1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/bitops.h>
40 #include <linux/blkdev.h>
41 #include <linux/compat.h>
42 #include <linux/context_tracking.h>
43 #include <linux/cpufreq.h>
44 #include <linux/cpuidle.h>
45 #include <linux/cpuset.h>
46 #include <linux/ctype.h>
47 #include <linux/debugfs.h>
48 #include <linux/delayacct.h>
49 #include <linux/energy_model.h>
50 #include <linux/init_task.h>
51 #include <linux/kprobes.h>
52 #include <linux/kthread.h>
53 #include <linux/membarrier.h>
54 #include <linux/migrate.h>
55 #include <linux/mmu_context.h>
56 #include <linux/nmi.h>
57 #include <linux/proc_fs.h>
58 #include <linux/prefetch.h>
59 #include <linux/profile.h>
60 #include <linux/psi.h>
61 #include <linux/rcupdate_wait.h>
62 #include <linux/security.h>
63 #include <linux/stop_machine.h>
64 #include <linux/suspend.h>
65 #include <linux/swait.h>
66 #include <linux/syscalls.h>
67 #include <linux/task_work.h>
68 #include <linux/tsacct_kern.h>
72 #ifdef CONFIG_PARAVIRT
73 # include <asm/paravirt.h>
77 #include "cpudeadline.h"
79 #include <trace/events/sched.h>
81 #ifdef CONFIG_SCHED_DEBUG
82 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
84 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
90 /* task_struct::on_rq states: */
91 #define TASK_ON_RQ_QUEUED 1
92 #define TASK_ON_RQ_MIGRATING 2
94 extern __read_mostly int scheduler_running;
96 extern unsigned long calc_load_update;
97 extern atomic_long_t calc_load_tasks;
99 extern void calc_global_load_tick(struct rq *this_rq);
100 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
102 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
104 * Helpers for converting nanosecond timing to jiffy resolution
106 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
109 * Increase resolution of nice-level calculations for 64-bit architectures.
110 * The extra resolution improves shares distribution and load balancing of
111 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
112 * hierarchies, especially on larger systems. This is not a user-visible change
113 * and does not change the user-interface for setting shares/weights.
115 * We increase resolution only if we have enough bits to allow this increased
116 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
117 * are pretty high and the returns do not justify the increased costs.
119 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
120 * increase coverage and consistency always enable it on 64-bit platforms.
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load_down(w) \
127 unsigned long __w = (w); \
129 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
133 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
134 # define scale_load(w) (w)
135 # define scale_load_down(w) (w)
139 * Task weight (visible to users) and its load (invisible to users) have
140 * independent resolution, but they should be well calibrated. We use
141 * scale_load() and scale_load_down(w) to convert between them. The
142 * following must be true:
144 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
147 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
150 * Single value that decides SCHED_DEADLINE internal math precision.
151 * 10 -> just above 1us
152 * 9 -> just above 0.5us
157 * Single value that denotes runtime == period, ie unlimited time.
159 #define RUNTIME_INF ((u64)~0ULL)
161 static inline int idle_policy(int policy)
163 return policy == SCHED_IDLE;
165 static inline int fair_policy(int policy)
167 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
170 static inline int rt_policy(int policy)
172 return policy == SCHED_FIFO || policy == SCHED_RR;
175 static inline int dl_policy(int policy)
177 return policy == SCHED_DEADLINE;
179 static inline bool valid_policy(int policy)
181 return idle_policy(policy) || fair_policy(policy) ||
182 rt_policy(policy) || dl_policy(policy);
185 static inline int task_has_idle_policy(struct task_struct *p)
187 return idle_policy(p->policy);
190 static inline int task_has_rt_policy(struct task_struct *p)
192 return rt_policy(p->policy);
195 static inline int task_has_dl_policy(struct task_struct *p)
197 return dl_policy(p->policy);
200 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
202 static inline void update_avg(u64 *avg, u64 sample)
204 s64 diff = sample - *avg;
209 * Shifting a value by an exponent greater *or equal* to the size of said value
210 * is UB; cap at size-1.
212 #define shr_bound(val, shift) \
213 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
216 * !! For sched_setattr_nocheck() (kernel) only !!
218 * This is actually gross. :(
220 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
221 * tasks, but still be able to sleep. We need this on platforms that cannot
222 * atomically change clock frequency. Remove once fast switching will be
223 * available on such platforms.
225 * SUGOV stands for SchedUtil GOVernor.
227 #define SCHED_FLAG_SUGOV 0x10000000
229 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
231 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
232 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
239 * Tells if entity @a should preempt entity @b.
242 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
244 return dl_entity_is_special(a) ||
245 dl_time_before(a->deadline, b->deadline);
249 * This is the priority-queue data structure of the RT scheduling class:
251 struct rt_prio_array {
252 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
253 struct list_head queue[MAX_RT_PRIO];
256 struct rt_bandwidth {
257 /* nests inside the rq lock: */
258 raw_spinlock_t rt_runtime_lock;
261 struct hrtimer rt_period_timer;
262 unsigned int rt_period_active;
265 void __dl_clear_params(struct task_struct *p);
267 struct dl_bandwidth {
268 raw_spinlock_t dl_runtime_lock;
273 static inline int dl_bandwidth_enabled(void)
275 return sysctl_sched_rt_runtime >= 0;
279 * To keep the bandwidth of -deadline tasks under control
280 * we need some place where:
281 * - store the maximum -deadline bandwidth of each cpu;
282 * - cache the fraction of bandwidth that is currently allocated in
285 * This is all done in the data structure below. It is similar to the
286 * one used for RT-throttling (rt_bandwidth), with the main difference
287 * that, since here we are only interested in admission control, we
288 * do not decrease any runtime while the group "executes", neither we
289 * need a timer to replenish it.
