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>
70 #include <asm-generic/vmlinux.lds.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[USER_PRIO(NICE_TO_PRIO(0))]) == 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 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
231 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
233 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
234 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
241 * Tells if entity @a should preempt entity @b.
244 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
246 return dl_entity_is_special(a) ||
247 dl_time_before(a->deadline, b->deadline);
251 * This is the priority-queue data structure of the RT scheduling class:
253 struct rt_prio_array {
254 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
255 struct list_head queue[MAX_RT_PRIO];
258 struct rt_bandwidth {
259 /* nests inside the rq lock: */
260 raw_spinlock_t rt_runtime_lock;
263 struct hrtimer rt_period_timer;
264 unsigned int rt_period_active;
267 void __dl_clear_params(struct task_struct *p);
269 struct dl_bandwidth {
270 raw_spinlock_t dl_runtime_lock;
275 static inline int dl_bandwidth_enabled(void)
277 return sysctl_sched_rt_runtime >= 0;
281 * To keep the bandwidth of -deadline tasks under control
282 * we need some place where:
283 * - store the maximum -deadline bandwidth of each cpu;
284 * - cache the fraction of bandwidth that is currently allocated in
287 * This is all done in the data structure below. It is similar to the
288 * one used for RT-throttling (rt_bandwidth), with the main difference
289 * that, since here we are only interested in admission control, we
290 * do not decrease any runtime while the group "executes", neither we
291 * need a timer to replenish it.
293 * With respect to SMP, bandwidth is given on a per root domain basis,
295 * - bw (< 100%) is the deadline bandwidth of each CPU;
296 * - total_bw is the currently allocated bandwidth in each root domain;
304 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
307 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
309 dl_b->total_bw -= tsk_bw;
310 __dl_update(dl_b, (s32)tsk_bw / cpus);
314 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
316 dl_b->total_bw += tsk_bw;
317 __dl_update(dl_b, -((s32)tsk_bw / cpus));
320 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
321 u64 old_bw, u64 new_bw)
323 return dl_b->bw != -1 &&
324 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
328 * Verify the fitness of task @p to run on @cpu taking into account the
329 * CPU original capacity and the runtime/deadline ratio of the task.
331 * The function will return true if the CPU original capacity of the
332 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
333 * task and false otherwise.
335 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
337 unsigned long cap = arch_scale_cpu_capacity(cpu);
339 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
342 extern void init_dl_bw(struct dl_bw *dl_b);
343 extern int sched_dl_global_validate(void);
344 extern void sched_dl_do_global(void);
345 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
346 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
347 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
348 extern bool __checkparam_dl(const struct sched_attr *attr);
349 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
350 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
351 extern int dl_bw_check_overflow(int cpu);
353 #ifdef CONFIG_CGROUP_SCHED
355 #include <linux/cgroup.h>
356 #include <linux/psi.h>
361 extern struct list_head task_groups;
363 struct cfs_bandwidth {
364 #ifdef CONFIG_CFS_BANDWIDTH
369 s64 hierarchical_quota;
374 struct hrtimer period_timer;
375 struct hrtimer slack_timer;
376 struct list_head throttled_cfs_rq;
385 /* Task group related information */
387 struct cgroup_subsys_state css;
389 #ifdef CONFIG_FAIR_GROUP_SCHED
390 /* schedulable entities of this group on each CPU */
391 struct sched_entity **se;
392 /* runqueue "owned" by this group on each CPU */
393 struct cfs_rq **cfs_rq;
394 unsigned long shares;
398 * load_avg can be heavily contended at clock tick time, so put
399 * it in its own cacheline separated from the fields above which
400 * will also be accessed at each tick.
402 atomic_long_t load_avg ____cacheline_aligned;
406 #ifdef CONFIG_RT_GROUP_SCHED
407 struct sched_rt_entity **rt_se;
408 struct rt_rq **rt_rq;
410 struct rt_bandwidth rt_bandwidth;
414 struct list_head list;
416 struct task_group *parent;
417 struct list_head siblings;
418 struct list_head children;
420 #ifdef CONFIG_SCHED_AUTOGROUP
421 struct autogroup *autogroup;
424 struct cfs_bandwidth cfs_bandwidth;
426 #ifdef CONFIG_UCLAMP_TASK_GROUP
427 /* The two decimal precision [%] value requested from user-space */
428 unsigned int uclamp_pct[UCLAMP_CNT];
429 /* Clamp values requested for a task group */
430 struct uclamp_se uclamp_req[UCLAMP_CNT];
431 /* Effective clamp values used for a task group */
432 struct uclamp_se uclamp[UCLAMP_CNT];
437 #ifdef CONFIG_FAIR_GROUP_SCHED
438 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
441 * A weight of 0 or 1 can cause arithmetics problems.
442 * A weight of a cfs_rq is the sum of weights of which entities
443 * are queued on this cfs_rq, so a weight of a entity should not be
444 * too large, so as the shares value of a task group.
445 * (The default weight is 1024 - so there's no practical
446 * limitation from this.)
448 #define MIN_SHARES (1UL << 1)
449 #define MAX_SHARES (1UL << 18)
452 typedef int (*tg_visitor)(struct task_group *, void *);
454 extern int walk_tg_tree_from(struct task_group *from,
455 tg_visitor down, tg_visitor up, void *data);
458 * Iterate the full tree, calling @down when first entering a node and @up when
459 * leaving it for the final time.
461 * Caller must hold rcu_lock or sufficient equivalent.
463 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
465 return walk_tg_tree_from(&root_task_group, down, up, data);
468 extern int tg_nop(struct task_group *tg, void *data);
470 extern void free_fair_sched_group(struct task_group *tg);
471 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
472 extern void online_fair_sched_group(struct task_group *tg);
473 extern void unregister_fair_sched_group(struct task_group *tg);
474 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
475 struct sched_entity *se, int cpu,
476 struct sched_entity *parent);
477 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
479 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
480 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
481 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
483 extern void free_rt_sched_group(struct task_group *tg);
484 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
485 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
486 struct sched_rt_entity *rt_se, int cpu,
487 struct sched_rt_entity *parent);
488 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
489 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
490 extern long sched_group_rt_runtime(struct task_group *tg);
491 extern long sched_group_rt_period(struct task_group *tg);
492 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
494 extern struct task_group *sched_create_group(struct task_group *parent);
495 extern void sched_online_group(struct task_group *tg,
496 struct task_group *parent);
497 extern void sched_destroy_group(struct task_group *tg);
498 extern void sched_offline_group(struct task_group *tg);
500 extern void sched_move_task(struct task_struct *tsk);
502 #ifdef CONFIG_FAIR_GROUP_SCHED
503 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
506 extern void set_task_rq_fair(struct sched_entity *se,
507 struct cfs_rq *prev, struct cfs_rq *next);
508 #else /* !CONFIG_SMP */
509 static inline void set_task_rq_fair(struct sched_entity *se,
510 struct cfs_rq *prev, struct cfs_rq *next) { }
511 #endif /* CONFIG_SMP */
512 #endif /* CONFIG_FAIR_GROUP_SCHED */
514 #else /* CONFIG_CGROUP_SCHED */
516 struct cfs_bandwidth { };
518 #endif /* CONFIG_CGROUP_SCHED */
520 /* CFS-related fields in a runqueue */
522 struct load_weight load;
523 unsigned int nr_running;
524 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
525 unsigned int idle_h_nr_running; /* SCHED_IDLE */
530 u64 min_vruntime_copy;
533 struct rb_root_cached tasks_timeline;
536 * 'curr' points to currently running entity on this cfs_rq.
537 * It is set to NULL otherwise (i.e when none are currently running).
539 struct sched_entity *curr;
540 struct sched_entity *next;
541 struct sched_entity *last;
542 struct sched_entity *skip;
544 #ifdef CONFIG_SCHED_DEBUG
545 unsigned int nr_spread_over;
552 struct sched_avg avg;
554 u64 load_last_update_time_copy;
557 raw_spinlock_t lock ____cacheline_aligned;
559 unsigned long load_avg;
560 unsigned long util_avg;
561 unsigned long runnable_avg;
564 #ifdef CONFIG_FAIR_GROUP_SCHED
565 unsigned long tg_load_avg_contrib;
567 long prop_runnable_sum;
570 * h_load = weight * f(tg)
572 * Where f(tg) is the recursive weight fraction assigned to
575 unsigned long h_load;
576 u64 last_h_load_update;
577 struct sched_entity *h_load_next;
578 #endif /* CONFIG_FAIR_GROUP_SCHED */
579 #endif /* CONFIG_SMP */
581 #ifdef CONFIG_FAIR_GROUP_SCHED
582 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
585 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
586 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
587 * (like users, containers etc.)
589 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
590 * This list is used during load balance.
593 struct list_head leaf_cfs_rq_list;
594 struct task_group *tg; /* group that "owns" this runqueue */
596 #ifdef CONFIG_CFS_BANDWIDTH
598 s64 runtime_remaining;
601 u64 throttled_clock_pelt;
602 u64 throttled_clock_pelt_time;
605 struct list_head throttled_list;
606 #endif /* CONFIG_CFS_BANDWIDTH */
607 #endif /* CONFIG_FAIR_GROUP_SCHED */
610 static inline int rt_bandwidth_enabled(void)
612 return sysctl_sched_rt_runtime >= 0;
615 /* RT IPI pull logic requires IRQ_WORK */
616 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
617 # define HAVE_RT_PUSH_IPI
620 /* Real-Time classes' related field in a runqueue: */
622 struct rt_prio_array active;
623 unsigned int rt_nr_running;
624 unsigned int rr_nr_running;
625 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
627 int curr; /* highest queued rt task prio */
629 int next; /* next highest */
634 unsigned long rt_nr_migratory;
635 unsigned long rt_nr_total;
637 struct plist_head pushable_tasks;
639 #endif /* CONFIG_SMP */
645 /* Nests inside the rq lock: */
646 raw_spinlock_t rt_runtime_lock;
648 #ifdef CONFIG_RT_GROUP_SCHED
649 unsigned long rt_nr_boosted;
652 struct task_group *tg;
656 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
658 return rt_rq->rt_queued && rt_rq->rt_nr_running;
661 /* Deadline class' related fields in a runqueue */
663 /* runqueue is an rbtree, ordered by deadline */
664 struct rb_root_cached root;
666 unsigned long dl_nr_running;
670 * Deadline values of the currently executing and the
671 * earliest ready task on this rq. Caching these facilitates
672 * the decision whether or not a ready but not running task
673 * should migrate somewhere else.
680 unsigned long dl_nr_migratory;
684 * Tasks on this rq that can be pushed away. They are kept in
685 * an rb-tree, ordered by tasks' deadlines, with caching
686 * of the leftmost (earliest deadline) element.
688 struct rb_root_cached pushable_dl_tasks_root;
693 * "Active utilization" for this runqueue: increased when a
694 * task wakes up (becomes TASK_RUNNING) and decreased when a
700 * Utilization of the tasks "assigned" to this runqueue (including
701 * the tasks that are in runqueue and the tasks that executed on this
702 * CPU and blocked). Increased when a task moves to this runqueue, and
703 * decreased when the task moves away (migrates, changes scheduling
704 * policy, or terminates).
705 * This is needed to compute the "inactive utilization" for the
706 * runqueue (inactive utilization = this_bw - running_bw).
712 * Inverse of the fraction of CPU utilization that can be reclaimed
713 * by the GRUB algorithm.
718 #ifdef CONFIG_FAIR_GROUP_SCHED
719 /* An entity is a task if it doesn't "own" a runqueue */
720 #define entity_is_task(se) (!se->my_q)
722 static inline void se_update_runnable(struct sched_entity *se)
724 if (!entity_is_task(se))
725 se->runnable_weight = se->my_q->h_nr_running;
728 static inline long se_runnable(struct sched_entity *se)
730 if (entity_is_task(se))
733 return se->runnable_weight;
737 #define entity_is_task(se) 1
739 static inline void se_update_runnable(struct sched_entity *se) {}
741 static inline long se_runnable(struct sched_entity *se)
749 * XXX we want to get rid of these helpers and use the full load resolution.
751 static inline long se_weight(struct sched_entity *se)
753 return scale_load_down(se->load.weight);
757 static inline bool sched_asym_prefer(int a, int b)
759 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
763 struct em_perf_domain *em_pd;
764 struct perf_domain *next;
768 /* Scheduling group status flags */
769 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
770 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
773 * We add the notion of a root-domain which will be used to define per-domain
774 * variables. Each exclusive cpuset essentially defines an island domain by
775 * fully partitioning the member CPUs from any other cpuset. Whenever a new
776 * exclusive cpuset is created, we also create and attach a new root-domain
785 cpumask_var_t online;
788 * Indicate pullable load on at least one CPU, e.g:
789 * - More than one runnable task
790 * - Running task is misfit
794 /* Indicate one or more cpus over-utilized (tipping point) */
798 * The bit corresponding to a CPU gets set here if such CPU has more
799 * than one runnable -deadline task (as it is below for RT tasks).
801 cpumask_var_t dlo_mask;
806 #ifdef HAVE_RT_PUSH_IPI
808 * For IPI pull requests, loop across the rto_mask.
810 struct irq_work rto_push_work;
811 raw_spinlock_t rto_lock;
812 /* These are only updated and read within rto_lock */
815 /* These atomics are updated outside of a lock */
816 atomic_t rto_loop_next;
817 atomic_t rto_loop_start;
820 * The "RT overload" flag: it gets set if a CPU has more than
821 * one runnable RT task.
823 cpumask_var_t rto_mask;
824 struct cpupri cpupri;
826 unsigned long max_cpu_capacity;
829 * NULL-terminated list of performance domains intersecting with the
830 * CPUs of the rd. Protected by RCU.
832 struct perf_domain __rcu *pd;
835 extern void init_defrootdomain(void);
836 extern int sched_init_domains(const struct cpumask *cpu_map);
837 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
838 extern void sched_get_rd(struct root_domain *rd);
839 extern void sched_put_rd(struct root_domain *rd);
841 #ifdef HAVE_RT_PUSH_IPI
842 extern void rto_push_irq_work_func(struct irq_work *work);
844 #endif /* CONFIG_SMP */
846 #ifdef CONFIG_UCLAMP_TASK
848 * struct uclamp_bucket - Utilization clamp bucket
849 * @value: utilization clamp value for tasks on this clamp bucket
850 * @tasks: number of RUNNABLE tasks on this clamp bucket
852 * Keep track of how many tasks are RUNNABLE for a given utilization
855 struct uclamp_bucket {
856 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
857 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
861 * struct uclamp_rq - rq's utilization clamp
862 * @value: currently active clamp values for a rq
863 * @bucket: utilization clamp buckets affecting a rq
865 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
866 * A clamp value is affecting a rq when there is at least one task RUNNABLE
867 * (or actually running) with that value.
869 * There are up to UCLAMP_CNT possible different clamp values, currently there
870 * are only two: minimum utilization and maximum utilization.
872 * All utilization clamping values are MAX aggregated, since:
873 * - for util_min: we want to run the CPU at least at the max of the minimum
874 * utilization required by its currently RUNNABLE tasks.
875 * - for util_max: we want to allow the CPU to run up to the max of the
876 * maximum utilization allowed by its currently RUNNABLE tasks.
878 * Since on each system we expect only a limited number of different
879 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
880 * the metrics required to compute all the per-rq utilization clamp values.
884 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
887 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
888 #endif /* CONFIG_UCLAMP_TASK */
891 * This is the main, per-CPU runqueue data structure.
893 * Locking rule: those places that want to lock multiple runqueues
894 * (such as the load balancing or the thread migration code), lock
895 * acquire operations must be ordered by ascending &runqueue.
902 * nr_running and cpu_load should be in the same cacheline because
903 * remote CPUs use both these fields when doing load calculation.
905 unsigned int nr_running;
906 #ifdef CONFIG_NUMA_BALANCING
907 unsigned int nr_numa_running;
908 unsigned int nr_preferred_running;
909 unsigned int numa_migrate_on;
911 #ifdef CONFIG_NO_HZ_COMMON
913 unsigned long last_blocked_load_update_tick;
914 unsigned int has_blocked_load;
915 call_single_data_t nohz_csd;
916 #endif /* CONFIG_SMP */
917 unsigned int nohz_tick_stopped;
919 #endif /* CONFIG_NO_HZ_COMMON */
922 unsigned int ttwu_pending;
926 #ifdef CONFIG_UCLAMP_TASK
927 /* Utilization clamp values based on CPU's RUNNABLE tasks */
928 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
929 unsigned int uclamp_flags;
930 #define UCLAMP_FLAG_IDLE 0x01
937 #ifdef CONFIG_FAIR_GROUP_SCHED
938 /* list of leaf cfs_rq on this CPU: */
939 struct list_head leaf_cfs_rq_list;
940 struct list_head *tmp_alone_branch;
941 #endif /* CONFIG_FAIR_GROUP_SCHED */
944 * This is part of a global counter where only the total sum
945 * over all CPUs matters. A task can increase this counter on
946 * one CPU and if it got migrated afterwards it may decrease
947 * it on another CPU. Always updated under the runqueue lock:
949 unsigned long nr_uninterruptible;
951 struct task_struct __rcu *curr;
952 struct task_struct *idle;
953 struct task_struct *stop;
954 unsigned long next_balance;
955 struct mm_struct *prev_mm;
957 unsigned int clock_update_flags;
959 /* Ensure that all clocks are in the same cache line */
960 u64 clock_task ____cacheline_aligned;
962 unsigned long lost_idle_time;
966 #ifdef CONFIG_MEMBARRIER
967 int membarrier_state;
971 struct root_domain *rd;
972 struct sched_domain __rcu *sd;
974 unsigned long cpu_capacity;
975 unsigned long cpu_capacity_orig;
976 unsigned long cpu_capacity_inverted;
978 struct callback_head *balance_callback;
980 unsigned char nohz_idle_balance;
981 unsigned char idle_balance;
983 unsigned long misfit_task_load;
985 /* For active balancing */
988 struct cpu_stop_work active_balance_work;
990 /* CPU of this runqueue: */
994 struct list_head cfs_tasks;
996 struct sched_avg avg_rt;
997 struct sched_avg avg_dl;
998 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
999 struct sched_avg avg_irq;
1001 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1002 struct sched_avg avg_thermal;
1007 /* This is used to determine avg_idle's max value */
1008 u64 max_idle_balance_cost;
1009 #endif /* CONFIG_SMP */
1011 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1014 #ifdef CONFIG_PARAVIRT
1015 u64 prev_steal_time;
1017 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1018 u64 prev_steal_time_rq;
1021 /* calc_load related fields */
1022 unsigned long calc_load_update;
1023 long calc_load_active;
1025 #ifdef CONFIG_SCHED_HRTICK
1027 call_single_data_t hrtick_csd;
1029 struct hrtimer hrtick_timer;
1030 ktime_t hrtick_time;
1033 #ifdef CONFIG_SCHEDSTATS
1035 struct sched_info rq_sched_info;
1036 unsigned long long rq_cpu_time;
1037 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1039 /* sys_sched_yield() stats */
1040 unsigned int yld_count;
1042 /* schedule() stats */
1043 unsigned int sched_count;
1044 unsigned int sched_goidle;
1046 /* try_to_wake_up() stats */
1047 unsigned int ttwu_count;
1048 unsigned int ttwu_local;
1051 #ifdef CONFIG_CPU_IDLE
1052 /* Must be inspected within a rcu lock section */
1053 struct cpuidle_state *idle_state;
1057 #ifdef CONFIG_FAIR_GROUP_SCHED
1059 /* CPU runqueue to which this cfs_rq is attached */
1060 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1067 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1069 return container_of(cfs_rq, struct rq, cfs);
1073 static inline int cpu_of(struct rq *rq)
1083 #ifdef CONFIG_SCHED_SMT
1084 extern void __update_idle_core(struct rq *rq);
1086 static inline void update_idle_core(struct rq *rq)
1088 if (static_branch_unlikely(&sched_smt_present))
1089 __update_idle_core(rq);
1093 static inline void update_idle_core(struct rq *rq) { }
1096 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1098 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1099 #define this_rq() this_cpu_ptr(&runqueues)
1100 #define task_rq(p) cpu_rq(task_cpu(p))
1101 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1102 #define raw_rq() raw_cpu_ptr(&runqueues)
1104 extern void update_rq_clock(struct rq *rq);
1106 static inline u64 __rq_clock_broken(struct rq *rq)
1108 return READ_ONCE(rq->clock);
1112 * rq::clock_update_flags bits
1114 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1115 * call to __schedule(). This is an optimisation to avoid
1116 * neighbouring rq clock updates.
