1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
76 #include "cpudeadline.h"
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED 1
89 #define TASK_ON_RQ_MIGRATING 2
91 extern __read_mostly int scheduler_running;
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
100 * Helpers for converting nanosecond timing to jiffy resolution
102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
105 * Increase resolution of nice-level calculations for 64-bit architectures.
106 * The extra resolution improves shares distribution and load balancing of
107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108 * hierarchies, especially on larger systems. This is not a user-visible change
109 * and does not change the user-interface for setting shares/weights.
111 * We increase resolution only if we have enough bits to allow this increased
112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113 * are pretty high and the returns do not justify the increased costs.
115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116 * increase coverage and consistency always enable it on 64-bit platforms.
119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) (w)
125 # define scale_load_down(w) (w)
129 * Task weight (visible to users) and its load (invisible to users) have
130 * independent resolution, but they should be well calibrated. We use
131 * scale_load() and scale_load_down(w) to convert between them. The
132 * following must be true:
134 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
137 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
140 * Single value that decides SCHED_DEADLINE internal math precision.
141 * 10 -> just above 1us
142 * 9 -> just above 0.5us
147 * Single value that denotes runtime == period, ie unlimited time.
149 #define RUNTIME_INF ((u64)~0ULL)
151 static inline int idle_policy(int policy)
153 return policy == SCHED_IDLE;
155 static inline int fair_policy(int policy)
157 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
160 static inline int rt_policy(int policy)
162 return policy == SCHED_FIFO || policy == SCHED_RR;
165 static inline int dl_policy(int policy)
167 return policy == SCHED_DEADLINE;
169 static inline bool valid_policy(int policy)
171 return idle_policy(policy) || fair_policy(policy) ||
172 rt_policy(policy) || dl_policy(policy);
175 static inline int task_has_idle_policy(struct task_struct *p)
177 return idle_policy(p->policy);
180 static inline int task_has_rt_policy(struct task_struct *p)
182 return rt_policy(p->policy);
185 static inline int task_has_dl_policy(struct task_struct *p)
187 return dl_policy(p->policy);
190 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
193 * !! For sched_setattr_nocheck() (kernel) only !!
195 * This is actually gross. :(
197 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
198 * tasks, but still be able to sleep. We need this on platforms that cannot
199 * atomically change clock frequency. Remove once fast switching will be
200 * available on such platforms.
202 * SUGOV stands for SchedUtil GOVernor.
204 #define SCHED_FLAG_SUGOV 0x10000000
206 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
208 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
209 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
216 * Tells if entity @a should preempt entity @b.
219 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
221 return dl_entity_is_special(a) ||
222 dl_time_before(a->deadline, b->deadline);
226 * This is the priority-queue data structure of the RT scheduling class:
228 struct rt_prio_array {
229 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
230 struct list_head queue[MAX_RT_PRIO];
233 struct rt_bandwidth {
234 /* nests inside the rq lock: */
235 raw_spinlock_t rt_runtime_lock;
238 struct hrtimer rt_period_timer;
239 unsigned int rt_period_active;
242 void __dl_clear_params(struct task_struct *p);
245 * To keep the bandwidth of -deadline tasks and groups under control
246 * we need some place where:
247 * - store the maximum -deadline bandwidth of the system (the group);
248 * - cache the fraction of that bandwidth that is currently allocated.
250 * This is all done in the data structure below. It is similar to the
251 * one used for RT-throttling (rt_bandwidth), with the main difference
252 * that, since here we are only interested in admission control, we
253 * do not decrease any runtime while the group "executes", neither we
254 * need a timer to replenish it.
256 * With respect to SMP, the bandwidth is given on a per-CPU basis,
258 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
259 * - dl_total_bw array contains, in the i-eth element, the currently
260 * allocated bandwidth on the i-eth CPU.
261 * Moreover, groups consume bandwidth on each CPU, while tasks only
262 * consume bandwidth on the CPU they're running on.
263 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
264 * that will be shown the next time the proc or cgroup controls will
265 * be red. It on its turn can be changed by writing on its own
268 struct dl_bandwidth {
269 raw_spinlock_t dl_runtime_lock;
274 static inline int dl_bandwidth_enabled(void)
276 return sysctl_sched_rt_runtime >= 0;
285 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
288 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
290 dl_b->total_bw -= tsk_bw;
291 __dl_update(dl_b, (s32)tsk_bw / cpus);
295 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
297 dl_b->total_bw += tsk_bw;
298 __dl_update(dl_b, -((s32)tsk_bw / cpus));
302 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
304 return dl_b->bw != -1 &&
305 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
308 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
309 extern void init_dl_bw(struct dl_bw *dl_b);
310 extern int sched_dl_global_validate(void);
311 extern void sched_dl_do_global(void);
312 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
313 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
314 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
315 extern bool __checkparam_dl(const struct sched_attr *attr);
316 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
317 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
318 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
319 extern bool dl_cpu_busy(unsigned int cpu);
321 #ifdef CONFIG_CGROUP_SCHED
323 #include <linux/cgroup.h>
324 #include <linux/psi.h>
329 extern struct list_head task_groups;
331 struct cfs_bandwidth {
332 #ifdef CONFIG_CFS_BANDWIDTH
337 s64 hierarchical_quota;
341 u8 distribute_running;
343 struct hrtimer period_timer;
344 struct hrtimer slack_timer;
345 struct list_head throttled_cfs_rq;
354 /* Task group related information */
356 struct cgroup_subsys_state css;
358 #ifdef CONFIG_FAIR_GROUP_SCHED
359 /* schedulable entities of this group on each CPU */
360 struct sched_entity **se;
361 /* runqueue "owned" by this group on each CPU */
362 struct cfs_rq **cfs_rq;
363 unsigned long shares;
367 * load_avg can be heavily contended at clock tick time, so put
368 * it in its own cacheline separated from the fields above which
369 * will also be accessed at each tick.
371 atomic_long_t load_avg ____cacheline_aligned;
375 #ifdef CONFIG_RT_GROUP_SCHED
376 struct sched_rt_entity **rt_se;
377 struct rt_rq **rt_rq;
379 struct rt_bandwidth rt_bandwidth;
383 struct list_head list;
385 struct task_group *parent;
386 struct list_head siblings;
387 struct list_head children;
389 #ifdef CONFIG_SCHED_AUTOGROUP
390 struct autogroup *autogroup;
393 struct cfs_bandwidth cfs_bandwidth;
396 #ifdef CONFIG_FAIR_GROUP_SCHED
397 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
400 * A weight of 0 or 1 can cause arithmetics problems.
401 * A weight of a cfs_rq is the sum of weights of which entities
402 * are queued on this cfs_rq, so a weight of a entity should not be
403 * too large, so as the shares value of a task group.
404 * (The default weight is 1024 - so there's no practical
405 * limitation from this.)
407 #define MIN_SHARES (1UL << 1)
408 #define MAX_SHARES (1UL << 18)
411 typedef int (*tg_visitor)(struct task_group *, void *);
413 extern int walk_tg_tree_from(struct task_group *from,
414 tg_visitor down, tg_visitor up, void *data);
417 * Iterate the full tree, calling @down when first entering a node and @up when
418 * leaving it for the final time.
420 * Caller must hold rcu_lock or sufficient equivalent.
422 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
424 return walk_tg_tree_from(&root_task_group, down, up, data);
427 extern int tg_nop(struct task_group *tg, void *data);
429 extern void free_fair_sched_group(struct task_group *tg);
430 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
431 extern void online_fair_sched_group(struct task_group *tg);
432 extern void unregister_fair_sched_group(struct task_group *tg);
433 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
434 struct sched_entity *se, int cpu,
435 struct sched_entity *parent);
436 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
438 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
439 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
440 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
442 extern void free_rt_sched_group(struct task_group *tg);
443 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
444 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
445 struct sched_rt_entity *rt_se, int cpu,
446 struct sched_rt_entity *parent);
447 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
448 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
449 extern long sched_group_rt_runtime(struct task_group *tg);
450 extern long sched_group_rt_period(struct task_group *tg);
451 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
453 extern struct task_group *sched_create_group(struct task_group *parent);
454 extern void sched_online_group(struct task_group *tg,
455 struct task_group *parent);
456 extern void sched_destroy_group(struct task_group *tg);
457 extern void sched_offline_group(struct task_group *tg);
459 extern void sched_move_task(struct task_struct *tsk);
461 #ifdef CONFIG_FAIR_GROUP_SCHED
462 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
465 extern void set_task_rq_fair(struct sched_entity *se,
466 struct cfs_rq *prev, struct cfs_rq *next);
467 #else /* !CONFIG_SMP */
468 static inline void set_task_rq_fair(struct sched_entity *se,
469 struct cfs_rq *prev, struct cfs_rq *next) { }
470 #endif /* CONFIG_SMP */
471 #endif /* CONFIG_FAIR_GROUP_SCHED */
473 #else /* CONFIG_CGROUP_SCHED */
475 struct cfs_bandwidth { };
477 #endif /* CONFIG_CGROUP_SCHED */
479 /* CFS-related fields in a runqueue */
481 struct load_weight load;
482 unsigned long runnable_weight;
483 unsigned int nr_running;
484 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
485 unsigned int idle_h_nr_running; /* SCHED_IDLE */
490 u64 min_vruntime_copy;
493 struct rb_root_cached tasks_timeline;
496 * 'curr' points to currently running entity on this cfs_rq.
497 * It is set to NULL otherwise (i.e when none are currently running).
499 struct sched_entity *curr;
500 struct sched_entity *next;
501 struct sched_entity *last;
502 struct sched_entity *skip;
504 #ifdef CONFIG_SCHED_DEBUG
505 unsigned int nr_spread_over;
512 struct sched_avg avg;
514 u64 load_last_update_time_copy;
517 raw_spinlock_t lock ____cacheline_aligned;
519 unsigned long load_avg;
520 unsigned long util_avg;
521 unsigned long runnable_sum;
524 #ifdef CONFIG_FAIR_GROUP_SCHED
525 unsigned long tg_load_avg_contrib;
527 long prop_runnable_sum;
530 * h_load = weight * f(tg)
532 * Where f(tg) is the recursive weight fraction assigned to
535 unsigned long h_load;
536 u64 last_h_load_update;
537 struct sched_entity *h_load_next;
538 #endif /* CONFIG_FAIR_GROUP_SCHED */
539 #endif /* CONFIG_SMP */
541 #ifdef CONFIG_FAIR_GROUP_SCHED
542 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
545 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
546 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
547 * (like users, containers etc.)
549 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
550 * This list is used during load balance.
553 struct list_head leaf_cfs_rq_list;
554 struct task_group *tg; /* group that "owns" this runqueue */
556 #ifdef CONFIG_CFS_BANDWIDTH
558 s64 runtime_remaining;
561 u64 throttled_clock_task;
562 u64 throttled_clock_task_time;
565 struct list_head throttled_list;
566 #endif /* CONFIG_CFS_BANDWIDTH */
567 #endif /* CONFIG_FAIR_GROUP_SCHED */
570 static inline int rt_bandwidth_enabled(void)
572 return sysctl_sched_rt_runtime >= 0;
575 /* RT IPI pull logic requires IRQ_WORK */
576 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
577 # define HAVE_RT_PUSH_IPI
580 /* Real-Time classes' related field in a runqueue: */
582 struct rt_prio_array active;
583 unsigned int rt_nr_running;
584 unsigned int rr_nr_running;
585 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
587 int curr; /* highest queued rt task prio */
589 int next; /* next highest */
594 unsigned long rt_nr_migratory;
595 unsigned long rt_nr_total;
597 struct plist_head pushable_tasks;
599 #endif /* CONFIG_SMP */
605 /* Nests inside the rq lock: */
606 raw_spinlock_t rt_runtime_lock;
608 #ifdef CONFIG_RT_GROUP_SCHED
609 unsigned long rt_nr_boosted;
612 struct task_group *tg;
616 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
618 return rt_rq->rt_queued && rt_rq->rt_nr_running;
621 /* Deadline class' related fields in a runqueue */
623 /* runqueue is an rbtree, ordered by deadline */
624 struct rb_root_cached root;
626 unsigned long dl_nr_running;
630 * Deadline values of the currently executing and the
631 * earliest ready task on this rq. Caching these facilitates
632 * the decision whether or not a ready but not running task
633 * should migrate somewhere else.
640 unsigned long dl_nr_migratory;
644 * Tasks on this rq that can be pushed away. They are kept in
645 * an rb-tree, ordered by tasks' deadlines, with caching
646 * of the leftmost (earliest deadline) element.
