1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Scheduler internal types and methods:
4 */
5 #include <linux/sched.h>
6
7 #include <linux/sched/autogroup.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/coredump.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/cputime.h>
12 #include <linux/sched/deadline.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/hotplug.h>
15 #include <linux/sched/idle.h>
16 #include <linux/sched/init.h>
17 #include <linux/sched/isolation.h>
18 #include <linux/sched/jobctl.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/mm.h>
21 #include <linux/sched/nohz.h>
22 #include <linux/sched/numa_balancing.h>
23 #include <linux/sched/prio.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/smt.h>
27 #include <linux/sched/stat.h>
28 #include <linux/sched/sysctl.h>
29 #include <linux/sched/task.h>
30 #include <linux/sched/task_stack.h>
31 #include <linux/sched/topology.h>
32 #include <linux/sched/user.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/sched/xacct.h>
35
36 #include <uapi/linux/sched/types.h>
37
38 #include <linux/binfmts.h>
39 #include <linux/blkdev.h>
40 #include <linux/compat.h>
41 #include <linux/context_tracking.h>
42 #include <linux/cpufreq.h>
43 #include <linux/cpuidle.h>
44 #include <linux/cpuset.h>
45 #include <linux/ctype.h>
46 #include <linux/debugfs.h>
47 #include <linux/delayacct.h>
48 #include <linux/energy_model.h>
49 #include <linux/init_task.h>
50 #include <linux/kprobes.h>
51 #include <linux/kthread.h>
52 #include <linux/membarrier.h>
53 #include <linux/migrate.h>
54 #include <linux/mmu_context.h>
55 #include <linux/nmi.h>
56 #include <linux/proc_fs.h>
57 #include <linux/prefetch.h>
58 #include <linux/profile.h>
59 #include <linux/psi.h>
60 #include <linux/rcupdate_wait.h>
61 #include <linux/security.h>
62 #include <linux/stop_machine.h>
63 #include <linux/suspend.h>
64 #include <linux/swait.h>
65 #include <linux/syscalls.h>
66 #include <linux/task_work.h>
67 #include <linux/tsacct_kern.h>
68
69 #include <asm/tlb.h>
70
71 #ifdef CONFIG_PARAVIRT
72 # include <asm/paravirt.h>
73 #endif
74
75 #include "cpupri.h"
76 #include "cpudeadline.h"
77
78 #ifdef CONFIG_SCHED_DEBUG
79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
80 #else
81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
82 #endif
83
84 struct rq;
85 struct cpuidle_state;
86
87 /* task_struct::on_rq states: */
88 #define TASK_ON_RQ_QUEUED 1
89 #define TASK_ON_RQ_MIGRATING 2
90
91 extern __read_mostly int scheduler_running;
92
93 extern unsigned long calc_load_update;
94 extern atomic_long_t calc_load_tasks;
95
96 extern void calc_global_load_tick(struct rq *this_rq);
97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
98
99 /*
100 * Helpers for converting nanosecond timing to jiffy resolution
101 */
102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
103
104 /*
105 * Increase resolution of nice-level calculations for 64-bit architectures.
106 * The extra resolution improves shares distribution and load balancing of
107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108 * hierarchies, especially on larger systems. This is not a user-visible change
109 * and does not change the user-interface for setting shares/weights.
110 *
111 * We increase resolution only if we have enough bits to allow this increased
112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113 * are pretty high and the returns do not justify the increased costs.
114 *
115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116 * increase coverage and consistency always enable it on 64-bit platforms.
117 */
118 #ifdef CONFIG_64BIT
119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w) \
122 ({ \
123 unsigned long __w = (w); \
124 if (__w) \
125 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
126 __w; \
127 })
128 #else
129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
130 # define scale_load(w) (w)
131 # define scale_load_down(w) (w)
132 #endif
133
134 /*
135 * Task weight (visible to users) and its load (invisible to users) have
136 * independent resolution, but they should be well calibrated. We use
137 * scale_load() and scale_load_down(w) to convert between them. The
138 * following must be true:
139 *
140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
141 *
142 */
143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
144
145 /*
146 * Single value that decides SCHED_DEADLINE internal math precision.
147 * 10 -> just above 1us
148 * 9 -> just above 0.5us
149 */
150 #define DL_SCALE 10
151
152 /*
153 * Single value that denotes runtime == period, ie unlimited time.
154 */
155 #define RUNTIME_INF ((u64)~0ULL)
156
idle_policy(int policy)157 static inline int idle_policy(int policy)
158 {
159 return policy == SCHED_IDLE;
160 }
fair_policy(int policy)161 static inline int fair_policy(int policy)
162 {
163 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
164 }
165
rt_policy(int policy)166 static inline int rt_policy(int policy)
167 {
168 return policy == SCHED_FIFO || policy == SCHED_RR;
169 }
170
dl_policy(int policy)171 static inline int dl_policy(int policy)
172 {
173 return policy == SCHED_DEADLINE;
174 }
valid_policy(int policy)175 static inline bool valid_policy(int policy)
176 {
177 return idle_policy(policy) || fair_policy(policy) ||
178 rt_policy(policy) || dl_policy(policy);
179 }
180
task_has_idle_policy(struct task_struct * p)181 static inline int task_has_idle_policy(struct task_struct *p)
182 {
183 return idle_policy(p->policy);
184 }
185
task_has_rt_policy(struct task_struct * p)186 static inline int task_has_rt_policy(struct task_struct *p)
187 {
188 return rt_policy(p->policy);
189 }
190
task_has_dl_policy(struct task_struct * p)191 static inline int task_has_dl_policy(struct task_struct *p)
192 {
193 return dl_policy(p->policy);
194 }
195
196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
197
198 /*
199 * !! For sched_setattr_nocheck() (kernel) only !!
200 *
201 * This is actually gross. :(
202 *
203 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
204 * tasks, but still be able to sleep. We need this on platforms that cannot
205 * atomically change clock frequency. Remove once fast switching will be
206 * available on such platforms.
207 *
208 * SUGOV stands for SchedUtil GOVernor.
209 */
210 #define SCHED_FLAG_SUGOV 0x10000000
211
212 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
213
dl_entity_is_special(struct sched_dl_entity * dl_se)214 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
215 {
216 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
217 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
218 #else
219 return false;
220 #endif
221 }
222
223 /*
224 * Tells if entity @a should preempt entity @b.
225 */
226 static inline bool
dl_entity_preempt(struct sched_dl_entity * a,struct sched_dl_entity * b)227 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
228 {
229 return dl_entity_is_special(a) ||
230 dl_time_before(a->deadline, b->deadline);
231 }
232
233 /*
234 * This is the priority-queue data structure of the RT scheduling class:
235 */
236 struct rt_prio_array {
237 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
238 struct list_head queue[MAX_RT_PRIO];
239 };
240
241 struct rt_bandwidth {
242 /* nests inside the rq lock: */
243 raw_spinlock_t rt_runtime_lock;
244 ktime_t rt_period;
245 u64 rt_runtime;
246 struct hrtimer rt_period_timer;
247 unsigned int rt_period_active;
248 };
249
250 void __dl_clear_params(struct task_struct *p);
251
252 struct dl_bandwidth {
253 raw_spinlock_t dl_runtime_lock;
254 u64 dl_runtime;
255 u64 dl_period;
256 };
257
dl_bandwidth_enabled(void)258 static inline int dl_bandwidth_enabled(void)
259 {
260 return sysctl_sched_rt_runtime >= 0;
261 }
262
263 /*
264 * To keep the bandwidth of -deadline tasks under control
265 * we need some place where:
266 * - store the maximum -deadline bandwidth of each cpu;
267 * - cache the fraction of bandwidth that is currently allocated in
268 * each root domain;
269 *
270 * This is all done in the data structure below. It is similar to the
271 * one used for RT-throttling (rt_bandwidth), with the main difference
272 * that, since here we are only interested in admission control, we
273 * do not decrease any runtime while the group "executes", neither we
274 * need a timer to replenish it.
275 *
276 * With respect to SMP, bandwidth is given on a per root domain basis,
277 * meaning that:
278 * - bw (< 100%) is the deadline bandwidth of each CPU;
279 * - total_bw is the currently allocated bandwidth in each root domain;
280 */
281 struct dl_bw {
282 raw_spinlock_t lock;
283 u64 bw;
284 u64 total_bw;
285 };
286
287 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
288
289 static inline
__dl_sub(struct dl_bw * dl_b,u64 tsk_bw,int cpus)290 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
291 {
292 dl_b->total_bw -= tsk_bw;
293 __dl_update(dl_b, (s32)tsk_bw / cpus);
294 }
295
296 static inline
__dl_add(struct dl_bw * dl_b,u64 tsk_bw,int cpus)297 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
298 {
299 dl_b->total_bw += tsk_bw;
300 __dl_update(dl_b, -((s32)tsk_bw / cpus));
301 }
302
303 static inline
__dl_overflow(struct dl_bw * dl_b,int cpus,u64 old_bw,u64 new_bw)304 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
305 {
306 return dl_b->bw != -1 &&
307 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
308 }
309
310 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
311 extern void init_dl_bw(struct dl_bw *dl_b);
312 extern int sched_dl_global_validate(void);
313 extern void sched_dl_do_global(void);
314 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
315 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
316 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
317 extern bool __checkparam_dl(const struct sched_attr *attr);
318 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
319 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
320 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
321 extern bool dl_cpu_busy(unsigned int cpu);
322
323 #ifdef CONFIG_CGROUP_SCHED
324
325 #include <linux/cgroup.h>
326 #include <linux/psi.h>
327
328 struct cfs_rq;
329 struct rt_rq;
330
331 extern struct list_head task_groups;
332
333 struct cfs_bandwidth {
334 #ifdef CONFIG_CFS_BANDWIDTH
335 raw_spinlock_t lock;
336 ktime_t period;
337 u64 quota;
338 u64 runtime;
339 s64 hierarchical_quota;
340
341 u8 idle;
342 u8 period_active;
343 u8 distribute_running;
344 u8 slack_started;
345 struct hrtimer period_timer;
346 struct hrtimer slack_timer;
347 struct list_head throttled_cfs_rq;
348
349 /* Statistics: */
350 int nr_periods;
351 int nr_throttled;
352 u64 throttled_time;
353 #endif
354 };
355
356 /* Task group related information */
357 struct task_group {
358 struct cgroup_subsys_state css;
359
360 #ifdef CONFIG_FAIR_GROUP_SCHED
361 /* schedulable entities of this group on each CPU */
362 struct sched_entity **se;
363 /* runqueue "owned" by this group on each CPU */
364 struct cfs_rq **cfs_rq;
365 unsigned long shares;
366
367 #ifdef CONFIG_SMP
368 /*
369 * load_avg can be heavily contended at clock tick time, so put
370 * it in its own cacheline separated from the fields above which
371 * will also be accessed at each tick.