291 * With respect to SMP, bandwidth is given on a per root domain basis,
293 * - bw (< 100%) is the deadline bandwidth of each CPU;
294 * - total_bw is the currently allocated bandwidth in each root domain;
302 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
305 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
307 dl_b->total_bw -= tsk_bw;
308 __dl_update(dl_b, (s32)tsk_bw / cpus);
312 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
314 dl_b->total_bw += tsk_bw;
315 __dl_update(dl_b, -((s32)tsk_bw / cpus));
318 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
319 u64 old_bw, u64 new_bw)
321 return dl_b->bw != -1 &&
322 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
326 * Verify the fitness of task @p to run on @cpu taking into account the
327 * CPU original capacity and the runtime/deadline ratio of the task.
329 * The function will return true if the CPU original capacity of the
330 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
331 * task and false otherwise.
333 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
335 unsigned long cap = arch_scale_cpu_capacity(cpu);
337 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
340 extern void init_dl_bw(struct dl_bw *dl_b);
341 extern int sched_dl_global_validate(void);
342 extern void sched_dl_do_global(void);
343 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
344 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
345 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
346 extern bool __checkparam_dl(const struct sched_attr *attr);
347 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
348 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
349 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
350 extern bool dl_cpu_busy(unsigned int cpu);
352 #ifdef CONFIG_CGROUP_SCHED
354 #include <linux/cgroup.h>
355 #include <linux/psi.h>
360 extern struct list_head task_groups;
362 struct cfs_bandwidth {
363 #ifdef CONFIG_CFS_BANDWIDTH
368 s64 hierarchical_quota;
373 struct hrtimer period_timer;
374 struct hrtimer slack_timer;
375 struct list_head throttled_cfs_rq;
384 /* Task group related information */
386 struct cgroup_subsys_state css;
388 #ifdef CONFIG_FAIR_GROUP_SCHED
389 /* schedulable entities of this group on each CPU */
390 struct sched_entity **se;
391 /* runqueue "owned" by this group on each CPU */
392 struct cfs_rq **cfs_rq;
393 unsigned long shares;
397 * load_avg can be heavily contended at clock tick time, so put
398 * it in its own cacheline separated from the fields above which
399 * will also be accessed at each tick.
401 atomic_long_t load_avg ____cacheline_aligned;
405 #ifdef CONFIG_RT_GROUP_SCHED
406 struct sched_rt_entity **rt_se;
407 struct rt_rq **rt_rq;
409 struct rt_bandwidth rt_bandwidth;
413 struct list_head list;
415 struct task_group *parent;
416 struct list_head siblings;
417 struct list_head children;
419 #ifdef CONFIG_SCHED_AUTOGROUP
420 struct autogroup *autogroup;
423 struct cfs_bandwidth cfs_bandwidth;
425 #ifdef CONFIG_UCLAMP_TASK_GROUP
426 /* The two decimal precision [%] value requested from user-space */
427 unsigned int uclamp_pct[UCLAMP_CNT];
428 /* Clamp values requested for a task group */
429 struct uclamp_se uclamp_req[UCLAMP_CNT];
430 /* Effective clamp values used for a task group */
431 struct uclamp_se uclamp[UCLAMP_CNT];
436 #ifdef CONFIG_FAIR_GROUP_SCHED
437 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
440 * A weight of 0 or 1 can cause arithmetics problems.
441 * A weight of a cfs_rq is the sum of weights of which entities
442 * are queued on this cfs_rq, so a weight of a entity should not be
443 * too large, so as the shares value of a task group.
444 * (The default weight is 1024 - so there's no practical
445 * limitation from this.)
447 #define MIN_SHARES (1UL << 1)
448 #define MAX_SHARES (1UL << 18)
451 typedef int (*tg_visitor)(struct task_group *, void *);
453 extern int walk_tg_tree_from(struct task_group *from,
454 tg_visitor down, tg_visitor up, void *data);
457 * Iterate the full tree, calling @down when first entering a node and @up when
458 * leaving it for the final time.
460 * Caller must hold rcu_lock or sufficient equivalent.
462 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
464 return walk_tg_tree_from(&root_task_group, down, up, data);
467 extern int tg_nop(struct task_group *tg, void *data);
469 extern void free_fair_sched_group(struct task_group *tg);
470 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
471 extern void online_fair_sched_group(struct task_group *tg);
472 extern void unregister_fair_sched_group(struct task_group *tg);
473 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
474 struct sched_entity *se, int cpu,
475 struct sched_entity *parent);
476 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
478 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
479 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
480 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
482 extern void free_rt_sched_group(struct task_group *tg);
483 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
484 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
485 struct sched_rt_entity *rt_se, int cpu,
486 struct sched_rt_entity *parent);
487 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
488 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
489 extern long sched_group_rt_runtime(struct task_group *tg);
490 extern long sched_group_rt_period(struct task_group *tg);
491 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
493 extern struct task_group *sched_create_group(struct task_group *parent);
494 extern void sched_online_group(struct task_group *tg,
495 struct task_group *parent);
496 extern void sched_destroy_group(struct task_group *tg);
497 extern void sched_offline_group(struct task_group *tg);
499 extern void sched_move_task(struct task_struct *tsk);
501 #ifdef CONFIG_FAIR_GROUP_SCHED
502 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
505 extern void set_task_rq_fair(struct sched_entity *se,
506 struct cfs_rq *prev, struct cfs_rq *next);
507 #else /* !CONFIG_SMP */
508 static inline void set_task_rq_fair(struct sched_entity *se,
509 struct cfs_rq *prev, struct cfs_rq *next) { }
510 #endif /* CONFIG_SMP */
511 #endif /* CONFIG_FAIR_GROUP_SCHED */
513 #else /* CONFIG_CGROUP_SCHED */
515 struct cfs_bandwidth { };
517 #endif /* CONFIG_CGROUP_SCHED */
519 /* CFS-related fields in a runqueue */
521 struct load_weight load;
522 unsigned int nr_running;
523 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
524 unsigned int idle_h_nr_running; /* SCHED_IDLE */
529 u64 min_vruntime_copy;
532 struct rb_root_cached tasks_timeline;
535 * 'curr' points to currently running entity on this cfs_rq.
536 * It is set to NULL otherwise (i.e when none are currently running).
538 struct sched_entity *curr;
539 struct sched_entity *next;
540 struct sched_entity *last;
541 struct sched_entity *skip;
543 #ifdef CONFIG_SCHED_DEBUG
544 unsigned int nr_spread_over;
551 struct sched_avg avg;
553 u64 load_last_update_time_copy;
556 raw_spinlock_t lock ____cacheline_aligned;
558 unsigned long load_avg;
559 unsigned long util_avg;
560 unsigned long runnable_avg;
563 #ifdef CONFIG_FAIR_GROUP_SCHED
564 unsigned long tg_load_avg_contrib;
566 long prop_runnable_sum;
569 * h_load = weight * f(tg)
571 * Where f(tg) is the recursive weight fraction assigned to
574 unsigned long h_load;
575 u64 last_h_load_update;
576 struct sched_entity *h_load_next;
577 #endif /* CONFIG_FAIR_GROUP_SCHED */
578 #endif /* CONFIG_SMP */
580 #ifdef CONFIG_FAIR_GROUP_SCHED
581 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
584 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
585 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
586 * (like users, containers etc.)
588 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
589 * This list is used during load balance.
592 struct list_head leaf_cfs_rq_list;
593 struct task_group *tg; /* group that "owns" this runqueue */
595 #ifdef CONFIG_CFS_BANDWIDTH
597 s64 runtime_remaining;
600 u64 throttled_clock_task;
601 u64 throttled_clock_task_time;
604 struct list_head throttled_list;
605 #endif /* CONFIG_CFS_BANDWIDTH */
606 #endif /* CONFIG_FAIR_GROUP_SCHED */
609 static inline int rt_bandwidth_enabled(void)
611 return sysctl_sched_rt_runtime >= 0;
614 /* RT IPI pull logic requires IRQ_WORK */
615 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
616 # define HAVE_RT_PUSH_IPI
619 /* Real-Time classes' related field in a runqueue: */
621 struct rt_prio_array active;
622 unsigned int rt_nr_running;
623 unsigned int rr_nr_running;
624 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
626 int curr; /* highest queued rt task prio */
628 int next; /* next highest */
633 unsigned long rt_nr_migratory;
634 unsigned long rt_nr_total;
636 struct plist_head pushable_tasks;
638 #endif /* CONFIG_SMP */
644 /* Nests inside the rq lock: */
645 raw_spinlock_t rt_runtime_lock;
647 #ifdef CONFIG_RT_GROUP_SCHED
648 unsigned long rt_nr_boosted;
651 struct task_group *tg;
655 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
657 return rt_rq->rt_queued && rt_rq->rt_nr_running;
660 /* Deadline class' related fields in a runqueue */
662 /* runqueue is an rbtree, ordered by deadline */
663 struct rb_root_cached root;
665 unsigned long dl_nr_running;
669 * Deadline values of the currently executing and the
670 * earliest ready task on this rq. Caching these facilitates
671 * the decision whether or not a ready but not running task
672 * should migrate somewhere else.
679 unsigned long dl_nr_migratory;
683 * Tasks on this rq that can be pushed away. They are kept in
684 * an rb-tree, ordered by tasks' deadlines, with caching
685 * of the leftmost (earliest deadline) element.
687 struct rb_root_cached pushable_dl_tasks_root;
692 * "Active utilization" for this runqueue: increased when a
693 * task wakes up (becomes TASK_RUNNING) and decreased when a
699 * Utilization of the tasks "assigned" to this runqueue (including
700 * the tasks that are in runqueue and the tasks that executed on this
701 * CPU and blocked). Increased when a task moves to this runqueue, and
702 * decreased when the task moves away (migrates, changes scheduling
703 * policy, or terminates).
704 * This is needed to compute the "inactive utilization" for the
705 * runqueue (inactive utilization = this_bw - running_bw).
711 * Inverse of the fraction of CPU utilization that can be reclaimed
712 * by the GRUB algorithm.
717 #ifdef CONFIG_FAIR_GROUP_SCHED
718 /* An entity is a task if it doesn't "own" a runqueue */
719 #define entity_is_task(se) (!se->my_q)
721 static inline void se_update_runnable(struct sched_entity *se)
723 if (!entity_is_task(se))
724 se->runnable_weight = se->my_q->h_nr_running;
727 static inline long se_runnable(struct sched_entity *se)
729 if (entity_is_task(se))
732 return se->runnable_weight;
736 #define entity_is_task(se) 1
738 static inline void se_update_runnable(struct sched_entity *se) {}
740 static inline long se_runnable(struct sched_entity *se)
748 * XXX we want to get rid of these helpers and use the full load resolution.
750 static inline long se_weight(struct sched_entity *se)
752 return scale_load_down(se->load.weight);
756 static inline bool sched_asym_prefer(int a, int b)
758 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
762 struct em_perf_domain *em_pd;
763 struct perf_domain *next;
767 /* Scheduling group status flags */
768 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
769 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
772 * We add the notion of a root-domain which will be used to define per-domain
773 * variables. Each exclusive cpuset essentially defines an island domain by
774 * fully partitioning the member CPUs from any other cpuset. Whenever a new
775 * exclusive cpuset is created, we also create and attach a new root-domain
784 cpumask_var_t online;
787 * Indicate pullable load on at least one CPU, e.g:
788 * - More than one runnable task
789 * - Running task is misfit
793 /* Indicate one or more cpus over-utilized (tipping point) */
797 * The bit corresponding to a CPU gets set here if such CPU has more
798 * than one runnable -deadline task (as it is below for RT tasks).
800 cpumask_var_t dlo_mask;
806 * Indicate whether a root_domain's dl_bw has been checked or
807 * updated. It's monotonously increasing value.
809 * Also, some corner cases, like 'wrap around' is dangerous, but given
810 * that u64 is 'big enough'. So that shouldn't be a concern.
814 #ifdef HAVE_RT_PUSH_IPI
816 * For IPI pull requests, loop across the rto_mask.
818 struct irq_work rto_push_work;
819 raw_spinlock_t rto_lock;
820 /* These are only updated and read within rto_lock */
823 /* These atomics are updated outside of a lock */
824 atomic_t rto_loop_next;
825 atomic_t rto_loop_start;
828 * The "RT overload" flag: it gets set if a CPU has more than
829 * one runnable RT task.
