1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #include <linux/sched.h>
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
36 #include <uapi/linux/sched/types.h>
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
76 #include "cpudeadline.h"
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED 1
89 #define TASK_ON_RQ_MIGRATING 2
91 extern __read_mostly int scheduler_running;
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
100 extern void cpu_load_update_active(struct rq *this_rq);
102 static inline void cpu_load_update_active(struct rq *this_rq) { }
106 * Helpers for converting nanosecond timing to jiffy resolution
108 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
111 * Increase resolution of nice-level calculations for 64-bit architectures.
112 * The extra resolution improves shares distribution and load balancing of
113 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
114 * hierarchies, especially on larger systems. This is not a user-visible change
115 * and does not change the user-interface for setting shares/weights.
117 * We increase resolution only if we have enough bits to allow this increased
118 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
119 * are pretty high and the returns do not justify the increased costs.
121 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
122 * increase coverage and consistency always enable it on 64-bit platforms.
125 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
127 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
130 # define scale_load(w) (w)
131 # define scale_load_down(w) (w)
135 * Task weight (visible to users) and its load (invisible to users) have
136 * independent resolution, but they should be well calibrated. We use
137 * scale_load() and scale_load_down(w) to convert between them. The
138 * following must be true:
140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
146 * Single value that decides SCHED_DEADLINE internal math precision.
147 * 10 -> just above 1us
148 * 9 -> just above 0.5us
153 * Single value that denotes runtime == period, ie unlimited time.
155 #define RUNTIME_INF ((u64)~0ULL)
157 static inline int idle_policy(int policy)
159 return policy == SCHED_IDLE;
161 static inline int fair_policy(int policy)
163 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
166 static inline int rt_policy(int policy)
168 return policy == SCHED_FIFO || policy == SCHED_RR;
171 static inline int dl_policy(int policy)
173 return policy == SCHED_DEADLINE;
175 static inline bool valid_policy(int policy)
177 return idle_policy(policy) || fair_policy(policy) ||
178 rt_policy(policy) || dl_policy(policy);
181 static inline int task_has_idle_policy(struct task_struct *p)
183 return idle_policy(p->policy);
186 static inline int task_has_rt_policy(struct task_struct *p)
188 return rt_policy(p->policy);
191 static inline int task_has_dl_policy(struct task_struct *p)
193 return dl_policy(p->policy);
196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
199 * !! For sched_setattr_nocheck() (kernel) only !!
201 * This is actually gross. :(
203 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
204 * tasks, but still be able to sleep. We need this on platforms that cannot
205 * atomically change clock frequency. Remove once fast switching will be
206 * available on such platforms.
208 * SUGOV stands for SchedUtil GOVernor.
210 #define SCHED_FLAG_SUGOV 0x10000000
212 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
214 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
215 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
222 * Tells if entity @a should preempt entity @b.
225 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
227 return dl_entity_is_special(a) ||
228 dl_time_before(a->deadline, b->deadline);
232 * This is the priority-queue data structure of the RT scheduling class:
234 struct rt_prio_array {
235 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
236 struct list_head queue[MAX_RT_PRIO];
239 struct rt_bandwidth {
240 /* nests inside the rq lock: */
241 raw_spinlock_t rt_runtime_lock;
244 struct hrtimer rt_period_timer;
245 unsigned int rt_period_active;
248 void __dl_clear_params(struct task_struct *p);
251 * To keep the bandwidth of -deadline tasks and groups under control
252 * we need some place where:
253 * - store the maximum -deadline bandwidth of the system (the group);
254 * - cache the fraction of that bandwidth that is currently allocated.
256 * This is all done in the data structure below. It is similar to the
257 * one used for RT-throttling (rt_bandwidth), with the main difference
258 * that, since here we are only interested in admission control, we
259 * do not decrease any runtime while the group "executes", neither we
260 * need a timer to replenish it.
262 * With respect to SMP, the bandwidth is given on a per-CPU basis,
264 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
265 * - dl_total_bw array contains, in the i-eth element, the currently
266 * allocated bandwidth on the i-eth CPU.
267 * Moreover, groups consume bandwidth on each CPU, while tasks only
268 * consume bandwidth on the CPU they're running on.
269 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
270 * that will be shown the next time the proc or cgroup controls will
271 * be red. It on its turn can be changed by writing on its own
274 struct dl_bandwidth {
275 raw_spinlock_t dl_runtime_lock;
280 static inline int dl_bandwidth_enabled(void)
282 return sysctl_sched_rt_runtime >= 0;
291 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
294 void __dl_sub(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 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
303 dl_b->total_bw += tsk_bw;
304 __dl_update(dl_b, -((s32)tsk_bw / cpus));
308 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
310 return dl_b->bw != -1 &&
311 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
314 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
315 extern void init_dl_bw(struct dl_bw *dl_b);
316 extern int sched_dl_global_validate(void);
317 extern void sched_dl_do_global(void);
318 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
319 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
320 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
321 extern bool __checkparam_dl(const struct sched_attr *attr);
322 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
323 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
324 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
325 extern bool dl_cpu_busy(unsigned int cpu);
327 #ifdef CONFIG_CGROUP_SCHED
329 #include <linux/cgroup.h>
330 #include <linux/psi.h>
335 extern struct list_head task_groups;
337 struct cfs_bandwidth {
338 #ifdef CONFIG_CFS_BANDWIDTH
343 s64 hierarchical_quota;
349 struct hrtimer period_timer;
350 struct hrtimer slack_timer;
351 struct list_head throttled_cfs_rq;
358 bool distribute_running;
362 /* Task group related information */
364 struct cgroup_subsys_state css;
366 #ifdef CONFIG_FAIR_GROUP_SCHED
367 /* schedulable entities of this group on each CPU */
368 struct sched_entity **se;
369 /* runqueue "owned" by this group on each CPU */
370 struct cfs_rq **cfs_rq;
371 unsigned long shares;
375 * load_avg can be heavily contended at clock tick time, so put
376 * it in its own cacheline separated from the fields above which
377 * will also be accessed at each tick.
379 atomic_long_t load_avg ____cacheline_aligned;
383 #ifdef CONFIG_RT_GROUP_SCHED
384 struct sched_rt_entity **rt_se;
385 struct rt_rq **rt_rq;
387 struct rt_bandwidth rt_bandwidth;
391 struct list_head list;
393 struct task_group *parent;
394 struct list_head siblings;
395 struct list_head children;
397 #ifdef CONFIG_SCHED_AUTOGROUP
398 struct autogroup *autogroup;
401 struct cfs_bandwidth cfs_bandwidth;
404 #ifdef CONFIG_FAIR_GROUP_SCHED
405 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
408 * A weight of 0 or 1 can cause arithmetics problems.
409 * A weight of a cfs_rq is the sum of weights of which entities
410 * are queued on this cfs_rq, so a weight of a entity should not be
411 * too large, so as the shares value of a task group.
412 * (The default weight is 1024 - so there's no practical
413 * limitation from this.)
415 #define MIN_SHARES (1UL << 1)
416 #define MAX_SHARES (1UL << 18)
419 typedef int (*tg_visitor)(struct task_group *, void *);
421 extern int walk_tg_tree_from(struct task_group *from,
422 tg_visitor down, tg_visitor up, void *data);
425 * Iterate the full tree, calling @down when first entering a node and @up when
426 * leaving it for the final time.
428 * Caller must hold rcu_lock or sufficient equivalent.
430 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
432 return walk_tg_tree_from(&root_task_group, down, up, data);
435 extern int tg_nop(struct task_group *tg, void *data);
437 extern void free_fair_sched_group(struct task_group *tg);
438 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
439 extern void online_fair_sched_group(struct task_group *tg);
440 extern void unregister_fair_sched_group(struct task_group *tg);
441 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
442 struct sched_entity *se, int cpu,
443 struct sched_entity *parent);
444 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
446 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
447 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
448 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
450 extern void free_rt_sched_group(struct task_group *tg);
451 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
452 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
453 struct sched_rt_entity *rt_se, int cpu,
454 struct sched_rt_entity *parent);
455 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
456 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
457 extern long sched_group_rt_runtime(struct task_group *tg);
458 extern long sched_group_rt_period(struct task_group *tg);
459 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
461 extern struct task_group *sched_create_group(struct task_group *parent);
462 extern void sched_online_group(struct task_group *tg,
463 struct task_group *parent);
464 extern void sched_destroy_group(struct task_group *tg);
465 extern void sched_offline_group(struct task_group *tg);
467 extern void sched_move_task(struct task_struct *tsk);
469 #ifdef CONFIG_FAIR_GROUP_SCHED
470 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
473 extern void set_task_rq_fair(struct sched_entity *se,
474 struct cfs_rq *prev, struct cfs_rq *next);
475 #else /* !CONFIG_SMP */
476 static inline void set_task_rq_fair(struct sched_entity *se,
477 struct cfs_rq *prev, struct cfs_rq *next) { }
478 #endif /* CONFIG_SMP */
479 #endif /* CONFIG_FAIR_GROUP_SCHED */
481 #else /* CONFIG_CGROUP_SCHED */
483 struct cfs_bandwidth { };
485 #endif /* CONFIG_CGROUP_SCHED */
487 /* CFS-related fields in a runqueue */
489 struct load_weight load;
490 unsigned long runnable_weight;
491 unsigned int nr_running;
492 unsigned int h_nr_running;
497 u64 min_vruntime_copy;
500 struct rb_root_cached tasks_timeline;
503 * 'curr' points to currently running entity on this cfs_rq.
504 * It is set to NULL otherwise (i.e when none are currently running).
506 struct sched_entity *curr;
507 struct sched_entity *next;
508 struct sched_entity *last;
509 struct sched_entity *skip;
511 #ifdef CONFIG_SCHED_DEBUG
512 unsigned int nr_spread_over;
519 struct sched_avg avg;
521 u64 load_last_update_time_copy;
524 raw_spinlock_t lock ____cacheline_aligned;
526 unsigned long load_avg;
527 unsigned long util_avg;
528 unsigned long runnable_sum;
531 #ifdef CONFIG_FAIR_GROUP_SCHED
532 unsigned long tg_load_avg_contrib;
534 long prop_runnable_sum;
537 * h_load = weight * f(tg)
539 * Where f(tg) is the recursive weight fraction assigned to
542 unsigned long h_load;
543 u64 last_h_load_update;
544 struct sched_entity *h_load_next;
545 #endif /* CONFIG_FAIR_GROUP_SCHED */
546 #endif /* CONFIG_SMP */
548 #ifdef CONFIG_FAIR_GROUP_SCHED
549 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
552 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
553 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
554 * (like users, containers etc.)
556 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
557 * This list is used during load balance.
560 struct list_head leaf_cfs_rq_list;
561 struct task_group *tg; /* group that "owns" this runqueue */
563 #ifdef CONFIG_CFS_BANDWIDTH
567 s64 runtime_remaining;
570 u64 throttled_clock_task;
571 u64 throttled_clock_task_time;
574 struct list_head throttled_list;
575 #endif /* CONFIG_CFS_BANDWIDTH */
576 #endif /* CONFIG_FAIR_GROUP_SCHED */
579 static inline int rt_bandwidth_enabled(void)
581 return sysctl_sched_rt_runtime >= 0;
584 /* RT IPI pull logic requires IRQ_WORK */
585 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
586 # define HAVE_RT_PUSH_IPI
589 /* Real-Time classes' related field in a runqueue: */
591 struct rt_prio_array active;
592 unsigned int rt_nr_running;
593 unsigned int rr_nr_running;
594 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
596 int curr; /* highest queued rt task prio */
598 int next; /* next highest */
603 unsigned long rt_nr_migratory;
604 unsigned long rt_nr_total;
606 struct plist_head pushable_tasks;
608 #endif /* CONFIG_SMP */
614 /* Nests inside the rq lock: */
615 raw_spinlock_t rt_runtime_lock;
617 #ifdef CONFIG_RT_GROUP_SCHED
618 unsigned long rt_nr_boosted;
621 struct task_group *tg;
625 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
627 return rt_rq->rt_queued && rt_rq->rt_nr_running;
630 /* Deadline class' related fields in a runqueue */
632 /* runqueue is an rbtree, ordered by deadline */
633 struct rb_root_cached root;
635 unsigned long dl_nr_running;
639 * Deadline values of the currently executing and the
640 * earliest ready task on this rq. Caching these facilitates
641 * the decision whether or not a ready but not running task
642 * should migrate somewhere else.
649 unsigned long dl_nr_migratory;
653 * Tasks on this rq that can be pushed away. They are kept in
654 * an rb-tree, ordered by tasks' deadlines, with caching
655 * of the leftmost (earliest deadline) element.
657 struct rb_root_cached pushable_dl_tasks_root;
662 * "Active utilization" for this runqueue: increased when a
663 * task wakes up (becomes TASK_RUNNING) and decreased when a
669 * Utilization of the tasks "assigned" to this runqueue (including
670 * the tasks that are in runqueue and the tasks that executed on this
671 * CPU and blocked). Increased when a task moves to this runqueue, and
672 * decreased when the task moves away (migrates, changes scheduling
673 * policy, or terminates).
674 * This is needed to compute the "inactive utilization" for the
675 * runqueue (inactive utilization = this_bw - running_bw).
681 * Inverse of the fraction of CPU utilization that can be reclaimed
682 * by the GRUB algorithm.
687 #ifdef CONFIG_FAIR_GROUP_SCHED
688 /* An entity is a task if it doesn't "own" a runqueue */
689 #define entity_is_task(se) (!se->my_q)
691 #define entity_is_task(se) 1
696 * XXX we want to get rid of these helpers and use the full load resolution.
698 static inline long se_weight(struct sched_entity *se)
700 return scale_load_down(se->load.weight);
703 static inline long se_runnable(struct sched_entity *se)
705 return scale_load_down(se->runnable_weight);
708 static inline bool sched_asym_prefer(int a, int b)
710 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
714 struct em_perf_domain *em_pd;
715 struct perf_domain *next;
719 /* Scheduling group status flags */
720 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
721 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
724 * We add the notion of a root-domain which will be used to define per-domain
725 * variables. Each exclusive cpuset essentially defines an island domain by
726 * fully partitioning the member CPUs from any other cpuset. Whenever a new
727 * exclusive cpuset is created, we also create and attach a new root-domain
736 cpumask_var_t online;
739 * Indicate pullable load on at least one CPU, e.g:
740 * - More than one runnable task
741 * - Running task is misfit
745 /* Indicate one or more cpus over-utilized (tipping point) */
749 * The bit corresponding to a CPU gets set here if such CPU has more
750 * than one runnable -deadline task (as it is below for RT tasks).
752 cpumask_var_t dlo_mask;
757 #ifdef HAVE_RT_PUSH_IPI
759 * For IPI pull requests, loop across the rto_mask.
761 struct irq_work rto_push_work;
762 raw_spinlock_t rto_lock;
763 /* These are only updated and read within rto_lock */
766 /* These atomics are updated outside of a lock */
767 atomic_t rto_loop_next;
768 atomic_t rto_loop_start;
771 * The "RT overload" flag: it gets set if a CPU has more than
772 * one runnable RT task.
774 cpumask_var_t rto_mask;
775 struct cpupri cpupri;
777 unsigned long max_cpu_capacity;
780 * NULL-terminated list of performance domains intersecting with the
781 * CPUs of the rd. Protected by RCU.
783 struct perf_domain __rcu *pd;
786 extern struct root_domain def_root_domain;
787 extern struct mutex sched_domains_mutex;
789 extern void init_defrootdomain(void);
790 extern int sched_init_domains(const struct cpumask *cpu_map);
791 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
792 extern void sched_get_rd(struct root_domain *rd);
793 extern void sched_put_rd(struct root_domain *rd);
795 #ifdef HAVE_RT_PUSH_IPI
796 extern void rto_push_irq_work_func(struct irq_work *work);
798 #endif /* CONFIG_SMP */
801 * This is the main, per-CPU runqueue data structure.
803 * Locking rule: those places that want to lock multiple runqueues
804 * (such as the load balancing or the thread migration code), lock
805 * acquire operations must be ordered by ascending &runqueue.
812 * nr_running and cpu_load should be in the same cacheline because
813 * remote CPUs use both these fields when doing load calculation.
815 unsigned int nr_running;
816 #ifdef CONFIG_NUMA_BALANCING
817 unsigned int nr_numa_running;
818 unsigned int nr_preferred_running;
819 unsigned int numa_migrate_on;
821 #define CPU_LOAD_IDX_MAX 5
822 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
823 #ifdef CONFIG_NO_HZ_COMMON
825 unsigned long last_load_update_tick;
826 unsigned long last_blocked_load_update_tick;
827 unsigned int has_blocked_load;
828 #endif /* CONFIG_SMP */
829 unsigned int nohz_tick_stopped;
831 #endif /* CONFIG_NO_HZ_COMMON */
833 /* capture load from *all* tasks on this CPU: */
834 struct load_weight load;
835 unsigned long nr_load_updates;
842 #ifdef CONFIG_FAIR_GROUP_SCHED
843 /* list of leaf cfs_rq on this CPU: */
844 struct list_head leaf_cfs_rq_list;
845 struct list_head *tmp_alone_branch;
846 #endif /* CONFIG_FAIR_GROUP_SCHED */
849 * This is part of a global counter where only the total sum
850 * over all CPUs matters. A task can increase this counter on
851 * one CPU and if it got migrated afterwards it may decrease
852 * it on another CPU. Always updated under the runqueue lock:
854 unsigned long nr_uninterruptible;
856 struct task_struct *curr;
857 struct task_struct *idle;
858 struct task_struct *stop;
859 unsigned long next_balance;
860 struct mm_struct *prev_mm;
862 unsigned int clock_update_flags;
864 /* Ensure that all clocks are in the same cache line */
865 u64 clock_task ____cacheline_aligned;
867 unsigned long lost_idle_time;
872 struct root_domain *rd;
873 struct sched_domain __rcu *sd;
875 unsigned long cpu_capacity;
876 unsigned long cpu_capacity_orig;
878 struct callback_head *balance_callback;
880 unsigned char idle_balance;
882 unsigned long misfit_task_load;
884 /* For active balancing */
887 struct cpu_stop_work active_balance_work;
889 /* CPU of this runqueue: */
893 struct list_head cfs_tasks;
895 struct sched_avg avg_rt;
896 struct sched_avg avg_dl;
897 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
898 struct sched_avg avg_irq;
903 /* This is used to determine avg_idle's max value */
904 u64 max_idle_balance_cost;
907 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
910 #ifdef CONFIG_PARAVIRT
913 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
914 u64 prev_steal_time_rq;
917 /* calc_load related fields */
918 unsigned long calc_load_update;
919 long calc_load_active;
921 #ifdef CONFIG_SCHED_HRTICK
923 int hrtick_csd_pending;
924 call_single_data_t hrtick_csd;
926 struct hrtimer hrtick_timer;
929 #ifdef CONFIG_SCHEDSTATS
931 struct sched_info rq_sched_info;
932 unsigned long long rq_cpu_time;
933 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
935 /* sys_sched_yield() stats */
936 unsigned int yld_count;
938 /* schedule() stats */
939 unsigned int sched_count;
940 unsigned int sched_goidle;
942 /* try_to_wake_up() stats */
943 unsigned int ttwu_count;
944 unsigned int ttwu_local;
948 struct llist_head wake_list;
951 #ifdef CONFIG_CPU_IDLE
952 /* Must be inspected within a rcu lock section */
953 struct cpuidle_state *idle_state;
957 #ifdef CONFIG_FAIR_GROUP_SCHED
959 /* CPU runqueue to which this cfs_rq is attached */
960 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
967 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
969 return container_of(cfs_rq, struct rq, cfs);
973 static inline int cpu_of(struct rq *rq)
983 #ifdef CONFIG_SCHED_SMT
984 extern void __update_idle_core(struct rq *rq);
986 static inline void update_idle_core(struct rq *rq)
988 if (static_branch_unlikely(&sched_smt_present))
989 __update_idle_core(rq);
993 static inline void update_idle_core(struct rq *rq) { }
996 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
998 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
999 #define this_rq() this_cpu_ptr(&runqueues)
1000 #define task_rq(p) cpu_rq(task_cpu(p))
1001 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1002 #define raw_rq() raw_cpu_ptr(&runqueues)
1004 extern void update_rq_clock(struct rq *rq);
1006 static inline u64 __rq_clock_broken(struct rq *rq)
1008 return READ_ONCE(rq->clock);
1012 * rq::clock_update_flags bits
1014 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1015 * call to __schedule(). This is an optimisation to avoid
1016 * neighbouring rq clock updates.
1018 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1019 * in effect and calls to update_rq_clock() are being ignored.
1021 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1022 * made to update_rq_clock() since the last time rq::lock was pinned.
1024 * If inside of __schedule(), clock_update_flags will have been
1025 * shifted left (a left shift is a cheap operation for the fast path
1026 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1028 * if (rq-clock_update_flags >= RQCF_UPDATED)
1030 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1031 * one position though, because the next rq_unpin_lock() will shift it
1034 #define RQCF_REQ_SKIP 0x01
1035 #define RQCF_ACT_SKIP 0x02
1036 #define RQCF_UPDATED 0x04
1038 static inline void assert_clock_updated(struct rq *rq)
1041 * The only reason for not seeing a clock update since the
1042 * last rq_pin_lock() is if we're currently skipping updates.
1044 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1047 static inline u64 rq_clock(struct rq *rq)
1049 lockdep_assert_held(&rq->lock);
1050 assert_clock_updated(rq);
1055 static inline u64 rq_clock_task(struct rq *rq)
1057 lockdep_assert_held(&rq->lock);
1058 assert_clock_updated(rq);
1060 return rq->clock_task;
1063 static inline void rq_clock_skip_update(struct rq *rq)
1065 lockdep_assert_held(&rq->lock);
1066 rq->clock_update_flags |= RQCF_REQ_SKIP;
1070 * See rt task throttling, which is the only time a skip
1071 * request is cancelled.
1073 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1075 lockdep_assert_held(&rq->lock);
1076 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1080 unsigned long flags;
1081 struct pin_cookie cookie;
1082 #ifdef CONFIG_SCHED_DEBUG
1084 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1085 * current pin context is stashed here in case it needs to be
1086 * restored in rq_repin_lock().
1088 unsigned int clock_update_flags;
1092 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1094 rf->cookie = lockdep_pin_lock(&rq->lock);
1096 #ifdef CONFIG_SCHED_DEBUG
1097 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1098 rf->clock_update_flags = 0;
1102 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1104 #ifdef CONFIG_SCHED_DEBUG
1105 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1106 rf->clock_update_flags = RQCF_UPDATED;
1109 lockdep_unpin_lock(&rq->lock, rf->cookie);
1112 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1114 lockdep_repin_lock(&rq->lock, rf->cookie);
1116 #ifdef CONFIG_SCHED_DEBUG
1118 * Restore the value we stashed in @rf for this pin context.
1120 rq->clock_update_flags |= rf->clock_update_flags;
1124 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1125 __acquires(rq->lock);
1127 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1128 __acquires(p->pi_lock)
1129 __acquires(rq->lock);
1131 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1132 __releases(rq->lock)
1134 rq_unpin_lock(rq, rf);
1135 raw_spin_unlock(&rq->lock);
1139 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1140 __releases(rq->lock)
1141 __releases(p->pi_lock)
1143 rq_unpin_lock(rq, rf);
1144 raw_spin_unlock(&rq->lock);
1145 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1149 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1150 __acquires(rq->lock)
1152 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1153 rq_pin_lock(rq, rf);
1157 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1158 __acquires(rq->lock)
1160 raw_spin_lock_irq(&rq->lock);
1161 rq_pin_lock(rq, rf);
1165 rq_lock(struct rq *rq, struct rq_flags *rf)
1166 __acquires(rq->lock)
1168 raw_spin_lock(&rq->lock);
1169 rq_pin_lock(rq, rf);
1173 rq_relock(struct rq *rq, struct rq_flags *rf)
1174 __acquires(rq->lock)
1176 raw_spin_lock(&rq->lock);
1177 rq_repin_lock(rq, rf);
1181 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1182 __releases(rq->lock)
1184 rq_unpin_lock(rq, rf);
1185 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1189 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1190 __releases(rq->lock)
1192 rq_unpin_lock(rq, rf);
1193 raw_spin_unlock_irq(&rq->lock);
1197 rq_unlock(struct rq *rq, struct rq_flags *rf)
1198 __releases(rq->lock)
1200 rq_unpin_lock(rq, rf);
1201 raw_spin_unlock(&rq->lock);
1204 static inline struct rq *
1205 this_rq_lock_irq(struct rq_flags *rf)
1206 __acquires(rq->lock)
1210 local_irq_disable();
1217 enum numa_topology_type {
1222 extern enum numa_topology_type sched_numa_topology_type;
1223 extern int sched_max_numa_distance;
1224 extern bool find_numa_distance(int distance);
1228 extern void sched_init_numa(void);
1229 extern void sched_domains_numa_masks_set(unsigned int cpu);
1230 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1232 static inline void sched_init_numa(void) { }
1233 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1234 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1237 #ifdef CONFIG_NUMA_BALANCING
1238 /* The regions in numa_faults array from task_struct */
1239 enum numa_faults_stats {
1245 extern void sched_setnuma(struct task_struct *p, int node);
1246 extern int migrate_task_to(struct task_struct *p, int cpu);
1247 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1249 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1252 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1255 #endif /* CONFIG_NUMA_BALANCING */
1260 queue_balance_callback(struct rq *rq,
1261 struct callback_head *head,
1262 void (*func)(struct rq *rq))
1264 lockdep_assert_held(&rq->lock);
1266 if (unlikely(head->next))
1269 head->func = (void (*)(struct callback_head *))func;
1270 head->next = rq->balance_callback;
1271 rq->balance_callback = head;
1274 extern void sched_ttwu_pending(void);
1276 #define rcu_dereference_check_sched_domain(p) \
1277 rcu_dereference_check((p), \
1278 lockdep_is_held(&sched_domains_mutex))
1281 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1282 * See destroy_sched_domains: call_rcu for details.
1284 * The domain tree of any CPU may only be accessed from within
1285 * preempt-disabled sections.
1287 #define for_each_domain(cpu, __sd) \
1288 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1289 __sd; __sd = __sd->parent)
1291 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1294 * highest_flag_domain - Return highest sched_domain containing flag.
1295 * @cpu: The CPU whose highest level of sched domain is to
1297 * @flag: The flag to check for the highest sched_domain
1298 * for the given CPU.
1300 * Returns the highest sched_domain of a CPU which contains the given flag.
1302 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1304 struct sched_domain *sd, *hsd = NULL;
1306 for_each_domain(cpu, sd) {
1307 if (!(sd->flags & flag))
1315 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1317 struct sched_domain *sd;
1319 for_each_domain(cpu, sd) {
1320 if (sd->flags & flag)
1327 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1328 DECLARE_PER_CPU(int, sd_llc_size);
1329 DECLARE_PER_CPU(int, sd_llc_id);
1330 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1331 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1332 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1333 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1334 extern struct static_key_false sched_asym_cpucapacity;
1336 struct sched_group_capacity {
1339 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1342 unsigned long capacity;
1343 unsigned long min_capacity; /* Min per-CPU capacity in group */
1344 unsigned long max_capacity; /* Max per-CPU capacity in group */
1345 unsigned long next_update;
1346 int imbalance; /* XXX unrelated to capacity but shared group state */
1348 #ifdef CONFIG_SCHED_DEBUG
1352 unsigned long cpumask[0]; /* Balance mask */
1355 struct sched_group {
1356 struct sched_group *next; /* Must be a circular list */
1359 unsigned int group_weight;
1360 struct sched_group_capacity *sgc;
1361 int asym_prefer_cpu; /* CPU of highest priority in group */
1364 * The CPUs this group covers.
1366 * NOTE: this field is variable length. (Allocated dynamically
1367 * by attaching extra space to the end of the structure,
1368 * depending on how many CPUs the kernel has booted up with)
1370 unsigned long cpumask[0];
1373 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1375 return to_cpumask(sg->cpumask);
1379 * See build_balance_mask().
1381 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1383 return to_cpumask(sg->sgc->cpumask);
1387 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1388 * @group: The group whose first CPU is to be returned.
1390 static inline unsigned int group_first_cpu(struct sched_group *group)
1392 return cpumask_first(sched_group_span(group));
1395 extern int group_balance_cpu(struct sched_group *sg);
1397 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1398 void register_sched_domain_sysctl(void);
1399 void dirty_sched_domain_sysctl(int cpu);
1400 void unregister_sched_domain_sysctl(void);
1402 static inline void register_sched_domain_sysctl(void)
1405 static inline void dirty_sched_domain_sysctl(int cpu)
1408 static inline void unregister_sched_domain_sysctl(void)
1415 static inline void sched_ttwu_pending(void) { }
1417 #endif /* CONFIG_SMP */
1420 #include "autogroup.h"
1422 #ifdef CONFIG_CGROUP_SCHED
1425 * Return the group to which this tasks belongs.
1427 * We cannot use task_css() and friends because the cgroup subsystem
1428 * changes that value before the cgroup_subsys::attach() method is called,
1429 * therefore we cannot pin it and might observe the wrong value.
1431 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1432 * core changes this before calling sched_move_task().
1434 * Instead we use a 'copy' which is updated from sched_move_task() while
1435 * holding both task_struct::pi_lock and rq::lock.
1437 static inline struct task_group *task_group(struct task_struct *p)
1439 return p->sched_task_group;
1442 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1443 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1445 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1446 struct task_group *tg = task_group(p);
1449 #ifdef CONFIG_FAIR_GROUP_SCHED
1450 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1451 p->se.cfs_rq = tg->cfs_rq[cpu];
1452 p->se.parent = tg->se[cpu];
1455 #ifdef CONFIG_RT_GROUP_SCHED
1456 p->rt.rt_rq = tg->rt_rq[cpu];
1457 p->rt.parent = tg->rt_se[cpu];
1461 #else /* CONFIG_CGROUP_SCHED */
1463 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1464 static inline struct task_group *task_group(struct task_struct *p)
1469 #endif /* CONFIG_CGROUP_SCHED */
1471 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1473 set_task_rq(p, cpu);
1476 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1477 * successfully executed on another CPU. We must ensure that updates of
1478 * per-task data have been completed by this moment.
1481 #ifdef CONFIG_THREAD_INFO_IN_TASK
1482 WRITE_ONCE(p->cpu, cpu);
1484 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1491 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1493 #ifdef CONFIG_SCHED_DEBUG
1494 # include <linux/static_key.h>
1495 # define const_debug __read_mostly
1497 # define const_debug const
1500 #define SCHED_FEAT(name, enabled) \
1501 __SCHED_FEAT_##name ,
1504 #include "features.h"
1510 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1513 * To support run-time toggling of sched features, all the translation units
1514 * (but core.c) reference the sysctl_sched_features defined in core.c.
1516 extern const_debug unsigned int sysctl_sched_features;
1518 #define SCHED_FEAT(name, enabled) \
1519 static __always_inline bool static_branch_##name(struct static_key *key) \
1521 return static_key_##enabled(key); \
1524 #include "features.h"
1527 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1528 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1530 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1533 * Each translation unit has its own copy of sysctl_sched_features to allow
1534 * constants propagation at compile time and compiler optimization based on
1537 #define SCHED_FEAT(name, enabled) \
1538 (1UL << __SCHED_FEAT_##name) * enabled |
1539 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1540 #include "features.h"
1544 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1546 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1548 extern struct static_key_false sched_numa_balancing;
1549 extern struct static_key_false sched_schedstats;
1551 static inline u64 global_rt_period(void)
1553 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1556 static inline u64 global_rt_runtime(void)
1558 if (sysctl_sched_rt_runtime < 0)
1561 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1564 static inline int task_current(struct rq *rq, struct task_struct *p)
1566 return rq->curr == p;
1569 static inline int task_running(struct rq *rq, struct task_struct *p)
1574 return task_current(rq, p);
1578 static inline int task_on_rq_queued(struct task_struct *p)
1580 return p->on_rq == TASK_ON_RQ_QUEUED;
1583 static inline int task_on_rq_migrating(struct task_struct *p)
1585 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1591 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1592 #define WF_FORK 0x02 /* Child wakeup after fork */
1593 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1596 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1597 * of tasks with abnormal "nice" values across CPUs the contribution that
1598 * each task makes to its run queue's load is weighted according to its
1599 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1600 * scaled version of the new time slice allocation that they receive on time
1604 #define WEIGHT_IDLEPRIO 3
1605 #define WMULT_IDLEPRIO 1431655765
1607 extern const int sched_prio_to_weight[40];
1608 extern const u32 sched_prio_to_wmult[40];
1611 * {de,en}queue flags:
1613 * DEQUEUE_SLEEP - task is no longer runnable
1614 * ENQUEUE_WAKEUP - task just became runnable
1616 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1617 * are in a known state which allows modification. Such pairs
1618 * should preserve as much state as possible.
1620 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1623 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1624 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1625 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1629 #define DEQUEUE_SLEEP 0x01
1630 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1631 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1632 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1634 #define ENQUEUE_WAKEUP 0x01
1635 #define ENQUEUE_RESTORE 0x02
1636 #define ENQUEUE_MOVE 0x04
1637 #define ENQUEUE_NOCLOCK 0x08
1639 #define ENQUEUE_HEAD 0x10
1640 #define ENQUEUE_REPLENISH 0x20
1642 #define ENQUEUE_MIGRATED 0x40
1644 #define ENQUEUE_MIGRATED 0x00
1647 #define RETRY_TASK ((void *)-1UL)
1649 struct sched_class {
1650 const struct sched_class *next;
1652 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1653 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1654 void (*yield_task) (struct rq *rq);
1655 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1657 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1660 * It is the responsibility of the pick_next_task() method that will
1661 * return the next task to call put_prev_task() on the @prev task or
1662 * something equivalent.
1664 * May return RETRY_TASK when it finds a higher prio class has runnable
1667 struct task_struct * (*pick_next_task)(struct rq *rq,
1668 struct task_struct *prev,
1669 struct rq_flags *rf);
1670 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1673 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1674 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1676 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1678 void (*set_cpus_allowed)(struct task_struct *p,
1679 const struct cpumask *newmask);
1681 void (*rq_online)(struct rq *rq);
1682 void (*rq_offline)(struct rq *rq);
1685 void (*set_curr_task)(struct rq *rq);
1686 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1687 void (*task_fork)(struct task_struct *p);
1688 void (*task_dead)(struct task_struct *p);
1691 * The switched_from() call is allowed to drop rq->lock, therefore we
1692 * cannot assume the switched_from/switched_to pair is serliazed by
1693 * rq->lock. They are however serialized by p->pi_lock.
1695 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1696 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1697 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1700 unsigned int (*get_rr_interval)(struct rq *rq,
1701 struct task_struct *task);
1703 void (*update_curr)(struct rq *rq);
1705 #define TASK_SET_GROUP 0
1706 #define TASK_MOVE_GROUP 1
1708 #ifdef CONFIG_FAIR_GROUP_SCHED
1709 void (*task_change_group)(struct task_struct *p, int type);
1713 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1715 prev->sched_class->put_prev_task(rq, prev);
1718 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1720 curr->sched_class->set_curr_task(rq);
1724 #define sched_class_highest (&stop_sched_class)
1726 #define sched_class_highest (&dl_sched_class)
1728 #define for_each_class(class) \
1729 for (class = sched_class_highest; class; class = class->next)
1731 extern const struct sched_class stop_sched_class;
1732 extern const struct sched_class dl_sched_class;
1733 extern const struct sched_class rt_sched_class;
1734 extern const struct sched_class fair_sched_class;
1735 extern const struct sched_class idle_sched_class;
1740 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1742 extern void trigger_load_balance(struct rq *rq);
1744 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1748 #ifdef CONFIG_CPU_IDLE
1749 static inline void idle_set_state(struct rq *rq,
1750 struct cpuidle_state *idle_state)
1752 rq->idle_state = idle_state;
1755 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1757 SCHED_WARN_ON(!rcu_read_lock_held());
1759 return rq->idle_state;
1762 static inline void idle_set_state(struct rq *rq,
1763 struct cpuidle_state *idle_state)
1767 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1773 extern void schedule_idle(void);
1775 extern void sysrq_sched_debug_show(void);
1776 extern void sched_init_granularity(void);
1777 extern void update_max_interval(void);
1779 extern void init_sched_dl_class(void);
1780 extern void init_sched_rt_class(void);
1781 extern void init_sched_fair_class(void);
1783 extern void reweight_task(struct task_struct *p, int prio);
1785 extern void resched_curr(struct rq *rq);
1786 extern void resched_cpu(int cpu);
1788 extern struct rt_bandwidth def_rt_bandwidth;
1789 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1791 extern struct dl_bandwidth def_dl_bandwidth;
1792 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1793 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1794 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1795 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1798 #define BW_UNIT (1 << BW_SHIFT)
1799 #define RATIO_SHIFT 8
1800 unsigned long to_ratio(u64 period, u64 runtime);
1802 extern void init_entity_runnable_average(struct sched_entity *se);
1803 extern void post_init_entity_util_avg(struct task_struct *p);
1805 #ifdef CONFIG_NO_HZ_FULL
1806 extern bool sched_can_stop_tick(struct rq *rq);
1807 extern int __init sched_tick_offload_init(void);
1810 * Tick may be needed by tasks in the runqueue depending on their policy and
1811 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1812 * nohz mode if necessary.
1814 static inline void sched_update_tick_dependency(struct rq *rq)
1818 if (!tick_nohz_full_enabled())
1823 if (!tick_nohz_full_cpu(cpu))
1826 if (sched_can_stop_tick(rq))
1827 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1829 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1832 static inline int sched_tick_offload_init(void) { return 0; }
1833 static inline void sched_update_tick_dependency(struct rq *rq) { }
1836 static inline void add_nr_running(struct rq *rq, unsigned count)
1838 unsigned prev_nr = rq->nr_running;
1840 rq->nr_running = prev_nr + count;
1843 if (prev_nr < 2 && rq->nr_running >= 2) {
1844 if (!READ_ONCE(rq->rd->overload))
1845 WRITE_ONCE(rq->rd->overload, 1);
1849 sched_update_tick_dependency(rq);
1852 static inline void sub_nr_running(struct rq *rq, unsigned count)
1854 rq->nr_running -= count;
1855 /* Check if we still need preemption */
1856 sched_update_tick_dependency(rq);
1859 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1860 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1862 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1864 extern const_debug unsigned int sysctl_sched_nr_migrate;
1865 extern const_debug unsigned int sysctl_sched_migration_cost;
1867 #ifdef CONFIG_SCHED_HRTICK
1871 * - enabled by features
1872 * - hrtimer is actually high res
1874 static inline int hrtick_enabled(struct rq *rq)
1876 if (!sched_feat(HRTICK))
1878 if (!cpu_active(cpu_of(rq)))
1880 return hrtimer_is_hres_active(&rq->hrtick_timer);
1883 void hrtick_start(struct rq *rq, u64 delay);
1887 static inline int hrtick_enabled(struct rq *rq)
1892 #endif /* CONFIG_SCHED_HRTICK */
1894 #ifndef arch_scale_freq_capacity
1895 static __always_inline
1896 unsigned long arch_scale_freq_capacity(int cpu)
1898 return SCHED_CAPACITY_SCALE;
1903 #ifdef CONFIG_PREEMPT
1905 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1908 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1909 * way at the expense of forcing extra atomic operations in all
1910 * invocations. This assures that the double_lock is acquired using the
1911 * same underlying policy as the spinlock_t on this architecture, which
1912 * reduces latency compared to the unfair variant below. However, it
1913 * also adds more overhead and therefore may reduce throughput.
1915 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1916 __releases(this_rq->lock)
1917 __acquires(busiest->lock)
1918 __acquires(this_rq->lock)
1920 raw_spin_unlock(&this_rq->lock);
1921 double_rq_lock(this_rq, busiest);
1928 * Unfair double_lock_balance: Optimizes throughput at the expense of
1929 * latency by eliminating extra atomic operations when the locks are
1930 * already in proper order on entry. This favors lower CPU-ids and will
1931 * grant the double lock to lower CPUs over higher ids under contention,
1932 * regardless of entry order into the function.
1934 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1935 __releases(this_rq->lock)
1936 __acquires(busiest->lock)
1937 __acquires(this_rq->lock)
1941 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1942 if (busiest < this_rq) {
1943 raw_spin_unlock(&this_rq->lock);
1944 raw_spin_lock(&busiest->lock);
1945 raw_spin_lock_nested(&this_rq->lock,
1946 SINGLE_DEPTH_NESTING);
1949 raw_spin_lock_nested(&busiest->lock,
1950 SINGLE_DEPTH_NESTING);
1955 #endif /* CONFIG_PREEMPT */
1958 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1960 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1962 if (unlikely(!irqs_disabled())) {
1963 /* printk() doesn't work well under rq->lock */
1964 raw_spin_unlock(&this_rq->lock);
1968 return _double_lock_balance(this_rq, busiest);
1971 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1972 __releases(busiest->lock)
1974 raw_spin_unlock(&busiest->lock);
1975 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1978 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1984 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1987 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1993 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1996 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2002 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2006 * double_rq_lock - safely lock two runqueues
2008 * Note this does not disable interrupts like task_rq_lock,
2009 * you need to do so manually before calling.
2011 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2012 __acquires(rq1->lock)
2013 __acquires(rq2->lock)
2015 BUG_ON(!irqs_disabled());
2017 raw_spin_lock(&rq1->lock);
2018 __acquire(rq2->lock); /* Fake it out ;) */
2021 raw_spin_lock(&rq1->lock);
2022 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2024 raw_spin_lock(&rq2->lock);
2025 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2031 * double_rq_unlock - safely unlock two runqueues
2033 * Note this does not restore interrupts like task_rq_unlock,
2034 * you need to do so manually after calling.
2036 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2037 __releases(rq1->lock)
2038 __releases(rq2->lock)
2040 raw_spin_unlock(&rq1->lock);
2042 raw_spin_unlock(&rq2->lock);
2044 __release(rq2->lock);
2047 extern void set_rq_online (struct rq *rq);
2048 extern void set_rq_offline(struct rq *rq);
2049 extern bool sched_smp_initialized;
2051 #else /* CONFIG_SMP */
2054 * double_rq_lock - safely lock two runqueues
2056 * Note this does not disable interrupts like task_rq_lock,
2057 * you need to do so manually before calling.
2059 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2060 __acquires(rq1->lock)
2061 __acquires(rq2->lock)
2063 BUG_ON(!irqs_disabled());
2065 raw_spin_lock(&rq1->lock);
2066 __acquire(rq2->lock); /* Fake it out ;) */
2070 * double_rq_unlock - safely unlock two runqueues
2072 * Note this does not restore interrupts like task_rq_unlock,
2073 * you need to do so manually after calling.
2075 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2076 __releases(rq1->lock)
2077 __releases(rq2->lock)
2080 raw_spin_unlock(&rq1->lock);
2081 __release(rq2->lock);
2086 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2087 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2089 #ifdef CONFIG_SCHED_DEBUG
2090 extern bool sched_debug_enabled;
2092 extern void print_cfs_stats(struct seq_file *m, int cpu);
2093 extern void print_rt_stats(struct seq_file *m, int cpu);
2094 extern void print_dl_stats(struct seq_file *m, int cpu);
2095 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2096 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2097 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2098 #ifdef CONFIG_NUMA_BALANCING
2100 show_numa_stats(struct task_struct *p, struct seq_file *m);
2102 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2103 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2104 #endif /* CONFIG_NUMA_BALANCING */
2105 #endif /* CONFIG_SCHED_DEBUG */
2107 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2108 extern void init_rt_rq(struct rt_rq *rt_rq);
2109 extern void init_dl_rq(struct dl_rq *dl_rq);
2111 extern void cfs_bandwidth_usage_inc(void);
2112 extern void cfs_bandwidth_usage_dec(void);
2114 #ifdef CONFIG_NO_HZ_COMMON
2115 #define NOHZ_BALANCE_KICK_BIT 0
2116 #define NOHZ_STATS_KICK_BIT 1
2118 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2119 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2121 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2123 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2125 extern void nohz_balance_exit_idle(struct rq *rq);
2127 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2133 void __dl_update(struct dl_bw *dl_b, s64 bw)
2135 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2138 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2139 "sched RCU must be held");
2140 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2141 struct rq *rq = cpu_rq(i);
2143 rq->dl.extra_bw += bw;
2148 void __dl_update(struct dl_bw *dl_b, s64 bw)
2150 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2157 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2162 struct u64_stats_sync sync;
2165 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2168 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2169 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2170 * and never move forward.
2172 static inline u64 irq_time_read(int cpu)
2174 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2179 seq = __u64_stats_fetch_begin(&irqtime->sync);
2180 total = irqtime->total;
2181 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2185 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2187 #ifdef CONFIG_CPU_FREQ
2188 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2191 * cpufreq_update_util - Take a note about CPU utilization changes.
2192 * @rq: Runqueue to carry out the update for.
2193 * @flags: Update reason flags.
2195 * This function is called by the scheduler on the CPU whose utilization is
2198 * It can only be called from RCU-sched read-side critical sections.
2200 * The way cpufreq is currently arranged requires it to evaluate the CPU
2201 * performance state (frequency/voltage) on a regular basis to prevent it from
2202 * being stuck in a completely inadequate performance level for too long.
2203 * That is not guaranteed to happen if the updates are only triggered from CFS
2204 * and DL, though, because they may not be coming in if only RT tasks are
2205 * active all the time (or there are RT tasks only).
2207 * As a workaround for that issue, this function is called periodically by the
2208 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2209 * but that really is a band-aid. Going forward it should be replaced with
2210 * solutions targeted more specifically at RT tasks.
2212 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2214 struct update_util_data *data;
2216 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2219 data->func(data, rq_clock(rq), flags);
2222 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2223 #endif /* CONFIG_CPU_FREQ */
2225 #ifdef arch_scale_freq_capacity
2226 # ifndef arch_scale_freq_invariant
2227 # define arch_scale_freq_invariant() true
2230 # define arch_scale_freq_invariant() false
2234 static inline unsigned long capacity_orig_of(int cpu)
2236 return cpu_rq(cpu)->cpu_capacity_orig;
2240 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2242 * enum schedutil_type - CPU utilization type
2243 * @FREQUENCY_UTIL: Utilization used to select frequency
2244 * @ENERGY_UTIL: Utilization used during energy calculation
2246 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2247 * need to be aggregated differently depending on the usage made of them. This
2248 * enum is used within schedutil_freq_util() to differentiate the types of
2249 * utilization expected by the callers, and adjust the aggregation accordingly.
2251 enum schedutil_type {
2256 unsigned long schedutil_freq_util(int cpu, unsigned long util_cfs,
2257 unsigned long max, enum schedutil_type type);
2259 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2261 unsigned long max = arch_scale_cpu_capacity(NULL, cpu);
2263 return schedutil_freq_util(cpu, cfs, max, ENERGY_UTIL);
2266 static inline unsigned long cpu_bw_dl(struct rq *rq)
2268 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2271 static inline unsigned long cpu_util_dl(struct rq *rq)
2273 return READ_ONCE(rq->avg_dl.util_avg);
2276 static inline unsigned long cpu_util_cfs(struct rq *rq)
2278 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2280 if (sched_feat(UTIL_EST)) {
2281 util = max_t(unsigned long, util,
2282 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2288 static inline unsigned long cpu_util_rt(struct rq *rq)
2290 return READ_ONCE(rq->avg_rt.util_avg);
2292 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2293 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2299 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2300 static inline unsigned long cpu_util_irq(struct rq *rq)
2302 return rq->avg_irq.util_avg;
2306 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2308 util *= (max - irq);
2315 static inline unsigned long cpu_util_irq(struct rq *rq)
2321 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2327 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2329 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2331 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2333 static inline bool sched_energy_enabled(void)
2335 return static_branch_unlikely(&sched_energy_present);
2338 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2340 #define perf_domain_span(pd) NULL
2341 static inline bool sched_energy_enabled(void) { return false; }
2343 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */