xref: /linux/kernel/sched/sched.h (revision e60e1ee60630cafef5e430c2ae364877e061d980)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #include <linux/sched.h>
4 #include <linux/sched/autogroup.h>
5 #include <linux/sched/sysctl.h>
6 #include <linux/sched/topology.h>
7 #include <linux/sched/rt.h>
8 #include <linux/sched/deadline.h>
9 #include <linux/sched/clock.h>
10 #include <linux/sched/wake_q.h>
11 #include <linux/sched/signal.h>
12 #include <linux/sched/numa_balancing.h>
13 #include <linux/sched/mm.h>
14 #include <linux/sched/cpufreq.h>
15 #include <linux/sched/stat.h>
16 #include <linux/sched/nohz.h>
17 #include <linux/sched/debug.h>
18 #include <linux/sched/hotplug.h>
19 #include <linux/sched/task.h>
20 #include <linux/sched/task_stack.h>
21 #include <linux/sched/cputime.h>
22 #include <linux/sched/init.h>
23 
24 #include <linux/u64_stats_sync.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/binfmts.h>
27 #include <linux/mutex.h>
28 #include <linux/spinlock.h>
29 #include <linux/stop_machine.h>
30 #include <linux/irq_work.h>
31 #include <linux/tick.h>
32 #include <linux/slab.h>
33 #include <linux/cgroup.h>
34 
35 #ifdef CONFIG_PARAVIRT
36 #include <asm/paravirt.h>
37 #endif
38 
39 #include "cpupri.h"
40 #include "cpudeadline.h"
41 
42 #ifdef CONFIG_SCHED_DEBUG
43 # define SCHED_WARN_ON(x)	WARN_ONCE(x, #x)
44 #else
45 # define SCHED_WARN_ON(x)	({ (void)(x), 0; })
46 #endif
47 
48 struct rq;
49 struct cpuidle_state;
50 
51 /* task_struct::on_rq states: */
52 #define TASK_ON_RQ_QUEUED	1
53 #define TASK_ON_RQ_MIGRATING	2
54 
55 extern __read_mostly int scheduler_running;
56 
57 extern unsigned long calc_load_update;
58 extern atomic_long_t calc_load_tasks;
59 
60 extern void calc_global_load_tick(struct rq *this_rq);
61 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
62 
63 #ifdef CONFIG_SMP
64 extern void cpu_load_update_active(struct rq *this_rq);
65 #else
66 static inline void cpu_load_update_active(struct rq *this_rq) { }
67 #endif
68 
69 /*
70  * Helpers for converting nanosecond timing to jiffy resolution
71  */
72 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
73 
74 /*
75  * Increase resolution of nice-level calculations for 64-bit architectures.
76  * The extra resolution improves shares distribution and load balancing of
77  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
78  * hierarchies, especially on larger systems. This is not a user-visible change
79  * and does not change the user-interface for setting shares/weights.
80  *
81  * We increase resolution only if we have enough bits to allow this increased
82  * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
83  * pretty high and the returns do not justify the increased costs.
84  *
85  * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
86  * increase coverage and consistency always enable it on 64bit platforms.
87  */
88 #ifdef CONFIG_64BIT
89 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
90 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
91 # define scale_load_down(w)	((w) >> SCHED_FIXEDPOINT_SHIFT)
92 #else
93 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
94 # define scale_load(w)		(w)
95 # define scale_load_down(w)	(w)
96 #endif
97 
98 /*
99  * Task weight (visible to users) and its load (invisible to users) have
100  * independent resolution, but they should be well calibrated. We use
101  * scale_load() and scale_load_down(w) to convert between them. The
102  * following must be true:
103  *
104  *  scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
105  *
106  */
107 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
108 
109 /*
110  * Single value that decides SCHED_DEADLINE internal math precision.
111  * 10 -> just above 1us
112  * 9  -> just above 0.5us
113  */
114 #define DL_SCALE (10)
115 
116 /*
117  * These are the 'tuning knobs' of the scheduler:
118  */
119 
120 /*
121  * single value that denotes runtime == period, ie unlimited time.
122  */
123 #define RUNTIME_INF	((u64)~0ULL)
124 
125 static inline int idle_policy(int policy)
126 {
127 	return policy == SCHED_IDLE;
128 }
129 static inline int fair_policy(int policy)
130 {
131 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
132 }
133 
134 static inline int rt_policy(int policy)
135 {
136 	return policy == SCHED_FIFO || policy == SCHED_RR;
137 }
138 
139 static inline int dl_policy(int policy)
140 {
141 	return policy == SCHED_DEADLINE;
142 }
143 static inline bool valid_policy(int policy)
144 {
145 	return idle_policy(policy) || fair_policy(policy) ||
146 		rt_policy(policy) || dl_policy(policy);
147 }
148 
149 static inline int task_has_rt_policy(struct task_struct *p)
150 {
151 	return rt_policy(p->policy);
152 }
153 
154 static inline int task_has_dl_policy(struct task_struct *p)
155 {
156 	return dl_policy(p->policy);
157 }
158 
159 /*
160  * Tells if entity @a should preempt entity @b.
161  */
162 static inline bool
163 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
164 {
165 	return dl_time_before(a->deadline, b->deadline);
166 }
167 
168 /*
169  * This is the priority-queue data structure of the RT scheduling class:
170  */
171 struct rt_prio_array {
172 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
173 	struct list_head queue[MAX_RT_PRIO];
174 };
175 
176 struct rt_bandwidth {
177 	/* nests inside the rq lock: */
178 	raw_spinlock_t		rt_runtime_lock;
179 	ktime_t			rt_period;
180 	u64			rt_runtime;
181 	struct hrtimer		rt_period_timer;
182 	unsigned int		rt_period_active;
183 };
184 
185 void __dl_clear_params(struct task_struct *p);
186 
187 /*
188  * To keep the bandwidth of -deadline tasks and groups under control
189  * we need some place where:
190  *  - store the maximum -deadline bandwidth of the system (the group);
191  *  - cache the fraction of that bandwidth that is currently allocated.
192  *
193  * This is all done in the data structure below. It is similar to the
194  * one used for RT-throttling (rt_bandwidth), with the main difference
195  * that, since here we are only interested in admission control, we
196  * do not decrease any runtime while the group "executes", neither we
197  * need a timer to replenish it.
198  *
199  * With respect to SMP, the bandwidth is given on a per-CPU basis,
200  * meaning that:
201  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
202  *  - dl_total_bw array contains, in the i-eth element, the currently
203  *    allocated bandwidth on the i-eth CPU.
204  * Moreover, groups consume bandwidth on each CPU, while tasks only
205  * consume bandwidth on the CPU they're running on.
206  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
207  * that will be shown the next time the proc or cgroup controls will
208  * be red. It on its turn can be changed by writing on its own
209  * control.
210  */
211 struct dl_bandwidth {
212 	raw_spinlock_t dl_runtime_lock;
213 	u64 dl_runtime;
214 	u64 dl_period;
215 };
216 
217 static inline int dl_bandwidth_enabled(void)
218 {
219 	return sysctl_sched_rt_runtime >= 0;
220 }
221 
222 struct dl_bw {
223 	raw_spinlock_t lock;
224 	u64 bw, total_bw;
225 };
226 
227 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
228 
229 static inline
230 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
231 {
232 	dl_b->total_bw -= tsk_bw;
233 	__dl_update(dl_b, (s32)tsk_bw / cpus);
234 }
235 
236 static inline
237 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
238 {
239 	dl_b->total_bw += tsk_bw;
240 	__dl_update(dl_b, -((s32)tsk_bw / cpus));
241 }
242 
243 static inline
244 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
245 {
246 	return dl_b->bw != -1 &&
247 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
248 }
249 
250 void dl_change_utilization(struct task_struct *p, u64 new_bw);
251 extern void init_dl_bw(struct dl_bw *dl_b);
252 extern int sched_dl_global_validate(void);
253 extern void sched_dl_do_global(void);
254 extern int sched_dl_overflow(struct task_struct *p, int policy,
255 			     const struct sched_attr *attr);
256 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
257 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
258 extern bool __checkparam_dl(const struct sched_attr *attr);
259 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
260 extern int dl_task_can_attach(struct task_struct *p,
261 			      const struct cpumask *cs_cpus_allowed);
262 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur,
263 					const struct cpumask *trial);
264 extern bool dl_cpu_busy(unsigned int cpu);
265 
266 #ifdef CONFIG_CGROUP_SCHED
267 
268 #include <linux/cgroup.h>
269 
270 struct cfs_rq;
271 struct rt_rq;
272 
273 extern struct list_head task_groups;
274 
275 struct cfs_bandwidth {
276 #ifdef CONFIG_CFS_BANDWIDTH
277 	raw_spinlock_t lock;
278 	ktime_t period;
279 	u64 quota, runtime;
280 	s64 hierarchical_quota;
281 	u64 runtime_expires;
282 
283 	int idle, period_active;
284 	struct hrtimer period_timer, slack_timer;
285 	struct list_head throttled_cfs_rq;
286 
287 	/* statistics */
288 	int nr_periods, nr_throttled;
289 	u64 throttled_time;
290 #endif
291 };
292 
293 /* task group related information */
294 struct task_group {
295 	struct cgroup_subsys_state css;
296 
297 #ifdef CONFIG_FAIR_GROUP_SCHED
298 	/* schedulable entities of this group on each cpu */
299 	struct sched_entity **se;
300 	/* runqueue "owned" by this group on each cpu */
301 	struct cfs_rq **cfs_rq;
302 	unsigned long shares;
303 
304 #ifdef	CONFIG_SMP
305 	/*
306 	 * load_avg can be heavily contended at clock tick time, so put
307 	 * it in its own cacheline separated from the fields above which
308 	 * will also be accessed at each tick.
309 	 */
310 	atomic_long_t load_avg ____cacheline_aligned;
311 #endif
312 #endif
313 
314 #ifdef CONFIG_RT_GROUP_SCHED
315 	struct sched_rt_entity **rt_se;
316 	struct rt_rq **rt_rq;
317 
318 	struct rt_bandwidth rt_bandwidth;
319 #endif
320 
321 	struct rcu_head rcu;
322 	struct list_head list;
323 
324 	struct task_group *parent;
325 	struct list_head siblings;
326 	struct list_head children;
327 
328 #ifdef CONFIG_SCHED_AUTOGROUP
329 	struct autogroup *autogroup;
330 #endif
331 
332 	struct cfs_bandwidth cfs_bandwidth;
333 };
334 
335 #ifdef CONFIG_FAIR_GROUP_SCHED
336 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
337 
338 /*
339  * A weight of 0 or 1 can cause arithmetics problems.
340  * A weight of a cfs_rq is the sum of weights of which entities
341  * are queued on this cfs_rq, so a weight of a entity should not be
342  * too large, so as the shares value of a task group.
343  * (The default weight is 1024 - so there's no practical
344  *  limitation from this.)
345  */
346 #define MIN_SHARES	(1UL <<  1)
347 #define MAX_SHARES	(1UL << 18)
348 #endif
349 
350 typedef int (*tg_visitor)(struct task_group *, void *);
351 
352 extern int walk_tg_tree_from(struct task_group *from,
353 			     tg_visitor down, tg_visitor up, void *data);
354 
355 /*
356  * Iterate the full tree, calling @down when first entering a node and @up when
357  * leaving it for the final time.
358  *
359  * Caller must hold rcu_lock or sufficient equivalent.
360  */
361 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
362 {
363 	return walk_tg_tree_from(&root_task_group, down, up, data);
364 }
365 
366 extern int tg_nop(struct task_group *tg, void *data);
367 
368 extern void free_fair_sched_group(struct task_group *tg);
369 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
370 extern void online_fair_sched_group(struct task_group *tg);
371 extern void unregister_fair_sched_group(struct task_group *tg);
372 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
373 			struct sched_entity *se, int cpu,
374 			struct sched_entity *parent);
375 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
376 
377 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
378 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
379 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
380 
381 extern void free_rt_sched_group(struct task_group *tg);
382 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
383 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
384 		struct sched_rt_entity *rt_se, int cpu,
385 		struct sched_rt_entity *parent);
386 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
387 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
388 extern long sched_group_rt_runtime(struct task_group *tg);
389 extern long sched_group_rt_period(struct task_group *tg);
390 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
391 
392 extern struct task_group *sched_create_group(struct task_group *parent);
393 extern void sched_online_group(struct task_group *tg,
394 			       struct task_group *parent);
395 extern void sched_destroy_group(struct task_group *tg);
396 extern void sched_offline_group(struct task_group *tg);
397 
398 extern void sched_move_task(struct task_struct *tsk);
399 
400 #ifdef CONFIG_FAIR_GROUP_SCHED
401 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
402 
403 #ifdef CONFIG_SMP
404 extern void set_task_rq_fair(struct sched_entity *se,
405 			     struct cfs_rq *prev, struct cfs_rq *next);
406 #else /* !CONFIG_SMP */
407 static inline void set_task_rq_fair(struct sched_entity *se,
408 			     struct cfs_rq *prev, struct cfs_rq *next) { }
409 #endif /* CONFIG_SMP */
410 #endif /* CONFIG_FAIR_GROUP_SCHED */
411 
412 #else /* CONFIG_CGROUP_SCHED */
413 
414 struct cfs_bandwidth { };
415 
416 #endif	/* CONFIG_CGROUP_SCHED */
417 
418 /* CFS-related fields in a runqueue */
419 struct cfs_rq {
420 	struct load_weight load;
421 	unsigned long runnable_weight;
422 	unsigned int nr_running, h_nr_running;
423 
424 	u64 exec_clock;
425 	u64 min_vruntime;
426 #ifndef CONFIG_64BIT
427 	u64 min_vruntime_copy;
428 #endif
429 
430 	struct rb_root_cached tasks_timeline;
431 
432 	/*
433 	 * 'curr' points to currently running entity on this cfs_rq.
434 	 * It is set to NULL otherwise (i.e when none are currently running).
435 	 */
436 	struct sched_entity *curr, *next, *last, *skip;
437 
438 #ifdef	CONFIG_SCHED_DEBUG
439 	unsigned int nr_spread_over;
440 #endif
441 
442 #ifdef CONFIG_SMP
443 	/*
444 	 * CFS load tracking
445 	 */
446 	struct sched_avg avg;
447 #ifndef CONFIG_64BIT
448 	u64 load_last_update_time_copy;
449 #endif
450 	struct {
451 		raw_spinlock_t	lock ____cacheline_aligned;
452 		int		nr;
453 		unsigned long	load_avg;
454 		unsigned long	util_avg;
455 		unsigned long	runnable_sum;
456 	} removed;
457 
458 #ifdef CONFIG_FAIR_GROUP_SCHED
459 	unsigned long tg_load_avg_contrib;
460 	long propagate;
461 	long prop_runnable_sum;
462 
463 	/*
464 	 *   h_load = weight * f(tg)
465 	 *
466 	 * Where f(tg) is the recursive weight fraction assigned to
467 	 * this group.
468 	 */
469 	unsigned long h_load;
470 	u64 last_h_load_update;
471 	struct sched_entity *h_load_next;
472 #endif /* CONFIG_FAIR_GROUP_SCHED */
473 #endif /* CONFIG_SMP */
474 
475 #ifdef CONFIG_FAIR_GROUP_SCHED
476 	struct rq *rq;	/* cpu runqueue to which this cfs_rq is attached */
477 
478 	/*
479 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
480 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
481 	 * (like users, containers etc.)
482 	 *
483 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
484 	 * list is used during load balance.
485 	 */
486 	int on_list;
487 	struct list_head leaf_cfs_rq_list;
488 	struct task_group *tg;	/* group that "owns" this runqueue */
489 
490 #ifdef CONFIG_CFS_BANDWIDTH
491 	int runtime_enabled;
492 	u64 runtime_expires;
493 	s64 runtime_remaining;
494 
495 	u64 throttled_clock, throttled_clock_task;
496 	u64 throttled_clock_task_time;
497 	int throttled, throttle_count;
498 	struct list_head throttled_list;
499 #endif /* CONFIG_CFS_BANDWIDTH */
500 #endif /* CONFIG_FAIR_GROUP_SCHED */
501 };
502 
503 static inline int rt_bandwidth_enabled(void)
504 {
505 	return sysctl_sched_rt_runtime >= 0;
506 }
507 
508 /* RT IPI pull logic requires IRQ_WORK */
509 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
510 # define HAVE_RT_PUSH_IPI
511 #endif
512 
513 /* Real-Time classes' related field in a runqueue: */
514 struct rt_rq {
515 	struct rt_prio_array active;
516 	unsigned int rt_nr_running;
517 	unsigned int rr_nr_running;
518 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
519 	struct {
520 		int curr; /* highest queued rt task prio */
521 #ifdef CONFIG_SMP
522 		int next; /* next highest */
523 #endif
524 	} highest_prio;
525 #endif
526 #ifdef CONFIG_SMP
527 	unsigned long rt_nr_migratory;
528 	unsigned long rt_nr_total;
529 	int overloaded;
530 	struct plist_head pushable_tasks;
531 #endif /* CONFIG_SMP */
532 	int rt_queued;
533 
534 	int rt_throttled;
535 	u64 rt_time;
536 	u64 rt_runtime;
537 	/* Nests inside the rq lock: */
538 	raw_spinlock_t rt_runtime_lock;
539 
540 #ifdef CONFIG_RT_GROUP_SCHED
541 	unsigned long rt_nr_boosted;
542 
543 	struct rq *rq;
544 	struct task_group *tg;
545 #endif
546 };
547 
548 /* Deadline class' related fields in a runqueue */
549 struct dl_rq {
550 	/* runqueue is an rbtree, ordered by deadline */
551 	struct rb_root_cached root;
552 
553 	unsigned long dl_nr_running;
554 
555 #ifdef CONFIG_SMP
556 	/*
557 	 * Deadline values of the currently executing and the
558 	 * earliest ready task on this rq. Caching these facilitates
559 	 * the decision wether or not a ready but not running task
560 	 * should migrate somewhere else.
561 	 */
562 	struct {
563 		u64 curr;
564 		u64 next;
565 	} earliest_dl;
566 
567 	unsigned long dl_nr_migratory;
568 	int overloaded;
569 
570 	/*
571 	 * Tasks on this rq that can be pushed away. They are kept in
572 	 * an rb-tree, ordered by tasks' deadlines, with caching
573 	 * of the leftmost (earliest deadline) element.
574 	 */
575 	struct rb_root_cached pushable_dl_tasks_root;
576 #else
577 	struct dl_bw dl_bw;
578 #endif
579 	/*
580 	 * "Active utilization" for this runqueue: increased when a
581 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
582 	 * task blocks
583 	 */
584 	u64 running_bw;
585 
586 	/*
587 	 * Utilization of the tasks "assigned" to this runqueue (including
588 	 * the tasks that are in runqueue and the tasks that executed on this
589 	 * CPU and blocked). Increased when a task moves to this runqueue, and
590 	 * decreased when the task moves away (migrates, changes scheduling
591 	 * policy, or terminates).
592 	 * This is needed to compute the "inactive utilization" for the
593 	 * runqueue (inactive utilization = this_bw - running_bw).
594 	 */
595 	u64 this_bw;
596 	u64 extra_bw;
597 
598 	/*
599 	 * Inverse of the fraction of CPU utilization that can be reclaimed
600 	 * by the GRUB algorithm.
601 	 */
602 	u64 bw_ratio;
603 };
604 
605 #ifdef CONFIG_SMP
606 
607 static inline bool sched_asym_prefer(int a, int b)
608 {
609 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
610 }
611 
612 /*
613  * We add the notion of a root-domain which will be used to define per-domain
614  * variables. Each exclusive cpuset essentially defines an island domain by
615  * fully partitioning the member cpus from any other cpuset. Whenever a new
616  * exclusive cpuset is created, we also create and attach a new root-domain
617  * object.
618  *
619  */
620 struct root_domain {
621 	atomic_t refcount;
622 	atomic_t rto_count;
623 	struct rcu_head rcu;
624 	cpumask_var_t span;
625 	cpumask_var_t online;
626 
627 	/* Indicate more than one runnable task for any CPU */
628 	bool overload;
629 
630 	/*
631 	 * The bit corresponding to a CPU gets set here if such CPU has more
632 	 * than one runnable -deadline task (as it is below for RT tasks).
633 	 */
634 	cpumask_var_t dlo_mask;
635 	atomic_t dlo_count;
636 	struct dl_bw dl_bw;
637 	struct cpudl cpudl;
638 
639 #ifdef HAVE_RT_PUSH_IPI
640 	/*
641 	 * For IPI pull requests, loop across the rto_mask.
642 	 */
643 	struct irq_work rto_push_work;
644 	raw_spinlock_t rto_lock;
645 	/* These are only updated and read within rto_lock */
646 	int rto_loop;
647 	int rto_cpu;
648 	/* These atomics are updated outside of a lock */
649 	atomic_t rto_loop_next;
650 	atomic_t rto_loop_start;
651 #endif
652 	/*
653 	 * The "RT overload" flag: it gets set if a CPU has more than
654 	 * one runnable RT task.
655 	 */
656 	cpumask_var_t rto_mask;
657 	struct cpupri cpupri;
658 
659 	unsigned long max_cpu_capacity;
660 };
661 
662 extern struct root_domain def_root_domain;
663 extern struct mutex sched_domains_mutex;
664 
665 extern void init_defrootdomain(void);
666 extern int sched_init_domains(const struct cpumask *cpu_map);
667 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
668 
669 #ifdef HAVE_RT_PUSH_IPI
670 extern void rto_push_irq_work_func(struct irq_work *work);
671 #endif
672 #endif /* CONFIG_SMP */
673 
674 /*
675  * This is the main, per-CPU runqueue data structure.
676  *
677  * Locking rule: those places that want to lock multiple runqueues
678  * (such as the load balancing or the thread migration code), lock
679  * acquire operations must be ordered by ascending &runqueue.
680  */
681 struct rq {
682 	/* runqueue lock: */
683 	raw_spinlock_t lock;
684 
685 	/*
686 	 * nr_running and cpu_load should be in the same cacheline because
687 	 * remote CPUs use both these fields when doing load calculation.
688 	 */
689 	unsigned int nr_running;
690 #ifdef CONFIG_NUMA_BALANCING
691 	unsigned int nr_numa_running;
692 	unsigned int nr_preferred_running;
693 #endif
694 	#define CPU_LOAD_IDX_MAX 5
695 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
696 #ifdef CONFIG_NO_HZ_COMMON
697 #ifdef CONFIG_SMP
698 	unsigned long last_load_update_tick;
699 #endif /* CONFIG_SMP */
700 	unsigned long nohz_flags;
701 #endif /* CONFIG_NO_HZ_COMMON */
702 #ifdef CONFIG_NO_HZ_FULL
703 	unsigned long last_sched_tick;
704 #endif
705 	/* capture load from *all* tasks on this cpu: */
706 	struct load_weight load;
707 	unsigned long nr_load_updates;
708 	u64 nr_switches;
709 
710 	struct cfs_rq cfs;
711 	struct rt_rq rt;
712 	struct dl_rq dl;
713 
714 #ifdef CONFIG_FAIR_GROUP_SCHED
715 	/* list of leaf cfs_rq on this cpu: */
716 	struct list_head leaf_cfs_rq_list;
717 	struct list_head *tmp_alone_branch;
718 #endif /* CONFIG_FAIR_GROUP_SCHED */
719 
720 	/*
721 	 * This is part of a global counter where only the total sum
722 	 * over all CPUs matters. A task can increase this counter on
723 	 * one CPU and if it got migrated afterwards it may decrease
724 	 * it on another CPU. Always updated under the runqueue lock:
725 	 */
726 	unsigned long nr_uninterruptible;
727 
728 	struct task_struct *curr, *idle, *stop;
729 	unsigned long next_balance;
730 	struct mm_struct *prev_mm;
731 
732 	unsigned int clock_update_flags;
733 	u64 clock;
734 	u64 clock_task;
735 
736 	atomic_t nr_iowait;
737 
738 #ifdef CONFIG_SMP
739 	struct root_domain *rd;
740 	struct sched_domain *sd;
741 
742 	unsigned long cpu_capacity;
743 	unsigned long cpu_capacity_orig;
744 
745 	struct callback_head *balance_callback;
746 
747 	unsigned char idle_balance;
748 	/* For active balancing */
749 	int active_balance;
750 	int push_cpu;
751 	struct cpu_stop_work active_balance_work;
752 	/* cpu of this runqueue: */
753 	int cpu;
754 	int online;
755 
756 	struct list_head cfs_tasks;
757 
758 	u64 rt_avg;
759 	u64 age_stamp;
760 	u64 idle_stamp;
761 	u64 avg_idle;
762 
763 	/* This is used to determine avg_idle's max value */
764 	u64 max_idle_balance_cost;
765 #endif
766 
767 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
768 	u64 prev_irq_time;
769 #endif
770 #ifdef CONFIG_PARAVIRT
771 	u64 prev_steal_time;
772 #endif
773 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
774 	u64 prev_steal_time_rq;
775 #endif
776 
777 	/* calc_load related fields */
778 	unsigned long calc_load_update;
779 	long calc_load_active;
780 
781 #ifdef CONFIG_SCHED_HRTICK
782 #ifdef CONFIG_SMP
783 	int hrtick_csd_pending;
784 	call_single_data_t hrtick_csd;
785 #endif
786 	struct hrtimer hrtick_timer;
787 #endif
788 
789 #ifdef CONFIG_SCHEDSTATS
790 	/* latency stats */
791 	struct sched_info rq_sched_info;
792 	unsigned long long rq_cpu_time;
793 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
794 
795 	/* sys_sched_yield() stats */
796 	unsigned int yld_count;
797 
798 	/* schedule() stats */
799 	unsigned int sched_count;
800 	unsigned int sched_goidle;
801 
802 	/* try_to_wake_up() stats */
803 	unsigned int ttwu_count;
804 	unsigned int ttwu_local;
805 #endif
806 
807 #ifdef CONFIG_SMP
808 	struct llist_head wake_list;
809 #endif
810 
811 #ifdef CONFIG_CPU_IDLE
812 	/* Must be inspected within a rcu lock section */
813 	struct cpuidle_state *idle_state;
814 #endif
815 };
816 
817 static inline int cpu_of(struct rq *rq)
818 {
819 #ifdef CONFIG_SMP
820 	return rq->cpu;
821 #else
822 	return 0;
823 #endif
824 }
825 
826 
827 #ifdef CONFIG_SCHED_SMT
828 
829 extern struct static_key_false sched_smt_present;
830 
831 extern void __update_idle_core(struct rq *rq);
832 
833 static inline void update_idle_core(struct rq *rq)
834 {
835 	if (static_branch_unlikely(&sched_smt_present))
836 		__update_idle_core(rq);
837 }
838 
839 #else
840 static inline void update_idle_core(struct rq *rq) { }
841 #endif
842 
843 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
844 
845 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
846 #define this_rq()		this_cpu_ptr(&runqueues)
847 #define task_rq(p)		cpu_rq(task_cpu(p))
848 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
849 #define raw_rq()		raw_cpu_ptr(&runqueues)
850 
851 static inline u64 __rq_clock_broken(struct rq *rq)
852 {
853 	return READ_ONCE(rq->clock);
854 }
855 
856 /*
857  * rq::clock_update_flags bits
858  *
859  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
860  *  call to __schedule(). This is an optimisation to avoid
861  *  neighbouring rq clock updates.
862  *
863  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
864  *  in effect and calls to update_rq_clock() are being ignored.
865  *
866  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
867  *  made to update_rq_clock() since the last time rq::lock was pinned.
868  *
869  * If inside of __schedule(), clock_update_flags will have been
870  * shifted left (a left shift is a cheap operation for the fast path
871  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
872  *
873  *	if (rq-clock_update_flags >= RQCF_UPDATED)
874  *
875  * to check if %RQCF_UPADTED is set. It'll never be shifted more than
876  * one position though, because the next rq_unpin_lock() will shift it
877  * back.
878  */
879 #define RQCF_REQ_SKIP	0x01
880 #define RQCF_ACT_SKIP	0x02
881 #define RQCF_UPDATED	0x04
882 
883 static inline void assert_clock_updated(struct rq *rq)
884 {
885 	/*
886 	 * The only reason for not seeing a clock update since the
887 	 * last rq_pin_lock() is if we're currently skipping updates.
888 	 */
889 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
890 }
891 
892 static inline u64 rq_clock(struct rq *rq)
893 {
894 	lockdep_assert_held(&rq->lock);
895 	assert_clock_updated(rq);
896 
897 	return rq->clock;
898 }
899 
900 static inline u64 rq_clock_task(struct rq *rq)
901 {
902 	lockdep_assert_held(&rq->lock);
903 	assert_clock_updated(rq);
904 
905 	return rq->clock_task;
906 }
907 
908 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
909 {
910 	lockdep_assert_held(&rq->lock);
911 	if (skip)
912 		rq->clock_update_flags |= RQCF_REQ_SKIP;
913 	else
914 		rq->clock_update_flags &= ~RQCF_REQ_SKIP;
915 }
916 
917 struct rq_flags {
918 	unsigned long flags;
919 	struct pin_cookie cookie;
920 #ifdef CONFIG_SCHED_DEBUG
921 	/*
922 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
923 	 * current pin context is stashed here in case it needs to be
924 	 * restored in rq_repin_lock().
925 	 */
926 	unsigned int clock_update_flags;
927 #endif
928 };
929 
930 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
931 {
932 	rf->cookie = lockdep_pin_lock(&rq->lock);
933 
934 #ifdef CONFIG_SCHED_DEBUG
935 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
936 	rf->clock_update_flags = 0;
937 #endif
938 }
939 
940 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
941 {
942 #ifdef CONFIG_SCHED_DEBUG
943 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
944 		rf->clock_update_flags = RQCF_UPDATED;
945 #endif
946 
947 	lockdep_unpin_lock(&rq->lock, rf->cookie);
948 }
949 
950 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
951 {
952 	lockdep_repin_lock(&rq->lock, rf->cookie);
953 
954 #ifdef CONFIG_SCHED_DEBUG
955 	/*
956 	 * Restore the value we stashed in @rf for this pin context.
957 	 */
958 	rq->clock_update_flags |= rf->clock_update_flags;
959 #endif
960 }
961 
962 #ifdef CONFIG_NUMA
963 enum numa_topology_type {
964 	NUMA_DIRECT,
965 	NUMA_GLUELESS_MESH,
966 	NUMA_BACKPLANE,
967 };
968 extern enum numa_topology_type sched_numa_topology_type;
969 extern int sched_max_numa_distance;
970 extern bool find_numa_distance(int distance);
971 #endif
972 
973 #ifdef CONFIG_NUMA
974 extern void sched_init_numa(void);
975 extern void sched_domains_numa_masks_set(unsigned int cpu);
976 extern void sched_domains_numa_masks_clear(unsigned int cpu);
977 #else
978 static inline void sched_init_numa(void) { }
979 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
980 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
981 #endif
982 
983 #ifdef CONFIG_NUMA_BALANCING
984 /* The regions in numa_faults array from task_struct */
985 enum numa_faults_stats {
986 	NUMA_MEM = 0,
987 	NUMA_CPU,
988 	NUMA_MEMBUF,
989 	NUMA_CPUBUF
990 };
991 extern void sched_setnuma(struct task_struct *p, int node);
992 extern int migrate_task_to(struct task_struct *p, int cpu);
993 extern int migrate_swap(struct task_struct *, struct task_struct *);
994 #endif /* CONFIG_NUMA_BALANCING */
995 
996 #ifdef CONFIG_SMP
997 
998 static inline void
999 queue_balance_callback(struct rq *rq,
1000 		       struct callback_head *head,
1001 		       void (*func)(struct rq *rq))
1002 {
1003 	lockdep_assert_held(&rq->lock);
1004 
1005 	if (unlikely(head->next))
1006 		return;
1007 
1008 	head->func = (void (*)(struct callback_head *))func;
1009 	head->next = rq->balance_callback;
1010 	rq->balance_callback = head;
1011 }
1012 
1013 extern void sched_ttwu_pending(void);
1014 
1015 #define rcu_dereference_check_sched_domain(p) \
1016 	rcu_dereference_check((p), \
1017 			      lockdep_is_held(&sched_domains_mutex))
1018 
1019 /*
1020  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1021  * See detach_destroy_domains: synchronize_sched for details.
1022  *
1023  * The domain tree of any CPU may only be accessed from within
1024  * preempt-disabled sections.
1025  */
1026 #define for_each_domain(cpu, __sd) \
1027 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1028 			__sd; __sd = __sd->parent)
1029 
1030 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1031 
1032 /**
1033  * highest_flag_domain - Return highest sched_domain containing flag.
1034  * @cpu:	The cpu whose highest level of sched domain is to
1035  *		be returned.
1036  * @flag:	The flag to check for the highest sched_domain
1037  *		for the given cpu.
1038  *
1039  * Returns the highest sched_domain of a cpu which contains the given flag.
1040  */
1041 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1042 {
1043 	struct sched_domain *sd, *hsd = NULL;
1044 
1045 	for_each_domain(cpu, sd) {
1046 		if (!(sd->flags & flag))
1047 			break;
1048 		hsd = sd;
1049 	}
1050 
1051 	return hsd;
1052 }
1053 
1054 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1055 {
1056 	struct sched_domain *sd;
1057 
1058 	for_each_domain(cpu, sd) {
1059 		if (sd->flags & flag)
1060 			break;
1061 	}
1062 
1063 	return sd;
1064 }
1065 
1066 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1067 DECLARE_PER_CPU(int, sd_llc_size);
1068 DECLARE_PER_CPU(int, sd_llc_id);
1069 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1070 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1071 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1072 
1073 struct sched_group_capacity {
1074 	atomic_t ref;
1075 	/*
1076 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1077 	 * for a single CPU.
1078 	 */
1079 	unsigned long capacity;
1080 	unsigned long min_capacity; /* Min per-CPU capacity in group */
1081 	unsigned long next_update;
1082 	int imbalance; /* XXX unrelated to capacity but shared group state */
1083 
1084 #ifdef CONFIG_SCHED_DEBUG
1085 	int id;
1086 #endif
1087 
1088 	unsigned long cpumask[0]; /* balance mask */
1089 };
1090 
1091 struct sched_group {
1092 	struct sched_group *next;	/* Must be a circular list */
1093 	atomic_t ref;
1094 
1095 	unsigned int group_weight;
1096 	struct sched_group_capacity *sgc;
1097 	int asym_prefer_cpu;		/* cpu of highest priority in group */
1098 
1099 	/*
1100 	 * The CPUs this group covers.
1101 	 *
1102 	 * NOTE: this field is variable length. (Allocated dynamically
1103 	 * by attaching extra space to the end of the structure,
1104 	 * depending on how many CPUs the kernel has booted up with)
1105 	 */
1106 	unsigned long cpumask[0];
1107 };
1108 
1109 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1110 {
1111 	return to_cpumask(sg->cpumask);
1112 }
1113 
1114 /*
1115  * See build_balance_mask().
1116  */
1117 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1118 {
1119 	return to_cpumask(sg->sgc->cpumask);
1120 }
1121 
1122 /**
1123  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1124  * @group: The group whose first cpu is to be returned.
1125  */
1126 static inline unsigned int group_first_cpu(struct sched_group *group)
1127 {
1128 	return cpumask_first(sched_group_span(group));
1129 }
1130 
1131 extern int group_balance_cpu(struct sched_group *sg);
1132 
1133 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1134 void register_sched_domain_sysctl(void);
1135 void dirty_sched_domain_sysctl(int cpu);
1136 void unregister_sched_domain_sysctl(void);
1137 #else
1138 static inline void register_sched_domain_sysctl(void)
1139 {
1140 }
1141 static inline void dirty_sched_domain_sysctl(int cpu)
1142 {
1143 }
1144 static inline void unregister_sched_domain_sysctl(void)
1145 {
1146 }
1147 #endif
1148 
1149 #else
1150 
1151 static inline void sched_ttwu_pending(void) { }
1152 
1153 #endif /* CONFIG_SMP */
1154 
1155 #include "stats.h"
1156 #include "autogroup.h"
1157 
1158 #ifdef CONFIG_CGROUP_SCHED
1159 
1160 /*
1161  * Return the group to which this tasks belongs.
1162  *
1163  * We cannot use task_css() and friends because the cgroup subsystem
1164  * changes that value before the cgroup_subsys::attach() method is called,
1165  * therefore we cannot pin it and might observe the wrong value.
1166  *
1167  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1168  * core changes this before calling sched_move_task().
1169  *
1170  * Instead we use a 'copy' which is updated from sched_move_task() while
1171  * holding both task_struct::pi_lock and rq::lock.
1172  */
1173 static inline struct task_group *task_group(struct task_struct *p)
1174 {
1175 	return p->sched_task_group;
1176 }
1177 
1178 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1179 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1180 {
1181 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1182 	struct task_group *tg = task_group(p);
1183 #endif
1184 
1185 #ifdef CONFIG_FAIR_GROUP_SCHED
1186 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1187 	p->se.cfs_rq = tg->cfs_rq[cpu];
1188 	p->se.parent = tg->se[cpu];
1189 #endif
1190 
1191 #ifdef CONFIG_RT_GROUP_SCHED
1192 	p->rt.rt_rq  = tg->rt_rq[cpu];
1193 	p->rt.parent = tg->rt_se[cpu];
1194 #endif
1195 }
1196 
1197 #else /* CONFIG_CGROUP_SCHED */
1198 
1199 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1200 static inline struct task_group *task_group(struct task_struct *p)
1201 {
1202 	return NULL;
1203 }
1204 
1205 #endif /* CONFIG_CGROUP_SCHED */
1206 
1207 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1208 {
1209 	set_task_rq(p, cpu);
1210 #ifdef CONFIG_SMP
1211 	/*
1212 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1213 	 * successfuly executed on another CPU. We must ensure that updates of
1214 	 * per-task data have been completed by this moment.
1215 	 */
1216 	smp_wmb();
1217 #ifdef CONFIG_THREAD_INFO_IN_TASK
1218 	p->cpu = cpu;
1219 #else
1220 	task_thread_info(p)->cpu = cpu;
1221 #endif
1222 	p->wake_cpu = cpu;
1223 #endif
1224 }
1225 
1226 /*
1227  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1228  */
1229 #ifdef CONFIG_SCHED_DEBUG
1230 # include <linux/static_key.h>
1231 # define const_debug __read_mostly
1232 #else
1233 # define const_debug const
1234 #endif
1235 
1236 #define SCHED_FEAT(name, enabled)	\
1237 	__SCHED_FEAT_##name ,
1238 
1239 enum {
1240 #include "features.h"
1241 	__SCHED_FEAT_NR,
1242 };
1243 
1244 #undef SCHED_FEAT
1245 
1246 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1247 
1248 /*
1249  * To support run-time toggling of sched features, all the translation units
1250  * (but core.c) reference the sysctl_sched_features defined in core.c.
1251  */
1252 extern const_debug unsigned int sysctl_sched_features;
1253 
1254 #define SCHED_FEAT(name, enabled)					\
1255 static __always_inline bool static_branch_##name(struct static_key *key) \
1256 {									\
1257 	return static_key_##enabled(key);				\
1258 }
1259 
1260 #include "features.h"
1261 #undef SCHED_FEAT
1262 
1263 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1264 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1265 
1266 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1267 
1268 /*
1269  * Each translation unit has its own copy of sysctl_sched_features to allow
1270  * constants propagation at compile time and compiler optimization based on
1271  * features default.
1272  */
1273 #define SCHED_FEAT(name, enabled)	\
1274 	(1UL << __SCHED_FEAT_##name) * enabled |
1275 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1276 #include "features.h"
1277 	0;
1278 #undef SCHED_FEAT
1279 
1280 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1281 
1282 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1283 
1284 extern struct static_key_false sched_numa_balancing;
1285 extern struct static_key_false sched_schedstats;
1286 
1287 static inline u64 global_rt_period(void)
1288 {
1289 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1290 }
1291 
1292 static inline u64 global_rt_runtime(void)
1293 {
1294 	if (sysctl_sched_rt_runtime < 0)
1295 		return RUNTIME_INF;
1296 
1297 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1298 }
1299 
1300 static inline int task_current(struct rq *rq, struct task_struct *p)
1301 {
1302 	return rq->curr == p;
1303 }
1304 
1305 static inline int task_running(struct rq *rq, struct task_struct *p)
1306 {
1307 #ifdef CONFIG_SMP
1308 	return p->on_cpu;
1309 #else
1310 	return task_current(rq, p);
1311 #endif
1312 }
1313 
1314 static inline int task_on_rq_queued(struct task_struct *p)
1315 {
1316 	return p->on_rq == TASK_ON_RQ_QUEUED;
1317 }
1318 
1319 static inline int task_on_rq_migrating(struct task_struct *p)
1320 {
1321 	return p->on_rq == TASK_ON_RQ_MIGRATING;
1322 }
1323 
1324 #ifndef prepare_arch_switch
1325 # define prepare_arch_switch(next)	do { } while (0)
1326 #endif
1327 #ifndef finish_arch_post_lock_switch
1328 # define finish_arch_post_lock_switch()	do { } while (0)
1329 #endif
1330 
1331 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1332 {
1333 #ifdef CONFIG_SMP
1334 	/*
1335 	 * We can optimise this out completely for !SMP, because the
1336 	 * SMP rebalancing from interrupt is the only thing that cares
1337 	 * here.
1338 	 */
1339 	next->on_cpu = 1;
1340 #endif
1341 }
1342 
1343 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1344 {
1345 #ifdef CONFIG_SMP
1346 	/*
1347 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1348 	 * We must ensure this doesn't happen until the switch is completely
1349 	 * finished.
1350 	 *
1351 	 * In particular, the load of prev->state in finish_task_switch() must
1352 	 * happen before this.
1353 	 *
1354 	 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1355 	 */
1356 	smp_store_release(&prev->on_cpu, 0);
1357 #endif
1358 #ifdef CONFIG_DEBUG_SPINLOCK
1359 	/* this is a valid case when another task releases the spinlock */
1360 	rq->lock.owner = current;
1361 #endif
1362 	/*
1363 	 * If we are tracking spinlock dependencies then we have to
1364 	 * fix up the runqueue lock - which gets 'carried over' from
1365 	 * prev into current:
1366 	 */
1367 	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1368 
1369 	raw_spin_unlock_irq(&rq->lock);
1370 }
1371 
1372 /*
1373  * wake flags
1374  */
1375 #define WF_SYNC		0x01		/* waker goes to sleep after wakeup */
1376 #define WF_FORK		0x02		/* child wakeup after fork */
1377 #define WF_MIGRATED	0x4		/* internal use, task got migrated */
1378 
1379 /*
1380  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1381  * of tasks with abnormal "nice" values across CPUs the contribution that
1382  * each task makes to its run queue's load is weighted according to its
1383  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1384  * scaled version of the new time slice allocation that they receive on time
1385  * slice expiry etc.
1386  */
1387 
1388 #define WEIGHT_IDLEPRIO                3
1389 #define WMULT_IDLEPRIO         1431655765
1390 
1391 extern const int sched_prio_to_weight[40];
1392 extern const u32 sched_prio_to_wmult[40];
1393 
1394 /*
1395  * {de,en}queue flags:
1396  *
1397  * DEQUEUE_SLEEP  - task is no longer runnable
1398  * ENQUEUE_WAKEUP - task just became runnable
1399  *
1400  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1401  *                are in a known state which allows modification. Such pairs
1402  *                should preserve as much state as possible.
1403  *
1404  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1405  *        in the runqueue.
1406  *
1407  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
1408  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1409  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
1410  *
1411  */
1412 
1413 #define DEQUEUE_SLEEP		0x01
1414 #define DEQUEUE_SAVE		0x02 /* matches ENQUEUE_RESTORE */
1415 #define DEQUEUE_MOVE		0x04 /* matches ENQUEUE_MOVE */
1416 #define DEQUEUE_NOCLOCK		0x08 /* matches ENQUEUE_NOCLOCK */
1417 
1418 #define ENQUEUE_WAKEUP		0x01
1419 #define ENQUEUE_RESTORE		0x02
1420 #define ENQUEUE_MOVE		0x04
1421 #define ENQUEUE_NOCLOCK		0x08
1422 
1423 #define ENQUEUE_HEAD		0x10
1424 #define ENQUEUE_REPLENISH	0x20
1425 #ifdef CONFIG_SMP
1426 #define ENQUEUE_MIGRATED	0x40
1427 #else
1428 #define ENQUEUE_MIGRATED	0x00
1429 #endif
1430 
1431 #define RETRY_TASK		((void *)-1UL)
1432 
1433 struct sched_class {
1434 	const struct sched_class *next;
1435 
1436 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1437 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1438 	void (*yield_task) (struct rq *rq);
1439 	bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1440 
1441 	void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1442 
1443 	/*
1444 	 * It is the responsibility of the pick_next_task() method that will
1445 	 * return the next task to call put_prev_task() on the @prev task or
1446 	 * something equivalent.
1447 	 *
1448 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1449 	 * tasks.
1450 	 */
1451 	struct task_struct * (*pick_next_task) (struct rq *rq,
1452 						struct task_struct *prev,
1453 						struct rq_flags *rf);
1454 	void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1455 
1456 #ifdef CONFIG_SMP
1457 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1458 	void (*migrate_task_rq)(struct task_struct *p);
1459 
1460 	void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1461 
1462 	void (*set_cpus_allowed)(struct task_struct *p,
1463 				 const struct cpumask *newmask);
1464 
1465 	void (*rq_online)(struct rq *rq);
1466 	void (*rq_offline)(struct rq *rq);
1467 #endif
1468 
1469 	void (*set_curr_task) (struct rq *rq);
1470 	void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1471 	void (*task_fork) (struct task_struct *p);
1472 	void (*task_dead) (struct task_struct *p);
1473 
1474 	/*
1475 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1476 	 * cannot assume the switched_from/switched_to pair is serliazed by
1477 	 * rq->lock. They are however serialized by p->pi_lock.
1478 	 */
1479 	void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1480 	void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1481 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1482 			     int oldprio);
1483 
1484 	unsigned int (*get_rr_interval) (struct rq *rq,
1485 					 struct task_struct *task);
1486 
1487 	void (*update_curr) (struct rq *rq);
1488 
1489 #define TASK_SET_GROUP  0
1490 #define TASK_MOVE_GROUP	1
1491 
1492 #ifdef CONFIG_FAIR_GROUP_SCHED
1493 	void (*task_change_group) (struct task_struct *p, int type);
1494 #endif
1495 };
1496 
1497 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1498 {
1499 	prev->sched_class->put_prev_task(rq, prev);
1500 }
1501 
1502 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1503 {
1504 	curr->sched_class->set_curr_task(rq);
1505 }
1506 
1507 #ifdef CONFIG_SMP
1508 #define sched_class_highest (&stop_sched_class)
1509 #else
1510 #define sched_class_highest (&dl_sched_class)
1511 #endif
1512 #define for_each_class(class) \
1513    for (class = sched_class_highest; class; class = class->next)
1514 
1515 extern const struct sched_class stop_sched_class;
1516 extern const struct sched_class dl_sched_class;
1517 extern const struct sched_class rt_sched_class;
1518 extern const struct sched_class fair_sched_class;
1519 extern const struct sched_class idle_sched_class;
1520 
1521 
1522 #ifdef CONFIG_SMP
1523 
1524 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1525 
1526 extern void trigger_load_balance(struct rq *rq);
1527 
1528 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1529 
1530 #endif
1531 
1532 #ifdef CONFIG_CPU_IDLE
1533 static inline void idle_set_state(struct rq *rq,
1534 				  struct cpuidle_state *idle_state)
1535 {
1536 	rq->idle_state = idle_state;
1537 }
1538 
1539 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1540 {
1541 	SCHED_WARN_ON(!rcu_read_lock_held());
1542 	return rq->idle_state;
1543 }
1544 #else
1545 static inline void idle_set_state(struct rq *rq,
1546 				  struct cpuidle_state *idle_state)
1547 {
1548 }
1549 
1550 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1551 {
1552 	return NULL;
1553 }
1554 #endif
1555 
1556 extern void schedule_idle(void);
1557 
1558 extern void sysrq_sched_debug_show(void);
1559 extern void sched_init_granularity(void);
1560 extern void update_max_interval(void);
1561 
1562 extern void init_sched_dl_class(void);
1563 extern void init_sched_rt_class(void);
1564 extern void init_sched_fair_class(void);
1565 
1566 extern void reweight_task(struct task_struct *p, int prio);
1567 
1568 extern void resched_curr(struct rq *rq);
1569 extern void resched_cpu(int cpu);
1570 
1571 extern struct rt_bandwidth def_rt_bandwidth;
1572 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1573 
1574 extern struct dl_bandwidth def_dl_bandwidth;
1575 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1576 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1577 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1578 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1579 
1580 #define BW_SHIFT	20
1581 #define BW_UNIT		(1 << BW_SHIFT)
1582 #define RATIO_SHIFT	8
1583 unsigned long to_ratio(u64 period, u64 runtime);
1584 
1585 extern void init_entity_runnable_average(struct sched_entity *se);
1586 extern void post_init_entity_util_avg(struct sched_entity *se);
1587 
1588 #ifdef CONFIG_NO_HZ_FULL
1589 extern bool sched_can_stop_tick(struct rq *rq);
1590 
1591 /*
1592  * Tick may be needed by tasks in the runqueue depending on their policy and
1593  * requirements. If tick is needed, lets send the target an IPI to kick it out of
1594  * nohz mode if necessary.
1595  */
1596 static inline void sched_update_tick_dependency(struct rq *rq)
1597 {
1598 	int cpu;
1599 
1600 	if (!tick_nohz_full_enabled())
1601 		return;
1602 
1603 	cpu = cpu_of(rq);
1604 
1605 	if (!tick_nohz_full_cpu(cpu))
1606 		return;
1607 
1608 	if (sched_can_stop_tick(rq))
1609 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1610 	else
1611 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1612 }
1613 #else
1614 static inline void sched_update_tick_dependency(struct rq *rq) { }
1615 #endif
1616 
1617 static inline void add_nr_running(struct rq *rq, unsigned count)
1618 {
1619 	unsigned prev_nr = rq->nr_running;
1620 
1621 	rq->nr_running = prev_nr + count;
1622 
1623 	if (prev_nr < 2 && rq->nr_running >= 2) {
1624 #ifdef CONFIG_SMP
1625 		if (!rq->rd->overload)
1626 			rq->rd->overload = true;
1627 #endif
1628 	}
1629 
1630 	sched_update_tick_dependency(rq);
1631 }
1632 
1633 static inline void sub_nr_running(struct rq *rq, unsigned count)
1634 {
1635 	rq->nr_running -= count;
1636 	/* Check if we still need preemption */
1637 	sched_update_tick_dependency(rq);
1638 }
1639 
1640 static inline void rq_last_tick_reset(struct rq *rq)
1641 {
1642 #ifdef CONFIG_NO_HZ_FULL
1643 	rq->last_sched_tick = jiffies;
1644 #endif
1645 }
1646 
1647 extern void update_rq_clock(struct rq *rq);
1648 
1649 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1650 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1651 
1652 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1653 
1654 extern const_debug unsigned int sysctl_sched_time_avg;
1655 extern const_debug unsigned int sysctl_sched_nr_migrate;
1656 extern const_debug unsigned int sysctl_sched_migration_cost;
1657 
1658 static inline u64 sched_avg_period(void)
1659 {
1660 	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1661 }
1662 
1663 #ifdef CONFIG_SCHED_HRTICK
1664 
1665 /*
1666  * Use hrtick when:
1667  *  - enabled by features
1668  *  - hrtimer is actually high res
1669  */
1670 static inline int hrtick_enabled(struct rq *rq)
1671 {
1672 	if (!sched_feat(HRTICK))
1673 		return 0;
1674 	if (!cpu_active(cpu_of(rq)))
1675 		return 0;
1676 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1677 }
1678 
1679 void hrtick_start(struct rq *rq, u64 delay);
1680 
1681 #else
1682 
1683 static inline int hrtick_enabled(struct rq *rq)
1684 {
1685 	return 0;
1686 }
1687 
1688 #endif /* CONFIG_SCHED_HRTICK */
1689 
1690 #ifdef CONFIG_SMP
1691 extern void sched_avg_update(struct rq *rq);
1692 
1693 #ifndef arch_scale_freq_capacity
1694 static __always_inline
1695 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1696 {
1697 	return SCHED_CAPACITY_SCALE;
1698 }
1699 #endif
1700 
1701 #ifndef arch_scale_cpu_capacity
1702 static __always_inline
1703 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1704 {
1705 	if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1706 		return sd->smt_gain / sd->span_weight;
1707 
1708 	return SCHED_CAPACITY_SCALE;
1709 }
1710 #endif
1711 
1712 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1713 {
1714 	rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1715 	sched_avg_update(rq);
1716 }
1717 #else
1718 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1719 static inline void sched_avg_update(struct rq *rq) { }
1720 #endif
1721 
1722 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1723 	__acquires(rq->lock);
1724 
1725 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1726 	__acquires(p->pi_lock)
1727 	__acquires(rq->lock);
1728 
1729 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1730 	__releases(rq->lock)
1731 {
1732 	rq_unpin_lock(rq, rf);
1733 	raw_spin_unlock(&rq->lock);
1734 }
1735 
1736 static inline void
1737 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1738 	__releases(rq->lock)
1739 	__releases(p->pi_lock)
1740 {
1741 	rq_unpin_lock(rq, rf);
1742 	raw_spin_unlock(&rq->lock);
1743 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1744 }
1745 
1746 static inline void
1747 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1748 	__acquires(rq->lock)
1749 {
1750 	raw_spin_lock_irqsave(&rq->lock, rf->flags);
1751 	rq_pin_lock(rq, rf);
1752 }
1753 
1754 static inline void
1755 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1756 	__acquires(rq->lock)
1757 {
1758 	raw_spin_lock_irq(&rq->lock);
1759 	rq_pin_lock(rq, rf);
1760 }
1761 
1762 static inline void
1763 rq_lock(struct rq *rq, struct rq_flags *rf)
1764 	__acquires(rq->lock)
1765 {
1766 	raw_spin_lock(&rq->lock);
1767 	rq_pin_lock(rq, rf);
1768 }
1769 
1770 static inline void
1771 rq_relock(struct rq *rq, struct rq_flags *rf)
1772 	__acquires(rq->lock)
1773 {
1774 	raw_spin_lock(&rq->lock);
1775 	rq_repin_lock(rq, rf);
1776 }
1777 
1778 static inline void
1779 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1780 	__releases(rq->lock)
1781 {
1782 	rq_unpin_lock(rq, rf);
1783 	raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1784 }
1785 
1786 static inline void
1787 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1788 	__releases(rq->lock)
1789 {
1790 	rq_unpin_lock(rq, rf);
1791 	raw_spin_unlock_irq(&rq->lock);
1792 }
1793 
1794 static inline void
1795 rq_unlock(struct rq *rq, struct rq_flags *rf)
1796 	__releases(rq->lock)
1797 {
1798 	rq_unpin_lock(rq, rf);
1799 	raw_spin_unlock(&rq->lock);
1800 }
1801 
1802 #ifdef CONFIG_SMP
1803 #ifdef CONFIG_PREEMPT
1804 
1805 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1806 
1807 /*
1808  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1809  * way at the expense of forcing extra atomic operations in all
1810  * invocations.  This assures that the double_lock is acquired using the
1811  * same underlying policy as the spinlock_t on this architecture, which
1812  * reduces latency compared to the unfair variant below.  However, it
1813  * also adds more overhead and therefore may reduce throughput.
1814  */
1815 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1816 	__releases(this_rq->lock)
1817 	__acquires(busiest->lock)
1818 	__acquires(this_rq->lock)
1819 {
1820 	raw_spin_unlock(&this_rq->lock);
1821 	double_rq_lock(this_rq, busiest);
1822 
1823 	return 1;
1824 }
1825 
1826 #else
1827 /*
1828  * Unfair double_lock_balance: Optimizes throughput at the expense of
1829  * latency by eliminating extra atomic operations when the locks are
1830  * already in proper order on entry.  This favors lower cpu-ids and will
1831  * grant the double lock to lower cpus over higher ids under contention,
1832  * regardless of entry order into the function.
1833  */
1834 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1835 	__releases(this_rq->lock)
1836 	__acquires(busiest->lock)
1837 	__acquires(this_rq->lock)
1838 {
1839 	int ret = 0;
1840 
1841 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1842 		if (busiest < this_rq) {
1843 			raw_spin_unlock(&this_rq->lock);
1844 			raw_spin_lock(&busiest->lock);
1845 			raw_spin_lock_nested(&this_rq->lock,
1846 					      SINGLE_DEPTH_NESTING);
1847 			ret = 1;
1848 		} else
1849 			raw_spin_lock_nested(&busiest->lock,
1850 					      SINGLE_DEPTH_NESTING);
1851 	}
1852 	return ret;
1853 }
1854 
1855 #endif /* CONFIG_PREEMPT */
1856 
1857 /*
1858  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1859  */
1860 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1861 {
1862 	if (unlikely(!irqs_disabled())) {
1863 		/* printk() doesn't work good under rq->lock */
1864 		raw_spin_unlock(&this_rq->lock);
1865 		BUG_ON(1);
1866 	}
1867 
1868 	return _double_lock_balance(this_rq, busiest);
1869 }
1870 
1871 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1872 	__releases(busiest->lock)
1873 {
1874 	raw_spin_unlock(&busiest->lock);
1875 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1876 }
1877 
1878 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1879 {
1880 	if (l1 > l2)
1881 		swap(l1, l2);
1882 
1883 	spin_lock(l1);
1884 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1885 }
1886 
1887 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1888 {
1889 	if (l1 > l2)
1890 		swap(l1, l2);
1891 
1892 	spin_lock_irq(l1);
1893 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1894 }
1895 
1896 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1897 {
1898 	if (l1 > l2)
1899 		swap(l1, l2);
1900 
1901 	raw_spin_lock(l1);
1902 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1903 }
1904 
1905 /*
1906  * double_rq_lock - safely lock two runqueues
1907  *
1908  * Note this does not disable interrupts like task_rq_lock,
1909  * you need to do so manually before calling.
1910  */
1911 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1912 	__acquires(rq1->lock)
1913 	__acquires(rq2->lock)
1914 {
1915 	BUG_ON(!irqs_disabled());
1916 	if (rq1 == rq2) {
1917 		raw_spin_lock(&rq1->lock);
1918 		__acquire(rq2->lock);	/* Fake it out ;) */
1919 	} else {
1920 		if (rq1 < rq2) {
1921 			raw_spin_lock(&rq1->lock);
1922 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1923 		} else {
1924 			raw_spin_lock(&rq2->lock);
1925 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1926 		}
1927 	}
1928 }
1929 
1930 /*
1931  * double_rq_unlock - safely unlock two runqueues
1932  *
1933  * Note this does not restore interrupts like task_rq_unlock,
1934  * you need to do so manually after calling.
1935  */
1936 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1937 	__releases(rq1->lock)
1938 	__releases(rq2->lock)
1939 {
1940 	raw_spin_unlock(&rq1->lock);
1941 	if (rq1 != rq2)
1942 		raw_spin_unlock(&rq2->lock);
1943 	else
1944 		__release(rq2->lock);
1945 }
1946 
1947 extern void set_rq_online (struct rq *rq);
1948 extern void set_rq_offline(struct rq *rq);
1949 extern bool sched_smp_initialized;
1950 
1951 #else /* CONFIG_SMP */
1952 
1953 /*
1954  * double_rq_lock - safely lock two runqueues
1955  *
1956  * Note this does not disable interrupts like task_rq_lock,
1957  * you need to do so manually before calling.
1958  */
1959 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1960 	__acquires(rq1->lock)
1961 	__acquires(rq2->lock)
1962 {
1963 	BUG_ON(!irqs_disabled());
1964 	BUG_ON(rq1 != rq2);
1965 	raw_spin_lock(&rq1->lock);
1966 	__acquire(rq2->lock);	/* Fake it out ;) */
1967 }
1968 
1969 /*
1970  * double_rq_unlock - safely unlock two runqueues
1971  *
1972  * Note this does not restore interrupts like task_rq_unlock,
1973  * you need to do so manually after calling.
1974  */
1975 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1976 	__releases(rq1->lock)
1977 	__releases(rq2->lock)
1978 {
1979 	BUG_ON(rq1 != rq2);
1980 	raw_spin_unlock(&rq1->lock);
1981 	__release(rq2->lock);
1982 }
1983 
1984 #endif
1985 
1986 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1987 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1988 
1989 #ifdef	CONFIG_SCHED_DEBUG
1990 extern bool sched_debug_enabled;
1991 
1992 extern void print_cfs_stats(struct seq_file *m, int cpu);
1993 extern void print_rt_stats(struct seq_file *m, int cpu);
1994 extern void print_dl_stats(struct seq_file *m, int cpu);
1995 extern void
1996 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1997 #ifdef CONFIG_NUMA_BALANCING
1998 extern void
1999 show_numa_stats(struct task_struct *p, struct seq_file *m);
2000 extern void
2001 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2002 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2003 #endif /* CONFIG_NUMA_BALANCING */
2004 #endif /* CONFIG_SCHED_DEBUG */
2005 
2006 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2007 extern void init_rt_rq(struct rt_rq *rt_rq);
2008 extern void init_dl_rq(struct dl_rq *dl_rq);
2009 
2010 extern void cfs_bandwidth_usage_inc(void);
2011 extern void cfs_bandwidth_usage_dec(void);
2012 
2013 #ifdef CONFIG_NO_HZ_COMMON
2014 enum rq_nohz_flag_bits {
2015 	NOHZ_TICK_STOPPED,
2016 	NOHZ_BALANCE_KICK,
2017 };
2018 
2019 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2020 
2021 extern void nohz_balance_exit_idle(unsigned int cpu);
2022 #else
2023 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
2024 #endif
2025 
2026 
2027 #ifdef CONFIG_SMP
2028 static inline
2029 void __dl_update(struct dl_bw *dl_b, s64 bw)
2030 {
2031 	struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2032 	int i;
2033 
2034 	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2035 			 "sched RCU must be held");
2036 	for_each_cpu_and(i, rd->span, cpu_active_mask) {
2037 		struct rq *rq = cpu_rq(i);
2038 
2039 		rq->dl.extra_bw += bw;
2040 	}
2041 }
2042 #else
2043 static inline
2044 void __dl_update(struct dl_bw *dl_b, s64 bw)
2045 {
2046 	struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2047 
2048 	dl->extra_bw += bw;
2049 }
2050 #endif
2051 
2052 
2053 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2054 struct irqtime {
2055 	u64			total;
2056 	u64			tick_delta;
2057 	u64			irq_start_time;
2058 	struct u64_stats_sync	sync;
2059 };
2060 
2061 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2062 
2063 /*
2064  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2065  * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2066  * and never move forward.
2067  */
2068 static inline u64 irq_time_read(int cpu)
2069 {
2070 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2071 	unsigned int seq;
2072 	u64 total;
2073 
2074 	do {
2075 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2076 		total = irqtime->total;
2077 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2078 
2079 	return total;
2080 }
2081 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2082 
2083 #ifdef CONFIG_CPU_FREQ
2084 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2085 
2086 /**
2087  * cpufreq_update_util - Take a note about CPU utilization changes.
2088  * @rq: Runqueue to carry out the update for.
2089  * @flags: Update reason flags.
2090  *
2091  * This function is called by the scheduler on the CPU whose utilization is
2092  * being updated.
2093  *
2094  * It can only be called from RCU-sched read-side critical sections.
2095  *
2096  * The way cpufreq is currently arranged requires it to evaluate the CPU
2097  * performance state (frequency/voltage) on a regular basis to prevent it from
2098  * being stuck in a completely inadequate performance level for too long.
2099  * That is not guaranteed to happen if the updates are only triggered from CFS,
2100  * though, because they may not be coming in if RT or deadline tasks are active
2101  * all the time (or there are RT and DL tasks only).
2102  *
2103  * As a workaround for that issue, this function is called by the RT and DL
2104  * sched classes to trigger extra cpufreq updates to prevent it from stalling,
2105  * but that really is a band-aid.  Going forward it should be replaced with
2106  * solutions targeted more specifically at RT and DL tasks.
2107  */
2108 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2109 {
2110 	struct update_util_data *data;
2111 
2112 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2113 						  cpu_of(rq)));
2114 	if (data)
2115 		data->func(data, rq_clock(rq), flags);
2116 }
2117 #else
2118 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2119 #endif /* CONFIG_CPU_FREQ */
2120 
2121 #ifdef arch_scale_freq_capacity
2122 #ifndef arch_scale_freq_invariant
2123 #define arch_scale_freq_invariant()	(true)
2124 #endif
2125 #else /* arch_scale_freq_capacity */
2126 #define arch_scale_freq_invariant()	(false)
2127 #endif
2128