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/stat.h>
27 #include <linux/sched/sysctl.h>
28 #include <linux/sched/task.h>
29 #include <linux/sched/task_stack.h>
30 #include <linux/sched/topology.h>
31 #include <linux/sched/user.h>
32 #include <linux/sched/wake_q.h>
33 #include <linux/sched/xacct.h>
35 #include <uapi/linux/sched/types.h>
37 #include <linux/binfmts.h>
38 #include <linux/blkdev.h>
39 #include <linux/compat.h>
40 #include <linux/context_tracking.h>
41 #include <linux/cpufreq.h>
42 #include <linux/cpuidle.h>
43 #include <linux/cpuset.h>
44 #include <linux/ctype.h>
45 #include <linux/debugfs.h>
46 #include <linux/delayacct.h>
47 #include <linux/init_task.h>
48 #include <linux/kprobes.h>
49 #include <linux/kthread.h>
50 #include <linux/membarrier.h>
51 #include <linux/migrate.h>
52 #include <linux/mmu_context.h>
53 #include <linux/nmi.h>
54 #include <linux/proc_fs.h>
55 #include <linux/prefetch.h>
56 #include <linux/profile.h>
57 #include <linux/psi.h>
58 #include <linux/rcupdate_wait.h>
59 #include <linux/security.h>
60 #include <linux/stop_machine.h>
61 #include <linux/suspend.h>
62 #include <linux/swait.h>
63 #include <linux/syscalls.h>
64 #include <linux/task_work.h>
65 #include <linux/tsacct_kern.h>
69 #ifdef CONFIG_PARAVIRT
70 # include <asm/paravirt.h>
74 #include "cpudeadline.h"
76 #ifdef CONFIG_SCHED_DEBUG
77 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
79 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
85 /* task_struct::on_rq states: */
86 #define TASK_ON_RQ_QUEUED 1
87 #define TASK_ON_RQ_MIGRATING 2
89 extern __read_mostly int scheduler_running;
91 extern unsigned long calc_load_update;
92 extern atomic_long_t calc_load_tasks;
94 extern void calc_global_load_tick(struct rq *this_rq);
95 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
98 extern void cpu_load_update_active(struct rq *this_rq);
100 static inline void cpu_load_update_active(struct rq *this_rq) { }
104 * Helpers for converting nanosecond timing to jiffy resolution
106 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
109 * Increase resolution of nice-level calculations for 64-bit architectures.
110 * The extra resolution improves shares distribution and load balancing of
111 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
112 * hierarchies, especially on larger systems. This is not a user-visible change
113 * and does not change the user-interface for setting shares/weights.
115 * We increase resolution only if we have enough bits to allow this increased
116 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
117 * are pretty high and the returns do not justify the increased costs.
119 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
120 * increase coverage and consistency always enable it on 64-bit platforms.
123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
127 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
128 # define scale_load(w) (w)
129 # define scale_load_down(w) (w)
133 * Task weight (visible to users) and its load (invisible to users) have
134 * independent resolution, but they should be well calibrated. We use
135 * scale_load() and scale_load_down(w) to convert between them. The
136 * following must be true:
138 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
141 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
144 * Single value that decides SCHED_DEADLINE internal math precision.
145 * 10 -> just above 1us
146 * 9 -> just above 0.5us
151 * Single value that denotes runtime == period, ie unlimited time.
153 #define RUNTIME_INF ((u64)~0ULL)
155 static inline int idle_policy(int policy)
157 return policy == SCHED_IDLE;
159 static inline int fair_policy(int policy)
161 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
164 static inline int rt_policy(int policy)
166 return policy == SCHED_FIFO || policy == SCHED_RR;
169 static inline int dl_policy(int policy)
171 return policy == SCHED_DEADLINE;
173 static inline bool valid_policy(int policy)
175 return idle_policy(policy) || fair_policy(policy) ||
176 rt_policy(policy) || dl_policy(policy);
179 static inline int task_has_rt_policy(struct task_struct *p)
181 return rt_policy(p->policy);
184 static inline int task_has_dl_policy(struct task_struct *p)
186 return dl_policy(p->policy);
189 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
192 * !! For sched_setattr_nocheck() (kernel) only !!
194 * This is actually gross. :(
196 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
197 * tasks, but still be able to sleep. We need this on platforms that cannot
198 * atomically change clock frequency. Remove once fast switching will be
199 * available on such platforms.
201 * SUGOV stands for SchedUtil GOVernor.
203 #define SCHED_FLAG_SUGOV 0x10000000
205 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
207 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
208 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
215 * Tells if entity @a should preempt entity @b.
218 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
220 return dl_entity_is_special(a) ||
221 dl_time_before(a->deadline, b->deadline);
225 * This is the priority-queue data structure of the RT scheduling class:
227 struct rt_prio_array {
228 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
229 struct list_head queue[MAX_RT_PRIO];
232 struct rt_bandwidth {
233 /* nests inside the rq lock: */
234 raw_spinlock_t rt_runtime_lock;
237 struct hrtimer rt_period_timer;
238 unsigned int rt_period_active;
241 void __dl_clear_params(struct task_struct *p);
244 * To keep the bandwidth of -deadline tasks and groups under control
245 * we need some place where:
246 * - store the maximum -deadline bandwidth of the system (the group);
247 * - cache the fraction of that bandwidth that is currently allocated.
249 * This is all done in the data structure below. It is similar to the
250 * one used for RT-throttling (rt_bandwidth), with the main difference
251 * that, since here we are only interested in admission control, we
252 * do not decrease any runtime while the group "executes", neither we
253 * need a timer to replenish it.
255 * With respect to SMP, the bandwidth is given on a per-CPU basis,
257 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
258 * - dl_total_bw array contains, in the i-eth element, the currently
259 * allocated bandwidth on the i-eth CPU.
260 * Moreover, groups consume bandwidth on each CPU, while tasks only
261 * consume bandwidth on the CPU they're running on.
262 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
263 * that will be shown the next time the proc or cgroup controls will
264 * be red. It on its turn can be changed by writing on its own
267 struct dl_bandwidth {
268 raw_spinlock_t dl_runtime_lock;
273 static inline int dl_bandwidth_enabled(void)
275 return sysctl_sched_rt_runtime >= 0;
284 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
287 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
289 dl_b->total_bw -= tsk_bw;
290 __dl_update(dl_b, (s32)tsk_bw / cpus);
294 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
296 dl_b->total_bw += tsk_bw;
297 __dl_update(dl_b, -((s32)tsk_bw / cpus));
301 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
303 return dl_b->bw != -1 &&
304 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
307 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
308 extern void init_dl_bw(struct dl_bw *dl_b);
309 extern int sched_dl_global_validate(void);
310 extern void sched_dl_do_global(void);
311 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
312 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
313 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
314 extern bool __checkparam_dl(const struct sched_attr *attr);
315 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
316 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
317 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
318 extern bool dl_cpu_busy(unsigned int cpu);
320 #ifdef CONFIG_CGROUP_SCHED
322 #include <linux/cgroup.h>
323 #include <linux/psi.h>
328 extern struct list_head task_groups;
330 struct cfs_bandwidth {
331 #ifdef CONFIG_CFS_BANDWIDTH
336 s64 hierarchical_quota;
342 struct hrtimer period_timer;
343 struct hrtimer slack_timer;
344 struct list_head throttled_cfs_rq;
351 bool distribute_running;
355 /* Task group related information */
357 struct cgroup_subsys_state css;
359 #ifdef CONFIG_FAIR_GROUP_SCHED
360 /* schedulable entities of this group on each CPU */
361 struct sched_entity **se;
362 /* runqueue "owned" by this group on each CPU */
363 struct cfs_rq **cfs_rq;
364 unsigned long shares;
368 * load_avg can be heavily contended at clock tick time, so put
369 * it in its own cacheline separated from the fields above which
370 * will also be accessed at each tick.
372 atomic_long_t load_avg ____cacheline_aligned;
376 #ifdef CONFIG_RT_GROUP_SCHED
377 struct sched_rt_entity **rt_se;
378 struct rt_rq **rt_rq;
380 struct rt_bandwidth rt_bandwidth;
384 struct list_head list;
386 struct task_group *parent;
387 struct list_head siblings;
388 struct list_head children;
390 #ifdef CONFIG_SCHED_AUTOGROUP
391 struct autogroup *autogroup;
394 struct cfs_bandwidth cfs_bandwidth;
397 #ifdef CONFIG_FAIR_GROUP_SCHED
398 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
401 * A weight of 0 or 1 can cause arithmetics problems.
402 * A weight of a cfs_rq is the sum of weights of which entities
403 * are queued on this cfs_rq, so a weight of a entity should not be
404 * too large, so as the shares value of a task group.
405 * (The default weight is 1024 - so there's no practical
406 * limitation from this.)
408 #define MIN_SHARES (1UL << 1)
409 #define MAX_SHARES (1UL << 18)
412 typedef int (*tg_visitor)(struct task_group *, void *);
414 extern int walk_tg_tree_from(struct task_group *from,
415 tg_visitor down, tg_visitor up, void *data);
418 * Iterate the full tree, calling @down when first entering a node and @up when
419 * leaving it for the final time.
421 * Caller must hold rcu_lock or sufficient equivalent.
423 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
425 return walk_tg_tree_from(&root_task_group, down, up, data);
428 extern int tg_nop(struct task_group *tg, void *data);
430 extern void free_fair_sched_group(struct task_group *tg);
431 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
432 extern void online_fair_sched_group(struct task_group *tg);
433 extern void unregister_fair_sched_group(struct task_group *tg);
434 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
435 struct sched_entity *se, int cpu,
436 struct sched_entity *parent);
437 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
439 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
440 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
441 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
443 extern void free_rt_sched_group(struct task_group *tg);
444 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
445 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
446 struct sched_rt_entity *rt_se, int cpu,
447 struct sched_rt_entity *parent);
448 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
449 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
450 extern long sched_group_rt_runtime(struct task_group *tg);
451 extern long sched_group_rt_period(struct task_group *tg);
452 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
454 extern struct task_group *sched_create_group(struct task_group *parent);
455 extern void sched_online_group(struct task_group *tg,
456 struct task_group *parent);
457 extern void sched_destroy_group(struct task_group *tg);
458 extern void sched_offline_group(struct task_group *tg);
460 extern void sched_move_task(struct task_struct *tsk);
462 #ifdef CONFIG_FAIR_GROUP_SCHED
463 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
466 extern void set_task_rq_fair(struct sched_entity *se,
467 struct cfs_rq *prev, struct cfs_rq *next);
468 #else /* !CONFIG_SMP */
469 static inline void set_task_rq_fair(struct sched_entity *se,
470 struct cfs_rq *prev, struct cfs_rq *next) { }
471 #endif /* CONFIG_SMP */
472 #endif /* CONFIG_FAIR_GROUP_SCHED */
474 #else /* CONFIG_CGROUP_SCHED */
476 struct cfs_bandwidth { };
478 #endif /* CONFIG_CGROUP_SCHED */
480 /* CFS-related fields in a runqueue */
482 struct load_weight load;
483 unsigned long runnable_weight;
484 unsigned int nr_running;
485 unsigned int h_nr_running;
490 u64 min_vruntime_copy;
493 struct rb_root_cached tasks_timeline;
496 * 'curr' points to currently running entity on this cfs_rq.
497 * It is set to NULL otherwise (i.e when none are currently running).
499 struct sched_entity *curr;
500 struct sched_entity *next;
501 struct sched_entity *last;
502 struct sched_entity *skip;
504 #ifdef CONFIG_SCHED_DEBUG
505 unsigned int nr_spread_over;
512 struct sched_avg avg;
514 u64 load_last_update_time_copy;
517 raw_spinlock_t lock ____cacheline_aligned;
519 unsigned long load_avg;
520 unsigned long util_avg;
521 unsigned long runnable_sum;
524 #ifdef CONFIG_FAIR_GROUP_SCHED
525 unsigned long tg_load_avg_contrib;
527 long prop_runnable_sum;
530 * h_load = weight * f(tg)
532 * Where f(tg) is the recursive weight fraction assigned to
535 unsigned long h_load;
536 u64 last_h_load_update;
537 struct sched_entity *h_load_next;
538 #endif /* CONFIG_FAIR_GROUP_SCHED */
539 #endif /* CONFIG_SMP */
541 #ifdef CONFIG_FAIR_GROUP_SCHED
542 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
545 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
546 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
547 * (like users, containers etc.)
549 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
550 * This list is used during load balance.
553 struct list_head leaf_cfs_rq_list;
554 struct task_group *tg; /* group that "owns" this runqueue */
556 #ifdef CONFIG_CFS_BANDWIDTH
560 s64 runtime_remaining;
563 u64 throttled_clock_task;
564 u64 throttled_clock_task_time;
567 struct list_head throttled_list;
568 #endif /* CONFIG_CFS_BANDWIDTH */
569 #endif /* CONFIG_FAIR_GROUP_SCHED */
572 static inline int rt_bandwidth_enabled(void)
574 return sysctl_sched_rt_runtime >= 0;
577 /* RT IPI pull logic requires IRQ_WORK */
578 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
579 # define HAVE_RT_PUSH_IPI
582 /* Real-Time classes' related field in a runqueue: */
584 struct rt_prio_array active;
585 unsigned int rt_nr_running;
586 unsigned int rr_nr_running;
587 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
589 int curr; /* highest queued rt task prio */
591 int next; /* next highest */
596 unsigned long rt_nr_migratory;
597 unsigned long rt_nr_total;
599 struct plist_head pushable_tasks;
601 #endif /* CONFIG_SMP */
607 /* Nests inside the rq lock: */
608 raw_spinlock_t rt_runtime_lock;
610 #ifdef CONFIG_RT_GROUP_SCHED
611 unsigned long rt_nr_boosted;
614 struct task_group *tg;
618 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
620 return rt_rq->rt_queued && rt_rq->rt_nr_running;
623 /* Deadline class' related fields in a runqueue */
625 /* runqueue is an rbtree, ordered by deadline */
626 struct rb_root_cached root;
628 unsigned long dl_nr_running;
632 * Deadline values of the currently executing and the
633 * earliest ready task on this rq. Caching these facilitates
634 * the decision wether or not a ready but not running task
635 * should migrate somewhere else.
642 unsigned long dl_nr_migratory;
646 * Tasks on this rq that can be pushed away. They are kept in
647 * an rb-tree, ordered by tasks' deadlines, with caching
648 * of the leftmost (earliest deadline) element.
650 struct rb_root_cached pushable_dl_tasks_root;
655 * "Active utilization" for this runqueue: increased when a
656 * task wakes up (becomes TASK_RUNNING) and decreased when a
662 * Utilization of the tasks "assigned" to this runqueue (including
663 * the tasks that are in runqueue and the tasks that executed on this
664 * CPU and blocked). Increased when a task moves to this runqueue, and
665 * decreased when the task moves away (migrates, changes scheduling
666 * policy, or terminates).
667 * This is needed to compute the "inactive utilization" for the
668 * runqueue (inactive utilization = this_bw - running_bw).
674 * Inverse of the fraction of CPU utilization that can be reclaimed
675 * by the GRUB algorithm.
680 #ifdef CONFIG_FAIR_GROUP_SCHED
681 /* An entity is a task if it doesn't "own" a runqueue */
682 #define entity_is_task(se) (!se->my_q)
684 #define entity_is_task(se) 1
689 * XXX we want to get rid of these helpers and use the full load resolution.
691 static inline long se_weight(struct sched_entity *se)
693 return scale_load_down(se->load.weight);
696 static inline long se_runnable(struct sched_entity *se)
698 return scale_load_down(se->runnable_weight);
701 static inline bool sched_asym_prefer(int a, int b)
703 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
707 * We add the notion of a root-domain which will be used to define per-domain
708 * variables. Each exclusive cpuset essentially defines an island domain by
709 * fully partitioning the member CPUs from any other cpuset. Whenever a new
710 * exclusive cpuset is created, we also create and attach a new root-domain
719 cpumask_var_t online;
722 * Indicate pullable load on at least one CPU, e.g:
723 * - More than one runnable task
724 * - Running task is misfit
729 * The bit corresponding to a CPU gets set here if such CPU has more
730 * than one runnable -deadline task (as it is below for RT tasks).
732 cpumask_var_t dlo_mask;
737 #ifdef HAVE_RT_PUSH_IPI
739 * For IPI pull requests, loop across the rto_mask.
741 struct irq_work rto_push_work;
742 raw_spinlock_t rto_lock;
743 /* These are only updated and read within rto_lock */
746 /* These atomics are updated outside of a lock */
747 atomic_t rto_loop_next;
748 atomic_t rto_loop_start;
751 * The "RT overload" flag: it gets set if a CPU has more than
752 * one runnable RT task.
754 cpumask_var_t rto_mask;
755 struct cpupri cpupri;
757 unsigned long max_cpu_capacity;
760 extern struct root_domain def_root_domain;
761 extern struct mutex sched_domains_mutex;
763 extern void init_defrootdomain(void);
764 extern int sched_init_domains(const struct cpumask *cpu_map);
765 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
766 extern void sched_get_rd(struct root_domain *rd);
767 extern void sched_put_rd(struct root_domain *rd);
769 #ifdef HAVE_RT_PUSH_IPI
770 extern void rto_push_irq_work_func(struct irq_work *work);
772 #endif /* CONFIG_SMP */
775 * This is the main, per-CPU runqueue data structure.
777 * Locking rule: those places that want to lock multiple runqueues
778 * (such as the load balancing or the thread migration code), lock
779 * acquire operations must be ordered by ascending &runqueue.
786 * nr_running and cpu_load should be in the same cacheline because
787 * remote CPUs use both these fields when doing load calculation.
789 unsigned int nr_running;
790 #ifdef CONFIG_NUMA_BALANCING
791 unsigned int nr_numa_running;
792 unsigned int nr_preferred_running;
793 unsigned int numa_migrate_on;
795 #define CPU_LOAD_IDX_MAX 5
796 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
797 #ifdef CONFIG_NO_HZ_COMMON
799 unsigned long last_load_update_tick;
800 unsigned long last_blocked_load_update_tick;
801 unsigned int has_blocked_load;
802 #endif /* CONFIG_SMP */
803 unsigned int nohz_tick_stopped;
805 #endif /* CONFIG_NO_HZ_COMMON */
807 /* capture load from *all* tasks on this CPU: */
808 struct load_weight load;
809 unsigned long nr_load_updates;
816 #ifdef CONFIG_FAIR_GROUP_SCHED
817 /* list of leaf cfs_rq on this CPU: */
818 struct list_head leaf_cfs_rq_list;
819 struct list_head *tmp_alone_branch;
820 #endif /* CONFIG_FAIR_GROUP_SCHED */
823 * This is part of a global counter where only the total sum
824 * over all CPUs matters. A task can increase this counter on
825 * one CPU and if it got migrated afterwards it may decrease
826 * it on another CPU. Always updated under the runqueue lock:
828 unsigned long nr_uninterruptible;
830 struct task_struct *curr;
831 struct task_struct *idle;
832 struct task_struct *stop;
833 unsigned long next_balance;
834 struct mm_struct *prev_mm;
836 unsigned int clock_update_flags;
843 struct root_domain *rd;
844 struct sched_domain *sd;
846 unsigned long cpu_capacity;
847 unsigned long cpu_capacity_orig;
849 struct callback_head *balance_callback;
851 unsigned char idle_balance;
853 unsigned long misfit_task_load;
855 /* For active balancing */
858 struct cpu_stop_work active_balance_work;
860 /* CPU of this runqueue: */
864 struct list_head cfs_tasks;
866 struct sched_avg avg_rt;
867 struct sched_avg avg_dl;
868 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
869 struct sched_avg avg_irq;
874 /* This is used to determine avg_idle's max value */
875 u64 max_idle_balance_cost;
878 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
881 #ifdef CONFIG_PARAVIRT
884 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
885 u64 prev_steal_time_rq;
888 /* calc_load related fields */
889 unsigned long calc_load_update;
890 long calc_load_active;
892 #ifdef CONFIG_SCHED_HRTICK
894 int hrtick_csd_pending;
895 call_single_data_t hrtick_csd;
897 struct hrtimer hrtick_timer;
900 #ifdef CONFIG_SCHEDSTATS
902 struct sched_info rq_sched_info;
903 unsigned long long rq_cpu_time;
904 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
906 /* sys_sched_yield() stats */
907 unsigned int yld_count;
909 /* schedule() stats */
910 unsigned int sched_count;
911 unsigned int sched_goidle;
913 /* try_to_wake_up() stats */
914 unsigned int ttwu_count;
915 unsigned int ttwu_local;
919 struct llist_head wake_list;
922 #ifdef CONFIG_CPU_IDLE
923 /* Must be inspected within a rcu lock section */
924 struct cpuidle_state *idle_state;
928 static inline int cpu_of(struct rq *rq)
938 #ifdef CONFIG_SCHED_SMT
940 extern struct static_key_false sched_smt_present;
942 extern void __update_idle_core(struct rq *rq);
944 static inline void update_idle_core(struct rq *rq)
946 if (static_branch_unlikely(&sched_smt_present))
947 __update_idle_core(rq);
951 static inline void update_idle_core(struct rq *rq) { }
954 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
956 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
957 #define this_rq() this_cpu_ptr(&runqueues)
958 #define task_rq(p) cpu_rq(task_cpu(p))
959 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
960 #define raw_rq() raw_cpu_ptr(&runqueues)
962 extern void update_rq_clock(struct rq *rq);
964 static inline u64 __rq_clock_broken(struct rq *rq)
966 return READ_ONCE(rq->clock);
970 * rq::clock_update_flags bits
972 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
973 * call to __schedule(). This is an optimisation to avoid
974 * neighbouring rq clock updates.
976 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
977 * in effect and calls to update_rq_clock() are being ignored.
979 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
980 * made to update_rq_clock() since the last time rq::lock was pinned.
982 * If inside of __schedule(), clock_update_flags will have been
983 * shifted left (a left shift is a cheap operation for the fast path
984 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
986 * if (rq-clock_update_flags >= RQCF_UPDATED)
988 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
989 * one position though, because the next rq_unpin_lock() will shift it
992 #define RQCF_REQ_SKIP 0x01
993 #define RQCF_ACT_SKIP 0x02
994 #define RQCF_UPDATED 0x04
996 static inline void assert_clock_updated(struct rq *rq)
999 * The only reason for not seeing a clock update since the
1000 * last rq_pin_lock() is if we're currently skipping updates.
1002 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1005 static inline u64 rq_clock(struct rq *rq)
1007 lockdep_assert_held(&rq->lock);
1008 assert_clock_updated(rq);
1013 static inline u64 rq_clock_task(struct rq *rq)
1015 lockdep_assert_held(&rq->lock);
1016 assert_clock_updated(rq);
1018 return rq->clock_task;
1021 static inline void rq_clock_skip_update(struct rq *rq)
1023 lockdep_assert_held(&rq->lock);
1024 rq->clock_update_flags |= RQCF_REQ_SKIP;
1028 * See rt task throttling, which is the only time a skip
1029 * request is cancelled.
1031 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1033 lockdep_assert_held(&rq->lock);
1034 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1038 unsigned long flags;
1039 struct pin_cookie cookie;
1040 #ifdef CONFIG_SCHED_DEBUG
1042 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1043 * current pin context is stashed here in case it needs to be
1044 * restored in rq_repin_lock().
1046 unsigned int clock_update_flags;
1050 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1052 rf->cookie = lockdep_pin_lock(&rq->lock);
1054 #ifdef CONFIG_SCHED_DEBUG
1055 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1056 rf->clock_update_flags = 0;
1060 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1062 #ifdef CONFIG_SCHED_DEBUG
1063 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1064 rf->clock_update_flags = RQCF_UPDATED;
1067 lockdep_unpin_lock(&rq->lock, rf->cookie);
1070 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1072 lockdep_repin_lock(&rq->lock, rf->cookie);
1074 #ifdef CONFIG_SCHED_DEBUG
1076 * Restore the value we stashed in @rf for this pin context.
1078 rq->clock_update_flags |= rf->clock_update_flags;
1082 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1083 __acquires(rq->lock);
1085 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1086 __acquires(p->pi_lock)
1087 __acquires(rq->lock);
1089 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1090 __releases(rq->lock)
1092 rq_unpin_lock(rq, rf);
1093 raw_spin_unlock(&rq->lock);
1097 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1098 __releases(rq->lock)
1099 __releases(p->pi_lock)
1101 rq_unpin_lock(rq, rf);
1102 raw_spin_unlock(&rq->lock);
1103 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1107 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1108 __acquires(rq->lock)
1110 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1111 rq_pin_lock(rq, rf);
1115 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1116 __acquires(rq->lock)
1118 raw_spin_lock_irq(&rq->lock);
1119 rq_pin_lock(rq, rf);
1123 rq_lock(struct rq *rq, struct rq_flags *rf)
1124 __acquires(rq->lock)
1126 raw_spin_lock(&rq->lock);
1127 rq_pin_lock(rq, rf);
1131 rq_relock(struct rq *rq, struct rq_flags *rf)
1132 __acquires(rq->lock)
1134 raw_spin_lock(&rq->lock);
1135 rq_repin_lock(rq, rf);
1139 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1140 __releases(rq->lock)
1142 rq_unpin_lock(rq, rf);
1143 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1147 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1148 __releases(rq->lock)
1150 rq_unpin_lock(rq, rf);
1151 raw_spin_unlock_irq(&rq->lock);
1155 rq_unlock(struct rq *rq, struct rq_flags *rf)
1156 __releases(rq->lock)
1158 rq_unpin_lock(rq, rf);
1159 raw_spin_unlock(&rq->lock);
1162 static inline struct rq *
1163 this_rq_lock_irq(struct rq_flags *rf)
1164 __acquires(rq->lock)
1168 local_irq_disable();
1175 enum numa_topology_type {
1180 extern enum numa_topology_type sched_numa_topology_type;
1181 extern int sched_max_numa_distance;
1182 extern bool find_numa_distance(int distance);
1186 extern void sched_init_numa(void);
1187 extern void sched_domains_numa_masks_set(unsigned int cpu);
1188 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1190 static inline void sched_init_numa(void) { }
1191 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1192 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1195 #ifdef CONFIG_NUMA_BALANCING
1196 /* The regions in numa_faults array from task_struct */
1197 enum numa_faults_stats {
1203 extern void sched_setnuma(struct task_struct *p, int node);
1204 extern int migrate_task_to(struct task_struct *p, int cpu);
1205 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1207 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1210 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1213 #endif /* CONFIG_NUMA_BALANCING */
1218 queue_balance_callback(struct rq *rq,
1219 struct callback_head *head,
1220 void (*func)(struct rq *rq))
1222 lockdep_assert_held(&rq->lock);
1224 if (unlikely(head->next))
1227 head->func = (void (*)(struct callback_head *))func;
1228 head->next = rq->balance_callback;
1229 rq->balance_callback = head;
1232 extern void sched_ttwu_pending(void);
1234 #define rcu_dereference_check_sched_domain(p) \
1235 rcu_dereference_check((p), \
1236 lockdep_is_held(&sched_domains_mutex))
1239 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1240 * See detach_destroy_domains: synchronize_sched for details.
1242 * The domain tree of any CPU may only be accessed from within
1243 * preempt-disabled sections.
1245 #define for_each_domain(cpu, __sd) \
1246 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1247 __sd; __sd = __sd->parent)
1249 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1252 * highest_flag_domain - Return highest sched_domain containing flag.
1253 * @cpu: The CPU whose highest level of sched domain is to
1255 * @flag: The flag to check for the highest sched_domain
1256 * for the given CPU.
1258 * Returns the highest sched_domain of a CPU which contains the given flag.
1260 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1262 struct sched_domain *sd, *hsd = NULL;
1264 for_each_domain(cpu, sd) {
1265 if (!(sd->flags & flag))
1273 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1275 struct sched_domain *sd;
1277 for_each_domain(cpu, sd) {
1278 if (sd->flags & flag)
1285 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1286 DECLARE_PER_CPU(int, sd_llc_size);
1287 DECLARE_PER_CPU(int, sd_llc_id);
1288 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1289 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1290 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1291 extern struct static_key_false sched_asym_cpucapacity;
1293 struct sched_group_capacity {
1296 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1299 unsigned long capacity;
1300 unsigned long min_capacity; /* Min per-CPU capacity in group */
1301 unsigned long max_capacity; /* Max per-CPU capacity in group */
1302 unsigned long next_update;
1303 int imbalance; /* XXX unrelated to capacity but shared group state */
1305 #ifdef CONFIG_SCHED_DEBUG
1309 unsigned long cpumask[0]; /* Balance mask */
1312 struct sched_group {
1313 struct sched_group *next; /* Must be a circular list */
1316 unsigned int group_weight;
1317 struct sched_group_capacity *sgc;
1318 int asym_prefer_cpu; /* CPU of highest priority in group */
1321 * The CPUs this group covers.
1323 * NOTE: this field is variable length. (Allocated dynamically
1324 * by attaching extra space to the end of the structure,
1325 * depending on how many CPUs the kernel has booted up with)
1327 unsigned long cpumask[0];
1330 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1332 return to_cpumask(sg->cpumask);
1336 * See build_balance_mask().
1338 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1340 return to_cpumask(sg->sgc->cpumask);
1344 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1345 * @group: The group whose first CPU is to be returned.
1347 static inline unsigned int group_first_cpu(struct sched_group *group)
1349 return cpumask_first(sched_group_span(group));
1352 extern int group_balance_cpu(struct sched_group *sg);
1354 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1355 void register_sched_domain_sysctl(void);
1356 void dirty_sched_domain_sysctl(int cpu);
1357 void unregister_sched_domain_sysctl(void);
1359 static inline void register_sched_domain_sysctl(void)
1362 static inline void dirty_sched_domain_sysctl(int cpu)
1365 static inline void unregister_sched_domain_sysctl(void)
1372 static inline void sched_ttwu_pending(void) { }
1374 #endif /* CONFIG_SMP */
1377 #include "autogroup.h"
1379 #ifdef CONFIG_CGROUP_SCHED
1382 * Return the group to which this tasks belongs.
1384 * We cannot use task_css() and friends because the cgroup subsystem
1385 * changes that value before the cgroup_subsys::attach() method is called,
1386 * therefore we cannot pin it and might observe the wrong value.
1388 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1389 * core changes this before calling sched_move_task().
1391 * Instead we use a 'copy' which is updated from sched_move_task() while
1392 * holding both task_struct::pi_lock and rq::lock.
1394 static inline struct task_group *task_group(struct task_struct *p)
1396 return p->sched_task_group;
1399 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1400 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1402 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1403 struct task_group *tg = task_group(p);
1406 #ifdef CONFIG_FAIR_GROUP_SCHED
1407 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1408 p->se.cfs_rq = tg->cfs_rq[cpu];
1409 p->se.parent = tg->se[cpu];
1412 #ifdef CONFIG_RT_GROUP_SCHED
1413 p->rt.rt_rq = tg->rt_rq[cpu];
1414 p->rt.parent = tg->rt_se[cpu];
1418 #else /* CONFIG_CGROUP_SCHED */
1420 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1421 static inline struct task_group *task_group(struct task_struct *p)
1426 #endif /* CONFIG_CGROUP_SCHED */
1428 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1430 set_task_rq(p, cpu);
1433 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1434 * successfuly executed on another CPU. We must ensure that updates of
1435 * per-task data have been completed by this moment.
1438 #ifdef CONFIG_THREAD_INFO_IN_TASK
1441 task_thread_info(p)->cpu = cpu;
1448 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1450 #ifdef CONFIG_SCHED_DEBUG
1451 # include <linux/static_key.h>
1452 # define const_debug __read_mostly
1454 # define const_debug const
1457 #define SCHED_FEAT(name, enabled) \
1458 __SCHED_FEAT_##name ,
1461 #include "features.h"
1467 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1470 * To support run-time toggling of sched features, all the translation units
1471 * (but core.c) reference the sysctl_sched_features defined in core.c.
1473 extern const_debug unsigned int sysctl_sched_features;
1475 #define SCHED_FEAT(name, enabled) \
1476 static __always_inline bool static_branch_##name(struct static_key *key) \
1478 return static_key_##enabled(key); \
1481 #include "features.h"
1484 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1485 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1487 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1490 * Each translation unit has its own copy of sysctl_sched_features to allow
1491 * constants propagation at compile time and compiler optimization based on
1494 #define SCHED_FEAT(name, enabled) \
1495 (1UL << __SCHED_FEAT_##name) * enabled |
1496 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1497 #include "features.h"
1501 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1503 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1505 extern struct static_key_false sched_numa_balancing;
1506 extern struct static_key_false sched_schedstats;
1508 static inline u64 global_rt_period(void)
1510 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1513 static inline u64 global_rt_runtime(void)
1515 if (sysctl_sched_rt_runtime < 0)
1518 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1521 static inline int task_current(struct rq *rq, struct task_struct *p)
1523 return rq->curr == p;
1526 static inline int task_running(struct rq *rq, struct task_struct *p)
1531 return task_current(rq, p);
1535 static inline int task_on_rq_queued(struct task_struct *p)
1537 return p->on_rq == TASK_ON_RQ_QUEUED;
1540 static inline int task_on_rq_migrating(struct task_struct *p)
1542 return p->on_rq == TASK_ON_RQ_MIGRATING;
1548 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1549 #define WF_FORK 0x02 /* Child wakeup after fork */
1550 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1553 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1554 * of tasks with abnormal "nice" values across CPUs the contribution that
1555 * each task makes to its run queue's load is weighted according to its
1556 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1557 * scaled version of the new time slice allocation that they receive on time
1561 #define WEIGHT_IDLEPRIO 3
1562 #define WMULT_IDLEPRIO 1431655765
1564 extern const int sched_prio_to_weight[40];
1565 extern const u32 sched_prio_to_wmult[40];
1568 * {de,en}queue flags:
1570 * DEQUEUE_SLEEP - task is no longer runnable
1571 * ENQUEUE_WAKEUP - task just became runnable
1573 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1574 * are in a known state which allows modification. Such pairs
1575 * should preserve as much state as possible.
1577 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1580 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1581 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1582 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1586 #define DEQUEUE_SLEEP 0x01
1587 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1588 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1589 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1591 #define ENQUEUE_WAKEUP 0x01
1592 #define ENQUEUE_RESTORE 0x02
1593 #define ENQUEUE_MOVE 0x04
1594 #define ENQUEUE_NOCLOCK 0x08
1596 #define ENQUEUE_HEAD 0x10
1597 #define ENQUEUE_REPLENISH 0x20
1599 #define ENQUEUE_MIGRATED 0x40
1601 #define ENQUEUE_MIGRATED 0x00
1604 #define RETRY_TASK ((void *)-1UL)
1606 struct sched_class {
1607 const struct sched_class *next;
1609 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1610 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1611 void (*yield_task) (struct rq *rq);
1612 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1614 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1617 * It is the responsibility of the pick_next_task() method that will
1618 * return the next task to call put_prev_task() on the @prev task or
1619 * something equivalent.
1621 * May return RETRY_TASK when it finds a higher prio class has runnable
1624 struct task_struct * (*pick_next_task)(struct rq *rq,
1625 struct task_struct *prev,
1626 struct rq_flags *rf);
1627 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1630 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1631 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1633 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1635 void (*set_cpus_allowed)(struct task_struct *p,
1636 const struct cpumask *newmask);
1638 void (*rq_online)(struct rq *rq);
1639 void (*rq_offline)(struct rq *rq);
1642 void (*set_curr_task)(struct rq *rq);
1643 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1644 void (*task_fork)(struct task_struct *p);
1645 void (*task_dead)(struct task_struct *p);
1648 * The switched_from() call is allowed to drop rq->lock, therefore we
1649 * cannot assume the switched_from/switched_to pair is serliazed by
1650 * rq->lock. They are however serialized by p->pi_lock.
1652 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1653 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1654 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1657 unsigned int (*get_rr_interval)(struct rq *rq,
1658 struct task_struct *task);
1660 void (*update_curr)(struct rq *rq);
1662 #define TASK_SET_GROUP 0
1663 #define TASK_MOVE_GROUP 1
1665 #ifdef CONFIG_FAIR_GROUP_SCHED
1666 void (*task_change_group)(struct task_struct *p, int type);
1670 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1672 prev->sched_class->put_prev_task(rq, prev);
1675 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1677 curr->sched_class->set_curr_task(rq);
1681 #define sched_class_highest (&stop_sched_class)
1683 #define sched_class_highest (&dl_sched_class)
1685 #define for_each_class(class) \
1686 for (class = sched_class_highest; class; class = class->next)
1688 extern const struct sched_class stop_sched_class;
1689 extern const struct sched_class dl_sched_class;
1690 extern const struct sched_class rt_sched_class;
1691 extern const struct sched_class fair_sched_class;
1692 extern const struct sched_class idle_sched_class;
1697 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1699 extern void trigger_load_balance(struct rq *rq);
1701 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1705 #ifdef CONFIG_CPU_IDLE
1706 static inline void idle_set_state(struct rq *rq,
1707 struct cpuidle_state *idle_state)
1709 rq->idle_state = idle_state;
1712 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1714 SCHED_WARN_ON(!rcu_read_lock_held());
1716 return rq->idle_state;
1719 static inline void idle_set_state(struct rq *rq,
1720 struct cpuidle_state *idle_state)
1724 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1730 extern void schedule_idle(void);
1732 extern void sysrq_sched_debug_show(void);
1733 extern void sched_init_granularity(void);
1734 extern void update_max_interval(void);
1736 extern void init_sched_dl_class(void);
1737 extern void init_sched_rt_class(void);
1738 extern void init_sched_fair_class(void);
1740 extern void reweight_task(struct task_struct *p, int prio);
1742 extern void resched_curr(struct rq *rq);
1743 extern void resched_cpu(int cpu);
1745 extern struct rt_bandwidth def_rt_bandwidth;
1746 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1748 extern struct dl_bandwidth def_dl_bandwidth;
1749 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1750 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1751 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1752 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1755 #define BW_UNIT (1 << BW_SHIFT)
1756 #define RATIO_SHIFT 8
1757 unsigned long to_ratio(u64 period, u64 runtime);
1759 extern void init_entity_runnable_average(struct sched_entity *se);
1760 extern void post_init_entity_util_avg(struct sched_entity *se);
1762 #ifdef CONFIG_NO_HZ_FULL
1763 extern bool sched_can_stop_tick(struct rq *rq);
1764 extern int __init sched_tick_offload_init(void);
1767 * Tick may be needed by tasks in the runqueue depending on their policy and
1768 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1769 * nohz mode if necessary.
1771 static inline void sched_update_tick_dependency(struct rq *rq)
1775 if (!tick_nohz_full_enabled())
1780 if (!tick_nohz_full_cpu(cpu))
1783 if (sched_can_stop_tick(rq))
1784 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1786 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1789 static inline int sched_tick_offload_init(void) { return 0; }
1790 static inline void sched_update_tick_dependency(struct rq *rq) { }
1793 static inline void add_nr_running(struct rq *rq, unsigned count)
1795 unsigned prev_nr = rq->nr_running;
1797 rq->nr_running = prev_nr + count;
1799 if (prev_nr < 2 && rq->nr_running >= 2) {
1801 if (!READ_ONCE(rq->rd->overload))
1802 WRITE_ONCE(rq->rd->overload, 1);
1806 sched_update_tick_dependency(rq);
1809 static inline void sub_nr_running(struct rq *rq, unsigned count)
1811 rq->nr_running -= count;
1812 /* Check if we still need preemption */
1813 sched_update_tick_dependency(rq);
1816 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1817 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1819 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1821 extern const_debug unsigned int sysctl_sched_nr_migrate;
1822 extern const_debug unsigned int sysctl_sched_migration_cost;
1824 #ifdef CONFIG_SCHED_HRTICK
1828 * - enabled by features
1829 * - hrtimer is actually high res
1831 static inline int hrtick_enabled(struct rq *rq)
1833 if (!sched_feat(HRTICK))
1835 if (!cpu_active(cpu_of(rq)))
1837 return hrtimer_is_hres_active(&rq->hrtick_timer);
1840 void hrtick_start(struct rq *rq, u64 delay);
1844 static inline int hrtick_enabled(struct rq *rq)
1849 #endif /* CONFIG_SCHED_HRTICK */
1851 #ifndef arch_scale_freq_capacity
1852 static __always_inline
1853 unsigned long arch_scale_freq_capacity(int cpu)
1855 return SCHED_CAPACITY_SCALE;
1860 #ifndef arch_scale_cpu_capacity
1861 static __always_inline
1862 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1864 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1865 return sd->smt_gain / sd->span_weight;
1867 return SCHED_CAPACITY_SCALE;
1871 #ifndef arch_scale_cpu_capacity
1872 static __always_inline
1873 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
1875 return SCHED_CAPACITY_SCALE;
1881 #ifdef CONFIG_PREEMPT
1883 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1886 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1887 * way at the expense of forcing extra atomic operations in all
1888 * invocations. This assures that the double_lock is acquired using the
1889 * same underlying policy as the spinlock_t on this architecture, which
1890 * reduces latency compared to the unfair variant below. However, it
1891 * also adds more overhead and therefore may reduce throughput.
1893 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1894 __releases(this_rq->lock)
1895 __acquires(busiest->lock)
1896 __acquires(this_rq->lock)
1898 raw_spin_unlock(&this_rq->lock);
1899 double_rq_lock(this_rq, busiest);
1906 * Unfair double_lock_balance: Optimizes throughput at the expense of
1907 * latency by eliminating extra atomic operations when the locks are
1908 * already in proper order on entry. This favors lower CPU-ids and will
1909 * grant the double lock to lower CPUs over higher ids under contention,
1910 * regardless of entry order into the function.
1912 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1913 __releases(this_rq->lock)
1914 __acquires(busiest->lock)
1915 __acquires(this_rq->lock)
1919 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1920 if (busiest < this_rq) {
1921 raw_spin_unlock(&this_rq->lock);
1922 raw_spin_lock(&busiest->lock);
1923 raw_spin_lock_nested(&this_rq->lock,
1924 SINGLE_DEPTH_NESTING);
1927 raw_spin_lock_nested(&busiest->lock,
1928 SINGLE_DEPTH_NESTING);
1933 #endif /* CONFIG_PREEMPT */
1936 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1938 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1940 if (unlikely(!irqs_disabled())) {
1941 /* printk() doesn't work well under rq->lock */
1942 raw_spin_unlock(&this_rq->lock);
1946 return _double_lock_balance(this_rq, busiest);
1949 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1950 __releases(busiest->lock)
1952 raw_spin_unlock(&busiest->lock);
1953 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1956 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1962 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1965 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1971 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1974 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1980 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1984 * double_rq_lock - safely lock two runqueues
1986 * Note this does not disable interrupts like task_rq_lock,
1987 * you need to do so manually before calling.
1989 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1990 __acquires(rq1->lock)
1991 __acquires(rq2->lock)
1993 BUG_ON(!irqs_disabled());
1995 raw_spin_lock(&rq1->lock);
1996 __acquire(rq2->lock); /* Fake it out ;) */
1999 raw_spin_lock(&rq1->lock);
2000 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2002 raw_spin_lock(&rq2->lock);
2003 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2009 * double_rq_unlock - safely unlock two runqueues
2011 * Note this does not restore interrupts like task_rq_unlock,
2012 * you need to do so manually after calling.
2014 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2015 __releases(rq1->lock)
2016 __releases(rq2->lock)
2018 raw_spin_unlock(&rq1->lock);
2020 raw_spin_unlock(&rq2->lock);
2022 __release(rq2->lock);
2025 extern void set_rq_online (struct rq *rq);
2026 extern void set_rq_offline(struct rq *rq);
2027 extern bool sched_smp_initialized;
2029 #else /* CONFIG_SMP */
2032 * double_rq_lock - safely lock two runqueues
2034 * Note this does not disable interrupts like task_rq_lock,
2035 * you need to do so manually before calling.
2037 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2038 __acquires(rq1->lock)
2039 __acquires(rq2->lock)
2041 BUG_ON(!irqs_disabled());
2043 raw_spin_lock(&rq1->lock);
2044 __acquire(rq2->lock); /* Fake it out ;) */
2048 * double_rq_unlock - safely unlock two runqueues
2050 * Note this does not restore interrupts like task_rq_unlock,
2051 * you need to do so manually after calling.
2053 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2054 __releases(rq1->lock)
2055 __releases(rq2->lock)
2058 raw_spin_unlock(&rq1->lock);
2059 __release(rq2->lock);
2064 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2065 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2067 #ifdef CONFIG_SCHED_DEBUG
2068 extern bool sched_debug_enabled;
2070 extern void print_cfs_stats(struct seq_file *m, int cpu);
2071 extern void print_rt_stats(struct seq_file *m, int cpu);
2072 extern void print_dl_stats(struct seq_file *m, int cpu);
2073 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2074 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2075 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2076 #ifdef CONFIG_NUMA_BALANCING
2078 show_numa_stats(struct task_struct *p, struct seq_file *m);
2080 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2081 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2082 #endif /* CONFIG_NUMA_BALANCING */
2083 #endif /* CONFIG_SCHED_DEBUG */
2085 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2086 extern void init_rt_rq(struct rt_rq *rt_rq);
2087 extern void init_dl_rq(struct dl_rq *dl_rq);
2089 extern void cfs_bandwidth_usage_inc(void);
2090 extern void cfs_bandwidth_usage_dec(void);
2092 #ifdef CONFIG_NO_HZ_COMMON
2093 #define NOHZ_BALANCE_KICK_BIT 0
2094 #define NOHZ_STATS_KICK_BIT 1
2096 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2097 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2099 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2101 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2103 extern void nohz_balance_exit_idle(struct rq *rq);
2105 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2111 void __dl_update(struct dl_bw *dl_b, s64 bw)
2113 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2116 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2117 "sched RCU must be held");
2118 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2119 struct rq *rq = cpu_rq(i);
2121 rq->dl.extra_bw += bw;
2126 void __dl_update(struct dl_bw *dl_b, s64 bw)
2128 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2135 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2140 struct u64_stats_sync sync;
2143 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2146 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2147 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2148 * and never move forward.
2150 static inline u64 irq_time_read(int cpu)
2152 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2157 seq = __u64_stats_fetch_begin(&irqtime->sync);
2158 total = irqtime->total;
2159 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2163 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2165 #ifdef CONFIG_CPU_FREQ
2166 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2169 * cpufreq_update_util - Take a note about CPU utilization changes.
2170 * @rq: Runqueue to carry out the update for.
2171 * @flags: Update reason flags.
2173 * This function is called by the scheduler on the CPU whose utilization is
2176 * It can only be called from RCU-sched read-side critical sections.
2178 * The way cpufreq is currently arranged requires it to evaluate the CPU
2179 * performance state (frequency/voltage) on a regular basis to prevent it from
2180 * being stuck in a completely inadequate performance level for too long.
2181 * That is not guaranteed to happen if the updates are only triggered from CFS
2182 * and DL, though, because they may not be coming in if only RT tasks are
2183 * active all the time (or there are RT tasks only).
2185 * As a workaround for that issue, this function is called periodically by the
2186 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2187 * but that really is a band-aid. Going forward it should be replaced with
2188 * solutions targeted more specifically at RT tasks.
2190 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2192 struct update_util_data *data;
2194 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2197 data->func(data, rq_clock(rq), flags);
2200 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2201 #endif /* CONFIG_CPU_FREQ */
2203 #ifdef arch_scale_freq_capacity
2204 # ifndef arch_scale_freq_invariant
2205 # define arch_scale_freq_invariant() true
2208 # define arch_scale_freq_invariant() false
2211 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2212 static inline unsigned long cpu_bw_dl(struct rq *rq)
2214 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2217 static inline unsigned long cpu_util_dl(struct rq *rq)
2219 return READ_ONCE(rq->avg_dl.util_avg);
2222 static inline unsigned long cpu_util_cfs(struct rq *rq)
2224 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2226 if (sched_feat(UTIL_EST)) {
2227 util = max_t(unsigned long, util,
2228 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2234 static inline unsigned long cpu_util_rt(struct rq *rq)
2236 return READ_ONCE(rq->avg_rt.util_avg);
2240 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2241 static inline unsigned long cpu_util_irq(struct rq *rq)
2243 return rq->avg_irq.util_avg;
2247 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2249 util *= (max - irq);
2256 static inline unsigned long cpu_util_irq(struct rq *rq)
2262 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)