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(const struct sched_dl_entity * dl_se)251 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
252 {
253 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
254 	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
255 #else
256 	return false;
257 #endif
258 }
259 
260 /*
261  * Tells if entity @a should preempt entity @b.
262  */
dl_entity_preempt(const struct sched_dl_entity * a,const struct sched_dl_entity * b)263 static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
264 				     const struct sched_dl_entity *b)
265 {
266 	return dl_entity_is_special(a) ||
267 	       dl_time_before(a->deadline, b->deadline);
268 }
269 
270 /*
271  * This is the priority-queue data structure of the RT scheduling class:
272  */
273 struct rt_prio_array {
274 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
275 	struct list_head queue[MAX_RT_PRIO];
276 };
277 
278 struct rt_bandwidth {
279 	/* nests inside the rq lock: */
280 	raw_spinlock_t		rt_runtime_lock;
281 	ktime_t			rt_period;
282 	u64			rt_runtime;
283 	struct hrtimer		rt_period_timer;
284 	unsigned int		rt_period_active;
285 };
286 
287 void __dl_clear_params(struct task_struct *p);
288 
dl_bandwidth_enabled(void)289 static inline int dl_bandwidth_enabled(void)
290 {
291 	return sysctl_sched_rt_runtime >= 0;
292 }
293 
294 /*
295  * To keep the bandwidth of -deadline tasks under control
296  * we need some place where:
297  *  - store the maximum -deadline bandwidth of each cpu;
298  *  - cache the fraction of bandwidth that is currently allocated in
299  *    each root domain;
300  *
301  * This is all done in the data structure below. It is similar to the
302  * one used for RT-throttling (rt_bandwidth), with the main difference
303  * that, since here we are only interested in admission control, we
304  * do not decrease any runtime while the group "executes", neither we
305  * need a timer to replenish it.
306  *
307  * With respect to SMP, bandwidth is given on a per root domain basis,
308  * meaning that:
309  *  - bw (< 100%) is the deadline bandwidth of each CPU;
310  *  - total_bw is the currently allocated bandwidth in each root domain;
311  */
312 struct dl_bw {
313 	raw_spinlock_t		lock;
314 	u64			bw;
315 	u64			total_bw;
316 };
317 
318 extern void init_dl_bw(struct dl_bw *dl_b);
319 extern int  sched_dl_global_validate(void);
320 extern void sched_dl_do_global(void);
321 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
322 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
323 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
324 extern bool __checkparam_dl(const struct sched_attr *attr);
325 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
326 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
327 extern int  dl_bw_check_overflow(int cpu);
328 
329 #ifdef CONFIG_CGROUP_SCHED
330 
331 struct cfs_rq;
332 struct rt_rq;
333 
334 extern struct list_head task_groups;
335 
336 struct cfs_bandwidth {
337 #ifdef CONFIG_CFS_BANDWIDTH
338 	raw_spinlock_t		lock;
339 	ktime_t			period;
340 	u64			quota;
341 	u64			runtime;
342 	u64			burst;
343 	u64			runtime_snap;
344 	s64			hierarchical_quota;
345 
346 	u8			idle;
347 	u8			period_active;
348 	u8			slack_started;
349 	struct hrtimer		period_timer;
350 	struct hrtimer		slack_timer;
351 	struct list_head	throttled_cfs_rq;
352 
353 	/* Statistics: */
354 	int			nr_periods;
355 	int			nr_throttled;
356 	int			nr_burst;
357 	u64			throttled_time;
358 	u64			burst_time;
359 #endif
360 };
361 
362 /* Task group related information */
363 struct task_group {
364 	struct cgroup_subsys_state css;
365 
366 #ifdef CONFIG_FAIR_GROUP_SCHED
367 	/* schedulable entities of this group on each CPU */
368 	struct sched_entity	**se;
369 	/* runqueue "owned" by this group on each CPU */
370 	struct cfs_rq		**cfs_rq;
371 	unsigned long		shares;
372 
373 	/* A positive value indicates that this is a SCHED_IDLE group. */
374 	int			idle;
375 
376 #ifdef	CONFIG_SMP
377 	/*
378 	 * load_avg can be heavily contended at clock tick time, so put
379 	 * it in its own cacheline separated from the fields above which
380 	 * will also be accessed at each tick.
381 	 */
382 	atomic_long_t		load_avg ____cacheline_aligned;
383 #endif
384 #endif
385 
386 #ifdef CONFIG_RT_GROUP_SCHED
387 	struct sched_rt_entity	**rt_se;
388 	struct rt_rq		**rt_rq;
389 
390 	struct rt_bandwidth	rt_bandwidth;
391 #endif
392 
393 	struct rcu_head		rcu;
394 	struct list_head	list;
395 
396 	struct task_group	*parent;
397 	struct list_head	siblings;
398 	struct list_head	children;
399 
400 #ifdef CONFIG_SCHED_AUTOGROUP
401 	struct autogroup	*autogroup;
402 #endif
403 
404 	struct cfs_bandwidth	cfs_bandwidth;
405 
406 #ifdef CONFIG_UCLAMP_TASK_GROUP
407 	/* The two decimal precision [%] value requested from user-space */
408 	unsigned int		uclamp_pct[UCLAMP_CNT];
409 	/* Clamp values requested for a task group */
410 	struct uclamp_se	uclamp_req[UCLAMP_CNT];
411 	/* Effective clamp values used for a task group */
412 	struct uclamp_se	uclamp[UCLAMP_CNT];
413 #endif
414 
415 };
416 
417 #ifdef CONFIG_FAIR_GROUP_SCHED
418 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
419 
420 /*
421  * A weight of 0 or 1 can cause arithmetics problems.
422  * A weight of a cfs_rq is the sum of weights of which entities
423  * are queued on this cfs_rq, so a weight of a entity should not be
424  * too large, so as the shares value of a task group.
425  * (The default weight is 1024 - so there's no practical
426  *  limitation from this.)
427  */
428 #define MIN_SHARES		(1UL <<  1)
429 #define MAX_SHARES		(1UL << 18)
430 #endif
431 
432 typedef int (*tg_visitor)(struct task_group *, void *);
433 
434 extern int walk_tg_tree_from(struct task_group *from,
435 			     tg_visitor down, tg_visitor up, void *data);
436 
437 /*
438  * Iterate the full tree, calling @down when first entering a node and @up when
439  * leaving it for the final time.
440  *
441  * Caller must hold rcu_lock or sufficient equivalent.
442  */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)443 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
444 {
445 	return walk_tg_tree_from(&root_task_group, down, up, data);
446 }
447 
448 extern int tg_nop(struct task_group *tg, void *data);
449 
450 extern void free_fair_sched_group(struct task_group *tg);
451 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
452 extern void online_fair_sched_group(struct task_group *tg);
453 extern void unregister_fair_sched_group(struct task_group *tg);
454 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
455 			struct sched_entity *se, int cpu,
456 			struct sched_entity *parent);
457 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
458 
459 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
460 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
461 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
462 extern bool cfs_task_bw_constrained(struct task_struct *p);
463 
464 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
465 		struct sched_rt_entity *rt_se, int cpu,
466 		struct sched_rt_entity *parent);
467 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
468 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
469 extern long sched_group_rt_runtime(struct task_group *tg);
470 extern long sched_group_rt_period(struct task_group *tg);
471 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
472 
473 extern struct task_group *sched_create_group(struct task_group *parent);
474 extern void sched_online_group(struct task_group *tg,
475 			       struct task_group *parent);
476 extern void sched_destroy_group(struct task_group *tg);
477 extern void sched_release_group(struct task_group *tg);
478 
479 extern void sched_move_task(struct task_struct *tsk);
480 
481 #ifdef CONFIG_FAIR_GROUP_SCHED
482 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
483 
484 extern int sched_group_set_idle(struct task_group *tg, long idle);
485 
486 #ifdef CONFIG_SMP
487 extern void set_task_rq_fair(struct sched_entity *se,
488 			     struct cfs_rq *prev, struct cfs_rq *next);
489 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)490 static inline void set_task_rq_fair(struct sched_entity *se,
491 			     struct cfs_rq *prev, struct cfs_rq *next) { }
492 #endif /* CONFIG_SMP */
493 #endif /* CONFIG_FAIR_GROUP_SCHED */
494 
495 #else /* CONFIG_CGROUP_SCHED */
496 
497 struct cfs_bandwidth { };
cfs_task_bw_constrained(struct task_struct * p)498 static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
499 
500 #endif	/* CONFIG_CGROUP_SCHED */
501 
502 extern void unregister_rt_sched_group(struct task_group *tg);
503 extern void free_rt_sched_group(struct task_group *tg);
504 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
505 
506 /*
507  * u64_u32_load/u64_u32_store
508  *
509  * Use a copy of a u64 value to protect against data race. This is only
510  * applicable for 32-bits architectures.
511  */
512 #ifdef CONFIG_64BIT
513 # define u64_u32_load_copy(var, copy)       var
514 # define u64_u32_store_copy(var, copy, val) (var = val)
515 #else
516 # define u64_u32_load_copy(var, copy)					\
517 ({									\
518 	u64 __val, __val_copy;						\
519 	do {								\
520 		__val_copy = copy;					\
521 		/*							\
522 		 * paired with u64_u32_store_copy(), ordering access	\
523 		 * to var and copy.					\
524 		 */							\
525 		smp_rmb();						\
526 		__val = var;						\
527 	} while (__val != __val_copy);					\
528 	__val;								\
529 })
530 # define u64_u32_store_copy(var, copy, val)				\
531 do {									\
532 	typeof(val) __val = (val);					\
533 	var = __val;							\
534 	/*								\
535 	 * paired with u64_u32_load_copy(), ordering access to var and	\
536 	 * copy.							\
537 	 */								\
538 	smp_wmb();							\
539 	copy = __val;							\
540 } while (0)
541 #endif
542 # define u64_u32_load(var)      u64_u32_load_copy(var, var##_copy)
543 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
544 
545 /* CFS-related fields in a runqueue */
546 struct cfs_rq {
547 	struct load_weight	load;
548 	unsigned int		nr_running;
549 	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
550 	unsigned int		idle_nr_running;   /* SCHED_IDLE */
551 	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
552 
553 	s64			avg_vruntime;
554 	u64			avg_load;
555 
556 	u64			exec_clock;
557 	u64			min_vruntime;
558 #ifdef CONFIG_SCHED_CORE
559 	unsigned int		forceidle_seq;
560 	u64			min_vruntime_fi;
561 #endif
562 
563 #ifndef CONFIG_64BIT
564 	u64			min_vruntime_copy;
565 #endif
566 
567 	struct rb_root_cached	tasks_timeline;
568 
569 	/*
570 	 * 'curr' points to currently running entity on this cfs_rq.
571 	 * It is set to NULL otherwise (i.e when none are currently running).
572 	 */
573 	struct sched_entity	*curr;
574 	struct sched_entity	*next;
575 
576 #ifdef	CONFIG_SCHED_DEBUG
577 	unsigned int		nr_spread_over;
578 #endif
579 
580 #ifdef CONFIG_SMP
581 	/*
582 	 * CFS load tracking
583 	 */
584 	struct sched_avg	avg;
585 #ifndef CONFIG_64BIT
586 	u64			last_update_time_copy;
587 #endif
588 	struct {
589 		raw_spinlock_t	lock ____cacheline_aligned;
590 		int		nr;
591 		unsigned long	load_avg;
592 		unsigned long	util_avg;
593 		unsigned long	runnable_avg;
594 	} removed;
595 
596 #ifdef CONFIG_FAIR_GROUP_SCHED
597 	unsigned long		tg_load_avg_contrib;
598 	long			propagate;
599 	long			prop_runnable_sum;
600 
601 	/*
602 	 *   h_load = weight * f(tg)
603 	 *
604 	 * Where f(tg) is the recursive weight fraction assigned to
605 	 * this group.
606 	 */
607 	unsigned long		h_load;
608 	u64			last_h_load_update;
609 	struct sched_entity	*h_load_next;
610 #endif /* CONFIG_FAIR_GROUP_SCHED */
611 #endif /* CONFIG_SMP */
612 
613 #ifdef CONFIG_FAIR_GROUP_SCHED
614 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
615 
616 	/*
617 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
618 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
619 	 * (like users, containers etc.)
620 	 *
621 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
622 	 * This list is used during load balance.
623 	 */
624 	int			on_list;
625 	struct list_head	leaf_cfs_rq_list;
626 	struct task_group	*tg;	/* group that "owns" this runqueue */
627 
628 	/* Locally cached copy of our task_group's idle value */
629 	int			idle;
630 
631 #ifdef CONFIG_CFS_BANDWIDTH
632 	int			runtime_enabled;
633 	s64			runtime_remaining;
634 
635 	u64			throttled_pelt_idle;
636 #ifndef CONFIG_64BIT
637 	u64                     throttled_pelt_idle_copy;
638 #endif
639 	u64			throttled_clock;
640 	u64			throttled_clock_pelt;
641 	u64			throttled_clock_pelt_time;
642 	u64			throttled_clock_self;
643 	u64			throttled_clock_self_time;
644 	int			throttled;
645 	int			throttle_count;
646 	struct list_head	throttled_list;
647 #ifdef CONFIG_SMP
648 	struct list_head	throttled_csd_list;
649 #endif
650 #endif /* CONFIG_CFS_BANDWIDTH */
651 #endif /* CONFIG_FAIR_GROUP_SCHED */
652 };
653 
rt_bandwidth_enabled(void)654 static inline int rt_bandwidth_enabled(void)
655 {
656 	return sysctl_sched_rt_runtime >= 0;
657 }
658 
659 /* RT IPI pull logic requires IRQ_WORK */
660 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
661 # define HAVE_RT_PUSH_IPI
662 #endif
663 
664 /* Real-Time classes' related field in a runqueue: */
665 struct rt_rq {
666 	struct rt_prio_array	active;
667 	unsigned int		rt_nr_running;
668 	unsigned int		rr_nr_running;
669 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
670 	struct {
671 		int		curr; /* highest queued rt task prio */
672 #ifdef CONFIG_SMP
673 		int		next; /* next highest */
674 #endif
675 	} highest_prio;
676 #endif
677 #ifdef CONFIG_SMP
678 	unsigned int		rt_nr_migratory;
679 	unsigned int		rt_nr_total;
680 	int			overloaded;
681 	struct plist_head	pushable_tasks;
682 
683 #endif /* CONFIG_SMP */
684 	int			rt_queued;
685 
686 	int			rt_throttled;
687 	u64			rt_time;
688 	u64			rt_runtime;
689 	/* Nests inside the rq lock: */
690 	raw_spinlock_t		rt_runtime_lock;
691 
692 #ifdef CONFIG_RT_GROUP_SCHED
693 	unsigned int		rt_nr_boosted;
694 
695 	struct rq		*rq;
696 	struct task_group	*tg;
697 #endif
698 };
699 
rt_rq_is_runnable(struct rt_rq * rt_rq)700 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
701 {
702 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
703 }
704 
705 /* Deadline class' related fields in a runqueue */
706 struct dl_rq {
707 	/* runqueue is an rbtree, ordered by deadline */
708 	struct rb_root_cached	root;
709 
710 	unsigned int		dl_nr_running;
711 
712 #ifdef CONFIG_SMP
713 	/*
714 	 * Deadline values of the currently executing and the
715 	 * earliest ready task on this rq. Caching these facilitates
716 	 * the decision whether or not a ready but not running task
717 	 * should migrate somewhere else.
718 	 */
719 	struct {
720 		u64		curr;
721 		u64		next;
722 	} earliest_dl;
723 
724 	unsigned int		dl_nr_migratory;
725 	int			overloaded;
726 
727 	/*
728 	 * Tasks on this rq that can be pushed away. They are kept in
729 	 * an rb-tree, ordered by tasks' deadlines, with caching
730 	 * of the leftmost (earliest deadline) element.
731 	 */
732 	struct rb_root_cached	pushable_dl_tasks_root;
733 #else
734 	struct dl_bw		dl_bw;
735 #endif
736 	/*
737 	 * "Active utilization" for this runqueue: increased when a
738 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
739 	 * task blocks
740 	 */
741 	u64			running_bw;
742 
743 	/*
744 	 * Utilization of the tasks "assigned" to this runqueue (including
745 	 * the tasks that are in runqueue and the tasks that executed on this
746 	 * CPU and blocked). Increased when a task moves to this runqueue, and
747 	 * decreased when the task moves away (migrates, changes scheduling
748 	 * policy, or terminates).
749 	 * This is needed to compute the "inactive utilization" for the
750 	 * runqueue (inactive utilization = this_bw - running_bw).
751 	 */
752 	u64			this_bw;
753 	u64			extra_bw;
754 
755 	/*
756 	 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
757 	 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
758 	 */
759 	u64			max_bw;
760 
761 	/*
762 	 * Inverse of the fraction of CPU utilization that can be reclaimed
763 	 * by the GRUB algorithm.
764 	 */
765 	u64			bw_ratio;
766 };
767 
768 #ifdef CONFIG_FAIR_GROUP_SCHED
769 /* An entity is a task if it doesn't "own" a runqueue */
770 #define entity_is_task(se)	(!se->my_q)
771 
se_update_runnable(struct sched_entity * se)772 static inline void se_update_runnable(struct sched_entity *se)
773 {
774 	if (!entity_is_task(se))
775 		se->runnable_weight = se->my_q->h_nr_running;
776 }
777 
se_runnable(struct sched_entity * se)778 static inline long se_runnable(struct sched_entity *se)
779 {
780 	if (entity_is_task(se))
781 		return !!se->on_rq;
782 	else
783 		return se->runnable_weight;
784 }
785 
786 #else
787 #define entity_is_task(se)	1
788 
se_update_runnable(struct sched_entity * se)789 static inline void se_update_runnable(struct sched_entity *se) {}
790 
se_runnable(struct sched_entity * se)791 static inline long se_runnable(struct sched_entity *se)
792 {
793 	return !!se->on_rq;
794 }
795 #endif
796 
797 #ifdef CONFIG_SMP
798 /*
799  * XXX we want to get rid of these helpers and use the full load resolution.
800  */
se_weight(struct sched_entity * se)801 static inline long se_weight(struct sched_entity *se)
802 {
803 	return scale_load_down(se->load.weight);
804 }
805 
806 
sched_asym_prefer(int a,int b)807 static inline bool sched_asym_prefer(int a, int b)
808 {
809 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
810 }
811 
812 struct perf_domain {
813 	struct em_perf_domain *em_pd;
814 	struct perf_domain *next;
815 	struct rcu_head rcu;
816 };
817 
818 /* Scheduling group status flags */
819 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
820 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
821 
822 /*
823  * We add the notion of a root-domain which will be used to define per-domain
824  * variables. Each exclusive cpuset essentially defines an island domain by
825  * fully partitioning the member CPUs from any other cpuset. Whenever a new
826  * exclusive cpuset is created, we also create and attach a new root-domain
827  * object.
828  *
829  */
830 struct root_domain {
831 	atomic_t		refcount;
832 	atomic_t		rto_count;
833 	struct rcu_head		rcu;
834 	cpumask_var_t		span;
835 	cpumask_var_t		online;
836 
837 	/*
838 	 * Indicate pullable load on at least one CPU, e.g:
839 	 * - More than one runnable task
840 	 * - Running task is misfit
841 	 */
842 	int			overload;
843 
844 	/* Indicate one or more cpus over-utilized (tipping point) */
845 	int			overutilized;
846 
847 	/*
848 	 * The bit corresponding to a CPU gets set here if such CPU has more
849 	 * than one runnable -deadline task (as it is below for RT tasks).
850 	 */
851 	cpumask_var_t		dlo_mask;
852 	atomic_t		dlo_count;
853 	struct dl_bw		dl_bw;
854 	struct cpudl		cpudl;
855 
856 	/*
857 	 * Indicate whether a root_domain's dl_bw has been checked or
858 	 * updated. It's monotonously increasing value.
859 	 *
860 	 * Also, some corner cases, like 'wrap around' is dangerous, but given
861 	 * that u64 is 'big enough'. So that shouldn't be a concern.
862 	 */
863 	u64 visit_gen;
864 
865 #ifdef HAVE_RT_PUSH_IPI
866 	/*
867 	 * For IPI pull requests, loop across the rto_mask.
868 	 */
869 	struct irq_work		rto_push_work;
870 	raw_spinlock_t		rto_lock;
871 	/* These are only updated and read within rto_lock */
872 	int			rto_loop;
873 	int			rto_cpu;
874 	/* These atomics are updated outside of a lock */
875 	atomic_t		rto_loop_next;
876 	atomic_t		rto_loop_start;
877 #endif
878 	/*
879 	 * The "RT overload" flag: it gets set if a CPU has more than
880 	 * one runnable RT task.
881 	 */
882 	cpumask_var_t		rto_mask;
883 	struct cpupri		cpupri;
884 
885 	unsigned long		max_cpu_capacity;
886 
887 	/*
888 	 * NULL-terminated list of performance domains intersecting with the
889 	 * CPUs of the rd. Protected by RCU.
890 	 */
891 	struct perf_domain __rcu *pd;
892 };
893 
894 extern void init_defrootdomain(void);
895 extern int sched_init_domains(const struct cpumask *cpu_map);
896 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
897 extern void sched_get_rd(struct root_domain *rd);
898 extern void sched_put_rd(struct root_domain *rd);
899 
900 #ifdef HAVE_RT_PUSH_IPI
901 extern void rto_push_irq_work_func(struct irq_work *work);
902 #endif
903 #endif /* CONFIG_SMP */
904 
905 #ifdef CONFIG_UCLAMP_TASK
906 /*
907  * struct uclamp_bucket - Utilization clamp bucket
908  * @value: utilization clamp value for tasks on this clamp bucket
909  * @tasks: number of RUNNABLE tasks on this clamp bucket
910  *
911  * Keep track of how many tasks are RUNNABLE for a given utilization
912  * clamp value.
913  */
914 struct uclamp_bucket {
915 	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
916 	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
917 };
918 
919 /*
920  * struct uclamp_rq - rq's utilization clamp
921  * @value: currently active clamp values for a rq
922  * @bucket: utilization clamp buckets affecting a rq
923  *
924  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
925  * A clamp value is affecting a rq when there is at least one task RUNNABLE
926  * (or actually running) with that value.
927  *
928  * There are up to UCLAMP_CNT possible different clamp values, currently there
929  * are only two: minimum utilization and maximum utilization.
930  *
931  * All utilization clamping values are MAX aggregated, since:
932  * - for util_min: we want to run the CPU at least at the max of the minimum
933  *   utilization required by its currently RUNNABLE tasks.
934  * - for util_max: we want to allow the CPU to run up to the max of the
935  *   maximum utilization allowed by its currently RUNNABLE tasks.
936  *
937  * Since on each system we expect only a limited number of different
938  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
939  * the metrics required to compute all the per-rq utilization clamp values.
940  */
941 struct uclamp_rq {
942 	unsigned int value;
943 	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
944 };
945 
946 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
947 #endif /* CONFIG_UCLAMP_TASK */
948 
949 struct rq;
950 struct balance_callback {
951 	struct balance_callback *next;
952 	void (*func)(struct rq *rq);
953 };
954 
955 /*
956  * This is the main, per-CPU runqueue data structure.
957  *
958  * Locking rule: those places that want to lock multiple runqueues
959  * (such as the load balancing or the thread migration code), lock
960  * acquire operations must be ordered by ascending &runqueue.
961  */
962 struct rq {
963 	/* runqueue lock: */
964 	raw_spinlock_t		__lock;
965 
966 	/*
967 	 * nr_running and cpu_load should be in the same cacheline because
968 	 * remote CPUs use both these fields when doing load calculation.
969 	 */
970 	unsigned int		nr_running;
971 #ifdef CONFIG_NUMA_BALANCING
972 	unsigned int		nr_numa_running;
973 	unsigned int		nr_preferred_running;
974 	unsigned int		numa_migrate_on;
975 #endif
976 #ifdef CONFIG_NO_HZ_COMMON
977 #ifdef CONFIG_SMP
978 	unsigned long		last_blocked_load_update_tick;
979 	unsigned int		has_blocked_load;
980 	call_single_data_t	nohz_csd;
981 #endif /* CONFIG_SMP */
982 	unsigned int		nohz_tick_stopped;
983 	atomic_t		nohz_flags;
984 #endif /* CONFIG_NO_HZ_COMMON */
985 
986 #ifdef CONFIG_SMP
987 	unsigned int		ttwu_pending;
988 #endif
989 	u64			nr_switches;
990 
991 #ifdef CONFIG_UCLAMP_TASK
992 	/* Utilization clamp values based on CPU's RUNNABLE tasks */
993 	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
994 	unsigned int		uclamp_flags;
995 #define UCLAMP_FLAG_IDLE 0x01
996 #endif
997 
998 	struct cfs_rq		cfs;
999 	struct rt_rq		rt;
1000 	struct dl_rq		dl;
1001 
1002 #ifdef CONFIG_FAIR_GROUP_SCHED
1003 	/* list of leaf cfs_rq on this CPU: */
1004 	struct list_head	leaf_cfs_rq_list;
1005 	struct list_head	*tmp_alone_branch;
1006 #endif /* CONFIG_FAIR_GROUP_SCHED */
1007 
1008 	/*
1009 	 * This is part of a global counter where only the total sum
1010 	 * over all CPUs matters. A task can increase this counter on
1011 	 * one CPU and if it got migrated afterwards it may decrease
1012 	 * it on another CPU. Always updated under the runqueue lock:
1013 	 */
1014 	unsigned int		nr_uninterruptible;
1015 
1016 	struct task_struct __rcu	*curr;
1017 	struct task_struct	*idle;
1018 	struct task_struct	*stop;
1019 	unsigned long		next_balance;
1020 	struct mm_struct	*prev_mm;
1021 
1022 	unsigned int		clock_update_flags;
1023 	u64			clock;
1024 	/* Ensure that all clocks are in the same cache line */
1025 	u64			clock_task ____cacheline_aligned;
1026 	u64			clock_pelt;
1027 	unsigned long		lost_idle_time;
1028 	u64			clock_pelt_idle;
1029 	u64			clock_idle;
1030 #ifndef CONFIG_64BIT
1031 	u64			clock_pelt_idle_copy;
1032 	u64			clock_idle_copy;
1033 #endif
1034 
1035 	atomic_t		nr_iowait;
1036 
1037 #ifdef CONFIG_SCHED_DEBUG
1038 	u64 last_seen_need_resched_ns;
1039 	int ticks_without_resched;
1040 #endif
1041 
1042 #ifdef CONFIG_MEMBARRIER
1043 	int membarrier_state;
1044 #endif
1045 
1046 #ifdef CONFIG_SMP
1047 	struct root_domain		*rd;
1048 	struct sched_domain __rcu	*sd;
1049 
1050 	unsigned long		cpu_capacity;
1051 	unsigned long		cpu_capacity_orig;
1052 
1053 	struct balance_callback *balance_callback;
1054 
1055 	unsigned char		nohz_idle_balance;
1056 	unsigned char		idle_balance;
1057 
1058 	unsigned long		misfit_task_load;
1059 
1060 	/* For active balancing */
1061 	int			active_balance;
1062 	int			push_cpu;
1063 	struct cpu_stop_work	active_balance_work;
1064 
1065 	/* CPU of this runqueue: */
1066 	int			cpu;
1067 	int			online;
1068 
1069 	struct list_head cfs_tasks;
1070 
1071 	struct sched_avg	avg_rt;
1072 	struct sched_avg	avg_dl;
1073 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1074 	struct sched_avg	avg_irq;
1075 #endif
1076 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1077 	struct sched_avg	avg_thermal;
1078 #endif
1079 	u64			idle_stamp;
1080 	u64			avg_idle;
1081 
1082 	unsigned long		wake_stamp;
1083 	u64			wake_avg_idle;
1084 
1085 	/* This is used to determine avg_idle's max value */
1086 	u64			max_idle_balance_cost;
1087 
1088 #ifdef CONFIG_HOTPLUG_CPU
1089 	struct rcuwait		hotplug_wait;
1090 #endif
1091 #endif /* CONFIG_SMP */
1092 
1093 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1094 	u64			prev_irq_time;
1095 #endif
1096 #ifdef CONFIG_PARAVIRT
1097 	u64			prev_steal_time;
1098 #endif
1099 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1100 	u64			prev_steal_time_rq;
1101 #endif
1102 
1103 	/* calc_load related fields */
1104 	unsigned long		calc_load_update;
1105 	long			calc_load_active;
1106 
1107 #ifdef CONFIG_SCHED_HRTICK
1108 #ifdef CONFIG_SMP
1109 	call_single_data_t	hrtick_csd;
1110 #endif
1111 	struct hrtimer		hrtick_timer;
1112 	ktime_t 		hrtick_time;
1113 #endif
1114 
1115 #ifdef CONFIG_SCHEDSTATS
1116 	/* latency stats */
1117 	struct sched_info	rq_sched_info;
1118 	unsigned long long	rq_cpu_time;
1119 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1120 
1121 	/* sys_sched_yield() stats */
1122 	unsigned int		yld_count;
1123 
1124 	/* schedule() stats */
1125 	unsigned int		sched_count;
1126 	unsigned int		sched_goidle;
1127 
1128 	/* try_to_wake_up() stats */
1129 	unsigned int		ttwu_count;
1130 	unsigned int		ttwu_local;
1131 #endif
1132 
1133 #ifdef CONFIG_CPU_IDLE
1134 	/* Must be inspected within a rcu lock section */
1135 	struct cpuidle_state	*idle_state;
1136 #endif
1137 
1138 #ifdef CONFIG_SMP
1139 	unsigned int		nr_pinned;
1140 #endif
1141 	unsigned int		push_busy;
1142 	struct cpu_stop_work	push_work;
1143 
1144 #ifdef CONFIG_SCHED_CORE
1145 	/* per rq */
1146 	struct rq		*core;
1147 	struct task_struct	*core_pick;
1148 	unsigned int		core_enabled;
1149 	unsigned int		core_sched_seq;
1150 	struct rb_root		core_tree;
1151 
1152 	/* shared state -- careful with sched_core_cpu_deactivate() */
1153 	unsigned int		core_task_seq;
1154 	unsigned int		core_pick_seq;
1155 	unsigned long		core_cookie;
1156 	unsigned int		core_forceidle_count;
1157 	unsigned int		core_forceidle_seq;
1158 	unsigned int		core_forceidle_occupation;
1159 	u64			core_forceidle_start;
1160 #endif
1161 
1162 	/* Scratch cpumask to be temporarily used under rq_lock */
1163 	cpumask_var_t		scratch_mask;
1164 
1165 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1166 	call_single_data_t	cfsb_csd;
1167 	struct list_head	cfsb_csd_list;
1168 #endif
1169 };
1170 
1171 #ifdef CONFIG_FAIR_GROUP_SCHED
1172 
1173 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1174 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1175 {
1176 	return cfs_rq->rq;
1177 }
1178 
1179 #else
1180 
rq_of(struct cfs_rq * cfs_rq)1181 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1182 {
1183 	return container_of(cfs_rq, struct rq, cfs);
1184 }
1185 #endif
1186 
cpu_of(struct rq * rq)1187 static inline int cpu_of(struct rq *rq)
1188 {
1189 #ifdef CONFIG_SMP
1190 	return rq->cpu;
1191 #else
1192 	return 0;
1193 #endif
1194 }
1195 
1196 #define MDF_PUSH	0x01
1197 
is_migration_disabled(struct task_struct * p)1198 static inline bool is_migration_disabled(struct task_struct *p)
1199 {
1200 #ifdef CONFIG_SMP
1201 	return p->migration_disabled;
1202 #else
1203 	return false;
1204 #endif
1205 }
1206 
1207 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1208 
1209 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1210 #define this_rq()		this_cpu_ptr(&runqueues)
1211 #define task_rq(p)		cpu_rq(task_cpu(p))
1212 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1213 #define raw_rq()		raw_cpu_ptr(&runqueues)
1214 
1215 struct sched_group;
1216 #ifdef CONFIG_SCHED_CORE
1217 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1218 
1219 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1220 
sched_core_enabled(struct rq * rq)1221 static inline bool sched_core_enabled(struct rq *rq)
1222 {
1223 	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1224 }
1225 
sched_core_disabled(void)1226 static inline bool sched_core_disabled(void)
1227 {
1228 	return !static_branch_unlikely(&__sched_core_enabled);
1229 }
1230 
1231 /*
1232  * Be careful with this function; not for general use. The return value isn't
1233  * stable unless you actually hold a relevant rq->__lock.
1234  */
rq_lockp(struct rq * rq)1235 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1236 {
1237 	if (sched_core_enabled(rq))
1238 		return &rq->core->__lock;
1239 
1240 	return &rq->__lock;
1241 }
1242 
__rq_lockp(struct rq * rq)1243 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1244 {
1245 	if (rq->core_enabled)
1246 		return &rq->core->__lock;
1247 
1248 	return &rq->__lock;
1249 }
1250 
1251 bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1252 			bool fi);
1253 void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1254 
1255 /*
1256  * Helpers to check if the CPU's core cookie matches with the task's cookie
1257  * when core scheduling is enabled.
1258  * A special case is that the task's cookie always matches with CPU's core
1259  * cookie if the CPU is in an idle core.
1260  */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1261 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1262 {
1263 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1264 	if (!sched_core_enabled(rq))
1265 		return true;
1266 
1267 	return rq->core->core_cookie == p->core_cookie;
1268 }
1269 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1270 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1271 {
1272 	bool idle_core = true;
1273 	int cpu;
1274 
1275 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1276 	if (!sched_core_enabled(rq))
1277 		return true;
1278 
1279 	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1280 		if (!available_idle_cpu(cpu)) {
1281 			idle_core = false;
1282 			break;
1283 		}
1284 	}
1285 
1286 	/*
1287 	 * A CPU in an idle core is always the best choice for tasks with
1288 	 * cookies.
1289 	 */
1290 	return idle_core || rq->core->core_cookie == p->core_cookie;
1291 }
1292 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1293 static inline bool sched_group_cookie_match(struct rq *rq,
1294 					    struct task_struct *p,
1295 					    struct sched_group *group)
1296 {
1297 	int cpu;
1298 
1299 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1300 	if (!sched_core_enabled(rq))
1301 		return true;
1302 
1303 	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1304 		if (sched_core_cookie_match(cpu_rq(cpu), p))
1305 			return true;
1306 	}
1307 	return false;
1308 }
1309 
sched_core_enqueued(struct task_struct * p)1310 static inline bool sched_core_enqueued(struct task_struct *p)
1311 {
1312 	return !RB_EMPTY_NODE(&p->core_node);
1313 }
1314 
1315 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1316 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1317 
1318 extern void sched_core_get(void);
1319 extern void sched_core_put(void);
1320 
1321 #else /* !CONFIG_SCHED_CORE */
1322 
sched_core_enabled(struct rq * rq)1323 static inline bool sched_core_enabled(struct rq *rq)
1324 {
1325 	return false;
1326 }
1327 
sched_core_disabled(void)1328 static inline bool sched_core_disabled(void)
1329 {
1330 	return true;
1331 }
1332 
rq_lockp(struct rq * rq)1333 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1334 {
1335 	return &rq->__lock;
1336 }
1337 
__rq_lockp(struct rq * rq)1338 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1339 {
1340 	return &rq->__lock;
1341 }
1342 
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1343 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1344 {
1345 	return true;
1346 }
1347 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1348 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1349 {
1350 	return true;
1351 }
1352 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1353 static inline bool sched_group_cookie_match(struct rq *rq,
1354 					    struct task_struct *p,
1355 					    struct sched_group *group)
1356 {
1357 	return true;
1358 }
1359 #endif /* CONFIG_SCHED_CORE */
1360 
lockdep_assert_rq_held(struct rq * rq)1361 static inline void lockdep_assert_rq_held(struct rq *rq)
1362 {
1363 	lockdep_assert_held(__rq_lockp(rq));
1364 }
1365 
1366 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1367 extern bool raw_spin_rq_trylock(struct rq *rq);
1368 extern void raw_spin_rq_unlock(struct rq *rq);
1369 
raw_spin_rq_lock(struct rq * rq)1370 static inline void raw_spin_rq_lock(struct rq *rq)
1371 {
1372 	raw_spin_rq_lock_nested(rq, 0);
1373 }
1374 
raw_spin_rq_lock_irq(struct rq * rq)1375 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1376 {
1377 	local_irq_disable();
1378 	raw_spin_rq_lock(rq);
1379 }
1380 
raw_spin_rq_unlock_irq(struct rq * rq)1381 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1382 {
1383 	raw_spin_rq_unlock(rq);
1384 	local_irq_enable();
1385 }
1386 
_raw_spin_rq_lock_irqsave(struct rq * rq)1387 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1388 {
1389 	unsigned long flags;
1390 	local_irq_save(flags);
1391 	raw_spin_rq_lock(rq);
1392 	return flags;
1393 }
1394 
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1395 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1396 {
1397 	raw_spin_rq_unlock(rq);
1398 	local_irq_restore(flags);
1399 }
1400 
1401 #define raw_spin_rq_lock_irqsave(rq, flags)	\
1402 do {						\
1403 	flags = _raw_spin_rq_lock_irqsave(rq);	\
1404 } while (0)
1405 
1406 #ifdef CONFIG_SCHED_SMT
1407 extern void __update_idle_core(struct rq *rq);
1408 
update_idle_core(struct rq * rq)1409 static inline void update_idle_core(struct rq *rq)
1410 {
1411 	if (static_branch_unlikely(&sched_smt_present))
1412 		__update_idle_core(rq);
1413 }
1414 
1415 #else
update_idle_core(struct rq * rq)1416 static inline void update_idle_core(struct rq *rq) { }
1417 #endif
1418 
1419 #ifdef CONFIG_FAIR_GROUP_SCHED
task_of(struct sched_entity * se)1420 static inline struct task_struct *task_of(struct sched_entity *se)
1421 {
1422 	SCHED_WARN_ON(!entity_is_task(se));
1423 	return container_of(se, struct task_struct, se);
1424 }
1425 
task_cfs_rq(struct task_struct * p)1426 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1427 {
1428 	return p->se.cfs_rq;
1429 }
1430 
1431 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(const struct sched_entity * se)1432 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1433 {
1434 	return se->cfs_rq;
1435 }
1436 
1437 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1438 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1439 {
1440 	return grp->my_q;
1441 }
1442 
1443 #else
1444 
1445 #define task_of(_se)	container_of(_se, struct task_struct, se)
1446 
task_cfs_rq(const struct task_struct * p)1447 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1448 {
1449 	return &task_rq(p)->cfs;
1450 }
1451 
cfs_rq_of(const struct sched_entity * se)1452 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1453 {
1454 	const struct task_struct *p = task_of(se);
1455 	struct rq *rq = task_rq(p);
1456 
1457 	return &rq->cfs;
1458 }
1459 
1460 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1461 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1462 {
1463 	return NULL;
1464 }
1465 #endif
1466 
1467 extern void update_rq_clock(struct rq *rq);
1468 
1469 /*
1470  * rq::clock_update_flags bits
1471  *
1472  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1473  *  call to __schedule(). This is an optimisation to avoid
1474  *  neighbouring rq clock updates.
1475  *
1476  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1477  *  in effect and calls to update_rq_clock() are being ignored.
1478  *
1479  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1480  *  made to update_rq_clock() since the last time rq::lock was pinned.
1481  *
1482  * If inside of __schedule(), clock_update_flags will have been
1483  * shifted left (a left shift is a cheap operation for the fast path
1484  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1485  *
1486  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1487  *
1488  * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1489  * one position though, because the next rq_unpin_lock() will shift it
1490  * back.
1491  */
1492 #define RQCF_REQ_SKIP		0x01
1493 #define RQCF_ACT_SKIP		0x02
1494 #define RQCF_UPDATED		0x04
1495 
assert_clock_updated(struct rq * rq)1496 static inline void assert_clock_updated(struct rq *rq)
1497 {
1498 	/*
1499 	 * The only reason for not seeing a clock update since the
1500 	 * last rq_pin_lock() is if we're currently skipping updates.
1501 	 */
1502 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1503 }
1504 
rq_clock(struct rq * rq)1505 static inline u64 rq_clock(struct rq *rq)
1506 {
1507 	lockdep_assert_rq_held(rq);
1508 	assert_clock_updated(rq);
1509 
1510 	return rq->clock;
1511 }
1512 
rq_clock_task(struct rq * rq)1513 static inline u64 rq_clock_task(struct rq *rq)
1514 {
1515 	lockdep_assert_rq_held(rq);
1516 	assert_clock_updated(rq);
1517 
1518 	return rq->clock_task;
1519 }
1520 
1521 /**
1522  * By default the decay is the default pelt decay period.
1523  * The decay shift can change the decay period in
1524  * multiples of 32.
1525  *  Decay shift		Decay period(ms)
1526  *	0			32
1527  *	1			64
1528  *	2			128
1529  *	3			256
1530  *	4			512
1531  */
1532 extern int sched_thermal_decay_shift;
1533 
rq_clock_thermal(struct rq * rq)1534 static inline u64 rq_clock_thermal(struct rq *rq)
1535 {
1536 	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1537 }
1538 
rq_clock_skip_update(struct rq * rq)1539 static inline void rq_clock_skip_update(struct rq *rq)
1540 {
1541 	lockdep_assert_rq_held(rq);
1542 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1543 }
1544 
1545 /*
1546  * See rt task throttling, which is the only time a skip
1547  * request is canceled.
1548  */
rq_clock_cancel_skipupdate(struct rq * rq)1549 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1550 {
1551 	lockdep_assert_rq_held(rq);
1552 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1553 }
1554 
1555 /*
1556  * During cpu offlining and rq wide unthrottling, we can trigger
1557  * an update_rq_clock() for several cfs and rt runqueues (Typically
1558  * when using list_for_each_entry_*)
1559  * rq_clock_start_loop_update() can be called after updating the clock
1560  * once and before iterating over the list to prevent multiple update.
1561  * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1562  * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1563  */
rq_clock_start_loop_update(struct rq * rq)1564 static inline void rq_clock_start_loop_update(struct rq *rq)
1565 {
1566 	lockdep_assert_rq_held(rq);
1567 	SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1568 	rq->clock_update_flags |= RQCF_ACT_SKIP;
1569 }
1570 
rq_clock_stop_loop_update(struct rq * rq)1571 static inline void rq_clock_stop_loop_update(struct rq *rq)
1572 {
1573 	lockdep_assert_rq_held(rq);
1574 	rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1575 }
1576 
1577 struct rq_flags {
1578 	unsigned long flags;
1579 	struct pin_cookie cookie;
1580 #ifdef CONFIG_SCHED_DEBUG
1581 	/*
1582 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1583 	 * current pin context is stashed here in case it needs to be
1584 	 * restored in rq_repin_lock().
1585 	 */
1586 	unsigned int clock_update_flags;
1587 #endif
1588 };
1589 
1590 extern struct balance_callback balance_push_callback;
1591 
1592 /*
1593  * Lockdep annotation that avoids accidental unlocks; it's like a
1594  * sticky/continuous lockdep_assert_held().
1595  *
1596  * This avoids code that has access to 'struct rq *rq' (basically everything in
1597  * the scheduler) from accidentally unlocking the rq if they do not also have a
1598  * copy of the (on-stack) 'struct rq_flags rf'.
1599  *
1600  * Also see Documentation/locking/lockdep-design.rst.
1601  */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1602 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1603 {
1604 	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1605 
1606 #ifdef CONFIG_SCHED_DEBUG
1607 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1608 	rf->clock_update_flags = 0;
1609 #ifdef CONFIG_SMP
1610 	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1611 #endif
1612 #endif
1613 }
1614 
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1615 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1616 {
1617 #ifdef CONFIG_SCHED_DEBUG
1618 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1619 		rf->clock_update_flags = RQCF_UPDATED;
1620 #endif
1621 
1622 	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1623 }
1624 
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1625 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1626 {
1627 	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1628 
1629 #ifdef CONFIG_SCHED_DEBUG
1630 	/*
1631 	 * Restore the value we stashed in @rf for this pin context.
1632 	 */
1633 	rq->clock_update_flags |= rf->clock_update_flags;
1634 #endif
1635 }
1636 
1637 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1638 	__acquires(rq->lock);
1639 
1640 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1641 	__acquires(p->pi_lock)
1642 	__acquires(rq->lock);
1643 
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1644 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1645 	__releases(rq->lock)
1646 {
1647 	rq_unpin_lock(rq, rf);
1648 	raw_spin_rq_unlock(rq);
1649 }
1650 
1651 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1652 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1653 	__releases(rq->lock)
1654 	__releases(p->pi_lock)
1655 {
1656 	rq_unpin_lock(rq, rf);
1657 	raw_spin_rq_unlock(rq);
1658 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1659 }
1660 
1661 static inline void
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1662 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1663 	__acquires(rq->lock)
1664 {
1665 	raw_spin_rq_lock_irqsave(rq, rf->flags);
1666 	rq_pin_lock(rq, rf);
1667 }
1668 
1669 static inline void
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1670 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1671 	__acquires(rq->lock)
1672 {
1673 	raw_spin_rq_lock_irq(rq);
1674 	rq_pin_lock(rq, rf);
1675 }
1676 
1677 static inline void
rq_lock(struct rq * rq,struct rq_flags * rf)1678 rq_lock(struct rq *rq, struct rq_flags *rf)
1679 	__acquires(rq->lock)
1680 {
1681 	raw_spin_rq_lock(rq);
1682 	rq_pin_lock(rq, rf);
1683 }
1684 
1685 static inline void
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1686 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1687 	__releases(rq->lock)
1688 {
1689 	rq_unpin_lock(rq, rf);
1690 	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1691 }
1692 
1693 static inline void
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1694 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1695 	__releases(rq->lock)
1696 {
1697 	rq_unpin_lock(rq, rf);
1698 	raw_spin_rq_unlock_irq(rq);
1699 }
1700 
1701 static inline void
rq_unlock(struct rq * rq,struct rq_flags * rf)1702 rq_unlock(struct rq *rq, struct rq_flags *rf)
1703 	__releases(rq->lock)
1704 {
1705 	rq_unpin_lock(rq, rf);
1706 	raw_spin_rq_unlock(rq);
1707 }
1708 
1709 DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1710 		    rq_lock(_T->lock, &_T->rf),
1711 		    rq_unlock(_T->lock, &_T->rf),
1712 		    struct rq_flags rf)
1713 
1714 DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1715 		    rq_lock_irq(_T->lock, &_T->rf),
1716 		    rq_unlock_irq(_T->lock, &_T->rf),
1717 		    struct rq_flags rf)
1718 
1719 DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1720 		    rq_lock_irqsave(_T->lock, &_T->rf),
1721 		    rq_unlock_irqrestore(_T->lock, &_T->rf),
1722 		    struct rq_flags rf)
1723 
1724 static inline struct rq *
this_rq_lock_irq(struct rq_flags * rf)1725 this_rq_lock_irq(struct rq_flags *rf)
1726 	__acquires(rq->lock)
1727 {
1728 	struct rq *rq;
1729 
1730 	local_irq_disable();
1731 	rq = this_rq();
1732 	rq_lock(rq, rf);
1733 	return rq;
1734 }
1735 
1736 #ifdef CONFIG_NUMA
1737 enum numa_topology_type {
1738 	NUMA_DIRECT,
1739 	NUMA_GLUELESS_MESH,
1740 	NUMA_BACKPLANE,
1741 };
1742 extern enum numa_topology_type sched_numa_topology_type;
1743 extern int sched_max_numa_distance;
1744 extern bool find_numa_distance(int distance);
1745 extern void sched_init_numa(int offline_node);
1746 extern void sched_update_numa(int cpu, bool online);
1747 extern void sched_domains_numa_masks_set(unsigned int cpu);
1748 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1749 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1750 #else
sched_init_numa(int offline_node)1751 static inline void sched_init_numa(int offline_node) { }
sched_update_numa(int cpu,bool online)1752 static inline void sched_update_numa(int cpu, bool online) { }
sched_domains_numa_masks_set(unsigned int cpu)1753 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1754 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1755 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1756 {
1757 	return nr_cpu_ids;
1758 }
1759 #endif
1760 
1761 #ifdef CONFIG_NUMA_BALANCING
1762 /* The regions in numa_faults array from task_struct */
1763 enum numa_faults_stats {
1764 	NUMA_MEM = 0,
1765 	NUMA_CPU,
1766 	NUMA_MEMBUF,
1767 	NUMA_CPUBUF
1768 };
1769 extern void sched_setnuma(struct task_struct *p, int node);
1770 extern int migrate_task_to(struct task_struct *p, int cpu);
1771 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1772 			int cpu, int scpu);
1773 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1774 #else
1775 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1776 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1777 {
1778 }
1779 #endif /* CONFIG_NUMA_BALANCING */
1780 
1781 #ifdef CONFIG_SMP
1782 
1783 static inline void
queue_balance_callback(struct rq * rq,struct balance_callback * head,void (* func)(struct rq * rq))1784 queue_balance_callback(struct rq *rq,
1785 		       struct balance_callback *head,
1786 		       void (*func)(struct rq *rq))
1787 {
1788 	lockdep_assert_rq_held(rq);
1789 
1790 	/*
1791 	 * Don't (re)queue an already queued item; nor queue anything when
1792 	 * balance_push() is active, see the comment with
1793 	 * balance_push_callback.
1794 	 */
1795 	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1796 		return;
1797 
1798 	head->func = func;
1799 	head->next = rq->balance_callback;
1800 	rq->balance_callback = head;
1801 }
1802 
1803 #define rcu_dereference_check_sched_domain(p) \
1804 	rcu_dereference_check((p), \
1805 			      lockdep_is_held(&sched_domains_mutex))
1806 
1807 /*
1808  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1809  * See destroy_sched_domains: call_rcu for details.
1810  *
1811  * The domain tree of any CPU may only be accessed from within
1812  * preempt-disabled sections.
1813  */
1814 #define for_each_domain(cpu, __sd) \
1815 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1816 			__sd; __sd = __sd->parent)
1817 
1818 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1819 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1820 static const unsigned int SD_SHARED_CHILD_MASK =
1821 #include <linux/sched/sd_flags.h>
1822 0;
1823 #undef SD_FLAG
1824 
1825 /**
1826  * highest_flag_domain - Return highest sched_domain containing flag.
1827  * @cpu:	The CPU whose highest level of sched domain is to
1828  *		be returned.
1829  * @flag:	The flag to check for the highest sched_domain
1830  *		for the given CPU.
1831  *
1832  * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1833  * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1834  */
highest_flag_domain(int cpu,int flag)1835 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1836 {
1837 	struct sched_domain *sd, *hsd = NULL;
1838 
1839 	for_each_domain(cpu, sd) {
1840 		if (sd->flags & flag) {
1841 			hsd = sd;
1842 			continue;
1843 		}
1844 
1845 		/*
1846 		 * Stop the search if @flag is known to be shared at lower
1847 		 * levels. It will not be found further up.
1848 		 */
1849 		if (flag & SD_SHARED_CHILD_MASK)
1850 			break;
1851 	}
1852 
1853 	return hsd;
1854 }
1855 
lowest_flag_domain(int cpu,int flag)1856 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1857 {
1858 	struct sched_domain *sd;
1859 
1860 	for_each_domain(cpu, sd) {
1861 		if (sd->flags & flag)
1862 			break;
1863 	}
1864 
1865 	return sd;
1866 }
1867 
1868 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1869 DECLARE_PER_CPU(int, sd_llc_size);
1870 DECLARE_PER_CPU(int, sd_llc_id);
1871 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1872 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1873 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1874 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1875 extern struct static_key_false sched_asym_cpucapacity;
1876 
sched_asym_cpucap_active(void)1877 static __always_inline bool sched_asym_cpucap_active(void)
1878 {
1879 	return static_branch_unlikely(&sched_asym_cpucapacity);
1880 }
1881 
1882 struct sched_group_capacity {
1883 	atomic_t		ref;
1884 	/*
1885 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1886 	 * for a single CPU.
1887 	 */
1888 	unsigned long		capacity;
1889 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1890 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1891 	unsigned long		next_update;
1892 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1893 
1894 #ifdef CONFIG_SCHED_DEBUG
1895 	int			id;
1896 #endif
1897 
1898 	unsigned long		cpumask[];		/* Balance mask */
1899 };
1900 
1901 struct sched_group {
1902 	struct sched_group	*next;			/* Must be a circular list */
1903 	atomic_t		ref;
1904 
1905 	unsigned int		group_weight;
1906 	unsigned int		cores;
1907 	struct sched_group_capacity *sgc;
1908 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1909 	int			flags;
1910 
1911 	/*
1912 	 * The CPUs this group covers.
1913 	 *
1914 	 * NOTE: this field is variable length. (Allocated dynamically
1915 	 * by attaching extra space to the end of the structure,
1916 	 * depending on how many CPUs the kernel has booted up with)
1917 	 */
1918 	unsigned long		cpumask[];
1919 };
1920 
sched_group_span(struct sched_group * sg)1921 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1922 {
1923 	return to_cpumask(sg->cpumask);
1924 }
1925 
1926 /*
1927  * See build_balance_mask().
1928  */
group_balance_mask(struct sched_group * sg)1929 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1930 {
1931 	return to_cpumask(sg->sgc->cpumask);
1932 }
1933 
1934 extern int group_balance_cpu(struct sched_group *sg);
1935 
1936 #ifdef CONFIG_SCHED_DEBUG
1937 void update_sched_domain_debugfs(void);
1938 void dirty_sched_domain_sysctl(int cpu);
1939 #else
update_sched_domain_debugfs(void)1940 static inline void update_sched_domain_debugfs(void)
1941 {
1942 }
dirty_sched_domain_sysctl(int cpu)1943 static inline void dirty_sched_domain_sysctl(int cpu)
1944 {
1945 }
1946 #endif
1947 
1948 extern int sched_update_scaling(void);
1949 
task_user_cpus(struct task_struct * p)1950 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1951 {
1952 	if (!p->user_cpus_ptr)
1953 		return cpu_possible_mask; /* &init_task.cpus_mask */
1954 	return p->user_cpus_ptr;
1955 }
1956 #endif /* CONFIG_SMP */
1957 
1958 #include "stats.h"
1959 
1960 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1961 
1962 extern void __sched_core_account_forceidle(struct rq *rq);
1963 
sched_core_account_forceidle(struct rq * rq)1964 static inline void sched_core_account_forceidle(struct rq *rq)
1965 {
1966 	if (schedstat_enabled())
1967 		__sched_core_account_forceidle(rq);
1968 }
1969 
1970 extern void __sched_core_tick(struct rq *rq);
1971 
sched_core_tick(struct rq * rq)1972 static inline void sched_core_tick(struct rq *rq)
1973 {
1974 	if (sched_core_enabled(rq) && schedstat_enabled())
1975 		__sched_core_tick(rq);
1976 }
1977 
1978 #else
1979 
sched_core_account_forceidle(struct rq * rq)1980 static inline void sched_core_account_forceidle(struct rq *rq) {}
1981 
sched_core_tick(struct rq * rq)1982 static inline void sched_core_tick(struct rq *rq) {}
1983 
1984 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1985 
1986 #ifdef CONFIG_CGROUP_SCHED
1987 
1988 /*
1989  * Return the group to which this tasks belongs.
1990  *
1991  * We cannot use task_css() and friends because the cgroup subsystem
1992  * changes that value before the cgroup_subsys::attach() method is called,
1993  * therefore we cannot pin it and might observe the wrong value.
1994  *
1995  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1996  * core changes this before calling sched_move_task().
1997  *
1998  * Instead we use a 'copy' which is updated from sched_move_task() while
1999  * holding both task_struct::pi_lock and rq::lock.
2000  */
task_group(struct task_struct * p)2001 static inline struct task_group *task_group(struct task_struct *p)
2002 {
2003 	return p->sched_task_group;
2004 }
2005 
2006 /* 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)2007 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2008 {
2009 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2010 	struct task_group *tg = task_group(p);
2011 #endif
2012 
2013 #ifdef CONFIG_FAIR_GROUP_SCHED
2014 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2015 	p->se.cfs_rq = tg->cfs_rq[cpu];
2016 	p->se.parent = tg->se[cpu];
2017 	p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2018 #endif
2019 
2020 #ifdef CONFIG_RT_GROUP_SCHED
2021 	p->rt.rt_rq  = tg->rt_rq[cpu];
2022 	p->rt.parent = tg->rt_se[cpu];
2023 #endif
2024 }
2025 
2026 #else /* CONFIG_CGROUP_SCHED */
2027 
set_task_rq(struct task_struct * p,unsigned int cpu)2028 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
task_group(struct task_struct * p)2029 static inline struct task_group *task_group(struct task_struct *p)
2030 {
2031 	return NULL;
2032 }
2033 
2034 #endif /* CONFIG_CGROUP_SCHED */
2035 
__set_task_cpu(struct task_struct * p,unsigned int cpu)2036 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2037 {
2038 	set_task_rq(p, cpu);
2039 #ifdef CONFIG_SMP
2040 	/*
2041 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2042 	 * successfully executed on another CPU. We must ensure that updates of
2043 	 * per-task data have been completed by this moment.
2044 	 */
2045 	smp_wmb();
2046 	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2047 	p->wake_cpu = cpu;
2048 #endif
2049 }
2050 
2051 /*
2052  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2053  */
2054 #ifdef CONFIG_SCHED_DEBUG
2055 # define const_debug __read_mostly
2056 #else
2057 # define const_debug const
2058 #endif
2059 
2060 #define SCHED_FEAT(name, enabled)	\
2061 	__SCHED_FEAT_##name ,
2062 
2063 enum {
2064 #include "features.h"
2065 	__SCHED_FEAT_NR,
2066 };
2067 
2068 #undef SCHED_FEAT
2069 
2070 #ifdef CONFIG_SCHED_DEBUG
2071 
2072 /*
2073  * To support run-time toggling of sched features, all the translation units
2074  * (but core.c) reference the sysctl_sched_features defined in core.c.
2075  */
2076 extern const_debug unsigned int sysctl_sched_features;
2077 
2078 #ifdef CONFIG_JUMP_LABEL
2079 #define SCHED_FEAT(name, enabled)					\
2080 static __always_inline bool static_branch_##name(struct static_key *key) \
2081 {									\
2082 	return static_key_##enabled(key);				\
2083 }
2084 
2085 #include "features.h"
2086 #undef SCHED_FEAT
2087 
2088 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2089 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2090 
2091 #else /* !CONFIG_JUMP_LABEL */
2092 
2093 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2094 
2095 #endif /* CONFIG_JUMP_LABEL */
2096 
2097 #else /* !SCHED_DEBUG */
2098 
2099 /*
2100  * Each translation unit has its own copy of sysctl_sched_features to allow
2101  * constants propagation at compile time and compiler optimization based on
2102  * features default.
2103  */
2104 #define SCHED_FEAT(name, enabled)	\
2105 	(1UL << __SCHED_FEAT_##name) * enabled |
2106 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2107 #include "features.h"
2108 	0;
2109 #undef SCHED_FEAT
2110 
2111 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2112 
2113 #endif /* SCHED_DEBUG */
2114 
2115 extern struct static_key_false sched_numa_balancing;
2116 extern struct static_key_false sched_schedstats;
2117 
global_rt_period(void)2118 static inline u64 global_rt_period(void)
2119 {
2120 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2121 }
2122 
global_rt_runtime(void)2123 static inline u64 global_rt_runtime(void)
2124 {
2125 	if (sysctl_sched_rt_runtime < 0)
2126 		return RUNTIME_INF;
2127 
2128 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2129 }
2130 
task_current(struct rq * rq,struct task_struct * p)2131 static inline int task_current(struct rq *rq, struct task_struct *p)
2132 {
2133 	return rq->curr == p;
2134 }
2135 
task_on_cpu(struct rq * rq,struct task_struct * p)2136 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2137 {
2138 #ifdef CONFIG_SMP
2139 	return p->on_cpu;
2140 #else
2141 	return task_current(rq, p);
2142 #endif
2143 }
2144 
task_on_rq_queued(struct task_struct * p)2145 static inline int task_on_rq_queued(struct task_struct *p)
2146 {
2147 	return p->on_rq == TASK_ON_RQ_QUEUED;
2148 }
2149 
task_on_rq_migrating(struct task_struct * p)2150 static inline int task_on_rq_migrating(struct task_struct *p)
2151 {
2152 	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2153 }
2154 
2155 /* Wake flags. The first three directly map to some SD flag value */
2156 #define WF_EXEC         0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2157 #define WF_FORK         0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2158 #define WF_TTWU         0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2159 
2160 #define WF_SYNC         0x10 /* Waker goes to sleep after wakeup */
2161 #define WF_MIGRATED     0x20 /* Internal use, task got migrated */
2162 #define WF_CURRENT_CPU  0x40 /* Prefer to move the wakee to the current CPU. */
2163 
2164 #ifdef CONFIG_SMP
2165 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2166 static_assert(WF_FORK == SD_BALANCE_FORK);
2167 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2168 #endif
2169 
2170 /*
2171  * To aid in avoiding the subversion of "niceness" due to uneven distribution
2172  * of tasks with abnormal "nice" values across CPUs the contribution that
2173  * each task makes to its run queue's load is weighted according to its
2174  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2175  * scaled version of the new time slice allocation that they receive on time
2176  * slice expiry etc.
2177  */
2178 
2179 #define WEIGHT_IDLEPRIO		3
2180 #define WMULT_IDLEPRIO		1431655765
2181 
2182 extern const int		sched_prio_to_weight[40];
2183 extern const u32		sched_prio_to_wmult[40];
2184 
2185 /*
2186  * {de,en}queue flags:
2187  *
2188  * DEQUEUE_SLEEP  - task is no longer runnable
2189  * ENQUEUE_WAKEUP - task just became runnable
2190  *
2191  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2192  *                are in a known state which allows modification. Such pairs
2193  *                should preserve as much state as possible.
2194  *
2195  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2196  *        in the runqueue.
2197  *
2198  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2199  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2200  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2201  *
2202  */
2203 
2204 #define DEQUEUE_SLEEP		0x01
2205 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2206 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2207 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
2208 
2209 #define ENQUEUE_WAKEUP		0x01
2210 #define ENQUEUE_RESTORE		0x02
2211 #define ENQUEUE_MOVE		0x04
2212 #define ENQUEUE_NOCLOCK		0x08
2213 
2214 #define ENQUEUE_HEAD		0x10
2215 #define ENQUEUE_REPLENISH	0x20
2216 #ifdef CONFIG_SMP
2217 #define ENQUEUE_MIGRATED	0x40
2218 #else
2219 #define ENQUEUE_MIGRATED	0x00
2220 #endif
2221 #define ENQUEUE_INITIAL		0x80
2222 
2223 #define RETRY_TASK		((void *)-1UL)
2224 
2225 struct affinity_context {
2226 	const struct cpumask *new_mask;
2227 	struct cpumask *user_mask;
2228 	unsigned int flags;
2229 };
2230 
2231 struct sched_class {
2232 
2233 #ifdef CONFIG_UCLAMP_TASK
2234 	int uclamp_enabled;
2235 #endif
2236 
2237 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2238 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2239 	void (*yield_task)   (struct rq *rq);
2240 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2241 
2242 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2243 
2244 	struct task_struct *(*pick_next_task)(struct rq *rq);
2245 
2246 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2247 	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2248 
2249 #ifdef CONFIG_SMP
2250 	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2251 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2252 
2253 	struct task_struct * (*pick_task)(struct rq *rq);
2254 
2255 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2256 
2257 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2258 
2259 	void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2260 
2261 	void (*rq_online)(struct rq *rq);
2262 	void (*rq_offline)(struct rq *rq);
2263 
2264 	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2265 #endif
2266 
2267 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2268 	void (*task_fork)(struct task_struct *p);
2269 	void (*task_dead)(struct task_struct *p);
2270 
2271 	/*
2272 	 * The switched_from() call is allowed to drop rq->lock, therefore we
2273 	 * cannot assume the switched_from/switched_to pair is serialized by
2274 	 * rq->lock. They are however serialized by p->pi_lock.
2275 	 */
2276 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2277 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2278 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2279 			      int oldprio);
2280 
2281 	unsigned int (*get_rr_interval)(struct rq *rq,
2282 					struct task_struct *task);
2283 
2284 	void (*update_curr)(struct rq *rq);
2285 
2286 #ifdef CONFIG_FAIR_GROUP_SCHED
2287 	void (*task_change_group)(struct task_struct *p);
2288 #endif
2289 
2290 #ifdef CONFIG_SCHED_CORE
2291 	int (*task_is_throttled)(struct task_struct *p, int cpu);
2292 #endif
2293 };
2294 
put_prev_task(struct rq * rq,struct task_struct * prev)2295 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2296 {
2297 	WARN_ON_ONCE(rq->curr != prev);
2298 	prev->sched_class->put_prev_task(rq, prev);
2299 }
2300 
set_next_task(struct rq * rq,struct task_struct * next)2301 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2302 {
2303 	next->sched_class->set_next_task(rq, next, false);
2304 }
2305 
2306 
2307 /*
2308  * Helper to define a sched_class instance; each one is placed in a separate
2309  * section which is ordered by the linker script:
2310  *
2311  *   include/asm-generic/vmlinux.lds.h
2312  *
2313  * *CAREFUL* they are laid out in *REVERSE* order!!!
2314  *
2315  * Also enforce alignment on the instance, not the type, to guarantee layout.
2316  */
2317 #define DEFINE_SCHED_CLASS(name) \
2318 const struct sched_class name##_sched_class \
2319 	__aligned(__alignof__(struct sched_class)) \
2320 	__section("__" #name "_sched_class")
2321 
2322 /* Defined in include/asm-generic/vmlinux.lds.h */
2323 extern struct sched_class __sched_class_highest[];
2324 extern struct sched_class __sched_class_lowest[];
2325 
2326 #define for_class_range(class, _from, _to) \
2327 	for (class = (_from); class < (_to); class++)
2328 
2329 #define for_each_class(class) \
2330 	for_class_range(class, __sched_class_highest, __sched_class_lowest)
2331 
2332 #define sched_class_above(_a, _b)	((_a) < (_b))
2333 
2334 extern const struct sched_class stop_sched_class;
2335 extern const struct sched_class dl_sched_class;
2336 extern const struct sched_class rt_sched_class;
2337 extern const struct sched_class fair_sched_class;
2338 extern const struct sched_class idle_sched_class;
2339 
sched_stop_runnable(struct rq * rq)2340 static inline bool sched_stop_runnable(struct rq *rq)
2341 {
2342 	return rq->stop && task_on_rq_queued(rq->stop);
2343 }
2344 
sched_dl_runnable(struct rq * rq)2345 static inline bool sched_dl_runnable(struct rq *rq)
2346 {
2347 	return rq->dl.dl_nr_running > 0;
2348 }
2349 
sched_rt_runnable(struct rq * rq)2350 static inline bool sched_rt_runnable(struct rq *rq)
2351 {
2352 	return rq->rt.rt_queued > 0;
2353 }
2354 
sched_fair_runnable(struct rq * rq)2355 static inline bool sched_fair_runnable(struct rq *rq)
2356 {
2357 	return rq->cfs.nr_running > 0;
2358 }
2359 
2360 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2361 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2362 
2363 #define SCA_CHECK		0x01
2364 #define SCA_MIGRATE_DISABLE	0x02
2365 #define SCA_MIGRATE_ENABLE	0x04
2366 #define SCA_USER		0x08
2367 
2368 #ifdef CONFIG_SMP
2369 
2370 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2371 
2372 extern void trigger_load_balance(struct rq *rq);
2373 
2374 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2375 
get_push_task(struct rq * rq)2376 static inline struct task_struct *get_push_task(struct rq *rq)
2377 {
2378 	struct task_struct *p = rq->curr;
2379 
2380 	lockdep_assert_rq_held(rq);
2381 
2382 	if (rq->push_busy)
2383 		return NULL;
2384 
2385 	if (p->nr_cpus_allowed == 1)
2386 		return NULL;
2387 
2388 	if (p->migration_disabled)
2389 		return NULL;
2390 
2391 	rq->push_busy = true;
2392 	return get_task_struct(p);
2393 }
2394 
2395 extern int push_cpu_stop(void *arg);
2396 
2397 #endif
2398 
2399 #ifdef CONFIG_CPU_IDLE
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2400 static inline void idle_set_state(struct rq *rq,
2401 				  struct cpuidle_state *idle_state)
2402 {
2403 	rq->idle_state = idle_state;
2404 }
2405 
idle_get_state(struct rq * rq)2406 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2407 {
2408 	SCHED_WARN_ON(!rcu_read_lock_held());
2409 
2410 	return rq->idle_state;
2411 }
2412 #else
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2413 static inline void idle_set_state(struct rq *rq,
2414 				  struct cpuidle_state *idle_state)
2415 {
2416 }
2417 
idle_get_state(struct rq * rq)2418 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2419 {
2420 	return NULL;
2421 }
2422 #endif
2423 
2424 extern void schedule_idle(void);
2425 asmlinkage void schedule_user(void);
2426 
2427 extern void sysrq_sched_debug_show(void);
2428 extern void sched_init_granularity(void);
2429 extern void update_max_interval(void);
2430 
2431 extern void init_sched_dl_class(void);
2432 extern void init_sched_rt_class(void);
2433 extern void init_sched_fair_class(void);
2434 
2435 extern void reweight_task(struct task_struct *p, int prio);
2436 
2437 extern void resched_curr(struct rq *rq);
2438 extern void resched_cpu(int cpu);
2439 
2440 extern struct rt_bandwidth def_rt_bandwidth;
2441 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2442 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2443 
2444 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2445 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2446 
2447 #define BW_SHIFT		20
2448 #define BW_UNIT			(1 << BW_SHIFT)
2449 #define RATIO_SHIFT		8
2450 #define MAX_BW_BITS		(64 - BW_SHIFT)
2451 #define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2452 unsigned long to_ratio(u64 period, u64 runtime);
2453 
2454 extern void init_entity_runnable_average(struct sched_entity *se);
2455 extern void post_init_entity_util_avg(struct task_struct *p);
2456 
2457 #ifdef CONFIG_NO_HZ_FULL
2458 extern bool sched_can_stop_tick(struct rq *rq);
2459 extern int __init sched_tick_offload_init(void);
2460 
2461 /*
2462  * Tick may be needed by tasks in the runqueue depending on their policy and
2463  * requirements. If tick is needed, lets send the target an IPI to kick it out of
2464  * nohz mode if necessary.
2465  */
sched_update_tick_dependency(struct rq * rq)2466 static inline void sched_update_tick_dependency(struct rq *rq)
2467 {
2468 	int cpu = cpu_of(rq);
2469 
2470 	if (!tick_nohz_full_cpu(cpu))
2471 		return;
2472 
2473 	if (sched_can_stop_tick(rq))
2474 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2475 	else
2476 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2477 }
2478 #else
sched_tick_offload_init(void)2479 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2480 static inline void sched_update_tick_dependency(struct rq *rq) { }
2481 #endif
2482 
add_nr_running(struct rq * rq,unsigned count)2483 static inline void add_nr_running(struct rq *rq, unsigned count)
2484 {
2485 	unsigned prev_nr = rq->nr_running;
2486 
2487 	rq->nr_running = prev_nr + count;
2488 	if (trace_sched_update_nr_running_tp_enabled()) {
2489 		call_trace_sched_update_nr_running(rq, count);
2490 	}
2491 
2492 #ifdef CONFIG_SMP
2493 	if (prev_nr < 2 && rq->nr_running >= 2) {
2494 		if (!READ_ONCE(rq->rd->overload))
2495 			WRITE_ONCE(rq->rd->overload, 1);
2496 	}
2497 #endif
2498 
2499 	sched_update_tick_dependency(rq);
2500 }
2501 
sub_nr_running(struct rq * rq,unsigned count)2502 static inline void sub_nr_running(struct rq *rq, unsigned count)
2503 {
2504 	rq->nr_running -= count;
2505 	if (trace_sched_update_nr_running_tp_enabled()) {
2506 		call_trace_sched_update_nr_running(rq, -count);
2507 	}
2508 
2509 	/* Check if we still need preemption */
2510 	sched_update_tick_dependency(rq);
2511 }
2512 
2513 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2514 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2515 
2516 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2517 
2518 #ifdef CONFIG_PREEMPT_RT
2519 #define SCHED_NR_MIGRATE_BREAK 8
2520 #else
2521 #define SCHED_NR_MIGRATE_BREAK 32
2522 #endif
2523 
2524 extern const_debug unsigned int sysctl_sched_nr_migrate;
2525 extern const_debug unsigned int sysctl_sched_migration_cost;
2526 
2527 extern unsigned int sysctl_sched_base_slice;
2528 
2529 #ifdef CONFIG_SCHED_DEBUG
2530 extern int sysctl_resched_latency_warn_ms;
2531 extern int sysctl_resched_latency_warn_once;
2532 
2533 extern unsigned int sysctl_sched_tunable_scaling;
2534 
2535 extern unsigned int sysctl_numa_balancing_scan_delay;
2536 extern unsigned int sysctl_numa_balancing_scan_period_min;
2537 extern unsigned int sysctl_numa_balancing_scan_period_max;
2538 extern unsigned int sysctl_numa_balancing_scan_size;
2539 extern unsigned int sysctl_numa_balancing_hot_threshold;
2540 #endif
2541 
2542 #ifdef CONFIG_SCHED_HRTICK
2543 
2544 /*
2545  * Use hrtick when:
2546  *  - enabled by features
2547  *  - hrtimer is actually high res
2548  */
hrtick_enabled(struct rq * rq)2549 static inline int hrtick_enabled(struct rq *rq)
2550 {
2551 	if (!cpu_active(cpu_of(rq)))
2552 		return 0;
2553 	return hrtimer_is_hres_active(&rq->hrtick_timer);
2554 }
2555 
hrtick_enabled_fair(struct rq * rq)2556 static inline int hrtick_enabled_fair(struct rq *rq)
2557 {
2558 	if (!sched_feat(HRTICK))
2559 		return 0;
2560 	return hrtick_enabled(rq);
2561 }
2562 
hrtick_enabled_dl(struct rq * rq)2563 static inline int hrtick_enabled_dl(struct rq *rq)
2564 {
2565 	if (!sched_feat(HRTICK_DL))
2566 		return 0;
2567 	return hrtick_enabled(rq);
2568 }
2569 
2570 void hrtick_start(struct rq *rq, u64 delay);
2571 
2572 #else
2573 
hrtick_enabled_fair(struct rq * rq)2574 static inline int hrtick_enabled_fair(struct rq *rq)
2575 {
2576 	return 0;
2577 }
2578 
hrtick_enabled_dl(struct rq * rq)2579 static inline int hrtick_enabled_dl(struct rq *rq)
2580 {
2581 	return 0;
2582 }
2583 
hrtick_enabled(struct rq * rq)2584 static inline int hrtick_enabled(struct rq *rq)
2585 {
2586 	return 0;
2587 }
2588 
2589 #endif /* CONFIG_SCHED_HRTICK */
2590 
2591 #ifndef arch_scale_freq_tick
2592 static __always_inline
arch_scale_freq_tick(void)2593 void arch_scale_freq_tick(void)
2594 {
2595 }
2596 #endif
2597 
2598 #ifndef arch_scale_freq_capacity
2599 /**
2600  * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2601  * @cpu: the CPU in question.
2602  *
2603  * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2604  *
2605  *     f_curr
2606  *     ------ * SCHED_CAPACITY_SCALE
2607  *     f_max
2608  */
2609 static __always_inline
arch_scale_freq_capacity(int cpu)2610 unsigned long arch_scale_freq_capacity(int cpu)
2611 {
2612 	return SCHED_CAPACITY_SCALE;
2613 }
2614 #endif
2615 
2616 #ifdef CONFIG_SCHED_DEBUG
2617 /*
2618  * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2619  * acquire rq lock instead of rq_lock(). So at the end of these two functions
2620  * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2621  * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2622  */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2623 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2624 {
2625 	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2626 	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2627 #ifdef CONFIG_SMP
2628 	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2629 #endif
2630 }
2631 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2632 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2633 #endif
2634 
2635 #define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...)		\
2636 __DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2637 static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2638 { class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t;	\
2639   _lock; return _t; }
2640 
2641 #ifdef CONFIG_SMP
2642 
rq_order_less(struct rq * rq1,struct rq * rq2)2643 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2644 {
2645 #ifdef CONFIG_SCHED_CORE
2646 	/*
2647 	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2648 	 * order by core-id first and cpu-id second.
2649 	 *
2650 	 * Notably:
2651 	 *
2652 	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2653 	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2654 	 *
2655 	 * when only cpu-id is considered.
2656 	 */
2657 	if (rq1->core->cpu < rq2->core->cpu)
2658 		return true;
2659 	if (rq1->core->cpu > rq2->core->cpu)
2660 		return false;
2661 
2662 	/*
2663 	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2664 	 */
2665 #endif
2666 	return rq1->cpu < rq2->cpu;
2667 }
2668 
2669 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2670 
2671 #ifdef CONFIG_PREEMPTION
2672 
2673 /*
2674  * fair double_lock_balance: Safely acquires both rq->locks in a fair
2675  * way at the expense of forcing extra atomic operations in all
2676  * invocations.  This assures that the double_lock is acquired using the
2677  * same underlying policy as the spinlock_t on this architecture, which
2678  * reduces latency compared to the unfair variant below.  However, it
2679  * also adds more overhead and therefore may reduce throughput.
2680  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2681 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2682 	__releases(this_rq->lock)
2683 	__acquires(busiest->lock)
2684 	__acquires(this_rq->lock)
2685 {
2686 	raw_spin_rq_unlock(this_rq);
2687 	double_rq_lock(this_rq, busiest);
2688 
2689 	return 1;
2690 }
2691 
2692 #else
2693 /*
2694  * Unfair double_lock_balance: Optimizes throughput at the expense of
2695  * latency by eliminating extra atomic operations when the locks are
2696  * already in proper order on entry.  This favors lower CPU-ids and will
2697  * grant the double lock to lower CPUs over higher ids under contention,
2698  * regardless of entry order into the function.
2699  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2700 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2701 	__releases(this_rq->lock)
2702 	__acquires(busiest->lock)
2703 	__acquires(this_rq->lock)
2704 {
2705 	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2706 	    likely(raw_spin_rq_trylock(busiest))) {
2707 		double_rq_clock_clear_update(this_rq, busiest);
2708 		return 0;
2709 	}
2710 
2711 	if (rq_order_less(this_rq, busiest)) {
2712 		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2713 		double_rq_clock_clear_update(this_rq, busiest);
2714 		return 0;
2715 	}
2716 
2717 	raw_spin_rq_unlock(this_rq);
2718 	double_rq_lock(this_rq, busiest);
2719 
2720 	return 1;
2721 }
2722 
2723 #endif /* CONFIG_PREEMPTION */
2724 
2725 /*
2726  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2727  */
double_lock_balance(struct rq * this_rq,struct rq * busiest)2728 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2729 {
2730 	lockdep_assert_irqs_disabled();
2731 
2732 	return _double_lock_balance(this_rq, busiest);
2733 }
2734 
double_unlock_balance(struct rq * this_rq,struct rq * busiest)2735 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2736 	__releases(busiest->lock)
2737 {
2738 	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2739 		raw_spin_rq_unlock(busiest);
2740 	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2741 }
2742 
double_lock(spinlock_t * l1,spinlock_t * l2)2743 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2744 {
2745 	if (l1 > l2)
2746 		swap(l1, l2);
2747 
2748 	spin_lock(l1);
2749 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2750 }
2751 
double_lock_irq(spinlock_t * l1,spinlock_t * l2)2752 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2753 {
2754 	if (l1 > l2)
2755 		swap(l1, l2);
2756 
2757 	spin_lock_irq(l1);
2758 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2759 }
2760 
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)2761 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2762 {
2763 	if (l1 > l2)
2764 		swap(l1, l2);
2765 
2766 	raw_spin_lock(l1);
2767 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2768 }
2769 
double_raw_unlock(raw_spinlock_t * l1,raw_spinlock_t * l2)2770 static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2771 {
2772 	raw_spin_unlock(l1);
2773 	raw_spin_unlock(l2);
2774 }
2775 
2776 DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
2777 		    double_raw_lock(_T->lock, _T->lock2),
2778 		    double_raw_unlock(_T->lock, _T->lock2))
2779 
2780 /*
2781  * double_rq_unlock - safely unlock two runqueues
2782  *
2783  * Note this does not restore interrupts like task_rq_unlock,
2784  * you need to do so manually after calling.
2785  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2786 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2787 	__releases(rq1->lock)
2788 	__releases(rq2->lock)
2789 {
2790 	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2791 		raw_spin_rq_unlock(rq2);
2792 	else
2793 		__release(rq2->lock);
2794 	raw_spin_rq_unlock(rq1);
2795 }
2796 
2797 extern void set_rq_online (struct rq *rq);
2798 extern void set_rq_offline(struct rq *rq);
2799 extern bool sched_smp_initialized;
2800 
2801 #else /* CONFIG_SMP */
2802 
2803 /*
2804  * double_rq_lock - safely lock two runqueues
2805  *
2806  * Note this does not disable interrupts like task_rq_lock,
2807  * you need to do so manually before calling.
2808  */
double_rq_lock(struct rq * rq1,struct rq * rq2)2809 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2810 	__acquires(rq1->lock)
2811 	__acquires(rq2->lock)
2812 {
2813 	WARN_ON_ONCE(!irqs_disabled());
2814 	WARN_ON_ONCE(rq1 != rq2);
2815 	raw_spin_rq_lock(rq1);
2816 	__acquire(rq2->lock);	/* Fake it out ;) */
2817 	double_rq_clock_clear_update(rq1, rq2);
2818 }
2819 
2820 /*
2821  * double_rq_unlock - safely unlock two runqueues
2822  *
2823  * Note this does not restore interrupts like task_rq_unlock,
2824  * you need to do so manually after calling.
2825  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2826 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2827 	__releases(rq1->lock)
2828 	__releases(rq2->lock)
2829 {
2830 	WARN_ON_ONCE(rq1 != rq2);
2831 	raw_spin_rq_unlock(rq1);
2832 	__release(rq2->lock);
2833 }
2834 
2835 #endif
2836 
2837 DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
2838 		    double_rq_lock(_T->lock, _T->lock2),
2839 		    double_rq_unlock(_T->lock, _T->lock2))
2840 
2841 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2842 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2843 
2844 #ifdef	CONFIG_SCHED_DEBUG
2845 extern bool sched_debug_verbose;
2846 
2847 extern void print_cfs_stats(struct seq_file *m, int cpu);
2848 extern void print_rt_stats(struct seq_file *m, int cpu);
2849 extern void print_dl_stats(struct seq_file *m, int cpu);
2850 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2851 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2852 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2853 
2854 extern void resched_latency_warn(int cpu, u64 latency);
2855 #ifdef CONFIG_NUMA_BALANCING
2856 extern void
2857 show_numa_stats(struct task_struct *p, struct seq_file *m);
2858 extern void
2859 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2860 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2861 #endif /* CONFIG_NUMA_BALANCING */
2862 #else
resched_latency_warn(int cpu,u64 latency)2863 static inline void resched_latency_warn(int cpu, u64 latency) {}
2864 #endif /* CONFIG_SCHED_DEBUG */
2865 
2866 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2867 extern void init_rt_rq(struct rt_rq *rt_rq);
2868 extern void init_dl_rq(struct dl_rq *dl_rq);
2869 
2870 extern void cfs_bandwidth_usage_inc(void);
2871 extern void cfs_bandwidth_usage_dec(void);
2872 
2873 #ifdef CONFIG_NO_HZ_COMMON
2874 #define NOHZ_BALANCE_KICK_BIT	0
2875 #define NOHZ_STATS_KICK_BIT	1
2876 #define NOHZ_NEWILB_KICK_BIT	2
2877 #define NOHZ_NEXT_KICK_BIT	3
2878 
2879 /* Run rebalance_domains() */
2880 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2881 /* Update blocked load */
2882 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2883 /* Update blocked load when entering idle */
2884 #define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2885 /* Update nohz.next_balance */
2886 #define NOHZ_NEXT_KICK		BIT(NOHZ_NEXT_KICK_BIT)
2887 
2888 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2889 
2890 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2891 
2892 extern void nohz_balance_exit_idle(struct rq *rq);
2893 #else
nohz_balance_exit_idle(struct rq * rq)2894 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2895 #endif
2896 
2897 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2898 extern void nohz_run_idle_balance(int cpu);
2899 #else
nohz_run_idle_balance(int cpu)2900 static inline void nohz_run_idle_balance(int cpu) { }
2901 #endif
2902 
2903 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2904 struct irqtime {
2905 	u64			total;
2906 	u64			tick_delta;
2907 	u64			irq_start_time;
2908 	struct u64_stats_sync	sync;
2909 };
2910 
2911 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2912 
2913 /*
2914  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2915  * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2916  * and never move forward.
2917  */
irq_time_read(int cpu)2918 static inline u64 irq_time_read(int cpu)
2919 {
2920 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2921 	unsigned int seq;
2922 	u64 total;
2923 
2924 	do {
2925 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2926 		total = irqtime->total;
2927 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2928 
2929 	return total;
2930 }
2931 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2932 
2933 #ifdef CONFIG_CPU_FREQ
2934 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2935 
2936 /**
2937  * cpufreq_update_util - Take a note about CPU utilization changes.
2938  * @rq: Runqueue to carry out the update for.
2939  * @flags: Update reason flags.
2940  *
2941  * This function is called by the scheduler on the CPU whose utilization is
2942  * being updated.
2943  *
2944  * It can only be called from RCU-sched read-side critical sections.
2945  *
2946  * The way cpufreq is currently arranged requires it to evaluate the CPU
2947  * performance state (frequency/voltage) on a regular basis to prevent it from
2948  * being stuck in a completely inadequate performance level for too long.
2949  * That is not guaranteed to happen if the updates are only triggered from CFS
2950  * and DL, though, because they may not be coming in if only RT tasks are
2951  * active all the time (or there are RT tasks only).
2952  *
2953  * As a workaround for that issue, this function is called periodically by the
2954  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2955  * but that really is a band-aid.  Going forward it should be replaced with
2956  * solutions targeted more specifically at RT tasks.
2957  */
cpufreq_update_util(struct rq * rq,unsigned int flags)2958 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2959 {
2960 	struct update_util_data *data;
2961 
2962 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2963 						  cpu_of(rq)));
2964 	if (data)
2965 		data->func(data, rq_clock(rq), flags);
2966 }
2967 #else
cpufreq_update_util(struct rq * rq,unsigned int flags)2968 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2969 #endif /* CONFIG_CPU_FREQ */
2970 
2971 #ifdef arch_scale_freq_capacity
2972 # ifndef arch_scale_freq_invariant
2973 #  define arch_scale_freq_invariant()	true
2974 # endif
2975 #else
2976 # define arch_scale_freq_invariant()	false
2977 #endif
2978 
2979 #ifdef CONFIG_SMP
capacity_orig_of(int cpu)2980 static inline unsigned long capacity_orig_of(int cpu)
2981 {
2982 	return cpu_rq(cpu)->cpu_capacity_orig;
2983 }
2984 
2985 /**
2986  * enum cpu_util_type - CPU utilization type
2987  * @FREQUENCY_UTIL:	Utilization used to select frequency
2988  * @ENERGY_UTIL:	Utilization used during energy calculation
2989  *
2990  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2991  * need to be aggregated differently depending on the usage made of them. This
2992  * enum is used within effective_cpu_util() to differentiate the types of
2993  * utilization expected by the callers, and adjust the aggregation accordingly.
2994  */
2995 enum cpu_util_type {
2996 	FREQUENCY_UTIL,
2997 	ENERGY_UTIL,
2998 };
2999 
3000 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3001 				 enum cpu_util_type type,
3002 				 struct task_struct *p);
3003 
3004 /*
3005  * Verify the fitness of task @p to run on @cpu taking into account the
3006  * CPU original capacity and the runtime/deadline ratio of the task.
3007  *
3008  * The function will return true if the original capacity of @cpu is
3009  * greater than or equal to task's deadline density right shifted by
3010  * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3011  */
dl_task_fits_capacity(struct task_struct * p,int cpu)3012 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3013 {
3014 	unsigned long cap = arch_scale_cpu_capacity(cpu);
3015 
3016 	return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3017 }
3018 
cpu_bw_dl(struct rq * rq)3019 static inline unsigned long cpu_bw_dl(struct rq *rq)
3020 {
3021 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3022 }
3023 
cpu_util_dl(struct rq * rq)3024 static inline unsigned long cpu_util_dl(struct rq *rq)
3025 {
3026 	return READ_ONCE(rq->avg_dl.util_avg);
3027 }
3028 
3029 
3030 extern unsigned long cpu_util_cfs(int cpu);
3031 extern unsigned long cpu_util_cfs_boost(int cpu);
3032 
cpu_util_rt(struct rq * rq)3033 static inline unsigned long cpu_util_rt(struct rq *rq)
3034 {
3035 	return READ_ONCE(rq->avg_rt.util_avg);
3036 }
3037 #endif
3038 
3039 #ifdef CONFIG_UCLAMP_TASK
3040 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3041 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3042 static inline unsigned long uclamp_rq_get(struct rq *rq,
3043 					  enum uclamp_id clamp_id)
3044 {
3045 	return READ_ONCE(rq->uclamp[clamp_id].value);
3046 }
3047 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3048 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3049 				 unsigned int value)
3050 {
3051 	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3052 }
3053 
uclamp_rq_is_idle(struct rq * rq)3054 static inline bool uclamp_rq_is_idle(struct rq *rq)
3055 {
3056 	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3057 }
3058 
3059 /**
3060  * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3061  * @rq:		The rq to clamp against. Must not be NULL.
3062  * @util:	The util value to clamp.
3063  * @p:		The task to clamp against. Can be NULL if you want to clamp
3064  *		against @rq only.
3065  *
3066  * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3067  *
3068  * If sched_uclamp_used static key is disabled, then just return the util
3069  * without any clamping since uclamp aggregation at the rq level in the fast
3070  * path is disabled, rendering this operation a NOP.
3071  *
3072  * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3073  * will return the correct effective uclamp value of the task even if the
3074  * static key is disabled.
3075  */
3076 static __always_inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)3077 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3078 				  struct task_struct *p)
3079 {
3080 	unsigned long min_util = 0;
3081 	unsigned long max_util = 0;
3082 
3083 	if (!static_branch_likely(&sched_uclamp_used))
3084 		return util;
3085 
3086 	if (p) {
3087 		min_util = uclamp_eff_value(p, UCLAMP_MIN);
3088 		max_util = uclamp_eff_value(p, UCLAMP_MAX);
3089 
3090 		/*
3091 		 * Ignore last runnable task's max clamp, as this task will
3092 		 * reset it. Similarly, no need to read the rq's min clamp.
3093 		 */
3094 		if (uclamp_rq_is_idle(rq))
3095 			goto out;
3096 	}
3097 
3098 	min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3099 	max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3100 out:
3101 	/*
3102 	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3103 	 * RUNNABLE tasks with _different_ clamps, we can end up with an
3104 	 * inversion. Fix it now when the clamps are applied.
3105 	 */
3106 	if (unlikely(min_util >= max_util))
3107 		return min_util;
3108 
3109 	return clamp(util, min_util, max_util);
3110 }
3111 
3112 /* Is the rq being capped/throttled by uclamp_max? */
uclamp_rq_is_capped(struct rq * rq)3113 static inline bool uclamp_rq_is_capped(struct rq *rq)
3114 {
3115 	unsigned long rq_util;
3116 	unsigned long max_util;
3117 
3118 	if (!static_branch_likely(&sched_uclamp_used))
3119 		return false;
3120 
3121 	rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3122 	max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3123 
3124 	return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3125 }
3126 
3127 /*
3128  * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3129  * by default in the fast path and only gets turned on once userspace performs
3130  * an operation that requires it.
3131  *
3132  * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3133  * hence is active.
3134  */
uclamp_is_used(void)3135 static inline bool uclamp_is_used(void)
3136 {
3137 	return static_branch_likely(&sched_uclamp_used);
3138 }
3139 #else /* CONFIG_UCLAMP_TASK */
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)3140 static inline unsigned long uclamp_eff_value(struct task_struct *p,
3141 					     enum uclamp_id clamp_id)
3142 {
3143 	if (clamp_id == UCLAMP_MIN)
3144 		return 0;
3145 
3146 	return SCHED_CAPACITY_SCALE;
3147 }
3148 
3149 static inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)3150 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3151 				  struct task_struct *p)
3152 {
3153 	return util;
3154 }
3155 
uclamp_rq_is_capped(struct rq * rq)3156 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3157 
uclamp_is_used(void)3158 static inline bool uclamp_is_used(void)
3159 {
3160 	return false;
3161 }
3162 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3163 static inline unsigned long uclamp_rq_get(struct rq *rq,
3164 					  enum uclamp_id clamp_id)
3165 {
3166 	if (clamp_id == UCLAMP_MIN)
3167 		return 0;
3168 
3169 	return SCHED_CAPACITY_SCALE;
3170 }
3171 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3172 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3173 				 unsigned int value)
3174 {
3175 }
3176 
uclamp_rq_is_idle(struct rq * rq)3177 static inline bool uclamp_rq_is_idle(struct rq *rq)
3178 {
3179 	return false;
3180 }
3181 #endif /* CONFIG_UCLAMP_TASK */
3182 
3183 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
cpu_util_irq(struct rq * rq)3184 static inline unsigned long cpu_util_irq(struct rq *rq)
3185 {
3186 	return rq->avg_irq.util_avg;
3187 }
3188 
3189 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3190 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3191 {
3192 	util *= (max - irq);
3193 	util /= max;
3194 
3195 	return util;
3196 
3197 }
3198 #else
cpu_util_irq(struct rq * rq)3199 static inline unsigned long cpu_util_irq(struct rq *rq)
3200 {
3201 	return 0;
3202 }
3203 
3204 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3205 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3206 {
3207 	return util;
3208 }
3209 #endif
3210 
3211 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3212 
3213 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3214 
3215 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3216 
sched_energy_enabled(void)3217 static inline bool sched_energy_enabled(void)
3218 {
3219 	return static_branch_unlikely(&sched_energy_present);
3220 }
3221 
3222 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3223 
3224 #define perf_domain_span(pd) NULL
sched_energy_enabled(void)3225 static inline bool sched_energy_enabled(void) { return false; }
3226 
3227 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3228 
3229 #ifdef CONFIG_MEMBARRIER
3230 /*
3231  * The scheduler provides memory barriers required by membarrier between:
3232  * - prior user-space memory accesses and store to rq->membarrier_state,
3233  * - store to rq->membarrier_state and following user-space memory accesses.
3234  * In the same way it provides those guarantees around store to rq->curr.
3235  */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3236 static inline void membarrier_switch_mm(struct rq *rq,
3237 					struct mm_struct *prev_mm,
3238 					struct mm_struct *next_mm)
3239 {
3240 	int membarrier_state;
3241 
3242 	if (prev_mm == next_mm)
3243 		return;
3244 
3245 	membarrier_state = atomic_read(&next_mm->membarrier_state);
3246 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3247 		return;
3248 
3249 	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3250 }
3251 #else
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3252 static inline void membarrier_switch_mm(struct rq *rq,
3253 					struct mm_struct *prev_mm,
3254 					struct mm_struct *next_mm)
3255 {
3256 }
3257 #endif
3258 
3259 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3260 static inline bool is_per_cpu_kthread(struct task_struct *p)
3261 {
3262 	if (!(p->flags & PF_KTHREAD))
3263 		return false;
3264 
3265 	if (p->nr_cpus_allowed != 1)
3266 		return false;
3267 
3268 	return true;
3269 }
3270 #endif
3271 
3272 extern void swake_up_all_locked(struct swait_queue_head *q);
3273 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3274 
3275 extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3276 
3277 #ifdef CONFIG_PREEMPT_DYNAMIC
3278 extern int preempt_dynamic_mode;
3279 extern int sched_dynamic_mode(const char *str);
3280 extern void sched_dynamic_update(int mode);
3281 #endif
3282 
update_current_exec_runtime(struct task_struct * curr,u64 now,u64 delta_exec)3283 static inline void update_current_exec_runtime(struct task_struct *curr,
3284 						u64 now, u64 delta_exec)
3285 {
3286 	curr->se.sum_exec_runtime += delta_exec;
3287 	account_group_exec_runtime(curr, delta_exec);
3288 
3289 	curr->se.exec_start = now;
3290 	cgroup_account_cputime(curr, delta_exec);
3291 }
3292 
3293 #ifdef CONFIG_SCHED_MM_CID
3294 
3295 #define SCHED_MM_CID_PERIOD_NS	(100ULL * 1000000)	/* 100ms */
3296 #define MM_CID_SCAN_DELAY	100			/* 100ms */
3297 
3298 extern raw_spinlock_t cid_lock;
3299 extern int use_cid_lock;
3300 
3301 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3302 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3303 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3304 extern void init_sched_mm_cid(struct task_struct *t);
3305 
__mm_cid_put(struct mm_struct * mm,int cid)3306 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3307 {
3308 	if (cid < 0)
3309 		return;
3310 	cpumask_clear_cpu(cid, mm_cidmask(mm));
3311 }
3312 
3313 /*
3314  * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3315  * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3316  * be held to transition to other states.
3317  *
3318  * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3319  * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3320  */
mm_cid_put_lazy(struct task_struct * t)3321 static inline void mm_cid_put_lazy(struct task_struct *t)
3322 {
3323 	struct mm_struct *mm = t->mm;
3324 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3325 	int cid;
3326 
3327 	lockdep_assert_irqs_disabled();
3328 	cid = __this_cpu_read(pcpu_cid->cid);
3329 	if (!mm_cid_is_lazy_put(cid) ||
3330 	    !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3331 		return;
3332 	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3333 }
3334 
mm_cid_pcpu_unset(struct mm_struct * mm)3335 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3336 {
3337 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3338 	int cid, res;
3339 
3340 	lockdep_assert_irqs_disabled();
3341 	cid = __this_cpu_read(pcpu_cid->cid);
3342 	for (;;) {
3343 		if (mm_cid_is_unset(cid))
3344 			return MM_CID_UNSET;
3345 		/*
3346 		 * Attempt transition from valid or lazy-put to unset.
3347 		 */
3348 		res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3349 		if (res == cid)
3350 			break;
3351 		cid = res;
3352 	}
3353 	return cid;
3354 }
3355 
mm_cid_put(struct mm_struct * mm)3356 static inline void mm_cid_put(struct mm_struct *mm)
3357 {
3358 	int cid;
3359 
3360 	lockdep_assert_irqs_disabled();
3361 	cid = mm_cid_pcpu_unset(mm);
3362 	if (cid == MM_CID_UNSET)
3363 		return;
3364 	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3365 }
3366 
__mm_cid_try_get(struct mm_struct * mm)3367 static inline int __mm_cid_try_get(struct mm_struct *mm)
3368 {
3369 	struct cpumask *cpumask;
3370 	int cid;
3371 
3372 	cpumask = mm_cidmask(mm);
3373 	/*
3374 	 * Retry finding first zero bit if the mask is temporarily
3375 	 * filled. This only happens during concurrent remote-clear
3376 	 * which owns a cid without holding a rq lock.
3377 	 */
3378 	for (;;) {
3379 		cid = cpumask_first_zero(cpumask);
3380 		if (cid < nr_cpu_ids)
3381 			break;
3382 		cpu_relax();
3383 	}
3384 	if (cpumask_test_and_set_cpu(cid, cpumask))
3385 		return -1;
3386 	return cid;
3387 }
3388 
3389 /*
3390  * Save a snapshot of the current runqueue time of this cpu
3391  * with the per-cpu cid value, allowing to estimate how recently it was used.
3392  */
mm_cid_snapshot_time(struct rq * rq,struct mm_struct * mm)3393 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3394 {
3395 	struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3396 
3397 	lockdep_assert_rq_held(rq);
3398 	WRITE_ONCE(pcpu_cid->time, rq->clock);
3399 }
3400 
__mm_cid_get(struct rq * rq,struct mm_struct * mm)3401 static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3402 {
3403 	int cid;
3404 
3405 	/*
3406 	 * All allocations (even those using the cid_lock) are lock-free. If
3407 	 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3408 	 * guarantee forward progress.
3409 	 */
3410 	if (!READ_ONCE(use_cid_lock)) {
3411 		cid = __mm_cid_try_get(mm);
3412 		if (cid >= 0)
3413 			goto end;
3414 		raw_spin_lock(&cid_lock);
3415 	} else {
3416 		raw_spin_lock(&cid_lock);
3417 		cid = __mm_cid_try_get(mm);
3418 		if (cid >= 0)
3419 			goto unlock;
3420 	}
3421 
3422 	/*
3423 	 * cid concurrently allocated. Retry while forcing following
3424 	 * allocations to use the cid_lock to ensure forward progress.
3425 	 */
3426 	WRITE_ONCE(use_cid_lock, 1);
3427 	/*
3428 	 * Set use_cid_lock before allocation. Only care about program order
3429 	 * because this is only required for forward progress.
3430 	 */
3431 	barrier();
3432 	/*
3433 	 * Retry until it succeeds. It is guaranteed to eventually succeed once
3434 	 * all newcoming allocations observe the use_cid_lock flag set.
3435 	 */
3436 	do {
3437 		cid = __mm_cid_try_get(mm);
3438 		cpu_relax();
3439 	} while (cid < 0);
3440 	/*
3441 	 * Allocate before clearing use_cid_lock. Only care about
3442 	 * program order because this is for forward progress.
3443 	 */
3444 	barrier();
3445 	WRITE_ONCE(use_cid_lock, 0);
3446 unlock:
3447 	raw_spin_unlock(&cid_lock);
3448 end:
3449 	mm_cid_snapshot_time(rq, mm);
3450 	return cid;
3451 }
3452 
mm_cid_get(struct rq * rq,struct mm_struct * mm)3453 static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3454 {
3455 	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3456 	struct cpumask *cpumask;
3457 	int cid;
3458 
3459 	lockdep_assert_rq_held(rq);
3460 	cpumask = mm_cidmask(mm);
3461 	cid = __this_cpu_read(pcpu_cid->cid);
3462 	if (mm_cid_is_valid(cid)) {
3463 		mm_cid_snapshot_time(rq, mm);
3464 		return cid;
3465 	}
3466 	if (mm_cid_is_lazy_put(cid)) {
3467 		if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3468 			__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3469 	}
3470 	cid = __mm_cid_get(rq, mm);
3471 	__this_cpu_write(pcpu_cid->cid, cid);
3472 	return cid;
3473 }
3474 
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3475 static inline void switch_mm_cid(struct rq *rq,
3476 				 struct task_struct *prev,
3477 				 struct task_struct *next)
3478 {
3479 	/*
3480 	 * Provide a memory barrier between rq->curr store and load of
3481 	 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3482 	 *
3483 	 * Should be adapted if context_switch() is modified.
3484 	 */
3485 	if (!next->mm) {                                // to kernel
3486 		/*
3487 		 * user -> kernel transition does not guarantee a barrier, but
3488 		 * we can use the fact that it performs an atomic operation in
3489 		 * mmgrab().
3490 		 */
3491 		if (prev->mm)                           // from user
3492 			smp_mb__after_mmgrab();
3493 		/*
3494 		 * kernel -> kernel transition does not change rq->curr->mm
3495 		 * state. It stays NULL.
3496 		 */
3497 	} else {                                        // to user
3498 		/*
3499 		 * kernel -> user transition does not provide a barrier
3500 		 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3501 		 * Provide it here.
3502 		 */
3503 		if (!prev->mm)                          // from kernel
3504 			smp_mb();
3505 		/*
3506 		 * user -> user transition guarantees a memory barrier through
3507 		 * switch_mm() when current->mm changes. If current->mm is
3508 		 * unchanged, no barrier is needed.
3509 		 */
3510 	}
3511 	if (prev->mm_cid_active) {
3512 		mm_cid_snapshot_time(rq, prev->mm);
3513 		mm_cid_put_lazy(prev);
3514 		prev->mm_cid = -1;
3515 	}
3516 	if (next->mm_cid_active)
3517 		next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3518 }
3519 
3520 #else
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3521 static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
sched_mm_cid_migrate_from(struct task_struct * t)3522 static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
sched_mm_cid_migrate_to(struct rq * dst_rq,struct task_struct * t)3523 static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
task_tick_mm_cid(struct rq * rq,struct task_struct * curr)3524 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
init_sched_mm_cid(struct task_struct * t)3525 static inline void init_sched_mm_cid(struct task_struct *t) { }
3526 #endif
3527 
3528 extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3529 extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3530 
3531 #endif /* _KERNEL_SCHED_SCHED_H */
3532