• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Scheduler internal types and methods:
4  */
5 #include <linux/sched.h>
6 
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
35 
36 #include <uapi/linux/sched/types.h>
37 
38 #include <linux/binfmts.h>
39 #include <linux/bitops.h>
40 #include <linux/blkdev.h>
41 #include <linux/compat.h>
42 #include <linux/context_tracking.h>
43 #include <linux/cpufreq.h>
44 #include <linux/cpuidle.h>
45 #include <linux/cpuset.h>
46 #include <linux/ctype.h>
47 #include <linux/debugfs.h>
48 #include <linux/delayacct.h>
49 #include <linux/energy_model.h>
50 #include <linux/init_task.h>
51 #include <linux/kprobes.h>
52 #include <linux/kthread.h>
53 #include <linux/membarrier.h>
54 #include <linux/migrate.h>
55 #include <linux/mmu_context.h>
56 #include <linux/nmi.h>
57 #include <linux/proc_fs.h>
58 #include <linux/prefetch.h>
59 #include <linux/profile.h>
60 #include <linux/psi.h>
61 #include <linux/ratelimit.h>
62 #include <linux/rcupdate_wait.h>
63 #include <linux/security.h>
64 #include <linux/stop_machine.h>
65 #include <linux/suspend.h>
66 #include <linux/swait.h>
67 #include <linux/syscalls.h>
68 #include <linux/task_work.h>
69 #include <linux/tsacct_kern.h>
70 #include <linux/android_vendor.h>
71 #include <linux/android_kabi.h>
72 
73 #include <asm/tlb.h>
74 
75 #ifdef CONFIG_PARAVIRT
76 # include <asm/paravirt.h>
77 #endif
78 
79 #include "cpupri.h"
80 #include "cpudeadline.h"
81 
82 #include <trace/events/sched.h>
83 
84 #ifdef CONFIG_SCHED_DEBUG
85 # define SCHED_WARN_ON(x)	WARN_ONCE(x, #x)
86 #else
87 # define SCHED_WARN_ON(x)	({ (void)(x), 0; })
88 #endif
89 
90 struct rq;
91 struct cpuidle_state;
92 
93 /* task_struct::on_rq states: */
94 #define TASK_ON_RQ_QUEUED	1
95 #define TASK_ON_RQ_MIGRATING	2
96 
97 extern __read_mostly int scheduler_running;
98 
99 extern unsigned long calc_load_update;
100 extern atomic_long_t calc_load_tasks;
101 
102 extern void calc_global_load_tick(struct rq *this_rq);
103 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
104 
105 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
106 /*
107  * Helpers for converting nanosecond timing to jiffy resolution
108  */
109 #define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
110 
111 /*
112  * Increase resolution of nice-level calculations for 64-bit architectures.
113  * The extra resolution improves shares distribution and load balancing of
114  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
115  * hierarchies, especially on larger systems. This is not a user-visible change
116  * and does not change the user-interface for setting shares/weights.
117  *
118  * We increase resolution only if we have enough bits to allow this increased
119  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
120  * are pretty high and the returns do not justify the increased costs.
121  *
122  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
123  * increase coverage and consistency always enable it on 64-bit platforms.
124  */
125 #ifdef CONFIG_64BIT
126 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
127 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
128 # define scale_load_down(w) \
129 ({ \
130 	unsigned long __w = (w); \
131 	if (__w) \
132 		__w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
133 	__w; \
134 })
135 #else
136 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
137 # define scale_load(w)		(w)
138 # define scale_load_down(w)	(w)
139 #endif
140 
141 /*
142  * Task weight (visible to users) and its load (invisible to users) have
143  * independent resolution, but they should be well calibrated. We use
144  * scale_load() and scale_load_down(w) to convert between them. The
145  * following must be true:
146  *
147  *  scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
148  *
149  */
150 #define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
151 
152 /*
153  * Single value that decides SCHED_DEADLINE internal math precision.
154  * 10 -> just above 1us
155  * 9  -> just above 0.5us
156  */
157 #define DL_SCALE		10
158 
159 /*
160  * Single value that denotes runtime == period, ie unlimited time.
161  */
162 #define RUNTIME_INF		((u64)~0ULL)
163 
idle_policy(int policy)164 static inline int idle_policy(int policy)
165 {
166 	return policy == SCHED_IDLE;
167 }
fair_policy(int policy)168 static inline int fair_policy(int policy)
169 {
170 	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
171 }
172 
rt_policy(int policy)173 static inline int rt_policy(int policy)
174 {
175 	return policy == SCHED_FIFO || policy == SCHED_RR;
176 }
177 
dl_policy(int policy)178 static inline int dl_policy(int policy)
179 {
180 	return policy == SCHED_DEADLINE;
181 }
valid_policy(int policy)182 static inline bool valid_policy(int policy)
183 {
184 	return idle_policy(policy) || fair_policy(policy) ||
185 		rt_policy(policy) || dl_policy(policy);
186 }
187 
task_has_idle_policy(struct task_struct * p)188 static inline int task_has_idle_policy(struct task_struct *p)
189 {
190 	return idle_policy(p->policy);
191 }
192 
task_has_rt_policy(struct task_struct * p)193 static inline int task_has_rt_policy(struct task_struct *p)
194 {
195 	return rt_policy(p->policy);
196 }
197 
task_has_dl_policy(struct task_struct * p)198 static inline int task_has_dl_policy(struct task_struct *p)
199 {
200 	return dl_policy(p->policy);
201 }
202 
203 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
204 
update_avg(u64 * avg,u64 sample)205 static inline void update_avg(u64 *avg, u64 sample)
206 {
207 	s64 diff = sample - *avg;
208 	*avg += diff / 8;
209 }
210 
211 /*
212  * Shifting a value by an exponent greater *or equal* to the size of said value
213  * is UB; cap at size-1.
214  */
215 #define shr_bound(val, shift)							\
216 	(val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
217 
218 /*
219  * !! For sched_setattr_nocheck() (kernel) only !!
220  *
221  * This is actually gross. :(
222  *
223  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
224  * tasks, but still be able to sleep. We need this on platforms that cannot
225  * atomically change clock frequency. Remove once fast switching will be
226  * available on such platforms.
227  *
228  * SUGOV stands for SchedUtil GOVernor.
229  */
230 #define SCHED_FLAG_SUGOV	0x10000000
231 
232 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
233 
dl_entity_is_special(struct sched_dl_entity * dl_se)234 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
235 {
236 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
237 	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
238 #else
239 	return false;
240 #endif
241 }
242 
243 /*
244  * Tells if entity @a should preempt entity @b.
245  */
246 static inline bool
dl_entity_preempt(struct sched_dl_entity * a,struct sched_dl_entity * b)247 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
248 {
249 	return dl_entity_is_special(a) ||
250 	       dl_time_before(a->deadline, b->deadline);
251 }
252 
253 /*
254  * This is the priority-queue data structure of the RT scheduling class:
255  */
256 struct rt_prio_array {
257 	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
258 	struct list_head queue[MAX_RT_PRIO];
259 };
260 
261 struct rt_bandwidth {
262 	/* nests inside the rq lock: */
263 	raw_spinlock_t		rt_runtime_lock;
264 	ktime_t			rt_period;
265 	u64			rt_runtime;
266 	struct hrtimer		rt_period_timer;
267 	unsigned int		rt_period_active;
268 };
269 
270 void __dl_clear_params(struct task_struct *p);
271 
272 struct dl_bandwidth {
273 	raw_spinlock_t		dl_runtime_lock;
274 	u64			dl_runtime;
275 	u64			dl_period;
276 };
277 
dl_bandwidth_enabled(void)278 static inline int dl_bandwidth_enabled(void)
279 {
280 	return sysctl_sched_rt_runtime >= 0;
281 }
282 
283 /*
284  * To keep the bandwidth of -deadline tasks under control
285  * we need some place where:
286  *  - store the maximum -deadline bandwidth of each cpu;
287  *  - cache the fraction of bandwidth that is currently allocated in
288  *    each root domain;
289  *
290  * This is all done in the data structure below. It is similar to the
291  * one used for RT-throttling (rt_bandwidth), with the main difference
292  * that, since here we are only interested in admission control, we
293  * do not decrease any runtime while the group "executes", neither we
294  * need a timer to replenish it.
295  *
296  * With respect to SMP, bandwidth is given on a per root domain basis,
297  * meaning that:
298  *  - bw (< 100%) is the deadline bandwidth of each CPU;
299  *  - total_bw is the currently allocated bandwidth in each root domain;
300  */
301 struct dl_bw {
302 	raw_spinlock_t		lock;
303 	u64			bw;
304 	u64			total_bw;
305 };
306 
307 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
308 
309 static inline
__dl_sub(struct dl_bw * dl_b,u64 tsk_bw,int cpus)310 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
311 {
312 	dl_b->total_bw -= tsk_bw;
313 	__dl_update(dl_b, (s32)tsk_bw / cpus);
314 }
315 
316 static inline
__dl_add(struct dl_bw * dl_b,u64 tsk_bw,int cpus)317 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
318 {
319 	dl_b->total_bw += tsk_bw;
320 	__dl_update(dl_b, -((s32)tsk_bw / cpus));
321 }
322 
__dl_overflow(struct dl_bw * dl_b,unsigned long cap,u64 old_bw,u64 new_bw)323 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
324 				 u64 old_bw, u64 new_bw)
325 {
326 	return dl_b->bw != -1 &&
327 	       cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
328 }
329 
330 /*
331  * Verify the fitness of task @p to run on @cpu taking into account the
332  * CPU original capacity and the runtime/deadline ratio of the task.
333  *
334  * The function will return true if the CPU original capacity of the
335  * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
336  * task and false otherwise.
337  */
dl_task_fits_capacity(struct task_struct * p,int cpu)338 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
339 {
340 	unsigned long cap = arch_scale_cpu_capacity(cpu);
341 
342 	return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
343 }
344 
345 extern void init_dl_bw(struct dl_bw *dl_b);
346 extern int  sched_dl_global_validate(void);
347 extern void sched_dl_do_global(void);
348 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
349 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
350 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
351 extern bool __checkparam_dl(const struct sched_attr *attr);
352 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
353 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
354 extern int  dl_bw_check_overflow(int cpu);
355 
356 #ifdef CONFIG_CGROUP_SCHED
357 
358 #include <linux/cgroup.h>
359 #include <linux/psi.h>
360 
361 struct cfs_rq;
362 struct rt_rq;
363 
364 extern struct list_head task_groups;
365 
366 struct cfs_bandwidth {
367 #ifdef CONFIG_CFS_BANDWIDTH
368 	raw_spinlock_t		lock;
369 	ktime_t			period;
370 	u64			quota;
371 	u64			runtime;
372 	u64			burst;
373 	s64			hierarchical_quota;
374 
375 	u8			idle;
376 	u8			period_active;
377 	u8			slack_started;
378 	struct hrtimer		period_timer;
379 	struct hrtimer		slack_timer;
380 	struct list_head	throttled_cfs_rq;
381 
382 	/* Statistics: */
383 	int			nr_periods;
384 	int			nr_throttled;
385 	u64			throttled_time;
386 #endif
387 };
388 
389 /* Task group related information */
390 struct task_group {
391 	struct cgroup_subsys_state css;
392 
393 #ifdef CONFIG_FAIR_GROUP_SCHED
394 	/* schedulable entities of this group on each CPU */
395 	struct sched_entity	**se;
396 	/* runqueue "owned" by this group on each CPU */
397 	struct cfs_rq		**cfs_rq;
398 	unsigned long		shares;
399 
400 	/* A positive value indicates that this is a SCHED_IDLE group. */
401 	int			idle;
402 
403 #ifdef	CONFIG_SMP
404 	/*
405 	 * load_avg can be heavily contended at clock tick time, so put
406 	 * it in its own cacheline separated from the fields above which
407 	 * will also be accessed at each tick.
408 	 */
409 	atomic_long_t		load_avg ____cacheline_aligned;
410 #endif
411 #endif
412 
413 #ifdef CONFIG_RT_GROUP_SCHED
414 	struct sched_rt_entity	**rt_se;
415 	struct rt_rq		**rt_rq;
416 
417 	struct rt_bandwidth	rt_bandwidth;
418 #endif
419 
420 	struct rcu_head		rcu;
421 	struct list_head	list;
422 
423 	struct task_group	*parent;
424 	struct list_head	siblings;
425 	struct list_head	children;
426 
427 #ifdef CONFIG_SCHED_AUTOGROUP
428 	struct autogroup	*autogroup;
429 #endif
430 
431 	struct cfs_bandwidth	cfs_bandwidth;
432 
433 #ifdef CONFIG_UCLAMP_TASK_GROUP
434 	/* The two decimal precision [%] value requested from user-space */
435 	unsigned int		uclamp_pct[UCLAMP_CNT];
436 	/* Clamp values requested for a task group */
437 	struct uclamp_se	uclamp_req[UCLAMP_CNT];
438 	/* Effective clamp values used for a task group */
439 	struct uclamp_se	uclamp[UCLAMP_CNT];
440 	/* Latency-sensitive flag used for a task group */
441 	unsigned int		latency_sensitive;
442 
443 	ANDROID_VENDOR_DATA_ARRAY(1, 4);
444 #endif
445 
446 	ANDROID_KABI_RESERVE(1);
447 	ANDROID_KABI_RESERVE(2);
448 	ANDROID_KABI_RESERVE(3);
449 	ANDROID_KABI_RESERVE(4);
450 };
451 
452 #ifdef CONFIG_FAIR_GROUP_SCHED
453 #define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
454 
455 /*
456  * A weight of 0 or 1 can cause arithmetics problems.
