struct rq *rq)
1176 lockdep_assert_held(&rq->lock);
1177 assert_clock_updated(rq);
1182 static inline u64 rq_clock_task(struct rq *rq)
1184 lockdep_assert_held(&rq->lock);
1185 assert_clock_updated(rq);
1187 return rq->clock_task;
1191 * By default the decay is the default pelt decay period.
1192 * The decay shift can change the decay period in
1194 * Decay shift Decay period(ms)
1201 extern int sched_thermal_decay_shift;
1203 static inline u64 rq_clock_thermal(struct rq *rq)
1205 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1208 static inline void rq_clock_skip_update(struct rq *rq)
1210 lockdep_assert_held(&rq->lock);
1211 rq->clock_update_flags |= RQCF_REQ_SKIP;
1215 * See rt task throttling, which is the only time a skip
1216 * request is cancelled.
1218 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1220 lockdep_assert_held(&rq->lock);
1221 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1225 unsigned long flags;
1226 struct pin_cookie cookie;
1227 #ifdef CONFIG_SCHED_DEBUG
1229 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1230 * current pin context is stashed here in case it needs to be
1231 * restored in rq_repin_lock().
1233 unsigned int clock_update_flags;
1237 extern struct callback_head balance_push_callback;
1240 * Lockdep annotation that avoids accidental unlocks; it's like a
1241 * sticky/continuous lockdep_assert_held().
1243 * This avoids code that has access to 'struct rq *rq' (basically everything in
1244 * the scheduler) from accidentally unlocking the rq if they do not also have a
1245 * copy of the (on-stack) 'struct rq_flags rf'.
1247 * Also see Documentation/locking/lockdep-design.rst.
1249 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1251 rf->cookie = lockdep_pin_lock(&rq->lock);
1253 #ifdef CONFIG_SCHED_DEBUG
1254 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1255 rf->clock_update_flags = 0;
1257 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1262 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1264 #ifdef CONFIG_SCHED_DEBUG
1265 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1266 rf->clock_update_flags = RQCF_UPDATED;
1269 lockdep_unpin_lock(&rq->lock, rf->cookie);
1272 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1274 lockdep_repin_lock(&rq->lock, rf->cookie);
1276 #ifdef CONFIG_SCHED_DEBUG
1278 * Restore the value we stashed in @rf for this pin context.
1280 rq->clock_update_flags |= rf->clock_update_flags;
1284 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1285 __acquires(rq->lock);
1287 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1288 __acquires(p->pi_lock)
1289 __acquires(rq->lock);
1291 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1292 __releases(rq->lock)
1294 rq_unpin_lock(rq, rf);
1295 raw_spin_unlock(&rq->lock);
1299 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1300 __releases(rq->lock)
1301 __releases(p->pi_lock)
1303 rq_unpin_lock(rq, rf);
1304 raw_spin_unlock(&rq->lock);
1305 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1309 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1310 __acquires(rq->lock)
1312 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1313 rq_pin_lock(rq, rf);
1317 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1318 __acquires(rq->lock)
1320 raw_spin_lock_irq(&rq->lock);
1321 rq_pin_lock(rq, rf);
1325 rq_lock(struct rq *rq, struct rq_flags *rf)
1326 __acquires(rq->lock)
1328 raw_spin_lock(&rq->lock);
1329 rq_pin_lock(rq, rf);
1333 rq_relock(struct rq *rq, struct rq_flags *rf)
1334 __acquires(rq->lock)
1336 raw_spin_lock(&rq->lock);
1337 rq_repin_lock(rq, rf);
1341 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1342 __releases(rq->lock)
1344 rq_unpin_lock(rq, rf);
1345 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1349 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1350 __releases(rq->lock)
1352 rq_unpin_lock(rq, rf);
1353 raw_spin_unlock_irq(&rq->lock);
1357 rq_unlock(struct rq *rq, struct rq_flags *rf)
1358 __releases(rq->lock)
1360 rq_unpin_lock(rq, rf);
1361 raw_spin_unlock(&rq->lock);
1364 static inline struct rq *
1365 this_rq_lock_irq(struct rq_flags *rf)
1366 __acquires(rq->lock)
1370 local_irq_disable();
1377 enum numa_topology_type {
1382 extern enum numa_topology_type sched_numa_topology_type;
1383 extern int sched_max_numa_distance;
1384 extern bool find_numa_distance(int distance);
1385 extern void sched_init_numa(void);
1386 extern void sched_domains_numa_masks_set(unsigned int cpu);
1387 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1388 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1390 static inline void sched_init_numa(void) { }
1391 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1392 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1393 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1399 #ifdef CONFIG_NUMA_BALANCING
1400 /* The regions in numa_faults array from task_struct */
1401 enum numa_faults_stats {
1407 extern void sched_setnuma(struct task_struct *p, int node);
1408 extern int migrate_task_to(struct task_struct *p, int cpu);
1409 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1411 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1414 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1417 #endif /* CONFIG_NUMA_BALANCING */
1422 queue_balance_callback(struct rq *rq,
1423 struct callback_head *head,
1424 void (*func)(struct rq *rq))
1426 lockdep_assert_held(&rq->lock);
1428 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1431 head->func = (void (*)(struct callback_head *))func;
1432 head->next = rq->balance_callback;
1433 rq->balance_callback = head;
1436 #define rcu_dereference_check_sched_domain(p) \
1437 rcu_dereference_check((p), \
1438 lockdep_is_held(&sched_domains_mutex))
1441 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1442 * See destroy_sched_domains: call_rcu for details.
