xref: /openbmc/linux/kernel/sched/sched.h (revision b830f94f)
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 /*
100  * Helpers for converting nanosecond timing to jiffy resolution
101  */
102 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
103 
104 /*
105  * Increase resolution of nice-level calculations for 64-bit architectures.
106  * The extra resolution improves shares distribution and load balancing of
107  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108  * hierarchies, especially on larger systems. This is not a user-visible change
109  * and does not change the user-interface for setting shares/weights.
110  *
111  * We increase resolution only if we have enough bits to allow this increased
112  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113  * are pretty high and the returns do not justify the increased costs.
114  *
115  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116  * increase coverage and consistency always enable it on 64-bit platforms.
117  */
118 #ifdef CONFIG_64BIT
119 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w)	((w) >> SCHED_FIXEDPOINT_SHIFT)
122 #else
123 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w)		(w)
125 # define scale_load_down(w)	(w)
126 #endif
127 
128 /*
129  * Task weight (visible to users) and its load (invisible to users) have
130  * independent resolution, but they should be well calibrated. We use
131  * scale_load() and scale_load_down(w) to convert between them. The
132  * following must be true:
133  *
134  *  scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
135  *
136  */
137 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
138 
139 /*
140  * Single value that decides SCHED_DEADLINE internal math precision.
141  * 10 -> just above 1us
142  * 9  -> just above 0.5us
143  */
144 #define DL_SCALE		10
145 
146 /*
147  * Single value that denotes runtime == period, ie unlimited time.
148  */
149 #define RUNTIME_INF		((u64)~0ULL)
150 
151 static inline int idle_policy(int policy)
152 {
153 	return policy == SCHED_IDLE;
154 }
155 static inline int fair_policy(int policy)
156 {
157 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
158 }
159 
160 static inline int rt_policy(int policy)
161 {
162 	return policy == SCHED_FIFO || policy == SCHED_RR;
163 }
164 
165 static inline int dl_policy(int policy)
166 {
167 	return policy == SCHED_DEADLINE;
168 }
169 static inline bool valid_policy(int policy)
170 {
171 	return idle_policy(policy) || fair_policy(policy) ||
172 		rt_policy(policy) || dl_policy(policy);
173 }
174 
175 static inline int task_has_idle_policy(struct task_struct *p)
176 {
177 	return idle_policy(p->policy);
178 }
179 
180 static inline int task_has_rt_policy(struct task_struct *p)
181 {
182 	return rt_policy(p->policy);
183 }
184 
185 static inline int task_has_dl_policy(struct task_struct *p)
186 {
187 	return dl_policy(p->policy);
188 }
189 
190 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
191 
192 /*
193  * !! For sched_setattr_nocheck() (kernel) only !!
194  *
195  * This is actually gross. :(
196  *
197  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
198  * tasks, but still be able to sleep. We need this on platforms that cannot
199  * atomically change clock frequency. Remove once fast switching will be
200  * available on such platforms.
201  *
202  * SUGOV stands for SchedUtil GOVernor.
203  */
204 #define SCHED_FLAG_SUGOV	0x10000000
205 
206 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
207 {
208 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
209 	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
210 #else
211 	return false;
212 #endif
213 }
214 
215 /*
216  * Tells if entity @a should preempt entity @b.
217  */
218 static inline bool
219 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
220 {
221 	return dl_entity_is_special(a) ||
222 	       dl_time_before(a->deadline, b->deadline);
223 }
224 
225 /*
226  * This is the priority-queue data structure of the RT scheduling class:
227  */
228 struct rt_prio_array {
229 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
230 	struct list_head queue[MAX_RT_PRIO];
231 };
232 
233 struct rt_bandwidth {
234 	/* nests inside the rq lock: */
235 	raw_spinlock_t		rt_runtime_lock;
236 	ktime_t			rt_period;
237 	u64			rt_runtime;
238 	struct hrtimer		rt_period_timer;
239 	unsigned int		rt_period_active;
240 };
241 
242 void __dl_clear_params(struct task_struct *p);
243 
244 /*
245  * To keep the bandwidth of -deadline tasks and groups under control
246  * we need some place where:
247  *  - store the maximum -deadline bandwidth of the system (the group);
248  *  - cache the fraction of that bandwidth that is currently allocated.
249  *
250  * This is all done in the data structure below. It is similar to the
251  * one used for RT-throttling (rt_bandwidth), with the main difference
252  * that, since here we are only interested in admission control, we
253  * do not decrease any runtime while the group "executes", neither we
254  * need a timer to replenish it.
255  *
256  * With respect to SMP, the bandwidth is given on a per-CPU basis,
257  * meaning that:
258  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
259  *  - dl_total_bw array contains, in the i-eth element, the currently
260  *    allocated bandwidth on the i-eth CPU.
261  * Moreover, groups consume bandwidth on each CPU, while tasks only
262  * consume bandwidth on the CPU they're running on.
263  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
264  * that will be shown the next time the proc or cgroup controls will
265  * be red. It on its turn can be changed by writing on its own
266  * control.
267  */
268 struct dl_bandwidth {
269 	raw_spinlock_t		dl_runtime_lock;
270 	u64			dl_runtime;
271 	u64			dl_period;
272 };
273 
274 static inline int dl_bandwidth_enabled(void)
275 {
276 	return sysctl_sched_rt_runtime >= 0;
277 }
278 
279 struct dl_bw {
280 	raw_spinlock_t		lock;
281 	u64			bw;
282 	u64			total_bw;
283 };
284 
285 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
286 
287 static inline
288 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
289 {
290 	dl_b->total_bw -= tsk_bw;
291 	__dl_update(dl_b, (s32)tsk_bw / cpus);
292 }
293 
294 static inline
295 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
296 {
297 	dl_b->total_bw += tsk_bw;
298 	__dl_update(dl_b, -((s32)tsk_bw / cpus));
299 }
300 
301 static inline
302 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
303 {
304 	return dl_b->bw != -1 &&
305 	       dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
306 }
307 
308 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
309 extern void init_dl_bw(struct dl_bw *dl_b);
310 extern int  sched_dl_global_validate(void);
311 extern void sched_dl_do_global(void);
312 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
313 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
314 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
315 extern bool __checkparam_dl(const struct sched_attr *attr);
316 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
317 extern int  dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
318 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
319 extern bool dl_cpu_busy(unsigned int cpu);
320 
321 #ifdef CONFIG_CGROUP_SCHED
322 
323 #include <linux/cgroup.h>
324 #include <linux/psi.h>
325 
326 struct cfs_rq;
327 struct rt_rq;
328 
329 extern struct list_head task_groups;
330 
331 struct cfs_bandwidth {
332 #ifdef CONFIG_CFS_BANDWIDTH
333 	raw_spinlock_t		lock;
334 	ktime_t			period;
335 	u64			quota;
336 	u64			runtime;
337 	s64			hierarchical_quota;
338 	u64			runtime_expires;
339 	int			expires_seq;
340 
341 	u8			idle;
342 	u8			period_active;
343 	u8			distribute_running;
344 	u8			slack_started;
345 	struct hrtimer		period_timer;
346 	struct hrtimer		slack_timer;
347 	struct list_head	throttled_cfs_rq;
348 
349 	/* Statistics: */
350 	int			nr_periods;
351 	int			nr_throttled;
352 	u64			throttled_time;
353 #endif
354 };
355 
356 /* Task group related information */
357 struct task_group {
358 	struct cgroup_subsys_state css;
359 
360 #ifdef CONFIG_FAIR_GROUP_SCHED
361 	/* schedulable entities of this group on each CPU */
362 	struct sched_entity	**se;
363 	/* runqueue "owned" by this group on each CPU */
364 	struct cfs_rq		**cfs_rq;
365 	unsigned long		shares;
366 
367 #ifdef	CONFIG_SMP
368 	/*
369 	 * load_avg can be heavily contended at clock tick time, so put
370 	 * it in its own cacheline separated from the fields above which
371 	 * will also be accessed at each tick.
372 	 */
373 	atomic_long_t		load_avg ____cacheline_aligned;
374 #endif
375 #endif
376 
377 #ifdef CONFIG_RT_GROUP_SCHED
378 	struct sched_rt_entity	**rt_se;
379 	struct rt_rq		**rt_rq;
380 
381 	struct rt_bandwidth	rt_bandwidth;
382 #endif
383 
384 	struct rcu_head		rcu;
385 	struct list_head	list;
386 
387 	struct task_group	*parent;
388 	struct list_head	siblings;
389 	struct list_head	children;
390 
391 #ifdef CONFIG_SCHED_AUTOGROUP
392 	struct autogroup	*autogroup;
393 #endif
394 
395 	struct cfs_bandwidth	cfs_bandwidth;
396 };
397 
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
400 
401 /*
402  * A weight of 0 or 1 can cause arithmetics problems.
403  * A weight of a cfs_rq is the sum of weights of which entities
404  * are queued on this cfs_rq, so a weight of a entity should not be
405  * too large, so as the shares value of a task group.
406  * (The default weight is 1024 - so there's no practical
407  *  limitation from this.)
408  */
409 #define MIN_SHARES		(1UL <<  1)
410 #define MAX_SHARES		(1UL << 18)
411 #endif
412 
413 typedef int (*tg_visitor)(struct task_group *, void *);
414 
415 extern int walk_tg_tree_from(struct task_group *from,
416 			     tg_visitor down, tg_visitor up, void *data);
417 
418 /*
419  * Iterate the full tree, calling @down when first entering a node and @up when
420  * leaving it for the final time.
421  *
422  * Caller must hold rcu_lock or sufficient equivalent.
423  */
424 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
425 {
426 	return walk_tg_tree_from(&root_task_group, down, up, data);
427 }
428 
429 extern int tg_nop(struct task_group *tg, void *data);
430 
431 extern void free_fair_sched_group(struct task_group *tg);
432 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
433 extern void online_fair_sched_group(struct task_group *tg);
434 extern void unregister_fair_sched_group(struct task_group *tg);
435 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
436 			struct sched_entity *se, int cpu,
437 			struct sched_entity *parent);
438 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
439 
440 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
441 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
442 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
443 
444 extern void free_rt_sched_group(struct task_group *tg);
445 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
446 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
447 		struct sched_rt_entity *rt_se, int cpu,
448 		struct sched_rt_entity *parent);
449 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
450 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
451 extern long sched_group_rt_runtime(struct task_group *tg);
452 extern long sched_group_rt_period(struct task_group *tg);
453 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
454 
455 extern struct task_group *sched_create_group(struct task_group *parent);
456 extern void sched_online_group(struct task_group *tg,
457 			       struct task_group *parent);
458 extern void sched_destroy_group(struct task_group *tg);
459 extern void sched_offline_group(struct task_group *tg);
460 
461 extern void sched_move_task(struct task_struct *tsk);
462 
463 #ifdef CONFIG_FAIR_GROUP_SCHED
464 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
465 
466 #ifdef CONFIG_SMP
467 extern void set_task_rq_fair(struct sched_entity *se,
468 			     struct cfs_rq *prev, struct cfs_rq *next);
469 #else /* !CONFIG_SMP */
470 static inline void set_task_rq_fair(struct sched_entity *se,
471 			     struct cfs_rq *prev, struct cfs_rq *next) { }
472 #endif /* CONFIG_SMP */
473 #endif /* CONFIG_FAIR_GROUP_SCHED */
474 
475 #else /* CONFIG_CGROUP_SCHED */
476 
477 struct cfs_bandwidth { };
478 
479 #endif	/* CONFIG_CGROUP_SCHED */
480 
481 /* CFS-related fields in a runqueue */
482 struct cfs_rq {
483 	struct load_weight	load;
484 	unsigned long		runnable_weight;
485 	unsigned int		nr_running;
486 	unsigned int		h_nr_running;
487 
488 	u64			exec_clock;
489 	u64			min_vruntime;
490 #ifndef CONFIG_64BIT
491 	u64			min_vruntime_copy;
492 #endif
493 
494 	struct rb_root_cached	tasks_timeline;
495 
496 	/*
497 	 * 'curr' points to currently running entity on this cfs_rq.
498 	 * It is set to NULL otherwise (i.e when none are currently running).
499 	 */
500 	struct sched_entity	*curr;
501 	struct sched_entity	*next;
502 	struct sched_entity	*last;
503 	struct sched_entity	*skip;
504 
505 #ifdef	CONFIG_SCHED_DEBUG
506 	unsigned int		nr_spread_over;
507 #endif
508 
509 #ifdef CONFIG_SMP
510 	/*
511 	 * CFS load tracking
512 	 */
513 	struct sched_avg	avg;
514 #ifndef CONFIG_64BIT
515 	u64			load_last_update_time_copy;
516 #endif
517 	struct {
518 		raw_spinlock_t	lock ____cacheline_aligned;
519 		int		nr;
520 		unsigned long	load_avg;
521 		unsigned long	util_avg;
522 		unsigned long	runnable_sum;
523 	} removed;
524 
525 #ifdef CONFIG_FAIR_GROUP_SCHED
526 	unsigned long		tg_load_avg_contrib;
527 	long			propagate;
528 	long			prop_runnable_sum;
529 
530 	/*
531 	 *   h_load = weight * f(tg)
532 	 *
533 	 * Where f(tg) is the recursive weight fraction assigned to
534 	 * this group.
535 	 */
536 	unsigned long		h_load;
537 	u64			last_h_load_update;
538 	struct sched_entity	*h_load_next;
539 #endif /* CONFIG_FAIR_GROUP_SCHED */
540 #endif /* CONFIG_SMP */
541 
542 #ifdef CONFIG_FAIR_GROUP_SCHED
543 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
544 
545 	/*
546 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
547 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
548 	 * (like users, containers etc.)
549 	 *
550 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
551 	 * This list is used during load balance.
552 	 */
553 	int			on_list;
554 	struct list_head	leaf_cfs_rq_list;
555 	struct task_group	*tg;	/* group that "owns" this runqueue */
556 
557 #ifdef CONFIG_CFS_BANDWIDTH
558 	int			runtime_enabled;
559 	int			expires_seq;
560 	u64			runtime_expires;
561 	s64			runtime_remaining;
562 
563 	u64			throttled_clock;
564 	u64			throttled_clock_task;
565 	u64			throttled_clock_task_time;
566 	int			throttled;
567 	int			throttle_count;
568 	struct list_head	throttled_list;
569 #endif /* CONFIG_CFS_BANDWIDTH */
570 #endif /* CONFIG_FAIR_GROUP_SCHED */
571 };
572 
573 static inline int rt_bandwidth_enabled(void)
574 {
575 	return sysctl_sched_rt_runtime >= 0;
576 }
577 
578 /* RT IPI pull logic requires IRQ_WORK */
579 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
580 # define HAVE_RT_PUSH_IPI
581 #endif
582 
583 /* Real-Time classes' related field in a runqueue: */
584 struct rt_rq {
585 	struct rt_prio_array	active;
586 	unsigned int		rt_nr_running;
587 	unsigned int		rr_nr_running;
588 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
589 	struct {
590 		int		curr; /* highest queued rt task prio */
591 #ifdef CONFIG_SMP
592 		int		next; /* next highest */
593 #endif
594 	} highest_prio;
595 #endif
596 #ifdef CONFIG_SMP
597 	unsigned long		rt_nr_migratory;
598 	unsigned long		rt_nr_total;
599 	int			overloaded;
600 	struct plist_head	pushable_tasks;
601 
602 #endif /* CONFIG_SMP */
603 	int			rt_queued;
604 
605 	int			rt_throttled;
606 	u64			rt_time;
607 	u64			rt_runtime;
608 	/* Nests inside the rq lock: */
609 	raw_spinlock_t		rt_runtime_lock;
610 
611 #ifdef CONFIG_RT_GROUP_SCHED
612 	unsigned long		rt_nr_boosted;
613 
614 	struct rq		*rq;
615 	struct task_group	*tg;
616 #endif
617 };
618 
619 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
620 {
621 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
622 }
623 
624 /* Deadline class' related fields in a runqueue */
625 struct dl_rq {
626 	/* runqueue is an rbtree, ordered by deadline */
627 	struct rb_root_cached	root;
628 
629 	unsigned long		dl_nr_running;
630 
631 #ifdef CONFIG_SMP
632 	/*
633 	 * Deadline values of the currently executing and the
634 	 * earliest ready task on this rq. Caching these facilitates
635 	 * the decision whether or not a ready but not running task
636 	 * should migrate somewhere else.
637 	 */
638 	struct {
639 		u64		curr;
640 		u64		next;
641 	} earliest_dl;
642 
643 	unsigned long		dl_nr_migratory;
644 	int			overloaded;
645 
646 	/*
647 	 * Tasks on this rq that can be pushed away. They are kept in
648 	 * an rb-tree, ordered by tasks' deadlines, with caching
649 	 * of the leftmost (earliest deadline) element.
650 	 */
651 	struct rb_root_cached	pushable_dl_tasks_root;
652 #else
653 	struct dl_bw		dl_bw;
654 #endif
655 	/*
656 	 * "Active utilization" for this runqueue: increased when a
657 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
658 	 * task blocks
659 	 */
660 	u64			running_bw;
661 
662 	/*
663 	 * Utilization of the tasks "assigned" to this runqueue (including
664 	 * the tasks that are in runqueue and the tasks that executed on this
665 	 * CPU and blocked). Increased when a task moves to this runqueue, and
666 	 * decreased when the task moves away (migrates, changes scheduling
667 	 * policy, or terminates).
668 	 * This is needed to compute the "inactive utilization" for the
669 	 * runqueue (inactive utilization = this_bw - running_bw).
670 	 */
671 	u64			this_bw;
672 	u64			extra_bw;
673 
674 	/*
675 	 * Inverse of the fraction of CPU utilization that can be reclaimed
676 	 * by the GRUB algorithm.
677 	 */
678 	u64			bw_ratio;
679 };
680 
681 #ifdef CONFIG_FAIR_GROUP_SCHED
682 /* An entity is a task if it doesn't "own" a runqueue */
683 #define entity_is_task(se)	(!se->my_q)
684 #else
685 #define entity_is_task(se)	1
686 #endif
687 
688 #ifdef CONFIG_SMP
689 /*
690  * XXX we want to get rid of these helpers and use the full load resolution.
691  */
692 static inline long se_weight(struct sched_entity *se)
693 {
694 	return scale_load_down(se->load.weight);
695 }
696 
697 static inline long se_runnable(struct sched_entity *se)
698 {
699 	return scale_load_down(se->runnable_weight);
700 }
701 
702 static inline bool sched_asym_prefer(int a, int b)
703 {
704 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
705 }
706 
707 struct perf_domain {
708 	struct em_perf_domain *em_pd;
709 	struct perf_domain *next;
710 	struct rcu_head rcu;
711 };
712 
713 /* Scheduling group status flags */
714 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
715 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
716 
717 /*
718  * We add the notion of a root-domain which will be used to define per-domain
719  * variables. Each exclusive cpuset essentially defines an island domain by
720  * fully partitioning the member CPUs from any other cpuset. Whenever a new
721  * exclusive cpuset is created, we also create and attach a new root-domain
722  * object.
723  *
724  */
725 struct root_domain {
726 	atomic_t		refcount;
727 	atomic_t		rto_count;
728 	struct rcu_head		rcu;
729 	cpumask_var_t		span;
730 	cpumask_var_t		online;
731 
732 	/*
733 	 * Indicate pullable load on at least one CPU, e.g:
734 	 * - More than one runnable task
735 	 * - Running task is misfit
736 	 */
737 	int			overload;
738 
739 	/* Indicate one or more cpus over-utilized (tipping point) */
740 	int			overutilized;
741 
742 	/*
743 	 * The bit corresponding to a CPU gets set here if such CPU has more
744 	 * than one runnable -deadline task (as it is below for RT tasks).
745 	 */
746 	cpumask_var_t		dlo_mask;
747 	atomic_t		dlo_count;
748 	struct dl_bw		dl_bw;
749 	struct cpudl		cpudl;
750 
751 #ifdef HAVE_RT_PUSH_IPI
752 	/*
753 	 * For IPI pull requests, loop across the rto_mask.
754 	 */
755 	struct irq_work		rto_push_work;
756 	raw_spinlock_t		rto_lock;
757 	/* These are only updated and read within rto_lock */
758 	int			rto_loop;
759 	int			rto_cpu;
760 	/* These atomics are updated outside of a lock */
761 	atomic_t		rto_loop_next;
762 	atomic_t		rto_loop_start;
763 #endif
764 	/*
765 	 * The "RT overload" flag: it gets set if a CPU has more than
766 	 * one runnable RT task.
767 	 */
768 	cpumask_var_t		rto_mask;
769 	struct cpupri		cpupri;
770 
771 	unsigned long		max_cpu_capacity;
772 
773 	/*
774 	 * NULL-terminated list of performance domains intersecting with the
775 	 * CPUs of the rd. Protected by RCU.
776 	 */
777 	struct perf_domain __rcu *pd;
778 };
779 
780 extern struct root_domain def_root_domain;
781 extern struct mutex sched_domains_mutex;
782 
783 extern void init_defrootdomain(void);
784 extern int sched_init_domains(const struct cpumask *cpu_map);
785 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
786 extern void sched_get_rd(struct root_domain *rd);
787 extern void sched_put_rd(struct root_domain *rd);
788 
789 #ifdef HAVE_RT_PUSH_IPI
790 extern void rto_push_irq_work_func(struct irq_work *work);
791 #endif
792 #endif /* CONFIG_SMP */
793 
794 #ifdef CONFIG_UCLAMP_TASK
795 /*
796  * struct uclamp_bucket - Utilization clamp bucket
797  * @value: utilization clamp value for tasks on this clamp bucket
798  * @tasks: number of RUNNABLE tasks on this clamp bucket
799  *
800  * Keep track of how many tasks are RUNNABLE for a given utilization
801  * clamp value.
802  */
803 struct uclamp_bucket {
804 	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
805 	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
806 };
807 
808 /*
809  * struct uclamp_rq - rq's utilization clamp
810  * @value: currently active clamp values for a rq
811  * @bucket: utilization clamp buckets affecting a rq
812  *
813  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
814  * A clamp value is affecting a rq when there is at least one task RUNNABLE
815  * (or actually running) with that value.
816  *
817  * There are up to UCLAMP_CNT possible different clamp values, currently there
818  * are only two: minimum utilization and maximum utilization.
819  *
820  * All utilization clamping values are MAX aggregated, since:
821  * - for util_min: we want to run the CPU at least at the max of the minimum
822  *   utilization required by its currently RUNNABLE tasks.
823  * - for util_max: we want to allow the CPU to run up to the max of the
824  *   maximum utilization allowed by its currently RUNNABLE tasks.
825  *
826  * Since on each system we expect only a limited number of different
827  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
828  * the metrics required to compute all the per-rq utilization clamp values.
829  */
830 struct uclamp_rq {
831 	unsigned int value;
832 	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
833 };
834 #endif /* CONFIG_UCLAMP_TASK */
835 
836 /*
837  * This is the main, per-CPU runqueue data structure.
838  *
839  * Locking rule: those places that want to lock multiple runqueues
840  * (such as the load balancing or the thread migration code), lock
841  * acquire operations must be ordered by ascending &runqueue.
842  */
843 struct rq {
844 	/* runqueue lock: */
845 	raw_spinlock_t		lock;
846 
847 	/*
848 	 * nr_running and cpu_load should be in the same cacheline because
849 	 * remote CPUs use both these fields when doing load calculation.
850 	 */
851 	unsigned int		nr_running;
852 #ifdef CONFIG_NUMA_BALANCING
853 	unsigned int		nr_numa_running;
854 	unsigned int		nr_preferred_running;
855 	unsigned int		numa_migrate_on;
856 #endif
857 #ifdef CONFIG_NO_HZ_COMMON
858 #ifdef CONFIG_SMP
859 	unsigned long		last_load_update_tick;
860 	unsigned long		last_blocked_load_update_tick;
861 	unsigned int		has_blocked_load;
862 #endif /* CONFIG_SMP */
863 	unsigned int		nohz_tick_stopped;
864 	atomic_t nohz_flags;
865 #endif /* CONFIG_NO_HZ_COMMON */
866 
867 	unsigned long		nr_load_updates;
868 	u64			nr_switches;
869 
870 #ifdef CONFIG_UCLAMP_TASK
871 	/* Utilization clamp values based on CPU's RUNNABLE tasks */
872 	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
873 	unsigned int		uclamp_flags;
874 #define UCLAMP_FLAG_IDLE 0x01
875 #endif
876 
877 	struct cfs_rq		cfs;
878 	struct rt_rq		rt;
879 	struct dl_rq		dl;
880 
881 #ifdef CONFIG_FAIR_GROUP_SCHED
882 	/* list of leaf cfs_rq on this CPU: */
883 	struct list_head	leaf_cfs_rq_list;
884 	struct list_head	*tmp_alone_branch;
885 #endif /* CONFIG_FAIR_GROUP_SCHED */
886 
887 	/*
888 	 * This is part of a global counter where only the total sum
889 	 * over all CPUs matters. A task can increase this counter on
890 	 * one CPU and if it got migrated afterwards it may decrease
891 	 * it on another CPU. Always updated under the runqueue lock:
892 	 */
893 	unsigned long		nr_uninterruptible;
894 
895 	struct task_struct	*curr;
896 	struct task_struct	*idle;
897 	struct task_struct	*stop;
898 	unsigned long		next_balance;
899 	struct mm_struct	*prev_mm;
900 
901 	unsigned int		clock_update_flags;
902 	u64			clock;
903 	/* Ensure that all clocks are in the same cache line */
904 	u64			clock_task ____cacheline_aligned;
905 	u64			clock_pelt;
906 	unsigned long		lost_idle_time;
907 
908 	atomic_t		nr_iowait;
909 
910 #ifdef CONFIG_SMP
911 	struct root_domain		*rd;
912 	struct sched_domain __rcu	*sd;
913 
914 	unsigned long		cpu_capacity;
915 	unsigned long		cpu_capacity_orig;
916 
917 	struct callback_head	*balance_callback;
918 
919 	unsigned char		idle_balance;
920 
921 	unsigned long		misfit_task_load;
922 
923 	/* For active balancing */
924 	int			active_balance;
925 	int			push_cpu;
926 	struct cpu_stop_work	active_balance_work;
927 
928 	/* CPU of this runqueue: */
929 	int			cpu;
930 	int			online;
931 
932 	struct list_head cfs_tasks;
933 
934 	struct sched_avg	avg_rt;
935 	struct sched_avg	avg_dl;
936 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
937 	struct sched_avg	avg_irq;
938 #endif
939 	u64			idle_stamp;
940 	u64			avg_idle;
941 
942 	/* This is used to determine avg_idle's max value */
943 	u64			max_idle_balance_cost;
944 #endif
945 
946 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
947 	u64			prev_irq_time;
948 #endif
949 #ifdef CONFIG_PARAVIRT
950 	u64			prev_steal_time;
951 #endif
952 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
953 	u64			prev_steal_time_rq;
954 #endif
955 
956 	/* calc_load related fields */
957 	unsigned long		calc_load_update;
958 	long			calc_load_active;
959 
960 #ifdef CONFIG_SCHED_HRTICK
961 #ifdef CONFIG_SMP
962 	int			hrtick_csd_pending;
963 	call_single_data_t	hrtick_csd;
964 #endif
965 	struct hrtimer		hrtick_timer;
966 #endif
967 
968 #ifdef CONFIG_SCHEDSTATS
969 	/* latency stats */
970 	struct sched_info	rq_sched_info;
971 	unsigned long long	rq_cpu_time;
972 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
973 
974 	/* sys_sched_yield() stats */
975 	unsigned int		yld_count;
976 
977 	/* schedule() stats */
978 	unsigned int		sched_count;
979 	unsigned int		sched_goidle;
980 
981 	/* try_to_wake_up() stats */
982 	unsigned int		ttwu_count;
983 	unsigned int		ttwu_local;
984 #endif
985 
986 #ifdef CONFIG_SMP
987 	struct llist_head	wake_list;
988 #endif
989 
990 #ifdef CONFIG_CPU_IDLE
991 	/* Must be inspected within a rcu lock section */
992 	struct cpuidle_state	*idle_state;
993 #endif
994 };
995 
996 #ifdef CONFIG_FAIR_GROUP_SCHED
997 
998 /* CPU runqueue to which this cfs_rq is attached */
999 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1000 {
1001 	return cfs_rq->rq;
1002 }
1003 
1004 #else
1005 
1006 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1007 {
1008 	return container_of(cfs_rq, struct rq, cfs);
1009 }
1010 #endif
1011 
1012 static inline int cpu_of(struct rq *rq)
1013 {
1014 #ifdef CONFIG_SMP
1015 	return rq->cpu;
1016 #else
1017 	return 0;
1018 #endif
1019 }
1020 
1021 
1022 #ifdef CONFIG_SCHED_SMT
1023 extern void __update_idle_core(struct rq *rq);
1024 
1025 static inline void update_idle_core(struct rq *rq)
1026 {
1027 	if (static_branch_unlikely(&sched_smt_present))
1028 		__update_idle_core(rq);
1029 }
1030 
1031 #else
1032 static inline void update_idle_core(struct rq *rq) { }
1033 #endif
1034 
1035 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1036 
1037 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1038 #define this_rq()		this_cpu_ptr(&runqueues)
1039 #define task_rq(p)		cpu_rq(task_cpu(p))
1040 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1041 #define raw_rq()		raw_cpu_ptr(&runqueues)
1042 
1043 extern void update_rq_clock(struct rq *rq);
1044 
1045 static inline u64 __rq_clock_broken(struct rq *rq)
1046 {
1047 	return READ_ONCE(rq->clock);
1048 }
1049 
1050 /*
1051  * rq::clock_update_flags bits
1052  *
1053  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1054  *  call to __schedule(). This is an optimisation to avoid
1055  *  neighbouring rq clock updates.
1056  *
1057  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1058  *  in effect and calls to update_rq_clock() are being ignored.
1059  *
1060  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1061  *  made to update_rq_clock() since the last time rq::lock was pinned.
1062  *
1063  * If inside of __schedule(), clock_update_flags will have been
1064  * shifted left (a left shift is a cheap operation for the fast path
1065  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1066  *
1067  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1068  *
1069  * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1070  * one position though, because the next rq_unpin_lock() will shift it
1071  * back.
1072  */
1073 #define RQCF_REQ_SKIP		0x01
1074 #define RQCF_ACT_SKIP		0x02
1075 #define RQCF_UPDATED		0x04
1076 
1077 static inline void assert_clock_updated(struct rq *rq)
1078 {
1079 	/*
1080 	 * The only reason for not seeing a clock update since the
1081 	 * last rq_pin_lock() is if we're currently skipping updates.
1082 	 */
1083 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1084 }
1085 
1086 static inline u64 rq_clock(struct rq *rq)
1087 {
1088 	lockdep_assert_held(&rq->lock);
1089 	assert_clock_updated(rq);
1090 
1091 	return rq->clock;
1092 }
1093 
1094 static inline u64 rq_clock_task(struct rq *rq)
1095 {
1096 	lockdep_assert_held(&rq->lock);
1097 	assert_clock_updated(rq);
1098 
1099 	return rq->clock_task;
1100 }
1101 
1102 static inline void rq_clock_skip_update(struct rq *rq)
1103 {
1104 	lockdep_assert_held(&rq->lock);
1105 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1106 }
1107 
1108 /*
1109  * See rt task throttling, which is the only time a skip
1110  * request is cancelled.
1111  */
1112 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1113 {
1114 	lockdep_assert_held(&rq->lock);
1115 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1116 }
1117 
1118 struct rq_flags {
1119 	unsigned long flags;
1120 	struct pin_cookie cookie;
1121 #ifdef CONFIG_SCHED_DEBUG
1122 	/*
1123 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1124 	 * current pin context is stashed here in case it needs to be
1125 	 * restored in rq_repin_lock().
1126 	 */
1127 	unsigned int clock_update_flags;
1128 #endif
1129 };
1130 
1131 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1132 {
1133 	rf->cookie = lockdep_pin_lock(&rq->lock);
1134 
1135 #ifdef CONFIG_SCHED_DEBUG
1136 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1137 	rf->clock_update_flags = 0;
1138 #endif
1139 }
1140 
1141 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1142 {
1143 #ifdef CONFIG_SCHED_DEBUG
1144 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1145 		rf->clock_update_flags = RQCF_UPDATED;
1146 #endif
1147 
1148 	lockdep_unpin_lock(&rq->lock, rf->cookie);
1149 }
1150 
1151 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1152 {
1153 	lockdep_repin_lock(&rq->lock, rf->cookie);
1154 
1155 #ifdef CONFIG_SCHED_DEBUG
1156 	/*
1157 	 * Restore the value we stashed in @rf for this pin context.
1158 	 */
1159 	rq->clock_update_flags |= rf->clock_update_flags;
1160 #endif
1161 }
1162 
1163 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1164 	__acquires(rq->lock);
1165 
1166 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1167 	__acquires(p->pi_lock)
1168 	__acquires(rq->lock);
1169 
1170 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1171 	__releases(rq->lock)
1172 {
1173 	rq_unpin_lock(rq, rf);
1174 	raw_spin_unlock(&rq->lock);
1175 }
1176 
1177 static inline void
1178 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1179 	__releases(rq->lock)
1180 	__releases(p->pi_lock)
1181 {
1182 	rq_unpin_lock(rq, rf);
1183 	raw_spin_unlock(&rq->lock);
1184 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1185 }
1186 
1187 static inline void
1188 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1189 	__acquires(rq->lock)
1190 {
1191 	raw_spin_lock_irqsave(&rq->lock, rf->flags);
1192 	rq_pin_lock(rq, rf);
1193 }
1194 
1195 static inline void
1196 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1197 	__acquires(rq->lock)
1198 {
1199 	raw_spin_lock_irq(&rq->lock);
1200 	rq_pin_lock(rq, rf);
1201 }
1202 
1203 static inline void
1204 rq_lock(struct rq *rq, struct rq_flags *rf)
1205 	__acquires(rq->lock)
1206 {
1207 	raw_spin_lock(&rq->lock);
1208 	rq_pin_lock(rq, rf);
1209 }
1210 
1211 static inline void
1212 rq_relock(struct rq *rq, struct rq_flags *rf)
1213 	__acquires(rq->lock)
1214 {
1215 	raw_spin_lock(&rq->lock);
1216 	rq_repin_lock(rq, rf);
1217 }
1218 
1219 static inline void
1220 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1221 	__releases(rq->lock)
1222 {
1223 	rq_unpin_lock(rq, rf);
1224 	raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1225 }
1226 
1227 static inline void
1228 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1229 	__releases(rq->lock)
1230 {
1231 	rq_unpin_lock(rq, rf);
1232 	raw_spin_unlock_irq(&rq->lock);
1233 }
1234 
1235 static inline void
1236 rq_unlock(struct rq *rq, struct rq_flags *rf)
1237 	__releases(rq->lock)
1238 {
1239 	rq_unpin_lock(rq, rf);
1240 	raw_spin_unlock(&rq->lock);
1241 }
1242 
1243 static inline struct rq *
1244 this_rq_lock_irq(struct rq_flags *rf)
1245 	__acquires(rq->lock)
1246 {
1247 	struct rq *rq;
1248 
1249 	local_irq_disable();
1250 	rq = this_rq();
1251 	rq_lock(rq, rf);
1252 	return rq;
1253 }
1254 
1255 #ifdef CONFIG_NUMA
1256 enum numa_topology_type {
1257 	NUMA_DIRECT,
1258 	NUMA_GLUELESS_MESH,
1259 	NUMA_BACKPLANE,
1260 };
1261 extern enum numa_topology_type sched_numa_topology_type;
1262 extern int sched_max_numa_distance;
1263 extern bool find_numa_distance(int distance);
1264 #endif
1265 
1266 #ifdef CONFIG_NUMA
1267 extern void sched_init_numa(void);
1268 extern void sched_domains_numa_masks_set(unsigned int cpu);
1269 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1270 #else
1271 static inline void sched_init_numa(void) { }
1272 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1273 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1274 #endif
1275 
1276 #ifdef CONFIG_NUMA_BALANCING
1277 /* The regions in numa_faults array from task_struct */
1278 enum numa_faults_stats {
1279 	NUMA_MEM = 0,
1280 	NUMA_CPU,
1281 	NUMA_MEMBUF,
1282 	NUMA_CPUBUF
1283 };
1284 extern void sched_setnuma(struct task_struct *p, int node);
1285 extern int migrate_task_to(struct task_struct *p, int cpu);
1286 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1287 			int cpu, int scpu);
1288 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1289 #else
1290 static inline void
1291 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1292 {
1293 }
1294 #endif /* CONFIG_NUMA_BALANCING */
1295 
1296 #ifdef CONFIG_SMP
1297 
1298 static inline void
1299 queue_balance_callback(struct rq *rq,
1300 		       struct callback_head *head,
1301 		       void (*func)(struct rq *rq))
1302 {
1303 	lockdep_assert_held(&rq->lock);
1304 
1305 	if (unlikely(head->next))
1306 		return;
1307 
1308 	head->func = (void (*)(struct callback_head *))func;
1309 	head->next = rq->balance_callback;
1310 	rq->balance_callback = head;
1311 }
1312 
1313 extern void sched_ttwu_pending(void);
1314 
1315 #define rcu_dereference_check_sched_domain(p) \
1316 	rcu_dereference_check((p), \
1317 			      lockdep_is_held(&sched_domains_mutex))
1318 
1319 /*
1320  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1321  * See destroy_sched_domains: call_rcu for details.
1322  *
1323  * The domain tree of any CPU may only be accessed from within
1324  * preempt-disabled sections.
1325  */
1326 #define for_each_domain(cpu, __sd) \
1327 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1328 			__sd; __sd = __sd->parent)
1329 
1330 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1331 
1332 /**
1333  * highest_flag_domain - Return highest sched_domain containing flag.
1334  * @cpu:	The CPU whose highest level of sched domain is to
1335  *		be returned.
1336  * @flag:	The flag to check for the highest sched_domain
1337  *		for the given CPU.
1338  *
1339  * Returns the highest sched_domain of a CPU which contains the given flag.
1340  */
1341 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1342 {
1343 	struct sched_domain *sd, *hsd = NULL;
1344 
1345 	for_each_domain(cpu, sd) {
1346 		if (!(sd->flags & flag))
1347 			break;
1348 		hsd = sd;
1349 	}
1350 
1351 	return hsd;
1352 }
1353 
1354 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1355 {
1356 	struct sched_domain *sd;
1357 
1358 	for_each_domain(cpu, sd) {
1359 		if (sd->flags & flag)
1360 			break;
1361 	}
1362 
1363 	return sd;
1364 }
1365 
1366 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1367 DECLARE_PER_CPU(int, sd_llc_size);
1368 DECLARE_PER_CPU(int, sd_llc_id);
1369 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1370 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1371 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1372 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1373 extern struct static_key_false sched_asym_cpucapacity;
1374 
1375 struct sched_group_capacity {
1376 	atomic_t		ref;
1377 	/*
1378 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1379 	 * for a single CPU.
1380 	 */
1381 	unsigned long		capacity;
1382 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1383 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1384 	unsigned long		next_update;
1385 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1386 
1387 #ifdef CONFIG_SCHED_DEBUG
1388 	int			id;
1389 #endif
1390 
1391 	unsigned long		cpumask[0];		/* Balance mask */
1392 };
1393 
1394 struct sched_group {
1395 	struct sched_group	*next;			/* Must be a circular list */
1396 	atomic_t		ref;
1397 
1398 	unsigned int		group_weight;
1399 	struct sched_group_capacity *sgc;
1400 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1401 
1402 	/*
1403 	 * The CPUs this group covers.
1404 	 *
1405 	 * NOTE: this field is variable length. (Allocated dynamically
1406 	 * by attaching extra space to the end of the structure,
1407 	 * depending on how many CPUs the kernel has booted up with)
1408 	 */
1409 	unsigned long		cpumask[0];
1410 };
1411 
1412 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1413 {
1414 	return to_cpumask(sg->cpumask);
1415 }
1416 
1417 /*
1418  * See build_balance_mask().
1419  */
1420 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1421 {
1422 	return to_cpumask(sg->sgc->cpumask);
1423 }
1424 
1425 /**
1426  * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1427  * @group: The group whose first CPU is to be returned.
1428  */
1429 static inline unsigned int group_first_cpu(struct sched_group *group)
1430 {
1431 	return cpumask_first(sched_group_span(group));
1432 }
1433 
1434 extern int group_balance_cpu(struct sched_group *sg);
1435 
1436 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1437 void register_sched_domain_sysctl(void);
1438 void dirty_sched_domain_sysctl(int cpu);
1439 void unregister_sched_domain_sysctl(void);
1440 #else
1441 static inline void register_sched_domain_sysctl(void)
1442 {
1443 }
1444 static inline void dirty_sched_domain_sysctl(int cpu)
1445 {
1446 }
1447 static inline void unregister_sched_domain_sysctl(void)
1448 {
1449 }
1450 #endif
1451 
1452 #else
1453 
1454 static inline void sched_ttwu_pending(void) { }
1455 
1456 #endif /* CONFIG_SMP */
1457 
1458 #include "stats.h"
1459 #include "autogroup.h"
1460 
1461 #ifdef CONFIG_CGROUP_SCHED
1462 
1463 /*
1464  * Return the group to which this tasks belongs.
1465  *
1466  * We cannot use task_css() and friends because the cgroup subsystem
1467  * changes that value before the cgroup_subsys::attach() method is called,
1468  * therefore we cannot pin it and might observe the wrong value.
1469  *
1470  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1471  * core changes this before calling sched_move_task().
1472  *
1473  * Instead we use a 'copy' which is updated from sched_move_task() while
1474  * holding both task_struct::pi_lock and rq::lock.
1475  */
1476 static inline struct task_group *task_group(struct task_struct *p)
1477 {
1478 	return p->sched_task_group;
1479 }
1480 
1481 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1482 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1483 {
1484 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1485 	struct task_group *tg = task_group(p);
1486 #endif
1487 
1488 #ifdef CONFIG_FAIR_GROUP_SCHED
1489 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1490 	p->se.cfs_rq = tg->cfs_rq[cpu];
1491 	p->se.parent = tg->se[cpu];
1492 #endif
1493 
1494 #ifdef CONFIG_RT_GROUP_SCHED
1495 	p->rt.rt_rq  = tg->rt_rq[cpu];
1496 	p->rt.parent = tg->rt_se[cpu];
1497 #endif
1498 }
1499 
1500 #else /* CONFIG_CGROUP_SCHED */
1501 
1502 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1503 static inline struct task_group *task_group(struct task_struct *p)
1504 {
1505 	return NULL;
1506 }
1507 
1508 #endif /* CONFIG_CGROUP_SCHED */
1509 
1510 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1511 {
1512 	set_task_rq(p, cpu);
1513 #ifdef CONFIG_SMP
1514 	/*
1515 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1516 	 * successfully executed on another CPU. We must ensure that updates of
1517 	 * per-task data have been completed by this moment.
1518 	 */
1519 	smp_wmb();
1520 #ifdef CONFIG_THREAD_INFO_IN_TASK
1521 	WRITE_ONCE(p->cpu, cpu);
1522 #else
1523 	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1524 #endif
1525 	p->wake_cpu = cpu;
1526 #endif
1527 }
1528 
1529 /*
1530  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1531  */
1532 #ifdef CONFIG_SCHED_DEBUG
1533 # include <linux/static_key.h>
1534 # define const_debug __read_mostly
1535 #else
1536 # define const_debug const
1537 #endif
1538 
1539 #define SCHED_FEAT(name, enabled)	\
1540 	__SCHED_FEAT_##name ,
1541 
1542 enum {
1543 #include "features.h"
1544 	__SCHED_FEAT_NR,
1545 };
1546 
1547 #undef SCHED_FEAT
1548 
1549 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1550 
1551 /*
1552  * To support run-time toggling of sched features, all the translation units
1553  * (but core.c) reference the sysctl_sched_features defined in core.c.
1554  */
1555 extern const_debug unsigned int sysctl_sched_features;
1556 
1557 #define SCHED_FEAT(name, enabled)					\
1558 static __always_inline bool static_branch_##name(struct static_key *key) \
1559 {									\
1560 	return static_key_##enabled(key);				\
1561 }
1562 
1563 #include "features.h"
1564 #undef SCHED_FEAT
1565 
1566 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1567 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1568 
1569 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1570 
1571 /*
1572  * Each translation unit has its own copy of sysctl_sched_features to allow
1573  * constants propagation at compile time and compiler optimization based on
1574  * features default.
1575  */
1576 #define SCHED_FEAT(name, enabled)	\
1577 	(1UL << __SCHED_FEAT_##name) * enabled |
1578 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1579 #include "features.h"
1580 	0;
1581 #undef SCHED_FEAT
1582 
1583 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1584 
1585 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1586 
1587 extern struct static_key_false sched_numa_balancing;
1588 extern struct static_key_false sched_schedstats;
1589 
1590 static inline u64 global_rt_period(void)
1591 {
1592 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1593 }
1594 
1595 static inline u64 global_rt_runtime(void)
1596 {
1597 	if (sysctl_sched_rt_runtime < 0)
1598 		return RUNTIME_INF;
1599 
1600 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1601 }
1602 
1603 static inline int task_current(struct rq *rq, struct task_struct *p)
1604 {
1605 	return rq->curr == p;
1606 }
1607 
1608 static inline int task_running(struct rq *rq, struct task_struct *p)
1609 {
1610 #ifdef CONFIG_SMP
1611 	return p->on_cpu;
1612 #else
1613 	return task_current(rq, p);
1614 #endif
1615 }
1616 
1617 static inline int task_on_rq_queued(struct task_struct *p)
1618 {
1619 	return p->on_rq == TASK_ON_RQ_QUEUED;
1620 }
1621 
1622 static inline int task_on_rq_migrating(struct task_struct *p)
1623 {
1624 	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1625 }
1626 
1627 /*
1628  * wake flags
1629  */
1630 #define WF_SYNC			0x01		/* Waker goes to sleep after wakeup */
1631 #define WF_FORK			0x02		/* Child wakeup after fork */
1632 #define WF_MIGRATED		0x4		/* Internal use, task got migrated */
1633 
1634 /*
1635  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1636  * of tasks with abnormal "nice" values across CPUs the contribution that
1637  * each task makes to its run queue's load is weighted according to its
1638  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1639  * scaled version of the new time slice allocation that they receive on time
1640  * slice expiry etc.
1641  */
1642 
1643 #define WEIGHT_IDLEPRIO		3
1644 #define WMULT_IDLEPRIO		1431655765
1645 
1646 extern const int		sched_prio_to_weight[40];
1647 extern const u32		sched_prio_to_wmult[40];
1648 
1649 /*
1650  * {de,en}queue flags:
1651  *
1652  * DEQUEUE_SLEEP  - task is no longer runnable
1653  * ENQUEUE_WAKEUP - task just became runnable
1654  *
1655  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1656  *                are in a known state which allows modification. Such pairs
1657  *                should preserve as much state as possible.
1658  *
1659  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1660  *        in the runqueue.
1661  *
1662  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
1663  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1664  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
1665  *
1666  */
1667 
1668 #define DEQUEUE_SLEEP		0x01
1669 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
1670 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
1671 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
1672 
1673 #define ENQUEUE_WAKEUP		0x01
1674 #define ENQUEUE_RESTORE		0x02
1675 #define ENQUEUE_MOVE		0x04
1676 #define ENQUEUE_NOCLOCK		0x08
1677 
1678 #define ENQUEUE_HEAD		0x10
1679 #define ENQUEUE_REPLENISH	0x20
1680 #ifdef CONFIG_SMP
1681 #define ENQUEUE_MIGRATED	0x40
1682 #else
1683 #define ENQUEUE_MIGRATED	0x00
1684 #endif
1685 
1686 #define RETRY_TASK		((void *)-1UL)
1687 
1688 struct sched_class {
1689 	const struct sched_class *next;
1690 
1691 #ifdef CONFIG_UCLAMP_TASK
1692 	int uclamp_enabled;
1693 #endif
1694 
1695 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1696 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1697 	void (*yield_task)   (struct rq *rq);
1698 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1699 
1700 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1701 
1702 	/*
1703 	 * It is the responsibility of the pick_next_task() method that will
1704 	 * return the next task to call put_prev_task() on the @prev task or
1705 	 * something equivalent.
1706 	 *
1707 	 * May return RETRY_TASK when it finds a higher prio class has runnable
1708 	 * tasks.
1709 	 */
1710 	struct task_struct * (*pick_next_task)(struct rq *rq,
1711 					       struct task_struct *prev,
1712 					       struct rq_flags *rf);
1713 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1714 
1715 #ifdef CONFIG_SMP
1716 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1717 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1718 
1719 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1720 
1721 	void (*set_cpus_allowed)(struct task_struct *p,
1722 				 const struct cpumask *newmask);
1723 
1724 	void (*rq_online)(struct rq *rq);
1725 	void (*rq_offline)(struct rq *rq);
1726 #endif
1727 
1728 	void (*set_curr_task)(struct rq *rq);
1729 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1730 	void (*task_fork)(struct task_struct *p);
1731 	void (*task_dead)(struct task_struct *p);
1732 
1733 	/*
1734 	 * The switched_from() call is allowed to drop rq->lock, therefore we
1735 	 * cannot assume the switched_from/switched_to pair is serliazed by
1736 	 * rq->lock. They are however serialized by p->pi_lock.
1737 	 */
1738 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1739 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
1740 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1741 			      int oldprio);
1742 
1743 	unsigned int (*get_rr_interval)(struct rq *rq,
1744 					struct task_struct *task);
1745 
1746 	void (*update_curr)(struct rq *rq);
1747 
1748 #define TASK_SET_GROUP		0
1749 #define TASK_MOVE_GROUP		1
1750 
1751 #ifdef CONFIG_FAIR_GROUP_SCHED
1752 	void (*task_change_group)(struct task_struct *p, int type);
1753 #endif
1754 };
1755 
1756 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1757 {
1758 	prev->sched_class->put_prev_task(rq, prev);
1759 }
1760 
1761 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1762 {
1763 	curr->sched_class->set_curr_task(rq);
1764 }
1765 
1766 #ifdef CONFIG_SMP
1767 #define sched_class_highest (&stop_sched_class)
1768 #else
1769 #define sched_class_highest (&dl_sched_class)
1770 #endif
1771 #define for_each_class(class) \
1772    for (class = sched_class_highest; class; class = class->next)
1773 
1774 extern const struct sched_class stop_sched_class;
1775 extern const struct sched_class dl_sched_class;
1776 extern const struct sched_class rt_sched_class;
1777 extern const struct sched_class fair_sched_class;
1778 extern const struct sched_class idle_sched_class;
1779 
1780 
1781 #ifdef CONFIG_SMP
1782 
1783 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1784 
1785 extern void trigger_load_balance(struct rq *rq);
1786 
1787 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1788 
1789 #endif
1790 
1791 #ifdef CONFIG_CPU_IDLE
1792 static inline void idle_set_state(struct rq *rq,
1793 				  struct cpuidle_state *idle_state)
1794 {
1795 	rq->idle_state = idle_state;
1796 }
1797 
1798 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1799 {
1800 	SCHED_WARN_ON(!rcu_read_lock_held());
1801 
1802 	return rq->idle_state;
1803 }
1804 #else
1805 static inline void idle_set_state(struct rq *rq,
1806 				  struct cpuidle_state *idle_state)
1807 {
1808 }
1809 
1810 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1811 {
1812 	return NULL;
1813 }
1814 #endif
1815 
1816 extern void schedule_idle(void);
1817 
1818 extern void sysrq_sched_debug_show(void);
1819 extern void sched_init_granularity(void);
1820 extern void update_max_interval(void);
1821 
1822 extern void init_sched_dl_class(void);
1823 extern void init_sched_rt_class(void);
1824 extern void init_sched_fair_class(void);
1825 
1826 extern void reweight_task(struct task_struct *p, int prio);
1827 
1828 extern void resched_curr(struct rq *rq);
1829 extern void resched_cpu(int cpu);
1830 
1831 extern struct rt_bandwidth def_rt_bandwidth;
1832 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1833 
1834 extern struct dl_bandwidth def_dl_bandwidth;
1835 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1836 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1837 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1838 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1839 
1840 #define BW_SHIFT		20
1841 #define BW_UNIT			(1 << BW_SHIFT)
1842 #define RATIO_SHIFT		8
1843 unsigned long to_ratio(u64 period, u64 runtime);
1844 
1845 extern void init_entity_runnable_average(struct sched_entity *se);
1846 extern void post_init_entity_util_avg(struct task_struct *p);
1847 
1848 #ifdef CONFIG_NO_HZ_FULL
1849 extern bool sched_can_stop_tick(struct rq *rq);
1850 extern int __init sched_tick_offload_init(void);
1851 
1852 /*
1853  * Tick may be needed by tasks in the runqueue depending on their policy and
1854  * requirements. If tick is needed, lets send the target an IPI to kick it out of
1855  * nohz mode if necessary.
1856  */
1857 static inline void sched_update_tick_dependency(struct rq *rq)
1858 {
1859 	int cpu;
1860 
1861 	if (!tick_nohz_full_enabled())
1862 		return;
1863 
1864 	cpu = cpu_of(rq);
1865 
1866 	if (!tick_nohz_full_cpu(cpu))
1867 		return;
1868 
1869 	if (sched_can_stop_tick(rq))
1870 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1871 	else
1872 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1873 }
1874 #else
1875 static inline int sched_tick_offload_init(void) { return 0; }
1876 static inline void sched_update_tick_dependency(struct rq *rq) { }
1877 #endif
1878 
1879 static inline void add_nr_running(struct rq *rq, unsigned count)
1880 {
1881 	unsigned prev_nr = rq->nr_running;
1882 
1883 	rq->nr_running = prev_nr + count;
1884 
1885 #ifdef CONFIG_SMP
1886 	if (prev_nr < 2 && rq->nr_running >= 2) {
1887 		if (!READ_ONCE(rq->rd->overload))
1888 			WRITE_ONCE(rq->rd->overload, 1);
1889 	}
1890 #endif
1891 
1892 	sched_update_tick_dependency(rq);
1893 }
1894 
1895 static inline void sub_nr_running(struct rq *rq, unsigned count)
1896 {
1897 	rq->nr_running -= count;
1898 	/* Check if we still need preemption */
1899 	sched_update_tick_dependency(rq);
1900 }
1901 
1902 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1903 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1904 
1905 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1906 
1907 extern const_debug unsigned int sysctl_sched_nr_migrate;
1908 extern const_debug unsigned int sysctl_sched_migration_cost;
1909 
1910 #ifdef CONFIG_SCHED_HRTICK
1911 
1912 /*
1913  * Use hrtick when:
1914  *  - enabled by features
1915  *  - hrtimer is actually high res
1916  */
1917 static inline int hrtick_enabled(struct rq *rq)
1918 {
1919 	if (!sched_feat(HRTICK))
1920 		return 0;
1921 	if (!cpu_active(cpu_of(rq)))
1922 		return 0;
1923 	return hrtimer_is_hres_active(&rq->hrtick_timer);
1924 }
1925 
1926 void hrtick_start(struct rq *rq, u64 delay);
1927 
1928 #else
1929 
1930 static inline int hrtick_enabled(struct rq *rq)
1931 {
1932 	return 0;
1933 }
1934 
1935 #endif /* CONFIG_SCHED_HRTICK */
1936 
1937 #ifndef arch_scale_freq_capacity
1938 static __always_inline
1939 unsigned long arch_scale_freq_capacity(int cpu)
1940 {
1941 	return SCHED_CAPACITY_SCALE;
1942 }
1943 #endif
1944 
1945 #ifdef CONFIG_SMP
1946 #ifdef CONFIG_PREEMPT
1947 
1948 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1949 
1950 /*
1951  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1952  * way at the expense of forcing extra atomic operations in all
1953  * invocations.  This assures that the double_lock is acquired using the
1954  * same underlying policy as the spinlock_t on this architecture, which
1955  * reduces latency compared to the unfair variant below.  However, it
1956  * also adds more overhead and therefore may reduce throughput.
1957  */
1958 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1959 	__releases(this_rq->lock)
1960 	__acquires(busiest->lock)
1961 	__acquires(this_rq->lock)
1962 {
1963 	raw_spin_unlock(&this_rq->lock);
1964 	double_rq_lock(this_rq, busiest);
1965 
1966 	return 1;
1967 }
1968 
1969 #else
1970 /*
1971  * Unfair double_lock_balance: Optimizes throughput at the expense of
1972  * latency by eliminating extra atomic operations when the locks are
1973  * already in proper order on entry.  This favors lower CPU-ids and will
1974  * grant the double lock to lower CPUs over higher ids under contention,
1975  * regardless of entry order into the function.
1976  */
1977 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1978 	__releases(this_rq->lock)
1979 	__acquires(busiest->lock)
1980 	__acquires(this_rq->lock)
1981 {
1982 	int ret = 0;
1983 
1984 	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1985 		if (busiest < this_rq) {
1986 			raw_spin_unlock(&this_rq->lock);
1987 			raw_spin_lock(&busiest->lock);
1988 			raw_spin_lock_nested(&this_rq->lock,
1989 					      SINGLE_DEPTH_NESTING);
1990 			ret = 1;
1991 		} else
1992 			raw_spin_lock_nested(&busiest->lock,
1993 					      SINGLE_DEPTH_NESTING);
1994 	}
1995 	return ret;
1996 }
1997 
1998 #endif /* CONFIG_PREEMPT */
1999 
2000 /*
2001  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2002  */
2003 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2004 {
2005 	if (unlikely(!irqs_disabled())) {
2006 		/* printk() doesn't work well under rq->lock */
2007 		raw_spin_unlock(&this_rq->lock);
2008 		BUG_ON(1);
2009 	}
2010 
2011 	return _double_lock_balance(this_rq, busiest);
2012 }
2013 
2014 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2015 	__releases(busiest->lock)
2016 {
2017 	raw_spin_unlock(&busiest->lock);
2018 	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2019 }
2020 
2021 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2022 {
2023 	if (l1 > l2)
2024 		swap(l1, l2);
2025 
2026 	spin_lock(l1);
2027 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2028 }
2029 
2030 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2031 {
2032 	if (l1 > l2)
2033 		swap(l1, l2);
2034 
2035 	spin_lock_irq(l1);
2036 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2037 }
2038 
2039 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2040 {
2041 	if (l1 > l2)
2042 		swap(l1, l2);
2043 
2044 	raw_spin_lock(l1);
2045 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2046 }
2047 
2048 /*
2049  * double_rq_lock - safely lock two runqueues
2050  *
2051  * Note this does not disable interrupts like task_rq_lock,
2052  * you need to do so manually before calling.
2053  */
2054 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2055 	__acquires(rq1->lock)
2056 	__acquires(rq2->lock)
2057 {
2058 	BUG_ON(!irqs_disabled());
2059 	if (rq1 == rq2) {
2060 		raw_spin_lock(&rq1->lock);
2061 		__acquire(rq2->lock);	/* Fake it out ;) */
2062 	} else {
2063 		if (rq1 < rq2) {
2064 			raw_spin_lock(&rq1->lock);
2065 			raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2066 		} else {
2067 			raw_spin_lock(&rq2->lock);
2068 			raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2069 		}
2070 	}
2071 }
2072 
2073 /*
2074  * double_rq_unlock - safely unlock two runqueues
2075  *
2076  * Note this does not restore interrupts like task_rq_unlock,
2077  * you need to do so manually after calling.
2078  */
2079 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2080 	__releases(rq1->lock)
2081 	__releases(rq2->lock)
2082 {
2083 	raw_spin_unlock(&rq1->lock);
2084 	if (rq1 != rq2)
2085 		raw_spin_unlock(&rq2->lock);
2086 	else
2087 		__release(rq2->lock);
2088 }
2089 
2090 extern void set_rq_online (struct rq *rq);
2091 extern void set_rq_offline(struct rq *rq);
2092 extern bool sched_smp_initialized;
2093 
2094 #else /* CONFIG_SMP */
2095 
2096 /*
2097  * double_rq_lock - safely lock two runqueues
2098  *
2099  * Note this does not disable interrupts like task_rq_lock,
2100  * you need to do so manually before calling.
2101  */
2102 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2103 	__acquires(rq1->lock)
2104 	__acquires(rq2->lock)
2105 {
2106 	BUG_ON(!irqs_disabled());
2107 	BUG_ON(rq1 != rq2);
2108 	raw_spin_lock(&rq1->lock);
2109 	__acquire(rq2->lock);	/* Fake it out ;) */
2110 }
2111 
2112 /*
2113  * double_rq_unlock - safely unlock two runqueues
2114  *
2115  * Note this does not restore interrupts like task_rq_unlock,
2116  * you need to do so manually after calling.
2117  */
2118 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2119 	__releases(rq1->lock)
2120 	__releases(rq2->lock)
2121 {
2122 	BUG_ON(rq1 != rq2);
2123 	raw_spin_unlock(&rq1->lock);
2124 	__release(rq2->lock);
2125 }
2126 
2127 #endif
2128 
2129 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2130 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2131 
2132 #ifdef	CONFIG_SCHED_DEBUG
2133 extern bool sched_debug_enabled;
2134 
2135 extern void print_cfs_stats(struct seq_file *m, int cpu);
2136 extern void print_rt_stats(struct seq_file *m, int cpu);
2137 extern void print_dl_stats(struct seq_file *m, int cpu);
2138 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2139 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2140 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2141 #ifdef CONFIG_NUMA_BALANCING
2142 extern void
2143 show_numa_stats(struct task_struct *p, struct seq_file *m);
2144 extern void
2145 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2146 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2147 #endif /* CONFIG_NUMA_BALANCING */
2148 #endif /* CONFIG_SCHED_DEBUG */
2149 
2150 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2151 extern void init_rt_rq(struct rt_rq *rt_rq);
2152 extern void init_dl_rq(struct dl_rq *dl_rq);
2153 
2154 extern void cfs_bandwidth_usage_inc(void);
2155 extern void cfs_bandwidth_usage_dec(void);
2156 
2157 #ifdef CONFIG_NO_HZ_COMMON
2158 #define NOHZ_BALANCE_KICK_BIT	0
2159 #define NOHZ_STATS_KICK_BIT	1
2160 
2161 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2162 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2163 
2164 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2165 
2166 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2167 
2168 extern void nohz_balance_exit_idle(struct rq *rq);
2169 #else
2170 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2171 #endif
2172 
2173 
2174 #ifdef CONFIG_SMP
2175 static inline
2176 void __dl_update(struct dl_bw *dl_b, s64 bw)
2177 {
2178 	struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2179 	int i;
2180 
2181 	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2182 			 "sched RCU must be held");
2183 	for_each_cpu_and(i, rd->span, cpu_active_mask) {
2184 		struct rq *rq = cpu_rq(i);
2185 
2186 		rq->dl.extra_bw += bw;
2187 	}
2188 }
2189 #else
2190 static inline
2191 void __dl_update(struct dl_bw *dl_b, s64 bw)
2192 {
2193 	struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2194 
2195 	dl->extra_bw += bw;
2196 }
2197 #endif
2198 
2199 
2200 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2201 struct irqtime {
2202 	u64			total;
2203 	u64			tick_delta;
2204 	u64			irq_start_time;
2205 	struct u64_stats_sync	sync;
2206 };
2207 
2208 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2209 
2210 /*
2211  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2212  * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2213  * and never move forward.
2214  */
2215 static inline u64 irq_time_read(int cpu)
2216 {
2217 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2218 	unsigned int seq;
2219 	u64 total;
2220 
2221 	do {
2222 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2223 		total = irqtime->total;
2224 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2225 
2226 	return total;
2227 }
2228 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2229 
2230 #ifdef CONFIG_CPU_FREQ
2231 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2232 
2233 /**
2234  * cpufreq_update_util - Take a note about CPU utilization changes.
2235  * @rq: Runqueue to carry out the update for.
2236  * @flags: Update reason flags.
2237  *
2238  * This function is called by the scheduler on the CPU whose utilization is
2239  * being updated.
2240  *
2241  * It can only be called from RCU-sched read-side critical sections.
2242  *
2243  * The way cpufreq is currently arranged requires it to evaluate the CPU
2244  * performance state (frequency/voltage) on a regular basis to prevent it from
2245  * being stuck in a completely inadequate performance level for too long.
2246  * That is not guaranteed to happen if the updates are only triggered from CFS
2247  * and DL, though, because they may not be coming in if only RT tasks are
2248  * active all the time (or there are RT tasks only).
2249  *
2250  * As a workaround for that issue, this function is called periodically by the
2251  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2252  * but that really is a band-aid.  Going forward it should be replaced with
2253  * solutions targeted more specifically at RT tasks.
2254  */
2255 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2256 {
2257 	struct update_util_data *data;
2258 
2259 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2260 						  cpu_of(rq)));
2261 	if (data)
2262 		data->func(data, rq_clock(rq), flags);
2263 }
2264 #else
2265 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2266 #endif /* CONFIG_CPU_FREQ */
2267 
2268 #ifdef CONFIG_UCLAMP_TASK
2269 unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id);
2270 
2271 static __always_inline
2272 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2273 			      struct task_struct *p)
2274 {
2275 	unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2276 	unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2277 
2278 	if (p) {
2279 		min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2280 		max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2281 	}
2282 
2283 	/*
2284 	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2285 	 * RUNNABLE tasks with _different_ clamps, we can end up with an
2286 	 * inversion. Fix it now when the clamps are applied.
2287 	 */
2288 	if (unlikely(min_util >= max_util))
2289 		return min_util;
2290 
2291 	return clamp(util, min_util, max_util);
2292 }
2293 
2294 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2295 {
2296 	return uclamp_util_with(rq, util, NULL);
2297 }
2298 #else /* CONFIG_UCLAMP_TASK */
2299 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2300 					    struct task_struct *p)
2301 {
2302 	return util;
2303 }
2304 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2305 {
2306 	return util;
2307 }
2308 #endif /* CONFIG_UCLAMP_TASK */
2309 
2310 #ifdef arch_scale_freq_capacity
2311 # ifndef arch_scale_freq_invariant
2312 #  define arch_scale_freq_invariant()	true
2313 # endif
2314 #else
2315 # define arch_scale_freq_invariant()	false
2316 #endif
2317 
2318 #ifdef CONFIG_SMP
2319 static inline unsigned long capacity_orig_of(int cpu)
2320 {
2321 	return cpu_rq(cpu)->cpu_capacity_orig;
2322 }
2323 #endif
2324 
2325 /**
2326  * enum schedutil_type - CPU utilization type
2327  * @FREQUENCY_UTIL:	Utilization used to select frequency
2328  * @ENERGY_UTIL:	Utilization used during energy calculation
2329  *
2330  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2331  * need to be aggregated differently depending on the usage made of them. This
2332  * enum is used within schedutil_freq_util() to differentiate the types of
2333  * utilization expected by the callers, and adjust the aggregation accordingly.
2334  */
2335 enum schedutil_type {
2336 	FREQUENCY_UTIL,
2337 	ENERGY_UTIL,
2338 };
2339 
2340 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2341 
2342 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2343 				 unsigned long max, enum schedutil_type type,
2344 				 struct task_struct *p);
2345 
2346 static inline unsigned long cpu_bw_dl(struct rq *rq)
2347 {
2348 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2349 }
2350 
2351 static inline unsigned long cpu_util_dl(struct rq *rq)
2352 {
2353 	return READ_ONCE(rq->avg_dl.util_avg);
2354 }
2355 
2356 static inline unsigned long cpu_util_cfs(struct rq *rq)
2357 {
2358 	unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2359 
2360 	if (sched_feat(UTIL_EST)) {
2361 		util = max_t(unsigned long, util,
2362 			     READ_ONCE(rq->cfs.avg.util_est.enqueued));
2363 	}
2364 
2365 	return util;
2366 }
2367 
2368 static inline unsigned long cpu_util_rt(struct rq *rq)
2369 {
2370 	return READ_ONCE(rq->avg_rt.util_avg);
2371 }
2372 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2373 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2374 				 unsigned long max, enum schedutil_type type,
2375 				 struct task_struct *p)
2376 {
2377 	return 0;
2378 }
2379 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2380 
2381 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2382 static inline unsigned long cpu_util_irq(struct rq *rq)
2383 {
2384 	return rq->avg_irq.util_avg;
2385 }
2386 
2387 static inline
2388 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2389 {
2390 	util *= (max - irq);
2391 	util /= max;
2392 
2393 	return util;
2394 
2395 }
2396 #else
2397 static inline unsigned long cpu_util_irq(struct rq *rq)
2398 {
2399 	return 0;
2400 }
2401 
2402 static inline
2403 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2404 {
2405 	return util;
2406 }
2407 #endif
2408 
2409 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2410 
2411 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2412 
2413 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2414 
2415 static inline bool sched_energy_enabled(void)
2416 {
2417 	return static_branch_unlikely(&sched_energy_present);
2418 }
2419 
2420 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2421 
2422 #define perf_domain_span(pd) NULL
2423 static inline bool sched_energy_enabled(void) { return false; }
2424 
2425 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2426