xref: /openbmc/linux/kernel/sched/sched.h (revision 1cac4f26)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Scheduler internal types and methods:
4  */
5 #include <linux/sched.h>
6 
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
35 
36 #include <uapi/linux/sched/types.h>
37 
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
68 
69 #include <asm/tlb.h>
70 
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
73 #endif
74 
75 #include "cpupri.h"
76 #include "cpudeadline.h"
77 
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x)	WARN_ONCE(x, #x)
80 #else
81 # define SCHED_WARN_ON(x)	({ (void)(x), 0; })
82 #endif
83 
84 struct rq;
85 struct cpuidle_state;
86 
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED	1
89 #define TASK_ON_RQ_MIGRATING	2
90 
91 extern __read_mostly int scheduler_running;
92 
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
95 
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
98 
99 #ifdef CONFIG_SMP
100 extern void cpu_load_update_active(struct rq *this_rq);
101 #else
102 static inline void cpu_load_update_active(struct rq *this_rq) { }
103 #endif
104 
105 /*
106  * Helpers for converting nanosecond timing to jiffy resolution
107  */
108 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
109 
110 /*
111  * Increase resolution of nice-level calculations for 64-bit architectures.
112  * The extra resolution improves shares distribution and load balancing of
113  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
114  * hierarchies, especially on larger systems. This is not a user-visible change
115  * and does not change the user-interface for setting shares/weights.
116  *
117  * We increase resolution only if we have enough bits to allow this increased
118  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
119  * are pretty high and the returns do not justify the increased costs.
120  *
121  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
122  * increase coverage and consistency always enable it on 64-bit platforms.
123  */
124 #ifdef CONFIG_64BIT
125 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
127 # define scale_load_down(w)	((w) >> SCHED_FIXEDPOINT_SHIFT)
128 #else
129 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
130 # define scale_load(w)		(w)
131 # define scale_load_down(w)	(w)
132 #endif
133 
134 /*
135  * Task weight (visible to users) and its load (invisible to users) have
136  * independent resolution, but they should be well calibrated. We use
137  * scale_load() and scale_load_down(w) to convert between them. The
138  * following must be true:
139  *
140  *  scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
141  *
142  */
143 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
144 
145 /*
146  * Single value that decides SCHED_DEADLINE internal math precision.
147  * 10 -> just above 1us
148  * 9  -> just above 0.5us
149  */
150 #define DL_SCALE		10
151 
152 /*
153  * Single value that denotes runtime == period, ie unlimited time.
154  */
155 #define RUNTIME_INF		((u64)~0ULL)
156 
157 static inline int idle_policy(int policy)
158 {
159 	return policy == SCHED_IDLE;
160 }
161 static inline int fair_policy(int policy)
162 {
163 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
164 }
165 
166 static inline int rt_policy(int policy)
167 {
168 	return policy == SCHED_FIFO || policy == SCHED_RR;
169 }
170 
171 static inline int dl_policy(int policy)
172 {
173 	return policy == SCHED_DEADLINE;
174 }
175 static inline bool valid_policy(int policy)
176 {
177 	return idle_policy(policy) || fair_policy(policy) ||
178 		rt_policy(policy) || dl_policy(policy);
179 }
180 
181 static inline int task_has_idle_policy(struct task_struct *p)
182 {
183 	return idle_policy(p->policy);
184 }
185 
186 static inline int task_has_rt_policy(struct task_struct *p)
187 {
188 	return rt_policy(p->policy);
189 }
190 
191 static inline int task_has_dl_policy(struct task_struct *p)
192 {
193 	return dl_policy(p->policy);
194 }
195 
196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
197 
198 /*
199  * !! For sched_setattr_nocheck() (kernel) only !!
200  *
201  * This is actually gross. :(
202  *
203  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
204  * tasks, but still be able to sleep. We need this on platforms that cannot
205  * atomically change clock frequency. Remove once fast switching will be
206  * available on such platforms.
207  *
208  * SUGOV stands for SchedUtil GOVernor.
209  */
210 #define SCHED_FLAG_SUGOV	0x10000000
211 
212 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
213 {
214 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
215 	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
216 #else
217 	return false;
218 #endif
219 }
220 
221 /*
222  * Tells if entity @a should preempt entity @b.
223  */
224 static inline bool
225 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
226 {
227 	return dl_entity_is_special(a) ||
228 	       dl_time_before(a->deadline, b->deadline);
229 }
230 
231 /*
232  * This is the priority-queue data structure of the RT scheduling class:
233  */
234 struct rt_prio_array {
235 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
236 	struct list_head queue[MAX_RT_PRIO];
237 };
238 
239 struct rt_bandwidth {
240 	/* nests inside the rq lock: */
241 	raw_spinlock_t		rt_runtime_lock;
242 	ktime_t			rt_period;
243 	u64			rt_runtime;
244 	struct hrtimer		rt_period_timer;
245 	unsigned int		rt_period_active;
246 };
247 
248 void __dl_clear_params(struct task_struct *p);
249 
250 /*
251  * To keep the bandwidth of -deadline tasks and groups under control
252  * we need some place where:
253  *  - store the maximum -deadline bandwidth of the system (the group);
254  *  - cache the fraction of that bandwidth that is currently allocated.
255  *
256  * This is all done in the data structure below. It is similar to the
257  * one used for RT-throttling (rt_bandwidth), with the main difference
258  * that, since here we are only interested in admission control, we
259  * do not decrease any runtime while the group "executes", neither we
260  * need a timer to replenish it.
261  *
262  * With respect to SMP, the bandwidth is given on a per-CPU basis,
263  * meaning that:
264  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
265  *  - dl_total_bw array contains, in the i-eth element, the currently
266  *    allocated bandwidth on the i-eth CPU.
267  * Moreover, groups consume bandwidth on each CPU, while tasks only
268  * consume bandwidth on the CPU they're running on.
269  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
270  * that will be shown the next time the proc or cgroup controls will
271  * be red. It on its turn can be changed by writing on its own
272  * control.
273  */
274 struct dl_bandwidth {
275 	raw_spinlock_t		dl_runtime_lock;
276 	u64			dl_runtime;
277 	u64			dl_period;
278 };
279 
280 static inline int dl_bandwidth_enabled(void)
281 {
282 	return sysctl_sched_rt_runtime >= 0;
283 }
284 
285 struct dl_bw {
286 	raw_spinlock_t		lock;
287 	u64			bw;
288 	u64			total_bw;
289 };
290 
291 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
292 
293 static inline
294 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
295 {
296 	dl_b->total_bw -= tsk_bw;
297 	__dl_update(dl_b, (s32)tsk_bw / cpus);
298 }
299 
300 static inline
301 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
302 {
303 	dl_b->total_bw += tsk_bw;
304 	__dl_update(dl_b, -((s32)tsk_bw / cpus));
305 }
306 
307 static inline
308 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
309 {
310 	return dl_b->bw != -1 &&
311 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
312 }
313 
314 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
315 extern void init_dl_bw(struct dl_bw *dl_b);
316 extern int  sched_dl_global_validate(void);
317 extern void sched_dl_do_global(void);
318 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
319 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
320 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
321 extern bool __checkparam_dl(const struct sched_attr *attr);
322 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
323 extern int  dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
324 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
325 extern bool dl_cpu_busy(unsigned int cpu);
326 
327 #ifdef CONFIG_CGROUP_SCHED
328 
329 #include <linux/cgroup.h>
330 #include <linux/psi.h>
331 
332 struct cfs_rq;
333 struct rt_rq;
334 
335 extern struct list_head task_groups;
336 
337 struct cfs_bandwidth {
338 #ifdef CONFIG_CFS_BANDWIDTH
339 	raw_spinlock_t		lock;
340 	ktime_t			period;
341 	u64			quota;
342 	u64			runtime;
343 	s64			hierarchical_quota;
344 	u64			runtime_expires;
345 	int			expires_seq;
346 
347 	short			idle;
348 	short			period_active;
349 	struct hrtimer		period_timer;
350 	struct hrtimer		slack_timer;
351 	struct list_head	throttled_cfs_rq;
352 
353 	/* Statistics: */
354 	int			nr_periods;
355 	int			nr_throttled;
356 	u64			throttled_time;
357 
358 	bool                    distribute_running;
359 #endif
360 };
361 
362 /* Task group related information */
363 struct task_group {
364 	struct cgroup_subsys_state css;
365 
366 #ifdef CONFIG_FAIR_GROUP_SCHED
367 	/* schedulable entities of this group on each CPU */
368 	struct sched_entity	**se;
369 	/* runqueue "owned" by this group on each CPU */
370 	struct cfs_rq		**cfs_rq;
371 	unsigned long		shares;
372 
373 #ifdef	CONFIG_SMP
374 	/*
375 	 * load_avg can be heavily contended at clock tick time, so put
376 	 * it in its own cacheline separated from the fields above which
377 	 * will also be accessed at each tick.
378 	 */
379 	atomic_long_t		load_avg ____cacheline_aligned;
380 #endif
381 #endif
382 
383 #ifdef CONFIG_RT_GROUP_SCHED
384 	struct sched_rt_entity	**rt_se;
385 	struct rt_rq		**rt_rq;
386 
387 	struct rt_bandwidth	rt_bandwidth;
388 #endif
389 
390 	struct rcu_head		rcu;
391 	struct list_head	list;
392 
393 	struct task_group	*parent;
394 	struct list_head	siblings;
395 	struct list_head	children;
396 
397 #ifdef CONFIG_SCHED_AUTOGROUP
398 	struct autogroup	*autogroup;
399 #endif
400 
401 	struct cfs_bandwidth	cfs_bandwidth;
402 };
403 
404 #ifdef CONFIG_FAIR_GROUP_SCHED
405 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
406 
407 /*
408  * A weight of 0 or 1 can cause arithmetics problems.
409  * A weight of a cfs_rq is the sum of weights of which entities
410  * are queued on this cfs_rq, so a weight of a entity should not be
411  * too large, so as the shares value of a task group.
412  * (The default weight is 1024 - so there's no practical
413  *  limitation from this.)
414  */
415 #define MIN_SHARES		(1UL <<  1)
416 #define MAX_SHARES		(1UL << 18)
417 #endif
418 
419 typedef int (*tg_visitor)(struct task_group *, void *);
420 
421 extern int walk_tg_tree_from(struct task_group *from,
422 			     tg_visitor down, tg_visitor up, void *data);
423 
424 /*
425  * Iterate the full tree, calling @down when first entering a node and @up when
426  * leaving it for the final time.
427  *
428  * Caller must hold rcu_lock or sufficient equivalent.
429  */
430 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
431 {
432 	return walk_tg_tree_from(&root_task_group, down, up, data);
433 }
434 
435 extern int tg_nop(struct task_group *tg, void *data);
436 
437 extern void free_fair_sched_group(struct task_group *tg);
438 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
439 extern void online_fair_sched_group(struct task_group *tg);
440 extern void unregister_fair_sched_group(struct task_group *tg);
441 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
442 			struct sched_entity *se, int cpu,
443 			struct sched_entity *parent);
444 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
445 
446 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
447 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
448 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
449 
450 extern void free_rt_sched_group(struct task_group *tg);
451 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
452 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
453 		struct sched_rt_entity *rt_se, int cpu,
454 		struct sched_rt_entity *parent);
455 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
456 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
457 extern long sched_group_rt_runtime(struct task_group *tg);
458 extern long sched_group_rt_period(struct task_group *tg);
459 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
460 
461 extern struct task_group *sched_create_group(struct task_group *parent);
462 extern void sched_online_group(struct task_group *tg,
463 			       struct task_group *parent);
464 extern void sched_destroy_group(struct task_group *tg);
465 extern void sched_offline_group(struct task_group *tg);
466 
467 extern void sched_move_task(struct task_struct *tsk);
468 
469 #ifdef CONFIG_FAIR_GROUP_SCHED
470 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
471 
472 #ifdef CONFIG_SMP
473 extern void set_task_rq_fair(struct sched_entity *se,
474 			     struct cfs_rq *prev, struct cfs_rq *next);
475 #else /* !CONFIG_SMP */
476 static inline void set_task_rq_fair(struct sched_entity *se,
477 			     struct cfs_rq *prev, struct cfs_rq *next) { }
478 #endif /* CONFIG_SMP */
479 #endif /* CONFIG_FAIR_GROUP_SCHED */
480 
481 #else /* CONFIG_CGROUP_SCHED */
482 
483 struct cfs_bandwidth { };
484 
485 #endif	/* CONFIG_CGROUP_SCHED */
486 
487 /* CFS-related fields in a runqueue */
488 struct cfs_rq {
489 	struct load_weight	load;
490 	unsigned long		runnable_weight;
491 	unsigned int		nr_running;
492 	unsigned int		h_nr_running;
493 
494 	u64			exec_clock;
495 	u64			min_vruntime;
496 #ifndef CONFIG_64BIT
497 	u64			min_vruntime_copy;
498 #endif
499 
500 	struct rb_root_cached	tasks_timeline;
501 
502 	/*
503 	 * 'curr' points to currently running entity on this cfs_rq.
504 	 * It is set to NULL otherwise (i.e when none are currently running).
505 	 */
506 	struct sched_entity	*curr;
507 	struct sched_entity	*next;
508 	struct sched_entity	*last;
509 	struct sched_entity	*skip;
510 
511 #ifdef	CONFIG_SCHED_DEBUG
512 	unsigned int		nr_spread_over;
513 #endif
514 
515 #ifdef CONFIG_SMP
516 	/*
517 	 * CFS load tracking
518 	 */
519 	struct sched_avg	avg;
520 #ifndef CONFIG_64BIT
521 	u64			load_last_update_time_copy;
522 #endif
523 	struct {
524 		raw_spinlock_t	lock ____cacheline_aligned;
525 		int		nr;
526 		unsigned long	load_avg;
527 		unsigned long	util_avg;
528 		unsigned long	runnable_sum;
529 	} removed;
530 
531 #ifdef CONFIG_FAIR_GROUP_SCHED
532 	unsigned long		tg_load_avg_contrib;
533 	long			propagate;
534 	long			prop_runnable_sum;
535 
536 	/*
537 	 *   h_load = weight * f(tg)
538 	 *
539 	 * Where f(tg) is the recursive weight fraction assigned to
540 	 * this group.
541 	 */
542 	unsigned long		h_load;
543 	u64			last_h_load_update;
544 	struct sched_entity	*h_load_next;
545 #endif /* CONFIG_FAIR_GROUP_SCHED */
546 #endif /* CONFIG_SMP */
547 
548 #ifdef CONFIG_FAIR_GROUP_SCHED
549 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
550 
551 	/*
552 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
553 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
554 	 * (like users, containers etc.)
555 	 *
556 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
557 	 * This list is used during load balance.
558 	 */
559 	int			on_list;
560 	struct list_head	leaf_cfs_rq_list;
561 	struct task_group	*tg;	/* group that "owns" this runqueue */
562 
563 #ifdef CONFIG_CFS_BANDWIDTH
564 	int			runtime_enabled;
565 	int			expires_seq;
566 	u64			runtime_expires;
567 	s64			runtime_remaining;
568 
569 	u64			throttled_clock;
570 	u64			throttled_clock_task;
571 	u64			throttled_clock_task_time;
572 	int			throttled;
573 	int			throttle_count;
574 	struct list_head	throttled_list;
575 #endif /* CONFIG_CFS_BANDWIDTH */
576 #endif /* CONFIG_FAIR_GROUP_SCHED */
577 };
578 
579 static inline int rt_bandwidth_enabled(void)
580 {
581 	return sysctl_sched_rt_runtime >= 0;
582 }
583 
584 /* RT IPI pull logic requires IRQ_WORK */
585 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
586 # define HAVE_RT_PUSH_IPI
587 #endif
588 
589 /* Real-Time classes' related field in a runqueue: */
590 struct rt_rq {
591 	struct rt_prio_array	active;
592 	unsigned int		rt_nr_running;
593 	unsigned int		rr_nr_running;
594 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
595 	struct {
596 		int		curr; /* highest queued rt task prio */
597 #ifdef CONFIG_SMP
598 		int		next; /* next highest */
599 #endif
600 	} highest_prio;
601 #endif
602 #ifdef CONFIG_SMP
603 	unsigned long		rt_nr_migratory;
604 	unsigned long		rt_nr_total;
605 	int			overloaded;
606 	struct plist_head	pushable_tasks;
607 
608 #endif /* CONFIG_SMP */
609 	int			rt_queued;
610 
611 	int			rt_throttled;
612 	u64			rt_time;
613 	u64			rt_runtime;
614 	/* Nests inside the rq lock: */
615 	raw_spinlock_t		rt_runtime_lock;
616 
617 #ifdef CONFIG_RT_GROUP_SCHED
618 	unsigned long		rt_nr_boosted;
619 
620 	struct rq		*rq;
621 	struct task_group	*tg;
622 #endif
623 };
624 
625 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
626 {
627 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
628 }
629 
630 /* Deadline class' related fields in a runqueue */
631 struct dl_rq {
632 	/* runqueue is an rbtree, ordered by deadline */
633 	struct rb_root_cached	root;
634 
635 	unsigned long		dl_nr_running;
636 
637 #ifdef CONFIG_SMP
638 	/*
639 	 * Deadline values of the currently executing and the
640 	 * earliest ready task on this rq. Caching these facilitates
641 	 * the decision whether or not a ready but not running task
642 	 * should migrate somewhere else.
