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 
idle_policy(int policy)181 static inline int idle_policy(int policy)
182 {
183 	return policy == SCHED_IDLE;
184 }
fair_policy(int policy)185 static inline int fair_policy(int policy)
186 {
187 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
188 }
189 
rt_policy(int policy)190 static inline int rt_policy(int policy)
191 {
192 	return policy == SCHED_FIFO || policy == SCHED_RR;
193 }
194 
dl_policy(int policy)195 static inline int dl_policy(int policy)
196 {
197 	return policy == SCHED_DEADLINE;
198 }
valid_policy(int policy)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 
task_has_idle_policy(struct task_struct * p)205 static inline int task_has_idle_policy(struct task_struct *p)
206 {
207 	return idle_policy(p->policy);
208 }
209 
task_has_rt_policy(struct task_struct * p)210 static inline int task_has_rt_policy(struct task_struct *p)
211 {
212 	return rt_policy(p->policy);
213 }
214 
task_has_dl_policy(struct task_struct * p)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 
update_avg(u64 * avg,u64 sample)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 
dl_entity_is_special(struct sched_dl_entity * dl_se)251 static inline bool dl_entity_is_special(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(struct sched_dl_entity * a,struct sched_dl_entity * b)264 dl_entity_preempt(struct sched_dl_entity *a, 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 struct dl_bandwidth {
290 	raw_spinlock_t		dl_runtime_lock;
291 	u64			dl_runtime;
292 	u64			dl_period;
293 };
294 
dl_bandwidth_enabled(void)295 static inline int dl_bandwidth_enabled(void)
296 {
297 	return sysctl_sched_rt_runtime >= 0;
298 }
299 
300 /*
301  * To keep the bandwidth of -deadline tasks under control
302  * we need some place where:
303  *  - store the maximum -deadline bandwidth of each cpu;
304  *  - cache the fraction of bandwidth that is currently allocated in
305  *    each root domain;
306  *
307  * This is all done in the data structure below. It is similar to the
308  * one used for RT-throttling (rt_bandwidth), with the main difference
309  * that, since here we are only interested in admission control, we
310  * do not decrease any runtime while the group "executes", neither we
311  * need a timer to replenish it.
312  *
313  * With respect to SMP, bandwidth is given on a per root domain basis,
314  * meaning that:
315  *  - bw (< 100%) is the deadline bandwidth of each CPU;
316  *  - total_bw is the currently allocated bandwidth in each root domain;
317  */
318 struct dl_bw {
319 	raw_spinlock_t		lock;
320 	u64			bw;
321 	u64			total_bw;
322 };
323 
324 extern void init_dl_bw(struct dl_bw *dl_b);
325 extern int  sched_dl_global_validate(void);
326 extern void sched_dl_do_global(void);
327 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
328 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
329 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
330 extern bool __checkparam_dl(const struct sched_attr *attr);
331 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
332 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
333 extern int  dl_cpu_busy(int cpu, struct task_struct *p);
334 
335 #ifdef CONFIG_CGROUP_SCHED
336 
337 struct cfs_rq;
338 struct rt_rq;
339 
340 extern struct list_head task_groups;
341 
342 struct cfs_bandwidth {
343 #ifdef CONFIG_CFS_BANDWIDTH
344 	raw_spinlock_t		lock;
345 	ktime_t			period;
346 	u64			quota;
347 	u64			runtime;
348 	u64			burst;
349 	u64			runtime_snap;
350 	s64			hierarchical_quota;
351 
352 	u8			idle;
353 	u8			period_active;
354 	u8			slack_started;
355 	struct hrtimer		period_timer;
356 	struct hrtimer		slack_timer;
357 	struct list_head	throttled_cfs_rq;
358 
359 	/* Statistics: */
360 	int			nr_periods;
361 	int			nr_throttled;
362 	int			nr_burst;
363 	u64			throttled_time;
364 	u64			burst_time;
365 #endif
366 };
367 
368 /* Task group related information */
369 struct task_group {
370 	struct cgroup_subsys_state css;
371 
372 #ifdef CONFIG_FAIR_GROUP_SCHED
373 	/* schedulable entities of this group on each CPU */
374 	struct sched_entity	**se;
375 	/* runqueue "owned" by this group on each CPU */
376 	struct cfs_rq		**cfs_rq;
377 	unsigned long		shares;
378 
379 	/* A positive value indicates that this is a SCHED_IDLE group. */
380 	int			idle;
381 
382 #ifdef	CONFIG_SMP
383 	/*
384 	 * load_avg can be heavily contended at clock tick time, so put
385 	 * it in its own cacheline separated from the fields above which
386 	 * will also be accessed at each tick.
387 	 */
388 	atomic_long_t		load_avg ____cacheline_aligned;
389 #endif
390 #endif
391 
392 #ifdef CONFIG_RT_GROUP_SCHED
393 	struct sched_rt_entity	**rt_se;
394 	struct rt_rq		**rt_rq;
395 
396 	struct rt_bandwidth	rt_bandwidth;
397 #endif
398 
399 	struct rcu_head		rcu;
400 	struct list_head	list;
401 
402 	struct task_group	*parent;
403 	struct list_head	siblings;
404 	struct list_head	children;
405 
406 #ifdef CONFIG_SCHED_AUTOGROUP
407 	struct autogroup	*autogroup;
408 #endif
409 
410 	struct cfs_bandwidth	cfs_bandwidth;
411 
412 #ifdef CONFIG_UCLAMP_TASK_GROUP
413 	/* The two decimal precision [%] value requested from user-space */
414 	unsigned int		uclamp_pct[UCLAMP_CNT];
415 	/* Clamp values requested for a task group */
416 	struct uclamp_se	uclamp_req[UCLAMP_CNT];
417 	/* Effective clamp values used for a task group */
418 	struct uclamp_se	uclamp[UCLAMP_CNT];
419 #endif
420 
421 };
422 
423 #ifdef CONFIG_FAIR_GROUP_SCHED
424 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
425 
426 /*
427  * A weight of 0 or 1 can cause arithmetics problems.
428  * A weight of a cfs_rq is the sum of weights of which entities
429  * are queued on this cfs_rq, so a weight of a entity should not be
430  * too large, so as the shares value of a task group.
431  * (The default weight is 1024 - so there's no practical
432  *  limitation from this.)
433  */
434 #define MIN_SHARES		(1UL <<  1)
435 #define MAX_SHARES		(1UL << 18)
436 #endif
437 
438 typedef int (*tg_visitor)(struct task_group *, void *);
439 
440 extern int walk_tg_tree_from(struct task_group *from,
441 			     tg_visitor down, tg_visitor up, void *data);
442 
443 /*
444  * Iterate the full tree, calling @down when first entering a node and @up when
445  * leaving it for the final time.
446  *
447  * Caller must hold rcu_lock or sufficient equivalent.
448  */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)449 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
450 {
451 	return walk_tg_tree_from(&root_task_group, down, up, data);
452 }
453 
454 extern int tg_nop(struct task_group *tg, void *data);
455 
456 extern void free_fair_sched_group(struct task_group *tg);
457 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
458 extern void online_fair_sched_group(struct task_group *tg);
459 extern void unregister_fair_sched_group(struct task_group *tg);
460 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
461 			struct sched_entity *se, int cpu,
462 			struct sched_entity *parent);
463 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
464 
465 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
466 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
467 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
468 
469 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
470 		struct sched_rt_entity *rt_se, int cpu,
471 		struct sched_rt_entity *parent);
472 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
473 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
474 extern long sched_group_rt_runtime(struct task_group *tg);
475 extern long sched_group_rt_period(struct task_group *tg);
476 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
477 
478 extern struct task_group *sched_create_group(struct task_group *parent);
479 extern void sched_online_group(struct task_group *tg,
480 			       struct task_group *parent);
481 extern void sched_destroy_group(struct task_group *tg);
482 extern void sched_release_group(struct task_group *tg);
483 
484 extern void sched_move_task(struct task_struct *tsk);
485 
486 #ifdef CONFIG_FAIR_GROUP_SCHED
487 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
488 
489 extern int sched_group_set_idle(struct task_group *tg, long idle);
490 
491 #ifdef CONFIG_SMP
492 extern void set_task_rq_fair(struct sched_entity *se,
493 			     struct cfs_rq *prev, struct cfs_rq *next);
494 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)495 static inline void set_task_rq_fair(struct sched_entity *se,
496 			     struct cfs_rq *prev, struct cfs_rq *next) { }
497 #endif /* CONFIG_SMP */
498 #endif /* CONFIG_FAIR_GROUP_SCHED */
499 
500 #else /* CONFIG_CGROUP_SCHED */
501 
502 struct cfs_bandwidth { };
503 
504 #endif	/* CONFIG_CGROUP_SCHED */
505 
506 extern void unregister_rt_sched_group(struct task_group *tg);
507 extern void free_rt_sched_group(struct task_group *tg);
508 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
509 
510 /*
511  * u64_u32_load/u64_u32_store
512  *
513  * Use a copy of a u64 value to protect against data race. This is only
514  * applicable for 32-bits architectures.
515  */
516 #ifdef CONFIG_64BIT
517 # define u64_u32_load_copy(var, copy)       var
518 # define u64_u32_store_copy(var, copy, val) (var = val)
519 #else
520 # define u64_u32_load_copy(var, copy)					\
521 ({									\
522 	u64 __val, __val_copy;						\
523 	do {								\
524 		__val_copy = copy;					\
525 		/*							\
526 		 * paired with u64_u32_store_copy(), ordering access	\
527 		 * to var and copy.					\
528 		 */							\
529 		smp_rmb();						\
530 		__val = var;						\
531 	} while (__val != __val_copy);					\
532 	__val;								\
533 })
534 # define u64_u32_store_copy(var, copy, val)				\
535 do {									\
536 	typeof(val) __val = (val);					\
537 	var = __val;							\
538 	/*								\
539 	 * paired with u64_u32_load_copy(), ordering access to var and	\
540 	 * copy.							\
541 	 */								\
542 	smp_wmb();							\
543 	copy = __val;							\
544 } while (0)
545 #endif
546 # define u64_u32_load(var)      u64_u32_load_copy(var, var##_copy)
547 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
548 
549 /* CFS-related fields in a runqueue */
550 struct cfs_rq {
551 	struct load_weight	load;
552 	unsigned int		nr_running;
553 	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
554 	unsigned int		idle_nr_running;   /* SCHED_IDLE */
555 	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
556 
557 	u64			exec_clock;
558 	u64			min_vruntime;
559 #ifdef CONFIG_SCHED_CORE
560 	unsigned int		forceidle_seq;
561 	u64			min_vruntime_fi;
562 #endif
563 
564 #ifndef CONFIG_64BIT
565 	u64			min_vruntime_copy;
566 #endif
567 
568 	struct rb_root_cached	tasks_timeline;
569 
570 	/*
571 	 * 'curr' points to currently running entity on this cfs_rq.
572 	 * It is set to NULL otherwise (i.e when none are currently running).
573 	 */
574 	struct sched_entity	*curr;
575 	struct sched_entity	*next;
576 	struct sched_entity	*last;
577 	struct sched_entity	*skip;
578 
579 #ifdef	CONFIG_SCHED_DEBUG
580 	unsigned int		nr_spread_over;
581 #endif
582 
583 #ifdef CONFIG_SMP
584 	/*
585 	 * CFS load tracking
586 	 */
587 	struct sched_avg	avg;
588 #ifndef CONFIG_64BIT
589 	u64			last_update_time_copy;
590 #endif
591 	struct {
592 		raw_spinlock_t	lock ____cacheline_aligned;
593 		int		nr;
594 		unsigned long	load_avg;
595 		unsigned long	util_avg;
596 		unsigned long	runnable_avg;
597 	} removed;
598 
599 #ifdef CONFIG_FAIR_GROUP_SCHED
600 	unsigned long		tg_load_avg_contrib;
601 	long			propagate;
602 	long			prop_runnable_sum;
603 
604 	/*
605 	 *   h_load = weight * f(tg)
606 	 *
607 	 * Where f(tg) is the recursive weight fraction assigned to
608 	 * this group.
609 	 */
610 	unsigned long		h_load;
611 	u64			last_h_load_update;
612 	struct sched_entity	*h_load_next;
613 #endif /* CONFIG_FAIR_GROUP_SCHED */
614 #endif /* CONFIG_SMP */
615 
616 #ifdef CONFIG_FAIR_GROUP_SCHED
617 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
618 
619 	/*
620 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
621 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
622 	 * (like users, containers etc.)
623 	 *
624 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
625 	 * This list is used during load balance.
626 	 */
627 	int			on_list;
628 	struct list_head	leaf_cfs_rq_list;
629 	struct task_group	*tg;	/* group that "owns" this runqueue */
630 
631 	/* Locally cached copy of our task_group's idle value */
632 	int			idle;
633 
634 #ifdef CONFIG_CFS_BANDWIDTH
635 	int			runtime_enabled;
636 	s64			runtime_remaining;
637 
638 	u64			throttled_pelt_idle;
639 #ifndef CONFIG_64BIT
640 	u64                     throttled_pelt_idle_copy;
641 #endif
642 	u64			throttled_clock;
643 	u64			throttled_clock_pelt;
644 	u64			throttled_clock_pelt_time;
645 	int			throttled;
646 	int			throttle_count;
647 	struct list_head	throttled_list;
648 #endif /* CONFIG_CFS_BANDWIDTH */
649 #endif /* CONFIG_FAIR_GROUP_SCHED */
650 };
651 
rt_bandwidth_enabled(void)652 static inline int rt_bandwidth_enabled(void)
653 {
654 	return sysctl_sched_rt_runtime >= 0;
655 }
656 
657 /* RT IPI pull logic requires IRQ_WORK */
658 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
659 # define HAVE_RT_PUSH_IPI
660 #endif
661 
662 /* Real-Time classes' related field in a runqueue: */
663 struct rt_rq {
664 	struct rt_prio_array	active;
665 	unsigned int		rt_nr_running;
666 	unsigned int		rr_nr_running;
667 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
668 	struct {
669 		int		curr; /* highest queued rt task prio */
670 #ifdef CONFIG_SMP
671 		int		next; /* next highest */
672 #endif
673 	} highest_prio;
674 #endif
675 #ifdef CONFIG_SMP
676 	unsigned int		rt_nr_migratory;
677 	unsigned int		rt_nr_total;
678 	int			overloaded;
679 	struct plist_head	pushable_tasks;
680 
681 #endif /* CONFIG_SMP */
682 	int			rt_queued;
683 
684 	int			rt_throttled;
685 	u64			rt_time;
686 	u64			rt_runtime;
687 	/* Nests inside the rq lock: */
688 	raw_spinlock_t		rt_runtime_lock;
689 
690 #ifdef CONFIG_RT_GROUP_SCHED
691 	unsigned int		rt_nr_boosted;
692 
693 	struct rq		*rq;
694 	struct task_group	*tg;
695 #endif
696 };
697 
rt_rq_is_runnable(struct rt_rq * rt_rq)698 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
699 {
700 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
701 }
702 
703 /* Deadline class' related fields in a runqueue */
704 struct dl_rq {
705 	/* runqueue is an rbtree, ordered by deadline */
706 	struct rb_root_cached	root;
707 
708 	unsigned int		dl_nr_running;
709 
710 #ifdef CONFIG_SMP
711 	/*
712 	 * Deadline values of the currently executing and the
713 	 * earliest ready task on this rq. Caching these facilitates
714 	 * the decision whether or not a ready but not running task
715 	 * should migrate somewhere else.
716 	 */
717 	struct {
718 		u64		curr;
719 		u64		next;
720 	} earliest_dl;
721 
722 	unsigned int		dl_nr_migratory;
723 	int			overloaded;
724 
725 	/*
726 	 * Tasks on this rq that can be pushed away. They are kept in
727 	 * an rb-tree, ordered by tasks' deadlines, with caching
728 	 * of the leftmost (earliest deadline) element.
729 	 */
730 	struct rb_root_cached	pushable_dl_tasks_root;
731 #else
732 	struct dl_bw		dl_bw;
733 #endif
734 	/*
735 	 * "Active utilization" for this runqueue: increased when a
736 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
737 	 * task blocks
738 	 */
739 	u64			running_bw;
740 
741 	/*
742 	 * Utilization of the tasks "assigned" to this runqueue (including
743 	 * the tasks that are in runqueue and the tasks that executed on this
744 	 * CPU and blocked). Increased when a task moves to this runqueue, and
745 	 * decreased when the task moves away (migrates, changes scheduling
746 	 * policy, or terminates).
747 	 * This is needed to compute the "inactive utilization" for the
748 	 * runqueue (inactive utilization = this_bw - running_bw).
749 	 */
750 	u64			this_bw;
751 	u64			extra_bw;
752 
753 	/*
754 	 * Inverse of the fraction of CPU utilization that can be reclaimed
755 	 * by the GRUB algorithm.
756 	 */
757 	u64			bw_ratio;
758 };
759 
760 #ifdef CONFIG_FAIR_GROUP_SCHED
761 /* An entity is a task if it doesn't "own" a runqueue */
762 #define entity_is_task(se)	(!se->my_q)
763 
se_update_runnable(struct sched_entity * se)764 static inline void se_update_runnable(struct sched_entity *se)
765 {
766 	if (!entity_is_task(se))
767 		se->runnable_weight = se->my_q->h_nr_running;
768 }
769 
se_runnable(struct sched_entity * se)770 static inline long se_runnable(struct sched_entity *se)
771 {
772 	if (entity_is_task(se))
773 		return !!se->on_rq;
774 	else
775 		return se->runnable_weight;
776 }
777 
778 #else
779 #define entity_is_task(se)	1
780 
se_update_runnable(struct sched_entity * se)781 static inline void se_update_runnable(struct sched_entity *se) {}
782 
se_runnable(struct sched_entity * se)783 static inline long se_runnable(struct sched_entity *se)
784 {
785 	return !!se->on_rq;
786 }
787 #endif
788 
789 #ifdef CONFIG_SMP
790 /*
791  * XXX we want to get rid of these helpers and use the full load resolution.
792  */
se_weight(struct sched_entity * se)793 static inline long se_weight(struct sched_entity *se)
794 {
795 	return scale_load_down(se->load.weight);
796 }
797 
798 
sched_asym_prefer(int a,int b)799 static inline bool sched_asym_prefer(int a, int b)
800 {
801 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
802 }
803 
804 struct perf_domain {
805 	struct em_perf_domain *em_pd;
806 	struct perf_domain *next;
807 	struct rcu_head rcu;
808 };
809 
810 /* Scheduling group status flags */
811 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
812 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
813 
814 /*
815  * We add the notion of a root-domain which will be used to define per-domain
816  * variables. Each exclusive cpuset essentially defines an island domain by
817  * fully partitioning the member CPUs from any other cpuset. Whenever a new
818  * exclusive cpuset is created, we also create and attach a new root-domain
819  * object.
820  *
821  */
822 struct root_domain {
823 	atomic_t		refcount;
824 	atomic_t		rto_count;
825 	struct rcu_head		rcu;
826 	cpumask_var_t		span;
827 	cpumask_var_t		online;
828 
829 	/*
830 	 * Indicate pullable load on at least one CPU, e.g:
831 	 * - More than one runnable task
832 	 * - Running task is misfit
833 	 */
834 	int			overload;
835 
836 	/* Indicate one or more cpus over-utilized (tipping point) */
837 	int			overutilized;
838 
839 	/*
840 	 * The bit corresponding to a CPU gets set here if such CPU has more
841 	 * than one runnable -deadline task (as it is below for RT tasks).
842 	 */
843 	cpumask_var_t		dlo_mask;
844 	atomic_t		dlo_count;
845 	struct dl_bw		dl_bw;
846 	struct cpudl		cpudl;
847 
848 	/*
849 	 * Indicate whether a root_domain's dl_bw has been checked or
850 	 * updated. It's monotonously increasing value.
851 	 *
852 	 * Also, some corner cases, like 'wrap around' is dangerous, but given
853 	 * that u64 is 'big enough'. So that shouldn't be a concern.
854 	 */
855 	u64 visit_gen;
856 
857 #ifdef HAVE_RT_PUSH_IPI
858 	/*
859 	 * For IPI pull requests, loop across the rto_mask.
860 	 */
861 	struct irq_work		rto_push_work;
862 	raw_spinlock_t		rto_lock;
863 	/* These are only updated and read within rto_lock */
864 	int			rto_loop;
865 	int			rto_cpu;
866 	/* These atomics are updated outside of a lock */
867 	atomic_t		rto_loop_next;
868 	atomic_t		rto_loop_start;
869 #endif
870 	/*
871 	 * The "RT overload" flag: it gets set if a CPU has more than
872 	 * one runnable RT task.
873 	 */
874 	cpumask_var_t		rto_mask;
875 	struct cpupri		cpupri;
876 
877 	unsigned long		max_cpu_capacity;
878 
879 	/*
880 	 * NULL-terminated list of performance domains intersecting with the
881 	 * CPUs of the rd. Protected by RCU.
882 	 */
883 	struct perf_domain __rcu *pd;
884 };
885 
886 extern void init_defrootdomain(void);
887 extern int sched_init_domains(const struct cpumask *cpu_map);
888 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
889 extern void sched_get_rd(struct root_domain *rd);
890 extern void sched_put_rd(struct root_domain *rd);
891 
892 #ifdef HAVE_RT_PUSH_IPI
893 extern void rto_push_irq_work_func(struct irq_work *work);
894 #endif
895 #endif /* CONFIG_SMP */
896 
897 #ifdef CONFIG_UCLAMP_TASK
898 /*
899  * struct uclamp_bucket - Utilization clamp bucket
900  * @value: utilization clamp value for tasks on this clamp bucket
901  * @tasks: number of RUNNABLE tasks on this clamp bucket
902  *
903  * Keep track of how many tasks are RUNNABLE for a given utilization
904  * clamp value.
905  */
906 struct uclamp_bucket {
907 	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
908 	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
909 };
910 
911 /*
912  * struct uclamp_rq - rq's utilization clamp
913  * @value: currently active clamp values for a rq
914  * @bucket: utilization clamp buckets affecting a rq
915  *
916  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
917  * A clamp value is affecting a rq when there is at least one task RUNNABLE
918  * (or actually running) with that value.
919  *
920  * There are up to UCLAMP_CNT possible different clamp values, currently there
921  * are only two: minimum utilization and maximum utilization.
922  *
923  * All utilization clamping values are MAX aggregated, since:
924  * - for util_min: we want to run the CPU at least at the max of the minimum
925  *   utilization required by its currently RUNNABLE tasks.
926  * - for util_max: we want to allow the CPU to run up to the max of the
927  *   maximum utilization allowed by its currently RUNNABLE tasks.
928  *
929  * Since on each system we expect only a limited number of different
930  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
931  * the metrics required to compute all the per-rq utilization clamp values.
932  */
933 struct uclamp_rq {
934 	unsigned int value;
935 	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
936 };
937 
938 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
939 #endif /* CONFIG_UCLAMP_TASK */
940 
941 struct rq;
942 struct balance_callback {
943 	struct balance_callback *next;
944 	void (*func)(struct rq *rq);
945 };
946 
947 /*
948  * This is the main, per-CPU runqueue data structure.
949  *
950  * Locking rule: those places that want to lock multiple runqueues
951  * (such as the load balancing or the thread migration code), lock
952  * acquire operations must be ordered by ascending &runqueue.
953  */
954 struct rq {
955 	/* runqueue lock: */
956 	raw_spinlock_t		__lock;
957 
958 	/*
959 	 * nr_running and cpu_load should be in the same cacheline because
960 	 * remote CPUs use both these fields when doing load calculation.
961 	 */
962 	unsigned int		nr_running;
963 #ifdef CONFIG_NUMA_BALANCING
964 	unsigned int		nr_numa_running;
965 	unsigned int		nr_preferred_running;
966 	unsigned int		numa_migrate_on;
967 #endif
968 #ifdef CONFIG_NO_HZ_COMMON
969 #ifdef CONFIG_SMP
970 	unsigned long		last_blocked_load_update_tick;
971 	unsigned int		has_blocked_load;
972 	call_single_data_t	nohz_csd;
973 #endif /* CONFIG_SMP */
974 	unsigned int		nohz_tick_stopped;
975 	atomic_t		nohz_flags;
976 #endif /* CONFIG_NO_HZ_COMMON */
977 
978 #ifdef CONFIG_SMP
979 	unsigned int		ttwu_pending;
980 #endif
981 	u64			nr_switches;
982 
983 #ifdef CONFIG_UCLAMP_TASK
984 	/* Utilization clamp values based on CPU's RUNNABLE tasks */
985 	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
986 	unsigned int		uclamp_flags;
987 #define UCLAMP_FLAG_IDLE 0x01
988 #endif
989 
990 	struct cfs_rq		cfs;
991 	struct rt_rq		rt;
992 	struct dl_rq		dl;
993 
994 #ifdef CONFIG_FAIR_GROUP_SCHED
995 	/* list of leaf cfs_rq on this CPU: */
996 	struct list_head	leaf_cfs_rq_list;
997 	struct list_head	*tmp_alone_branch;
998 #endif /* CONFIG_FAIR_GROUP_SCHED */
999 
1000 	/*
1001 	 * This is part of a global counter where only the total sum
1002 	 * over all CPUs matters. A task can increase this counter on
1003 	 * one CPU and if it got migrated afterwards it may decrease
1004 	 * it on another CPU. Always updated under the runqueue lock:
1005 	 */
1006 	unsigned int		nr_uninterruptible;
1007 
1008 	struct task_struct __rcu	*curr;
1009 	struct task_struct	*idle;
1010 	struct task_struct	*stop;
1011 	unsigned long		next_balance;
1012 	struct mm_struct	*prev_mm;
1013 
1014 	unsigned int		clock_update_flags;
1015 	u64			clock;
1016 	/* Ensure that all clocks are in the same cache line */
1017 	u64			clock_task ____cacheline_aligned;
1018 	u64			clock_pelt;
1019 	unsigned long		lost_idle_time;
1020 	u64			clock_pelt_idle;
1021 	u64			clock_idle;
1022 #ifndef CONFIG_64BIT
1023 	u64			clock_pelt_idle_copy;
1024 	u64			clock_idle_copy;
1025 #endif
1026 
1027 	atomic_t		nr_iowait;
1028 
1029 #ifdef CONFIG_SCHED_DEBUG
1030 	u64 last_seen_need_resched_ns;
1031 	int ticks_without_resched;
1032 #endif
1033 
1034 #ifdef CONFIG_MEMBARRIER
1035 	int membarrier_state;
1036 #endif
1037 
1038 #ifdef CONFIG_SMP
1039 	struct root_domain		*rd;
1040 	struct sched_domain __rcu	*sd;
1041 
1042 	unsigned long		cpu_capacity;
1043 	unsigned long		cpu_capacity_orig;
1044 
1045 	struct balance_callback *balance_callback;
1046 
1047 	unsigned char		nohz_idle_balance;
1048 	unsigned char		idle_balance;
1049 
1050 	unsigned long		misfit_task_load;
1051 
1052 	/* For active balancing */
1053 	int			active_balance;
1054 	int			push_cpu;
1055 	struct cpu_stop_work	active_balance_work;
1056 
1057 	/* CPU of this runqueue: */
1058 	int			cpu;
1059 	int			online;
1060 
1061 	struct list_head cfs_tasks;
1062 
1063 	struct sched_avg	avg_rt;
1064 	struct sched_avg	avg_dl;
1065 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1066 	struct sched_avg	avg_irq;
1067 #endif
1068 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1069 	struct sched_avg	avg_thermal;
1070 #endif
1071 	u64			idle_stamp;
1072 	u64			avg_idle;
1073 
1074 	unsigned long		wake_stamp;
1075 	u64			wake_avg_idle;
1076 
1077 	/* This is used to determine avg_idle's max value */
1078 	u64			max_idle_balance_cost;
1079 
1080 #ifdef CONFIG_HOTPLUG_CPU
1081 	struct rcuwait		hotplug_wait;
1082 #endif
1083 #endif /* CONFIG_SMP */
1084 
1085 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1086 	u64			prev_irq_time;
1087 #endif
1088 #ifdef CONFIG_PARAVIRT
1089 	u64			prev_steal_time;
1090 #endif
1091 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1092 	u64			prev_steal_time_rq;
1093 #endif
1094 
1095 	/* calc_load related fields */
1096 	unsigned long		calc_load_update;
1097 	long			calc_load_active;
1098 
1099 #ifdef CONFIG_SCHED_HRTICK
1100 #ifdef CONFIG_SMP
1101 	call_single_data_t	hrtick_csd;
1102 #endif
1103 	struct hrtimer		hrtick_timer;
1104 	ktime_t 		hrtick_time;
1105 #endif
1106 
1107 #ifdef CONFIG_SCHEDSTATS
1108 	/* latency stats */
1109 	struct sched_info	rq_sched_info;
1110 	unsigned long long	rq_cpu_time;
1111 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1112 
1113 	/* sys_sched_yield() stats */
1114 	unsigned int		yld_count;
1115 
1116 	/* schedule() stats */
1117 	unsigned int		sched_count;
1118 	unsigned int		sched_goidle;
1119 
1120 	/* try_to_wake_up() stats */
1121 	unsigned int		ttwu_count;
1122 	unsigned int		ttwu_local;
1123 #endif
1124 
1125 #ifdef CONFIG_CPU_IDLE
1126 	/* Must be inspected within a rcu lock section */
1127 	struct cpuidle_state	*idle_state;
1128 #endif
1129 
1130 #ifdef CONFIG_SMP
1131 	unsigned int		nr_pinned;
1132 #endif
1133 	unsigned int		push_busy;
1134 	struct cpu_stop_work	push_work;
1135 
1136 #ifdef CONFIG_SCHED_CORE
1137 	/* per rq */
1138 	struct rq		*core;
1139 	struct task_struct	*core_pick;
1140 	unsigned int		core_enabled;
1141 	unsigned int		core_sched_seq;
1142 	struct rb_root		core_tree;
1143 
1144 	/* shared state -- careful with sched_core_cpu_deactivate() */
1145 	unsigned int		core_task_seq;
1146 	unsigned int		core_pick_seq;
1147 	unsigned long		core_cookie;
1148 	unsigned int		core_forceidle_count;
1149 	unsigned int		core_forceidle_seq;
1150 	unsigned int		core_forceidle_occupation;
1151 	u64			core_forceidle_start;
1152 #endif
1153 };
1154 
1155 #ifdef CONFIG_FAIR_GROUP_SCHED
1156 
1157 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1158 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1159 {
1160 	return cfs_rq->rq;
1161 }
1162 
1163 #else
1164 
rq_of(struct cfs_rq * cfs_rq)1165 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1166 {
1167 	return container_of(cfs_rq, struct rq, cfs);
1168 }
1169 #endif
1170 
cpu_of(struct rq * rq)1171 static inline int cpu_of(struct rq *rq)
1172 {
1173 #ifdef CONFIG_SMP
1174 	return rq->cpu;
1175 #else
1176 	return 0;
1177 #endif
1178 }
1179 
1180 #define MDF_PUSH	0x01
1181 
is_migration_disabled(struct task_struct * p)1182 static inline bool is_migration_disabled(struct task_struct *p)
1183 {
1184 #ifdef CONFIG_SMP
1185 	return p->migration_disabled;
1186 #else
1187 	return false;
1188 #endif
1189 }
1190 
1191 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1192 
1193 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1194 #define this_rq()		this_cpu_ptr(&runqueues)
1195 #define task_rq(p)		cpu_rq(task_cpu(p))
1196 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1197 #define raw_rq()		raw_cpu_ptr(&runqueues)
1198 
1199 struct sched_group;
1200 #ifdef CONFIG_SCHED_CORE
1201 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1202 
1203 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1204 
sched_core_enabled(struct rq * rq)1205 static inline bool sched_core_enabled(struct rq *rq)
1206 {
1207 	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1208 }
1209 
sched_core_disabled(void)1210 static inline bool sched_core_disabled(void)
1211 {
1212 	return !static_branch_unlikely(&__sched_core_enabled);
1213 }
1214 
1215 /*
1216  * Be careful with this function; not for general use. The return value isn't
1217  * stable unless you actually hold a relevant rq->__lock.
1218  */
rq_lockp(struct rq * rq)1219 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1220 {
1221 	if (sched_core_enabled(rq))
1222 		return &rq->core->__lock;
1223 
1224 	return &rq->__lock;
1225 }
1226 
__rq_lockp(struct rq * rq)1227 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1228 {
1229 	if (rq->core_enabled)
1230 		return &rq->core->__lock;
1231 
1232 	return &rq->__lock;
1233 }
1234 
1235 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1236 
1237 /*
1238  * Helpers to check if the CPU's core cookie matches with the task's cookie
1239  * when core scheduling is enabled.
1240  * A special case is that the task's cookie always matches with CPU's core
1241  * cookie if the CPU is in an idle core.
1242  */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1243 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1244 {
1245 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1246 	if (!sched_core_enabled(rq))
1247 		return true;
1248 
1249 	return rq->core->core_cookie == p->core_cookie;
1250 }
1251 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1252 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1253 {
1254 	bool idle_core = true;
1255 	int cpu;
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 	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1262 		if (!available_idle_cpu(cpu)) {
1263 			idle_core = false;
1264 			break;
1265 		}
1266 	}
1267 
1268 	/*
1269 	 * A CPU in an idle core is always the best choice for tasks with
1270 	 * cookies.
1271 	 */
1272 	return idle_core || rq->core->core_cookie == p->core_cookie;
1273 }
1274 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1275 static inline bool sched_group_cookie_match(struct rq *rq,
1276 					    struct task_struct *p,
1277 					    struct sched_group *group)
1278 {
1279 	int cpu;
1280 
1281 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1282 	if (!sched_core_enabled(rq))
1283 		return true;
1284 
1285 	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1286 		if (sched_core_cookie_match(cpu_rq(cpu), p))
1287 			return true;
1288 	}
1289 	return false;
1290 }
1291 
sched_core_enqueued(struct task_struct * p)1292 static inline bool sched_core_enqueued(struct task_struct *p)
1293 {
1294 	return !RB_EMPTY_NODE(&p->core_node);
1295 }
1296 
1297 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1298 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1299 
1300 extern void sched_core_get(void);
1301 extern void sched_core_put(void);
1302 
1303 #else /* !CONFIG_SCHED_CORE */
1304 
sched_core_enabled(struct rq * rq)1305 static inline bool sched_core_enabled(struct rq *rq)
1306 {
1307 	return false;
1308 }
1309 
sched_core_disabled(void)1310 static inline bool sched_core_disabled(void)
1311 {
1312 	return true;
1313 }
1314 
rq_lockp(struct rq * rq)1315 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1316 {
1317 	return &rq->__lock;
1318 }
1319 
__rq_lockp(struct rq * rq)1320 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1321 {
1322 	return &rq->__lock;
1323 }
1324 
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1325 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1326 {
1327 	return true;
1328 }
1329 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1330 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1331 {
1332 	return true;
1333 }
1334 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1335 static inline bool sched_group_cookie_match(struct rq *rq,
1336 					    struct task_struct *p,
1337 					    struct sched_group *group)
1338 {
1339 	return true;
1340 }
1341 #endif /* CONFIG_SCHED_CORE */
1342 
lockdep_assert_rq_held(struct rq * rq)1343 static inline void lockdep_assert_rq_held(struct rq *rq)
1344 {
1345 	lockdep_assert_held(__rq_lockp(rq));
1346 }
1347 
1348 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1349 extern bool raw_spin_rq_trylock(struct rq *rq);
1350 extern void raw_spin_rq_unlock(struct rq *rq);
1351 
raw_spin_rq_lock(struct rq * rq)1352 static inline void raw_spin_rq_lock(struct rq *rq)
1353 {
1354 	raw_spin_rq_lock_nested(rq, 0);
1355 }
1356 
raw_spin_rq_lock_irq(struct rq * rq)1357 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1358 {
1359 	local_irq_disable();
1360 	raw_spin_rq_lock(rq);
1361 }
1362 
raw_spin_rq_unlock_irq(struct rq * rq)1363 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1364 {
1365 	raw_spin_rq_unlock(rq);
1366 	local_irq_enable();
1367 }
1368 
_raw_spin_rq_lock_irqsave(struct rq * rq)1369 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1370 {
1371 	unsigned long flags;
1372 	local_irq_save(flags);
1373 	raw_spin_rq_lock(rq);
1374 	return flags;
1375 }
1376 
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1377 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1378 {
1379 	raw_spin_rq_unlock(rq);
1380 	local_irq_restore(flags);
1381 }
1382 
1383 #define raw_spin_rq_lock_irqsave(rq, flags)	\
1384 do {						\
1385 	flags = _raw_spin_rq_lock_irqsave(rq);	\
1386 } while (0)
1387 
1388 #ifdef CONFIG_SCHED_SMT
1389 extern void __update_idle_core(struct rq *rq);
1390 
update_idle_core(struct rq * rq)1391 static inline void update_idle_core(struct rq *rq)
1392 {
1393 	if (static_branch_unlikely(&sched_smt_present))
1394 		__update_idle_core(rq);
1395 }
1396 
1397 #else
update_idle_core(struct rq * rq)1398 static inline void update_idle_core(struct rq *rq) { }
1399 #endif
1400 
1401 #ifdef CONFIG_FAIR_GROUP_SCHED
task_of(struct sched_entity * se)1402 static inline struct task_struct *task_of(struct sched_entity *se)
1403 {
1404 	SCHED_WARN_ON(!entity_is_task(se));
1405 	return container_of(se, struct task_struct, se);
1406 }
1407 
task_cfs_rq(struct task_struct * p)1408 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1409 {
1410 	return p->se.cfs_rq;
1411 }
1412 
1413 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(struct sched_entity * se)1414 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1415 {
1416 	return se->cfs_rq;
1417 }
1418 
1419 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1420 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1421 {
1422 	return grp->my_q;
1423 }
1424 
1425 #else
1426 
task_of(struct sched_entity * se)1427 static inline struct task_struct *task_of(struct sched_entity *se)
1428 {
1429 	return container_of(se, struct task_struct, se);
1430 }
1431 
task_cfs_rq(struct task_struct * p)1432 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1433 {
1434 	return &task_rq(p)->cfs;
1435 }
1436 
cfs_rq_of(struct sched_entity * se)1437 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1438 {
1439 	struct task_struct *p = task_of(se);
1440 	struct rq *rq = task_rq(p);
1441 
1442 	return &rq->cfs;
1443 }
1444 
1445 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1446 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1447 {
1448 	return NULL;
1449 }
1450 #endif
1451 
1452 extern void update_rq_clock(struct rq *rq);
1453 
1454 /*
1455  * rq::clock_update_flags bits
1456  *
1457  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1458  *  call to __schedule(). This is an optimisation to avoid
1459  *  neighbouring rq clock updates.
1460  *
1461  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1462  *  in effect and calls to update_rq_clock() are being ignored.
1463  *
1464  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1465  *  made to update_rq_clock() since the last time rq::lock was pinned.
1466  *
1467  * If inside of __schedule(), clock_update_flags will have been
1468  * shifted left (a left shift is a cheap operation for the fast path
1469  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1470  *
1471  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1472  *
1473  * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1474  * one position though, because the next rq_unpin_lock() will shift it
1475  * back.
1476  */
1477 #define RQCF_REQ_SKIP		0x01
1478 #define RQCF_ACT_SKIP		0x02
1479 #define RQCF_UPDATED		0x04
1480 
assert_clock_updated(struct rq * rq)1481 static inline void assert_clock_updated(struct rq *rq)
1482 {
1483 	/*
1484 	 * The only reason for not seeing a clock update since the
1485 	 * last rq_pin_lock() is if we're currently skipping updates.
1486 	 */
1487 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1488 }
1489 
rq_clock(struct rq * rq)1490 static inline u64 rq_clock(struct rq *rq)
1491 {
1492 	lockdep_assert_rq_held(rq);
1493 	assert_clock_updated(rq);
1494 
1495 	return rq->clock;
1496 }
1497 
rq_clock_task(struct rq * rq)1498 static inline u64 rq_clock_task(struct rq *rq)
1499 {
1500 	lockdep_assert_rq_held(rq);
1501 	assert_clock_updated(rq);
1502 
1503 	return rq->clock_task;
1504 }
1505 
1506 /**
1507  * By default the decay is the default pelt decay period.
1508  * The decay shift can change the decay period in
1509  * multiples of 32.
1510  *  Decay shift		Decay period(ms)
1511  *	0			32
1512  *	1			64
1513  *	2			128
1514  *	3			256
1515  *	4			512
1516  */
1517 extern int sched_thermal_decay_shift;
1518 
rq_clock_thermal(struct rq * rq)1519 static inline u64 rq_clock_thermal(struct rq *rq)
1520 {
1521 	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1522 }
1523 
rq_clock_skip_update(struct rq * rq)1524 static inline void rq_clock_skip_update(struct rq *rq)
1525 {
1526 	lockdep_assert_rq_held(rq);
1527 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1528 }
1529 
1530 /*
1531  * See rt task throttling, which is the only time a skip
1532  * request is canceled.
1533  */
rq_clock_cancel_skipupdate(struct rq * rq)1534 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1535 {
1536 	lockdep_assert_rq_held(rq);
1537 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1538 }
1539 
1540 struct rq_flags {
1541 	unsigned long flags;
1542 	struct pin_cookie cookie;
1543 #ifdef CONFIG_SCHED_DEBUG
1544 	/*
1545 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1546 	 * current pin context is stashed here in case it needs to be
1547 	 * restored in rq_repin_lock().
1548 	 */
1549 	unsigned int clock_update_flags;
1550 #endif
1551 };
1552 
1553 extern struct balance_callback balance_push_callback;
1554 
1555 /*
1556  * Lockdep annotation that avoids accidental unlocks; it's like a
1557  * sticky/continuous lockdep_assert_held().
1558  *
1559  * This avoids code that has access to 'struct rq *rq' (basically everything in
1560  * the scheduler) from accidentally unlocking the rq if they do not also have a
1561  * copy of the (on-stack) 'struct rq_flags rf'.
1562  *
1563  * Also see Documentation/locking/lockdep-design.rst.
1564  */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1565 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1566 {
1567 	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1568 
1569 #ifdef CONFIG_SCHED_DEBUG
1570 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1571 	rf->clock_update_flags = 0;
1572 #ifdef CONFIG_SMP
1573 	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1574 #endif
1575 #endif
1576 }
1577 
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1578 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1579 {
1580 #ifdef CONFIG_SCHED_DEBUG
1581 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1582 		rf->clock_update_flags = RQCF_UPDATED;
1583 #endif
1584 
1585 	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1586 }
1587 
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1588 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1589 {
1590 	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1591 
1592 #ifdef CONFIG_SCHED_DEBUG
1593 	/*
1594 	 * Restore the value we stashed in @rf for this pin context.
1595 	 */
1596 	rq->clock_update_flags |= rf->clock_update_flags;
1597 #endif
1598 }
1599 
1600 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1601 	__acquires(rq->lock);
1602 
1603 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1604 	__acquires(p->pi_lock)
1605 	__acquires(rq->lock);
1606 
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1607 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1608 	__releases(rq->lock)
1609 {
1610 	rq_unpin_lock(rq, rf);
1611 	raw_spin_rq_unlock(rq);
1612 }
1613 
1614 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1615 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1616 	__releases(rq->lock)
1617 	__releases(p->pi_lock)
1618 {
1619 	rq_unpin_lock(rq, rf);
1620 	raw_spin_rq_unlock(rq);
1621 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1622 }
1623 
1624 static inline void
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1625 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1626 	__acquires(rq->lock)
1627 {
1628 	raw_spin_rq_lock_irqsave(rq, rf->flags);
1629 	rq_pin_lock(rq, rf);
1630 }
1631 
1632 static inline void
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1633 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1634 	__acquires(rq->lock)
1635 {
1636 	raw_spin_rq_lock_irq(rq);
1637 	rq_pin_lock(rq, rf);
1638 }
1639 
1640 static inline void
rq_lock(struct rq * rq,struct rq_flags * rf)1641 rq_lock(struct rq *rq, struct rq_flags *rf)
1642 	__acquires(rq->lock)
1643 {
1644 	raw_spin_rq_lock(rq);
1645 	rq_pin_lock(rq, rf);
1646 }
1647 
1648 static inline void
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1649 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1650 	__releases(rq->lock)
1651 {
1652 	rq_unpin_lock(rq, rf);
1653 	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1654 }
1655 
1656 static inline void
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1657 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1658 	__releases(rq->lock)
1659 {
1660 	rq_unpin_lock(rq, rf);
1661 	raw_spin_rq_unlock_irq(rq);
1662 }
1663 
1664 static inline void
rq_unlock(struct rq * rq,struct rq_flags * rf)1665 rq_unlock(struct rq *rq, struct rq_flags *rf)
1666 	__releases(rq->lock)
1667 {
1668 	rq_unpin_lock(rq, rf);
1669 	raw_spin_rq_unlock(rq);
1670 }
1671 
1672 static inline struct rq *
this_rq_lock_irq(struct rq_flags * rf)1673 this_rq_lock_irq(struct rq_flags *rf)
1674 	__acquires(rq->lock)
1675 {
1676 	struct rq *rq;
1677 
1678 	local_irq_disable();
1679 	rq = this_rq();
1680 	rq_lock(rq, rf);
1681 	return rq;
1682 }
1683 
1684 #ifdef CONFIG_NUMA
1685 enum numa_topology_type {
1686 	NUMA_DIRECT,
1687 	NUMA_GLUELESS_MESH,
1688 	NUMA_BACKPLANE,
1689 };
1690 extern enum numa_topology_type sched_numa_topology_type;
1691 extern int sched_max_numa_distance;
1692 extern bool find_numa_distance(int distance);
1693 extern void sched_init_numa(int offline_node);
1694 extern void sched_update_numa(int cpu, bool online);
1695 extern void sched_domains_numa_masks_set(unsigned int cpu);
1696 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1697 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1698 #else
sched_init_numa(int offline_node)1699 static inline void sched_init_numa(int offline_node) { }
sched_update_numa(int cpu,bool online)1700 static inline void sched_update_numa(int cpu, bool online) { }
sched_domains_numa_masks_set(unsigned int cpu)1701 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1702 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1703 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1704 {
1705 	return nr_cpu_ids;
1706 }
1707 #endif
1708 
1709 #ifdef CONFIG_NUMA_BALANCING
1710 /* The regions in numa_faults array from task_struct */
1711 enum numa_faults_stats {
1712 	NUMA_MEM = 0,
1713 	NUMA_CPU,
1714 	NUMA_MEMBUF,
1715 	NUMA_CPUBUF
1716 };
1717 extern void sched_setnuma(struct task_struct *p, int node);
1718 extern int migrate_task_to(struct task_struct *p, int cpu);
1719 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1720 			int cpu, int scpu);
1721 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1722 #else
1723 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1724 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1725 {
1726 }
1727 #endif /* CONFIG_NUMA_BALANCING */
1728 
1729 #ifdef CONFIG_SMP
1730 
1731 static inline void
queue_balance_callback(struct rq * rq,struct balance_callback * head,void (* func)(struct rq * rq))1732 queue_balance_callback(struct rq *rq,
1733 		       struct balance_callback *head,
1734 		       void (*func)(struct rq *rq))
1735 {
1736 	lockdep_assert_rq_held(rq);
1737 
1738 	/*
1739 	 * Don't (re)queue an already queued item; nor queue anything when
1740 	 * balance_push() is active, see the comment with
1741 	 * balance_push_callback.
1742 	 */
1743 	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1744 		return;
1745 
1746 	head->func = func;
1747 	head->next = rq->balance_callback;
1748 	rq->balance_callback = head;
1749 }
1750 
1751 #define rcu_dereference_check_sched_domain(p) \
1752 	rcu_dereference_check((p), \
1753 			      lockdep_is_held(&sched_domains_mutex))
1754 
1755 /*
1756  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1757  * See destroy_sched_domains: call_rcu for details.
1758  *
1759  * The domain tree of any CPU may only be accessed from within
1760  * preempt-disabled sections.
1761  */
1762 #define for_each_domain(cpu, __sd) \
1763 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1764 			__sd; __sd = __sd->parent)
1765 
1766 /**
1767  * highest_flag_domain - Return highest sched_domain containing flag.
1768  * @cpu:	The CPU whose highest level of sched domain is to
1769  *		be returned.
1770  * @flag:	The flag to check for the highest sched_domain
1771  *		for the given CPU.
1772  *
1773  * Returns the highest sched_domain of a CPU which contains the given flag.
1774  */
highest_flag_domain(int cpu,int flag)1775 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1776 {
1777 	struct sched_domain *sd, *hsd = NULL;
1778 
1779 	for_each_domain(cpu, sd) {
1780 		if (!(sd->flags & flag))
1781 			break;
1782 		hsd = sd;
1783 	}
1784 
1785 	return hsd;
1786 }
1787 
lowest_flag_domain(int cpu,int flag)1788 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1789 {
1790 	struct sched_domain *sd;
1791 
1792 	for_each_domain(cpu, sd) {
1793 		if (sd->flags & flag)
1794 			break;
1795 	}
1796 
1797 	return sd;
1798 }
1799 
1800 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1801 DECLARE_PER_CPU(int, sd_llc_size);
1802 DECLARE_PER_CPU(int, sd_llc_id);
1803 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1804 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1805 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1806 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1807 extern struct static_key_false sched_asym_cpucapacity;
1808 
sched_asym_cpucap_active(void)1809 static __always_inline bool sched_asym_cpucap_active(void)
1810 {
1811 	return static_branch_unlikely(&sched_asym_cpucapacity);
1812 }
1813 
1814 struct sched_group_capacity {
1815 	atomic_t		ref;
1816 	/*
1817 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1818 	 * for a single CPU.
1819 	 */
1820 	unsigned long		capacity;
1821 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1822 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1823 	unsigned long		next_update;
1824 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1825 
1826 #ifdef CONFIG_SCHED_DEBUG
1827 	int			id;
1828 #endif
1829 
1830 	unsigned long		cpumask[];		/* Balance mask */
1831 };
1832 
1833 struct sched_group {
1834 	struct sched_group	*next;			/* Must be a circular list */
1835 	atomic_t		ref;
1836 
1837 	unsigned int		group_weight;
1838 	struct sched_group_capacity *sgc;
1839 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1840 	int			flags;
1841 
1842 	/*
1843 	 * The CPUs this group covers.
1844 	 *
1845 	 * NOTE: this field is variable length. (Allocated dynamically
1846 	 * by attaching extra space to the end of the structure,
1847 	 * depending on how many CPUs the kernel has booted up with)
1848 	 */
1849 	unsigned long		cpumask[];
1850 };
1851 
sched_group_span(struct sched_group * sg)1852 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1853 {
1854 	return to_cpumask(sg->cpumask);
1855 }
1856 
1857 /*
1858  * See build_balance_mask().
1859  */
group_balance_mask(struct sched_group * sg)1860 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1861 {
1862 	return to_cpumask(sg->sgc->cpumask);
1863 }
1864 
1865 extern int group_balance_cpu(struct sched_group *sg);
1866 
1867 #ifdef CONFIG_SCHED_DEBUG
1868 void update_sched_domain_debugfs(void);
1869 void dirty_sched_domain_sysctl(int cpu);
1870 #else
update_sched_domain_debugfs(void)1871 static inline void update_sched_domain_debugfs(void)
1872 {
1873 }
dirty_sched_domain_sysctl(int cpu)1874 static inline void dirty_sched_domain_sysctl(int cpu)
1875 {
1876 }
1877 #endif
1878 
1879 extern int sched_update_scaling(void);
1880 #endif /* CONFIG_SMP */
1881 
1882 #include "stats.h"
1883 
1884 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1885 
1886 extern void __sched_core_account_forceidle(struct rq *rq);
1887 
sched_core_account_forceidle(struct rq * rq)1888 static inline void sched_core_account_forceidle(struct rq *rq)
1889 {
1890 	if (schedstat_enabled())
1891 		__sched_core_account_forceidle(rq);
1892 }
1893 
1894 extern void __sched_core_tick(struct rq *rq);
1895 
sched_core_tick(struct rq * rq)1896 static inline void sched_core_tick(struct rq *rq)
1897 {
1898 	if (sched_core_enabled(rq) && schedstat_enabled())
1899 		__sched_core_tick(rq);
1900 }
1901 
1902 #else
1903 
sched_core_account_forceidle(struct rq * rq)1904 static inline void sched_core_account_forceidle(struct rq *rq) {}
1905 
sched_core_tick(struct rq * rq)1906 static inline void sched_core_tick(struct rq *rq) {}
1907 
1908 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1909 
1910 #ifdef CONFIG_CGROUP_SCHED
1911 
1912 /*
1913  * Return the group to which this tasks belongs.
1914  *
1915  * We cannot use task_css() and friends because the cgroup subsystem
1916  * changes that value before the cgroup_subsys::attach() method is called,
1917  * therefore we cannot pin it and might observe the wrong value.
1918  *
1919  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1920  * core changes this before calling sched_move_task().
1921  *
1922  * Instead we use a 'copy' which is updated from sched_move_task() while
1923  * holding both task_struct::pi_lock and rq::lock.
1924  */
task_group(struct task_struct * p)1925 static inline struct task_group *task_group(struct task_struct *p)
1926 {
1927 	return p->sched_task_group;
1928 }
1929 
1930 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
set_task_rq(struct task_struct * p,unsigned int cpu)1931 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1932 {
1933 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1934 	struct task_group *tg = task_group(p);
1935 #endif
1936 
1937 #ifdef CONFIG_FAIR_GROUP_SCHED
1938 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1939 	p->se.cfs_rq = tg->cfs_rq[cpu];
1940 	p->se.parent = tg->se[cpu];
1941 	p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
1942 #endif
1943 
1944 #ifdef CONFIG_RT_GROUP_SCHED
1945 	p->rt.rt_rq  = tg->rt_rq[cpu];
1946 	p->rt.parent = tg->rt_se[cpu];
1947 #endif
1948 }
1949 
1950 #else /* CONFIG_CGROUP_SCHED */
1951 
set_task_rq(struct task_struct * p,unsigned int cpu)1952 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
task_group(struct task_struct * p)1953 static inline struct task_group *task_group(struct task_struct *p)
1954 {
1955 	return NULL;
1956 }
1957 
1958 #endif /* CONFIG_CGROUP_SCHED */
1959 
__set_task_cpu(struct task_struct * p,unsigned int cpu)1960 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1961 {
1962 	set_task_rq(p, cpu);
1963 #ifdef CONFIG_SMP
1964 	/*
1965 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1966 	 * successfully executed on another CPU. We must ensure that updates of
1967 	 * per-task data have been completed by this moment.
1968 	 */
1969 	smp_wmb();
1970 	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1971 	p->wake_cpu = cpu;
1972 #endif
1973 }
1974 
1975 /*
1976  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1977  */
1978 #ifdef CONFIG_SCHED_DEBUG
1979 # define const_debug __read_mostly
1980 #else
1981 # define const_debug const
1982 #endif
1983 
1984 #define SCHED_FEAT(name, enabled)	\
1985 	__SCHED_FEAT_##name ,
1986 
1987 enum {
1988 #include "features.h"
1989 	__SCHED_FEAT_NR,
1990 };
1991 
1992 #undef SCHED_FEAT
1993 
1994 #ifdef CONFIG_SCHED_DEBUG
1995 
1996 /*
1997  * To support run-time toggling of sched features, all the translation units
1998  * (but core.c) reference the sysctl_sched_features defined in core.c.
1999  */
2000 extern const_debug unsigned int sysctl_sched_features;
2001 
2002 #ifdef CONFIG_JUMP_LABEL
2003 #define SCHED_FEAT(name, enabled)					\
2004 static __always_inline bool static_branch_##name(struct static_key *key) \
2005 {									\
2006 	return static_key_##enabled(key);				\
2007 }
2008 
2009 #include "features.h"
2010 #undef SCHED_FEAT
2011 
2012 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2013 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2014 
2015 #else /* !CONFIG_JUMP_LABEL */
2016 
2017 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2018 
2019 #endif /* CONFIG_JUMP_LABEL */
2020 
2021 #else /* !SCHED_DEBUG */
2022 
2023 /*
2024  * Each translation unit has its own copy of sysctl_sched_features to allow
2025  * constants propagation at compile time and compiler optimization based on
2026  * features default.
2027  */
2028 #define SCHED_FEAT(name, enabled)	\
2029 	(1UL << __SCHED_FEAT_##name) * enabled |
2030 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2031 #include "features.h"
2032 	0;
2033 #undef SCHED_FEAT
2034 
2035 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2036 
2037 #endif /* SCHED_DEBUG */
2038 
2039 extern struct static_key_false sched_numa_balancing;
2040 extern struct static_key_false sched_schedstats;
2041 
global_rt_period(void)2042 static inline u64 global_rt_period(void)
2043 {
2044 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2045 }
2046 
global_rt_runtime(void)2047 static inline u64 global_rt_runtime(void)
2048 {
2049 	if (sysctl_sched_rt_runtime < 0)
2050 		return RUNTIME_INF;
2051 
2052 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2053 }
2054 
task_current(struct rq * rq,struct task_struct * p)2055 static inline int task_current(struct rq *rq, struct task_struct *p)
2056 {
2057 	return rq->curr == p;
2058 }
2059 
task_on_cpu(struct rq * rq,struct task_struct * p)2060 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2061 {
2062 #ifdef CONFIG_SMP
2063 	return p->on_cpu;
2064 #else
2065 	return task_current(rq, p);
2066 #endif
2067 }
2068 
task_on_rq_queued(struct task_struct * p)2069 static inline int task_on_rq_queued(struct task_struct *p)
2070 {
2071 	return p->on_rq == TASK_ON_RQ_QUEUED;
2072 }
2073 
task_on_rq_migrating(struct task_struct * p)2074 static inline int task_on_rq_migrating(struct task_struct *p)
2075 {
2076 	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2077 }
2078 
2079 /* Wake flags. The first three directly map to some SD flag value */
2080 #define WF_EXEC     0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2081 #define WF_FORK     0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2082 #define WF_TTWU     0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2083 
2084 #define WF_SYNC     0x10 /* Waker goes to sleep after wakeup */
2085 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2086 
2087 #ifdef CONFIG_SMP
2088 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2089 static_assert(WF_FORK == SD_BALANCE_FORK);
2090 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2091 #endif
2092 
2093 /*
2094  * To aid in avoiding the subversion of "niceness" due to uneven distribution
2095  * of tasks with abnormal "nice" values across CPUs the contribution that
2096  * each task makes to its run queue's load is weighted according to its
2097  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2098  * scaled version of the new time slice allocation that they receive on time
2099  * slice expiry etc.
2100  */
2101 
2102 #define WEIGHT_IDLEPRIO		3
2103 #define WMULT_IDLEPRIO		1431655765
2104 
2105 extern const int		sched_prio_to_weight[40];
2106 extern const u32		sched_prio_to_wmult[40];
2107 
2108 /*
2109  * {de,en}queue flags:
2110  *
2111  * DEQUEUE_SLEEP  - task is no longer runnable
2112  * ENQUEUE_WAKEUP - task just became runnable
2113  *
2114  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2115  *                are in a known state which allows modification. Such pairs
2116  *                should preserve as much state as possible.
2117  *
2118  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2119  *        in the runqueue.
2120  *
2121  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2122  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2123  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2124  *
2125  */
2126 
2127 #define DEQUEUE_SLEEP		0x01
2128 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2129 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2130 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
2131 
2132 #define ENQUEUE_WAKEUP		0x01
2133 #define ENQUEUE_RESTORE		0x02
2134 #define ENQUEUE_MOVE		0x04
2135 #define ENQUEUE_NOCLOCK		0x08
2136 
2137 #define ENQUEUE_HEAD		0x10
2138 #define ENQUEUE_REPLENISH	0x20
2139 #ifdef CONFIG_SMP
2140 #define ENQUEUE_MIGRATED	0x40
2141 #else
2142 #define ENQUEUE_MIGRATED	0x00
2143 #endif
2144 
2145 #define RETRY_TASK		((void *)-1UL)
2146 
2147 struct sched_class {
2148 
2149 #ifdef CONFIG_UCLAMP_TASK
2150 	int uclamp_enabled;
2151 #endif
2152 
2153 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2154 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2155 	void (*yield_task)   (struct rq *rq);
2156 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2157 
2158 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2159 
2160 	struct task_struct *(*pick_next_task)(struct rq *rq);
2161 
2162 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2163 	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2164 
2165 #ifdef CONFIG_SMP
2166 	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2167 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2168 
2169 	struct task_struct * (*pick_task)(struct rq *rq);
2170 
2171 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2172 
2173 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2174 
2175 	void (*set_cpus_allowed)(struct task_struct *p,
2176 				 const struct cpumask *newmask,
2177 				 u32 flags);
2178 
2179 	void (*rq_online)(struct rq *rq);
2180 	void (*rq_offline)(struct rq *rq);
2181 
2182 	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2183 #endif
2184 
2185 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2186 	void (*task_fork)(struct task_struct *p);
2187 	void (*task_dead)(struct task_struct *p);
2188 
2189 	/*
2190 	 * The switched_from() call is allowed to drop rq->lock, therefore we
2191 	 * cannot assume the switched_from/switched_to pair is serialized by
2192 	 * rq->lock. They are however serialized by p->pi_lock.
2193 	 */
2194 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2195 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2196 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2197 			      int oldprio);
2198 
2199 	unsigned int (*get_rr_interval)(struct rq *rq,
2200 					struct task_struct *task);
2201 
2202 	void (*update_curr)(struct rq *rq);
2203 
2204 #ifdef CONFIG_FAIR_GROUP_SCHED
2205 	void (*task_change_group)(struct task_struct *p);
2206 #endif
2207 };
2208 
put_prev_task(struct rq * rq,struct task_struct * prev)2209 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2210 {
2211 	WARN_ON_ONCE(rq->curr != prev);
2212 	prev->sched_class->put_prev_task(rq, prev);
2213 }
2214 
set_next_task(struct rq * rq,struct task_struct * next)2215 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2216 {
2217 	next->sched_class->set_next_task(rq, next, false);
2218 }
2219 
2220 
2221 /*
2222  * Helper to define a sched_class instance; each one is placed in a separate
2223  * section which is ordered by the linker script:
2224  *
2225  *   include/asm-generic/vmlinux.lds.h
2226  *
2227  * *CAREFUL* they are laid out in *REVERSE* order!!!
2228  *
2229  * Also enforce alignment on the instance, not the type, to guarantee layout.
2230  */
2231 #define DEFINE_SCHED_CLASS(name) \
2232 const struct sched_class name##_sched_class \
2233 	__aligned(__alignof__(struct sched_class)) \
2234 	__section("__" #name "_sched_class")
2235 
2236 /* Defined in include/asm-generic/vmlinux.lds.h */
2237 extern struct sched_class __sched_class_highest[];
2238 extern struct sched_class __sched_class_lowest[];
2239 
2240 #define for_class_range(class, _from, _to) \
2241 	for (class = (_from); class < (_to); class++)
2242 
2243 #define for_each_class(class) \
2244 	for_class_range(class, __sched_class_highest, __sched_class_lowest)
2245 
2246 #define sched_class_above(_a, _b)	((_a) < (_b))
2247 
2248 extern const struct sched_class stop_sched_class;
2249 extern const struct sched_class dl_sched_class;
2250 extern const struct sched_class rt_sched_class;
2251 extern const struct sched_class fair_sched_class;
2252 extern const struct sched_class idle_sched_class;
2253 
sched_stop_runnable(struct rq * rq)2254 static inline bool sched_stop_runnable(struct rq *rq)
2255 {
2256 	return rq->stop && task_on_rq_queued(rq->stop);
2257 }
2258 
sched_dl_runnable(struct rq * rq)2259 static inline bool sched_dl_runnable(struct rq *rq)
2260 {
2261 	return rq->dl.dl_nr_running > 0;
2262 }
2263 
sched_rt_runnable(struct rq * rq)2264 static inline bool sched_rt_runnable(struct rq *rq)
2265 {
2266 	return rq->rt.rt_queued > 0;
2267 }
2268 
sched_fair_runnable(struct rq * rq)2269 static inline bool sched_fair_runnable(struct rq *rq)
2270 {
2271 	return rq->cfs.nr_running > 0;
2272 }
2273 
2274 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2275 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2276 
2277 #define SCA_CHECK		0x01
2278 #define SCA_MIGRATE_DISABLE	0x02
2279 #define SCA_MIGRATE_ENABLE	0x04
2280 #define SCA_USER		0x08
2281 
2282 #ifdef CONFIG_SMP
2283 
2284 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2285 
2286 extern void trigger_load_balance(struct rq *rq);
2287 
2288 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2289 
get_push_task(struct rq * rq)2290 static inline struct task_struct *get_push_task(struct rq *rq)
2291 {
2292 	struct task_struct *p = rq->curr;
2293 
2294 	lockdep_assert_rq_held(rq);
2295 
2296 	if (rq->push_busy)
2297 		return NULL;
2298 
2299 	if (p->nr_cpus_allowed == 1)
2300 		return NULL;
2301 
2302 	if (p->migration_disabled)
2303 		return NULL;
2304 
2305 	rq->push_busy = true;
2306 	return get_task_struct(p);
2307 }
2308 
2309 extern int push_cpu_stop(void *arg);
2310 
2311 #endif
2312 
2313 #ifdef CONFIG_CPU_IDLE
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2314 static inline void idle_set_state(struct rq *rq,
2315 				  struct cpuidle_state *idle_state)
2316 {
2317 	rq->idle_state = idle_state;
2318 }
2319 
idle_get_state(struct rq * rq)2320 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2321 {
2322 	SCHED_WARN_ON(!rcu_read_lock_held());
2323 
2324 	return rq->idle_state;
2325 }
2326 #else
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2327 static inline void idle_set_state(struct rq *rq,
2328 				  struct cpuidle_state *idle_state)
2329 {
2330 }
2331 
idle_get_state(struct rq * rq)2332 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2333 {
2334 	return NULL;
2335 }
2336 #endif
2337 
2338 extern void schedule_idle(void);
2339 
2340 extern void sysrq_sched_debug_show(void);
2341 extern void sched_init_granularity(void);
2342 extern void update_max_interval(void);
2343 
2344 extern void init_sched_dl_class(void);
2345 extern void init_sched_rt_class(void);
2346 extern void init_sched_fair_class(void);
2347 
2348 extern void reweight_task(struct task_struct *p, int prio);
2349 
2350 extern void resched_curr(struct rq *rq);
2351 extern void resched_cpu(int cpu);
2352 
2353 extern struct rt_bandwidth def_rt_bandwidth;
2354 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2355 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2356 
2357 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2358 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2359 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2360 
2361 #define BW_SHIFT		20
2362 #define BW_UNIT			(1 << BW_SHIFT)
2363 #define RATIO_SHIFT		8
2364 #define MAX_BW_BITS		(64 - BW_SHIFT)
2365 #define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2366 unsigned long to_ratio(u64 period, u64 runtime);
2367 
2368 extern void init_entity_runnable_average(struct sched_entity *se);
2369 extern void post_init_entity_util_avg(struct task_struct *p);
2370 
2371 #ifdef CONFIG_NO_HZ_FULL
2372 extern bool sched_can_stop_tick(struct rq *rq);
2373 extern int __init sched_tick_offload_init(void);
2374 
2375 /*
2376  * Tick may be needed by tasks in the runqueue depending on their policy and
2377  * requirements. If tick is needed, lets send the target an IPI to kick it out of
2378  * nohz mode if necessary.
2379  */
sched_update_tick_dependency(struct rq * rq)2380 static inline void sched_update_tick_dependency(struct rq *rq)
2381 {
2382 	int cpu = cpu_of(rq);
2383 
2384 	if (!tick_nohz_full_cpu(cpu))
2385 		return;
2386 
2387 	if (sched_can_stop_tick(rq))
2388 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2389 	else
2390 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2391 }
2392 #else
sched_tick_offload_init(void)2393 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2394 static inline void sched_update_tick_dependency(struct rq *rq) { }
2395 #endif
2396 
add_nr_running(struct rq * rq,unsigned count)2397 static inline void add_nr_running(struct rq *rq, unsigned count)
2398 {
2399 	unsigned prev_nr = rq->nr_running;
2400 
2401 	rq->nr_running = prev_nr + count;
2402 	if (trace_sched_update_nr_running_tp_enabled()) {
2403 		call_trace_sched_update_nr_running(rq, count);
2404 	}
2405 
2406 #ifdef CONFIG_SMP
2407 	if (prev_nr < 2 && rq->nr_running >= 2) {
2408 		if (!READ_ONCE(rq->rd->overload))
2409 			WRITE_ONCE(rq->rd->overload, 1);
2410 	}
2411 #endif
2412 
2413 	sched_update_tick_dependency(rq);
2414 }
2415 
sub_nr_running(struct rq * rq,unsigned count)2416 static inline void sub_nr_running(struct rq *rq, unsigned count)
2417 {
2418 	rq->nr_running -= count;
2419 	if (trace_sched_update_nr_running_tp_enabled()) {
2420 		call_trace_sched_update_nr_running(rq, -count);
2421 	}
2422 
2423 	/* Check if we still need preemption */
2424 	sched_update_tick_dependency(rq);
2425 }
2426 
2427 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2428 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2429 
2430 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2431 
2432 #ifdef CONFIG_PREEMPT_RT
2433 #define SCHED_NR_MIGRATE_BREAK 8
2434 #else
2435 #define SCHED_NR_MIGRATE_BREAK 32
2436 #endif
2437 
2438 extern const_debug unsigned int sysctl_sched_nr_migrate;
2439 extern const_debug unsigned int sysctl_sched_migration_cost;
2440 
2441 #ifdef CONFIG_SCHED_DEBUG
2442 extern unsigned int sysctl_sched_latency;
2443 extern unsigned int sysctl_sched_min_granularity;
2444 extern unsigned int sysctl_sched_idle_min_granularity;
2445 extern unsigned int sysctl_sched_wakeup_granularity;
2446 extern int sysctl_resched_latency_warn_ms;
2447 extern int sysctl_resched_latency_warn_once;
2448 
2449 extern unsigned int sysctl_sched_tunable_scaling;
2450 
2451 extern unsigned int sysctl_numa_balancing_scan_delay;
2452 extern unsigned int sysctl_numa_balancing_scan_period_min;
2453 extern unsigned int sysctl_numa_balancing_scan_period_max;
2454 extern unsigned int sysctl_numa_balancing_scan_size;
2455 extern unsigned int sysctl_numa_balancing_hot_threshold;
2456 #endif
2457 
2458 #ifdef CONFIG_SCHED_HRTICK
2459 
2460 /*
2461  * Use hrtick when:
2462  *  - enabled by features
2463  *  - hrtimer is actually high res
2464  */
hrtick_enabled(struct rq * rq)2465 static inline int hrtick_enabled(struct rq *rq)
2466 {
2467 	if (!cpu_active(cpu_of(rq)))
2468 		return 0;
2469 	return hrtimer_is_hres_active(&rq->hrtick_timer);
2470 }
2471 
hrtick_enabled_fair(struct rq * rq)2472 static inline int hrtick_enabled_fair(struct rq *rq)
2473 {
2474 	if (!sched_feat(HRTICK))
2475 		return 0;
2476 	return hrtick_enabled(rq);
2477 }
2478 
hrtick_enabled_dl(struct rq * rq)2479 static inline int hrtick_enabled_dl(struct rq *rq)
2480 {
2481 	if (!sched_feat(HRTICK_DL))
2482 		return 0;
2483 	return hrtick_enabled(rq);
2484 }
2485 
2486 void hrtick_start(struct rq *rq, u64 delay);
2487 
2488 #else
2489 
hrtick_enabled_fair(struct rq * rq)2490 static inline int hrtick_enabled_fair(struct rq *rq)
2491 {
2492 	return 0;
2493 }
2494 
hrtick_enabled_dl(struct rq * rq)2495 static inline int hrtick_enabled_dl(struct rq *rq)
2496 {
2497 	return 0;
2498 }
2499 
hrtick_enabled(struct rq * rq)2500 static inline int hrtick_enabled(struct rq *rq)
2501 {
2502 	return 0;
2503 }
2504 
2505 #endif /* CONFIG_SCHED_HRTICK */
2506 
2507 #ifndef arch_scale_freq_tick
2508 static __always_inline
arch_scale_freq_tick(void)2509 void arch_scale_freq_tick(void)
2510 {
2511 }
2512 #endif
2513 
2514 #ifndef arch_scale_freq_capacity
2515 /**
2516  * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2517  * @cpu: the CPU in question.
2518  *
2519  * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2520  *
2521  *     f_curr
2522  *     ------ * SCHED_CAPACITY_SCALE
2523  *     f_max
2524  */
2525 static __always_inline
arch_scale_freq_capacity(int cpu)2526 unsigned long arch_scale_freq_capacity(int cpu)
2527 {
2528 	return SCHED_CAPACITY_SCALE;
2529 }
2530 #endif
2531 
2532 #ifdef CONFIG_SCHED_DEBUG
2533 /*
2534  * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2535  * acquire rq lock instead of rq_lock(). So at the end of these two functions
2536  * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2537  * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2538  */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2539 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2540 {
2541 	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2542 	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2543 #ifdef CONFIG_SMP
2544 	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2545 #endif
2546 }
2547 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2548 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2549 #endif
2550 
2551 #ifdef CONFIG_SMP
2552 
rq_order_less(struct rq * rq1,struct rq * rq2)2553 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2554 {
2555 #ifdef CONFIG_SCHED_CORE
2556 	/*
2557 	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2558 	 * order by core-id first and cpu-id second.
2559 	 *
2560 	 * Notably:
2561 	 *
2562 	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2563 	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2564 	 *
2565 	 * when only cpu-id is considered.
2566 	 */
2567 	if (rq1->core->cpu < rq2->core->cpu)
2568 		return true;
2569 	if (rq1->core->cpu > rq2->core->cpu)
2570 		return false;
2571 
2572 	/*
2573 	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2574 	 */
2575 #endif
2576 	return rq1->cpu < rq2->cpu;
2577 }
2578 
2579 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2580 
2581 #ifdef CONFIG_PREEMPTION
2582 
2583 /*
2584  * fair double_lock_balance: Safely acquires both rq->locks in a fair
2585  * way at the expense of forcing extra atomic operations in all
2586  * invocations.  This assures that the double_lock is acquired using the
2587  * same underlying policy as the spinlock_t on this architecture, which
2588  * reduces latency compared to the unfair variant below.  However, it
2589  * also adds more overhead and therefore may reduce throughput.
2590  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2591 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2592 	__releases(this_rq->lock)
2593 	__acquires(busiest->lock)
2594 	__acquires(this_rq->lock)
2595 {
2596 	raw_spin_rq_unlock(this_rq);
2597 	double_rq_lock(this_rq, busiest);
2598 
2599 	return 1;
2600 }
2601 
2602 #else
2603 /*
2604  * Unfair double_lock_balance: Optimizes throughput at the expense of
2605  * latency by eliminating extra atomic operations when the locks are
2606  * already in proper order on entry.  This favors lower CPU-ids and will
2607  * grant the double lock to lower CPUs over higher ids under contention,
2608  * regardless of entry order into the function.
2609  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2610 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2611 	__releases(this_rq->lock)
2612 	__acquires(busiest->lock)
2613 	__acquires(this_rq->lock)
2614 {
2615 	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2616 	    likely(raw_spin_rq_trylock(busiest))) {
2617 		double_rq_clock_clear_update(this_rq, busiest);
2618 		return 0;
2619 	}
2620 
2621 	if (rq_order_less(this_rq, busiest)) {
2622 		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2623 		double_rq_clock_clear_update(this_rq, busiest);
2624 		return 0;
2625 	}
2626 
2627 	raw_spin_rq_unlock(this_rq);
2628 	double_rq_lock(this_rq, busiest);
2629 
2630 	return 1;
2631 }
2632 
2633 #endif /* CONFIG_PREEMPTION */
2634 
2635 /*
2636  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2637  */
double_lock_balance(struct rq * this_rq,struct rq * busiest)2638 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2639 {
2640 	lockdep_assert_irqs_disabled();
2641 
2642 	return _double_lock_balance(this_rq, busiest);
2643 }
2644 
double_unlock_balance(struct rq * this_rq,struct rq * busiest)2645 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2646 	__releases(busiest->lock)
2647 {
2648 	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2649 		raw_spin_rq_unlock(busiest);
2650 	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2651 }
2652 
double_lock(spinlock_t * l1,spinlock_t * l2)2653 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2654 {
2655 	if (l1 > l2)
2656 		swap(l1, l2);
2657 
2658 	spin_lock(l1);
2659 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2660 }
2661 
double_lock_irq(spinlock_t * l1,spinlock_t * l2)2662 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2663 {
2664 	if (l1 > l2)
2665 		swap(l1, l2);
2666 
2667 	spin_lock_irq(l1);
2668 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2669 }
2670 
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)2671 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2672 {
2673 	if (l1 > l2)
2674 		swap(l1, l2);
2675 
2676 	raw_spin_lock(l1);
2677 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2678 }
2679 
2680 /*
2681  * double_rq_unlock - safely unlock two runqueues
2682  *
2683  * Note this does not restore interrupts like task_rq_unlock,
2684  * you need to do so manually after calling.
2685  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2686 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2687 	__releases(rq1->lock)
2688 	__releases(rq2->lock)
2689 {
2690 	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2691 		raw_spin_rq_unlock(rq2);
2692 	else
2693 		__release(rq2->lock);
2694 	raw_spin_rq_unlock(rq1);
2695 }
2696 
2697 extern void set_rq_online (struct rq *rq);
2698 extern void set_rq_offline(struct rq *rq);
2699 extern bool sched_smp_initialized;
2700 
2701 #else /* CONFIG_SMP */
2702 
2703 /*
2704  * double_rq_lock - safely lock two runqueues
2705  *
2706  * Note this does not disable interrupts like task_rq_lock,
2707  * you need to do so manually before calling.
2708  */
double_rq_lock(struct rq * rq1,struct rq * rq2)2709 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2710 	__acquires(rq1->lock)
2711 	__acquires(rq2->lock)
2712 {
2713 	WARN_ON_ONCE(!irqs_disabled());
2714 	WARN_ON_ONCE(rq1 != rq2);
2715 	raw_spin_rq_lock(rq1);
2716 	__acquire(rq2->lock);	/* Fake it out ;) */
2717 	double_rq_clock_clear_update(rq1, rq2);
2718 }
2719 
2720 /*
2721  * double_rq_unlock - safely unlock two runqueues
2722  *
2723  * Note this does not restore interrupts like task_rq_unlock,
2724  * you need to do so manually after calling.
2725  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2726 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2727 	__releases(rq1->lock)
2728 	__releases(rq2->lock)
2729 {
2730 	WARN_ON_ONCE(rq1 != rq2);
2731 	raw_spin_rq_unlock(rq1);
2732 	__release(rq2->lock);
2733 }
2734 
2735 #endif
2736 
2737 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2738 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2739 
2740 #ifdef	CONFIG_SCHED_DEBUG
2741 extern bool sched_debug_verbose;
2742 
2743 extern void print_cfs_stats(struct seq_file *m, int cpu);
2744 extern void print_rt_stats(struct seq_file *m, int cpu);
2745 extern void print_dl_stats(struct seq_file *m, int cpu);
2746 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2747 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2748 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2749 
2750 extern void resched_latency_warn(int cpu, u64 latency);
2751 #ifdef CONFIG_NUMA_BALANCING
2752 extern void
2753 show_numa_stats(struct task_struct *p, struct seq_file *m);
2754 extern void
2755 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2756 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2757 #endif /* CONFIG_NUMA_BALANCING */
2758 #else
resched_latency_warn(int cpu,u64 latency)2759 static inline void resched_latency_warn(int cpu, u64 latency) {}
2760 #endif /* CONFIG_SCHED_DEBUG */
2761 
2762 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2763 extern void init_rt_rq(struct rt_rq *rt_rq);
2764 extern void init_dl_rq(struct dl_rq *dl_rq);
2765 
2766 extern void cfs_bandwidth_usage_inc(void);
2767 extern void cfs_bandwidth_usage_dec(void);
2768 
2769 #ifdef CONFIG_NO_HZ_COMMON
2770 #define NOHZ_BALANCE_KICK_BIT	0
2771 #define NOHZ_STATS_KICK_BIT	1
2772 #define NOHZ_NEWILB_KICK_BIT	2
2773 #define NOHZ_NEXT_KICK_BIT	3
2774 
2775 /* Run rebalance_domains() */
2776 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2777 /* Update blocked load */
2778 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2779 /* Update blocked load when entering idle */
2780 #define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2781 /* Update nohz.next_balance */
2782 #define NOHZ_NEXT_KICK		BIT(NOHZ_NEXT_KICK_BIT)
2783 
2784 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2785 
2786 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2787 
2788 extern void nohz_balance_exit_idle(struct rq *rq);
2789 #else
nohz_balance_exit_idle(struct rq * rq)2790 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2791 #endif
2792 
2793 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2794 extern void nohz_run_idle_balance(int cpu);
2795 #else
nohz_run_idle_balance(int cpu)2796 static inline void nohz_run_idle_balance(int cpu) { }
2797 #endif
2798 
2799 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2800 struct irqtime {
2801 	u64			total;
2802 	u64			tick_delta;
2803 	u64			irq_start_time;
2804 	struct u64_stats_sync	sync;
2805 };
2806 
2807 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2808 
2809 /*
2810  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2811  * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2812  * and never move forward.
2813  */
irq_time_read(int cpu)2814 static inline u64 irq_time_read(int cpu)
2815 {
2816 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2817 	unsigned int seq;
2818 	u64 total;
2819 
2820 	do {
2821 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2822 		total = irqtime->total;
2823 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2824 
2825 	return total;
2826 }
2827 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2828 
2829 #ifdef CONFIG_CPU_FREQ
2830 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2831 
2832 /**
2833  * cpufreq_update_util - Take a note about CPU utilization changes.
2834  * @rq: Runqueue to carry out the update for.
2835  * @flags: Update reason flags.
2836  *
2837  * This function is called by the scheduler on the CPU whose utilization is
2838  * being updated.
2839  *
2840  * It can only be called from RCU-sched read-side critical sections.
2841  *
2842  * The way cpufreq is currently arranged requires it to evaluate the CPU
2843  * performance state (frequency/voltage) on a regular basis to prevent it from
2844  * being stuck in a completely inadequate performance level for too long.
2845  * That is not guaranteed to happen if the updates are only triggered from CFS
2846  * and DL, though, because they may not be coming in if only RT tasks are
2847  * active all the time (or there are RT tasks only).
2848  *
2849  * As a workaround for that issue, this function is called periodically by the
2850  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2851  * but that really is a band-aid.  Going forward it should be replaced with
2852  * solutions targeted more specifically at RT tasks.
2853  */
cpufreq_update_util(struct rq * rq,unsigned int flags)2854 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2855 {
2856 	struct update_util_data *data;
2857 
2858 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2859 						  cpu_of(rq)));
2860 	if (data)
2861 		data->func(data, rq_clock(rq), flags);
2862 }
2863 #else
cpufreq_update_util(struct rq * rq,unsigned int flags)2864 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2865 #endif /* CONFIG_CPU_FREQ */
2866 
2867 #ifdef arch_scale_freq_capacity
2868 # ifndef arch_scale_freq_invariant
2869 #  define arch_scale_freq_invariant()	true
2870 # endif
2871 #else
2872 # define arch_scale_freq_invariant()	false
2873 #endif
2874 
2875 #ifdef CONFIG_SMP
capacity_orig_of(int cpu)2876 static inline unsigned long capacity_orig_of(int cpu)
2877 {
2878 	return cpu_rq(cpu)->cpu_capacity_orig;
2879 }
2880 
2881 /**
2882  * enum cpu_util_type - CPU utilization type
2883  * @FREQUENCY_UTIL:	Utilization used to select frequency
2884  * @ENERGY_UTIL:	Utilization used during energy calculation
2885  *
2886  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2887  * need to be aggregated differently depending on the usage made of them. This
2888  * enum is used within effective_cpu_util() to differentiate the types of
2889  * utilization expected by the callers, and adjust the aggregation accordingly.
2890  */
2891 enum cpu_util_type {
2892 	FREQUENCY_UTIL,
2893 	ENERGY_UTIL,
2894 };
2895 
2896 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2897 				 enum cpu_util_type type,
2898 				 struct task_struct *p);
2899 
2900 /*
2901  * Verify the fitness of task @p to run on @cpu taking into account the
2902  * CPU original capacity and the runtime/deadline ratio of the task.
2903  *
2904  * The function will return true if the original capacity of @cpu is
2905  * greater than or equal to task's deadline density right shifted by
2906  * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2907  */
dl_task_fits_capacity(struct task_struct * p,int cpu)2908 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2909 {
2910 	unsigned long cap = arch_scale_cpu_capacity(cpu);
2911 
2912 	return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2913 }
2914 
cpu_bw_dl(struct rq * rq)2915 static inline unsigned long cpu_bw_dl(struct rq *rq)
2916 {
2917 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2918 }
2919 
cpu_util_dl(struct rq * rq)2920 static inline unsigned long cpu_util_dl(struct rq *rq)
2921 {
2922 	return READ_ONCE(rq->avg_dl.util_avg);
2923 }
2924 
2925 /**
2926  * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2927  * @cpu: the CPU to get the utilization for.
2928  *
2929  * The unit of the return value must be the same as the one of CPU capacity
2930  * so that CPU utilization can be compared with CPU capacity.
2931  *
2932  * CPU utilization is the sum of running time of runnable tasks plus the
2933  * recent utilization of currently non-runnable tasks on that CPU.
2934  * It represents the amount of CPU capacity currently used by CFS tasks in
2935  * the range [0..max CPU capacity] with max CPU capacity being the CPU
2936  * capacity at f_max.
2937  *
2938  * The estimated CPU utilization is defined as the maximum between CPU
2939  * utilization and sum of the estimated utilization of the currently
2940  * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2941  * previously-executed tasks, which helps better deduce how busy a CPU will
2942  * be when a long-sleeping task wakes up. The contribution to CPU utilization
2943  * of such a task would be significantly decayed at this point of time.
2944  *
2945  * CPU utilization can be higher than the current CPU capacity
2946  * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2947  * of rounding errors as well as task migrations or wakeups of new tasks.
2948  * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2949  * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2950  * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2951  * capacity. CPU utilization is allowed to overshoot current CPU capacity
2952  * though since this is useful for predicting the CPU capacity required
2953  * after task migrations (scheduler-driven DVFS).
2954  *
2955  * Return: (Estimated) utilization for the specified CPU.
2956  */
cpu_util_cfs(int cpu)2957 static inline unsigned long cpu_util_cfs(int cpu)
2958 {
2959 	struct cfs_rq *cfs_rq;
2960 	unsigned long util;
2961 
2962 	cfs_rq = &cpu_rq(cpu)->cfs;
2963 	util = READ_ONCE(cfs_rq->avg.util_avg);
2964 
2965 	if (sched_feat(UTIL_EST)) {
2966 		util = max_t(unsigned long, util,
2967 			     READ_ONCE(cfs_rq->avg.util_est.enqueued));
2968 	}
2969 
2970 	return min(util, capacity_orig_of(cpu));
2971 }
2972 
cpu_util_rt(struct rq * rq)2973 static inline unsigned long cpu_util_rt(struct rq *rq)
2974 {
2975 	return READ_ONCE(rq->avg_rt.util_avg);
2976 }
2977 #endif
2978 
2979 #ifdef CONFIG_UCLAMP_TASK
2980 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2981 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)2982 static inline unsigned long uclamp_rq_get(struct rq *rq,
2983 					  enum uclamp_id clamp_id)
2984 {
2985 	return READ_ONCE(rq->uclamp[clamp_id].value);
2986 }
2987 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)2988 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
2989 				 unsigned int value)
2990 {
2991 	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
2992 }
2993 
uclamp_rq_is_idle(struct rq * rq)2994 static inline bool uclamp_rq_is_idle(struct rq *rq)
2995 {
2996 	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
2997 }
2998 
2999 /**
3000  * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3001  * @rq:		The rq to clamp against. Must not be NULL.
3002  * @util:	The util value to clamp.
3003  * @p:		The task to clamp against. Can be NULL if you want to clamp
3004  *		against @rq only.
3005  *
3006  * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3007  *
3008  * If sched_uclamp_used static key is disabled, then just return the util
3009  * without any clamping since uclamp aggregation at the rq level in the fast
3010  * path is disabled, rendering this operation a NOP.
3011  *
3012  * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3013  * will return the correct effective uclamp value of the task even if the
3014  * static key is disabled.
3015  */
3016 static __always_inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)3017 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3018 				  struct task_struct *p)
3019 {
3020 	unsigned long min_util = 0;
3021 	unsigned long max_util = 0;
3022 
3023 	if (!static_branch_likely(&sched_uclamp_used))
3024 		return util;
3025 
3026 	if (p) {
3027 		min_util = uclamp_eff_value(p, UCLAMP_MIN);
3028 		max_util = uclamp_eff_value(p, UCLAMP_MAX);
3029 
3030 		/*
3031 		 * Ignore last runnable task's max clamp, as this task will
3032 		 * reset it. Similarly, no need to read the rq's min clamp.
3033 		 */
3034 		if (uclamp_rq_is_idle(rq))
3035 			goto out;
3036 	}
3037 
3038 	min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3039 	max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3040 out:
3041 	/*
3042 	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3043 	 * RUNNABLE tasks with _different_ clamps, we can end up with an
3044 	 * inversion. Fix it now when the clamps are applied.
3045 	 */
3046 	if (unlikely(min_util >= max_util))
3047 		return min_util;
3048 
3049 	return clamp(util, min_util, max_util);
3050 }
3051 
3052 /* Is the rq being capped/throttled by uclamp_max? */
uclamp_rq_is_capped(struct rq * rq)3053 static inline bool uclamp_rq_is_capped(struct rq *rq)
3054 {
3055 	unsigned long rq_util;
3056 	unsigned long max_util;
3057 
3058 	if (!static_branch_likely(&sched_uclamp_used))
3059 		return false;
3060 
3061 	rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3062 	max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3063 
3064 	return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3065 }
3066 
3067 /*
3068  * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3069  * by default in the fast path and only gets turned on once userspace performs
3070  * an operation that requires it.
3071  *
3072  * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3073  * hence is active.
3074  */
uclamp_is_used(void)3075 static inline bool uclamp_is_used(void)
3076 {
3077 	return static_branch_likely(&sched_uclamp_used);
3078 }
3079 #else /* CONFIG_UCLAMP_TASK */
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)3080 static inline unsigned long uclamp_eff_value(struct task_struct *p,
3081 					     enum uclamp_id clamp_id)
3082 {
3083 	if (clamp_id == UCLAMP_MIN)
3084 		return 0;
3085 
3086 	return SCHED_CAPACITY_SCALE;
3087 }
3088 
3089 static inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)3090 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3091 				  struct task_struct *p)
3092 {
3093 	return util;
3094 }
3095 
uclamp_rq_is_capped(struct rq * rq)3096 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3097 
uclamp_is_used(void)3098 static inline bool uclamp_is_used(void)
3099 {
3100 	return false;
3101 }
3102 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3103 static inline unsigned long uclamp_rq_get(struct rq *rq,
3104 					  enum uclamp_id clamp_id)
3105 {
3106 	if (clamp_id == UCLAMP_MIN)
3107 		return 0;
3108 
3109 	return SCHED_CAPACITY_SCALE;
3110 }
3111 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3112 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3113 				 unsigned int value)
3114 {
3115 }
3116 
uclamp_rq_is_idle(struct rq * rq)3117 static inline bool uclamp_rq_is_idle(struct rq *rq)
3118 {
3119 	return false;
3120 }
3121 #endif /* CONFIG_UCLAMP_TASK */
3122 
3123 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
cpu_util_irq(struct rq * rq)3124 static inline unsigned long cpu_util_irq(struct rq *rq)
3125 {
3126 	return rq->avg_irq.util_avg;
3127 }
3128 
3129 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3130 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3131 {
3132 	util *= (max - irq);
3133 	util /= max;
3134 
3135 	return util;
3136 
3137 }
3138 #else
cpu_util_irq(struct rq * rq)3139 static inline unsigned long cpu_util_irq(struct rq *rq)
3140 {
3141 	return 0;
3142 }
3143 
3144 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3145 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3146 {
3147 	return util;
3148 }
3149 #endif
3150 
3151 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3152 
3153 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3154 
3155 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3156 
sched_energy_enabled(void)3157 static inline bool sched_energy_enabled(void)
3158 {
3159 	return static_branch_unlikely(&sched_energy_present);
3160 }
3161 
3162 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3163 
3164 #define perf_domain_span(pd) NULL
sched_energy_enabled(void)3165 static inline bool sched_energy_enabled(void) { return false; }
3166 
3167 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3168 
3169 #ifdef CONFIG_MEMBARRIER
3170 /*
3171  * The scheduler provides memory barriers required by membarrier between:
3172  * - prior user-space memory accesses and store to rq->membarrier_state,
3173  * - store to rq->membarrier_state and following user-space memory accesses.
3174  * In the same way it provides those guarantees around store to rq->curr.
3175  */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3176 static inline void membarrier_switch_mm(struct rq *rq,
3177 					struct mm_struct *prev_mm,
3178 					struct mm_struct *next_mm)
3179 {
3180 	int membarrier_state;
3181 
3182 	if (prev_mm == next_mm)
3183 		return;
3184 
3185 	membarrier_state = atomic_read(&next_mm->membarrier_state);
3186 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3187 		return;
3188 
3189 	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3190 }
3191 #else
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3192 static inline void membarrier_switch_mm(struct rq *rq,
3193 					struct mm_struct *prev_mm,
3194 					struct mm_struct *next_mm)
3195 {
3196 }
3197 #endif
3198 
3199 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3200 static inline bool is_per_cpu_kthread(struct task_struct *p)
3201 {
3202 	if (!(p->flags & PF_KTHREAD))
3203 		return false;
3204 
3205 	if (p->nr_cpus_allowed != 1)
3206 		return false;
3207 
3208 	return true;
3209 }
3210 #endif
3211 
3212 extern void swake_up_all_locked(struct swait_queue_head *q);
3213 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3214 
3215 #ifdef CONFIG_PREEMPT_DYNAMIC
3216 extern int preempt_dynamic_mode;
3217 extern int sched_dynamic_mode(const char *str);
3218 extern void sched_dynamic_update(int mode);
3219 #endif
3220 
update_current_exec_runtime(struct task_struct * curr,u64 now,u64 delta_exec)3221 static inline void update_current_exec_runtime(struct task_struct *curr,
3222 						u64 now, u64 delta_exec)
3223 {
3224 	curr->se.sum_exec_runtime += delta_exec;
3225 	account_group_exec_runtime(curr, delta_exec);
3226 
3227 	curr->se.exec_start = now;
3228 	cgroup_account_cputime(curr, delta_exec);
3229 }
3230 
3231 #endif /* _KERNEL_SCHED_SCHED_H */
3232