Line data Source code
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 : #include <trace/events/sched.h>
79 :
80 : #ifdef CONFIG_SCHED_DEBUG
81 : # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
82 : #else
83 : # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
84 : #endif
85 :
86 : struct rq;
87 : struct cpuidle_state;
88 :
89 : /* task_struct::on_rq states: */
90 : #define TASK_ON_RQ_QUEUED 1
91 : #define TASK_ON_RQ_MIGRATING 2
92 :
93 : extern __read_mostly int scheduler_running;
94 :
95 : extern unsigned long calc_load_update;
96 : extern atomic_long_t calc_load_tasks;
97 :
98 : extern void calc_global_load_tick(struct rq *this_rq);
99 : extern long calc_load_fold_active(struct rq *this_rq, long adjust);
100 :
101 : extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
102 : /*
103 : * Helpers for converting nanosecond timing to jiffy resolution
104 : */
105 : #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
106 :
107 : /*
108 : * Increase resolution of nice-level calculations for 64-bit architectures.
109 : * The extra resolution improves shares distribution and load balancing of
110 : * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
111 : * hierarchies, especially on larger systems. This is not a user-visible change
112 : * and does not change the user-interface for setting shares/weights.
113 : *
114 : * We increase resolution only if we have enough bits to allow this increased
115 : * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
116 : * are pretty high and the returns do not justify the increased costs.
117 : *
118 : * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
119 : * increase coverage and consistency always enable it on 64-bit platforms.
120 : */
121 : #ifdef CONFIG_64BIT
122 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
123 : # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
124 : # define scale_load_down(w) \
125 : ({ \
126 : unsigned long __w = (w); \
127 : if (__w) \
128 : __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
129 : __w; \
130 : })
131 : #else
132 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
133 : # define scale_load(w) (w)
134 : # define scale_load_down(w) (w)
135 : #endif
136 :
137 : /*
138 : * Task weight (visible to users) and its load (invisible to users) have
139 : * independent resolution, but they should be well calibrated. We use
140 : * scale_load() and scale_load_down(w) to convert between them. The
141 : * following must be true:
142 : *
143 : * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
144 : *
145 : */
146 : #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
147 :
148 : /*
149 : * Single value that decides SCHED_DEADLINE internal math precision.
150 : * 10 -> just above 1us
151 : * 9 -> just above 0.5us
152 : */
153 : #define DL_SCALE 10
154 :
155 : /*
156 : * Single value that denotes runtime == period, ie unlimited time.
157 : */
158 : #define RUNTIME_INF ((u64)~0ULL)
159 :
160 46717 : static inline int idle_policy(int policy)
161 : {
162 46636 : return policy == SCHED_IDLE;
163 : }
164 110 : static inline int fair_policy(int policy)
165 : {
166 55 : return policy == SCHED_NORMAL || policy == SCHED_BATCH;
167 : }
168 :
169 162 : static inline int rt_policy(int policy)
170 : {
171 106 : return policy == SCHED_FIFO || policy == SCHED_RR;
172 : }
173 :
174 219 : static inline int dl_policy(int policy)
175 : {
176 169 : return policy == SCHED_DEADLINE;
177 : }
178 55 : static inline bool valid_policy(int policy)
179 : {
180 55 : return idle_policy(policy) || fair_policy(policy) ||
181 59 : rt_policy(policy) || dl_policy(policy);
182 : }
183 :
184 46662 : static inline int task_has_idle_policy(struct task_struct *p)
185 : {
186 46662 : return idle_policy(p->policy);
187 : }
188 :
189 52 : static inline int task_has_rt_policy(struct task_struct *p)
190 : {
191 52 : return rt_policy(p->policy);
192 : }
193 :
194 50 : static inline int task_has_dl_policy(struct task_struct *p)
195 : {
196 50 : return dl_policy(p->policy);
197 : }
198 :
199 : #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
200 :
201 6465 : static inline void update_avg(u64 *avg, u64 sample)
202 : {
203 6465 : s64 diff = sample - *avg;
204 6465 : *avg += diff / 8;
205 0 : }
206 :
207 : /*
208 : * !! For sched_setattr_nocheck() (kernel) only !!
209 : *
210 : * This is actually gross. :(
211 : *
212 : * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
213 : * tasks, but still be able to sleep. We need this on platforms that cannot
214 : * atomically change clock frequency. Remove once fast switching will be
215 : * available on such platforms.
216 : *
217 : * SUGOV stands for SchedUtil GOVernor.
218 : */
219 : #define SCHED_FLAG_SUGOV 0x10000000
220 :
221 0 : static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
222 : {
223 : #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
224 : return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
225 : #else
226 0 : return false;
227 : #endif
228 : }
229 :
230 : /*
231 : * Tells if entity @a should preempt entity @b.
232 : */
233 : static inline bool
234 0 : dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
235 : {
236 0 : return dl_entity_is_special(a) ||
237 0 : dl_time_before(a->deadline, b->deadline);
238 : }
239 :
240 : /*
241 : * This is the priority-queue data structure of the RT scheduling class:
242 : */
243 : struct rt_prio_array {
244 : DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
245 : struct list_head queue[MAX_RT_PRIO];
246 : };
247 :
248 : struct rt_bandwidth {
249 : /* nests inside the rq lock: */
250 : raw_spinlock_t rt_runtime_lock;
251 : ktime_t rt_period;
252 : u64 rt_runtime;
253 : struct hrtimer rt_period_timer;
254 : unsigned int rt_period_active;
255 : };
256 :
257 : void __dl_clear_params(struct task_struct *p);
258 :
259 : struct dl_bandwidth {
260 : raw_spinlock_t dl_runtime_lock;
261 : u64 dl_runtime;
262 : u64 dl_period;
263 : };
264 :
265 0 : static inline int dl_bandwidth_enabled(void)
266 : {
267 0 : return sysctl_sched_rt_runtime >= 0;
268 : }
269 :
270 : /*
271 : * To keep the bandwidth of -deadline tasks under control
272 : * we need some place where:
273 : * - store the maximum -deadline bandwidth of each cpu;
274 : * - cache the fraction of bandwidth that is currently allocated in
275 : * each root domain;
276 : *
277 : * This is all done in the data structure below. It is similar to the
278 : * one used for RT-throttling (rt_bandwidth), with the main difference
279 : * that, since here we are only interested in admission control, we
280 : * do not decrease any runtime while the group "executes", neither we
281 : * need a timer to replenish it.
282 : *
283 : * With respect to SMP, bandwidth is given on a per root domain basis,
284 : * meaning that:
285 : * - bw (< 100%) is the deadline bandwidth of each CPU;
286 : * - total_bw is the currently allocated bandwidth in each root domain;
287 : */
288 : struct dl_bw {
289 : raw_spinlock_t lock;
290 : u64 bw;
291 : u64 total_bw;
292 : };
293 :
294 : static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
295 :
296 : static inline
297 0 : void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
298 : {
299 0 : dl_b->total_bw -= tsk_bw;
300 0 : __dl_update(dl_b, (s32)tsk_bw / cpus);
301 0 : }
302 :
303 : static inline
304 0 : void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
305 : {
306 0 : dl_b->total_bw += tsk_bw;
307 0 : __dl_update(dl_b, -((s32)tsk_bw / cpus));
308 : }
309 :
310 0 : static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
311 : u64 old_bw, u64 new_bw)
312 : {
313 0 : return dl_b->bw != -1 &&
314 0 : cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
315 : }
316 :
317 : /*
318 : * Verify the fitness of task @p to run on @cpu taking into account the
319 : * CPU original capacity and the runtime/deadline ratio of the task.
320 : *
321 : * The function will return true if the CPU original capacity of the
322 : * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
323 : * task and false otherwise.
324 : */
325 0 : static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
326 : {
327 0 : unsigned long cap = arch_scale_cpu_capacity(cpu);
328 :
329 0 : return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
330 : }
331 :
332 : extern void init_dl_bw(struct dl_bw *dl_b);
333 : extern int sched_dl_global_validate(void);
334 : extern void sched_dl_do_global(void);
335 : extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
336 : extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
337 : extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
338 : extern bool __checkparam_dl(const struct sched_attr *attr);
339 : extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
340 : extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
341 : extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
342 : extern bool dl_cpu_busy(unsigned int cpu);
343 :
344 : #ifdef CONFIG_CGROUP_SCHED
345 :
346 : #include <linux/cgroup.h>
347 : #include <linux/psi.h>
348 :
349 : struct cfs_rq;
350 : struct rt_rq;
351 :
352 : extern struct list_head task_groups;
353 :
354 : struct cfs_bandwidth {
355 : #ifdef CONFIG_CFS_BANDWIDTH
356 : raw_spinlock_t lock;
357 : ktime_t period;
358 : u64 quota;
359 : u64 runtime;
360 : s64 hierarchical_quota;
361 :
362 : u8 idle;
363 : u8 period_active;
364 : u8 slack_started;
365 : struct hrtimer period_timer;
366 : struct hrtimer slack_timer;
367 : struct list_head throttled_cfs_rq;
368 :
369 : /* Statistics: */
370 : int nr_periods;
371 : int nr_throttled;
372 : u64 throttled_time;
373 : #endif
374 : };
375 :
376 : /* Task group related information */
377 : struct task_group {
378 : struct cgroup_subsys_state css;
379 :
380 : #ifdef CONFIG_FAIR_GROUP_SCHED
381 : /* schedulable entities of this group on each CPU */
382 : struct sched_entity **se;
383 : /* runqueue "owned" by this group on each CPU */
384 : struct cfs_rq **cfs_rq;
385 : unsigned long shares;
386 :
387 : #ifdef CONFIG_SMP
388 : /*
389 : * load_avg can be heavily contended at clock tick time, so put
390 : * it in its own cacheline separated from the fields above which
391 : * will also be accessed at each tick.
392 : */
393 : atomic_long_t load_avg ____cacheline_aligned;
394 : #endif
395 : #endif
396 :
397 : #ifdef CONFIG_RT_GROUP_SCHED
398 : struct sched_rt_entity **rt_se;
399 : struct rt_rq **rt_rq;
400 :
401 : struct rt_bandwidth rt_bandwidth;
402 : #endif
403 :
404 : struct rcu_head rcu;
405 : struct list_head list;
406 :
407 : struct task_group *parent;
408 : struct list_head siblings;
409 : struct list_head children;
410 :
411 : #ifdef CONFIG_SCHED_AUTOGROUP
412 : struct autogroup *autogroup;
413 : #endif
414 :
415 : struct cfs_bandwidth cfs_bandwidth;
416 :
417 : #ifdef CONFIG_UCLAMP_TASK_GROUP
418 : /* The two decimal precision [%] value requested from user-space */
419 : unsigned int uclamp_pct[UCLAMP_CNT];
420 : /* Clamp values requested for a task group */
421 : struct uclamp_se uclamp_req[UCLAMP_CNT];
422 : /* Effective clamp values used for a task group */
423 : struct uclamp_se uclamp[UCLAMP_CNT];
424 : #endif
425 :
426 : };
427 :
428 : #ifdef CONFIG_FAIR_GROUP_SCHED
429 : #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
430 :
431 : /*
432 : * A weight of 0 or 1 can cause arithmetics problems.