648 struct rb_root_cached pushable_dl_tasks_root;
653 * "Active utilization" for this runqueue: increased when a
654 * task wakes up (becomes TASK_RUNNING) and decreased when a
660 * Utilization of the tasks "assigned" to this runqueue (including
661 * the tasks that are in runqueue and the tasks that executed on this
662 * CPU and blocked). Increased when a task moves to this runqueue, and
663 * decreased when the task moves away (migrates, changes scheduling
664 * policy, or terminates).
665 * This is needed to compute the "inactive utilization" for the
666 * runqueue (inactive utilization = this_bw - running_bw).
672 * Inverse of the fraction of CPU utilization that can be reclaimed
673 * by the GRUB algorithm.
678 #ifdef CONFIG_FAIR_GROUP_SCHED
679 /* An entity is a task if it doesn't "own" a runqueue */
680 #define entity_is_task(se) (!se->my_q)
682 #define entity_is_task(se) 1
687 * XXX we want to get rid of these helpers and use the full load resolution.
689 static inline long se_weight(struct sched_entity *se)
691 return scale_load_down(se->load.weight);
694 static inline long se_runnable(struct sched_entity *se)
696 return scale_load_down(se->runnable_weight);
699 static inline bool sched_asym_prefer(int a, int b)
701 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
705 struct em_perf_domain *em_pd;
706 struct perf_domain *next;
710 /* Scheduling group status flags */
711 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
712 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
715 * We add the notion of a root-domain which will be used to define per-domain
716 * variables. Each exclusive cpuset essentially defines an island domain by
717 * fully partitioning the member CPUs from any other cpuset. Whenever a new
718 * exclusive cpuset is created, we also create and attach a new root-domain
727 cpumask_var_t online;
730 * Indicate pullable load on at least one CPU, e.g:
731 * - More than one runnable task
732 * - Running task is misfit
736 /* Indicate one or more cpus over-utilized (tipping point) */
740 * The bit corresponding to a CPU gets set here if such CPU has more
741 * than one runnable -deadline task (as it is below for RT tasks).
743 cpumask_var_t dlo_mask;
748 #ifdef HAVE_RT_PUSH_IPI
750 * For IPI pull requests, loop across the rto_mask.
752 struct irq_work rto_push_work;
753 raw_spinlock_t rto_lock;
754 /* These are only updated and read within rto_lock */
757 /* These atomics are updated outside of a lock */
758 atomic_t rto_loop_next;
759 atomic_t rto_loop_start;
762 * The "RT overload" flag: it gets set if a CPU has more than
763 * one runnable RT task.
765 cpumask_var_t rto_mask;
766 struct cpupri cpupri;
768 unsigned long max_cpu_capacity;
771 * NULL-terminated list of performance domains intersecting with the
772 * CPUs of the rd. Protected by RCU.
774 struct perf_domain __rcu *pd;
777 extern void init_defrootdomain(void);
778 extern int sched_init_domains(const struct cpumask *cpu_map);
779 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
780 extern void sched_get_rd(struct root_domain *rd);
781 extern void sched_put_rd(struct root_domain *rd);
783 #ifdef HAVE_RT_PUSH_IPI
784 extern void rto_push_irq_work_func(struct irq_work *work);
786 #endif /* CONFIG_SMP */
788 #ifdef CONFIG_UCLAMP_TASK
790 * struct uclamp_bucket - Utilization clamp bucket
791 * @value: utilization clamp value for tasks on this clamp bucket
792 * @tasks: number of RUNNABLE tasks on this clamp bucket
794 * Keep track of how many tasks are RUNNABLE for a given utilization
797 struct uclamp_bucket {
798 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
799 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
803 * struct uclamp_rq - rq's utilization clamp
804 * @value: currently active clamp values for a rq
805 * @bucket: utilization clamp buckets affecting a rq
807 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
808 * A clamp value is affecting a rq when there is at least one task RUNNABLE
809 * (or actually running) with that value.
811 * There are up to UCLAMP_CNT possible different clamp values, currently there
812 * are only two: minimum utilization and maximum utilization.
814 * All utilization clamping values are MAX aggregated, since:
815 * - for util_min: we want to run the CPU at least at the max of the minimum
816 * utilization required by its currently RUNNABLE tasks.
817 * - for util_max: we want to allow the CPU to run up to the max of the
818 * maximum utilization allowed by its currently RUNNABLE tasks.
820 * Since on each system we expect only a limited number of different
821 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
822 * the metrics required to compute all the per-rq utilization clamp values.
826 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
828 #endif /* CONFIG_UCLAMP_TASK */
831 * This is the main, per-CPU runqueue data structure.
833 * Locking rule: those places that want to lock multiple runqueues
834 * (such as the load balancing or the thread migration code), lock
835 * acquire operations must be ordered by ascending &runqueue.
842 * nr_running and cpu_load should be in the same cacheline because
843 * remote CPUs use both these fields when doing load calculation.
845 unsigned int nr_running;
846 #ifdef CONFIG_NUMA_BALANCING
847 unsigned int nr_numa_running;
848 unsigned int nr_preferred_running;
849 unsigned int numa_migrate_on;
851 #ifdef CONFIG_NO_HZ_COMMON
853 unsigned long last_load_update_tick;
854 unsigned long last_blocked_load_update_tick;
855 unsigned int has_blocked_load;
856 #endif /* CONFIG_SMP */
857 unsigned int nohz_tick_stopped;
859 #endif /* CONFIG_NO_HZ_COMMON */
861 unsigned long nr_load_updates;
864 #ifdef CONFIG_UCLAMP_TASK
865 /* Utilization clamp values based on CPU's RUNNABLE tasks */
866 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
867 unsigned int uclamp_flags;
868 #define UCLAMP_FLAG_IDLE 0x01
875 #ifdef CONFIG_FAIR_GROUP_SCHED
876 /* list of leaf cfs_rq on this CPU: */
877 struct list_head leaf_cfs_rq_list;
878 struct list_head *tmp_alone_branch;
879 #endif /* CONFIG_FAIR_GROUP_SCHED */
882 * This is part of a global counter where only the total sum
883 * over all CPUs matters. A task can increase this counter on
884 * one CPU and if it got migrated afterwards it may decrease
885 * it on another CPU. Always updated under the runqueue lock:
887 unsigned long nr_uninterruptible;
889 struct task_struct *curr;
890 struct task_struct *idle;
891 struct task_struct *stop;
892 unsigned long next_balance;
893 struct mm_struct *prev_mm;
895 unsigned int clock_update_flags;
897 /* Ensure that all clocks are in the same cache line */
898 u64 clock_task ____cacheline_aligned;
900 unsigned long lost_idle_time;
905 struct root_domain *rd;
906 struct sched_domain __rcu *sd;
908 unsigned long cpu_capacity;
909 unsigned long cpu_capacity_orig;
911 struct callback_head *balance_callback;
913 unsigned char idle_balance;
915 unsigned long misfit_task_load;
917 /* For active balancing */
920 struct cpu_stop_work active_balance_work;
922 /* CPU of this runqueue: */
926 struct list_head cfs_tasks;
928 struct sched_avg avg_rt;
929 struct sched_avg avg_dl;
930 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
931 struct sched_avg avg_irq;
936 /* This is used to determine avg_idle's max value */
937 u64 max_idle_balance_cost;
940 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
943 #ifdef CONFIG_PARAVIRT
946 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
947 u64 prev_steal_time_rq;
950 /* calc_load related fields */
951 unsigned long calc_load_update;
952 long calc_load_active;
954 #ifdef CONFIG_SCHED_HRTICK
956 int hrtick_csd_pending;
957 call_single_data_t hrtick_csd;
959 struct hrtimer hrtick_timer;
962 #ifdef CONFIG_SCHEDSTATS
964 struct sched_info rq_sched_info;
965 unsigned long long rq_cpu_time;
966 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
968 /* sys_sched_yield() stats */
969 unsigned int yld_count;
971 /* schedule() stats */
972 unsigned int sched_count;
973 unsigned int sched_goidle;
975 /* try_to_wake_up() stats */
976 unsigned int ttwu_count;
977 unsigned int ttwu_local;
981 struct llist_head wake_list;
984 #ifdef CONFIG_CPU_IDLE
985 /* Must be inspected within a rcu lock section */
986 struct cpuidle_state *idle_state;
990 #ifdef CONFIG_FAIR_GROUP_SCHED
992 /* CPU runqueue to which this cfs_rq is attached */
993 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1000 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1002 return container_of(cfs_rq, struct rq, cfs);
1006 static inline int cpu_of(struct rq *rq)
1016 #ifdef CONFIG_SCHED_SMT
1017 extern void __update_idle_core(struct rq *rq);
1019 static inline void update_idle_core(struct rq *rq)
1021 if (static_branch_unlikely(&sched_smt_present))
1022 __update_idle_core(rq);
1026 static inline void update_idle_core(struct rq *rq) { }
1029 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1031 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1032 #define this_rq() this_cpu_ptr(&runqueues)
1033 #define task_rq(p) cpu_rq(task_cpu(p))
1034 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1035 #define raw_rq() raw_cpu_ptr(&runqueues)
1037 extern void update_rq_clock(struct rq *rq);
1039 static inline u64 __rq_clock_broken(struct rq *rq)
1041 return READ_ONCE(rq->clock);
1045 * rq::clock_update_flags bits
1047 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1048 * call to __schedule(). This is an optimisation to avoid
1049 * neighbouring rq clock updates.
1051 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1052 * in effect and calls to update_rq_clock() are being ignored.
1054 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1055 * made to update_rq_clock() since the last time rq::lock was pinned.
1057 * If inside of __schedule(), clock_update_flags will have been
1058 * shifted left (a left shift is a cheap operation for the fast path
1059 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1061 * if (rq-clock_update_flags >= RQCF_UPDATED)
1063 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1064 * one position though, because the next rq_unpin_lock() will shift it
1067 #define RQCF_REQ_SKIP 0x01
1068 #define RQCF_ACT_SKIP 0x02
1069 #define RQCF_UPDATED 0x04
1071 static inline void assert_clock_updated(struct rq *rq)
1074 * The only reason for not seeing a clock update since the
1075 * last rq_pin_lock() is if we're currently skipping updates.
1077 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1080 static inline u64 rq_clock(struct rq *rq)
1082 lockdep_assert_held(&rq->lock);
1083 assert_clock_updated(rq);
1088 static inline u64 rq_clock_task(struct rq *rq)
1090 lockdep_assert_held(&rq->lock);
1091 assert_clock_updated(rq);
1093 return rq->clock_task;
1096 static inline void rq_clock_skip_update(struct rq *rq)
1098 lockdep_assert_held(&rq->lock);
1099 rq->clock_update_flags |= RQCF_REQ_SKIP;
1103 * See rt task throttling, which is the only time a skip
1104 * request is cancelled.
1106 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1108 lockdep_assert_held(&rq->lock);
1109 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1113 unsigned long flags;
1114 struct pin_cookie cookie;
1115 #ifdef CONFIG_SCHED_DEBUG
1117 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1118 * current pin context is stashed here in case it needs to be
1119 * restored in rq_repin_lock().
1121 unsigned int clock_update_flags;
1125 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1127 rf->cookie = lockdep_pin_lock(&rq->lock);
1129 #ifdef CONFIG_SCHED_DEBUG
1130 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1131 rf->clock_update_flags = 0;
1135 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1137 #ifdef CONFIG_SCHED_DEBUG
1138 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1139 rf->clock_update_flags = RQCF_UPDATED;
1142 lockdep_unpin_lock(&rq->lock, rf->cookie);
1145 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1147 lockdep_repin_lock(&rq->lock, rf->cookie);
1149 #ifdef CONFIG_SCHED_DEBUG
1151 * Restore the value we stashed in @rf for this pin context.
1153 rq->clock_update_flags |= rf->clock_update_flags;
1157 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1158 __acquires(rq->lock);
1160 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1161 __acquires(p->pi_lock)
1162 __acquires(rq->lock);
1164 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1165 __releases(rq->lock)
1167 rq_unpin_lock(rq, rf);
1168 raw_spin_unlock(&rq->lock);
1172 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1173 __releases(rq->lock)
1174 __releases(p->pi_lock)
1176 rq_unpin_lock(rq, rf);
1177 raw_spin_unlock(&rq->lock);
1178 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1182 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1183 __acquires(rq->lock)
1185 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1186 rq_pin_lock(rq, rf);
1190 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1191 __acquires(rq->lock)
1193 raw_spin_lock_irq(&rq->lock);
1194 rq_pin_lock(rq, rf);
1198 rq_lock(struct rq *rq, struct rq_flags *rf)
1199 __acquires(rq->lock)
1201 raw_spin_lock(&rq->lock);
1202 rq_pin_lock(rq, rf);
1206 rq_relock(struct rq *rq, struct rq_flags *rf)
1207 __acquires(rq->lock)
1209 raw_spin_lock(&rq->lock);
1210 rq_repin_lock(rq, rf);
1214 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1215 __releases(rq->lock)
1217 rq_unpin_lock(rq, rf);
1218 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1222 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1223 __releases(rq->lock)
1225 rq_unpin_lock(rq, rf);
1226 raw_spin_unlock_irq(&rq->lock);
1230 rq_unlock(struct rq *rq, struct rq_flags *rf)
1231 __releases(rq->lock)
1233 rq_unpin_lock(rq, rf);
1234 raw_spin_unlock(&rq->lock);
1237 static inline struct rq *
1238 this_rq_lock_irq(struct rq_flags *rf)
1239 __acquires(rq->lock)
1243 local_irq_disable();
1250 enum numa_topology_type {
1255 extern enum numa_topology_type sched_numa_topology_type;
1256 extern int sched_max_numa_distance;
1257 extern bool find_numa_distance(int distance);
1258 extern void sched_init_numa(void);
1259 extern void sched_domains_numa_masks_set(unsigned int cpu);
1260 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1261 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1263 static inline void sched_init_numa(void) { }
1264 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1265 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1266 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1272 #ifdef CONFIG_NUMA_BALANCING
1273 /* The regions in numa_faults array from task_struct */
1274 enum numa_faults_stats {
1280 extern void sched_setnuma(struct task_struct *p, int node);
1281 extern int migrate_task_to(struct task_struct *p, int cpu);
1282 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1284 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1287 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1290 #endif /* CONFIG_NUMA_BALANCING */
1295 queue_balance_callback(struct rq *rq,
1296 struct callback_head *head,
1297 void (*func)(struct rq *rq))
1299 lockdep_assert_held(&rq->lock);
1301 if (unlikely(head->next))
1304 head->func = (void (*)(struct callback_head *))func;
1305 head->next = rq->balance_callback;
1306 rq->balance_callback = head;
1309 extern void sched_ttwu_pending(void);
1311 #define rcu_dereference_check_sched_domain(p) \
1312 rcu_dereference_check((p), \
1313 lockdep_is_held(&sched_domains_mutex))
1316 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1317 * See destroy_sched_domains: call_rcu for details.
1319 * The domain tree of any CPU may only be accessed from within
1320 * preempt-disabled sections.
1322 #define for_each_domain(cpu, __sd) \
1323 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1324 __sd; __sd = __sd->parent)
1326 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1329 * highest_flag_domain - Return highest sched_domain containing flag.
1330 * @cpu: The CPU whose highest level of sched domain is to
1332 * @flag: The flag to check for the highest sched_domain
1333 * for the given CPU.
1335 * Returns the highest sched_domain of a CPU which contains the given flag.
1337 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1339 struct sched_domain *sd, *hsd = NULL;
1341 for_each_domain(cpu, sd) {
1342 if (!(sd->flags & flag))
1350 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1352 struct sched_domain *sd;
1354 for_each_domain(cpu, sd) {
1355 if (sd->flags & flag)
1362 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1363 DECLARE_PER_CPU(int, sd_llc_size);
1364 DECLARE_PER_CPU(int, sd_llc_id);
1365 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1366 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1367 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1368 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1369 extern struct static_key_false sched_asym_cpucapacity;
1371 struct sched_group_capacity {
1374 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1377 unsigned long capacity;
1378 unsigned long min_capacity; /* Min per-CPU capacity in group */
1379 unsigned long max_capacity; /* Max per-CPU capacity in group */
1380 unsigned long next_update;
1381 int imbalance; /* XXX unrelated to capacity but shared group state */
1383 #ifdef CONFIG_SCHED_DEBUG
1387 unsigned long cpumask[0]; /* Balance mask */
1390 struct sched_group {
1391 struct sched_group *next; /* Must be a circular list */
1394 unsigned int group_weight;
1395 struct sched_group_capacity *sgc;
1396 int asym_prefer_cpu; /* CPU of highest priority in group */
1399 * The CPUs this group covers.
1401 * NOTE: this field is variable length. (Allocated dynamically
1402 * by attaching extra space to the end of the structure,
1403 * depending on how many CPUs the kernel has booted up with)
1405 unsigned long cpumask[0];
1408 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1410 return to_cpumask(sg->cpumask);
1414 * See build_balance_mask().
1416 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1418 return to_cpumask(sg->sgc->cpumask);
1422 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1423 * @group: The group whose first CPU is to be returned.
1425 static inline unsigned int group_first_cpu(struct sched_group *group)
1427 return cpumask_first(sched_group_span(group));
1430 extern int group_balance_cpu(struct sched_group *sg);
1432 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1433 void register_sched_domain_sysctl(void);
1434 void dirty_sched_domain_sysctl(int cpu);
1435 void unregister_sched_domain_sysctl(void);
1437 static inline void register_sched_domain_sysctl(void)
1440 static inline void dirty_sched_domain_sysctl(int cpu)
1443 static inline void unregister_sched_domain_sysctl(void)
1450 static inline void sched_ttwu_pending(void) { }
1452 #endif /* CONFIG_SMP */
1455 #include "autogroup.h"
1457 #ifdef CONFIG_CGROUP_SCHED
1460 * Return the group to which this tasks belongs.
1462 * We cannot use task_css() and friends because the cgroup subsystem
1463 * changes that value before the cgroup_subsys::attach() method is called,
1464 * therefore we cannot pin it and might observe the wrong value.
1466 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1467 * core changes this before calling sched_move_task().
1469 * Instead we use a 'copy' which is updated from sched_move_task() while
1470 * holding both task_struct::pi_lock and rq::lock.
1472 static inline struct task_group *task_group(struct task_struct *p)
1474 return p->sched_task_group;
1477 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1478 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1480 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1481 struct task_group *tg = task_group(p);
1484 #ifdef CONFIG_FAIR_GROUP_SCHED
1485 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1486 p->se.cfs_rq = tg->cfs_rq[cpu];
1487 p->se.parent = tg->se[cpu];
1490 #ifdef CONFIG_RT_GROUP_SCHED
1491 p->rt.rt_rq = tg->rt_rq[cpu];
1492 p->rt.parent = tg->rt_se[cpu];
1496 #else /* CONFIG_CGROUP_SCHED */
1498 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1499 static inline struct task_group *task_group(struct task_struct *p)
1504 #endif /* CONFIG_CGROUP_SCHED */
1506 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1508 set_task_rq(p, cpu);
1511 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1512 * successfully executed on another CPU. We must ensure that updates of
1513 * per-task data have been completed by this moment.
