xref: /openbmc/linux/kernel/sched/sched.h (revision 4f6cce39)
1 
2 #include <linux/sched.h>
3 #include <linux/sched/autogroup.h>
4 #include <linux/sched/sysctl.h>
5 #include <linux/sched/topology.h>
6 #include <linux/sched/rt.h>
7 #include <linux/sched/deadline.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/wake_q.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/numa_balancing.h>
12 #include <linux/sched/mm.h>
13 #include <linux/sched/cpufreq.h>
14 #include <linux/sched/stat.h>
15 #include <linux/sched/nohz.h>
16 #include <linux/sched/debug.h>
17 #include <linux/sched/hotplug.h>
18 #include <linux/sched/task.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/sched/cputime.h>
21 #include <linux/sched/init.h>
22 
23 #include <linux/u64_stats_sync.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/binfmts.h>
26 #include <linux/mutex.h>
27 #include <linux/spinlock.h>
28 #include <linux/stop_machine.h>
29 #include <linux/irq_work.h>
30 #include <linux/tick.h>
31 #include <linux/slab.h>
32 
33 #ifdef CONFIG_PARAVIRT
34 #include <asm/paravirt.h>
35 #endif
36 
37 #include "cpupri.h"
38 #include "cpudeadline.h"
39 #include "cpuacct.h"
40 
41 #ifdef CONFIG_SCHED_DEBUG
42 #define SCHED_WARN_ON(x)	WARN_ONCE(x, #x)
43 #else
44 #define SCHED_WARN_ON(x)	((void)(x))
45 #endif
46 
47 struct rq;
48 struct cpuidle_state;
49 
50 /* task_struct::on_rq states: */
51 #define TASK_ON_RQ_QUEUED	1
52 #define TASK_ON_RQ_MIGRATING	2
53 
54 extern __read_mostly int scheduler_running;
55 
56 extern unsigned long calc_load_update;
57 extern atomic_long_t calc_load_tasks;
58 
59 extern void calc_global_load_tick(struct rq *this_rq);
60 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
61 
62 #ifdef CONFIG_SMP
63 extern void cpu_load_update_active(struct rq *this_rq);
64 #else
65 static inline void cpu_load_update_active(struct rq *this_rq) { }
66 #endif
67 
68 /*
69  * Helpers for converting nanosecond timing to jiffy resolution
70  */
71 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
72 
73 /*
74  * Increase resolution of nice-level calculations for 64-bit architectures.
75  * The extra resolution improves shares distribution and load balancing of
76  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
77  * hierarchies, especially on larger systems. This is not a user-visible change
78  * and does not change the user-interface for setting shares/weights.
79  *
80  * We increase resolution only if we have enough bits to allow this increased
81  * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
82  * pretty high and the returns do not justify the increased costs.
83  *
84  * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
85  * increase coverage and consistency always enable it on 64bit platforms.
86  */
87 #ifdef CONFIG_64BIT
88 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
89 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
90 # define scale_load_down(w)	((w) >> SCHED_FIXEDPOINT_SHIFT)
91 #else
92 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
93 # define scale_load(w)		(w)
94 # define scale_load_down(w)	(w)
95 #endif
96 
97 /*
98  * Task weight (visible to users) and its load (invisible to users) have
99  * independent resolution, but they should be well calibrated. We use
100  * scale_load() and scale_load_down(w) to convert between them. The
101  * following must be true:
102  *
103  *  scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
104  *
105  */
106 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
107 
108 /*
109  * Single value that decides SCHED_DEADLINE internal math precision.
110  * 10 -> just above 1us
111  * 9  -> just above 0.5us
112  */
113 #define DL_SCALE (10)
114 
115 /*
116  * These are the 'tuning knobs' of the scheduler:
117  */
118 
119 /*
120  * single value that denotes runtime == period, ie unlimited time.
121  */
122 #define RUNTIME_INF	((u64)~0ULL)
123 
124 static inline int idle_policy(int policy)
125 {
126 	return policy == SCHED_IDLE;
127 }
128 static inline int fair_policy(int policy)
129 {
130 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
131 }
132 
133 static inline int rt_policy(int policy)
134 {
135 	return policy == SCHED_FIFO || policy == SCHED_RR;
136 }
137 
138 static inline int dl_policy(int policy)
139 {
140 	return policy == SCHED_DEADLINE;
141 }
142 static inline bool valid_policy(int policy)
143 {
144 	return idle_policy(policy) || fair_policy(policy) ||
145 		rt_policy(policy) || dl_policy(policy);
146 }
147 
148 static inline int task_has_rt_policy(struct task_struct *p)
149 {
150 	return rt_policy(p->policy);
151 }
152 
153 static inline int task_has_dl_policy(struct task_struct *p)
154 {
155 	return dl_policy(p->policy);
156 }
157 
158 /*
159  * Tells if entity @a should preempt entity @b.
160  */
161 static inline bool
162 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
163 {
164 	return dl_time_before(a->deadline, b->deadline);
165 }
166 
167 /*
168  * This is the priority-queue data structure of the RT scheduling class:
169  */
170 struct rt_prio_array {
171 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
172 	struct list_head queue[MAX_RT_PRIO];
173 };
174 
175 struct rt_bandwidth {
176 	/* nests inside the rq lock: */
177 	raw_spinlock_t		rt_runtime_lock;
178 	ktime_t			rt_period;
179 	u64			rt_runtime;
180 	struct hrtimer		rt_period_timer;
181 	unsigned int		rt_period_active;
182 };
183 
184 void __dl_clear_params(struct task_struct *p);
185 
186 /*
187  * To keep the bandwidth of -deadline tasks and groups under control
188  * we need some place where:
189  *  - store the maximum -deadline bandwidth of the system (the group);
190  *  - cache the fraction of that bandwidth that is currently allocated.
191  *
192  * This is all done in the data structure below. It is similar to the
193  * one used for RT-throttling (rt_bandwidth), with the main difference
194  * that, since here we are only interested in admission control, we
195  * do not decrease any runtime while the group "executes", neither we
196  * need a timer to replenish it.
197  *
198  * With respect to SMP, the bandwidth is given on a per-CPU basis,
199  * meaning that:
200  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
201  *  - dl_total_bw array contains, in the i-eth element, the currently
202  *    allocated bandwidth on the i-eth CPU.
203  * Moreover, groups consume bandwidth on each CPU, while tasks only
204  * consume bandwidth on the CPU they're running on.
205  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
206  * that will be shown the next time the proc or cgroup controls will
207  * be red. It on its turn can be changed by writing on its own
208  * control.
209  */
210 struct dl_bandwidth {
211 	raw_spinlock_t dl_runtime_lock;
212 	u64 dl_runtime;
213 	u64 dl_period;
214 };
215 
216 static inline int dl_bandwidth_enabled(void)
217 {
218 	return sysctl_sched_rt_runtime >= 0;
219 }
220 
221 extern struct dl_bw *dl_bw_of(int i);
222 
223 struct dl_bw {
224 	raw_spinlock_t lock;
225 	u64 bw, total_bw;
226 };
227 
228 static inline
229 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
230 {
231 	dl_b->total_bw -= tsk_bw;
232 }
233 
234 static inline
235 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
236 {
237 	dl_b->total_bw += tsk_bw;
238 }
239 
240 static inline
241 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
242 {
243 	return dl_b->bw != -1 &&
244 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
245 }
246 
247 extern void init_dl_bw(struct dl_bw *dl_b);
248 
249 #ifdef CONFIG_CGROUP_SCHED
250 
251 #include <linux/cgroup.h>
252 
253 struct cfs_rq;
254 struct rt_rq;
255 
256 extern struct list_head task_groups;
257 
258 struct cfs_bandwidth {
259 #ifdef CONFIG_CFS_BANDWIDTH
260 	raw_spinlock_t lock;
261 	ktime_t period;
262 	u64 quota, runtime;
263 	s64 hierarchical_quota;
264 	u64 runtime_expires;
265 
266 	int idle, period_active;
267 	struct hrtimer period_timer, slack_timer;
268 	struct list_head throttled_cfs_rq;
269 
270 	/* statistics */
271 	int nr_periods, nr_throttled;
272 	u64 throttled_time;
273 #endif
274 };
275 
276 /* task group related information */
277 struct task_group {
278 	struct cgroup_subsys_state css;
279 
280 #ifdef CONFIG_FAIR_GROUP_SCHED
281 	/* schedulable entities of this group on each cpu */
282 	struct sched_entity **se;
283 	/* runqueue "owned" by this group on each cpu */
284 	struct cfs_rq **cfs_rq;
285 	unsigned long shares;
286 
287 #ifdef	CONFIG_SMP
288 	/*
289 	 * load_avg can be heavily contended at clock tick time, so put
290 	 * it in its own cacheline separated from the fields above which
291 	 * will also be accessed at each tick.
292 	 */
293 	atomic_long_t load_avg ____cacheline_aligned;
294 #endif
295 #endif
296 
297 #ifdef CONFIG_RT_GROUP_SCHED
298 	struct sched_rt_entity **rt_se;
299 	struct rt_rq **rt_rq;
300 
301 	struct rt_bandwidth rt_bandwidth;
302 #endif
303 
304 	struct rcu_head rcu;
305 	struct list_head list;
306 
307 	struct task_group *parent;
308 	struct list_head siblings;
309 	struct list_head children;
310 
311 #ifdef CONFIG_SCHED_AUTOGROUP
312 	struct autogroup *autogroup;
313 #endif
314 
315 	struct cfs_bandwidth cfs_bandwidth;
316 };
317 
318 #ifdef CONFIG_FAIR_GROUP_SCHED
319 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
320 
321 /*
322  * A weight of 0 or 1 can cause arithmetics problems.
323  * A weight of a cfs_rq is the sum of weights of which entities
324  * are queued on this cfs_rq, so a weight of a entity should not be
325  * too large, so as the shares value of a task group.
326  * (The default weight is 1024 - so there's no practical
327  *  limitation from this.)
328  */
329 #define MIN_SHARES	(1UL <<  1)
330 #define MAX_SHARES	(1UL << 18)
331 #endif
332 
333 typedef int (*tg_visitor)(struct task_group *, void *);
334 
335 extern int walk_tg_tree_from(struct task_group *from,
336 			     tg_visitor down, tg_visitor up, void *data);
337 
338 /*
339  * Iterate the full tree, calling @down when first entering a node and @up when
340  * leaving it for the final time.
341  *
342  * Caller must hold rcu_lock or sufficient equivalent.
343  */
344 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
345 {
346 	return walk_tg_tree_from(&root_task_group, down, up, data);
347 }
348 
349 extern int tg_nop(struct task_group *tg, void *data);
350 
351 extern void free_fair_sched_group(struct task_group *tg);
352 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
353 extern void online_fair_sched_group(struct task_group *tg);
354 extern void unregister_fair_sched_group(struct task_group *tg);
355 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
356 			struct sched_entity *se, int cpu,
357 			struct sched_entity *parent);
358 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
359 
360 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
361 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
362 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
363 
364 extern void free_rt_sched_group(struct task_group *tg);
365 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
366 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
367 		struct sched_rt_entity *rt_se, int cpu,
368 		struct sched_rt_entity *parent);
369 
370 extern struct task_group *sched_create_group(struct task_group *parent);
371 extern void sched_online_group(struct task_group *tg,
372 			       struct task_group *parent);
373 extern void sched_destroy_group(struct task_group *tg);
374 extern void sched_offline_group(struct task_group *tg);
375 
376 extern void sched_move_task(struct task_struct *tsk);
377 
378 #ifdef CONFIG_FAIR_GROUP_SCHED
379 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
380 
381 #ifdef CONFIG_SMP
382 extern void set_task_rq_fair(struct sched_entity *se,
383 			     struct cfs_rq *prev, struct cfs_rq *next);
384 #else /* !CONFIG_SMP */
385 static inline void set_task_rq_fair(struct sched_entity *se,
386 			     struct cfs_rq *prev, struct cfs_rq *next) { }
387 #endif /* CONFIG_SMP */
388 #endif /* CONFIG_FAIR_GROUP_SCHED */
389 
390 #else /* CONFIG_CGROUP_SCHED */
391 
392 struct cfs_bandwidth { };
393 
394 #endif	/* CONFIG_CGROUP_SCHED */
395 
396 /* CFS-related fields in a runqueue */
397 struct cfs_rq {
398 	struct load_weight load;
399 	unsigned int nr_running, h_nr_running;
400 
401 	u64 exec_clock;
402 	u64 min_vruntime;
403 #ifndef CONFIG_64BIT
404 	u64 min_vruntime_copy;
405 #endif
406 
407 	struct rb_root tasks_timeline;
408 	struct rb_node *rb_leftmost;
409 
410 	/*
411 	 * 'curr' points to currently running entity on this cfs_rq.
412 	 * It is set to NULL otherwise (i.e when none are currently running).
413 	 */
414 	struct sched_entity *curr, *next, *last, *skip;
415 
416 #ifdef	CONFIG_SCHED_DEBUG
417 	unsigned int nr_spread_over;
418 #endif
419 
420 #ifdef CONFIG_SMP
421 	/*
422 	 * CFS load tracking
423 	 */
424 	struct sched_avg avg;
425 	u64 runnable_load_sum;
426 	unsigned long runnable_load_avg;
427 #ifdef CONFIG_FAIR_GROUP_SCHED
428 	unsigned long tg_load_avg_contrib;
429 	unsigned long propagate_avg;
430 #endif
431 	atomic_long_t removed_load_avg, removed_util_avg;
432 #ifndef CONFIG_64BIT
433 	u64 load_last_update_time_copy;
434 #endif
435 
436 #ifdef CONFIG_FAIR_GROUP_SCHED
437 	/*
438 	 *   h_load = weight * f(tg)
439 	 *
440 	 * Where f(tg) is the recursive weight fraction assigned to
441 	 * this group.
442 	 */
443 	unsigned long h_load;
444 	u64 last_h_load_update;
445 	struct sched_entity *h_load_next;
446 #endif /* CONFIG_FAIR_GROUP_SCHED */
447 #endif /* CONFIG_SMP */
448 
449 #ifdef CONFIG_FAIR_GROUP_SCHED
450 	struct rq *rq;	/* cpu runqueue to which this cfs_rq is attached */
451 
452 	/*
453 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
454 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
455 	 * (like users, containers etc.)
456 	 *
457 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
458 	 * list is used during load balance.
459 	 */
460 	int on_list;
461 	struct list_head leaf_cfs_rq_list;
462 	struct task_group *tg;	/* group that "owns" this runqueue */
463 
464 #ifdef CONFIG_CFS_BANDWIDTH
465 	int runtime_enabled;
466 	u64 runtime_expires;
467 	s64 runtime_remaining;
468 
469 	u64 throttled_clock, throttled_clock_task;
470 	u64 throttled_clock_task_time;
471 	int throttled, throttle_count;
472 	struct list_head throttled_list;
473 #endif /* CONFIG_CFS_BANDWIDTH */
474 #endif /* CONFIG_FAIR_GROUP_SCHED */
475 };
476 
477 static inline int rt_bandwidth_enabled(void)
478 {
479 	return sysctl_sched_rt_runtime >= 0;
480 }
481 
482 /* RT IPI pull logic requires IRQ_WORK */
483 #ifdef CONFIG_IRQ_WORK
484 # define HAVE_RT_PUSH_IPI
485 #endif
486 
487 /* Real-Time classes' related field in a runqueue: */
488 struct rt_rq {
489 	struct rt_prio_array active;
490 	unsigned int rt_nr_running;
491 	unsigned int rr_nr_running;
492 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
493 	struct {
494 		int curr; /* highest queued rt task prio */
495 #ifdef CONFIG_SMP
496 		int next; /* next highest */
497 #endif
498 	} highest_prio;
499 #endif
500 #ifdef CONFIG_SMP
501 	unsigned long rt_nr_migratory;
502 	unsigned long rt_nr_total;
503 	int overloaded;
504 	struct plist_head pushable_tasks;
505 #ifdef HAVE_RT_PUSH_IPI
506 	int push_flags;
507 	int push_cpu;
508 	struct irq_work push_work;
509 	raw_spinlock_t push_lock;
510 #endif
511 #endif /* CONFIG_SMP */
512 	int rt_queued;
513 
514 	int rt_throttled;
515 	u64 rt_time;
516 	u64 rt_runtime;
517 	/* Nests inside the rq lock: */
518 	raw_spinlock_t rt_runtime_lock;
519 
520 #ifdef CONFIG_RT_GROUP_SCHED
521 	unsigned long rt_nr_boosted;
522 
523 	struct rq *rq;
524 	struct task_group *tg;
525 #endif
526 };
527 
528 /* Deadline class' related fields in a runqueue */
529 struct dl_rq {
530 	/* runqueue is an rbtree, ordered by deadline */
531 	struct rb_root rb_root;
532 	struct rb_node *rb_leftmost;
533 
534 	unsigned long dl_nr_running;
535 
536 #ifdef CONFIG_SMP
537 	/*
538 	 * Deadline values of the currently executing and the
539 	 * earliest ready task on this rq. Caching these facilitates
540 	 * the decision wether or not a ready but not running task
541 	 * should migrate somewhere else.
542 	 */
543 	struct {
544 		u64 curr;
545 		u64 next;
546 	} earliest_dl;
547 
548 	unsigned long dl_nr_migratory;
549 	int overloaded;
550 
551 	/*
552 	 * Tasks on this rq that can be pushed away. They are kept in
553 	 * an rb-tree, ordered by tasks' deadlines, with caching
554 	 * of the leftmost (earliest deadline) element.
555 	 */
556 	struct rb_root pushable_dl_tasks_root;
557 	struct rb_node *pushable_dl_tasks_leftmost;
558 #else
559 	struct dl_bw dl_bw;
560 #endif
561 };
562 
563 #ifdef CONFIG_SMP
564 
565 static inline bool sched_asym_prefer(int a, int b)
566 {
567 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
568 }
569 
570 /*
571  * We add the notion of a root-domain which will be used to define per-domain
572  * variables. Each exclusive cpuset essentially defines an island domain by
573  * fully partitioning the member cpus from any other cpuset. Whenever a new
574  * exclusive cpuset is created, we also create and attach a new root-domain
575  * object.
576  *
577  */
578 struct root_domain {
579 	atomic_t refcount;
580 	atomic_t rto_count;
581 	struct rcu_head rcu;
582 	cpumask_var_t span;
583 	cpumask_var_t online;
584 
585 	/* Indicate more than one runnable task for any CPU */
586 	bool overload;
587 
588 	/*
589 	 * The bit corresponding to a CPU gets set here if such CPU has more
590 	 * than one runnable -deadline task (as it is below for RT tasks).
591 	 */
592 	cpumask_var_t dlo_mask;
593 	atomic_t dlo_count;
594 	struct dl_bw dl_bw;
595 	struct cpudl cpudl;
596 
597 	/*
598 	 * The "RT overload" flag: it gets set if a CPU has more than
599 	 * one runnable RT task.
600 	 */
601 	cpumask_var_t rto_mask;
602 	struct cpupri cpupri;
603 
604 	unsigned long max_cpu_capacity;
605 };
606 
607 extern struct root_domain def_root_domain;
608 extern struct mutex sched_domains_mutex;
609 extern cpumask_var_t fallback_doms;
610 extern cpumask_var_t sched_domains_tmpmask;
611 
612 extern void init_defrootdomain(void);
613 extern int init_sched_domains(const struct cpumask *cpu_map);
614 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
615 
616 #endif /* CONFIG_SMP */
617 
618 /*
619  * This is the main, per-CPU runqueue data structure.
620  *
621  * Locking rule: those places that want to lock multiple runqueues
622  * (such as the load balancing or the thread migration code), lock
623  * acquire operations must be ordered by ascending &runqueue.
624  */
625 struct rq {
626 	/* runqueue lock: */
627 	raw_spinlock_t lock;
628 
629 	/*
630 	 * nr_running and cpu_load should be in the same cacheline because
631 	 * remote CPUs use both these fields when doing load calculation.
632 	 */
633 	unsigned int nr_running;
634 #ifdef CONFIG_NUMA_BALANCING
635 	unsigned int nr_numa_running;
636 	unsigned int nr_preferred_running;
637 #endif
638 	#define CPU_LOAD_IDX_MAX 5
639 	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
640 #ifdef CONFIG_NO_HZ_COMMON
641 #ifdef CONFIG_SMP
642 	unsigned long last_load_update_tick;
643 #endif /* CONFIG_SMP */
644 	unsigned long nohz_flags;
645 #endif /* CONFIG_NO_HZ_COMMON */
646 #ifdef CONFIG_NO_HZ_FULL
647 	unsigned long last_sched_tick;
648 #endif
649 	/* capture load from *all* tasks on this cpu: */
650 	struct load_weight load;
651 	unsigned long nr_load_updates;
652 	u64 nr_switches;
653 
654 	struct cfs_rq cfs;
655 	struct rt_rq rt;
656 	struct dl_rq dl;
657 
658 #ifdef CONFIG_FAIR_GROUP_SCHED
659 	/* list of leaf cfs_rq on this cpu: */
660 	struct list_head leaf_cfs_rq_list;
661 	struct list_head *tmp_alone_branch;
662 #endif /* CONFIG_FAIR_GROUP_SCHED */
663 
664 	/*
665 	 * This is part of a global counter where only the total sum
666 	 * over all CPUs matters. A task can increase this counter on
667 	 * one CPU and if it got migrated afterwards it may decrease
668 	 * it on another CPU. Always updated under the runqueue lock:
669 	 */
670 	unsigned long nr_uninterruptible;
671 
672 	struct task_struct *curr, *idle, *stop;
673 	unsigned long next_balance;
674 	struct mm_struct *prev_mm;
675 
676 	unsigned int clock_update_flags;
677 	u64 clock;
678 	u64 clock_task;
679 
680 	atomic_t nr_iowait;
681 
682 #ifdef CONFIG_SMP
683 	struct root_domain *rd;
684 	struct sched_domain *sd;
685 
686 	unsigned long cpu_capacity;
687 	unsigned long cpu_capacity_orig;
688 
689 	struct callback_head *balance_callback;
690 
691 	unsigned char idle_balance;
692 	/* For active balancing */
693 	int active_balance;
694 	int push_cpu;
695 	struct cpu_stop_work active_balance_work;
696 	/* cpu of this runqueue: */
697 	int cpu;
698 	int online;
699 
700 	struct list_head cfs_tasks;
701 
702 	u64 rt_avg;
703 	u64 age_stamp;
704 	u64 idle_stamp;
705 	u64 avg_idle;
706 
707 	/* This is used to determine avg_idle's max value */
708 	u64 max_idle_balance_cost;
709 #endif
710 
711 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
712 	u64 prev_irq_time;
713 #endif
714 #ifdef CONFIG_PARAVIRT
715 	u64 prev_steal_time;
716 #endif
717 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
718 	u64 prev_steal_time_rq;
719 #endif
720 
721 	/* calc_load related fields */
722 	unsigned long calc_load_update;
723 	long calc_load_active;
724 
725 #ifdef CONFIG_SCHED_HRTICK
726 #ifdef CONFIG_SMP
727 	int hrtick_csd_pending;
728 	struct call_single_data hrtick_csd;
729 #endif
730 	struct hrtimer hrtick_timer;
731 #endif
732 
733 #ifdef CONFIG_SCHEDSTATS
734 	/* latency stats */
735 	struct sched_info rq_sched_info;
736 	unsigned long long rq_cpu_time;
737 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
738 
739 	/* sys_sched_yield() stats */
740 	unsigned int yld_count;
741 
742 	/* schedule() stats */
743 	unsigned int sched_count;
744 	unsigned int sched_goidle;
745 
746 	/* try_to_wake_up() stats */
747 	unsigned int ttwu_count;
748 	unsigned int ttwu_local;
749 #endif
750 
751 #ifdef CONFIG_SMP
752 	struct llist_head wake_list;
753 #endif
754 
755 #ifdef CONFIG_CPU_IDLE
756 	/* Must be inspected within a rcu lock section */
757 	struct cpuidle_state *idle_state;
758 #endif
759 };
760 
761 static inline int cpu_of(struct rq *rq)
762 {
763 #ifdef CONFIG_SMP
764 	return rq->cpu;
765 #else
766 	return 0;
767 #endif
768 }
769 
770 
771 #ifdef CONFIG_SCHED_SMT
772 
773 extern struct static_key_false sched_smt_present;
774 
775 extern void __update_idle_core(struct rq *rq);
776 
777 static inline void update_idle_core(struct rq *rq)
778 {
779 	if (static_branch_unlikely(&sched_smt_present))
780 		__update_idle_core(rq);
781 }
782 
783 #else
784 static inline void update_idle_core(struct rq *rq) { }
785 #endif
786 
787 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
788 
789 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
790 #define this_rq()		this_cpu_ptr(&runqueues)
791 #define task_rq(p)		cpu_rq(task_cpu(p))
792 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
793 #define raw_rq()		raw_cpu_ptr(&runqueues)
794 
795 static inline u64 __rq_clock_broken(struct rq *rq)
796 {
797 	return READ_ONCE(rq->clock);
798 }
799 
800 /*
801  * rq::clock_update_flags bits
802  *
803  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
804  *  call to __schedule(). This is an optimisation to avoid
805  *  neighbouring rq clock updates.
806  *
807  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
808  *  in effect and calls to update_rq_clock() are being ignored.
809  *
810  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
811  *  made to update_rq_clock() since the last time rq::lock was pinned.
812  *
813  * If inside of __schedule(), clock_update_flags will have been
814  * shifted left (a left shift is a cheap operation for the fast path
815  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
816  *
817  *	if (rq-clock_update_flags >= RQCF_UPDATED)
818  *
819  * to check if %RQCF_UPADTED is set. It'll never be shifted more than
820  * one position though, because the next rq_unpin_lock() will shift it
821  * back.
822  */
823 #define RQCF_REQ_SKIP	0x01
824 #define RQCF_ACT_SKIP	0x02
825 #define RQCF_UPDATED	0x04
826 
827 static inline void assert_clock_updated(struct rq *rq)
828 {
829 	/*
830 	 * The only reason for not seeing a clock update since the
831 	 * last rq_pin_lock() is if we're currently skipping updates.
832 	 */
833 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
834 }
835 
836 static inline u64 rq_clock(struct rq *rq)
837 {
838 	lockdep_assert_held(&rq->lock);
839 	assert_clock_updated(rq);
840 
841 	return rq->clock;
842 }
843 
844 static inline u64 rq_clock_task(struct rq *rq)
845 {
846 	lockdep_assert_held(&rq->lock);
847 	assert_clock_updated(rq);
848 
849 	return rq->clock_task;
850 }
851 
852 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
853 {
854 	lockdep_assert_held(&rq->lock);
855 	if (skip)
856 		rq->clock_update_flags |= RQCF_REQ_SKIP;
857 	else
858 		rq->clock_update_flags &= ~RQCF_REQ_SKIP;
859 }
860 
861 struct rq_flags {
862 	unsigned long flags;
863 	struct pin_cookie cookie;
864 #ifdef CONFIG_SCHED_DEBUG
865 	/*
866 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
867 	 * current pin context is stashed here in case it needs to be
868 	 * restored in rq_repin_lock().
869 	 */
870 	unsigned int clock_update_flags;
871 #endif
872 };
873 
874 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
875 {
876 	rf->cookie = lockdep_pin_lock(&rq->lock);
877 
878 #ifdef CONFIG_SCHED_DEBUG
879 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
880 	rf->clock_update_flags = 0;
881 #endif
882 }
883 
884 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
885 {
886 #ifdef CONFIG_SCHED_DEBUG
887 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
888 		rf->clock_update_flags = RQCF_UPDATED;
889 #endif
890 
891 	lockdep_unpin_lock(&rq->lock, rf->cookie);
892 }
893 
894 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
895 {
896 	lockdep_repin_lock(&rq->lock, rf->cookie);
897 
898 #ifdef CONFIG_SCHED_DEBUG
899 	/*
900 	 * Restore the value we stashed in @rf for this pin context.
901 	 */
902 	rq->clock_update_flags |= rf->clock_update_flags;
903 #endif
904 }
905 
906 #ifdef CONFIG_NUMA
907 enum numa_topology_type {
908 	NUMA_DIRECT,
909 	NUMA_GLUELESS_MESH,
910 	NUMA_BACKPLANE,
911 };
912 extern enum numa_topology_type sched_numa_topology_type;
913 extern int sched_max_numa_distance;
914 extern bool find_numa_distance(int distance);
915 #endif
916 
917 #ifdef CONFIG_NUMA
918 extern void sched_init_numa(void);
919 extern void sched_domains_numa_masks_set(unsigned int cpu);
920 extern void sched_domains_numa_masks_clear(unsigned int cpu);
921 #else
922 static inline void sched_init_numa(void) { }
923 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
924 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
925 #endif
926 
927 #ifdef CONFIG_NUMA_BALANCING
928 /* The regions in numa_faults array from task_struct */
929 enum numa_faults_stats {
930 	NUMA_MEM = 0,
931 	NUMA_CPU,
932 	NUMA_MEMBUF,
933 	NUMA_CPUBUF
934 };
935 extern void sched_setnuma(struct task_struct *p, int node);
936 extern int migrate_task_to(struct task_struct *p, int cpu);
937 extern int migrate_swap(struct task_struct *, struct task_struct *);
938 #endif /* CONFIG_NUMA_BALANCING */
939 
940 #ifdef CONFIG_SMP
941 
942 static inline void
943 queue_balance_callback(struct rq *rq,
944 		       struct callback_head *head,
945 		       void (*func)(struct rq *rq))
946 {
947 	lockdep_assert_held(&rq->lock);
948 
949 	if (unlikely(head->next))
950 		return;
951 
952 	head->func = (void (*)(struct callback_head *))func;
953 	head->next = rq->balance_callback;
954 	rq->balance_callback = head;
955 }
956 
957 extern void sched_ttwu_pending(void);
958 
959 #define rcu_dereference_check_sched_domain(p) \
960 	rcu_dereference_check((p), \
961 			      lockdep_is_held(&sched_domains_mutex))
962 
963 /*
964  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
965  * See detach_destroy_domains: synchronize_sched for details.
966  *
967  * The domain tree of any CPU may only be accessed from within
968  * preempt-disabled sections.
969  */
970 #define for_each_domain(cpu, __sd) \
971 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
972 			__sd; __sd = __sd->parent)
973 
974 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
975 
976 /**
977  * highest_flag_domain - Return highest sched_domain containing flag.
978  * @cpu:	The cpu whose highest level of sched domain is to
979  *		be returned.
980  * @flag:	The flag to check for the highest sched_domain
981  *		for the given cpu.
982  *
983  * Returns the highest sched_domain of a cpu which contains the given flag.
984  */
985 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
986 {
987 	struct sched_domain *sd, *hsd = NULL;
988 
989 	for_each_domain(cpu, sd) {
990 		if (!(sd->flags & flag))
991 			break;
992 		hsd = sd;
993 	}
994 
995 	return hsd;
996 }
997 
998 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
999 {
1000 	struct sched_domain *sd;
1001 
1002 	for_each_domain(cpu, sd) {
1003 		if (sd->flags & flag)
1004 			break;
1005 	}
1006 
1007 	return sd;
1008 }
1009 
1010 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1011 DECLARE_PER_CPU(int, sd_llc_size);
1012 DECLARE_PER_CPU(int, sd_llc_id);
1013 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1014 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1015 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
1016 
1017 struct sched_group_capacity {
1018 	atomic_t ref;
1019 	/*
1020 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1021 	 * for a single CPU.
1022 	 */
1023 	unsigned long capacity;
1024 	unsigned long min_capacity; /* Min per-CPU capacity in group */
1025 	unsigned long next_update;
1026 	int imbalance; /* XXX unrelated to capacity but shared group state */
1027 
1028 	unsigned long cpumask[0]; /* iteration mask */
1029 };
1030 
1031 struct sched_group {
1032 	struct sched_group *next;	/* Must be a circular list */
1033 	atomic_t ref;
1034 
1035 	unsigned int group_weight;
1036 	struct sched_group_capacity *sgc;
1037 	int asym_prefer_cpu;		/* cpu of highest priority in group */
1038 
1039 	/*
1040 	 * The CPUs this group covers.
1041 	 *
1042 	 * NOTE: this field is variable length. (Allocated dynamically
1043 	 * by attaching extra space to the end of the structure,
1044 	 * depending on how many CPUs the kernel has booted up with)
1045 	 */
1046 	unsigned long cpumask[0];
1047 };
1048 
1049 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
1050 {
1051 	return to_cpumask(sg->cpumask);
1052 }
1053 
1054 /*
1055  * cpumask masking which cpus in the group are allowed to iterate up the domain
1056  * tree.
1057  */
1058 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
1059 {
1060 	return to_cpumask(sg->sgc->cpumask);
1061 }
1062 
1063 /**
1064  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1065  * @group: The group whose first cpu is to be returned.
1066  */
1067 static inline unsigned int group_first_cpu(struct sched_group *group)
1068 {
1069 	return cpumask_first(sched_group_cpus(group));
1070 }
1071 
1072 extern int group_balance_cpu(struct sched_group *sg);
1073 
1074 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1075 void register_sched_domain_sysctl(void);
1076 void unregister_sched_domain_sysctl(void);
1077 #else
1078 static inline void register_sched_domain_sysctl(void)
1079 {
1080 }
1081 static inline void unregister_sched_domain_sysctl(void)
1082 {
1083 }
1084 #endif
1085 
1086 #else
1087 
1088 static inline void sched_ttwu_pending(void) { }
1089 
1090 #endif /* CONFIG_SMP */
1091 
1092 #include "stats.h"
1093 #include "autogroup.h"
1094 
1095 #ifdef CONFIG_CGROUP_SCHED
1096 
1097 /*
1098  * Return the group to which this tasks belongs.
1099  *
1100  * We cannot use task_css() and friends because the cgroup subsystem
1101  * changes that value before the cgroup_subsys::attach() method is called,
1102  * therefore we cannot pin it and might observe the wrong value.
1103  *
1104  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1105  * core changes this before calling sched_move_task().
1106  *
1107  * Instead we use a 'copy' which is updated from sched_move_task() while
1108  * holding both task_struct::pi_lock and rq::lock.
1109  */
1110 static inline struct task_group *task_group(struct task_struct *p)
1111 {
1112 	return p->sched_task_group;
1113 }
1114 
1115 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1116 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1117 {
1118 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1119 	struct task_group *tg = task_group(p);
1120 #endif
1121 
1122 #ifdef CONFIG_FAIR_GROUP_SCHED
1123 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1124 	p->se.cfs_rq = tg->cfs_rq[cpu];
1125 	p->se.parent = tg->se[cpu];
1126 #endif
1127 
1128 #ifdef CONFIG_RT_GROUP_SCHED
1129 	p->rt.rt_rq  = tg->rt_rq[cpu];
1130 	p->rt.parent = tg->rt_se[cpu];
1131 #endif
1132 }
1133 
1134 #else /* CONFIG_CGROUP_SCHED */
1135 
1136 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1137 static inline struct task_group *task_group(struct task_struct *p)
1138 {
1139 	return NULL;
1140 }
1141 
1142 #endif /* CONFIG_CGROUP_SCHED */
1143 
1144 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1145 {
1146 	set_task_rq(p, cpu);
1147 #ifdef CONFIG_SMP
1148 	/*
1149 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1150 	 * successfuly executed on another CPU. We must ensure that updates of
1151 	 * per-task data have been completed by this moment.
1152 	 */
1153 	smp_wmb();
1154 #ifdef CONFIG_THREAD_INFO_IN_TASK
1155 	p->cpu = cpu;
1156 #else
1157 	task_thread_info(p)->cpu = cpu;
1158 #endif
1159 	p->wake_cpu = cpu;
1160 #endif
1161 }
1162 
1163 /*
1164  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1165  */
1166 #ifdef CONFIG_SCHED_DEBUG
1167 # include <linux/static_key.h>
1168 # define const_debug __read_mostly
1169 #else
1170 # define const_debug const
1171 #endif
1172 
1173 extern const_debug unsigned int sysctl_sched_features;
1174 
1175 #define SCHED_FEAT(name, enabled)	\
1176 	__SCHED_FEAT_##name ,
1177 
1178 enum {
1179 #include "features.h"
1180 	__SCHED_FEAT_NR,
1181 };
1182 
1183 #undef SCHED_FEAT
1184 
1185 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1186 #define SCHED_FEAT(name, enabled)					\
1187 static __always_inline bool static_branch_##name(struct static_key *key) \
1188 {									\
1189 	return static_key_##enabled(key);				\
1190 }
1191 
1192 #include "features.h"
1193 
1194 #undef SCHED_FEAT
1195 
1196 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1197 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1198 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1199 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1200 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1201 
1202 extern struct static_key_false sched_numa_balancing;
1203 extern struct static_key_false sched_schedstats;
1204 
1205 static inline u64 global_rt_period(void)
1206 {
1207 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1208 }
1209 
1210 static inline u64 global_rt_runtime(void)
1211 {
1212 	if (sysctl_sched_rt_runtime < 0)
1213 		return RUNTIME_INF;
1214 
1215 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1216 }
1217 
1218 static inline int task_current(struct rq *rq, struct task_struct *p)
1219 {
1220 	return rq->curr == p;
1221 }
1222 
1223 static inline int task_running(struct rq *rq, struct task_struct *p)
1224 {
1225 #ifdef CONFIG_SMP
1226 	return p->on_cpu;
1227 #else
1228 	return task_current(rq, p);
1229 #endif
1230 }
1231 
1232 static inline int task_on_rq_queued(struct task_struct *p)
1233 {
1234 	return p->on_rq == TASK_ON_RQ_QUEUED;
1235 }
1236 
1237 static inline int task_on_rq_migrating(struct task_struct *p)
1238 {
1239 	return p->on_rq == TASK_ON_RQ_MIGRATING;
1240 }
1241 
1242 #ifndef prepare_arch_switch
1243 # define prepare_arch_switch(next)	do { } while (0)
1244 #endif
1245 #ifndef finish_arch_post_lock_switch
1246 # define finish_arch_post_lock_switch()	do { } while (0)
1247 #endif
1248 
1249 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1250 {
1251 #ifdef CONFIG_SMP
1252 	/*
1253 	 * We can optimise this out completely for !SMP, because the
1254 	 * SMP rebalancing from interrupt is the only thing that cares
1255 	 * here.
1256 	 */
1257 	next->on_cpu = 1;
1258 #endif
1259 }
1260 
1261 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1262 {
1263 #ifdef CONFIG_SMP
1264 	/*
1265 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1266 	 * We must ensure this doesn't happen until the switch is completely
1267 	 * finished.
1268 	 *
1269 	 * In particular, the load of prev->state in finish_task_switch() must
1270 	 * happen before this.
1271 	 *
1272 	 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1273 	 */
1274 	smp_store_release(&prev->on_cpu, 0);
1275 #endif
1276 #ifdef CONFIG_DEBUG_SPINLOCK
1277 	/* this is a valid case when another task releases the spinlock */
1278 	rq->lock.owner = current;
1279 #endif
1280 	/*
1281 	 * If we are tracking spinlock dependencies then we have to
1282 	 * fix up the runqueue lock - which gets 'carried over' from
1283 	 * prev into current:
1284 	 */
1285 	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1286 
1287 	raw_spin_unlock_irq(&rq->lock);
1288 }
1289 
1290 /*
1291  * wake flags
1292  */
1293 #define WF_SYNC		0x01		/* waker goes to sleep after wakeup */
1294 #define WF_FORK		0x02		/* child wakeup after fork */
1295 #define WF_MIGRATED	0x4		/* internal use, task got migrated */
1296 
1297 /*
1298  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1299  * of tasks with abnormal "nice" values across CPUs the contribution that
1300  * each task makes to its run queue's load is weighted according to its
1301  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1302  * scaled version of the new time slice allocation that they receive on time
1303  * slice expiry etc.
1304  */
1305 
1306 #define WEIGHT_IDLEPRIO                3
1307 #define WMULT_IDLEPRIO         1431655765
1308 
1309 extern const int sched_prio_to_weight[40];
1310 extern const u32 sched_prio_to_wmult[40];
1311 
1312 /*
1313  * {de,en}queue flags:
1314  *
1315  * DEQUEUE_SLEEP  - task is no longer runnable
1316  * ENQUEUE_WAKEUP - task just became runnable
1317  *
1318  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1319  *                are in a known state which allows modification. Such pairs
1320  *                should preserve as much state as possible.
1321  *
1322  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1323  *        in the runqueue.
1324  *
1325  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
1326  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1327  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
1328  *
1329  */
1330 
1331 #define DEQUEUE_SLEEP		0x01
1332 #define DEQUEUE_SAVE		0x02 /* matches ENQUEUE_RESTORE */
1333 #define DEQUEUE_MOVE		0x04 /* matches ENQUEUE_MOVE */
1334 
1335 #define ENQUEUE_WAKEUP		0x01
1336 #define ENQUEUE_RESTORE		0x02
1337 #define ENQUEUE_MOVE		0x04
1338 
1339 #define ENQUEUE_HEAD		0x08
1340 #define ENQUEUE_REPLENISH	0x10
1341 #ifdef CONFIG_SMP
1342 #define ENQUEUE_MIGRATED	0x20
1343 #else
1344 #define ENQUEUE_MIGRATED	0x00
1345 #endif
1346 
1347 #define RETRY_TASK		((void *)-1UL)
1348 
1349 struct sched_class {
1350 	const struct sched_class *next;
1351 
1352 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1353 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1354 	void (*yield_task) (struct rq *rq);
1355 	bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1356 
1357 	void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1358 
1359 	/*
1360 	 * It is the responsibility of the pick_next_task() method that will
1361 	 * return the next task to call put_prev_task() on the @prev task or
1362 	 * something equivalent.
1363 	 *
1364 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1365 	 * tasks.
1366 	 */
1367 	struct task_struct * (*pick_next_task) (struct rq *rq,
1368 						struct task_struct *prev,
1369 						struct rq_flags *rf);
1370 	void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1371 
1372 #ifdef CONFIG_SMP
1373 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1374 	void (*migrate_task_rq)(struct task_struct *p);
1375 
1376 	void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1377 
1378 	void (*set_cpus_allowed)(struct task_struct *p,
1379 				 const struct cpumask *newmask);
1380 
1381 	void (*rq_online)(struct rq *rq);
1382 	void (*rq_offline)(struct rq *rq);
1383 #endif
1384 
1385 	void (*set_curr_task) (struct rq *rq);
1386 	void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1387 	void (*task_fork) (struct task_struct *p);
1388 	void (*task_dead) (struct task_struct *p);
1389 
1390 	/*
1391 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1392 	 * cannot assume the switched_from/switched_to pair is serliazed by
1393 	 * rq->lock. They are however serialized by p->pi_lock.
1394 	 */
1395 	void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1396 	void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1397 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1398 			     int oldprio);
1399 
1400 	unsigned int (*get_rr_interval) (struct rq *rq,
1401 					 struct task_struct *task);
1402 
1403 	void (*update_curr) (struct rq *rq);
1404 
1405 #define TASK_SET_GROUP  0
1406 #define TASK_MOVE_GROUP	1
1407 
1408 #ifdef CONFIG_FAIR_GROUP_SCHED
1409 	void (*task_change_group) (struct task_struct *p, int type);
1410 #endif
1411 };
1412 
1413 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1414 {
1415 	prev->sched_class->put_prev_task(rq, prev);
1416 }
1417 
1418 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1419 {
1420 	curr->sched_class->set_curr_task(rq);
1421 }
1422 
1423 #define sched_class_highest (&stop_sched_class)
1424 #define for_each_class(class) \
1425    for (class = sched_class_highest; class; class = class->next)
1426 
1427 extern const struct sched_class stop_sched_class;
1428 extern const struct sched_class dl_sched_class;
1429 extern const struct sched_class rt_sched_class;
1430 extern const struct sched_class fair_sched_class;
1431 extern const struct sched_class idle_sched_class;
1432 
1433 
1434 #ifdef CONFIG_SMP
1435 
1436 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1437 
1438 extern void trigger_load_balance(struct rq *rq);
1439 
1440 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1441 
1442 #endif
1443 
1444 #ifdef CONFIG_CPU_IDLE
1445 static inline void idle_set_state(struct rq *rq,
1446 				  struct cpuidle_state *idle_state)
1447 {
1448 	rq->idle_state = idle_state;
1449 }
1450 
1451 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1452 {
1453 	SCHED_WARN_ON(!rcu_read_lock_held());
1454 	return rq->idle_state;
1455 }
1456 #else
1457 static inline void idle_set_state(struct rq *rq,
1458 				  struct cpuidle_state *idle_state)
1459 {
1460 }
1461 
1462 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1463 {
1464 	return NULL;
1465 }
1466 #endif
1467 
1468 extern void sysrq_sched_debug_show(void);
1469 extern void sched_init_granularity(void);
1470 extern void update_max_interval(void);
1471 
1472 extern void init_sched_dl_class(void);
1473 extern void init_sched_rt_class(void);
1474 extern void init_sched_fair_class(void);
1475 
1476 extern void resched_curr(struct rq *rq);
1477 extern void resched_cpu(int cpu);
1478 
1479 extern struct rt_bandwidth def_rt_bandwidth;
1480 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1481 
1482 extern struct dl_bandwidth def_dl_bandwidth;
1483 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1484 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1485 
1486 unsigned long to_ratio(u64 period, u64 runtime);
1487 
1488 extern void init_entity_runnable_average(struct sched_entity *se);
1489 extern void post_init_entity_util_avg(struct sched_entity *se);
1490 
1491 #ifdef CONFIG_NO_HZ_FULL
1492 extern bool sched_can_stop_tick(struct rq *rq);
1493 
1494 /*
1495  * Tick may be needed by tasks in the runqueue depending on their policy and
1496  * requirements. If tick is needed, lets send the target an IPI to kick it out of
1497  * nohz mode if necessary.
1498  */
1499 static inline void sched_update_tick_dependency(struct rq *rq)
1500 {
1501 	int cpu;
1502 
1503 	if (!tick_nohz_full_enabled())
1504 		return;
1505 
1506 	cpu = cpu_of(rq);
1507 
1508 	if (!tick_nohz_full_cpu(cpu))
1509 		return;
1510 
1511 	if (sched_can_stop_tick(rq))
1512 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1513 	else
1514 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1515 }
1516 #else
1517 static inline void sched_update_tick_dependency(struct rq *rq) { }
1518 #endif
1519 
1520 static inline void add_nr_running(struct rq *rq, unsigned count)
1521 {
1522 	unsigned prev_nr = rq->nr_running;
1523 
1524 	rq->nr_running = prev_nr + count;
1525 
1526 	if (prev_nr < 2 && rq->nr_running >= 2) {
1527 #ifdef CONFIG_SMP
1528 		if (!rq->rd->overload)
1529 			rq->rd->overload = true;
1530 #endif
1531 	}
1532 
1533 	sched_update_tick_dependency(rq);
1534 }
1535 
1536 static inline void sub_nr_running(struct rq *rq, unsigned count)
1537 {
1538 	rq->nr_running -= count;
1539 	/* Check if we still need preemption */
1540 	sched_update_tick_dependency(rq);
1541 }
1542 
1543 static inline void rq_last_tick_reset(struct rq *rq)
1544 {
1545 #ifdef CONFIG_NO_HZ_FULL
1546 	rq->last_sched_tick = jiffies;
1547 #endif
1548 }
1549 
1550 extern void update_rq_clock(struct rq *rq);
1551 
1552 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1553 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1554 
1555 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1556 
1557 extern const_debug unsigned int sysctl_sched_time_avg;
1558 extern const_debug unsigned int sysctl_sched_nr_migrate;
1559 extern const_debug unsigned int sysctl_sched_migration_cost;
1560 
1561 static inline u64 sched_avg_period(void)
1562 {
1563 	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1564 }
1565 
1566 #ifdef CONFIG_SCHED_HRTICK
1567 
1568 /*
1569  * Use hrtick when:
1570  *  - enabled by features
1571  *  - hrtimer is actually high res
1572  */
1573 static inline int hrtick_enabled(struct rq *rq)
1574 {
1575 	if (!sched_feat(HRTICK))
1576 		return 0;
1577 	if (!cpu_active(cpu_of(rq)))
1578 		return 0;
1579 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1580 }
1581 
1582 void hrtick_start(struct rq *rq, u64 delay);
1583 
1584 #else
1585 
1586 static inline int hrtick_enabled(struct rq *rq)
1587 {
1588 	return 0;
1589 }
1590 
1591 #endif /* CONFIG_SCHED_HRTICK */
1592 
1593 #ifdef CONFIG_SMP
1594 extern void sched_avg_update(struct rq *rq);
1595 
1596 #ifndef arch_scale_freq_capacity
1597 static __always_inline
1598 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1599 {
1600 	return SCHED_CAPACITY_SCALE;
1601 }
1602 #endif
1603 
1604 #ifndef arch_scale_cpu_capacity
1605 static __always_inline
1606 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1607 {
1608 	if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1609 		return sd->smt_gain / sd->span_weight;
1610 
1611 	return SCHED_CAPACITY_SCALE;
1612 }
1613 #endif
1614 
1615 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1616 {
1617 	rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1618 	sched_avg_update(rq);
1619 }
1620 #else
1621 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1622 static inline void sched_avg_update(struct rq *rq) { }
1623 #endif
1624 
1625 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1626 	__acquires(rq->lock);
1627 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1628 	__acquires(p->pi_lock)
1629 	__acquires(rq->lock);
1630 
1631 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1632 	__releases(rq->lock)
1633 {
1634 	rq_unpin_lock(rq, rf);
1635 	raw_spin_unlock(&rq->lock);
1636 }
1637 
1638 static inline void
1639 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1640 	__releases(rq->lock)
1641 	__releases(p->pi_lock)
1642 {
1643 	rq_unpin_lock(rq, rf);
1644 	raw_spin_unlock(&rq->lock);
1645 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1646 }
1647 
1648 #ifdef CONFIG_SMP
1649 #ifdef CONFIG_PREEMPT
1650 
1651 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1652 
1653 /*
1654  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1655  * way at the expense of forcing extra atomic operations in all
1656  * invocations.  This assures that the double_lock is acquired using the
1657  * same underlying policy as the spinlock_t on this architecture, which
1658  * reduces latency compared to the unfair variant below.  However, it
1659  * also adds more overhead and therefore may reduce throughput.
1660  */
1661 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1662 	__releases(this_rq->lock)
1663 	__acquires(busiest->lock)
1664 	__acquires(this_rq->lock)
1665 {
1666 	raw_spin_unlock(&this_rq->lock);
1667 	double_rq_lock(this_rq, busiest);
1668 
1669 	return 1;
1670 }
1671 
1672 #else
1673 /*
1674  * Unfair double_lock_balance: Optimizes throughput at the expense of
1675  * latency by eliminating extra atomic operations when the locks are
1676  * already in proper order on entry.  This favors lower cpu-ids and will
1677  * grant the double lock to lower cpus over higher ids under contention,
1678  * regardless of entry order into the function.
1679  */
1680 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1681 	__releases(this_rq->lock)
1682 	__acquires(busiest->lock)
1683 	__acquires(this_rq->lock)
1684 {
1685 	int ret = 0;
1686 
1687 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1688 		if (busiest < this_rq) {
1689 			raw_spin_unlock(&this_rq->lock);
1690 			raw_spin_lock(&busiest->lock);
1691 			raw_spin_lock_nested(&this_rq->lock,
1692 					      SINGLE_DEPTH_NESTING);
1693 			ret = 1;
1694 		} else
1695 			raw_spin_lock_nested(&busiest->lock,
1696 					      SINGLE_DEPTH_NESTING);
1697 	}
1698 	return ret;
1699 }
1700 
1701 #endif /* CONFIG_PREEMPT */
1702 
1703 /*
1704  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1705  */
1706 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1707 {
1708 	if (unlikely(!irqs_disabled())) {
1709 		/* printk() doesn't work good under rq->lock */
1710 		raw_spin_unlock(&this_rq->lock);
1711 		BUG_ON(1);
1712 	}
1713 
1714 	return _double_lock_balance(this_rq, busiest);
1715 }
1716 
1717 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1718 	__releases(busiest->lock)
1719 {
1720 	raw_spin_unlock(&busiest->lock);
1721 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1722 }
1723 
1724 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1725 {
1726 	if (l1 > l2)
1727 		swap(l1, l2);
1728 
1729 	spin_lock(l1);
1730 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1731 }
1732 
1733 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1734 {
1735 	if (l1 > l2)
1736 		swap(l1, l2);
1737 
1738 	spin_lock_irq(l1);
1739 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1740 }
1741 
1742 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1743 {
1744 	if (l1 > l2)
1745 		swap(l1, l2);
1746 
1747 	raw_spin_lock(l1);
1748 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1749 }
1750 
1751 /*
1752  * double_rq_lock - safely lock two runqueues
1753  *
1754  * Note this does not disable interrupts like task_rq_lock,
1755  * you need to do so manually before calling.
1756  */
1757 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1758 	__acquires(rq1->lock)
1759 	__acquires(rq2->lock)
1760 {
1761 	BUG_ON(!irqs_disabled());
1762 	if (rq1 == rq2) {
1763 		raw_spin_lock(&rq1->lock);
1764 		__acquire(rq2->lock);	/* Fake it out ;) */
1765 	} else {
1766 		if (rq1 < rq2) {
1767 			raw_spin_lock(&rq1->lock);
1768 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1769 		} else {
1770 			raw_spin_lock(&rq2->lock);
1771 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1772 		}
1773 	}
1774 }
1775 
1776 /*
1777  * double_rq_unlock - safely unlock two runqueues
1778  *
1779  * Note this does not restore interrupts like task_rq_unlock,
1780  * you need to do so manually after calling.
1781  */
1782 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1783 	__releases(rq1->lock)
1784 	__releases(rq2->lock)
1785 {
1786 	raw_spin_unlock(&rq1->lock);
1787 	if (rq1 != rq2)
1788 		raw_spin_unlock(&rq2->lock);
1789 	else
1790 		__release(rq2->lock);
1791 }
1792 
1793 extern void set_rq_online (struct rq *rq);
1794 extern void set_rq_offline(struct rq *rq);
1795 extern bool sched_smp_initialized;
1796 
1797 #else /* CONFIG_SMP */
1798 
1799 /*
1800  * double_rq_lock - safely lock two runqueues
1801  *
1802  * Note this does not disable interrupts like task_rq_lock,
1803  * you need to do so manually before calling.
1804  */
1805 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1806 	__acquires(rq1->lock)
1807 	__acquires(rq2->lock)
1808 {
1809 	BUG_ON(!irqs_disabled());
1810 	BUG_ON(rq1 != rq2);
1811 	raw_spin_lock(&rq1->lock);
1812 	__acquire(rq2->lock);	/* Fake it out ;) */
1813 }
1814 
1815 /*
1816  * double_rq_unlock - safely unlock two runqueues
1817  *
1818  * Note this does not restore interrupts like task_rq_unlock,
1819  * you need to do so manually after calling.
1820  */
1821 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1822 	__releases(rq1->lock)
1823 	__releases(rq2->lock)
1824 {
1825 	BUG_ON(rq1 != rq2);
1826 	raw_spin_unlock(&rq1->lock);
1827 	__release(rq2->lock);
1828 }
1829 
1830 #endif
1831 
1832 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1833 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1834 
1835 #ifdef	CONFIG_SCHED_DEBUG
1836 extern void print_cfs_stats(struct seq_file *m, int cpu);
1837 extern void print_rt_stats(struct seq_file *m, int cpu);
1838 extern void print_dl_stats(struct seq_file *m, int cpu);
1839 extern void
1840 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1841 #ifdef CONFIG_NUMA_BALANCING
1842 extern void
1843 show_numa_stats(struct task_struct *p, struct seq_file *m);
1844 extern void
1845 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1846 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
1847 #endif /* CONFIG_NUMA_BALANCING */
1848 #endif /* CONFIG_SCHED_DEBUG */
1849 
1850 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1851 extern void init_rt_rq(struct rt_rq *rt_rq);
1852 extern void init_dl_rq(struct dl_rq *dl_rq);
1853 
1854 extern void cfs_bandwidth_usage_inc(void);
1855 extern void cfs_bandwidth_usage_dec(void);
1856 
1857 #ifdef CONFIG_NO_HZ_COMMON
1858 enum rq_nohz_flag_bits {
1859 	NOHZ_TICK_STOPPED,
1860 	NOHZ_BALANCE_KICK,
1861 };
1862 
1863 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
1864 
1865 extern void nohz_balance_exit_idle(unsigned int cpu);
1866 #else
1867 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1868 #endif
1869 
1870 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1871 struct irqtime {
1872 	u64			tick_delta;
1873 	u64			irq_start_time;
1874 	struct u64_stats_sync	sync;
1875 };
1876 
1877 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
1878 
1879 static inline u64 irq_time_read(int cpu)
1880 {
1881 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
1882 	u64 *cpustat = kcpustat_cpu(cpu).cpustat;
1883 	unsigned int seq;
1884 	u64 total;
1885 
1886 	do {
1887 		seq = __u64_stats_fetch_begin(&irqtime->sync);
1888 		total = cpustat[CPUTIME_SOFTIRQ] + cpustat[CPUTIME_IRQ];
1889 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
1890 
1891 	return total;
1892 }
1893 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1894 
1895 #ifdef CONFIG_CPU_FREQ
1896 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1897 
1898 /**
1899  * cpufreq_update_util - Take a note about CPU utilization changes.
1900  * @rq: Runqueue to carry out the update for.
1901  * @flags: Update reason flags.
1902  *
1903  * This function is called by the scheduler on the CPU whose utilization is
1904  * being updated.
1905  *
1906  * It can only be called from RCU-sched read-side critical sections.
1907  *
1908  * The way cpufreq is currently arranged requires it to evaluate the CPU
1909  * performance state (frequency/voltage) on a regular basis to prevent it from
1910  * being stuck in a completely inadequate performance level for too long.
1911  * That is not guaranteed to happen if the updates are only triggered from CFS,
1912  * though, because they may not be coming in if RT or deadline tasks are active
1913  * all the time (or there are RT and DL tasks only).
1914  *
1915  * As a workaround for that issue, this function is called by the RT and DL
1916  * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1917  * but that really is a band-aid.  Going forward it should be replaced with
1918  * solutions targeted more specifically at RT and DL tasks.
1919  */
1920 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
1921 {
1922 	struct update_util_data *data;
1923 
1924 	data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1925 	if (data)
1926 		data->func(data, rq_clock(rq), flags);
1927 }
1928 
1929 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags)
1930 {
1931 	if (cpu_of(rq) == smp_processor_id())
1932 		cpufreq_update_util(rq, flags);
1933 }
1934 #else
1935 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
1936 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {}
1937 #endif /* CONFIG_CPU_FREQ */
1938 
1939 #ifdef arch_scale_freq_capacity
1940 #ifndef arch_scale_freq_invariant
1941 #define arch_scale_freq_invariant()	(true)
1942 #endif
1943 #else /* arch_scale_freq_capacity */
1944 #define arch_scale_freq_invariant()	(false)
1945 #endif
1946