1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/bitops.h>
40 #include <linux/blkdev.h>
41 #include <linux/compat.h>
42 #include <linux/context_tracking.h>
43 #include <linux/cpufreq.h>
44 #include <linux/cpuidle.h>
45 #include <linux/cpuset.h>
46 #include <linux/ctype.h>
47 #include <linux/debugfs.h>
48 #include <linux/delayacct.h>
49 #include <linux/energy_model.h>
50 #include <linux/init_task.h>
51 #include <linux/kprobes.h>
52 #include <linux/kthread.h>
53 #include <linux/membarrier.h>
54 #include <linux/migrate.h>
55 #include <linux/mmu_context.h>
56 #include <linux/nmi.h>
57 #include <linux/proc_fs.h>
58 #include <linux/prefetch.h>
59 #include <linux/profile.h>
60 #include <linux/psi.h>
61 #include <linux/ratelimit.h>
62 #include <linux/rcupdate_wait.h>
63 #include <linux/security.h>
64 #include <linux/stop_machine.h>
65 #include <linux/suspend.h>
66 #include <linux/swait.h>
67 #include <linux/syscalls.h>
68 #include <linux/task_work.h>
69 #include <linux/tsacct_kern.h>
73 #ifdef CONFIG_PARAVIRT
74 # include <asm/paravirt.h>
78 #include "cpudeadline.h"
80 #include <trace/events/sched.h>
82 #ifdef CONFIG_SCHED_DEBUG
83 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
85 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
91 /* task_struct::on_rq states: */
92 #define TASK_ON_RQ_QUEUED 1
93 #define TASK_ON_RQ_MIGRATING 2
95 extern __read_mostly int scheduler_running;
97 extern unsigned long calc_load_update;
98 extern atomic_long_t calc_load_tasks;
100 extern void calc_global_load_tick(struct rq *this_rq);
101 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
103 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
105 * Helpers for converting nanosecond timing to jiffy resolution
107 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
116 * We increase resolution only if we have enough bits to allow this increased
117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
124 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load_down(w) \
128 unsigned long __w = (w); \
130 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
134 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
135 # define scale_load(w) (w)
136 # define scale_load_down(w) (w)
140 * Task weight (visible to users) and its load (invisible to users) have
141 * independent resolution, but they should be well calibrated. We use
142 * scale_load() and scale_load_down(w) to convert between them. The
143 * following must be true:
145 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
148 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
151 * Single value that decides SCHED_DEADLINE internal math precision.
152 * 10 -> just above 1us
153 * 9 -> just above 0.5us
158 * Single value that denotes runtime == period, ie unlimited time.
160 #define RUNTIME_INF ((u64)~0ULL)
162 static inline int idle_policy(int policy)
164 return policy == SCHED_IDLE;
166 static inline int fair_policy(int policy)
168 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
171 static inline int rt_policy(int policy)
173 return policy == SCHED_FIFO || policy == SCHED_RR;
176 static inline int dl_policy(int policy)
178 return policy == SCHED_DEADLINE;
180 static inline bool valid_policy(int policy)
182 return idle_policy(policy) || fair_policy(policy) ||
183 rt_policy(policy) || dl_policy(policy);
186 static inline int task_has_idle_policy(struct task_struct *p)
188 return idle_policy(p->policy);
191 static inline int task_has_rt_policy(struct task_struct *p)
193 return rt_policy(p->policy);
196 static inline int task_has_dl_policy(struct task_struct *p)
198 return dl_policy(p->policy);
201 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
203 static inline void update_avg(u64 *avg, u64 sample)
205 s64 diff = sample - *avg;
210 * Shifting a value by an exponent greater *or equal* to the size of said value
211 * is UB; cap at size-1.
213 #define shr_bound(val, shift) \
214 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
217 * !! For sched_setattr_nocheck() (kernel) only !!
219 * This is actually gross. :(
221 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
222 * tasks, but still be able to sleep. We need this on platforms that cannot
223 * atomically change clock frequency. Remove once fast switching will be
224 * available on such platforms.
226 * SUGOV stands for SchedUtil GOVernor.
228 #define SCHED_FLAG_SUGOV 0x10000000
230 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
232 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
233 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
240 * Tells if entity @a should preempt entity @b.
243 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
245 return dl_entity_is_special(a) ||
246 dl_time_before(a->deadline, b->deadline);
250 * This is the priority-queue data structure of the RT scheduling class:
252 struct rt_prio_array {
253 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
254 struct list_head queue[MAX_RT_PRIO];
257 struct rt_bandwidth {
258 /* nests inside the rq lock: */
259 raw_spinlock_t rt_runtime_lock;
262 struct hrtimer rt_period_timer;
263 unsigned int rt_period_active;
266 void __dl_clear_params(struct task_struct *p);
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;
280 * To keep the bandwidth of -deadline tasks under control
281 * we need some place where:
282 * - store the maximum -deadline bandwidth of each cpu;
283 * - cache the fraction of bandwidth that is currently allocated in
286 * This is all done in the data structure below. It is similar to the
287 * one used for RT-throttling (rt_bandwidth), with the main difference
288 * that, since here we are only interested in admission control, we
289 * do not decrease any runtime while the group "executes", neither we
290 * need a timer to replenish it.
292 * With respect to SMP, bandwidth is given on a per root domain basis,
294 * - bw (< 100%) is the deadline bandwidth of each CPU;
295 * - total_bw is the currently allocated bandwidth in each root domain;
303 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
306 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
308 dl_b->total_bw -= tsk_bw;
309 __dl_update(dl_b, (s32)tsk_bw / cpus);
313 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
315 dl_b->total_bw += tsk_bw;
316 __dl_update(dl_b, -((s32)tsk_bw / cpus));
319 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
320 u64 old_bw, u64 new_bw)
322 return dl_b->bw != -1 &&
323 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
327 * Verify the fitness of task @p to run on @cpu taking into account the
328 * CPU original capacity and the runtime/deadline ratio of the task.
330 * The function will return true if the CPU original capacity of the
331 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
332 * task and false otherwise.
334 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
336 unsigned long cap = arch_scale_cpu_capacity(cpu);
338 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
341 extern void init_dl_bw(struct dl_bw *dl_b);
342 extern int sched_dl_global_validate(void);
343 extern void sched_dl_do_global(void);
344 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
345 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
346 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
347 extern bool __checkparam_dl(const struct sched_attr *attr);
348 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
349 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
350 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
351 extern bool dl_cpu_busy(unsigned 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
370 s64 hierarchical_quota;
375 struct hrtimer period_timer;
376 struct hrtimer slack_timer;
377 struct list_head throttled_cfs_rq;
386 /* Task group related information */
388 struct cgroup_subsys_state css;
390 #ifdef CONFIG_FAIR_GROUP_SCHED
391 /* schedulable entities of this group on each CPU */
392 struct sched_entity **se;
393 /* runqueue "owned" by this group on each CPU */
394 struct cfs_rq **cfs_rq;
395 unsigned long shares;
399 * load_avg can be heavily contended at clock tick time, so put
400 * it in its own cacheline separated from the fields above which
401 * will also be accessed at each tick.
403 atomic_long_t load_avg ____cacheline_aligned;
407 #ifdef CONFIG_RT_GROUP_SCHED
408 struct sched_rt_entity **rt_se;
409 struct rt_rq **rt_rq;
411 struct rt_bandwidth rt_bandwidth;
415 struct list_head list;
417 struct task_group *parent;
418 struct list_head siblings;
419 struct list_head children;
421 #ifdef CONFIG_SCHED_AUTOGROUP
422 struct autogroup *autogroup;
425 struct cfs_bandwidth cfs_bandwidth;
427 #ifdef CONFIG_UCLAMP_TASK_GROUP
428 /* The two decimal precision [%] value requested from user-space */
429 unsigned int uclamp_pct[UCLAMP_CNT];
430 /* Clamp values requested for a task group */
431 struct uclamp_se uclamp_req[UCLAMP_CNT];
432 /* Effective clamp values used for a task group */
433 struct uclamp_se uclamp[UCLAMP_CNT];
438 #ifdef CONFIG_FAIR_GROUP_SCHED
439 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
442 * A weight of 0 or 1 can cause arithmetics problems.
443 * A weight of a cfs_rq is the sum of weights of which entities
444 * are queued on this cfs_rq, so a weight of a entity should not be
445 * too large, so as the shares value of a task group.
446 * (The default weight is 1024 - so there's no practical
447 * limitation from this.)
449 #define MIN_SHARES (1UL << 1)
450 #define MAX_SHARES (1UL << 18)
453 typedef int (*tg_visitor)(struct task_group *, void *);
455 extern int walk_tg_tree_from(struct task_group *from,
456 tg_visitor down, tg_visitor up, void *data);
459 * Iterate the full tree, calling @down when first entering a node and @up when
460 * leaving it for the final time.
462 * Caller must hold rcu_lock or sufficient equivalent.
464 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
466 return walk_tg_tree_from(&root_task_group, down, up, data);
469 extern int tg_nop(struct task_group *tg, void *data);
471 extern void free_fair_sched_group(struct task_group *tg);
472 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
473 extern void online_fair_sched_group(struct task_group *tg);
474 extern void unregister_fair_sched_group(struct task_group *tg);
475 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
476 struct sched_entity *se, int cpu,
477 struct sched_entity *parent);
478 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
480 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
481 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
482 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
484 extern void free_rt_sched_group(struct task_group *tg);
485 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
486 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
487 struct sched_rt_entity *rt_se, int cpu,
488 struct sched_rt_entity *parent);
489 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
490 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
491 extern long sched_group_rt_runtime(struct task_group *tg);
492 extern long sched_group_rt_period(struct task_group *tg);
493 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
495 extern struct task_group *sched_create_group(struct task_group *parent);
496 extern void sched_online_group(struct task_group *tg,
497 struct task_group *parent);
498 extern void sched_destroy_group(struct task_group *tg);
499 extern void sched_offline_group(struct task_group *tg);
501 extern void sched_move_task(struct task_struct *tsk);
503 #ifdef CONFIG_FAIR_GROUP_SCHED
504 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
507 extern void set_task_rq_fair(struct sched_entity *se,
508 struct cfs_rq *prev, struct cfs_rq *next);
509 #else /* !CONFIG_SMP */
510 static inline void set_task_rq_fair(struct sched_entity *se,
511 struct cfs_rq *prev, struct cfs_rq *next) { }
512 #endif /* CONFIG_SMP */
513 #endif /* CONFIG_FAIR_GROUP_SCHED */
515 #else /* CONFIG_CGROUP_SCHED */
517 struct cfs_bandwidth { };
519 #endif /* CONFIG_CGROUP_SCHED */
521 /* CFS-related fields in a runqueue */
523 struct load_weight load;
524 unsigned int nr_running;
525 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
526 unsigned int idle_h_nr_running; /* SCHED_IDLE */
530 #ifdef CONFIG_SCHED_CORE
531 unsigned int forceidle_seq;
536 u64 min_vruntime_copy;
539 struct rb_root_cached tasks_timeline;
542 * 'curr' points to currently running entity on this cfs_rq.
543 * It is set to NULL otherwise (i.e when none are currently running).
545 struct sched_entity *curr;
546 struct sched_entity *next;
547 struct sched_entity *last;
548 struct sched_entity *skip;
550 #ifdef CONFIG_SCHED_DEBUG
551 unsigned int nr_spread_over;
558 struct sched_avg avg;
560 u64 load_last_update_time_copy;
563 raw_spinlock_t lock ____cacheline_aligned;
565 unsigned long load_avg;
566 unsigned long util_avg;
567 unsigned long runnable_avg;
570 #ifdef CONFIG_FAIR_GROUP_SCHED
571 unsigned long tg_load_avg_contrib;
573 long prop_runnable_sum;
576 * h_load = weight * f(tg)
578 * Where f(tg) is the recursive weight fraction assigned to
581 unsigned long h_load;
582 u64 last_h_load_update;
583 struct sched_entity *h_load_next;
584 #endif /* CONFIG_FAIR_GROUP_SCHED */
585 #endif /* CONFIG_SMP */
587 #ifdef CONFIG_FAIR_GROUP_SCHED
588 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
591 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
592 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
593 * (like users, containers etc.)
595 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
596 * This list is used during load balance.
599 struct list_head leaf_cfs_rq_list;
600 struct task_group *tg; /* group that "owns" this runqueue */
602 #ifdef CONFIG_CFS_BANDWIDTH
604 s64 runtime_remaining;
607 u64 throttled_clock_task;
608 u64 throttled_clock_task_time;
611 struct list_head throttled_list;
612 #endif /* CONFIG_CFS_BANDWIDTH */
613 #endif /* CONFIG_FAIR_GROUP_SCHED */
616 static inline int rt_bandwidth_enabled(void)
618 return sysctl_sched_rt_runtime >= 0;
621 /* RT IPI pull logic requires IRQ_WORK */
622 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
623 # define HAVE_RT_PUSH_IPI
626 /* Real-Time classes' related field in a runqueue: */
628 struct rt_prio_array active;
629 unsigned int rt_nr_running;
630 unsigned int rr_nr_running;
631 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
633 int curr; /* highest queued rt task prio */
635 int next; /* next highest */
640 unsigned int rt_nr_migratory;
641 unsigned int rt_nr_total;
643 struct plist_head pushable_tasks;
645 #endif /* CONFIG_SMP */
651 /* Nests inside the rq lock: */
652 raw_spinlock_t rt_runtime_lock;
654 #ifdef CONFIG_RT_GROUP_SCHED
655 unsigned int rt_nr_boosted;
658 struct task_group *tg;
662 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
664 return rt_rq->rt_queued && rt_rq->rt_nr_running;
667 /* Deadline class' related fields in a runqueue */
669 /* runqueue is an rbtree, ordered by deadline */
670 struct rb_root_cached root;
672 unsigned int dl_nr_running;
676 * Deadline values of the currently executing and the
677 * earliest ready task on this rq. Caching these facilitates
678 * the decision whether or not a ready but not running task
679 * should migrate somewhere else.
686 unsigned int dl_nr_migratory;
690 * Tasks on this rq that can be pushed away. They are kept in
691 * an rb-tree, ordered by tasks' deadlines, with caching
692 * of the leftmost (earliest deadline) element.
694 struct rb_root_cached pushable_dl_tasks_root;
699 * "Active utilization" for this runqueue: increased when a
700 * task wakes up (becomes TASK_RUNNING) and decreased when a
706 * Utilization of the tasks "assigned" to this runqueue (including
707 * the tasks that are in runqueue and the tasks that executed on this
708 * CPU and blocked). Increased when a task moves to this runqueue, and
709 * decreased when the task moves away (migrates, changes scheduling
710 * policy, or terminates).
711 * This is needed to compute the "inactive utilization" for the
712 * runqueue (inactive utilization = this_bw - running_bw).
718 * Inverse of the fraction of CPU utilization that can be reclaimed
719 * by the GRUB algorithm.
724 #ifdef CONFIG_FAIR_GROUP_SCHED
725 /* An entity is a task if it doesn't "own" a runqueue */
726 #define entity_is_task(se) (!se->my_q)
728 static inline void se_update_runnable(struct sched_entity *se)
730 if (!entity_is_task(se))
731 se->runnable_weight = se->my_q->h_nr_running;
734 static inline long se_runnable(struct sched_entity *se)
736 if (entity_is_task(se))
739 return se->runnable_weight;
743 #define entity_is_task(se) 1
745 static inline void se_update_runnable(struct sched_entity *se) {}
747 static inline long se_runnable(struct sched_entity *se)
755 * XXX we want to get rid of these helpers and use the full load resolution.