1118 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1119 * in effect and calls to update_rq_clock() are being ignored.
1121 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1122 * made to update_rq_clock() since the last time rq::lock was pinned.
1124 * If inside of __schedule(), clock_update_flags will have been
1125 * shifted left (a left shift is a cheap operation for the fast path
1126 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1128 * if (rq-clock_update_flags >= RQCF_UPDATED)
1130 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1131 * one position though, because the next rq_unpin_lock() will shift it
1134 #define RQCF_REQ_SKIP 0x01
1135 #define RQCF_ACT_SKIP 0x02
1136 #define RQCF_UPDATED 0x04
1138 static inline void assert_clock_updated(struct rq *rq)
1141 * The only reason for not seeing a clock update since the
1142 * last rq_pin_lock() is if we're currently skipping updates.
1144 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1147 static inline u64 rq_clock(struct rq *rq)
1149 lockdep_assert_held(&rq->lock);
1150 assert_clock_updated(rq);
1155 static inline u64 rq_clock_task(struct rq *rq)
1157 lockdep_assert_held(&rq->lock);
1158 assert_clock_updated(rq);
1160 return rq->clock_task;
1164 * By default the decay is the default pelt decay period.
1165 * The decay shift can change the decay period in
1167 * Decay shift Decay period(ms)
1174 extern int sched_thermal_decay_shift;
1176 static inline u64 rq_clock_thermal(struct rq *rq)
1178 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1181 static inline void rq_clock_skip_update(struct rq *rq)
1183 lockdep_assert_held(&rq->lock);
1184 rq->clock_update_flags |= RQCF_REQ_SKIP;
1188 * See rt task throttling, which is the only time a skip
1189 * request is cancelled.
1191 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1193 lockdep_assert_held(&rq->lock);
1194 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1198 unsigned long flags;
1199 struct pin_cookie cookie;
1200 #ifdef CONFIG_SCHED_DEBUG
1202 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1203 * current pin context is stashed here in case it needs to be
1204 * restored in rq_repin_lock().
1206 unsigned int clock_update_flags;
1211 * Lockdep annotation that avoids accidental unlocks; it's like a
1212 * sticky/continuous lockdep_assert_held().
1214 * This avoids code that has access to 'struct rq *rq' (basically everything in
1215 * the scheduler) from accidentally unlocking the rq if they do not also have a
1216 * copy of the (on-stack) 'struct rq_flags rf'.
1218 * Also see Documentation/locking/lockdep-design.rst.
1220 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1222 rf->cookie = lockdep_pin_lock(&rq->lock);
1224 #ifdef CONFIG_SCHED_DEBUG
1225 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1226 rf->clock_update_flags = 0;
1230 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1232 #ifdef CONFIG_SCHED_DEBUG
1233 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1234 rf->clock_update_flags = RQCF_UPDATED;
1237 lockdep_unpin_lock(&rq->lock, rf->cookie);
1240 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1242 lockdep_repin_lock(&rq->lock, rf->cookie);
1244 #ifdef CONFIG_SCHED_DEBUG
1246 * Restore the value we stashed in @rf for this pin context.
1248 rq->clock_update_flags |= rf->clock_update_flags;
1252 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1253 __acquires(rq->lock);
1255 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1256 __acquires(p->pi_lock)
1257 __acquires(rq->lock);
1259 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1260 __releases(rq->lock)
1262 rq_unpin_lock(rq, rf);
1263 raw_spin_unlock(&rq->lock);
1267 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1268 __releases(rq->lock)
1269 __releases(p->pi_lock)
1271 rq_unpin_lock(rq, rf);
1272 raw_spin_unlock(&rq->lock);
1273 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1277 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1278 __acquires(rq->lock)
1280 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1281 rq_pin_lock(rq, rf);
1285 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1286 __acquires(rq->lock)
1288 raw_spin_lock_irq(&rq->lock);
1289 rq_pin_lock(rq, rf);
1293 rq_lock(struct rq *rq, struct rq_flags *rf)
1294 __acquires(rq->lock)
1296 raw_spin_lock(&rq->lock);
1297 rq_pin_lock(rq, rf);
1301 rq_relock(struct rq *rq, struct rq_flags *rf)
1302 __acquires(rq->lock)
1304 raw_spin_lock(&rq->lock);
1305 rq_repin_lock(rq, rf);
1309 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1310 __releases(rq->lock)
1312 rq_unpin_lock(rq, rf);
1313 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1317 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1318 __releases(rq->lock)
1320 rq_unpin_lock(rq, rf);
1321 raw_spin_unlock_irq(&rq->lock);
1325 rq_unlock(struct rq *rq, struct rq_flags *rf)
1326 __releases(rq->lock)
1328 rq_unpin_lock(rq, rf);
1329 raw_spin_unlock(&rq->lock);
1332 static inline struct rq *
1333 this_rq_lock_irq(struct rq_flags *rf)
1334 __acquires(rq->lock)
1338 local_irq_disable();
1345 enum numa_topology_type {
1350 extern enum numa_topology_type sched_numa_topology_type;
1351 extern int sched_max_numa_distance;
1352 extern bool find_numa_distance(int distance);
1353 extern void sched_init_numa(void);
1354 extern void sched_domains_numa_masks_set(unsigned int cpu);
1355 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1356 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1358 static inline void sched_init_numa(void) { }
1359 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1360 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1361 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1367 #ifdef CONFIG_NUMA_BALANCING
1368 /* The regions in numa_faults array from task_struct */
1369 enum numa_faults_stats {
1375 extern void sched_setnuma(struct task_struct *p, int node);
1376 extern int migrate_task_to(struct task_struct *p, int cpu);
1377 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1379 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1382 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1385 #endif /* CONFIG_NUMA_BALANCING */
1390 queue_balance_callback(struct rq *rq,
1391 struct callback_head *head,
1392 void (*func)(struct rq *rq))
1394 lockdep_assert_held(&rq->lock);
1396 if (unlikely(head->next))
1399 head->func = (void (*)(struct callback_head *))func;
1400 head->next = rq->balance_callback;
1401 rq->balance_callback = head;
1404 #define rcu_dereference_check_sched_domain(p) \
1405 rcu_dereference_check((p), \
1406 lockdep_is_held(&sched_domains_mutex))
1409 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1410 * See destroy_sched_domains: call_rcu for details.
1412 * The domain tree of any CPU may only be accessed from within
1413 * preempt-disabled sections.
1415 #define for_each_domain(cpu, __sd) \
1416 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1417 __sd; __sd = __sd->parent)
1420 * highest_flag_domain - Return highest sched_domain containing flag.
1421 * @cpu: The CPU whose highest level of sched domain is to
1423 * @flag: The flag to check for the highest sched_domain
1424 * for the given CPU.
1426 * Returns the highest sched_domain of a CPU which contains the given flag.
1428 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1430 struct sched_domain *sd, *hsd = NULL;
1432 for_each_domain(cpu, sd) {
1433 if (!(sd->flags & flag))
1441 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1443 struct sched_domain *sd;
1445 for_each_domain(cpu, sd) {
1446 if (sd->flags & flag)
1453 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1454 DECLARE_PER_CPU(int, sd_llc_size);
1455 DECLARE_PER_CPU(int, sd_llc_id);
1456 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1457 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1458 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1459 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1460 extern struct static_key_false sched_asym_cpucapacity;
1462 struct sched_group_capacity {
1465 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1468 unsigned long capacity;
1469 unsigned long min_capacity; /* Min per-CPU capacity in group */
1470 unsigned long max_capacity; /* Max per-CPU capacity in group */
1471 unsigned long next_update;
1472 int imbalance; /* XXX unrelated to capacity but shared group state */
1474 #ifdef CONFIG_SCHED_DEBUG
1478 unsigned long cpumask[]; /* Balance mask */
1481 struct sched_group {
1482 struct sched_group *next; /* Must be a circular list */
1485 unsigned int group_weight;
1486 struct sched_group_capacity *sgc;
1487 int asym_prefer_cpu; /* CPU of highest priority in group */
1490 * The CPUs this group covers.
1492 * NOTE: this field is variable length. (Allocated dynamically
1493 * by attaching extra space to the end of the structure,
1494 * depending on how many CPUs the kernel has booted up with)
1496 unsigned long cpumask[];
1499 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1501 return to_cpumask(sg->cpumask);
1505 * See build_balance_mask().
1507 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1509 return to_cpumask(sg->sgc->cpumask);
1513 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1514 * @group: The group whose first CPU is to be returned.
1516 static inline unsigned int group_first_cpu(struct sched_group *group)
1518 return cpumask_first(sched_group_span(group));
1521 extern int group_balance_cpu(struct sched_group *sg);
1523 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1524 void register_sched_domain_sysctl(void);
1525 void dirty_sched_domain_sysctl(int cpu);
1526 void unregister_sched_domain_sysctl(void);
1528 static inline void register_sched_domain_sysctl(void)
1531 static inline void dirty_sched_domain_sysctl(int cpu)
1534 static inline void unregister_sched_domain_sysctl(void)
1539 extern void flush_smp_call_function_from_idle(void);
1541 #else /* !CONFIG_SMP: */
1542 static inline void flush_smp_call_function_from_idle(void) { }
1546 #include "autogroup.h"
1548 #ifdef CONFIG_CGROUP_SCHED
1551 * Return the group to which this tasks belongs.
1553 * We cannot use task_css() and friends because the cgroup subsystem
1554 * changes that value before the cgroup_subsys::attach() method is called,
1555 * therefore we cannot pin it and might observe the wrong value.
1557 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1558 * core changes this before calling sched_move_task().
1560 * Instead we use a 'copy' which is updated from sched_move_task() while
1561 * holding both task_struct::pi_lock and rq::lock.
1563 static inline struct task_group *task_group(struct task_struct *p)
1565 return p->sched_task_group;
1568 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1569 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1571 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1572 struct task_group *tg = task_group(p);
1575 #ifdef CONFIG_FAIR_GROUP_SCHED
1576 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1577 p->se.cfs_rq = tg->cfs_rq[cpu];
1578 p->se.parent = tg->se[cpu];
1581 #ifdef CONFIG_RT_GROUP_SCHED
1582 p->rt.rt_rq = tg->rt_rq[cpu];
1583 p->rt.parent = tg->rt_se[cpu];
1587 #else /* CONFIG_CGROUP_SCHED */
1589 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1590 static inline struct task_group *task_group(struct task_struct *p)
1595 #endif /* CONFIG_CGROUP_SCHED */
1597 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1599 set_task_rq(p, cpu);
1602 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1603 * successfully executed on another CPU. We must ensure that updates of
1604 * per-task data have been completed by this moment.
1607 #ifdef CONFIG_THREAD_INFO_IN_TASK
1608 WRITE_ONCE(p->cpu, cpu);
1610 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1617 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1619 #ifdef CONFIG_SCHED_DEBUG
1620 # include <linux/static_key.h>
1621 # define const_debug __read_mostly
1623 # define const_debug const
1626 #define SCHED_FEAT(name, enabled) \
1627 __SCHED_FEAT_##name ,
1630 #include "features.h"
1636 #ifdef CONFIG_SCHED_DEBUG
1639 * To support run-time toggling of sched features, all the translation units
1640 * (but core.c) reference the sysctl_sched_features defined in core.c.
1642 extern const_debug unsigned int sysctl_sched_features;
1644 #ifdef CONFIG_JUMP_LABEL
1645 #define SCHED_FEAT(name, enabled) \
1646 static __always_inline bool static_branch_##name(struct static_key *key) \
1648 return static_key_##enabled(key); \
1651 #include "features.h"
1654 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1655 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1657 #else /* !CONFIG_JUMP_LABEL */
1659 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1661 #endif /* CONFIG_JUMP_LABEL */
1663 #else /* !SCHED_DEBUG */
1666 * Each translation unit has its own copy of sysctl_sched_features to allow
1667 * constants propagation at compile time and compiler optimization based on
1670 #define SCHED_FEAT(name, enabled) \
1671 (1UL << __SCHED_FEAT_##name) * enabled |
1672 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1673 #include "features.h"
1677 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1679 #endif /* SCHED_DEBUG */
1681 extern struct static_key_false sched_numa_balancing;
1682 extern struct static_key_false sched_schedstats;
1684 static inline u64 global_rt_period(void)
1686 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1689 static inline u64 global_rt_runtime(void)
1691 if (sysctl_sched_rt_runtime < 0)
1694 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1697 static inline int task_current(struct rq *rq, struct task_struct *p)
1699 return rq->curr == p;
1702 static inline int task_running(struct rq *rq, struct task_struct *p)
1707 return task_current(rq, p);
1711 static inline int task_on_rq_queued(struct task_struct *p)
1713 return p->on_rq == TASK_ON_RQ_QUEUED;
1716 static inline int task_on_rq_migrating(struct task_struct *p)
1718 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1724 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1725 #define WF_FORK 0x02 /* Child wakeup after fork */
1726 #define WF_MIGRATED 0x04 /* Internal use, task got migrated */
1727 #define WF_ON_CPU 0x08 /* Wakee is on_cpu */
1730 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1731 * of tasks with abnormal "nice" values across CPUs the contribution that
1732 * each task makes to its run queue's load is weighted according to its
1733 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1734 * scaled version of the new time slice allocation that they receive on time
1738 #define WEIGHT_IDLEPRIO 3
1739 #define WMULT_IDLEPRIO 1431655765
1741 extern const int sched_prio_to_weight[40];
1742 extern const u32 sched_prio_to_wmult[40];
1745 * {de,en}queue flags:
1747 * DEQUEUE_SLEEP - task is no longer runnable
1748 * ENQUEUE_WAKEUP - task just became runnable
1750 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1751 * are in a known state which allows modification. Such pairs
1752 * should preserve as much state as possible.
