xref: /openbmc/linux/kernel/sched/sched.h (revision 4bce6fce)
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 long calc_load_fold_active(struct rq *this_rq);
30 extern void update_cpu_load_active(struct rq *this_rq);
31 
32 /*
33  * Helpers for converting nanosecond timing to jiffy resolution
34  */
35 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
36 
37 /*
38  * Increase resolution of nice-level calculations for 64-bit architectures.
39  * The extra resolution improves shares distribution and load balancing of
40  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
41  * hierarchies, especially on larger systems. This is not a user-visible change
42  * and does not change the user-interface for setting shares/weights.
43  *
44  * We increase resolution only if we have enough bits to allow this increased
45  * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
46  * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
47  * increased costs.
48  */
49 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load  */
50 # define SCHED_LOAD_RESOLUTION	10
51 # define scale_load(w)		((w) << SCHED_LOAD_RESOLUTION)
52 # define scale_load_down(w)	((w) >> SCHED_LOAD_RESOLUTION)
53 #else
54 # define SCHED_LOAD_RESOLUTION	0
55 # define scale_load(w)		(w)
56 # define scale_load_down(w)	(w)
57 #endif
58 
59 #define SCHED_LOAD_SHIFT	(10 + SCHED_LOAD_RESOLUTION)
60 #define SCHED_LOAD_SCALE	(1L << SCHED_LOAD_SHIFT)
61 
62 #define NICE_0_LOAD		SCHED_LOAD_SCALE
63 #define NICE_0_SHIFT		SCHED_LOAD_SHIFT
64 
65 /*
66  * Single value that decides SCHED_DEADLINE internal math precision.
67  * 10 -> just above 1us
68  * 9  -> just above 0.5us
69  */
70 #define DL_SCALE (10)
71 
72 /*
73  * These are the 'tuning knobs' of the scheduler:
74  */
75 
76 /*
77  * single value that denotes runtime == period, ie unlimited time.
78  */
79 #define RUNTIME_INF	((u64)~0ULL)
80 
81 static inline int fair_policy(int policy)
82 {
83 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
84 }
85 
86 static inline int rt_policy(int policy)
87 {
88 	return policy == SCHED_FIFO || policy == SCHED_RR;
89 }
90 
91 static inline int dl_policy(int policy)
92 {
93 	return policy == SCHED_DEADLINE;
94 }
95 
96 static inline int task_has_rt_policy(struct task_struct *p)
97 {
98 	return rt_policy(p->policy);
99 }
100 
101 static inline int task_has_dl_policy(struct task_struct *p)
102 {
103 	return dl_policy(p->policy);
104 }
105 
106 static inline bool dl_time_before(u64 a, u64 b)
107 {
108 	return (s64)(a - b) < 0;
109 }
110 
111 /*
112  * Tells if entity @a should preempt entity @b.
113  */
114 static inline bool
115 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
116 {
117 	return dl_time_before(a->deadline, b->deadline);
118 }
119 
120 /*
121  * This is the priority-queue data structure of the RT scheduling class:
122  */
123 struct rt_prio_array {
124 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
125 	struct list_head queue[MAX_RT_PRIO];
126 };
127 
128 struct rt_bandwidth {
129 	/* nests inside the rq lock: */
130 	raw_spinlock_t		rt_runtime_lock;
131 	ktime_t			rt_period;
132 	u64			rt_runtime;
133 	struct hrtimer		rt_period_timer;
134 };
135 
136 void __dl_clear_params(struct task_struct *p);
137 
138 /*
139  * To keep the bandwidth of -deadline tasks and groups under control
140  * we need some place where:
141  *  - store the maximum -deadline bandwidth of the system (the group);
142  *  - cache the fraction of that bandwidth that is currently allocated.
143  *
144  * This is all done in the data structure below. It is similar to the
145  * one used for RT-throttling (rt_bandwidth), with the main difference
146  * that, since here we are only interested in admission control, we
147  * do not decrease any runtime while the group "executes", neither we
148  * need a timer to replenish it.
149  *
150  * With respect to SMP, the bandwidth is given on a per-CPU basis,
151  * meaning that:
152  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
153  *  - dl_total_bw array contains, in the i-eth element, the currently
154  *    allocated bandwidth on the i-eth CPU.
155  * Moreover, groups consume bandwidth on each CPU, while tasks only
156  * consume bandwidth on the CPU they're running on.
157  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
158  * that will be shown the next time the proc or cgroup controls will
159  * be red. It on its turn can be changed by writing on its own
160  * control.
161  */
162 struct dl_bandwidth {
163 	raw_spinlock_t dl_runtime_lock;
164 	u64 dl_runtime;
165 	u64 dl_period;
166 };
167 
168 static inline int dl_bandwidth_enabled(void)
169 {
170 	return sysctl_sched_rt_runtime >= 0;
171 }
172 
173 extern struct dl_bw *dl_bw_of(int i);
174 
175 struct dl_bw {
176 	raw_spinlock_t lock;
177 	u64 bw, total_bw;
178 };
179 
180 static inline
181 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
182 {
183 	dl_b->total_bw -= tsk_bw;
184 }
185 
186 static inline
187 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
188 {
189 	dl_b->total_bw += tsk_bw;
190 }
191 
192 static inline
193 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
194 {
195 	return dl_b->bw != -1 &&
196 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
197 }
198 
199 extern struct mutex sched_domains_mutex;
200 
201 #ifdef CONFIG_CGROUP_SCHED
202 
203 #include <linux/cgroup.h>
204 
205 struct cfs_rq;
206 struct rt_rq;
207 
208 extern struct list_head task_groups;
209 
210 struct cfs_bandwidth {
211 #ifdef CONFIG_CFS_BANDWIDTH
212 	raw_spinlock_t lock;
213 	ktime_t period;
214 	u64 quota, runtime;
215 	s64 hierarchical_quota;
216 	u64 runtime_expires;
217 
218 	int idle, timer_active;
219 	struct hrtimer period_timer, slack_timer;
220 	struct list_head throttled_cfs_rq;
221 
222 	/* statistics */
223 	int nr_periods, nr_throttled;
224 	u64 throttled_time;
225 #endif
226 };
227 
228 /* task group related information */
229 struct task_group {
230 	struct cgroup_subsys_state css;
231 
232 #ifdef CONFIG_FAIR_GROUP_SCHED
233 	/* schedulable entities of this group on each cpu */
234 	struct sched_entity **se;
235 	/* runqueue "owned" by this group on each cpu */
236 	struct cfs_rq **cfs_rq;
237 	unsigned long shares;
238 
239 #ifdef	CONFIG_SMP
240 	atomic_long_t load_avg;
241 	atomic_t runnable_avg;
242 #endif
243 #endif
244 
245 #ifdef CONFIG_RT_GROUP_SCHED
246 	struct sched_rt_entity **rt_se;
247 	struct rt_rq **rt_rq;
248 
249 	struct rt_bandwidth rt_bandwidth;
250 #endif
251 
252 	struct rcu_head rcu;
253 	struct list_head list;
254 
255 	struct task_group *parent;
256 	struct list_head siblings;
257 	struct list_head children;
258 
259 #ifdef CONFIG_SCHED_AUTOGROUP
260 	struct autogroup *autogroup;
261 #endif
262 
263 	struct cfs_bandwidth cfs_bandwidth;
264 };
265 
266 #ifdef CONFIG_FAIR_GROUP_SCHED
267 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
268 
269 /*
270  * A weight of 0 or 1 can cause arithmetics problems.
271  * A weight of a cfs_rq is the sum of weights of which entities
272  * are queued on this cfs_rq, so a weight of a entity should not be
273  * too large, so as the shares value of a task group.
274  * (The default weight is 1024 - so there's no practical
275  *  limitation from this.)
276  */
277 #define MIN_SHARES	(1UL <<  1)
278 #define MAX_SHARES	(1UL << 18)
279 #endif
280 
281 typedef int (*tg_visitor)(struct task_group *, void *);
282 
283 extern int walk_tg_tree_from(struct task_group *from,
284 			     tg_visitor down, tg_visitor up, void *data);
285 
286 /*
287  * Iterate the full tree, calling @down when first entering a node and @up when
288  * leaving it for the final time.
289  *
290  * Caller must hold rcu_lock or sufficient equivalent.
291  */
292 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
293 {
294 	return walk_tg_tree_from(&root_task_group, down, up, data);
295 }
296 
297 extern int tg_nop(struct task_group *tg, void *data);
298 
299 extern void free_fair_sched_group(struct task_group *tg);
300 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
301 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
302 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
303 			struct sched_entity *se, int cpu,
304 			struct sched_entity *parent);
305 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
306 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
307 
308 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
309 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force);
310 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
311 
312 extern void free_rt_sched_group(struct task_group *tg);
313 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
314 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
315 		struct sched_rt_entity *rt_se, int cpu,
316 		struct sched_rt_entity *parent);
317 
318 extern struct task_group *sched_create_group(struct task_group *parent);
319 extern void sched_online_group(struct task_group *tg,
320 			       struct task_group *parent);
321 extern void sched_destroy_group(struct task_group *tg);
322 extern void sched_offline_group(struct task_group *tg);
323 
324 extern void sched_move_task(struct task_struct *tsk);
325 
326 #ifdef CONFIG_FAIR_GROUP_SCHED
327 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
328 #endif
329 
330 #else /* CONFIG_CGROUP_SCHED */
331 
332 struct cfs_bandwidth { };
333 
334 #endif	/* CONFIG_CGROUP_SCHED */
335 
336 /* CFS-related fields in a runqueue */
337 struct cfs_rq {
338 	struct load_weight load;
339 	unsigned int nr_running, h_nr_running;
340 
341 	u64 exec_clock;
342 	u64 min_vruntime;
343 #ifndef CONFIG_64BIT
344 	u64 min_vruntime_copy;
345 #endif
346 
347 	struct rb_root tasks_timeline;
348 	struct rb_node *rb_leftmost;
349 
350 	/*
351 	 * 'curr' points to currently running entity on this cfs_rq.
352 	 * It is set to NULL otherwise (i.e when none are currently running).
353 	 */
354 	struct sched_entity *curr, *next, *last, *skip;
355 
356 #ifdef	CONFIG_SCHED_DEBUG
357 	unsigned int nr_spread_over;
358 #endif
359 
360 #ifdef CONFIG_SMP
361 	/*
362 	 * CFS Load tracking
363 	 * Under CFS, load is tracked on a per-entity basis and aggregated up.
364 	 * This allows for the description of both thread and group usage (in
365 	 * the FAIR_GROUP_SCHED case).
366 	 * runnable_load_avg is the sum of the load_avg_contrib of the
367 	 * sched_entities on the rq.
368 	 * blocked_load_avg is similar to runnable_load_avg except that its
369 	 * the blocked sched_entities on the rq.
370 	 * utilization_load_avg is the sum of the average running time of the
371 	 * sched_entities on the rq.
372 	 */
373 	unsigned long runnable_load_avg, blocked_load_avg, utilization_load_avg;
374 	atomic64_t decay_counter;
375 	u64 last_decay;
376 	atomic_long_t removed_load;
377 
378 #ifdef CONFIG_FAIR_GROUP_SCHED
379 	/* Required to track per-cpu representation of a task_group */
380 	u32 tg_runnable_contrib;
381 	unsigned long tg_load_contrib;
382 
383 	/*
384 	 *   h_load = weight * f(tg)
385 	 *
386 	 * Where f(tg) is the recursive weight fraction assigned to
387 	 * this group.
388 	 */
389 	unsigned long h_load;
390 	u64 last_h_load_update;
391 	struct sched_entity *h_load_next;
392 #endif /* CONFIG_FAIR_GROUP_SCHED */
393 #endif /* CONFIG_SMP */
394 
395 #ifdef CONFIG_FAIR_GROUP_SCHED
396 	struct rq *rq;	/* cpu runqueue to which this cfs_rq is attached */
397 
398 	/*
399 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
400 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
401 	 * (like users, containers etc.)
402 	 *
403 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
404 	 * list is used during load balance.