372 */
373 atomic_long_t load_avg ____cacheline_aligned;
374 #endif
375 #endif
376
377 #ifdef CONFIG_RT_GROUP_SCHED
378 struct sched_rt_entity **rt_se;
379 struct rt_rq **rt_rq;
380
381 struct rt_bandwidth rt_bandwidth;
382 #endif
383
384 struct rcu_head rcu;
385 struct list_head list;
386
387 struct task_group *parent;
388 struct list_head siblings;
389 struct list_head children;
390
391 #ifdef CONFIG_SCHED_AUTOGROUP
392 struct autogroup *autogroup;
393 #endif
394
395 struct cfs_bandwidth cfs_bandwidth;
396
397 #ifdef CONFIG_UCLAMP_TASK_GROUP
398 /* The two decimal precision [%] value requested from user-space */
399 unsigned int uclamp_pct[UCLAMP_CNT];
400 /* Clamp values requested for a task group */
401 struct uclamp_se uclamp_req[UCLAMP_CNT];
402 /* Effective clamp values used for a task group */
403 struct uclamp_se uclamp[UCLAMP_CNT];
404 /* Latency-sensitive flag used for a task group */
405 unsigned int latency_sensitive;
406 #endif
407
408 };
409
410 #ifdef CONFIG_FAIR_GROUP_SCHED
411 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
412
413 /*
414 * A weight of 0 or 1 can cause arithmetics problems.
415 * A weight of a cfs_rq is the sum of weights of which entities
416 * are queued on this cfs_rq, so a weight of a entity should not be
417 * too large, so as the shares value of a task group.
418 * (The default weight is 1024 - so there's no practical
419 * limitation from this.)
420 */
421 #define MIN_SHARES (1UL << 1)
422 #define MAX_SHARES (1UL << 18)
423 #endif
424
425 typedef int (*tg_visitor)(struct task_group *, void *);
426
427 extern int walk_tg_tree_from(struct task_group *from,
428 tg_visitor down, tg_visitor up, void *data);
429
430 /*
431 * Iterate the full tree, calling @down when first entering a node and @up when
432 * leaving it for the final time.
433 *
434 * Caller must hold rcu_lock or sufficient equivalent.
435 */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)436 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
437 {
438 return walk_tg_tree_from(&root_task_group, down, up, data);
439 }
440
441 extern int tg_nop(struct task_group *tg, void *data);
442
443 extern void free_fair_sched_group(struct task_group *tg);
444 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
445 extern void online_fair_sched_group(struct task_group *tg);
446 extern void unregister_fair_sched_group(struct task_group *tg);
447 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
448 struct sched_entity *se, int cpu,
449 struct sched_entity *parent);
450 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
451
452 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
453 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
454 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
455
456 extern void free_rt_sched_group(struct task_group *tg);
457 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
458 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
459 struct sched_rt_entity *rt_se, int cpu,
460 struct sched_rt_entity *parent);
461 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
462 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
463 extern long sched_group_rt_runtime(struct task_group *tg);
464 extern long sched_group_rt_period(struct task_group *tg);
465 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
466
467 extern struct task_group *sched_create_group(struct task_group *parent);
468 extern void sched_online_group(struct task_group *tg,
469 struct task_group *parent);
470 extern void sched_destroy_group(struct task_group *tg);
471 extern void sched_offline_group(struct task_group *tg);
472
473 extern void sched_move_task(struct task_struct *tsk);
474
475 #ifdef CONFIG_FAIR_GROUP_SCHED
476 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
477
478 #ifdef CONFIG_SMP
479 extern void set_task_rq_fair(struct sched_entity *se,
480 struct cfs_rq *prev, struct cfs_rq *next);
481 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)482 static inline void set_task_rq_fair(struct sched_entity *se,
483 struct cfs_rq *prev, struct cfs_rq *next) { }
484 #endif /* CONFIG_SMP */
485 #endif /* CONFIG_FAIR_GROUP_SCHED */
486
487 #else /* CONFIG_CGROUP_SCHED */
488
489 struct cfs_bandwidth { };
490
491 #endif /* CONFIG_CGROUP_SCHED */
492
493 /* CFS-related fields in a runqueue */
494 struct cfs_rq {
495 struct load_weight load;
496 unsigned long runnable_weight;
497 unsigned int nr_running;
498 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
499 unsigned int idle_h_nr_running; /* SCHED_IDLE */
500
501 u64 exec_clock;
502 u64 min_vruntime;
503 #ifndef CONFIG_64BIT
504 u64 min_vruntime_copy;
505 #endif
506
507 struct rb_root_cached tasks_timeline;
508
509 /*
510 * 'curr' points to currently running entity on this cfs_rq.
511 * It is set to NULL otherwise (i.e when none are currently running).
512 */
513 struct sched_entity *curr;
514 struct sched_entity *next;
515 struct sched_entity *last;
516 struct sched_entity *skip;
517
518 #ifdef CONFIG_SCHED_DEBUG
519 unsigned int nr_spread_over;
520 #endif
521
522 #ifdef CONFIG_SMP
523 /*
524 * CFS load tracking
525 */
526 struct sched_avg avg;
527 #ifndef CONFIG_64BIT
528 u64 load_last_update_time_copy;
529 #endif
530 struct {
531 raw_spinlock_t lock ____cacheline_aligned;
532 int nr;
533 unsigned long load_avg;
534 unsigned long util_avg;
535 unsigned long runnable_sum;
536 } removed;
537
538 #ifdef CONFIG_FAIR_GROUP_SCHED
539 unsigned long tg_load_avg_contrib;
540 long propagate;
541 long prop_runnable_sum;
542
543 /*
544 * h_load = weight * f(tg)
545 *
546 * Where f(tg) is the recursive weight fraction assigned to
547 * this group.
548 */
549 unsigned long h_load;
550 u64 last_h_load_update;
551 struct sched_entity *h_load_next;
552 #endif /* CONFIG_FAIR_GROUP_SCHED */
553 #endif /* CONFIG_SMP */
554
555 #ifdef CONFIG_FAIR_GROUP_SCHED
556 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
557
558 /*
559 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
560 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
561 * (like users, containers etc.)
562 *
563 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
564 * This list is used during load balance.
565 */
566 int on_list;
567 struct list_head leaf_cfs_rq_list;
568 struct task_group *tg; /* group that "owns" this runqueue */
569
570 #ifdef CONFIG_CFS_BANDWIDTH
571 int runtime_enabled;
572 s64 runtime_remaining;
573
574 u64 throttled_clock;
575 u64 throttled_clock_pelt;
576 u64 throttled_clock_pelt_time;
577 int throttled;
578 int throttle_count;
579 struct list_head throttled_list;
580 #endif /* CONFIG_CFS_BANDWIDTH */
581 #endif /* CONFIG_FAIR_GROUP_SCHED */
582 };
583
rt_bandwidth_enabled(void)584 static inline int rt_bandwidth_enabled(void)
585 {
586 return sysctl_sched_rt_runtime >= 0;
587 }
588
589 /* RT IPI pull logic requires IRQ_WORK */
590 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
591 # define HAVE_RT_PUSH_IPI
592 #endif
593
594 /* Real-Time classes' related field in a runqueue: */
595 struct rt_rq {
596 struct rt_prio_array active;
597 unsigned int rt_nr_running;
598 unsigned int rr_nr_running;
599 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
600 struct {
601 int curr; /* highest queued rt task prio */
602 #ifdef CONFIG_SMP
603 int next; /* next highest */
604 #endif
605 } highest_prio;
606 #endif
607 #ifdef CONFIG_SMP
608 unsigned long rt_nr_migratory;
609 unsigned long rt_nr_total;
610 int overloaded;
611 struct plist_head pushable_tasks;
612
613 #endif /* CONFIG_SMP */
614 int rt_queued;
615
616 int rt_throttled;
617 u64 rt_time;
618 u64 rt_runtime;
619 /* Nests inside the rq lock: */
620 raw_spinlock_t rt_runtime_lock;
621
622 #ifdef CONFIG_RT_GROUP_SCHED
623 unsigned long rt_nr_boosted;
624
625 struct rq *rq;
626 struct task_group *tg;
627 #endif
628 };
629
rt_rq_is_runnable(struct rt_rq * rt_rq)630 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
631 {
632 return rt_rq->rt_queued && rt_rq->rt_nr_running;
633 }
634
635 /* Deadline class' related fields in a runqueue */
636 struct dl_rq {
637 /* runqueue is an rbtree, ordered by deadline */
638 struct rb_root_cached root;
639
640 unsigned long dl_nr_running;
641
642 #ifdef CONFIG_SMP
643 /*
644 * Deadline values of the currently executing and the
645 * earliest ready task on this rq. Caching these facilitates
646 * the decision whether or not a ready but not running task
647 * should migrate somewhere else.
648 */
649 struct {
650 u64 curr;
651 u64 next;
652 } earliest_dl;
653
654 unsigned long dl_nr_migratory;
655 int overloaded;
656
657 /*
658 * Tasks on this rq that can be pushed away. They are kept in
659 * an rb-tree, ordered by tasks' deadlines, with caching
660 * of the leftmost (earliest deadline) element.
661 */
662 struct rb_root_cached pushable_dl_tasks_root;
663 #else
664 struct dl_bw dl_bw;
665 #endif
666 /*
667 * "Active utilization" for this runqueue: increased when a
668 * task wakes up (becomes TASK_RUNNING) and decreased when a
669 * task blocks
670 */
671 u64 running_bw;
672
673 /*
674 * Utilization of the tasks "assigned" to this runqueue (including
675 * the tasks that are in runqueue and the tasks that executed on this
676 * CPU and blocked). Increased when a task moves to this runqueue, and
677 * decreased when the task moves away (migrates, changes scheduling
678 * policy, or terminates).
679 * This is needed to compute the "inactive utilization" for the
680 * runqueue (inactive utilization = this_bw - running_bw).
681 */
682 u64 this_bw;
683 u64 extra_bw;
684
685 /*
686 * Inverse of the fraction of CPU utilization that can be reclaimed
687 * by the GRUB algorithm.
688 */
689 u64 bw_ratio;
690 };
691
692 #ifdef CONFIG_FAIR_GROUP_SCHED
693 /* An entity is a task if it doesn't "own" a runqueue */
694 #define entity_is_task(se) (!se->my_q)
695 #else
696 #define entity_is_task(se) 1
697 #endif
698
699 #ifdef CONFIG_SMP
700 /*
701 * XXX we want to get rid of these helpers and use the full load resolution.
702 */
se_weight(struct sched_entity * se)703 static inline long se_weight(struct sched_entity *se)
704 {
705 return scale_load_down(se->load.weight);
706 }
707
se_runnable(struct sched_entity * se)708 static inline long se_runnable(struct sched_entity *se)
709 {
710 return scale_load_down(se->runnable_weight);
711 }
712
sched_asym_prefer(int a,int b)713 static inline bool sched_asym_prefer(int a, int b)
714 {
715 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
716 }
717
718 struct perf_domain {
719 struct em_perf_domain *em_pd;
720 struct perf_domain *next;
721 struct rcu_head rcu;
722 };
723
724 struct max_cpu_capacity {
725 raw_spinlock_t lock;
726 unsigned long val;
727 int cpu;
728 };
729
730 /* Scheduling group status flags */
731 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
732 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
733
734 /*
735 * We add the notion of a root-domain which will be used to define per-domain
736 * variables. Each exclusive cpuset essentially defines an island domain by
737 * fully partitioning the member CPUs from any other cpuset. Whenever a new
738 * exclusive cpuset is created, we also create and attach a new root-domain
739 * object.
740 *
741 */
742 struct root_domain {
743 atomic_t refcount;
744 atomic_t rto_count;
745 struct rcu_head rcu;
746 cpumask_var_t span;
747 cpumask_var_t online;
748
749 /*
750 * Indicate pullable load on at least one CPU, e.g:
751 * - More than one runnable task
752 * - Running task is misfit
753 */
754 int overload;
755
756 /* Indicate one or more cpus over-utilized (tipping point) */
757 int overutilized;
758
759 /*
760 * The bit corresponding to a CPU gets set here if such CPU has more
761 * than one runnable -deadline task (as it is below for RT tasks).
762 */
763 cpumask_var_t dlo_mask;
764 atomic_t dlo_count;
765 struct dl_bw dl_bw;
766 struct cpudl cpudl;
767
768 #ifdef HAVE_RT_PUSH_IPI
769 /*
770 * For IPI pull requests, loop across the rto_mask.
771 */
772 struct irq_work rto_push_work;
773 raw_spinlock_t rto_lock;
774 /* These are only updated and read within rto_lock */
775 int rto_loop;
776 int rto_cpu;
777 /* These atomics are updated outside of a lock */
778 atomic_t rto_loop_next;
779 atomic_t rto_loop_start;
780 #endif
781 /*
782 * The "RT overload" flag: it gets set if a CPU has more than
783 * one runnable RT task.
784 */
785 cpumask_var_t rto_mask;
786 struct cpupri cpupri;
787
788 /* Maximum cpu capacity in the system. */
789 struct max_cpu_capacity max_cpu_capacity;
790
791 /*
792 * NULL-terminated list of performance domains intersecting with the
793 * CPUs of the rd. Protected by RCU.
794 */
795 struct perf_domain __rcu *pd;
796 };
797
798 extern void init_defrootdomain(void);
799 extern void init_max_cpu_capacity(struct max_cpu_capacity *mcc);
800 extern int sched_init_domains(const struct cpumask *cpu_map);
801 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
802 extern void sched_get_rd(struct root_domain *rd);
803 extern void sched_put_rd(struct root_domain *rd);
804
805 #ifdef HAVE_RT_PUSH_IPI
806 extern void rto_push_irq_work_func(struct irq_work *work);
807 #endif
808 #endif /* CONFIG_SMP */
809
810 #ifdef CONFIG_UCLAMP_TASK
811 /*
812 * struct uclamp_bucket - Utilization clamp bucket
813 * @value: utilization clamp value for tasks on this clamp bucket
814 * @tasks: number of RUNNABLE tasks on this clamp bucket
815 *
816 * Keep track of how many tasks are RUNNABLE for a given utilization
817 * clamp value.
818 */
819 struct uclamp_bucket {
820 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
821 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
822 };
823
824 /*
825 * struct uclamp_rq - rq's utilization clamp
826 * @value: currently active clamp values for a rq
827 * @bucket: utilization clamp buckets affecting a rq
828 *
829 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
830 * A clamp value is affecting a rq when there is at least one task RUNNABLE
831 * (or actually running) with that value.
832 *
833 * There are up to UCLAMP_CNT possible different clamp values, currently there
834 * are only two: minimum utilization and maximum utilization.
835 *
836 * All utilization clamping values are MAX aggregated, since:
837 * - for util_min: we want to run the CPU at least at the max of the minimum
838 * utilization required by its currently RUNNABLE tasks.
839 * - for util_max: we want to allow the CPU to run up to the max of the
840 * maximum utilization allowed by its currently RUNNABLE tasks.
841 *
842 * Since on each system we expect only a limited number of different
843 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
844 * the metrics required to compute all the per-rq utilization clamp values.
845 */
846 struct uclamp_rq {
847 unsigned int value;
848 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
849 };
850
851 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
852 #endif /* CONFIG_UCLAMP_TASK */
853
854 /*
855 * This is the main, per-CPU runqueue data structure.
856 *
857 * Locking rule: those places that want to lock multiple runqueues
858 * (such as the load balancing or the thread migration code), lock
859 * acquire operations must be ordered by ascending &runqueue.
860 */
861 struct rq {
862 /* runqueue lock: */
863 raw_spinlock_t lock;
864
865 /*
866 * nr_running and cpu_load should be in the same cacheline because
867 * remote CPUs use both these fields when doing load calculation.
868 */
869 unsigned int nr_running;
870 #ifdef CONFIG_NUMA_BALANCING
871 unsigned int nr_numa_running;
872 unsigned int nr_preferred_running;
873 unsigned int numa_migrate_on;
874 #endif
875 #ifdef CONFIG_NO_HZ_COMMON
876 #ifdef CONFIG_SMP
877 unsigned long last_load_update_tick;
878 unsigned long last_blocked_load_update_tick;
879 unsigned int has_blocked_load;
880 #endif /* CONFIG_SMP */
881 unsigned int nohz_tick_stopped;
882 atomic_t nohz_flags;
883 #endif /* CONFIG_NO_HZ_COMMON */
884
885 unsigned long nr_load_updates;
886 u64 nr_switches;
887
888 #ifdef CONFIG_UCLAMP_TASK
889 /* Utilization clamp values based on CPU's RUNNABLE tasks */
890 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
891 unsigned int uclamp_flags;
892 #define UCLAMP_FLAG_IDLE 0x01
893 #endif
894
895 struct cfs_rq cfs;
896 struct rt_rq rt;
897 struct dl_rq dl;
898
899 #ifdef CONFIG_FAIR_GROUP_SCHED
900 /* list of leaf cfs_rq on this CPU: */
901 struct list_head leaf_cfs_rq_list;
902 struct list_head *tmp_alone_branch;
903 #endif /* CONFIG_FAIR_GROUP_SCHED */
904
905 /*
906 * This is part of a global counter where only the total sum
907 * over all CPUs matters. A task can increase this counter on
908 * one CPU and if it got migrated afterwards it may decrease
909 * it on another CPU. Always updated under the runqueue lock:
910 */
911 unsigned long nr_uninterruptible;
912
913 struct task_struct *curr;
914 struct task_struct *idle;
915 struct task_struct *stop;
916 unsigned long next_balance;
917 struct mm_struct *prev_mm;
918
919 unsigned int clock_update_flags;
920 u64 clock;
921 /* Ensure that all clocks are in the same cache line */
922 u64 clock_task ____cacheline_aligned;
923 u64 clock_pelt;
924 unsigned long lost_idle_time;
925
926 atomic_t nr_iowait;
927
928 #ifdef CONFIG_MEMBARRIER
929 int membarrier_state;
930 #endif
931
932 #ifdef CONFIG_SMP
933 struct root_domain *rd;
934 struct sched_domain __rcu *sd;
935
936 unsigned long cpu_capacity;
937 unsigned long cpu_capacity_orig;
938
939 struct callback_head *balance_callback;
940
941 unsigned char idle_balance;
942
943 unsigned long misfit_task_load;
944
945 /* For active balancing */
946 int active_balance;
947 int push_cpu;
948 struct cpu_stop_work active_balance_work;
949
950 /* CPU of this runqueue: */
951 int cpu;
952 int online;
953
954 struct list_head cfs_tasks;
955
956 struct sched_avg avg_rt;
957 struct sched_avg avg_dl;
958 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
959 struct sched_avg avg_irq;
960 #endif
961 u64 idle_stamp;
962 u64 avg_idle;
963
964 /* This is used to determine avg_idle's max value */
965 u64 max_idle_balance_cost;
966 #endif
967
968 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
969 u64 prev_irq_time;
970 #endif
971 #ifdef CONFIG_PARAVIRT
972 u64 prev_steal_time;
973 #endif
974 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
975 u64 prev_steal_time_rq;
976 #endif
977
978 /* calc_load related fields */
979 unsigned long calc_load_update;
980 long calc_load_active;
981
982 #ifdef CONFIG_SCHED_HRTICK
983 #ifdef CONFIG_SMP
984 int hrtick_csd_pending;
985 call_single_data_t hrtick_csd;
986 #endif
987 struct hrtimer hrtick_timer;
988 #endif
989
990 #ifdef CONFIG_SCHEDSTATS
991 /* latency stats */
992 struct sched_info rq_sched_info;
993 unsigned long long rq_cpu_time;
994 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
995
996 /* sys_sched_yield() stats */
997 unsigned int yld_count;
998
999 /* schedule() stats */
1000 unsigned int sched_count;
1001 unsigned int sched_goidle;
1002
1003 /* try_to_wake_up() stats */
1004 unsigned int ttwu_count;
1005 unsigned int ttwu_local;
1006 #endif
1007
1008 #ifdef CONFIG_SMP
1009 struct llist_head wake_list;
1010 #endif
1011
1012 #ifdef CONFIG_CPU_IDLE
1013 /* Must be inspected within a rcu lock section */
1014 struct cpuidle_state *idle_state;
1015 #endif
1016 };
1017
1018 #ifdef CONFIG_FAIR_GROUP_SCHED
1019
1020 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1021 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1022 {
1023 return cfs_rq->rq;
1024 }
1025
1026 #else
1027
rq_of(struct cfs_rq * cfs_rq)1028 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1029 {
1030 return container_of(cfs_rq, struct rq, cfs);
1031 }
1032 #endif
1033
cpu_of(struct rq * rq)1034 static inline int cpu_of(struct rq *rq)
1035 {
1036 #ifdef CONFIG_SMP
1037 return rq->cpu;
1038 #else
1039 return 0;
1040 #endif
1041 }
1042
1043
1044 #ifdef CONFIG_SCHED_SMT
1045 extern void __update_idle_core(struct rq *rq);
1046
update_idle_core(struct rq * rq)1047 static inline void update_idle_core(struct rq *rq)
1048 {
1049 if (static_branch_unlikely(&sched_smt_present))
1050 __update_idle_core(rq);
1051 }
1052
1053 #else
update_idle_core(struct rq * rq)1054 static inline void update_idle_core(struct rq *rq) { }
1055 #endif
1056
1057 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1058
1059 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1060 #define this_rq() this_cpu_ptr(&runqueues)
1061 #define task_rq(p) cpu_rq(task_cpu(p))
1062 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1063 #define raw_rq() raw_cpu_ptr(&runqueues)
1064
1065 extern void update_rq_clock(struct rq *rq);
1066
__rq_clock_broken(struct rq * rq)1067 static inline u64 __rq_clock_broken(struct rq *rq)
1068 {
1069 return READ_ONCE(rq->clock);
1070 }
1071
1072 /*
1073 * rq::clock_update_flags bits
1074 *
1075 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1076 * call to __schedule(). This is an optimisation to avoid
1077 * neighbouring rq clock updates.
1078 *
1079 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1080 * in effect and calls to update_rq_clock() are being ignored.
1081 *
1082 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1083 * made to update_rq_clock() since the last time rq::lock was pinned.
1084 *
1085 * If inside of __schedule(), clock_update_flags will have been
1086 * shifted left (a left shift is a cheap operation for the fast path
1087 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1088 *
1089 * if (rq-clock_update_flags >= RQCF_UPDATED)
1090 *
1091 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1092 * one position though, because the next rq_unpin_lock() will shift it
1093 * back.
1094 */
1095 #define RQCF_REQ_SKIP 0x01
1096 #define RQCF_ACT_SKIP 0x02
1097 #define RQCF_UPDATED 0x04
1098
assert_clock_updated(struct rq * rq)1099 static inline void assert_clock_updated(struct rq *rq)
1100 {
1101 /*
1102 * The only reason for not seeing a clock update since the
1103 * last rq_pin_lock() is if we're currently skipping updates.
1104 */
1105 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1106 }
1107
rq_clock(struct rq * rq)1108 static inline u64 rq_clock(struct rq *rq)
1109 {
1110 lockdep_assert_held(&rq->lock);
1111 assert_clock_updated(rq);
1112
1113 return rq->clock;
1114 }
1115
rq_clock_task(struct rq * rq)1116 static inline u64 rq_clock_task(struct rq *rq)
1117 {
1118 lockdep_assert_held(&rq->lock);
1119 assert_clock_updated(rq);
1120
1121 return rq->clock_task;
1122 }
1123
rq_clock_skip_update(struct rq * rq)1124 static inline void rq_clock_skip_update(struct rq *rq)
1125 {
1126 lockdep_assert_held(&rq->lock);
1127 rq->clock_update_flags |= RQCF_REQ_SKIP;
1128 }
1129
1130 /*
1131 * See rt task throttling, which is the only time a skip
1132 * request is cancelled.
1133 */
rq_clock_cancel_skipupdate(struct rq * rq)1134 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1135 {
1136 lockdep_assert_held(&rq->lock);
1137 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1138 }
1139
1140 struct rq_flags {
1141 unsigned long flags;
1142 struct pin_cookie cookie;
1143 #ifdef CONFIG_SCHED_DEBUG
1144 /*
1145 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1146 * current pin context is stashed here in case it needs to be
1147 * restored in rq_repin_lock().
1148 */
1149 unsigned int clock_update_flags;
1150 #endif
1151 };
1152
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1153 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1154 {
1155 rf->cookie = lockdep_pin_lock(&rq->lock);
1156
1157 #ifdef CONFIG_SCHED_DEBUG
1158 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1159 rf->clock_update_flags = 0;
1160 #endif
1161 }
1162
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1163 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1164 {
1165 #ifdef CONFIG_SCHED_DEBUG
1166 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1167 rf->clock_update_flags = RQCF_UPDATED;
1168 #endif
1169
1170 lockdep_unpin_lock(&rq->lock, rf->cookie);
1171 }
1172
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1173 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1174 {
1175 lockdep_repin_lock(&rq->lock, rf->cookie);
1176
1177 #ifdef CONFIG_SCHED_DEBUG
1178 /*
1179 * Restore the value we stashed in @rf for this pin context.
1180 */
1181 rq->clock_update_flags |= rf->clock_update_flags;
1182 #endif
1183 }
1184
1185 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1186 __acquires(rq->lock);
1187
1188 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1189 __acquires(p->pi_lock)
1190 __acquires(rq->lock);
1191
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1192 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1193 __releases(rq->lock)
1194 {
1195 rq_unpin_lock(rq, rf);
1196 raw_spin_unlock(&rq->lock);
1197 }
1198
1199 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1200 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1201 __releases(rq->lock)
1202 __releases(p->pi_lock)
1203 {
1204 rq_unpin_lock(rq, rf);
1205 raw_spin_unlock(&rq->lock);
1206 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1207 }
1208
1209 static inline void
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1210 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1211 __acquires(rq->lock)
1212 {
1213 raw_spin_lock_irqsave(&rq->lock, rf->flags);
1214 rq_pin_lock(rq, rf);
1215 }
1216
1217 static inline void
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1218 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1219 __acquires(rq->lock)
1220 {
1221 raw_spin_lock_irq(&rq->lock);
1222 rq_pin_lock(rq, rf);
1223 }
1224
1225 static inline void
rq_lock(struct rq * rq,struct rq_flags * rf)1226 rq_lock(struct rq *rq, struct rq_flags *rf)
1227 __acquires(rq->lock)
1228 {
1229 raw_spin_lock(&rq->lock);
1230 rq_pin_lock(rq, rf);
1231 }
1232
1233 static inline void
rq_relock(struct rq * rq,struct rq_flags * rf)1234 rq_relock(struct rq *rq, struct rq_flags *rf)
1235 __acquires(rq->lock)
1236 {
1237 raw_spin_lock(&rq->lock);
1238 rq_repin_lock(rq, rf);
1239 }
1240
1241 static inline void
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1242 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1243 __releases(rq->lock)
1244 {
1245 rq_unpin_lock(rq, rf);
1246 raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1247 }
1248
1249 static inline void
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1250 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1251 __releases(rq->lock)
1252 {
1253 rq_unpin_lock(rq, rf);
1254 raw_spin_unlock_irq(&rq->lock);
1255 }
1256
1257 static inline void
rq_unlock(struct rq * rq,struct rq_flags * rf)1258 rq_unlock(struct rq *rq, struct rq_flags *rf)
1259 __releases(rq->lock)
1260 {
1261 rq_unpin_lock(rq, rf);
1262 raw_spin_unlock(&rq->lock);
1263 }
1264
1265 static inline struct rq *
this_rq_lock_irq(struct rq_flags * rf)1266 this_rq_lock_irq(struct rq_flags *rf)
1267 __acquires(rq->lock)
1268 {
1269 struct rq *rq;
1270
1271 local_irq_disable();
1272 rq = this_rq();
1273 rq_lock(rq, rf);
1274 return rq;
1275 }
1276
1277 #ifdef CONFIG_NUMA
1278 enum numa_topology_type {
1279 NUMA_DIRECT,
1280 NUMA_GLUELESS_MESH,
1281 NUMA_BACKPLANE,
1282 };
1283 extern enum numa_topology_type sched_numa_topology_type;
1284 extern int sched_max_numa_distance;
1285 extern bool find_numa_distance(int distance);
1286 extern void sched_init_numa(void);
1287 extern void sched_domains_numa_masks_set(unsigned int cpu);
1288 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1289 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1290 #else
sched_init_numa(void)1291 static inline void sched_init_numa(void) { }
sched_domains_numa_masks_set(unsigned int cpu)1292 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1293 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1294 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1295 {
1296 return nr_cpu_ids;
1297 }
1298 #endif
1299
1300 #ifdef CONFIG_NUMA_BALANCING
1301 /* The regions in numa_faults array from task_struct */
1302 enum numa_faults_stats {
1303 NUMA_MEM = 0,
1304 NUMA_CPU,
1305 NUMA_MEMBUF,
1306 NUMA_CPUBUF
1307 };
1308 extern void sched_setnuma(struct task_struct *p, int node);
1309 extern int migrate_task_to(struct task_struct *p, int cpu);
1310 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1311 int cpu, int scpu);
1312 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1313 #else
1314 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1315 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1316 {
1317 }
1318 #endif /* CONFIG_NUMA_BALANCING */
1319
1320 #ifdef CONFIG_SMP
1321
1322 static inline void
queue_balance_callback(struct rq * rq,struct callback_head * head,void (* func)(struct rq * rq))1323 queue_balance_callback(struct rq *rq,
1324 struct callback_head *head,
1325 void (*func)(struct rq *rq))
1326 {
1327 lockdep_assert_held(&rq->lock);
1328
1329 if (unlikely(head->next))
1330 return;
1331
1332 head->func = (void (*)(struct callback_head *))func;
1333 head->next = rq->balance_callback;
1334 rq->balance_callback = head;
1335 }
1336
1337 extern void sched_ttwu_pending(void);
1338
1339 #define rcu_dereference_check_sched_domain(p) \
1340 rcu_dereference_check((p), \
1341 lockdep_is_held(&sched_domains_mutex))
1342
1343 /*
1344 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1345 * See destroy_sched_domains: call_rcu for details.