1754 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1757 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1758 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1759 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1763 #define DEQUEUE_SLEEP 0x01
1764 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1765 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1766 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1768 #define ENQUEUE_WAKEUP 0x01
1769 #define ENQUEUE_RESTORE 0x02
1770 #define ENQUEUE_MOVE 0x04
1771 #define ENQUEUE_NOCLOCK 0x08
1773 #define ENQUEUE_HEAD 0x10
1774 #define ENQUEUE_REPLENISH 0x20
1776 #define ENQUEUE_MIGRATED 0x40
1778 #define ENQUEUE_MIGRATED 0x00
1781 #define RETRY_TASK ((void *)-1UL)
1783 struct sched_class {
1785 #ifdef CONFIG_UCLAMP_TASK
1789 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1790 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1791 void (*yield_task) (struct rq *rq);
1792 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1794 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1796 struct task_struct *(*pick_next_task)(struct rq *rq);
1798 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1799 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1802 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1803 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1804 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1806 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1808 void (*set_cpus_allowed)(struct task_struct *p,
1809 const struct cpumask *newmask);
1811 void (*rq_online)(struct rq *rq);
1812 void (*rq_offline)(struct rq *rq);
1815 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1816 void (*task_fork)(struct task_struct *p);
1817 void (*task_dead)(struct task_struct *p);
1820 * The switched_from() call is allowed to drop rq->lock, therefore we
1821 * cannot assume the switched_from/switched_to pair is serliazed by
1822 * rq->lock. They are however serialized by p->pi_lock.
1824 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1825 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1826 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1829 unsigned int (*get_rr_interval)(struct rq *rq,
1830 struct task_struct *task);
1832 void (*update_curr)(struct rq *rq);
1834 #define TASK_SET_GROUP 0
1835 #define TASK_MOVE_GROUP 1
1837 #ifdef CONFIG_FAIR_GROUP_SCHED
1838 void (*task_change_group)(struct task_struct *p, int type);
1840 } __aligned(STRUCT_ALIGNMENT); /* STRUCT_ALIGN(), vmlinux.lds.h */
1842 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1844 WARN_ON_ONCE(rq->curr != prev);
1845 prev->sched_class->put_prev_task(rq, prev);
1848 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1850 WARN_ON_ONCE(rq->curr != next);
1851 next->sched_class->set_next_task(rq, next, false);
1854 /* Defined in include/asm-generic/vmlinux.lds.h */
1855 extern struct sched_class __begin_sched_classes[];
1856 extern struct sched_class __end_sched_classes[];
1858 #define sched_class_highest (__end_sched_classes - 1)
1859 #define sched_class_lowest (__begin_sched_classes - 1)
1861 #define for_class_range(class, _from, _to) \
1862 for (class = (_from); class != (_to); class--)
1864 #define for_each_class(class) \
1865 for_class_range(class, sched_class_highest, sched_class_lowest)
1867 extern const struct sched_class stop_sched_class;
1868 extern const struct sched_class dl_sched_class;
1869 extern const struct sched_class rt_sched_class;
1870 extern const struct sched_class fair_sched_class;
1871 extern const struct sched_class idle_sched_class;
1873 static inline bool sched_stop_runnable(struct rq *rq)
1875 return rq->stop && task_on_rq_queued(rq->stop);
1878 static inline bool sched_dl_runnable(struct rq *rq)
1880 return rq->dl.dl_nr_running > 0;
1883 static inline bool sched_rt_runnable(struct rq *rq)
1885 return rq->rt.rt_queued > 0;
1888 static inline bool sched_fair_runnable(struct rq *rq)
1890 return rq->cfs.nr_running > 0;
1893 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1894 extern struct task_struct *pick_next_task_idle(struct rq *rq);
1898 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1900 extern void trigger_load_balance(struct rq *rq);
1902 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1906 #ifdef CONFIG_CPU_IDLE
1907 static inline void idle_set_state(struct rq *rq,
1908 struct cpuidle_state *idle_state)
1910 rq->idle_state = idle_state;
1913 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1915 SCHED_WARN_ON(!rcu_read_lock_held());
1917 return rq->idle_state;
1920 static inline void idle_set_state(struct rq *rq,
1921 struct cpuidle_state *idle_state)
1925 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1931 extern void schedule_idle(void);
1933 extern void sysrq_sched_debug_show(void);
1934 extern void sched_init_granularity(void);
1935 extern void update_max_interval(void);
1937 extern void init_sched_dl_class(void);
1938 extern void init_sched_rt_class(void);
1939 extern void init_sched_fair_class(void);
1941 extern void reweight_task(struct task_struct *p, int prio);
1943 extern void resched_curr(struct rq *rq);
1944 extern void resched_cpu(int cpu);
1946 extern struct rt_bandwidth def_rt_bandwidth;
1947 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1949 extern struct dl_bandwidth def_dl_bandwidth;
1950 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1951 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1952 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1955 #define BW_UNIT (1 << BW_SHIFT)
1956 #define RATIO_SHIFT 8
1957 #define MAX_BW_BITS (64 - BW_SHIFT)
1958 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
1959 unsigned long to_ratio(u64 period, u64 runtime);
1961 extern void init_entity_runnable_average(struct sched_entity *se);
1962 extern void post_init_entity_util_avg(struct task_struct *p);
1964 #ifdef CONFIG_NO_HZ_FULL
1965 extern bool sched_can_stop_tick(struct rq *rq);
1966 extern int __init sched_tick_offload_init(void);
1969 * Tick may be needed by tasks in the runqueue depending on their policy and
1970 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1971 * nohz mode if necessary.
1973 static inline void sched_update_tick_dependency(struct rq *rq)
1975 int cpu = cpu_of(rq);
1977 if (!tick_nohz_full_cpu(cpu))
1980 if (sched_can_stop_tick(rq))
1981 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1983 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1986 static inline int sched_tick_offload_init(void) { return 0; }
1987 static inline void sched_update_tick_dependency(struct rq *rq) { }
1990 static inline void add_nr_running(struct rq *rq, unsigned count)
1992 unsigned prev_nr = rq->nr_running;
1994 rq->nr_running = prev_nr + count;
1995 if (trace_sched_update_nr_running_tp_enabled()) {
1996 call_trace_sched_update_nr_running(rq, count);
2000 if (prev_nr < 2 && rq->nr_running >= 2) {
2001 if (!READ_ONCE(rq->rd->overload))
2002 WRITE_ONCE(rq->rd->overload, 1);
2006 sched_update_tick_dependency(rq);
2009 static inline void sub_nr_running(struct rq *rq, unsigned count)
2011 rq->nr_running -= count;
2012 if (trace_sched_update_nr_running_tp_enabled()) {
2013 call_trace_sched_update_nr_running(rq, -count);
2016 /* Check if we still need preemption */
2017 sched_update_tick_dependency(rq);
2020 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2021 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2023 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2025 extern const_debug unsigned int sysctl_sched_nr_migrate;
2026 extern const_debug unsigned int sysctl_sched_migration_cost;
2028 #ifdef CONFIG_SCHED_HRTICK
2032 * - enabled by features
2033 * - hrtimer is actually high res
2035 static inline int hrtick_enabled(struct rq *rq)
2037 if (!sched_feat(HRTICK))
2039 if (!cpu_active(cpu_of(rq)))
2041 return hrtimer_is_hres_active(&rq->hrtick_timer);
2044 void hrtick_start(struct rq *rq, u64 delay);
2048 static inline int hrtick_enabled(struct rq *rq)
2053 #endif /* CONFIG_SCHED_HRTICK */
2055 #ifndef arch_scale_freq_tick
2056 static __always_inline
2057 void arch_scale_freq_tick(void)
2062 #ifndef arch_scale_freq_capacity
2064 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2065 * @cpu: the CPU in question.
