xref: /linux/kernel/sched/sched.h (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
1 
2 #include <linux/sched.h>
3 #include <linux/mutex.h>
4 #include <linux/spinlock.h>
5 #include <linux/stop_machine.h>
6 
7 #include "cpupri.h"
8 
9 extern __read_mostly int scheduler_running;
10 
11 /*
12  * Convert user-nice values [ -20 ... 0 ... 19 ]
13  * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
14  * and back.
15  */
16 #define NICE_TO_PRIO(nice)	(MAX_RT_PRIO + (nice) + 20)
17 #define PRIO_TO_NICE(prio)	((prio) - MAX_RT_PRIO - 20)
18 #define TASK_NICE(p)		PRIO_TO_NICE((p)->static_prio)
19 
20 /*
21  * 'User priority' is the nice value converted to something we
22  * can work with better when scaling various scheduler parameters,
23  * it's a [ 0 ... 39 ] range.
24  */
25 #define USER_PRIO(p)		((p)-MAX_RT_PRIO)
26 #define TASK_USER_PRIO(p)	USER_PRIO((p)->static_prio)
27 #define MAX_USER_PRIO		(USER_PRIO(MAX_PRIO))
28 
29 /*
30  * Helpers for converting nanosecond timing to jiffy resolution
31  */
32 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
33 
34 #define NICE_0_LOAD		SCHED_LOAD_SCALE
35 #define NICE_0_SHIFT		SCHED_LOAD_SHIFT
36 
37 /*
38  * These are the 'tuning knobs' of the scheduler:
39  */
40 
41 /*
42  * single value that denotes runtime == period, ie unlimited time.
43  */
44 #define RUNTIME_INF	((u64)~0ULL)
45 
46 static inline int rt_policy(int policy)
47 {
48 	if (policy == SCHED_FIFO || policy == SCHED_RR)
49 		return 1;
50 	return 0;
51 }
52 
53 static inline int task_has_rt_policy(struct task_struct *p)
54 {
55 	return rt_policy(p->policy);
56 }
57 
58 /*
59  * This is the priority-queue data structure of the RT scheduling class:
60  */
61 struct rt_prio_array {
62 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
63 	struct list_head queue[MAX_RT_PRIO];
64 };
65 
66 struct rt_bandwidth {
67 	/* nests inside the rq lock: */
68 	raw_spinlock_t		rt_runtime_lock;
69 	ktime_t			rt_period;
70 	u64			rt_runtime;
71 	struct hrtimer		rt_period_timer;
72 };
73 
74 extern struct mutex sched_domains_mutex;
75 
76 #ifdef CONFIG_CGROUP_SCHED
77 
78 #include <linux/cgroup.h>
79 
80 struct cfs_rq;
81 struct rt_rq;
82 
83 extern struct list_head task_groups;
84 
85 struct cfs_bandwidth {
86 #ifdef CONFIG_CFS_BANDWIDTH
87 	raw_spinlock_t lock;
88 	ktime_t period;
89 	u64 quota, runtime;
90 	s64 hierarchal_quota;
91 	u64 runtime_expires;
92 
93 	int idle, timer_active;
94 	struct hrtimer period_timer, slack_timer;
95 	struct list_head throttled_cfs_rq;
96 
97 	/* statistics */
98 	int nr_periods, nr_throttled;
99 	u64 throttled_time;
100 #endif
101 };
102 
103 /* task group related information */
104 struct task_group {
105 	struct cgroup_subsys_state css;
106 
107 #ifdef CONFIG_FAIR_GROUP_SCHED
108 	/* schedulable entities of this group on each cpu */
109 	struct sched_entity **se;
110 	/* runqueue "owned" by this group on each cpu */
111 	struct cfs_rq **cfs_rq;
112 	unsigned long shares;
113 
114 	atomic_t load_weight;
115 	atomic64_t load_avg;
116 	atomic_t runnable_avg;
117 #endif
118 
119 #ifdef CONFIG_RT_GROUP_SCHED
120 	struct sched_rt_entity **rt_se;
121 	struct rt_rq **rt_rq;
122 
123 	struct rt_bandwidth rt_bandwidth;
124 #endif
125 
126 	struct rcu_head rcu;
127 	struct list_head list;
128 
129 	struct task_group *parent;
130 	struct list_head siblings;
131 	struct list_head children;
132 
133 #ifdef CONFIG_SCHED_AUTOGROUP
134 	struct autogroup *autogroup;
135 #endif
136 
137 	struct cfs_bandwidth cfs_bandwidth;
138 };
139 
140 #ifdef CONFIG_FAIR_GROUP_SCHED
141 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
142 
143 /*
144  * A weight of 0 or 1 can cause arithmetics problems.
145  * A weight of a cfs_rq is the sum of weights of which entities
146  * are queued on this cfs_rq, so a weight of a entity should not be
147  * too large, so as the shares value of a task group.
148  * (The default weight is 1024 - so there's no practical
149  *  limitation from this.)
150  */
151 #define MIN_SHARES	(1UL <<  1)
152 #define MAX_SHARES	(1UL << 18)
153 #endif
154 
155 /* Default task group.
156  *	Every task in system belong to this group at bootup.
157  */
158 extern struct task_group root_task_group;
159 
160 typedef int (*tg_visitor)(struct task_group *, void *);
161 
162 extern int walk_tg_tree_from(struct task_group *from,
163 			     tg_visitor down, tg_visitor up, void *data);
164 
165 /*
166  * Iterate the full tree, calling @down when first entering a node and @up when
167  * leaving it for the final time.
168  *
169  * Caller must hold rcu_lock or sufficient equivalent.
170  */
171 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
172 {
173 	return walk_tg_tree_from(&root_task_group, down, up, data);
174 }
175 
176 extern int tg_nop(struct task_group *tg, void *data);
177 
178 extern void free_fair_sched_group(struct task_group *tg);
179 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
180 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
181 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
182 			struct sched_entity *se, int cpu,
183 			struct sched_entity *parent);
184 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
185 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
186 
187 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
188 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
189 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
190 
191 extern void free_rt_sched_group(struct task_group *tg);
192 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
193 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
194 		struct sched_rt_entity *rt_se, int cpu,
195 		struct sched_rt_entity *parent);
196 
197 #else /* CONFIG_CGROUP_SCHED */
198 
199 struct cfs_bandwidth { };
200 
201 #endif	/* CONFIG_CGROUP_SCHED */
202 
203 /* CFS-related fields in a runqueue */
204 struct cfs_rq {
205 	struct load_weight load;
206 	unsigned int nr_running, h_nr_running;
207 
208 	u64 exec_clock;
209 	u64 min_vruntime;
210 #ifndef CONFIG_64BIT
211 	u64 min_vruntime_copy;
212 #endif
213 
214 	struct rb_root tasks_timeline;
215 	struct rb_node *rb_leftmost;
216 
217 	/*
218 	 * 'curr' points to currently running entity on this cfs_rq.
219 	 * It is set to NULL otherwise (i.e when none are currently running).
220 	 */
221 	struct sched_entity *curr, *next, *last, *skip;
222 
223 #ifdef	CONFIG_SCHED_DEBUG
224 	unsigned int nr_spread_over;
225 #endif
226 
227 #ifdef CONFIG_SMP
228 /*
229  * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
230  * removed when useful for applications beyond shares distribution (e.g.
231  * load-balance).
232  */
233 #ifdef CONFIG_FAIR_GROUP_SCHED
234 	/*
235 	 * CFS Load tracking
236 	 * Under CFS, load is tracked on a per-entity basis and aggregated up.
237 	 * This allows for the description of both thread and group usage (in
238 	 * the FAIR_GROUP_SCHED case).
239 	 */
240 	u64 runnable_load_avg, blocked_load_avg;
241 	atomic64_t decay_counter, removed_load;
242 	u64 last_decay;
243 #endif /* CONFIG_FAIR_GROUP_SCHED */
244 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
245 #ifdef CONFIG_FAIR_GROUP_SCHED
246 	u32 tg_runnable_contrib;
247 	u64 tg_load_contrib;
248 #endif /* CONFIG_FAIR_GROUP_SCHED */
249 
250 	/*
251 	 *   h_load = weight * f(tg)
252 	 *
253 	 * Where f(tg) is the recursive weight fraction assigned to
254 	 * this group.
255 	 */
256 	unsigned long h_load;
257 #endif /* CONFIG_SMP */
258 
259 #ifdef CONFIG_FAIR_GROUP_SCHED
260 	struct rq *rq;	/* cpu runqueue to which this cfs_rq is attached */
261 
262 	/*
263 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
264 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
265 	 * (like users, containers etc.)
266 	 *
267 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
268 	 * list is used during load balance.
269 	 */
270 	int on_list;
271 	struct list_head leaf_cfs_rq_list;
272 	struct task_group *tg;	/* group that "owns" this runqueue */
273 
274 #ifdef CONFIG_CFS_BANDWIDTH
275 	int runtime_enabled;
276 	u64 runtime_expires;
277 	s64 runtime_remaining;
278 
279 	u64 throttled_clock, throttled_clock_task;
280 	u64 throttled_clock_task_time;
281 	int throttled, throttle_count;
282 	struct list_head throttled_list;
283 #endif /* CONFIG_CFS_BANDWIDTH */
284 #endif /* CONFIG_FAIR_GROUP_SCHED */
285 };
286 
287 static inline int rt_bandwidth_enabled(void)
288 {
289 	return sysctl_sched_rt_runtime >= 0;
290 }
291 
292 /* Real-Time classes' related field in a runqueue: */
293 struct rt_rq {
294 	struct rt_prio_array active;
295 	unsigned int rt_nr_running;
296 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
297 	struct {
298 		int curr; /* highest queued rt task prio */
299 #ifdef CONFIG_SMP
300 		int next; /* next highest */
301 #endif
302 	} highest_prio;
303 #endif
304 #ifdef CONFIG_SMP
305 	unsigned long rt_nr_migratory;
306 	unsigned long rt_nr_total;
307 	int overloaded;
308 	struct plist_head pushable_tasks;
309 #endif
310 	int rt_throttled;
311 	u64 rt_time;
312 	u64 rt_runtime;
313 	/* Nests inside the rq lock: */
314 	raw_spinlock_t rt_runtime_lock;
315 
316 #ifdef CONFIG_RT_GROUP_SCHED
317 	unsigned long rt_nr_boosted;
318 
319 	struct rq *rq;
320 	struct list_head leaf_rt_rq_list;
321 	struct task_group *tg;
322 #endif
323 };
324 
325 #ifdef CONFIG_SMP
326 
327 /*
328  * We add the notion of a root-domain which will be used to define per-domain
329  * variables. Each exclusive cpuset essentially defines an island domain by
330  * fully partitioning the member cpus from any other cpuset. Whenever a new
331  * exclusive cpuset is created, we also create and attach a new root-domain
332  * object.
333  *
334  */
335 struct root_domain {
336 	atomic_t refcount;
337 	atomic_t rto_count;
338 	struct rcu_head rcu;
339 	cpumask_var_t span;
340 	cpumask_var_t online;
341 
342 	/*
343 	 * The "RT overload" flag: it gets set if a CPU has more than
344 	 * one runnable RT task.
345 	 */
346 	cpumask_var_t rto_mask;
347 	struct cpupri cpupri;
348 };
349 
350 extern struct root_domain def_root_domain;
351 
352 #endif /* CONFIG_SMP */
353 
354 /*
355  * This is the main, per-CPU runqueue data structure.
356  *
357  * Locking rule: those places that want to lock multiple runqueues
358  * (such as the load balancing or the thread migration code), lock
359  * acquire operations must be ordered by ascending &runqueue.
360  */
361 struct rq {
362 	/* runqueue lock: */
363 	raw_spinlock_t lock;
364 
365 	/*
366 	 * nr_running and cpu_load should be in the same cacheline because
367 	 * remote CPUs use both these fields when doing load calculation.
368 	 */
369 	unsigned int nr_running;
370 	#define CPU_LOAD_IDX_MAX 5
371 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
372 	unsigned long last_load_update_tick;
373 #ifdef CONFIG_NO_HZ
374 	u64 nohz_stamp;
375 	unsigned long nohz_flags;
376 #endif
377 	int skip_clock_update;
378 
379 	/* capture load from *all* tasks on this cpu: */
380 	struct load_weight load;
381 	unsigned long nr_load_updates;
382 	u64 nr_switches;
383 
384 	struct cfs_rq cfs;
385 	struct rt_rq rt;
386 
387 #ifdef CONFIG_FAIR_GROUP_SCHED
388 	/* list of leaf cfs_rq on this cpu: */
389 	struct list_head leaf_cfs_rq_list;
390 #ifdef CONFIG_SMP
391 	unsigned long h_load_throttle;
392 #endif /* CONFIG_SMP */
393 #endif /* CONFIG_FAIR_GROUP_SCHED */
394 
395 #ifdef CONFIG_RT_GROUP_SCHED
396 	struct list_head leaf_rt_rq_list;
397 #endif
398 
399 	/*
400 	 * This is part of a global counter where only the total sum
401 	 * over all CPUs matters. A task can increase this counter on
402 	 * one CPU and if it got migrated afterwards it may decrease
403 	 * it on another CPU. Always updated under the runqueue lock:
404 	 */
405 	unsigned long nr_uninterruptible;
406 
407 	struct task_struct *curr, *idle, *stop;
408 	unsigned long next_balance;
409 	struct mm_struct *prev_mm;
410 
411 	u64 clock;
412 	u64 clock_task;
413 
414 	atomic_t nr_iowait;
415 
416 #ifdef CONFIG_SMP
417 	struct root_domain *rd;
418 	struct sched_domain *sd;
419 
420 	unsigned long cpu_power;
421 
422 	unsigned char idle_balance;
423 	/* For active balancing */
424 	int post_schedule;
425 	int active_balance;
426 	int push_cpu;
427 	struct cpu_stop_work active_balance_work;
428 	/* cpu of this runqueue: */
429 	int cpu;
430 	int online;
431 
432 	struct list_head cfs_tasks;
433 
434 	u64 rt_avg;
435 	u64 age_stamp;
436 	u64 idle_stamp;
437 	u64 avg_idle;
438 #endif
439 
440 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
441 	u64 prev_irq_time;
442 #endif
443 #ifdef CONFIG_PARAVIRT
444 	u64 prev_steal_time;
445 #endif
446 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
447 	u64 prev_steal_time_rq;
448 #endif
449 
450 	/* calc_load related fields */
451 	unsigned long calc_load_update;
452 	long calc_load_active;
453 
454 #ifdef CONFIG_SCHED_HRTICK
455 #ifdef CONFIG_SMP
456 	int hrtick_csd_pending;
457 	struct call_single_data hrtick_csd;
458 #endif
459 	struct hrtimer hrtick_timer;
460 #endif
461 
462 #ifdef CONFIG_SCHEDSTATS
463 	/* latency stats */
464 	struct sched_info rq_sched_info;
465 	unsigned long long rq_cpu_time;
466 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
467 
468 	/* sys_sched_yield() stats */
469 	unsigned int yld_count;
470 
471 	/* schedule() stats */
472 	unsigned int sched_count;
473 	unsigned int sched_goidle;
474 
475 	/* try_to_wake_up() stats */
476 	unsigned int ttwu_count;
477 	unsigned int ttwu_local;
478 #endif
479 
480 #ifdef CONFIG_SMP
481 	struct llist_head wake_list;
482 #endif
483 
484 	struct sched_avg avg;
485 };
486 
487 static inline int cpu_of(struct rq *rq)
488 {
489 #ifdef CONFIG_SMP
490 	return rq->cpu;
491 #else
492 	return 0;
493 #endif
494 }
495 
496 DECLARE_PER_CPU(struct rq, runqueues);
497 
498 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
499 #define this_rq()		(&__get_cpu_var(runqueues))
500 #define task_rq(p)		cpu_rq(task_cpu(p))
501 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
502 #define raw_rq()		(&__raw_get_cpu_var(runqueues))
503 
504 #ifdef CONFIG_SMP
505 
506 #define rcu_dereference_check_sched_domain(p) \
507 	rcu_dereference_check((p), \
508 			      lockdep_is_held(&sched_domains_mutex))
509 
510 /*
511  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
512  * See detach_destroy_domains: synchronize_sched for details.
513  *
514  * The domain tree of any CPU may only be accessed from within
515  * preempt-disabled sections.
516  */
517 #define for_each_domain(cpu, __sd) \
518 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
519 			__sd; __sd = __sd->parent)
520 
521 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
522 
523 /**
524  * highest_flag_domain - Return highest sched_domain containing flag.
525  * @cpu:	The cpu whose highest level of sched domain is to
526  *		be returned.
527  * @flag:	The flag to check for the highest sched_domain
528  *		for the given cpu.
529  *
530  * Returns the highest sched_domain of a cpu which contains the given flag.
531  */
532 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
533 {
534 	struct sched_domain *sd, *hsd = NULL;
535 
536 	for_each_domain(cpu, sd) {
537 		if (!(sd->flags & flag))
538 			break;
539 		hsd = sd;
540 	}
541 
542 	return hsd;
543 }
544 
545 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
546 DECLARE_PER_CPU(int, sd_llc_id);
547 
548 extern int group_balance_cpu(struct sched_group *sg);
549 
550 #endif /* CONFIG_SMP */
551 
552 #include "stats.h"
553 #include "auto_group.h"
554 
555 #ifdef CONFIG_CGROUP_SCHED
556 
557 /*
558  * Return the group to which this tasks belongs.
559  *
560  * We cannot use task_subsys_state() and friends because the cgroup
561  * subsystem changes that value before the cgroup_subsys::attach() method
562  * is called, therefore we cannot pin it and might observe the wrong value.
563  *
564  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
565  * core changes this before calling sched_move_task().
566  *
567  * Instead we use a 'copy' which is updated from sched_move_task() while
568  * holding both task_struct::pi_lock and rq::lock.
569  */
570 static inline struct task_group *task_group(struct task_struct *p)
571 {
572 	return p->sched_task_group;
573 }
574 
575 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
576 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
577 {
578 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
579 	struct task_group *tg = task_group(p);
580 #endif
581 
582 #ifdef CONFIG_FAIR_GROUP_SCHED
583 	p->se.cfs_rq = tg->cfs_rq[cpu];
584 	p->se.parent = tg->se[cpu];
585 #endif
586 
587 #ifdef CONFIG_RT_GROUP_SCHED
588 	p->rt.rt_rq  = tg->rt_rq[cpu];
589 	p->rt.parent = tg->rt_se[cpu];
590 #endif
591 }
592 
593 #else /* CONFIG_CGROUP_SCHED */
594 
595 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
596 static inline struct task_group *task_group(struct task_struct *p)
597 {
598 	return NULL;
599 }
600 
601 #endif /* CONFIG_CGROUP_SCHED */
602 
603 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
604 {
605 	set_task_rq(p, cpu);
606 #ifdef CONFIG_SMP
607 	/*
608 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
609 	 * successfuly executed on another CPU. We must ensure that updates of
610 	 * per-task data have been completed by this moment.
611 	 */
612 	smp_wmb();
613 	task_thread_info(p)->cpu = cpu;
614 #endif
615 }
616 
617 /*
618  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
619  */
620 #ifdef CONFIG_SCHED_DEBUG
621 # include <linux/static_key.h>
622 # define const_debug __read_mostly
623 #else
624 # define const_debug const
625 #endif
626 
627 extern const_debug unsigned int sysctl_sched_features;
628 
629 #define SCHED_FEAT(name, enabled)	\
630 	__SCHED_FEAT_##name ,
631 
632 enum {
633 #include "features.h"
634 	__SCHED_FEAT_NR,
635 };
636 
637 #undef SCHED_FEAT
638 
639 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
640 static __always_inline bool static_branch__true(struct static_key *key)
641 {
642 	return static_key_true(key); /* Not out of line branch. */
643 }
644 
645 static __always_inline bool static_branch__false(struct static_key *key)
646 {
647 	return static_key_false(key); /* Out of line branch. */
648 }
649 
650 #define SCHED_FEAT(name, enabled)					\
651 static __always_inline bool static_branch_##name(struct static_key *key) \
652 {									\
653 	return static_branch__##enabled(key);				\
654 }
655 
656 #include "features.h"
657 
658 #undef SCHED_FEAT
659 
660 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
661 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
662 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
663 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
664 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
665 
666 #ifdef CONFIG_NUMA_BALANCING
667 #define sched_feat_numa(x) sched_feat(x)
668 #ifdef CONFIG_SCHED_DEBUG
669 #define numabalancing_enabled sched_feat_numa(NUMA)
670 #else
671 extern bool numabalancing_enabled;
672 #endif /* CONFIG_SCHED_DEBUG */
673 #else
674 #define sched_feat_numa(x) (0)
675 #define numabalancing_enabled (0)
676 #endif /* CONFIG_NUMA_BALANCING */
677 
678 static inline u64 global_rt_period(void)
679 {
680 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
681 }
682 
683 static inline u64 global_rt_runtime(void)
684 {
685 	if (sysctl_sched_rt_runtime < 0)
686 		return RUNTIME_INF;
687 
688 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
689 }
690 
691 
692 
693 static inline int task_current(struct rq *rq, struct task_struct *p)
694 {
695 	return rq->curr == p;
696 }
697 
698 static inline int task_running(struct rq *rq, struct task_struct *p)
699 {
700 #ifdef CONFIG_SMP
701 	return p->on_cpu;
702 #else
703 	return task_current(rq, p);
704 #endif
705 }
706 
707 
708 #ifndef prepare_arch_switch
709 # define prepare_arch_switch(next)	do { } while (0)
710 #endif
711 #ifndef finish_arch_switch
712 # define finish_arch_switch(prev)	do { } while (0)
713 #endif
714 #ifndef finish_arch_post_lock_switch
715 # define finish_arch_post_lock_switch()	do { } while (0)
716 #endif
717 
718 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
719 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
720 {
721 #ifdef CONFIG_SMP
722 	/*
723 	 * We can optimise this out completely for !SMP, because the
724 	 * SMP rebalancing from interrupt is the only thing that cares
725 	 * here.
726 	 */
727 	next->on_cpu = 1;
728 #endif
729 }
730 
731 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
732 {
733 #ifdef CONFIG_SMP
734 	/*
735 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
736 	 * We must ensure this doesn't happen until the switch is completely
737 	 * finished.
738 	 */
739 	smp_wmb();
740 	prev->on_cpu = 0;
741 #endif
742 #ifdef CONFIG_DEBUG_SPINLOCK
743 	/* this is a valid case when another task releases the spinlock */
744 	rq->lock.owner = current;
745 #endif
746 	/*
747 	 * If we are tracking spinlock dependencies then we have to
748 	 * fix up the runqueue lock - which gets 'carried over' from
749 	 * prev into current:
750 	 */
751 	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
752 
753 	raw_spin_unlock_irq(&rq->lock);
754 }
755 
756 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
757 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
758 {
759 #ifdef CONFIG_SMP
760 	/*
761 	 * We can optimise this out completely for !SMP, because the
762 	 * SMP rebalancing from interrupt is the only thing that cares
763 	 * here.
764 	 */
765 	next->on_cpu = 1;
766 #endif
767 	raw_spin_unlock(&rq->lock);
768 }
769 
770 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
771 {
772 #ifdef CONFIG_SMP
773 	/*
774 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
775 	 * We must ensure this doesn't happen until the switch is completely
776 	 * finished.
777 	 */
778 	smp_wmb();
779 	prev->on_cpu = 0;
780 #endif
781 	local_irq_enable();
782 }
783 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
784 
785 
786 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
787 {
788 	lw->weight += inc;
789 	lw->inv_weight = 0;
790 }
791 
792 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
793 {
794 	lw->weight -= dec;
795 	lw->inv_weight = 0;
796 }
797 
798 static inline void update_load_set(struct load_weight *lw, unsigned long w)
799 {
800 	lw->weight = w;
801 	lw->inv_weight = 0;
802 }
803 
804 /*
805  * To aid in avoiding the subversion of "niceness" due to uneven distribution
806  * of tasks with abnormal "nice" values across CPUs the contribution that
807  * each task makes to its run queue's load is weighted according to its
808  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
809  * scaled version of the new time slice allocation that they receive on time
810  * slice expiry etc.
811  */
812 
813 #define WEIGHT_IDLEPRIO                3
814 #define WMULT_IDLEPRIO         1431655765
815 
816 /*
817  * Nice levels are multiplicative, with a gentle 10% change for every
818  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
819  * nice 1, it will get ~10% less CPU time than another CPU-bound task
820  * that remained on nice 0.
821  *
822  * The "10% effect" is relative and cumulative: from _any_ nice level,
823  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
824  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
825  * If a task goes up by ~10% and another task goes down by ~10% then
826  * the relative distance between them is ~25%.)
827  */
828 static const int prio_to_weight[40] = {
829  /* -20 */     88761,     71755,     56483,     46273,     36291,
830  /* -15 */     29154,     23254,     18705,     14949,     11916,
831  /* -10 */      9548,      7620,      6100,      4904,      3906,
832  /*  -5 */      3121,      2501,      1991,      1586,      1277,
833  /*   0 */      1024,       820,       655,       526,       423,
834  /*   5 */       335,       272,       215,       172,       137,
835  /*  10 */       110,        87,        70,        56,        45,
836  /*  15 */        36,        29,        23,        18,        15,
837 };
838 
839 /*
840  * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
841  *
842  * In cases where the weight does not change often, we can use the
843  * precalculated inverse to speed up arithmetics by turning divisions
844  * into multiplications:
845  */
846 static const u32 prio_to_wmult[40] = {
847  /* -20 */     48388,     59856,     76040,     92818,    118348,
848  /* -15 */    147320,    184698,    229616,    287308,    360437,
849  /* -10 */    449829,    563644,    704093,    875809,   1099582,
850  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
851  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
852  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
853  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
854  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
855 };
856 
857 /* Time spent by the tasks of the cpu accounting group executing in ... */
858 enum cpuacct_stat_index {
859 	CPUACCT_STAT_USER,	/* ... user mode */
860 	CPUACCT_STAT_SYSTEM,	/* ... kernel mode */
861 
862 	CPUACCT_STAT_NSTATS,
863 };
864 
865 
866 #define sched_class_highest (&stop_sched_class)
867 #define for_each_class(class) \
868    for (class = sched_class_highest; class; class = class->next)
869 
870 extern const struct sched_class stop_sched_class;
871 extern const struct sched_class rt_sched_class;
872 extern const struct sched_class fair_sched_class;
873 extern const struct sched_class idle_sched_class;
874 
875 
876 #ifdef CONFIG_SMP
877 
878 extern void trigger_load_balance(struct rq *rq, int cpu);
879 extern void idle_balance(int this_cpu, struct rq *this_rq);
880 
881 #else	/* CONFIG_SMP */
882 
883 static inline void idle_balance(int cpu, struct rq *rq)
884 {
885 }
886 
887 #endif
888 
889 extern void sysrq_sched_debug_show(void);
890 extern void sched_init_granularity(void);
891 extern void update_max_interval(void);
892 extern void update_group_power(struct sched_domain *sd, int cpu);
893 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
894 extern void init_sched_rt_class(void);
895 extern void init_sched_fair_class(void);
896 
897 extern void resched_task(struct task_struct *p);
898 extern void resched_cpu(int cpu);
899 
900 extern struct rt_bandwidth def_rt_bandwidth;
901 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
902 
903 extern void update_idle_cpu_load(struct rq *this_rq);
904 
905 #ifdef CONFIG_CGROUP_CPUACCT
906 #include <linux/cgroup.h>
907 /* track cpu usage of a group of tasks and its child groups */
908 struct cpuacct {
909 	struct cgroup_subsys_state css;
910 	/* cpuusage holds pointer to a u64-type object on every cpu */
911 	u64 __percpu *cpuusage;
912 	struct kernel_cpustat __percpu *cpustat;
913 };
914 
915 extern struct cgroup_subsys cpuacct_subsys;
916 extern struct cpuacct root_cpuacct;
917 
918 /* return cpu accounting group corresponding to this container */
919 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
920 {
921 	return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
922 			    struct cpuacct, css);
923 }
924 
925 /* return cpu accounting group to which this task belongs */
926 static inline struct cpuacct *task_ca(struct task_struct *tsk)
927 {
928 	return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
929 			    struct cpuacct, css);
930 }
931 
932 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
933 {
934 	if (!ca || !ca->css.cgroup->parent)
935 		return NULL;
936 	return cgroup_ca(ca->css.cgroup->parent);
937 }
938 
939 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
940 #else
941 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
942 #endif
943 
944 #ifdef CONFIG_PARAVIRT
945 static inline u64 steal_ticks(u64 steal)
946 {
947 	if (unlikely(steal > NSEC_PER_SEC))
948 		return div_u64(steal, TICK_NSEC);
949 
950 	return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
951 }
952 #endif
953 
954 static inline void inc_nr_running(struct rq *rq)
955 {
956 	rq->nr_running++;
957 }
958 
959 static inline void dec_nr_running(struct rq *rq)
960 {
961 	rq->nr_running--;
962 }
963 
964 extern void update_rq_clock(struct rq *rq);
965 
966 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
967 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
968 
969 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
970 
971 extern const_debug unsigned int sysctl_sched_time_avg;
972 extern const_debug unsigned int sysctl_sched_nr_migrate;
973 extern const_debug unsigned int sysctl_sched_migration_cost;
974 
975 static inline u64 sched_avg_period(void)
976 {
977 	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
978 }
979 
980 #ifdef CONFIG_SCHED_HRTICK
981 
982 /*
983  * Use hrtick when:
984  *  - enabled by features
985  *  - hrtimer is actually high res
986  */
987 static inline int hrtick_enabled(struct rq *rq)
988 {
989 	if (!sched_feat(HRTICK))
990 		return 0;
991 	if (!cpu_active(cpu_of(rq)))
992 		return 0;
993 	return hrtimer_is_hres_active(&rq->hrtick_timer);
994 }
995 
996 void hrtick_start(struct rq *rq, u64 delay);
997 
998 #else
999 
1000 static inline int hrtick_enabled(struct rq *rq)
1001 {
1002 	return 0;
1003 }
1004 
1005 #endif /* CONFIG_SCHED_HRTICK */
1006 
1007 #ifdef CONFIG_SMP
1008 extern void sched_avg_update(struct rq *rq);
1009 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1010 {
1011 	rq->rt_avg += rt_delta;
1012 	sched_avg_update(rq);
1013 }
1014 #else
1015 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1016 static inline void sched_avg_update(struct rq *rq) { }
1017 #endif
1018 
1019 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1020 
1021 #ifdef CONFIG_SMP
1022 #ifdef CONFIG_PREEMPT
1023 
1024 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1025 
1026 /*
1027  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1028  * way at the expense of forcing extra atomic operations in all
1029  * invocations.  This assures that the double_lock is acquired using the
1030  * same underlying policy as the spinlock_t on this architecture, which
1031  * reduces latency compared to the unfair variant below.  However, it
1032  * also adds more overhead and therefore may reduce throughput.
1033  */
1034 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1035 	__releases(this_rq->lock)
1036 	__acquires(busiest->lock)
1037 	__acquires(this_rq->lock)
1038 {
1039 	raw_spin_unlock(&this_rq->lock);
1040 	double_rq_lock(this_rq, busiest);
1041 
1042 	return 1;
1043 }
1044 
1045 #else
1046 /*
1047  * Unfair double_lock_balance: Optimizes throughput at the expense of
1048  * latency by eliminating extra atomic operations when the locks are
1049  * already in proper order on entry.  This favors lower cpu-ids and will
1050  * grant the double lock to lower cpus over higher ids under contention,
1051  * regardless of entry order into the function.
1052  */
1053 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1054 	__releases(this_rq->lock)
1055 	__acquires(busiest->lock)
1056 	__acquires(this_rq->lock)
1057 {
1058 	int ret = 0;
1059 
1060 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1061 		if (busiest < this_rq) {
1062 			raw_spin_unlock(&this_rq->lock);
1063 			raw_spin_lock(&busiest->lock);
1064 			raw_spin_lock_nested(&this_rq->lock,
1065 					      SINGLE_DEPTH_NESTING);
1066 			ret = 1;
1067 		} else
1068 			raw_spin_lock_nested(&busiest->lock,
1069 					      SINGLE_DEPTH_NESTING);
1070 	}
1071 	return ret;
1072 }
1073 
1074 #endif /* CONFIG_PREEMPT */
1075 
1076 /*
1077  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1078  */
1079 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1080 {
1081 	if (unlikely(!irqs_disabled())) {
1082 		/* printk() doesn't work good under rq->lock */
1083 		raw_spin_unlock(&this_rq->lock);
1084 		BUG_ON(1);
1085 	}
1086 
1087 	return _double_lock_balance(this_rq, busiest);
1088 }
1089 
1090 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1091 	__releases(busiest->lock)
1092 {
1093 	raw_spin_unlock(&busiest->lock);
1094 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1095 }
1096 
1097 /*
1098  * double_rq_lock - safely lock two runqueues
1099  *
1100  * Note this does not disable interrupts like task_rq_lock,
1101  * you need to do so manually before calling.
1102  */
1103 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1104 	__acquires(rq1->lock)
1105 	__acquires(rq2->lock)
1106 {
1107 	BUG_ON(!irqs_disabled());
1108 	if (rq1 == rq2) {
1109 		raw_spin_lock(&rq1->lock);
1110 		__acquire(rq2->lock);	/* Fake it out ;) */
1111 	} else {
1112 		if (rq1 < rq2) {
1113 			raw_spin_lock(&rq1->lock);
1114 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1115 		} else {
1116 			raw_spin_lock(&rq2->lock);
1117 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1118 		}
1119 	}
1120 }
1121 
1122 /*
1123  * double_rq_unlock - safely unlock two runqueues
1124  *
1125  * Note this does not restore interrupts like task_rq_unlock,
1126  * you need to do so manually after calling.
1127  */
1128 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1129 	__releases(rq1->lock)
1130 	__releases(rq2->lock)
1131 {
1132 	raw_spin_unlock(&rq1->lock);
1133 	if (rq1 != rq2)
1134 		raw_spin_unlock(&rq2->lock);
1135 	else
1136 		__release(rq2->lock);
1137 }
1138 
1139 #else /* CONFIG_SMP */
1140 
1141 /*
1142  * double_rq_lock - safely lock two runqueues
1143  *
1144  * Note this does not disable interrupts like task_rq_lock,
1145  * you need to do so manually before calling.
1146  */
1147 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1148 	__acquires(rq1->lock)
1149 	__acquires(rq2->lock)
1150 {
1151 	BUG_ON(!irqs_disabled());
1152 	BUG_ON(rq1 != rq2);
1153 	raw_spin_lock(&rq1->lock);
1154 	__acquire(rq2->lock);	/* Fake it out ;) */
1155 }
1156 
1157 /*
1158  * double_rq_unlock - safely unlock two runqueues
1159  *
1160  * Note this does not restore interrupts like task_rq_unlock,
1161  * you need to do so manually after calling.
1162  */
1163 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1164 	__releases(rq1->lock)
1165 	__releases(rq2->lock)
1166 {
1167 	BUG_ON(rq1 != rq2);
1168 	raw_spin_unlock(&rq1->lock);
1169 	__release(rq2->lock);
1170 }
1171 
1172 #endif
1173 
1174 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1175 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1176 extern void print_cfs_stats(struct seq_file *m, int cpu);
1177 extern void print_rt_stats(struct seq_file *m, int cpu);
1178 
1179 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1180 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1181 
1182 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1183 
1184 #ifdef CONFIG_NO_HZ
1185 enum rq_nohz_flag_bits {
1186 	NOHZ_TICK_STOPPED,
1187 	NOHZ_BALANCE_KICK,
1188 	NOHZ_IDLE,
1189 };
1190 
1191 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
1192 #endif
1193 
1194 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1195 
1196 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1197 DECLARE_PER_CPU(u64, cpu_softirq_time);
1198 
1199 #ifndef CONFIG_64BIT
1200 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1201 
1202 static inline void irq_time_write_begin(void)
1203 {
1204 	__this_cpu_inc(irq_time_seq.sequence);
1205 	smp_wmb();
1206 }
1207 
1208 static inline void irq_time_write_end(void)
1209 {
1210 	smp_wmb();
1211 	__this_cpu_inc(irq_time_seq.sequence);
1212 }
1213 
1214 static inline u64 irq_time_read(int cpu)
1215 {
1216 	u64 irq_time;
1217 	unsigned seq;
1218 
1219 	do {
1220 		seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1221 		irq_time = per_cpu(cpu_softirq_time, cpu) +
1222 			   per_cpu(cpu_hardirq_time, cpu);
1223 	} while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1224 
1225 	return irq_time;
1226 }
1227 #else /* CONFIG_64BIT */
1228 static inline void irq_time_write_begin(void)
1229 {
1230 }
1231 
1232 static inline void irq_time_write_end(void)
1233 {
1234 }
1235 
1236 static inline u64 irq_time_read(int cpu)
1237 {
1238 	return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1239 }
1240 #endif /* CONFIG_64BIT */
1241 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1242