405 	 */
406 	int on_list;
407 	struct list_head leaf_cfs_rq_list;
408 	struct task_group *tg;	/* group that "owns" this runqueue */
409 
410 #ifdef CONFIG_CFS_BANDWIDTH
411 	int runtime_enabled;
412 	u64 runtime_expires;
413 	s64 runtime_remaining;
414 
415 	u64 throttled_clock, throttled_clock_task;
416 	u64 throttled_clock_task_time;
417 	int throttled, throttle_count;
418 	struct list_head throttled_list;
419 #endif /* CONFIG_CFS_BANDWIDTH */
420 #endif /* CONFIG_FAIR_GROUP_SCHED */
421 };
422 
423 static inline int rt_bandwidth_enabled(void)
424 {
425 	return sysctl_sched_rt_runtime >= 0;
426 }
427 
428 /* RT IPI pull logic requires IRQ_WORK */
429 #ifdef CONFIG_IRQ_WORK
430 # define HAVE_RT_PUSH_IPI
431 #endif
432 
433 /* Real-Time classes' related field in a runqueue: */
434 struct rt_rq {
435 	struct rt_prio_array active;
436 	unsigned int rt_nr_running;
437 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
438 	struct {
439 		int curr; /* highest queued rt task prio */
440 #ifdef CONFIG_SMP
441 		int next; /* next highest */
442 #endif
443 	} highest_prio;
444 #endif
445 #ifdef CONFIG_SMP
446 	unsigned long rt_nr_migratory;
447 	unsigned long rt_nr_total;
448 	int overloaded;
449 	struct plist_head pushable_tasks;
450 #ifdef HAVE_RT_PUSH_IPI
451 	int push_flags;
452 	int push_cpu;
453 	struct irq_work push_work;
454 	raw_spinlock_t push_lock;
455 #endif
456 #endif /* CONFIG_SMP */
457 	int rt_queued;
458 
459 	int rt_throttled;
460 	u64 rt_time;
461 	u64 rt_runtime;
462 	/* Nests inside the rq lock: */
463 	raw_spinlock_t rt_runtime_lock;
464 
465 #ifdef CONFIG_RT_GROUP_SCHED
466 	unsigned long rt_nr_boosted;
467 
468 	struct rq *rq;
469 	struct task_group *tg;
470 #endif
471 };
472 
473 /* Deadline class' related fields in a runqueue */
474 struct dl_rq {
475 	/* runqueue is an rbtree, ordered by deadline */
476 	struct rb_root rb_root;
477 	struct rb_node *rb_leftmost;
478 
479 	unsigned long dl_nr_running;
480 
481 #ifdef CONFIG_SMP
482 	/*
483 	 * Deadline values of the currently executing and the
484 	 * earliest ready task on this rq. Caching these facilitates
485 	 * the decision wether or not a ready but not running task
486 	 * should migrate somewhere else.
487 	 */
488 	struct {
489 		u64 curr;
490 		u64 next;
491 	} earliest_dl;
492 
493 	unsigned long dl_nr_migratory;
494 	int overloaded;
495 
496 	/*
497 	 * Tasks on this rq that can be pushed away. They are kept in
498 	 * an rb-tree, ordered by tasks' deadlines, with caching
499 	 * of the leftmost (earliest deadline) element.
500 	 */
501 	struct rb_root pushable_dl_tasks_root;
502 	struct rb_node *pushable_dl_tasks_leftmost;
503 #else
504 	struct dl_bw dl_bw;
505 #endif
506 };
507 
508 #ifdef CONFIG_SMP
509 
510 /*
511  * We add the notion of a root-domain which will be used to define per-domain
512  * variables. Each exclusive cpuset essentially defines an island domain by
513  * fully partitioning the member cpus from any other cpuset. Whenever a new
514  * exclusive cpuset is created, we also create and attach a new root-domain
515  * object.
516  *
517  */
518 struct root_domain {
519 	atomic_t refcount;
520 	atomic_t rto_count;
521 	struct rcu_head rcu;
522 	cpumask_var_t span;
523 	cpumask_var_t online;
524 
525 	/* Indicate more than one runnable task for any CPU */
526 	bool overload;
527 
528 	/*
529 	 * The bit corresponding to a CPU gets set here if such CPU has more
530 	 * than one runnable -deadline task (as it is below for RT tasks).
531 	 */
532 	cpumask_var_t dlo_mask;
533 	atomic_t dlo_count;
534 	struct dl_bw dl_bw;
535 	struct cpudl cpudl;
536 
537 	/*
538 	 * The "RT overload" flag: it gets set if a CPU has more than
539 	 * one runnable RT task.
540 	 */
541 	cpumask_var_t rto_mask;
542 	struct cpupri cpupri;
543 };
544 
545 extern struct root_domain def_root_domain;
546 
547 #endif /* CONFIG_SMP */
548 
549 /*
550  * This is the main, per-CPU runqueue data structure.
551  *
552  * Locking rule: those places that want to lock multiple runqueues
553  * (such as the load balancing or the thread migration code), lock
554  * acquire operations must be ordered by ascending &runqueue.
555  */
556 struct rq {
557 	/* runqueue lock: */
558 	raw_spinlock_t lock;
559 
560 	/*
561 	 * nr_running and cpu_load should be in the same cacheline because
562 	 * remote CPUs use both these fields when doing load calculation.
563 	 */
564 	unsigned int nr_running;
565 #ifdef CONFIG_NUMA_BALANCING
566 	unsigned int nr_numa_running;
567 	unsigned int nr_preferred_running;
568 #endif
569 	#define CPU_LOAD_IDX_MAX 5
570 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
571 	unsigned long last_load_update_tick;
572 #ifdef CONFIG_NO_HZ_COMMON
573 	u64 nohz_stamp;
574 	unsigned long nohz_flags;
575 #endif
576 #ifdef CONFIG_NO_HZ_FULL
577 	unsigned long last_sched_tick;
578 #endif
579 	/* capture load from *all* tasks on this cpu: */
580 	struct load_weight load;
581 	unsigned long nr_load_updates;
582 	u64 nr_switches;
583 
584 	struct cfs_rq cfs;
585 	struct rt_rq rt;
586 	struct dl_rq dl;
587 
588 #ifdef CONFIG_FAIR_GROUP_SCHED
589 	/* list of leaf cfs_rq on this cpu: */
590 	struct list_head leaf_cfs_rq_list;
591 
592 	struct sched_avg avg;
593 #endif /* CONFIG_FAIR_GROUP_SCHED */
594 
595 	/*
596 	 * This is part of a global counter where only the total sum
597 	 * over all CPUs matters. A task can increase this counter on
598 	 * one CPU and if it got migrated afterwards it may decrease
599 	 * it on another CPU. Always updated under the runqueue lock:
600 	 */
601 	unsigned long nr_uninterruptible;
602 
603 	struct task_struct *curr, *idle, *stop;
604 	unsigned long next_balance;
605 	struct mm_struct *prev_mm;
606 
607 	unsigned int clock_skip_update;
608 	u64 clock;
609 	u64 clock_task;
610 
611 	atomic_t nr_iowait;
612 
613 #ifdef CONFIG_SMP
614 	struct root_domain *rd;
615 	struct sched_domain *sd;
616 
617 	unsigned long cpu_capacity;
618 	unsigned long cpu_capacity_orig;
619 
620 	unsigned char idle_balance;
621 	/* For active balancing */
622 	int post_schedule;
623 	int active_balance;
624 	int push_cpu;
625 	struct cpu_stop_work active_balance_work;
626 	/* cpu of this runqueue: */
627 	int cpu;
628 	int online;
629 
630 	struct list_head cfs_tasks;
631 
632 	u64 rt_avg;
633 	u64 age_stamp;
634 	u64 idle_stamp;
635 	u64 avg_idle;
636 
637 	/* This is used to determine avg_idle's max value */
638 	u64 max_idle_balance_cost;
639 #endif
640 
641 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
642 	u64 prev_irq_time;
643 #endif
644 #ifdef CONFIG_PARAVIRT
645 	u64 prev_steal_time;
646 #endif
647 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
648 	u64 prev_steal_time_rq;
649 #endif
650 
651 	/* calc_load related fields */
652 	unsigned long calc_load_update;
653 	long calc_load_active;
654 
655 #ifdef CONFIG_SCHED_HRTICK
656 #ifdef CONFIG_SMP
657 	int hrtick_csd_pending;
658 	struct call_single_data hrtick_csd;
659 #endif
660 	struct hrtimer hrtick_timer;
661 #endif
662 
663 #ifdef CONFIG_SCHEDSTATS
664 	/* latency stats */
665 	struct sched_info rq_sched_info;
666 	unsigned long long rq_cpu_time;
667 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
668 
669 	/* sys_sched_yield() stats */
670 	unsigned int yld_count;
671 
672 	/* schedule() stats */
673 	unsigned int sched_count;
674 	unsigned int sched_goidle;
675 
676 	/* try_to_wake_up() stats */
677 	unsigned int ttwu_count;
678 	unsigned int ttwu_local;
679 #endif
680 
681 #ifdef CONFIG_SMP
682 	struct llist_head wake_list;
683 #endif
684 
685 #ifdef CONFIG_CPU_IDLE
686 	/* Must be inspected within a rcu lock section */
687 	struct cpuidle_state *idle_state;
688 #endif
689 };
690 
691 static inline int cpu_of(struct rq *rq)
692 {
693 #ifdef CONFIG_SMP
694 	return rq->cpu;
695 #else
696 	return 0;
697 #endif
698 }
699 
700 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
701 
702 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
703 #define this_rq()		this_cpu_ptr(&runqueues)
704 #define task_rq(p)		cpu_rq(task_cpu(p))
705 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
706 #define raw_rq()		raw_cpu_ptr(&runqueues)
707 
708 static inline u64 __rq_clock_broken(struct rq *rq)
709 {
710 	return ACCESS_ONCE(rq->clock);
711 }
712 
713 static inline u64 rq_clock(struct rq *rq)
714 {
715 	lockdep_assert_held(&rq->lock);
716 	return rq->clock;
717 }
718 
719 static inline u64 rq_clock_task(struct rq *rq)
720 {
721 	lockdep_assert_held(&rq->lock);
722 	return rq->clock_task;
723 }
724 
725 #define RQCF_REQ_SKIP	0x01
726 #define RQCF_ACT_SKIP	0x02
727 
728 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
729 {
730 	lockdep_assert_held(&rq->lock);
731 	if (skip)
732 		rq->clock_skip_update |= RQCF_REQ_SKIP;
733 	else
734 		rq->clock_skip_update &= ~RQCF_REQ_SKIP;
735 }
736 
737 #ifdef CONFIG_NUMA
738 enum numa_topology_type {
739 	NUMA_DIRECT,
740 	NUMA_GLUELESS_MESH,
741 	NUMA_BACKPLANE,
742 };
743 extern enum numa_topology_type sched_numa_topology_type;
744 extern int sched_max_numa_distance;
745 extern bool find_numa_distance(int distance);
746 #endif
747 
748 #ifdef CONFIG_NUMA_BALANCING
749 /* The regions in numa_faults array from task_struct */
750 enum numa_faults_stats {
751 	NUMA_MEM = 0,
752 	NUMA_CPU,
753 	NUMA_MEMBUF,
754 	NUMA_CPUBUF
755 };
756 extern void sched_setnuma(struct task_struct *p, int node);
757 extern int migrate_task_to(struct task_struct *p, int cpu);
758 extern int migrate_swap(struct task_struct *, struct task_struct *);
759 #endif /* CONFIG_NUMA_BALANCING */
760 
761 #ifdef CONFIG_SMP
762 
763 extern void sched_ttwu_pending(void);
764 
765 #define rcu_dereference_check_sched_domain(p) \
766 	rcu_dereference_check((p), \
767 			      lockdep_is_held(&sched_domains_mutex))
768 
769 /*
770  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
771  * See detach_destroy_domains: synchronize_sched for details.
772  *
773  * The domain tree of any CPU may only be accessed from within
774  * preempt-disabled sections.
775  */
776 #define for_each_domain(cpu, __sd) \
777 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
778 			__sd; __sd = __sd->parent)
779 
780 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
781 
782 /**
783  * highest_flag_domain - Return highest sched_domain containing flag.
784  * @cpu:	The cpu whose highest level of sched domain is to
785  *		be returned.
786  * @flag:	The flag to check for the highest sched_domain
787  *		for the given cpu.
788  *
789  * Returns the highest sched_domain of a cpu which contains the given flag.
790  */
791 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
792 {
793 	struct sched_domain *sd, *hsd = NULL;
794 
795 	for_each_domain(cpu, sd) {
796 		if (!(sd->flags & flag))
797 			break;
798 		hsd = sd;
799 	}
800 
801 	return hsd;
802 }
803 
804 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
805 {
806 	struct sched_domain *sd;
807 
808 	for_each_domain(cpu, sd) {
809 		if (sd->flags & flag)
810 			break;
811 	}
812 
813 	return sd;
814 }
815 
816 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
817 DECLARE_PER_CPU(int, sd_llc_size);
818 DECLARE_PER_CPU(int, sd_llc_id);
819 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
820 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
821 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
822 
823 struct sched_group_capacity {
824 	atomic_t ref;
825 	/*
826 	 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
827 	 * for a single CPU.
828 	 */
829 	unsigned int capacity;
830 	unsigned long next_update;
831 	int imbalance; /* XXX unrelated to capacity but shared group state */
832 	/*
833 	 * Number of busy cpus in this group.
834 	 */
835 	atomic_t nr_busy_cpus;
836 
837 	unsigned long cpumask[0]; /* iteration mask */
838 };
839 
840 struct sched_group {
841 	struct sched_group *next;	/* Must be a circular list */
842 	atomic_t ref;
843 
844 	unsigned int group_weight;
845 	struct sched_group_capacity *sgc;
846 
847 	/*
848 	 * The CPUs this group covers.
849 	 *
850 	 * NOTE: this field is variable length. (Allocated dynamically
851 	 * by attaching extra space to the end of the structure,
852 	 * depending on how many CPUs the kernel has booted up with)
853 	 */
854 	unsigned long cpumask[0];
855 };
856 
857 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
858 {
859 	return to_cpumask(sg->cpumask);
860 }
861 
862 /*
863  * cpumask masking which cpus in the group are allowed to iterate up the domain
864  * tree.
865  */
866 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
867 {
868 	return to_cpumask(sg->sgc->cpumask);
869 }
870 
871 /**
872  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
873  * @group: The group whose first cpu is to be returned.
874  */
875 static inline unsigned int group_first_cpu(struct sched_group *group)
876 {
877 	return cpumask_first(sched_group_cpus(group));
878 }
879 
880 extern int group_balance_cpu(struct sched_group *sg);
881 
882 #else
883 
884 static inline void sched_ttwu_pending(void) { }
885 
886 #endif /* CONFIG_SMP */
887 
888 #include "stats.h"
889 #include "auto_group.h"
890 
891 #ifdef CONFIG_CGROUP_SCHED
892 
893 /*
894  * Return the group to which this tasks belongs.
895  *
896  * We cannot use task_css() and friends because the cgroup subsystem
897  * changes that value before the cgroup_subsys::attach() method is called,
898  * therefore we cannot pin it and might observe the wrong value.
899  *
900  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
901  * core changes this before calling sched_move_task().
902  *
903  * Instead we use a 'copy' which is updated from sched_move_task() while
904  * holding both task_struct::pi_lock and rq::lock.
905  */
906 static inline struct task_group *task_group(struct task_struct *p)
907 {
908 	return p->sched_task_group;
909 }
910 
911 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
912 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
913 {
914 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
915 	struct task_group *tg = task_group(p);
916 #endif
917 
918 #ifdef CONFIG_FAIR_GROUP_SCHED
919 	p->se.cfs_rq = tg->cfs_rq[cpu];
920 	p->se.parent = tg->se[cpu];
921 #endif
922 
923 #ifdef CONFIG_RT_GROUP_SCHED
924 	p->rt.rt_rq  = tg->rt_rq[cpu];
925 	p->rt.parent = tg->rt_se[cpu];
926 #endif
927 }
928 
929 #else /* CONFIG_CGROUP_SCHED */
930 
931 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
932 static inline struct task_group *task_group(struct task_struct *p)
933 {
934 	return NULL;
935 }
936 
937 #endif /* CONFIG_CGROUP_SCHED */
938 
939 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
940 {
941 	set_task_rq(p, cpu);
942 #ifdef CONFIG_SMP
943 	/*
944 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
945 	 * successfuly executed on another CPU. We must ensure that updates of
946 	 * per-task data have been completed by this moment.
947 	 */
948 	smp_wmb();
949 	task_thread_info(p)->cpu = cpu;
950 	p->wake_cpu = cpu;
951 #endif
952 }
953 
954 /*
955  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
956  */
957 #ifdef CONFIG_SCHED_DEBUG
958 # include <linux/static_key.h>
959 # define const_debug __read_mostly
960 #else
961 # define const_debug const
962 #endif
963 
964 extern const_debug unsigned int sysctl_sched_features;
965 
966 #define SCHED_FEAT(name, enabled)	\
967 	__SCHED_FEAT_##name ,
968 
969 enum {
970 #include "features.h"
971 	__SCHED_FEAT_NR,
972 };
973 
974 #undef SCHED_FEAT
975 
976 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
977 #define SCHED_FEAT(name, enabled)					\
978 static __always_inline bool static_branch_##name(struct static_key *key) \
979 {									\
980 	return static_key_##enabled(key);				\
981 }
982 
983 #include "features.h"
984 
985 #undef SCHED_FEAT
986 
987 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
988 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
989 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
990 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
991 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
992 
993 #ifdef CONFIG_NUMA_BALANCING
994 #define sched_feat_numa(x) sched_feat(x)
995 #ifdef CONFIG_SCHED_DEBUG
996 #define numabalancing_enabled sched_feat_numa(NUMA)
997 #else
998 extern bool numabalancing_enabled;
999 #endif /* CONFIG_SCHED_DEBUG */
1000 #else
1001 #define sched_feat_numa(x) (0)
1002 #define numabalancing_enabled (0)
1003 #endif /* CONFIG_NUMA_BALANCING */
1004 
1005 static inline u64 global_rt_period(void)
1006 {
1007 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1008 }
1009 
1010 static inline u64 global_rt_runtime(void)
1011 {
1012 	if (sysctl_sched_rt_runtime < 0)
1013 		return RUNTIME_INF;
1014 
1015 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1016 }
1017 
1018 static inline int task_current(struct rq *rq, struct task_struct *p)
1019 {
1020 	return rq->curr == p;
1021 }
1022 
1023 static inline int task_running(struct rq *rq, struct task_struct *p)
1024 {
1025 #ifdef CONFIG_SMP
1026 	return p->on_cpu;
1027 #else
1028 	return task_current(rq, p);
1029 #endif
1030 }
1031 
1032 static inline int task_on_rq_queued(struct task_struct *p)
1033 {
1034 	return p->on_rq == TASK_ON_RQ_QUEUED;
1035 }
1036 
1037 static inline int task_on_rq_migrating(struct task_struct *p)
1038 {
1039 	return p->on_rq == TASK_ON_RQ_MIGRATING;
1040 }
1041 
1042 #ifndef prepare_arch_switch
1043 # define prepare_arch_switch(next)	do { } while (0)
1044 #endif
1045 #ifndef finish_arch_switch
1046 # define finish_arch_switch(prev)	do { } while (0)
1047 #endif
1048 #ifndef finish_arch_post_lock_switch
1049 # define finish_arch_post_lock_switch()	do { } while (0)
1050 #endif
1051 
1052 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1053 {
1054 #ifdef CONFIG_SMP
1055 	/*
1056 	 * We can optimise this out completely for !SMP, because the
1057 	 * SMP rebalancing from interrupt is the only thing that cares
1058 	 * here.
1059 	 */
1060 	next->on_cpu = 1;
1061 #endif
1062 }
1063 
1064 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1065 {
1066 #ifdef CONFIG_SMP
1067 	/*
1068 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1069 	 * We must ensure this doesn't happen until the switch is completely
1070 	 * finished.
1071 	 */
1072 	smp_wmb();
1073 	prev->on_cpu = 0;
1074 #endif
1075 #ifdef CONFIG_DEBUG_SPINLOCK
1076 	/* this is a valid case when another task releases the spinlock */
1077 	rq->lock.owner = current;
1078 #endif
1079 	/*
1080 	 * If we are tracking spinlock dependencies then we have to
1081 	 * fix up the runqueue lock - which gets 'carried over' from
1082 	 * prev into current:
1083 	 */
1084 	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1085 
1086 	raw_spin_unlock_irq(&rq->lock);
1087 }
1088 
1089 /*
1090  * wake flags
1091  */
1092 #define WF_SYNC		0x01		/* waker goes to sleep after wakeup */
1093 #define WF_FORK		0x02		/* child wakeup after fork */
1094 #define WF_MIGRATED	0x4		/* internal use, task got migrated */
1095 
1096 /*
1097  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1098  * of tasks with abnormal "nice" values across CPUs the contribution that
1099  * each task makes to its run queue's load is weighted according to its
1100  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1101  * scaled version of the new time slice allocation that they receive on time
1102  * slice expiry etc.
1103  */
1104 
1105 #define WEIGHT_IDLEPRIO                3
1106 #define WMULT_IDLEPRIO         1431655765
1107 
1108 /*
1109  * Nice levels are multiplicative, with a gentle 10% change for every
1110  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1111  * nice 1, it will get ~10% less CPU time than another CPU-bound task
1112  * that remained on nice 0.
1113  *
1114  * The "10% effect" is relative and cumulative: from _any_ nice level,
1115  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1116  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1117  * If a task goes up by ~10% and another task goes down by ~10% then
1118  * the relative distance between them is ~25%.)
1119  */
1120 static const int prio_to_weight[40] = {
1121  /* -20 */     88761,     71755,     56483,     46273,     36291,
1122  /* -15 */     29154,     23254,     18705,     14949,     11916,
1123  /* -10 */      9548,      7620,      6100,      4904,      3906,
1124  /*  -5 */      3121,      2501,      1991,      1586,      1277,
1125  /*   0 */      1024,       820,       655,       526,       423,
1126  /*   5 */       335,       272,       215,       172,       137,
1127  /*  10 */       110,        87,        70,        56,        45,
1128  /*  15 */        36,        29,        23,        18,        15,
1129 };
1130 
1131 /*
1132  * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1133  *
1134  * In cases where the weight does not change often, we can use the
1135  * precalculated inverse to speed up arithmetics by turning divisions
1136  * into multiplications:
1137  */
1138 static const u32 prio_to_wmult[40] = {
1139  /* -20 */     48388,     59856,     76040,     92818,    118348,
1140  /* -15 */    147320,    184698,    229616,    287308,    360437,
1141  /* -10 */    449829,    563644,    704093,    875809,   1099582,
1142  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
1143  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
1144  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
1145  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
1146  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1147 };
1148 
1149 #define ENQUEUE_WAKEUP		1
1150 #define ENQUEUE_HEAD		2
1151 #ifdef CONFIG_SMP
1152 #define ENQUEUE_WAKING		4	/* sched_class::task_waking was called */
1153 #else
1154 #define ENQUEUE_WAKING		0
1155 #endif
1156 #define ENQUEUE_REPLENISH	8
1157 
1158 #define DEQUEUE_SLEEP		1
1159 
1160 #define RETRY_TASK		((void *)-1UL)
1161 
1162 struct sched_class {
1163 	const struct sched_class *next;
1164 
1165 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1166 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1167 	void (*yield_task) (struct rq *rq);
1168 	bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1169 
1170 	void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1171 
1172 	/*
1173 	 * It is the responsibility of the pick_next_task() method that will
1174 	 * return the next task to call put_prev_task() on the @prev task or
1175 	 * something equivalent.
1176 	 *
1177 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1178 	 * tasks.
1179 	 */
1180 	struct task_struct * (*pick_next_task) (struct rq *rq,
1181 						struct task_struct *prev);
1182 	void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1183 
1184 #ifdef CONFIG_SMP
1185 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1186 	void (*migrate_task_rq)(struct task_struct *p, int next_cpu);
1187 
1188 	void (*post_schedule) (struct rq *this_rq);
1189 	void (*task_waking) (struct task_struct *task);
1190 	void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1191 
1192 	void (*set_cpus_allowed)(struct task_struct *p,
1193 				 const struct cpumask *newmask);
1194 
1195 	void (*rq_online)(struct rq *rq);
1196 	void (*rq_offline)(struct rq *rq);
1197 #endif
1198 
1199 	void (*set_curr_task) (struct rq *rq);
1200 	void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1201 	void (*task_fork) (struct task_struct *p);
1202 	void (*task_dead) (struct task_struct *p);
1203 
1204 	/*
1205 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1206 	 * cannot assume the switched_from/switched_to pair is serliazed by
1207 	 * rq->lock. They are however serialized by p->pi_lock.
1208 	 */
1209 	void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1210 	void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1211 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1212 			     int oldprio);
1213 
1214 	unsigned int (*get_rr_interval) (struct rq *rq,
1215 					 struct task_struct *task);
1216 
1217 	void (*update_curr) (struct rq *rq);
1218 
1219 #ifdef CONFIG_FAIR_GROUP_SCHED
1220 	void (*task_move_group) (struct task_struct *p, int on_rq);
1221 #endif
1222 };
1223 
1224 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1225 {
1226 	prev->sched_class->put_prev_task(rq, prev);
1227 }
1228 
1229 #define sched_class_highest (&stop_sched_class)
1230 #define for_each_class(class) \
1231    for (class = sched_class_highest; class; class = class->next)
1232 
1233 extern const struct sched_class stop_sched_class;
1234 extern const struct sched_class dl_sched_class;
1235 extern const struct sched_class rt_sched_class;
1236 extern const struct sched_class fair_sched_class;
1237 extern const struct sched_class idle_sched_class;
1238 
1239 
1240 #ifdef CONFIG_SMP
1241 
1242 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1243 
1244 extern void trigger_load_balance(struct rq *rq);
1245 
1246 extern void idle_enter_fair(struct rq *this_rq);
1247 extern void idle_exit_fair(struct rq *this_rq);
1248 
1249 #else
1250 
1251 static inline void idle_enter_fair(struct rq *rq) { }
1252 static inline void idle_exit_fair(struct rq *rq) { }
1253 
1254 #endif
1255 
1256 #ifdef CONFIG_CPU_IDLE
1257 static inline void idle_set_state(struct rq *rq,
1258 				  struct cpuidle_state *idle_state)
1259 {
1260 	rq->idle_state = idle_state;
1261 }
1262 
1263 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1264 {
1265 	WARN_ON(!rcu_read_lock_held());
1266 	return rq->idle_state;
1267 }
1268 #else
1269 static inline void idle_set_state(struct rq *rq,
1270 				  struct cpuidle_state *idle_state)
1271 {
1272 }
1273 
1274 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1275 {
1276 	return NULL;
1277 }
1278 #endif
1279 
1280 extern void sysrq_sched_debug_show(void);
1281 extern void sched_init_granularity(void);
1282 extern void update_max_interval(void);
1283 
1284 extern void init_sched_dl_class(void);
1285 extern void init_sched_rt_class(void);
1286 extern void init_sched_fair_class(void);
1287 extern void init_sched_dl_class(void);
1288 
1289 extern void resched_curr(struct rq *rq);
1290 extern void resched_cpu(int cpu);
1291 
1292 extern struct rt_bandwidth def_rt_bandwidth;
1293 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1294 
1295 extern struct dl_bandwidth def_dl_bandwidth;
1296 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1297 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1298 
1299 unsigned long to_ratio(u64 period, u64 runtime);
1300 
1301 extern void update_idle_cpu_load(struct rq *this_rq);
1302 
1303 extern void init_task_runnable_average(struct task_struct *p);
1304 
1305 static inline void add_nr_running(struct rq *rq, unsigned count)
1306 {
1307 	unsigned prev_nr = rq->nr_running;
1308 
1309 	rq->nr_running = prev_nr + count;
1310 
1311 	if (prev_nr < 2 && rq->nr_running >= 2) {
1312 #ifdef CONFIG_SMP
1313 		if (!rq->rd->overload)
1314 			rq->rd->overload = true;
1315 #endif
1316 
1317 #ifdef CONFIG_NO_HZ_FULL
1318 		if (tick_nohz_full_cpu(rq->cpu)) {
1319 			/*
1320 			 * Tick is needed if more than one task runs on a CPU.
1321 			 * Send the target an IPI to kick it out of nohz mode.
1322 			 *
1323 			 * We assume that IPI implies full memory barrier and the
1324 			 * new value of rq->nr_running is visible on reception
1325 			 * from the target.
1326 			 */
1327 			tick_nohz_full_kick_cpu(rq->cpu);
1328 		}
1329 #endif
1330 	}
1331 }
1332 
1333 static inline void sub_nr_running(struct rq *rq, unsigned count)
1334 {
1335 	rq->nr_running -= count;
1336 }
1337 
1338 static inline void rq_last_tick_reset(struct rq *rq)
1339 {
1340 #ifdef CONFIG_NO_HZ_FULL
1341 	rq->last_sched_tick = jiffies;
1342 #endif
1343 }
1344 
1345 extern void update_rq_clock(struct rq *rq);
1346 
1347 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1348 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1349 
1350 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1351 
1352 extern const_debug unsigned int sysctl_sched_time_avg;
1353 extern const_debug unsigned int sysctl_sched_nr_migrate;
1354 extern const_debug unsigned int sysctl_sched_migration_cost;
1355 
1356 static inline u64 sched_avg_period(void)
1357 {
1358 	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1359 }
1360 
1361 #ifdef CONFIG_SCHED_HRTICK
1362 
1363 /*
1364  * Use hrtick when:
1365  *  - enabled by features
1366  *  - hrtimer is actually high res
1367  */
1368 static inline int hrtick_enabled(struct rq *rq)
1369 {
1370 	if (!sched_feat(HRTICK))
1371 		return 0;
1372 	if (!cpu_active(cpu_of(rq)))
1373 		return 0;
1374 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1375 }
1376 
1377 void hrtick_start(struct rq *rq, u64 delay);
1378 
1379 #else
1380 
1381 static inline int hrtick_enabled(struct rq *rq)
1382 {
1383 	return 0;
1384 }
1385 
1386 #endif /* CONFIG_SCHED_HRTICK */
1387 
1388 #ifdef CONFIG_SMP
1389 extern void sched_avg_update(struct rq *rq);
1390 
1391 #ifndef arch_scale_freq_capacity
1392 static __always_inline
1393 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1394 {
1395 	return SCHED_CAPACITY_SCALE;
1396 }
1397 #endif
1398 
1399 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1400 {
1401 	rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1402 	sched_avg_update(rq);
1403 }
1404 #else
1405 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1406 static inline void sched_avg_update(struct rq *rq) { }
1407 #endif
1408 
1409 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1410 
1411 /*
1412  * __task_rq_lock - lock the rq @p resides on.
1413  */
1414 static inline struct rq *__task_rq_lock(struct task_struct *p)
1415 	__acquires(rq->lock)
1416 {
1417 	struct rq *rq;
1418 
1419 	lockdep_assert_held(&p->pi_lock);
1420 
1421 	for (;;) {
1422 		rq = task_rq(p);
1423 		raw_spin_lock(&rq->lock);
1424 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p)))
1425 			return rq;
1426 		raw_spin_unlock(&rq->lock);
1427 
1428 		while (unlikely(task_on_rq_migrating(p)))
1429 			cpu_relax();
1430 	}
1431 }
1432 
1433 /*
1434  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1435  */
1436 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1437 	__acquires(p->pi_lock)
1438 	__acquires(rq->lock)
1439 {
1440 	struct rq *rq;
1441 
1442 	for (;;) {
1443 		raw_spin_lock_irqsave(&p->pi_lock, *flags);
1444 		rq = task_rq(p);
1445 		raw_spin_lock(&rq->lock);
1446 		/*
1447 		 *	move_queued_task()		task_rq_lock()
1448 		 *
1449 		 *	ACQUIRE (rq->lock)
1450 		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
1451 		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
1452 		 *	[S] ->cpu = new_cpu		[L] task_rq()
1453 		 *					[L] ->on_rq
1454 		 *	RELEASE (rq->lock)
1455 		 *
1456 		 * If we observe the old cpu in task_rq_lock, the acquire of
1457 		 * the old rq->lock will fully serialize against the stores.
1458 		 *
1459 		 * If we observe the new cpu in task_rq_lock, the acquire will
1460 		 * pair with the WMB to ensure we must then also see migrating.
1461 		 */
1462 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p)))
1463 			return rq;
1464 		raw_spin_unlock(&rq->lock);
1465 		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1466 
1467 		while (unlikely(task_on_rq_migrating(p)))
1468 			cpu_relax();
1469 	}
1470 }
1471 
1472 static inline void __task_rq_unlock(struct rq *rq)
1473 	__releases(rq->lock)
1474 {
1475 	raw_spin_unlock(&rq->lock);
1476 }
1477 
1478 static inline void
1479 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1480 	__releases(rq->lock)
1481 	__releases(p->pi_lock)
1482 {
1483 	raw_spin_unlock(&rq->lock);
1484 	raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1485 }
1486 
1487 #ifdef CONFIG_SMP
1488 #ifdef CONFIG_PREEMPT
1489 
1490 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1491 
1492 /*
1493  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1494  * way at the expense of forcing extra atomic operations in all
1495  * invocations.  This assures that the double_lock is acquired using the
1496  * same underlying policy as the spinlock_t on this architecture, which
1497  * reduces latency compared to the unfair variant below.  However, it
1498  * also adds more overhead and therefore may reduce throughput.
1499  */
1500 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1501 	__releases(this_rq->lock)
1502 	__acquires(busiest->lock)
1503 	__acquires(this_rq->lock)
1504 {
1505 	raw_spin_unlock(&this_rq->lock);
1506 	double_rq_lock(this_rq, busiest);
1507 
1508 	return 1;
1509 }
1510 
1511 #else
1512 /*
1513  * Unfair double_lock_balance: Optimizes throughput at the expense of
1514  * latency by eliminating extra atomic operations when the locks are
1515  * already in proper order on entry.  This favors lower cpu-ids and will
1516  * grant the double lock to lower cpus over higher ids under contention,
1517  * regardless of entry order into the function.
1518  */
1519 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1520 	__releases(this_rq->lock)
1521 	__acquires(busiest->lock)
1522 	__acquires(this_rq->lock)
1523 {
1524 	int ret = 0;
1525 
1526 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1527 		if (busiest < this_rq) {
1528 			raw_spin_unlock(&this_rq->lock);
1529 			raw_spin_lock(&busiest->lock);
1530 			raw_spin_lock_nested(&this_rq->lock,
1531 					      SINGLE_DEPTH_NESTING);
1532 			ret = 1;
1533 		} else
1534 			raw_spin_lock_nested(&busiest->lock,
1535 					      SINGLE_DEPTH_NESTING);
1536 	}
1537 	return ret;
1538 }
1539 
1540 #endif /* CONFIG_PREEMPT */
1541 
1542 /*
1543  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1544  */
1545 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1546 {
1547 	if (unlikely(!irqs_disabled())) {
1548 		/* printk() doesn't work good under rq->lock */
1549 		raw_spin_unlock(&this_rq->lock);
1550 		BUG_ON(1);
1551 	}
1552 
1553 	return _double_lock_balance(this_rq, busiest);
1554 }
1555 
1556 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1557 	__releases(busiest->lock)
1558 {
1559 	raw_spin_unlock(&busiest->lock);
1560 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1561 }
1562 
1563 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1564 {
1565 	if (l1 > l2)
1566 		swap(l1, l2);
1567 
1568 	spin_lock(l1);
1569 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1570 }
1571 
1572 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1573 {
1574 	if (l1 > l2)
1575 		swap(l1, l2);
1576 
1577 	spin_lock_irq(l1);
1578 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1579 }
1580 
1581 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1582 {
1583 	if (l1 > l2)
1584 		swap(l1, l2);
1585 
1586 	raw_spin_lock(l1);
1587 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1588 }
1589 
1590 /*
1591  * double_rq_lock - safely lock two runqueues
1592  *
1593  * Note this does not disable interrupts like task_rq_lock,
1594  * you need to do so manually before calling.
1595  */
1596 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1597 	__acquires(rq1->lock)
1598 	__acquires(rq2->lock)
1599 {
1600 	BUG_ON(!irqs_disabled());
1601 	if (rq1 == rq2) {
1602 		raw_spin_lock(&rq1->lock);
1603 		__acquire(rq2->lock);	/* Fake it out ;) */
1604 	} else {
1605 		if (rq1 < rq2) {
1606 			raw_spin_lock(&rq1->lock);
1607 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1608 		} else {
1609 			raw_spin_lock(&rq2->lock);
1610 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1611 		}
1612 	}
1613 }
1614 
1615 /*
1616  * double_rq_unlock - safely unlock two runqueues
1617  *
1618  * Note this does not restore interrupts like task_rq_unlock,
1619  * you need to do so manually after calling.
1620  */
1621 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1622 	__releases(rq1->lock)
1623 	__releases(rq2->lock)
1624 {
1625 	raw_spin_unlock(&rq1->lock);
1626 	if (rq1 != rq2)
1627 		raw_spin_unlock(&rq2->lock);
1628 	else
1629 		__release(rq2->lock);
1630 }
1631 
1632 #else /* CONFIG_SMP */
1633 
1634 /*
1635  * double_rq_lock - safely lock two runqueues
1636  *
1637  * Note this does not disable interrupts like task_rq_lock,
1638  * you need to do so manually before calling.
1639  */
1640 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1641 	__acquires(rq1->lock)
1642 	__acquires(rq2->lock)
1643 {
1644 	BUG_ON(!irqs_disabled());
1645 	BUG_ON(rq1 != rq2);
1646 	raw_spin_lock(&rq1->lock);
1647 	__acquire(rq2->lock);	/* Fake it out ;) */
1648 }
1649 
1650 /*
1651  * double_rq_unlock - safely unlock two runqueues
1652  *
1653  * Note this does not restore interrupts like task_rq_unlock,
1654  * you need to do so manually after calling.
1655  */
1656 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1657 	__releases(rq1->lock)
1658 	__releases(rq2->lock)
1659 {
1660 	BUG_ON(rq1 != rq2);
1661 	raw_spin_unlock(&rq1->lock);
1662 	__release(rq2->lock);
1663 }
1664 
1665 #endif
1666 
1667 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1668 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1669 extern void print_cfs_stats(struct seq_file *m, int cpu);
1670 extern void print_rt_stats(struct seq_file *m, int cpu);
1671 extern void print_dl_stats(struct seq_file *m, int cpu);
1672 
1673 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1674 extern void init_rt_rq(struct rt_rq *rt_rq);
1675 extern void init_dl_rq(struct dl_rq *dl_rq);
1676 
1677 extern void cfs_bandwidth_usage_inc(void);
1678 extern void cfs_bandwidth_usage_dec(void);
1679 
1680 #ifdef CONFIG_NO_HZ_COMMON
1681 enum rq_nohz_flag_bits {
1682 	NOHZ_TICK_STOPPED,
1683 	NOHZ_BALANCE_KICK,
1684 };
1685 
1686 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
1687 #endif
1688 
1689 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1690 
1691 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1692 DECLARE_PER_CPU(u64, cpu_softirq_time);
1693 
1694 #ifndef CONFIG_64BIT
1695 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1696 
1697 static inline void irq_time_write_begin(void)
1698 {
1699 	__this_cpu_inc(irq_time_seq.sequence);
1700 	smp_wmb();
1701 }
1702 
1703 static inline void irq_time_write_end(void)
1704 {
1705 	smp_wmb();
1706 	__this_cpu_inc(irq_time_seq.sequence);
1707 }
1708 
1709 static inline u64 irq_time_read(int cpu)
1710 {
1711 	u64 irq_time;
1712 	unsigned seq;
1713 
1714 	do {
1715 		seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1716 		irq_time = per_cpu(cpu_softirq_time, cpu) +
1717 			   per_cpu(cpu_hardirq_time, cpu);
1718 	} while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1719 
1720 	return irq_time;
1721 }
1722 #else /* CONFIG_64BIT */
1723 static inline void irq_time_write_begin(void)
1724 {
1725 }
1726 
1727 static inline void irq_time_write_end(void)
1728 {
1729 }
1730 
1731 static inline u64 irq_time_read(int cpu)
1732 {
1733 	return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1734 }
1735 #endif /* CONFIG_64BIT */
1736 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1737