1516 #ifdef CONFIG_THREAD_INFO_IN_TASK
1517 WRITE_ONCE(p->cpu, cpu);
1519 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1526 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1528 #ifdef CONFIG_SCHED_DEBUG
1529 # include <linux/static_key.h>
1530 # define const_debug __read_mostly
1532 # define const_debug const
1535 #define SCHED_FEAT(name, enabled) \
1536 __SCHED_FEAT_##name ,
1539 #include "features.h"
1545 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1548 * To support run-time toggling of sched features, all the translation units
1549 * (but core.c) reference the sysctl_sched_features defined in core.c.
1551 extern const_debug unsigned int sysctl_sched_features;
1553 #define SCHED_FEAT(name, enabled) \
1554 static __always_inline bool static_branch_##name(struct static_key *key) \
1556 return static_key_##enabled(key); \
1559 #include "features.h"
1562 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1563 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1565 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1568 * Each translation unit has its own copy of sysctl_sched_features to allow
1569 * constants propagation at compile time and compiler optimization based on
1572 #define SCHED_FEAT(name, enabled) \
1573 (1UL << __SCHED_FEAT_##name) * enabled |
1574 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1575 #include "features.h"
1579 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1581 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1583 extern struct static_key_false sched_numa_balancing;
1584 extern struct static_key_false sched_schedstats;
1586 static inline u64 global_rt_period(void)
1588 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1591 static inline u64 global_rt_runtime(void)
1593 if (sysctl_sched_rt_runtime < 0)
1596 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1599 static inline int task_current(struct rq *rq, struct task_struct *p)
1601 return rq->curr == p;
1604 static inline int task_running(struct rq *rq, struct task_struct *p)
1609 return task_current(rq, p);
1613 static inline int task_on_rq_queued(struct task_struct *p)
1615 return p->on_rq == TASK_ON_RQ_QUEUED;
1618 static inline int task_on_rq_migrating(struct task_struct *p)
1620 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1626 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1627 #define WF_FORK 0x02 /* Child wakeup after fork */
1628 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1631 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1632 * of tasks with abnormal "nice" values across CPUs the contribution that
1633 * each task makes to its run queue's load is weighted according to its
1634 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1635 * scaled version of the new time slice allocation that they receive on time
1639 #define WEIGHT_IDLEPRIO 3
1640 #define WMULT_IDLEPRIO 1431655765
1642 extern const int sched_prio_to_weight[40];
1643 extern const u32 sched_prio_to_wmult[40];
1646 * {de,en}queue flags:
1648 * DEQUEUE_SLEEP - task is no longer runnable
1649 * ENQUEUE_WAKEUP - task just became runnable
1651 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1652 * are in a known state which allows modification. Such pairs
1653 * should preserve as much state as possible.
1655 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1658 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1659 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1660 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1664 #define DEQUEUE_SLEEP 0x01
1665 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1666 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1667 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1669 #define ENQUEUE_WAKEUP 0x01
1670 #define ENQUEUE_RESTORE 0x02
1671 #define ENQUEUE_MOVE 0x04
1672 #define ENQUEUE_NOCLOCK 0x08
1674 #define ENQUEUE_HEAD 0x10
1675 #define ENQUEUE_REPLENISH 0x20
1677 #define ENQUEUE_MIGRATED 0x40
1679 #define ENQUEUE_MIGRATED 0x00
1682 #define RETRY_TASK ((void *)-1UL)
1684 struct sched_class {
1685 const struct sched_class *next;
1687 #ifdef CONFIG_UCLAMP_TASK
1691 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1692 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1693 void (*yield_task) (struct rq *rq);
1694 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1696 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1699 * It is the responsibility of the pick_next_task() method that will
1700 * return the next task to call put_prev_task() on the @prev task or
1701 * something equivalent.
1703 * May return RETRY_TASK when it finds a higher prio class has runnable
1706 struct task_struct * (*pick_next_task)(struct rq *rq,
1707 struct task_struct *prev,
1708 struct rq_flags *rf);
1709 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1710 void (*set_next_task)(struct rq *rq, struct task_struct *p);
1713 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1714 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1716 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1718 void (*set_cpus_allowed)(struct task_struct *p,
1719 const struct cpumask *newmask);
1721 void (*rq_online)(struct rq *rq);
1722 void (*rq_offline)(struct rq *rq);
1725 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1726 void (*task_fork)(struct task_struct *p);
1727 void (*task_dead)(struct task_struct *p);
1730 * The switched_from() call is allowed to drop rq->lock, therefore we
1731 * cannot assume the switched_from/switched_to pair is serliazed by
1732 * rq->lock. They are however serialized by p->pi_lock.
1734 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1735 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1736 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1739 unsigned int (*get_rr_interval)(struct rq *rq,
1740 struct task_struct *task);
1742 void (*update_curr)(struct rq *rq);
1744 #define TASK_SET_GROUP 0
1745 #define TASK_MOVE_GROUP 1
1747 #ifdef CONFIG_FAIR_GROUP_SCHED
1748 void (*task_change_group)(struct task_struct *p, int type);
1752 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1754 WARN_ON_ONCE(rq->curr != prev);
1755 prev->sched_class->put_prev_task(rq, prev);
1758 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1760 WARN_ON_ONCE(rq->curr != next);
1761 next->sched_class->set_next_task(rq, next);
1765 #define sched_class_highest (&stop_sched_class)
1767 #define sched_class_highest (&dl_sched_class)
1769 #define for_each_class(class) \
1770 for (class = sched_class_highest; class; class = class->next)
1772 extern const struct sched_class stop_sched_class;
1773 extern const struct sched_class dl_sched_class;
1774 extern const struct sched_class rt_sched_class;
1775 extern const struct sched_class fair_sched_class;
1776 extern const struct sched_class idle_sched_class;
1781 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1783 extern void trigger_load_balance(struct rq *rq);
1785 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1789 #ifdef CONFIG_CPU_IDLE
1790 static inline void idle_set_state(struct rq *rq,
1791 struct cpuidle_state *idle_state)
1793 rq->idle_state = idle_state;
1796 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1798 SCHED_WARN_ON(!rcu_read_lock_held());
1800 return rq->idle_state;
1803 static inline void idle_set_state(struct rq *rq,
1804 struct cpuidle_state *idle_state)
1808 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1814 extern void schedule_idle(void);
1816 extern void sysrq_sched_debug_show(void);
1817 extern void sched_init_granularity(void);
1818 extern void update_max_interval(void);
1820 extern void init_sched_dl_class(void);
1821 extern void init_sched_rt_class(void);
1822 extern void init_sched_fair_class(void);
1824 extern void reweight_task(struct task_struct *p, int prio);
1826 extern void resched_curr(struct rq *rq);
1827 extern void resched_cpu(int cpu);
1829 extern struct rt_bandwidth def_rt_bandwidth;
1830 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1832 extern struct dl_bandwidth def_dl_bandwidth;
1833 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1834 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1835 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1836 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1839 #define BW_UNIT (1 << BW_SHIFT)
1840 #define RATIO_SHIFT 8
1841 unsigned long to_ratio(u64 period, u64 runtime);
1843 extern void init_entity_runnable_average(struct sched_entity *se);
1844 extern void post_init_entity_util_avg(struct task_struct *p);
1846 #ifdef CONFIG_NO_HZ_FULL
1847 extern bool sched_can_stop_tick(struct rq *rq);
1848 extern int __init sched_tick_offload_init(void);
1851 * Tick may be needed by tasks in the runqueue depending on their policy and
1852 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1853 * nohz mode if necessary.