2067 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2070 * ------ * SCHED_CAPACITY_SCALE
2073 static __always_inline
2074 unsigned long arch_scale_freq_capacity(int cpu)
2076 return SCHED_CAPACITY_SCALE;
2081 #ifdef CONFIG_PREEMPTION
2083 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2086 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2087 * way at the expense of forcing extra atomic operations in all
2088 * invocations. This assures that the double_lock is acquired using the
2089 * same underlying policy as the spinlock_t on this architecture, which
2090 * reduces latency compared to the unfair variant below. However, it
2091 * also adds more overhead and therefore may reduce throughput.
2093 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2094 __releases(this_rq->lock)
2095 __acquires(busiest->lock)
2096 __acquires(this_rq->lock)
2098 raw_spin_unlock(&this_rq->lock);
2099 double_rq_lock(this_rq, busiest);
2106 * Unfair double_lock_balance: Optimizes throughput at the expense of
2107 * latency by eliminating extra atomic operations when the locks are
2108 * already in proper order on entry. This favors lower CPU-ids and will
2109 * grant the double lock to lower CPUs over higher ids under contention,
2110 * regardless of entry order into the function.
2112 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2113 __releases(this_rq->lock)
2114 __acquires(busiest->lock)
2115 __acquires(this_rq->lock)
2119 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2120 if (busiest < this_rq) {
2121 raw_spin_unlock(&this_rq->lock);
2122 raw_spin_lock(&busiest->lock);
2123 raw_spin_lock_nested(&this_rq->lock,
2124 SINGLE_DEPTH_NESTING);
2127 raw_spin_lock_nested(&busiest->lock,
2128 SINGLE_DEPTH_NESTING);
2133 #endif /* CONFIG_PREEMPTION */
2136 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2138 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2140 if (unlikely(!irqs_disabled())) {
2141 /* printk() doesn't work well under rq->lock */
2142 raw_spin_unlock(&this_rq->lock);
2146 return _double_lock_balance(this_rq, busiest);
2149 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2150 __releases(busiest->lock)
2152 raw_spin_unlock(&busiest->lock);
2153 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2156 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2162 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2165 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2171 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2174 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2180 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2184 * double_rq_lock - safely lock two runqueues
2186 * Note this does not disable interrupts like task_rq_lock,
2187 * you need to do so manually before calling.
2189 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2190 __acquires(rq1->lock)
2191 __acquires(rq2->lock)
2193 BUG_ON(!irqs_disabled());
2195 raw_spin_lock(&rq1->lock);
2196 __acquire(rq2->lock); /* Fake it out ;) */
2199 raw_spin_lock(&rq1->lock);
2200 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2202 raw_spin_lock(&rq2->lock);
2203 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2209 * double_rq_unlock - safely unlock two runqueues
2211 * Note this does not restore interrupts like task_rq_unlock,
2212 * you need to do so manually after calling.
2214 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2215 __releases(rq1->lock)
2216 __releases(rq2->lock)
2218 raw_spin_unlock(&rq1->lock);
2220 raw_spin_unlock(&rq2->lock);
2222 __release(rq2->lock);
2225 extern void set_rq_online (struct rq *rq);
2226 extern void set_rq_offline(struct rq *rq);
2227 extern bool sched_smp_initialized;
2229 #else /* CONFIG_SMP */
2232 * double_rq_lock - safely lock two runqueues
2234 * Note this does not disable interrupts like task_rq_lock,
2235 * you need to do so manually before calling.
2237 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2238 __acquires(rq1->lock)
2239 __acquires(rq2->lock)
2241 BUG_ON(!irqs_disabled());
2243 raw_spin_lock(&rq1->lock);
2244 __acquire(rq2->lock); /* Fake it out ;) */
2248 * double_rq_unlock - safely unlock two runqueues
2250 * Note this does not restore interrupts like task_rq_unlock,
2251 * you need to do so manually after calling.
2253 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2254 __releases(rq1->lock)
2255 __releases(rq2->lock)
2258 raw_spin_unlock(&rq1->lock);
2259 __release(rq2->lock);
2264 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2265 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2267 #ifdef CONFIG_SCHED_DEBUG
2268 extern bool sched_debug_enabled;
2270 extern void print_cfs_stats(struct seq_file *m, int cpu);
2271 extern void print_rt_stats(struct seq_file *m, int cpu);
2272 extern void print_dl_stats(struct seq_file *m, int cpu);
2273 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2274 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2275 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2276 #ifdef CONFIG_NUMA_BALANCING
2278 show_numa_stats(struct task_struct *p, struct seq_file *m);
2280 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2281 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2282 #endif /* CONFIG_NUMA_BALANCING */
2283 #endif /* CONFIG_SCHED_DEBUG */
2285 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2286 extern void init_rt_rq(struct rt_rq *rt_rq);
2287 extern void init_dl_rq(struct dl_rq *dl_rq);
2289 extern void cfs_bandwidth_usage_inc(void);
2290 extern void cfs_bandwidth_usage_dec(void);
2292 #ifdef CONFIG_NO_HZ_COMMON
2293 #define NOHZ_BALANCE_KICK_BIT 0
2294 #define NOHZ_STATS_KICK_BIT 1
2296 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2297 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2299 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2301 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2303 extern void nohz_balance_exit_idle(struct rq *rq);
2305 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2311 void __dl_update(struct dl_bw *dl_b, s64 bw)
2313 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2316 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2317 "sched RCU must be held");
2318 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2319 struct rq *rq = cpu_rq(i);
2321 rq->dl.extra_bw += bw;
2326 void __dl_update(struct dl_bw *dl_b, s64 bw)
2328 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2335 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2340 struct u64_stats_sync sync;
2343 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2346 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2347 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2348 * and never move forward.
2350 static inline u64 irq_time_read(int cpu)
2352 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2357 seq = __u64_stats_fetch_begin(&irqtime->sync);
2358 total = irqtime->total;
2359 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2363 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2365 #ifdef CONFIG_CPU_FREQ
2366 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2369 * cpufreq_update_util - Take a note about CPU utilization changes.
2370 * @rq: Runqueue to carry out the update for.
2371 * @flags: Update reason flags.
2373 * This function is called by the scheduler on the CPU whose utilization is
2376 * It can only be called from RCU-sched read-side critical sections.