1346 *
1347 * The domain tree of any CPU may only be accessed from within
1348 * preempt-disabled sections.
1349 */
1350 #define for_each_domain(cpu, __sd) \
1351 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1352 __sd; __sd = __sd->parent)
1353
1354 /**
1355 * highest_flag_domain - Return highest sched_domain containing flag.
1356 * @cpu: The CPU whose highest level of sched domain is to
1357 * be returned.
1358 * @flag: The flag to check for the highest sched_domain
1359 * for the given CPU.
1360 *
1361 * Returns the highest sched_domain of a CPU which contains the given flag.
1362 */
highest_flag_domain(int cpu,int flag)1363 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1364 {
1365 struct sched_domain *sd, *hsd = NULL;
1366
1367 for_each_domain(cpu, sd) {
1368 if (!(sd->flags & flag))
1369 break;
1370 hsd = sd;
1371 }
1372
1373 return hsd;
1374 }
1375
lowest_flag_domain(int cpu,int flag)1376 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1377 {
1378 struct sched_domain *sd;
1379
1380 for_each_domain(cpu, sd) {
1381 if (sd->flags & flag)
1382 break;
1383 }
1384
1385 return sd;
1386 }
1387
1388 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1389 DECLARE_PER_CPU(int, sd_llc_size);
1390 DECLARE_PER_CPU(int, sd_llc_id);
1391 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1392 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1393 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1394 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1395 extern struct static_key_false sched_asym_cpucapacity;
1396
1397 struct sched_group_capacity {
1398 atomic_t ref;
1399 /*
1400 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1401 * for a single CPU.
1402 */
1403 unsigned long capacity;
1404 unsigned long min_capacity; /* Min per-CPU capacity in group */
1405 unsigned long max_capacity; /* Max per-CPU capacity in group */
1406 unsigned long next_update;
1407 int imbalance; /* XXX unrelated to capacity but shared group state */
1408
1409 #ifdef CONFIG_SCHED_DEBUG
1410 int id;
1411 #endif
1412
1413 unsigned long cpumask[0]; /* Balance mask */
1414 };
1415
1416 struct sched_group {
1417 struct sched_group *next; /* Must be a circular list */
1418 atomic_t ref;
1419
1420 unsigned int group_weight;
1421 struct sched_group_capacity *sgc;
1422 int asym_prefer_cpu; /* CPU of highest priority in group */
1423
1424 /*
1425 * The CPUs this group covers.
1426 *
1427 * NOTE: this field is variable length. (Allocated dynamically
1428 * by attaching extra space to the end of the structure,
1429 * depending on how many CPUs the kernel has booted up with)
1430 */
1431 unsigned long cpumask[0];
1432 };
1433
sched_group_span(struct sched_group * sg)1434 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1435 {
1436 return to_cpumask(sg->cpumask);
1437 }
1438
1439 /*
1440 * See build_balance_mask().
1441 */
group_balance_mask(struct sched_group * sg)1442 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1443 {
1444 return to_cpumask(sg->sgc->cpumask);
1445 }
1446
1447 /**
1448 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1449 * @group: The group whose first CPU is to be returned.
1450 */
group_first_cpu(struct sched_group * group)1451 static inline unsigned int group_first_cpu(struct sched_group *group)
1452 {
1453 return cpumask_first(sched_group_span(group));
1454 }
1455
1456 extern int group_balance_cpu(struct sched_group *sg);
1457
1458 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1459 void register_sched_domain_sysctl(void);
1460 void dirty_sched_domain_sysctl(int cpu);
1461 void unregister_sched_domain_sysctl(void);
1462 #else
register_sched_domain_sysctl(void)1463 static inline void register_sched_domain_sysctl(void)
1464 {
1465 }
dirty_sched_domain_sysctl(int cpu)1466 static inline void dirty_sched_domain_sysctl(int cpu)
1467 {
1468 }
unregister_sched_domain_sysctl(void)1469 static inline void unregister_sched_domain_sysctl(void)
1470 {
1471 }
1472 #endif
1473
1474 extern int newidle_balance(struct rq *this_rq, struct rq_flags *rf);
1475
1476 #else
1477
sched_ttwu_pending(void)1478 static inline void sched_ttwu_pending(void) { }
1479
newidle_balance(struct rq * this_rq,struct rq_flags * rf)1480 static inline int newidle_balance(struct rq *this_rq, struct rq_flags *rf) { return 0; }
1481
1482 #endif /* CONFIG_SMP */
1483
1484 #include "stats.h"
1485 #include "autogroup.h"
1486
1487 #ifdef CONFIG_CGROUP_SCHED
1488
1489 /*
1490 * Return the group to which this tasks belongs.
1491 *
1492 * We cannot use task_css() and friends because the cgroup subsystem
1493 * changes that value before the cgroup_subsys::attach() method is called,
1494 * therefore we cannot pin it and might observe the wrong value.
1495 *
1496 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1497 * core changes this before calling sched_move_task().
1498 *
1499 * Instead we use a 'copy' which is updated from sched_move_task() while
1500 * holding both task_struct::pi_lock and rq::lock.
1501 */
task_group(struct task_struct * p)1502 static inline struct task_group *task_group(struct task_struct *p)
1503 {
1504 return p->sched_task_group;
1505 }
1506
1507 /* 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)1508 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1509 {
1510 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1511 struct task_group *tg = task_group(p);
1512 #endif
1513
1514 #ifdef CONFIG_FAIR_GROUP_SCHED
1515 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1516 p->se.cfs_rq = tg->cfs_rq[cpu];
1517 p->se.parent = tg->se[cpu];
1518 #endif
1519
1520 #ifdef CONFIG_RT_GROUP_SCHED
1521 p->rt.rt_rq = tg->rt_rq[cpu];
1522 p->rt.parent = tg->rt_se[cpu];
1523 #endif
1524 }
1525
1526 #else /* CONFIG_CGROUP_SCHED */
1527
set_task_rq(struct task_struct * p,unsigned int cpu)1528 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
task_group(struct task_struct * p)1529 static inline struct task_group *task_group(struct task_struct *p)
1530 {
1531 return NULL;
1532 }
1533
1534 #endif /* CONFIG_CGROUP_SCHED */
1535
__set_task_cpu(struct task_struct * p,unsigned int cpu)1536 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1537 {
1538 set_task_rq(p, cpu);
1539 #ifdef CONFIG_SMP
1540 /*
1541 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1542 * successfully executed on another CPU. We must ensure that updates of
1543 * per-task data have been completed by this moment.
1544 */
1545 smp_wmb();
1546 #ifdef CONFIG_THREAD_INFO_IN_TASK
1547 WRITE_ONCE(p->cpu, cpu);
1548 #else
1549 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1550 #endif
1551 p->wake_cpu = cpu;
1552 #endif
1553 }
1554
1555 /*
1556 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1557 */
1558 #ifdef CONFIG_SCHED_DEBUG
1559 # include <linux/static_key.h>
1560 # define const_debug __read_mostly
1561 #else
1562 # define const_debug const
1563 #endif
1564
1565 #define SCHED_FEAT(name, enabled) \
1566 __SCHED_FEAT_##name ,
1567
1568 enum {
1569 #include "features.h"
1570 __SCHED_FEAT_NR,
1571 };
1572
1573 #undef SCHED_FEAT
1574
1575 #ifdef CONFIG_SCHED_DEBUG
1576
1577 /*
1578 * To support run-time toggling of sched features, all the translation units
1579 * (but core.c) reference the sysctl_sched_features defined in core.c.
1580 */
1581 extern const_debug unsigned int sysctl_sched_features;
1582
1583 #ifdef CONFIG_JUMP_LABEL
1584 #define SCHED_FEAT(name, enabled) \
1585 static __always_inline bool static_branch_##name(struct static_key *key) \
1586 { \
1587 return static_key_##enabled(key); \
1588 }
1589
1590 #include "features.h"
1591 #undef SCHED_FEAT
1592
1593 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1594 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1595
1596 #else /* !CONFIG_JUMP_LABEL */
1597
1598 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1599
1600 #endif /* CONFIG_JUMP_LABEL */
1601
1602 #else /* !SCHED_DEBUG */
1603
1604 /*
1605 * Each translation unit has its own copy of sysctl_sched_features to allow
1606 * constants propagation at compile time and compiler optimization based on
1607 * features default.
1608 */
1609 #define SCHED_FEAT(name, enabled) \
1610 (1UL << __SCHED_FEAT_##name) * enabled |
1611 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1612 #include "features.h"
1613 0;
1614 #undef SCHED_FEAT
1615
1616 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1617
1618 #endif /* SCHED_DEBUG */
1619
1620 extern struct static_key_false sched_numa_balancing;
1621 extern struct static_key_false sched_schedstats;
1622
global_rt_period(void)1623 static inline u64 global_rt_period(void)
1624 {
1625 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1626 }
1627
global_rt_runtime(void)1628 static inline u64 global_rt_runtime(void)
1629 {
1630 if (sysctl_sched_rt_runtime < 0)
1631 return RUNTIME_INF;
1632
1633 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1634 }
1635
task_current(struct rq * rq,struct task_struct * p)1636 static inline int task_current(struct rq *rq, struct task_struct *p)
1637 {
1638 return rq->curr == p;
1639 }
1640
task_running(struct rq * rq,struct task_struct * p)1641 static inline int task_running(struct rq *rq, struct task_struct *p)
1642 {
1643 #ifdef CONFIG_SMP
1644 return p->on_cpu;
1645 #else
1646 return task_current(rq, p);
1647 #endif
1648 }
1649
task_on_rq_queued(struct task_struct * p)1650 static inline int task_on_rq_queued(struct task_struct *p)
1651 {
1652 return p->on_rq == TASK_ON_RQ_QUEUED;
1653 }
1654
task_on_rq_migrating(struct task_struct * p)1655 static inline int task_on_rq_migrating(struct task_struct *p)
1656 {
1657 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1658 }
1659
1660 /*
1661 * wake flags
1662 */
1663 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */
1664 #define WF_FORK 0x02 /* Child wakeup after fork */
1665 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */
1666
1667 /*
1668 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1669 * of tasks with abnormal "nice" values across CPUs the contribution that
1670 * each task makes to its run queue's load is weighted according to its
1671 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1672 * scaled version of the new time slice allocation that they receive on time
1673 * slice expiry etc.
1674 */
1675
1676 #define WEIGHT_IDLEPRIO 3
1677 #define WMULT_IDLEPRIO 1431655765
1678
1679 extern const int sched_prio_to_weight[40];
1680 extern const u32 sched_prio_to_wmult[40];
1681
1682 /*
1683 * {de,en}queue flags:
1684 *
1685 * DEQUEUE_SLEEP - task is no longer runnable
1686 * ENQUEUE_WAKEUP - task just became runnable
1687 *
1688 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1689 * are in a known state which allows modification. Such pairs
1690 * should preserve as much state as possible.
1691 *
1692 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1693 * in the runqueue.
1694 *
1695 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1696 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1697 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1698 *
1699 */
1700
1701 #define DEQUEUE_SLEEP 0x01
1702 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1703 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1704 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1705
1706 #define ENQUEUE_WAKEUP 0x01
1707 #define ENQUEUE_RESTORE 0x02
1708 #define ENQUEUE_MOVE 0x04
1709 #define ENQUEUE_NOCLOCK 0x08
1710
1711 #define ENQUEUE_HEAD 0x10
1712 #define ENQUEUE_REPLENISH 0x20
1713 #ifdef CONFIG_SMP
1714 #define ENQUEUE_MIGRATED 0x40
1715 #else
1716 #define ENQUEUE_MIGRATED 0x00
1717 #endif
1718
1719 #define RETRY_TASK ((void *)-1UL)
1720
1721 struct sched_class {
1722 const struct sched_class *next;
1723
1724 #ifdef CONFIG_UCLAMP_TASK
1725 int uclamp_enabled;
1726 #endif
1727
1728 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1729 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1730 void (*yield_task) (struct rq *rq);
1731 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1732
1733 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1734
1735 /*
1736 * Both @prev and @rf are optional and may be NULL, in which case the
1737 * caller must already have invoked put_prev_task(rq, prev, rf).
1738 *
1739 * Otherwise it is the responsibility of the pick_next_task() to call
1740 * put_prev_task() on the @prev task or something equivalent, IFF it
1741 * returns a next task.
1742 *
1743 * In that case (@rf != NULL) it may return RETRY_TASK when it finds a
1744 * higher prio class has runnable tasks.
1745 */
1746 struct task_struct * (*pick_next_task)(struct rq *rq,
1747 struct task_struct *prev,
1748 struct rq_flags *rf);
1749 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1750 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1751
1752 #ifdef CONFIG_SMP
1753 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1754 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1755 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1756
1757 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1758
1759 void (*set_cpus_allowed)(struct task_struct *p,
1760 const struct cpumask *newmask);
1761
1762 void (*rq_online)(struct rq *rq);
1763 void (*rq_offline)(struct rq *rq);
1764 #endif
1765
1766 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1767 void (*task_fork)(struct task_struct *p);
1768 void (*task_dead)(struct task_struct *p);
1769
1770 /*
1771 * The switched_from() call is allowed to drop rq->lock, therefore we
1772 * cannot assume the switched_from/switched_to pair is serliazed by
1773 * rq->lock. They are however serialized by p->pi_lock.
1774 */
1775 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1776 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1777 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1778 int oldprio);
1779
1780 unsigned int (*get_rr_interval)(struct rq *rq,
1781 struct task_struct *task);
1782
1783 void (*update_curr)(struct rq *rq);
1784
1785 #define TASK_SET_GROUP 0
1786 #define TASK_MOVE_GROUP 1
1787
1788 #ifdef CONFIG_FAIR_GROUP_SCHED
1789 void (*task_change_group)(struct task_struct *p, int type);
1790 #endif
1791 };
1792
put_prev_task(struct rq * rq,struct task_struct * prev)1793 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1794 {
1795 WARN_ON_ONCE(rq->curr != prev);
1796 prev->sched_class->put_prev_task(rq, prev);
1797 }
1798
set_next_task(struct rq * rq,struct task_struct * next)1799 static inline void set_next_task(struct rq *rq, struct task_struct *next)
1800 {
1801 WARN_ON_ONCE(rq->curr != next);
1802 next->sched_class->set_next_task(rq, next, false);
1803 }
1804
1805 #ifdef CONFIG_SMP
1806 #define sched_class_highest (&stop_sched_class)
1807 #else
1808 #define sched_class_highest (&dl_sched_class)
1809 #endif
1810
1811 #define for_class_range(class, _from, _to) \
1812 for (class = (_from); class != (_to); class = class->next)
1813
1814 #define for_each_class(class) \
1815 for_class_range(class, sched_class_highest, NULL)
1816
1817 extern const struct sched_class stop_sched_class;
1818 extern const struct sched_class dl_sched_class;
1819 extern const struct sched_class rt_sched_class;
1820 extern const struct sched_class fair_sched_class;
1821 extern const struct sched_class idle_sched_class;
1822
sched_stop_runnable(struct rq * rq)1823 static inline bool sched_stop_runnable(struct rq *rq)
1824 {
1825 return rq->stop && task_on_rq_queued(rq->stop);
1826 }
1827
sched_dl_runnable(struct rq * rq)1828 static inline bool sched_dl_runnable(struct rq *rq)
1829 {
1830 return rq->dl.dl_nr_running > 0;
1831 }
1832
sched_rt_runnable(struct rq * rq)1833 static inline bool sched_rt_runnable(struct rq *rq)
1834 {
1835 return rq->rt.rt_queued > 0;
1836 }
1837
sched_fair_runnable(struct rq * rq)1838 static inline bool sched_fair_runnable(struct rq *rq)
1839 {
1840 return rq->cfs.nr_running > 0;
1841 }
1842
1843 #ifdef CONFIG_SMP
1844
1845 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1846
1847 extern void trigger_load_balance(struct rq *rq);
1848
1849 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1850
1851 #endif
1852
1853 #ifdef CONFIG_CPU_IDLE
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)1854 static inline void idle_set_state(struct rq *rq,
1855 struct cpuidle_state *idle_state)
1856 {
1857 rq->idle_state = idle_state;
1858 }
1859
idle_get_state(struct rq * rq)1860 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1861 {
1862 SCHED_WARN_ON(!rcu_read_lock_held());
1863
1864 return rq->idle_state;
1865 }
1866 #else
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)1867 static inline void idle_set_state(struct rq *rq,
1868 struct cpuidle_state *idle_state)
1869 {
1870 }
1871
idle_get_state(struct rq * rq)1872 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1873 {
1874 return NULL;
1875 }
1876 #endif
1877
1878 extern void schedule_idle(void);
1879
1880 extern void sysrq_sched_debug_show(void);
1881 extern void sched_init_granularity(void);
1882 extern void update_max_interval(void);
1883
1884 extern void init_sched_dl_class(void);
1885 extern void init_sched_rt_class(void);
1886 extern void init_sched_fair_class(void);
1887
1888 extern void reweight_task(struct task_struct *p, int prio);
1889
1890 extern void resched_curr(struct rq *rq);
1891 extern void resched_cpu(int cpu);
1892
1893 extern struct rt_bandwidth def_rt_bandwidth;
1894 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1895
1896 extern struct dl_bandwidth def_dl_bandwidth;
1897 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1898 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1899 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1900 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1901
1902 #define BW_SHIFT 20
1903 #define BW_UNIT (1 << BW_SHIFT)
1904 #define RATIO_SHIFT 8
1905 #define MAX_BW_BITS (64 - BW_SHIFT)
1906 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
1907 unsigned long to_ratio(u64 period, u64 runtime);
1908
1909 extern void init_entity_runnable_average(struct sched_entity *se);
1910 extern void post_init_entity_util_avg(struct task_struct *p);
1911
1912 #ifdef CONFIG_NO_HZ_FULL
1913 extern bool sched_can_stop_tick(struct rq *rq);
1914 extern int __init sched_tick_offload_init(void);
1915
1916 /*
1917 * Tick may be needed by tasks in the runqueue depending on their policy and
1918 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1919 * nohz mode if necessary.