643 	 */
644 	struct {
645 		u64		curr;
646 		u64		next;
647 	} earliest_dl;
648 
649 	unsigned long		dl_nr_migratory;
650 	int			overloaded;
651 
652 	/*
653 	 * Tasks on this rq that can be pushed away. They are kept in
654 	 * an rb-tree, ordered by tasks' deadlines, with caching
655 	 * of the leftmost (earliest deadline) element.
656 	 */
657 	struct rb_root_cached	pushable_dl_tasks_root;
658 #else
659 	struct dl_bw		dl_bw;
660 #endif
661 	/*
662 	 * "Active utilization" for this runqueue: increased when a
663 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
664 	 * task blocks
665 	 */
666 	u64			running_bw;
667 
668 	/*
669 	 * Utilization of the tasks "assigned" to this runqueue (including
670 	 * the tasks that are in runqueue and the tasks that executed on this
671 	 * CPU and blocked). Increased when a task moves to this runqueue, and
672 	 * decreased when the task moves away (migrates, changes scheduling
673 	 * policy, or terminates).
674 	 * This is needed to compute the "inactive utilization" for the
675 	 * runqueue (inactive utilization = this_bw - running_bw).
676 	 */
677 	u64			this_bw;
678 	u64			extra_bw;
679 
680 	/*
681 	 * Inverse of the fraction of CPU utilization that can be reclaimed
682 	 * by the GRUB algorithm.
683 	 */
684 	u64			bw_ratio;
685 };
686 
687 #ifdef CONFIG_FAIR_GROUP_SCHED
688 /* An entity is a task if it doesn't "own" a runqueue */
689 #define entity_is_task(se)	(!se->my_q)
690 #else
691 #define entity_is_task(se)	1
692 #endif
693 
694 #ifdef CONFIG_SMP
695 /*
696  * XXX we want to get rid of these helpers and use the full load resolution.
697  */
698 static inline long se_weight(struct sched_entity *se)
699 {
700 	return scale_load_down(se->load.weight);
701 }
702 
703 static inline long se_runnable(struct sched_entity *se)
704 {
705 	return scale_load_down(se->runnable_weight);
706 }
707 
708 static inline bool sched_asym_prefer(int a, int b)
709 {
710 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
711 }
712 
713 struct perf_domain {
714 	struct em_perf_domain *em_pd;
715 	struct perf_domain *next;
716 	struct rcu_head rcu;
717 };
718 
719 /* Scheduling group status flags */
720 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
721 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
722 
723 /*
724  * We add the notion of a root-domain which will be used to define per-domain
725  * variables. Each exclusive cpuset essentially defines an island domain by
726  * fully partitioning the member CPUs from any other cpuset. Whenever a new
727  * exclusive cpuset is created, we also create and attach a new root-domain
728  * object.
729  *
730  */
731 struct root_domain {
732 	atomic_t		refcount;
733 	atomic_t		rto_count;
734 	struct rcu_head		rcu;
735 	cpumask_var_t		span;
736 	cpumask_var_t		online;
737 
738 	/*
739 	 * Indicate pullable load on at least one CPU, e.g:
740 	 * - More than one runnable task
741 	 * - Running task is misfit
742 	 */
743 	int			overload;
744 
745 	/* Indicate one or more cpus over-utilized (tipping point) */
746 	int			overutilized;
747 
748 	/*
749 	 * The bit corresponding to a CPU gets set here if such CPU has more
750 	 * than one runnable -deadline task (as it is below for RT tasks).
751 	 */
752 	cpumask_var_t		dlo_mask;
753 	atomic_t		dlo_count;
754 	struct dl_bw		dl_bw;
755 	struct cpudl		cpudl;
756 
757 #ifdef HAVE_RT_PUSH_IPI
758 	/*
759 	 * For IPI pull requests, loop across the rto_mask.
760 	 */
761 	struct irq_work		rto_push_work;
762 	raw_spinlock_t		rto_lock;
763 	/* These are only updated and read within rto_lock */
764 	int			rto_loop;
765 	int			rto_cpu;
766 	/* These atomics are updated outside of a lock */
767 	atomic_t		rto_loop_next;
768 	atomic_t		rto_loop_start;
769 #endif
770 	/*
771 	 * The "RT overload" flag: it gets set if a CPU has more than
772 	 * one runnable RT task.
773 	 */
774 	cpumask_var_t		rto_mask;
775 	struct cpupri		cpupri;
776 
777 	unsigned long		max_cpu_capacity;
778 
779 	/*
780 	 * NULL-terminated list of performance domains intersecting with the
781 	 * CPUs of the rd. Protected by RCU.
782 	 */
783 	struct perf_domain	*pd;
784 };
785 
786 extern struct root_domain def_root_domain;
787 extern struct mutex sched_domains_mutex;
788 
789 extern void init_defrootdomain(void);
790 extern int sched_init_domains(const struct cpumask *cpu_map);
791 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
792 extern void sched_get_rd(struct root_domain *rd);
793 extern void sched_put_rd(struct root_domain *rd);
794 
795 #ifdef HAVE_RT_PUSH_IPI
796 extern void rto_push_irq_work_func(struct irq_work *work);
797 #endif
798 #endif /* CONFIG_SMP */
799 
800 /*
801  * This is the main, per-CPU runqueue data structure.
802  *
803  * Locking rule: those places that want to lock multiple runqueues
804  * (such as the load balancing or the thread migration code), lock
805  * acquire operations must be ordered by ascending &runqueue.
806  */
807 struct rq {
808 	/* runqueue lock: */
809 	raw_spinlock_t		lock;
810 
811 	/*
812 	 * nr_running and cpu_load should be in the same cacheline because
813 	 * remote CPUs use both these fields when doing load calculation.
814 	 */
815 	unsigned int		nr_running;
816 #ifdef CONFIG_NUMA_BALANCING
817 	unsigned int		nr_numa_running;
818 	unsigned int		nr_preferred_running;
819 	unsigned int		numa_migrate_on;
820 #endif
821 	#define CPU_LOAD_IDX_MAX 5
822 	unsigned long		cpu_load[CPU_LOAD_IDX_MAX];
823 #ifdef CONFIG_NO_HZ_COMMON
824 #ifdef CONFIG_SMP
825 	unsigned long		last_load_update_tick;
826 	unsigned long		last_blocked_load_update_tick;
827 	unsigned int		has_blocked_load;
828 #endif /* CONFIG_SMP */
829 	unsigned int		nohz_tick_stopped;
830 	atomic_t nohz_flags;
831 #endif /* CONFIG_NO_HZ_COMMON */
832 
833 	/* capture load from *all* tasks on this CPU: */
834 	struct load_weight	load;
835 	unsigned long		nr_load_updates;
836 	u64			nr_switches;
837 
838 	struct cfs_rq		cfs;
839 	struct rt_rq		rt;
840 	struct dl_rq		dl;
841 
842 #ifdef CONFIG_FAIR_GROUP_SCHED
843 	/* list of leaf cfs_rq on this CPU: */
844 	struct list_head	leaf_cfs_rq_list;
845 	struct list_head	*tmp_alone_branch;
846 #endif /* CONFIG_FAIR_GROUP_SCHED */
847 
848 	/*
849 	 * This is part of a global counter where only the total sum
850 	 * over all CPUs matters. A task can increase this counter on
851 	 * one CPU and if it got migrated afterwards it may decrease
852 	 * it on another CPU. Always updated under the runqueue lock:
853 	 */
854 	unsigned long		nr_uninterruptible;
855 
856 	struct task_struct	*curr;
857 	struct task_struct	*idle;
858 	struct task_struct	*stop;
859 	unsigned long		next_balance;
860 	struct mm_struct	*prev_mm;
861 
862 	unsigned int		clock_update_flags;
863 	u64			clock;
864 	u64			clock_task;
865 
866 	atomic_t		nr_iowait;
867 
868 #ifdef CONFIG_SMP
869 	struct root_domain	*rd;
870 	struct sched_domain	*sd;
871 
872 	unsigned long		cpu_capacity;
873 	unsigned long		cpu_capacity_orig;
874 
875 	struct callback_head	*balance_callback;
876 
877 	unsigned char		idle_balance;
878 
879 	unsigned long		misfit_task_load;
880 
881 	/* For active balancing */
882 	int			active_balance;
883 	int			push_cpu;
884 	struct cpu_stop_work	active_balance_work;
885 
886 	/* CPU of this runqueue: */
887 	int			cpu;
888 	int			online;
889 
890 	struct list_head cfs_tasks;
891 
892 	struct sched_avg	avg_rt;
893 	struct sched_avg	avg_dl;
894 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
895 	struct sched_avg	avg_irq;
896 #endif
897 	u64			idle_stamp;
898 	u64			avg_idle;
899 
900 	/* This is used to determine avg_idle's max value */
901 	u64			max_idle_balance_cost;
902 #endif
903 
904 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
905 	u64			prev_irq_time;
906 #endif
907 #ifdef CONFIG_PARAVIRT
908 	u64			prev_steal_time;
909 #endif
910 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
911 	u64			prev_steal_time_rq;
912 #endif
913 
914 	/* calc_load related fields */
915 	unsigned long		calc_load_update;
916 	long			calc_load_active;
917 
918 #ifdef CONFIG_SCHED_HRTICK
919 #ifdef CONFIG_SMP
920 	int			hrtick_csd_pending;
921 	call_single_data_t	hrtick_csd;
922 #endif
923 	struct hrtimer		hrtick_timer;
924 #endif
925 
926 #ifdef CONFIG_SCHEDSTATS
927 	/* latency stats */
928 	struct sched_info	rq_sched_info;
929 	unsigned long long	rq_cpu_time;
930 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
931 
932 	/* sys_sched_yield() stats */
933 	unsigned int		yld_count;
934 
935 	/* schedule() stats */
936 	unsigned int		sched_count;
937 	unsigned int		sched_goidle;
938 
939 	/* try_to_wake_up() stats */
940 	unsigned int		ttwu_count;
941 	unsigned int		ttwu_local;
942 #endif
943 
944 #ifdef CONFIG_SMP
945 	struct llist_head	wake_list;
946 #endif
947 
948 #ifdef CONFIG_CPU_IDLE
949 	/* Must be inspected within a rcu lock section */
950 	struct cpuidle_state	*idle_state;
951 #endif
952 };
953 
954 static inline int cpu_of(struct rq *rq)
955 {
956 #ifdef CONFIG_SMP
957 	return rq->cpu;
958 #else
959 	return 0;
960 #endif
961 }
962 
963 
964 #ifdef CONFIG_SCHED_SMT
965 extern void __update_idle_core(struct rq *rq);
966 
967 static inline void update_idle_core(struct rq *rq)
968 {
969 	if (static_branch_unlikely(&sched_smt_present))
970 		__update_idle_core(rq);
971 }
972 
973 #else
974 static inline void update_idle_core(struct rq *rq) { }
975 #endif
976 
977 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
978 
979 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
980 #define this_rq()		this_cpu_ptr(&runqueues)
981 #define task_rq(p)		cpu_rq(task_cpu(p))
982 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
983 #define raw_rq()		raw_cpu_ptr(&runqueues)
984 
985 extern void update_rq_clock(struct rq *rq);
986 
987 static inline u64 __rq_clock_broken(struct rq *rq)
988 {
989 	return READ_ONCE(rq->clock);
990 }
991 
992 /*
993  * rq::clock_update_flags bits
994  *
995  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
996  *  call to __schedule(). This is an optimisation to avoid
997  *  neighbouring rq clock updates.
998  *
999  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1000  *  in effect and calls to update_rq_clock() are being ignored.
1001  *
1002  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1003  *  made to update_rq_clock() since the last time rq::lock was pinned.
1004  *
1005  * If inside of __schedule(), clock_update_flags will have been
1006  * shifted left (a left shift is a cheap operation for the fast path
1007  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1008  *
1009  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1010  *
1011  * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1012  * one position though, because the next rq_unpin_lock() will shift it
1013  * back.
1014  */
1015 #define RQCF_REQ_SKIP		0x01
1016 #define RQCF_ACT_SKIP		0x02
1017 #define RQCF_UPDATED		0x04
1018 
1019 static inline void assert_clock_updated(struct rq *rq)
1020 {
1021 	/*
1022 	 * The only reason for not seeing a clock update since the
1023 	 * last rq_pin_lock() is if we're currently skipping updates.
1024 	 */
1025 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1026 }
1027 
1028 static inline u64 rq_clock(struct rq *rq)
1029 {
1030 	lockdep_assert_held(&rq->lock);
1031 	assert_clock_updated(rq);
1032 
1033 	return rq->clock;
1034 }
1035 
1036 static inline u64 rq_clock_task(struct rq *rq)
1037 {
1038 	lockdep_assert_held(&rq->lock);
1039 	assert_clock_updated(rq);
1040 
1041 	return rq->clock_task;
1042 }
1043 
1044 static inline void rq_clock_skip_update(struct rq *rq)
1045 {
1046 	lockdep_assert_held(&rq->lock);
1047 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1048 }
1049 
1050 /*
1051  * See rt task throttling, which is the only time a skip
1052  * request is cancelled.
1053  */
1054 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1055 {
1056 	lockdep_assert_held(&rq->lock);
1057 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1058 }
1059 
1060 struct rq_flags {
1061 	unsigned long flags;
1062 	struct pin_cookie cookie;
1063 #ifdef CONFIG_SCHED_DEBUG
1064 	/*
1065 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1066 	 * current pin context is stashed here in case it needs to be
1067 	 * restored in rq_repin_lock().
1068 	 */
1069 	unsigned int clock_update_flags;
1070 #endif
1071 };
1072 
1073 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1074 {
1075 	rf->cookie = lockdep_pin_lock(&rq->lock);
1076 
1077 #ifdef CONFIG_SCHED_DEBUG
1078 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1079 	rf->clock_update_flags = 0;
1080 #endif
1081 }
1082 
1083 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1084 {
1085 #ifdef CONFIG_SCHED_DEBUG
1086 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1087 		rf->clock_update_flags = RQCF_UPDATED;
1088 #endif
1089 
1090 	lockdep_unpin_lock(&rq->lock, rf->cookie);
1091 }
1092 
1093 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1094 {
1095 	lockdep_repin_lock(&rq->lock, rf->cookie);
1096 
1097 #ifdef CONFIG_SCHED_DEBUG
1098 	/*
1099 	 * Restore the value we stashed in @rf for this pin context.
1100 	 */
1101 	rq->clock_update_flags |= rf->clock_update_flags;
1102 #endif
1103 }
1104 
1105 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1106 	__acquires(rq->lock);
1107 
1108 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1109 	__acquires(p->pi_lock)
1110 	__acquires(rq->lock);
1111 
1112 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1113 	__releases(rq->lock)
1114 {
1115 	rq_unpin_lock(rq, rf);
1116 	raw_spin_unlock(&rq->lock);
1117 }
1118 
1119 static inline void
1120 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1121 	__releases(rq->lock)
1122 	__releases(p->pi_lock)
1123 {
1124 	rq_unpin_lock(rq, rf);
1125 	raw_spin_unlock(&rq->lock);
1126 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1127 }
1128 
1129 static inline void
1130 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1131 	__acquires(rq->lock)
1132 {
1133 	raw_spin_lock_irqsave(&rq->lock, rf->flags);
1134 	rq_pin_lock(rq, rf);
1135 }
1136 
1137 static inline void
1138 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1139 	__acquires(rq->lock)
1140 {
1141 	raw_spin_lock_irq(&rq->lock);
1142 	rq_pin_lock(rq, rf);
1143 }
1144 
1145 static inline void
1146 rq_lock(struct rq *rq, struct rq_flags *rf)
1147 	__acquires(rq->lock)
1148 {
1149 	raw_spin_lock(&rq->lock);
1150 	rq_pin_lock(rq, rf);
1151 }
1152 
1153 static inline void
1154 rq_relock(struct rq *rq, struct rq_flags *rf)
1155 	__acquires(rq->lock)
1156 {
1157 	raw_spin_lock(&rq->lock);
1158 	rq_repin_lock(rq, rf);
1159 }
1160 
1161 static inline void
1162 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1163 	__releases(rq->lock)
1164 {
1165 	rq_unpin_lock(rq, rf);
1166 	raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1167 }
1168 
1169 static inline void
1170 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1171 	__releases(rq->lock)
1172 {
1173 	rq_unpin_lock(rq, rf);
1174 	raw_spin_unlock_irq(&rq->lock);
1175 }
1176 
1177 static inline void
1178 rq_unlock(struct rq *rq, struct rq_flags *rf)
1179 	__releases(rq->lock)
1180 {
1181 	rq_unpin_lock(rq, rf);
1182 	raw_spin_unlock(&rq->lock);
1183 }
1184 
1185 static inline struct rq *
1186 this_rq_lock_irq(struct rq_flags *rf)
1187 	__acquires(rq->lock)
1188 {
1189 	struct rq *rq;
1190 
1191 	local_irq_disable();
1192 	rq = this_rq();
1193 	rq_lock(rq, rf);
1194 	return rq;
1195 }
1196 
1197 #ifdef CONFIG_NUMA
1198 enum numa_topology_type {
1199 	NUMA_DIRECT,
1200 	NUMA_GLUELESS_MESH,
1201 	NUMA_BACKPLANE,
1202 };
1203 extern enum numa_topology_type sched_numa_topology_type;
1204 extern int sched_max_numa_distance;
1205 extern bool find_numa_distance(int distance);
1206 #endif
1207 
1208 #ifdef CONFIG_NUMA
1209 extern void sched_init_numa(void);
1210 extern void sched_domains_numa_masks_set(unsigned int cpu);
1211 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1212 #else
1213 static inline void sched_init_numa(void) { }
1214 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1215 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1216 #endif
1217 
1218 #ifdef CONFIG_NUMA_BALANCING
1219 /* The regions in numa_faults array from task_struct */
1220 enum numa_faults_stats {
1221 	NUMA_MEM = 0,
1222 	NUMA_CPU,
1223 	NUMA_MEMBUF,
1224 	NUMA_CPUBUF
1225 };
1226 extern void sched_setnuma(struct task_struct *p, int node);
1227 extern int migrate_task_to(struct task_struct *p, int cpu);
1228 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1229 			int cpu, int scpu);
1230 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1231 #else
1232 static inline void
1233 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1234 {
1235 }
1236 #endif /* CONFIG_NUMA_BALANCING */
1237 
1238 #ifdef CONFIG_SMP
1239 
1240 static inline void
1241 queue_balance_callback(struct rq *rq,
1242 		       struct callback_head *head,
1243 		       void (*func)(struct rq *rq))
1244 {
1245 	lockdep_assert_held(&rq->lock);
1246 
1247 	if (unlikely(head->next))
1248 		return;
1249 
1250 	head->func = (void (*)(struct callback_head *))func;
1251 	head->next = rq->balance_callback;
1252 	rq->balance_callback = head;
1253 }
1254 
1255 extern void sched_ttwu_pending(void);
1256 
1257 #define rcu_dereference_check_sched_domain(p) \
1258 	rcu_dereference_check((p), \
1259 			      lockdep_is_held(&sched_domains_mutex))
1260 
1261 /*
1262  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1263  * See detach_destroy_domains: synchronize_sched for details.
1264  *
1265  * The domain tree of any CPU may only be accessed from within
1266  * preempt-disabled sections.
1267  */
1268 #define for_each_domain(cpu, __sd) \
1269 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1270 			__sd; __sd = __sd->parent)
1271 
1272 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1273 
1274 /**
1275  * highest_flag_domain - Return highest sched_domain containing flag.
1276  * @cpu:	The CPU whose highest level of sched domain is to
1277  *		be returned.
1278  * @flag:	The flag to check for the highest sched_domain
1279  *		for the given CPU.
1280  *
1281  * Returns the highest sched_domain of a CPU which contains the given flag.
1282  */
1283 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1284 {
1285 	struct sched_domain *sd, *hsd = NULL;
1286 
1287 	for_each_domain(cpu, sd) {
1288 		if (!(sd->flags & flag))
1289 			break;
1290 		hsd = sd;
1291 	}
1292 
1293 	return hsd;
1294 }
1295 
1296 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1297 {
1298 	struct sched_domain *sd;
1299 
1300 	for_each_domain(cpu, sd) {
1301 		if (sd->flags & flag)
1302 			break;
1303 	}
1304 
1305 	return sd;
1306 }
1307 
1308 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
1309 DECLARE_PER_CPU(int, sd_llc_size);
1310 DECLARE_PER_CPU(int, sd_llc_id);
1311 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
1312 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
1313 DECLARE_PER_CPU(struct sched_domain *, sd_asym_packing);
1314 DECLARE_PER_CPU(struct sched_domain *, sd_asym_cpucapacity);
1315 extern struct static_key_false sched_asym_cpucapacity;
1316 
1317 struct sched_group_capacity {
1318 	atomic_t		ref;
1319 	/*
1320 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1321 	 * for a single CPU.
1322 	 */
1323 	unsigned long		capacity;
1324 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1325 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1326 	unsigned long		next_update;
1327 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1328 
1329 #ifdef CONFIG_SCHED_DEBUG
1330 	int			id;
1331 #endif
1332 
1333 	unsigned long		cpumask[0];		/* Balance mask */
1334 };
1335 
1336 struct sched_group {
1337 	struct sched_group	*next;			/* Must be a circular list */
1338 	atomic_t		ref;
1339 
1340 	unsigned int		group_weight;
1341 	struct sched_group_capacity *sgc;
1342 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1343 
1344 	/*
1345 	 * The CPUs this group covers.
1346 	 *
1347 	 * NOTE: this field is variable length. (Allocated dynamically
1348 	 * by attaching extra space to the end of the structure,
1349 	 * depending on how many CPUs the kernel has booted up with)
1350 	 */
1351 	unsigned long		cpumask[0];
1352 };
1353 
1354 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1355 {
1356 	return to_cpumask(sg->cpumask);
1357 }
1358 
1359 /*
1360  * See build_balance_mask().
1361  */
1362 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1363 {
1364 	return to_cpumask(sg->sgc->cpumask);
1365 }
1366 
1367 /**
1368  * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1369  * @group: The group whose first CPU is to be returned.
1370  */
1371 static inline unsigned int group_first_cpu(struct sched_group *group)
1372 {
1373 	return cpumask_first(sched_group_span(group));
1374 }
1375 
1376 extern int group_balance_cpu(struct sched_group *sg);
1377 
1378 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1379 void register_sched_domain_sysctl(void);
1380 void dirty_sched_domain_sysctl(int cpu);
1381 void unregister_sched_domain_sysctl(void);
1382 #else
1383 static inline void register_sched_domain_sysctl(void)
1384 {
1385 }
1386 static inline void dirty_sched_domain_sysctl(int cpu)
1387 {
1388 }
1389 static inline void unregister_sched_domain_sysctl(void)
1390 {
1391 }
1392 #endif
1393 
1394 #else
1395 
1396 static inline void sched_ttwu_pending(void) { }
1397 
1398 #endif /* CONFIG_SMP */
1399 
1400 #include "stats.h"
1401 #include "autogroup.h"
1402 
1403 #ifdef CONFIG_CGROUP_SCHED
1404 
1405 /*
1406  * Return the group to which this tasks belongs.
1407  *
1408  * We cannot use task_css() and friends because the cgroup subsystem
1409  * changes that value before the cgroup_subsys::attach() method is called,
1410  * therefore we cannot pin it and might observe the wrong value.
1411  *
1412  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1413  * core changes this before calling sched_move_task().
1414  *
1415  * Instead we use a 'copy' which is updated from sched_move_task() while
1416  * holding both task_struct::pi_lock and rq::lock.
1417  */
1418 static inline struct task_group *task_group(struct task_struct *p)
1419 {
1420 	return p->sched_task_group;
1421 }
1422 
1423 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1424 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1425 {
1426 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1427 	struct task_group *tg = task_group(p);
1428 #endif
1429 
1430 #ifdef CONFIG_FAIR_GROUP_SCHED
1431 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1432 	p->se.cfs_rq = tg->cfs_rq[cpu];
1433 	p->se.parent = tg->se[cpu];
1434 #endif
1435 
1436 #ifdef CONFIG_RT_GROUP_SCHED
1437 	p->rt.rt_rq  = tg->rt_rq[cpu];
1438 	p->rt.parent = tg->rt_se[cpu];
1439 #endif
1440 }
1441 
1442 #else /* CONFIG_CGROUP_SCHED */
1443 
1444 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1445 static inline struct task_group *task_group(struct task_struct *p)
1446 {
1447 	return NULL;
1448 }
1449 
1450 #endif /* CONFIG_CGROUP_SCHED */
1451 
1452 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1453 {
1454 	set_task_rq(p, cpu);
1455 #ifdef CONFIG_SMP
1456 	/*
1457 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1458 	 * successfully executed on another CPU. We must ensure that updates of
1459 	 * per-task data have been completed by this moment.
1460 	 */
1461 	smp_wmb();
1462 #ifdef CONFIG_THREAD_INFO_IN_TASK
1463 	p->cpu = cpu;
1464 #else
1465 	task_thread_info(p)->cpu = cpu;
1466 #endif
1467 	p->wake_cpu = cpu;
1468 #endif
1469 }
1470 
1471 /*
1472  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1473  */
1474 #ifdef CONFIG_SCHED_DEBUG
1475 # include <linux/static_key.h>
1476 # define const_debug __read_mostly
1477 #else
1478 # define const_debug const
1479 #endif
1480 
1481 #define SCHED_FEAT(name, enabled)	\
1482 	__SCHED_FEAT_##name ,
1483 
1484 enum {
1485 #include "features.h"
1486 	__SCHED_FEAT_NR,
1487 };
1488 
1489 #undef SCHED_FEAT
1490 
1491 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1492 
1493 /*
1494  * To support run-time toggling of sched features, all the translation units
1495  * (but core.c) reference the sysctl_sched_features defined in core.c.
1496  */
1497 extern const_debug unsigned int sysctl_sched_features;
1498 
1499 #define SCHED_FEAT(name, enabled)					\
1500 static __always_inline bool static_branch_##name(struct static_key *key) \
1501 {									\
1502 	return static_key_##enabled(key);				\
1503 }
1504 
1505 #include "features.h"
1506 #undef SCHED_FEAT
1507 
1508 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1509 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1510 
1511 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1512 
1513 /*
1514  * Each translation unit has its own copy of sysctl_sched_features to allow
1515  * constants propagation at compile time and compiler optimization based on
1516  * features default.
1517  */
1518 #define SCHED_FEAT(name, enabled)	\
1519 	(1UL << __SCHED_FEAT_##name) * enabled |
1520 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1521 #include "features.h"
1522 	0;
1523 #undef SCHED_FEAT
1524 
1525 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1526 
1527 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1528 
1529 extern struct static_key_false sched_numa_balancing;
1530 extern struct static_key_false sched_schedstats;
1531 
1532 static inline u64 global_rt_period(void)
1533 {
1534 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1535 }
1536 
1537 static inline u64 global_rt_runtime(void)
1538 {
1539 	if (sysctl_sched_rt_runtime < 0)
1540 		return RUNTIME_INF;
1541 
1542 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1543 }
1544 
1545 static inline int task_current(struct rq *rq, struct task_struct *p)
1546 {
1547 	return rq->curr == p;
1548 }
1549 
1550 static inline int task_running(struct rq *rq, struct task_struct *p)
1551 {
1552 #ifdef CONFIG_SMP
1553 	return p->on_cpu;
1554 #else
1555 	return task_current(rq, p);
1556 #endif
1557 }
1558 
1559 static inline int task_on_rq_queued(struct task_struct *p)
1560 {
1561 	return p->on_rq == TASK_ON_RQ_QUEUED;
1562 }
1563 
1564 static inline int task_on_rq_migrating(struct task_struct *p)
1565 {
1566 	return p->on_rq == TASK_ON_RQ_MIGRATING;
1567 }
1568 
1569 /*
1570  * wake flags
1571  */
1572 #define WF_SYNC			0x01		/* Waker goes to sleep after wakeup */
1573 #define WF_FORK			0x02		/* Child wakeup after fork */
1574 #define WF_MIGRATED		0x4		/* Internal use, task got migrated */
1575 
1576 /*
1577  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1578  * of tasks with abnormal "nice" values across CPUs the contribution that
1579  * each task makes to its run queue's load is weighted according to its
1580  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1581  * scaled version of the new time slice allocation that they receive on time
1582  * slice expiry etc.
1583  */
1584 
1585 #define WEIGHT_IDLEPRIO		3
1586 #define WMULT_IDLEPRIO		1431655765
1587 
1588 extern const int		sched_prio_to_weight[40];
1589 extern const u32		sched_prio_to_wmult[40];
1590 
1591 /*
1592  * {de,en}queue flags:
1593  *
1594  * DEQUEUE_SLEEP  - task is no longer runnable
1595  * ENQUEUE_WAKEUP - task just became runnable
1596  *
1597  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1598  *                are in a known state which allows modification. Such pairs
1599  *                should preserve as much state as possible.
1600  *
1601  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1602  *        in the runqueue.
1603  *
1604  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
1605  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1606  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
1607  *
1608  */
1609 
1610 #define DEQUEUE_SLEEP		0x01
1611 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
1612 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
1613 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
1614 
1615 #define ENQUEUE_WAKEUP		0x01
1616 #define ENQUEUE_RESTORE		0x02
1617 #define ENQUEUE_MOVE		0x04
1618 #define ENQUEUE_NOCLOCK		0x08
1619 
1620 #define ENQUEUE_HEAD		0x10
1621 #define ENQUEUE_REPLENISH	0x20
1622 #ifdef CONFIG_SMP
1623 #define ENQUEUE_MIGRATED	0x40
1624 #else
1625 #define ENQUEUE_MIGRATED	0x00
1626 #endif
1627 
1628 #define RETRY_TASK		((void *)-1UL)
1629 
1630 struct sched_class {
1631 	const struct sched_class *next;
1632 
1633 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1634 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1635 	void (*yield_task)   (struct rq *rq);
1636 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1637 
1638 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1639 
1640 	/*
1641 	 * It is the responsibility of the pick_next_task() method that will
1642 	 * return the next task to call put_prev_task() on the @prev task or
1643 	 * something equivalent.
1644 	 *
1645 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1646 	 * tasks.
1647 	 */
1648 	struct task_struct * (*pick_next_task)(struct rq *rq,
1649 					       struct task_struct *prev,
1650 					       struct rq_flags *rf);
1651 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1652 
1653 #ifdef CONFIG_SMP
1654 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1655 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1656 
1657 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1658 
1659 	void (*set_cpus_allowed)(struct task_struct *p,
1660 				 const struct cpumask *newmask);
1661 
1662 	void (*rq_online)(struct rq *rq);
1663 	void (*rq_offline)(struct rq *rq);
1664 #endif
1665 
1666 	void (*set_curr_task)(struct rq *rq);
1667 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1668 	void (*task_fork)(struct task_struct *p);
1669 	void (*task_dead)(struct task_struct *p);
1670 
1671 	/*
1672 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1673 	 * cannot assume the switched_from/switched_to pair is serliazed by
1674 	 * rq->lock. They are however serialized by p->pi_lock.
1675 	 */
1676 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1677 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
1678 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1679 			      int oldprio);
1680 
1681 	unsigned int (*get_rr_interval)(struct rq *rq,
1682 					struct task_struct *task);
1683 
1684 	void (*update_curr)(struct rq *rq);
1685 
1686 #define TASK_SET_GROUP		0
1687 #define TASK_MOVE_GROUP		1
1688 
1689 #ifdef CONFIG_FAIR_GROUP_SCHED
1690 	void (*task_change_group)(struct task_struct *p, int type);
1691 #endif
1692 };
1693 
1694 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1695 {
1696 	prev->sched_class->put_prev_task(rq, prev);
1697 }
1698 
1699 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1700 {
1701 	curr->sched_class->set_curr_task(rq);
1702 }
1703 
1704 #ifdef CONFIG_SMP
1705 #define sched_class_highest (&stop_sched_class)
1706 #else
1707 #define sched_class_highest (&dl_sched_class)
1708 #endif
1709 #define for_each_class(class) \
1710    for (class = sched_class_highest; class; class = class->next)
1711 
1712 extern const struct sched_class stop_sched_class;
1713 extern const struct sched_class dl_sched_class;
1714 extern const struct sched_class rt_sched_class;
1715 extern const struct sched_class fair_sched_class;
1716 extern const struct sched_class idle_sched_class;
1717 
1718 
1719 #ifdef CONFIG_SMP
1720 
1721 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1722 
1723 extern void trigger_load_balance(struct rq *rq);
1724 
1725 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1726 
1727 #endif
1728 
1729 #ifdef CONFIG_CPU_IDLE
1730 static inline void idle_set_state(struct rq *rq,
1731 				  struct cpuidle_state *idle_state)
1732 {
1733 	rq->idle_state = idle_state;
1734 }
1735 
1736 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1737 {
1738 	SCHED_WARN_ON(!rcu_read_lock_held());
1739 
1740 	return rq->idle_state;
1741 }
1742 #else
1743 static inline void idle_set_state(struct rq *rq,
1744 				  struct cpuidle_state *idle_state)
1745 {
1746 }
1747 
1748 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1749 {
1750 	return NULL;
1751 }
1752 #endif
1753 
1754 extern void schedule_idle(void);
1755 
1756 extern void sysrq_sched_debug_show(void);
1757 extern void sched_init_granularity(void);
1758 extern void update_max_interval(void);
1759 
1760 extern void init_sched_dl_class(void);
1761 extern void init_sched_rt_class(void);
1762 extern void init_sched_fair_class(void);
1763 
1764 extern void reweight_task(struct task_struct *p, int prio);
1765 
1766 extern void resched_curr(struct rq *rq);
1767 extern void resched_cpu(int cpu);
1768 
1769 extern struct rt_bandwidth def_rt_bandwidth;
1770 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1771 
1772 extern struct dl_bandwidth def_dl_bandwidth;
1773 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1774 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1775 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1776 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1777 
1778 #define BW_SHIFT		20
1779 #define BW_UNIT			(1 << BW_SHIFT)
1780 #define RATIO_SHIFT		8
1781 unsigned long to_ratio(u64 period, u64 runtime);
1782 
1783 extern void init_entity_runnable_average(struct sched_entity *se);
1784 extern void post_init_entity_util_avg(struct sched_entity *se);
1785 
1786 #ifdef CONFIG_NO_HZ_FULL
1787 extern bool sched_can_stop_tick(struct rq *rq);
1788 extern int __init sched_tick_offload_init(void);
1789 
1790 /*
1791  * Tick may be needed by tasks in the runqueue depending on their policy and
1792  * requirements. If tick is needed, lets send the target an IPI to kick it out of
1793  * nohz mode if necessary.
1794  */
1795 static inline void sched_update_tick_dependency(struct rq *rq)
1796 {
1797 	int cpu;
1798 
1799 	if (!tick_nohz_full_enabled())
1800 		return;
1801 
1802 	cpu = cpu_of(rq);
1803 
1804 	if (!tick_nohz_full_cpu(cpu))
1805 		return;
1806 
1807 	if (sched_can_stop_tick(rq))
1808 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1809 	else
1810 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1811 }
1812 #else
1813 static inline int sched_tick_offload_init(void) { return 0; }
1814 static inline void sched_update_tick_dependency(struct rq *rq) { }
1815 #endif
1816 
1817 static inline void add_nr_running(struct rq *rq, unsigned count)
1818 {
1819 	unsigned prev_nr = rq->nr_running;
1820 
1821 	rq->nr_running = prev_nr + count;
1822 
1823 #ifdef CONFIG_SMP
1824 	if (prev_nr < 2 && rq->nr_running >= 2) {
1825 		if (!READ_ONCE(rq->rd->overload))
1826 			WRITE_ONCE(rq->rd->overload, 1);
1827 	}
1828 #endif
1829 
1830 	sched_update_tick_dependency(rq);
1831 }
1832 
1833 static inline void sub_nr_running(struct rq *rq, unsigned count)
1834 {
1835 	rq->nr_running -= count;
1836 	/* Check if we still need preemption */
1837 	sched_update_tick_dependency(rq);
1838 }
1839 
1840 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1841 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1842 
1843 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1844 
1845 extern const_debug unsigned int sysctl_sched_nr_migrate;
1846 extern const_debug unsigned int sysctl_sched_migration_cost;
1847 
1848 #ifdef CONFIG_SCHED_HRTICK
1849 
1850 /*
1851  * Use hrtick when:
1852  *  - enabled by features
1853  *  - hrtimer is actually high res
1854  */
1855 static inline int hrtick_enabled(struct rq *rq)
1856 {
1857 	if (!sched_feat(HRTICK))
1858 		return 0;
1859 	if (!cpu_active(cpu_of(rq)))
1860 		return 0;
1861 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1862 }
1863 
1864 void hrtick_start(struct rq *rq, u64 delay);
1865 
1866 #else
1867 
1868 static inline int hrtick_enabled(struct rq *rq)
1869 {
1870 	return 0;
1871 }
1872 
1873 #endif /* CONFIG_SCHED_HRTICK */
1874 
1875 #ifndef arch_scale_freq_capacity
1876 static __always_inline
1877 unsigned long arch_scale_freq_capacity(int cpu)
1878 {
1879 	return SCHED_CAPACITY_SCALE;
1880 }
1881 #endif
1882 
1883 #ifdef CONFIG_SMP
1884 #ifdef CONFIG_PREEMPT
1885 
1886 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1887 
1888 /*
1889  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1890  * way at the expense of forcing extra atomic operations in all
1891  * invocations.  This assures that the double_lock is acquired using the
1892  * same underlying policy as the spinlock_t on this architecture, which
1893  * reduces latency compared to the unfair variant below.  However, it
1894  * also adds more overhead and therefore may reduce throughput.
1895  */
1896 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1897 	__releases(this_rq->lock)
1898 	__acquires(busiest->lock)
1899 	__acquires(this_rq->lock)
1900 {
1901 	raw_spin_unlock(&this_rq->lock);
1902 	double_rq_lock(this_rq, busiest);
1903 
1904 	return 1;
1905 }
1906 
1907 #else
1908 /*
1909  * Unfair double_lock_balance: Optimizes throughput at the expense of
1910  * latency by eliminating extra atomic operations when the locks are
1911  * already in proper order on entry.  This favors lower CPU-ids and will
1912  * grant the double lock to lower CPUs over higher ids under contention,
1913  * regardless of entry order into the function.
1914  */
1915 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1916 	__releases(this_rq->lock)
1917 	__acquires(busiest->lock)
1918 	__acquires(this_rq->lock)
1919 {
1920 	int ret = 0;
1921 
1922 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1923 		if (busiest < this_rq) {
1924 			raw_spin_unlock(&this_rq->lock);
1925 			raw_spin_lock(&busiest->lock);
1926 			raw_spin_lock_nested(&this_rq->lock,
1927 					      SINGLE_DEPTH_NESTING);
1928 			ret = 1;
1929 		} else
1930 			raw_spin_lock_nested(&busiest->lock,
1931 					      SINGLE_DEPTH_NESTING);
1932 	}
1933 	return ret;
1934 }
1935 
1936 #endif /* CONFIG_PREEMPT */
1937 
1938 /*
1939  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1940  */
1941 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1942 {
1943 	if (unlikely(!irqs_disabled())) {
1944 		/* printk() doesn't work well under rq->lock */
1945 		raw_spin_unlock(&this_rq->lock);
1946 		BUG_ON(1);
1947 	}
1948 
1949 	return _double_lock_balance(this_rq, busiest);
1950 }
1951 
1952 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1953 	__releases(busiest->lock)
1954 {
1955 	raw_spin_unlock(&busiest->lock);
1956 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1957 }
1958 
1959 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1960 {
1961 	if (l1 > l2)
1962 		swap(l1, l2);
1963 
1964 	spin_lock(l1);
1965 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1966 }
1967 
1968 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1969 {
1970 	if (l1 > l2)
1971 		swap(l1, l2);
1972 
1973 	spin_lock_irq(l1);
1974 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1975 }
1976 
1977 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1978 {
1979 	if (l1 > l2)
1980 		swap(l1, l2);
1981 
1982 	raw_spin_lock(l1);
1983 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1984 }
1985 
1986 /*
1987  * double_rq_lock - safely lock two runqueues
1988  *
1989  * Note this does not disable interrupts like task_rq_lock,
1990  * you need to do so manually before calling.
1991  */
1992 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1993 	__acquires(rq1->lock)
1994 	__acquires(rq2->lock)
1995 {
1996 	BUG_ON(!irqs_disabled());
1997 	if (rq1 == rq2) {
1998 		raw_spin_lock(&rq1->lock);
1999 		__acquire(rq2->lock);	/* Fake it out ;) */
2000 	} else {
2001 		if (rq1 < rq2) {
2002 			raw_spin_lock(&rq1->lock);
2003 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2004 		} else {
2005 			raw_spin_lock(&rq2->lock);
2006 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2007 		}
2008 	}
2009 }
2010 
2011 /*
2012  * double_rq_unlock - safely unlock two runqueues
2013  *
2014  * Note this does not restore interrupts like task_rq_unlock,
2015  * you need to do so manually after calling.
2016  */
2017 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2018 	__releases(rq1->lock)
2019 	__releases(rq2->lock)
2020 {
2021 	raw_spin_unlock(&rq1->lock);
2022 	if (rq1 != rq2)
2023 		raw_spin_unlock(&rq2->lock);
2024 	else
2025 		__release(rq2->lock);
2026 }
2027 
2028 extern void set_rq_online (struct rq *rq);
2029 extern void set_rq_offline(struct rq *rq);
2030 extern bool sched_smp_initialized;
2031 
2032 #else /* CONFIG_SMP */
2033 
2034 /*
2035  * double_rq_lock - safely lock two runqueues
2036  *
2037  * Note this does not disable interrupts like task_rq_lock,
2038  * you need to do so manually before calling.
2039  */
2040 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2041 	__acquires(rq1->lock)
2042 	__acquires(rq2->lock)
2043 {
2044 	BUG_ON(!irqs_disabled());
2045 	BUG_ON(rq1 != rq2);
2046 	raw_spin_lock(&rq1->lock);
2047 	__acquire(rq2->lock);	/* Fake it out ;) */
2048 }
2049 
2050 /*
2051  * double_rq_unlock - safely unlock two runqueues
2052  *
2053  * Note this does not restore interrupts like task_rq_unlock,
2054  * you need to do so manually after calling.
2055  */
2056 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2057 	__releases(rq1->lock)
2058 	__releases(rq2->lock)
2059 {
2060 	BUG_ON(rq1 != rq2);
2061 	raw_spin_unlock(&rq1->lock);
2062 	__release(rq2->lock);
2063 }
2064 
2065 #endif
2066 
2067 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2068 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2069 
2070 #ifdef	CONFIG_SCHED_DEBUG
2071 extern bool sched_debug_enabled;
2072 
2073 extern void print_cfs_stats(struct seq_file *m, int cpu);
2074 extern void print_rt_stats(struct seq_file *m, int cpu);
2075 extern void print_dl_stats(struct seq_file *m, int cpu);
2076 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2077 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2078 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2079 #ifdef CONFIG_NUMA_BALANCING
2080 extern void
2081 show_numa_stats(struct task_struct *p, struct seq_file *m);
2082 extern void
2083 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2084 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2085 #endif /* CONFIG_NUMA_BALANCING */
2086 #endif /* CONFIG_SCHED_DEBUG */
2087 
2088 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2089 extern void init_rt_rq(struct rt_rq *rt_rq);
2090 extern void init_dl_rq(struct dl_rq *dl_rq);
2091 
2092 extern void cfs_bandwidth_usage_inc(void);
2093 extern void cfs_bandwidth_usage_dec(void);
2094 
2095 #ifdef CONFIG_NO_HZ_COMMON
2096 #define NOHZ_BALANCE_KICK_BIT	0
2097 #define NOHZ_STATS_KICK_BIT	1
2098 
2099 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2100 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2101 
2102 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2103 
2104 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2105 
2106 extern void nohz_balance_exit_idle(struct rq *rq);
2107 #else
2108 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2109 #endif
2110 
2111 
2112 #ifdef CONFIG_SMP
2113 static inline
2114 void __dl_update(struct dl_bw *dl_b, s64 bw)
2115 {
2116 	struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2117 	int i;
2118 
2119 	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2120 			 "sched RCU must be held");
2121 	for_each_cpu_and(i, rd->span, cpu_active_mask) {
2122 		struct rq *rq = cpu_rq(i);
2123 
2124 		rq->dl.extra_bw += bw;
2125 	}
2126 }
2127 #else
2128 static inline
2129 void __dl_update(struct dl_bw *dl_b, s64 bw)
2130 {
2131 	struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2132 
2133 	dl->extra_bw += bw;
2134 }
2135 #endif
2136 
2137 
2138 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2139 struct irqtime {
2140 	u64			total;
2141 	u64			tick_delta;
2142 	u64			irq_start_time;
2143 	struct u64_stats_sync	sync;
2144 };
2145 
2146 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2147 
2148 /*
2149  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2150  * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2151  * and never move forward.
2152  */
2153 static inline u64 irq_time_read(int cpu)
2154 {
2155 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2156 	unsigned int seq;
2157 	u64 total;
2158 
2159 	do {
2160 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2161 		total = irqtime->total;
2162 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2163 
2164 	return total;
2165 }
2166 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2167 
2168 #ifdef CONFIG_CPU_FREQ
2169 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2170 
2171 /**
2172  * cpufreq_update_util - Take a note about CPU utilization changes.
2173  * @rq: Runqueue to carry out the update for.
2174  * @flags: Update reason flags.
2175  *
2176  * This function is called by the scheduler on the CPU whose utilization is
2177  * being updated.
2178  *
2179  * It can only be called from RCU-sched read-side critical sections.
2180  *
2181  * The way cpufreq is currently arranged requires it to evaluate the CPU
2182  * performance state (frequency/voltage) on a regular basis to prevent it from
2183  * being stuck in a completely inadequate performance level for too long.
2184  * That is not guaranteed to happen if the updates are only triggered from CFS
2185  * and DL, though, because they may not be coming in if only RT tasks are
2186  * active all the time (or there are RT tasks only).
2187  *
2188  * As a workaround for that issue, this function is called periodically by the
2189  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2190  * but that really is a band-aid.  Going forward it should be replaced with
2191  * solutions targeted more specifically at RT tasks.
2192  */
2193 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2194 {
2195 	struct update_util_data *data;
2196 
2197 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2198 						  cpu_of(rq)));
2199 	if (data)
2200 		data->func(data, rq_clock(rq), flags);
2201 }
2202 #else
2203 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2204 #endif /* CONFIG_CPU_FREQ */
2205 
2206 #ifdef arch_scale_freq_capacity
2207 # ifndef arch_scale_freq_invariant
2208 #  define arch_scale_freq_invariant()	true
2209 # endif
2210 #else
2211 # define arch_scale_freq_invariant()	false
2212 #endif
2213 
2214 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2215 /**
2216  * enum schedutil_type - CPU utilization type
2217  * @FREQUENCY_UTIL:	Utilization used to select frequency
2218  * @ENERGY_UTIL:	Utilization used during energy calculation
2219  *
2220  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2221  * need to be aggregated differently depending on the usage made of them. This
2222  * enum is used within schedutil_freq_util() to differentiate the types of
2223  * utilization expected by the callers, and adjust the aggregation accordingly.
2224  */
2225 enum schedutil_type {
2226 	FREQUENCY_UTIL,
2227 	ENERGY_UTIL,
2228 };
2229 
2230 unsigned long schedutil_freq_util(int cpu, unsigned long util_cfs,
2231 				  unsigned long max, enum schedutil_type type);
2232 
2233 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2234 {
2235 	unsigned long max = arch_scale_cpu_capacity(NULL, cpu);
2236 
2237 	return schedutil_freq_util(cpu, cfs, max, ENERGY_UTIL);
2238 }
2239 
2240 static inline unsigned long cpu_bw_dl(struct rq *rq)
2241 {
2242 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2243 }
2244 
2245 static inline unsigned long cpu_util_dl(struct rq *rq)
2246 {
2247 	return READ_ONCE(rq->avg_dl.util_avg);
2248 }
2249 
2250 static inline unsigned long cpu_util_cfs(struct rq *rq)
2251 {
2252 	unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2253 
2254 	if (sched_feat(UTIL_EST)) {
2255 		util = max_t(unsigned long, util,
2256 			     READ_ONCE(rq->cfs.avg.util_est.enqueued));
2257 	}
2258 
2259 	return util;
2260 }
2261 
2262 static inline unsigned long cpu_util_rt(struct rq *rq)
2263 {
2264 	return READ_ONCE(rq->avg_rt.util_avg);
2265 }
2266 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2267 static inline unsigned long schedutil_energy_util(int cpu, unsigned long cfs)
2268 {
2269 	return cfs;
2270 }
2271 #endif
2272 
2273 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2274 static inline unsigned long cpu_util_irq(struct rq *rq)
2275 {
2276 	return rq->avg_irq.util_avg;
2277 }
2278 
2279 static inline
2280 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2281 {
2282 	util *= (max - irq);
2283 	util /= max;
2284 
2285 	return util;
2286 
2287 }
2288 #else
2289 static inline unsigned long cpu_util_irq(struct rq *rq)
2290 {
2291 	return 0;
2292 }
2293 
2294 static inline
2295 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2296 {
2297 	return util;
2298 }
2299 #endif
2300 
2301 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2302 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2303 #else
2304 #define perf_domain_span(pd) NULL
2305 #endif
2306 
2307 #ifdef CONFIG_SMP
2308 extern struct static_key_false sched_energy_present;
2309 #endif
2310