433 : * A weight of a cfs_rq is the sum of weights of which entities
434 : * are queued on this cfs_rq, so a weight of a entity should not be
435 : * too large, so as the shares value of a task group.
436 : * (The default weight is 1024 - so there's no practical
437 : * limitation from this.)
438 : */
439 : #define MIN_SHARES (1UL << 1)
440 : #define MAX_SHARES (1UL << 18)
441 : #endif
442 :
443 : typedef int (*tg_visitor)(struct task_group *, void *);
444 :
445 : extern int walk_tg_tree_from(struct task_group *from,
446 : tg_visitor down, tg_visitor up, void *data);
447 :
448 : /*
449 : * Iterate the full tree, calling @down when first entering a node and @up when
450 : * leaving it for the final time.
451 : *
452 : * Caller must hold rcu_lock or sufficient equivalent.
453 : */
454 : static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
455 : {
456 : return walk_tg_tree_from(&root_task_group, down, up, data);
457 : }
458 :
459 : extern int tg_nop(struct task_group *tg, void *data);
460 :
461 : extern void free_fair_sched_group(struct task_group *tg);
462 : extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
463 : extern void online_fair_sched_group(struct task_group *tg);
464 : extern void unregister_fair_sched_group(struct task_group *tg);
465 : extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
466 : struct sched_entity *se, int cpu,
467 : struct sched_entity *parent);
468 : extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
469 :
470 : extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
471 : extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
472 : extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
473 :
474 : extern void free_rt_sched_group(struct task_group *tg);
475 : extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
476 : extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
477 : struct sched_rt_entity *rt_se, int cpu,
478 : struct sched_rt_entity *parent);
479 : extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
480 : extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
481 : extern long sched_group_rt_runtime(struct task_group *tg);
482 : extern long sched_group_rt_period(struct task_group *tg);
483 : extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
484 :
485 : extern struct task_group *sched_create_group(struct task_group *parent);
486 : extern void sched_online_group(struct task_group *tg,
487 : struct task_group *parent);
488 : extern void sched_destroy_group(struct task_group *tg);
489 : extern void sched_offline_group(struct task_group *tg);
490 :
491 : extern void sched_move_task(struct task_struct *tsk);
492 :
493 : #ifdef CONFIG_FAIR_GROUP_SCHED
494 : extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
495 :
496 : #ifdef CONFIG_SMP
497 : extern void set_task_rq_fair(struct sched_entity *se,
498 : struct cfs_rq *prev, struct cfs_rq *next);
499 : #else /* !CONFIG_SMP */
500 : static inline void set_task_rq_fair(struct sched_entity *se,
501 : struct cfs_rq *prev, struct cfs_rq *next) { }
502 : #endif /* CONFIG_SMP */
503 : #endif /* CONFIG_FAIR_GROUP_SCHED */
504 :
505 : #else /* CONFIG_CGROUP_SCHED */
506 :
507 : struct cfs_bandwidth { };
508 :
509 : #endif /* CONFIG_CGROUP_SCHED */
510 :
511 : /* CFS-related fields in a runqueue */
512 : struct cfs_rq {
513 : struct load_weight load;
514 : unsigned int nr_running;
515 : unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
516 : unsigned int idle_h_nr_running; /* SCHED_IDLE */
517 :
518 : u64 exec_clock;
519 : u64 min_vruntime;
520 : #ifndef CONFIG_64BIT
521 : u64 min_vruntime_copy;
522 : #endif
523 :
524 : struct rb_root_cached tasks_timeline;
525 :
526 : /*
527 : * 'curr' points to currently running entity on this cfs_rq.
528 : * It is set to NULL otherwise (i.e when none are currently running).
529 : */
530 : struct sched_entity *curr;
531 : struct sched_entity *next;
532 : struct sched_entity *last;
533 : struct sched_entity *skip;
534 :
535 : #ifdef CONFIG_SCHED_DEBUG
536 : unsigned int nr_spread_over;
537 : #endif
538 :
539 : #ifdef CONFIG_SMP
540 : /*
541 : * CFS load tracking
542 : */
543 : struct sched_avg avg;
544 : #ifndef CONFIG_64BIT
545 : u64 load_last_update_time_copy;
546 : #endif
547 : struct {
548 : raw_spinlock_t lock ____cacheline_aligned;
549 : int nr;
550 : unsigned long load_avg;
551 : unsigned long util_avg;
552 : unsigned long runnable_avg;
553 : } removed;
554 :
555 : #ifdef CONFIG_FAIR_GROUP_SCHED
556 : unsigned long tg_load_avg_contrib;
557 : long propagate;
558 : long prop_runnable_sum;
559 :
560 : /*
561 : * h_load = weight * f(tg)
562 : *
563 : * Where f(tg) is the recursive weight fraction assigned to
564 : * this group.
565 : */
566 : unsigned long h_load;
567 : u64 last_h_load_update;
568 : struct sched_entity *h_load_next;
569 : #endif /* CONFIG_FAIR_GROUP_SCHED */
570 : #endif /* CONFIG_SMP */
571 :
572 : #ifdef CONFIG_FAIR_GROUP_SCHED
573 : struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
574 :
575 : /*
576 : * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
577 : * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
578 : * (like users, containers etc.)
579 : *
580 : * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
581 : * This list is used during load balance.
582 : */
583 : int on_list;
584 : struct list_head leaf_cfs_rq_list;
585 : struct task_group *tg; /* group that "owns" this runqueue */
586 :
587 : #ifdef CONFIG_CFS_BANDWIDTH
588 : int runtime_enabled;
589 : s64 runtime_remaining;
590 :
591 : u64 throttled_clock;
592 : u64 throttled_clock_task;
593 : u64 throttled_clock_task_time;
594 : int throttled;
595 : int throttle_count;
596 : struct list_head throttled_list;
597 : #endif /* CONFIG_CFS_BANDWIDTH */
598 : #endif /* CONFIG_FAIR_GROUP_SCHED */
599 : };
600 :
601 0 : static inline int rt_bandwidth_enabled(void)
602 : {
603 0 : return sysctl_sched_rt_runtime >= 0;
604 : }
605 :
606 : /* RT IPI pull logic requires IRQ_WORK */
607 : #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
608 : # define HAVE_RT_PUSH_IPI
609 : #endif
610 :
611 : /* Real-Time classes' related field in a runqueue: */
612 : struct rt_rq {
613 : struct rt_prio_array active;
614 : unsigned int rt_nr_running;
615 : unsigned int rr_nr_running;
616 : #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
617 : struct {
618 : int curr; /* highest queued rt task prio */
619 : #ifdef CONFIG_SMP
620 : int next; /* next highest */
621 : #endif
622 : } highest_prio;
623 : #endif
624 : #ifdef CONFIG_SMP
625 : unsigned long rt_nr_migratory;
626 : unsigned long rt_nr_total;
627 : int overloaded;
628 : struct plist_head pushable_tasks;
629 :
630 : #endif /* CONFIG_SMP */
631 : int rt_queued;
632 :
633 : int rt_throttled;
634 : u64 rt_time;
635 : u64 rt_runtime;
636 : /* Nests inside the rq lock: */
637 : raw_spinlock_t rt_runtime_lock;
638 :
639 : #ifdef CONFIG_RT_GROUP_SCHED
640 : unsigned long rt_nr_boosted;
641 :
642 : struct rq *rq;
643 : struct task_group *tg;
644 : #endif
645 : };
646 :
647 0 : static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
648 : {
649 0 : return rt_rq->rt_queued && rt_rq->rt_nr_running;
650 : }
651 :
652 : /* Deadline class' related fields in a runqueue */
653 : struct dl_rq {
654 : /* runqueue is an rbtree, ordered by deadline */
655 : struct rb_root_cached root;
656 :
657 : unsigned long dl_nr_running;
658 :
659 : #ifdef CONFIG_SMP
660 : /*
661 : * Deadline values of the currently executing and the
662 : * earliest ready task on this rq. Caching these facilitates
663 : * the decision whether or not a ready but not running task
664 : * should migrate somewhere else.
665 : */
666 : struct {
667 : u64 curr;
668 : u64 next;
669 : } earliest_dl;
670 :
671 : unsigned long dl_nr_migratory;
672 : int overloaded;
673 :
674 : /*
675 : * Tasks on this rq that can be pushed away. They are kept in
676 : * an rb-tree, ordered by tasks' deadlines, with caching
677 : * of the leftmost (earliest deadline) element.
678 : */
679 : struct rb_root_cached pushable_dl_tasks_root;
680 : #else
681 : struct dl_bw dl_bw;
682 : #endif
683 : /*
684 : * "Active utilization" for this runqueue: increased when a
685 : * task wakes up (becomes TASK_RUNNING) and decreased when a
686 : * task blocks
687 : */
688 : u64 running_bw;
689 :
690 : /*
691 : * Utilization of the tasks "assigned" to this runqueue (including
692 : * the tasks that are in runqueue and the tasks that executed on this
693 : * CPU and blocked). Increased when a task moves to this runqueue, and
694 : * decreased when the task moves away (migrates, changes scheduling
695 : * policy, or terminates).
696 : * This is needed to compute the "inactive utilization" for the
697 : * runqueue (inactive utilization = this_bw - running_bw).
698 : */
699 : u64 this_bw;
700 : u64 extra_bw;
701 :
702 : /*
703 : * Inverse of the fraction of CPU utilization that can be reclaimed
704 : * by the GRUB algorithm.