757 static inline long se_weight(struct sched_entity *se)
759 return scale_load_down(se->load.weight);
763 static inline bool sched_asym_prefer(int a, int b)
765 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
769 struct em_perf_domain *em_pd;
770 struct perf_domain *next;
774 /* Scheduling group status flags */
775 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
776 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
779 * We add the notion of a root-domain which will be used to define per-domain
780 * variables. Each exclusive cpuset essentially defines an island domain by
781 * fully partitioning the member CPUs from any other cpuset. Whenever a new
782 * exclusive cpuset is created, we also create and attach a new root-domain
791 cpumask_var_t online;
794 * Indicate pullable load on at least one CPU, e.g:
795 * - More than one runnable task
796 * - Running task is misfit
800 /* Indicate one or more cpus over-utilized (tipping point) */
804 * The bit corresponding to a CPU gets set here if such CPU has more
805 * than one runnable -deadline task (as it is below for RT tasks).
807 cpumask_var_t dlo_mask;
813 * Indicate whether a root_domain's dl_bw has been checked or
814 * updated. It's monotonously increasing value.
816 * Also, some corner cases, like 'wrap around' is dangerous, but given
817 * that u64 is 'big enough'. So that shouldn't be a concern.
821 #ifdef HAVE_RT_PUSH_IPI
823 * For IPI pull requests, loop across the rto_mask.
825 struct irq_work rto_push_work;
826 raw_spinlock_t rto_lock;
827 /* These are only updated and read within rto_lock */
830 /* These atomics are updated outside of a lock */
831 atomic_t rto_loop_next;
832 atomic_t rto_loop_start;
835 * The "RT overload" flag: it gets set if a CPU has more than
836 * one runnable RT task.
838 cpumask_var_t rto_mask;
839 struct cpupri cpupri;
841 unsigned long max_cpu_capacity;
844 * NULL-terminated list of performance domains intersecting with the
845 * CPUs of the rd. Protected by RCU.
847 struct perf_domain __rcu *pd;
850 extern void init_defrootdomain(void);
851 extern int sched_init_domains(const struct cpumask *cpu_map);
852 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
853 extern void sched_get_rd(struct root_domain *rd);
854 extern void sched_put_rd(struct root_domain *rd);
856 #ifdef HAVE_RT_PUSH_IPI
857 extern void rto_push_irq_work_func(struct irq_work *work);
859 #endif /* CONFIG_SMP */
861 #ifdef CONFIG_UCLAMP_TASK
863 * struct uclamp_bucket - Utilization clamp bucket
864 * @value: utilization clamp value for tasks on this clamp bucket
865 * @tasks: number of RUNNABLE tasks on this clamp bucket
867 * Keep track of how many tasks are RUNNABLE for a given utilization
870 struct uclamp_bucket {
871 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
872 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
876 * struct uclamp_rq - rq's utilization clamp
877 * @value: currently active clamp values for a rq
878 * @bucket: utilization clamp buckets affecting a rq
880 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
881 * A clamp value is affecting a rq when there is at least one task RUNNABLE
882 * (or actually running) with that value.
884 * There are up to UCLAMP_CNT possible different clamp values, currently there
885 * are only two: minimum utilization and maximum utilization.
887 * All utilization clamping values are MAX aggregated, since:
888 * - for util_min: we want to run the CPU at least at the max of the minimum
889 * utilization required by its currently RUNNABLE tasks.
890 * - for util_max: we want to allow the CPU to run up to the max of the
891 * maximum utilization allowed by its currently RUNNABLE tasks.
893 * Since on each system we expect only a limited number of different
894 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
895 * the metrics required to compute all the per-rq utilization clamp values.
899 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
902 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
903 #endif /* CONFIG_UCLAMP_TASK */
906 * This is the main, per-CPU runqueue data structure.
908 * Locking rule: those places that want to lock multiple runqueues
909 * (such as the load balancing or the thread migration code), lock
910 * acquire operations must be ordered by ascending &runqueue.
914 raw_spinlock_t __lock;
917 * nr_running and cpu_load should be in the same cacheline because
918 * remote CPUs use both these fields when doing load calculation.
920 unsigned int nr_running;
921 #ifdef CONFIG_NUMA_BALANCING
922 unsigned int nr_numa_running;
923 unsigned int nr_preferred_running;
924 unsigned int numa_migrate_on;
926 #ifdef CONFIG_NO_HZ_COMMON
928 unsigned long last_blocked_load_update_tick;
929 unsigned int has_blocked_load;
930 call_single_data_t nohz_csd;
931 #endif /* CONFIG_SMP */
932 unsigned int nohz_tick_stopped;
934 #endif /* CONFIG_NO_HZ_COMMON */
937 unsigned int ttwu_pending;
941 #ifdef CONFIG_UCLAMP_TASK
942 /* Utilization clamp values based on CPU's RUNNABLE tasks */
943 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
944 unsigned int uclamp_flags;
945 #define UCLAMP_FLAG_IDLE 0x01
952 #ifdef CONFIG_FAIR_GROUP_SCHED
953 /* list of leaf cfs_rq on this CPU: */
954 struct list_head leaf_cfs_rq_list;
955 struct list_head *tmp_alone_branch;
956 #endif /* CONFIG_FAIR_GROUP_SCHED */
959 * This is part of a global counter where only the total sum
960 * over all CPUs matters. A task can increase this counter on
961 * one CPU and if it got migrated afterwards it may decrease
962 * it on another CPU. Always updated under the runqueue lock:
964 unsigned int nr_uninterruptible;
966 struct task_struct __rcu *curr;
967 struct task_struct *idle;
968 struct task_struct *stop;
969 unsigned long next_balance;
970 struct mm_struct *prev_mm;
972 unsigned int clock_update_flags;
974 /* Ensure that all clocks are in the same cache line */
975 u64 clock_task ____cacheline_aligned;
977 unsigned long lost_idle_time;
981 #ifdef CONFIG_SCHED_DEBUG
982 u64 last_seen_need_resched_ns;
983 int ticks_without_resched;
986 #ifdef CONFIG_MEMBARRIER
987 int membarrier_state;
991 struct root_domain *rd;
992 struct sched_domain __rcu *sd;
994 unsigned long cpu_capacity;
995 unsigned long cpu_capacity_orig;
997 struct callback_head *balance_callback;
999 unsigned char nohz_idle_balance;
1000 unsigned char idle_balance;
1002 unsigned long misfit_task_load;
1004 /* For active balancing */
1007 struct cpu_stop_work active_balance_work;
1009 /* CPU of this runqueue: */
1013 struct list_head cfs_tasks;
1015 struct sched_avg avg_rt;
1016 struct sched_avg avg_dl;
1017 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1018 struct sched_avg avg_irq;
1020 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1021 struct sched_avg avg_thermal;
1026 unsigned long wake_stamp;
1029 /* This is used to determine avg_idle's max value */
1030 u64 max_idle_balance_cost;
1032 #ifdef CONFIG_HOTPLUG_CPU
1033 struct rcuwait hotplug_wait;
1035 #endif /* CONFIG_SMP */
1037 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1040 #ifdef CONFIG_PARAVIRT
1041 u64 prev_steal_time;
1043 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1044 u64 prev_steal_time_rq;
1047 /* calc_load related fields */
1048 unsigned long calc_load_update;
1049 long calc_load_active;
1051 #ifdef CONFIG_SCHED_HRTICK
1053 call_single_data_t hrtick_csd;
1055 struct hrtimer hrtick_timer;
1056 ktime_t hrtick_time;
1059 #ifdef CONFIG_SCHEDSTATS
1061 struct sched_info rq_sched_info;
1062 unsigned long long rq_cpu_time;
1063 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1065 /* sys_sched_yield() stats */
1066 unsigned int yld_count;
1068 /* schedule() stats */
1069 unsigned int sched_count;
1070 unsigned int sched_goidle;
1072 /* try_to_wake_up() stats */
1073 unsigned int ttwu_count;
1074 unsigned int ttwu_local;
1077 #ifdef CONFIG_CPU_IDLE
1078 /* Must be inspected within a rcu lock section */
1079 struct cpuidle_state *idle_state;
1083 unsigned int nr_pinned;
1085 unsigned int push_busy;
1086 struct cpu_stop_work push_work;
1088 #ifdef CONFIG_SCHED_CORE
1091 struct task_struct *core_pick;
1092 unsigned int core_enabled;
1093 unsigned int core_sched_seq;
1094 struct rb_root core_tree;
1097 unsigned int core_task_seq;
1098 unsigned int core_pick_seq;
1099 unsigned long core_cookie;
1100 unsigned char core_forceidle;
1101 unsigned int core_forceidle_seq;
1105 #ifdef CONFIG_FAIR_GROUP_SCHED
1107 /* CPU runqueue to which this cfs_rq is attached */
1108 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1115 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1117 return container_of(cfs_rq, struct rq, cfs);
1121 static inline int cpu_of(struct rq *rq)
1130 #define MDF_PUSH 0x01
1132 static inline bool is_migration_disabled(struct task_struct *p)
1135 return p->migration_disabled;
1142 #ifdef CONFIG_SCHED_CORE
1143 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1145 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1147 static inline bool sched_core_enabled(struct rq *rq)
1149 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1152 static inline bool sched_core_disabled(void)
1154 return !static_branch_unlikely(&__sched_core_enabled);
1158 * Be careful with this function; not for general use. The return value isn't
1159 * stable unless you actually hold a relevant rq->__lock.
1161 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1163 if (sched_core_enabled(rq))
1164 return &rq->core->__lock;
1169 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1171 if (rq->core_enabled)
1172 return &rq->core->__lock;
1177 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1180 * Helpers to check if the CPU's core cookie matches with the task's cookie
1181 * when core scheduling is enabled.
1182 * A special case is that the task's cookie always matches with CPU's core
1183 * cookie if the CPU is in an idle core.
1185 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1187 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1188 if (!sched_core_enabled(rq))
1191 return rq->core->core_cookie == p->core_cookie;
1194 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1196 bool idle_core = true;
1199 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1200 if (!sched_core_enabled(rq))
1203 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1204 if (!available_idle_cpu(cpu)) {
1211 * A CPU in an idle core is always the best choice for tasks with
1214 return idle_core || rq->core->core_cookie == p->core_cookie;
1217 static inline bool sched_group_cookie_match(struct rq *rq,
1218 struct task_struct *p,
1219 struct sched_group *group)
1223 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1224 if (!sched_core_enabled(rq))
1227 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1228 if (sched_core_cookie_match(rq, p))
1234 extern void queue_core_balance(struct rq *rq);
1236 static inline bool sched_core_enqueued(struct task_struct *p)
1238 return !RB_EMPTY_NODE(&p->core_node);
1241 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1242 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1244 extern void sched_core_get(void);
1245 extern void sched_core_put(void);
1247 extern unsigned long sched_core_alloc_cookie(void);
1248 extern void sched_core_put_cookie(unsigned long cookie);
1249 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1250 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1252 #else /* !CONFIG_SCHED_CORE */
1254 static inline bool sched_core_enabled(struct rq *rq)
1259 static inline bool sched_core_disabled(void)
1264 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1269 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1274 static inline void queue_core_balance(struct rq *rq)
1278 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1283 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1288 static inline bool sched_group_cookie_match(struct rq *rq,
1289 struct task_struct *p,
1290 struct sched_group *group)
1294 #endif /* CONFIG_SCHED_CORE */
1296 static inline void lockdep_assert_rq_held(struct rq *rq)
1298 lockdep_assert_held(__rq_lockp(rq));
1301 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1302 extern bool raw_spin_rq_trylock(struct rq *rq);
1303 extern void raw_spin_rq_unlock(struct rq *rq);
1305 static inline void raw_spin_rq_lock(struct rq *rq)
1307 raw_spin_rq_lock_nested(rq, 0);
1310 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1312 local_irq_disable();
1313 raw_spin_rq_lock(rq);
1316 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1318 raw_spin_rq_unlock(rq);
1322 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1324 unsigned long flags;
1325 local_irq_save(flags);
1326 raw_spin_rq_lock(rq);
1330 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1332 raw_spin_rq_unlock(rq);
1333 local_irq_restore(flags);
1336 #define raw_spin_rq_lock_irqsave(rq, flags) \
1338 flags = _raw_spin_rq_lock_irqsave(rq); \
1341 #ifdef CONFIG_SCHED_SMT
1342 extern void __update_idle_core(struct rq *rq);
1344 static inline void update_idle_core(struct rq *rq)
1346 if (static_branch_unlikely(&sched_smt_present))
1347 __update_idle_core(rq);
1351 static inline void update_idle_core(struct rq *rq) { }
1354 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1356 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1357 #define this_rq() this_cpu_ptr(&runqueues)
1358 #define task_rq(p) cpu_rq(task_cpu(p))
1359 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1360 #define raw_rq() raw_cpu_ptr(&runqueues)
1362 #ifdef CONFIG_FAIR_GROUP_SCHED
1363 static inline struct task_struct *task_of(struct sched_entity *se)
1365 SCHED_WARN_ON(!entity_is_task(se));
1366 return container_of(se, struct task_struct, se);
1369 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1371 return p->se.cfs_rq;
1374 /* runqueue on which this entity is (to be) queued */
1375 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1380 /* runqueue "owned" by this group */
1381 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1388 static inline struct task_struct *task_of(struct sched_entity *se)
1390 return container_of(se, struct task_struct, se);
1393 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1395 return &task_rq(p)->cfs;
1398 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1400 struct task_struct *p = task_of(se);
1401 struct rq *rq = task_rq(p);
1406 /* runqueue "owned" by this group */
1407 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1413 extern void update_rq_clock(struct rq *rq);
1415 static inline u64 __rq_clock_broken(struct rq *rq)
1417 return READ_ONCE(rq->clock);
1421 * rq::clock_update_flags bits
1423 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1424 * call to __schedule(). This is an optimisation to avoid
1425 * neighbouring rq clock updates.
1427 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1428 * in effect and calls to update_rq_clock() are being ignored.
1430 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1431 * made to update_rq_clock() since the last time rq::lock was pinned.
1433 * If inside of __schedule(), clock_update_flags will have been
1434 * shifted left (a left shift is a cheap operation for the fast path
1435 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1437 * if (rq-clock_update_flags >= RQCF_UPDATED)
1439 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1440 * one position though, because the next rq_unpin_lock() will shift it
1443 #define RQCF_REQ_SKIP 0x01
1444 #define RQCF_ACT_SKIP 0x02
1445 #define RQCF_UPDATED 0x04
1447 static inline void assert_clock_updated(struct rq *rq)
1450 * The only reason for not seeing a clock update since the
1451 * last rq_pin_lock() is if we're currently skipping updates.
1453 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1456 static inline u64 rq_clock(struct rq *rq)
1458 lockdep_assert_rq_held(rq);
1459 assert_clock_updated(rq);
1464 static inline u64 rq_clock_task(struct rq *rq)
1466 lockdep_assert_rq_held(rq);
1467 assert_clock_updated(rq);
1469 return rq->clock_task;
1473 * By default the decay is the default pelt decay period.
1474 * The decay shift can change the decay period in
1476 * Decay shift Decay period(ms)
1483 extern int sched_thermal_decay_shift;
1485 static inline u64 rq_clock_thermal(struct rq *rq)
1487 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1490 static inline void rq_clock_skip_update(struct rq *rq)
1492 lockdep_assert_rq_held(rq);
1493 rq->clock_update_flags |= RQCF_REQ_SKIP;
1497 * See rt task throttling, which is the only time a skip
1498 * request is canceled.
1500 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1502 lockdep_assert_rq_held(rq);
1503 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1507 unsigned long flags;
1508 struct pin_cookie cookie;
1509 #ifdef CONFIG_SCHED_DEBUG
1511 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1512 * current pin context is stashed here in case it needs to be
1513 * restored in rq_repin_lock().
1515 unsigned int clock_update_flags;
1519 extern struct callback_head balance_push_callback;
1522 * Lockdep annotation that avoids accidental unlocks; it's like a
1523 * sticky/continuous lockdep_assert_held().
1525 * This avoids code that has access to 'struct rq *rq' (basically everything in
1526 * the scheduler) from accidentally unlocking the rq if they do not also have a
1527 * copy of the (on-stack) 'struct rq_flags rf'.
1529 * Also see Documentation/locking/lockdep-design.rst.
1531 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1533 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1535 #ifdef CONFIG_SCHED_DEBUG
1536 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1537 rf->clock_update_flags = 0;
1539 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1544 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1546 #ifdef CONFIG_SCHED_DEBUG
1547 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1548 rf->clock_update_flags = RQCF_UPDATED;
1551 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1554 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1556 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1558 #ifdef CONFIG_SCHED_DEBUG
1560 * Restore the value we stashed in @rf for this pin context.
1562 rq->clock_update_flags |= rf->clock_update_flags;
1566 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1567 __acquires(rq->lock);
1569 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1570 __acquires(p->pi_lock)
1571 __acquires(rq->lock);
1573 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1574 __releases(rq->lock)
1576 rq_unpin_lock(rq, rf);
1577 raw_spin_rq_unlock(rq);
1581 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1582 __releases(rq->lock)
1583 __releases(p->pi_lock)
1585 rq_unpin_lock(rq, rf);
1586 raw_spin_rq_unlock(rq);
1587 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1591 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1592 __acquires(rq->lock)
1594 raw_spin_rq_lock_irqsave(rq, rf->flags);
1595 rq_pin_lock(rq, rf);
1599 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1600 __acquires(rq->lock)
1602 raw_spin_rq_lock_irq(rq);
1603 rq_pin_lock(rq, rf);
1607 rq_lock(struct rq *rq, struct rq_flags *rf)
1608 __acquires(rq->lock)
1610 raw_spin_rq_lock(rq);
1611 rq_pin_lock(rq, rf);
1615 rq_relock(struct rq *rq, struct rq_flags *rf)
1616 __acquires(rq->lock)
1618 raw_spin_rq_lock(rq);
1619 rq_repin_lock(rq, rf);
1623 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1624 __releases(rq->lock)
1626 rq_unpin_lock(rq, rf);
1627 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1631 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1632 __releases(rq->lock)
1634 rq_unpin_lock(rq, rf);
1635 raw_spin_rq_unlock_irq(rq);
1639 rq_unlock(struct rq *rq, struct rq_flags *rf)
1640 __releases(rq->lock)
1642 rq_unpin_lock(rq, rf);
1643 raw_spin_rq_unlock(rq);
1646 static inline struct rq *
1647 this_rq_lock_irq(struct rq_flags *rf)
1648 __acquires(rq->lock)
1652 local_irq_disable();
1659 enum numa_topology_type {
1664 extern enum numa_topology_type sched_numa_topology_type;
1665 extern int sched_max_numa_distance;
1666 extern bool find_numa_distance(int distance);
1667 extern void sched_init_numa(void);
1668 extern void sched_domains_numa_masks_set(unsigned int cpu);
1669 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1670 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1672 static inline void sched_init_numa(void) { }
1673 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1674 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1675 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1681 #ifdef CONFIG_NUMA_BALANCING
1682 /* The regions in numa_faults array from task_struct */
1683 enum numa_faults_stats {
1689 extern void sched_setnuma(struct task_struct *p, int node);
1690 extern int migrate_task_to(struct task_struct *p, int cpu);
1691 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1693 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1696 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1699 #endif /* CONFIG_NUMA_BALANCING */
1704 queue_balance_callback(struct rq *rq,
1705 struct callback_head *head,
1706 void (*func)(struct rq *rq))
1708 lockdep_assert_rq_held(rq);
1710 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1713 head->func = (void (*)(struct callback_head *))func;
1714 head->next = rq->balance_callback;
1715 rq->balance_callback = head;
1718 #define rcu_dereference_check_sched_domain(p) \
1719 rcu_dereference_check((p), \
1720 lockdep_is_held(&sched_domains_mutex))
1723 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1724 * See destroy_sched_domains: call_rcu for details.
1726 * The domain tree of any CPU may only be accessed from within
1727 * preempt-disabled sections.
1729 #define for_each_domain(cpu, __sd) \
1730 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1731 __sd; __sd = __sd->parent)
1734 * highest_flag_domain - Return highest sched_domain containing flag.
1735 * @cpu: The CPU whose highest level of sched domain is to
1737 * @flag: The flag to check for the highest sched_domain
1738 * for the given CPU.
1740 * Returns the highest sched_domain of a CPU which contains the given flag.
1742 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1744 struct sched_domain *sd, *hsd = NULL;
1746 for_each_domain(cpu, sd) {
1747 if (!(sd->flags & flag))
1755 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1757 struct sched_domain *sd;
1759 for_each_domain(cpu, sd) {
1760 if (sd->flags & flag)
1767 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1768 DECLARE_PER_CPU(int, sd_llc_size);
1769 DECLARE_PER_CPU(int, sd_llc_id);
1770 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1771 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1772 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1773 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1774 extern struct static_key_false sched_asym_cpucapacity;
1776 struct sched_group_capacity {
1779 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1782 unsigned long capacity;
1783 unsigned long min_capacity; /* Min per-CPU capacity in group */
1784 unsigned long max_capacity; /* Max per-CPU capacity in group */
1785 unsigned long next_update;
1786 int imbalance; /* XXX unrelated to capacity but shared group state */
1788 #ifdef CONFIG_SCHED_DEBUG
1792 unsigned long cpumask[]; /* Balance mask */
1795 struct sched_group {
1796 struct sched_group *next; /* Must be a circular list */
1799 unsigned int group_weight;
1800 struct sched_group_capacity *sgc;
1801 int asym_prefer_cpu; /* CPU of highest priority in group */
1804 * The CPUs this group covers.
1806 * NOTE: this field is variable length. (Allocated dynamically
1807 * by attaching extra space to the end of the structure,
1808 * depending on how many CPUs the kernel has booted up with)
1810 unsigned long cpumask[];
1813 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1815 return to_cpumask(sg->cpumask);
1819 * See build_balance_mask().
1821 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1823 return to_cpumask(sg->sgc->cpumask);
1827 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1828 * @group: The group whose first CPU is to be returned.
1830 static inline unsigned int group_first_cpu(struct sched_group *group)
1832 return cpumask_first(sched_group_span(group));
1835 extern int group_balance_cpu(struct sched_group *sg);
1837 #ifdef CONFIG_SCHED_DEBUG
1838 void update_sched_domain_debugfs(void);
1839 void dirty_sched_domain_sysctl(int cpu);
1841 static inline void update_sched_domain_debugfs(void)
1844 static inline void dirty_sched_domain_sysctl(int cpu)
1849 extern int sched_update_scaling(void);
1851 extern void flush_smp_call_function_from_idle(void);
1853 #else /* !CONFIG_SMP: */
1854 static inline void flush_smp_call_function_from_idle(void) { }
1858 #include "autogroup.h"
1860 #ifdef CONFIG_CGROUP_SCHED
1863 * Return the group to which this tasks belongs.
1865 * We cannot use task_css() and friends because the cgroup subsystem
1866 * changes that value before the cgroup_subsys::attach() method is called,
1867 * therefore we cannot pin it and might observe the wrong value.
1869 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1870 * core changes this before calling sched_move_task().
1872 * Instead we use a 'copy' which is updated from sched_move_task() while
1873 * holding both task_struct::pi_lock and rq::lock.
1875 static inline struct task_group *task_group(struct task_struct *p)
1877 return p->sched_task_group;
1880 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1881 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1883 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1884 struct task_group *tg = task_group(p);
1887 #ifdef CONFIG_FAIR_GROUP_SCHED
1888 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1889 p->se.cfs_rq = tg->cfs_rq[cpu];
1890 p->se.parent = tg->se[cpu];
1893 #ifdef CONFIG_RT_GROUP_SCHED
1894 p->rt.rt_rq = tg->rt_rq[cpu];
1895 p->rt.parent = tg->rt_se[cpu];
1899 #else /* CONFIG_CGROUP_SCHED */
1901 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1902 static inline struct task_group *task_group(struct task_struct *p)
1907 #endif /* CONFIG_CGROUP_SCHED */
1909 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1911 set_task_rq(p, cpu);
1914 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1915 * successfully executed on another CPU. We must ensure that updates of
1916 * per-task data have been completed by this moment.
1919 #ifdef CONFIG_THREAD_INFO_IN_TASK
1920 WRITE_ONCE(p->cpu, cpu);
1922 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1929 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1931 #ifdef CONFIG_SCHED_DEBUG
1932 # include <linux/static_key.h>
1933 # define const_debug __read_mostly
1935 # define const_debug const
1938 #define SCHED_FEAT(name, enabled) \
1939 __SCHED_FEAT_##name ,
1942 #include "features.h"
1948 #ifdef CONFIG_SCHED_DEBUG
1951 * To support run-time toggling of sched features, all the translation units
1952 * (but core.c) reference the sysctl_sched_features defined in core.c.
1954 extern const_debug unsigned int sysctl_sched_features;
1956 #ifdef CONFIG_JUMP_LABEL
1957 #define SCHED_FEAT(name, enabled) \
1958 static __always_inline bool static_branch_##name(struct static_key *key) \
1960 return static_key_##enabled(key); \
1963 #include "features.h"
1966 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1967 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1969 #else /* !CONFIG_JUMP_LABEL */
1971 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1973 #endif /* CONFIG_JUMP_LABEL */
1975 #else /* !SCHED_DEBUG */
1978 * Each translation unit has its own copy of sysctl_sched_features to allow
1979 * constants propagation at compile time and compiler optimization based on
1982 #define SCHED_FEAT(name, enabled) \
1983 (1UL << __SCHED_FEAT_##name) * enabled |
1984 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1985 #include "features.h"
1989 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1991 #endif /* SCHED_DEBUG */
1993 extern struct static_key_false sched_numa_balancing;
1994 extern struct static_key_false sched_schedstats;
1996 static inline u64 global_rt_period(void)
1998 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2001 static inline u64 global_rt_runtime(void)
2003 if (sysctl_sched_rt_runtime < 0)
2006 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2009 static inline int task_current(struct rq *rq, struct task_struct *p)
2011 return rq->curr == p;
2014 static inline int task_running(struct rq *rq, struct task_struct *p)
2019 return task_current(rq, p);
2023 static inline int task_on_rq_queued(struct task_struct *p)
2025 return p->on_rq == TASK_ON_RQ_QUEUED;
2028 static inline int task_on_rq_migrating(struct task_struct *p)
2030 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2033 /* Wake flags. The first three directly map to some SD flag value */
2034 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2035 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2036 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2038 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2039 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2040 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2043 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2044 static_assert(WF_FORK == SD_BALANCE_FORK);
2045 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2049 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2050 * of tasks with abnormal "nice" values across CPUs the contribution that
2051 * each task makes to its run queue's load is weighted according to its
2052 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2053 * scaled version of the new time slice allocation that they receive on time
2057 #define WEIGHT_IDLEPRIO 3
2058 #define WMULT_IDLEPRIO 1431655765
2060 extern const int sched_prio_to_weight[40];
2061 extern const u32 sched_prio_to_wmult[40];
2064 * {de,en}queue flags:
2066 * DEQUEUE_SLEEP - task is no longer runnable
2067 * ENQUEUE_WAKEUP - task just became runnable
2069 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2070 * are in a known state which allows modification. Such pairs
2071 * should preserve as much state as possible.
2073 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2076 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2077 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2078 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2082 #define DEQUEUE_SLEEP 0x01
2083 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2084 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2085 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2087 #define ENQUEUE_WAKEUP 0x01
2088 #define ENQUEUE_RESTORE 0x02
2089 #define ENQUEUE_MOVE 0x04
2090 #define ENQUEUE_NOCLOCK 0x08
2092 #define ENQUEUE_HEAD 0x10
2093 #define ENQUEUE_REPLENISH 0x20
2095 #define ENQUEUE_MIGRATED 0x40
2097 #define ENQUEUE_MIGRATED 0x00
2100 #define RETRY_TASK ((void *)-1UL)
2102 struct sched_class {
2104 #ifdef CONFIG_UCLAMP_TASK
2108 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2109 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2110 void (*yield_task) (struct rq *rq);
2111 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2113 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2115 struct task_struct *(*pick_next_task)(struct rq *rq);
2117 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2118 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2121 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2122 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2124 struct task_struct * (*pick_task)(struct rq *rq);
2126 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2128 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2130 void (*set_cpus_allowed)(struct task_struct *p,
2131 const struct cpumask *newmask,
2134 void (*rq_online)(struct rq *rq);
2135 void (*rq_offline)(struct rq *rq);
2137 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2140 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2141 void (*task_fork)(struct task_struct *p);
2142 void (*task_dead)(struct task_struct *p);
2145 * The switched_from() call is allowed to drop rq->lock, therefore we
2146 * cannot assume the switched_from/switched_to pair is serialized by
2147 * rq->lock. They are however serialized by p->pi_lock.
2149 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2150 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2151 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2154 unsigned int (*get_rr_interval)(struct rq *rq,
2155 struct task_struct *task);
2157 void (*update_curr)(struct rq *rq);
2159 #define TASK_SET_GROUP 0
2160 #define TASK_MOVE_GROUP 1
2162 #ifdef CONFIG_FAIR_GROUP_SCHED
2163 void (*task_change_group)(struct task_struct *p, int type);
2167 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2169 WARN_ON_ONCE(rq->curr != prev);
2170 prev->sched_class->put_prev_task(rq, prev);
2173 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2175 next->sched_class->set_next_task(rq, next, false);
2180 * Helper to define a sched_class instance; each one is placed in a separate
2181 * section which is ordered by the linker script:
2183 * include/asm-generic/vmlinux.lds.h
2185 * Also enforce alignment on the instance, not the type, to guarantee layout.
2187 #define DEFINE_SCHED_CLASS(name) \
2188 const struct sched_class name##_sched_class \
2189 __aligned(__alignof__(struct sched_class)) \
2190 __section("__" #name "_sched_class")
2192 /* Defined in include/asm-generic/vmlinux.lds.h */
2193 extern struct sched_class __begin_sched_classes[];
2194 extern struct sched_class __end_sched_classes[];
2196 #define sched_class_highest (__end_sched_classes - 1)
2197 #define sched_class_lowest (__begin_sched_classes - 1)
2199 #define for_class_range(class, _from, _to) \
2200 for (class = (_from); class != (_to); class--)
2202 #define for_each_class(class) \
2203 for_class_range(class, sched_class_highest, sched_class_lowest)
2205 extern const struct sched_class stop_sched_class;
2206 extern const struct sched_class dl_sched_class;
2207 extern const struct sched_class rt_sched_class;
2208 extern const struct sched_class fair_sched_class;
2209 extern const struct sched_class idle_sched_class;
2211 static inline bool sched_stop_runnable(struct rq *rq)
2213 return rq->stop && task_on_rq_queued(rq->stop);
2216 static inline bool sched_dl_runnable(struct rq *rq)
2218 return rq->dl.dl_nr_running > 0;
2221 static inline bool sched_rt_runnable(struct rq *rq)
2223 return rq->rt.rt_queued > 0;
2226 static inline bool sched_fair_runnable(struct rq *rq)
2228 return rq->cfs.nr_running > 0;
2231 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2232 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2234 #define SCA_CHECK 0x01
2235 #define SCA_MIGRATE_DISABLE 0x02
2236 #define SCA_MIGRATE_ENABLE 0x04
2240 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2242 extern void trigger_load_balance(struct rq *rq);
2244 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2246 static inline struct task_struct *get_push_task(struct rq *rq)
2248 struct task_struct *p = rq->curr;
2250 lockdep_assert_rq_held(rq);
2255 if (p->nr_cpus_allowed == 1)
2258 rq->push_busy = true;
2259 return get_task_struct(p);
2262 extern int push_cpu_stop(void *arg);
2266 #ifdef CONFIG_CPU_IDLE
2267 static inline void idle_set_state(struct rq *rq,
2268 struct cpuidle_state *idle_state)
2270 rq->idle_state = idle_state;
2273 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2275 SCHED_WARN_ON(!rcu_read_lock_held());
2277 return rq->idle_state;
2280 static inline void idle_set_state(struct rq *rq,
2281 struct cpuidle_state *idle_state)
2285 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2291 extern void schedule_idle(void);
2293 extern void sysrq_sched_debug_show(void);
2294 extern void sched_init_granularity(void);
2295 extern void update_max_interval(void);
2297 extern void init_sched_dl_class(void);
2298 extern void init_sched_rt_class(void);
2299 extern void init_sched_fair_class(void);
2301 extern void reweight_task(struct task_struct *p, int prio);
2303 extern void resched_curr(struct rq *rq);
2304 extern void resched_cpu(int cpu);
2306 extern struct rt_bandwidth def_rt_bandwidth;
2307 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2309 extern struct dl_bandwidth def_dl_bandwidth;
2310 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2311 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2312 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2315 #define BW_UNIT (1 << BW_SHIFT)
2316 #define RATIO_SHIFT 8
2317 #define MAX_BW_BITS (64 - BW_SHIFT)
2318 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2319 unsigned long to_ratio(u64 period, u64 runtime);
2321 extern void init_entity_runnable_average(struct sched_entity *se);
2322 extern void post_init_entity_util_avg(struct task_struct *p);
2324 #ifdef CONFIG_NO_HZ_FULL
2325 extern bool sched_can_stop_tick(struct rq *rq);
2326 extern int __init sched_tick_offload_init(void);
2329 * Tick may be needed by tasks in the runqueue depending on their policy and
2330 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2331 * nohz mode if necessary.
2333 static inline void sched_update_tick_dependency(struct rq *rq)
2335 int cpu = cpu_of(rq);
2337 if (!tick_nohz_full_cpu(cpu))
2340 if (sched_can_stop_tick(rq))
2341 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2343 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2346 static inline int sched_tick_offload_init(void) { return 0; }
2347 static inline void sched_update_tick_dependency(struct rq *rq) { }
2350 static inline void add_nr_running(struct rq *rq, unsigned count)
2352 unsigned prev_nr = rq->nr_running;
2354 rq->nr_running = prev_nr + count;
2355 if (trace_sched_update_nr_running_tp_enabled()) {
2356 call_trace_sched_update_nr_running(rq, count);
2360 if (prev_nr < 2 && rq->nr_running >= 2) {
2361 if (!READ_ONCE(rq->rd->overload))
2362 WRITE_ONCE(rq->rd->overload, 1);
2366 sched_update_tick_dependency(rq);
2369 static inline void sub_nr_running(struct rq *rq, unsigned count)
2371 rq->nr_running -= count;
2372 if (trace_sched_update_nr_running_tp_enabled()) {
2373 call_trace_sched_update_nr_running(rq, -count);
2376 /* Check if we still need preemption */
2377 sched_update_tick_dependency(rq);
2380 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2381 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2383 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2385 extern const_debug unsigned int sysctl_sched_nr_migrate;
2386 extern const_debug unsigned int sysctl_sched_migration_cost;
2388 #ifdef CONFIG_SCHED_HRTICK
2392 * - enabled by features
2393 * - hrtimer is actually high res
2395 static inline int hrtick_enabled(struct rq *rq)
2397 if (!cpu_active(cpu_of(rq)))
2399 return hrtimer_is_hres_active(&rq->hrtick_timer);
2402 static inline int hrtick_enabled_fair(struct rq *rq)
2404 if (!sched_feat(HRTICK))
2406 return hrtick_enabled(rq);
2409 static inline int hrtick_enabled_dl(struct rq *rq)
2411 if (!sched_feat(HRTICK_DL))
2413 return hrtick_enabled(rq);
2416 void hrtick_start(struct rq *rq, u64 delay);
2420 static inline int hrtick_enabled_fair(struct rq *rq)
2425 static inline int hrtick_enabled_dl(struct rq *rq)
2430 static inline int hrtick_enabled(struct rq *rq)
2435 #endif /* CONFIG_SCHED_HRTICK */
2437 #ifndef arch_scale_freq_tick
2438 static __always_inline
2439 void arch_scale_freq_tick(void)
2444 #ifndef arch_scale_freq_capacity
2446 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2447 * @cpu: the CPU in question.
2449 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2452 * ------ * SCHED_CAPACITY_SCALE
2455 static __always_inline
2456 unsigned long arch_scale_freq_capacity(int cpu)
2458 return SCHED_CAPACITY_SCALE;
2465 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2467 #ifdef CONFIG_SCHED_CORE
2469 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2470 * order by core-id first and cpu-id second.
2474 * double_rq_lock(0,3); will take core-0, core-1 lock
2475 * double_rq_lock(1,2); will take core-1, core-0 lock
2477 * when only cpu-id is considered.
2479 if (rq1->core->cpu < rq2->core->cpu)
2481 if (rq1->core->cpu > rq2->core->cpu)
2485 * __sched_core_flip() relies on SMT having cpu-id lock order.
2488 return rq1->cpu < rq2->cpu;
2491 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2493 #ifdef CONFIG_PREEMPTION
2496 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2497 * way at the expense of forcing extra atomic operations in all
2498 * invocations. This assures that the double_lock is acquired using the
2499 * same underlying policy as the spinlock_t on this architecture, which
2500 * reduces latency compared to the unfair variant below. However, it
2501 * also adds more overhead and therefore may reduce throughput.
2503 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2504 __releases(this_rq->lock)
2505 __acquires(busiest->lock)
2506 __acquires(this_rq->lock)
2508 raw_spin_rq_unlock(this_rq);
2509 double_rq_lock(this_rq, busiest);
2516 * Unfair double_lock_balance: Optimizes throughput at the expense of
2517 * latency by eliminating extra atomic operations when the locks are
2518 * already in proper order on entry. This favors lower CPU-ids and will
2519 * grant the double lock to lower CPUs over higher ids under contention,
2520 * regardless of entry order into the function.
2522 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2523 __releases(this_rq->lock)
2524 __acquires(busiest->lock)
2525 __acquires(this_rq->lock)
2527 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2530 if (likely(raw_spin_rq_trylock(busiest)))
2533 if (rq_order_less(this_rq, busiest)) {
2534 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2538 raw_spin_rq_unlock(this_rq);
2539 double_rq_lock(this_rq, busiest);
2544 #endif /* CONFIG_PREEMPTION */
2547 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2549 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2551 lockdep_assert_irqs_disabled();
2553 return _double_lock_balance(this_rq, busiest);
2556 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2557 __releases(busiest->lock)
2559 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2560 raw_spin_rq_unlock(busiest);
2561 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2564 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2570 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2573 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2579 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2582 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2588 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2592 * double_rq_unlock - safely unlock two runqueues
2594 * Note this does not restore interrupts like task_rq_unlock,
2595 * you need to do so manually after calling.
2597 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2598 __releases(rq1->lock)
2599 __releases(rq2->lock)
2601 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2602 raw_spin_rq_unlock(rq2);
2604 __release(rq2->lock);
2605 raw_spin_rq_unlock(rq1);
2608 extern void set_rq_online (struct rq *rq);
2609 extern void set_rq_offline(struct rq *rq);
2610 extern bool sched_smp_initialized;
2612 #else /* CONFIG_SMP */
2615 * double_rq_lock - safely lock two runqueues
2617 * Note this does not disable interrupts like task_rq_lock,
2618 * you need to do so manually before calling.
2620 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2621 __acquires(rq1->lock)
2622 __acquires(rq2->lock)
2624 BUG_ON(!irqs_disabled());
2626 raw_spin_rq_lock(rq1);
2627 __acquire(rq2->lock); /* Fake it out ;) */
2631 * double_rq_unlock - safely unlock two runqueues
2633 * Note this does not restore interrupts like task_rq_unlock,
2634 * you need to do so manually after calling.
