xref: /linux/kernel/sched/sched.h (revision 2363088eba2ecccfb643725e4864af73c4226a04)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Scheduler internal types and methods:
4  */
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
7 
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
23 
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/context_tracking.h>
31 #include <linux/cpufreq.h>
32 #include <linux/cpumask_api.h>
33 #include <linux/ctype.h>
34 #include <linux/file.h>
35 #include <linux/fs_api.h>
36 #include <linux/hrtimer_api.h>
37 #include <linux/interrupt.h>
38 #include <linux/irq_work.h>
39 #include <linux/jiffies.h>
40 #include <linux/kref_api.h>
41 #include <linux/kthread.h>
42 #include <linux/ktime_api.h>
43 #include <linux/lockdep_api.h>
44 #include <linux/lockdep.h>
45 #include <linux/minmax.h>
46 #include <linux/mm.h>
47 #include <linux/module.h>
48 #include <linux/mutex_api.h>
49 #include <linux/plist.h>
50 #include <linux/poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/psi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/seq_file.h>
56 #include <linux/seqlock.h>
57 #include <linux/softirq.h>
58 #include <linux/spinlock_api.h>
59 #include <linux/static_key.h>
60 #include <linux/stop_machine.h>
61 #include <linux/syscalls_api.h>
62 #include <linux/syscalls.h>
63 #include <linux/tick.h>
64 #include <linux/topology.h>
65 #include <linux/types.h>
66 #include <linux/u64_stats_sync_api.h>
67 #include <linux/uaccess.h>
68 #include <linux/wait_api.h>
69 #include <linux/wait_bit.h>
70 #include <linux/workqueue_api.h>
71 
72 #include <trace/events/power.h>
73 #include <trace/events/sched.h>
74 
75 #include "../workqueue_internal.h"
76 
77 #ifdef CONFIG_CGROUP_SCHED
78 #include <linux/cgroup.h>
79 #include <linux/psi.h>
80 #endif
81 
82 #ifdef CONFIG_SCHED_DEBUG
83 # include <linux/static_key.h>
84 #endif
85 
86 #ifdef CONFIG_PARAVIRT
87 # include <asm/paravirt.h>
88 # include <asm/paravirt_api_clock.h>
89 #endif
90 
91 #include "cpupri.h"
92 #include "cpudeadline.h"
93 
94 #ifdef CONFIG_SCHED_DEBUG
95 # define SCHED_WARN_ON(x)      WARN_ONCE(x, #x)
96 #else
97 # define SCHED_WARN_ON(x)      ({ (void)(x), 0; })
98 #endif
99 
100 struct rq;
101 struct cpuidle_state;
102 
103 /* task_struct::on_rq states: */
104 #define TASK_ON_RQ_QUEUED	1
105 #define TASK_ON_RQ_MIGRATING	2
106 
107 extern __read_mostly int scheduler_running;
108 
109 extern unsigned long calc_load_update;
110 extern atomic_long_t calc_load_tasks;
111 
112 extern unsigned int sysctl_sched_child_runs_first;
113 
114 extern void calc_global_load_tick(struct rq *this_rq);
115 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
116 
117 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
118 
119 extern unsigned int sysctl_sched_rt_period;
120 extern int sysctl_sched_rt_runtime;
121 extern int sched_rr_timeslice;
122 
123 /*
124  * Helpers for converting nanosecond timing to jiffy resolution
125  */
126 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
127 
128 /*
129  * Increase resolution of nice-level calculations for 64-bit architectures.
130  * The extra resolution improves shares distribution and load balancing of
131  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
132  * hierarchies, especially on larger systems. This is not a user-visible change
133  * and does not change the user-interface for setting shares/weights.
134  *
135  * We increase resolution only if we have enough bits to allow this increased
136  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
137  * are pretty high and the returns do not justify the increased costs.
138  *
139  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
140  * increase coverage and consistency always enable it on 64-bit platforms.
141  */
142 #ifdef CONFIG_64BIT
143 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
144 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
145 # define scale_load_down(w) \
146 ({ \
147 	unsigned long __w = (w); \
148 	if (__w) \
149 		__w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
150 	__w; \
151 })
152 #else
153 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
154 # define scale_load(w)		(w)
155 # define scale_load_down(w)	(w)
156 #endif
157 
158 /*
159  * Task weight (visible to users) and its load (invisible to users) have
160  * independent resolution, but they should be well calibrated. We use
161  * scale_load() and scale_load_down(w) to convert between them. The
162  * following must be true:
163  *
164  *  scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
165  *
166  */
167 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
168 
169 /*
170  * Single value that decides SCHED_DEADLINE internal math precision.
171  * 10 -> just above 1us
172  * 9  -> just above 0.5us
173  */
174 #define DL_SCALE		10
175 
176 /*
177  * Single value that denotes runtime == period, ie unlimited time.
178  */
179 #define RUNTIME_INF		((u64)~0ULL)
180 
181 static inline int idle_policy(int policy)
182 {
183 	return policy == SCHED_IDLE;
184 }
185 static inline int fair_policy(int policy)
186 {
187 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
188 }
189 
190 static inline int rt_policy(int policy)
191 {
192 	return policy == SCHED_FIFO || policy == SCHED_RR;
193 }
194 
195 static inline int dl_policy(int policy)
196 {
197 	return policy == SCHED_DEADLINE;
198 }
199 static inline bool valid_policy(int policy)
200 {
201 	return idle_policy(policy) || fair_policy(policy) ||
202 		rt_policy(policy) || dl_policy(policy);
203 }
204 
205 static inline int task_has_idle_policy(struct task_struct *p)
206 {
207 	return idle_policy(p->policy);
208 }
209 
210 static inline int task_has_rt_policy(struct task_struct *p)
211 {
212 	return rt_policy(p->policy);
213 }
214 
215 static inline int task_has_dl_policy(struct task_struct *p)
216 {
217 	return dl_policy(p->policy);
218 }
219 
220 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
221 
222 static inline void update_avg(u64 *avg, u64 sample)
223 {
224 	s64 diff = sample - *avg;
225 	*avg += diff / 8;
226 }
227 
228 /*
229  * Shifting a value by an exponent greater *or equal* to the size of said value
230  * is UB; cap at size-1.
231  */
232 #define shr_bound(val, shift)							\
233 	(val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
234 
235 /*
236  * !! For sched_setattr_nocheck() (kernel) only !!
237  *
238  * This is actually gross. :(
239  *
240  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
241  * tasks, but still be able to sleep. We need this on platforms that cannot
242  * atomically change clock frequency. Remove once fast switching will be
243  * available on such platforms.
244  *
245  * SUGOV stands for SchedUtil GOVernor.
246  */
247 #define SCHED_FLAG_SUGOV	0x10000000
248 
249 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
250 
251 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
252 {
253 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
254 	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
255 #else
256 	return false;
257 #endif
258 }
259 
260 /*
261  * Tells if entity @a should preempt entity @b.
262  */
263 static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
264 				     const struct sched_dl_entity *b)
265 {
266 	return dl_entity_is_special(a) ||
267 	       dl_time_before(a->deadline, b->deadline);
268 }
269 
270 /*
271  * This is the priority-queue data structure of the RT scheduling class:
272  */
273 struct rt_prio_array {
274 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
275 	struct list_head queue[MAX_RT_PRIO];
276 };
277 
278 struct rt_bandwidth {
279 	/* nests inside the rq lock: */
280 	raw_spinlock_t		rt_runtime_lock;
281 	ktime_t			rt_period;
282 	u64			rt_runtime;
283 	struct hrtimer		rt_period_timer;
284 	unsigned int		rt_period_active;
285 };
286 
287 void __dl_clear_params(struct task_struct *p);
288 
289 static inline int dl_bandwidth_enabled(void)
290 {
291 	return sysctl_sched_rt_runtime >= 0;
292 }
293 
294 /*
295  * To keep the bandwidth of -deadline tasks under control
296  * we need some place where:
297  *  - store the maximum -deadline bandwidth of each cpu;
298  *  - cache the fraction of bandwidth that is currently allocated in
299  *    each root domain;
300  *
301  * This is all done in the data structure below. It is similar to the
302  * one used for RT-throttling (rt_bandwidth), with the main difference
303  * that, since here we are only interested in admission control, we
304  * do not decrease any runtime while the group "executes", neither we
305  * need a timer to replenish it.
306  *
307  * With respect to SMP, bandwidth is given on a per root domain basis,
308  * meaning that:
309  *  - bw (< 100%) is the deadline bandwidth of each CPU;
310  *  - total_bw is the currently allocated bandwidth in each root domain;
311  */
312 struct dl_bw {
313 	raw_spinlock_t		lock;
314 	u64			bw;
315 	u64			total_bw;
316 };
317 
318 extern void init_dl_bw(struct dl_bw *dl_b);
319 extern int  sched_dl_global_validate(void);
320 extern void sched_dl_do_global(void);
321 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
322 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
323 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
324 extern bool __checkparam_dl(const struct sched_attr *attr);
325 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
326 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
327 extern int  dl_bw_check_overflow(int cpu);
328 
329 #ifdef CONFIG_CGROUP_SCHED
330 
331 struct cfs_rq;
332 struct rt_rq;
333 
334 extern struct list_head task_groups;
335 
336 struct cfs_bandwidth {
337 #ifdef CONFIG_CFS_BANDWIDTH
338 	raw_spinlock_t		lock;
339 	ktime_t			period;
340 	u64			quota;
341 	u64			runtime;
342 	u64			burst;
343 	u64			runtime_snap;
344 	s64			hierarchical_quota;
345 
346 	u8			idle;
347 	u8			period_active;
348 	u8			slack_started;
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 	int			nr_burst;
357 	u64			throttled_time;
358 	u64			burst_time;
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 	/* A positive value indicates that this is a SCHED_IDLE group. */
374 	int			idle;
375 
376 #ifdef	CONFIG_SMP
377 	/*
378 	 * load_avg can be heavily contended at clock tick time, so put
379 	 * it in its own cacheline separated from the fields above which
380 	 * will also be accessed at each tick.
381 	 */
382 	atomic_long_t		load_avg ____cacheline_aligned;
383 #endif
384 #endif
385 
386 #ifdef CONFIG_RT_GROUP_SCHED
387 	struct sched_rt_entity	**rt_se;
388 	struct rt_rq		**rt_rq;
389 
390 	struct rt_bandwidth	rt_bandwidth;
391 #endif
392 
393 	struct rcu_head		rcu;
394 	struct list_head	list;
395 
396 	struct task_group	*parent;
397 	struct list_head	siblings;
398 	struct list_head	children;
399 
400 #ifdef CONFIG_SCHED_AUTOGROUP
401 	struct autogroup	*autogroup;
402 #endif
403 
404 	struct cfs_bandwidth	cfs_bandwidth;
405 
406 #ifdef CONFIG_UCLAMP_TASK_GROUP
407 	/* The two decimal precision [%] value requested from user-space */
408 	unsigned int		uclamp_pct[UCLAMP_CNT];
409 	/* Clamp values requested for a task group */
410 	struct uclamp_se	uclamp_req[UCLAMP_CNT];
411 	/* Effective clamp values used for a task group */
412 	struct uclamp_se	uclamp[UCLAMP_CNT];
413 #endif
414 
415 };
416 
417 #ifdef CONFIG_FAIR_GROUP_SCHED
418 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
419 
420 /*
421  * A weight of 0 or 1 can cause arithmetics problems.
422  * A weight of a cfs_rq is the sum of weights of which entities
423  * are queued on this cfs_rq, so a weight of a entity should not be
424  * too large, so as the shares value of a task group.
425  * (The default weight is 1024 - so there's no practical
426  *  limitation from this.)
427  */
428 #define MIN_SHARES		(1UL <<  1)
429 #define MAX_SHARES		(1UL << 18)
430 #endif
431 
432 typedef int (*tg_visitor)(struct task_group *, void *);
433 
434 extern int walk_tg_tree_from(struct task_group *from,
435 			     tg_visitor down, tg_visitor up, void *data);
436 
437 /*
438  * Iterate the full tree, calling @down when first entering a node and @up when
439  * leaving it for the final time.
440  *
441  * Caller must hold rcu_lock or sufficient equivalent.
