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