705 : */
706 : u64 bw_ratio;
707 : };
708 :
709 : #ifdef CONFIG_FAIR_GROUP_SCHED
710 : /* An entity is a task if it doesn't "own" a runqueue */
711 : #define entity_is_task(se) (!se->my_q)
712 :
713 : static inline void se_update_runnable(struct sched_entity *se)
714 : {
715 : if (!entity_is_task(se))
716 : se->runnable_weight = se->my_q->h_nr_running;
717 : }
718 :
719 : static inline long se_runnable(struct sched_entity *se)
720 : {
721 : if (entity_is_task(se))
722 : return !!se->on_rq;
723 : else
724 : return se->runnable_weight;
725 : }
726 :
727 : #else
728 : #define entity_is_task(se) 1
729 :
730 31532 : static inline void se_update_runnable(struct sched_entity *se) {}
731 :
732 74410 : static inline long se_runnable(struct sched_entity *se)
733 : {
734 74410 : return !!se->on_rq;
735 : }
736 : #endif
737 :
738 : #ifdef CONFIG_SMP
739 : /*
740 : * XXX we want to get rid of these helpers and use the full load resolution.
741 : */
742 45952 : static inline long se_weight(struct sched_entity *se)
743 : {
744 43970 : return scale_load_down(se->load.weight);
745 : }
746 :
747 :
748 0 : static inline bool sched_asym_prefer(int a, int b)
749 : {
750 0 : return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
751 : }
752 :
753 : struct perf_domain {
754 : struct em_perf_domain *em_pd;
755 : struct perf_domain *next;
756 : struct rcu_head rcu;
757 : };
758 :
759 : /* Scheduling group status flags */
760 : #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
761 : #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
762 :
763 : /*
764 : * We add the notion of a root-domain which will be used to define per-domain
765 : * variables. Each exclusive cpuset essentially defines an island domain by
766 : * fully partitioning the member CPUs from any other cpuset. Whenever a new
767 : * exclusive cpuset is created, we also create and attach a new root-domain
768 : * object.
769 : *
770 : */
771 : struct root_domain {
772 : atomic_t refcount;
773 : atomic_t rto_count;
774 : struct rcu_head rcu;
775 : cpumask_var_t span;
776 : cpumask_var_t online;
777 :
778 : /*
779 : * Indicate pullable load on at least one CPU, e.g:
780 : * - More than one runnable task
781 : * - Running task is misfit
782 : */
783 : int overload;
784 :
785 : /* Indicate one or more cpus over-utilized (tipping point) */
786 : int overutilized;
787 :
788 : /*
789 : * The bit corresponding to a CPU gets set here if such CPU has more
790 : * than one runnable -deadline task (as it is below for RT tasks).
791 : */
792 : cpumask_var_t dlo_mask;
793 : atomic_t dlo_count;
794 : struct dl_bw dl_bw;
795 : struct cpudl cpudl;
796 :
797 : /*
798 : * Indicate whether a root_domain's dl_bw has been checked or
799 : * updated. It's monotonously increasing value.
800 : *
801 : * Also, some corner cases, like 'wrap around' is dangerous, but given
802 : * that u64 is 'big enough'. So that shouldn't be a concern.
803 : */
804 : u64 visit_gen;
805 :
806 : #ifdef HAVE_RT_PUSH_IPI
807 : /*
808 : * For IPI pull requests, loop across the rto_mask.
809 : */
810 : struct irq_work rto_push_work;
811 : raw_spinlock_t rto_lock;
812 : /* These are only updated and read within rto_lock */
813 : int rto_loop;
814 : int rto_cpu;
815 : /* These atomics are updated outside of a lock */
816 : atomic_t rto_loop_next;
817 : atomic_t rto_loop_start;
818 : #endif
819 : /*
820 : * The "RT overload" flag: it gets set if a CPU has more than
821 : * one runnable RT task.
822 : */
823 : cpumask_var_t rto_mask;
824 : struct cpupri cpupri;
825 :
826 : unsigned long max_cpu_capacity;
827 :
828 : /*
829 : * NULL-terminated list of performance domains intersecting with the
830 : * CPUs of the rd. Protected by RCU.
831 : */
832 : struct perf_domain __rcu *pd;
833 : };
834 :
835 : extern void init_defrootdomain(void);
836 : extern int sched_init_domains(const struct cpumask *cpu_map);
837 : extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
838 : extern void sched_get_rd(struct root_domain *rd);
839 : extern void sched_put_rd(struct root_domain *rd);
840 :
841 : #ifdef HAVE_RT_PUSH_IPI
842 : extern void rto_push_irq_work_func(struct irq_work *work);
843 : #endif
844 : #endif /* CONFIG_SMP */
845 :
846 : #ifdef CONFIG_UCLAMP_TASK
847 : /*
848 : * struct uclamp_bucket - Utilization clamp bucket
849 : * @value: utilization clamp value for tasks on this clamp bucket
850 : * @tasks: number of RUNNABLE tasks on this clamp bucket
851 : *
852 : * Keep track of how many tasks are RUNNABLE for a given utilization
853 : * clamp value.
854 : */
855 : struct uclamp_bucket {
856 : unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
857 : unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
858 : };
859 :
860 : /*
861 : * struct uclamp_rq - rq's utilization clamp
862 : * @value: currently active clamp values for a rq
863 : * @bucket: utilization clamp buckets affecting a rq
864 : *
865 : * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
866 : * A clamp value is affecting a rq when there is at least one task RUNNABLE
867 : * (or actually running) with that value.
868 : *
869 : * There are up to UCLAMP_CNT possible different clamp values, currently there
870 : * are only two: minimum utilization and maximum utilization.
871 : *
872 : * All utilization clamping values are MAX aggregated, since:
873 : * - for util_min: we want to run the CPU at least at the max of the minimum
874 : * utilization required by its currently RUNNABLE tasks.
875 : * - for util_max: we want to allow the CPU to run up to the max of the
876 : * maximum utilization allowed by its currently RUNNABLE tasks.
877 : *
878 : * Since on each system we expect only a limited number of different
879 : * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
880 : * the metrics required to compute all the per-rq utilization clamp values.
881 : */
882 : struct uclamp_rq {
883 : unsigned int value;
884 : struct uclamp_bucket bucket[UCLAMP_BUCKETS];
885 : };
886 :
887 : DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
888 : #endif /* CONFIG_UCLAMP_TASK */
889 :
890 : /*
891 : * This is the main, per-CPU runqueue data structure.
892 : *
893 : * Locking rule: those places that want to lock multiple runqueues
894 : * (such as the load balancing or the thread migration code), lock
895 : * acquire operations must be ordered by ascending &runqueue.
896 : */
897 : struct rq {
898 : /* runqueue lock: */
899 : raw_spinlock_t lock;
900 :
901 : /*
902 : * nr_running and cpu_load should be in the same cacheline because
903 : * remote CPUs use both these fields when doing load calculation.
904 : */
905 : unsigned int nr_running;
906 : #ifdef CONFIG_NUMA_BALANCING
907 : unsigned int nr_numa_running;
908 : unsigned int nr_preferred_running;
909 : unsigned int numa_migrate_on;
910 : #endif
911 : #ifdef CONFIG_NO_HZ_COMMON
912 : #ifdef CONFIG_SMP
913 : unsigned long last_blocked_load_update_tick;
914 : unsigned int has_blocked_load;
915 : call_single_data_t nohz_csd;
916 : #endif /* CONFIG_SMP */
917 : unsigned int nohz_tick_stopped;
918 : atomic_t nohz_flags;
919 : #endif /* CONFIG_NO_HZ_COMMON */
920 :
921 : #ifdef CONFIG_SMP
922 : unsigned int ttwu_pending;
923 : #endif
924 : u64 nr_switches;
925 :
926 : #ifdef CONFIG_UCLAMP_TASK
927 : /* Utilization clamp values based on CPU's RUNNABLE tasks */
928 : struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
929 : unsigned int uclamp_flags;
930 : #define UCLAMP_FLAG_IDLE 0x01
931 : #endif
932 :
933 : struct cfs_rq cfs;
934 : struct rt_rq rt;
935 : struct dl_rq dl;
936 :
937 : #ifdef CONFIG_FAIR_GROUP_SCHED
938 : /* list of leaf cfs_rq on this CPU: */
939 : struct list_head leaf_cfs_rq_list;
940 : struct list_head *tmp_alone_branch;
941 : #endif /* CONFIG_FAIR_GROUP_SCHED */
942 :
943 : /*
944 : * This is part of a global counter where only the total sum
945 : * over all CPUs matters. A task can increase this counter on
946 : * one CPU and if it got migrated afterwards it may decrease
947 : * it on another CPU. Always updated under the runqueue lock:
948 : */
949 : unsigned long nr_uninterruptible;
950 :
951 : struct task_struct __rcu *curr;
952 : struct task_struct *idle;
953 : struct task_struct *stop;
954 : unsigned long next_balance;
955 : struct mm_struct *prev_mm;
956 :
957 : unsigned int clock_update_flags;
958 : u64 clock;
959 : /* Ensure that all clocks are in the same cache line */
960 : u64 clock_task ____cacheline_aligned;
961 : u64 clock_pelt;
962 : unsigned long lost_idle_time;
963 :
964 : atomic_t nr_iowait;
965 :
966 : #ifdef CONFIG_MEMBARRIER
967 : int membarrier_state;
968 : #endif
969 :
970 : #ifdef CONFIG_SMP
971 : struct root_domain *rd;
972 : struct sched_domain __rcu *sd;
973 :
974 : unsigned long cpu_capacity;
975 : unsigned long cpu_capacity_orig;
976 :
977 : struct callback_head *balance_callback;
978 : unsigned char balance_push;
979 :
980 : unsigned char nohz_idle_balance;
981 : unsigned char idle_balance;
982 :
983 : unsigned long misfit_task_load;
984 :
985 : /* For active balancing */
986 : int active_balance;
987 : int push_cpu;
988 : struct cpu_stop_work active_balance_work;
989 :
990 : /* CPU of this runqueue: */
991 : int cpu;
992 : int online;
993 :
994 : struct list_head cfs_tasks;
995 :
996 : struct sched_avg avg_rt;
997 : struct sched_avg avg_dl;
998 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
999 : struct sched_avg avg_irq;
1000 : #endif
1001 : #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1002 : struct sched_avg avg_thermal;
1003 : #endif
1004 : u64 idle_stamp;
1005 : u64 avg_idle;
1006 :
1007 : /* This is used to determine avg_idle's max value */
1008 : u64 max_idle_balance_cost;
1009 :
1010 : #ifdef CONFIG_HOTPLUG_CPU
1011 : struct rcuwait hotplug_wait;
1012 : #endif
1013 : #endif /* CONFIG_SMP */
1014 :
1015 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1016 : u64 prev_irq_time;
1017 : #endif
1018 : #ifdef CONFIG_PARAVIRT
1019 : u64 prev_steal_time;
1020 : #endif
1021 : #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1022 : u64 prev_steal_time_rq;
1023 : #endif
1024 :
1025 : /* calc_load related fields */
1026 : unsigned long calc_load_update;
1027 : long calc_load_active;
1028 :
1029 : #ifdef CONFIG_SCHED_HRTICK
1030 : #ifdef CONFIG_SMP
1031 : call_single_data_t hrtick_csd;
1032 : #endif
1033 : struct hrtimer hrtick_timer;
1034 : ktime_t hrtick_time;
1035 : #endif
1036 :
1037 : #ifdef CONFIG_SCHEDSTATS
1038 : /* latency stats */
1039 : struct sched_info rq_sched_info;
1040 : unsigned long long rq_cpu_time;
1041 : /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1042 :
1043 : /* sys_sched_yield() stats */
1044 : unsigned int yld_count;
1045 :
1046 : /* schedule() stats */
1047 : unsigned int sched_count;
1048 : unsigned int sched_goidle;
1049 :
1050 : /* try_to_wake_up() stats */
1051 : unsigned int ttwu_count;
1052 : unsigned int ttwu_local;
1053 : #endif
1054 :
1055 : #ifdef CONFIG_CPU_IDLE
1056 : /* Must be inspected within a rcu lock section */
1057 : struct cpuidle_state *idle_state;
1058 : #endif
1059 :
1060 : #ifdef CONFIG_SMP
1061 : unsigned int nr_pinned;
1062 : #endif
1063 : unsigned int push_busy;
1064 : struct cpu_stop_work push_work;
1065 : };
1066 :
1067 : #ifdef CONFIG_FAIR_GROUP_SCHED
1068 :
1069 : /* CPU runqueue to which this cfs_rq is attached */
1070 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1071 : {
1072 : return cfs_rq->rq;
1073 : }
1074 :
1075 : #else
1076 :
1077 212266 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1078 : {
1079 205255 : return container_of(cfs_rq, struct rq, cfs);
1080 : }
1081 : #endif
1082 :
1083 313536 : static inline int cpu_of(struct rq *rq)
1084 : {
1085 : #ifdef CONFIG_SMP
1086 307512 : return rq->cpu;
1087 : #else
1088 : return 0;
1089 : #endif
1090 : }
1091 :
1092 : #define MDF_PUSH 0x01
1093 :
1094 27005 : static inline bool is_migration_disabled(struct task_struct *p)
1095 : {
1096 : #ifdef CONFIG_SMP
1097 27005 : return p->migration_disabled;
1098 : #else
1099 : return false;
1100 : #endif
1101 : }
1102 :
1103 : #ifdef CONFIG_SCHED_SMT
1104 : extern void __update_idle_core(struct rq *rq);
1105 :
1106 6965 : static inline void update_idle_core(struct rq *rq)
1107 : {
1108 6965 : if (static_branch_unlikely(&sched_smt_present))
1109 0 : __update_idle_core(rq);
1110 6967 : }
1111 :
1112 : #else
1113 : static inline void update_idle_core(struct rq *rq) { }
1114 : #endif
1115 :
1116 : DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1117 :
1118 : #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1119 : #define this_rq() this_cpu_ptr(&runqueues)
1120 : #define task_rq(p) cpu_rq(task_cpu(p))
1121 : #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1122 : #define raw_rq() raw_cpu_ptr(&runqueues)
1123 :
1124 : extern void update_rq_clock(struct rq *rq);
1125 :
1126 : static inline u64 __rq_clock_broken(struct rq *rq)
1127 : {
1128 : return READ_ONCE(rq->clock);
1129 : }
1130 :
1131 : /*
1132 : * rq::clock_update_flags bits
1133 : *
1134 : * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1135 : * call to __schedule(). This is an optimisation to avoid
1136 : * neighbouring rq clock updates.
1137 : *
1138 : * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1139 : * in effect and calls to update_rq_clock() are being ignored.
1140 : *
1141 : * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1142 : * made to update_rq_clock() since the last time rq::lock was pinned.
1143 : *
1144 : * If inside of __schedule(), clock_update_flags will have been
1145 : * shifted left (a left shift is a cheap operation for the fast path
1146 : * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1147 : *
1148 : * if (rq-clock_update_flags >= RQCF_UPDATED)
1149 : *
1150 : * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1151 : * one position though, because the next rq_unpin_lock() will shift it
1152 : * back.
1153 : */
1154 : #define RQCF_REQ_SKIP 0x01
1155 : #define RQCF_ACT_SKIP 0x02
1156 : #define RQCF_UPDATED 0x04
1157 :
1158 335948 : static inline void assert_clock_updated(struct rq *rq)
1159 : {
1160 : /*
1161 : * The only reason for not seeing a clock update since the
1162 : * last rq_pin_lock() is if we're currently skipping updates.
1163 : */
1164 335948 : SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1165 : }
1166 :
1167 89226 : static inline u64 rq_clock(struct rq *rq)
1168 : {
1169 178465 : lockdep_assert_held(&rq->lock);
1170 89229 : assert_clock_updated(rq);
1171 :
1172 89229 : return rq->clock;
1173 : }
1174 :
1175 146457 : static inline u64 rq_clock_task(struct rq *rq)
1176 : {
1177 293254 : lockdep_assert_held(&rq->lock);
1178 146714 : assert_clock_updated(rq);
1179 :
1180 146714 : return rq->clock_task;
1181 : }
1182 :
1183 : /**
1184 : * By default the decay is the default pelt decay period.
1185 : * The decay shift can change the decay period in
1186 : * multiples of 32.
1187 : * Decay shift Decay period(ms)
1188 : * 0 32
1189 : * 1 64
1190 : * 2 128
1191 : * 3 256
1192 : * 4 512
1193 : */
1194 : extern int sched_thermal_decay_shift;
1195 :
1196 37292 : static inline u64 rq_clock_thermal(struct rq *rq)
1197 : {
1198 37292 : return rq_clock_task(rq) >> sched_thermal_decay_shift;
1199 : }
1200 :
1201 12988 : static inline void rq_clock_skip_update(struct rq *rq)
1202 : {
1203 25976 : lockdep_assert_held(&rq->lock);
1204 12987 : rq->clock_update_flags |= RQCF_REQ_SKIP;
1205 12987 : }
1206 :
1207 : /*
1208 : * See rt task throttling, which is the only time a skip
1209 : * request is cancelled.
1210 : */
1211 0 : static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1212 : {
1213 0 : lockdep_assert_held(&rq->lock);
1214 0 : rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1215 0 : }
1216 :
1217 : struct rq_flags {
1218 : unsigned long flags;
1219 : struct pin_cookie cookie;
1220 : #ifdef CONFIG_SCHED_DEBUG
1221 : /*
1222 : * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1223 : * current pin context is stashed here in case it needs to be
1224 : * restored in rq_repin_lock().
1225 : */
1226 : unsigned int clock_update_flags;
1227 : #endif
1228 : };
1229 :
1230 : extern struct callback_head balance_push_callback;
1231 :
1232 : /*
1233 : * Lockdep annotation that avoids accidental unlocks; it's like a
1234 : * sticky/continuous lockdep_assert_held().
1235 : *
1236 : * This avoids code that has access to 'struct rq *rq' (basically everything in
1237 : * the scheduler) from accidentally unlocking the rq if they do not also have a
1238 : * copy of the (on-stack) 'struct rq_flags rf'.
1239 : *
1240 : * Also see Documentation/locking/lockdep-design.rst.
1241 : */
1242 85422 : static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1243 : {
1244 18279 : rf->cookie = lockdep_pin_lock(&rq->lock);
1245 :
1246 : #ifdef CONFIG_SCHED_DEBUG
1247 : rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1248 : rf->clock_update_flags = 0;
1249 : #ifdef CONFIG_SMP
1250 : SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1251 : #endif
1252 : #endif
1253 : }
1254 :
1255 92427 : static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1256 : {
1257 : #ifdef CONFIG_SCHED_DEBUG
1258 : if (rq->clock_update_flags > RQCF_ACT_SKIP)
1259 : rf->clock_update_flags = RQCF_UPDATED;
1260 : #endif
1261 :
1262 35422 : lockdep_unpin_lock(&rq->lock, rf->cookie);
1263 26511 : }
1264 :
1265 7457 : static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1266 : {
1267 7457 : lockdep_repin_lock(&rq->lock, rf->cookie);
1268 :
1269 : #ifdef CONFIG_SCHED_DEBUG
1270 : /*
1271 : * Restore the value we stashed in @rf for this pin context.
1272 : */
1273 : rq->clock_update_flags |= rf->clock_update_flags;
1274 : #endif
1275 36 : }
1276 :
1277 : struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1278 : __acquires(rq->lock);
1279 :
1280 : struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1281 : __acquires(p->pi_lock)
1282 : __acquires(rq->lock);
1283 :
1284 171 : static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1285 : __releases(rq->lock)
1286 : {
1287 171 : rq_unpin_lock(rq, rf);
1288 171 : raw_spin_unlock(&rq->lock);
1289 : }
1290 :
1291 : static inline void
1292 1279 : task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1293 : __releases(rq->lock)
1294 : __releases(p->pi_lock)
1295 : {
1296 1279 : rq_unpin_lock(rq, rf);
1297 1279 : raw_spin_unlock(&rq->lock);
1298 1279 : raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1299 1279 : }
1300 :
1301 : static inline void
1302 19125 : rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1303 : __acquires(rq->lock)
1304 : {
1305 19125 : raw_spin_lock_irqsave(&rq->lock, rf->flags);
1306 19222 : rq_pin_lock(rq, rf);
1307 19198 : }
1308 :
1309 : static inline void
1310 1 : rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1311 : __acquires(rq->lock)
1312 : {
1313 1 : raw_spin_lock_irq(&rq->lock);
1314 1 : rq_pin_lock(rq, rf);
1315 : }
1316 :
1317 : static inline void
1318 63874 : rq_lock(struct rq *rq, struct rq_flags *rf)
1319 : __acquires(rq->lock)
1320 : {
1321 63874 : raw_spin_lock(&rq->lock);
1322 64780 : rq_pin_lock(rq, rf);
1323 : }
1324 :
1325 : static inline void
1326 : rq_relock(struct rq *rq, struct rq_flags *rf)
1327 : __acquires(rq->lock)
1328 : {
1329 : raw_spin_lock(&rq->lock);
1330 : rq_repin_lock(rq, rf);
1331 : }
1332 :
1333 : static inline void
1334 16552 : rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1335 : __releases(rq->lock)
1336 : {
1337 16552 : rq_unpin_lock(rq, rf);
1338 16597 : raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1339 16574 : }
1340 :
1341 : static inline void
1342 0 : rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1343 : __releases(rq->lock)
1344 : {
1345 0 : rq_unpin_lock(rq, rf);
1346 0 : raw_spin_unlock_irq(&rq->lock);
1347 : }
1348 :
1349 : static inline void
1350 39003 : rq_unlock(struct rq *rq, struct rq_flags *rf)
1351 : __releases(rq->lock)
1352 : {
1353 39003 : rq_unpin_lock(rq, rf);
1354 39289 : raw_spin_unlock(&rq->lock);
1355 0 : }
1356 :
1357 : static inline struct rq *
1358 0 : this_rq_lock_irq(struct rq_flags *rf)
1359 : __acquires(rq->lock)
1360 : {
1361 0 : struct rq *rq;
1362 :
1363 0 : local_irq_disable();
1364 0 : rq = this_rq();
1365 0 : rq_lock(rq, rf);
1366 0 : return rq;
1367 : }
1368 :
1369 : #ifdef CONFIG_NUMA
1370 : enum numa_topology_type {
1371 : NUMA_DIRECT,
1372 : NUMA_GLUELESS_MESH,
1373 : NUMA_BACKPLANE,
1374 : };
1375 : extern enum numa_topology_type sched_numa_topology_type;
1376 : extern int sched_max_numa_distance;
1377 : extern bool find_numa_distance(int distance);
1378 : extern void sched_init_numa(void);
1379 : extern void sched_domains_numa_masks_set(unsigned int cpu);
1380 : extern void sched_domains_numa_masks_clear(unsigned int cpu);
1381 : extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1382 : #else
1383 : static inline void sched_init_numa(void) { }
1384 : static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1385 : static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1386 : static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1387 : {
1388 : return nr_cpu_ids;
1389 : }
1390 : #endif
1391 :
1392 : #ifdef CONFIG_NUMA_BALANCING
1393 : /* The regions in numa_faults array from task_struct */
1394 : enum numa_faults_stats {
1395 : NUMA_MEM = 0,
1396 : NUMA_CPU,
1397 : NUMA_MEMBUF,
1398 : NUMA_CPUBUF
1399 : };
1400 : extern void sched_setnuma(struct task_struct *p, int node);
1401 : extern int migrate_task_to(struct task_struct *p, int cpu);
1402 : extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1403 : int cpu, int scpu);
1404 : extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1405 : #else
1406 : static inline void
1407 990 : init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1408 : {
1409 990 : }
1410 : #endif /* CONFIG_NUMA_BALANCING */
1411 :
1412 : #ifdef CONFIG_SMP
1413 :
1414 : static inline void
1415 0 : queue_balance_callback(struct rq *rq,
1416 : struct callback_head *head,
1417 : void (*func)(struct rq *rq))
1418 : {
1419 0 : lockdep_assert_held(&rq->lock);
1420 :
1421 0 : if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1422 : return;
1423 :
1424 0 : head->func = (void (*)(struct callback_head *))func;
1425 0 : head->next = rq->balance_callback;
1426 0 : rq->balance_callback = head;
1427 : }
1428 :
1429 : #define rcu_dereference_check_sched_domain(p) \
1430 : rcu_dereference_check((p), \
1431 : lockdep_is_held(&sched_domains_mutex))
1432 :
1433 : /*
1434 : * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1435 : * See destroy_sched_domains: call_rcu for details.
1436 : *
1437 : * The domain tree of any CPU may only be accessed from within
1438 : * preempt-disabled sections.
1439 : */
1440 : #define for_each_domain(cpu, __sd) \
1441 : for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1442 : __sd; __sd = __sd->parent)
1443 :
1444 : /**
1445 : * highest_flag_domain - Return highest sched_domain containing flag.
1446 : * @cpu: The CPU whose highest level of sched domain is to
1447 : * be returned.
1448 : * @flag: The flag to check for the highest sched_domain
1449 : * for the given CPU.
1450 : *
1451 : * Returns the highest sched_domain of a CPU which contains the given flag.
1452 : */
1453 8 : static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1454 : {
1455 8 : struct sched_domain *sd, *hsd = NULL;
1456 :
1457 16 : for_each_domain(cpu, sd) {
1458 8 : if (!(sd->flags & flag))
1459 : break;
1460 0 : hsd = sd;
1461 : }
1462 :
1463 8 : return hsd;
1464 : }
1465 :
1466 8 : static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1467 : {
1468 8 : struct sched_domain *sd;
1469 :
1470 24 : for_each_domain(cpu, sd) {
1471 8 : if (sd->flags & flag)
1472 : break;
1473 : }
1474 :
1475 8 : return sd;
1476 : }
1477 :
1478 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1479 : DECLARE_PER_CPU(int, sd_llc_size);
1480 : DECLARE_PER_CPU(int, sd_llc_id);
1481 : DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1482 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1483 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1484 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1485 : extern struct static_key_false sched_asym_cpucapacity;
1486 :
1487 : struct sched_group_capacity {
1488 : atomic_t ref;
1489 : /*
1490 : * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1491 : * for a single CPU.
1492 : */
1493 : unsigned long capacity;
1494 : unsigned long min_capacity; /* Min per-CPU capacity in group */
1495 : unsigned long max_capacity; /* Max per-CPU capacity in group */
1496 : unsigned long next_update;
1497 : int imbalance; /* XXX unrelated to capacity but shared group state */
1498 :
1499 : #ifdef CONFIG_SCHED_DEBUG
1500 : int id;
1501 : #endif
1502 :
1503 : unsigned long cpumask[]; /* Balance mask */
1504 : };
1505 :
1506 : struct sched_group {
1507 : struct sched_group *next; /* Must be a circular list */
1508 : atomic_t ref;
1509 :
1510 : unsigned int group_weight;
1511 : struct sched_group_capacity *sgc;
1512 : int asym_prefer_cpu; /* CPU of highest priority in group */
1513 :
1514 : /*
1515 : * The CPUs this group covers.
1516 : *
1517 : * NOTE: this field is variable length. (Allocated dynamically
1518 : * by attaching extra space to the end of the structure,
1519 : * depending on how many CPUs the kernel has booted up with)
1520 : */
1521 : unsigned long cpumask[];
1522 : };
1523 :
1524 190951 : static inline struct cpumask *sched_group_span(struct sched_group *sg)
1525 : {
1526 184607 : return to_cpumask(sg->cpumask);
1527 : }
1528 :
1529 : /*
1530 : * See build_balance_mask().
1531 : */
1532 9844 : static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1533 : {
1534 9844 : return to_cpumask(sg->sgc->cpumask);
1535 : }
1536 :
1537 : /**
1538 : * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1539 : * @group: The group whose first CPU is to be returned.
1540 : */
1541 : static inline unsigned int group_first_cpu(struct sched_group *group)
1542 : {
1543 : return cpumask_first(sched_group_span(group));
1544 : }
1545 :
1546 : extern int group_balance_cpu(struct sched_group *sg);
1547 :
1548 : #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1549 : void register_sched_domain_sysctl(void);
1550 : void dirty_sched_domain_sysctl(int cpu);
1551 : void unregister_sched_domain_sysctl(void);
1552 : #else
1553 1 : static inline void register_sched_domain_sysctl(void)
1554 : {
1555 1 : }
1556 4 : static inline void dirty_sched_domain_sysctl(int cpu)
1557 : {
1558 4 : }
1559 0 : static inline void unregister_sched_domain_sysctl(void)
1560 : {
1561 0 : }
1562 : #endif
1563 :
1564 : extern void flush_smp_call_function_from_idle(void);
1565 :
1566 : #else /* !CONFIG_SMP: */
1567 : static inline void flush_smp_call_function_from_idle(void) { }
1568 : #endif
1569 :
1570 : #include "stats.h"
1571 : #include "autogroup.h"
1572 :
1573 : #ifdef CONFIG_CGROUP_SCHED
1574 :
1575 : /*
1576 : * Return the group to which this tasks belongs.
1577 : *
1578 : * We cannot use task_css() and friends because the cgroup subsystem
1579 : * changes that value before the cgroup_subsys::attach() method is called,
1580 : * therefore we cannot pin it and might observe the wrong value.
1581 : *
1582 : * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1583 : * core changes this before calling sched_move_task().
1584 : *
1585 : * Instead we use a 'copy' which is updated from sched_move_task() while
1586 : * holding both task_struct::pi_lock and rq::lock.
1587 : */
1588 : static inline struct task_group *task_group(struct task_struct *p)
1589 : {
1590 : return p->sched_task_group;
1591 : }
1592 :
1593 : /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1594 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1595 : {
1596 : #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1597 : struct task_group *tg = task_group(p);
1598 : #endif
1599 :
1600 : #ifdef CONFIG_FAIR_GROUP_SCHED
1601 : set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1602 : p->se.cfs_rq = tg->cfs_rq[cpu];
1603 : p->se.parent = tg->se[cpu];
1604 : #endif
1605 :
1606 : #ifdef CONFIG_RT_GROUP_SCHED
1607 : p->rt.rt_rq = tg->rt_rq[cpu];
1608 : p->rt.parent = tg->rt_se[cpu];
1609 : #endif
1610 : }
1611 :
1612 : #else /* CONFIG_CGROUP_SCHED */
1613 :
1614 2962 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1615 5145 : static inline struct task_group *task_group(struct task_struct *p)
1616 : {
1617 5145 : return NULL;
1618 : }
1619 :
1620 : #endif /* CONFIG_CGROUP_SCHED */
1621 :
1622 2962 : static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1623 : {
1624 2962 : set_task_rq(p, cpu);
1625 : #ifdef CONFIG_SMP
1626 : /*
1627 : * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1628 : * successfully executed on another CPU. We must ensure that updates of
1629 : * per-task data have been completed by this moment.
1630 : */
1631 2962 : smp_wmb();
1632 : #ifdef CONFIG_THREAD_INFO_IN_TASK
1633 2962 : WRITE_ONCE(p->cpu, cpu);
1634 : #else
1635 : WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1636 : #endif
1637 1967 : p->wake_cpu = cpu;
1638 : #endif
1639 : }
1640 :
1641 : /*
1642 : * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1643 : */
1644 : #ifdef CONFIG_SCHED_DEBUG
1645 : # include <linux/static_key.h>
1646 : # define const_debug __read_mostly
1647 : #else
1648 : # define const_debug const
1649 : #endif
1650 :
1651 : #define SCHED_FEAT(name, enabled) \
1652 : __SCHED_FEAT_##name ,
1653 :
1654 : enum {
1655 : #include "features.h"
1656 : __SCHED_FEAT_NR,
1657 : };
1658 :
1659 : #undef SCHED_FEAT
1660 :
1661 : #ifdef CONFIG_SCHED_DEBUG
1662 :
1663 : /*
1664 : * To support run-time toggling of sched features, all the translation units
1665 : * (but core.c) reference the sysctl_sched_features defined in core.c.
1666 : */
1667 : extern const_debug unsigned int sysctl_sched_features;
1668 :
1669 : #ifdef CONFIG_JUMP_LABEL
1670 : #define SCHED_FEAT(name, enabled) \
1671 : static __always_inline bool static_branch_##name(struct static_key *key) \
1672 : { \
1673 : return static_key_##enabled(key); \
1674 : }
1675 :
1676 : #include "features.h"
1677 : #undef SCHED_FEAT
1678 :
1679 : extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1680 : #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1681 :
1682 : #else /* !CONFIG_JUMP_LABEL */
1683 :
1684 : #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1685 :
1686 : #endif /* CONFIG_JUMP_LABEL */
1687 :
1688 : #else /* !SCHED_DEBUG */
1689 :
1690 : /*
1691 : * Each translation unit has its own copy of sysctl_sched_features to allow
1692 : * constants propagation at compile time and compiler optimization based on
1693 : * features default.
1694 : */
1695 : #define SCHED_FEAT(name, enabled) \
1696 : (1UL << __SCHED_FEAT_##name) * enabled |
1697 : static const_debug __maybe_unused unsigned int sysctl_sched_features =
1698 : #include "features.h"
1699 : 0;
1700 : #undef SCHED_FEAT
1701 :
1702 : #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1703 :
1704 : #endif /* SCHED_DEBUG */
1705 :
1706 : extern struct static_key_false sched_numa_balancing;
1707 : extern struct static_key_false sched_schedstats;
1708 :
1709 12 : static inline u64 global_rt_period(void)
1710 : {
1711 6 : return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1712 : }
1713 :
1714 18 : static inline u64 global_rt_runtime(void)
1715 : {
1716 8 : if (sysctl_sched_rt_runtime < 0)
1717 : return RUNTIME_INF;
1718 :
1719 12 : return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1720 : }
1721 :
1722 156 : static inline int task_current(struct rq *rq, struct task_struct *p)
1723 : {
1724 156 : return rq->curr == p;
1725 : }
1726 :
1727 3978 : static inline int task_running(struct rq *rq, struct task_struct *p)
1728 : {
1729 : #ifdef CONFIG_SMP
1730 3978 : return p->on_cpu;
1731 : #else
1732 : return task_current(rq, p);
1733 : #endif
1734 : }
1735 :
1736 44378 : static inline int task_on_rq_queued(struct task_struct *p)
1737 : {
1738 44298 : return p->on_rq == TASK_ON_RQ_QUEUED;
1739 : }
1740 :
1741 1419 : static inline int task_on_rq_migrating(struct task_struct *p)
1742 : {
1743 1419 : return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1744 : }
1745 :
1746 : /* Wake flags. The first three directly map to some SD flag value */
1747 : #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
1748 : #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
1749 : #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
1750 :
1751 : #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
1752 : #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
1753 : #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
1754 :
1755 : #ifdef CONFIG_SMP
1756 : static_assert(WF_EXEC == SD_BALANCE_EXEC);
1757 : static_assert(WF_FORK == SD_BALANCE_FORK);
1758 : static_assert(WF_TTWU == SD_BALANCE_WAKE);
1759 : #endif
1760 :
1761 : /*
1762 : * To aid in avoiding the subversion of "niceness" due to uneven distribution
1763 : * of tasks with abnormal "nice" values across CPUs the contribution that
1764 : * each task makes to its run queue's load is weighted according to its
1765 : * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1766 : * scaled version of the new time slice allocation that they receive on time
1767 : * slice expiry etc.
1768 : */
1769 :
1770 : #define WEIGHT_IDLEPRIO 3
1771 : #define WMULT_IDLEPRIO 1431655765
1772 :
1773 : extern const int sched_prio_to_weight[40];
1774 : extern const u32 sched_prio_to_wmult[40];
1775 :
1776 : /*
1777 : * {de,en}queue flags:
1778 : *
1779 : * DEQUEUE_SLEEP - task is no longer runnable
1780 : * ENQUEUE_WAKEUP - task just became runnable
1781 : *
1782 : * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1783 : * are in a known state which allows modification. Such pairs
1784 : * should preserve as much state as possible.
1785 : *
1786 : * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1787 : * in the runqueue.
1788 : *
1789 : * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1790 : * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1791 : * ENQUEUE_MIGRATED - the task was migrated during wakeup
1792 : *
1793 : */
1794 :
1795 : #define DEQUEUE_SLEEP 0x01
1796 : #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
1797 : #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
1798 : #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
1799 :
1800 : #define ENQUEUE_WAKEUP 0x01
1801 : #define ENQUEUE_RESTORE 0x02
1802 : #define ENQUEUE_MOVE 0x04
1803 : #define ENQUEUE_NOCLOCK 0x08
1804 :
1805 : #define ENQUEUE_HEAD 0x10
1806 : #define ENQUEUE_REPLENISH 0x20
1807 : #ifdef CONFIG_SMP
1808 : #define ENQUEUE_MIGRATED 0x40
1809 : #else
1810 : #define ENQUEUE_MIGRATED 0x00
1811 : #endif
1812 :
1813 : #define RETRY_TASK ((void *)-1UL)
1814 :
1815 : struct sched_class {
1816 :
1817 : #ifdef CONFIG_UCLAMP_TASK
1818 : int uclamp_enabled;
1819 : #endif
1820 :
1821 : void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1822 : void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1823 : void (*yield_task) (struct rq *rq);
1824 : bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
1825 :
1826 : void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1827 :
1828 : struct task_struct *(*pick_next_task)(struct rq *rq);
1829 :
1830 : void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1831 : void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
1832 :
1833 : #ifdef CONFIG_SMP
1834 : int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1835 : int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
1836 : void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1837 :
1838 : void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1839 :
1840 : void (*set_cpus_allowed)(struct task_struct *p,
1841 : const struct cpumask *newmask,
1842 : u32 flags);
1843 :
1844 : void (*rq_online)(struct rq *rq);
1845 : void (*rq_offline)(struct rq *rq);
1846 :
1847 : struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
1848 : #endif
1849 :
1850 : void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1851 : void (*task_fork)(struct task_struct *p);
1852 : void (*task_dead)(struct task_struct *p);
1853 :
1854 : /*
1855 : * The switched_from() call is allowed to drop rq->lock, therefore we
1856 : * cannot assume the switched_from/switched_to pair is serliazed by
1857 : * rq->lock. They are however serialized by p->pi_lock.
1858 : */
1859 : void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1860 : void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1861 : void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1862 : int oldprio);
1863 :
1864 : unsigned int (*get_rr_interval)(struct rq *rq,
1865 : struct task_struct *task);
1866 :
1867 : void (*update_curr)(struct rq *rq);
1868 :
1869 : #define TASK_SET_GROUP 0
1870 : #define TASK_MOVE_GROUP 1
1871 :
1872 : #ifdef CONFIG_FAIR_GROUP_SCHED
1873 : void (*task_change_group)(struct task_struct *p, int type);
1874 : #endif
1875 : };
1876 :
1877 28065 : static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1878 : {
1879 28065 : WARN_ON_ONCE(rq->curr != prev);
1880 28065 : prev->sched_class->put_prev_task(rq, prev);
1881 28065 : }
1882 :
1883 15 : static inline void set_next_task(struct rq *rq, struct task_struct *next)
1884 : {
1885 15 : WARN_ON_ONCE(rq->curr != next);
1886 15 : next->sched_class->set_next_task(rq, next, false);
1887 15 : }
1888 :
1889 :
1890 : /*
1891 : * Helper to define a sched_class instance; each one is placed in a separate
1892 : * section which is ordered by the linker script:
1893 : *
1894 : * include/asm-generic/vmlinux.lds.h
1895 : *
1896 : * Also enforce alignment on the instance, not the type, to guarantee layout.