442  */
443 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
444 {
445 	return walk_tg_tree_from(&root_task_group, down, up, data);
446 }
447 
448 extern int tg_nop(struct task_group *tg, void *data);
449 
450 extern void free_fair_sched_group(struct task_group *tg);
451 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
452 extern void online_fair_sched_group(struct task_group *tg);
453 extern void unregister_fair_sched_group(struct task_group *tg);
454 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
455 			struct sched_entity *se, int cpu,
456 			struct sched_entity *parent);
457 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
458 
459 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
460 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
461 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
462 
463 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
464 		struct sched_rt_entity *rt_se, int cpu,
465 		struct sched_rt_entity *parent);
466 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
467 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
468 extern long sched_group_rt_runtime(struct task_group *tg);
469 extern long sched_group_rt_period(struct task_group *tg);
470 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
471 
472 extern struct task_group *sched_create_group(struct task_group *parent);
473 extern void sched_online_group(struct task_group *tg,
474 			       struct task_group *parent);
475 extern void sched_destroy_group(struct task_group *tg);
476 extern void sched_release_group(struct task_group *tg);
477 
478 extern void sched_move_task(struct task_struct *tsk);
479 
480 #ifdef CONFIG_FAIR_GROUP_SCHED
481 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
482 
483 extern int sched_group_set_idle(struct task_group *tg, long idle);
484 
485 #ifdef CONFIG_SMP
486 extern void set_task_rq_fair(struct sched_entity *se,
487 			     struct cfs_rq *prev, struct cfs_rq *next);
488 #else /* !CONFIG_SMP */
489 static inline void set_task_rq_fair(struct sched_entity *se,
490 			     struct cfs_rq *prev, struct cfs_rq *next) { }
491 #endif /* CONFIG_SMP */
492 #endif /* CONFIG_FAIR_GROUP_SCHED */
493 
494 #else /* CONFIG_CGROUP_SCHED */
495 
496 struct cfs_bandwidth { };
497 
498 #endif	/* CONFIG_CGROUP_SCHED */
499 
500 extern void unregister_rt_sched_group(struct task_group *tg);
501 extern void free_rt_sched_group(struct task_group *tg);
502 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
503 
504 /*
505  * u64_u32_load/u64_u32_store
506  *
507  * Use a copy of a u64 value to protect against data race. This is only
508  * applicable for 32-bits architectures.
509  */
510 #ifdef CONFIG_64BIT
511 # define u64_u32_load_copy(var, copy)       var
512 # define u64_u32_store_copy(var, copy, val) (var = val)
513 #else
514 # define u64_u32_load_copy(var, copy)					\
515 ({									\
516 	u64 __val, __val_copy;						\
517 	do {								\
518 		__val_copy = copy;					\
519 		/*							\
520 		 * paired with u64_u32_store_copy(), ordering access	\
521 		 * to var and copy.					\
522 		 */							\
523 		smp_rmb();						\
524 		__val = var;						\
525 	} while (__val != __val_copy);					\
526 	__val;								\
527 })
528 # define u64_u32_store_copy(var, copy, val)				\
529 do {									\
530 	typeof(val) __val = (val);					\
531 	var = __val;							\
532 	/*								\
533 	 * paired with u64_u32_load_copy(), ordering access to var and	\
534 	 * copy.							\
535 	 */								\
536 	smp_wmb();							\
537 	copy = __val;							\
538 } while (0)
539 #endif
540 # define u64_u32_load(var)      u64_u32_load_copy(var, var##_copy)
541 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
542 
543 /* CFS-related fields in a runqueue */
544 struct cfs_rq {
545 	struct load_weight	load;
546 	unsigned int		nr_running;
547 	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
548 	unsigned int		idle_nr_running;   /* SCHED_IDLE */
549 	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
550 
551 	u64			exec_clock;
552 	u64			min_vruntime;
553 #ifdef CONFIG_SCHED_CORE
554 	unsigned int		forceidle_seq;
555 	u64			min_vruntime_fi;
556 #endif
557 
558 #ifndef CONFIG_64BIT
559 	u64			min_vruntime_copy;
560 #endif
561 
562 	struct rb_root_cached	tasks_timeline;
563 
564 	/*
565 	 * 'curr' points to currently running entity on this cfs_rq.
566 	 * It is set to NULL otherwise (i.e when none are currently running).
567 	 */
568 	struct sched_entity	*curr;
569 	struct sched_entity	*next;
570 	struct sched_entity	*last;
571 	struct sched_entity	*skip;
572 
573 #ifdef	CONFIG_SCHED_DEBUG
574 	unsigned int		nr_spread_over;
575 #endif
576 
577 #ifdef CONFIG_SMP
578 	/*
579 	 * CFS load tracking
580 	 */
581 	struct sched_avg	avg;
582 #ifndef CONFIG_64BIT
583 	u64			last_update_time_copy;
584 #endif
585 	struct {
586 		raw_spinlock_t	lock ____cacheline_aligned;
587 		int		nr;
588 		unsigned long	load_avg;
589 		unsigned long	util_avg;
590 		unsigned long	runnable_avg;
591 	} removed;
592 
593 #ifdef CONFIG_FAIR_GROUP_SCHED
594 	unsigned long		tg_load_avg_contrib;
595 	long			propagate;
596 	long			prop_runnable_sum;
597 
598 	/*
599 	 *   h_load = weight * f(tg)
600 	 *
601 	 * Where f(tg) is the recursive weight fraction assigned to
602 	 * this group.
603 	 */
604 	unsigned long		h_load;
605 	u64			last_h_load_update;
606 	struct sched_entity	*h_load_next;
607 #endif /* CONFIG_FAIR_GROUP_SCHED */
608 #endif /* CONFIG_SMP */
609 
610 #ifdef CONFIG_FAIR_GROUP_SCHED
611 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
612 
613 	/*
614 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
615 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
616 	 * (like users, containers etc.)
617 	 *
618 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
619 	 * This list is used during load balance.
620 	 */
621 	int			on_list;
622 	struct list_head	leaf_cfs_rq_list;
623 	struct task_group	*tg;	/* group that "owns" this runqueue */
624 
625 	/* Locally cached copy of our task_group's idle value */
626 	int			idle;
627 
628 #ifdef CONFIG_CFS_BANDWIDTH
629 	int			runtime_enabled;
630 	s64			runtime_remaining;
631 
632 	u64			throttled_pelt_idle;
633 #ifndef CONFIG_64BIT
634 	u64                     throttled_pelt_idle_copy;
635 #endif
636 	u64			throttled_clock;
637 	u64			throttled_clock_pelt;
638 	u64			throttled_clock_pelt_time;
639 	int			throttled;
640 	int			throttle_count;
641 	struct list_head	throttled_list;
642 #ifdef CONFIG_SMP
643 	struct list_head	throttled_csd_list;
644 #endif
645 #endif /* CONFIG_CFS_BANDWIDTH */
646 #endif /* CONFIG_FAIR_GROUP_SCHED */
647 };
648 
649 static inline int rt_bandwidth_enabled(void)
650 {
651 	return sysctl_sched_rt_runtime >= 0;
652 }
653 
654 /* RT IPI pull logic requires IRQ_WORK */
655 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
656 # define HAVE_RT_PUSH_IPI
657 #endif
658 
659 /* Real-Time classes' related field in a runqueue: */
660 struct rt_rq {
661 	struct rt_prio_array	active;
662 	unsigned int		rt_nr_running;
663 	unsigned int		rr_nr_running;
664 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
665 	struct {
666 		int		curr; /* highest queued rt task prio */
667 #ifdef CONFIG_SMP
668 		int		next; /* next highest */
669 #endif
670 	} highest_prio;
671 #endif
672 #ifdef CONFIG_SMP
673 	unsigned int		rt_nr_migratory;
674 	unsigned int		rt_nr_total;
675 	int			overloaded;
676 	struct plist_head	pushable_tasks;
677 
678 #endif /* CONFIG_SMP */
679 	int			rt_queued;
680 
681 	int			rt_throttled;
682 	u64			rt_time;
683 	u64			rt_runtime;
684 	/* Nests inside the rq lock: */
685 	raw_spinlock_t		rt_runtime_lock;
686 
687 #ifdef CONFIG_RT_GROUP_SCHED
688 	unsigned int		rt_nr_boosted;
689 
690 	struct rq		*rq;
691 	struct task_group	*tg;
692 #endif
693 };
694 
695 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
696 {
697 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
698 }
699 
700 /* Deadline class' related fields in a runqueue */
701 struct dl_rq {
702 	/* runqueue is an rbtree, ordered by deadline */
703 	struct rb_root_cached	root;
704 
705 	unsigned int		dl_nr_running;
706 
707 #ifdef CONFIG_SMP
708 	/*
709 	 * Deadline values of the currently executing and the
710 	 * earliest ready task on this rq. Caching these facilitates
711 	 * the decision whether or not a ready but not running task
712 	 * should migrate somewhere else.
713 	 */
714 	struct {
715 		u64		curr;
716 		u64		next;
717 	} earliest_dl;
718 
719 	unsigned int		dl_nr_migratory;
720 	int			overloaded;
721 
722 	/*
723 	 * Tasks on this rq that can be pushed away. They are kept in
724 	 * an rb-tree, ordered by tasks' deadlines, with caching
725 	 * of the leftmost (earliest deadline) element.
726 	 */
727 	struct rb_root_cached	pushable_dl_tasks_root;
728 #else
729 	struct dl_bw		dl_bw;
730 #endif
731 	/*
732 	 * "Active utilization" for this runqueue: increased when a
733 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
734 	 * task blocks
735 	 */
736 	u64			running_bw;
737 
738 	/*
739 	 * Utilization of the tasks "assigned" to this runqueue (including
740 	 * the tasks that are in runqueue and the tasks that executed on this
741 	 * CPU and blocked). Increased when a task moves to this runqueue, and
742 	 * decreased when the task moves away (migrates, changes scheduling
743 	 * policy, or terminates).
744 	 * This is needed to compute the "inactive utilization" for the
745 	 * runqueue (inactive utilization = this_bw - running_bw).
746 	 */
747 	u64			this_bw;
748 	u64			extra_bw;
749 
750 	/*
751 	 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
752 	 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
753 	 */
754 	u64			max_bw;
755 
756 	/*
757 	 * Inverse of the fraction of CPU utilization that can be reclaimed
758 	 * by the GRUB algorithm.
759 	 */
760 	u64			bw_ratio;
761 };
762 
763 #ifdef CONFIG_FAIR_GROUP_SCHED
764 /* An entity is a task if it doesn't "own" a runqueue */
765 #define entity_is_task(se)	(!se->my_q)
766 
767 static inline void se_update_runnable(struct sched_entity *se)
768 {
769 	if (!entity_is_task(se))
770 		se->runnable_weight = se->my_q->h_nr_running;
771 }
772 
773 static inline long se_runnable(struct sched_entity *se)
774 {
775 	if (entity_is_task(se))
776 		return !!se->on_rq;
777 	else
778 		return se->runnable_weight;
779 }
780 
781 #else
782 #define entity_is_task(se)	1
783 
784 static inline void se_update_runnable(struct sched_entity *se) {}
785 
786 static inline long se_runnable(struct sched_entity *se)
787 {
788 	return !!se->on_rq;
789 }
790 #endif
791 
792 #ifdef CONFIG_SMP
793 /*
794  * XXX we want to get rid of these helpers and use the full load resolution.
795  */
796 static inline long se_weight(struct sched_entity *se)
797 {
798 	return scale_load_down(se->load.weight);
799 }
800 
801 
802 static inline bool sched_asym_prefer(int a, int b)
803 {
804 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
805 }
806 
807 struct perf_domain {
808 	struct em_perf_domain *em_pd;
809 	struct perf_domain *next;
810 	struct rcu_head rcu;
811 };
812 
813 /* Scheduling group status flags */
814 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
815 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
816 
817 /*
818  * We add the notion of a root-domain which will be used to define per-domain
819  * variables. Each exclusive cpuset essentially defines an island domain by
820  * fully partitioning the member CPUs from any other cpuset. Whenever a new
821  * exclusive cpuset is created, we also create and attach a new root-domain
822  * object.
823  *
824  */
825 struct root_domain {
826 	atomic_t		refcount;
827 	atomic_t		rto_count;
828 	struct rcu_head		rcu;
829 	cpumask_var_t		span;
830 	cpumask_var_t		online;
831 
832 	/*
833 	 * Indicate pullable load on at least one CPU, e.g:
834 	 * - More than one runnable task
835 	 * - Running task is misfit
836 	 */
837 	int			overload;
838 
839 	/* Indicate one or more cpus over-utilized (tipping point) */
840 	int			overutilized;
841 
842 	/*
843 	 * The bit corresponding to a CPU gets set here if such CPU has more
844 	 * than one runnable -deadline task (as it is below for RT tasks).
845 	 */
846 	cpumask_var_t		dlo_mask;
847 	atomic_t		dlo_count;
848 	struct dl_bw		dl_bw;
849 	struct cpudl		cpudl;
850 
851 	/*
852 	 * Indicate whether a root_domain's dl_bw has been checked or
853 	 * updated. It's monotonously increasing value.
854 	 *
855 	 * Also, some corner cases, like 'wrap around' is dangerous, but given
856 	 * that u64 is 'big enough'. So that shouldn't be a concern.
857 	 */
858 	u64 visit_gen;
859 
860 #ifdef HAVE_RT_PUSH_IPI
861 	/*
862 	 * For IPI pull requests, loop across the rto_mask.
863 	 */
864 	struct irq_work		rto_push_work;
865 	raw_spinlock_t		rto_lock;
866 	/* These are only updated and read within rto_lock */
867 	int			rto_loop;
868 	int			rto_cpu;
869 	/* These atomics are updated outside of a lock */
870 	atomic_t		rto_loop_next;
871 	atomic_t		rto_loop_start;
872 #endif
873 	/*
874 	 * The "RT overload" flag: it gets set if a CPU has more than
875 	 * one runnable RT task.
876 	 */
877 	cpumask_var_t		rto_mask;
878 	struct cpupri		cpupri;
879 
880 	unsigned long		max_cpu_capacity;
881 
882 	/*
883 	 * NULL-terminated list of performance domains intersecting with the
884 	 * CPUs of the rd. Protected by RCU.
885 	 */
886 	struct perf_domain __rcu *pd;
887 };
888 
889 extern void init_defrootdomain(void);
890 extern int sched_init_domains(const struct cpumask *cpu_map);
891 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
892 extern void sched_get_rd(struct root_domain *rd);
893 extern void sched_put_rd(struct root_domain *rd);
894 
895 #ifdef HAVE_RT_PUSH_IPI
896 extern void rto_push_irq_work_func(struct irq_work *work);
897 #endif
898 #endif /* CONFIG_SMP */
899 
900 #ifdef CONFIG_UCLAMP_TASK
901 /*
902  * struct uclamp_bucket - Utilization clamp bucket
903  * @value: utilization clamp value for tasks on this clamp bucket
904  * @tasks: number of RUNNABLE tasks on this clamp bucket
905  *
906  * Keep track of how many tasks are RUNNABLE for a given utilization
907  * clamp value.
908  */
909 struct uclamp_bucket {
910 	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
911 	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
912 };
913 
914 /*
915  * struct uclamp_rq - rq's utilization clamp
916  * @value: currently active clamp values for a rq
917  * @bucket: utilization clamp buckets affecting a rq
918  *
919  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
920  * A clamp value is affecting a rq when there is at least one task RUNNABLE
921  * (or actually running) with that value.
922  *
923  * There are up to UCLAMP_CNT possible different clamp values, currently there
924  * are only two: minimum utilization and maximum utilization.
925  *
926  * All utilization clamping values are MAX aggregated, since:
927  * - for util_min: we want to run the CPU at least at the max of the minimum
928  *   utilization required by its currently RUNNABLE tasks.
929  * - for util_max: we want to allow the CPU to run up to the max of the
930  *   maximum utilization allowed by its currently RUNNABLE tasks.
931  *
932  * Since on each system we expect only a limited number of different
933  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
934  * the metrics required to compute all the per-rq utilization clamp values.
935  */
936 struct uclamp_rq {
937 	unsigned int value;
938 	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
939 };
940 
941 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
942 #endif /* CONFIG_UCLAMP_TASK */
943 
944 struct rq;
945 struct balance_callback {
946 	struct balance_callback *next;
947 	void (*func)(struct rq *rq);
948 };
949 
950 /*
951  * This is the main, per-CPU runqueue data structure.
952  *
953  * Locking rule: those places that want to lock multiple runqueues
954  * (such as the load balancing or the thread migration code), lock
955  * acquire operations must be ordered by ascending &runqueue.
956  */
957 struct rq {
958 	/* runqueue lock: */
959 	raw_spinlock_t		__lock;
960 
961 	/*
962 	 * nr_running and cpu_load should be in the same cacheline because
963 	 * remote CPUs use both these fields when doing load calculation.
964 	 */
965 	unsigned int		nr_running;
966 #ifdef CONFIG_NUMA_BALANCING
967 	unsigned int		nr_numa_running;
968 	unsigned int		nr_preferred_running;
969 	unsigned int		numa_migrate_on;
970 #endif
971 #ifdef CONFIG_NO_HZ_COMMON
972 #ifdef CONFIG_SMP
973 	unsigned long		last_blocked_load_update_tick;
974 	unsigned int		has_blocked_load;
975 	call_single_data_t	nohz_csd;
976 #endif /* CONFIG_SMP */
977 	unsigned int		nohz_tick_stopped;
978 	atomic_t		nohz_flags;
979 #endif /* CONFIG_NO_HZ_COMMON */
980 
981 #ifdef CONFIG_SMP
982 	unsigned int		ttwu_pending;
983 #endif
984 	u64			nr_switches;
985 
986 #ifdef CONFIG_UCLAMP_TASK
987 	/* Utilization clamp values based on CPU's RUNNABLE tasks */
988 	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
989 	unsigned int		uclamp_flags;
990 #define UCLAMP_FLAG_IDLE 0x01
991 #endif
992 
993 	struct cfs_rq		cfs;
994 	struct rt_rq		rt;
995 	struct dl_rq		dl;
996 
997 #ifdef CONFIG_FAIR_GROUP_SCHED
998 	/* list of leaf cfs_rq on this CPU: */
999 	struct list_head	leaf_cfs_rq_list;
1000 	struct list_head	*tmp_alone_branch;
1001 #endif /* CONFIG_FAIR_GROUP_SCHED */
1002 
1003 	/*
1004 	 * This is part of a global counter where only the total sum
1005 	 * over all CPUs matters. A task can increase this counter on
1006 	 * one CPU and if it got migrated afterwards it may decrease
1007 	 * it on another CPU. Always updated under the runqueue lock:
1008 	 */
1009 	unsigned int		nr_uninterruptible;
1010 
1011 	struct task_struct __rcu	*curr;
1012 	struct task_struct	*idle;
1013 	struct task_struct	*stop;
1014 	unsigned long		next_balance;
1015 	struct mm_struct	*prev_mm;
1016 
1017 	unsigned int		clock_update_flags;
1018 	u64			clock;
1019 	/* Ensure that all clocks are in the same cache line */
1020 	u64			clock_task ____cacheline_aligned;
1021 	u64			clock_pelt;
1022 	unsigned long		lost_idle_time;
1023 	u64			clock_pelt_idle;
1024 	u64			clock_idle;
1025 #ifndef CONFIG_64BIT
1026 	u64			clock_pelt_idle_copy;
1027 	u64			clock_idle_copy;
1028 #endif
1029 
1030 	atomic_t		nr_iowait;
1031 
1032 #ifdef CONFIG_SCHED_DEBUG
1033 	u64 last_seen_need_resched_ns;
1034 	int ticks_without_resched;
1035 #endif
1036 
1037 #ifdef CONFIG_MEMBARRIER
1038 	int membarrier_state;
1039 #endif
1040 
1041 #ifdef CONFIG_SMP
1042 	struct root_domain		*rd;
1043 	struct sched_domain __rcu	*sd;
1044 
1045 	unsigned long		cpu_capacity;
1046 	unsigned long		cpu_capacity_orig;
1047 
1048 	struct balance_callback *balance_callback;
1049 
1050 	unsigned char		nohz_idle_balance;
1051 	unsigned char		idle_balance;
1052 
1053 	unsigned long		misfit_task_load;
1054 
1055 	/* For active balancing */
1056 	int			active_balance;
1057 	int			push_cpu;
1058 	struct cpu_stop_work	active_balance_work;
1059 
1060 	/* CPU of this runqueue: */
1061 	int			cpu;
1062 	int			online;
1063 
1064 	struct list_head cfs_tasks;
1065 
1066 	struct sched_avg	avg_rt;
1067 	struct sched_avg	avg_dl;
1068 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1069 	struct sched_avg	avg_irq;
1070 #endif
1071 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1072 	struct sched_avg	avg_thermal;
1073 #endif
1074 	u64			idle_stamp;
1075 	u64			avg_idle;
1076 
1077 	unsigned long		wake_stamp;
1078 	u64			wake_avg_idle;
1079 
1080 	/* This is used to determine avg_idle's max value */
1081 	u64			max_idle_balance_cost;
1082 
1083 #ifdef CONFIG_HOTPLUG_CPU
1084 	struct rcuwait		hotplug_wait;
1085 #endif
1086 #endif /* CONFIG_SMP */
1087 
1088 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1089 	u64			prev_irq_time;
1090 #endif
1091 #ifdef CONFIG_PARAVIRT
1092 	u64			prev_steal_time;
1093 #endif
1094 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1095 	u64			prev_steal_time_rq;
1096 #endif
1097 
1098 	/* calc_load related fields */
1099 	unsigned long		calc_load_update;
1100 	long			calc_load_active;
1101 
1102 #ifdef CONFIG_SCHED_HRTICK
1103 #ifdef CONFIG_SMP
1104 	call_single_data_t	hrtick_csd;
1105 #endif
1106 	struct hrtimer		hrtick_timer;
1107 	ktime_t 		hrtick_time;
1108 #endif
1109 
1110 #ifdef CONFIG_SCHEDSTATS
1111 	/* latency stats */
1112 	struct sched_info	rq_sched_info;
1113 	unsigned long long	rq_cpu_time;
1114 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1115 
1116 	/* sys_sched_yield() stats */
1117 	unsigned int		yld_count;
1118 
1119 	/* schedule() stats */
1120 	unsigned int		sched_count;
1121 	unsigned int		sched_goidle;
1122 
1123 	/* try_to_wake_up() stats */
1124 	unsigned int		ttwu_count;
1125 	unsigned int		ttwu_local;
1126 #endif
1127 
1128 #ifdef CONFIG_CPU_IDLE
1129 	/* Must be inspected within a rcu lock section */
1130 	struct cpuidle_state	*idle_state;
1131 #endif
1132 
1133 #ifdef CONFIG_SMP
1134 	unsigned int		nr_pinned;
1135 #endif
1136 	unsigned int		push_busy;
1137 	struct cpu_stop_work	push_work;
1138 
1139 #ifdef CONFIG_SCHED_CORE
1140 	/* per rq */
1141 	struct rq		*core;
1142 	struct task_struct	*core_pick;
1143 	unsigned int		core_enabled;
1144 	unsigned int		core_sched_seq;
1145 	struct rb_root		core_tree;
1146 
1147 	/* shared state -- careful with sched_core_cpu_deactivate() */
1148 	unsigned int		core_task_seq;
1149 	unsigned int		core_pick_seq;
1150 	unsigned long		core_cookie;
1151 	unsigned int		core_forceidle_count;
1152 	unsigned int		core_forceidle_seq;
1153 	unsigned int		core_forceidle_occupation;
1154 	u64			core_forceidle_start;
1155 #endif
1156 
1157 	/* Scratch cpumask to be temporarily used under rq_lock */
1158 	cpumask_var_t		scratch_mask;
1159 
1160 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1161 	call_single_data_t	cfsb_csd;
1162 	struct list_head	cfsb_csd_list;
1163 #endif
1164 };
1165 
1166 #ifdef CONFIG_FAIR_GROUP_SCHED
1167 
1168 /* CPU runqueue to which this cfs_rq is attached */
1169 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1170 {
1171 	return cfs_rq->rq;
1172 }
1173 
1174 #else
1175 
1176 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1177 {
1178 	return container_of(cfs_rq, struct rq, cfs);
1179 }
1180 #endif
1181 
1182 static inline int cpu_of(struct rq *rq)
1183 {
1184 #ifdef CONFIG_SMP
1185 	return rq->cpu;
1186 #else
1187 	return 0;
1188 #endif
1189 }
1190 
1191 #define MDF_PUSH	0x01
1192 
1193 static inline bool is_migration_disabled(struct task_struct *p)
1194 {
1195 #ifdef CONFIG_SMP
1196 	return p->migration_disabled;
1197 #else
1198 	return false;
1199 #endif
1200 }
1201 
1202 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1203 
1204 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1205 #define this_rq()		this_cpu_ptr(&runqueues)
1206 #define task_rq(p)		cpu_rq(task_cpu(p))
1207 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1208 #define raw_rq()		raw_cpu_ptr(&runqueues)
1209 
1210 struct sched_group;
1211 #ifdef CONFIG_SCHED_CORE
1212 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1213 
1214 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1215 
1216 static inline bool sched_core_enabled(struct rq *rq)
1217 {
1218 	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1219 }
1220 
1221 static inline bool sched_core_disabled(void)
1222 {
1223 	return !static_branch_unlikely(&__sched_core_enabled);
1224 }
1225 
1226 /*
1227  * Be careful with this function; not for general use. The return value isn't
1228  * stable unless you actually hold a relevant rq->__lock.
1229  */
1230 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1231 {
1232 	if (sched_core_enabled(rq))
1233 		return &rq->core->__lock;
1234 
1235 	return &rq->__lock;
1236 }
1237 
1238 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1239 {
1240 	if (rq->core_enabled)
1241 		return &rq->core->__lock;
1242 
1243 	return &rq->__lock;
1244 }
1245 
1246 bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1247 			bool fi);
1248 
1249 /*
1250  * Helpers to check if the CPU's core cookie matches with the task's cookie
1251  * when core scheduling is enabled.
1252  * A special case is that the task's cookie always matches with CPU's core
1253  * cookie if the CPU is in an idle core.
1254  */
1255 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1256 {
1257 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1258 	if (!sched_core_enabled(rq))
1259 		return true;
1260 
1261 	return rq->core->core_cookie == p->core_cookie;
1262 }
1263 
1264 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1265 {
1266 	bool idle_core = true;
1267 	int cpu;
1268 
1269 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1270 	if (!sched_core_enabled(rq))
1271 		return true;
1272 
1273 	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1274 		if (!available_idle_cpu(cpu)) {
1275 			idle_core = false;
1276 			break;
1277 		}
1278 	}
1279 
1280 	/*
1281 	 * A CPU in an idle core is always the best choice for tasks with
1282 	 * cookies.
1283 	 */
1284 	return idle_core || rq->core->core_cookie == p->core_cookie;
1285 }
1286 
1287 static inline bool sched_group_cookie_match(struct rq *rq,
1288 					    struct task_struct *p,
1289 					    struct sched_group *group)
1290 {
1291 	int cpu;
1292 
1293 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1294 	if (!sched_core_enabled(rq))
1295 		return true;
1296 
1297 	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1298 		if (sched_core_cookie_match(cpu_rq(cpu), p))
1299 			return true;
1300 	}
1301 	return false;
1302 }
1303 
1304 static inline bool sched_core_enqueued(struct task_struct *p)
1305 {
1306 	return !RB_EMPTY_NODE(&p->core_node);
1307 }
1308 
1309 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1310 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1311 
1312 extern void sched_core_get(void);
1313 extern void sched_core_put(void);
1314 
1315 #else /* !CONFIG_SCHED_CORE */
1316 
1317 static inline bool sched_core_enabled(struct rq *rq)
1318 {
1319 	return false;
1320 }
1321 
1322 static inline bool sched_core_disabled(void)
1323 {
1324 	return true;
1325 }
1326 
1327 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1328 {
1329 	return &rq->__lock;
1330 }
1331 
1332 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1333 {
1334 	return &rq->__lock;
1335 }
1336 
1337 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1338 {
1339 	return true;
1340 }
1341 
1342 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1343 {
1344 	return true;
1345 }
1346 
1347 static inline bool sched_group_cookie_match(struct rq *rq,
1348 					    struct task_struct *p,
1349 					    struct sched_group *group)
1350 {
1351 	return true;
1352 }
1353 #endif /* CONFIG_SCHED_CORE */
1354 
1355 static inline void lockdep_assert_rq_held(struct rq *rq)
1356 {
1357 	lockdep_assert_held(__rq_lockp(rq));
1358 }
1359 
1360 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1361 extern bool raw_spin_rq_trylock(struct rq *rq);
1362 extern void raw_spin_rq_unlock(struct rq *rq);
1363 
1364 static inline void raw_spin_rq_lock(struct rq *rq)
1365 {
1366 	raw_spin_rq_lock_nested(rq, 0);
1367 }
1368 
1369 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1370 {
1371 	local_irq_disable();
1372 	raw_spin_rq_lock(rq);
1373 }
1374 
1375 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1376 {
1377 	raw_spin_rq_unlock(rq);
1378 	local_irq_enable();
1379 }
1380 
1381 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1382 {
1383 	unsigned long flags;
1384 	local_irq_save(flags);
1385 	raw_spin_rq_lock(rq);
1386 	return flags;
1387 }
1388 
1389 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1390 {
1391 	raw_spin_rq_unlock(rq);
1392 	local_irq_restore(flags);
1393 }
1394 
1395 #define raw_spin_rq_lock_irqsave(rq, flags)	\
1396 do {						\
1397 	flags = _raw_spin_rq_lock_irqsave(rq);	\
1398 } while (0)
1399 
1400 #ifdef CONFIG_SCHED_SMT
1401 extern void __update_idle_core(struct rq *rq);
1402 
1403 static inline void update_idle_core(struct rq *rq)
1404 {
1405 	if (static_branch_unlikely(&sched_smt_present))
1406 		__update_idle_core(rq);
1407 }
1408 
1409 #else
1410 static inline void update_idle_core(struct rq *rq) { }
1411 #endif
1412 
1413 #ifdef CONFIG_FAIR_GROUP_SCHED
1414 static inline struct task_struct *task_of(struct sched_entity *se)
1415 {
1416 	SCHED_WARN_ON(!entity_is_task(se));
1417 	return container_of(se, struct task_struct, se);
1418 }
1419 
1420 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1421 {
1422 	return p->se.cfs_rq;
1423 }
1424 
1425 /* runqueue on which this entity is (to be) queued */
1426 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1427 {
1428 	return se->cfs_rq;
1429 }
1430 
1431 /* runqueue "owned" by this group */
1432 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1433 {
1434 	return grp->my_q;
1435 }
1436 
1437 #else
1438 
1439 #define task_of(_se)	container_of(_se, struct task_struct, se)
1440 
1441 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1442 {
1443 	return &task_rq(p)->cfs;
1444 }
1445 
1446 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1447 {
1448 	const struct task_struct *p = task_of(se);
1449 	struct rq *rq = task_rq(p);
1450 
1451 	return &rq->cfs;
1452 }
1453 
1454 /* runqueue "owned" by this group */
1455 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1456 {
1457 	return NULL;
1458 }
1459 #endif
1460 
1461 extern void update_rq_clock(struct rq *rq);
1462 
1463 /*
1464  * rq::clock_update_flags bits
1465  *
1466  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1467  *  call to __schedule(). This is an optimisation to avoid
1468  *  neighbouring rq clock updates.
1469  *
1470  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1471  *  in effect and calls to update_rq_clock() are being ignored.
1472  *
1473  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1474  *  made to update_rq_clock() since the last time rq::lock was pinned.
1475  *
1476  * If inside of __schedule(), clock_update_flags will have been
1477  * shifted left (a left shift is a cheap operation for the fast path
1478  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1479  *
1480  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1481  *
1482  * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1483  * one position though, because the next rq_unpin_lock() will shift it
1484  * back.
1485  */
1486 #define RQCF_REQ_SKIP		0x01
1487 #define RQCF_ACT_SKIP		0x02
1488 #define RQCF_UPDATED		0x04
1489 
1490 static inline void assert_clock_updated(struct rq *rq)
1491 {
1492 	/*
1493 	 * The only reason for not seeing a clock update since the
1494 	 * last rq_pin_lock() is if we're currently skipping updates.
1495 	 */
1496 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1497 }
1498 
1499 static inline u64 rq_clock(struct rq *rq)
1500 {
1501 	lockdep_assert_rq_held(rq);
1502 	assert_clock_updated(rq);
1503 
1504 	return rq->clock;
1505 }
1506 
1507 static inline u64 rq_clock_task(struct rq *rq)
1508 {
1509 	lockdep_assert_rq_held(rq);
1510 	assert_clock_updated(rq);
1511 
1512 	return rq->clock_task;
1513 }
1514 
1515 /**
1516  * By default the decay is the default pelt decay period.
1517  * The decay shift can change the decay period in
1518  * multiples of 32.
1519  *  Decay shift		Decay period(ms)
1520  *	0			32
1521  *	1			64
1522  *	2			128
1523  *	3			256
1524  *	4			512
1525  */
1526 extern int sched_thermal_decay_shift;
1527 
1528 static inline u64 rq_clock_thermal(struct rq *rq)
1529 {
1530 	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1531 }
1532 
1533 static inline void rq_clock_skip_update(struct rq *rq)
1534 {
1535 	lockdep_assert_rq_held(rq);
1536 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1537 }
1538 
1539 /*
1540  * See rt task throttling, which is the only time a skip
1541  * request is canceled.
1542  */
1543 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1544 {
1545 	lockdep_assert_rq_held(rq);
1546 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1547 }
1548 
1549 /*
1550  * During cpu offlining and rq wide unthrottling, we can trigger
1551  * an update_rq_clock() for several cfs and rt runqueues (Typically
1552  * when using list_for_each_entry_*)
1553  * rq_clock_start_loop_update() can be called after updating the clock
1554  * once and before iterating over the list to prevent multiple update.
1555  * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1556  * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1557  */
1558 static inline void rq_clock_start_loop_update(struct rq *rq)
1559 {
1560 	lockdep_assert_rq_held(rq);
1561 	SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1562 	rq->clock_update_flags |= RQCF_ACT_SKIP;
1563 }
1564 
1565 static inline void rq_clock_stop_loop_update(struct rq *rq)
1566 {
1567 	lockdep_assert_rq_held(rq);
1568 	rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1569 }
1570 
1571 struct rq_flags {
1572 	unsigned long flags;
1573 	struct pin_cookie cookie;
1574 #ifdef CONFIG_SCHED_DEBUG
1575 	/*
1576 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1577 	 * current pin context is stashed here in case it needs to be
1578 	 * restored in rq_repin_lock().
1579 	 */
1580 	unsigned int clock_update_flags;
1581 #endif
1582 };
1583 
1584 extern struct balance_callback balance_push_callback;
1585 
1586 /*
1587  * Lockdep annotation that avoids accidental unlocks; it's like a
1588  * sticky/continuous lockdep_assert_held().
1589  *
1590  * This avoids code that has access to 'struct rq *rq' (basically everything in
1591  * the scheduler) from accidentally unlocking the rq if they do not also have a
1592  * copy of the (on-stack) 'struct rq_flags rf'.
1593  *
1594  * Also see Documentation/locking/lockdep-design.rst.
1595  */
1596 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1597 {
1598 	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1599 
1600 #ifdef CONFIG_SCHED_DEBUG
1601 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1602 	rf->clock_update_flags = 0;
1603 #ifdef CONFIG_SMP
1604 	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1605 #endif
1606 #endif
1607 }
1608 
1609 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1610 {
1611 #ifdef CONFIG_SCHED_DEBUG
1612 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1613 		rf->clock_update_flags = RQCF_UPDATED;
1614 #endif
1615 
1616 	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1617 }
1618 
1619 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1620 {
1621 	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1622 
1623 #ifdef CONFIG_SCHED_DEBUG
1624 	/*
1625 	 * Restore the value we stashed in @rf for this pin context.
1626 	 */
1627 	rq->clock_update_flags |= rf->clock_update_flags;
1628 #endif
1629 }
1630 
1631 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1632 	__acquires(rq->lock);
1633 
1634 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1635 	__acquires(p->pi_lock)
1636 	__acquires(rq->lock);
1637 
1638 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1639 	__releases(rq->lock)
1640 {
1641 	rq_unpin_lock(rq, rf);
1642 	raw_spin_rq_unlock(rq);
1643 }
1644 
1645 static inline void
1646 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1647 	__releases(rq->lock)
1648 	__releases(p->pi_lock)
1649 {
1650 	rq_unpin_lock(rq, rf);
1651 	raw_spin_rq_unlock(rq);
1652 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1653 }
1654 
1655 static inline void
1656 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1657 	__acquires(rq->lock)
1658 {
1659 	raw_spin_rq_lock_irqsave(rq, rf->flags);
1660 	rq_pin_lock(rq, rf);
1661 }
1662 
1663 static inline void
1664 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1665 	__acquires(rq->lock)
1666 {
1667 	raw_spin_rq_lock_irq(rq);
1668 	rq_pin_lock(rq, rf);
1669 }
1670 
1671 static inline void
1672 rq_lock(struct rq *rq, struct rq_flags *rf)
1673 	__acquires(rq->lock)
1674 {
1675 	raw_spin_rq_lock(rq);
1676 	rq_pin_lock(rq, rf);
1677 }
1678 
1679 static inline void
1680 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1681 	__releases(rq->lock)
1682 {
1683 	rq_unpin_lock(rq, rf);
1684 	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1685 }
1686 
1687 static inline void
1688 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1689 	__releases(rq->lock)
1690 {
1691 	rq_unpin_lock(rq, rf);
1692 	raw_spin_rq_unlock_irq(rq);
1693 }
1694 
1695 static inline void
1696 rq_unlock(struct rq *rq, struct rq_flags *rf)
1697 	__releases(rq->lock)
1698 {
1699 	rq_unpin_lock(rq, rf);
1700 	raw_spin_rq_unlock(rq);
1701 }
1702 
1703 static inline struct rq *
1704 this_rq_lock_irq(struct rq_flags *rf)
1705 	__acquires(rq->lock)
1706 {
1707 	struct rq *rq;
1708 
1709 	local_irq_disable();
1710 	rq = this_rq();
1711 	rq_lock(rq, rf);
1712 	return rq;
1713 }
1714 
1715 #ifdef CONFIG_NUMA
1716 enum numa_topology_type {
1717 	NUMA_DIRECT,
1718 	NUMA_GLUELESS_MESH,
1719 	NUMA_BACKPLANE,
1720 };
1721 extern enum numa_topology_type sched_numa_topology_type;
1722 extern int sched_max_numa_distance;
1723 extern bool find_numa_distance(int distance);
1724 extern void sched_init_numa(int offline_node);
1725 extern void sched_update_numa(int cpu, bool online);
1726 extern void sched_domains_numa_masks_set(unsigned int cpu);
1727 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1728 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1729 #else
1730 static inline void sched_init_numa(int offline_node) { }
1731 static inline void sched_update_numa(int cpu, bool online) { }
1732 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1733 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1734 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1735 {
1736 	return nr_cpu_ids;
1737 }
1738 #endif
1739 
1740 #ifdef CONFIG_NUMA_BALANCING
1741 /* The regions in numa_faults array from task_struct */
1742 enum numa_faults_stats {
1743 	NUMA_MEM = 0,
1744 	NUMA_CPU,
1745 	NUMA_MEMBUF,
1746 	NUMA_CPUBUF
1747 };
1748 extern void sched_setnuma(struct task_struct *p, int node);
1749 extern int migrate_task_to(struct task_struct *p, int cpu);
1750 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1751 			int cpu, int scpu);
1752 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1753 #else
1754 static inline void
1755 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1756 {
1757 }
1758 #endif /* CONFIG_NUMA_BALANCING */
1759 
1760 #ifdef CONFIG_SMP
1761 
1762 static inline void
1763 queue_balance_callback(struct rq *rq,
1764 		       struct balance_callback *head,
1765 		       void (*func)(struct rq *rq))
1766 {
1767 	lockdep_assert_rq_held(rq);
1768 
1769 	/*
1770 	 * Don't (re)queue an already queued item; nor queue anything when
1771 	 * balance_push() is active, see the comment with
1772 	 * balance_push_callback.
1773 	 */
1774 	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1775 		return;
1776 
1777 	head->func = func;
1778 	head->next = rq->balance_callback;
1779 	rq->balance_callback = head;
1780 }
1781 
1782 #define rcu_dereference_check_sched_domain(p) \
1783 	rcu_dereference_check((p), \
1784 			      lockdep_is_held(&sched_domains_mutex))
1785 
1786 /*
1787  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1788  * See destroy_sched_domains: call_rcu for details.
1789  *
1790  * The domain tree of any CPU may only be accessed from within
1791  * preempt-disabled sections.
1792  */
1793 #define for_each_domain(cpu, __sd) \
1794 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1795 			__sd; __sd = __sd->parent)
1796 
1797 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1798 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1799 static const unsigned int SD_SHARED_CHILD_MASK =
1800 #include <linux/sched/sd_flags.h>
1801 0;
1802 #undef SD_FLAG
1803 
1804 /**
1805  * highest_flag_domain - Return highest sched_domain containing flag.
1806  * @cpu:	The CPU whose highest level of sched domain is to
1807  *		be returned.
1808  * @flag:	The flag to check for the highest sched_domain
1809  *		for the given CPU.
1810  *
1811  * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1812  * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1813  */
1814 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1815 {
1816 	struct sched_domain *sd, *hsd = NULL;
1817 
1818 	for_each_domain(cpu, sd) {
1819 		if (sd->flags & flag) {
1820 			hsd = sd;
1821 			continue;
1822 		}
1823 
1824 		/*
1825 		 * Stop the search if @flag is known to be shared at lower
1826 		 * levels. It will not be found further up.
1827 		 */
1828 		if (flag & SD_SHARED_CHILD_MASK)
1829 			break;
1830 	}
1831 
1832 	return hsd;
1833 }
1834 
1835 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1836 {
1837 	struct sched_domain *sd;
1838 
1839 	for_each_domain(cpu, sd) {
1840 		if (sd->flags & flag)
1841 			break;
1842 	}
1843 
1844 	return sd;
1845 }
1846 
1847 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1848 DECLARE_PER_CPU(int, sd_llc_size);
1849 DECLARE_PER_CPU(int, sd_llc_id);
1850 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1851 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1852 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1853 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1854 extern struct static_key_false sched_asym_cpucapacity;
1855 
1856 static __always_inline bool sched_asym_cpucap_active(void)
1857 {
1858 	return static_branch_unlikely(&sched_asym_cpucapacity);
1859 }
1860 
1861 struct sched_group_capacity {
1862 	atomic_t		ref;
1863 	/*
1864 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1865 	 * for a single CPU.
1866 	 */
1867 	unsigned long		capacity;
1868 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1869 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1870 	unsigned long		next_update;
1871 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1872 
1873 #ifdef CONFIG_SCHED_DEBUG
1874 	int			id;
1875 #endif
1876 
1877 	unsigned long		cpumask[];		/* Balance mask */
1878 };
1879 
1880 struct sched_group {
1881 	struct sched_group	*next;			/* Must be a circular list */
1882 	atomic_t		ref;
1883 
1884 	unsigned int		group_weight;
1885 	struct sched_group_capacity *sgc;
1886 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1887 	int			flags;
1888 
1889 	/*
1890 	 * The CPUs this group covers.
1891 	 *
1892 	 * NOTE: this field is variable length. (Allocated dynamically
1893 	 * by attaching extra space to the end of the structure,
1894 	 * depending on how many CPUs the kernel has booted up with)
1895 	 */
1896 	unsigned long		cpumask[];
1897 };
1898 
1899 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1900 {
1901 	return to_cpumask(sg->cpumask);
1902 }
1903 
1904 /*
1905  * See build_balance_mask().
1906  */
1907 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1908 {
1909 	return to_cpumask(sg->sgc->cpumask);
1910 }
1911 
1912 extern int group_balance_cpu(struct sched_group *sg);
1913 
1914 #ifdef CONFIG_SCHED_DEBUG
1915 void update_sched_domain_debugfs(void);
1916 void dirty_sched_domain_sysctl(int cpu);
1917 #else
1918 static inline void update_sched_domain_debugfs(void)
1919 {
1920 }
1921 static inline void dirty_sched_domain_sysctl(int cpu)
1922 {
1923 }
1924 #endif
1925 
1926 extern int sched_update_scaling(void);
1927 
1928 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1929 {
1930 	if (!p->user_cpus_ptr)
1931 		return cpu_possible_mask; /* &init_task.cpus_mask */
1932 	return p->user_cpus_ptr;
1933 }
1934 #endif /* CONFIG_SMP */
1935 
1936 #include "stats.h"
1937 
1938 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1939 
1940 extern void __sched_core_account_forceidle(struct rq *rq);
1941 
1942 static inline void sched_core_account_forceidle(struct rq *rq)
1943 {
1944 	if (schedstat_enabled())
1945 		__sched_core_account_forceidle(rq);
1946 }
1947 
1948 extern void __sched_core_tick(struct rq *rq);
1949 
1950 static inline void sched_core_tick(struct rq *rq)
1951 {
1952 	if (sched_core_enabled(rq) && schedstat_enabled())
1953 		__sched_core_tick(rq);
1954 }
1955 
1956 #else
1957 
1958 static inline void sched_core_account_forceidle(struct rq *rq) {}
1959 
1960 static inline void sched_core_tick(struct rq *rq) {}
1961 
1962 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1963 
1964 #ifdef CONFIG_CGROUP_SCHED
1965 
1966 /*
1967  * Return the group to which this tasks belongs.
1968  *
1969  * We cannot use task_css() and friends because the cgroup subsystem
1970  * changes that value before the cgroup_subsys::attach() method is called,
1971  * therefore we cannot pin it and might observe the wrong value.
1972  *
1973  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1974  * core changes this before calling sched_move_task().
1975  *
1976  * Instead we use a 'copy' which is updated from sched_move_task() while
1977  * holding both task_struct::pi_lock and rq::lock.
1978  */
1979 static inline struct task_group *task_group(struct task_struct *p)
1980 {
1981 	return p->sched_task_group;
1982 }
1983 
1984 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1985 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1986 {
1987 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1988 	struct task_group *tg = task_group(p);
1989 #endif
1990 
1991 #ifdef CONFIG_FAIR_GROUP_SCHED
1992 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1993 	p->se.cfs_rq = tg->cfs_rq[cpu];
1994 	p->se.parent = tg->se[cpu];
1995 	p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
1996 #endif
1997 
1998 #ifdef CONFIG_RT_GROUP_SCHED
1999 	p->rt.rt_rq  = tg->rt_rq[cpu];
2000 	p->rt.parent = tg->rt_se[cpu];
2001 #endif
2002 }
2003 
2004 #else /* CONFIG_CGROUP_SCHED */
2005 
2006 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2007 static inline struct task_group *task_group(struct task_struct *p)
2008 {
2009 	return NULL;
2010 }
2011 
2012 #endif /* CONFIG_CGROUP_SCHED */
2013 
2014 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2015 {
2016 	set_task_rq(p, cpu);
2017 #ifdef CONFIG_SMP
2018 	/*
2019 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2020 	 * successfully executed on another CPU. We must ensure that updates of
2021 	 * per-task data have been completed by this moment.
2022 	 */
2023 	smp_wmb();
2024 	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2025 	p->wake_cpu = cpu;
2026 #endif
2027 }
2028 
2029 /*
2030  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2031  */
2032 #ifdef CONFIG_SCHED_DEBUG
2033 # define const_debug __read_mostly
2034 #else
2035 # define const_debug const
2036 #endif
2037 
2038 #define SCHED_FEAT(name, enabled)	\
2039 	__SCHED_FEAT_##name ,
2040 
2041 enum {
2042 #include "features.h"
2043 	__SCHED_FEAT_NR,
2044 };
2045 
2046 #undef SCHED_FEAT
2047 
2048 #ifdef CONFIG_SCHED_DEBUG
2049 
2050 /*
2051  * To support run-time toggling of sched features, all the translation units
2052  * (but core.c) reference the sysctl_sched_features defined in core.c.
2053  */
2054 extern const_debug unsigned int sysctl_sched_features;
2055 
2056 #ifdef CONFIG_JUMP_LABEL
2057 #define SCHED_FEAT(name, enabled)					\
2058 static __always_inline bool static_branch_##name(struct static_key *key) \
2059 {									\
2060 	return static_key_##enabled(key);				\
2061 }
2062 
2063 #include "features.h"
2064 #undef SCHED_FEAT
2065 
2066 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2067 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2068 
2069 #else /* !CONFIG_JUMP_LABEL */
2070 
2071 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2072 
2073 #endif /* CONFIG_JUMP_LABEL */
2074 
2075 #else /* !SCHED_DEBUG */
2076 
2077 /*
2078  * Each translation unit has its own copy of sysctl_sched_features to allow
2079  * constants propagation at compile time and compiler optimization based on
2080  * features default.
2081  */
2082 #define SCHED_FEAT(name, enabled)	\
2083 	(1UL << __SCHED_FEAT_##name) * enabled |
2084 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2085 #include "features.h"
2086 	0;
2087 #undef SCHED_FEAT
2088 
2089 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2090 
2091 #endif /* SCHED_DEBUG */
2092 
2093 extern struct static_key_false sched_numa_balancing;
2094 extern struct static_key_false sched_schedstats;
2095 
2096 static inline u64 global_rt_period(void)
2097 {
2098 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2099 }
2100 
2101 static inline u64 global_rt_runtime(void)
2102 {
2103 	if (sysctl_sched_rt_runtime < 0)
2104 		return RUNTIME_INF;
2105 
2106 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2107 }
2108 
2109 static inline int task_current(struct rq *rq, struct task_struct *p)
2110 {
2111 	return rq->curr == p;
2112 }
2113 
2114 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2115 {
2116 #ifdef CONFIG_SMP
2117 	return p->on_cpu;
2118 #else
2119 	return task_current(rq, p);
2120 #endif
2121 }
2122 
2123 static inline int task_on_rq_queued(struct task_struct *p)
2124 {
2125 	return p->on_rq == TASK_ON_RQ_QUEUED;
2126 }
2127 
2128 static inline int task_on_rq_migrating(struct task_struct *p)
2129 {
2130 	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2131 }
2132 
2133 /* Wake flags. The first three directly map to some SD flag value */
2134 #define WF_EXEC     0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2135 #define WF_FORK     0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2136 #define WF_TTWU     0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2137 
2138 #define WF_SYNC     0x10 /* Waker goes to sleep after wakeup */
2139 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2140 
2141 #ifdef CONFIG_SMP
2142 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2143 static_assert(WF_FORK == SD_BALANCE_FORK);
2144 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2145 #endif
2146 
2147 /*
2148  * To aid in avoiding the subversion of "niceness" due to uneven distribution
2149  * of tasks with abnormal "nice" values across CPUs the contribution that
2150  * each task makes to its run queue's load is weighted according to its
2151  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2152  * scaled version of the new time slice allocation that they receive on time
2153  * slice expiry etc.
2154  */
2155 
2156 #define WEIGHT_IDLEPRIO		3
2157 #define WMULT_IDLEPRIO		1431655765
2158 
2159 extern const int		sched_prio_to_weight[40];
2160 extern const u32		sched_prio_to_wmult[40];
2161 
2162 /*
2163  * {de,en}queue flags:
2164  *
2165  * DEQUEUE_SLEEP  - task is no longer runnable
2166  * ENQUEUE_WAKEUP - task just became runnable
2167  *
2168  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2169  *                are in a known state which allows modification. Such pairs
2170  *                should preserve as much state as possible.
2171  *
2172  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2173  *        in the runqueue.
2174  *
2175  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2176  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2177  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2178  *
2179  */
2180 
2181 #define DEQUEUE_SLEEP		0x01
2182 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2183 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2184 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
2185 
2186 #define ENQUEUE_WAKEUP		0x01
2187 #define ENQUEUE_RESTORE		0x02
2188 #define ENQUEUE_MOVE		0x04
2189 #define ENQUEUE_NOCLOCK		0x08
2190 
2191 #define ENQUEUE_HEAD		0x10
2192 #define ENQUEUE_REPLENISH	0x20
2193 #ifdef CONFIG_SMP
2194 #define ENQUEUE_MIGRATED	0x40
2195 #else
2196 #define ENQUEUE_MIGRATED	0x00
2197 #endif
2198 
2199 #define RETRY_TASK		((void *)-1UL)
2200 
2201 struct affinity_context {
2202 	const struct cpumask *new_mask;
2203 	struct cpumask *user_mask;
2204 	unsigned int flags;
2205 };
2206 
2207 struct sched_class {
2208 
2209 #ifdef CONFIG_UCLAMP_TASK
2210 	int uclamp_enabled;
2211 #endif
2212 
2213 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2214 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2215 	void (*yield_task)   (struct rq *rq);
2216 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2217 
2218 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2219 
2220 	struct task_struct *(*pick_next_task)(struct rq *rq);
2221 
2222 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2223 	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2224 
2225 #ifdef CONFIG_SMP
2226 	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2227 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2228 
2229 	struct task_struct * (*pick_task)(struct rq *rq);
2230 
2231 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2232 
2233 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2234 
2235 	void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2236 
2237 	void (*rq_online)(struct rq *rq);
2238 	void (*rq_offline)(struct rq *rq);
2239 
2240 	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2241 #endif
2242 
2243 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2244 	void (*task_fork)(struct task_struct *p);
2245 	void (*task_dead)(struct task_struct *p);
2246 
2247 	/*
2248 	 * The switched_from() call is allowed to drop rq->lock, therefore we
2249 	 * cannot assume the switched_from/switched_to pair is serialized by
2250 	 * rq->lock. They are however serialized by p->pi_lock.
2251 	 */
2252 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2253 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2254 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2255 			      int oldprio);
2256 
2257 	unsigned int (*get_rr_interval)(struct rq *rq,
2258 					struct task_struct *task);
2259 
2260 	void (*update_curr)(struct rq *rq);
2261 
2262 #ifdef CONFIG_FAIR_GROUP_SCHED
2263 	void (*task_change_group)(struct task_struct *p);
2264 #endif
2265 
2266 #ifdef CONFIG_SCHED_CORE
2267 	int (*task_is_throttled)(struct task_struct *p, int cpu);
2268 #endif
2269 };
2270 
2271 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2272 {
2273 	WARN_ON_ONCE(rq->curr != prev);
2274 	prev->sched_class->put_prev_task(rq, prev);
2275 }
2276 
2277 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2278 {
2279 	next->sched_class->set_next_task(rq, next, false);
2280 }
2281 
2282 
2283 /*
2284  * Helper to define a sched_class instance; each one is placed in a separate
2285  * section which is ordered by the linker script:
2286  *
2287  *   include/asm-generic/vmlinux.lds.h
2288  *
2289  * *CAREFUL* they are laid out in *REVERSE* order!!!
2290  *
2291  * Also enforce alignment on the instance, not the type, to guarantee layout.
2292  */
2293 #define DEFINE_SCHED_CLASS(name) \
2294 const struct sched_class name##_sched_class \
2295 	__aligned(__alignof__(struct sched_class)) \
2296 	__section("__" #name "_sched_class")
2297 
2298 /* Defined in include/asm-generic/vmlinux.lds.h */
2299 extern struct sched_class __sched_class_highest[];
2300 extern struct sched_class __sched_class_lowest[];
2301 
2302 #define for_class_range(class, _from, _to) \
2303 	for (class = (_from); class < (_to); class++)
2304 
2305 #define for_each_class(class) \
2306 	for_class_range(class, __sched_class_highest, __sched_class_lowest)
2307 
2308 #define sched_class_above(_a, _b)	((_a) < (_b))
2309 
2310 extern const struct sched_class stop_sched_class;
2311 extern const struct sched_class dl_sched_class;
2312 extern const struct sched_class rt_sched_class;
2313 extern const struct sched_class fair_sched_class;
2314 extern const struct sched_class idle_sched_class;
2315 
2316 static inline bool sched_stop_runnable(struct rq *rq)
2317 {
2318 	return rq->stop && task_on_rq_queued(rq->stop);
2319 }
2320 
2321 static inline bool sched_dl_runnable(struct rq *rq)
2322 {
2323 	return rq->dl.dl_nr_running > 0;
2324 }
2325 
2326 static inline bool sched_rt_runnable(struct rq *rq)
2327 {
2328 	return rq->rt.rt_queued > 0;
2329 }
2330 
2331 static inline bool sched_fair_runnable(struct rq *rq)
2332 {
2333 	return rq->cfs.nr_running > 0;
2334 }
2335 
2336 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2337 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2338 
2339 #define SCA_CHECK		0x01
2340 #define SCA_MIGRATE_DISABLE	0x02
2341 #define SCA_MIGRATE_ENABLE	0x04
2342 #define SCA_USER		0x08
2343 
2344 #ifdef CONFIG_SMP
2345 
2346 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2347 
2348 extern void trigger_load_balance(struct rq *rq);
2349 
2350 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2351 
2352 static inline struct task_struct *get_push_task(struct rq *rq)
2353 {
2354 	struct task_struct *p = rq->curr;
2355 
2356 	lockdep_assert_rq_held(rq);
2357 
2358 	if (rq->push_busy)
2359 		return NULL;
2360 
2361 	if (p->nr_cpus_allowed == 1)
2362 		return NULL;
2363 
2364 	if (p->migration_disabled)
2365 		return NULL;
2366 
2367 	rq->push_busy = true;
2368 	return get_task_struct(p);
2369 }
2370 
2371 extern int push_cpu_stop(void *arg);
2372 
2373 #endif
2374 
2375 #ifdef CONFIG_CPU_IDLE
2376 static inline void idle_set_state(struct rq *rq,
2377 				  struct cpuidle_state *idle_state)
2378 {
2379 	rq->idle_state = idle_state;
2380 }
2381 
2382 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2383 {
2384 	SCHED_WARN_ON(!rcu_read_lock_held());
2385 
2386 	return rq->idle_state;
2387 }
2388 #else
2389 static inline void idle_set_state(struct rq *rq,
2390 				  struct cpuidle_state *idle_state)
2391 {
2392 }
2393 
2394 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2395 {
2396 	return NULL;
2397 }
2398 #endif
2399 
2400 extern void schedule_idle(void);
2401 
2402 extern void sysrq_sched_debug_show(void);
2403 extern void sched_init_granularity(void);
2404 extern void update_max_interval(void);
2405 
2406 extern void init_sched_dl_class(void);
2407 extern void init_sched_rt_class(void);
2408 extern void init_sched_fair_class(void);
2409 
2410 extern void reweight_task(struct task_struct *p, int prio);
2411 
2412 extern void resched_curr(struct rq *rq);
2413 extern void resched_cpu(int cpu);
2414 
2415 extern struct rt_bandwidth def_rt_bandwidth;
2416 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2417 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2418 
2419 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2420 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2421 
2422 #define BW_SHIFT		20
2423 #define BW_UNIT			(1 << BW_SHIFT)
2424 #define RATIO_SHIFT		8
2425 #define MAX_BW_BITS		(64 - BW_SHIFT)
2426 #define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2427 unsigned long to_ratio(u64 period, u64 runtime);
2428 
2429 extern void init_entity_runnable_average(struct sched_entity *se);
2430 extern void post_init_entity_util_avg(struct task_struct *p);
2431 
2432 #ifdef CONFIG_NO_HZ_FULL
2433 extern bool sched_can_stop_tick(struct rq *rq);
2434 extern int __init sched_tick_offload_init(void);
2435 
2436 /*
2437  * Tick may be needed by tasks in the runqueue depending on their policy and
2438  * requirements. If tick is needed, lets send the target an IPI to kick it out of
2439  * nohz mode if necessary.
2440  */
2441 static inline void sched_update_tick_dependency(struct rq *rq)
2442 {
2443 	int cpu = cpu_of(rq);
2444 
2445 	if (!tick_nohz_full_cpu(cpu))
2446 		return;
2447 
2448 	if (sched_can_stop_tick(rq))
2449 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2450 	else
2451 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2452 }
2453 #else
2454 static inline int sched_tick_offload_init(void) { return 0; }
2455 static inline void sched_update_tick_dependency(struct rq *rq) { }
2456 #endif
2457 
2458 static inline void add_nr_running(struct rq *rq, unsigned count)
2459 {
2460 	unsigned prev_nr = rq->nr_running;
2461 
2462 	rq->nr_running = prev_nr + count;
2463 	if (trace_sched_update_nr_running_tp_enabled()) {
2464 		call_trace_sched_update_nr_running(rq, count);
2465 	}
2466 
2467 #ifdef CONFIG_SMP
2468 	if (prev_nr < 2 && rq->nr_running >= 2) {
2469 		if (!READ_ONCE(rq->rd->overload))
2470 			WRITE_ONCE(rq->rd->overload, 1);
2471 	}
2472 #endif
2473 
2474 	sched_update_tick_dependency(rq);
2475 }
2476 
2477 static inline void sub_nr_running(struct rq *rq, unsigned count)
2478 {
2479 	rq->nr_running -= count;
2480 	if (trace_sched_update_nr_running_tp_enabled()) {
2481 		call_trace_sched_update_nr_running(rq, -count);
2482 	}
2483 
2484 	/* Check if we still need preemption */
2485 	sched_update_tick_dependency(rq);
2486 }
2487 
2488 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2489 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2490 
2491 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2492 
2493 #ifdef CONFIG_PREEMPT_RT
2494 #define SCHED_NR_MIGRATE_BREAK 8
2495 #else
2496 #define SCHED_NR_MIGRATE_BREAK 32
2497 #endif
2498 
2499 extern const_debug unsigned int sysctl_sched_nr_migrate;
2500 extern const_debug unsigned int sysctl_sched_migration_cost;
2501 
2502 #ifdef CONFIG_SCHED_DEBUG
2503 extern unsigned int sysctl_sched_latency;
2504 extern unsigned int sysctl_sched_min_granularity;
2505 extern unsigned int sysctl_sched_idle_min_granularity;
2506 extern unsigned int sysctl_sched_wakeup_granularity;
2507 extern int sysctl_resched_latency_warn_ms;
2508 extern int sysctl_resched_latency_warn_once;
2509 
2510 extern unsigned int sysctl_sched_tunable_scaling;
2511 
2512 extern unsigned int sysctl_numa_balancing_scan_delay;
2513 extern unsigned int sysctl_numa_balancing_scan_period_min;
2514 extern unsigned int sysctl_numa_balancing_scan_period_max;
2515 extern unsigned int sysctl_numa_balancing_scan_size;
2516 extern unsigned int sysctl_numa_balancing_hot_threshold;
2517 #endif
2518 
2519 #ifdef CONFIG_SCHED_HRTICK
2520 
2521 /*
2522  * Use hrtick when:
2523  *  - enabled by features
2524  *  - hrtimer is actually high res
2525  */
2526 static inline int hrtick_enabled(struct rq *rq)
2527 {
2528 	if (!cpu_active(cpu_of(rq)))
2529 		return 0;
2530 	return hrtimer_is_hres_active(&rq->hrtick_timer);
2531 }
2532 
2533 static inline int hrtick_enabled_fair(struct rq *rq)
2534 {
2535 	if (!sched_feat(HRTICK))
2536 		return 0;
2537 	return hrtick_enabled(rq);
2538 }
2539 
2540 static inline int hrtick_enabled_dl(struct rq *rq)
2541 {
2542 	if (!sched_feat(HRTICK_DL))
2543 		return 0;
2544 	return hrtick_enabled(rq);
2545 }
2546 
2547 void hrtick_start(struct rq *rq, u64 delay);
2548 
2549 #else
2550 
2551 static inline int hrtick_enabled_fair(struct rq *rq)
2552 {
2553 	return 0;
2554 }
2555 
2556 static inline int hrtick_enabled_dl(struct rq *rq)
2557 {
2558 	return 0;
2559 }
2560 
2561 static inline int hrtick_enabled(struct rq *rq)
2562 {
2563 	return 0;
2564 }
2565 
2566 #endif /* CONFIG_SCHED_HRTICK */
2567 
2568 #ifndef arch_scale_freq_tick
2569 static __always_inline
2570 void arch_scale_freq_tick(void)
2571 {
2572 }
2573 #endif
2574 
2575 #ifndef arch_scale_freq_capacity
2576 /**
2577  * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2578  * @cpu: the CPU in question.
2579  *
2580  * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2581  *
2582  *     f_curr
2583  *     ------ * SCHED_CAPACITY_SCALE
2584  *     f_max
2585  */
2586 static __always_inline
2587 unsigned long arch_scale_freq_capacity(int cpu)
2588 {
2589 	return SCHED_CAPACITY_SCALE;
2590 }
2591 #endif
2592 
2593 #ifdef CONFIG_SCHED_DEBUG
2594 /*
2595  * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2596  * acquire rq lock instead of rq_lock(). So at the end of these two functions
2597  * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2598  * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2599  */
2600 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2601 {
2602 	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2603 	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2604 #ifdef CONFIG_SMP
2605 	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2606 #endif
2607 }
2608 #else
2609 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2610 #endif
2611 
2612 #ifdef CONFIG_SMP
2613 
2614 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2615 {
2616 #ifdef CONFIG_SCHED_CORE
2617 	/*
2618 	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2619 	 * order by core-id first and cpu-id second.
2620 	 *
2621 	 * Notably:
2622 	 *
2623 	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2624 	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2625 	 *
2626 	 * when only cpu-id is considered.
2627 	 */
2628 	if (rq1->core->cpu < rq2->core->cpu)
2629 		return true;
2630 	if (rq1->core->cpu > rq2->core->cpu)
2631 		return false;
2632 
2633 	/*
2634 	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2635 	 */
2636 #endif
2637 	return rq1->cpu < rq2->cpu;
2638 }
2639 
2640 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2641 
2642 #ifdef CONFIG_PREEMPTION
2643 
2644 /*
2645  * fair double_lock_balance: Safely acquires both rq->locks in a fair
2646  * way at the expense of forcing extra atomic operations in all
2647  * invocations.  This assures that the double_lock is acquired using the
2648  * same underlying policy as the spinlock_t on this architecture, which
2649  * reduces latency compared to the unfair variant below.  However, it
2650  * also adds more overhead and therefore may reduce throughput.
2651  */
2652 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2653 	__releases(this_rq->lock)
2654 	__acquires(busiest->lock)
2655 	__acquires(this_rq->lock)
2656 {
2657 	raw_spin_rq_unlock(this_rq);
2658 	double_rq_lock(this_rq, busiest);
2659 
2660 	return 1;
2661 }
2662 
2663 #else
2664 /*
2665  * Unfair double_lock_balance: Optimizes throughput at the expense of
2666  * latency by eliminating extra atomic operations when the locks are
2667  * already in proper order on entry.  This favors lower CPU-ids and will
2668  * grant the double lock to lower CPUs over higher ids under contention,
2669  * regardless of entry order into the function.
2670  */
2671 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2672 	__releases(this_rq->lock)
2673 	__acquires(busiest->lock)
2674 	__acquires(this_rq->lock)
2675 {
2676 	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2677 	    likely(raw_spin_rq_trylock(busiest))) {
2678 		double_rq_clock_clear_update(this_rq, busiest);
2679 		return 0;
2680 	}
2681 
2682 	if (rq_order_less(this_rq, busiest)) {
2683 		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2684 		double_rq_clock_clear_update(this_rq, busiest);
2685 		return 0;
2686 	}
2687 
2688 	raw_spin_rq_unlock(this_rq);
2689 	double_rq_lock(this_rq, busiest);
2690 
2691 	return 1;
2692 }
2693 
2694 #endif /* CONFIG_PREEMPTION */
2695 
2696 /*
2697  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2698  */
2699 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2700 {
2701 	lockdep_assert_irqs_disabled();
2702 
2703 	return _double_lock_balance(this_rq, busiest);
2704 }
2705 
2706 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2707 	__releases(busiest->lock)
2708 {
2709 	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2710 		raw_spin_rq_unlock(busiest);
2711 	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2712 }
2713 
2714 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2715 {
2716 	if (l1 > l2)
2717 		swap(l1, l2);
2718 
2719 	spin_lock(l1);
2720 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2721 }
2722 
2723 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2724 {
2725 	if (l1 > l2)
2726 		swap(l1, l2);
2727 
2728 	spin_lock_irq(l1);
2729 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2730 }
2731 
2732 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2733 {
2734 	if (l1 > l2)
2735 		swap(l1, l2);
2736 
2737 	raw_spin_lock(l1);
2738 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2739 }
2740 
2741 /*
2742  * double_rq_unlock - safely unlock two runqueues
2743  *
2744  * Note this does not restore interrupts like task_rq_unlock,
2745  * you need to do so manually after calling.
2746  */
2747 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2748 	__releases(rq1->lock)
2749 	__releases(rq2->lock)
2750 {
2751 	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2752 		raw_spin_rq_unlock(rq2);
2753 	else
2754 		__release(rq2->lock);
2755 	raw_spin_rq_unlock(rq1);
2756 }
2757 
2758 extern void set_rq_online (struct rq *rq);
2759 extern void set_rq_offline(struct rq *rq);
2760 extern bool sched_smp_initialized;
2761 
2762 #else /* CONFIG_SMP */
2763 
2764 /*
2765  * double_rq_lock - safely lock two runqueues
2766  *
2767  * Note this does not disable interrupts like task_rq_lock,
2768  * you need to do so manually before calling.
2769  */
2770 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2771 	__acquires(rq1->lock)
2772 	__acquires(rq2->lock)
2773 {
2774 	WARN_ON_ONCE(!irqs_disabled());
2775 	WARN_ON_ONCE(rq1 != rq2);
2776 	raw_spin_rq_lock(rq1);
2777 	__acquire(rq2->lock);	/* Fake it out ;) */
2778 	double_rq_clock_clear_update(rq1, rq2);
2779 }
2780 
2781 /*
2782  * double_rq_unlock - safely unlock two runqueues
2783  *
2784  * Note this does not restore interrupts like task_rq_unlock,
2785  * you need to do so manually after calling.
2786  */
2787 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2788 	__releases(rq1->lock)
2789 	__releases(rq2->lock)
2790 {
2791 	WARN_ON_ONCE(rq1 != rq2);
2792 	raw_spin_rq_unlock(rq1);
2793 	__release(rq2->lock);
2794 }
2795 
2796 #endif
2797 
2798 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2799 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2800 
2801 #ifdef	CONFIG_SCHED_DEBUG
2802 extern bool sched_debug_verbose;
2803 
2804 extern void print_cfs_stats(struct seq_file *m, int cpu);
2805 extern void print_rt_stats(struct seq_file *m, int cpu);
2806 extern void print_dl_stats(struct seq_file *m, int cpu);
2807 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2808 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2809 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2810 
2811 extern void resched_latency_warn(int cpu, u64 latency);
2812 #ifdef CONFIG_NUMA_BALANCING
2813 extern void
2814 show_numa_stats(struct task_struct *p, struct seq_file *m);
2815 extern void
2816 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2817 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2818 #endif /* CONFIG_NUMA_BALANCING */
2819 #else
2820 static inline void resched_latency_warn(int cpu, u64 latency) {}
2821 #endif /* CONFIG_SCHED_DEBUG */
2822 
2823 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2824 extern void init_rt_rq(struct rt_rq *rt_rq);
2825 extern void init_dl_rq(struct dl_rq *dl_rq);
2826 
2827 extern void cfs_bandwidth_usage_inc(void);
2828 extern void cfs_bandwidth_usage_dec(void);
2829 
2830 #ifdef CONFIG_NO_HZ_COMMON
2831 #define NOHZ_BALANCE_KICK_BIT	0
2832 #define NOHZ_STATS_KICK_BIT	1
2833 #define NOHZ_NEWILB_KICK_BIT	2
2834 #define NOHZ_NEXT_KICK_BIT	3
2835 
2836 /* Run rebalance_domains() */
2837 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2838 /* Update blocked load */
2839 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2840 /* Update blocked load when entering idle */
2841 #define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2842 /* Update nohz.next_balance */
2843 #define NOHZ_NEXT_KICK		BIT(NOHZ_NEXT_KICK_BIT)
2844 
2845 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2846 
2847 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2848 
2849 extern void nohz_balance_exit_idle(struct rq *rq);
2850 #else
2851 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2852 #endif
2853 
2854 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2855 extern void nohz_run_idle_balance(int cpu);
2856 #else
2857 static inline void nohz_run_idle_balance(int cpu) { }
2858 #endif
2859 
2860 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2861 struct irqtime {
2862 	u64			total;
2863 	u64			tick_delta;
2864 	u64			irq_start_time;
2865 	struct u64_stats_sync	sync;
2866 };
2867 
2868 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2869 
2870 /*
2871  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2872  * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2873  * and never move forward.
2874  */
2875 static inline u64 irq_time_read(int cpu)
2876 {
2877 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2878 	unsigned int seq;
2879 	u64 total;
2880 
2881 	do {
2882 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2883 		total = irqtime->total;
2884 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2885 
2886 	return total;
2887 }
2888 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2889 
2890 #ifdef CONFIG_CPU_FREQ
2891 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2892 
2893 /**
2894  * cpufreq_update_util - Take a note about CPU utilization changes.
2895  * @rq: Runqueue to carry out the update for.
2896  * @flags: Update reason flags.
2897  *
2898  * This function is called by the scheduler on the CPU whose utilization is
2899  * being updated.
2900  *
2901  * It can only be called from RCU-sched read-side critical sections.
2902  *
2903  * The way cpufreq is currently arranged requires it to evaluate the CPU
2904  * performance state (frequency/voltage) on a regular basis to prevent it from
2905  * being stuck in a completely inadequate performance level for too long.
2906  * That is not guaranteed to happen if the updates are only triggered from CFS
2907  * and DL, though, because they may not be coming in if only RT tasks are
2908  * active all the time (or there are RT tasks only).
2909  *
2910  * As a workaround for that issue, this function is called periodically by the
2911  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2912  * but that really is a band-aid.  Going forward it should be replaced with
2913  * solutions targeted more specifically at RT tasks.
2914  */
2915 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2916 {
2917 	struct update_util_data *data;
2918 
2919 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2920 						  cpu_of(rq)));
2921 	if (data)
2922 		data->func(data, rq_clock(rq), flags);
2923 }
2924 #else
2925 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2926 #endif /* CONFIG_CPU_FREQ */
2927 
2928 #ifdef arch_scale_freq_capacity
2929 # ifndef arch_scale_freq_invariant
2930 #  define arch_scale_freq_invariant()	true
2931 # endif
2932 #else
2933 # define arch_scale_freq_invariant()	false
2934 #endif
2935 
2936 #ifdef CONFIG_SMP
2937 static inline unsigned long capacity_orig_of(int cpu)
2938 {
2939 	return cpu_rq(cpu)->cpu_capacity_orig;
2940 }
2941 
2942 /**
2943  * enum cpu_util_type - CPU utilization type
2944  * @FREQUENCY_UTIL:	Utilization used to select frequency
2945  * @ENERGY_UTIL:	Utilization used during energy calculation
2946  *
2947  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2948  * need to be aggregated differently depending on the usage made of them. This
2949  * enum is used within effective_cpu_util() to differentiate the types of
2950  * utilization expected by the callers, and adjust the aggregation accordingly.
2951  */
2952 enum cpu_util_type {
2953 	FREQUENCY_UTIL,
2954 	ENERGY_UTIL,
2955 };
2956 
2957 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2958 				 enum cpu_util_type type,
2959 				 struct task_struct *p);
2960 
2961 /*
2962  * Verify the fitness of task @p to run on @cpu taking into account the
2963  * CPU original capacity and the runtime/deadline ratio of the task.
2964  *
2965  * The function will return true if the original capacity of @cpu is
2966  * greater than or equal to task's deadline density right shifted by
2967  * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2968  */
2969 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2970 {
2971 	unsigned long cap = arch_scale_cpu_capacity(cpu);
2972 
2973 	return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2974 }
2975 
2976 static inline unsigned long cpu_bw_dl(struct rq *rq)
2977 {
2978 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2979 }
2980 
2981 static inline unsigned long cpu_util_dl(struct rq *rq)
2982 {
2983 	return READ_ONCE(rq->avg_dl.util_avg);
2984 }
2985 
2986 
2987 extern unsigned long cpu_util_cfs(int cpu);
2988 extern unsigned long cpu_util_cfs_boost(int cpu);
2989 
2990 static inline unsigned long cpu_util_rt(struct rq *rq)
2991 {
2992 	return READ_ONCE(rq->avg_rt.util_avg);
2993 }
2994 #endif
2995 
2996 #ifdef CONFIG_UCLAMP_TASK
2997 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2998 
2999 static inline unsigned long uclamp_rq_get(struct rq *rq,
3000 					  enum uclamp_id clamp_id)
3001 {
3002 	return READ_ONCE(rq->uclamp[clamp_id].value);
3003 }
3004 
3005 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3006 				 unsigned int value)
3007 {
3008 	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3009 }
3010 
3011 static inline bool uclamp_rq_is_idle(struct rq *rq)
3012 {
3013 	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3014 }
3015 
3016 /**
3017  * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3018  * @rq:		The rq to clamp against. Must not be NULL.
3019  * @util:	The util value to clamp.
3020  * @p:		The task to clamp against. Can be NULL if you want to clamp
3021  *		against @rq only.
3022  *
3023  * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3024  *
3025  * If sched_uclamp_used static key is disabled, then just return the util
3026  * without any clamping since uclamp aggregation at the rq level in the fast
3027  * path is disabled, rendering this operation a NOP.
3028  *
3029  * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3030  * will return the correct effective uclamp value of the task even if the
3031  * static key is disabled.
3032  */
3033 static __always_inline
3034 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3035 				  struct task_struct *p)
3036 {
3037 	unsigned long min_util = 0;
3038 	unsigned long max_util = 0;
3039 
3040 	if (!static_branch_likely(&sched_uclamp_used))
3041 		return util;
3042 
3043 	if (p) {
3044 		min_util = uclamp_eff_value(p, UCLAMP_MIN);
3045 		max_util = uclamp_eff_value(p, UCLAMP_MAX);
3046 
3047 		/*
3048 		 * Ignore last runnable task's max clamp, as this task will
3049 		 * reset it. Similarly, no need to read the rq's min clamp.
3050 		 */
3051 		if (uclamp_rq_is_idle(rq))
3052 			goto out;
3053 	}
3054 
3055 	min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3056 	max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3057 out:
3058 	/*
3059 	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3060 	 * RUNNABLE tasks with _different_ clamps, we can end up with an
3061 	 * inversion. Fix it now when the clamps are applied.
3062 	 */
3063 	if (unlikely(min_util >= max_util))
3064 		return min_util;
3065 
3066 	return clamp(util, min_util, max_util);
3067 }
3068 
3069 /* Is the rq being capped/throttled by uclamp_max? */
3070 static inline bool uclamp_rq_is_capped(struct rq *rq)
3071 {
3072 	unsigned long rq_util;
3073 	unsigned long max_util;
3074 
3075 	if (!static_branch_likely(&sched_uclamp_used))
3076 		return false;
3077 
3078 	rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3079 	max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3080 
3081 	return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3082 }
3083 
3084 /*
3085  * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3086  * by default in the fast path and only gets turned on once userspace performs
3087  * an operation that requires it.
3088  *
3089  * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3090  * hence is active.
3091  */
3092 static inline bool uclamp_is_used(void)
3093 {
3094 	return static_branch_likely(&sched_uclamp_used);
3095 }
3096 #else /* CONFIG_UCLAMP_TASK */
3097 static inline unsigned long uclamp_eff_value(struct task_struct *p,
3098 					     enum uclamp_id clamp_id)
3099 {
3100 	if (clamp_id == UCLAMP_MIN)
3101 		return 0;
3102 
3103 	return SCHED_CAPACITY_SCALE;
3104 }
3105 
3106 static inline
3107 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3108 				  struct task_struct *p)
3109 {
3110 	return util;
3111 }
3112 
3113 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3114 
3115 static inline bool uclamp_is_used(void)
3116 {
3117 	return false;
3118 }
3119 
3120 static inline unsigned long uclamp_rq_get(struct rq *rq,
3121 					  enum uclamp_id clamp_id)
3122 {
3123 	if (clamp_id == UCLAMP_MIN)
3124 		return 0;
3125 
3126 	return SCHED_CAPACITY_SCALE;
3127 }
3128 
3129 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3130 				 unsigned int value)
3131 {
3132 }
3133 
3134 static inline bool uclamp_rq_is_idle(struct rq *rq)
3135 {
3136 	return false;
3137 }
3138 #endif /* CONFIG_UCLAMP_TASK */
3139 
3140 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3141 static inline unsigned long cpu_util_irq(struct rq *rq)
3142 {
3143 	return rq->avg_irq.util_avg;
3144 }
3145 
3146 static inline
3147 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3148 {
3149 	util *= (max - irq);
3150 	util /= max;
3151 
3152 	return util;
3153 
3154 }
3155 #else
3156 static inline unsigned long cpu_util_irq(struct rq *rq)
3157 {
3158 	return 0;
3159 }
3160 
3161 static inline
3162 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3163 {
3164 	return util;
3165 }
3166 #endif
3167 
3168 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3169 
3170 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3171 
3172 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3173 
3174 static inline bool sched_energy_enabled(void)
3175 {
3176 	return static_branch_unlikely(&sched_energy_present);
3177 }
3178 
3179 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3180 
3181 #define perf_domain_span(pd) NULL
3182 static inline bool sched_energy_enabled(void) { return false; }
3183 
3184 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3185 
3186 #ifdef CONFIG_MEMBARRIER
3187 /*
3188  * The scheduler provides memory barriers required by membarrier between:
3189  * - prior user-space memory accesses and store to rq->membarrier_state,
3190  * - store to rq->membarrier_state and following user-space memory accesses.
3191  * In the same way it provides those guarantees around store to rq->curr.
3192  */
3193 static inline void membarrier_switch_mm(struct rq *rq,
3194 					struct mm_struct *prev_mm,
3195 					struct mm_struct *next_mm)
3196 {
3197 	int membarrier_state;
3198 
3199 	if (prev_mm == next_mm)
3200 		return;
3201 
3202 	membarrier_state = atomic_read(&next_mm->membarrier_state);
3203 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3204 		return;
3205 
3206 	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3207 }
3208 #else
3209 static inline void membarrier_switch_mm(struct rq *rq,
3210 					struct mm_struct *prev_mm,
3211 					struct mm_struct *next_mm)
3212 {
3213 }
3214 #endif
3215 
3216 #ifdef CONFIG_SMP
3217 static inline bool is_per_cpu_kthread(struct task_struct *p)
3218 {
3219 	if (!(p->flags & PF_KTHREAD))
3220 		return false;
3221 
3222 	if (p->nr_cpus_allowed != 1)
3223 		return false;
3224 
3225 	return true;
3226 }
3227 #endif
3228 
3229 extern void swake_up_all_locked(struct swait_queue_head *q);
3230 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3231 
3232 #ifdef CONFIG_PREEMPT_DYNAMIC
3233 extern int preempt_dynamic_mode;
3234 extern int sched_dynamic_mode(const char *str);
3235 extern void sched_dynamic_update(int mode);
3236 #endif
3237 
3238 static inline void update_current_exec_runtime(struct task_struct *curr,
3239 						u64 now, u64 delta_exec)
3240 {
3241 	curr->se.sum_exec_runtime += delta_exec;
3242 	account_group_exec_runtime(curr, delta_exec);
3243 
3244 	curr->se.exec_start = now;
3245 	cgroup_account_cputime(curr, delta_exec);
3246 }
3247 
3248 #ifdef CONFIG_SCHED_MM_CID
3249 
3250 #define SCHED_MM_CID_PERIOD_NS	(100ULL * 1000000)	/* 100ms */
3251 #define MM_CID_SCAN_DELAY	100			/* 100ms */
3252 
3253 extern raw_spinlock_t cid_lock;
3254 extern int use_cid_lock;
3255 
3256 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3257 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3258 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3259 extern void init_sched_mm_cid(struct task_struct *t);
3260 
3261 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3262 {
3263 	if (cid < 0)
3264 		return;
3265 	cpumask_clear_cpu(cid, mm_cidmask(mm));
3266 }
3267 
3268 /*
3269  * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3270  * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3271  * be held to transition to other states.
3272  *
3273  * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3274  * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3275  */
3276 static inline void mm_cid_put_lazy(struct task_struct *t)
3277 {
3278 	struct mm_struct *mm = t->mm;
3279 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3280 	int cid;
3281 
3282 	lockdep_assert_irqs_disabled();
3283 	cid = __this_cpu_read(pcpu_cid->cid);
3284 	if (!mm_cid_is_lazy_put(cid) ||
3285 	    !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3286 		return;
3287 	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3288 }
3289 
3290 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3291 {
3292 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3293 	int cid, res;
3294 
3295 	lockdep_assert_irqs_disabled();
3296 	cid = __this_cpu_read(pcpu_cid->cid);
3297 	for (;;) {
3298 		if (mm_cid_is_unset(cid))
3299 			return MM_CID_UNSET;
3300 		/*
3301 		 * Attempt transition from valid or lazy-put to unset.
3302 		 */
3303 		res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3304 		if (res == cid)
3305 			break;
3306 		cid = res;
3307 	}
3308 	return cid;
3309 }
3310 
3311 static inline void mm_cid_put(struct mm_struct *mm)
3312 {
3313 	int cid;
3314 
3315 	lockdep_assert_irqs_disabled();
3316 	cid = mm_cid_pcpu_unset(mm);
3317 	if (cid == MM_CID_UNSET)
3318 		return;
3319 	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3320 }
3321 
3322 static inline int __mm_cid_try_get(struct mm_struct *mm)
3323 {
3324 	struct cpumask *cpumask;
3325 	int cid;
3326 
3327 	cpumask = mm_cidmask(mm);
3328 	/*
3329 	 * Retry finding first zero bit if the mask is temporarily
3330 	 * filled. This only happens during concurrent remote-clear
3331 	 * which owns a cid without holding a rq lock.
3332 	 */
3333 	for (;;) {
3334 		cid = cpumask_first_zero(cpumask);
3335 		if (cid < nr_cpu_ids)
3336 			break;
3337 		cpu_relax();
3338 	}
3339 	if (cpumask_test_and_set_cpu(cid, cpumask))
3340 		return -1;
3341 	return cid;
3342 }
3343 
3344 /*
3345  * Save a snapshot of the current runqueue time of this cpu
3346  * with the per-cpu cid value, allowing to estimate how recently it was used.
3347  */
3348 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3349 {
3350 	struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3351 
3352 	lockdep_assert_rq_held(rq);
3353 	WRITE_ONCE(pcpu_cid->time, rq->clock);
3354 }
3355 
3356 static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3357 {
3358 	int cid;
3359 
3360 	/*
3361 	 * All allocations (even those using the cid_lock) are lock-free. If
3362 	 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3363 	 * guarantee forward progress.
3364 	 */
3365 	if (!READ_ONCE(use_cid_lock)) {
3366 		cid = __mm_cid_try_get(mm);
3367 		if (cid >= 0)
3368 			goto end;
3369 		raw_spin_lock(&cid_lock);
3370 	} else {
3371 		raw_spin_lock(&cid_lock);
3372 		cid = __mm_cid_try_get(mm);
3373 		if (cid >= 0)
3374 			goto unlock;
3375 	}
3376 
3377 	/*
3378 	 * cid concurrently allocated. Retry while forcing following
3379 	 * allocations to use the cid_lock to ensure forward progress.
3380 	 */
3381 	WRITE_ONCE(use_cid_lock, 1);
3382 	/*
3383 	 * Set use_cid_lock before allocation. Only care about program order
3384 	 * because this is only required for forward progress.
3385 	 */
3386 	barrier();
3387 	/*
3388 	 * Retry until it succeeds. It is guaranteed to eventually succeed once
3389 	 * all newcoming allocations observe the use_cid_lock flag set.
3390 	 */
3391 	do {
3392 		cid = __mm_cid_try_get(mm);
3393 		cpu_relax();
3394 	} while (cid < 0);
3395 	/*
3396 	 * Allocate before clearing use_cid_lock. Only care about
3397 	 * program order because this is for forward progress.
3398 	 */
3399 	barrier();
3400 	WRITE_ONCE(use_cid_lock, 0);
3401 unlock:
3402 	raw_spin_unlock(&cid_lock);
3403 end:
3404 	mm_cid_snapshot_time(rq, mm);
3405 	return cid;
3406 }
3407 
3408 static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3409 {
3410 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3411 	struct cpumask *cpumask;
3412 	int cid;
3413 
3414 	lockdep_assert_rq_held(rq);
3415 	cpumask = mm_cidmask(mm);
3416 	cid = __this_cpu_read(pcpu_cid->cid);
3417 	if (mm_cid_is_valid(cid)) {
3418 		mm_cid_snapshot_time(rq, mm);
3419 		return cid;
3420 	}
3421 	if (mm_cid_is_lazy_put(cid)) {
3422 		if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3423 			__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3424 	}
3425 	cid = __mm_cid_get(rq, mm);
3426 	__this_cpu_write(pcpu_cid->cid, cid);
3427 	return cid;
3428 }
3429 
3430 static inline void switch_mm_cid(struct rq *rq,
3431 				 struct task_struct *prev,
3432 				 struct task_struct *next)
3433 {
3434 	/*
3435 	 * Provide a memory barrier between rq->curr store and load of
3436 	 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3437 	 *
3438 	 * Should be adapted if context_switch() is modified.
3439 	 */
3440 	if (!next->mm) {                                // to kernel
3441 		/*
3442 		 * user -> kernel transition does not guarantee a barrier, but
3443 		 * we can use the fact that it performs an atomic operation in
3444 		 * mmgrab().
3445 		 */
3446 		if (prev->mm)                           // from user
3447 			smp_mb__after_mmgrab();
3448 		/*
3449 		 * kernel -> kernel transition does not change rq->curr->mm
3450 		 * state. It stays NULL.
3451 		 */
3452 	} else {                                        // to user
3453 		/*
3454 		 * kernel -> user transition does not provide a barrier
3455 		 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3456 		 * Provide it here.
3457 		 */
3458 		if (!prev->mm)                          // from kernel
3459 			smp_mb();
3460 		/*
3461 		 * user -> user transition guarantees a memory barrier through
3462 		 * switch_mm() when current->mm changes. If current->mm is
3463 		 * unchanged, no barrier is needed.
3464 		 */
3465 	}
3466 	if (prev->mm_cid_active) {
3467 		mm_cid_snapshot_time(rq, prev->mm);
3468 		mm_cid_put_lazy(prev);
3469 		prev->mm_cid = -1;
3470 	}
3471 	if (next->mm_cid_active)
3472 		next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3473 }
3474 
3475 #else
3476 static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3477 static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3478 static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3479 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3480 static inline void init_sched_mm_cid(struct task_struct *t) { }
3481 #endif
3482 
3483 #endif /* _KERNEL_SCHED_SCHED_H */
3484