1855 static inline void sched_update_tick_dependency(struct rq *rq)
1859 if (!tick_nohz_full_enabled())
1864 if (!tick_nohz_full_cpu(cpu))
1867 if (sched_can_stop_tick(rq))
1868 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1870 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1873 static inline int sched_tick_offload_init(void) { return 0; }
1874 static inline void sched_update_tick_dependency(struct rq *rq) { }
1877 static inline void add_nr_running(struct rq *rq, unsigned count)
1879 unsigned prev_nr = rq->nr_running;
1881 rq->nr_running = prev_nr + count;
1884 if (prev_nr < 2 && rq->nr_running >= 2) {
1885 if (!READ_ONCE(rq->rd->overload))
1886 WRITE_ONCE(rq->rd->overload, 1);
1890 sched_update_tick_dependency(rq);
1893 static inline void sub_nr_running(struct rq *rq, unsigned count)
1895 rq->nr_running -= count;
1896 /* Check if we still need preemption */
1897 sched_update_tick_dependency(rq);
1900 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1901 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1903 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1905 extern const_debug unsigned int sysctl_sched_nr_migrate;
1906 extern const_debug unsigned int sysctl_sched_migration_cost;
1908 #ifdef CONFIG_SCHED_HRTICK
1912 * - enabled by features
1913 * - hrtimer is actually high res
1915 static inline int hrtick_enabled(struct rq *rq)
1917 if (!sched_feat(HRTICK))
1919 if (!cpu_active(cpu_of(rq)))
1921 return hrtimer_is_hres_active(&rq->hrtick_timer);
1924 void hrtick_start(struct rq *rq, u64 delay);
1928 static inline int hrtick_enabled(struct rq *rq)
1933 #endif /* CONFIG_SCHED_HRTICK */
1935 #ifndef arch_scale_freq_capacity
1936 static __always_inline
1937 unsigned long arch_scale_freq_capacity(int cpu)
1939 return SCHED_CAPACITY_SCALE;
1944 #ifdef CONFIG_PREEMPT
1946 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1949 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1950 * way at the expense of forcing extra atomic operations in all
1951 * invocations. This assures that the double_lock is acquired using the
1952 * same underlying policy as the spinlock_t on this architecture, which
1953 * reduces latency compared to the unfair variant below. However, it
1954 * also adds more overhead and therefore may reduce throughput.
1956 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1957 __releases(this_rq->lock)
1958 __acquires(busiest->lock)
1959 __acquires(this_rq->lock)
1961 raw_spin_unlock(&this_rq->lock);
1962 double_rq_lock(this_rq, busiest);
1969 * Unfair double_lock_balance: Optimizes throughput at the expense of
1970 * latency by eliminating extra atomic operations when the locks are
1971 * already in proper order on entry. This favors lower CPU-ids and will
1972 * grant the double lock to lower CPUs over higher ids under contention,
1973 * regardless of entry order into the function.
1975 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1976 __releases(this_rq->lock)
1977 __acquires(busiest->lock)
1978 __acquires(this_rq->lock)
1982 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1983 if (busiest < this_rq) {
1984 raw_spin_unlock(&this_rq->lock);
1985 raw_spin_lock(&busiest->lock);
1986 raw_spin_lock_nested(&this_rq->lock,
1987 SINGLE_DEPTH_NESTING);
1990 raw_spin_lock_nested(&busiest->lock,
1991 SINGLE_DEPTH_NESTING);
1996 #endif /* CONFIG_PREEMPT */
1999 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2001 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2003 if (unlikely(!irqs_disabled())) {
2004 /* printk() doesn't work well under rq->lock */
2005 raw_spin_unlock(&this_rq->lock);
2009 return _double_lock_balance(this_rq, busiest);
2012 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2013 __releases(busiest->lock)
2015 raw_spin_unlock(&busiest->lock);
2016 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2019 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2025 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2028 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2034 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2037 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2043 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2047 * double_rq_lock - safely lock two runqueues
2049 * Note this does not disable interrupts like task_rq_lock,
2050 * you need to do so manually before calling.
2052 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2053 __acquires(rq1->lock)
2054 __acquires(rq2->lock)
2056 BUG_ON(!irqs_disabled());
2058 raw_spin_lock(&rq1->lock);
2059 __acquire(rq2->lock); /* Fake it out ;) */
2062 raw_spin_lock(&rq1->lock);
2063 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2065 raw_spin_lock(&rq2->lock);
2066 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2072 * double_rq_unlock - safely unlock two runqueues
2074 * Note this does not restore interrupts like task_rq_unlock,
2075 * you need to do so manually after calling.
2077 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2078 __releases(rq1->lock)
2079 __releases(rq2->lock)
2081 raw_spin_unlock(&rq1->lock);
2083 raw_spin_unlock(&rq2->lock);
2085 __release(rq2->lock);
2088 extern void set_rq_online (struct rq *rq);
2089 extern void set_rq_offline(struct rq *rq);
2090 extern bool sched_smp_initialized;
2092 #else /* CONFIG_SMP */
2095 * double_rq_lock - safely lock two runqueues
2097 * Note this does not disable interrupts like task_rq_lock,
2098 * you need to do so manually before calling.
2100 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2101 __acquires(rq1->lock)
2102 __acquires(rq2->lock)
2104 BUG_ON(!irqs_disabled());
2106 raw_spin_lock(&rq1->lock);
2107 __acquire(rq2->lock); /* Fake it out ;) */
2111 * double_rq_unlock - safely unlock two runqueues
2113 * Note this does not restore interrupts like task_rq_unlock,
2114 * you need to do so manually after calling.
2116 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2117 __releases(rq1->lock)
2118 __releases(rq2->lock)
2121 raw_spin_unlock(&rq1->lock);
2122 __release(rq2->lock);
2127 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2128 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2130 #ifdef CONFIG_SCHED_DEBUG
2131 extern bool sched_debug_enabled;
2133 extern void print_cfs_stats(struct seq_file *m, int cpu);
2134 extern void print_rt_stats(struct seq_file *m, int cpu);
2135 extern void print_dl_stats(struct seq_file *m, int cpu);
2136 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2137 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2138 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2139 #ifdef CONFIG_NUMA_BALANCING
2141 show_numa_stats(struct task_struct *p, struct seq_file *m);
2143 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2144 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2145 #endif /* CONFIG_NUMA_BALANCING */
2146 #endif /* CONFIG_SCHED_DEBUG */
2148 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2149 extern void init_rt_rq(struct rt_rq *rt_rq);
2150 extern void init_dl_rq(struct dl_rq *dl_rq);
2152 extern void cfs_bandwidth_usage_inc(void);
2153 extern void cfs_bandwidth_usage_dec(void);
2155 #ifdef CONFIG_NO_HZ_COMMON
2156 #define NOHZ_BALANCE_KICK_BIT 0
2157 #define NOHZ_STATS_KICK_BIT 1
2159 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2160 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2162 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2164 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2166 extern void nohz_balance_exit_idle(struct rq *rq);
2168 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2174 void __dl_update(struct dl_bw *dl_b, s64 bw)
2176 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2179 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2180 "sched RCU must be held");
2181 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2182 struct rq *rq = cpu_rq(i);
2184 rq->dl.extra_bw += bw;
2189 void __dl_update(struct dl_bw *dl_b, s64 bw)
2191 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2198 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2203 struct u64_stats_sync sync;
2206 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2209 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2210 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2211 * and never move forward.
2213 static inline u64 irq_time_read(int cpu)
2215 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2220 seq = __u64_stats_fetch_begin(&irqtime->sync);
2221 total = irqtime->total;
2222 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2226 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2228 #ifdef CONFIG_CPU_FREQ
2229 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2232 * cpufreq_update_util - Take a note about CPU utilization changes.
2233 * @rq: Runqueue to carry out the update for.
2234 * @flags: Update reason flags.
2236 * This function is called by the scheduler on the CPU whose utilization is
2239 * It can only be called from RCU-sched read-side critical sections.
2241 * The way cpufreq is currently arranged requires it to evaluate the CPU
2242 * performance state (frequency/voltage) on a regular basis to prevent it from
2243 * being stuck in a completely inadequate performance level for too long.
2244 * That is not guaranteed to happen if the updates are only triggered from CFS
2245 * and DL, though, because they may not be coming in if only RT tasks are
2246 * active all the time (or there are RT tasks only).
2248 * As a workaround for that issue, this function is called periodically by the
2249 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2250 * but that really is a band-aid. Going forward it should be replaced with
2251 * solutions targeted more specifically at RT tasks.
2253 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2255 struct update_util_data *data;
2257 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2260 data->func(data, rq_clock(rq), flags);
2263 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2264 #endif /* CONFIG_CPU_FREQ */
2266 #ifdef CONFIG_UCLAMP_TASK
2267 unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id);
2269 static __always_inline
2270 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2271 struct task_struct *p)
2273 unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2274 unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2277 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2278 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2282 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2283 * RUNNABLE tasks with _different_ clamps, we can end up with an
2284 * inversion. Fix it now when the clamps are applied.
2286 if (unlikely(min_util >= max_util))
2289 return clamp(util, min_util, max_util);
2292 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2294 return uclamp_util_with(rq, util, NULL);
2296 #else /* CONFIG_UCLAMP_TASK */
2297 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2298 struct task_struct *p)
2302 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2306 #endif /* CONFIG_UCLAMP_TASK */
2308 #ifdef arch_scale_freq_capacity
2309 # ifndef arch_scale_freq_invariant
2310 # define arch_scale_freq_invariant() true
2313 # define arch_scale_freq_invariant() false
2317 static inline unsigned long capacity_orig_of(int cpu)
2319 return cpu_rq(cpu)->cpu_capacity_orig;
2324 * enum schedutil_type - CPU utilization type
2325 * @FREQUENCY_UTIL: Utilization used to select frequency
2326 * @ENERGY_UTIL: Utilization used during energy calculation
2328 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2329 * need to be aggregated differently depending on the usage made of them. This
2330 * enum is used within schedutil_freq_util() to differentiate the types of
2331 * utilization expected by the callers, and adjust the aggregation accordingly.
2333 enum schedutil_type {
2338 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2340 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2341 unsigned long max, enum schedutil_type type,
2342 struct task_struct *p);
2344 static inline unsigned long cpu_bw_dl(struct rq *rq)
2346 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2349 static inline unsigned long cpu_util_dl(struct rq *rq)
2351 return READ_ONCE(rq->avg_dl.util_avg);
2354 static inline unsigned long cpu_util_cfs(struct rq *rq)
2356 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2358 if (sched_feat(UTIL_EST)) {
2359 util = max_t(unsigned long, util,
2360 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2366 static inline unsigned long cpu_util_rt(struct rq *rq)
2368 return READ_ONCE(rq->avg_rt.util_avg);
2370 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2371 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2372 unsigned long max, enum schedutil_type type,
2373 struct task_struct *p)
2377 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2379 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2380 static inline unsigned long cpu_util_irq(struct rq *rq)
2382 return rq->avg_irq.util_avg;
2386 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2388 util *= (max - irq);
2395 static inline unsigned long cpu_util_irq(struct rq *rq)
2401 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2407 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2409 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2411 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2413 static inline bool sched_energy_enabled(void)
2415 return static_branch_unlikely(&sched_energy_present);
2418 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2420 #define perf_domain_span(pd) NULL
2421 static inline bool sched_energy_enabled(void) { return false; }
2423 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */