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 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
232 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
234 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
235 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
242 * Tells if entity @a should preempt entity @b.
245 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
247 return dl_entity_is_special(a) ||
248 dl_time_before(a->deadline, b->deadline);
252 * This is the priority-queue data structure of the RT scheduling class:
254 struct rt_prio_array {
255 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
256 struct list_head queue[MAX_RT_PRIO];
259 struct rt_bandwidth {
260 /* nests inside the rq lock: */
261 raw_spinlock_t rt_runtime_lock;
264 struct hrtimer rt_period_timer;
265 unsigned int rt_period_active;
268 void __dl_clear_params(struct task_struct *p);
270 struct dl_bandwidth {
271 raw_spinlock_t dl_runtime_lock;
276 static inline int dl_bandwidth_enabled(void)
278 return sysctl_sched_rt_runtime >= 0;
282 * To keep the bandwidth of -deadline tasks under control
283 * we need some place where:
284 * - store the maximum -deadline bandwidth of each cpu;
285 * - cache the fraction of bandwidth that is currently allocated in
288 * This is all done in the data structure below. It is similar to the
289 * one used for RT-throttling (rt_bandwidth), with the main difference
290 * that, since here we are only interested in admission control, we
291 * do not decrease any runtime while the group "executes", neither we
292 * need a timer to replenish it.
294 * With respect to SMP, bandwidth is given on a per root domain basis,
296 * - bw (< 100%) is the deadline bandwidth of each CPU;
297 * - total_bw is the currently allocated bandwidth in each root domain;
305 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
308 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
310 dl_b->total_bw -= tsk_bw;
311 __dl_update(dl_b, (s32)tsk_bw / cpus);
315 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
317 dl_b->total_bw += tsk_bw;
318 __dl_update(dl_b, -((s32)tsk_bw / cpus));
321 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
322 u64 old_bw, u64 new_bw)
324 return dl_b->bw != -1 &&
325 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
329 * Verify the fitness of task @p to run on @cpu taking into account the
330 * CPU original capacity and the runtime/deadline ratio of the task.
332 * The function will return true if the CPU original capacity of the
333 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
334 * task and false otherwise.
336 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
338 unsigned long cap = arch_scale_cpu_capacity(cpu);
340 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
343 extern void init_dl_bw(struct dl_bw *dl_b);
344 extern int sched_dl_global_validate(void);
345 extern void sched_dl_do_global(void);
346 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
347 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
348 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
349 extern bool __checkparam_dl(const struct sched_attr *attr);
350 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
351 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
352 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
353 extern bool dl_cpu_busy(unsigned int cpu);
355 #ifdef CONFIG_CGROUP_SCHED
357 #include <linux/cgroup.h>
358 #include <linux/psi.h>
363 extern struct list_head task_groups;
365 struct cfs_bandwidth {
366 #ifdef CONFIG_CFS_BANDWIDTH
372 s64 hierarchical_quota;
377 struct hrtimer period_timer;
378 struct hrtimer slack_timer;
379 struct list_head throttled_cfs_rq;
388 /* Task group related information */
390 struct cgroup_subsys_state css;
392 #ifdef CONFIG_FAIR_GROUP_SCHED
393 /* schedulable entities of this group on each CPU */
394 struct sched_entity **se;
395 /* runqueue "owned" by this group on each CPU */
396 struct cfs_rq **cfs_rq;
397 unsigned long shares;
401 * load_avg can be heavily contended at clock tick time, so put
402 * it in its own cacheline separated from the fields above which
403 * will also be accessed at each tick.
405 atomic_long_t load_avg ____cacheline_aligned;
409 #ifdef CONFIG_RT_GROUP_SCHED
410 struct sched_rt_entity **rt_se;
411 struct rt_rq **rt_rq;
413 struct rt_bandwidth rt_bandwidth;
417 struct list_head list;
419 struct task_group *parent;
420 struct list_head siblings;
421 struct list_head children;
423 #ifdef CONFIG_SCHED_AUTOGROUP
424 struct autogroup *autogroup;
427 struct cfs_bandwidth cfs_bandwidth;
429 #ifdef CONFIG_UCLAMP_TASK_GROUP
430 /* The two decimal precision [%] value requested from user-space */
431 unsigned int uclamp_pct[UCLAMP_CNT];
432 /* Clamp values requested for a task group */
433 struct uclamp_se uclamp_req[UCLAMP_CNT];
434 /* Effective clamp values used for a task group */
435 struct uclamp_se uclamp[UCLAMP_CNT];
440 #ifdef CONFIG_FAIR_GROUP_SCHED
441 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
444 * A weight of 0 or 1 can cause arithmetics problems.
445 * A weight of a cfs_rq is the sum of weights of which entities
446 * are queued on this cfs_rq, so a weight of a entity should not be
447 * too large, so as the shares value of a task group.
448 * (The default weight is 1024 - so there's no practical
449 * limitation from this.)
451 #define MIN_SHARES (1UL << 1)
452 #define MAX_SHARES (1UL << 18)
455 typedef int (*tg_visitor)(struct task_group *, void *);
457 extern int walk_tg_tree_from(struct task_group *from,
458 tg_visitor down, tg_visitor up, void *data);
461 * Iterate the full tree, calling @down when first entering a node and @up when
462 * leaving it for the final time.
464 * Caller must hold rcu_lock or sufficient equivalent.
466 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
468 return walk_tg_tree_from(&root_task_group, down, up, data);
471 extern int tg_nop(struct task_group *tg, void *data);
473 extern void free_fair_sched_group(struct task_group *tg);
474 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
475 extern void online_fair_sched_group(struct task_group *tg);
476 extern void unregister_fair_sched_group(struct task_group *tg);
477 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
478 struct sched_entity *se, int cpu,
479 struct sched_entity *parent);
480 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
482 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
483 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
484 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
486 extern void free_rt_sched_group(struct task_group *tg);
487 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
488 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
489 struct sched_rt_entity *rt_se, int cpu,
490 struct sched_rt_entity *parent);
491 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
492 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
493 extern long sched_group_rt_runtime(struct task_group *tg);
494 extern long sched_group_rt_period(struct task_group *tg);
495 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
497 extern struct task_group *sched_create_group(struct task_group *parent);
498 extern void sched_online_group(struct task_group *tg,
499 struct task_group *parent);
500 extern void sched_destroy_group(struct task_group *tg);
501 extern void sched_offline_group(struct task_group *tg);
503 extern void sched_move_task(struct task_struct *tsk);
505 #ifdef CONFIG_FAIR_GROUP_SCHED
506 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
509 extern void set_task_rq_fair(struct sched_entity *se,
510 struct cfs_rq *prev, struct cfs_rq *next);
511 #else /* !CONFIG_SMP */
512 static inline void set_task_rq_fair(struct sched_entity *se,
513 struct cfs_rq *prev, struct cfs_rq *next) { }
514 #endif /* CONFIG_SMP */
515 #endif /* CONFIG_FAIR_GROUP_SCHED */
517 #else /* CONFIG_CGROUP_SCHED */
519 struct cfs_bandwidth { };
521 #endif /* CONFIG_CGROUP_SCHED */
523 /* CFS-related fields in a runqueue */
525 struct load_weight load;
526 unsigned int nr_running;
527 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
528 unsigned int idle_h_nr_running; /* SCHED_IDLE */
532 #ifdef CONFIG_SCHED_CORE
533 unsigned int forceidle_seq;
538 u64 min_vruntime_copy;
541 struct rb_root_cached tasks_timeline;
544 * 'curr' points to currently running entity on this cfs_rq.
545 * It is set to NULL otherwise (i.e when none are currently running).
547 struct sched_entity *curr;
548 struct sched_entity *next;
549 struct sched_entity *last;
550 struct sched_entity *skip;
552 #ifdef CONFIG_SCHED_DEBUG
553 unsigned int nr_spread_over;
560 struct sched_avg avg;
562 u64 load_last_update_time_copy;
565 raw_spinlock_t lock ____cacheline_aligned;
567 unsigned long load_avg;
568 unsigned long util_avg;
569 unsigned long runnable_avg;
572 #ifdef CONFIG_FAIR_GROUP_SCHED
573 unsigned long tg_load_avg_contrib;
575 long prop_runnable_sum;
578 * h_load = weight * f(tg)
580 * Where f(tg) is the recursive weight fraction assigned to
583 unsigned long h_load;
584 u64 last_h_load_update;
585 struct sched_entity *h_load_next;
586 #endif /* CONFIG_FAIR_GROUP_SCHED */
587 #endif /* CONFIG_SMP */
589 #ifdef CONFIG_FAIR_GROUP_SCHED
590 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
593 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
594 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
595 * (like users, containers etc.)
597 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
598 * This list is used during load balance.
601 struct list_head leaf_cfs_rq_list;
602 struct task_group *tg; /* group that "owns" this runqueue */
604 #ifdef CONFIG_CFS_BANDWIDTH
606 s64 runtime_remaining;
609 u64 throttled_clock_task;
610 u64 throttled_clock_task_time;
613 struct list_head throttled_list;
614 #endif /* CONFIG_CFS_BANDWIDTH */
615 #endif /* CONFIG_FAIR_GROUP_SCHED */
618 static inline int rt_bandwidth_enabled(void)
620 return sysctl_sched_rt_runtime >= 0;
623 /* RT IPI pull logic requires IRQ_WORK */
624 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
625 # define HAVE_RT_PUSH_IPI
628 /* Real-Time classes' related field in a runqueue: */
630 struct rt_prio_array active;
631 unsigned int rt_nr_running;
632 unsigned int rr_nr_running;
633 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
635 int curr; /* highest queued rt task prio */
637 int next; /* next highest */
642 unsigned int rt_nr_migratory;
643 unsigned int rt_nr_total;
645 struct plist_head pushable_tasks;
647 #endif /* CONFIG_SMP */
653 /* Nests inside the rq lock: */
654 raw_spinlock_t rt_runtime_lock;
656 #ifdef CONFIG_RT_GROUP_SCHED
657 unsigned int rt_nr_boosted;
660 struct task_group *tg;
664 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
666 return rt_rq->rt_queued && rt_rq->rt_nr_running;
669 /* Deadline class' related fields in a runqueue */
671 /* runqueue is an rbtree, ordered by deadline */
672 struct rb_root_cached root;
674 unsigned int dl_nr_running;
678 * Deadline values of the currently executing and the
679 * earliest ready task on this rq. Caching these facilitates
680 * the decision whether or not a ready but not running task
681 * should migrate somewhere else.
688 unsigned int dl_nr_migratory;
692 * Tasks on this rq that can be pushed away. They are kept in
693 * an rb-tree, ordered by tasks' deadlines, with caching
694 * of the leftmost (earliest deadline) element.
696 struct rb_root_cached pushable_dl_tasks_root;
701 * "Active utilization" for this runqueue: increased when a
702 * task wakes up (becomes TASK_RUNNING) and decreased when a
708 * Utilization of the tasks "assigned" to this runqueue (including
709 * the tasks that are in runqueue and the tasks that executed on this
710 * CPU and blocked). Increased when a task moves to this runqueue, and
711 * decreased when the task moves away (migrates, changes scheduling
712 * policy, or terminates).
713 * This is needed to compute the "inactive utilization" for the
714 * runqueue (inactive utilization = this_bw - running_bw).
720 * Inverse of the fraction of CPU utilization that can be reclaimed
721 * by the GRUB algorithm.
726 #ifdef CONFIG_FAIR_GROUP_SCHED
727 /* An entity is a task if it doesn't "own" a runqueue */
728 #define entity_is_task(se) (!se->my_q)
730 static inline void se_update_runnable(struct sched_entity *se)
732 if (!entity_is_task(se))
733 se->runnable_weight = se->my_q->h_nr_running;
736 static inline long se_runnable(struct sched_entity *se)
738 if (entity_is_task(se))
741 return se->runnable_weight;
745 #define entity_is_task(se) 1
747 static inline void se_update_runnable(struct sched_entity *se) {}
749 static inline long se_runnable(struct sched_entity *se)
757 * XXX we want to get rid of these helpers and use the full load resolution.
759 static inline long se_weight(struct sched_entity *se)
761 return scale_load_down(se->load.weight);
765 static inline bool sched_asym_prefer(int a, int b)
767 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
771 struct em_perf_domain *em_pd;
772 struct perf_domain *next;
776 /* Scheduling group status flags */
777 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
778 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
781 * We add the notion of a root-domain which will be used to define per-domain
782 * variables. Each exclusive cpuset essentially defines an island domain by
783 * fully partitioning the member CPUs from any other cpuset. Whenever a new
784 * exclusive cpuset is created, we also create and attach a new root-domain
793 cpumask_var_t online;
796 * Indicate pullable load on at least one CPU, e.g:
797 * - More than one runnable task
798 * - Running task is misfit
802 /* Indicate one or more cpus over-utilized (tipping point) */
806 * The bit corresponding to a CPU gets set here if such CPU has more
807 * than one runnable -deadline task (as it is below for RT tasks).
809 cpumask_var_t dlo_mask;
815 * Indicate whether a root_domain's dl_bw has been checked or
816 * updated. It's monotonously increasing value.
818 * Also, some corner cases, like 'wrap around' is dangerous, but given
819 * that u64 is 'big enough'. So that shouldn't be a concern.
823 #ifdef HAVE_RT_PUSH_IPI
825 * For IPI pull requests, loop across the rto_mask.
827 struct irq_work rto_push_work;
828 raw_spinlock_t rto_lock;
829 /* These are only updated and read within rto_lock */
832 /* These atomics are updated outside of a lock */
833 atomic_t rto_loop_next;
834 atomic_t rto_loop_start;
837 * The "RT overload" flag: it gets set if a CPU has more than
838 * one runnable RT task.
840 cpumask_var_t rto_mask;
841 struct cpupri cpupri;
843 unsigned long max_cpu_capacity;
846 * NULL-terminated list of performance domains intersecting with the
847 * CPUs of the rd. Protected by RCU.
849 struct perf_domain __rcu *pd;
852 extern void init_defrootdomain(void);
853 extern int sched_init_domains(const struct cpumask *cpu_map);
854 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
855 extern void sched_get_rd(struct root_domain *rd);
856 extern void sched_put_rd(struct root_domain *rd);
858 #ifdef HAVE_RT_PUSH_IPI
859 extern void rto_push_irq_work_func(struct irq_work *work);
861 #endif /* CONFIG_SMP */
863 #ifdef CONFIG_UCLAMP_TASK
865 * struct uclamp_bucket - Utilization clamp bucket
866 * @value: utilization clamp value for tasks on this clamp bucket
867 * @tasks: number of RUNNABLE tasks on this clamp bucket
869 * Keep track of how many tasks are RUNNABLE for a given utilization
872 struct uclamp_bucket {
873 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
874 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
878 * struct uclamp_rq - rq's utilization clamp
879 * @value: currently active clamp values for a rq
880 * @bucket: utilization clamp buckets affecting a rq
882 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
883 * A clamp value is affecting a rq when there is at least one task RUNNABLE
884 * (or actually running) with that value.
886 * There are up to UCLAMP_CNT possible different clamp values, currently there
887 * are only two: minimum utilization and maximum utilization.
889 * All utilization clamping values are MAX aggregated, since:
890 * - for util_min: we want to run the CPU at least at the max of the minimum
891 * utilization required by its currently RUNNABLE tasks.
892 * - for util_max: we want to allow the CPU to run up to the max of the
893 * maximum utilization allowed by its currently RUNNABLE tasks.
895 * Since on each system we expect only a limited number of different
896 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
897 * the metrics required to compute all the per-rq utilization clamp values.
901 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
904 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
905 #endif /* CONFIG_UCLAMP_TASK */
908 * This is the main, per-CPU runqueue data structure.
910 * Locking rule: those places that want to lock multiple runqueues
911 * (such as the load balancing or the thread migration code), lock
912 * acquire operations must be ordered by ascending &runqueue.
916 raw_spinlock_t __lock;
919 * nr_running and cpu_load should be in the same cacheline because
920 * remote CPUs use both these fields when doing load calculation.
922 unsigned int nr_running;
923 #ifdef CONFIG_NUMA_BALANCING
924 unsigned int nr_numa_running;
925 unsigned int nr_preferred_running;
926 unsigned int numa_migrate_on;
928 #ifdef CONFIG_NO_HZ_COMMON
930 unsigned long last_blocked_load_update_tick;
931 unsigned int has_blocked_load;
932 call_single_data_t nohz_csd;
933 #endif /* CONFIG_SMP */
934 unsigned int nohz_tick_stopped;
936 #endif /* CONFIG_NO_HZ_COMMON */
939 unsigned int ttwu_pending;
943 #ifdef CONFIG_UCLAMP_TASK
944 /* Utilization clamp values based on CPU's RUNNABLE tasks */
945 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
946 unsigned int uclamp_flags;
947 #define UCLAMP_FLAG_IDLE 0x01
954 #ifdef CONFIG_FAIR_GROUP_SCHED
955 /* list of leaf cfs_rq on this CPU: */
956 struct list_head leaf_cfs_rq_list;
957 struct list_head *tmp_alone_branch;
958 #endif /* CONFIG_FAIR_GROUP_SCHED */
961 * This is part of a global counter where only the total sum
962 * over all CPUs matters. A task can increase this counter on
963 * one CPU and if it got migrated afterwards it may decrease
964 * it on another CPU. Always updated under the runqueue lock:
966 unsigned int nr_uninterruptible;
968 struct task_struct __rcu *curr;
969 struct task_struct *idle;
970 struct task_struct *stop;
971 unsigned long next_balance;
972 struct mm_struct *prev_mm;
974 unsigned int clock_update_flags;
976 /* Ensure that all clocks are in the same cache line */
977 u64 clock_task ____cacheline_aligned;
979 unsigned long lost_idle_time;
983 #ifdef CONFIG_SCHED_DEBUG
984 u64 last_seen_need_resched_ns;
985 int ticks_without_resched;
988 #ifdef CONFIG_MEMBARRIER
989 int membarrier_state;
993 struct root_domain *rd;
994 struct sched_domain __rcu *sd;
996 unsigned long cpu_capacity;
997 unsigned long cpu_capacity_orig;
999 struct callback_head *balance_callback;
1001 unsigned char nohz_idle_balance;
1002 unsigned char idle_balance;
1004 unsigned long misfit_task_load;
1006 /* For active balancing */
1009 struct cpu_stop_work active_balance_work;
1011 /* CPU of this runqueue: */
1015 struct list_head cfs_tasks;
1017 struct sched_avg avg_rt;
1018 struct sched_avg avg_dl;
1019 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1020 struct sched_avg avg_irq;
1022 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1023 struct sched_avg avg_thermal;
1028 unsigned long wake_stamp;
1031 /* This is used to determine avg_idle's max value */
1032 u64 max_idle_balance_cost;
1034 #ifdef CONFIG_HOTPLUG_CPU
1035 struct rcuwait hotplug_wait;
1037 #endif /* CONFIG_SMP */
1039 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1042 #ifdef CONFIG_PARAVIRT
1043 u64 prev_steal_time;
1045 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1046 u64 prev_steal_time_rq;
1049 /* calc_load related fields */
1050 unsigned long calc_load_update;
1051 long calc_load_active;
1053 #ifdef CONFIG_SCHED_HRTICK
1055 call_single_data_t hrtick_csd;
1057 struct hrtimer hrtick_timer;
1058 ktime_t hrtick_time;
1061 #ifdef CONFIG_SCHEDSTATS
1063 struct sched_info rq_sched_info;
1064 unsigned long long rq_cpu_time;
1065 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1067 /* sys_sched_yield() stats */
1068 unsigned int yld_count;
1070 /* schedule() stats */
1071 unsigned int sched_count;
1072 unsigned int sched_goidle;
1074 /* try_to_wake_up() stats */
1075 unsigned int ttwu_count;
1076 unsigned int ttwu_local;
1079 #ifdef CONFIG_CPU_IDLE
1080 /* Must be inspected within a rcu lock section */
1081 struct cpuidle_state *idle_state;
1085 unsigned int nr_pinned;
1087 unsigned int push_busy;
1088 struct cpu_stop_work push_work;
1090 #ifdef CONFIG_SCHED_CORE
1093 struct task_struct *core_pick;
1094 unsigned int core_enabled;
1095 unsigned int core_sched_seq;
1096 struct rb_root core_tree;
1099 unsigned int core_task_seq;
1100 unsigned int core_pick_seq;
1101 unsigned long core_cookie;
1102 unsigned char core_forceidle;
1103 unsigned int core_forceidle_seq;
1107 #ifdef CONFIG_FAIR_GROUP_SCHED
1109 /* CPU runqueue to which this cfs_rq is attached */
1110 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1117 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1119 return container_of(cfs_rq, struct rq, cfs);
1123 static inline int cpu_of(struct rq *rq)
1132 #define MDF_PUSH 0x01
1134 static inline bool is_migration_disabled(struct task_struct *p)
1137 return p->migration_disabled;
1144 #ifdef CONFIG_SCHED_CORE
1145 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1147 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1149 static inline bool sched_core_enabled(struct rq *rq)
1151 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1154 static inline bool sched_core_disabled(void)
1156 return !static_branch_unlikely(&__sched_core_enabled);
1160 * Be careful with this function; not for general use. The return value isn't
1161 * stable unless you actually hold a relevant rq->__lock.
1163 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1165 if (sched_core_enabled(rq))
1166 return &rq->core->__lock;
1171 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1173 if (rq->core_enabled)
1174 return &rq->core->__lock;
1179 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1182 * Helpers to check if the CPU's core cookie matches with the task's cookie
1183 * when core scheduling is enabled.
1184 * A special case is that the task's cookie always matches with CPU's core
1185 * cookie if the CPU is in an idle core.
1187 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1189 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1190 if (!sched_core_enabled(rq))
1193 return rq->core->core_cookie == p->core_cookie;
1196 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1198 bool idle_core = true;
1201 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1202 if (!sched_core_enabled(rq))
1205 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1206 if (!available_idle_cpu(cpu)) {
1213 * A CPU in an idle core is always the best choice for tasks with
1216 return idle_core || rq->core->core_cookie == p->core_cookie;
1219 static inline bool sched_group_cookie_match(struct rq *rq,
1220 struct task_struct *p,
1221 struct sched_group *group)
1225 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1226 if (!sched_core_enabled(rq))
1229 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1230 if (sched_core_cookie_match(rq, p))
1236 extern void queue_core_balance(struct rq *rq);
1238 static inline bool sched_core_enqueued(struct task_struct *p)
1240 return !RB_EMPTY_NODE(&p->core_node);
1243 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1244 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1246 extern void sched_core_get(void);
1247 extern void sched_core_put(void);
1249 extern unsigned long sched_core_alloc_cookie(void);
1250 extern void sched_core_put_cookie(unsigned long cookie);
1251 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1252 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1254 #else /* !CONFIG_SCHED_CORE */
1256 static inline bool sched_core_enabled(struct rq *rq)
1261 static inline bool sched_core_disabled(void)
1266 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1271 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1276 static inline void queue_core_balance(struct rq *rq)
1280 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1285 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1290 static inline bool sched_group_cookie_match(struct rq *rq,
1291 struct task_struct *p,
1292 struct sched_group *group)
1296 #endif /* CONFIG_SCHED_CORE */
1298 static inline void lockdep_assert_rq_held(struct rq *rq)
1300 lockdep_assert_held(__rq_lockp(rq));
1303 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1304 extern bool raw_spin_rq_trylock(struct rq *rq);
1305 extern void raw_spin_rq_unlock(struct rq *rq);
1307 static inline void raw_spin_rq_lock(struct rq *rq)
1309 raw_spin_rq_lock_nested(rq, 0);
1312 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1314 local_irq_disable();
1315 raw_spin_rq_lock(rq);
1318 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1320 raw_spin_rq_unlock(rq);
1324 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1326 unsigned long flags;
1327 local_irq_save(flags);
1328 raw_spin_rq_lock(rq);
1332 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1334 raw_spin_rq_unlock(rq);
1335 local_irq_restore(flags);
1338 #define raw_spin_rq_lock_irqsave(rq, flags) \
1340 flags = _raw_spin_rq_lock_irqsave(rq); \
1343 #ifdef CONFIG_SCHED_SMT
1344 extern void __update_idle_core(struct rq *rq);
1346 static inline void update_idle_core(struct rq *rq)
1348 if (static_branch_unlikely(&sched_smt_present))
1349 __update_idle_core(rq);
1353 static inline void update_idle_core(struct rq *rq) { }
1356 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1358 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1359 #define this_rq() this_cpu_ptr(&runqueues)
1360 #define task_rq(p) cpu_rq(task_cpu(p))
1361 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1362 #define raw_rq() raw_cpu_ptr(&runqueues)
1364 #ifdef CONFIG_FAIR_GROUP_SCHED
1365 static inline struct task_struct *task_of(struct sched_entity *se)
1367 SCHED_WARN_ON(!entity_is_task(se));
1368 return container_of(se, struct task_struct, se);
1371 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1373 return p->se.cfs_rq;
1376 /* runqueue on which this entity is (to be) queued */
1377 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1382 /* runqueue "owned" by this group */
1383 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1390 static inline struct task_struct *task_of(struct sched_entity *se)
1392 return container_of(se, struct task_struct, se);
1395 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1397 return &task_rq(p)->cfs;
1400 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1402 struct task_struct *p = task_of(se);
1403 struct rq *rq = task_rq(p);
1408 /* runqueue "owned" by this group */
1409 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1415 extern void update_rq_clock(struct rq *rq);
1417 static inline u64 __rq_clock_broken(struct rq *rq)
1419 return READ_ONCE(rq->clock);
1423 * rq::clock_update_flags bits
1425 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1426 * call to __schedule(). This is an optimisation to avoid
1427 * neighbouring rq clock updates.
1429 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1430 * in effect and calls to update_rq_clock() are being ignored.
1432 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1433 * made to update_rq_clock() since the last time rq::lock was pinned.
1435 * If inside of __schedule(), clock_update_flags will have been
1436 * shifted left (a left shift is a cheap operation for the fast path
1437 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1439 * if (rq-clock_update_flags >= RQCF_UPDATED)
1441 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1442 * one position though, because the next rq_unpin_lock() will shift it
1445 #define RQCF_REQ_SKIP 0x01
1446 #define RQCF_ACT_SKIP 0x02
1447 #define RQCF_UPDATED 0x04
1449 static inline void assert_clock_updated(struct rq *rq)
1452 * The only reason for not seeing a clock update since the
1453 * last rq_pin_lock() is if we're currently skipping updates.
1455 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1458 static inline u64 rq_clock(struct rq *rq)
1460 lockdep_assert_rq_held(rq);
1461 assert_clock_updated(rq);
1466 static inline u64 rq_clock_task(struct rq *rq)
1468 lockdep_assert_rq_held(rq);
1469 assert_clock_updated(rq);
1471 return rq->clock_task;
1475 * By default the decay is the default pelt decay period.
1476 * The decay shift can change the decay period in
1478 * Decay shift Decay period(ms)
1485 extern int sched_thermal_decay_shift;
1487 static inline u64 rq_clock_thermal(struct rq *rq)
1489 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1492 static inline void rq_clock_skip_update(struct rq *rq)
1494 lockdep_assert_rq_held(rq);
1495 rq->clock_update_flags |= RQCF_REQ_SKIP;
1499 * See rt task throttling, which is the only time a skip
1500 * request is canceled.
1502 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1504 lockdep_assert_rq_held(rq);
1505 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1509 unsigned long flags;
1510 struct pin_cookie cookie;
1511 #ifdef CONFIG_SCHED_DEBUG
1513 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1514 * current pin context is stashed here in case it needs to be
1515 * restored in rq_repin_lock().
1517 unsigned int clock_update_flags;
1521 extern struct callback_head balance_push_callback;
1524 * Lockdep annotation that avoids accidental unlocks; it's like a
1525 * sticky/continuous lockdep_assert_held().
1527 * This avoids code that has access to 'struct rq *rq' (basically everything in
1528 * the scheduler) from accidentally unlocking the rq if they do not also have a
1529 * copy of the (on-stack) 'struct rq_flags rf'.
1531 * Also see Documentation/locking/lockdep-design.rst.
1533 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1535 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1537 #ifdef CONFIG_SCHED_DEBUG
1538 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1539 rf->clock_update_flags = 0;
1541 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1546 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1548 #ifdef CONFIG_SCHED_DEBUG
1549 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1550 rf->clock_update_flags = RQCF_UPDATED;
1553 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1556 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1558 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1560 #ifdef CONFIG_SCHED_DEBUG
1562 * Restore the value we stashed in @rf for this pin context.
1564 rq->clock_update_flags |= rf->clock_update_flags;
1568 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1569 __acquires(rq->lock);
1571 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1572 __acquires(p->pi_lock)
1573 __acquires(rq->lock);
1575 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1576 __releases(rq->lock)
1578 rq_unpin_lock(rq, rf);
1579 raw_spin_rq_unlock(rq);
1583 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1584 __releases(rq->lock)
1585 __releases(p->pi_lock)
1587 rq_unpin_lock(rq, rf);
1588 raw_spin_rq_unlock(rq);
1589 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1593 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1594 __acquires(rq->lock)
1596 raw_spin_rq_lock_irqsave(rq, rf->flags);
1597 rq_pin_lock(rq, rf);
1601 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1602 __acquires(rq->lock)
1604 raw_spin_rq_lock_irq(rq);
1605 rq_pin_lock(rq, rf);
1609 rq_lock(struct rq *rq, struct rq_flags *rf)
1610 __acquires(rq->lock)
1612 raw_spin_rq_lock(rq);
1613 rq_pin_lock(rq, rf);
1617 rq_relock(struct rq *rq, struct rq_flags *rf)
1618 __acquires(rq->lock)
1620 raw_spin_rq_lock(rq);
1621 rq_repin_lock(rq, rf);
1625 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1626 __releases(rq->lock)
1628 rq_unpin_lock(rq, rf);
1629 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1633 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1634 __releases(rq->lock)
1636 rq_unpin_lock(rq, rf);
1637 raw_spin_rq_unlock_irq(rq);
1641 rq_unlock(struct rq *rq, struct rq_flags *rf)
1642 __releases(rq->lock)
1644 rq_unpin_lock(rq, rf);
1645 raw_spin_rq_unlock(rq);
1648 static inline struct rq *
1649 this_rq_lock_irq(struct rq_flags *rf)
1650 __acquires(rq->lock)
1654 local_irq_disable();
1661 enum numa_topology_type {
1666 extern enum numa_topology_type sched_numa_topology_type;
1667 extern int sched_max_numa_distance;
1668 extern bool find_numa_distance(int distance);
1669 extern void sched_init_numa(void);
1670 extern void sched_domains_numa_masks_set(unsigned int cpu);
1671 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1672 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1674 static inline void sched_init_numa(void) { }
1675 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1676 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1677 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1683 #ifdef CONFIG_NUMA_BALANCING
1684 /* The regions in numa_faults array from task_struct */
1685 enum numa_faults_stats {
1691 extern void sched_setnuma(struct task_struct *p, int node);
1692 extern int migrate_task_to(struct task_struct *p, int cpu);
1693 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1695 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1698 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1701 #endif /* CONFIG_NUMA_BALANCING */
1706 queue_balance_callback(struct rq *rq,
1707 struct callback_head *head,
1708 void (*func)(struct rq *rq))
1710 lockdep_assert_rq_held(rq);
1712 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1715 head->func = (void (*)(struct callback_head *))func;
1716 head->next = rq->balance_callback;
1717 rq->balance_callback = head;
1720 #define rcu_dereference_check_sched_domain(p) \
1721 rcu_dereference_check((p), \
1722 lockdep_is_held(&sched_domains_mutex))
1725 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1726 * See destroy_sched_domains: call_rcu for details.
1728 * The domain tree of any CPU may only be accessed from within
1729 * preempt-disabled sections.
1731 #define for_each_domain(cpu, __sd) \
1732 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1733 __sd; __sd = __sd->parent)
1736 * highest_flag_domain - Return highest sched_domain containing flag.
1737 * @cpu: The CPU whose highest level of sched domain is to
1739 * @flag: The flag to check for the highest sched_domain
1740 * for the given CPU.
1742 * Returns the highest sched_domain of a CPU which contains the given flag.
1744 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1746 struct sched_domain *sd, *hsd = NULL;
1748 for_each_domain(cpu, sd) {
1749 if (!(sd->flags & flag))
1757 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1759 struct sched_domain *sd;
1761 for_each_domain(cpu, sd) {
1762 if (sd->flags & flag)
1769 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1770 DECLARE_PER_CPU(int, sd_llc_size);
1771 DECLARE_PER_CPU(int, sd_llc_id);
1772 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1773 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1774 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1775 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1776 extern struct static_key_false sched_asym_cpucapacity;
1778 struct sched_group_capacity {
1781 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1784 unsigned long capacity;
1785 unsigned long min_capacity; /* Min per-CPU capacity in group */
1786 unsigned long max_capacity; /* Max per-CPU capacity in group */
1787 unsigned long next_update;
1788 int imbalance; /* XXX unrelated to capacity but shared group state */
1790 #ifdef CONFIG_SCHED_DEBUG
1794 unsigned long cpumask[]; /* Balance mask */
1797 struct sched_group {
1798 struct sched_group *next; /* Must be a circular list */
1801 unsigned int group_weight;
1802 struct sched_group_capacity *sgc;
1803 int asym_prefer_cpu; /* CPU of highest priority in group */
1806 * The CPUs this group covers.
1808 * NOTE: this field is variable length. (Allocated dynamically
1809 * by attaching extra space to the end of the structure,
1810 * depending on how many CPUs the kernel has booted up with)
1812 unsigned long cpumask[];
1815 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1817 return to_cpumask(sg->cpumask);
1821 * See build_balance_mask().
1823 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1825 return to_cpumask(sg->sgc->cpumask);
1829 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1830 * @group: The group whose first CPU is to be returned.
1832 static inline unsigned int group_first_cpu(struct sched_group *group)
1834 return cpumask_first(sched_group_span(group));
1837 extern int group_balance_cpu(struct sched_group *sg);
1839 #ifdef CONFIG_SCHED_DEBUG
1840 void update_sched_domain_debugfs(void);
1841 void dirty_sched_domain_sysctl(int cpu);
1843 static inline void update_sched_domain_debugfs(void)
1846 static inline void dirty_sched_domain_sysctl(int cpu)
1851 extern int sched_update_scaling(void);
1853 extern void flush_smp_call_function_from_idle(void);
1855 #else /* !CONFIG_SMP: */
1856 static inline void flush_smp_call_function_from_idle(void) { }
1860 #include "autogroup.h"
1862 #ifdef CONFIG_CGROUP_SCHED
1865 * Return the group to which this tasks belongs.
1867 * We cannot use task_css() and friends because the cgroup subsystem
1868 * changes that value before the cgroup_subsys::attach() method is called,
1869 * therefore we cannot pin it and might observe the wrong value.
1871 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1872 * core changes this before calling sched_move_task().
1874 * Instead we use a 'copy' which is updated from sched_move_task() while
1875 * holding both task_struct::pi_lock and rq::lock.
1877 static inline struct task_group *task_group(struct task_struct *p)
1879 return p->sched_task_group;
1882 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1883 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1885 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1886 struct task_group *tg = task_group(p);
1889 #ifdef CONFIG_FAIR_GROUP_SCHED
1890 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1891 p->se.cfs_rq = tg->cfs_rq[cpu];
1892 p->se.parent = tg->se[cpu];
1895 #ifdef CONFIG_RT_GROUP_SCHED
1896 p->rt.rt_rq = tg->rt_rq[cpu];
1897 p->rt.parent = tg->rt_se[cpu];
1901 #else /* CONFIG_CGROUP_SCHED */
1903 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1904 static inline struct task_group *task_group(struct task_struct *p)
1909 #endif /* CONFIG_CGROUP_SCHED */
1911 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1913 set_task_rq(p, cpu);
1916 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1917 * successfully executed on another CPU. We must ensure that updates of
1918 * per-task data have been completed by this moment.
1921 #ifdef CONFIG_THREAD_INFO_IN_TASK
1922 WRITE_ONCE(p->cpu, cpu);
1924 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1931 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1933 #ifdef CONFIG_SCHED_DEBUG
1934 # include <linux/static_key.h>
1935 # define const_debug __read_mostly
1937 # define const_debug const
1940 #define SCHED_FEAT(name, enabled) \
1941 __SCHED_FEAT_##name ,
1944 #include "features.h"
1950 #ifdef CONFIG_SCHED_DEBUG
1953 * To support run-time toggling of sched features, all the translation units
1954 * (but core.c) reference the sysctl_sched_features defined in core.c.
1956 extern const_debug unsigned int sysctl_sched_features;
1958 #ifdef CONFIG_JUMP_LABEL
1959 #define SCHED_FEAT(name, enabled) \
1960 static __always_inline bool static_branch_##name(struct static_key *key) \
1962 return static_key_##enabled(key); \
1965 #include "features.h"
1968 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1969 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1971 #else /* !CONFIG_JUMP_LABEL */
1973 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1975 #endif /* CONFIG_JUMP_LABEL */
1977 #else /* !SCHED_DEBUG */
1980 * Each translation unit has its own copy of sysctl_sched_features to allow
1981 * constants propagation at compile time and compiler optimization based on
1984 #define SCHED_FEAT(name, enabled) \
1985 (1UL << __SCHED_FEAT_##name) * enabled |
1986 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1987 #include "features.h"
1991 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1993 #endif /* SCHED_DEBUG */
1995 extern struct static_key_false sched_numa_balancing;
1996 extern struct static_key_false sched_schedstats;
1998 static inline u64 global_rt_period(void)
2000 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2003 static inline u64 global_rt_runtime(void)
2005 if (sysctl_sched_rt_runtime < 0)
2008 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2011 static inline int task_current(struct rq *rq, struct task_struct *p)
2013 return rq->curr == p;
2016 static inline int task_running(struct rq *rq, struct task_struct *p)
2021 return task_current(rq, p);
2025 static inline int task_on_rq_queued(struct task_struct *p)
2027 return p->on_rq == TASK_ON_RQ_QUEUED;
2030 static inline int task_on_rq_migrating(struct task_struct *p)
2032 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2035 /* Wake flags. The first three directly map to some SD flag value */
2036 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2037 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2038 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2040 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2041 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2042 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2045 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2046 static_assert(WF_FORK == SD_BALANCE_FORK);
2047 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2051 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2052 * of tasks with abnormal "nice" values across CPUs the contribution that
2053 * each task makes to its run queue's load is weighted according to its
2054 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2055 * scaled version of the new time slice allocation that they receive on time
2059 #define WEIGHT_IDLEPRIO 3
2060 #define WMULT_IDLEPRIO 1431655765
2062 extern const int sched_prio_to_weight[40];
2063 extern const u32 sched_prio_to_wmult[40];
2066 * {de,en}queue flags:
2068 * DEQUEUE_SLEEP - task is no longer runnable
2069 * ENQUEUE_WAKEUP - task just became runnable
2071 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2072 * are in a known state which allows modification. Such pairs
2073 * should preserve as much state as possible.
2075 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2078 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2079 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2080 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2084 #define DEQUEUE_SLEEP 0x01
2085 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2086 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2087 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2089 #define ENQUEUE_WAKEUP 0x01
2090 #define ENQUEUE_RESTORE 0x02
2091 #define ENQUEUE_MOVE 0x04
2092 #define ENQUEUE_NOCLOCK 0x08
2094 #define ENQUEUE_HEAD 0x10
2095 #define ENQUEUE_REPLENISH 0x20
2097 #define ENQUEUE_MIGRATED 0x40
2099 #define ENQUEUE_MIGRATED 0x00
2102 #define RETRY_TASK ((void *)-1UL)
2104 struct sched_class {
2106 #ifdef CONFIG_UCLAMP_TASK
2110 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2111 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2112 void (*yield_task) (struct rq *rq);
2113 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2115 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2117 struct task_struct *(*pick_next_task)(struct rq *rq);
2119 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2120 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2123 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2124 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2126 struct task_struct * (*pick_task)(struct rq *rq);
2128 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2130 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2132 void (*set_cpus_allowed)(struct task_struct *p,
2133 const struct cpumask *newmask,
2136 void (*rq_online)(struct rq *rq);
2137 void (*rq_offline)(struct rq *rq);
2139 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2142 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2143 void (*task_fork)(struct task_struct *p);
2144 void (*task_dead)(struct task_struct *p);
2147 * The switched_from() call is allowed to drop rq->lock, therefore we
2148 * cannot assume the switched_from/switched_to pair is serialized by
2149 * rq->lock. They are however serialized by p->pi_lock.
2151 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2152 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2153 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2156 unsigned int (*get_rr_interval)(struct rq *rq,
2157 struct task_struct *task);
2159 void (*update_curr)(struct rq *rq);
2161 #define TASK_SET_GROUP 0
2162 #define TASK_MOVE_GROUP 1
2164 #ifdef CONFIG_FAIR_GROUP_SCHED
2165 void (*task_change_group)(struct task_struct *p, int type);
2169 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2171 WARN_ON_ONCE(rq->curr != prev);
2172 prev->sched_class->put_prev_task(rq, prev);
2175 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2177 next->sched_class->set_next_task(rq, next, false);
2182 * Helper to define a sched_class instance; each one is placed in a separate
2183 * section which is ordered by the linker script:
2185 * include/asm-generic/vmlinux.lds.h
2187 * Also enforce alignment on the instance, not the type, to guarantee layout.
2189 #define DEFINE_SCHED_CLASS(name) \
2190 const struct sched_class name##_sched_class \
2191 __aligned(__alignof__(struct sched_class)) \
2192 __section("__" #name "_sched_class")
2194 /* Defined in include/asm-generic/vmlinux.lds.h */
2195 extern struct sched_class __begin_sched_classes[];
2196 extern struct sched_class __end_sched_classes[];
2198 #define sched_class_highest (__end_sched_classes - 1)
2199 #define sched_class_lowest (__begin_sched_classes - 1)
2201 #define for_class_range(class, _from, _to) \
2202 for (class = (_from); class != (_to); class--)
2204 #define for_each_class(class) \
2205 for_class_range(class, sched_class_highest, sched_class_lowest)
2207 extern const struct sched_class stop_sched_class;
2208 extern const struct sched_class dl_sched_class;
2209 extern const struct sched_class rt_sched_class;
2210 extern const struct sched_class fair_sched_class;
2211 extern const struct sched_class idle_sched_class;
2213 static inline bool sched_stop_runnable(struct rq *rq)
2215 return rq->stop && task_on_rq_queued(rq->stop);
2218 static inline bool sched_dl_runnable(struct rq *rq)
2220 return rq->dl.dl_nr_running > 0;
2223 static inline bool sched_rt_runnable(struct rq *rq)
2225 return rq->rt.rt_queued > 0;
2228 static inline bool sched_fair_runnable(struct rq *rq)
2230 return rq->cfs.nr_running > 0;
2233 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2234 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2236 #define SCA_CHECK 0x01
2237 #define SCA_MIGRATE_DISABLE 0x02
2238 #define SCA_MIGRATE_ENABLE 0x04
2242 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2244 extern void trigger_load_balance(struct rq *rq);
2246 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2248 static inline struct task_struct *get_push_task(struct rq *rq)
2250 struct task_struct *p = rq->curr;
2252 lockdep_assert_rq_held(rq);
2257 if (p->nr_cpus_allowed == 1)
2260 rq->push_busy = true;
2261 return get_task_struct(p);
2264 extern int push_cpu_stop(void *arg);
2268 #ifdef CONFIG_CPU_IDLE
2269 static inline void idle_set_state(struct rq *rq,
2270 struct cpuidle_state *idle_state)
2272 rq->idle_state = idle_state;
2275 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2277 SCHED_WARN_ON(!rcu_read_lock_held());
2279 return rq->idle_state;
2282 static inline void idle_set_state(struct rq *rq,
2283 struct cpuidle_state *idle_state)
2287 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2293 extern void schedule_idle(void);
2295 extern void sysrq_sched_debug_show(void);
2296 extern void sched_init_granularity(void);
2297 extern void update_max_interval(void);
2299 extern void init_sched_dl_class(void);
2300 extern void init_sched_rt_class(void);
2301 extern void init_sched_fair_class(void);
2303 extern void reweight_task(struct task_struct *p, int prio);
2305 extern void resched_curr(struct rq *rq);
2306 extern void resched_cpu(int cpu);
2308 extern struct rt_bandwidth def_rt_bandwidth;
2309 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2311 extern struct dl_bandwidth def_dl_bandwidth;
2312 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2313 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2314 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2317 #define BW_UNIT (1 << BW_SHIFT)
2318 #define RATIO_SHIFT 8
2319 #define MAX_BW_BITS (64 - BW_SHIFT)
2320 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2321 unsigned long to_ratio(u64 period, u64 runtime);
2323 extern void init_entity_runnable_average(struct sched_entity *se);
2324 extern void post_init_entity_util_avg(struct task_struct *p);
2326 #ifdef CONFIG_NO_HZ_FULL
2327 extern bool sched_can_stop_tick(struct rq *rq);
2328 extern int __init sched_tick_offload_init(void);
2331 * Tick may be needed by tasks in the runqueue depending on their policy and
2332 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2333 * nohz mode if necessary.
2335 static inline void sched_update_tick_dependency(struct rq *rq)
2337 int cpu = cpu_of(rq);
2339 if (!tick_nohz_full_cpu(cpu))
2342 if (sched_can_stop_tick(rq))
2343 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2345 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2348 static inline int sched_tick_offload_init(void) { return 0; }
2349 static inline void sched_update_tick_dependency(struct rq *rq) { }
2352 static inline void add_nr_running(struct rq *rq, unsigned count)
2354 unsigned prev_nr = rq->nr_running;
2356 rq->nr_running = prev_nr + count;
2357 if (trace_sched_update_nr_running_tp_enabled()) {
2358 call_trace_sched_update_nr_running(rq, count);
2362 if (prev_nr < 2 && rq->nr_running >= 2) {
2363 if (!READ_ONCE(rq->rd->overload))
2364 WRITE_ONCE(rq->rd->overload, 1);
2368 sched_update_tick_dependency(rq);
2371 static inline void sub_nr_running(struct rq *rq, unsigned count)
2373 rq->nr_running -= count;
2374 if (trace_sched_update_nr_running_tp_enabled()) {
2375 call_trace_sched_update_nr_running(rq, -count);
2378 /* Check if we still need preemption */
2379 sched_update_tick_dependency(rq);
2382 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2383 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2385 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2387 extern const_debug unsigned int sysctl_sched_nr_migrate;
2388 extern const_debug unsigned int sysctl_sched_migration_cost;
2390 #ifdef CONFIG_SCHED_DEBUG
2391 extern unsigned int sysctl_sched_latency;
2392 extern unsigned int sysctl_sched_min_granularity;
2393 extern unsigned int sysctl_sched_wakeup_granularity;
2394 extern int sysctl_resched_latency_warn_ms;
2395 extern int sysctl_resched_latency_warn_once;
2397 extern unsigned int sysctl_sched_tunable_scaling;
2399 extern unsigned int sysctl_numa_balancing_scan_delay;
2400 extern unsigned int sysctl_numa_balancing_scan_period_min;
2401 extern unsigned int sysctl_numa_balancing_scan_period_max;
2402 extern unsigned int sysctl_numa_balancing_scan_size;
2405 #ifdef CONFIG_SCHED_HRTICK
2409 * - enabled by features
2410 * - hrtimer is actually high res
2412 static inline int hrtick_enabled(struct rq *rq)
2414 if (!cpu_active(cpu_of(rq)))
2416 return hrtimer_is_hres_active(&rq->hrtick_timer);
2419 static inline int hrtick_enabled_fair(struct rq *rq)
2421 if (!sched_feat(HRTICK))
2423 return hrtick_enabled(rq);
2426 static inline int hrtick_enabled_dl(struct rq *rq)
2428 if (!sched_feat(HRTICK_DL))
2430 return hrtick_enabled(rq);
2433 void hrtick_start(struct rq *rq, u64 delay);
2437 static inline int hrtick_enabled_fair(struct rq *rq)
2442 static inline int hrtick_enabled_dl(struct rq *rq)
2447 static inline int hrtick_enabled(struct rq *rq)
2452 #endif /* CONFIG_SCHED_HRTICK */
2454 #ifndef arch_scale_freq_tick
2455 static __always_inline
2456 void arch_scale_freq_tick(void)
2461 #ifndef arch_scale_freq_capacity
2463 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2464 * @cpu: the CPU in question.
2466 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2469 * ------ * SCHED_CAPACITY_SCALE
2472 static __always_inline
2473 unsigned long arch_scale_freq_capacity(int cpu)
2475 return SCHED_CAPACITY_SCALE;
2482 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2484 #ifdef CONFIG_SCHED_CORE
2486 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2487 * order by core-id first and cpu-id second.
2491 * double_rq_lock(0,3); will take core-0, core-1 lock
2492 * double_rq_lock(1,2); will take core-1, core-0 lock
2494 * when only cpu-id is considered.
2496 if (rq1->core->cpu < rq2->core->cpu)
2498 if (rq1->core->cpu > rq2->core->cpu)
2502 * __sched_core_flip() relies on SMT having cpu-id lock order.
2505 return rq1->cpu < rq2->cpu;
2508 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2510 #ifdef CONFIG_PREEMPTION
2513 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2514 * way at the expense of forcing extra atomic operations in all
2515 * invocations. This assures that the double_lock is acquired using the
2516 * same underlying policy as the spinlock_t on this architecture, which
2517 * reduces latency compared to the unfair variant below. However, it
2518 * also adds more overhead and therefore may reduce throughput.
2520 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2521 __releases(this_rq->lock)
2522 __acquires(busiest->lock)
2523 __acquires(this_rq->lock)
2525 raw_spin_rq_unlock(this_rq);
2526 double_rq_lock(this_rq, busiest);
2533 * Unfair double_lock_balance: Optimizes throughput at the expense of
2534 * latency by eliminating extra atomic operations when the locks are
2535 * already in proper order on entry. This favors lower CPU-ids and will
2536 * grant the double lock to lower CPUs over higher ids under contention,
2537 * regardless of entry order into the function.
2539 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2540 __releases(this_rq->lock)
2541 __acquires(busiest->lock)
2542 __acquires(this_rq->lock)
2544 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2547 if (likely(raw_spin_rq_trylock(busiest)))
2550 if (rq_order_less(this_rq, busiest)) {
2551 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2555 raw_spin_rq_unlock(this_rq);
2556 double_rq_lock(this_rq, busiest);
2561 #endif /* CONFIG_PREEMPTION */
2564 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2566 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2568 lockdep_assert_irqs_disabled();
2570 return _double_lock_balance(this_rq, busiest);
2573 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2574 __releases(busiest->lock)
2576 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2577 raw_spin_rq_unlock(busiest);
2578 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2581 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2587 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2590 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2596 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2599 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2605 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2609 * double_rq_unlock - safely unlock two runqueues
2611 * Note this does not restore interrupts like task_rq_unlock,
2612 * you need to do so manually after calling.
2614 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2615 __releases(rq1->lock)
2616 __releases(rq2->lock)
2618 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2619 raw_spin_rq_unlock(rq2);
2621 __release(rq2->lock);
2622 raw_spin_rq_unlock(rq1);
2625 extern void set_rq_online (struct rq *rq);
2626 extern void set_rq_offline(struct rq *rq);
2627 extern bool sched_smp_initialized;
2629 #else /* CONFIG_SMP */
2632 * double_rq_lock - safely lock two runqueues
2634 * Note this does not disable interrupts like task_rq_lock,
2635 * you need to do so manually before calling.
2637 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2638 __acquires(rq1->lock)
2639 __acquires(rq2->lock)
2641 BUG_ON(!irqs_disabled());
2643 raw_spin_rq_lock(rq1);
2644 __acquire(rq2->lock); /* Fake it out ;) */
2648 * double_rq_unlock - safely unlock two runqueues
2650 * Note this does not restore interrupts like task_rq_unlock,
2651 * you need to do so manually after calling.
2653 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2654 __releases(rq1->lock)
2655 __releases(rq2->lock)
2658 raw_spin_rq_unlock(rq1);
2659 __release(rq2->lock);
2664 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2665 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2667 #ifdef CONFIG_SCHED_DEBUG
2668 extern bool sched_debug_verbose;
2670 extern void print_cfs_stats(struct seq_file *m, int cpu);
2671 extern void print_rt_stats(struct seq_file *m, int cpu);
2672 extern void print_dl_stats(struct seq_file *m, int cpu);
2673 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2674 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2675 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2677 extern void resched_latency_warn(int cpu, u64 latency);
2678 #ifdef CONFIG_NUMA_BALANCING
2680 show_numa_stats(struct task_struct *p, struct seq_file *m);
2682 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2683 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2684 #endif /* CONFIG_NUMA_BALANCING */
2686 static inline void resched_latency_warn(int cpu, u64 latency) {}
2687 #endif /* CONFIG_SCHED_DEBUG */
2689 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2690 extern void init_rt_rq(struct rt_rq *rt_rq);
2691 extern void init_dl_rq(struct dl_rq *dl_rq);
2693 extern void cfs_bandwidth_usage_inc(void);
2694 extern void cfs_bandwidth_usage_dec(void);
2696 #ifdef CONFIG_NO_HZ_COMMON
2697 #define NOHZ_BALANCE_KICK_BIT 0
2698 #define NOHZ_STATS_KICK_BIT 1
2699 #define NOHZ_NEWILB_KICK_BIT 2
2701 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2702 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2703 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2705 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2707 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2709 extern void nohz_balance_exit_idle(struct rq *rq);
2711 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2714 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2715 extern void nohz_run_idle_balance(int cpu);
2717 static inline void nohz_run_idle_balance(int cpu) { }
2722 void __dl_update(struct dl_bw *dl_b, s64 bw)
2724 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2727 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2728 "sched RCU must be held");
2729 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2730 struct rq *rq = cpu_rq(i);
2732 rq->dl.extra_bw += bw;
2737 void __dl_update(struct dl_bw *dl_b, s64 bw)
2739 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2746 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2751 struct u64_stats_sync sync;
2754 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2757 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2758 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2759 * and never move forward.
2761 static inline u64 irq_time_read(int cpu)
2763 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2768 seq = __u64_stats_fetch_begin(&irqtime->sync);
2769 total = irqtime->total;
2770 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2774 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2776 #ifdef CONFIG_CPU_FREQ
2777 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2780 * cpufreq_update_util - Take a note about CPU utilization changes.
2781 * @rq: Runqueue to carry out the update for.
2782 * @flags: Update reason flags.
2784 * This function is called by the scheduler on the CPU whose utilization is
2787 * It can only be called from RCU-sched read-side critical sections.
2789 * The way cpufreq is currently arranged requires it to evaluate the CPU
2790 * performance state (frequency/voltage) on a regular basis to prevent it from
2791 * being stuck in a completely inadequate performance level for too long.
2792 * That is not guaranteed to happen if the updates are only triggered from CFS
2793 * and DL, though, because they may not be coming in if only RT tasks are
2794 * active all the time (or there are RT tasks only).
2796 * As a workaround for that issue, this function is called periodically by the
2797 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2798 * but that really is a band-aid. Going forward it should be replaced with
2799 * solutions targeted more specifically at RT tasks.
2801 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2803 struct update_util_data *data;
2805 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2808 data->func(data, rq_clock(rq), flags);
2811 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2812 #endif /* CONFIG_CPU_FREQ */
2814 #ifdef CONFIG_UCLAMP_TASK
2815 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2818 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2819 * @rq: The rq to clamp against. Must not be NULL.
2820 * @util: The util value to clamp.
2821 * @p: The task to clamp against. Can be NULL if you want to clamp
2824 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2826 * If sched_uclamp_used static key is disabled, then just return the util
2827 * without any clamping since uclamp aggregation at the rq level in the fast
2828 * path is disabled, rendering this operation a NOP.
2830 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2831 * will return the correct effective uclamp value of the task even if the
2832 * static key is disabled.
2834 static __always_inline
2835 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2836 struct task_struct *p)
2838 unsigned long min_util;
2839 unsigned long max_util;
2841 if (!static_branch_likely(&sched_uclamp_used))
2844 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2845 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2848 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2849 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2853 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2854 * RUNNABLE tasks with _different_ clamps, we can end up with an
2855 * inversion. Fix it now when the clamps are applied.
2857 if (unlikely(min_util >= max_util))
2860 return clamp(util, min_util, max_util);
2864 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2865 * by default in the fast path and only gets turned on once userspace performs
2866 * an operation that requires it.
2868 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2871 static inline bool uclamp_is_used(void)
2873 return static_branch_likely(&sched_uclamp_used);
2875 #else /* CONFIG_UCLAMP_TASK */
2877 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2878 struct task_struct *p)
2883 static inline bool uclamp_is_used(void)
2887 #endif /* CONFIG_UCLAMP_TASK */
2889 #ifdef arch_scale_freq_capacity
2890 # ifndef arch_scale_freq_invariant
2891 # define arch_scale_freq_invariant() true
2894 # define arch_scale_freq_invariant() false
2898 static inline unsigned long capacity_orig_of(int cpu)
2900 return cpu_rq(cpu)->cpu_capacity_orig;
2904 * enum cpu_util_type - CPU utilization type
2905 * @FREQUENCY_UTIL: Utilization used to select frequency
2906 * @ENERGY_UTIL: Utilization used during energy calculation
2908 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2909 * need to be aggregated differently depending on the usage made of them. This
2910 * enum is used within effective_cpu_util() to differentiate the types of
2911 * utilization expected by the callers, and adjust the aggregation accordingly.
2913 enum cpu_util_type {
2918 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2919 unsigned long max, enum cpu_util_type type,
2920 struct task_struct *p);
2922 static inline unsigned long cpu_bw_dl(struct rq *rq)
2924 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2927 static inline unsigned long cpu_util_dl(struct rq *rq)
2929 return READ_ONCE(rq->avg_dl.util_avg);
2932 static inline unsigned long cpu_util_cfs(struct rq *rq)
2934 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2936 if (sched_feat(UTIL_EST)) {
2937 util = max_t(unsigned long, util,
2938 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2944 static inline unsigned long cpu_util_rt(struct rq *rq)
2946 return READ_ONCE(rq->avg_rt.util_avg);
2950 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2951 static inline unsigned long cpu_util_irq(struct rq *rq)
2953 return rq->avg_irq.util_avg;
2957 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2959 util *= (max - irq);
2966 static inline unsigned long cpu_util_irq(struct rq *rq)
2972 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2978 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2980 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2982 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2984 static inline bool sched_energy_enabled(void)
2986 return static_branch_unlikely(&sched_energy_present);
2989 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2991 #define perf_domain_span(pd) NULL
2992 static inline bool sched_energy_enabled(void) { return false; }
2994 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2996 #ifdef CONFIG_MEMBARRIER
2998 * The scheduler provides memory barriers required by membarrier between:
2999 * - prior user-space memory accesses and store to rq->membarrier_state,
3000 * - store to rq->membarrier_state and following user-space memory accesses.
3001 * In the same way it provides those guarantees around store to rq->curr.
3003 static inline void membarrier_switch_mm(struct rq *rq,
3004 struct mm_struct *prev_mm,
3005 struct mm_struct *next_mm)
3007 int membarrier_state;
3009 if (prev_mm == next_mm)
3012 membarrier_state = atomic_read(&next_mm->membarrier_state);
3013 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3016 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3019 static inline void membarrier_switch_mm(struct rq *rq,
3020 struct mm_struct *prev_mm,
3021 struct mm_struct *next_mm)
3027 static inline bool is_per_cpu_kthread(struct task_struct *p)
3029 if (!(p->flags & PF_KTHREAD))
3032 if (p->nr_cpus_allowed != 1)
3039 extern void swake_up_all_locked(struct swait_queue_head *q);
3040 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3042 #ifdef CONFIG_PREEMPT_DYNAMIC
3043 extern int preempt_dynamic_mode;
3044 extern int sched_dynamic_mode(const char *str);
3045 extern void sched_dynamic_update(int mode);