2636 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2637 __releases(rq1->lock)
2638 __releases(rq2->lock)
2641 raw_spin_rq_unlock(rq1);
2642 __release(rq2->lock);
2647 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2648 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2650 #ifdef CONFIG_SCHED_DEBUG
2651 extern bool sched_debug_verbose;
2653 extern void print_cfs_stats(struct seq_file *m, int cpu);
2654 extern void print_rt_stats(struct seq_file *m, int cpu);
2655 extern void print_dl_stats(struct seq_file *m, int cpu);
2656 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2657 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2658 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2660 extern void resched_latency_warn(int cpu, u64 latency);
2661 #ifdef CONFIG_NUMA_BALANCING
2663 show_numa_stats(struct task_struct *p, struct seq_file *m);
2665 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2666 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2667 #endif /* CONFIG_NUMA_BALANCING */
2669 static inline void resched_latency_warn(int cpu, u64 latency) {}
2670 #endif /* CONFIG_SCHED_DEBUG */
2672 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2673 extern void init_rt_rq(struct rt_rq *rt_rq);
2674 extern void init_dl_rq(struct dl_rq *dl_rq);
2676 extern void cfs_bandwidth_usage_inc(void);
2677 extern void cfs_bandwidth_usage_dec(void);
2679 #ifdef CONFIG_NO_HZ_COMMON
2680 #define NOHZ_BALANCE_KICK_BIT 0
2681 #define NOHZ_STATS_KICK_BIT 1
2682 #define NOHZ_NEWILB_KICK_BIT 2
2684 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2685 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2686 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2688 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2690 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2692 extern void nohz_balance_exit_idle(struct rq *rq);
2694 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2697 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2698 extern void nohz_run_idle_balance(int cpu);
2700 static inline void nohz_run_idle_balance(int cpu) { }
2705 void __dl_update(struct dl_bw *dl_b, s64 bw)
2707 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2710 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2711 "sched RCU must be held");
2712 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2713 struct rq *rq = cpu_rq(i);
2715 rq->dl.extra_bw += bw;
2720 void __dl_update(struct dl_bw *dl_b, s64 bw)
2722 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2729 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2734 struct u64_stats_sync sync;
2737 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2740 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2741 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2742 * and never move forward.
2744 static inline u64 irq_time_read(int cpu)
2746 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2751 seq = __u64_stats_fetch_begin(&irqtime->sync);
2752 total = irqtime->total;
2753 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2757 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2759 #ifdef CONFIG_CPU_FREQ
2760 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2763 * cpufreq_update_util - Take a note about CPU utilization changes.
2764 * @rq: Runqueue to carry out the update for.
2765 * @flags: Update reason flags.
2767 * This function is called by the scheduler on the CPU whose utilization is
2770 * It can only be called from RCU-sched read-side critical sections.
2772 * The way cpufreq is currently arranged requires it to evaluate the CPU
2773 * performance state (frequency/voltage) on a regular basis to prevent it from
2774 * being stuck in a completely inadequate performance level for too long.
2775 * That is not guaranteed to happen if the updates are only triggered from CFS
2776 * and DL, though, because they may not be coming in if only RT tasks are
2777 * active all the time (or there are RT tasks only).
2779 * As a workaround for that issue, this function is called periodically by the
2780 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2781 * but that really is a band-aid. Going forward it should be replaced with
2782 * solutions targeted more specifically at RT tasks.
2784 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2786 struct update_util_data *data;
2788 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2791 data->func(data, rq_clock(rq), flags);
2794 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2795 #endif /* CONFIG_CPU_FREQ */
2797 #ifdef CONFIG_UCLAMP_TASK
2798 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2801 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2802 * @rq: The rq to clamp against. Must not be NULL.
2803 * @util: The util value to clamp.
2804 * @p: The task to clamp against. Can be NULL if you want to clamp
2807 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2809 * If sched_uclamp_used static key is disabled, then just return the util
2810 * without any clamping since uclamp aggregation at the rq level in the fast
2811 * path is disabled, rendering this operation a NOP.
2813 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2814 * will return the correct effective uclamp value of the task even if the
2815 * static key is disabled.
2817 static __always_inline
2818 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2819 struct task_struct *p)
2821 unsigned long min_util = 0;
2822 unsigned long max_util = 0;
2824 if (!static_branch_likely(&sched_uclamp_used))
2828 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2829 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2832 * Ignore last runnable task's max clamp, as this task will
2833 * reset it. Similarly, no need to read the rq's min clamp.
2835 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2839 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2840 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2843 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2844 * RUNNABLE tasks with _different_ clamps, we can end up with an
2845 * inversion. Fix it now when the clamps are applied.
2847 if (unlikely(min_util >= max_util))
2850 return clamp(util, min_util, max_util);
2854 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2855 * by default in the fast path and only gets turned on once userspace performs
2856 * an operation that requires it.
2858 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2861 static inline bool uclamp_is_used(void)
2863 return static_branch_likely(&sched_uclamp_used);
2865 #else /* CONFIG_UCLAMP_TASK */
2867 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2868 struct task_struct *p)
2873 static inline bool uclamp_is_used(void)
2877 #endif /* CONFIG_UCLAMP_TASK */
2879 #ifdef arch_scale_freq_capacity
2880 # ifndef arch_scale_freq_invariant
2881 # define arch_scale_freq_invariant() true
2884 # define arch_scale_freq_invariant() false
2888 static inline unsigned long capacity_orig_of(int cpu)
2890 return cpu_rq(cpu)->cpu_capacity_orig;
2894 * enum cpu_util_type - CPU utilization type
2895 * @FREQUENCY_UTIL: Utilization used to select frequency
2896 * @ENERGY_UTIL: Utilization used during energy calculation
2898 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2899 * need to be aggregated differently depending on the usage made of them. This
2900 * enum is used within effective_cpu_util() to differentiate the types of
2901 * utilization expected by the callers, and adjust the aggregation accordingly.
2903 enum cpu_util_type {
2908 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2909 unsigned long max, enum cpu_util_type type,
2910 struct task_struct *p);
2912 static inline unsigned long cpu_bw_dl(struct rq *rq)
2914 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2917 static inline unsigned long cpu_util_dl(struct rq *rq)
2919 return READ_ONCE(rq->avg_dl.util_avg);
2922 static inline unsigned long cpu_util_cfs(struct rq *rq)
2924 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2926 if (sched_feat(UTIL_EST)) {
2927 util = max_t(unsigned long, util,
2928 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2934 static inline unsigned long cpu_util_rt(struct rq *rq)
2936 return READ_ONCE(rq->avg_rt.util_avg);
2940 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2941 static inline unsigned long cpu_util_irq(struct rq *rq)
2943 return rq->avg_irq.util_avg;
2947 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2949 util *= (max - irq);
2956 static inline unsigned long cpu_util_irq(struct rq *rq)
2962 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2968 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2970 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2972 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2974 static inline bool sched_energy_enabled(void)
2976 return static_branch_unlikely(&sched_energy_present);
2979 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2981 #define perf_domain_span(pd) NULL
2982 static inline bool sched_energy_enabled(void) { return false; }
2984 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2986 #ifdef CONFIG_MEMBARRIER
2988 * The scheduler provides memory barriers required by membarrier between:
2989 * - prior user-space memory accesses and store to rq->membarrier_state,
2990 * - store to rq->membarrier_state and following user-space memory accesses.
2991 * In the same way it provides those guarantees around store to rq->curr.
2993 static inline void membarrier_switch_mm(struct rq *rq,
2994 struct mm_struct *prev_mm,
2995 struct mm_struct *next_mm)
2997 int membarrier_state;
2999 if (prev_mm == next_mm)
3002 membarrier_state = atomic_read(&next_mm->membarrier_state);
3003 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3006 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3009 static inline void membarrier_switch_mm(struct rq *rq,
3010 struct mm_struct *prev_mm,
3011 struct mm_struct *next_mm)
3017 static inline bool is_per_cpu_kthread(struct task_struct *p)
3019 if (!(p->flags & PF_KTHREAD))
3022 if (p->nr_cpus_allowed != 1)
3029 extern void swake_up_all_locked(struct swait_queue_head *q);
3030 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3032 #ifdef CONFIG_PREEMPT_DYNAMIC
3033 extern int preempt_dynamic_mode;
3034 extern int sched_dynamic_mode(const char *str);
3035 extern void sched_dynamic_update(int mode);