xref: /openbmc/linux/kernel/sched/sched.h (revision abfbd895)
1 
2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/sched/deadline.h>
6 #include <linux/mutex.h>
7 #include <linux/spinlock.h>
8 #include <linux/stop_machine.h>
9 #include <linux/irq_work.h>
10 #include <linux/tick.h>
11 #include <linux/slab.h>
12 
13 #include "cpupri.h"
14 #include "cpudeadline.h"
15 #include "cpuacct.h"
16 
17 struct rq;
18 struct cpuidle_state;
19 
20 /* task_struct::on_rq states: */
21 #define TASK_ON_RQ_QUEUED	1
22 #define TASK_ON_RQ_MIGRATING	2
23 
24 extern __read_mostly int scheduler_running;
25 
26 extern unsigned long calc_load_update;
27 extern atomic_long_t calc_load_tasks;
28 
29 extern void calc_global_load_tick(struct rq *this_rq);
30 extern long calc_load_fold_active(struct rq *this_rq);
31 
32 #ifdef CONFIG_SMP
33 extern void update_cpu_load_active(struct rq *this_rq);
34 #else
35 static inline void update_cpu_load_active(struct rq *this_rq) { }
36 #endif
37 
38 /*
39  * Helpers for converting nanosecond timing to jiffy resolution
40  */
41 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
42 
43 /*
44  * Increase resolution of nice-level calculations for 64-bit architectures.
45  * The extra resolution improves shares distribution and load balancing of
46  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
47  * hierarchies, especially on larger systems. This is not a user-visible change
48  * and does not change the user-interface for setting shares/weights.
49  *
50  * We increase resolution only if we have enough bits to allow this increased
51  * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
52  * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
53  * increased costs.
54  */
55 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load  */
56 # define SCHED_LOAD_RESOLUTION	10
57 # define scale_load(w)		((w) << SCHED_LOAD_RESOLUTION)
58 # define scale_load_down(w)	((w) >> SCHED_LOAD_RESOLUTION)
59 #else
60 # define SCHED_LOAD_RESOLUTION	0
61 # define scale_load(w)		(w)
62 # define scale_load_down(w)	(w)
63 #endif
64 
65 #define SCHED_LOAD_SHIFT	(10 + SCHED_LOAD_RESOLUTION)
66 #define SCHED_LOAD_SCALE	(1L << SCHED_LOAD_SHIFT)
67 
68 #define NICE_0_LOAD		SCHED_LOAD_SCALE
69 #define NICE_0_SHIFT		SCHED_LOAD_SHIFT
70 
71 /*
72  * Single value that decides SCHED_DEADLINE internal math precision.
73  * 10 -> just above 1us
74  * 9  -> just above 0.5us
75  */
76 #define DL_SCALE (10)
77 
78 /*
79  * These are the 'tuning knobs' of the scheduler:
80  */
81 
82 /*
83  * single value that denotes runtime == period, ie unlimited time.
84  */
85 #define RUNTIME_INF	((u64)~0ULL)
86 
87 static inline int idle_policy(int policy)
88 {
89 	return policy == SCHED_IDLE;
90 }
91 static inline int fair_policy(int policy)
92 {
93 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
94 }
95 
96 static inline int rt_policy(int policy)
97 {
98 	return policy == SCHED_FIFO || policy == SCHED_RR;
99 }
100 
101 static inline int dl_policy(int policy)
102 {
103 	return policy == SCHED_DEADLINE;
104 }
105 static inline bool valid_policy(int policy)
106 {
107 	return idle_policy(policy) || fair_policy(policy) ||
108 		rt_policy(policy) || dl_policy(policy);
109 }
110 
111 static inline int task_has_rt_policy(struct task_struct *p)
112 {
113 	return rt_policy(p->policy);
114 }
115 
116 static inline int task_has_dl_policy(struct task_struct *p)
117 {
118 	return dl_policy(p->policy);
119 }
120 
121 /*
122  * Tells if entity @a should preempt entity @b.
123  */
124 static inline bool
125 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
126 {
127 	return dl_time_before(a->deadline, b->deadline);
128 }
129 
130 /*
131  * This is the priority-queue data structure of the RT scheduling class:
132  */
133 struct rt_prio_array {
134 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
135 	struct list_head queue[MAX_RT_PRIO];
136 };
137 
138 struct rt_bandwidth {
139 	/* nests inside the rq lock: */
140 	raw_spinlock_t		rt_runtime_lock;
141 	ktime_t			rt_period;
142 	u64			rt_runtime;
143 	struct hrtimer		rt_period_timer;
144 	unsigned int		rt_period_active;
145 };
146 
147 void __dl_clear_params(struct task_struct *p);
148 
149 /*
150  * To keep the bandwidth of -deadline tasks and groups under control
151  * we need some place where:
152  *  - store the maximum -deadline bandwidth of the system (the group);
153  *  - cache the fraction of that bandwidth that is currently allocated.
154  *
155  * This is all done in the data structure below. It is similar to the
156  * one used for RT-throttling (rt_bandwidth), with the main difference
157  * that, since here we are only interested in admission control, we
158  * do not decrease any runtime while the group "executes", neither we
159  * need a timer to replenish it.
160  *
161  * With respect to SMP, the bandwidth is given on a per-CPU basis,
162  * meaning that:
163  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
164  *  - dl_total_bw array contains, in the i-eth element, the currently
165  *    allocated bandwidth on the i-eth CPU.
166  * Moreover, groups consume bandwidth on each CPU, while tasks only
167  * consume bandwidth on the CPU they're running on.
168  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
169  * that will be shown the next time the proc or cgroup controls will
170  * be red. It on its turn can be changed by writing on its own
171  * control.