457  * A weight of a cfs_rq is the sum of weights of which entities
458  * are queued on this cfs_rq, so a weight of a entity should not be
459  * too large, so as the shares value of a task group.
460  * (The default weight is 1024 - so there's no practical
461  *  limitation from this.)
462  */
463 #define MIN_SHARES		(1UL <<  1)
464 #define MAX_SHARES		(1UL << 18)
465 #endif
466 
467 typedef int (*tg_visitor)(struct task_group *, void *);
468 
469 extern int walk_tg_tree_from(struct task_group *from,
470 			     tg_visitor down, tg_visitor up, void *data);
471 
472 /*
473  * Iterate the full tree, calling @down when first entering a node and @up when
474  * leaving it for the final time.
475  *
476  * Caller must hold rcu_lock or sufficient equivalent.
477  */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)478 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
479 {
480 	return walk_tg_tree_from(&root_task_group, down, up, data);
481 }
482 
483 extern int tg_nop(struct task_group *tg, void *data);
484 
485 extern void free_fair_sched_group(struct task_group *tg);
486 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
487 extern void online_fair_sched_group(struct task_group *tg);
488 extern void unregister_fair_sched_group(struct task_group *tg);
489 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
490 			struct sched_entity *se, int cpu,
491 			struct sched_entity *parent);
492 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
493 
494 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
495 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
496 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
497 
498 extern void unregister_rt_sched_group(struct task_group *tg);
499 extern void free_rt_sched_group(struct task_group *tg);
500 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
501 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
502 		struct sched_rt_entity *rt_se, int cpu,
503 		struct sched_rt_entity *parent);
504 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
505 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
506 extern long sched_group_rt_runtime(struct task_group *tg);
507 extern long sched_group_rt_period(struct task_group *tg);
508 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
509 
510 extern struct task_group *sched_create_group(struct task_group *parent);
511 extern void sched_online_group(struct task_group *tg,
512 			       struct task_group *parent);
513 extern void sched_destroy_group(struct task_group *tg);
514 extern void sched_release_group(struct task_group *tg);
515 
516 extern void sched_move_task(struct task_struct *tsk);
517 
518 #ifdef CONFIG_FAIR_GROUP_SCHED
519 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
520 
521 extern int sched_group_set_idle(struct task_group *tg, long idle);
522 
523 #ifdef CONFIG_SMP
524 extern void set_task_rq_fair(struct sched_entity *se,
525 			     struct cfs_rq *prev, struct cfs_rq *next);
526 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)527 static inline void set_task_rq_fair(struct sched_entity *se,
528 			     struct cfs_rq *prev, struct cfs_rq *next) { }
529 #endif /* CONFIG_SMP */
530 #endif /* CONFIG_FAIR_GROUP_SCHED */
531 
532 #else /* CONFIG_CGROUP_SCHED */
533 
534 struct cfs_bandwidth { };
535 
536 #endif	/* CONFIG_CGROUP_SCHED */
537 
538 /* CFS-related fields in a runqueue */
539 struct cfs_rq {
540 	struct load_weight	load;
541 	unsigned int		nr_running;
542 	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
543 	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
544 
545 	u64			exec_clock;
546 	u64			min_vruntime;
547 #ifdef CONFIG_SCHED_CORE
548 	unsigned int		forceidle_seq;
549 	u64			min_vruntime_fi;
550 #endif
551 
552 #ifndef CONFIG_64BIT
553 	u64			min_vruntime_copy;
554 #endif
555 
556 	struct rb_root_cached	tasks_timeline;
557 
558 	/*
559 	 * 'curr' points to currently running entity on this cfs_rq.
560 	 * It is set to NULL otherwise (i.e when none are currently running).
561 	 */
562 	struct sched_entity	*curr;
563 	struct sched_entity	*next;
564 	struct sched_entity	*last;
565 	struct sched_entity	*skip;
566 
567 #ifdef	CONFIG_SCHED_DEBUG
568 	unsigned int		nr_spread_over;
569 #endif
570 
571 #ifdef CONFIG_SMP
572 	/*
573 	 * CFS load tracking
574 	 */
575 	struct sched_avg	avg;
576 #ifndef CONFIG_64BIT
577 	u64			load_last_update_time_copy;
578 #endif
579 	struct {
580 		raw_spinlock_t	lock ____cacheline_aligned;
581 		int		nr;
582 		unsigned long	load_avg;
583 		unsigned long	util_avg;
584 		unsigned long	runnable_avg;
585 	} removed;
586 
587 #ifdef CONFIG_FAIR_GROUP_SCHED
588 	unsigned long		tg_load_avg_contrib;
589 	long			propagate;
590 	long			prop_runnable_sum;
591 
592 	/*
593 	 *   h_load = weight * f(tg)
594 	 *
595 	 * Where f(tg) is the recursive weight fraction assigned to
596 	 * this group.
597 	 */
598 	unsigned long		h_load;
599 	u64			last_h_load_update;
600 	struct sched_entity	*h_load_next;
601 #endif /* CONFIG_FAIR_GROUP_SCHED */
602 #endif /* CONFIG_SMP */
603 
604 #ifdef CONFIG_FAIR_GROUP_SCHED
605 	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
606 
607 	/*
608 	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
609 	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
610 	 * (like users, containers etc.)
611 	 *
612 	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
613 	 * This list is used during load balance.
614 	 */
615 	int			on_list;
616 	struct list_head	leaf_cfs_rq_list;
617 	struct task_group	*tg;	/* group that "owns" this runqueue */
618 
619 	/* Locally cached copy of our task_group's idle value */
620 	int			idle;
621 
622 #ifdef CONFIG_CFS_BANDWIDTH
623 	int			runtime_enabled;
624 	s64			runtime_remaining;
625 
626 	u64			throttled_clock;
627 	u64			throttled_clock_pelt;
628 	u64			throttled_clock_pelt_time;
629 	int			throttled;
630 	int			throttle_count;
631 	struct list_head	throttled_list;
632 #endif /* CONFIG_CFS_BANDWIDTH */
633 #endif /* CONFIG_FAIR_GROUP_SCHED */
634 };
635 
rt_bandwidth_enabled(void)636 static inline int rt_bandwidth_enabled(void)
637 {
638 	return sysctl_sched_rt_runtime >= 0;
639 }
640 
641 /* RT IPI pull logic requires IRQ_WORK */
642 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
643 # define HAVE_RT_PUSH_IPI
644 #endif
645 
646 /* Real-Time classes' related field in a runqueue: */
647 struct rt_rq {
648 	struct rt_prio_array	active;
649 	unsigned int		rt_nr_running;
650 	unsigned int		rr_nr_running;
651 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
652 	struct {
653 		int		curr; /* highest queued rt task prio */
654 #ifdef CONFIG_SMP
655 		int		next; /* next highest */
656 #endif
657 	} highest_prio;
658 #endif
659 #ifdef CONFIG_SMP
660 	unsigned int		rt_nr_migratory;
661 	unsigned int		rt_nr_total;
662 	int			overloaded;
663 	struct plist_head	pushable_tasks;
664 
665 #endif /* CONFIG_SMP */
666 	int			rt_queued;
667 
668 	int			rt_throttled;
669 	u64			rt_time;
670 	u64			rt_runtime;
671 	/* Nests inside the rq lock: */
672 	raw_spinlock_t		rt_runtime_lock;
673 
674 #ifdef CONFIG_RT_GROUP_SCHED
675 	unsigned int		rt_nr_boosted;
676 
677 	struct rq		*rq;
678 	struct task_group	*tg;
679 #endif
680 };
681 
rt_rq_is_runnable(struct rt_rq * rt_rq)682 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
683 {
684 	return rt_rq->rt_queued && rt_rq->rt_nr_running;
685 }
686 
687 /* Deadline class' related fields in a runqueue */
688 struct dl_rq {
689 	/* runqueue is an rbtree, ordered by deadline */
690 	struct rb_root_cached	root;
691 
692 	unsigned int		dl_nr_running;
693 
694 #ifdef CONFIG_SMP
695 	/*
696 	 * Deadline values of the currently executing and the
697 	 * earliest ready task on this rq. Caching these facilitates
698 	 * the decision whether or not a ready but not running task
699 	 * should migrate somewhere else.
700 	 */
701 	struct {
702 		u64		curr;
703 		u64		next;
704 	} earliest_dl;
705 
706 	unsigned int		dl_nr_migratory;
707 	int			overloaded;
708 
709 	/*
710 	 * Tasks on this rq that can be pushed away. They are kept in
711 	 * an rb-tree, ordered by tasks' deadlines, with caching
712 	 * of the leftmost (earliest deadline) element.
713 	 */
714 	struct rb_root_cached	pushable_dl_tasks_root;
715 #else
716 	struct dl_bw		dl_bw;
717 #endif
718 	/*
719 	 * "Active utilization" for this runqueue: increased when a
720 	 * task wakes up (becomes TASK_RUNNING) and decreased when a
721 	 * task blocks
722 	 */
723 	u64			running_bw;
724 
725 	/*
726 	 * Utilization of the tasks "assigned" to this runqueue (including
727 	 * the tasks that are in runqueue and the tasks that executed on this
728 	 * CPU and blocked). Increased when a task moves to this runqueue, and
729 	 * decreased when the task moves away (migrates, changes scheduling
730 	 * policy, or terminates).
731 	 * This is needed to compute the "inactive utilization" for the
732 	 * runqueue (inactive utilization = this_bw - running_bw).
733 	 */
734 	u64			this_bw;
735 	u64			extra_bw;
736 
737 	/*
738 	 * Inverse of the fraction of CPU utilization that can be reclaimed
739 	 * by the GRUB algorithm.
740 	 */
741 	u64			bw_ratio;
742 };
743 
744 #ifdef CONFIG_FAIR_GROUP_SCHED
745 /* An entity is a task if it doesn't "own" a runqueue */
746 #define entity_is_task(se)	(!se->my_q)
747 
se_update_runnable(struct sched_entity * se)748 static inline void se_update_runnable(struct sched_entity *se)
749 {
750 	if (!entity_is_task(se))
751 		se->runnable_weight = se->my_q->h_nr_running;
752 }
753 
se_runnable(struct sched_entity * se)754 static inline long se_runnable(struct sched_entity *se)
755 {
756 	if (entity_is_task(se))
757 		return !!se->on_rq;
758 	else
759 		return se->runnable_weight;
760 }
761 
762 #else
763 #define entity_is_task(se)	1
764 
se_update_runnable(struct sched_entity * se)765 static inline void se_update_runnable(struct sched_entity *se) {}
766 
se_runnable(struct sched_entity * se)767 static inline long se_runnable(struct sched_entity *se)
768 {
769 	return !!se->on_rq;
770 }
771 #endif
772 
773 #ifdef CONFIG_SMP
774 /*
775  * XXX we want to get rid of these helpers and use the full load resolution.
776  */
se_weight(struct sched_entity * se)777 static inline long se_weight(struct sched_entity *se)
778 {
779 	return scale_load_down(se->load.weight);
780 }
781 
782 
sched_asym_prefer(int a,int b)783 static inline bool sched_asym_prefer(int a, int b)
784 {
785 	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
786 }
787 
788 struct perf_domain {
789 	struct em_perf_domain *em_pd;
790 	struct perf_domain *next;
791 	struct rcu_head rcu;
792 };
793 
794 /* Scheduling group status flags */
795 #define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
796 #define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
797 
798 /*
799  * We add the notion of a root-domain which will be used to define per-domain
800  * variables. Each exclusive cpuset essentially defines an island domain by
801  * fully partitioning the member CPUs from any other cpuset. Whenever a new
802  * exclusive cpuset is created, we also create and attach a new root-domain
803  * object.
804  *
805  */
806 struct root_domain {
807 	atomic_t		refcount;
808 	atomic_t		rto_count;
809 	struct rcu_head		rcu;
810 	cpumask_var_t		span;
811 	cpumask_var_t		online;
812 
813 	/*
814 	 * Indicate pullable load on at least one CPU, e.g:
815 	 * - More than one runnable task
816 	 * - Running task is misfit
817 	 */
818 	int			overload;
819 
820 	/* Indicate one or more cpus over-utilized (tipping point) */
821 	int			overutilized;
822 
823 	/*
824 	 * The bit corresponding to a CPU gets set here if such CPU has more
825 	 * than one runnable -deadline task (as it is below for RT tasks).