2378 * The way cpufreq is currently arranged requires it to evaluate the CPU
2379 * performance state (frequency/voltage) on a regular basis to prevent it from
2380 * being stuck in a completely inadequate performance level for too long.
2381 * That is not guaranteed to happen if the updates are only triggered from CFS
2382 * and DL, though, because they may not be coming in if only RT tasks are
2383 * active all the time (or there are RT tasks only).
2385 * As a workaround for that issue, this function is called periodically by the
2386 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2387 * but that really is a band-aid. Going forward it should be replaced with
2388 * solutions targeted more specifically at RT tasks.
2390 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2392 struct update_util_data *data;
2394 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2397 data->func(data, rq_clock(rq), flags);
2400 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2401 #endif /* CONFIG_CPU_FREQ */
2403 #ifdef CONFIG_UCLAMP_TASK
2404 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2406 static inline unsigned long uclamp_rq_get(struct rq *rq,
2407 enum uclamp_id clamp_id)
2409 return READ_ONCE(rq->uclamp[clamp_id].value);
2412 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
2415 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
2418 static inline bool uclamp_rq_is_idle(struct rq *rq)
2420 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
2424 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2425 * @rq: The rq to clamp against. Must not be NULL.
2426 * @util: The util value to clamp.
2427 * @p: The task to clamp against. Can be NULL if you want to clamp
2430 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2432 * If sched_uclamp_used static key is disabled, then just return the util
2433 * without any clamping since uclamp aggregation at the rq level in the fast
2434 * path is disabled, rendering this operation a NOP.
2436 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2437 * will return the correct effective uclamp value of the task even if the
2438 * static key is disabled.
2440 static __always_inline
2441 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2442 struct task_struct *p)
2444 unsigned long min_util = 0;
2445 unsigned long max_util = 0;
2447 if (!static_branch_likely(&sched_uclamp_used))
2451 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2452 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2455 * Ignore last runnable task's max clamp, as this task will
2456 * reset it. Similarly, no need to read the rq's min clamp.
2458 if (uclamp_rq_is_idle(rq))
2462 min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
2463 max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
2466 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2467 * RUNNABLE tasks with _different_ clamps, we can end up with an
2468 * inversion. Fix it now when the clamps are applied.
2470 if (unlikely(min_util >= max_util))
2473 return clamp(util, min_util, max_util);
2477 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2478 * by default in the fast path and only gets turned on once userspace performs
2479 * an operation that requires it.
2481 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2484 static inline bool uclamp_is_used(void)
2486 return static_branch_likely(&sched_uclamp_used);
2488 #else /* CONFIG_UCLAMP_TASK */
2489 static inline unsigned long uclamp_eff_value(struct task_struct *p,
2490 enum uclamp_id clamp_id)
2492 if (clamp_id == UCLAMP_MIN)
2495 return SCHED_CAPACITY_SCALE;
2499 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2500 struct task_struct *p)
2505 static inline bool uclamp_is_used(void)
2510 static inline unsigned long uclamp_rq_get(struct rq *rq,
2511 enum uclamp_id clamp_id)
2513 if (clamp_id == UCLAMP_MIN)
2516 return SCHED_CAPACITY_SCALE;
2519 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
2524 static inline bool uclamp_rq_is_idle(struct rq *rq)
2528 #endif /* CONFIG_UCLAMP_TASK */
2530 #ifdef arch_scale_freq_capacity
2531 # ifndef arch_scale_freq_invariant
2532 # define arch_scale_freq_invariant() true
2535 # define arch_scale_freq_invariant() false
2539 static inline unsigned long capacity_orig_of(int cpu)
2541 return cpu_rq(cpu)->cpu_capacity_orig;
2545 * Returns inverted capacity if the CPU is in capacity inversion state.
2548 * Capacity inversion detection only considers thermal impact where actual
2549 * performance points (OPPs) gets dropped.
2551 * Capacity inversion state happens when another performance domain that has
2552 * equal or lower capacity_orig_of() becomes effectively larger than the perf
2553 * domain this CPU belongs to due to thermal pressure throttling it hard.
2555 * See comment in update_cpu_capacity().
2557 static inline unsigned long cpu_in_capacity_inversion(int cpu)
2559 return cpu_rq(cpu)->cpu_capacity_inverted;
2564 * enum schedutil_type - CPU utilization type
2565 * @FREQUENCY_UTIL: Utilization used to select frequency
2566 * @ENERGY_UTIL: Utilization used during energy calculation
2568 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2569 * need to be aggregated differently depending on the usage made of them. This
2570 * enum is used within schedutil_freq_util() to differentiate the types of
2571 * utilization expected by the callers, and adjust the aggregation accordingly.
2573 enum schedutil_type {
2578 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2580 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2581 unsigned long max, enum schedutil_type type,
2582 struct task_struct *p);
2584 static inline unsigned long cpu_bw_dl(struct rq *rq)
2586 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2589 static inline unsigned long cpu_util_dl(struct rq *rq)
2591 return READ_ONCE(rq->avg_dl.util_avg);
2594 static inline unsigned long cpu_util_cfs(struct rq *rq)
2596 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2598 if (sched_feat(UTIL_EST)) {
2599 util = max_t(unsigned long, util,
2600 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2606 static inline unsigned long cpu_util_rt(struct rq *rq)
2608 return READ_ONCE(rq->avg_rt.util_avg);
2610 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2611 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2612 unsigned long max, enum schedutil_type type,
2613 struct task_struct *p)
2617 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2619 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2620 static inline unsigned long cpu_util_irq(struct rq *rq)
2622 return rq->avg_irq.util_avg;
2626 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2628 util *= (max - irq);
2635 static inline unsigned long cpu_util_irq(struct rq *rq)
2641 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2647 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2649 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2651 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2653 static inline bool sched_energy_enabled(void)
2655 return static_branch_unlikely(&sched_energy_present);
2658 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2660 #define perf_domain_span(pd) NULL
2661 static inline bool sched_energy_enabled(void) { return false; }
2663 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2665 #ifdef CONFIG_MEMBARRIER
2667 * The scheduler provides memory barriers required by membarrier between:
2668 * - prior user-space memory accesses and store to rq->membarrier_state,
2669 * - store to rq->membarrier_state and following user-space memory accesses.
2670 * In the same way it provides those guarantees around store to rq->curr.
2672 static inline void membarrier_switch_mm(struct rq *rq,
2673 struct mm_struct *prev_mm,
2674 struct mm_struct *next_mm)
2676 int membarrier_state;
2678 if (prev_mm == next_mm)
2681 membarrier_state = atomic_read(&next_mm->membarrier_state);
2682 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2685 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2688 static inline void membarrier_switch_mm(struct rq *rq,
2689 struct mm_struct *prev_mm,
2690 struct mm_struct *next_mm)
2696 static inline bool is_per_cpu_kthread(struct task_struct *p)
2698 if (!(p->flags & PF_KTHREAD))
2701 if (p->nr_cpus_allowed != 1)
2708 void swake_up_all_locked(struct swait_queue_head *q);
2709 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);