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;
399 /* A positive value indicates that this is a SCHED_IDLE group. */
404 * load_avg can be heavily contended at clock tick time, so put
405 * it in its own cacheline separated from the fields above which
406 * will also be accessed at each tick.
408 atomic_long_t load_avg ____cacheline_aligned;
412 #ifdef CONFIG_RT_GROUP_SCHED
413 struct sched_rt_entity **rt_se;
414 struct rt_rq **rt_rq;
416 struct rt_bandwidth rt_bandwidth;
420 struct list_head list;
422 struct task_group *parent;
423 struct list_head siblings;
424 struct list_head children;
426 #ifdef CONFIG_SCHED_AUTOGROUP
427 struct autogroup *autogroup;
430 struct cfs_bandwidth cfs_bandwidth;
432 #ifdef CONFIG_UCLAMP_TASK_GROUP
433 /* The two decimal precision [%] value requested from user-space */
434 unsigned int uclamp_pct[UCLAMP_CNT];
435 /* Clamp values requested for a task group */
436 struct uclamp_se uclamp_req[UCLAMP_CNT];
437 /* Effective clamp values used for a task group */
438 struct uclamp_se uclamp[UCLAMP_CNT];
443 #ifdef CONFIG_FAIR_GROUP_SCHED
444 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
447 * A weight of 0 or 1 can cause arithmetics problems.
448 * A weight of a cfs_rq is the sum of weights of which entities
449 * are queued on this cfs_rq, so a weight of a entity should not be
450 * too large, so as the shares value of a task group.
451 * (The default weight is 1024 - so there's no practical
452 * limitation from this.)
454 #define MIN_SHARES (1UL << 1)
455 #define MAX_SHARES (1UL << 18)
458 typedef int (*tg_visitor)(struct task_group *, void *);
460 extern int walk_tg_tree_from(struct task_group *from,
461 tg_visitor down, tg_visitor up, void *data);
464 * Iterate the full tree, calling @down when first entering a node and @up when
465 * leaving it for the final time.
467 * Caller must hold rcu_lock or sufficient equivalent.
469 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
471 return walk_tg_tree_from(&root_task_group, down, up, data);
474 extern int tg_nop(struct task_group *tg, void *data);
476 extern void free_fair_sched_group(struct task_group *tg);
477 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
478 extern void online_fair_sched_group(struct task_group *tg);
479 extern void unregister_fair_sched_group(struct task_group *tg);
480 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
481 struct sched_entity *se, int cpu,
482 struct sched_entity *parent);
483 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
485 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
486 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
487 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
489 extern void free_rt_sched_group(struct task_group *tg);
490 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
491 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
492 struct sched_rt_entity *rt_se, int cpu,
493 struct sched_rt_entity *parent);
494 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
495 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
496 extern long sched_group_rt_runtime(struct task_group *tg);
497 extern long sched_group_rt_period(struct task_group *tg);
498 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
500 extern struct task_group *sched_create_group(struct task_group *parent);
501 extern void sched_online_group(struct task_group *tg,
502 struct task_group *parent);
503 extern void sched_destroy_group(struct task_group *tg);
504 extern void sched_offline_group(struct task_group *tg);
506 extern void sched_move_task(struct task_struct *tsk);
508 #ifdef CONFIG_FAIR_GROUP_SCHED
509 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
511 extern int sched_group_set_idle(struct task_group *tg, long idle);
514 extern void set_task_rq_fair(struct sched_entity *se,
515 struct cfs_rq *prev, struct cfs_rq *next);
516 #else /* !CONFIG_SMP */
517 static inline void set_task_rq_fair(struct sched_entity *se,
518 struct cfs_rq *prev, struct cfs_rq *next) { }
519 #endif /* CONFIG_SMP */
520 #endif /* CONFIG_FAIR_GROUP_SCHED */
522 #else /* CONFIG_CGROUP_SCHED */
524 struct cfs_bandwidth { };
526 #endif /* CONFIG_CGROUP_SCHED */
528 /* CFS-related fields in a runqueue */
530 struct load_weight load;
531 unsigned int nr_running;
532 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
533 unsigned int idle_h_nr_running; /* SCHED_IDLE */
537 #ifdef CONFIG_SCHED_CORE
538 unsigned int forceidle_seq;
543 u64 min_vruntime_copy;
546 struct rb_root_cached tasks_timeline;
549 * 'curr' points to currently running entity on this cfs_rq.
550 * It is set to NULL otherwise (i.e when none are currently running).
552 struct sched_entity *curr;
553 struct sched_entity *next;
554 struct sched_entity *last;
555 struct sched_entity *skip;
557 #ifdef CONFIG_SCHED_DEBUG
558 unsigned int nr_spread_over;
565 struct sched_avg avg;
567 u64 load_last_update_time_copy;
570 raw_spinlock_t lock ____cacheline_aligned;
572 unsigned long load_avg;
573 unsigned long util_avg;
574 unsigned long runnable_avg;
577 #ifdef CONFIG_FAIR_GROUP_SCHED
578 unsigned long tg_load_avg_contrib;
580 long prop_runnable_sum;
583 * h_load = weight * f(tg)
585 * Where f(tg) is the recursive weight fraction assigned to
588 unsigned long h_load;
589 u64 last_h_load_update;
590 struct sched_entity *h_load_next;
591 #endif /* CONFIG_FAIR_GROUP_SCHED */
592 #endif /* CONFIG_SMP */
594 #ifdef CONFIG_FAIR_GROUP_SCHED
595 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
598 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
599 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
600 * (like users, containers etc.)
602 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
603 * This list is used during load balance.
606 struct list_head leaf_cfs_rq_list;
607 struct task_group *tg; /* group that "owns" this runqueue */
609 /* Locally cached copy of our task_group's idle value */
612 #ifdef CONFIG_CFS_BANDWIDTH
614 s64 runtime_remaining;
617 u64 throttled_clock_task;
618 u64 throttled_clock_task_time;
621 struct list_head throttled_list;
622 #endif /* CONFIG_CFS_BANDWIDTH */
623 #endif /* CONFIG_FAIR_GROUP_SCHED */
626 static inline int rt_bandwidth_enabled(void)
628 return sysctl_sched_rt_runtime >= 0;
631 /* RT IPI pull logic requires IRQ_WORK */
632 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
633 # define HAVE_RT_PUSH_IPI
636 /* Real-Time classes' related field in a runqueue: */
638 struct rt_prio_array active;
639 unsigned int rt_nr_running;
640 unsigned int rr_nr_running;
641 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
643 int curr; /* highest queued rt task prio */
645 int next; /* next highest */
650 unsigned int rt_nr_migratory;
651 unsigned int rt_nr_total;
653 struct plist_head pushable_tasks;
655 #endif /* CONFIG_SMP */
661 /* Nests inside the rq lock: */
662 raw_spinlock_t rt_runtime_lock;
664 #ifdef CONFIG_RT_GROUP_SCHED
665 unsigned int rt_nr_boosted;
668 struct task_group *tg;
672 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
674 return rt_rq->rt_queued && rt_rq->rt_nr_running;
677 /* Deadline class' related fields in a runqueue */
679 /* runqueue is an rbtree, ordered by deadline */
680 struct rb_root_cached root;
682 unsigned int dl_nr_running;
686 * Deadline values of the currently executing and the
687 * earliest ready task on this rq. Caching these facilitates
688 * the decision whether or not a ready but not running task
689 * should migrate somewhere else.
696 unsigned int dl_nr_migratory;
700 * Tasks on this rq that can be pushed away. They are kept in
701 * an rb-tree, ordered by tasks' deadlines, with caching
702 * of the leftmost (earliest deadline) element.
704 struct rb_root_cached pushable_dl_tasks_root;
709 * "Active utilization" for this runqueue: increased when a
710 * task wakes up (becomes TASK_RUNNING) and decreased when a
716 * Utilization of the tasks "assigned" to this runqueue (including
717 * the tasks that are in runqueue and the tasks that executed on this
718 * CPU and blocked). Increased when a task moves to this runqueue, and
719 * decreased when the task moves away (migrates, changes scheduling
720 * policy, or terminates).
721 * This is needed to compute the "inactive utilization" for the
722 * runqueue (inactive utilization = this_bw - running_bw).
728 * Inverse of the fraction of CPU utilization that can be reclaimed
729 * by the GRUB algorithm.
734 #ifdef CONFIG_FAIR_GROUP_SCHED
735 /* An entity is a task if it doesn't "own" a runqueue */
736 #define entity_is_task(se) (!se->my_q)
738 static inline void se_update_runnable(struct sched_entity *se)
740 if (!entity_is_task(se))
741 se->runnable_weight = se->my_q->h_nr_running;
744 static inline long se_runnable(struct sched_entity *se)
746 if (entity_is_task(se))
749 return se->runnable_weight;
753 #define entity_is_task(se) 1
755 static inline void se_update_runnable(struct sched_entity *se) {}
757 static inline long se_runnable(struct sched_entity *se)
765 * XXX we want to get rid of these helpers and use the full load resolution.
767 static inline long se_weight(struct sched_entity *se)
769 return scale_load_down(se->load.weight);
773 static inline bool sched_asym_prefer(int a, int b)
775 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
779 struct em_perf_domain *em_pd;
780 struct perf_domain *next;
784 /* Scheduling group status flags */
785 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
786 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
789 * We add the notion of a root-domain which will be used to define per-domain
790 * variables. Each exclusive cpuset essentially defines an island domain by
791 * fully partitioning the member CPUs from any other cpuset. Whenever a new
792 * exclusive cpuset is created, we also create and attach a new root-domain
801 cpumask_var_t online;
804 * Indicate pullable load on at least one CPU, e.g:
805 * - More than one runnable task
806 * - Running task is misfit
810 /* Indicate one or more cpus over-utilized (tipping point) */
814 * The bit corresponding to a CPU gets set here if such CPU has more
815 * than one runnable -deadline task (as it is below for RT tasks).
817 cpumask_var_t dlo_mask;
823 * Indicate whether a root_domain's dl_bw has been checked or
824 * updated. It's monotonously increasing value.
826 * Also, some corner cases, like 'wrap around' is dangerous, but given
827 * that u64 is 'big enough'. So that shouldn't be a concern.
831 #ifdef HAVE_RT_PUSH_IPI
833 * For IPI pull requests, loop across the rto_mask.
835 struct irq_work rto_push_work;
836 raw_spinlock_t rto_lock;
837 /* These are only updated and read within rto_lock */
840 /* These atomics are updated outside of a lock */
841 atomic_t rto_loop_next;
842 atomic_t rto_loop_start;
845 * The "RT overload" flag: it gets set if a CPU has more than
846 * one runnable RT task.
848 cpumask_var_t rto_mask;
849 struct cpupri cpupri;
851 unsigned long max_cpu_capacity;
854 * NULL-terminated list of performance domains intersecting with the
855 * CPUs of the rd. Protected by RCU.
857 struct perf_domain __rcu *pd;
860 extern void init_defrootdomain(void);
861 extern int sched_init_domains(const struct cpumask *cpu_map);
862 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
863 extern void sched_get_rd(struct root_domain *rd);
864 extern void sched_put_rd(struct root_domain *rd);
866 #ifdef HAVE_RT_PUSH_IPI
867 extern void rto_push_irq_work_func(struct irq_work *work);
869 #endif /* CONFIG_SMP */
871 #ifdef CONFIG_UCLAMP_TASK
873 * struct uclamp_bucket - Utilization clamp bucket
874 * @value: utilization clamp value for tasks on this clamp bucket
875 * @tasks: number of RUNNABLE tasks on this clamp bucket
877 * Keep track of how many tasks are RUNNABLE for a given utilization
880 struct uclamp_bucket {
881 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
882 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
886 * struct uclamp_rq - rq's utilization clamp
887 * @value: currently active clamp values for a rq
888 * @bucket: utilization clamp buckets affecting a rq
890 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
891 * A clamp value is affecting a rq when there is at least one task RUNNABLE
892 * (or actually running) with that value.
894 * There are up to UCLAMP_CNT possible different clamp values, currently there
895 * are only two: minimum utilization and maximum utilization.
897 * All utilization clamping values are MAX aggregated, since:
898 * - for util_min: we want to run the CPU at least at the max of the minimum
899 * utilization required by its currently RUNNABLE tasks.
900 * - for util_max: we want to allow the CPU to run up to the max of the
901 * maximum utilization allowed by its currently RUNNABLE tasks.
903 * Since on each system we expect only a limited number of different
904 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
905 * the metrics required to compute all the per-rq utilization clamp values.
909 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
912 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
913 #endif /* CONFIG_UCLAMP_TASK */
916 * This is the main, per-CPU runqueue data structure.
918 * Locking rule: those places that want to lock multiple runqueues
919 * (such as the load balancing or the thread migration code), lock
920 * acquire operations must be ordered by ascending &runqueue.
924 raw_spinlock_t __lock;
927 * nr_running and cpu_load should be in the same cacheline because
928 * remote CPUs use both these fields when doing load calculation.
930 unsigned int nr_running;
931 #ifdef CONFIG_NUMA_BALANCING
932 unsigned int nr_numa_running;
933 unsigned int nr_preferred_running;
934 unsigned int numa_migrate_on;
936 #ifdef CONFIG_NO_HZ_COMMON
938 unsigned long last_blocked_load_update_tick;
939 unsigned int has_blocked_load;
940 call_single_data_t nohz_csd;
941 #endif /* CONFIG_SMP */
942 unsigned int nohz_tick_stopped;
944 #endif /* CONFIG_NO_HZ_COMMON */
947 unsigned int ttwu_pending;
951 #ifdef CONFIG_UCLAMP_TASK
952 /* Utilization clamp values based on CPU's RUNNABLE tasks */
953 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
954 unsigned int uclamp_flags;
955 #define UCLAMP_FLAG_IDLE 0x01
962 #ifdef CONFIG_FAIR_GROUP_SCHED
963 /* list of leaf cfs_rq on this CPU: */
964 struct list_head leaf_cfs_rq_list;
965 struct list_head *tmp_alone_branch;
966 #endif /* CONFIG_FAIR_GROUP_SCHED */
969 * This is part of a global counter where only the total sum
970 * over all CPUs matters. A task can increase this counter on
971 * one CPU and if it got migrated afterwards it may decrease
972 * it on another CPU. Always updated under the runqueue lock:
974 unsigned int nr_uninterruptible;
976 struct task_struct __rcu *curr;
977 struct task_struct *idle;
978 struct task_struct *stop;
979 unsigned long next_balance;
980 struct mm_struct *prev_mm;
982 unsigned int clock_update_flags;
984 /* Ensure that all clocks are in the same cache line */
985 u64 clock_task ____cacheline_aligned;
987 unsigned long lost_idle_time;
991 #ifdef CONFIG_SCHED_DEBUG
992 u64 last_seen_need_resched_ns;
993 int ticks_without_resched;
996 #ifdef CONFIG_MEMBARRIER
997 int membarrier_state;
1001 struct root_domain *rd;
1002 struct sched_domain __rcu *sd;
1004 unsigned long cpu_capacity;
1005 unsigned long cpu_capacity_orig;
1007 struct callback_head *balance_callback;
1009 unsigned char nohz_idle_balance;
1010 unsigned char idle_balance;
1012 unsigned long misfit_task_load;
1014 /* For active balancing */
1017 struct cpu_stop_work active_balance_work;
1019 /* CPU of this runqueue: */
1023 struct list_head cfs_tasks;
1025 struct sched_avg avg_rt;
1026 struct sched_avg avg_dl;
1027 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1028 struct sched_avg avg_irq;
1030 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1031 struct sched_avg avg_thermal;
1036 unsigned long wake_stamp;
1039 /* This is used to determine avg_idle's max value */
1040 u64 max_idle_balance_cost;
1042 #ifdef CONFIG_HOTPLUG_CPU
1043 struct rcuwait hotplug_wait;
1045 #endif /* CONFIG_SMP */
1047 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1050 #ifdef CONFIG_PARAVIRT
1051 u64 prev_steal_time;
1053 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1054 u64 prev_steal_time_rq;
1057 /* calc_load related fields */
1058 unsigned long calc_load_update;
1059 long calc_load_active;
1061 #ifdef CONFIG_SCHED_HRTICK
1063 call_single_data_t hrtick_csd;
1065 struct hrtimer hrtick_timer;
1066 ktime_t hrtick_time;
1069 #ifdef CONFIG_SCHEDSTATS
1071 struct sched_info rq_sched_info;
1072 unsigned long long rq_cpu_time;
1073 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1075 /* sys_sched_yield() stats */
1076 unsigned int yld_count;
1078 /* schedule() stats */
1079 unsigned int sched_count;
1080 unsigned int sched_goidle;
1082 /* try_to_wake_up() stats */
1083 unsigned int ttwu_count;
1084 unsigned int ttwu_local;
1087 #ifdef CONFIG_CPU_IDLE
1088 /* Must be inspected within a rcu lock section */
1089 struct cpuidle_state *idle_state;
1093 unsigned int nr_pinned;
1095 unsigned int push_busy;
1096 struct cpu_stop_work push_work;
1098 #ifdef CONFIG_SCHED_CORE
1101 struct task_struct *core_pick;
1102 unsigned int core_enabled;
1103 unsigned int core_sched_seq;
1104 struct rb_root core_tree;
1107 unsigned int core_task_seq;
1108 unsigned int core_pick_seq;
1109 unsigned long core_cookie;
1110 unsigned char core_forceidle;
1111 unsigned int core_forceidle_seq;
1115 #ifdef CONFIG_FAIR_GROUP_SCHED
1117 /* CPU runqueue to which this cfs_rq is attached */
1118 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1125 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1127 return container_of(cfs_rq, struct rq, cfs);
1131 static inline int cpu_of(struct rq *rq)
1140 #define MDF_PUSH 0x01
1142 static inline bool is_migration_disabled(struct task_struct *p)
1145 return p->migration_disabled;
1152 #ifdef CONFIG_SCHED_CORE
1153 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1155 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1157 static inline bool sched_core_enabled(struct rq *rq)
1159 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1162 static inline bool sched_core_disabled(void)
1164 return !static_branch_unlikely(&__sched_core_enabled);
1168 * Be careful with this function; not for general use. The return value isn't
1169 * stable unless you actually hold a relevant rq->__lock.
1171 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1173 if (sched_core_enabled(rq))
1174 return &rq->core->__lock;
1179 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1181 if (rq->core_enabled)
1182 return &rq->core->__lock;
1187 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1190 * Helpers to check if the CPU's core cookie matches with the task's cookie
1191 * when core scheduling is enabled.
1192 * A special case is that the task's cookie always matches with CPU's core
1193 * cookie if the CPU is in an idle core.
1195 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1197 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1198 if (!sched_core_enabled(rq))
1201 return rq->core->core_cookie == p->core_cookie;
1204 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1206 bool idle_core = true;
1209 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1210 if (!sched_core_enabled(rq))
1213 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1214 if (!available_idle_cpu(cpu)) {
1221 * A CPU in an idle core is always the best choice for tasks with
1224 return idle_core || rq->core->core_cookie == p->core_cookie;
1227 static inline bool sched_group_cookie_match(struct rq *rq,
1228 struct task_struct *p,
1229 struct sched_group *group)
1233 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1234 if (!sched_core_enabled(rq))
1237 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1238 if (sched_core_cookie_match(rq, p))
1244 extern void queue_core_balance(struct rq *rq);
1246 static inline bool sched_core_enqueued(struct task_struct *p)
1248 return !RB_EMPTY_NODE(&p->core_node);
1251 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1252 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1254 extern void sched_core_get(void);
1255 extern void sched_core_put(void);
1257 extern unsigned long sched_core_alloc_cookie(void);
1258 extern void sched_core_put_cookie(unsigned long cookie);
1259 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1260 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1262 #else /* !CONFIG_SCHED_CORE */
1264 static inline bool sched_core_enabled(struct rq *rq)
1269 static inline bool sched_core_disabled(void)
1274 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1279 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1284 static inline void queue_core_balance(struct rq *rq)
1288 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1293 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1298 static inline bool sched_group_cookie_match(struct rq *rq,
1299 struct task_struct *p,
1300 struct sched_group *group)
1304 #endif /* CONFIG_SCHED_CORE */
1306 static inline void lockdep_assert_rq_held(struct rq *rq)
1308 lockdep_assert_held(__rq_lockp(rq));
1311 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1312 extern bool raw_spin_rq_trylock(struct rq *rq);
1313 extern void raw_spin_rq_unlock(struct rq *rq);
1315 static inline void raw_spin_rq_lock(struct rq *rq)
1317 raw_spin_rq_lock_nested(rq, 0);
1320 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1322 local_irq_disable();
1323 raw_spin_rq_lock(rq);
1326 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1328 raw_spin_rq_unlock(rq);
1332 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1334 unsigned long flags;
1335 local_irq_save(flags);
1336 raw_spin_rq_lock(rq);
1340 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1342 raw_spin_rq_unlock(rq);
1343 local_irq_restore(flags);
1346 #define raw_spin_rq_lock_irqsave(rq, flags) \
1348 flags = _raw_spin_rq_lock_irqsave(rq); \
1351 #ifdef CONFIG_SCHED_SMT
1352 extern void __update_idle_core(struct rq *rq);
1354 static inline void update_idle_core(struct rq *rq)
1356 if (static_branch_unlikely(&sched_smt_present))
1357 __update_idle_core(rq);
1361 static inline void update_idle_core(struct rq *rq) { }
1364 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1366 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1367 #define this_rq() this_cpu_ptr(&runqueues)
1368 #define task_rq(p) cpu_rq(task_cpu(p))
1369 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1370 #define raw_rq() raw_cpu_ptr(&runqueues)
1372 #ifdef CONFIG_FAIR_GROUP_SCHED
1373 static inline struct task_struct *task_of(struct sched_entity *se)
1375 SCHED_WARN_ON(!entity_is_task(se));
1376 return container_of(se, struct task_struct, se);
1379 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1381 return p->se.cfs_rq;
1384 /* runqueue on which this entity is (to be) queued */
1385 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1390 /* runqueue "owned" by this group */
1391 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1398 static inline struct task_struct *task_of(struct sched_entity *se)
1400 return container_of(se, struct task_struct, se);
1403 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1405 return &task_rq(p)->cfs;
1408 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1410 struct task_struct *p = task_of(se);
1411 struct rq *rq = task_rq(p);
1416 /* runqueue "owned" by this group */
1417 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1423 extern void update_rq_clock(struct rq *rq);
1425 static inline u64 __rq_clock_broken(struct rq *rq)
1427 return READ_ONCE(rq->clock);
1431 * rq::clock_update_flags bits
1433 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1434 * call to __schedule(). This is an optimisation to avoid
1435 * neighbouring rq clock updates.
1437 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1438 * in effect and calls to update_rq_clock() are being ignored.
1440 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1441 * made to update_rq_clock() since the last time rq::lock was pinned.
1443 * If inside of __schedule(), clock_update_flags will have been
1444 * shifted left (a left shift is a cheap operation for the fast path
1445 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1447 * if (rq-clock_update_flags >= RQCF_UPDATED)
1449 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1450 * one position though, because the next rq_unpin_lock() will shift it
1453 #define RQCF_REQ_SKIP 0x01
1454 #define RQCF_ACT_SKIP 0x02
1455 #define RQCF_UPDATED 0x04
1457 static inline void assert_clock_updated(struct rq *rq)
1460 * The only reason for not seeing a clock update since the
1461 * last rq_pin_lock() is if we're currently skipping updates.
1463 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1466 static inline u64 rq_clock(struct rq *rq)
1468 lockdep_assert_rq_held(rq);
1469 assert_clock_updated(rq);
1474 static inline u64 rq_clock_task(struct rq *rq)
1476 lockdep_assert_rq_held(rq);
1477 assert_clock_updated(rq);
1479 return rq->clock_task;
1483 * By default the decay is the default pelt decay period.
1484 * The decay shift can change the decay period in
1486 * Decay shift Decay period(ms)
1493 extern int sched_thermal_decay_shift;
1495 static inline u64 rq_clock_thermal(struct rq *rq)
1497 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1500 static inline void rq_clock_skip_update(struct rq *rq)
1502 lockdep_assert_rq_held(rq);
1503 rq->clock_update_flags |= RQCF_REQ_SKIP;
1507 * See rt task throttling, which is the only time a skip
1508 * request is canceled.
1510 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1512 lockdep_assert_rq_held(rq);
1513 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1517 unsigned long flags;
1518 struct pin_cookie cookie;
1519 #ifdef CONFIG_SCHED_DEBUG
1521 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1522 * current pin context is stashed here in case it needs to be
1523 * restored in rq_repin_lock().
1525 unsigned int clock_update_flags;
1529 extern struct callback_head balance_push_callback;
1532 * Lockdep annotation that avoids accidental unlocks; it's like a
1533 * sticky/continuous lockdep_assert_held().
1535 * This avoids code that has access to 'struct rq *rq' (basically everything in
1536 * the scheduler) from accidentally unlocking the rq if they do not also have a
1537 * copy of the (on-stack) 'struct rq_flags rf'.
1539 * Also see Documentation/locking/lockdep-design.rst.
1541 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1543 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1545 #ifdef CONFIG_SCHED_DEBUG
1546 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1547 rf->clock_update_flags = 0;
1549 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1554 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1556 #ifdef CONFIG_SCHED_DEBUG
1557 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1558 rf->clock_update_flags = RQCF_UPDATED;
1561 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1564 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1566 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1568 #ifdef CONFIG_SCHED_DEBUG
1570 * Restore the value we stashed in @rf for this pin context.
1572 rq->clock_update_flags |= rf->clock_update_flags;
1576 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1577 __acquires(rq->lock);
1579 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1580 __acquires(p->pi_lock)
1581 __acquires(rq->lock);
1583 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1584 __releases(rq->lock)
1586 rq_unpin_lock(rq, rf);
1587 raw_spin_rq_unlock(rq);
1591 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1592 __releases(rq->lock)
1593 __releases(p->pi_lock)
1595 rq_unpin_lock(rq, rf);
1596 raw_spin_rq_unlock(rq);
1597 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1601 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1602 __acquires(rq->lock)
1604 raw_spin_rq_lock_irqsave(rq, rf->flags);
1605 rq_pin_lock(rq, rf);
1609 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1610 __acquires(rq->lock)
1612 raw_spin_rq_lock_irq(rq);
1613 rq_pin_lock(rq, rf);
1617 rq_lock(struct rq *rq, struct rq_flags *rf)
1618 __acquires(rq->lock)
1620 raw_spin_rq_lock(rq);
1621 rq_pin_lock(rq, rf);
1625 rq_relock(struct rq *rq, struct rq_flags *rf)
1626 __acquires(rq->lock)
1628 raw_spin_rq_lock(rq);
1629 rq_repin_lock(rq, rf);
1633 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1634 __releases(rq->lock)
1636 rq_unpin_lock(rq, rf);
1637 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1641 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1642 __releases(rq->lock)
1644 rq_unpin_lock(rq, rf);
1645 raw_spin_rq_unlock_irq(rq);
1649 rq_unlock(struct rq *rq, struct rq_flags *rf)
1650 __releases(rq->lock)
1652 rq_unpin_lock(rq, rf);
1653 raw_spin_rq_unlock(rq);
1656 static inline struct rq *
1657 this_rq_lock_irq(struct rq_flags *rf)
1658 __acquires(rq->lock)
1662 local_irq_disable();
1669 enum numa_topology_type {
1674 extern enum numa_topology_type sched_numa_topology_type;
1675 extern int sched_max_numa_distance;
1676 extern bool find_numa_distance(int distance);
1677 extern void sched_init_numa(void);
1678 extern void sched_domains_numa_masks_set(unsigned int cpu);
1679 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1680 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1682 static inline void sched_init_numa(void) { }
1683 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1684 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1685 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1691 #ifdef CONFIG_NUMA_BALANCING
1692 /* The regions in numa_faults array from task_struct */
1693 enum numa_faults_stats {
1699 extern void sched_setnuma(struct task_struct *p, int node);
1700 extern int migrate_task_to(struct task_struct *p, int cpu);
1701 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1703 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1706 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1709 #endif /* CONFIG_NUMA_BALANCING */
1714 queue_balance_callback(struct rq *rq,
1715 struct callback_head *head,
1716 void (*func)(struct rq *rq))
1718 lockdep_assert_rq_held(rq);
1720 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1723 head->func = (void (*)(struct callback_head *))func;
1724 head->next = rq->balance_callback;
1725 rq->balance_callback = head;
1728 #define rcu_dereference_check_sched_domain(p) \
1729 rcu_dereference_check((p), \
1730 lockdep_is_held(&sched_domains_mutex))
1733 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1734 * See destroy_sched_domains: call_rcu for details.
1736 * The domain tree of any CPU may only be accessed from within
1737 * preempt-disabled sections.
1739 #define for_each_domain(cpu, __sd) \
1740 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1741 __sd; __sd = __sd->parent)
1744 * highest_flag_domain - Return highest sched_domain containing flag.
1745 * @cpu: The CPU whose highest level of sched domain is to
1747 * @flag: The flag to check for the highest sched_domain
1748 * for the given CPU.
1750 * Returns the highest sched_domain of a CPU which contains the given flag.
1752 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1754 struct sched_domain *sd, *hsd = NULL;
1756 for_each_domain(cpu, sd) {
1757 if (!(sd->flags & flag))
1765 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1767 struct sched_domain *sd;
1769 for_each_domain(cpu, sd) {
1770 if (sd->flags & flag)
1777 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1778 DECLARE_PER_CPU(int, sd_llc_size);
1779 DECLARE_PER_CPU(int, sd_llc_id);
1780 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1781 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1782 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1783 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1784 extern struct static_key_false sched_asym_cpucapacity;
1786 struct sched_group_capacity {
1789 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1792 unsigned long capacity;
1793 unsigned long min_capacity; /* Min per-CPU capacity in group */
1794 unsigned long max_capacity; /* Max per-CPU capacity in group */
1795 unsigned long next_update;
1796 int imbalance; /* XXX unrelated to capacity but shared group state */
1798 #ifdef CONFIG_SCHED_DEBUG
1802 unsigned long cpumask[]; /* Balance mask */
1805 struct sched_group {
1806 struct sched_group *next; /* Must be a circular list */
1809 unsigned int group_weight;
1810 struct sched_group_capacity *sgc;
1811 int asym_prefer_cpu; /* CPU of highest priority in group */
1814 * The CPUs this group covers.
1816 * NOTE: this field is variable length. (Allocated dynamically
1817 * by attaching extra space to the end of the structure,
1818 * depending on how many CPUs the kernel has booted up with)
1820 unsigned long cpumask[];
1823 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1825 return to_cpumask(sg->cpumask);
1829 * See build_balance_mask().
1831 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1833 return to_cpumask(sg->sgc->cpumask);
1837 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1838 * @group: The group whose first CPU is to be returned.
1840 static inline unsigned int group_first_cpu(struct sched_group *group)
1842 return cpumask_first(sched_group_span(group));
1845 extern int group_balance_cpu(struct sched_group *sg);
1847 #ifdef CONFIG_SCHED_DEBUG
1848 void update_sched_domain_debugfs(void);
1849 void dirty_sched_domain_sysctl(int cpu);
1851 static inline void update_sched_domain_debugfs(void)
1854 static inline void dirty_sched_domain_sysctl(int cpu)
1859 extern int sched_update_scaling(void);
1861 extern void flush_smp_call_function_from_idle(void);
1863 #else /* !CONFIG_SMP: */
1864 static inline void flush_smp_call_function_from_idle(void) { }
1868 #include "autogroup.h"
1870 #ifdef CONFIG_CGROUP_SCHED
1873 * Return the group to which this tasks belongs.
1875 * We cannot use task_css() and friends because the cgroup subsystem
1876 * changes that value before the cgroup_subsys::attach() method is called,
1877 * therefore we cannot pin it and might observe the wrong value.
1879 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1880 * core changes this before calling sched_move_task().
1882 * Instead we use a 'copy' which is updated from sched_move_task() while
1883 * holding both task_struct::pi_lock and rq::lock.
1885 static inline struct task_group *task_group(struct task_struct *p)
1887 return p->sched_task_group;
1890 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1891 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1893 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1894 struct task_group *tg = task_group(p);
1897 #ifdef CONFIG_FAIR_GROUP_SCHED
1898 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1899 p->se.cfs_rq = tg->cfs_rq[cpu];
1900 p->se.parent = tg->se[cpu];
1903 #ifdef CONFIG_RT_GROUP_SCHED
1904 p->rt.rt_rq = tg->rt_rq[cpu];
1905 p->rt.parent = tg->rt_se[cpu];
1909 #else /* CONFIG_CGROUP_SCHED */
1911 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1912 static inline struct task_group *task_group(struct task_struct *p)
1917 #endif /* CONFIG_CGROUP_SCHED */
1919 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1921 set_task_rq(p, cpu);
1924 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1925 * successfully executed on another CPU. We must ensure that updates of
1926 * per-task data have been completed by this moment.
1929 #ifdef CONFIG_THREAD_INFO_IN_TASK
1930 WRITE_ONCE(p->cpu, cpu);
1932 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1939 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1941 #ifdef CONFIG_SCHED_DEBUG
1942 # include <linux/static_key.h>
1943 # define const_debug __read_mostly
1945 # define const_debug const
1948 #define SCHED_FEAT(name, enabled) \
1949 __SCHED_FEAT_##name ,
1952 #include "features.h"
1958 #ifdef CONFIG_SCHED_DEBUG
1961 * To support run-time toggling of sched features, all the translation units
1962 * (but core.c) reference the sysctl_sched_features defined in core.c.
1964 extern const_debug unsigned int sysctl_sched_features;
1966 #ifdef CONFIG_JUMP_LABEL
1967 #define SCHED_FEAT(name, enabled) \
1968 static __always_inline bool static_branch_##name(struct static_key *key) \
1970 return static_key_##enabled(key); \
1973 #include "features.h"
1976 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1977 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1979 #else /* !CONFIG_JUMP_LABEL */
1981 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1983 #endif /* CONFIG_JUMP_LABEL */
1985 #else /* !SCHED_DEBUG */
1988 * Each translation unit has its own copy of sysctl_sched_features to allow
1989 * constants propagation at compile time and compiler optimization based on
1992 #define SCHED_FEAT(name, enabled) \
1993 (1UL << __SCHED_FEAT_##name) * enabled |
1994 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1995 #include "features.h"
1999 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2001 #endif /* SCHED_DEBUG */
2003 extern struct static_key_false sched_numa_balancing;
2004 extern struct static_key_false sched_schedstats;
2006 static inline u64 global_rt_period(void)
2008 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2011 static inline u64 global_rt_runtime(void)
2013 if (sysctl_sched_rt_runtime < 0)
2016 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2019 static inline int task_current(struct rq *rq, struct task_struct *p)
2021 return rq->curr == p;
2024 static inline int task_running(struct rq *rq, struct task_struct *p)
2029 return task_current(rq, p);
2033 static inline int task_on_rq_queued(struct task_struct *p)
2035 return p->on_rq == TASK_ON_RQ_QUEUED;
2038 static inline int task_on_rq_migrating(struct task_struct *p)
2040 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2043 /* Wake flags. The first three directly map to some SD flag value */
2044 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2045 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2046 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2048 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2049 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2050 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2053 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2054 static_assert(WF_FORK == SD_BALANCE_FORK);
2055 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2059 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2060 * of tasks with abnormal "nice" values across CPUs the contribution that
2061 * each task makes to its run queue's load is weighted according to its
2062 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2063 * scaled version of the new time slice allocation that they receive on time
2067 #define WEIGHT_IDLEPRIO 3
2068 #define WMULT_IDLEPRIO 1431655765
2070 extern const int sched_prio_to_weight[40];
2071 extern const u32 sched_prio_to_wmult[40];
2074 * {de,en}queue flags:
2076 * DEQUEUE_SLEEP - task is no longer runnable
2077 * ENQUEUE_WAKEUP - task just became runnable
2079 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2080 * are in a known state which allows modification. Such pairs
2081 * should preserve as much state as possible.
2083 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2086 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2087 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2088 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2092 #define DEQUEUE_SLEEP 0x01
2093 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2094 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2095 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2097 #define ENQUEUE_WAKEUP 0x01
2098 #define ENQUEUE_RESTORE 0x02
2099 #define ENQUEUE_MOVE 0x04
2100 #define ENQUEUE_NOCLOCK 0x08
2102 #define ENQUEUE_HEAD 0x10
2103 #define ENQUEUE_REPLENISH 0x20
2105 #define ENQUEUE_MIGRATED 0x40
2107 #define ENQUEUE_MIGRATED 0x00
2110 #define RETRY_TASK ((void *)-1UL)
2112 struct sched_class {
2114 #ifdef CONFIG_UCLAMP_TASK
2118 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2119 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2120 void (*yield_task) (struct rq *rq);
2121 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2123 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2125 struct task_struct *(*pick_next_task)(struct rq *rq);
2127 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2128 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2131 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2132 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2134 struct task_struct * (*pick_task)(struct rq *rq);
2136 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2138 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2140 void (*set_cpus_allowed)(struct task_struct *p,
2141 const struct cpumask *newmask,
2144 void (*rq_online)(struct rq *rq);
2145 void (*rq_offline)(struct rq *rq);
2147 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2150 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2151 void (*task_fork)(struct task_struct *p);
2152 void (*task_dead)(struct task_struct *p);
2155 * The switched_from() call is allowed to drop rq->lock, therefore we
2156 * cannot assume the switched_from/switched_to pair is serialized by
2157 * rq->lock. They are however serialized by p->pi_lock.
2159 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2160 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2161 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2164 unsigned int (*get_rr_interval)(struct rq *rq,
2165 struct task_struct *task);
2167 void (*update_curr)(struct rq *rq);
2169 #define TASK_SET_GROUP 0
2170 #define TASK_MOVE_GROUP 1
2172 #ifdef CONFIG_FAIR_GROUP_SCHED
2173 void (*task_change_group)(struct task_struct *p, int type);
2177 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2179 WARN_ON_ONCE(rq->curr != prev);
2180 prev->sched_class->put_prev_task(rq, prev);
2183 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2185 next->sched_class->set_next_task(rq, next, false);
2190 * Helper to define a sched_class instance; each one is placed in a separate
2191 * section which is ordered by the linker script:
2193 * include/asm-generic/vmlinux.lds.h
2195 * Also enforce alignment on the instance, not the type, to guarantee layout.
2197 #define DEFINE_SCHED_CLASS(name) \
2198 const struct sched_class name##_sched_class \
2199 __aligned(__alignof__(struct sched_class)) \
2200 __section("__" #name "_sched_class")
2202 /* Defined in include/asm-generic/vmlinux.lds.h */
2203 extern struct sched_class __begin_sched_classes[];
2204 extern struct sched_class __end_sched_classes[];
2206 #define sched_class_highest (__end_sched_classes - 1)
2207 #define sched_class_lowest (__begin_sched_classes - 1)
2209 #define for_class_range(class, _from, _to) \
2210 for (class = (_from); class != (_to); class--)
2212 #define for_each_class(class) \
2213 for_class_range(class, sched_class_highest, sched_class_lowest)
2215 extern const struct sched_class stop_sched_class;
2216 extern const struct sched_class dl_sched_class;
2217 extern const struct sched_class rt_sched_class;
2218 extern const struct sched_class fair_sched_class;
2219 extern const struct sched_class idle_sched_class;
2221 static inline bool sched_stop_runnable(struct rq *rq)
2223 return rq->stop && task_on_rq_queued(rq->stop);
2226 static inline bool sched_dl_runnable(struct rq *rq)
2228 return rq->dl.dl_nr_running > 0;
2231 static inline bool sched_rt_runnable(struct rq *rq)
2233 return rq->rt.rt_queued > 0;
2236 static inline bool sched_fair_runnable(struct rq *rq)
2238 return rq->cfs.nr_running > 0;
2241 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2242 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2244 #define SCA_CHECK 0x01
2245 #define SCA_MIGRATE_DISABLE 0x02
2246 #define SCA_MIGRATE_ENABLE 0x04
2247 #define SCA_USER 0x08
2251 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2253 extern void trigger_load_balance(struct rq *rq);
2255 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2257 static inline struct task_struct *get_push_task(struct rq *rq)
2259 struct task_struct *p = rq->curr;
2261 lockdep_assert_rq_held(rq);
2266 if (p->nr_cpus_allowed == 1)
2269 rq->push_busy = true;
2270 return get_task_struct(p);
2273 extern int push_cpu_stop(void *arg);
2277 #ifdef CONFIG_CPU_IDLE
2278 static inline void idle_set_state(struct rq *rq,
2279 struct cpuidle_state *idle_state)
2281 rq->idle_state = idle_state;
2284 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2286 SCHED_WARN_ON(!rcu_read_lock_held());
2288 return rq->idle_state;
2291 static inline void idle_set_state(struct rq *rq,
2292 struct cpuidle_state *idle_state)
2296 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2302 extern void schedule_idle(void);
2304 extern void sysrq_sched_debug_show(void);
2305 extern void sched_init_granularity(void);
2306 extern void update_max_interval(void);
2308 extern void init_sched_dl_class(void);
2309 extern void init_sched_rt_class(void);
2310 extern void init_sched_fair_class(void);
2312 extern void reweight_task(struct task_struct *p, int prio);
2314 extern void resched_curr(struct rq *rq);
2315 extern void resched_cpu(int cpu);
2317 extern struct rt_bandwidth def_rt_bandwidth;
2318 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2320 extern struct dl_bandwidth def_dl_bandwidth;
2321 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2322 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2323 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2326 #define BW_UNIT (1 << BW_SHIFT)
2327 #define RATIO_SHIFT 8
2328 #define MAX_BW_BITS (64 - BW_SHIFT)
2329 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2330 unsigned long to_ratio(u64 period, u64 runtime);
2332 extern void init_entity_runnable_average(struct sched_entity *se);
2333 extern void post_init_entity_util_avg(struct task_struct *p);
2335 #ifdef CONFIG_NO_HZ_FULL
2336 extern bool sched_can_stop_tick(struct rq *rq);
2337 extern int __init sched_tick_offload_init(void);
2340 * Tick may be needed by tasks in the runqueue depending on their policy and
2341 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2342 * nohz mode if necessary.
2344 static inline void sched_update_tick_dependency(struct rq *rq)
2346 int cpu = cpu_of(rq);
2348 if (!tick_nohz_full_cpu(cpu))
2351 if (sched_can_stop_tick(rq))
2352 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2354 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2357 static inline int sched_tick_offload_init(void) { return 0; }
2358 static inline void sched_update_tick_dependency(struct rq *rq) { }
2361 static inline void add_nr_running(struct rq *rq, unsigned count)
2363 unsigned prev_nr = rq->nr_running;
2365 rq->nr_running = prev_nr + count;
2366 if (trace_sched_update_nr_running_tp_enabled()) {
2367 call_trace_sched_update_nr_running(rq, count);
2371 if (prev_nr < 2 && rq->nr_running >= 2) {
2372 if (!READ_ONCE(rq->rd->overload))
2373 WRITE_ONCE(rq->rd->overload, 1);
2377 sched_update_tick_dependency(rq);
2380 static inline void sub_nr_running(struct rq *rq, unsigned count)
2382 rq->nr_running -= count;
2383 if (trace_sched_update_nr_running_tp_enabled()) {
2384 call_trace_sched_update_nr_running(rq, -count);
2387 /* Check if we still need preemption */
2388 sched_update_tick_dependency(rq);
2391 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2392 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2394 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2396 extern const_debug unsigned int sysctl_sched_nr_migrate;
2397 extern const_debug unsigned int sysctl_sched_migration_cost;
2399 #ifdef CONFIG_SCHED_DEBUG
2400 extern unsigned int sysctl_sched_latency;
2401 extern unsigned int sysctl_sched_min_granularity;
2402 extern unsigned int sysctl_sched_wakeup_granularity;
2403 extern int sysctl_resched_latency_warn_ms;
2404 extern int sysctl_resched_latency_warn_once;
2406 extern unsigned int sysctl_sched_tunable_scaling;
2408 extern unsigned int sysctl_numa_balancing_scan_delay;
2409 extern unsigned int sysctl_numa_balancing_scan_period_min;
2410 extern unsigned int sysctl_numa_balancing_scan_period_max;
2411 extern unsigned int sysctl_numa_balancing_scan_size;
2414 #ifdef CONFIG_SCHED_HRTICK
2418 * - enabled by features
2419 * - hrtimer is actually high res
2421 static inline int hrtick_enabled(struct rq *rq)
2423 if (!cpu_active(cpu_of(rq)))
2425 return hrtimer_is_hres_active(&rq->hrtick_timer);
2428 static inline int hrtick_enabled_fair(struct rq *rq)
2430 if (!sched_feat(HRTICK))
2432 return hrtick_enabled(rq);
2435 static inline int hrtick_enabled_dl(struct rq *rq)
2437 if (!sched_feat(HRTICK_DL))
2439 return hrtick_enabled(rq);
2442 void hrtick_start(struct rq *rq, u64 delay);
2446 static inline int hrtick_enabled_fair(struct rq *rq)
2451 static inline int hrtick_enabled_dl(struct rq *rq)
2456 static inline int hrtick_enabled(struct rq *rq)
2461 #endif /* CONFIG_SCHED_HRTICK */
2463 #ifndef arch_scale_freq_tick
2464 static __always_inline
2465 void arch_scale_freq_tick(void)
2470 #ifndef arch_scale_freq_capacity
2472 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2473 * @cpu: the CPU in question.
2475 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2478 * ------ * SCHED_CAPACITY_SCALE
2481 static __always_inline
2482 unsigned long arch_scale_freq_capacity(int cpu)
2484 return SCHED_CAPACITY_SCALE;
2491 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2493 #ifdef CONFIG_SCHED_CORE
2495 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2496 * order by core-id first and cpu-id second.
2500 * double_rq_lock(0,3); will take core-0, core-1 lock
2501 * double_rq_lock(1,2); will take core-1, core-0 lock
2503 * when only cpu-id is considered.
2505 if (rq1->core->cpu < rq2->core->cpu)
2507 if (rq1->core->cpu > rq2->core->cpu)
2511 * __sched_core_flip() relies on SMT having cpu-id lock order.
2514 return rq1->cpu < rq2->cpu;
2517 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2519 #ifdef CONFIG_PREEMPTION
2522 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2523 * way at the expense of forcing extra atomic operations in all
2524 * invocations. This assures that the double_lock is acquired using the
2525 * same underlying policy as the spinlock_t on this architecture, which
2526 * reduces latency compared to the unfair variant below. However, it
2527 * also adds more overhead and therefore may reduce throughput.
2529 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2530 __releases(this_rq->lock)
2531 __acquires(busiest->lock)
2532 __acquires(this_rq->lock)
2534 raw_spin_rq_unlock(this_rq);
2535 double_rq_lock(this_rq, busiest);
2542 * Unfair double_lock_balance: Optimizes throughput at the expense of
2543 * latency by eliminating extra atomic operations when the locks are
2544 * already in proper order on entry. This favors lower CPU-ids and will
2545 * grant the double lock to lower CPUs over higher ids under contention,
2546 * regardless of entry order into the function.
2548 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2549 __releases(this_rq->lock)
2550 __acquires(busiest->lock)
2551 __acquires(this_rq->lock)
2553 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2556 if (likely(raw_spin_rq_trylock(busiest)))
2559 if (rq_order_less(this_rq, busiest)) {
2560 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2564 raw_spin_rq_unlock(this_rq);
2565 double_rq_lock(this_rq, busiest);
2570 #endif /* CONFIG_PREEMPTION */
2573 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2575 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2577 lockdep_assert_irqs_disabled();
2579 return _double_lock_balance(this_rq, busiest);
2582 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2583 __releases(busiest->lock)
2585 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2586 raw_spin_rq_unlock(busiest);
2587 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2590 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2596 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2599 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2605 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2608 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2614 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2618 * double_rq_unlock - safely unlock two runqueues
2620 * Note this does not restore interrupts like task_rq_unlock,
2621 * you need to do so manually after calling.
2623 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2624 __releases(rq1->lock)
2625 __releases(rq2->lock)
2627 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2628 raw_spin_rq_unlock(rq2);
2630 __release(rq2->lock);
2631 raw_spin_rq_unlock(rq1);
2634 extern void set_rq_online (struct rq *rq);
2635 extern void set_rq_offline(struct rq *rq);
2636 extern bool sched_smp_initialized;
2638 #else /* CONFIG_SMP */
2641 * double_rq_lock - safely lock two runqueues
2643 * Note this does not disable interrupts like task_rq_lock,
2644 * you need to do so manually before calling.
2646 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2647 __acquires(rq1->lock)
2648 __acquires(rq2->lock)
2650 BUG_ON(!irqs_disabled());
2652 raw_spin_rq_lock(rq1);
2653 __acquire(rq2->lock); /* Fake it out ;) */
2657 * double_rq_unlock - safely unlock two runqueues
2659 * Note this does not restore interrupts like task_rq_unlock,
2660 * you need to do so manually after calling.
2662 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2663 __releases(rq1->lock)
2664 __releases(rq2->lock)
2667 raw_spin_rq_unlock(rq1);
2668 __release(rq2->lock);
2673 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2674 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2676 #ifdef CONFIG_SCHED_DEBUG
2677 extern bool sched_debug_verbose;
2679 extern void print_cfs_stats(struct seq_file *m, int cpu);
2680 extern void print_rt_stats(struct seq_file *m, int cpu);
2681 extern void print_dl_stats(struct seq_file *m, int cpu);
2682 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2683 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2684 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2686 extern void resched_latency_warn(int cpu, u64 latency);
2687 #ifdef CONFIG_NUMA_BALANCING
2689 show_numa_stats(struct task_struct *p, struct seq_file *m);
2691 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2692 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2693 #endif /* CONFIG_NUMA_BALANCING */
2695 static inline void resched_latency_warn(int cpu, u64 latency) {}
2696 #endif /* CONFIG_SCHED_DEBUG */
2698 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2699 extern void init_rt_rq(struct rt_rq *rt_rq);
2700 extern void init_dl_rq(struct dl_rq *dl_rq);
2702 extern void cfs_bandwidth_usage_inc(void);
2703 extern void cfs_bandwidth_usage_dec(void);
2705 #ifdef CONFIG_NO_HZ_COMMON
2706 #define NOHZ_BALANCE_KICK_BIT 0
2707 #define NOHZ_STATS_KICK_BIT 1
2708 #define NOHZ_NEWILB_KICK_BIT 2
2710 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2711 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2712 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2714 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2716 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2718 extern void nohz_balance_exit_idle(struct rq *rq);
2720 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2723 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2724 extern void nohz_run_idle_balance(int cpu);
2726 static inline void nohz_run_idle_balance(int cpu) { }
2731 void __dl_update(struct dl_bw *dl_b, s64 bw)
2733 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2736 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2737 "sched RCU must be held");
2738 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2739 struct rq *rq = cpu_rq(i);
2741 rq->dl.extra_bw += bw;
2746 void __dl_update(struct dl_bw *dl_b, s64 bw)
2748 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2755 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2760 struct u64_stats_sync sync;
2763 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2766 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2767 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2768 * and never move forward.
2770 static inline u64 irq_time_read(int cpu)
2772 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2777 seq = __u64_stats_fetch_begin(&irqtime->sync);
2778 total = irqtime->total;
2779 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2783 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2785 #ifdef CONFIG_CPU_FREQ
2786 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2789 * cpufreq_update_util - Take a note about CPU utilization changes.
2790 * @rq: Runqueue to carry out the update for.
2791 * @flags: Update reason flags.
2793 * This function is called by the scheduler on the CPU whose utilization is
2796 * It can only be called from RCU-sched read-side critical sections.
2798 * The way cpufreq is currently arranged requires it to evaluate the CPU
2799 * performance state (frequency/voltage) on a regular basis to prevent it from
2800 * being stuck in a completely inadequate performance level for too long.
2801 * That is not guaranteed to happen if the updates are only triggered from CFS
2802 * and DL, though, because they may not be coming in if only RT tasks are
2803 * active all the time (or there are RT tasks only).
2805 * As a workaround for that issue, this function is called periodically by the
2806 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2807 * but that really is a band-aid. Going forward it should be replaced with
2808 * solutions targeted more specifically at RT tasks.
2810 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2812 struct update_util_data *data;
2814 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2817 data->func(data, rq_clock(rq), flags);
2820 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2821 #endif /* CONFIG_CPU_FREQ */
2823 #ifdef CONFIG_UCLAMP_TASK
2824 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2827 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2828 * @rq: The rq to clamp against. Must not be NULL.
2829 * @util: The util value to clamp.
2830 * @p: The task to clamp against. Can be NULL if you want to clamp
2833 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2835 * If sched_uclamp_used static key is disabled, then just return the util
2836 * without any clamping since uclamp aggregation at the rq level in the fast
2837 * path is disabled, rendering this operation a NOP.
2839 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2840 * will return the correct effective uclamp value of the task even if the
2841 * static key is disabled.
2843 static __always_inline
2844 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2845 struct task_struct *p)
2847 unsigned long min_util = 0;
2848 unsigned long max_util = 0;
2850 if (!static_branch_likely(&sched_uclamp_used))
2854 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2855 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2858 * Ignore last runnable task's max clamp, as this task will
2859 * reset it. Similarly, no need to read the rq's min clamp.
2861 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2865 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2866 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2869 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2870 * RUNNABLE tasks with _different_ clamps, we can end up with an
2871 * inversion. Fix it now when the clamps are applied.
2873 if (unlikely(min_util >= max_util))
2876 return clamp(util, min_util, max_util);
2880 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2881 * by default in the fast path and only gets turned on once userspace performs
2882 * an operation that requires it.
2884 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2887 static inline bool uclamp_is_used(void)
2889 return static_branch_likely(&sched_uclamp_used);
2891 #else /* CONFIG_UCLAMP_TASK */
2893 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2894 struct task_struct *p)
2899 static inline bool uclamp_is_used(void)
2903 #endif /* CONFIG_UCLAMP_TASK */
2905 #ifdef arch_scale_freq_capacity
2906 # ifndef arch_scale_freq_invariant
2907 # define arch_scale_freq_invariant() true
2910 # define arch_scale_freq_invariant() false
2914 static inline unsigned long capacity_orig_of(int cpu)
2916 return cpu_rq(cpu)->cpu_capacity_orig;
2920 * enum cpu_util_type - CPU utilization type
2921 * @FREQUENCY_UTIL: Utilization used to select frequency
2922 * @ENERGY_UTIL: Utilization used during energy calculation
2924 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2925 * need to be aggregated differently depending on the usage made of them. This
2926 * enum is used within effective_cpu_util() to differentiate the types of
2927 * utilization expected by the callers, and adjust the aggregation accordingly.
2929 enum cpu_util_type {
2934 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2935 unsigned long max, enum cpu_util_type type,
2936 struct task_struct *p);
2938 static inline unsigned long cpu_bw_dl(struct rq *rq)
2940 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2943 static inline unsigned long cpu_util_dl(struct rq *rq)
2945 return READ_ONCE(rq->avg_dl.util_avg);
2948 static inline unsigned long cpu_util_cfs(struct rq *rq)
2950 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2952 if (sched_feat(UTIL_EST)) {
2953 util = max_t(unsigned long, util,
2954 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2960 static inline unsigned long cpu_util_rt(struct rq *rq)
2962 return READ_ONCE(rq->avg_rt.util_avg);
2966 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2967 static inline unsigned long cpu_util_irq(struct rq *rq)
2969 return rq->avg_irq.util_avg;
2973 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2975 util *= (max - irq);
2982 static inline unsigned long cpu_util_irq(struct rq *rq)
2988 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2994 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2996 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2998 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3000 static inline bool sched_energy_enabled(void)
3002 return static_branch_unlikely(&sched_energy_present);
3005 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3007 #define perf_domain_span(pd) NULL
3008 static inline bool sched_energy_enabled(void) { return false; }
3010 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3012 #ifdef CONFIG_MEMBARRIER
3014 * The scheduler provides memory barriers required by membarrier between:
3015 * - prior user-space memory accesses and store to rq->membarrier_state,
3016 * - store to rq->membarrier_state and following user-space memory accesses.
3017 * In the same way it provides those guarantees around store to rq->curr.
3019 static inline void membarrier_switch_mm(struct rq *rq,
3020 struct mm_struct *prev_mm,
3021 struct mm_struct *next_mm)
3023 int membarrier_state;
3025 if (prev_mm == next_mm)
3028 membarrier_state = atomic_read(&next_mm->membarrier_state);
3029 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3032 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3035 static inline void membarrier_switch_mm(struct rq *rq,
3036 struct mm_struct *prev_mm,
3037 struct mm_struct *next_mm)
3043 static inline bool is_per_cpu_kthread(struct task_struct *p)
3045 if (!(p->flags & PF_KTHREAD))
3048 if (p->nr_cpus_allowed != 1)
3055 extern void swake_up_all_locked(struct swait_queue_head *q);
3056 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3058 #ifdef CONFIG_PREEMPT_DYNAMIC
3059 extern int preempt_dynamic_mode;
3060 extern int sched_dynamic_mode(const char *str);
3061 extern void sched_dynamic_update(int mode);