projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
sched/fair: Start tracking SCHED_IDLE tasks count in cfs_rq
[linux-2.6-block.git] / kernel / sched / sched.h
... / ...
CommitLineData
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Scheduler internal types and methods:
4 */
5#include <linux/sched.h>
6
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>
35
36#include <uapi/linux/sched/types.h>
37
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>
68
69#include <asm/tlb.h>
70
71#ifdef CONFIG_PARAVIRT
72# include <asm/paravirt.h>
73#endif
74
75#include "cpupri.h"
76#include "cpudeadline.h"
77
78#ifdef CONFIG_SCHED_DEBUG
79# define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
80#else
81# define SCHED_WARN_ON(x) ({ (void)(x), 0; })
82#endif
83
84struct rq;
85struct cpuidle_state;
86
87/* task_struct::on_rq states: */
88#define TASK_ON_RQ_QUEUED 1
89#define TASK_ON_RQ_MIGRATING 2
90
91extern __read_mostly int scheduler_running;
92
93extern unsigned long calc_load_update;
94extern atomic_long_t calc_load_tasks;
95
96extern void calc_global_load_tick(struct rq *this_rq);
97extern long calc_load_fold_active(struct rq *this_rq, long adjust);
98
99/*
100 * Helpers for converting nanosecond timing to jiffy resolution
101 */
102#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
103
104/*
105 * Increase resolution of nice-level calculations for 64-bit architectures.
106 * The extra resolution improves shares distribution and load balancing of
107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108 * hierarchies, especially on larger systems. This is not a user-visible change
109 * and does not change the user-interface for setting shares/weights.
110 *
111 * We increase resolution only if we have enough bits to allow this increased
112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113 * are pretty high and the returns do not justify the increased costs.
114 *
115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116 * increase coverage and consistency always enable it on 64-bit platforms.
117 */
118#ifdef CONFIG_64BIT
119# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
121# define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
122#else
123# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
124# define scale_load(w) (w)
125# define scale_load_down(w) (w)
126#endif
127
128/*
129 * Task weight (visible to users) and its load (invisible to users) have
130 * independent resolution, but they should be well calibrated. We use
131 * scale_load() and scale_load_down(w) to convert between them. The
132 * following must be true:
133 *
134 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
135 *
136 */
137#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
138
139/*
140 * Single value that decides SCHED_DEADLINE internal math precision.
141 * 10 -> just above 1us
142 * 9 -> just above 0.5us
143 */
144#define DL_SCALE 10
145
146/*
147 * Single value that denotes runtime == period, ie unlimited time.
148 */
149#define RUNTIME_INF ((u64)~0ULL)
150
151static inline int idle_policy(int policy)
152{
153 return policy == SCHED_IDLE;
154}
155static inline int fair_policy(int policy)
156{
157 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
158}
159
160static inline int rt_policy(int policy)
161{
162 return policy == SCHED_FIFO || policy == SCHED_RR;
163}
164
165static inline int dl_policy(int policy)
166{
167 return policy == SCHED_DEADLINE;
168}
169static inline bool valid_policy(int policy)
170{
171 return idle_policy(policy) || fair_policy(policy) ||
172 rt_policy(policy) || dl_policy(policy);
173}
174
175static inline int task_has_idle_policy(struct task_struct *p)
176{
177 return idle_policy(p->policy);
178}
179
180static inline int task_has_rt_policy(struct task_struct *p)
181{
182 return rt_policy(p->policy);
183}
184
185static inline int task_has_dl_policy(struct task_struct *p)
186{
187 return dl_policy(p->policy);
188}
189
190#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
191
192/*
193 * !! For sched_setattr_nocheck() (kernel) only !!
194 *
195 * This is actually gross. :(
196 *
197 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
198 * tasks, but still be able to sleep. We need this on platforms that cannot
199 * atomically change clock frequency. Remove once fast switching will be
200 * available on such platforms.
201 *
202 * SUGOV stands for SchedUtil GOVernor.
203 */
204#define SCHED_FLAG_SUGOV 0x10000000
205
206static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
207{
208#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
209 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
210#else
211 return false;
212#endif
213}
214
215/*
216 * Tells if entity @a should preempt entity @b.
217 */
218static inline bool
219dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
220{
221 return dl_entity_is_special(a) ||
222 dl_time_before(a->deadline, b->deadline);
223}
224
225/*
226 * This is the priority-queue data structure of the RT scheduling class:
227 */
228struct rt_prio_array {
229 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
230 struct list_head queue[MAX_RT_PRIO];
231};
232
233struct rt_bandwidth {
234 /* nests inside the rq lock: */
235 raw_spinlock_t rt_runtime_lock;
236 ktime_t rt_period;
237 u64 rt_runtime;
238 struct hrtimer rt_period_timer;
239 unsigned int rt_period_active;
240};
241
242void __dl_clear_params(struct task_struct *p);
243
244/*
245 * To keep the bandwidth of -deadline tasks and groups under control
246 * we need some place where:
247 * - store the maximum -deadline bandwidth of the system (the group);
248 * - cache the fraction of that bandwidth that is currently allocated.
249 *
250 * This is all done in the data structure below. It is similar to the
251 * one used for RT-throttling (rt_bandwidth), with the main difference
252 * that, since here we are only interested in admission control, we
253 * do not decrease any runtime while the group "executes", neither we
254 * need a timer to replenish it.
255 *
256 * With respect to SMP, the bandwidth is given on a per-CPU basis,
257 * meaning that:
258 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
259 * - dl_total_bw array contains, in the i-eth element, the currently
260 * allocated bandwidth on the i-eth CPU.
261 * Moreover, groups consume bandwidth on each CPU, while tasks only
262 * consume bandwidth on the CPU they're running on.
263 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
264 * that will be shown the next time the proc or cgroup controls will
265 * be red. It on its turn can be changed by writing on its own
266 * control.
267 */
268struct dl_bandwidth {
269 raw_spinlock_t dl_runtime_lock;
270 u64 dl_runtime;
271 u64 dl_period;
272};
273
274static inline int dl_bandwidth_enabled(void)
275{
276 return sysctl_sched_rt_runtime >= 0;
277}
278
279struct dl_bw {
280 raw_spinlock_t lock;
281 u64 bw;
282 u64 total_bw;
283};
284
285static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
286
287static inline
288void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
289{
290 dl_b->total_bw -= tsk_bw;
291 __dl_update(dl_b, (s32)tsk_bw / cpus);
292}
293
294static inline
295void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
296{
297 dl_b->total_bw += tsk_bw;
298 __dl_update(dl_b, -((s32)tsk_bw / cpus));
299}
300
301static inline
302bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
303{
304 return dl_b->bw != -1 &&
305 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
306}
307
308extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
309extern void init_dl_bw(struct dl_bw *dl_b);
310extern int sched_dl_global_validate(void);
311extern void sched_dl_do_global(void);
312extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
313extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
314extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
315extern bool __checkparam_dl(const struct sched_attr *attr);
316extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
317extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
318extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
319extern bool dl_cpu_busy(unsigned int cpu);
320
321#ifdef CONFIG_CGROUP_SCHED
322
323#include <linux/cgroup.h>
324#include <linux/psi.h>
325
326struct cfs_rq;
327struct rt_rq;
328
329extern struct list_head task_groups;
330
331struct cfs_bandwidth {
332#ifdef CONFIG_CFS_BANDWIDTH
333 raw_spinlock_t lock;
334 ktime_t period;
335 u64 quota;
336 u64 runtime;
337 s64 hierarchical_quota;
338 u64 runtime_expires;
339 int expires_seq;
340
341 u8 idle;
342 u8 period_active;
343 u8 distribute_running;
344 u8 slack_started;
345 struct hrtimer period_timer;
346 struct hrtimer slack_timer;
347 struct list_head throttled_cfs_rq;
348
349 /* Statistics: */
350 int nr_periods;
351 int nr_throttled;
352 u64 throttled_time;
353#endif
354};
355
356/* Task group related information */
357struct task_group {
358 struct cgroup_subsys_state css;
359
360#ifdef CONFIG_FAIR_GROUP_SCHED
361 /* schedulable entities of this group on each CPU */
362 struct sched_entity **se;
363 /* runqueue "owned" by this group on each CPU */
364 struct cfs_rq **cfs_rq;
365 unsigned long shares;
366
367#ifdef CONFIG_SMP
368 /*
369 * load_avg can be heavily contended at clock tick time, so put
370 * it in its own cacheline separated from the fields above which
371 * will also be accessed at each tick.
372 */
373 atomic_long_t load_avg ____cacheline_aligned;
374#endif
375#endif
376
377#ifdef CONFIG_RT_GROUP_SCHED
378 struct sched_rt_entity **rt_se;
379 struct rt_rq **rt_rq;
380
381 struct rt_bandwidth rt_bandwidth;
382#endif
383
384 struct rcu_head rcu;
385 struct list_head list;
386
387 struct task_group *parent;
388 struct list_head siblings;
389 struct list_head children;
390
391#ifdef CONFIG_SCHED_AUTOGROUP
392 struct autogroup *autogroup;
393#endif
394
395 struct cfs_bandwidth cfs_bandwidth;
396};
397
398#ifdef CONFIG_FAIR_GROUP_SCHED
399#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
400
401/*
402 * A weight of 0 or 1 can cause arithmetics problems.
403 * A weight of a cfs_rq is the sum of weights of which entities
404 * are queued on this cfs_rq, so a weight of a entity should not be
405 * too large, so as the shares value of a task group.
406 * (The default weight is 1024 - so there's no practical
407 * limitation from this.)
408 */
409#define MIN_SHARES (1UL << 1)
410#define MAX_SHARES (1UL << 18)
411#endif
412
413typedef int (*tg_visitor)(struct task_group *, void *);
414
415extern int walk_tg_tree_from(struct task_group *from,
416 tg_visitor down, tg_visitor up, void *data);
417
418/*
419 * Iterate the full tree, calling @down when first entering a node and @up when
420 * leaving it for the final time.
421 *
422 * Caller must hold rcu_lock or sufficient equivalent.
423 */
424static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
425{
426 return walk_tg_tree_from(&root_task_group, down, up, data);
427}
428
429extern int tg_nop(struct task_group *tg, void *data);
430
431extern void free_fair_sched_group(struct task_group *tg);
432extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
433extern void online_fair_sched_group(struct task_group *tg);
434extern void unregister_fair_sched_group(struct task_group *tg);
435extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
436 struct sched_entity *se, int cpu,
437 struct sched_entity *parent);
438extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
439
440extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
441extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
442extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
443
444extern void free_rt_sched_group(struct task_group *tg);
445extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
446extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
447 struct sched_rt_entity *rt_se, int cpu,
448 struct sched_rt_entity *parent);
449extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
450extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
451extern long sched_group_rt_runtime(struct task_group *tg);
452extern long sched_group_rt_period(struct task_group *tg);
453extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
454
455extern struct task_group *sched_create_group(struct task_group *parent);
456extern void sched_online_group(struct task_group *tg,
457 struct task_group *parent);
458extern void sched_destroy_group(struct task_group *tg);
459extern void sched_offline_group(struct task_group *tg);
460
461extern void sched_move_task(struct task_struct *tsk);
462
463#ifdef CONFIG_FAIR_GROUP_SCHED
464extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
465
466#ifdef CONFIG_SMP
467extern void set_task_rq_fair(struct sched_entity *se,
468 struct cfs_rq *prev, struct cfs_rq *next);
469#else /* !CONFIG_SMP */
470static inline void set_task_rq_fair(struct sched_entity *se,
471 struct cfs_rq *prev, struct cfs_rq *next) { }
472#endif /* CONFIG_SMP */
473#endif /* CONFIG_FAIR_GROUP_SCHED */
474
475#else /* CONFIG_CGROUP_SCHED */
476
477struct cfs_bandwidth { };
478
479#endif /* CONFIG_CGROUP_SCHED */
480
481/* CFS-related fields in a runqueue */
482struct cfs_rq {
483 struct load_weight load;
484 unsigned long runnable_weight;
485 unsigned int nr_running;
486 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
487 unsigned int idle_h_nr_running; /* SCHED_IDLE */
488
489 u64 exec_clock;
490 u64 min_vruntime;
491#ifndef CONFIG_64BIT
492 u64 min_vruntime_copy;
493#endif
494
495 struct rb_root_cached tasks_timeline;
496
497 /*
498 * 'curr' points to currently running entity on this cfs_rq.
499 * It is set to NULL otherwise (i.e when none are currently running).
500 */
501 struct sched_entity *curr;
502 struct sched_entity *next;
503 struct sched_entity *last;
504 struct sched_entity *skip;
505
506#ifdef CONFIG_SCHED_DEBUG
507 unsigned int nr_spread_over;
508#endif
509
510#ifdef CONFIG_SMP
511 /*
512 * CFS load tracking
513 */
514 struct sched_avg avg;
515#ifndef CONFIG_64BIT
516 u64 load_last_update_time_copy;
517#endif
518 struct {
519 raw_spinlock_t lock ____cacheline_aligned;
520 int nr;
521 unsigned long load_avg;
522 unsigned long util_avg;
523 unsigned long runnable_sum;
524 } removed;
525
526#ifdef CONFIG_FAIR_GROUP_SCHED
527 unsigned long tg_load_avg_contrib;
528 long propagate;
529 long prop_runnable_sum;
530
531 /*
532 * h_load = weight * f(tg)
533 *
534 * Where f(tg) is the recursive weight fraction assigned to
535 * this group.
536 */
537 unsigned long h_load;
538 u64 last_h_load_update;
539 struct sched_entity *h_load_next;
540#endif /* CONFIG_FAIR_GROUP_SCHED */
541#endif /* CONFIG_SMP */
542
543#ifdef CONFIG_FAIR_GROUP_SCHED
544 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
545
546 /*
547 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
548 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
549 * (like users, containers etc.)
550 *
551 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
552 * This list is used during load balance.
553 */
554 int on_list;
555 struct list_head leaf_cfs_rq_list;
556 struct task_group *tg; /* group that "owns" this runqueue */
557
558#ifdef CONFIG_CFS_BANDWIDTH
559 int runtime_enabled;
560 int expires_seq;
561 u64 runtime_expires;
562 s64 runtime_remaining;
563
564 u64 throttled_clock;
565 u64 throttled_clock_task;
566 u64 throttled_clock_task_time;
567 int throttled;
568 int throttle_count;
569 struct list_head throttled_list;
570#endif /* CONFIG_CFS_BANDWIDTH */
571#endif /* CONFIG_FAIR_GROUP_SCHED */
572};
573
574static inline int rt_bandwidth_enabled(void)
575{
576 return sysctl_sched_rt_runtime >= 0;
577}
578
579/* RT IPI pull logic requires IRQ_WORK */
580#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
581# define HAVE_RT_PUSH_IPI
582#endif
583
584/* Real-Time classes' related field in a runqueue: */
585struct rt_rq {
586 struct rt_prio_array active;
587 unsigned int rt_nr_running;
588 unsigned int rr_nr_running;
589#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
590 struct {
591 int curr; /* highest queued rt task prio */
592#ifdef CONFIG_SMP
593 int next; /* next highest */
594#endif
595 } highest_prio;
596#endif
597#ifdef CONFIG_SMP
598 unsigned long rt_nr_migratory;
599 unsigned long rt_nr_total;
600 int overloaded;
601 struct plist_head pushable_tasks;
602
603#endif /* CONFIG_SMP */
604 int rt_queued;
605
606 int rt_throttled;
607 u64 rt_time;
608 u64 rt_runtime;
609 /* Nests inside the rq lock: */
610 raw_spinlock_t rt_runtime_lock;
611
612#ifdef CONFIG_RT_GROUP_SCHED
613 unsigned long rt_nr_boosted;
614
615 struct rq *rq;
616 struct task_group *tg;
617#endif
618};
619
620static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
621{
622 return rt_rq->rt_queued && rt_rq->rt_nr_running;
623}
624
625/* Deadline class' related fields in a runqueue */
626struct dl_rq {
627 /* runqueue is an rbtree, ordered by deadline */
628 struct rb_root_cached root;
629
630 unsigned long dl_nr_running;
631
632#ifdef CONFIG_SMP
633 /*
634 * Deadline values of the currently executing and the
635 * earliest ready task on this rq. Caching these facilitates
636 * the decision whether or not a ready but not running task
637 * should migrate somewhere else.
638 */
639 struct {
640 u64 curr;
641 u64 next;
642 } earliest_dl;
643
644 unsigned long dl_nr_migratory;
645 int overloaded;
646
647 /*
648 * Tasks on this rq that can be pushed away. They are kept in
649 * an rb-tree, ordered by tasks' deadlines, with caching
650 * of the leftmost (earliest deadline) element.
651 */
652 struct rb_root_cached pushable_dl_tasks_root;
653#else
654 struct dl_bw dl_bw;
655#endif
656 /*
657 * "Active utilization" for this runqueue: increased when a
658 * task wakes up (becomes TASK_RUNNING) and decreased when a
659 * task blocks
660 */
661 u64 running_bw;
662
663 /*
664 * Utilization of the tasks "assigned" to this runqueue (including
665 * the tasks that are in runqueue and the tasks that executed on this
666 * CPU and blocked). Increased when a task moves to this runqueue, and
667 * decreased when the task moves away (migrates, changes scheduling
668 * policy, or terminates).
669 * This is needed to compute the "inactive utilization" for the
670 * runqueue (inactive utilization = this_bw - running_bw).
671 */
672 u64 this_bw;
673 u64 extra_bw;
674
675 /*
676 * Inverse of the fraction of CPU utilization that can be reclaimed
677 * by the GRUB algorithm.
678 */
679 u64 bw_ratio;
680};
681
682#ifdef CONFIG_FAIR_GROUP_SCHED
683/* An entity is a task if it doesn't "own" a runqueue */
684#define entity_is_task(se) (!se->my_q)
685#else
686#define entity_is_task(se) 1
687#endif
688
689#ifdef CONFIG_SMP
690/*
691 * XXX we want to get rid of these helpers and use the full load resolution.
692 */
693static inline long se_weight(struct sched_entity *se)
694{
695 return scale_load_down(se->load.weight);
696}
697
698static inline long se_runnable(struct sched_entity *se)
699{
700 return scale_load_down(se->runnable_weight);
701}
702
703static inline bool sched_asym_prefer(int a, int b)
704{
705 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
706}
707
708struct perf_domain {
709 struct em_perf_domain *em_pd;
710 struct perf_domain *next;
711 struct rcu_head rcu;
712};
713
714/* Scheduling group status flags */
715#define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
716#define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
717
718/*
719 * We add the notion of a root-domain which will be used to define per-domain
720 * variables. Each exclusive cpuset essentially defines an island domain by
721 * fully partitioning the member CPUs from any other cpuset. Whenever a new
722 * exclusive cpuset is created, we also create and attach a new root-domain
723 * object.
724 *
725 */
726struct root_domain {
727 atomic_t refcount;
728 atomic_t rto_count;
729 struct rcu_head rcu;
730 cpumask_var_t span;
731 cpumask_var_t online;
732
733 /*
734 * Indicate pullable load on at least one CPU, e.g:
735 * - More than one runnable task
736 * - Running task is misfit
737 */
738 int overload;
739
740 /* Indicate one or more cpus over-utilized (tipping point) */
741 int overutilized;
742
743 /*
744 * The bit corresponding to a CPU gets set here if such CPU has more
745 * than one runnable -deadline task (as it is below for RT tasks).
746 */
747 cpumask_var_t dlo_mask;
748 atomic_t dlo_count;
749 struct dl_bw dl_bw;
750 struct cpudl cpudl;
751
752#ifdef HAVE_RT_PUSH_IPI
753 /*
754 * For IPI pull requests, loop across the rto_mask.
755 */
756 struct irq_work rto_push_work;
757 raw_spinlock_t rto_lock;
758 /* These are only updated and read within rto_lock */
759 int rto_loop;
760 int rto_cpu;
761 /* These atomics are updated outside of a lock */
762 atomic_t rto_loop_next;
763 atomic_t rto_loop_start;
764#endif
765 /*
766 * The "RT overload" flag: it gets set if a CPU has more than
767 * one runnable RT task.
768 */
769 cpumask_var_t rto_mask;
770 struct cpupri cpupri;
771
772 unsigned long max_cpu_capacity;
773
774 /*
775 * NULL-terminated list of performance domains intersecting with the
776 * CPUs of the rd. Protected by RCU.
777 */
778 struct perf_domain __rcu *pd;
779};
780
781extern struct root_domain def_root_domain;
782extern struct mutex sched_domains_mutex;
783
784extern void init_defrootdomain(void);
785extern int sched_init_domains(const struct cpumask *cpu_map);
786extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
787extern void sched_get_rd(struct root_domain *rd);
788extern void sched_put_rd(struct root_domain *rd);
789
790#ifdef HAVE_RT_PUSH_IPI
791extern void rto_push_irq_work_func(struct irq_work *work);
792#endif
793#endif /* CONFIG_SMP */
794
795#ifdef CONFIG_UCLAMP_TASK
796/*
797 * struct uclamp_bucket - Utilization clamp bucket
798 * @value: utilization clamp value for tasks on this clamp bucket
799 * @tasks: number of RUNNABLE tasks on this clamp bucket
800 *
801 * Keep track of how many tasks are RUNNABLE for a given utilization
802 * clamp value.
803 */
804struct uclamp_bucket {
805 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
806 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
807};
808
809/*
810 * struct uclamp_rq - rq's utilization clamp
811 * @value: currently active clamp values for a rq
812 * @bucket: utilization clamp buckets affecting a rq
813 *
814 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
815 * A clamp value is affecting a rq when there is at least one task RUNNABLE
816 * (or actually running) with that value.
817 *
818 * There are up to UCLAMP_CNT possible different clamp values, currently there
819 * are only two: minimum utilization and maximum utilization.
820 *
821 * All utilization clamping values are MAX aggregated, since:
822 * - for util_min: we want to run the CPU at least at the max of the minimum
823 * utilization required by its currently RUNNABLE tasks.
824 * - for util_max: we want to allow the CPU to run up to the max of the
825 * maximum utilization allowed by its currently RUNNABLE tasks.
826 *
827 * Since on each system we expect only a limited number of different
828 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
829 * the metrics required to compute all the per-rq utilization clamp values.
830 */
831struct uclamp_rq {
832 unsigned int value;
833 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
834};
835#endif /* CONFIG_UCLAMP_TASK */
836
837/*
838 * This is the main, per-CPU runqueue data structure.
839 *
840 * Locking rule: those places that want to lock multiple runqueues
841 * (such as the load balancing or the thread migration code), lock
842 * acquire operations must be ordered by ascending &runqueue.
843 */
844struct rq {
845 /* runqueue lock: */
846 raw_spinlock_t lock;
847
848 /*
849 * nr_running and cpu_load should be in the same cacheline because
850 * remote CPUs use both these fields when doing load calculation.
851 */
852 unsigned int nr_running;
853#ifdef CONFIG_NUMA_BALANCING
854 unsigned int nr_numa_running;
855 unsigned int nr_preferred_running;
856 unsigned int numa_migrate_on;
857#endif
858#ifdef CONFIG_NO_HZ_COMMON
859#ifdef CONFIG_SMP
860 unsigned long last_load_update_tick;
861 unsigned long last_blocked_load_update_tick;
862 unsigned int has_blocked_load;
863#endif /* CONFIG_SMP */
864 unsigned int nohz_tick_stopped;
865 atomic_t nohz_flags;
866#endif /* CONFIG_NO_HZ_COMMON */
867
868 unsigned long nr_load_updates;
869 u64 nr_switches;
870
871#ifdef CONFIG_UCLAMP_TASK
872 /* Utilization clamp values based on CPU's RUNNABLE tasks */
873 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
874 unsigned int uclamp_flags;
875#define UCLAMP_FLAG_IDLE 0x01
876#endif
877
878 struct cfs_rq cfs;
879 struct rt_rq rt;
880 struct dl_rq dl;
881
882#ifdef CONFIG_FAIR_GROUP_SCHED
883 /* list of leaf cfs_rq on this CPU: */
884 struct list_head leaf_cfs_rq_list;
885 struct list_head *tmp_alone_branch;
886#endif /* CONFIG_FAIR_GROUP_SCHED */
887
888 /*
889 * This is part of a global counter where only the total sum
890 * over all CPUs matters. A task can increase this counter on
891 * one CPU and if it got migrated afterwards it may decrease
892 * it on another CPU. Always updated under the runqueue lock:
893 */
894 unsigned long nr_uninterruptible;
895
896 struct task_struct *curr;
897 struct task_struct *idle;
898 struct task_struct *stop;
899 unsigned long next_balance;
900 struct mm_struct *prev_mm;
901
902 unsigned int clock_update_flags;
903 u64 clock;
904 /* Ensure that all clocks are in the same cache line */
905 u64 clock_task ____cacheline_aligned;
906 u64 clock_pelt;
907 unsigned long lost_idle_time;
908
909 atomic_t nr_iowait;
910
911#ifdef CONFIG_SMP
912 struct root_domain *rd;
913 struct sched_domain __rcu *sd;
914
915 unsigned long cpu_capacity;
916 unsigned long cpu_capacity_orig;
917
918 struct callback_head *balance_callback;
919
920 unsigned char idle_balance;
921
922 unsigned long misfit_task_load;
923
924 /* For active balancing */
925 int active_balance;
926 int push_cpu;
927 struct cpu_stop_work active_balance_work;
928
929 /* CPU of this runqueue: */
930 int cpu;
931 int online;
932
933 struct list_head cfs_tasks;
934
935 struct sched_avg avg_rt;
936 struct sched_avg avg_dl;
937#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
938 struct sched_avg avg_irq;
939#endif
940 u64 idle_stamp;
941 u64 avg_idle;
942
943 /* This is used to determine avg_idle's max value */
944 u64 max_idle_balance_cost;
945#endif
946
947#ifdef CONFIG_IRQ_TIME_ACCOUNTING
948 u64 prev_irq_time;
949#endif
950#ifdef CONFIG_PARAVIRT
951 u64 prev_steal_time;
952#endif
953#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
954 u64 prev_steal_time_rq;
955#endif
956
957 /* calc_load related fields */
958 unsigned long calc_load_update;
959 long calc_load_active;
960
961#ifdef CONFIG_SCHED_HRTICK
962#ifdef CONFIG_SMP
963 int hrtick_csd_pending;
964 call_single_data_t hrtick_csd;
965#endif
966 struct hrtimer hrtick_timer;
967#endif
968
969#ifdef CONFIG_SCHEDSTATS
970 /* latency stats */
971 struct sched_info rq_sched_info;
972 unsigned long long rq_cpu_time;
973 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
974
975 /* sys_sched_yield() stats */
976 unsigned int yld_count;
977
978 /* schedule() stats */
979 unsigned int sched_count;
980 unsigned int sched_goidle;
981
982 /* try_to_wake_up() stats */
983 unsigned int ttwu_count;
984 unsigned int ttwu_local;
985#endif
986
987#ifdef CONFIG_SMP
988 struct llist_head wake_list;
989#endif
990
991#ifdef CONFIG_CPU_IDLE
992 /* Must be inspected within a rcu lock section */
993 struct cpuidle_state *idle_state;
994#endif
995};
996
997#ifdef CONFIG_FAIR_GROUP_SCHED
998
999/* CPU runqueue to which this cfs_rq is attached */
1000static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1001{
1002 return cfs_rq->rq;
1003}
1004
1005#else
1006
1007static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1008{
1009 return container_of(cfs_rq, struct rq, cfs);
1010}
1011#endif
1012
1013static inline int cpu_of(struct rq *rq)
1014{
1015#ifdef CONFIG_SMP
1016 return rq->cpu;
1017#else
1018 return 0;
1019#endif
1020}
1021
1022
1023#ifdef CONFIG_SCHED_SMT
1024extern void __update_idle_core(struct rq *rq);
1025
1026static inline void update_idle_core(struct rq *rq)
1027{
1028 if (static_branch_unlikely(&sched_smt_present))
1029 __update_idle_core(rq);
1030}
1031
1032#else
1033static inline void update_idle_core(struct rq *rq) { }
1034#endif
1035
1036DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1037
1038#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1039#define this_rq() this_cpu_ptr(&runqueues)
1040#define task_rq(p) cpu_rq(task_cpu(p))
1041#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1042#define raw_rq() raw_cpu_ptr(&runqueues)
1043
1044extern void update_rq_clock(struct rq *rq);
1045
1046static inline u64 __rq_clock_broken(struct rq *rq)
1047{
1048 return READ_ONCE(rq->clock);
1049}
1050
1051/*
1052 * rq::clock_update_flags bits
1053 *
1054 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1055 * call to __schedule(). This is an optimisation to avoid
1056 * neighbouring rq clock updates.
1057 *
1058 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1059 * in effect and calls to update_rq_clock() are being ignored.
1060 *
1061 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1062 * made to update_rq_clock() since the last time rq::lock was pinned.
1063 *
1064 * If inside of __schedule(), clock_update_flags will have been
1065 * shifted left (a left shift is a cheap operation for the fast path
1066 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1067 *
1068 * if (rq-clock_update_flags >= RQCF_UPDATED)
1069 *
1070 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1071 * one position though, because the next rq_unpin_lock() will shift it
1072 * back.
1073 */
1074#define RQCF_REQ_SKIP 0x01
1075#define RQCF_ACT_SKIP 0x02
1076#define RQCF_UPDATED 0x04
1077
1078static inline void assert_clock_updated(struct rq *rq)
1079{
1080 /*
1081 * The only reason for not seeing a clock update since the
1082 * last rq_pin_lock() is if we're currently skipping updates.
1083 */
1084 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1085}
1086
1087static inline u64 rq_clock(struct rq *rq)
1088{
1089 lockdep_assert_held(&rq->lock);
1090 assert_clock_updated(rq);
1091
1092 return rq->clock;
1093}
1094
1095static inline u64 rq_clock_task(struct rq *rq)
1096{
1097 lockdep_assert_held(&rq->lock);
1098 assert_clock_updated(rq);
1099
1100 return rq->clock_task;
1101}
1102
1103static inline void rq_clock_skip_update(struct rq *rq)
1104{
1105 lockdep_assert_held(&rq->lock);
1106 rq->clock_update_flags |= RQCF_REQ_SKIP;
1107}
1108
1109/*
1110 * See rt task throttling, which is the only time a skip
1111 * request is cancelled.
1112 */
1113static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1114{
1115 lockdep_assert_held(&rq->lock);
1116 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1117}
1118
1119struct rq_flags {
1120 unsigned long flags;
1121 struct pin_cookie cookie;
1122#ifdef CONFIG_SCHED_DEBUG
1123 /*
1124 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1125 * current pin context is stashed here in case it needs to be
1126 * restored in rq_repin_lock().
1127 */
1128 unsigned int clock_update_flags;
1129#endif
1130};
1131
1132static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1133{
1134 rf->cookie = lockdep_pin_lock(&rq->lock);
1135
1136#ifdef CONFIG_SCHED_DEBUG
1137 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1138 rf->clock_update_flags = 0;
1139#endif
1140}
1141
1142static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1143{
1144#ifdef CONFIG_SCHED_DEBUG
1145 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1146 rf->clock_update_flags = RQCF_UPDATED;
1147#endif
1148
1149 lockdep_unpin_lock(&rq->lock, rf->cookie);
1150}
1151
1152static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1153{
1154 lockdep_repin_lock(&rq->lock, rf->cookie);
1155
1156#ifdef CONFIG_SCHED_DEBUG
1157 /*
1158 * Restore the value we stashed in @rf for this pin context.
1159 */
1160 rq->clock_update_flags |= rf->clock_update_flags;
1161#endif
1162}
1163
1164struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1165 __acquires(rq->lock);
1166
1167struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1168 __acquires(p->pi_lock)
1169 __acquires(rq->lock);
1170
1171static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1172 __releases(rq->lock)
1173{
1174 rq_unpin_lock(rq, rf);
1175 raw_spin_unlock(&rq->lock);
1176}
1177
1178static inline void
1179task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1180 __releases(rq->lock)
1181 __releases(p->pi_lock)
1182{
1183 rq_unpin_lock(rq, rf);
1184 raw_spin_unlock(&rq->lock);
1185 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1186}
1187
1188static inline void
1189rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1190 __acquires(rq->lock)
1191{
1192 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1193 rq_pin_lock(rq, rf);
1194}
1195
1196static inline void
1197rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1198 __acquires(rq->lock)
1199{
1200 raw_spin_lock_irq(&rq->lock);
1201 rq_pin_lock(rq, rf);
1202}
1203
1204static inline void
1205rq_lock(struct rq *rq, struct rq_flags *rf)
1206 __acquires(rq->lock)
1207{
1208 raw_spin_lock(&rq->lock);
1209 rq_pin_lock(rq, rf);
1210}
1211
1212static inline void
1213rq_relock(struct rq *rq, struct rq_flags *rf)
1214 __acquires(rq->lock)
1215{
1216 raw_spin_lock(&rq->lock);
1217 rq_repin_lock(rq, rf);
1218}
1219
1220static inline void
1221rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1222 __releases(rq->lock)
1223{
1224 rq_unpin_lock(rq, rf);
1225 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1226}
1227
1228static inline void
1229rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1230 __releases(rq->lock)
1231{
1232 rq_unpin_lock(rq, rf);
1233 raw_spin_unlock_irq(&rq->lock);
1234}
1235
1236static inline void
1237rq_unlock(struct rq *rq, struct rq_flags *rf)
1238 __releases(rq->lock)
1239{
1240 rq_unpin_lock(rq, rf);
1241 raw_spin_unlock(&rq->lock);
1242}
1243
1244static inline struct rq *
1245this_rq_lock_irq(struct rq_flags *rf)
1246 __acquires(rq->lock)
1247{
1248 struct rq *rq;
1249
1250 local_irq_disable();
1251 rq = this_rq();
1252 rq_lock(rq, rf);
1253 return rq;
1254}
1255
1256#ifdef CONFIG_NUMA
1257enum numa_topology_type {
1258 NUMA_DIRECT,
1259 NUMA_GLUELESS_MESH,
1260 NUMA_BACKPLANE,
1261};
1262extern enum numa_topology_type sched_numa_topology_type;
1263extern int sched_max_numa_distance;
1264extern bool find_numa_distance(int distance);
1265#endif
1266
1267#ifdef CONFIG_NUMA
1268extern void sched_init_numa(void);
1269extern void sched_domains_numa_masks_set(unsigned int cpu);
1270extern void sched_domains_numa_masks_clear(unsigned int cpu);
1271#else
1272static inline void sched_init_numa(void) { }
1273static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1274static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1275#endif
1276
1277#ifdef CONFIG_NUMA_BALANCING
1278/* The regions in numa_faults array from task_struct */
1279enum numa_faults_stats {
1280 NUMA_MEM = 0,
1281 NUMA_CPU,
1282 NUMA_MEMBUF,
1283 NUMA_CPUBUF
1284};
1285extern void sched_setnuma(struct task_struct *p, int node);
1286extern int migrate_task_to(struct task_struct *p, int cpu);
1287extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1288 int cpu, int scpu);
1289extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1290#else
1291static inline void
1292init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1293{
1294}
1295#endif /* CONFIG_NUMA_BALANCING */
1296
1297#ifdef CONFIG_SMP
1298
1299static inline void
1300queue_balance_callback(struct rq *rq,
1301 struct callback_head *head,
1302 void (*func)(struct rq *rq))
1303{
1304 lockdep_assert_held(&rq->lock);
1305
1306 if (unlikely(head->next))
1307 return;
1308
1309 head->func = (void (*)(struct callback_head *))func;
1310 head->next = rq->balance_callback;
1311 rq->balance_callback = head;
1312}
1313
1314extern void sched_ttwu_pending(void);
1315
1316#define rcu_dereference_check_sched_domain(p) \
1317 rcu_dereference_check((p), \
1318 lockdep_is_held(&sched_domains_mutex))
1319
1320/*
1321 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1322 * See destroy_sched_domains: call_rcu for details.
1323 *
1324 * The domain tree of any CPU may only be accessed from within
1325 * preempt-disabled sections.
1326 */
1327#define for_each_domain(cpu, __sd) \
1328 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1329 __sd; __sd = __sd->parent)
1330
1331#define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1332
1333/**
1334 * highest_flag_domain - Return highest sched_domain containing flag.
1335 * @cpu: The CPU whose highest level of sched domain is to
1336 * be returned.
1337 * @flag: The flag to check for the highest sched_domain
1338 * for the given CPU.
1339 *
1340 * Returns the highest sched_domain of a CPU which contains the given flag.
1341 */
1342static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1343{
1344 struct sched_domain *sd, *hsd = NULL;
1345
1346 for_each_domain(cpu, sd) {
1347 if (!(sd->flags & flag))
1348 break;
1349 hsd = sd;
1350 }
1351
1352 return hsd;
1353}
1354
1355static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1356{
1357 struct sched_domain *sd;
1358
1359 for_each_domain(cpu, sd) {
1360 if (sd->flags & flag)
1361 break;
1362 }
1363
1364 return sd;
1365}
1366
1367DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1368DECLARE_PER_CPU(int, sd_llc_size);
1369DECLARE_PER_CPU(int, sd_llc_id);
1370DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1371DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1372DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1373DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1374extern struct static_key_false sched_asym_cpucapacity;
1375
1376struct sched_group_capacity {
1377 atomic_t ref;
1378 /*
1379 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1380 * for a single CPU.
1381 */
1382 unsigned long capacity;
1383 unsigned long min_capacity; /* Min per-CPU capacity in group */
1384 unsigned long max_capacity; /* Max per-CPU capacity in group */
1385 unsigned long next_update;
1386 int imbalance; /* XXX unrelated to capacity but shared group state */
1387
1388#ifdef CONFIG_SCHED_DEBUG
1389 int id;
1390#endif
1391
1392 unsigned long cpumask[0]; /* Balance mask */
1393};
1394
1395struct sched_group {
1396 struct sched_group *next; /* Must be a circular list */
1397 atomic_t ref;
1398
1399 unsigned int group_weight;
1400 struct sched_group_capacity *sgc;
1401 int asym_prefer_cpu; /* CPU of highest priority in group */
1402
1403 /*
1404 * The CPUs this group covers.
1405 *
1406 * NOTE: this field is variable length. (Allocated dynamically
1407 * by attaching extra space to the end of the structure,
1408 * depending on how many CPUs the kernel has booted up with)
1409 */
1410 unsigned long cpumask[0];
1411};
1412
1413static inline struct cpumask *sched_group_span(struct sched_group *sg)
1414{
1415 return to_cpumask(sg->cpumask);
1416}
1417
1418/*
1419 * See build_balance_mask().
1420 */
1421static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1422{
1423 return to_cpumask(sg->sgc->cpumask);
1424}
1425
1426/**
1427 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1428 * @group: The group whose first CPU is to be returned.
1429 */
1430static inline unsigned int group_first_cpu(struct sched_group *group)
1431{
1432 return cpumask_first(sched_group_span(group));
1433}
1434
1435extern int group_balance_cpu(struct sched_group *sg);
1436
1437#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1438void register_sched_domain_sysctl(void);
1439void dirty_sched_domain_sysctl(int cpu);
1440void unregister_sched_domain_sysctl(void);
1441#else
1442static inline void register_sched_domain_sysctl(void)
1443{
1444}
1445static inline void dirty_sched_domain_sysctl(int cpu)
1446{
1447}
1448static inline void unregister_sched_domain_sysctl(void)
1449{
1450}
1451#endif
1452
1453#else
1454
1455static inline void sched_ttwu_pending(void) { }
1456
1457#endif /* CONFIG_SMP */
1458
1459#include "stats.h"
1460#include "autogroup.h"
1461
1462#ifdef CONFIG_CGROUP_SCHED
1463
1464/*
1465 * Return the group to which this tasks belongs.
1466 *
1467 * We cannot use task_css() and friends because the cgroup subsystem
1468 * changes that value before the cgroup_subsys::attach() method is called,
1469 * therefore we cannot pin it and might observe the wrong value.
1470 *
1471 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1472 * core changes this before calling sched_move_task().
1473 *
1474 * Instead we use a 'copy' which is updated from sched_move_task() while
1475 * holding both task_struct::pi_lock and rq::lock.
1476 */
1477static inline struct task_group *task_group(struct task_struct *p)
1478{
1479 return p->sched_task_group;
1480}
1481
1482/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1483static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1484{
1485#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1486 struct task_group *tg = task_group(p);
1487#endif
1488
1489#ifdef CONFIG_FAIR_GROUP_SCHED
1490 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1491 p->se.cfs_rq = tg->cfs_rq[cpu];
1492 p->se.parent = tg->se[cpu];
1493#endif
1494
1495#ifdef CONFIG_RT_GROUP_SCHED
1496 p->rt.rt_rq = tg->rt_rq[cpu];
1497 p->rt.parent = tg->rt_se[cpu];
1498#endif
1499}
1500
1501#else /* CONFIG_CGROUP_SCHED */
1502
1503static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1504static inline struct task_group *task_group(struct task_struct *p)
1505{
1506 return NULL;
1507}
1508
1509#endif /* CONFIG_CGROUP_SCHED */
1510
1511static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1512{
1513 set_task_rq(p, cpu);
1514#ifdef CONFIG_SMP
1515 /*
1516 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1517 * successfully executed on another CPU. We must ensure that updates of
1518 * per-task data have been completed by this moment.
1519 */
1520 smp_wmb();
1521#ifdef CONFIG_THREAD_INFO_IN_TASK
1522 WRITE_ONCE(p->cpu, cpu);
1523#else
1524 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1525#endif
1526 p->wake_cpu = cpu;
1527#endif
1528}
1529
1530/*
1531 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1532 */
1533#ifdef CONFIG_SCHED_DEBUG
1534# include <linux/static_key.h>
1535# define const_debug __read_mostly
1536#else
1537# define const_debug const
1538#endif
1539
1540#define SCHED_FEAT(name, enabled) \
1541 __SCHED_FEAT_##name ,
1542
1543enum {
1544#include "features.h"
1545 __SCHED_FEAT_NR,
1546};
1547
1548#undef SCHED_FEAT
1549
1550#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1551
1552/*
1553 * To support run-time toggling of sched features, all the translation units
1554 * (but core.c) reference the sysctl_sched_features defined in core.c.
1555 */
1556extern const_debug unsigned int sysctl_sched_features;
1557
1558#define SCHED_FEAT(name, enabled) \
1559static __always_inline bool static_branch_##name(struct static_key *key) \
1560{ \
1561 return static_key_##enabled(key); \
1562}
1563
1564#include "features.h"
1565#undef SCHED_FEAT
1566
1567extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1568#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1569
1570#else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1571
1572/*
1573 * Each translation unit has its own copy of sysctl_sched_features to allow
1574 * constants propagation at compile time and compiler optimization based on
1575 * features default.
1576 */
1577#define SCHED_FEAT(name, enabled) \
1578 (1UL << __SCHED_FEAT_##name) * enabled |
1579static const_debug __maybe_unused unsigned int sysctl_sched_features =
1580#include "features.h"
1581 0;
1582#undef SCHED_FEAT
1583
1584#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1585
1586#endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1587
1588extern struct static_key_false sched_numa_balancing;
1589extern struct static_key_false sched_schedstats;
1590
1591static inline u64 global_rt_period(void)
1592{
1593 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1594}
1595
1596static inline u64 global_rt_runtime(void)
1597{
1598 if (sysctl_sched_rt_runtime < 0)
1599 return RUNTIME_INF;
1600
1601 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1602}
1603
1604static inline int task_current(struct rq *rq, struct task_struct *p)
1605{
1606 return rq->curr == p;
1607}
1608
1609static inline int task_running(struct rq *rq, struct task_struct *p)
1610{
1611#ifdef CONFIG_SMP
1612 return p->on_cpu;
1613#else
1614 return task_current(rq, p);
1615#endif
1616}
1617
1618static inline int task_on_rq_queued(struct task_struct *p)
1619{
1620 return p->on_rq == TASK_ON_RQ_QUEUED;
1621}
1622
1623static inline int task_on_rq_migrating(struct task_struct *p)
1624{
1625 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1626}
1627
1628/*
1629 * wake flags
1630 */
1631#define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1632#define WF_FORK 0x02 /* Child wakeup after fork */
1633#define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1634
1635/*
1636 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1637 * of tasks with abnormal "nice" values across CPUs the contribution that
1638 * each task makes to its run queue's load is weighted according to its
1639 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1640 * scaled version of the new time slice allocation that they receive on time
1641 * slice expiry etc.
1642 */
1643
1644#define WEIGHT_IDLEPRIO 3
1645#define WMULT_IDLEPRIO 1431655765
1646
1647extern const int sched_prio_to_weight[40];
1648extern const u32 sched_prio_to_wmult[40];
1649
1650/*
1651 * {de,en}queue flags:
1652 *
1653 * DEQUEUE_SLEEP - task is no longer runnable
1654 * ENQUEUE_WAKEUP - task just became runnable
1655 *
1656 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1657 * are in a known state which allows modification. Such pairs
1658 * should preserve as much state as possible.
1659 *
1660 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1661 * in the runqueue.
1662 *
1663 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1664 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1665 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1666 *
1667 */
1668
1669#define DEQUEUE_SLEEP 0x01
1670#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1671#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1672#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1673
1674#define ENQUEUE_WAKEUP 0x01
1675#define ENQUEUE_RESTORE 0x02
1676#define ENQUEUE_MOVE 0x04
1677#define ENQUEUE_NOCLOCK 0x08
1678
1679#define ENQUEUE_HEAD 0x10
1680#define ENQUEUE_REPLENISH 0x20
1681#ifdef CONFIG_SMP
1682#define ENQUEUE_MIGRATED 0x40
1683#else
1684#define ENQUEUE_MIGRATED 0x00
1685#endif
1686
1687#define RETRY_TASK ((void *)-1UL)
1688
1689struct sched_class {
1690 const struct sched_class *next;
1691
1692#ifdef CONFIG_UCLAMP_TASK
1693 int uclamp_enabled;
1694#endif
1695
1696 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1697 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1698 void (*yield_task) (struct rq *rq);
1699 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1700
1701 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1702
1703 /*
1704 * It is the responsibility of the pick_next_task() method that will
1705 * return the next task to call put_prev_task() on the @prev task or
1706 * something equivalent.
1707 *
1708 * May return RETRY_TASK when it finds a higher prio class has runnable
1709 * tasks.
1710 */
1711 struct task_struct * (*pick_next_task)(struct rq *rq,
1712 struct task_struct *prev,
1713 struct rq_flags *rf);
1714 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1715
1716#ifdef CONFIG_SMP
1717 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1718 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1719
1720 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1721
1722 void (*set_cpus_allowed)(struct task_struct *p,
1723 const struct cpumask *newmask);
1724
1725 void (*rq_online)(struct rq *rq);
1726 void (*rq_offline)(struct rq *rq);
1727#endif
1728
1729 void (*set_curr_task)(struct rq *rq);
1730 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1731 void (*task_fork)(struct task_struct *p);
1732 void (*task_dead)(struct task_struct *p);
1733
1734 /*
1735 * The switched_from() call is allowed to drop rq->lock, therefore we
1736 * cannot assume the switched_from/switched_to pair is serliazed by
1737 * rq->lock. They are however serialized by p->pi_lock.
1738 */
1739 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1740 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1741 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1742 int oldprio);
1743
1744 unsigned int (*get_rr_interval)(struct rq *rq,
1745 struct task_struct *task);
1746
1747 void (*update_curr)(struct rq *rq);
1748
1749#define TASK_SET_GROUP 0
1750#define TASK_MOVE_GROUP 1
1751
1752#ifdef CONFIG_FAIR_GROUP_SCHED
1753 void (*task_change_group)(struct task_struct *p, int type);
1754#endif
1755};
1756
1757static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1758{
1759 prev->sched_class->put_prev_task(rq, prev);
1760}
1761
1762static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1763{
1764 curr->sched_class->set_curr_task(rq);
1765}
1766
1767#ifdef CONFIG_SMP
1768#define sched_class_highest (&stop_sched_class)
1769#else
1770#define sched_class_highest (&dl_sched_class)
1771#endif
1772#define for_each_class(class) \
1773 for (class = sched_class_highest; class; class = class->next)
1774
1775extern const struct sched_class stop_sched_class;
1776extern const struct sched_class dl_sched_class;
1777extern const struct sched_class rt_sched_class;
1778extern const struct sched_class fair_sched_class;
1779extern const struct sched_class idle_sched_class;
1780
1781
1782#ifdef CONFIG_SMP
1783
1784extern void update_group_capacity(struct sched_domain *sd, int cpu);
1785
1786extern void trigger_load_balance(struct rq *rq);
1787
1788extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1789
1790#endif
1791
1792#ifdef CONFIG_CPU_IDLE
1793static inline void idle_set_state(struct rq *rq,
1794 struct cpuidle_state *idle_state)
1795{
1796 rq->idle_state = idle_state;
1797}
1798
1799static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1800{
1801 SCHED_WARN_ON(!rcu_read_lock_held());
1802
1803 return rq->idle_state;
1804}
1805#else
1806static inline void idle_set_state(struct rq *rq,
1807 struct cpuidle_state *idle_state)
1808{
1809}
1810
1811static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1812{
1813 return NULL;
1814}
1815#endif
1816
1817extern void schedule_idle(void);
1818
1819extern void sysrq_sched_debug_show(void);
1820extern void sched_init_granularity(void);
1821extern void update_max_interval(void);
1822
1823extern void init_sched_dl_class(void);
1824extern void init_sched_rt_class(void);
1825extern void init_sched_fair_class(void);
1826
1827extern void reweight_task(struct task_struct *p, int prio);
1828
1829extern void resched_curr(struct rq *rq);
1830extern void resched_cpu(int cpu);
1831
1832extern struct rt_bandwidth def_rt_bandwidth;
1833extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1834
1835extern struct dl_bandwidth def_dl_bandwidth;
1836extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1837extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1838extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1839extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1840
1841#define BW_SHIFT 20
1842#define BW_UNIT (1 << BW_SHIFT)
1843#define RATIO_SHIFT 8
1844unsigned long to_ratio(u64 period, u64 runtime);
1845
1846extern void init_entity_runnable_average(struct sched_entity *se);
1847extern void post_init_entity_util_avg(struct task_struct *p);
1848
1849#ifdef CONFIG_NO_HZ_FULL
1850extern bool sched_can_stop_tick(struct rq *rq);
1851extern int __init sched_tick_offload_init(void);
1852
1853/*
1854 * Tick may be needed by tasks in the runqueue depending on their policy and
1855 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1856 * nohz mode if necessary.
1857 */
1858static inline void sched_update_tick_dependency(struct rq *rq)
1859{
1860 int cpu;
1861
1862 if (!tick_nohz_full_enabled())
1863 return;
1864
1865 cpu = cpu_of(rq);
1866
1867 if (!tick_nohz_full_cpu(cpu))
1868 return;
1869
1870 if (sched_can_stop_tick(rq))
1871 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1872 else
1873 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1874}
1875#else
1876static inline int sched_tick_offload_init(void) { return 0; }
1877static inline void sched_update_tick_dependency(struct rq *rq) { }
1878#endif
1879
1880static inline void add_nr_running(struct rq *rq, unsigned count)
1881{
1882 unsigned prev_nr = rq->nr_running;
1883
1884 rq->nr_running = prev_nr + count;
1885
1886#ifdef CONFIG_SMP
1887 if (prev_nr < 2 && rq->nr_running >= 2) {
1888 if (!READ_ONCE(rq->rd->overload))
1889 WRITE_ONCE(rq->rd->overload, 1);
1890 }
1891#endif
1892
1893 sched_update_tick_dependency(rq);
1894}
1895
1896static inline void sub_nr_running(struct rq *rq, unsigned count)
1897{
1898 rq->nr_running -= count;
1899 /* Check if we still need preemption */
1900 sched_update_tick_dependency(rq);
1901}
1902
1903extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1904extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1905
1906extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1907
1908extern const_debug unsigned int sysctl_sched_nr_migrate;
1909extern const_debug unsigned int sysctl_sched_migration_cost;
1910
1911#ifdef CONFIG_SCHED_HRTICK
1912
1913/*
1914 * Use hrtick when:
1915 * - enabled by features
1916 * - hrtimer is actually high res
1917 */
1918static inline int hrtick_enabled(struct rq *rq)
1919{
1920 if (!sched_feat(HRTICK))
1921 return 0;
1922 if (!cpu_active(cpu_of(rq)))
1923 return 0;
1924 return hrtimer_is_hres_active(&rq->hrtick_timer);
1925}
1926
1927void hrtick_start(struct rq *rq, u64 delay);
1928
1929#else
1930
1931static inline int hrtick_enabled(struct rq *rq)
1932{
1933 return 0;
1934}
1935
1936#endif /* CONFIG_SCHED_HRTICK */
1937
1938#ifndef arch_scale_freq_capacity
1939static __always_inline
1940unsigned long arch_scale_freq_capacity(int cpu)
1941{
1942 return SCHED_CAPACITY_SCALE;
1943}
1944#endif
1945
1946#ifdef CONFIG_SMP
1947#ifdef CONFIG_PREEMPT
1948
1949static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1950
1951/*
1952 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1953 * way at the expense of forcing extra atomic operations in all
1954 * invocations. This assures that the double_lock is acquired using the
1955 * same underlying policy as the spinlock_t on this architecture, which
1956 * reduces latency compared to the unfair variant below. However, it
1957 * also adds more overhead and therefore may reduce throughput.
1958 */
1959static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1960 __releases(this_rq->lock)
1961 __acquires(busiest->lock)
1962 __acquires(this_rq->lock)
1963{
1964 raw_spin_unlock(&this_rq->lock);
1965 double_rq_lock(this_rq, busiest);
1966
1967 return 1;
1968}
1969
1970#else
1971/*
1972 * Unfair double_lock_balance: Optimizes throughput at the expense of
1973 * latency by eliminating extra atomic operations when the locks are
1974 * already in proper order on entry. This favors lower CPU-ids and will
1975 * grant the double lock to lower CPUs over higher ids under contention,
1976 * regardless of entry order into the function.
1977 */
1978static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1979 __releases(this_rq->lock)
1980 __acquires(busiest->lock)
1981 __acquires(this_rq->lock)
1982{
1983 int ret = 0;
1984
1985 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1986 if (busiest < this_rq) {
1987 raw_spin_unlock(&this_rq->lock);
1988 raw_spin_lock(&busiest->lock);
1989 raw_spin_lock_nested(&this_rq->lock,
1990 SINGLE_DEPTH_NESTING);
1991 ret = 1;
1992 } else
1993 raw_spin_lock_nested(&busiest->lock,
1994 SINGLE_DEPTH_NESTING);
1995 }
1996 return ret;
1997}
1998
1999#endif /* CONFIG_PREEMPT */
2000
2001/*
2002 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2003 */
2004static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2005{
2006 if (unlikely(!irqs_disabled())) {
2007 /* printk() doesn't work well under rq->lock */
2008 raw_spin_unlock(&this_rq->lock);
2009 BUG_ON(1);
2010 }
2011
2012 return _double_lock_balance(this_rq, busiest);
2013}
2014
2015static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2016 __releases(busiest->lock)
2017{
2018 raw_spin_unlock(&busiest->lock);
2019 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2020}
2021
2022static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2023{
2024 if (l1 > l2)
2025 swap(l1, l2);
2026
2027 spin_lock(l1);
2028 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2029}
2030
2031static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2032{
2033 if (l1 > l2)
2034 swap(l1, l2);
2035
2036 spin_lock_irq(l1);
2037 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2038}
2039
2040static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2041{
2042 if (l1 > l2)
2043 swap(l1, l2);
2044
2045 raw_spin_lock(l1);
2046 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2047}
2048
2049/*
2050 * double_rq_lock - safely lock two runqueues
2051 *
2052 * Note this does not disable interrupts like task_rq_lock,
2053 * you need to do so manually before calling.
2054 */
2055static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2056 __acquires(rq1->lock)
2057 __acquires(rq2->lock)
2058{
2059 BUG_ON(!irqs_disabled());
2060 if (rq1 == rq2) {
2061 raw_spin_lock(&rq1->lock);
2062 __acquire(rq2->lock); /* Fake it out ;) */
2063 } else {
2064 if (rq1 < rq2) {
2065 raw_spin_lock(&rq1->lock);
2066 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2067 } else {
2068 raw_spin_lock(&rq2->lock);
2069 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2070 }
2071 }
2072}
2073
2074/*
2075 * double_rq_unlock - safely unlock two runqueues
2076 *
2077 * Note this does not restore interrupts like task_rq_unlock,
2078 * you need to do so manually after calling.
2079 */
2080static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2081 __releases(rq1->lock)
2082 __releases(rq2->lock)
2083{
2084 raw_spin_unlock(&rq1->lock);
2085 if (rq1 != rq2)
2086 raw_spin_unlock(&rq2->lock);
2087 else
2088 __release(rq2->lock);
2089}
2090
2091extern void set_rq_online (struct rq *rq);
2092extern void set_rq_offline(struct rq *rq);
2093extern bool sched_smp_initialized;
2094
2095#else /* CONFIG_SMP */
2096
2097/*
2098 * double_rq_lock - safely lock two runqueues
2099 *
2100 * Note this does not disable interrupts like task_rq_lock,
2101 * you need to do so manually before calling.
2102 */
2103static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2104 __acquires(rq1->lock)
2105 __acquires(rq2->lock)
2106{
2107 BUG_ON(!irqs_disabled());
2108 BUG_ON(rq1 != rq2);
2109 raw_spin_lock(&rq1->lock);
2110 __acquire(rq2->lock); /* Fake it out ;) */
2111}
2112
2113/*
2114 * double_rq_unlock - safely unlock two runqueues
2115 *
2116 * Note this does not restore interrupts like task_rq_unlock,
2117 * you need to do so manually after calling.
2118 */
2119static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2120 __releases(rq1->lock)
2121 __releases(rq2->lock)
2122{
2123 BUG_ON(rq1 != rq2);
2124 raw_spin_unlock(&rq1->lock);
2125 __release(rq2->lock);
2126}
2127
2128#endif
2129
2130extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2131extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2132
2133#ifdef CONFIG_SCHED_DEBUG
2134extern bool sched_debug_enabled;
2135
2136extern void print_cfs_stats(struct seq_file *m, int cpu);
2137extern void print_rt_stats(struct seq_file *m, int cpu);
2138extern void print_dl_stats(struct seq_file *m, int cpu);
2139extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2140extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2141extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2142#ifdef CONFIG_NUMA_BALANCING
2143extern void
2144show_numa_stats(struct task_struct *p, struct seq_file *m);
2145extern void
2146print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2147 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2148#endif /* CONFIG_NUMA_BALANCING */
2149#endif /* CONFIG_SCHED_DEBUG */
2150
2151extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2152extern void init_rt_rq(struct rt_rq *rt_rq);
2153extern void init_dl_rq(struct dl_rq *dl_rq);
2154
2155extern void cfs_bandwidth_usage_inc(void);
2156extern void cfs_bandwidth_usage_dec(void);
2157
2158#ifdef CONFIG_NO_HZ_COMMON
2159#define NOHZ_BALANCE_KICK_BIT 0
2160#define NOHZ_STATS_KICK_BIT 1
2161
2162#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2163#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2164
2165#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2166
2167#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2168
2169extern void nohz_balance_exit_idle(struct rq *rq);
2170#else
2171static inline void nohz_balance_exit_idle(struct rq *rq) { }
2172#endif
2173
2174
2175#ifdef CONFIG_SMP
2176static inline
2177void __dl_update(struct dl_bw *dl_b, s64 bw)
2178{
2179 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2180 int i;
2181
2182 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2183 "sched RCU must be held");
2184 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2185 struct rq *rq = cpu_rq(i);
2186
2187 rq->dl.extra_bw += bw;
2188 }
2189}
2190#else
2191static inline
2192void __dl_update(struct dl_bw *dl_b, s64 bw)
2193{
2194 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2195
2196 dl->extra_bw += bw;
2197}
2198#endif
2199
2200
2201#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2202struct irqtime {
2203 u64 total;
2204 u64 tick_delta;
2205 u64 irq_start_time;
2206 struct u64_stats_sync sync;
2207};
2208
2209DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2210
2211/*
2212 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2213 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2214 * and never move forward.
2215 */
2216static inline u64 irq_time_read(int cpu)
2217{
2218 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2219 unsigned int seq;
2220 u64 total;
2221
2222 do {
2223 seq = __u64_stats_fetch_begin(&irqtime->sync);
2224 total = irqtime->total;
2225 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2226
2227 return total;
2228}
2229#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2230
2231#ifdef CONFIG_CPU_FREQ
2232DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2233
2234/**
2235 * cpufreq_update_util - Take a note about CPU utilization changes.
2236 * @rq: Runqueue to carry out the update for.
2237 * @flags: Update reason flags.
2238 *
2239 * This function is called by the scheduler on the CPU whose utilization is
2240 * being updated.
2241 *
2242 * It can only be called from RCU-sched read-side critical sections.
2243 *
2244 * The way cpufreq is currently arranged requires it to evaluate the CPU
2245 * performance state (frequency/voltage) on a regular basis to prevent it from
2246 * being stuck in a completely inadequate performance level for too long.
2247 * That is not guaranteed to happen if the updates are only triggered from CFS
2248 * and DL, though, because they may not be coming in if only RT tasks are
2249 * active all the time (or there are RT tasks only).
2250 *
2251 * As a workaround for that issue, this function is called periodically by the
2252 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2253 * but that really is a band-aid. Going forward it should be replaced with
2254 * solutions targeted more specifically at RT tasks.
2255 */
2256static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2257{
2258 struct update_util_data *data;
2259
2260 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2261 cpu_of(rq)));
2262 if (data)
2263 data->func(data, rq_clock(rq), flags);
2264}
2265#else
2266static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2267#endif /* CONFIG_CPU_FREQ */
2268
2269#ifdef CONFIG_UCLAMP_TASK
2270unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id);
2271
2272static __always_inline
2273unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2274 struct task_struct *p)
2275{
2276 unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2277 unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2278
2279 if (p) {
2280 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2281 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2282 }
2283
2284 /*
2285 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2286 * RUNNABLE tasks with _different_ clamps, we can end up with an
2287 * inversion. Fix it now when the clamps are applied.
2288 */
2289 if (unlikely(min_util >= max_util))
2290 return min_util;
2291
2292 return clamp(util, min_util, max_util);
2293}
2294
2295static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2296{
2297 return uclamp_util_with(rq, util, NULL);
2298}
2299#else /* CONFIG_UCLAMP_TASK */
2300static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2301 struct task_struct *p)
2302{
2303 return util;
2304}
2305static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2306{
2307 return util;
2308}
2309#endif /* CONFIG_UCLAMP_TASK */
2310
2311#ifdef arch_scale_freq_capacity
2312# ifndef arch_scale_freq_invariant
2313# define arch_scale_freq_invariant() true
2314# endif
2315#else
2316# define arch_scale_freq_invariant() false
2317#endif
2318
2319#ifdef CONFIG_SMP
2320static inline unsigned long capacity_orig_of(int cpu)
2321{
2322 return cpu_rq(cpu)->cpu_capacity_orig;
2323}
2324#endif
2325
2326/**
2327 * enum schedutil_type - CPU utilization type
2328 * @FREQUENCY_UTIL: Utilization used to select frequency
2329 * @ENERGY_UTIL: Utilization used during energy calculation
2330 *
2331 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2332 * need to be aggregated differently depending on the usage made of them. This
2333 * enum is used within schedutil_freq_util() to differentiate the types of
2334 * utilization expected by the callers, and adjust the aggregation accordingly.
2335 */
2336enum schedutil_type {
2337 FREQUENCY_UTIL,
2338 ENERGY_UTIL,
2339};
2340
2341#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2342
2343unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2344 unsigned long max, enum schedutil_type type,
2345 struct task_struct *p);
2346
2347static inline unsigned long cpu_bw_dl(struct rq *rq)
2348{
2349 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2350}
2351
2352static inline unsigned long cpu_util_dl(struct rq *rq)
2353{
2354 return READ_ONCE(rq->avg_dl.util_avg);
2355}
2356
2357static inline unsigned long cpu_util_cfs(struct rq *rq)
2358{
2359 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2360
2361 if (sched_feat(UTIL_EST)) {
2362 util = max_t(unsigned long, util,
2363 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2364 }
2365
2366 return util;
2367}
2368
2369static inline unsigned long cpu_util_rt(struct rq *rq)
2370{
2371 return READ_ONCE(rq->avg_rt.util_avg);
2372}
2373#else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2374static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2375 unsigned long max, enum schedutil_type type,
2376 struct task_struct *p)
2377{
2378 return 0;
2379}
2380#endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2381
2382#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2383static inline unsigned long cpu_util_irq(struct rq *rq)
2384{
2385 return rq->avg_irq.util_avg;
2386}
2387
2388static inline
2389unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2390{
2391 util *= (max - irq);
2392 util /= max;
2393
2394 return util;
2395
2396}
2397#else
2398static inline unsigned long cpu_util_irq(struct rq *rq)
2399{
2400 return 0;
2401}
2402
2403static inline
2404unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2405{
2406 return util;
2407}
2408#endif
2409
2410#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2411
2412#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2413
2414DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2415
2416static inline bool sched_energy_enabled(void)
2417{
2418 return static_branch_unlikely(&sched_energy_present);
2419}
2420
2421#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2422
2423#define perf_domain_span(pd) NULL
2424static inline bool sched_energy_enabled(void) { return false; }
2425
2426#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
Linux 6.x block layer and io_uring tree(s)