1920 */
sched_update_tick_dependency(struct rq * rq)1921 static inline void sched_update_tick_dependency(struct rq *rq)
1922 {
1923 int cpu;
1924
1925 if (!tick_nohz_full_enabled())
1926 return;
1927
1928 cpu = cpu_of(rq);
1929
1930 if (!tick_nohz_full_cpu(cpu))
1931 return;
1932
1933 if (sched_can_stop_tick(rq))
1934 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1935 else
1936 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1937 }
1938 #else
sched_tick_offload_init(void)1939 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)1940 static inline void sched_update_tick_dependency(struct rq *rq) { }
1941 #endif
1942
add_nr_running(struct rq * rq,unsigned count)1943 static inline void add_nr_running(struct rq *rq, unsigned count)
1944 {
1945 unsigned prev_nr = rq->nr_running;
1946
1947 rq->nr_running = prev_nr + count;
1948
1949 #ifdef CONFIG_SMP
1950 if (prev_nr < 2 && rq->nr_running >= 2) {
1951 if (!READ_ONCE(rq->rd->overload))
1952 WRITE_ONCE(rq->rd->overload, 1);
1953 }
1954 #endif
1955
1956 sched_update_tick_dependency(rq);
1957 }
1958
sub_nr_running(struct rq * rq,unsigned count)1959 static inline void sub_nr_running(struct rq *rq, unsigned count)
1960 {
1961 rq->nr_running -= count;
1962 /* Check if we still need preemption */
1963 sched_update_tick_dependency(rq);
1964 }
1965
1966 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1967 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1968
1969 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1970
1971 extern const_debug unsigned int sysctl_sched_nr_migrate;
1972 extern const_debug unsigned int sysctl_sched_migration_cost;
1973
1974 #ifdef CONFIG_SCHED_HRTICK
1975
1976 /*
1977 * Use hrtick when:
1978 * - enabled by features
1979 * - hrtimer is actually high res
1980 */
hrtick_enabled(struct rq * rq)1981 static inline int hrtick_enabled(struct rq *rq)
1982 {
1983 if (!sched_feat(HRTICK))
1984 return 0;
1985 if (!cpu_active(cpu_of(rq)))
1986 return 0;
1987 return hrtimer_is_hres_active(&rq->hrtick_timer);
1988 }
1989
1990 void hrtick_start(struct rq *rq, u64 delay);
1991
1992 #else
1993
hrtick_enabled(struct rq * rq)1994 static inline int hrtick_enabled(struct rq *rq)
1995 {
1996 return 0;
1997 }
1998
1999 #endif /* CONFIG_SCHED_HRTICK */
2000
2001 #ifndef arch_scale_freq_capacity
2002 static __always_inline
arch_scale_freq_capacity(int cpu)2003 unsigned long arch_scale_freq_capacity(int cpu)
2004 {
2005 return SCHED_CAPACITY_SCALE;
2006 }
2007 #endif
2008
2009 #ifndef arch_scale_max_freq_capacity
2010 struct sched_domain;
2011 static __always_inline
arch_scale_max_freq_capacity(struct sched_domain * sd,int cpu)2012 unsigned long arch_scale_max_freq_capacity(struct sched_domain *sd, int cpu)
2013 {
2014 return SCHED_CAPACITY_SCALE;
2015 }
2016 #endif
2017
2018 #ifdef CONFIG_SMP
2019 #ifdef CONFIG_PREEMPTION
2020
2021 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2022
2023 /*
2024 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2025 * way at the expense of forcing extra atomic operations in all
2026 * invocations. This assures that the double_lock is acquired using the
2027 * same underlying policy as the spinlock_t on this architecture, which
2028 * reduces latency compared to the unfair variant below. However, it
2029 * also adds more overhead and therefore may reduce throughput.
2030 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2031 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2032 __releases(this_rq->lock)
2033 __acquires(busiest->lock)
2034 __acquires(this_rq->lock)
2035 {
2036 raw_spin_unlock(&this_rq->lock);
2037 double_rq_lock(this_rq, busiest);
2038
2039 return 1;
2040 }
2041
2042 #else
2043 /*
2044 * Unfair double_lock_balance: Optimizes throughput at the expense of
2045 * latency by eliminating extra atomic operations when the locks are
2046 * already in proper order on entry. This favors lower CPU-ids and will
2047 * grant the double lock to lower CPUs over higher ids under contention,
2048 * regardless of entry order into the function.
2049 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2050 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2051 __releases(this_rq->lock)
2052 __acquires(busiest->lock)
2053 __acquires(this_rq->lock)
2054 {
2055 int ret = 0;
2056
2057 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2058 if (busiest < this_rq) {
2059 raw_spin_unlock(&this_rq->lock);
2060 raw_spin_lock(&busiest->lock);
2061 raw_spin_lock_nested(&this_rq->lock,
2062 SINGLE_DEPTH_NESTING);
2063 ret = 1;
2064 } else
2065 raw_spin_lock_nested(&busiest->lock,
2066 SINGLE_DEPTH_NESTING);
2067 }
2068 return ret;
2069 }
2070
2071 #endif /* CONFIG_PREEMPTION */
2072
2073 /*
2074 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2075 */
double_lock_balance(struct rq * this_rq,struct rq * busiest)2076 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2077 {
2078 if (unlikely(!irqs_disabled())) {
2079 /* printk() doesn't work well under rq->lock */
2080 raw_spin_unlock(&this_rq->lock);
2081 BUG_ON(1);
2082 }
2083
2084 return _double_lock_balance(this_rq, busiest);
2085 }
2086
double_unlock_balance(struct rq * this_rq,struct rq * busiest)2087 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2088 __releases(busiest->lock)
2089 {
2090 raw_spin_unlock(&busiest->lock);
2091 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2092 }
2093
double_lock(spinlock_t * l1,spinlock_t * l2)2094 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2095 {
2096 if (l1 > l2)
2097 swap(l1, l2);
2098
2099 spin_lock(l1);
2100 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2101 }
2102
double_lock_irq(spinlock_t * l1,spinlock_t * l2)2103 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2104 {
2105 if (l1 > l2)
2106 swap(l1, l2);
2107
2108 spin_lock_irq(l1);
2109 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2110 }
2111
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)2112 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2113 {
2114 if (l1 > l2)
2115 swap(l1, l2);
2116
2117 raw_spin_lock(l1);
2118 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2119 }
2120
2121 /*
2122 * double_rq_lock - safely lock two runqueues
2123 *
2124 * Note this does not disable interrupts like task_rq_lock,
2125 * you need to do so manually before calling.
2126 */
double_rq_lock(struct rq * rq1,struct rq * rq2)2127 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2128 __acquires(rq1->lock)
2129 __acquires(rq2->lock)
2130 {
2131 BUG_ON(!irqs_disabled());
2132 if (rq1 == rq2) {
2133 raw_spin_lock(&rq1->lock);
2134 __acquire(rq2->lock); /* Fake it out ;) */
2135 } else {
2136 if (rq1 < rq2) {
2137 raw_spin_lock(&rq1->lock);
2138 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2139 } else {
2140 raw_spin_lock(&rq2->lock);
2141 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2142 }
2143 }
2144 }
2145
2146 /*
2147 * double_rq_unlock - safely unlock two runqueues
2148 *
2149 * Note this does not restore interrupts like task_rq_unlock,
2150 * you need to do so manually after calling.
2151 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2152 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2153 __releases(rq1->lock)
2154 __releases(rq2->lock)
2155 {
2156 raw_spin_unlock(&rq1->lock);
2157 if (rq1 != rq2)
2158 raw_spin_unlock(&rq2->lock);
2159 else
2160 __release(rq2->lock);
2161 }
2162
2163 extern void set_rq_online (struct rq *rq);
2164 extern void set_rq_offline(struct rq *rq);
2165 extern bool sched_smp_initialized;
2166
2167 #else /* CONFIG_SMP */
2168
2169 /*
2170 * double_rq_lock - safely lock two runqueues
2171 *
2172 * Note this does not disable interrupts like task_rq_lock,
2173 * you need to do so manually before calling.
2174 */
double_rq_lock(struct rq * rq1,struct rq * rq2)2175 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2176 __acquires(rq1->lock)
2177 __acquires(rq2->lock)
2178 {
2179 BUG_ON(!irqs_disabled());
2180 BUG_ON(rq1 != rq2);
2181 raw_spin_lock(&rq1->lock);
2182 __acquire(rq2->lock); /* Fake it out ;) */
2183 }
2184
2185 /*
2186 * double_rq_unlock - safely unlock two runqueues
2187 *
2188 * Note this does not restore interrupts like task_rq_unlock,
2189 * you need to do so manually after calling.
2190 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2191 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2192 __releases(rq1->lock)
2193 __releases(rq2->lock)
2194 {
2195 BUG_ON(rq1 != rq2);
2196 raw_spin_unlock(&rq1->lock);
2197 __release(rq2->lock);
2198 }
2199
2200 #endif
2201
2202 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2203 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2204
2205 #ifdef CONFIG_SCHED_DEBUG
2206 extern bool sched_debug_enabled;
2207
2208 extern void print_cfs_stats(struct seq_file *m, int cpu);
2209 extern void print_rt_stats(struct seq_file *m, int cpu);
2210 extern void print_dl_stats(struct seq_file *m, int cpu);
2211 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2212 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2213 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2214 #ifdef CONFIG_NUMA_BALANCING
2215 extern void
2216 show_numa_stats(struct task_struct *p, struct seq_file *m);
2217 extern void
2218 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2219 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2220 #endif /* CONFIG_NUMA_BALANCING */
2221 #endif /* CONFIG_SCHED_DEBUG */
2222
2223 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2224 extern void init_rt_rq(struct rt_rq *rt_rq);
2225 extern void init_dl_rq(struct dl_rq *dl_rq);
2226
2227 extern void cfs_bandwidth_usage_inc(void);
2228 extern void cfs_bandwidth_usage_dec(void);
2229
2230 #ifdef CONFIG_NO_HZ_COMMON
2231 #define NOHZ_BALANCE_KICK_BIT 0
2232 #define NOHZ_STATS_KICK_BIT 1
2233
2234 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2235 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2236
2237 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2238
2239 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2240
2241 extern void nohz_balance_exit_idle(struct rq *rq);
2242 #else
nohz_balance_exit_idle(struct rq * rq)2243 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2244 #endif
2245
2246
2247 #ifdef CONFIG_SMP
2248 static inline
__dl_update(struct dl_bw * dl_b,s64 bw)2249 void __dl_update(struct dl_bw *dl_b, s64 bw)
2250 {
2251 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2252 int i;
2253
2254 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2255 "sched RCU must be held");
2256 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2257 struct rq *rq = cpu_rq(i);
2258
2259 rq->dl.extra_bw += bw;
2260 }
2261 }
2262 #else
2263 static inline
__dl_update(struct dl_bw * dl_b,s64 bw)2264 void __dl_update(struct dl_bw *dl_b, s64 bw)
2265 {
2266 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2267
2268 dl->extra_bw += bw;
2269 }
2270 #endif
2271
2272
2273 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2274 struct irqtime {
2275 u64 total;
2276 u64 tick_delta;
2277 u64 irq_start_time;
2278 struct u64_stats_sync sync;
2279 };
2280
2281 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2282
2283 /*
2284 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2285 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2286 * and never move forward.
2287 */
irq_time_read(int cpu)2288 static inline u64 irq_time_read(int cpu)
2289 {
2290 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2291 unsigned int seq;
2292 u64 total;
2293
2294 do {
2295 seq = __u64_stats_fetch_begin(&irqtime->sync);
2296 total = irqtime->total;
2297 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2298
2299 return total;
2300 }
2301 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2302
2303 #ifdef CONFIG_CPU_FREQ
2304 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2305
2306 /**
2307 * cpufreq_update_util - Take a note about CPU utilization changes.
2308 * @rq: Runqueue to carry out the update for.
2309 * @flags: Update reason flags.
2310 *
2311 * This function is called by the scheduler on the CPU whose utilization is
2312 * being updated.
2313 *
2314 * It can only be called from RCU-sched read-side critical sections.
2315 *
2316 * The way cpufreq is currently arranged requires it to evaluate the CPU
2317 * performance state (frequency/voltage) on a regular basis to prevent it from
2318 * being stuck in a completely inadequate performance level for too long.
2319 * That is not guaranteed to happen if the updates are only triggered from CFS
2320 * and DL, though, because they may not be coming in if only RT tasks are
2321 * active all the time (or there are RT tasks only).
2322 *
2323 * As a workaround for that issue, this function is called periodically by the
2324 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2325 * but that really is a band-aid. Going forward it should be replaced with
2326 * solutions targeted more specifically at RT tasks.
2327 */
cpufreq_update_util(struct rq * rq,unsigned int flags)2328 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2329 {
2330 struct update_util_data *data;
2331
2332 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2333 cpu_of(rq)));
2334 if (data)
2335 data->func(data, rq_clock(rq), flags);
2336 }
2337 #else
cpufreq_update_util(struct rq * rq,unsigned int flags)2338 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2339 #endif /* CONFIG_CPU_FREQ */
2340
2341 #ifdef CONFIG_UCLAMP_TASK
2342 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2343
2344 /**
2345 * uclamp_util_with - clamp @util with @rq and @p effective uclamp values.
2346 * @rq: The rq to clamp against. Must not be NULL.
2347 * @util: The util value to clamp.
2348 * @p: The task to clamp against. Can be NULL if you want to clamp
2349 * against @rq only.
2350 *
2351 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2352 *
2353 * If sched_uclamp_used static key is disabled, then just return the util
2354 * without any clamping since uclamp aggregation at the rq level in the fast
2355 * path is disabled, rendering this operation a NOP.
2356 *
2357 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2358 * will return the correct effective uclamp value of the task even if the
2359 * static key is disabled.
2360 */
2361 static __always_inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)2362 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2363 struct task_struct *p)
2364 {
2365 unsigned long min_util;
2366 unsigned long max_util;
2367
2368 if (!static_branch_likely(&sched_uclamp_used))
2369 return util;
2370
2371 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2372 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2373
2374 if (p) {
2375 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2376 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2377 }
2378
2379 /*
2380 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2381 * RUNNABLE tasks with _different_ clamps, we can end up with an
2382 * inversion. Fix it now when the clamps are applied.
2383 */
2384 if (unlikely(min_util >= max_util))
2385 return min_util;
2386
2387 return clamp(util, min_util, max_util);
2388 }
2389
uclamp_boosted(struct task_struct * p)2390 static inline bool uclamp_boosted(struct task_struct *p)
2391 {
2392 return uclamp_eff_value(p, UCLAMP_MIN) > 0;
2393 }
2394
2395 /*
2396 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2397 * by default in the fast path and only gets turned on once userspace performs
2398 * an operation that requires it.
2399 *
2400 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2401 * hence is active.
2402 */
uclamp_is_used(void)2403 static inline bool uclamp_is_used(void)
2404 {
2405 return static_branch_likely(&sched_uclamp_used);
2406 }
2407 #else /* CONFIG_UCLAMP_TASK */
2408 static inline
uclamp_rq_util_with(struct rq * rq,unsigned long util,struct task_struct * p)2409 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2410 struct task_struct *p)
2411 {
2412 return util;
2413 }
uclamp_boosted(struct task_struct * p)2414 static inline bool uclamp_boosted(struct task_struct *p)
2415 {
2416 return false;
2417 }
2418
uclamp_is_used(void)2419 static inline bool uclamp_is_used(void)
2420 {
2421 return false;
2422 }
2423 #endif /* CONFIG_UCLAMP_TASK */
2424
2425 #ifdef CONFIG_UCLAMP_TASK_GROUP
uclamp_latency_sensitive(struct task_struct * p)2426 static inline bool uclamp_latency_sensitive(struct task_struct *p)
2427 {
2428 struct cgroup_subsys_state *css = task_css(p, cpu_cgrp_id);
2429 struct task_group *tg;
2430
2431 if (!css)
2432 return false;
2433 tg = container_of(css, struct task_group, css);
2434
2435 return tg->latency_sensitive;
2436 }
2437 #else
uclamp_latency_sensitive(struct task_struct * p)2438 static inline bool uclamp_latency_sensitive(struct task_struct *p)
2439 {
2440 return false;
2441 }
2442 #endif /* CONFIG_UCLAMP_TASK_GROUP */
2443
2444 #ifdef arch_scale_freq_capacity
2445 # ifndef arch_scale_freq_invariant
2446 # define arch_scale_freq_invariant() true
2447 # endif
2448 #else
2449 # define arch_scale_freq_invariant() false
2450 #endif
2451
2452 #ifdef CONFIG_SMP
capacity_orig_of(int cpu)2453 static inline unsigned long capacity_orig_of(int cpu)
2454 {
2455 return cpu_rq(cpu)->cpu_capacity_orig;
2456 }
2457 #endif
2458
2459 /**
2460 * enum schedutil_type - CPU utilization type
2461 * @FREQUENCY_UTIL: Utilization used to select frequency
2462 * @ENERGY_UTIL: Utilization used during energy calculation
2463 *
2464 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2465 * need to be aggregated differently depending on the usage made of them. This
2466 * enum is used within schedutil_freq_util() to differentiate the types of
2467 * utilization expected by the callers, and adjust the aggregation accordingly.
2468 */
2469 enum schedutil_type {
2470 FREQUENCY_UTIL,
2471 ENERGY_UTIL,
2472 };
2473
2474 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2475
2476 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2477 unsigned long max, enum schedutil_type type,
2478 struct task_struct *p);
2479
cpu_bw_dl(struct rq * rq)2480 static inline unsigned long cpu_bw_dl(struct rq *rq)
2481 {
2482 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2483 }
2484
cpu_util_dl(struct rq * rq)2485 static inline unsigned long cpu_util_dl(struct rq *rq)
2486 {
2487 return READ_ONCE(rq->avg_dl.util_avg);
2488 }
2489
cpu_util_cfs(struct rq * rq)2490 static inline unsigned long cpu_util_cfs(struct rq *rq)
2491 {
2492 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2493
2494 if (sched_feat(UTIL_EST)) {
2495 util = max_t(unsigned long, util,
2496 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2497 }
2498
2499 return util;
2500 }
2501
cpu_util_rt(struct rq * rq)2502 static inline unsigned long cpu_util_rt(struct rq *rq)
2503 {
2504 return READ_ONCE(rq->avg_rt.util_avg);
2505 }
2506 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
schedutil_cpu_util(int cpu,unsigned long util_cfs,unsigned long max,enum schedutil_type type,struct task_struct * p)2507 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2508 unsigned long max, enum schedutil_type type,
2509 struct task_struct *p)
2510 {
2511 return 0;
2512 }
2513 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2514
2515 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
cpu_util_irq(struct rq * rq)2516 static inline unsigned long cpu_util_irq(struct rq *rq)
2517 {
2518 return rq->avg_irq.util_avg;
2519 }
2520
2521 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)2522 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2523 {
2524 util *= (max - irq);
2525 util /= max;
2526
2527 return util;
2528
2529 }
2530 #else
cpu_util_irq(struct rq * rq)2531 static inline unsigned long cpu_util_irq(struct rq *rq)
2532 {
2533 return 0;
2534 }
2535
2536 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)2537 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2538 {
2539 return util;
2540 }
2541 #endif
2542
2543 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2544
2545 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2546
2547 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2548
sched_energy_enabled(void)2549 static inline bool sched_energy_enabled(void)
2550 {
2551 return static_branch_unlikely(&sched_energy_present);
2552 }
2553
2554 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2555
2556 #define perf_domain_span(pd) NULL
sched_energy_enabled(void)2557 static inline bool sched_energy_enabled(void) { return false; }
2558
2559 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2560
2561 #ifdef CONFIG_MEMBARRIER
2562 /*
2563 * The scheduler provides memory barriers required by membarrier between:
2564 * - prior user-space memory accesses and store to rq->membarrier_state,
2565 * - store to rq->membarrier_state and following user-space memory accesses.
2566 * In the same way it provides those guarantees around store to rq->curr.
2567 */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)2568 static inline void membarrier_switch_mm(struct rq *rq,
2569 struct mm_struct *prev_mm,
2570 struct mm_struct *next_mm)
2571 {
2572 int membarrier_state;
2573
2574 if (prev_mm == next_mm)
2575 return;
2576
2577 membarrier_state = atomic_read(&next_mm->membarrier_state);
2578 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2579 return;
2580
2581 WRITE_ONCE(rq->membarrier_state, membarrier_state);
2582 }
2583 #else
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)2584 static inline void membarrier_switch_mm(struct rq *rq,
2585 struct mm_struct *prev_mm,
2586 struct mm_struct *next_mm)
2587 {
2588 }
2589 #endif
2590