172  */
173 struct dl_bandwidth {
174 	raw_spinlock_t dl_runtime_lock;
175 	u64 dl_runtime;
176 	u64 dl_period;
177 };
178 
179 static inline int dl_bandwidth_enabled(void)
180 {
181 	return sysctl_sched_rt_runtime >= 0;
182 }
183 
184 extern struct dl_bw *dl_bw_of(int i);
185 
186 struct dl_bw {
187 	raw_spinlock_t lock;
188 	u64 bw, total_bw;
189 };
190 
191 static inline
192 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
193 {
194 	dl_b->total_bw -= tsk_bw;
195 }
196 
197 static inline
198 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
199 {
200 	dl_b->total_bw += tsk_bw;
201 }
202 
203 static inline
204 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
205 {
206 	return dl_b->bw != -1 &&
207 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
208 }
209 
210 extern struct mutex sched_domains_mutex;
211 
212 #ifdef CONFIG_CGROUP_SCHED
213 
214 #include <linux/cgroup.h>
215 
216 struct cfs_rq;
217 struct rt_rq;
218 
219 extern struct list_head task_groups;
220 
221 struct cfs_bandwidth {
222 #ifdef CONFIG_CFS_BANDWIDTH
223 	raw_spinlock_t lock;
224 	ktime_t period;
225 	u64 quota, runtime;
226 	s64 hierarchical_quota;
227 	u64 runtime_expires;
228 
229 	int idle, period_active;
230 	struct hrtimer period_timer, slack_timer;
231 	struct list_head throttled_cfs_rq;
232 
233 	/* statistics */
234 	int nr_periods, nr_throttled;
235 	u64 throttled_time;
236 #endif
237 };
238 
239 /* task group related information */
240 struct task_group {
241 	struct cgroup_subsys_state css;
242 
243 #ifdef CONFIG_FAIR_GROUP_SCHED
244 	/* schedulable entities of this group on each cpu */
245 	struct sched_entity **se;
246 	/* runqueue "owned" by this group on each cpu */
247 	struct cfs_rq **cfs_rq;
248 	unsigned long shares;
249 
250 #ifdef	CONFIG_SMP
251 	atomic_long_t load_avg;
252 #endif
253 #endif
254 
255 #ifdef CONFIG_RT_GROUP_SCHED
256 	struct sched_rt_entity **rt_se;
257 	struct rt_rq **rt_rq;
258 
259 	struct rt_bandwidth rt_bandwidth;
260 #endif
261 
262 	struct rcu_head rcu;
263 	struct list_head list;
264 
265 	struct task_group *parent;
266 	struct list_head siblings;
267 	struct list_head children;
268 
269 #ifdef CONFIG_SCHED_AUTOGROUP
270 	struct autogroup *autogroup;
271 #endif
272 
273 	struct cfs_bandwidth cfs_bandwidth;
274 };
275 
276 #ifdef CONFIG_FAIR_GROUP_SCHED
277 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
278 
279 /*
280  * A weight of 0 or 1 can cause arithmetics problems.
281  * A weight of a cfs_rq is the sum of weights of which entities
282  * are queued on this cfs_rq, so a weight of a entity should not be
283  * too large, so as the shares value of a task group.
284  * (The default weight is 1024 - so there's no practical
285  *  limitation from this.)
286  */
287 #define MIN_SHARES	(1UL <<  1)
288 #define MAX_SHARES	(1UL << 18)
289 #endif
290 
291 typedef int (*tg_visitor)(struct task_group *, void *);
292 
293 extern int walk_tg_tree_from(struct task_group *from,
294 			     tg_visitor down, tg_visitor up, void *data);
295 
296 /*
297  * Iterate the full tree, calling @down when first entering a node and @up when
298  * leaving it for the final time.
299  *
300  * Caller must hold rcu_lock or sufficient equivalent.
301  */
302 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
303 {
304 	return walk_tg_tree_from(&root_task_group, down, up, data);
305 }
306 
307 extern int tg_nop(struct task_group *tg, void *data);
308 
309 extern void free_fair_sched_group(struct task_group *tg);
310 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
311 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
312 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
313 			struct sched_entity *se, int cpu,
314 			struct sched_entity *parent);
315 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
316 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
317 
318 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
319 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
320 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
321 
322 extern void free_rt_sched_group(struct task_group *tg);
323 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
324 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
325 		struct sched_rt_entity *rt_se, int cpu,
326 		struct sched_rt_entity *parent);
327 
328 extern struct task_group *sched_create_group(struct task_group *parent);
329 extern void sched_online_group(struct task_group *tg,
330 			       struct task_group *parent);
331 extern void sched_destroy_group(struct task_group *tg);
332 extern void sched_offline_group(struct task_group *tg);
333 
334 extern void sched_move_task(struct task_struct *tsk);
335 
336 #ifdef CONFIG_FAIR_GROUP_SCHED
337 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
338 #endif
339 
340 #else /* CONFIG_CGROUP_SCHED */
341 
342 struct cfs_bandwidth { };
343 
344 #endif	/* CONFIG_CGROUP_SCHED */
345 
346 /* CFS-related fields in a runqueue */
347 struct cfs_rq {
348 	struct load_weight load;
349 	unsigned int nr_running, h_nr_running;
350 
351 	u64 exec_clock;
352 	u64 min_vruntime;
353 #ifndef CONFIG_64BIT
354 	u64 min_vruntime_copy;
355 #endif
356 
357 	struct rb_root tasks_timeline;
358 	struct rb_node *rb_leftmost;
359 
360 	/*
361 	 * 'curr' points to currently running entity on this cfs_rq.
362 	 * It is set to NULL otherwise (i.e when none are currently running).
363 	 */
364 	struct sched_entity *curr, *next, *last, *skip;
365 
366 #ifdef	CONFIG_SCHED_DEBUG
367 	unsigned int nr_spread_over;
368 #endif
369 
370 #ifdef CONFIG_SMP
371 	/*
372 	 * CFS load tracking
373 	 */
374 	struct sched_avg avg;
375 	u64 runnable_load_sum;
376 	unsigned long runnable_load_avg;
377 #ifdef CONFIG_FAIR_GROUP_SCHED
378 	unsigned long tg_load_avg_contrib;
379 #endif
380 	atomic_long_t removed_load_avg, removed_util_avg;
381 #ifndef CONFIG_64BIT
382 	u64 load_last_update_time_copy;
383 #endif
384 
385 #ifdef CONFIG_FAIR_GROUP_SCHED
386 	/*
387 	 *   h_load = weight * f(tg)
388 	 *
389 	 * Where f(tg) is the recursive weight fraction assigned to
390 	 * this group.
391 	 */
392 	unsigned long h_load;
393 	u64 last_h_load_update;
394 	struct sched_entity *h_load_next;
395 #endif /* CONFIG_FAIR_GROUP_SCHED */
396 #endif /* CONFIG_SMP */
397 
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 	struct rq *rq;	/* cpu runqueue to which this cfs_rq is attached */
400 
401 	/*
402 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
403 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
404 	 * (like users, containers etc.)
405 	 *
406 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
407 	 * list is used during load balance.
408 	 */
409 	int on_list;
410 	struct list_head leaf_cfs_rq_list;
411 	struct task_group *tg;	/* group that "owns" this runqueue */
412 
413 #ifdef CONFIG_CFS_BANDWIDTH
414 	int runtime_enabled;
415 	u64 runtime_expires;
416 	s64 runtime_remaining;
417 
418 	u64 throttled_clock, throttled_clock_task;
419 	u64 throttled_clock_task_time;
420 	int throttled, throttle_count;
421 	struct list_head throttled_list;
422 #endif /* CONFIG_CFS_BANDWIDTH */
423 #endif /* CONFIG_FAIR_GROUP_SCHED */
424 };
425 
426 static inline int rt_bandwidth_enabled(void)
427 {
428 	return sysctl_sched_rt_runtime >= 0;
429 }
430 
431 /* RT IPI pull logic requires IRQ_WORK */
432 #ifdef CONFIG_IRQ_WORK
433 # define HAVE_RT_PUSH_IPI
434 #endif
435 
436 /* Real-Time classes' related field in a runqueue: */
437 struct rt_rq {
438 	struct rt_prio_array active;
439 	unsigned int rt_nr_running;
440 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
441 	struct {
442 		int curr; /* highest queued rt task prio */
443 #ifdef CONFIG_SMP
444 		int next; /* next highest */
445 #endif
446 	} highest_prio;
447 #endif
448 #ifdef CONFIG_SMP
449 	unsigned long rt_nr_migratory;
450 	unsigned long rt_nr_total;
451 	int overloaded;
452 	struct plist_head pushable_tasks;
453 #ifdef HAVE_RT_PUSH_IPI
454 	int push_flags;
455 	int push_cpu;
456 	struct irq_work push_work;
457 	raw_spinlock_t push_lock;
458 #endif
459 #endif /* CONFIG_SMP */
460 	int rt_queued;
461 
462 	int rt_throttled;
463 	u64 rt_time;
464 	u64 rt_runtime;
465 	/* Nests inside the rq lock: */
466 	raw_spinlock_t rt_runtime_lock;
467 
468 #ifdef CONFIG_RT_GROUP_SCHED
469 	unsigned long rt_nr_boosted;
470 
471 	struct rq *rq;
472 	struct task_group *tg;
473 #endif
474 };
475 
476 /* Deadline class' related fields in a runqueue */
477 struct dl_rq {
478 	/* runqueue is an rbtree, ordered by deadline */
479 	struct rb_root rb_root;
480 	struct rb_node *rb_leftmost;
481 
482 	unsigned long dl_nr_running;
483 
484 #ifdef CONFIG_SMP
485 	/*
486 	 * Deadline values of the currently executing and the
487 	 * earliest ready task on this rq. Caching these facilitates
488 	 * the decision wether or not a ready but not running task
489 	 * should migrate somewhere else.
490 	 */
491 	struct {
492 		u64 curr;
493 		u64 next;
494 	} earliest_dl;
495 
496 	unsigned long dl_nr_migratory;
497 	int overloaded;
498 
499 	/*
500 	 * Tasks on this rq that can be pushed away. They are kept in
501 	 * an rb-tree, ordered by tasks' deadlines, with caching
502 	 * of the leftmost (earliest deadline) element.
503 	 */
504 	struct rb_root pushable_dl_tasks_root;
505 	struct rb_node *pushable_dl_tasks_leftmost;
506 #else
507 	struct dl_bw dl_bw;
508 #endif
509 };
510 
511 #ifdef CONFIG_SMP
512 
513 /*
514  * We add the notion of a root-domain which will be used to define per-domain
515  * variables. Each exclusive cpuset essentially defines an island domain by
516  * fully partitioning the member cpus from any other cpuset. Whenever a new
517  * exclusive cpuset is created, we also create and attach a new root-domain
518  * object.
519  *
520  */
521 struct root_domain {
522 	atomic_t refcount;
523 	atomic_t rto_count;
524 	struct rcu_head rcu;
525 	cpumask_var_t span;
526 	cpumask_var_t online;
527 
528 	/* Indicate more than one runnable task for any CPU */
529 	bool overload;
530 
531 	/*
532 	 * The bit corresponding to a CPU gets set here if such CPU has more
533 	 * than one runnable -deadline task (as it is below for RT tasks).
534 	 */
535 	cpumask_var_t dlo_mask;
536 	atomic_t dlo_count;
537 	struct dl_bw dl_bw;
538 	struct cpudl cpudl;
539 
540 	/*
541 	 * The "RT overload" flag: it gets set if a CPU has more than
542 	 * one runnable RT task.
543 	 */
544 	cpumask_var_t rto_mask;
545 	struct cpupri cpupri;
546 };
547 
548 extern struct root_domain def_root_domain;
549 
550 #endif /* CONFIG_SMP */
551 
552 /*
553  * This is the main, per-CPU runqueue data structure.
554  *
555  * Locking rule: those places that want to lock multiple runqueues
556  * (such as the load balancing or the thread migration code), lock
557  * acquire operations must be ordered by ascending &runqueue.
558  */
559 struct rq {
560 	/* runqueue lock: */
561 	raw_spinlock_t lock;
562 
563 	/*
564 	 * nr_running and cpu_load should be in the same cacheline because
565 	 * remote CPUs use both these fields when doing load calculation.
566 	 */
567 	unsigned int nr_running;
568 #ifdef CONFIG_NUMA_BALANCING
569 	unsigned int nr_numa_running;
570 	unsigned int nr_preferred_running;
571 #endif
572 	#define CPU_LOAD_IDX_MAX 5
573 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
574 	unsigned long last_load_update_tick;
575 #ifdef CONFIG_NO_HZ_COMMON
576 	u64 nohz_stamp;
577 	unsigned long nohz_flags;
578 #endif
579 #ifdef CONFIG_NO_HZ_FULL
580 	unsigned long last_sched_tick;
581 #endif
582 	/* capture load from *all* tasks on this cpu: */
583 	struct load_weight load;
584 	unsigned long nr_load_updates;
585 	u64 nr_switches;
586 
587 	struct cfs_rq cfs;
588 	struct rt_rq rt;
589 	struct dl_rq dl;
590 
591 #ifdef CONFIG_FAIR_GROUP_SCHED
592 	/* list of leaf cfs_rq on this cpu: */
593 	struct list_head leaf_cfs_rq_list;
594 #endif /* CONFIG_FAIR_GROUP_SCHED */
595 
596 	/*
597 	 * This is part of a global counter where only the total sum
598 	 * over all CPUs matters. A task can increase this counter on
599 	 * one CPU and if it got migrated afterwards it may decrease
600 	 * it on another CPU. Always updated under the runqueue lock:
601 	 */
602 	unsigned long nr_uninterruptible;
603 
604 	struct task_struct *curr, *idle, *stop;
605 	unsigned long next_balance;
606 	struct mm_struct *prev_mm;
607 
608 	unsigned int clock_skip_update;
609 	u64 clock;
610 	u64 clock_task;
611 
612 	atomic_t nr_iowait;
613 
614 #ifdef CONFIG_SMP
615 	struct root_domain *rd;
616 	struct sched_domain *sd;
617 
618 	unsigned long cpu_capacity;
619 	unsigned long cpu_capacity_orig;
620 
621 	struct callback_head *balance_callback;
622 
623 	unsigned char idle_balance;
624 	/* For active balancing */
625 	int active_balance;
626 	int push_cpu;
627 	struct cpu_stop_work active_balance_work;
628 	/* cpu of this runqueue: */
629 	int cpu;
630 	int online;
631 
632 	struct list_head cfs_tasks;
633 
634 	u64 rt_avg;
635 	u64 age_stamp;
636 	u64 idle_stamp;
637 	u64 avg_idle;
638 
639 	/* This is used to determine avg_idle's max value */
640 	u64 max_idle_balance_cost;
641 #endif
642 
643 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
644 	u64 prev_irq_time;
645 #endif
646 #ifdef CONFIG_PARAVIRT
647 	u64 prev_steal_time;
648 #endif
649 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
650 	u64 prev_steal_time_rq;
651 #endif
652 
653 	/* calc_load related fields */
654 	unsigned long calc_load_update;
655 	long calc_load_active;
656 
657 #ifdef CONFIG_SCHED_HRTICK
658 #ifdef CONFIG_SMP
659 	int hrtick_csd_pending;
660 	struct call_single_data hrtick_csd;
661 #endif
662 	struct hrtimer hrtick_timer;
663 #endif
664 
665 #ifdef CONFIG_SCHEDSTATS
666 	/* latency stats */
667 	struct sched_info rq_sched_info;
668 	unsigned long long rq_cpu_time;
669 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
670 
671 	/* sys_sched_yield() stats */
672 	unsigned int yld_count;
673 
674 	/* schedule() stats */
675 	unsigned int sched_count;
676 	unsigned int sched_goidle;
677 
678 	/* try_to_wake_up() stats */
679 	unsigned int ttwu_count;
680 	unsigned int ttwu_local;
681 #endif
682 
683 #ifdef CONFIG_SMP
684 	struct llist_head wake_list;
685 #endif
686 
687 #ifdef CONFIG_CPU_IDLE
688 	/* Must be inspected within a rcu lock section */
689 	struct cpuidle_state *idle_state;
690 #endif
691 };
692 
693 static inline int cpu_of(struct rq *rq)
694 {
695 #ifdef CONFIG_SMP
696 	return rq->cpu;
697 #else
698 	return 0;
699 #endif
700 }
701 
702 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
703 
704 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
705 #define this_rq()		this_cpu_ptr(&runqueues)
706 #define task_rq(p)		cpu_rq(task_cpu(p))
707 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
708 #define raw_rq()		raw_cpu_ptr(&runqueues)
709 
710 static inline u64 __rq_clock_broken(struct rq *rq)
711 {
712 	return READ_ONCE(rq->clock);
713 }
714 
715 static inline u64 rq_clock(struct rq *rq)
716 {
717 	lockdep_assert_held(&rq->lock);
718 	return rq->clock;
719 }
720 
721 static inline u64 rq_clock_task(struct rq *rq)
722 {
723 	lockdep_assert_held(&rq->lock);
724 	return rq->clock_task;
725 }
726 
727 #define RQCF_REQ_SKIP	0x01
728 #define RQCF_ACT_SKIP	0x02
729 
730 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
731 {
732 	lockdep_assert_held(&rq->lock);
733 	if (skip)
734 		rq->clock_skip_update |= RQCF_REQ_SKIP;
735 	else
736 		rq->clock_skip_update &= ~RQCF_REQ_SKIP;
737 }
738 
739 #ifdef CONFIG_NUMA
740 enum numa_topology_type {
741 	NUMA_DIRECT,
742 	NUMA_GLUELESS_MESH,
743 	NUMA_BACKPLANE,
744 };
745 extern enum numa_topology_type sched_numa_topology_type;
746 extern int sched_max_numa_distance;
747 extern bool find_numa_distance(int distance);
748 #endif
749 
750 #ifdef CONFIG_NUMA_BALANCING
751 /* The regions in numa_faults array from task_struct */
752 enum numa_faults_stats {
753 	NUMA_MEM = 0,
754 	NUMA_CPU,
755 	NUMA_MEMBUF,
756 	NUMA_CPUBUF
757 };
758 extern void sched_setnuma(struct task_struct *p, int node);
759 extern int migrate_task_to(struct task_struct *p, int cpu);
760 extern int migrate_swap(struct task_struct *, struct task_struct *);
761 #endif /* CONFIG_NUMA_BALANCING */
762 
763 #ifdef CONFIG_SMP
764 
765 static inline void
766 queue_balance_callback(struct rq *rq,
767 		       struct callback_head *head,
768 		       void (*func)(struct rq *rq))
769 {
770 	lockdep_assert_held(&rq->lock);
771 
772 	if (unlikely(head->next))
773 		return;
774 
775 	head->func = (void (*)(struct callback_head *))func;
776 	head->next = rq->balance_callback;
777 	rq->balance_callback = head;
778 }
779 
780 extern void sched_ttwu_pending(void);
781 
782 #define rcu_dereference_check_sched_domain(p) \
783 	rcu_dereference_check((p), \
784 			      lockdep_is_held(&sched_domains_mutex))
785 
786 /*
787  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
788  * See detach_destroy_domains: synchronize_sched for details.
789  *
790  * The domain tree of any CPU may only be accessed from within
791  * preempt-disabled sections.
792  */
793 #define for_each_domain(cpu, __sd) \
794 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
795 			__sd; __sd = __sd->parent)
796 
797 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
798 
799 /**
800  * highest_flag_domain - Return highest sched_domain containing flag.
801  * @cpu:	The cpu whose highest level of sched domain is to
802  *		be returned.
803  * @flag:	The flag to check for the highest sched_domain
804  *		for the given cpu.
805  *
806  * Returns the highest sched_domain of a cpu which contains the given flag.
807  */
808 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
809 {
810 	struct sched_domain *sd, *hsd = NULL;
811 
812 	for_each_domain(cpu, sd) {
813 		if (!(sd->flags & flag))
814 			break;
815 		hsd = sd;
816 	}
817 
818 	return hsd;
819 }
820 
821 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
822 {
823 	struct sched_domain *sd;
824 
825 	for_each_domain(cpu, sd) {
826 		if (sd->flags & flag)
827 			break;
828 	}
829 
830 	return sd;
831 }
832 
833 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
834 DECLARE_PER_CPU(int, sd_llc_size);
835 DECLARE_PER_CPU(int, sd_llc_id);
836 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
837 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
838 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
839 
840 struct sched_group_capacity {
841 	atomic_t ref;
842 	/*
843 	 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
844 	 * for a single CPU.
845 	 */
846 	unsigned int capacity;
847 	unsigned long next_update;
848 	int imbalance; /* XXX unrelated to capacity but shared group state */
849 	/*
850 	 * Number of busy cpus in this group.
851 	 */
852 	atomic_t nr_busy_cpus;
853 
854 	unsigned long cpumask[0]; /* iteration mask */
855 };
856 
857 struct sched_group {
858 	struct sched_group *next;	/* Must be a circular list */
859 	atomic_t ref;
860 
861 	unsigned int group_weight;
862 	struct sched_group_capacity *sgc;
863 
864 	/*
865 	 * The CPUs this group covers.
866 	 *
867 	 * NOTE: this field is variable length. (Allocated dynamically
868 	 * by attaching extra space to the end of the structure,
869 	 * depending on how many CPUs the kernel has booted up with)
870 	 */
871 	unsigned long cpumask[0];
872 };
873 
874 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
875 {
876 	return to_cpumask(sg->cpumask);
877 }
878 
879 /*
880  * cpumask masking which cpus in the group are allowed to iterate up the domain
881  * tree.
882  */
883 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
884 {
885 	return to_cpumask(sg->sgc->cpumask);
886 }
887 
888 /**
889  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
890  * @group: The group whose first cpu is to be returned.
891  */
892 static inline unsigned int group_first_cpu(struct sched_group *group)
893 {
894 	return cpumask_first(sched_group_cpus(group));
895 }
896 
897 extern int group_balance_cpu(struct sched_group *sg);
898 
899 #else
900 
901 static inline void sched_ttwu_pending(void) { }
902 
903 #endif /* CONFIG_SMP */
904 
905 #include "stats.h"
906 #include "auto_group.h"
907 
908 #ifdef CONFIG_CGROUP_SCHED
909 
910 /*
911  * Return the group to which this tasks belongs.
912  *
913  * We cannot use task_css() and friends because the cgroup subsystem
914  * changes that value before the cgroup_subsys::attach() method is called,
915  * therefore we cannot pin it and might observe the wrong value.
916  *
917  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
918  * core changes this before calling sched_move_task().
919  *
920  * Instead we use a 'copy' which is updated from sched_move_task() while
921  * holding both task_struct::pi_lock and rq::lock.
922  */
923 static inline struct task_group *task_group(struct task_struct *p)
924 {
925 	return p->sched_task_group;
926 }
927 
928 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
929 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
930 {
931 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
932 	struct task_group *tg = task_group(p);
933 #endif
934 
935 #ifdef CONFIG_FAIR_GROUP_SCHED
936 	p->se.cfs_rq = tg->cfs_rq[cpu];
937 	p->se.parent = tg->se[cpu];
938 #endif
939 
940 #ifdef CONFIG_RT_GROUP_SCHED
941 	p->rt.rt_rq  = tg->rt_rq[cpu];
942 	p->rt.parent = tg->rt_se[cpu];
943 #endif
944 }
945 
946 #else /* CONFIG_CGROUP_SCHED */
947 
948 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
949 static inline struct task_group *task_group(struct task_struct *p)
950 {
951 	return NULL;
952 }
953 
954 #endif /* CONFIG_CGROUP_SCHED */
955 
956 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
957 {
958 	set_task_rq(p, cpu);
959 #ifdef CONFIG_SMP
960 	/*
961 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
962 	 * successfuly executed on another CPU. We must ensure that updates of
963 	 * per-task data have been completed by this moment.
964 	 */
965 	smp_wmb();
966 	task_thread_info(p)->cpu = cpu;
967 	p->wake_cpu = cpu;
968 #endif
969 }
970 
971 /*
972  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
973  */
974 #ifdef CONFIG_SCHED_DEBUG
975 # include <linux/static_key.h>
976 # define const_debug __read_mostly
977 #else
978 # define const_debug const
979 #endif
980 
981 extern const_debug unsigned int sysctl_sched_features;
982 
983 #define SCHED_FEAT(name, enabled)	\
984 	__SCHED_FEAT_##name ,
985 
986 enum {
987 #include "features.h"
988 	__SCHED_FEAT_NR,
989 };
990 
991 #undef SCHED_FEAT
992 
993 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
994 #define SCHED_FEAT(name, enabled)					\
995 static __always_inline bool static_branch_##name(struct static_key *key) \
996 {									\
997 	return static_key_##enabled(key);				\
998 }
999 
1000 #include "features.h"
1001 
1002 #undef SCHED_FEAT
1003 
1004 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1005 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1006 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1007 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1008 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1009 
1010 extern struct static_key_false sched_numa_balancing;
1011 
1012 static inline u64 global_rt_period(void)
1013 {
1014 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1015 }
1016 
1017 static inline u64 global_rt_runtime(void)
1018 {
1019 	if (sysctl_sched_rt_runtime < 0)
1020 		return RUNTIME_INF;
1021 
1022 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1023 }
1024 
1025 static inline int task_current(struct rq *rq, struct task_struct *p)
1026 {
1027 	return rq->curr == p;
1028 }
1029 
1030 static inline int task_running(struct rq *rq, struct task_struct *p)
1031 {
1032 #ifdef CONFIG_SMP
1033 	return p->on_cpu;
1034 #else
1035 	return task_current(rq, p);
1036 #endif
1037 }
1038 
1039 static inline int task_on_rq_queued(struct task_struct *p)
1040 {
1041 	return p->on_rq == TASK_ON_RQ_QUEUED;
1042 }
1043 
1044 static inline int task_on_rq_migrating(struct task_struct *p)
1045 {
1046 	return p->on_rq == TASK_ON_RQ_MIGRATING;
1047 }
1048 
1049 #ifndef prepare_arch_switch
1050 # define prepare_arch_switch(next)	do { } while (0)
1051 #endif
1052 #ifndef finish_arch_post_lock_switch
1053 # define finish_arch_post_lock_switch()	do { } while (0)
1054 #endif
1055 
1056 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1057 {
1058 #ifdef CONFIG_SMP
1059 	/*
1060 	 * We can optimise this out completely for !SMP, because the
1061 	 * SMP rebalancing from interrupt is the only thing that cares
1062 	 * here.
1063 	 */
1064 	next->on_cpu = 1;
1065 #endif
1066 }
1067 
1068 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1069 {
1070 #ifdef CONFIG_SMP
1071 	/*
1072 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1073 	 * We must ensure this doesn't happen until the switch is completely
1074 	 * finished.
1075 	 *
1076 	 * In particular, the load of prev->state in finish_task_switch() must
1077 	 * happen before this.
1078 	 *
1079 	 * Pairs with the control dependency and rmb in try_to_wake_up().
1080 	 */
1081 	smp_store_release(&prev->on_cpu, 0);
1082 #endif
1083 #ifdef CONFIG_DEBUG_SPINLOCK
1084 	/* this is a valid case when another task releases the spinlock */
1085 	rq->lock.owner = current;
1086 #endif
1087 	/*
1088 	 * If we are tracking spinlock dependencies then we have to
1089 	 * fix up the runqueue lock - which gets 'carried over' from
1090 	 * prev into current:
1091 	 */
1092 	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1093 
1094 	raw_spin_unlock_irq(&rq->lock);
1095 }
1096 
1097 /*
1098  * wake flags
1099  */
1100 #define WF_SYNC		0x01		/* waker goes to sleep after wakeup */
1101 #define WF_FORK		0x02		/* child wakeup after fork */
1102 #define WF_MIGRATED	0x4		/* internal use, task got migrated */
1103 
1104 /*
1105  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1106  * of tasks with abnormal "nice" values across CPUs the contribution that
1107  * each task makes to its run queue's load is weighted according to its
1108  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1109  * scaled version of the new time slice allocation that they receive on time
1110  * slice expiry etc.
1111  */
1112 
1113 #define WEIGHT_IDLEPRIO                3
1114 #define WMULT_IDLEPRIO         1431655765
1115 
1116 /*
1117  * Nice levels are multiplicative, with a gentle 10% change for every
1118  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1119  * nice 1, it will get ~10% less CPU time than another CPU-bound task
1120  * that remained on nice 0.
1121  *
1122  * The "10% effect" is relative and cumulative: from _any_ nice level,
1123  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1124  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1125  * If a task goes up by ~10% and another task goes down by ~10% then
1126  * the relative distance between them is ~25%.)
1127  */
1128 static const int prio_to_weight[40] = {
1129  /* -20 */     88761,     71755,     56483,     46273,     36291,
1130  /* -15 */     29154,     23254,     18705,     14949,     11916,
1131  /* -10 */      9548,      7620,      6100,      4904,      3906,
1132  /*  -5 */      3121,      2501,      1991,      1586,      1277,
1133  /*   0 */      1024,       820,       655,       526,       423,
1134  /*   5 */       335,       272,       215,       172,       137,
1135  /*  10 */       110,        87,        70,        56,        45,
1136  /*  15 */        36,        29,        23,        18,        15,
1137 };
1138 
1139 /*
1140  * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1141  *
1142  * In cases where the weight does not change often, we can use the
1143  * precalculated inverse to speed up arithmetics by turning divisions
1144  * into multiplications:
1145  */
1146 static const u32 prio_to_wmult[40] = {
1147  /* -20 */     48388,     59856,     76040,     92818,    118348,
1148  /* -15 */    147320,    184698,    229616,    287308,    360437,
1149  /* -10 */    449829,    563644,    704093,    875809,   1099582,
1150  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
1151  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
1152  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
1153  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
1154  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1155 };
1156 
1157 #define ENQUEUE_WAKEUP		0x01
1158 #define ENQUEUE_HEAD		0x02
1159 #ifdef CONFIG_SMP
1160 #define ENQUEUE_WAKING		0x04	/* sched_class::task_waking was called */
1161 #else
1162 #define ENQUEUE_WAKING		0x00
1163 #endif
1164 #define ENQUEUE_REPLENISH	0x08
1165 #define ENQUEUE_RESTORE	0x10
1166 
1167 #define DEQUEUE_SLEEP		0x01
1168 #define DEQUEUE_SAVE		0x02
1169 
1170 #define RETRY_TASK		((void *)-1UL)
1171 
1172 struct sched_class {
1173 	const struct sched_class *next;
1174 
1175 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1176 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1177 	void (*yield_task) (struct rq *rq);
1178 	bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1179 
1180 	void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1181 
1182 	/*
1183 	 * It is the responsibility of the pick_next_task() method that will
1184 	 * return the next task to call put_prev_task() on the @prev task or
1185 	 * something equivalent.
1186 	 *
1187 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1188 	 * tasks.
1189 	 */
1190 	struct task_struct * (*pick_next_task) (struct rq *rq,
1191 						struct task_struct *prev);
1192 	void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1193 
1194 #ifdef CONFIG_SMP
1195 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1196 	void (*migrate_task_rq)(struct task_struct *p);
1197 
1198 	void (*task_waking) (struct task_struct *task);
1199 	void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1200 
1201 	void (*set_cpus_allowed)(struct task_struct *p,
1202 				 const struct cpumask *newmask);
1203 
1204 	void (*rq_online)(struct rq *rq);
1205 	void (*rq_offline)(struct rq *rq);
1206 #endif
1207 
1208 	void (*set_curr_task) (struct rq *rq);
1209 	void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1210 	void (*task_fork) (struct task_struct *p);
1211 	void (*task_dead) (struct task_struct *p);
1212 
1213 	/*
1214 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1215 	 * cannot assume the switched_from/switched_to pair is serliazed by
1216 	 * rq->lock. They are however serialized by p->pi_lock.
1217 	 */
1218 	void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1219 	void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1220 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1221 			     int oldprio);
1222 
1223 	unsigned int (*get_rr_interval) (struct rq *rq,
1224 					 struct task_struct *task);
1225 
1226 	void (*update_curr) (struct rq *rq);
1227 
1228 #ifdef CONFIG_FAIR_GROUP_SCHED
1229 	void (*task_move_group) (struct task_struct *p);
1230 #endif
1231 };
1232 
1233 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1234 {
1235 	prev->sched_class->put_prev_task(rq, prev);
1236 }
1237 
1238 #define sched_class_highest (&stop_sched_class)
1239 #define for_each_class(class) \
1240    for (class = sched_class_highest; class; class = class->next)
1241 
1242 extern const struct sched_class stop_sched_class;
1243 extern const struct sched_class dl_sched_class;
1244 extern const struct sched_class rt_sched_class;
1245 extern const struct sched_class fair_sched_class;
1246 extern const struct sched_class idle_sched_class;
1247 
1248 
1249 #ifdef CONFIG_SMP
1250 
1251 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1252 
1253 extern void trigger_load_balance(struct rq *rq);
1254 
1255 extern void idle_enter_fair(struct rq *this_rq);
1256 extern void idle_exit_fair(struct rq *this_rq);
1257 
1258 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1259 
1260 #else
1261 
1262 static inline void idle_enter_fair(struct rq *rq) { }
1263 static inline void idle_exit_fair(struct rq *rq) { }
1264 
1265 #endif
1266 
1267 #ifdef CONFIG_CPU_IDLE
1268 static inline void idle_set_state(struct rq *rq,
1269 				  struct cpuidle_state *idle_state)
1270 {
1271 	rq->idle_state = idle_state;
1272 }
1273 
1274 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1275 {
1276 	WARN_ON(!rcu_read_lock_held());
1277 	return rq->idle_state;
1278 }
1279 #else
1280 static inline void idle_set_state(struct rq *rq,
1281 				  struct cpuidle_state *idle_state)
1282 {
1283 }
1284 
1285 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1286 {
1287 	return NULL;
1288 }
1289 #endif
1290 
1291 extern void sysrq_sched_debug_show(void);
1292 extern void sched_init_granularity(void);
1293 extern void update_max_interval(void);
1294 
1295 extern void init_sched_dl_class(void);
1296 extern void init_sched_rt_class(void);
1297 extern void init_sched_fair_class(void);
1298 
1299 extern void resched_curr(struct rq *rq);
1300 extern void resched_cpu(int cpu);
1301 
1302 extern struct rt_bandwidth def_rt_bandwidth;
1303 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1304 
1305 extern struct dl_bandwidth def_dl_bandwidth;
1306 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1307 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1308 
1309 unsigned long to_ratio(u64 period, u64 runtime);
1310 
1311 extern void init_entity_runnable_average(struct sched_entity *se);
1312 
1313 static inline void add_nr_running(struct rq *rq, unsigned count)
1314 {
1315 	unsigned prev_nr = rq->nr_running;
1316 
1317 	rq->nr_running = prev_nr + count;
1318 
1319 	if (prev_nr < 2 && rq->nr_running >= 2) {
1320 #ifdef CONFIG_SMP
1321 		if (!rq->rd->overload)
1322 			rq->rd->overload = true;
1323 #endif
1324 
1325 #ifdef CONFIG_NO_HZ_FULL
1326 		if (tick_nohz_full_cpu(rq->cpu)) {
1327 			/*
1328 			 * Tick is needed if more than one task runs on a CPU.
1329 			 * Send the target an IPI to kick it out of nohz mode.
1330 			 *
1331 			 * We assume that IPI implies full memory barrier and the
1332 			 * new value of rq->nr_running is visible on reception
1333 			 * from the target.
1334 			 */
1335 			tick_nohz_full_kick_cpu(rq->cpu);
1336 		}
1337 #endif
1338 	}
1339 }
1340 
1341 static inline void sub_nr_running(struct rq *rq, unsigned count)
1342 {
1343 	rq->nr_running -= count;
1344 }
1345 
1346 static inline void rq_last_tick_reset(struct rq *rq)
1347 {
1348 #ifdef CONFIG_NO_HZ_FULL
1349 	rq->last_sched_tick = jiffies;
1350 #endif
1351 }
1352 
1353 extern void update_rq_clock(struct rq *rq);
1354 
1355 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1356 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1357 
1358 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1359 
1360 extern const_debug unsigned int sysctl_sched_time_avg;
1361 extern const_debug unsigned int sysctl_sched_nr_migrate;
1362 extern const_debug unsigned int sysctl_sched_migration_cost;
1363 
1364 static inline u64 sched_avg_period(void)
1365 {
1366 	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1367 }
1368 
1369 #ifdef CONFIG_SCHED_HRTICK
1370 
1371 /*
1372  * Use hrtick when:
1373  *  - enabled by features
1374  *  - hrtimer is actually high res
1375  */
1376 static inline int hrtick_enabled(struct rq *rq)
1377 {
1378 	if (!sched_feat(HRTICK))
1379 		return 0;
1380 	if (!cpu_active(cpu_of(rq)))
1381 		return 0;
1382 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1383 }
1384 
1385 void hrtick_start(struct rq *rq, u64 delay);
1386 
1387 #else
1388 
1389 static inline int hrtick_enabled(struct rq *rq)
1390 {
1391 	return 0;
1392 }
1393 
1394 #endif /* CONFIG_SCHED_HRTICK */
1395 
1396 #ifdef CONFIG_SMP
1397 extern void sched_avg_update(struct rq *rq);
1398 
1399 #ifndef arch_scale_freq_capacity
1400 static __always_inline
1401 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1402 {
1403 	return SCHED_CAPACITY_SCALE;
1404 }
1405 #endif
1406 
1407 #ifndef arch_scale_cpu_capacity
1408 static __always_inline
1409 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1410 {
1411 	if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1412 		return sd->smt_gain / sd->span_weight;
1413 
1414 	return SCHED_CAPACITY_SCALE;
1415 }
1416 #endif
1417 
1418 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1419 {
1420 	rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1421 	sched_avg_update(rq);
1422 }
1423 #else
1424 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1425 static inline void sched_avg_update(struct rq *rq) { }
1426 #endif
1427 
1428 /*
1429  * __task_rq_lock - lock the rq @p resides on.
1430  */
1431 static inline struct rq *__task_rq_lock(struct task_struct *p)
1432 	__acquires(rq->lock)
1433 {
1434 	struct rq *rq;
1435 
1436 	lockdep_assert_held(&p->pi_lock);
1437 
1438 	for (;;) {
1439 		rq = task_rq(p);
1440 		raw_spin_lock(&rq->lock);
1441 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1442 			lockdep_pin_lock(&rq->lock);
1443 			return rq;
1444 		}
1445 		raw_spin_unlock(&rq->lock);
1446 
1447 		while (unlikely(task_on_rq_migrating(p)))
1448 			cpu_relax();
1449 	}
1450 }
1451 
1452 /*
1453  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1454  */
1455 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1456 	__acquires(p->pi_lock)
1457 	__acquires(rq->lock)
1458 {
1459 	struct rq *rq;
1460 
1461 	for (;;) {
1462 		raw_spin_lock_irqsave(&p->pi_lock, *flags);
1463 		rq = task_rq(p);
1464 		raw_spin_lock(&rq->lock);
1465 		/*
1466 		 *	move_queued_task()		task_rq_lock()
1467 		 *
1468 		 *	ACQUIRE (rq->lock)
1469 		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
1470 		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
1471 		 *	[S] ->cpu = new_cpu		[L] task_rq()
1472 		 *					[L] ->on_rq
1473 		 *	RELEASE (rq->lock)
1474 		 *
1475 		 * If we observe the old cpu in task_rq_lock, the acquire of
1476 		 * the old rq->lock will fully serialize against the stores.
1477 		 *
1478 		 * If we observe the new cpu in task_rq_lock, the acquire will
1479 		 * pair with the WMB to ensure we must then also see migrating.
1480 		 */
1481 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1482 			lockdep_pin_lock(&rq->lock);
1483 			return rq;
1484 		}
1485 		raw_spin_unlock(&rq->lock);
1486 		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1487 
1488 		while (unlikely(task_on_rq_migrating(p)))
1489 			cpu_relax();
1490 	}
1491 }
1492 
1493 static inline void __task_rq_unlock(struct rq *rq)
1494 	__releases(rq->lock)
1495 {
1496 	lockdep_unpin_lock(&rq->lock);
1497 	raw_spin_unlock(&rq->lock);
1498 }
1499 
1500 static inline void
1501 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1502 	__releases(rq->lock)
1503 	__releases(p->pi_lock)
1504 {
1505 	lockdep_unpin_lock(&rq->lock);
1506 	raw_spin_unlock(&rq->lock);
1507 	raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1508 }
1509 
1510 #ifdef CONFIG_SMP
1511 #ifdef CONFIG_PREEMPT
1512 
1513 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1514 
1515 /*
1516  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1517  * way at the expense of forcing extra atomic operations in all
1518  * invocations.  This assures that the double_lock is acquired using the
1519  * same underlying policy as the spinlock_t on this architecture, which
1520  * reduces latency compared to the unfair variant below.  However, it
1521  * also adds more overhead and therefore may reduce throughput.
1522  */
1523 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1524 	__releases(this_rq->lock)
1525 	__acquires(busiest->lock)
1526 	__acquires(this_rq->lock)
1527 {
1528 	raw_spin_unlock(&this_rq->lock);
1529 	double_rq_lock(this_rq, busiest);
1530 
1531 	return 1;
1532 }
1533 
1534 #else
1535 /*
1536  * Unfair double_lock_balance: Optimizes throughput at the expense of
1537  * latency by eliminating extra atomic operations when the locks are
1538  * already in proper order on entry.  This favors lower cpu-ids and will
1539  * grant the double lock to lower cpus over higher ids under contention,
1540  * regardless of entry order into the function.
1541  */
1542 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1543 	__releases(this_rq->lock)
1544 	__acquires(busiest->lock)
1545 	__acquires(this_rq->lock)
1546 {
1547 	int ret = 0;
1548 
1549 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1550 		if (busiest < this_rq) {
1551 			raw_spin_unlock(&this_rq->lock);
1552 			raw_spin_lock(&busiest->lock);
1553 			raw_spin_lock_nested(&this_rq->lock,
1554 					      SINGLE_DEPTH_NESTING);
1555 			ret = 1;
1556 		} else
1557 			raw_spin_lock_nested(&busiest->lock,
1558 					      SINGLE_DEPTH_NESTING);
1559 	}
1560 	return ret;
1561 }
1562 
1563 #endif /* CONFIG_PREEMPT */
1564 
1565 /*
1566  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1567  */
1568 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1569 {
1570 	if (unlikely(!irqs_disabled())) {
1571 		/* printk() doesn't work good under rq->lock */
1572 		raw_spin_unlock(&this_rq->lock);
1573 		BUG_ON(1);
1574 	}
1575 
1576 	return _double_lock_balance(this_rq, busiest);
1577 }
1578 
1579 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1580 	__releases(busiest->lock)
1581 {
1582 	raw_spin_unlock(&busiest->lock);
1583 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1584 }
1585 
1586 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1587 {
1588 	if (l1 > l2)
1589 		swap(l1, l2);
1590 
1591 	spin_lock(l1);
1592 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1593 }
1594 
1595 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1596 {
1597 	if (l1 > l2)
1598 		swap(l1, l2);
1599 
1600 	spin_lock_irq(l1);
1601 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1602 }
1603 
1604 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1605 {
1606 	if (l1 > l2)
1607 		swap(l1, l2);
1608 
1609 	raw_spin_lock(l1);
1610 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1611 }
1612 
1613 /*
1614  * double_rq_lock - safely lock two runqueues
1615  *
1616  * Note this does not disable interrupts like task_rq_lock,
1617  * you need to do so manually before calling.
1618  */
1619 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1620 	__acquires(rq1->lock)
1621 	__acquires(rq2->lock)
1622 {
1623 	BUG_ON(!irqs_disabled());
1624 	if (rq1 == rq2) {
1625 		raw_spin_lock(&rq1->lock);
1626 		__acquire(rq2->lock);	/* Fake it out ;) */
1627 	} else {
1628 		if (rq1 < rq2) {
1629 			raw_spin_lock(&rq1->lock);
1630 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1631 		} else {
1632 			raw_spin_lock(&rq2->lock);
1633 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1634 		}
1635 	}
1636 }
1637 
1638 /*
1639  * double_rq_unlock - safely unlock two runqueues
1640  *
1641  * Note this does not restore interrupts like task_rq_unlock,
1642  * you need to do so manually after calling.
1643  */
1644 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1645 	__releases(rq1->lock)
1646 	__releases(rq2->lock)
1647 {
1648 	raw_spin_unlock(&rq1->lock);
1649 	if (rq1 != rq2)
1650 		raw_spin_unlock(&rq2->lock);
1651 	else
1652 		__release(rq2->lock);
1653 }
1654 
1655 #else /* CONFIG_SMP */
1656 
1657 /*
1658  * double_rq_lock - safely lock two runqueues
1659  *
1660  * Note this does not disable interrupts like task_rq_lock,
1661  * you need to do so manually before calling.
1662  */
1663 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1664 	__acquires(rq1->lock)
1665 	__acquires(rq2->lock)
1666 {
1667 	BUG_ON(!irqs_disabled());
1668 	BUG_ON(rq1 != rq2);
1669 	raw_spin_lock(&rq1->lock);
1670 	__acquire(rq2->lock);	/* Fake it out ;) */
1671 }
1672 
1673 /*
1674  * double_rq_unlock - safely unlock two runqueues
1675  *
1676  * Note this does not restore interrupts like task_rq_unlock,
1677  * you need to do so manually after calling.
1678  */
1679 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1680 	__releases(rq1->lock)
1681 	__releases(rq2->lock)
1682 {
1683 	BUG_ON(rq1 != rq2);
1684 	raw_spin_unlock(&rq1->lock);
1685 	__release(rq2->lock);
1686 }
1687 
1688 #endif
1689 
1690 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1691 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1692 
1693 #ifdef	CONFIG_SCHED_DEBUG
1694 extern void print_cfs_stats(struct seq_file *m, int cpu);
1695 extern void print_rt_stats(struct seq_file *m, int cpu);
1696 extern void print_dl_stats(struct seq_file *m, int cpu);
1697 extern void
1698 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1699 
1700 #ifdef CONFIG_NUMA_BALANCING
1701 extern void
1702 show_numa_stats(struct task_struct *p, struct seq_file *m);
1703 extern void
1704 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1705 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
1706 #endif /* CONFIG_NUMA_BALANCING */
1707 #endif /* CONFIG_SCHED_DEBUG */
1708 
1709 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1710 extern void init_rt_rq(struct rt_rq *rt_rq);
1711 extern void init_dl_rq(struct dl_rq *dl_rq);
1712 
1713 extern void cfs_bandwidth_usage_inc(void);
1714 extern void cfs_bandwidth_usage_dec(void);
1715 
1716 #ifdef CONFIG_NO_HZ_COMMON
1717 enum rq_nohz_flag_bits {
1718 	NOHZ_TICK_STOPPED,
1719 	NOHZ_BALANCE_KICK,
1720 };
1721 
1722 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
1723 #endif
1724 
1725 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1726 
1727 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1728 DECLARE_PER_CPU(u64, cpu_softirq_time);
1729 
1730 #ifndef CONFIG_64BIT
1731 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1732 
1733 static inline void irq_time_write_begin(void)
1734 {
1735 	__this_cpu_inc(irq_time_seq.sequence);
1736 	smp_wmb();
1737 }
1738 
1739 static inline void irq_time_write_end(void)
1740 {
1741 	smp_wmb();
1742 	__this_cpu_inc(irq_time_seq.sequence);
1743 }
1744 
1745 static inline u64 irq_time_read(int cpu)
1746 {
1747 	u64 irq_time;
1748 	unsigned seq;
1749 
1750 	do {
1751 		seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1752 		irq_time = per_cpu(cpu_softirq_time, cpu) +
1753 			   per_cpu(cpu_hardirq_time, cpu);
1754 	} while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1755 
1756 	return irq_time;
1757 }
1758 #else /* CONFIG_64BIT */
1759 static inline void irq_time_write_begin(void)
1760 {
1761 }
1762 
1763 static inline void irq_time_write_end(void)
1764 {
1765 }
1766 
1767 static inline u64 irq_time_read(int cpu)
1768 {
1769 	return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1770 }
1771 #endif /* CONFIG_64BIT */
1772 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1773