1444 * The domain tree of any CPU may only be accessed from within
1445 * preempt-disabled sections.
1447 #define for_each_domain(cpu, __sd) \
1448 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1449 __sd; __sd = __sd->parent)
1452 * highest_flag_domain - Return highest sched_domain containing flag.
1453 * @cpu: The CPU whose highest level of sched domain is to
1455 * @flag: The flag to check for the highest sched_domain
1456 * for the given CPU.
1458 * Returns the highest sched_domain of a CPU which contains the given flag.
1460 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1462 struct sched_domain *sd, *hsd = NULL;
1464 for_each_domain(cpu, sd) {
1465 if (!(sd->flags & flag))
1473 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1475 struct sched_domain *sd;
1477 for_each_domain(cpu, sd) {
1478 if (sd->flags & flag)
1485 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1486 DECLARE_PER_CPU(int, sd_llc_size);
1487 DECLARE_PER_CPU(int, sd_llc_id);
1488 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1489 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1490 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1491 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1492 extern struct static_key_false sched_asym_cpucapacity;
1494 struct sched_group_capacity {
1497 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1500 unsigned long capacity;
1501 unsigned long min_capacity; /* Min per-CPU capacity in group */
1502 unsigned long max_capacity; /* Max per-CPU capacity in group */
1503 unsigned long next_update;
1504 int imbalance; /* XXX unrelated to capacity but shared group state */
1506 #ifdef CONFIG_SCHED_DEBUG
1510 unsigned long cpumask[]; /* Balance mask */
1513 struct sched_group {
1514 struct sched_group *next; /* Must be a circular list */
1517 unsigned int group_weight;
1518 struct sched_group_capacity *sgc;
1519 int asym_prefer_cpu; /* CPU of highest priority in group */
1522 * The CPUs this group covers.
1524 * NOTE: this field is variable length. (Allocated dynamically
1525 * by attaching extra space to the end of the structure,
1526 * depending on how many CPUs the kernel has booted up with)
1528 unsigned long cpumask[];
1531 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1533 return to_cpumask(sg->cpumask);
1537 * See build_balance_mask().
1539 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1541 return to_cpumask(sg->sgc->cpumask);
1545 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1546 * @group: The group whose first CPU is to be returned.
1548 static inline unsigned int group_first_cpu(struct sched_group *group)
1550 return cpumask_first(sched_group_span(group));
1553 extern int group_balance_cpu(struct sched_group *sg);
1555 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1556 void register_sched_domain_sysctl(void);
1557 void dirty_sched_domain_sysctl(int cpu);
1558 void unregister_sched_domain_sysctl(void);
1560 static inline void register_sched_domain_sysctl(void)
1563 static inline void dirty_sched_domain_sysctl(int cpu)
1566 static inline void unregister_sched_domain_sysctl(void)
1571 extern void flush_smp_call_function_from_idle(void);
1573 #else /* !CONFIG_SMP: */
1574 static inline void flush_smp_call_function_from_idle(void) { }
1578 #include "autogroup.h"
1580 #ifdef CONFIG_CGROUP_SCHED
1583 * Return the group to which this tasks belongs.
1585 * We cannot use task_css() and friends because the cgroup subsystem
1586 * changes that value before the cgroup_subsys::attach() method is called,
1587 * therefore we cannot pin it and might observe the wrong value.
1589 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1590 * core changes this before calling sched_move_task().
1592 * Instead we use a 'copy' which is updated from sched_move_task() while
1593 * holding both task_struct::pi_lock and rq::lock.
1595 static inline struct task_group *task_group(struct task_struct *p)
1597 return p->sched_task_group;
1600 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1601 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1603 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1604 struct task_group *tg = task_group(p);
1607 #ifdef CONFIG_FAIR_GROUP_SCHED
1608 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1609 p->se.cfs_rq = tg->cfs_rq[cpu];
1610 p->se.parent = tg->se[cpu];
1613 #ifdef CONFIG_RT_GROUP_SCHED
1614 p->rt.rt_rq = tg->rt_rq[cpu];
1615 p->rt.parent = tg->rt_se[cpu];
1619 #else /* CONFIG_CGROUP_SCHED */
1621 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1622 static inline struct task_group *task_group(struct task_struct *p)
1627 #endif /* CONFIG_CGROUP_SCHED */
1629 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1631 set_task_rq(p, cpu);
1634 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1635 * successfully executed on another CPU. We must ensure that updates of
1636 * per-task data have been completed by this moment.
1639 #ifdef CONFIG_THREAD_INFO_IN_TASK
1640 WRITE_ONCE(p->cpu, cpu);
1642 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1649 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1651 #ifdef CONFIG_SCHED_DEBUG
1652 # include <linux/static_key.h>
1653 # define const_debug __read_mostly
1655 # define const_debug const
1658 #define SCHED_FEAT(name, enabled) \
1659 __SCHED_FEAT_##name ,
1662 #include "features.h"
1668 #ifdef CONFIG_SCHED_DEBUG
1671 * To support run-time toggling of sched features, all the translation units
1672 * (but core.c) reference the sysctl_sched_features defined in core.c.
1674 extern const_debug unsigned int sysctl_sched_features;
1676 #ifdef CONFIG_JUMP_LABEL
1677 #define SCHED_FEAT(name, enabled) \
1678 static __always_inline bool static_branch_##name(struct static_key *key) \
1680 return static_key_##enabled(key); \
1683 #include "features.h"
1686 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1687 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1689 #else /* !CONFIG_JUMP_LABEL */
1691 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1693 #endif /* CONFIG_JUMP_LABEL */
1695 #else /* !SCHED_DEBUG */
1698 * Each translation unit has its own copy of sysctl_sched_features to allow
1699 * constants propagation at compile time and compiler optimization based on
1702 #define SCHED_FEAT(name, enabled) \
1703 (1UL << __SCHED_FEAT_##name) * enabled |
1704 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1705 #include "features.h"
1709 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1711 #endif /* SCHED_DEBUG */
1713 extern struct static_key_false sched_numa_balancing;
1714 extern struct static_key_false sched_schedstats;
1716 static inline u64 global_rt_period(void)
1718 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1721 static inline u64 global_rt_runtime(void)
1723 if (sysctl_sched_rt_runtime < 0)
1726 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1729 static inline int task_current(struct rq *rq, struct task_struct *p)
1731 return rq->curr == p;
1734 static inline int task_running(struct rq *rq, struct task_struct *p)
1739 return task_current(rq, p);
1743 static inline int task_on_rq_queued(struct task_struct *p)
1745 return p->on_rq == TASK_ON_RQ_QUEUED;
1748 static inline int task_on_rq_migrating(struct task_struct *p)
1750 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1753 /* Wake flags. The first three directly map to some SD flag value */
1754 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
1755 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
1756 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
1758 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
1759 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
1760 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
1763 static_assert(WF_EXEC == SD_BALANCE_EXEC);
1764 static_assert(WF_FORK == SD_BALANCE_FORK);
1765 static_assert(WF_TTWU == SD_BALANCE_WAKE);
1769 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1770 * of tasks with abnormal "nice" values across CPUs the contribution that
1771 * each task makes to its run queue's load is weighted according to its
1772 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1773 * scaled version of the new time slice allocation that they receive on time
1777 #define WEIGHT_IDLEPRIO 3
1778 #define WMULT_IDLEPRIO 1431655765
1780 extern const int sched_prio_to_weight[40];
1781 extern const u32 sched_prio_to_wmult[40];
1784 * {de,en}queue flags:
1786 * DEQUEUE_SLEEP - task is no longer runnable
1787 * ENQUEUE_WAKEUP - task just became runnable
1789 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1790 * are in a known state which allows modification. Such pairs
1791 * should preserve as much state as possible.
1793 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1796 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1797 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1798 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1802 #define DEQUEUE_SLEEP 0x01
1803 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1804 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1805 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1807 #define ENQUEUE_WAKEUP 0x01
1808 #define ENQUEUE_RESTORE 0x02
1809 #define ENQUEUE_MOVE 0x04
1810 #define ENQUEUE_NOCLOCK 0x08
1812 #define ENQUEUE_HEAD 0x10
1813 #define ENQUEUE_REPLENISH 0x20
1815 #define ENQUEUE_MIGRATED 0x40
1817 #define ENQUEUE_MIGRATED 0x00
1820 #define RETRY_TASK ((void *)-1UL)
1822 struct sched_class {
1824 #ifdef CONFIG_UCLAMP_TASK
1828 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1829 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1830 void (*yield_task) (struct rq *rq);
1831 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1833 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1835 struct task_struct *(*pick_next_task)(struct rq *rq);
1837 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1838 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1841 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1842 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
1843 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1845 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1847 void (*set_cpus_allowed)(struct task_struct *p,
1848 const struct cpumask *newmask,
1851 void (*rq_online)(struct rq *rq);
1852 void (*rq_offline)(struct rq *rq);
1854 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
1857 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1858 void (*task_fork)(struct task_struct *p);
1859 void (*task_dead)(struct task_struct *p);
1862 * The switched_from() call is allowed to drop rq->lock, therefore we
1863 * cannot assume the switched_from/switched_to pair is serliazed by
1864 * rq->lock. They are however serialized by p->pi_lock.
1866 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1867 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1868 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1871 unsigned int (*get_rr_interval)(struct rq *rq,
1872 struct task_struct *task);
1874 void (*update_curr)(struct rq *rq);
1876 #define TASK_SET_GROUP 0
1877 #define TASK_MOVE_GROUP 1
1879 #ifdef CONFIG_FAIR_GROUP_SCHED
1880 void (*task_change_group)(struct task_struct *p, int type);
1884 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1886 WARN_ON_ONCE(rq->curr != prev);
1887 prev->sched_class->put_prev_task(rq, prev);
1890 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1892 WARN_ON_ONCE(rq->curr != next);
1893 next->sched_class->set_next_task(rq, next, false);
1898 * Helper to define a sched_class instance; each one is placed in a separate
1899 * section which is ordered by the linker script:
1901 * include/asm-generic/vmlinux.lds.h
1903 * Also enforce alignment on the instance, not the type, to guarantee layout.
1905 #define DEFINE_SCHED_CLASS(name) \
1906 const struct sched_class name##_sched_class \
1907 __aligned(__alignof__(struct sched_class)) \
1908 __section("__" #name "_sched_class")
1910 /* Defined in include/asm-generic/vmlinux.lds.h */
1911 extern struct sched_class __begin_sched_classes[];
1912 extern struct sched_class __end_sched_classes[];
1914 #define sched_class_highest (__end_sched_classes - 1)
1915 #define sched_class_lowest (__begin_sched_classes - 1)
1917 #define for_class_range(class, _from, _to) \
1918 for (class = (_from); class != (_to); class--)
1920 #define for_each_class(class) \
1921 for_class_range(class, sched_class_highest, sched_class_lowest)
1923 extern const struct sched_class stop_sched_class;
1924 extern const struct sched_class dl_sched_class;
1925 extern const struct sched_class rt_sched_class;
1926 extern const struct sched_class fair_sched_class;
1927 extern const struct sched_class idle_sched_class;
1929 static inline bool sched_stop_runnable(struct rq *rq)
1931 return rq->stop && task_on_rq_queued(rq->stop);
1934 static inline bool sched_dl_runnable(struct rq *rq)
1936 return rq->dl.dl_nr_running > 0;
1939 static inline bool sched_rt_runnable(struct rq *rq)
1941 return rq->rt.rt_queued > 0;
1944 static inline bool sched_fair_runnable(struct rq *rq)
1946 return rq->cfs.nr_running > 0;
1949 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1950 extern struct task_struct *pick_next_task_idle(struct rq *rq);
1952 #define SCA_CHECK 0x01
1953 #define SCA_MIGRATE_DISABLE 0x02
1954 #define SCA_MIGRATE_ENABLE 0x04
1958 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1960 extern void trigger_load_balance(struct rq *rq);
1962 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
1964 static inline struct task_struct *get_push_task(struct rq *rq)
1966 struct task_struct *p = rq->curr;
1968 lockdep_assert_held(&rq->lock);
1973 if (p->nr_cpus_allowed == 1)
1976 rq->push_busy = true;
1977 return get_task_struct(p);
1980 extern int push_cpu_stop(void *arg);
1984 #ifdef CONFIG_CPU_IDLE
1985 static inline void idle_set_state(struct rq *rq,
1986 struct cpuidle_state *idle_state)
1988 rq->idle_state = idle_state;
1991 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1993 SCHED_WARN_ON(!rcu_read_lock_held());
1995 return rq->idle_state;
1998 static inline void idle_set_state(struct rq *rq,
1999 struct cpuidle_state *idle_state)
2003 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2009 extern void schedule_idle(void);
2011 extern void sysrq_sched_debug_show(void);
2012 extern void sched_init_granularity(void);
2013 extern void update_max_interval(void);
2015 extern void init_sched_dl_class(void);
2016 extern void init_sched_rt_class(void);
2017 extern void init_sched_fair_class(void);
2019 extern void reweight_task(struct task_struct *p, int prio);
2021 extern void resched_curr(struct rq *rq);
2022 extern void resched_cpu(int cpu);
2024 extern struct rt_bandwidth def_rt_bandwidth;
2025 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2027 extern struct dl_bandwidth def_dl_bandwidth;
2028 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2029 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2030 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2033 #define BW_UNIT (1 << BW_SHIFT)
2034 #define RATIO_SHIFT 8
2035 #define MAX_BW_BITS (64 - BW_SHIFT)
2036 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2037 unsigned long to_ratio(u64 period, u64 runtime);
2039 extern void init_entity_runnable_average(struct sched_entity *se);
2040 extern void post_init_entity_util_avg(struct task_struct *p);
2042 #ifdef CONFIG_NO_HZ_FULL
2043 extern bool sched_can_stop_tick(struct rq *rq);
2044 extern int __init sched_tick_offload_init(void);
2047 * Tick may be needed by tasks in the runqueue depending on their policy and
2048 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2049 * nohz mode if necessary.
2051 static inline void sched_update_tick_dependency(struct rq *rq)
2053 int cpu = cpu_of(rq);
2055 if (!tick_nohz_full_cpu(cpu))
2058 if (sched_can_stop_tick(rq))
2059 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2061 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2064 static inline int sched_tick_offload_init(void) { return 0; }
2065 static inline void sched_update_tick_dependency(struct rq *rq) { }
2068 static inline void add_nr_running(struct rq *rq, unsigned count)
2070 unsigned prev_nr = rq->nr_running;
2072 rq->nr_running = prev_nr + count;
2073 if (trace_sched_update_nr_running_tp_enabled()) {
2074 call_trace_sched_update_nr_running(rq, count);
2078 if (prev_nr < 2 && rq->nr_running >= 2) {
2079 if (!READ_ONCE(rq->rd->overload))
2080 WRITE_ONCE(rq->rd->overload, 1);
2084 sched_update_tick_dependency(rq);
2087 static inline void sub_nr_running(struct rq *rq, unsigned count)
2089 rq->nr_running -= count;
2090 if (trace_sched_update_nr_running_tp_enabled()) {
2091 call_trace_sched_update_nr_running(rq, -count);
2094 /* Check if we still need preemption */
2095 sched_update_tick_dependency(rq);
2098 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2099 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2101 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2103 extern const_debug unsigned int sysctl_sched_nr_migrate;
2104 extern const_debug unsigned int sysctl_sched_migration_cost;
2106 #ifdef CONFIG_SCHED_HRTICK
2110 * - enabled by features
2111 * - hrtimer is actually high res
2113 static inline int hrtick_enabled(struct rq *rq)
2115 if (!cpu_active(cpu_of(rq)))
2117 return hrtimer_is_hres_active(&rq->hrtick_timer);
2120 static inline int hrtick_enabled_fair(struct rq *rq)
2122 if (!sched_feat(HRTICK))
2124 return hrtick_enabled(rq);
2127 static inline int hrtick_enabled_dl(struct rq *rq)
2129 if (!sched_feat(HRTICK_DL))
2131 return hrtick_enabled(rq);
2134 void hrtick_start(struct rq *rq, u64 delay);
2138 static inline int hrtick_enabled_fair(struct rq *rq)
2143 static inline int hrtick_enabled_dl(struct rq *rq)
2148 static inline int hrtick_enabled(struct rq *rq)
2153 #endif /* CONFIG_SCHED_HRTICK */
2155 #ifndef arch_scale_freq_tick
2156 static __always_inline
2157 void arch_scale_freq_tick(void)
2162 #ifndef arch_scale_freq_capacity
2164 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2165 * @cpu: the CPU in question.
2167 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2170 * ------ * SCHED_CAPACITY_SCALE
2173 static __always_inline
2174 unsigned long arch_scale_freq_capacity(int cpu)
2176 return SCHED_CAPACITY_SCALE;
2181 #ifdef CONFIG_PREEMPTION
2183 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2186 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2187 * way at the expense of forcing extra atomic operations in all
2188 * invocations. This assures that the double_lock is acquired using the
2189 * same underlying policy as the spinlock_t on this architecture, which
2190 * reduces latency compared to the unfair variant below. However, it
2191 * also adds more overhead and therefore may reduce throughput.
2193 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2194 __releases(this_rq->lock)
2195 __acquires(busiest->lock)
2196 __acquires(this_rq->lock)
2198 raw_spin_unlock(&this_rq->lock);
2199 double_rq_lock(this_rq, busiest);
2206 * Unfair double_lock_balance: Optimizes throughput at the expense of
2207 * latency by eliminating extra atomic operations when the locks are
2208 * already in proper order on entry. This favors lower CPU-ids and will
2209 * grant the double lock to lower CPUs over higher ids under contention,
2210 * regardless of entry order into the function.
2212 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2213 __releases(this_rq->lock)
2214 __acquires(busiest->lock)
2215 __acquires(this_rq->lock)
2219 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2220 if (busiest < this_rq) {
2221 raw_spin_unlock(&this_rq->lock);
2222 raw_spin_lock(&busiest->lock);
2223 raw_spin_lock_nested(&this_rq->lock,
2224 SINGLE_DEPTH_NESTING);
2227 raw_spin_lock_nested(&busiest->lock,
2228 SINGLE_DEPTH_NESTING);
2233 #endif /* CONFIG_PREEMPTION */
2236 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2238 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2240 if (unlikely(!irqs_disabled())) {
2241 /* printk() doesn't work well under rq->lock */
2242 raw_spin_unlock(&this_rq->lock);
2246 return _double_lock_balance(this_rq, busiest);
2249 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2250 __releases(busiest->lock)
2252 raw_spin_unlock(&busiest->lock);
2253 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2256 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2262 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2265 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2271 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2274 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2280 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2284 * double_rq_lock - safely lock two runqueues
2286 * Note this does not disable interrupts like task_rq_lock,
2287 * you need to do so manually before calling.
2289 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2290 __acquires(rq1->lock)
2291 __acquires(rq2->lock)
2293 BUG_ON(!irqs_disabled());
2295 raw_spin_lock(&rq1->lock);
2296 __acquire(rq2->lock); /* Fake it out ;) */
2299 raw_spin_lock(&rq1->lock);
2300 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2302 raw_spin_lock(&rq2->lock);
2303 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2309 * double_rq_unlock - safely unlock two runqueues
2311 * Note this does not restore interrupts like task_rq_unlock,
2312 * you need to do so manually after calling.
2314 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2315 __releases(rq1->lock)
2316 __releases(rq2->lock)
2318 raw_spin_unlock(&rq1->lock);
2320 raw_spin_unlock(&rq2->lock);
2322 __release(rq2->lock);
2325 extern void set_rq_online (struct rq *rq);
2326 extern void set_rq_offline(struct rq *rq);
2327 extern bool sched_smp_initialized;
2329 #else /* CONFIG_SMP */
2332 * double_rq_lock - safely lock two runqueues
2334 * Note this does not disable interrupts like task_rq_lock,
2335 * you need to do so manually before calling.
2337 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2338 __acquires(rq1->lock)
2339 __acquires(rq2->lock)
2341 BUG_ON(!irqs_disabled());
2343 raw_spin_lock(&rq1->lock);
2344 __acquire(rq2->lock); /* Fake it out ;) */
2348 * double_rq_unlock - safely unlock two runqueues
2350 * Note this does not restore interrupts like task_rq_unlock,
2351 * you need to do so manually after calling.
2353 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2354 __releases(rq1->lock)
2355 __releases(rq2->lock)
2358 raw_spin_unlock(&rq1->lock);
2359 __release(rq2->lock);
2364 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2365 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2367 #ifdef CONFIG_SCHED_DEBUG
2368 extern bool sched_debug_enabled;
2370 extern void print_cfs_stats(struct seq_file *m, int cpu);
2371 extern void print_rt_stats(struct seq_file *m, int cpu);
2372 extern void print_dl_stats(struct seq_file *m, int cpu);
2373 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2374 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2375 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2376 #ifdef CONFIG_NUMA_BALANCING
2378 show_numa_stats(struct task_struct *p, struct seq_file *m);
2380 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2381 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2382 #endif /* CONFIG_NUMA_BALANCING */
2383 #endif /* CONFIG_SCHED_DEBUG */
2385 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2386 extern void init_rt_rq(struct rt_rq *rt_rq);
2387 extern void init_dl_rq(struct dl_rq *dl_rq);
2389 extern void cfs_bandwidth_usage_inc(void);
2390 extern void cfs_bandwidth_usage_dec(void);
2392 #ifdef CONFIG_NO_HZ_COMMON
2393 #define NOHZ_BALANCE_KICK_BIT 0
2394 #define NOHZ_STATS_KICK_BIT 1
2395 #define NOHZ_NEWILB_KICK_BIT 2
2397 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2398 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2399 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2401 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2403 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2405 extern void nohz_balance_exit_idle(struct rq *rq);
2407 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2410 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2411 extern void nohz_run_idle_balance(int cpu);
2413 static inline void nohz_run_idle_balance(int cpu) { }
2418 void __dl_update(struct dl_bw *dl_b, s64 bw)
2420 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2423 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2424 "sched RCU must be held");
2425 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2426 struct rq *rq = cpu_rq(i);
2428 rq->dl.extra_bw += bw;
2433 void __dl_update(struct dl_bw *dl_b, s64 bw)
2435 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2442 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2447 struct u64_stats_sync sync;
2450 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2453 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2454 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2455 * and never move forward.
2457 static inline u64 irq_time_read(int cpu)
2459 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2464 seq = __u64_stats_fetch_begin(&irqtime->sync);
2465 total = irqtime->total;
2466 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2470 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2472 #ifdef CONFIG_CPU_FREQ
2473 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2476 * cpufreq_update_util - Take a note about CPU utilization changes.
2477 * @rq: Runqueue to carry out the update for.
2478 * @flags: Update reason flags.
2480 * This function is called by the scheduler on the CPU whose utilization is
2483 * It can only be called from RCU-sched read-side critical sections.
2485 * The way cpufreq is currently arranged requires it to evaluate the CPU
2486 * performance state (frequency/voltage) on a regular basis to prevent it from
2487 * being stuck in a completely inadequate performance level for too long.
2488 * That is not guaranteed to happen if the updates are only triggered from CFS
2489 * and DL, though, because they may not be coming in if only RT tasks are
2490 * active all the time (or there are RT tasks only).
2492 * As a workaround for that issue, this function is called periodically by the
2493 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2494 * but that really is a band-aid. Going forward it should be replaced with
2495 * solutions targeted more specifically at RT tasks.
2497 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2499 struct update_util_data *data;
2501 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2504 data->func(data, rq_clock(rq), flags);
2507 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2508 #endif /* CONFIG_CPU_FREQ */
2510 #ifdef CONFIG_UCLAMP_TASK
2511 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2514 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2515 * @rq: The rq to clamp against. Must not be NULL.
2516 * @util: The util value to clamp.