831 cpumask_var_t rto_mask;
832 struct cpupri cpupri;
834 unsigned long max_cpu_capacity;
837 * NULL-terminated list of performance domains intersecting with the
838 * CPUs of the rd. Protected by RCU.
840 struct perf_domain __rcu *pd;
843 extern void init_defrootdomain(void);
844 extern int sched_init_domains(const struct cpumask *cpu_map);
845 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
846 extern void sched_get_rd(struct root_domain *rd);
847 extern void sched_put_rd(struct root_domain *rd);
849 #ifdef HAVE_RT_PUSH_IPI
850 extern void rto_push_irq_work_func(struct irq_work *work);
852 #endif /* CONFIG_SMP */
854 #ifdef CONFIG_UCLAMP_TASK
856 * struct uclamp_bucket - Utilization clamp bucket
857 * @value: utilization clamp value for tasks on this clamp bucket
858 * @tasks: number of RUNNABLE tasks on this clamp bucket
860 * Keep track of how many tasks are RUNNABLE for a given utilization
863 struct uclamp_bucket {
864 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
865 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
869 * struct uclamp_rq - rq's utilization clamp
870 * @value: currently active clamp values for a rq
871 * @bucket: utilization clamp buckets affecting a rq
873 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
874 * A clamp value is affecting a rq when there is at least one task RUNNABLE
875 * (or actually running) with that value.
877 * There are up to UCLAMP_CNT possible different clamp values, currently there
878 * are only two: minimum utilization and maximum utilization.
880 * All utilization clamping values are MAX aggregated, since:
881 * - for util_min: we want to run the CPU at least at the max of the minimum
882 * utilization required by its currently RUNNABLE tasks.
883 * - for util_max: we want to allow the CPU to run up to the max of the
884 * maximum utilization allowed by its currently RUNNABLE tasks.
886 * Since on each system we expect only a limited number of different
887 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
888 * the metrics required to compute all the per-rq utilization clamp values.
892 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
895 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
896 #endif /* CONFIG_UCLAMP_TASK */
899 * This is the main, per-CPU runqueue data structure.
901 * Locking rule: those places that want to lock multiple runqueues
902 * (such as the load balancing or the thread migration code), lock
903 * acquire operations must be ordered by ascending &runqueue.
910 * nr_running and cpu_load should be in the same cacheline because
911 * remote CPUs use both these fields when doing load calculation.
913 unsigned int nr_running;
914 #ifdef CONFIG_NUMA_BALANCING
915 unsigned int nr_numa_running;
916 unsigned int nr_preferred_running;
917 unsigned int numa_migrate_on;
919 #ifdef CONFIG_NO_HZ_COMMON
921 unsigned long last_blocked_load_update_tick;
922 unsigned int has_blocked_load;
923 call_single_data_t nohz_csd;
924 #endif /* CONFIG_SMP */
925 unsigned int nohz_tick_stopped;
927 #endif /* CONFIG_NO_HZ_COMMON */
930 unsigned int ttwu_pending;
934 #ifdef CONFIG_UCLAMP_TASK
935 /* Utilization clamp values based on CPU's RUNNABLE tasks */
936 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
937 unsigned int uclamp_flags;
938 #define UCLAMP_FLAG_IDLE 0x01
945 #ifdef CONFIG_FAIR_GROUP_SCHED
946 /* list of leaf cfs_rq on this CPU: */
947 struct list_head leaf_cfs_rq_list;
948 struct list_head *tmp_alone_branch;
949 #endif /* CONFIG_FAIR_GROUP_SCHED */
952 * This is part of a global counter where only the total sum
953 * over all CPUs matters. A task can increase this counter on
954 * one CPU and if it got migrated afterwards it may decrease
955 * it on another CPU. Always updated under the runqueue lock:
957 unsigned long nr_uninterruptible;
959 struct task_struct __rcu *curr;
960 struct task_struct *idle;
961 struct task_struct *stop;
962 unsigned long next_balance;
963 struct mm_struct *prev_mm;
965 unsigned int clock_update_flags;
967 /* Ensure that all clocks are in the same cache line */
968 u64 clock_task ____cacheline_aligned;
970 unsigned long lost_idle_time;
974 #ifdef CONFIG_MEMBARRIER
975 int membarrier_state;
979 struct root_domain *rd;
980 struct sched_domain __rcu *sd;
982 unsigned long cpu_capacity;
983 unsigned long cpu_capacity_orig;
985 struct callback_head *balance_callback;
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_UPDATED 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 canceled.
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 int sched_update_scaling(void);
1573 extern void flush_smp_call_function_from_idle(void);
1575 #else /* !CONFIG_SMP: */
1576 static inline void flush_smp_call_function_from_idle(void) { }
1580 #include "autogroup.h"
1582 #ifdef CONFIG_CGROUP_SCHED
1585 * Return the group to which this tasks belongs.
1587 * We cannot use task_css() and friends because the cgroup subsystem
1588 * changes that value before the cgroup_subsys::attach() method is called,
1589 * therefore we cannot pin it and might observe the wrong value.
1591 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1592 * core changes this before calling sched_move_task().
1594 * Instead we use a 'copy' which is updated from sched_move_task() while
1595 * holding both task_struct::pi_lock and rq::lock.
1597 static inline struct task_group *task_group(struct task_struct *p)
1599 return p->sched_task_group;
1602 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1603 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1605 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1606 struct task_group *tg = task_group(p);
1609 #ifdef CONFIG_FAIR_GROUP_SCHED
1610 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1611 p->se.cfs_rq = tg->cfs_rq[cpu];
1612 p->se.parent = tg->se[cpu];
1615 #ifdef CONFIG_RT_GROUP_SCHED
1616 p->rt.rt_rq = tg->rt_rq[cpu];
1617 p->rt.parent = tg->rt_se[cpu];
1621 #else /* CONFIG_CGROUP_SCHED */
1623 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1624 static inline struct task_group *task_group(struct task_struct *p)
1629 #endif /* CONFIG_CGROUP_SCHED */
1631 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1633 set_task_rq(p, cpu);
1636 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1637 * successfully executed on another CPU. We must ensure that updates of
1638 * per-task data have been completed by this moment.
1641 #ifdef CONFIG_THREAD_INFO_IN_TASK
1642 WRITE_ONCE(p->cpu, cpu);
1644 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1651 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1653 #ifdef CONFIG_SCHED_DEBUG
1654 # include <linux/static_key.h>
1655 # define const_debug __read_mostly
1657 # define const_debug const
1660 #define SCHED_FEAT(name, enabled) \
1661 __SCHED_FEAT_##name ,
1664 #include "features.h"
1670 #ifdef CONFIG_SCHED_DEBUG
1673 * To support run-time toggling of sched features, all the translation units
1674 * (but core.c) reference the sysctl_sched_features defined in core.c.
1676 extern const_debug unsigned int sysctl_sched_features;
1678 #ifdef CONFIG_JUMP_LABEL
1679 #define SCHED_FEAT(name, enabled) \
1680 static __always_inline bool static_branch_##name(struct static_key *key) \
1682 return static_key_##enabled(key); \
1685 #include "features.h"
1688 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1689 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1691 #else /* !CONFIG_JUMP_LABEL */
1693 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1695 #endif /* CONFIG_JUMP_LABEL */
1697 #else /* !SCHED_DEBUG */
1700 * Each translation unit has its own copy of sysctl_sched_features to allow
1701 * constants propagation at compile time and compiler optimization based on
1704 #define SCHED_FEAT(name, enabled) \
1705 (1UL << __SCHED_FEAT_##name) * enabled |
1706 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1707 #include "features.h"
1711 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1713 #endif /* SCHED_DEBUG */
1715 extern struct static_key_false sched_numa_balancing;
1716 extern struct static_key_false sched_schedstats;
1718 static inline u64 global_rt_period(void)
1720 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1723 static inline u64 global_rt_runtime(void)
1725 if (sysctl_sched_rt_runtime < 0)
1728 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1731 static inline int task_current(struct rq *rq, struct task_struct *p)
1733 return rq->curr == p;
1736 static inline int task_running(struct rq *rq, struct task_struct *p)
1741 return task_current(rq, p);
1745 static inline int task_on_rq_queued(struct task_struct *p)
1747 return p->on_rq == TASK_ON_RQ_QUEUED;
1750 static inline int task_on_rq_migrating(struct task_struct *p)
1752 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1755 /* Wake flags. The first three directly map to some SD flag value */
1756 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
1757 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
1758 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
1760 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
1761 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
1762 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
1765 static_assert(WF_EXEC == SD_BALANCE_EXEC);
1766 static_assert(WF_FORK == SD_BALANCE_FORK);
1767 static_assert(WF_TTWU == SD_BALANCE_WAKE);
1771 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1772 * of tasks with abnormal "nice" values across CPUs the contribution that
1773 * each task makes to its run queue's load is weighted according to its
1774 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1775 * scaled version of the new time slice allocation that they receive on time
1779 #define WEIGHT_IDLEPRIO 3
1780 #define WMULT_IDLEPRIO 1431655765
1782 extern const int sched_prio_to_weight[40];
1783 extern const u32 sched_prio_to_wmult[40];
1786 * {de,en}queue flags:
1788 * DEQUEUE_SLEEP - task is no longer runnable
1789 * ENQUEUE_WAKEUP - task just became runnable
1791 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1792 * are in a known state which allows modification. Such pairs
1793 * should preserve as much state as possible.
1795 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1798 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1799 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1800 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1804 #define DEQUEUE_SLEEP 0x01
1805 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1806 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1807 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1809 #define ENQUEUE_WAKEUP 0x01
1810 #define ENQUEUE_RESTORE 0x02
1811 #define ENQUEUE_MOVE 0x04
1812 #define ENQUEUE_NOCLOCK 0x08
1814 #define ENQUEUE_HEAD 0x10
1815 #define ENQUEUE_REPLENISH 0x20
1817 #define ENQUEUE_MIGRATED 0x40
1819 #define ENQUEUE_MIGRATED 0x00
1822 #define RETRY_TASK ((void *)-1UL)
1824 struct sched_class {
1826 #ifdef CONFIG_UCLAMP_TASK
1830 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1831 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1832 void (*yield_task) (struct rq *rq);
1833 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1835 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1837 struct task_struct *(*pick_next_task)(struct rq *rq);
1839 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1840 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1843 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1844 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
1845 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1847 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1849 void (*set_cpus_allowed)(struct task_struct *p,
1850 const struct cpumask *newmask,
1853 void (*rq_online)(struct rq *rq);
1854 void (*rq_offline)(struct rq *rq);
1856 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
1859 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1860 void (*task_fork)(struct task_struct *p);
1861 void (*task_dead)(struct task_struct *p);
1864 * The switched_from() call is allowed to drop rq->lock, therefore we
1865 * cannot assume the switched_from/switched_to pair is serialized by
1866 * rq->lock. They are however serialized by p->pi_lock.
1868 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1869 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1870 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1873 unsigned int (*get_rr_interval)(struct rq *rq,
1874 struct task_struct *task);
1876 void (*update_curr)(struct rq *rq);
1878 #define TASK_SET_GROUP 0
1879 #define TASK_MOVE_GROUP 1
1881 #ifdef CONFIG_FAIR_GROUP_SCHED
1882 void (*task_change_group)(struct task_struct *p, int type);
1886 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1888 WARN_ON_ONCE(rq->curr != prev);
1889 prev->sched_class->put_prev_task(rq, prev);
1892 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1894 WARN_ON_ONCE(rq->curr != next);
1895 next->sched_class->set_next_task(rq, next, false);
1900 * Helper to define a sched_class instance; each one is placed in a separate
1901 * section which is ordered by the linker script:
1903 * include/asm-generic/vmlinux.lds.h
1905 * Also enforce alignment on the instance, not the type, to guarantee layout.
1907 #define DEFINE_SCHED_CLASS(name) \
1908 const struct sched_class name##_sched_class \
1909 __aligned(__alignof__(struct sched_class)) \
1910 __section("__" #name "_sched_class")
1912 /* Defined in include/asm-generic/vmlinux.lds.h */
1913 extern struct sched_class __begin_sched_classes[];
1914 extern struct sched_class __end_sched_classes[];
1916 #define sched_class_highest (__end_sched_classes - 1)
1917 #define sched_class_lowest (__begin_sched_classes - 1)
1919 #define for_class_range(class, _from, _to) \
1920 for (class = (_from); class != (_to); class--)
1922 #define for_each_class(class) \
1923 for_class_range(class, sched_class_highest, sched_class_lowest)
1925 extern const struct sched_class stop_sched_class;
1926 extern const struct sched_class dl_sched_class;
1927 extern const struct sched_class rt_sched_class;
1928 extern const struct sched_class fair_sched_class;
1929 extern const struct sched_class idle_sched_class;
1931 static inline bool sched_stop_runnable(struct rq *rq)
1933 return rq->stop && task_on_rq_queued(rq->stop);
1936 static inline bool sched_dl_runnable(struct rq *rq)
1938 return rq->dl.dl_nr_running > 0;
1941 static inline bool sched_rt_runnable(struct rq *rq)
1943 return rq->rt.rt_queued > 0;
1946 static inline bool sched_fair_runnable(struct rq *rq)
1948 return rq->cfs.nr_running > 0;
1951 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1952 extern struct task_struct *pick_next_task_idle(struct rq *rq);
1954 #define SCA_CHECK 0x01
1955 #define SCA_MIGRATE_DISABLE 0x02
1956 #define SCA_MIGRATE_ENABLE 0x04
1960 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1962 extern void trigger_load_balance(struct rq *rq);
1964 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
1966 static inline struct task_struct *get_push_task(struct rq *rq)
1968 struct task_struct *p = rq->curr;
1970 lockdep_assert_held(&rq->lock);
1975 if (p->nr_cpus_allowed == 1)
1978 rq->push_busy = true;
1979 return get_task_struct(p);
1982 extern int push_cpu_stop(void *arg);
1986 #ifdef CONFIG_CPU_IDLE
1987 static inline void idle_set_state(struct rq *rq,
1988 struct cpuidle_state *idle_state)
1990 rq->idle_state = idle_state;
1993 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1995 SCHED_WARN_ON(!rcu_read_lock_held());
1997 return rq->idle_state;
2000 static inline void idle_set_state(struct rq *rq,
2001 struct cpuidle_state *idle_state)
2005 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2011 extern void schedule_idle(void);
2013 extern void sysrq_sched_debug_show(void);
2014 extern void sched_init_granularity(void);
2015 extern void update_max_interval(void);
2017 extern void init_sched_dl_class(void);
2018 extern void init_sched_rt_class(void);
2019 extern void init_sched_fair_class(void);
2021 extern void reweight_task(struct task_struct *p, int prio);
2023 extern void resched_curr(struct rq *rq);
2024 extern void resched_cpu(int cpu);
2026 extern struct rt_bandwidth def_rt_bandwidth;
2027 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2029 extern struct dl_bandwidth def_dl_bandwidth;
2030 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2031 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2032 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2035 #define BW_UNIT (1 << BW_SHIFT)
2036 #define RATIO_SHIFT 8
2037 #define MAX_BW_BITS (64 - BW_SHIFT)
2038 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2039 unsigned long to_ratio(u64 period, u64 runtime);
2041 extern void init_entity_runnable_average(struct sched_entity *se);
2042 extern void post_init_entity_util_avg(struct task_struct *p);
2044 #ifdef CONFIG_NO_HZ_FULL
2045 extern bool sched_can_stop_tick(struct rq *rq);
2046 extern int __init sched_tick_offload_init(void);
2049 * Tick may be needed by tasks in the runqueue depending on their policy and
2050 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2051 * nohz mode if necessary.
2053 static inline void sched_update_tick_dependency(struct rq *rq)
2055 int cpu = cpu_of(rq);
2057 if (!tick_nohz_full_cpu(cpu))
2060 if (sched_can_stop_tick(rq))
2061 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2063 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2066 static inline int sched_tick_offload_init(void) { return 0; }
2067 static inline void sched_update_tick_dependency(struct rq *rq) { }
2070 static inline void add_nr_running(struct rq *rq, unsigned count)
2072 unsigned prev_nr = rq->nr_running;
2074 rq->nr_running = prev_nr + count;
2075 if (trace_sched_update_nr_running_tp_enabled()) {
2076 call_trace_sched_update_nr_running(rq, count);
2080 if (prev_nr < 2 && rq->nr_running >= 2) {
2081 if (!READ_ONCE(rq->rd->overload))
2082 WRITE_ONCE(rq->rd->overload, 1);
2086 sched_update_tick_dependency(rq);
2089 static inline void sub_nr_running(struct rq *rq, unsigned count)
2091 rq->nr_running -= count;
2092 if (trace_sched_update_nr_running_tp_enabled()) {
2093 call_trace_sched_update_nr_running(rq, -count);
2096 /* Check if we still need preemption */
2097 sched_update_tick_dependency(rq);
2100 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2101 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2103 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2105 extern const_debug unsigned int sysctl_sched_nr_migrate;
2106 extern const_debug unsigned int sysctl_sched_migration_cost;
2108 #ifdef CONFIG_SCHED_HRTICK
2112 * - enabled by features
2113 * - hrtimer is actually high res
2115 static inline int hrtick_enabled(struct rq *rq)
2117 if (!cpu_active(cpu_of(rq)))
2119 return hrtimer_is_hres_active(&rq->hrtick_timer);
2122 static inline int hrtick_enabled_fair(struct rq *rq)
2124 if (!sched_feat(HRTICK))
2126 return hrtick_enabled(rq);
2129 static inline int hrtick_enabled_dl(struct rq *rq)
2131 if (!sched_feat(HRTICK_DL))
2133 return hrtick_enabled(rq);
2136 void hrtick_start(struct rq *rq, u64 delay);
2140 static inline int hrtick_enabled_fair(struct rq *rq)
2145 static inline int hrtick_enabled_dl(struct rq *rq)
2150 static inline int hrtick_enabled(struct rq *rq)
2155 #endif /* CONFIG_SCHED_HRTICK */
2157 #ifndef arch_scale_freq_tick
2158 static __always_inline
2159 void arch_scale_freq_tick(void)
2164 #ifndef arch_scale_freq_capacity
2166 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2167 * @cpu: the CPU in question.
2169 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2172 * ------ * SCHED_CAPACITY_SCALE
2175 static __always_inline
2176 unsigned long arch_scale_freq_capacity(int cpu)
2178 return SCHED_CAPACITY_SCALE;
2183 #ifdef CONFIG_PREEMPTION
2185 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2188 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2189 * way at the expense of forcing extra atomic operations in all
2190 * invocations. This assures that the double_lock is acquired using the
2191 * same underlying policy as the spinlock_t on this architecture, which
2192 * reduces latency compared to the unfair variant below. However, it
2193 * also adds more overhead and therefore may reduce throughput.
2195 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2196 __releases(this_rq->lock)
2197 __acquires(busiest->lock)
2198 __acquires(this_rq->lock)
2200 raw_spin_unlock(&this_rq->lock);
2201 double_rq_lock(this_rq, busiest);
2208 * Unfair double_lock_balance: Optimizes throughput at the expense of
2209 * latency by eliminating extra atomic operations when the locks are
2210 * already in proper order on entry. This favors lower CPU-ids and will
2211 * grant the double lock to lower CPUs over higher ids under contention,
2212 * regardless of entry order into the function.
2214 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2215 __releases(this_rq->lock)
2216 __acquires(busiest->lock)
2217 __acquires(this_rq->lock)
2221 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2222 if (busiest < this_rq) {
2223 raw_spin_unlock(&this_rq->lock);
2224 raw_spin_lock(&busiest->lock);
2225 raw_spin_lock_nested(&this_rq->lock,
2226 SINGLE_DEPTH_NESTING);
2229 raw_spin_lock_nested(&busiest->lock,
2230 SINGLE_DEPTH_NESTING);
2235 #endif /* CONFIG_PREEMPTION */
2238 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2240 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2242 if (unlikely(!irqs_disabled())) {
2243 /* printk() doesn't work well under rq->lock */
2244 raw_spin_unlock(&this_rq->lock);
2248 return _double_lock_balance(this_rq, busiest);
2251 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2252 __releases(busiest->lock)
2254 raw_spin_unlock(&busiest->lock);
2255 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2258 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2264 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2267 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2273 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2276 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2282 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2286 * double_rq_lock - safely lock two runqueues
2288 * Note this does not disable interrupts like task_rq_lock,
2289 * you need to do so manually before calling.
2291 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2292 __acquires(rq1->lock)
2293 __acquires(rq2->lock)
2295 BUG_ON(!irqs_disabled());
2297 raw_spin_lock(&rq1->lock);
2298 __acquire(rq2->lock); /* Fake it out ;) */
2301 raw_spin_lock(&rq1->lock);
2302 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2304 raw_spin_lock(&rq2->lock);
2305 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2311 * double_rq_unlock - safely unlock two runqueues
2313 * Note this does not restore interrupts like task_rq_unlock,
2314 * you need to do so manually after calling.
2316 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2317 __releases(rq1->lock)
2318 __releases(rq2->lock)
2320 raw_spin_unlock(&rq1->lock);
2322 raw_spin_unlock(&rq2->lock);
2324 __release(rq2->lock);
2327 extern void set_rq_online (struct rq *rq);
2328 extern void set_rq_offline(struct rq *rq);
2329 extern bool sched_smp_initialized;
2331 #else /* CONFIG_SMP */
2334 * double_rq_lock - safely lock two runqueues
2336 * Note this does not disable interrupts like task_rq_lock,
2337 * you need to do so manually before calling.
2339 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2340 __acquires(rq1->lock)
2341 __acquires(rq2->lock)
2343 BUG_ON(!irqs_disabled());
2345 raw_spin_lock(&rq1->lock);
2346 __acquire(rq2->lock); /* Fake it out ;) */
2350 * double_rq_unlock - safely unlock two runqueues
2352 * Note this does not restore interrupts like task_rq_unlock,
2353 * you need to do so manually after calling.
2355 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2356 __releases(rq1->lock)
2357 __releases(rq2->lock)
2360 raw_spin_unlock(&rq1->lock);
2361 __release(rq2->lock);
2366 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2367 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2369 #ifdef CONFIG_SCHED_DEBUG
2370 extern bool sched_debug_enabled;
2372 extern void print_cfs_stats(struct seq_file *m, int cpu);
2373 extern void print_rt_stats(struct seq_file *m, int cpu);
2374 extern void print_dl_stats(struct seq_file *m, int cpu);
2375 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2376 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2377 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2378 #ifdef CONFIG_NUMA_BALANCING
2380 show_numa_stats(struct task_struct *p, struct seq_file *m);
2382 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2383 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2384 #endif /* CONFIG_NUMA_BALANCING */
2385 #endif /* CONFIG_SCHED_DEBUG */
2387 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2388 extern void init_rt_rq(struct rt_rq *rt_rq);
2389 extern void init_dl_rq(struct dl_rq *dl_rq);
2391 extern void cfs_bandwidth_usage_inc(void);
2392 extern void cfs_bandwidth_usage_dec(void);
2394 #ifdef CONFIG_NO_HZ_COMMON
2395 #define NOHZ_BALANCE_KICK_BIT 0
2396 #define NOHZ_STATS_KICK_BIT 1
2397 #define NOHZ_NEWILB_KICK_BIT 2
2399 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2400 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2401 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2403 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2405 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2407 extern void nohz_balance_exit_idle(struct rq *rq);
2409 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2412 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2413 extern void nohz_run_idle_balance(int cpu);
2415 static inline void nohz_run_idle_balance(int cpu) { }
2420 void __dl_update(struct dl_bw *dl_b, s64 bw)
2422 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2425 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2426 "sched RCU must be held");
2427 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2428 struct rq *rq = cpu_rq(i);
2430 rq->dl.extra_bw += bw;
2435 void __dl_update(struct dl_bw *dl_b, s64 bw)
2437 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2444 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2449 struct u64_stats_sync sync;
2452 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2455 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2456 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2457 * and never move forward.
2459 static inline u64 irq_time_read(int cpu)
2461 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2466 seq = __u64_stats_fetch_begin(&irqtime->sync);
2467 total = irqtime->total;
2468 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2472 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2474 #ifdef CONFIG_CPU_FREQ
2475 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2478 * cpufreq_update_util - Take a note about CPU utilization changes.
2479 * @rq: Runqueue to carry out the update for.
2480 * @flags: Update reason flags.
2482 * This function is called by the scheduler on the CPU whose utilization is
2485 * It can only be called from RCU-sched read-side critical sections.
2487 * The way cpufreq is currently arranged requires it to evaluate the CPU
2488 * performance state (frequency/voltage) on a regular basis to prevent it from
2489 * being stuck in a completely inadequate performance level for too long.
2490 * That is not guaranteed to happen if the updates are only triggered from CFS
2491 * and DL, though, because they may not be coming in if only RT tasks are
2492 * active all the time (or there are RT tasks only).
2494 * As a workaround for that issue, this function is called periodically by the
2495 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2496 * but that really is a band-aid. Going forward it should be replaced with
2497 * solutions targeted more specifically at RT tasks.
2499 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2501 struct update_util_data *data;
2503 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2506 data->func(data, rq_clock(rq), flags);
2509 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2510 #endif /* CONFIG_CPU_FREQ */
2512 #ifdef CONFIG_UCLAMP_TASK
2513 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2516 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2517 * @rq: The rq to clamp against. Must not be NULL.
2518 * @util: The util value to clamp.
2519 * @p: The task to clamp against. Can be NULL if you want to clamp
2522 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2524 * If sched_uclamp_used static key is disabled, then just return the util
2525 * without any clamping since uclamp aggregation at the rq level in the fast
2526 * path is disabled, rendering this operation a NOP.
2528 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2529 * will return the correct effective uclamp value of the task even if the
2530 * static key is disabled.
2532 static __always_inline
2533 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2534 struct task_struct *p)
2536 unsigned long min_util;
2537 unsigned long max_util;
2539 if (!static_branch_likely(&sched_uclamp_used))
2542 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2543 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2546 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2547 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2551 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2552 * RUNNABLE tasks with _different_ clamps, we can end up with an
2553 * inversion. Fix it now when the clamps are applied.
2555 if (unlikely(min_util >= max_util))
2558 return clamp(util, min_util, max_util);
2562 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2563 * by default in the fast path and only gets turned on once userspace performs
2564 * an operation that requires it.
2566 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2569 static inline bool uclamp_is_used(void)
2571 return static_branch_likely(&sched_uclamp_used);
2573 #else /* CONFIG_UCLAMP_TASK */
2575 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2576 struct task_struct *p)
2581 static inline bool uclamp_is_used(void)
2585 #endif /* CONFIG_UCLAMP_TASK */
2587 #ifdef arch_scale_freq_capacity
2588 # ifndef arch_scale_freq_invariant
2589 # define arch_scale_freq_invariant() true
2592 # define arch_scale_freq_invariant() false
2596 static inline unsigned long capacity_orig_of(int cpu)
2598 return cpu_rq(cpu)->cpu_capacity_orig;
2602 * enum cpu_util_type - CPU utilization type
2603 * @FREQUENCY_UTIL: Utilization used to select frequency
2604 * @ENERGY_UTIL: Utilization used during energy calculation
2606 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2607 * need to be aggregated differently depending on the usage made of them. This
2608 * enum is used within effective_cpu_util() to differentiate the types of
2609 * utilization expected by the callers, and adjust the aggregation accordingly.
2611 enum cpu_util_type {
2616 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2617 unsigned long max, enum cpu_util_type type,
2618 struct task_struct *p);
2620 static inline unsigned long cpu_bw_dl(struct rq *rq)
2622 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2625 static inline unsigned long cpu_util_dl(struct rq *rq)
2627 return READ_ONCE(rq->avg_dl.util_avg);
2630 static inline unsigned long cpu_util_cfs(struct rq *rq)
2632 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2634 if (sched_feat(UTIL_EST)) {
2635 util = max_t(unsigned long, util,
2636 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2642 static inline unsigned long cpu_util_rt(struct rq *rq)
2644 return READ_ONCE(rq->avg_rt.util_avg);
2648 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2649 static inline unsigned long cpu_util_irq(struct rq *rq)
2651 return rq->avg_irq.util_avg;
2655 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2657 util *= (max - irq);
2664 static inline unsigned long cpu_util_irq(struct rq *rq)
2670 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2676 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2678 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2680 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2682 static inline bool sched_energy_enabled(void)
2684 return static_branch_unlikely(&sched_energy_present);
2687 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2689 #define perf_domain_span(pd) NULL
2690 static inline bool sched_energy_enabled(void) { return false; }
2692 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2694 #ifdef CONFIG_MEMBARRIER
2696 * The scheduler provides memory barriers required by membarrier between:
2697 * - prior user-space memory accesses and store to rq->membarrier_state,
2698 * - store to rq->membarrier_state and following user-space memory accesses.
2699 * In the same way it provides those guarantees around store to rq->curr.
2701 static inline void membarrier_switch_mm(struct rq *rq,
2702 struct mm_struct *prev_mm,
2703 struct mm_struct *next_mm)
2705 int membarrier_state;
2707 if (prev_mm == next_mm)
2710 membarrier_state = atomic_read(&next_mm->membarrier_state);
2711 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2714 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2717 static inline void membarrier_switch_mm(struct rq *rq,
2718 struct mm_struct *prev_mm,
2719 struct mm_struct *next_mm)
2725 static inline bool is_per_cpu_kthread(struct task_struct *p)
2727 if (!(p->flags & PF_KTHREAD))
2730 if (p->nr_cpus_allowed != 1)
2737 extern void swake_up_all_locked(struct swait_queue_head *q);
2738 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
2740 #ifdef CONFIG_PREEMPT_DYNAMIC
2741 extern int preempt_dynamic_mode;
2742 extern int sched_dynamic_mode(const char *str);
2743 extern void sched_dynamic_update(int mode);