826 	 */
827 	cpumask_var_t		dlo_mask;
828 	atomic_t		dlo_count;
829 	struct dl_bw		dl_bw;
830 	struct cpudl		cpudl;
831 
832 	/*
833 	 * Indicate whether a root_domain's dl_bw has been checked or
834 	 * updated. It's monotonously increasing value.
835 	 *
836 	 * Also, some corner cases, like 'wrap around' is dangerous, but given
837 	 * that u64 is 'big enough'. So that shouldn't be a concern.
838 	 */
839 	u64 visit_gen;
840 
841 #ifdef HAVE_RT_PUSH_IPI
842 	/*
843 	 * For IPI pull requests, loop across the rto_mask.
844 	 */
845 	struct irq_work		rto_push_work;
846 	raw_spinlock_t		rto_lock;
847 	/* These are only updated and read within rto_lock */
848 	int			rto_loop;
849 	int			rto_cpu;
850 	/* These atomics are updated outside of a lock */
851 	atomic_t		rto_loop_next;
852 	atomic_t		rto_loop_start;
853 #endif
854 	/*
855 	 * The "RT overload" flag: it gets set if a CPU has more than
856 	 * one runnable RT task.
857 	 */
858 	cpumask_var_t		rto_mask;
859 	struct cpupri		cpupri;
860 
861 	unsigned long		max_cpu_capacity;
862 
863 	/*
864 	 * NULL-terminated list of performance domains intersecting with the
865 	 * CPUs of the rd. Protected by RCU.
866 	 */
867 	struct perf_domain __rcu *pd;
868 
869 	ANDROID_VENDOR_DATA_ARRAY(1, 4);
870 
871 	ANDROID_KABI_RESERVE(1);
872 	ANDROID_KABI_RESERVE(2);
873 	ANDROID_KABI_RESERVE(3);
874 	ANDROID_KABI_RESERVE(4);
875 };
876 
877 extern void init_defrootdomain(void);
878 extern int sched_init_domains(const struct cpumask *cpu_map);
879 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
880 extern void sched_get_rd(struct root_domain *rd);
881 extern void sched_put_rd(struct root_domain *rd);
882 
883 #ifdef HAVE_RT_PUSH_IPI
884 extern void rto_push_irq_work_func(struct irq_work *work);
885 #endif
886 extern struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu);
887 #endif /* CONFIG_SMP */
888 
889 #ifdef CONFIG_UCLAMP_TASK
890 /*
891  * struct uclamp_bucket - Utilization clamp bucket
892  * @value: utilization clamp value for tasks on this clamp bucket
893  * @tasks: number of RUNNABLE tasks on this clamp bucket
894  *
895  * Keep track of how many tasks are RUNNABLE for a given utilization
896  * clamp value.
897  */
898 struct uclamp_bucket {
899 	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
900 	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
901 };
902 
903 /*
904  * struct uclamp_rq - rq's utilization clamp
905  * @value: currently active clamp values for a rq
906  * @bucket: utilization clamp buckets affecting a rq
907  *
908  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
909  * A clamp value is affecting a rq when there is at least one task RUNNABLE
910  * (or actually running) with that value.
911  *
912  * There are up to UCLAMP_CNT possible different clamp values, currently there
913  * are only two: minimum utilization and maximum utilization.
914  *
915  * All utilization clamping values are MAX aggregated, since:
916  * - for util_min: we want to run the CPU at least at the max of the minimum
917  *   utilization required by its currently RUNNABLE tasks.
918  * - for util_max: we want to allow the CPU to run up to the max of the
919  *   maximum utilization allowed by its currently RUNNABLE tasks.
920  *
921  * Since on each system we expect only a limited number of different
922  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
923  * the metrics required to compute all the per-rq utilization clamp values.
924  */
925 struct uclamp_rq {
926 	unsigned int value;
927 	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
928 };
929 
930 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
931 #endif /* CONFIG_UCLAMP_TASK */
932 
933 /*
934  * This is the main, per-CPU runqueue data structure.
935  *
936  * Locking rule: those places that want to lock multiple runqueues
937  * (such as the load balancing or the thread migration code), lock
938  * acquire operations must be ordered by ascending &runqueue.
939  */
940 struct rq {
941 	/* runqueue lock: */
942 	raw_spinlock_t		__lock;
943 
944 	/*
945 	 * nr_running and cpu_load should be in the same cacheline because
946 	 * remote CPUs use both these fields when doing load calculation.
947 	 */
948 	unsigned int		nr_running;
949 #ifdef CONFIG_NUMA_BALANCING
950 	unsigned int		nr_numa_running;
951 	unsigned int		nr_preferred_running;
952 	unsigned int		numa_migrate_on;
953 #endif
954 #ifdef CONFIG_NO_HZ_COMMON
955 #ifdef CONFIG_SMP
956 	unsigned long		last_blocked_load_update_tick;
957 	unsigned int		has_blocked_load;
958 	call_single_data_t	nohz_csd;
959 #endif /* CONFIG_SMP */
960 	unsigned int		nohz_tick_stopped;
961 	atomic_t		nohz_flags;
962 #endif /* CONFIG_NO_HZ_COMMON */
963 
964 #ifdef CONFIG_SMP
965 	unsigned int		ttwu_pending;
966 #endif
967 	u64			nr_switches;
968 
969 #ifdef CONFIG_UCLAMP_TASK
970 	/* Utilization clamp values based on CPU's RUNNABLE tasks */
971 	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
972 	unsigned int		uclamp_flags;
973 #define UCLAMP_FLAG_IDLE 0x01
974 #endif
975 
976 	struct cfs_rq		cfs;
977 	struct rt_rq		rt;
978 	struct dl_rq		dl;
979 
980 #ifdef CONFIG_FAIR_GROUP_SCHED
981 	/* list of leaf cfs_rq on this CPU: */
982 	struct list_head	leaf_cfs_rq_list;
983 	struct list_head	*tmp_alone_branch;
984 #endif /* CONFIG_FAIR_GROUP_SCHED */
985 
986 	/*
987 	 * This is part of a global counter where only the total sum
988 	 * over all CPUs matters. A task can increase this counter on
989 	 * one CPU and if it got migrated afterwards it may decrease
990 	 * it on another CPU. Always updated under the runqueue lock:
991 	 */
992 	unsigned int		nr_uninterruptible;
993 
994 	struct task_struct __rcu	*curr;
995 	struct task_struct	*idle;
996 	struct task_struct	*stop;
997 	unsigned long		next_balance;
998 	struct mm_struct	*prev_mm;
999 
1000 	unsigned int		clock_update_flags;
1001 	u64			clock;
1002 	/* Ensure that all clocks are in the same cache line */
1003 	u64			clock_task ____cacheline_aligned;
1004 	u64			clock_task_mult;
1005 	u64			clock_pelt;
1006 	unsigned long		lost_idle_time;
1007 
1008 	atomic_t		nr_iowait;
1009 
1010 #ifdef CONFIG_SCHED_DEBUG
1011 	u64 last_seen_need_resched_ns;
1012 	int ticks_without_resched;
1013 #endif
1014 
1015 #ifdef CONFIG_MEMBARRIER
1016 	int membarrier_state;
1017 #endif
1018 
1019 #ifdef CONFIG_SMP
1020 	struct root_domain		*rd;
1021 	struct sched_domain __rcu	*sd;
1022 
1023 	unsigned long		cpu_capacity;
1024 	unsigned long		cpu_capacity_orig;
1025 
1026 	struct callback_head	*balance_callback;
1027 
1028 	unsigned char		nohz_idle_balance;
1029 	unsigned char		idle_balance;
1030 
1031 	unsigned long		misfit_task_load;
1032 
1033 	/* For active balancing */
1034 	int			active_balance;
1035 	int			push_cpu;
1036 	struct cpu_stop_work	active_balance_work;
1037 
1038 	/* CPU of this runqueue: */
1039 	int			cpu;
1040 	int			online;
1041 
1042 	struct list_head cfs_tasks;
1043 
1044 	struct sched_avg	avg_rt;
1045 	struct sched_avg	avg_dl;
1046 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1047 	struct sched_avg	avg_irq;
1048 #endif
1049 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1050 	struct sched_avg	avg_thermal;
1051 #endif
1052 	u64			idle_stamp;
1053 	u64			avg_idle;
1054 
1055 	unsigned long		wake_stamp;
1056 	u64			wake_avg_idle;
1057 
1058 	/* This is used to determine avg_idle's max value */
1059 	u64			max_idle_balance_cost;
1060 
1061 #ifdef CONFIG_HOTPLUG_CPU
1062 	struct rcuwait		hotplug_wait;
1063 #endif
1064 #endif /* CONFIG_SMP */
1065 
1066 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1067 	u64			prev_irq_time;
1068 #endif
1069 #ifdef CONFIG_PARAVIRT
1070 	u64			prev_steal_time;
1071 #endif
1072 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1073 	u64			prev_steal_time_rq;
1074 #endif
1075 
1076 	/* calc_load related fields */
1077 	unsigned long		calc_load_update;
1078 	long			calc_load_active;
1079 
1080 #ifdef CONFIG_SCHED_HRTICK
1081 #ifdef CONFIG_SMP
1082 	call_single_data_t	hrtick_csd;
1083 #endif
1084 	struct hrtimer		hrtick_timer;
1085 	ktime_t 		hrtick_time;
1086 #endif
1087 
1088 #ifdef CONFIG_SCHEDSTATS
1089 	/* latency stats */
1090 	struct sched_info	rq_sched_info;
1091 	unsigned long long	rq_cpu_time;
1092 	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1093 
1094 	/* sys_sched_yield() stats */
1095 	unsigned int		yld_count;
1096 
1097 	/* schedule() stats */
1098 	unsigned int		sched_count;
1099 	unsigned int		sched_goidle;
1100 
1101 	/* try_to_wake_up() stats */
1102 	unsigned int		ttwu_count;
1103 	unsigned int		ttwu_local;
1104 #endif
1105 
1106 #ifdef CONFIG_CPU_IDLE
1107 	/* Must be inspected within a rcu lock section */
1108 	struct cpuidle_state	*idle_state;
1109 #endif
1110 
1111 #ifdef CONFIG_SMP
1112 	unsigned int		nr_pinned;
1113 #endif
1114 	unsigned int		push_busy;
1115 	struct cpu_stop_work	push_work;
1116 
1117 #ifdef CONFIG_SCHED_CORE
1118 	/* per rq */
1119 	struct rq		*core;
1120 	struct task_struct	*core_pick;
1121 	unsigned int		core_enabled;
1122 	unsigned int		core_sched_seq;
1123 	struct rb_root		core_tree;
1124 
1125 	/* shared state -- careful with sched_core_cpu_deactivate() */
1126 	unsigned int		core_task_seq;
1127 	unsigned int		core_pick_seq;
1128 	unsigned long		core_cookie;
1129 	unsigned char		core_forceidle;
1130 	unsigned int		core_forceidle_seq;
1131 #endif
1132 
1133 	ANDROID_VENDOR_DATA_ARRAY(1, 96);
1134 	ANDROID_OEM_DATA_ARRAY(1, 16);
1135 
1136 	ANDROID_KABI_RESERVE(1);
1137 	ANDROID_KABI_RESERVE(2);
1138 	ANDROID_KABI_RESERVE(3);
1139 	ANDROID_KABI_RESERVE(4);
1140 };
1141 
1142 #ifdef CONFIG_FAIR_GROUP_SCHED
1143 
1144 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1145 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1146 {
1147 	return cfs_rq->rq;
1148 }
1149 
1150 #else
1151 
rq_of(struct cfs_rq * cfs_rq)1152 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1153 {
1154 	return container_of(cfs_rq, struct rq, cfs);
1155 }
1156 #endif
1157 
cpu_of(struct rq * rq)1158 static inline int cpu_of(struct rq *rq)
1159 {
1160 #ifdef CONFIG_SMP
1161 	return rq->cpu;
1162 #else
1163 	return 0;
1164 #endif
1165 }
1166 
1167 #define MDF_PUSH	0x01
1168 
is_migration_disabled(struct task_struct * p)1169 static inline bool is_migration_disabled(struct task_struct *p)
1170 {
1171 #ifdef CONFIG_SMP
1172 	return p->migration_disabled;
1173 #else
1174 	return false;
1175 #endif
1176 }
1177 
1178 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1179 
1180 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1181 #define this_rq()		this_cpu_ptr(&runqueues)
1182 #define task_rq(p)		cpu_rq(task_cpu(p))
1183 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1184 #define raw_rq()		raw_cpu_ptr(&runqueues)
1185 
1186 struct sched_group;
1187 #ifdef CONFIG_SCHED_CORE
1188 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1189 
1190 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1191 
sched_core_enabled(struct rq * rq)1192 static inline bool sched_core_enabled(struct rq *rq)
1193 {
1194 	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1195 }
1196 
sched_core_disabled(void)1197 static inline bool sched_core_disabled(void)
1198 {
1199 	return !static_branch_unlikely(&__sched_core_enabled);
1200 }
1201 
1202 /*
1203  * Be careful with this function; not for general use. The return value isn't
1204  * stable unless you actually hold a relevant rq->__lock.
1205  */
rq_lockp(struct rq * rq)1206 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1207 {
1208 	if (sched_core_enabled(rq))
1209 		return &rq->core->__lock;
1210 
1211 	return &rq->__lock;
1212 }
1213 
__rq_lockp(struct rq * rq)1214 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1215 {
1216 	if (rq->core_enabled)
1217 		return &rq->core->__lock;
1218 
1219 	return &rq->__lock;
1220 }
1221 
1222 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1223 
1224 /*
1225  * Helpers to check if the CPU's core cookie matches with the task's cookie
1226  * when core scheduling is enabled.
1227  * A special case is that the task's cookie always matches with CPU's core
1228  * cookie if the CPU is in an idle core.
1229  */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1230 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1231 {
1232 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1233 	if (!sched_core_enabled(rq))
1234 		return true;
1235 
1236 	return rq->core->core_cookie == p->core_cookie;
1237 }
1238 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1239 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1240 {
1241 	bool idle_core = true;
1242 	int cpu;
1243 
1244 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1245 	if (!sched_core_enabled(rq))
1246 		return true;
1247 
1248 	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1249 		if (!available_idle_cpu(cpu)) {
1250 			idle_core = false;
1251 			break;
1252 		}
1253 	}
1254 
1255 	/*
1256 	 * A CPU in an idle core is always the best choice for tasks with
1257 	 * cookies.
1258 	 */
1259 	return idle_core || rq->core->core_cookie == p->core_cookie;
1260 }
1261 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1262 static inline bool sched_group_cookie_match(struct rq *rq,
1263 					    struct task_struct *p,
1264 					    struct sched_group *group)
1265 {
1266 	int cpu;
1267 
1268 	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1269 	if (!sched_core_enabled(rq))
1270 		return true;
1271 
1272 	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1273 		if (sched_core_cookie_match(cpu_rq(cpu), p))
1274 			return true;
1275 	}
1276 	return false;
1277 }
1278 
1279 extern void queue_core_balance(struct rq *rq);
1280 
sched_core_enqueued(struct task_struct * p)1281 static inline bool sched_core_enqueued(struct task_struct *p)
1282 {
1283 	return !RB_EMPTY_NODE(&p->core_node);
1284 }
1285 
1286 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1287 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1288 
1289 extern void sched_core_get(void);
1290 extern void sched_core_put(void);
1291 
1292 extern unsigned long sched_core_alloc_cookie(void);
1293 extern void sched_core_put_cookie(unsigned long cookie);
1294 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1295 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1296 
1297 #else /* !CONFIG_SCHED_CORE */
1298 
sched_core_enabled(struct rq * rq)1299 static inline bool sched_core_enabled(struct rq *rq)
1300 {
1301 	return false;
1302 }
1303 
sched_core_disabled(void)1304 static inline bool sched_core_disabled(void)
1305 {
1306 	return true;
1307 }
1308 
rq_lockp(struct rq * rq)1309 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1310 {
1311 	return &rq->__lock;
1312 }
1313 
__rq_lockp(struct rq * rq)1314 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1315 {
1316 	return &rq->__lock;
1317 }
1318 
queue_core_balance(struct rq * rq)1319 static inline void queue_core_balance(struct rq *rq)
1320 {
1321 }
1322 
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1323 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1324 {
1325 	return true;
1326 }
1327 
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1328 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1329 {
1330 	return true;
1331 }
1332 
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1333 static inline bool sched_group_cookie_match(struct rq *rq,
1334 					    struct task_struct *p,
1335 					    struct sched_group *group)
1336 {
1337 	return true;
1338 }
1339 #endif /* CONFIG_SCHED_CORE */
1340 
lockdep_assert_rq_held(struct rq * rq)1341 static inline void lockdep_assert_rq_held(struct rq *rq)
1342 {
1343 	lockdep_assert_held(__rq_lockp(rq));
1344 }
1345 
1346 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1347 extern bool raw_spin_rq_trylock(struct rq *rq);
1348 extern void raw_spin_rq_unlock(struct rq *rq);
1349 
raw_spin_rq_lock(struct rq * rq)1350 static inline void raw_spin_rq_lock(struct rq *rq)
1351 {
1352 	raw_spin_rq_lock_nested(rq, 0);
1353 }
1354 
raw_spin_rq_lock_irq(struct rq * rq)1355 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1356 {
1357 	local_irq_disable();
1358 	raw_spin_rq_lock(rq);
1359 }
1360 
raw_spin_rq_unlock_irq(struct rq * rq)1361 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1362 {
1363 	raw_spin_rq_unlock(rq);
1364 	local_irq_enable();
1365 }
1366 
_raw_spin_rq_lock_irqsave(struct rq * rq)1367 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1368 {
1369 	unsigned long flags;
1370 	local_irq_save(flags);
1371 	raw_spin_rq_lock(rq);
1372 	return flags;
1373 }
1374 
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1375 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1376 {
1377 	raw_spin_rq_unlock(rq);
1378 	local_irq_restore(flags);
1379 }
1380 
1381 #define raw_spin_rq_lock_irqsave(rq, flags)	\
1382 do {						\
1383 	flags = _raw_spin_rq_lock_irqsave(rq);	\
1384 } while (0)
1385 
1386 #ifdef CONFIG_SCHED_SMT
1387 extern void __update_idle_core(struct rq *rq);
1388 
update_idle_core(struct rq * rq)1389 static inline void update_idle_core(struct rq *rq)
1390 {
1391 	if (static_branch_unlikely(&sched_smt_present))
1392 		__update_idle_core(rq);
1393 }
1394 
1395 #else
update_idle_core(struct rq * rq)1396 static inline void update_idle_core(struct rq *rq) { }
1397 #endif
1398 
1399 #ifdef CONFIG_FAIR_GROUP_SCHED
task_of(struct sched_entity * se)1400 static inline struct task_struct *task_of(struct sched_entity *se)
1401 {
1402 	SCHED_WARN_ON(!entity_is_task(se));
1403 	return container_of(se, struct task_struct, se);
1404 }
1405 
task_cfs_rq(struct task_struct * p)1406 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1407 {
1408 	return p->se.cfs_rq;
1409 }
1410 
1411 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(struct sched_entity * se)1412 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1413 {
1414 	return se->cfs_rq;
1415 }
1416 
1417 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1418 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1419 {
1420 	return grp->my_q;
1421 }
1422 
1423 #else
1424 
task_of(struct sched_entity * se)1425 static inline struct task_struct *task_of(struct sched_entity *se)
1426 {
1427 	return container_of(se, struct task_struct, se);
1428 }
1429 
task_cfs_rq(struct task_struct * p)1430 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1431 {
1432 	return &task_rq(p)->cfs;
1433 }
1434 
cfs_rq_of(struct sched_entity * se)1435 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1436 {
1437 	struct task_struct *p = task_of(se);
1438 	struct rq *rq = task_rq(p);
1439 
1440 	return &rq->cfs;
1441 }
1442 
1443 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1444 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1445 {
1446 	return NULL;
1447 }
1448 #endif
1449 
1450 extern void update_rq_clock(struct rq *rq);
1451 
__rq_clock_broken(struct rq * rq)1452 static inline u64 __rq_clock_broken(struct rq *rq)
1453 {
1454 	return READ_ONCE(rq->clock);
1455 }
1456 
1457 /*
1458  * rq::clock_update_flags bits
1459  *
1460  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1461  *  call to __schedule(). This is an optimisation to avoid
1462  *  neighbouring rq clock updates.
1463  *
1464  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1465  *  in effect and calls to update_rq_clock() are being ignored.
1466  *
1467  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1468  *  made to update_rq_clock() since the last time rq::lock was pinned.
1469  *
1470  * If inside of __schedule(), clock_update_flags will have been
1471  * shifted left (a left shift is a cheap operation for the fast path
1472  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1473  *
1474  *	if (rq-clock_update_flags >= RQCF_UPDATED)
1475  *
1476  * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1477  * one position though, because the next rq_unpin_lock() will shift it
1478  * back.
1479  */
1480 #define RQCF_REQ_SKIP		0x01
1481 #define RQCF_ACT_SKIP		0x02
1482 #define RQCF_UPDATED		0x04
1483 
assert_clock_updated(struct rq * rq)1484 static inline void assert_clock_updated(struct rq *rq)
1485 {
1486 	/*
1487 	 * The only reason for not seeing a clock update since the
1488 	 * last rq_pin_lock() is if we're currently skipping updates.
1489 	 */
1490 	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1491 }
1492 
rq_clock(struct rq * rq)1493 static inline u64 rq_clock(struct rq *rq)
1494 {
1495 	lockdep_assert_rq_held(rq);
1496 	assert_clock_updated(rq);
1497 
1498 	return rq->clock;
1499 }
1500 
rq_clock_task(struct rq * rq)1501 static inline u64 rq_clock_task(struct rq *rq)
1502 {
1503 	lockdep_assert_rq_held(rq);
1504 	assert_clock_updated(rq);
1505 
1506 	return rq->clock_task;
1507 }
1508 
rq_clock_task_mult(struct rq * rq)1509 static inline u64 rq_clock_task_mult(struct rq *rq)
1510 {
1511 	lockdep_assert_rq_held(rq);
1512 	assert_clock_updated(rq);
1513 
1514 	return rq->clock_task_mult;
1515 }
1516 
1517 /**
1518  * By default the decay is the default pelt decay period.
1519  * The decay shift can change the decay period in
1520  * multiples of 32.
1521  *  Decay shift		Decay period(ms)
1522  *	0			32
1523  *	1			64
1524  *	2			128
1525  *	3			256
1526  *	4			512
1527  */
1528 extern int sched_thermal_decay_shift;
1529 
rq_clock_thermal(struct rq * rq)1530 static inline u64 rq_clock_thermal(struct rq *rq)
1531 {
1532 	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1533 }
1534 
rq_clock_skip_update(struct rq * rq)1535 static inline void rq_clock_skip_update(struct rq *rq)
1536 {
1537 	lockdep_assert_rq_held(rq);
1538 	rq->clock_update_flags |= RQCF_REQ_SKIP;
1539 }
1540 
1541 /*
1542  * See rt task throttling, which is the only time a skip
1543  * request is canceled.
1544  */
rq_clock_cancel_skipupdate(struct rq * rq)1545 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1546 {
1547 	lockdep_assert_rq_held(rq);
1548 	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1549 }
1550 
1551 struct rq_flags {
1552 	unsigned long flags;
1553 	struct pin_cookie cookie;
1554 #ifdef CONFIG_SCHED_DEBUG
1555 	/*
1556 	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1557 	 * current pin context is stashed here in case it needs to be
1558 	 * restored in rq_repin_lock().
1559 	 */
1560 	unsigned int clock_update_flags;
1561 #endif
1562 };
1563 
1564 #ifdef CONFIG_SMP
1565 extern struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
1566 				 struct task_struct *p, int dest_cpu);
1567 #endif
1568 
1569 extern struct callback_head balance_push_callback;
1570 
1571 /*
1572  * Lockdep annotation that avoids accidental unlocks; it's like a
1573  * sticky/continuous lockdep_assert_held().
1574  *
1575  * This avoids code that has access to 'struct rq *rq' (basically everything in
1576  * the scheduler) from accidentally unlocking the rq if they do not also have a
1577  * copy of the (on-stack) 'struct rq_flags rf'.
1578  *
1579  * Also see Documentation/locking/lockdep-design.rst.
1580  */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1581 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1582 {
1583 	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1584 
1585 #ifdef CONFIG_SCHED_DEBUG
1586 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1587 	rf->clock_update_flags = 0;
1588 #ifdef CONFIG_SMP
1589 	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1590 #endif
1591 #endif
1592 }
1593 
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1594 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1595 {
1596 #ifdef CONFIG_SCHED_DEBUG
1597 	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1598 		rf->clock_update_flags = RQCF_UPDATED;
1599 #endif
1600 
1601 	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1602 }
1603 
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1604 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1605 {
1606 	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1607 
1608 #ifdef CONFIG_SCHED_DEBUG
1609 	/*
1610 	 * Restore the value we stashed in @rf for this pin context.
1611 	 */
1612 	rq->clock_update_flags |= rf->clock_update_flags;
1613 #endif
1614 }
1615 
1616 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1617 	__acquires(rq->lock);
1618 
1619 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1620 	__acquires(p->pi_lock)
1621 	__acquires(rq->lock);
1622 
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1623 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1624 	__releases(rq->lock)
1625 {
1626 	rq_unpin_lock(rq, rf);
1627 	raw_spin_rq_unlock(rq);
1628 }
1629 
1630 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1631 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1632 	__releases(rq->lock)
1633 	__releases(p->pi_lock)
1634 {
1635 	rq_unpin_lock(rq, rf);
1636 	raw_spin_rq_unlock(rq);
1637 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1638 }
1639 
1640 static inline void
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1641 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1642 	__acquires(rq->lock)
1643 {
1644 	raw_spin_rq_lock_irqsave(rq, rf->flags);
1645 	rq_pin_lock(rq, rf);
1646 }
1647 
1648 static inline void
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1649 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1650 	__acquires(rq->lock)
1651 {
1652 	raw_spin_rq_lock_irq(rq);
1653 	rq_pin_lock(rq, rf);
1654 }
1655 
1656 static inline void
rq_lock(struct rq * rq,struct rq_flags * rf)1657 rq_lock(struct rq *rq, struct rq_flags *rf)
1658 	__acquires(rq->lock)
1659 {
1660 	raw_spin_rq_lock(rq);
1661 	rq_pin_lock(rq, rf);
1662 }
1663 
1664 static inline void
rq_relock(struct rq * rq,struct rq_flags * rf)1665 rq_relock(struct rq *rq, struct rq_flags *rf)
1666 	__acquires(rq->lock)
1667 {
1668 	raw_spin_rq_lock(rq);
1669 	rq_repin_lock(rq, rf);
1670 }
1671 
1672 static inline void
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1673 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1674 	__releases(rq->lock)
1675 {
1676 	rq_unpin_lock(rq, rf);
1677 	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1678 }
1679 
1680 static inline void
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1681 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1682 	__releases(rq->lock)
1683 {
1684 	rq_unpin_lock(rq, rf);
1685 	raw_spin_rq_unlock_irq(rq);
1686 }
1687 
1688 static inline void
rq_unlock(struct rq * rq,struct rq_flags * rf)1689 rq_unlock(struct rq *rq, struct rq_flags *rf)
1690 	__releases(rq->lock)
1691 {
1692 	rq_unpin_lock(rq, rf);
1693 	raw_spin_rq_unlock(rq);
1694 }
1695 
1696 static inline struct rq *
this_rq_lock_irq(struct rq_flags * rf)1697 this_rq_lock_irq(struct rq_flags *rf)
1698 	__acquires(rq->lock)
1699 {
1700 	struct rq *rq;
1701 
1702 	local_irq_disable();
1703 	rq = this_rq();
1704 	rq_lock(rq, rf);
1705 	return rq;
1706 }
1707 
1708 #ifdef CONFIG_NUMA
1709 enum numa_topology_type {
1710 	NUMA_DIRECT,
1711 	NUMA_GLUELESS_MESH,
1712 	NUMA_BACKPLANE,
1713 };
1714 extern enum numa_topology_type sched_numa_topology_type;
1715 extern int sched_max_numa_distance;
1716 extern bool find_numa_distance(int distance);
1717 extern void sched_init_numa(void);
1718 extern void sched_domains_numa_masks_set(unsigned int cpu);
1719 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1720 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1721 #else
sched_init_numa(void)1722 static inline void sched_init_numa(void) { }
sched_domains_numa_masks_set(unsigned int cpu)1723 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1724 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1725 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1726 {
1727 	return nr_cpu_ids;
1728 }
1729 #endif
1730 
1731 #ifdef CONFIG_NUMA_BALANCING
1732 /* The regions in numa_faults array from task_struct */
1733 enum numa_faults_stats {
1734 	NUMA_MEM = 0,
1735 	NUMA_CPU,
1736 	NUMA_MEMBUF,
1737 	NUMA_CPUBUF
1738 };
1739 extern void sched_setnuma(struct task_struct *p, int node);
1740 extern int migrate_task_to(struct task_struct *p, int cpu);
1741 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1742 #else
1743 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1744 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1745 {
1746 }
1747 #endif /* CONFIG_NUMA_BALANCING */
1748 
1749 #ifdef CONFIG_SMP
1750 
1751 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1752 			int cpu, int scpu);
1753 
1754 static inline void
queue_balance_callback(struct rq * rq,struct callback_head * head,void (* func)(struct rq * rq))1755 queue_balance_callback(struct rq *rq,
1756 		       struct callback_head *head,
1757 		       void (*func)(struct rq *rq))
1758 {
1759 	lockdep_assert_rq_held(rq);
1760 
1761 	/*
1762 	 * Don't (re)queue an already queued item; nor queue anything when
1763 	 * balance_push() is active, see the comment with
1764 	 * balance_push_callback.
1765 	 */
1766 	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1767 		return;
1768 
1769 	head->func = (void (*)(struct callback_head *))func;
1770 	head->next = rq->balance_callback;
1771 	rq->balance_callback = head;
1772 }
1773 
1774 #define rcu_dereference_check_sched_domain(p) \
1775 	rcu_dereference_check((p), \
1776 			      lockdep_is_held(&sched_domains_mutex))
1777 
1778 /*
1779  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1780  * See destroy_sched_domains: call_rcu for details.
1781  *
1782  * The domain tree of any CPU may only be accessed from within
1783  * preempt-disabled sections.
1784  */
1785 #define for_each_domain(cpu, __sd) \
1786 	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1787 			__sd; __sd = __sd->parent)
1788 
1789 /**
1790  * highest_flag_domain - Return highest sched_domain containing flag.
1791  * @cpu:	The CPU whose highest level of sched domain is to
1792  *		be returned.
1793  * @flag:	The flag to check for the highest sched_domain
1794  *		for the given CPU.
1795  *
1796  * Returns the highest sched_domain of a CPU which contains the given flag.
1797  */
highest_flag_domain(int cpu,int flag)1798 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1799 {
1800 	struct sched_domain *sd, *hsd = NULL;
1801 
1802 	for_each_domain(cpu, sd) {
1803 		if (!(sd->flags & flag))
1804 			break;
1805 		hsd = sd;
1806 	}
1807 
1808 	return hsd;
1809 }
1810 
lowest_flag_domain(int cpu,int flag)1811 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1812 {
1813 	struct sched_domain *sd;
1814 
1815 	for_each_domain(cpu, sd) {
1816 		if (sd->flags & flag)
1817 			break;
1818 	}
1819 
1820 	return sd;
1821 }
1822 
1823 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1824 DECLARE_PER_CPU(int, sd_llc_size);
1825 DECLARE_PER_CPU(int, sd_llc_id);
1826 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1827 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1828 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1829 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1830 extern struct static_key_false sched_asym_cpucapacity;
1831 
sched_asym_cpucap_active(void)1832 static __always_inline bool sched_asym_cpucap_active(void)
1833 {
1834 	return static_branch_unlikely(&sched_asym_cpucapacity);
1835 }
1836 
1837 struct sched_group_capacity {
1838 	atomic_t		ref;
1839 	/*
1840 	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1841 	 * for a single CPU.
1842 	 */
1843 	unsigned long		capacity;
1844 	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1845 	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1846 	unsigned long		next_update;
1847 	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1848 
1849 #ifdef CONFIG_SCHED_DEBUG
1850 	int			id;
1851 #endif
1852 
1853 	unsigned long		cpumask[];		/* Balance mask */
1854 };
1855 
1856 struct sched_group {
1857 	struct sched_group	*next;			/* Must be a circular list */
1858 	atomic_t		ref;
1859 
1860 	unsigned int		group_weight;
1861 	struct sched_group_capacity *sgc;
1862 	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1863 
1864 	/*
1865 	 * The CPUs this group covers.
1866 	 *
1867 	 * NOTE: this field is variable length. (Allocated dynamically
1868 	 * by attaching extra space to the end of the structure,
1869 	 * depending on how many CPUs the kernel has booted up with)
1870 	 */
1871 	unsigned long		cpumask[];
1872 };
1873 
sched_group_span(struct sched_group * sg)1874 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1875 {
1876 	return to_cpumask(sg->cpumask);
1877 }
1878 
1879 /*
1880  * See build_balance_mask().
1881  */
group_balance_mask(struct sched_group * sg)1882 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1883 {
1884 	return to_cpumask(sg->sgc->cpumask);
1885 }
1886 
1887 /**
1888  * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1889  * @group: The group whose first CPU is to be returned.
1890  */
group_first_cpu(struct sched_group * group)1891 static inline unsigned int group_first_cpu(struct sched_group *group)
1892 {
1893 	return cpumask_first(sched_group_span(group));
1894 }
1895 
1896 extern int group_balance_cpu(struct sched_group *sg);
1897 
1898 #ifdef CONFIG_SCHED_DEBUG
1899 void update_sched_domain_debugfs(void);
1900 void dirty_sched_domain_sysctl(int cpu);
1901 #else
update_sched_domain_debugfs(void)1902 static inline void update_sched_domain_debugfs(void)
1903 {
1904 }
dirty_sched_domain_sysctl(int cpu)1905 static inline void dirty_sched_domain_sysctl(int cpu)
1906 {
1907 }
1908 #endif
1909 
1910 extern int sched_update_scaling(void);
1911 
1912 extern void flush_smp_call_function_from_idle(void);
1913 
1914 #else /* !CONFIG_SMP: */
flush_smp_call_function_from_idle(void)1915 static inline void flush_smp_call_function_from_idle(void) { }
1916 #endif
1917 
1918 #include "stats.h"
1919 #include "autogroup.h"
1920 
1921 #ifdef CONFIG_CGROUP_SCHED
1922 
1923 /*
1924  * Return the group to which this tasks belongs.
1925  *
1926  * We cannot use task_css() and friends because the cgroup subsystem
1927  * changes that value before the cgroup_subsys::attach() method is called,
1928  * therefore we cannot pin it and might observe the wrong value.
1929  *
1930  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1931  * core changes this before calling sched_move_task().
1932  *
1933  * Instead we use a 'copy' which is updated from sched_move_task() while
1934  * holding both task_struct::pi_lock and rq::lock.
1935  */
task_group(struct task_struct * p)1936 static inline struct task_group *task_group(struct task_struct *p)
1937 {
1938 	return p->sched_task_group;
1939 }
1940 
1941 /* 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)1942 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1943 {
1944 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1945 	struct task_group *tg = task_group(p);
1946 #endif
1947 
1948 #ifdef CONFIG_FAIR_GROUP_SCHED
1949 	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1950 	p->se.cfs_rq = tg->cfs_rq[cpu];
1951 	p->se.parent = tg->se[cpu];
1952 #endif
1953 
1954 #ifdef CONFIG_RT_GROUP_SCHED
1955 	p->rt.rt_rq  = tg->rt_rq[cpu];
1956 	p->rt.parent = tg->rt_se[cpu];
1957 #endif
1958 }
1959 
1960 #else /* CONFIG_CGROUP_SCHED */
1961 
set_task_rq(struct task_struct * p,unsigned int cpu)1962 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
task_group(struct task_struct * p)1963 static inline struct task_group *task_group(struct task_struct *p)
1964 {
1965 	return NULL;
1966 }
1967 
1968 #endif /* CONFIG_CGROUP_SCHED */
1969 
__set_task_cpu(struct task_struct * p,unsigned int cpu)1970 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1971 {
1972 	set_task_rq(p, cpu);
1973 #ifdef CONFIG_SMP
1974 	/*
1975 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1976 	 * successfully executed on another CPU. We must ensure that updates of
1977 	 * per-task data have been completed by this moment.
1978 	 */
1979 	smp_wmb();
1980 #ifdef CONFIG_THREAD_INFO_IN_TASK
1981 	WRITE_ONCE(p->cpu, cpu);
1982 #else
1983 	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1984 #endif
1985 	p->wake_cpu = cpu;
1986 #endif
1987 }
1988 
1989 /*
1990  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1991  */
1992 #ifdef CONFIG_SCHED_DEBUG
1993 # include <linux/static_key.h>
1994 # define const_debug __read_mostly
1995 #else
1996 # define const_debug const
1997 #endif
1998 
1999 #define SCHED_FEAT(name, enabled)	\
2000 	__SCHED_FEAT_##name ,
2001 
2002 enum {
2003 #include "features.h"
2004 	__SCHED_FEAT_NR,
2005 };
2006 
2007 #undef SCHED_FEAT
2008 
2009 #ifdef CONFIG_SCHED_DEBUG
2010 
2011 /*
2012  * To support run-time toggling of sched features, all the translation units
2013  * (but core.c) reference the sysctl_sched_features defined in core.c.
2014  */
2015 extern const_debug unsigned int sysctl_sched_features;
2016 
2017 #ifdef CONFIG_JUMP_LABEL
2018 #define SCHED_FEAT(name, enabled)					\
2019 static __always_inline bool static_branch_##name(struct static_key *key) \
2020 {									\
2021 	return static_key_##enabled(key);				\
2022 }
2023 
2024 #include "features.h"
2025 #undef SCHED_FEAT
2026 
2027 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2028 extern const char * const sched_feat_names[__SCHED_FEAT_NR];
2029 
2030 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2031 
2032 #else /* !CONFIG_JUMP_LABEL */
2033 
2034 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2035 
2036 #endif /* CONFIG_JUMP_LABEL */
2037 
2038 #else /* !SCHED_DEBUG */
2039 
2040 /*
2041  * Each translation unit has its own copy of sysctl_sched_features to allow
2042  * constants propagation at compile time and compiler optimization based on
2043  * features default.
2044  */
2045 #define SCHED_FEAT(name, enabled)	\
2046 	(1UL << __SCHED_FEAT_##name) * enabled |
2047 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2048 #include "features.h"
2049 	0;
2050 #undef SCHED_FEAT
2051 
2052 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2053 
2054 #endif /* SCHED_DEBUG */
2055 
2056 extern struct static_key_false sched_numa_balancing;
2057 extern struct static_key_false sched_schedstats;
2058 
global_rt_period(void)2059 static inline u64 global_rt_period(void)
2060 {
2061 	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2062 }
2063 
global_rt_runtime(void)2064 static inline u64 global_rt_runtime(void)
2065 {
2066 	if (sysctl_sched_rt_runtime < 0)
2067 		return RUNTIME_INF;
2068 
2069 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2070 }
2071 
task_current(struct rq * rq,struct task_struct * p)2072 static inline int task_current(struct rq *rq, struct task_struct *p)
2073 {
2074 	return rq->curr == p;
2075 }
2076 
task_running(struct rq * rq,struct task_struct * p)2077 static inline int task_running(struct rq *rq, struct task_struct *p)
2078 {
2079 #ifdef CONFIG_SMP
2080 	return p->on_cpu;
2081 #else
2082 	return task_current(rq, p);
2083 #endif
2084 }
2085 
task_on_rq_queued(struct task_struct * p)2086 static inline int task_on_rq_queued(struct task_struct *p)
2087 {
2088 	return p->on_rq == TASK_ON_RQ_QUEUED;
2089 }
2090 
task_on_rq_migrating(struct task_struct * p)2091 static inline int task_on_rq_migrating(struct task_struct *p)
2092 {
2093 	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2094 }
2095 
2096 /* Wake flags. The first three directly map to some SD flag value */
2097 #define WF_EXEC     0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2098 #define WF_FORK     0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2099 #define WF_TTWU     0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2100 
2101 #define WF_SYNC     0x10 /* Waker goes to sleep after wakeup */
2102 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2103 
2104 #define WF_ANDROID_VENDOR	0x1000 /* Vendor specific for Android */
2105 
2106 #ifdef CONFIG_SMP
2107 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2108 static_assert(WF_FORK == SD_BALANCE_FORK);
2109 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2110 #endif
2111 
2112 /*
2113  * To aid in avoiding the subversion of "niceness" due to uneven distribution
2114  * of tasks with abnormal "nice" values across CPUs the contribution that
2115  * each task makes to its run queue's load is weighted according to its
2116  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2117  * scaled version of the new time slice allocation that they receive on time
2118  * slice expiry etc.
2119  */
2120 
2121 #define WEIGHT_IDLEPRIO		3
2122 #define WMULT_IDLEPRIO		1431655765
2123 
2124 extern const int		sched_prio_to_weight[40];
2125 extern const u32		sched_prio_to_wmult[40];
2126 
2127 /*
2128  * {de,en}queue flags:
2129  *
2130  * DEQUEUE_SLEEP  - task is no longer runnable
2131  * ENQUEUE_WAKEUP - task just became runnable
2132  *
2133  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2134  *                are in a known state which allows modification. Such pairs
2135  *                should preserve as much state as possible.
2136  *
2137  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2138  *        in the runqueue.
2139  *
2140  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2141  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2142  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2143  *
2144  */
2145 
2146 #define DEQUEUE_SLEEP		0x01
2147 #define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2148 #define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2149 #define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
2150 
2151 #define ENQUEUE_WAKEUP		0x01
2152 #define ENQUEUE_RESTORE		0x02
2153 #define ENQUEUE_MOVE		0x04
2154 #define ENQUEUE_NOCLOCK		0x08
2155 
2156 #define ENQUEUE_HEAD		0x10
2157 #define ENQUEUE_REPLENISH	0x20
2158 #ifdef CONFIG_SMP
2159 #define ENQUEUE_MIGRATED	0x40
2160 #else
2161 #define ENQUEUE_MIGRATED	0x00
2162 #endif
2163 
2164 #define ENQUEUE_WAKEUP_SYNC	0x80
2165 
2166 #define RETRY_TASK		((void *)-1UL)
2167 
2168 struct sched_class {
2169 
2170 #ifdef CONFIG_UCLAMP_TASK
2171 	int uclamp_enabled;
2172 #endif
2173 
2174 	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2175 	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2176 	void (*yield_task)   (struct rq *rq);
2177 	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2178 
2179 	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2180 
2181 	struct task_struct *(*pick_next_task)(struct rq *rq);
2182 
2183 	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2184 	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2185 
2186 #ifdef CONFIG_SMP
2187 	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2188 	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2189 
2190 	struct task_struct * (*pick_task)(struct rq *rq);
2191 
2192 	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2193 
2194 	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2195 
2196 	void (*set_cpus_allowed)(struct task_struct *p,
2197 				 const struct cpumask *newmask,
2198 				 u32 flags);
2199 
2200 	void (*rq_online)(struct rq *rq);
2201 	void (*rq_offline)(struct rq *rq);
2202 
2203 	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2204 #endif
2205 
2206 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2207 	void (*task_fork)(struct task_struct *p);
2208 	void (*task_dead)(struct task_struct *p);
2209 
2210 	/*
2211 	 * The switched_from() call is allowed to drop rq->lock, therefore we
2212 	 * cannot assume the switched_from/switched_to pair is serialized by
2213 	 * rq->lock. They are however serialized by p->pi_lock.
2214 	 */
2215 	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2216 	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2217 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2218 			      int oldprio);
2219 
2220 	unsigned int (*get_rr_interval)(struct rq *rq,
2221 					struct task_struct *task);
2222 
2223 	void (*update_curr)(struct rq *rq);
2224 
2225 #define TASK_SET_GROUP		0
2226 #define TASK_MOVE_GROUP		1
2227 
2228 #ifdef CONFIG_FAIR_GROUP_SCHED
2229 	void (*task_change_group)(struct task_struct *p, int type);
2230 #endif
2231 };
2232 
put_prev_task(struct rq * rq,struct task_struct * prev)2233 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2234 {
2235 	WARN_ON_ONCE(rq->curr != prev);
2236 	prev->sched_class->put_prev_task(rq, prev);
2237 }
2238 
set_next_task(struct rq * rq,struct task_struct * next)2239 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2240 {
2241 	next->sched_class->set_next_task(rq, next, false);
2242 }
2243 
2244 
2245 /*
2246  * Helper to define a sched_class instance; each one is placed in a separate
2247  * section which is ordered by the linker script:
2248  *
2249  *   include/asm-generic/vmlinux.lds.h
2250  *
2251  * Also enforce alignment on the instance, not the type, to guarantee layout.
2252  */
2253 #define DEFINE_SCHED_CLASS(name) \
2254 const struct sched_class name##_sched_class \
2255 	__aligned(__alignof__(struct sched_class)) \
2256 	__section("__" #name "_sched_class")
2257 
2258 /* Defined in include/asm-generic/vmlinux.lds.h */
2259 extern struct sched_class __begin_sched_classes[];
2260 extern struct sched_class __end_sched_classes[];
2261 
2262 #define sched_class_highest (__end_sched_classes - 1)
2263 #define sched_class_lowest  (__begin_sched_classes - 1)
2264 
2265 #define for_class_range(class, _from, _to) \
2266 	for (class = (_from); class != (_to); class--)
2267 
2268 #define for_each_class(class) \
2269 	for_class_range(class, sched_class_highest, sched_class_lowest)
2270 
2271 extern const struct sched_class stop_sched_class;
2272 extern const struct sched_class dl_sched_class;
2273 extern const struct sched_class rt_sched_class;
2274 extern const struct sched_class fair_sched_class;
2275 extern const struct sched_class idle_sched_class;
2276 
sched_stop_runnable(struct rq * rq)2277 static inline bool sched_stop_runnable(struct rq *rq)
2278 {
2279 	return rq->stop && task_on_rq_queued(rq->stop);
2280 }
2281 
sched_dl_runnable(struct rq * rq)2282 static inline bool sched_dl_runnable(struct rq *rq)
2283 {
2284 	return rq->dl.dl_nr_running > 0;
2285 }
2286 
sched_rt_runnable(struct rq * rq)2287 static inline bool sched_rt_runnable(struct rq *rq)
2288 {
2289 	return rq->rt.rt_queued > 0;
2290 }
2291 
sched_fair_runnable(struct rq * rq)2292 static inline bool sched_fair_runnable(struct rq *rq)
2293 {
2294 	return rq->cfs.nr_running > 0;
2295 }
2296 
2297 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2298 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2299 
2300 #define SCA_CHECK		0x01
2301 #define SCA_MIGRATE_DISABLE	0x02
2302 #define SCA_MIGRATE_ENABLE	0x04
2303 #define SCA_USER		0x08
2304 
2305 #ifdef CONFIG_SMP
2306 
2307 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2308 
2309 extern void trigger_load_balance(struct rq *rq);
2310 
2311 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2312 
get_push_task(struct rq * rq)2313 static inline struct task_struct *get_push_task(struct rq *rq)
2314 {
2315 	struct task_struct *p = rq->curr;
2316 
2317 	lockdep_assert_rq_held(rq);
2318 
2319 	if (rq->push_busy)
2320 		return NULL;
2321 
2322 	if (p->nr_cpus_allowed == 1)
2323 		return NULL;
2324 
2325 	if (p->migration_disabled)
2326 		return NULL;
2327 
2328 	rq->push_busy = true;
2329 	return get_task_struct(p);
2330 }
2331 
2332 extern int push_cpu_stop(void *arg);
2333 
2334 extern unsigned long __read_mostly max_load_balance_interval;
2335 #endif
2336 
2337 #ifdef CONFIG_CPU_IDLE
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2338 static inline void idle_set_state(struct rq *rq,
2339 				  struct cpuidle_state *idle_state)
2340 {
2341 	rq->idle_state = idle_state;
2342 }
2343 
idle_get_state(struct rq * rq)2344 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2345 {
2346 	SCHED_WARN_ON(!rcu_read_lock_held());
2347 
2348 	return rq->idle_state;
2349 }
2350 #else
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2351 static inline void idle_set_state(struct rq *rq,
2352 				  struct cpuidle_state *idle_state)
2353 {
2354 }
2355 
idle_get_state(struct rq * rq)2356 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2357 {
2358 	return NULL;
2359 }
2360 #endif
2361 
2362 extern void schedule_idle(void);
2363 
2364 extern void sysrq_sched_debug_show(void);
2365 extern void sched_init_granularity(void);
2366 extern void update_max_interval(void);
2367 
2368 extern void init_sched_dl_class(void);
2369 extern void init_sched_rt_class(void);
2370 extern void init_sched_fair_class(void);
2371 
2372 extern void reweight_task(struct task_struct *p, int prio);
2373 
2374 extern void resched_curr(struct rq *rq);
2375 extern void resched_cpu(int cpu);
2376 
2377 extern struct rt_bandwidth def_rt_bandwidth;
2378 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2379 
2380 extern struct dl_bandwidth def_dl_bandwidth;
2381 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2382 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2383 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2384 
2385 #define BW_SHIFT		20
2386 #define BW_UNIT			(1 << BW_SHIFT)
2387 #define RATIO_SHIFT		8
2388 #define MAX_BW_BITS		(64 - BW_SHIFT)
2389 #define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2390 unsigned long to_ratio(u64 period, u64 runtime);
2391 
2392 extern void init_entity_runnable_average(struct sched_entity *se);
2393 extern void post_init_entity_util_avg(struct task_struct *p);
2394 
2395 #ifdef CONFIG_NO_HZ_FULL
2396 extern bool sched_can_stop_tick(struct rq *rq);
2397 extern int __init sched_tick_offload_init(void);
2398 
2399 /*
2400  * Tick may be needed by tasks in the runqueue depending on their policy and
2401  * requirements. If tick is needed, lets send the target an IPI to kick it out of
2402  * nohz mode if necessary.
2403  */
sched_update_tick_dependency(struct rq * rq)2404 static inline void sched_update_tick_dependency(struct rq *rq)
2405 {
2406 	int cpu = cpu_of(rq);
2407 
2408 	if (!tick_nohz_full_cpu(cpu))
2409 		return;
2410 
2411 	if (sched_can_stop_tick(rq))
2412 		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2413 	else
2414 		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2415 }
2416 #else
sched_tick_offload_init(void)2417 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2418 static inline void sched_update_tick_dependency(struct rq *rq) { }
2419 #endif
2420 
add_nr_running(struct rq * rq,unsigned count)2421 static inline void add_nr_running(struct rq *rq, unsigned count)
2422 {
2423 	unsigned prev_nr = rq->nr_running;
2424 
2425 	rq->nr_running = prev_nr + count;
2426 	if (trace_sched_update_nr_running_tp_enabled()) {
2427 		call_trace_sched_update_nr_running(rq, count);
2428 	}
2429 
2430 #ifdef CONFIG_SMP
2431 	if (prev_nr < 2 && rq->nr_running >= 2) {
2432 		if (!READ_ONCE(rq->rd->overload))
2433 			WRITE_ONCE(rq->rd->overload, 1);
2434 	}
2435 #endif
2436 
2437 	sched_update_tick_dependency(rq);
2438 }
2439 
sub_nr_running(struct rq * rq,unsigned count)2440 static inline void sub_nr_running(struct rq *rq, unsigned count)
2441 {
2442 	rq->nr_running -= count;
2443 	if (trace_sched_update_nr_running_tp_enabled()) {
2444 		call_trace_sched_update_nr_running(rq, -count);
2445 	}
2446 
2447 	/* Check if we still need preemption */
2448 	sched_update_tick_dependency(rq);
2449 }
2450 
2451 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2452 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2453 
2454 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2455 
2456 extern const_debug unsigned int sysctl_sched_nr_migrate;
2457 extern const_debug unsigned int sysctl_sched_migration_cost;
2458 
2459 #ifdef CONFIG_SCHED_DEBUG
2460 extern unsigned int sysctl_sched_latency;
2461 extern unsigned int sysctl_sched_min_granularity;
2462 extern unsigned int sysctl_sched_wakeup_granularity;
2463 extern int sysctl_resched_latency_warn_ms;
2464 extern int sysctl_resched_latency_warn_once;
2465 
2466 extern unsigned int sysctl_sched_tunable_scaling;
2467 
2468 extern unsigned int sysctl_numa_balancing_scan_delay;
2469 extern unsigned int sysctl_numa_balancing_scan_period_min;
2470 extern unsigned int sysctl_numa_balancing_scan_period_max;
2471 extern unsigned int sysctl_numa_balancing_scan_size;
2472 #endif
2473 
2474 #ifdef CONFIG_SCHED_HRTICK
2475 
2476 /*
2477  * Use hrtick when:
2478  *  - enabled by features
2479  *  - hrtimer is actually high res
2480  */
hrtick_enabled(struct rq * rq)2481 static inline int hrtick_enabled(struct rq *rq)
2482 {
2483 	if (!cpu_active(cpu_of(rq)))
2484 		return 0;
2485 	return hrtimer_is_hres_active(&rq->hrtick_timer);
2486 }
2487 
hrtick_enabled_fair(struct rq * rq)2488 static inline int hrtick_enabled_fair(struct rq *rq)
2489 {
2490 	if (!sched_feat(HRTICK))
2491 		return 0;
2492 	return hrtick_enabled(rq);
2493 }
2494 
hrtick_enabled_dl(struct rq * rq)2495 static inline int hrtick_enabled_dl(struct rq *rq)
2496 {
2497 	if (!sched_feat(HRTICK_DL))
2498 		return 0;
2499 	return hrtick_enabled(rq);
2500 }
2501 
2502 void hrtick_start(struct rq *rq, u64 delay);
2503 
2504 #else
2505 
hrtick_enabled_fair(struct rq * rq)2506 static inline int hrtick_enabled_fair(struct rq *rq)
2507 {
2508 	return 0;
2509 }
2510 
hrtick_enabled_dl(struct rq * rq)2511 static inline int hrtick_enabled_dl(struct rq *rq)
2512 {
2513 	return 0;
2514 }
2515 
hrtick_enabled(struct rq * rq)2516 static inline int hrtick_enabled(struct rq *rq)
2517 {
2518 	return 0;
2519 }
2520 
2521 #endif /* CONFIG_SCHED_HRTICK */
2522 
2523 #ifndef arch_scale_freq_tick
2524 static __always_inline
arch_scale_freq_tick(void)2525 void arch_scale_freq_tick(void)
2526 {
2527 }
2528 #endif
2529 
2530 #ifndef arch_scale_freq_capacity
2531 /**
2532  * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2533  * @cpu: the CPU in question.
2534  *
2535  * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2536  *
2537  *     f_curr
2538  *     ------ * SCHED_CAPACITY_SCALE
2539  *     f_max
2540  */
2541 static __always_inline
arch_scale_freq_capacity(int cpu)2542 unsigned long arch_scale_freq_capacity(int cpu)
2543 {
2544 	return SCHED_CAPACITY_SCALE;
2545 }
2546 #endif
2547 
2548 #ifdef CONFIG_SCHED_DEBUG
2549 /*
2550  * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2551  * acquire rq lock instead of rq_lock(). So at the end of these two functions
2552  * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2553  * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2554  */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2555 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2556 {
2557 	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2558 	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2559 #ifdef CONFIG_SMP
2560 	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2561 #endif
2562 }
2563 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2564 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2565 #endif
2566 
2567 #ifdef CONFIG_SMP
2568 
rq_order_less(struct rq * rq1,struct rq * rq2)2569 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2570 {
2571 #ifdef CONFIG_SCHED_CORE
2572 	/*
2573 	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2574 	 * order by core-id first and cpu-id second.
2575 	 *
2576 	 * Notably:
2577 	 *
2578 	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2579 	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2580 	 *
2581 	 * when only cpu-id is considered.
2582 	 */
2583 	if (rq1->core->cpu < rq2->core->cpu)
2584 		return true;
2585 	if (rq1->core->cpu > rq2->core->cpu)
2586 		return false;
2587 
2588 	/*
2589 	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2590 	 */
2591 #endif
2592 	return rq1->cpu < rq2->cpu;
2593 }
2594 
2595 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2596 
2597 #ifdef CONFIG_PREEMPTION
2598 
2599 /*
2600  * fair double_lock_balance: Safely acquires both rq->locks in a fair
2601  * way at the expense of forcing extra atomic operations in all
2602  * invocations.  This assures that the double_lock is acquired using the
2603  * same underlying policy as the spinlock_t on this architecture, which
2604  * reduces latency compared to the unfair variant below.  However, it
2605  * also adds more overhead and therefore may reduce throughput.
2606  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2607 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2608 	__releases(this_rq->lock)
2609 	__acquires(busiest->lock)
2610 	__acquires(this_rq->lock)
2611 {
2612 	raw_spin_rq_unlock(this_rq);
2613 	double_rq_lock(this_rq, busiest);
2614 
2615 	return 1;
2616 }
2617 
2618 #else
2619 /*
2620  * Unfair double_lock_balance: Optimizes throughput at the expense of
2621  * latency by eliminating extra atomic operations when the locks are
2622  * already in proper order on entry.  This favors lower CPU-ids and will
2623  * grant the double lock to lower CPUs over higher ids under contention,
2624  * regardless of entry order into the function.
2625  */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2626 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2627 	__releases(this_rq->lock)
2628 	__acquires(busiest->lock)
2629 	__acquires(this_rq->lock)
2630 {
2631 	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2632 	    likely(raw_spin_rq_trylock(busiest))) {
2633 		double_rq_clock_clear_update(this_rq, busiest);
2634 		return 0;
2635 	}
2636 
2637 	if (rq_order_less(this_rq, busiest)) {
2638 		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2639 		double_rq_clock_clear_update(this_rq, busiest);
2640 		return 0;
2641 	}
2642 
2643 	raw_spin_rq_unlock(this_rq);
2644 	double_rq_lock(this_rq, busiest);
2645 
2646 	return 1;
2647 }
2648 
2649 #endif /* CONFIG_PREEMPTION */
2650 
2651 /*
2652  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2653  */
double_lock_balance(struct rq * this_rq,struct rq * busiest)2654 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2655 {
2656 	lockdep_assert_irqs_disabled();
2657 
2658 	return _double_lock_balance(this_rq, busiest);
2659 }
2660 
double_unlock_balance(struct rq * this_rq,struct rq * busiest)2661 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2662 	__releases(busiest->lock)
2663 {
2664 	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2665 		raw_spin_rq_unlock(busiest);
2666 	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2667 }
2668 
double_lock(spinlock_t * l1,spinlock_t * l2)2669 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2670 {
2671 	if (l1 > l2)
2672 		swap(l1, l2);
2673 
2674 	spin_lock(l1);
2675 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2676 }
2677 
double_lock_irq(spinlock_t * l1,spinlock_t * l2)2678 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2679 {
2680 	if (l1 > l2)
2681 		swap(l1, l2);
2682 
2683 	spin_lock_irq(l1);
2684 	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2685 }
2686 
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)2687 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2688 {
2689 	if (l1 > l2)
2690 		swap(l1, l2);
2691 
2692 	raw_spin_lock(l1);
2693 	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2694 }
2695 
2696 /*
2697  * double_rq_unlock - safely unlock two runqueues
2698  *
2699  * Note this does not restore interrupts like task_rq_unlock,
2700  * you need to do so manually after calling.
2701  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2702 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2703 	__releases(rq1->lock)
2704 	__releases(rq2->lock)
2705 {
2706 	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2707 		raw_spin_rq_unlock(rq2);
2708 	else
2709 		__release(rq2->lock);
2710 	raw_spin_rq_unlock(rq1);
2711 }
2712 
2713 extern void set_rq_online (struct rq *rq);
2714 extern void set_rq_offline(struct rq *rq);
2715 extern bool sched_smp_initialized;
2716 
2717 #else /* CONFIG_SMP */
2718 
2719 /*
2720  * double_rq_lock - safely lock two runqueues
2721  *
2722  * Note this does not disable interrupts like task_rq_lock,
2723  * you need to do so manually before calling.
2724  */
double_rq_lock(struct rq * rq1,struct rq * rq2)2725 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2726 	__acquires(rq1->lock)
2727 	__acquires(rq2->lock)
2728 {
2729 	BUG_ON(!irqs_disabled());
2730 	BUG_ON(rq1 != rq2);
2731 	raw_spin_rq_lock(rq1);
2732 	__acquire(rq2->lock);	/* Fake it out ;) */
2733 	double_rq_clock_clear_update(rq1, rq2);
2734 }
2735 
2736 /*
2737  * double_rq_unlock - safely unlock two runqueues
2738  *
2739  * Note this does not restore interrupts like task_rq_unlock,
2740  * you need to do so manually after calling.
2741  */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2742 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2743 	__releases(rq1->lock)
2744 	__releases(rq2->lock)
2745 {
2746 	BUG_ON(rq1 != rq2);
2747 	raw_spin_rq_unlock(rq1);
2748 	__release(rq2->lock);
2749 }
2750 
2751 #endif
2752 
2753 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2754 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2755 
2756 #ifdef	CONFIG_SCHED_DEBUG
2757 extern bool sched_debug_verbose;
2758 
2759 extern void print_cfs_stats(struct seq_file *m, int cpu);
2760 extern void print_rt_stats(struct seq_file *m, int cpu);
2761 extern void print_dl_stats(struct seq_file *m, int cpu);
2762 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2763 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2764 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2765 
2766 extern void resched_latency_warn(int cpu, u64 latency);
2767 #ifdef CONFIG_NUMA_BALANCING
2768 extern void
2769 show_numa_stats(struct task_struct *p, struct seq_file *m);
2770 extern void
2771 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2772 	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2773 #endif /* CONFIG_NUMA_BALANCING */
2774 #else
resched_latency_warn(int cpu,u64 latency)2775 static inline void resched_latency_warn(int cpu, u64 latency) {}
2776 #endif /* CONFIG_SCHED_DEBUG */
2777 
2778 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2779 extern void init_rt_rq(struct rt_rq *rt_rq);
2780 extern void init_dl_rq(struct dl_rq *dl_rq);
2781 
2782 extern void cfs_bandwidth_usage_inc(void);
2783 extern void cfs_bandwidth_usage_dec(void);
2784 
2785 #ifdef CONFIG_NO_HZ_COMMON
2786 #define NOHZ_BALANCE_KICK_BIT	0
2787 #define NOHZ_STATS_KICK_BIT	1
2788 #define NOHZ_NEWILB_KICK_BIT	2
2789 
2790 #define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2791 #define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2792 #define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2793 
2794 #define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2795 
2796 #define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2797 
2798 extern void nohz_balance_exit_idle(struct rq *rq);
2799 #else
nohz_balance_exit_idle(struct rq * rq)2800 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2801 #endif
2802 
2803 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2804 extern void nohz_run_idle_balance(int cpu);
2805 #else
nohz_run_idle_balance(int cpu)2806 static inline void nohz_run_idle_balance(int cpu) { }
2807 #endif
2808 
2809 #ifdef CONFIG_SMP
2810 static inline
__dl_update(struct dl_bw * dl_b,s64 bw)2811 void __dl_update(struct dl_bw *dl_b, s64 bw)
2812 {
2813 	struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2814 	int i;
2815 
2816 	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2817 			 "sched RCU must be held");
2818 	for_each_cpu_and(i, rd->span, cpu_active_mask) {
2819 		struct rq *rq = cpu_rq(i);
2820 
2821 		rq->dl.extra_bw += bw;
2822 	}
2823 }
2824 #else
2825 static inline
__dl_update(struct dl_bw * dl_b,s64 bw)2826 void __dl_update(struct dl_bw *dl_b, s64 bw)
2827 {
2828 	struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2829 
2830 	dl->extra_bw += bw;
2831 }
2832 #endif
2833 
2834 
2835 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2836 struct irqtime {
2837 	u64			total;
2838 	u64			tick_delta;
2839 	u64			irq_start_time;
2840 	struct u64_stats_sync	sync;
2841 };
2842 
2843 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2844 
2845 /*
2846  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2847  * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2848  * and never move forward.
2849  */
irq_time_read(int cpu)2850 static inline u64 irq_time_read(int cpu)
2851 {
2852 	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2853 	unsigned int seq;
2854 	u64 total;
2855 
2856 	do {
2857 		seq = __u64_stats_fetch_begin(&irqtime->sync);
2858 		total = irqtime->total;
2859 	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2860 
2861 	return total;
2862 }
2863 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2864 
2865 #ifdef CONFIG_CPU_FREQ
2866 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2867 
2868 /**
2869  * cpufreq_update_util - Take a note about CPU utilization changes.
2870  * @rq: Runqueue to carry out the update for.
2871  * @flags: Update reason flags.
2872  *
2873  * This function is called by the scheduler on the CPU whose utilization is
2874  * being updated.
2875  *
2876  * It can only be called from RCU-sched read-side critical sections.
2877  *
2878  * The way cpufreq is currently arranged requires it to evaluate the CPU
2879  * performance state (frequency/voltage) on a regular basis to prevent it from
2880  * being stuck in a completely inadequate performance level for too long.
2881  * That is not guaranteed to happen if the updates are only triggered from CFS
2882  * and DL, though, because they may not be coming in if only RT tasks are
2883  * active all the time (or there are RT tasks only).
2884  *
2885  * As a workaround for that issue, this function is called periodically by the
2886  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2887  * but that really is a band-aid.  Going forward it should be replaced with
2888  * solutions targeted more specifically at RT tasks.
2889  */
cpufreq_update_util(struct rq * rq,unsigned int flags)2890 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2891 {
2892 	struct update_util_data *data;
2893 
2894 	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2895 						  cpu_of(rq)));
2896 	if (data)
2897 		data->func(data, rq_clock(rq), flags);
2898 }
2899 #else
cpufreq_update_util(struct rq * rq,unsigned int flags)2900 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2901 #endif /* CONFIG_CPU_FREQ */
2902 
2903 #ifdef CONFIG_UCLAMP_TASK
2904 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2905 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)2906 static inline unsigned long uclamp_rq_get(struct rq *rq,
2907 					  enum uclamp_id clamp_id)
2908 {
2909 	return READ_ONCE(rq->uclamp[clamp_id].value);
2910 }
2911 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)2912 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
2913 				 unsigned int value)
2914 {
2915 	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
2916 }
2917 
uclamp_rq_is_idle(struct rq * rq)2918 static inline bool uclamp_rq_is_idle(struct rq *rq)
2919 {
2920 	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
2921 }
2922 
2923 /**
2924  * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2925  * @rq:		The rq to clamp against. Must not be NULL.
2926  * @util:	The util value to clamp.
2927  * @p:		The task to clamp against. Can be NULL if you want to clamp
2928  *		against @rq only.
2929  *
2930  * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2931  *
2932  * If sched_uclamp_used static key is disabled, then just return the util
2933  * without any clamping since uclamp aggregation at the rq level in the fast
2934  * path is disabled, rendering this operation a NOP.
2935  *
2936  * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2937  * will return the correct effective uclamp value of the task even if the
2938  * static key is disabled.
2939  */
2940 static __always_inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)2941 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2942 				  struct task_struct *p)
2943 {
2944 	unsigned long min_util = 0;
2945 	unsigned long max_util = 0;
2946 
2947 	if (!static_branch_likely(&sched_uclamp_used))
2948 		return util;
2949 
2950 	if (p) {
2951 		min_util = uclamp_eff_value(p, UCLAMP_MIN);
2952 		max_util = uclamp_eff_value(p, UCLAMP_MAX);
2953 
2954 		/*
2955 		 * Ignore last runnable task's max clamp, as this task will
2956 		 * reset it. Similarly, no need to read the rq's min clamp.
2957 		 */
2958 		if (uclamp_rq_is_idle(rq))
2959 			goto out;
2960 	}
2961 
2962 	min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
2963 	max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
2964 out:
2965 	/*
2966 	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2967 	 * RUNNABLE tasks with _different_ clamps, we can end up with an
2968 	 * inversion. Fix it now when the clamps are applied.
2969 	 */
2970 	if (unlikely(min_util >= max_util))
2971 		return min_util;
2972 
2973 	return clamp(util, min_util, max_util);
2974 }
2975 
uclamp_boosted(struct task_struct * p)2976 static inline bool uclamp_boosted(struct task_struct *p)
2977 {
2978 	return uclamp_eff_value(p, UCLAMP_MIN) > 0;
2979 }
2980 
2981 /*
2982  * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2983  * by default in the fast path and only gets turned on once userspace performs
2984  * an operation that requires it.
2985  *
2986  * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2987  * hence is active.
2988  */
uclamp_is_used(void)2989 static inline bool uclamp_is_used(void)
2990 {
2991 	return static_branch_likely(&sched_uclamp_used);
2992 }
2993 #else /* CONFIG_UCLAMP_TASK */
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)2994 static inline unsigned long uclamp_eff_value(struct task_struct *p,
2995 					     enum uclamp_id clamp_id)
2996 {
2997 	if (clamp_id == UCLAMP_MIN)
2998 		return 0;
2999 
3000 	return SCHED_CAPACITY_SCALE;
3001 }
3002 
3003 static inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)3004 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3005 				  struct task_struct *p)
3006 {
3007 	return util;
3008 }
3009 
uclamp_boosted(struct task_struct * p)3010 static inline bool uclamp_boosted(struct task_struct *p)
3011 {
3012 	return false;
3013 }
3014 
uclamp_is_used(void)3015 static inline bool uclamp_is_used(void)
3016 {
3017 	return false;
3018 }
3019 
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3020 static inline unsigned long uclamp_rq_get(struct rq *rq,
3021 					  enum uclamp_id clamp_id)
3022 {
3023 	if (clamp_id == UCLAMP_MIN)
3024 		return 0;
3025 
3026 	return SCHED_CAPACITY_SCALE;
3027 }
3028 
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3029 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3030 				 unsigned int value)
3031 {
3032 }
3033 
uclamp_rq_is_idle(struct rq * rq)3034 static inline bool uclamp_rq_is_idle(struct rq *rq)
3035 {
3036 	return false;
3037 }
3038 #endif /* CONFIG_UCLAMP_TASK */
3039 
3040 #ifdef CONFIG_UCLAMP_TASK_GROUP
uclamp_latency_sensitive(struct task_struct * p)3041 static inline bool uclamp_latency_sensitive(struct task_struct *p)
3042 {
3043 	struct cgroup_subsys_state *css = task_css(p, cpu_cgrp_id);
3044 	struct task_group *tg;
3045 
3046 	if (!css)
3047 		return false;
3048 	tg = container_of(css, struct task_group, css);
3049 
3050 	return tg->latency_sensitive;
3051 }
3052 #else
uclamp_latency_sensitive(struct task_struct * p)3053 static inline bool uclamp_latency_sensitive(struct task_struct *p)
3054 {
3055 	return false;
3056 }
3057 #endif /* CONFIG_UCLAMP_TASK_GROUP */
3058 
3059 #ifdef arch_scale_freq_capacity
3060 # ifndef arch_scale_freq_invariant
3061 #  define arch_scale_freq_invariant()	true
3062 # endif
3063 #else
3064 # define arch_scale_freq_invariant()	false
3065 #endif
3066 
3067 #ifdef CONFIG_SMP
capacity_orig_of(int cpu)3068 static inline unsigned long capacity_orig_of(int cpu)
3069 {
3070 	return cpu_rq(cpu)->cpu_capacity_orig;
3071 }
3072 
3073 /**
3074  * enum cpu_util_type - CPU utilization type
3075  * @FREQUENCY_UTIL:	Utilization used to select frequency
3076  * @ENERGY_UTIL:	Utilization used during energy calculation
3077  *
3078  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
3079  * need to be aggregated differently depending on the usage made of them. This
3080  * enum is used within effective_cpu_util() to differentiate the types of
3081  * utilization expected by the callers, and adjust the aggregation accordingly.
3082  */
3083 enum cpu_util_type {
3084 	FREQUENCY_UTIL,
3085 	ENERGY_UTIL,
3086 };
3087 
3088 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3089 				 unsigned long max, enum cpu_util_type type,
3090 				 struct task_struct *p);
3091 
cpu_bw_dl(struct rq * rq)3092 static inline unsigned long cpu_bw_dl(struct rq *rq)
3093 {
3094 	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3095 }
3096 
cpu_util_dl(struct rq * rq)3097 static inline unsigned long cpu_util_dl(struct rq *rq)
3098 {
3099 	return READ_ONCE(rq->avg_dl.util_avg);
3100 }
3101 
cpu_util_cfs(struct rq * rq)3102 static inline unsigned long cpu_util_cfs(struct rq *rq)
3103 {
3104 	unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
3105 
3106 	if (sched_feat(UTIL_EST)) {
3107 		util = max_t(unsigned long, util,
3108 			     READ_ONCE(rq->cfs.avg.util_est.enqueued));
3109 	}
3110 
3111 	return util;
3112 }
3113 
cpu_util_rt(struct rq * rq)3114 static inline unsigned long cpu_util_rt(struct rq *rq)
3115 {
3116 	return READ_ONCE(rq->avg_rt.util_avg);
3117 }
3118 #endif
3119 
3120 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
cpu_util_irq(struct rq * rq)3121 static inline unsigned long cpu_util_irq(struct rq *rq)
3122 {
3123 	return rq->avg_irq.util_avg;
3124 }
3125 
3126 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3127 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3128 {
3129 	util *= (max - irq);
3130 	util /= max;
3131 
3132 	return util;
3133 
3134 }
3135 #else
cpu_util_irq(struct rq * rq)3136 static inline unsigned long cpu_util_irq(struct rq *rq)
3137 {
3138 	return 0;
3139 }
3140 
3141 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3142 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3143 {
3144 	return util;
3145 }
3146 #endif
3147 
3148 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3149 
3150 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3151 
3152 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3153 
sched_energy_enabled(void)3154 static inline bool sched_energy_enabled(void)
3155 {
3156 	return static_branch_unlikely(&sched_energy_present);
3157 }
3158 
3159 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3160 
3161 #define perf_domain_span(pd) NULL
sched_energy_enabled(void)3162 static inline bool sched_energy_enabled(void) { return false; }
3163 
3164 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3165 
3166 #ifdef CONFIG_MEMBARRIER
3167 /*
3168  * The scheduler provides memory barriers required by membarrier between:
3169  * - prior user-space memory accesses and store to rq->membarrier_state,
3170  * - store to rq->membarrier_state and following user-space memory accesses.
3171  * In the same way it provides those guarantees around store to rq->curr.
3172  */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3173 static inline void membarrier_switch_mm(struct rq *rq,
3174 					struct mm_struct *prev_mm,
3175 					struct mm_struct *next_mm)
3176 {
3177 	int membarrier_state;
3178 
3179 	if (prev_mm == next_mm)
3180 		return;
3181 
3182 	membarrier_state = atomic_read(&next_mm->membarrier_state);
3183 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3184 		return;
3185 
3186 	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3187 }
3188 #else
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3189 static inline void membarrier_switch_mm(struct rq *rq,
3190 					struct mm_struct *prev_mm,
3191 					struct mm_struct *next_mm)
3192 {
3193 }
3194 #endif
3195 
3196 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3197 static inline bool is_per_cpu_kthread(struct task_struct *p)
3198 {
3199 	if (!(p->flags & PF_KTHREAD))
3200 		return false;
3201 
3202 	if (p->nr_cpus_allowed != 1)
3203 		return false;
3204 
3205 	return true;
3206 }
3207 #endif
3208 
3209 extern void swake_up_all_locked(struct swait_queue_head *q);
3210 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3211 
3212 #ifdef CONFIG_PREEMPT_DYNAMIC
3213 extern int preempt_dynamic_mode;
3214 extern int sched_dynamic_mode(const char *str);
3215 extern void sched_dynamic_update(int mode);
3216 #endif
3217 
3218 /*
3219  * task_may_not_preempt - check whether a task may not be preemptible soon
3220  */
3221 #ifdef CONFIG_RT_SOFTINT_OPTIMIZATION
3222 extern bool task_may_not_preempt(struct task_struct *task, int cpu);
3223 #else
task_may_not_preempt(struct task_struct * task,int cpu)3224 static inline bool task_may_not_preempt(struct task_struct *task, int cpu)
3225 {
3226 	return false;
3227 }
3228 #endif /* CONFIG_RT_SOFTINT_OPTIMIZATION */
3229