2517 * @p: The task to clamp against. Can be NULL if you want to clamp
2520 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2522 * If sched_uclamp_used static key is disabled, then just return the util
2523 * without any clamping since uclamp aggregation at the rq level in the fast
2524 * path is disabled, rendering this operation a NOP.
2526 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2527 * will return the correct effective uclamp value of the task even if the
2528 * static key is disabled.
2530 static __always_inline
2531 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2532 struct task_struct *p)
2534 unsigned long min_util;
2535 unsigned long max_util;
2537 if (!static_branch_likely(&sched_uclamp_used))
2540 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2541 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2544 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2545 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2549 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2550 * RUNNABLE tasks with _different_ clamps, we can end up with an
2551 * inversion. Fix it now when the clamps are applied.
2553 if (unlikely(min_util >= max_util))
2556 return clamp(util, min_util, max_util);
2560 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2561 * by default in the fast path and only gets turned on once userspace performs
2562 * an operation that requires it.
2564 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2567 static inline bool uclamp_is_used(void)
2569 return static_branch_likely(&sched_uclamp_used);
2571 #else /* CONFIG_UCLAMP_TASK */
2573 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2574 struct task_struct *p)
2579 static inline bool uclamp_is_used(void)
2583 #endif /* CONFIG_UCLAMP_TASK */
2585 #ifdef arch_scale_freq_capacity
2586 # ifndef arch_scale_freq_invariant
2587 # define arch_scale_freq_invariant() true
2590 # define arch_scale_freq_invariant() false
2594 static inline unsigned long capacity_orig_of(int cpu)
2596 return cpu_rq(cpu)->cpu_capacity_orig;
2600 * enum cpu_util_type - CPU utilization type
2601 * @FREQUENCY_UTIL: Utilization used to select frequency
2602 * @ENERGY_UTIL: Utilization used during energy calculation
2604 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2605 * need to be aggregated differently depending on the usage made of them. This
2606 * enum is used within effective_cpu_util() to differentiate the types of
2607 * utilization expected by the callers, and adjust the aggregation accordingly.
2609 enum cpu_util_type {
2614 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2615 unsigned long max, enum cpu_util_type type,
2616 struct task_struct *p);
2618 static inline unsigned long cpu_bw_dl(struct rq *rq)
2620 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2623 static inline unsigned long cpu_util_dl(struct rq *rq)
2625 return READ_ONCE(rq->avg_dl.util_avg);
2628 static inline unsigned long cpu_util_cfs(struct rq *rq)
2630 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2632 if (sched_feat(UTIL_EST)) {
2633 util = max_t(unsigned long, util,
2634 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2640 static inline unsigned long cpu_util_rt(struct rq *rq)
2642 return READ_ONCE(rq->avg_rt.util_avg);
2646 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2647 static inline unsigned long cpu_util_irq(struct rq *rq)
2649 return rq->avg_irq.util_avg;
2653 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2655 util *= (max - irq);
2662 static inline unsigned long cpu_util_irq(struct rq *rq)
2668 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2674 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2676 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2678 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2680 static inline bool sched_energy_enabled(void)
2682 return static_branch_unlikely(&sched_energy_present);
2685 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2687 #define perf_domain_span(pd) NULL
2688 static inline bool sched_energy_enabled(void) { return false; }
2690 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2692 #ifdef CONFIG_MEMBARRIER
2694 * The scheduler provides memory barriers required by membarrier between:
2695 * - prior user-space memory accesses and store to rq->membarrier_state,
2696 * - store to rq->membarrier_state and following user-space memory accesses.
2697 * In the same way it provides those guarantees around store to rq->curr.
2699 static inline void membarrier_switch_mm(struct rq *rq,
2700 struct mm_struct *prev_mm,
2701 struct mm_struct *next_mm)
2703 int membarrier_state;
2705 if (prev_mm == next_mm)
2708 membarrier_state = atomic_read(&next_mm->membarrier_state);
2709 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2712 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2715 static inline void membarrier_switch_mm(struct rq *rq,
2716 struct mm_struct *prev_mm,
2717 struct mm_struct *next_mm)
2723 static inline bool is_per_cpu_kthread(struct task_struct *p)
2725 if (!(p->flags & PF_KTHREAD))
2728 if (p->nr_cpus_allowed != 1)
2735 void swake_up_all_locked(struct swait_queue_head *q);
2736 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);