1897 : */
1898 : #define DEFINE_SCHED_CLASS(name) \
1899 : const struct sched_class name##_sched_class \
1900 : __aligned(__alignof__(struct sched_class)) \
1901 : __section("__" #name "_sched_class")
1902 :
1903 : /* Defined in include/asm-generic/vmlinux.lds.h */
1904 : extern struct sched_class __begin_sched_classes[];
1905 : extern struct sched_class __end_sched_classes[];
1906 :
1907 : #define sched_class_highest (__end_sched_classes - 1)
1908 : #define sched_class_lowest (__begin_sched_classes - 1)
1909 :
1910 : #define for_class_range(class, _from, _to) \
1911 : for (class = (_from); class != (_to); class--)
1912 :
1913 : #define for_each_class(class) \
1914 : for_class_range(class, sched_class_highest, sched_class_lowest)
1915 :
1916 : extern const struct sched_class stop_sched_class;
1917 : extern const struct sched_class dl_sched_class;
1918 : extern const struct sched_class rt_sched_class;
1919 : extern const struct sched_class fair_sched_class;
1920 : extern const struct sched_class idle_sched_class;
1921 :
1922 197 : static inline bool sched_stop_runnable(struct rq *rq)
1923 : {
1924 316 : return rq->stop && task_on_rq_queued(rq->stop);
1925 : }
1926 :
1927 119 : static inline bool sched_dl_runnable(struct rq *rq)
1928 : {
1929 119 : return rq->dl.dl_nr_running > 0;
1930 : }
1931 :
1932 78 : static inline bool sched_rt_runnable(struct rq *rq)
1933 : {
1934 78 : return rq->rt.rt_queued > 0;
1935 : }
1936 :
1937 28495 : static inline bool sched_fair_runnable(struct rq *rq)
1938 : {
1939 28495 : return rq->cfs.nr_running > 0;
1940 : }
1941 :
1942 : extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
1943 : extern struct task_struct *pick_next_task_idle(struct rq *rq);
1944 :
1945 : #define SCA_CHECK 0x01
1946 : #define SCA_MIGRATE_DISABLE 0x02
1947 : #define SCA_MIGRATE_ENABLE 0x04
1948 :
1949 : #ifdef CONFIG_SMP
1950 :
1951 : extern void update_group_capacity(struct sched_domain *sd, int cpu);
1952 :
1953 : extern void trigger_load_balance(struct rq *rq);
1954 :
1955 : extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
1956 :
1957 0 : static inline struct task_struct *get_push_task(struct rq *rq)
1958 : {
1959 0 : struct task_struct *p = rq->curr;
1960 :
1961 0 : lockdep_assert_held(&rq->lock);
1962 :
1963 0 : if (rq->push_busy)
1964 : return NULL;
1965 :
1966 0 : if (p->nr_cpus_allowed == 1)
1967 : return NULL;
1968 :
1969 0 : rq->push_busy = true;
1970 0 : return get_task_struct(p);
1971 : }
1972 :
1973 : extern int push_cpu_stop(void *arg);
1974 :
1975 : #endif
1976 :
1977 : #ifdef CONFIG_CPU_IDLE
1978 : static inline void idle_set_state(struct rq *rq,
1979 : struct cpuidle_state *idle_state)
1980 : {
1981 : rq->idle_state = idle_state;
1982 : }
1983 :
1984 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1985 : {
1986 : SCHED_WARN_ON(!rcu_read_lock_held());
1987 :
1988 : return rq->idle_state;
1989 : }
1990 : #else
1991 0 : static inline void idle_set_state(struct rq *rq,
1992 : struct cpuidle_state *idle_state)
1993 : {
1994 0 : }
1995 :
1996 0 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1997 : {
1998 0 : return NULL;
1999 : }
2000 : #endif
2001 :
2002 : extern void schedule_idle(void);
2003 :
2004 : extern void sysrq_sched_debug_show(void);
2005 : extern void sched_init_granularity(void);
2006 : extern void update_max_interval(void);
2007 :
2008 : extern void init_sched_dl_class(void);
2009 : extern void init_sched_rt_class(void);
2010 : extern void init_sched_fair_class(void);
2011 :
2012 : extern void reweight_task(struct task_struct *p, int prio);
2013 :
2014 : extern void resched_curr(struct rq *rq);
2015 : extern void resched_cpu(int cpu);
2016 :
2017 : extern struct rt_bandwidth def_rt_bandwidth;
2018 : extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2019 :
2020 : extern struct dl_bandwidth def_dl_bandwidth;
2021 : extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2022 : extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2023 : extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2024 :
2025 : #define BW_SHIFT 20
2026 : #define BW_UNIT (1 << BW_SHIFT)
2027 : #define RATIO_SHIFT 8
2028 : #define MAX_BW_BITS (64 - BW_SHIFT)
2029 : #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2030 : unsigned long to_ratio(u64 period, u64 runtime);
2031 :
2032 : extern void init_entity_runnable_average(struct sched_entity *se);
2033 : extern void post_init_entity_util_avg(struct task_struct *p);
2034 :
2035 : #ifdef CONFIG_NO_HZ_FULL
2036 : extern bool sched_can_stop_tick(struct rq *rq);
2037 : extern int __init sched_tick_offload_init(void);
2038 :
2039 : /*
2040 : * Tick may be needed by tasks in the runqueue depending on their policy and
2041 : * requirements. If tick is needed, lets send the target an IPI to kick it out of
2042 : * nohz mode if necessary.
2043 : */
2044 : static inline void sched_update_tick_dependency(struct rq *rq)
2045 : {
2046 : int cpu = cpu_of(rq);
2047 :
2048 : if (!tick_nohz_full_cpu(cpu))
2049 : return;
2050 :
2051 : if (sched_can_stop_tick(rq))
2052 : tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2053 : else
2054 : tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2055 : }
2056 : #else
2057 : static inline int sched_tick_offload_init(void) { return 0; }
2058 31614 : static inline void sched_update_tick_dependency(struct rq *rq) { }
2059 : #endif
2060 :
2061 15802 : static inline void add_nr_running(struct rq *rq, unsigned count)
2062 : {
2063 15802 : unsigned prev_nr = rq->nr_running;
2064 :
2065 15802 : rq->nr_running = prev_nr + count;
2066 15802 : if (trace_sched_update_nr_running_tp_enabled()) {
2067 0 : call_trace_sched_update_nr_running(rq, count);
2068 : }
2069 :
2070 : #ifdef CONFIG_SMP
2071 15805 : if (prev_nr < 2 && rq->nr_running >= 2) {
2072 6005 : if (!READ_ONCE(rq->rd->overload))
2073 2027 : WRITE_ONCE(rq->rd->overload, 1);
2074 : }
2075 : #endif
2076 :
2077 15805 : sched_update_tick_dependency(rq);
2078 15805 : }
2079 :
2080 15808 : static inline void sub_nr_running(struct rq *rq, unsigned count)
2081 : {
2082 15808 : rq->nr_running -= count;
2083 15808 : if (trace_sched_update_nr_running_tp_enabled()) {
2084 0 : call_trace_sched_update_nr_running(rq, -count);
2085 : }
2086 :
2087 : /* Check if we still need preemption */
2088 15809 : sched_update_tick_dependency(rq);
2089 15809 : }
2090 :
2091 : extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2092 : extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2093 :
2094 : extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2095 :
2096 : extern const_debug unsigned int sysctl_sched_nr_migrate;
2097 : extern const_debug unsigned int sysctl_sched_migration_cost;
2098 :
2099 : #ifdef CONFIG_SCHED_HRTICK
2100 :
2101 : /*
2102 : * Use hrtick when:
2103 : * - enabled by features
2104 : * - hrtimer is actually high res
2105 : */
2106 : static inline int hrtick_enabled(struct rq *rq)
2107 : {
2108 : if (!cpu_active(cpu_of(rq)))
2109 : return 0;
2110 : return hrtimer_is_hres_active(&rq->hrtick_timer);
2111 : }
2112 :
2113 : static inline int hrtick_enabled_fair(struct rq *rq)
2114 : {
2115 : if (!sched_feat(HRTICK))
2116 : return 0;
2117 : return hrtick_enabled(rq);
2118 : }
2119 :
2120 : static inline int hrtick_enabled_dl(struct rq *rq)
2121 : {
2122 : if (!sched_feat(HRTICK_DL))
2123 : return 0;
2124 : return hrtick_enabled(rq);
2125 : }
2126 :
2127 : void hrtick_start(struct rq *rq, u64 delay);
2128 :
2129 : #else
2130 :
2131 21076 : static inline int hrtick_enabled_fair(struct rq *rq)
2132 : {
2133 21076 : return 0;
2134 : }
2135 :
2136 0 : static inline int hrtick_enabled_dl(struct rq *rq)
2137 : {
2138 0 : return 0;
2139 : }
2140 :
2141 : static inline int hrtick_enabled(struct rq *rq)
2142 : {
2143 : return 0;
2144 : }
2145 :
2146 : #endif /* CONFIG_SCHED_HRTICK */
2147 :
2148 : #ifndef arch_scale_freq_tick
2149 : static __always_inline
2150 : void arch_scale_freq_tick(void)
2151 : {
2152 : }
2153 : #endif
2154 :
2155 : #ifndef arch_scale_freq_capacity
2156 : /**
2157 : * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2158 : * @cpu: the CPU in question.
2159 : *
2160 : * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2161 : *
2162 : * f_curr
2163 : * ------ * SCHED_CAPACITY_SCALE
2164 : * f_max
2165 : */
2166 : static __always_inline
2167 : unsigned long arch_scale_freq_capacity(int cpu)
2168 : {
2169 : return SCHED_CAPACITY_SCALE;
2170 : }
2171 : #endif
2172 :
2173 : #ifdef CONFIG_SMP
2174 : #ifdef CONFIG_PREEMPTION
2175 :
2176 : static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
2177 :
2178 : /*
2179 : * fair double_lock_balance: Safely acquires both rq->locks in a fair
2180 : * way at the expense of forcing extra atomic operations in all
2181 : * invocations. This assures that the double_lock is acquired using the
2182 : * same underlying policy as the spinlock_t on this architecture, which
2183 : * reduces latency compared to the unfair variant below. However, it
2184 : * also adds more overhead and therefore may reduce throughput.
2185 : */
2186 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2187 : __releases(this_rq->lock)
2188 : __acquires(busiest->lock)
2189 : __acquires(this_rq->lock)
2190 : {
2191 : raw_spin_unlock(&this_rq->lock);
2192 : double_rq_lock(this_rq, busiest);
2193 :
2194 : return 1;
2195 : }
2196 :
2197 : #else
2198 : /*
2199 : * Unfair double_lock_balance: Optimizes throughput at the expense of
2200 : * latency by eliminating extra atomic operations when the locks are
2201 : * already in proper order on entry. This favors lower CPU-ids and will
2202 : * grant the double lock to lower CPUs over higher ids under contention,
2203 : * regardless of entry order into the function.
2204 : */
2205 0 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2206 : __releases(this_rq->lock)
2207 : __acquires(busiest->lock)
2208 : __acquires(this_rq->lock)
2209 : {
2210 0 : int ret = 0;
2211 :
2212 0 : if (unlikely(!raw_spin_trylock(&busiest->lock))) {
2213 0 : if (busiest < this_rq) {
2214 0 : raw_spin_unlock(&this_rq->lock);
2215 0 : raw_spin_lock(&busiest->lock);
2216 0 : raw_spin_lock_nested(&this_rq->lock,
2217 : SINGLE_DEPTH_NESTING);
2218 0 : ret = 1;
2219 : } else
2220 0 : raw_spin_lock_nested(&busiest->lock,
2221 : SINGLE_DEPTH_NESTING);
2222 : }
2223 0 : return ret;
2224 : }
2225 :
2226 : #endif /* CONFIG_PREEMPTION */
2227 :
2228 : /*
2229 : * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2230 : */
2231 0 : static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2232 : {
2233 0 : if (unlikely(!irqs_disabled())) {
2234 : /* printk() doesn't work well under rq->lock */
2235 0 : raw_spin_unlock(&this_rq->lock);
2236 0 : BUG_ON(1);
2237 : }
2238 :
2239 0 : return _double_lock_balance(this_rq, busiest);
2240 : }
2241 :
2242 0 : static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2243 : __releases(busiest->lock)
2244 : {
2245 0 : raw_spin_unlock(&busiest->lock);
2246 0 : lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2247 0 : }
2248 :
2249 : static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2250 : {
2251 : if (l1 > l2)
2252 : swap(l1, l2);
2253 :
2254 : spin_lock(l1);
2255 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2256 : }
2257 :
2258 : static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2259 : {
2260 : if (l1 > l2)
2261 : swap(l1, l2);
2262 :
2263 : spin_lock_irq(l1);
2264 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2265 : }
2266 :
2267 : static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2268 : {
2269 : if (l1 > l2)
2270 : swap(l1, l2);
2271 :
2272 : raw_spin_lock(l1);
2273 : raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2274 : }
2275 :
2276 : /*
2277 : * double_rq_lock - safely lock two runqueues
2278 : *
2279 : * Note this does not disable interrupts like task_rq_lock,
2280 : * you need to do so manually before calling.
2281 : */
2282 0 : static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2283 : __acquires(rq1->lock)
2284 : __acquires(rq2->lock)
2285 : {
2286 0 : BUG_ON(!irqs_disabled());
2287 0 : if (rq1 == rq2) {
2288 0 : raw_spin_lock(&rq1->lock);
2289 0 : __acquire(rq2->lock); /* Fake it out ;) */
2290 : } else {
2291 0 : if (rq1 < rq2) {
2292 0 : raw_spin_lock(&rq1->lock);
2293 0 : raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2294 : } else {
2295 0 : raw_spin_lock(&rq2->lock);
2296 0 : raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2297 : }
2298 : }
2299 0 : }
2300 :
2301 : /*
2302 : * double_rq_unlock - safely unlock two runqueues
2303 : *
2304 : * Note this does not restore interrupts like task_rq_unlock,
2305 : * you need to do so manually after calling.
2306 : */
2307 0 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2308 : __releases(rq1->lock)
2309 : __releases(rq2->lock)
2310 : {
2311 0 : raw_spin_unlock(&rq1->lock);
2312 0 : if (rq1 != rq2)
2313 0 : raw_spin_unlock(&rq2->lock);
2314 : else
2315 0 : __release(rq2->lock);
2316 0 : }
2317 :
2318 : extern void set_rq_online (struct rq *rq);
2319 : extern void set_rq_offline(struct rq *rq);
2320 : extern bool sched_smp_initialized;
2321 :
2322 : #else /* CONFIG_SMP */
2323 :
2324 : /*
2325 : * double_rq_lock - safely lock two runqueues
2326 : *
2327 : * Note this does not disable interrupts like task_rq_lock,
2328 : * you need to do so manually before calling.
2329 : */
2330 : static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2331 : __acquires(rq1->lock)
2332 : __acquires(rq2->lock)
2333 : {
2334 : BUG_ON(!irqs_disabled());
2335 : BUG_ON(rq1 != rq2);
2336 : raw_spin_lock(&rq1->lock);
2337 : __acquire(rq2->lock); /* Fake it out ;) */
2338 : }
2339 :
2340 : /*
2341 : * double_rq_unlock - safely unlock two runqueues
2342 : *
2343 : * Note this does not restore interrupts like task_rq_unlock,
2344 : * you need to do so manually after calling.
2345 : */
2346 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2347 : __releases(rq1->lock)
2348 : __releases(rq2->lock)
2349 : {
2350 : BUG_ON(rq1 != rq2);
2351 : raw_spin_unlock(&rq1->lock);
2352 : __release(rq2->lock);
2353 : }
2354 :
2355 : #endif
2356 :
2357 : extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2358 : extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2359 :
2360 : #ifdef CONFIG_SCHED_DEBUG
2361 : extern bool sched_debug_enabled;
2362 :
2363 : extern void print_cfs_stats(struct seq_file *m, int cpu);
2364 : extern void print_rt_stats(struct seq_file *m, int cpu);
2365 : extern void print_dl_stats(struct seq_file *m, int cpu);
2366 : extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2367 : extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2368 : extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2369 : #ifdef CONFIG_NUMA_BALANCING
2370 : extern void
2371 : show_numa_stats(struct task_struct *p, struct seq_file *m);
2372 : extern void
2373 : print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2374 : unsigned long tpf, unsigned long gsf, unsigned long gpf);
2375 : #endif /* CONFIG_NUMA_BALANCING */
2376 : #endif /* CONFIG_SCHED_DEBUG */
2377 :
2378 : extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2379 : extern void init_rt_rq(struct rt_rq *rt_rq);
2380 : extern void init_dl_rq(struct dl_rq *dl_rq);
2381 :
2382 : extern void cfs_bandwidth_usage_inc(void);
2383 : extern void cfs_bandwidth_usage_dec(void);
2384 :
2385 : #ifdef CONFIG_NO_HZ_COMMON
2386 : #define NOHZ_BALANCE_KICK_BIT 0
2387 : #define NOHZ_STATS_KICK_BIT 1
2388 :
2389 : #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2390 : #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2391 :
2392 : #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2393 :
2394 : #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2395 :
2396 : extern void nohz_balance_exit_idle(struct rq *rq);
2397 : #else
2398 : static inline void nohz_balance_exit_idle(struct rq *rq) { }
2399 : #endif
2400 :
2401 :
2402 : #ifdef CONFIG_SMP
2403 : static inline
2404 0 : void __dl_update(struct dl_bw *dl_b, s64 bw)
2405 : {
2406 0 : struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2407 0 : int i;
2408 :
2409 0 : RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2410 : "sched RCU must be held");
2411 0 : for_each_cpu_and(i, rd->span, cpu_active_mask) {
2412 0 : struct rq *rq = cpu_rq(i);
2413 :
2414 0 : rq->dl.extra_bw += bw;
2415 : }
2416 0 : }
2417 : #else
2418 : static inline
2419 : void __dl_update(struct dl_bw *dl_b, s64 bw)
2420 : {
2421 : struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2422 :
2423 : dl->extra_bw += bw;
2424 : }
2425 : #endif
2426 :
2427 :
2428 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2429 : struct irqtime {
2430 : u64 total;
2431 : u64 tick_delta;
2432 : u64 irq_start_time;
2433 : struct u64_stats_sync sync;
2434 : };
2435 :
2436 : DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2437 :
2438 : /*
2439 : * Returns the irqtime minus the softirq time computed by ksoftirqd.
2440 : * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2441 : * and never move forward.
2442 : */
2443 : static inline u64 irq_time_read(int cpu)
2444 : {
2445 : struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2446 : unsigned int seq;
2447 : u64 total;
2448 :
2449 : do {
2450 : seq = __u64_stats_fetch_begin(&irqtime->sync);
2451 : total = irqtime->total;
2452 : } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2453 :
2454 : return total;
2455 : }
2456 : #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2457 :
2458 : #ifdef CONFIG_CPU_FREQ
2459 : DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2460 :
2461 : /**
2462 : * cpufreq_update_util - Take a note about CPU utilization changes.
2463 : * @rq: Runqueue to carry out the update for.
2464 : * @flags: Update reason flags.
2465 : *
2466 : * This function is called by the scheduler on the CPU whose utilization is
2467 : * being updated.
2468 : *
2469 : * It can only be called from RCU-sched read-side critical sections.
2470 : *
2471 : * The way cpufreq is currently arranged requires it to evaluate the CPU
2472 : * performance state (frequency/voltage) on a regular basis to prevent it from
2473 : * being stuck in a completely inadequate performance level for too long.
2474 : * That is not guaranteed to happen if the updates are only triggered from CFS
2475 : * and DL, though, because they may not be coming in if only RT tasks are
2476 : * active all the time (or there are RT tasks only).
2477 : *
2478 : * As a workaround for that issue, this function is called periodically by the
2479 : * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2480 : * but that really is a band-aid. Going forward it should be replaced with
2481 : * solutions targeted more specifically at RT tasks.
2482 : */
2483 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2484 : {
2485 : struct update_util_data *data;
2486 :
2487 : data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2488 : cpu_of(rq)));
2489 : if (data)
2490 : data->func(data, rq_clock(rq), flags);
2491 : }
2492 : #else
2493 2874 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2494 : #endif /* CONFIG_CPU_FREQ */
2495 :
2496 : #ifdef CONFIG_UCLAMP_TASK
2497 : unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2498 :
2499 : /**
2500 : * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2501 : * @rq: The rq to clamp against. Must not be NULL.
2502 : * @util: The util value to clamp.
2503 : * @p: The task to clamp against. Can be NULL if you want to clamp
2504 : * against @rq only.
2505 : *
2506 : * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2507 : *
2508 : * If sched_uclamp_used static key is disabled, then just return the util
2509 : * without any clamping since uclamp aggregation at the rq level in the fast
2510 : * path is disabled, rendering this operation a NOP.
2511 : *
2512 : * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2513 : * will return the correct effective uclamp value of the task even if the
2514 : * static key is disabled.
2515 : */
2516 : static __always_inline
2517 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2518 : struct task_struct *p)
2519 : {
2520 : unsigned long min_util;
2521 : unsigned long max_util;
2522 :
2523 : if (!static_branch_likely(&sched_uclamp_used))
2524 : return util;
2525 :
2526 : min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2527 : max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2528 :
2529 : if (p) {
2530 : min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2531 : max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2532 : }
2533 :
2534 : /*
2535 : * Since CPU's {min,max}_util clamps are MAX aggregated considering
2536 : * RUNNABLE tasks with _different_ clamps, we can end up with an
2537 : * inversion. Fix it now when the clamps are applied.
2538 : */
2539 : if (unlikely(min_util >= max_util))
2540 : return min_util;
2541 :
2542 : return clamp(util, min_util, max_util);
2543 : }
2544 :
2545 : /*
2546 : * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2547 : * by default in the fast path and only gets turned on once userspace performs
2548 : * an operation that requires it.
2549 : *
2550 : * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2551 : * hence is active.
2552 : */
2553 : static inline bool uclamp_is_used(void)
2554 : {
2555 : return static_branch_likely(&sched_uclamp_used);
2556 : }
2557 : #else /* CONFIG_UCLAMP_TASK */
2558 : static inline
2559 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2560 : struct task_struct *p)
2561 : {
2562 : return util;
2563 : }
2564 :
2565 0 : static inline bool uclamp_is_used(void)
2566 : {
2567 0 : return false;
2568 : }
2569 : #endif /* CONFIG_UCLAMP_TASK */
2570 :
2571 : #ifdef arch_scale_freq_capacity
2572 : # ifndef arch_scale_freq_invariant
2573 : # define arch_scale_freq_invariant() true
2574 : # endif
2575 : #else
2576 : # define arch_scale_freq_invariant() false
2577 : #endif
2578 :
2579 : #ifdef CONFIG_SMP
2580 95762 : static inline unsigned long capacity_orig_of(int cpu)
2581 : {
2582 6649 : return cpu_rq(cpu)->cpu_capacity_orig;
2583 : }
2584 :
2585 : /**
2586 : * enum cpu_util_type - CPU utilization type
2587 : * @FREQUENCY_UTIL: Utilization used to select frequency
2588 : * @ENERGY_UTIL: Utilization used during energy calculation
2589 : *
2590 : * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2591 : * need to be aggregated differently depending on the usage made of them. This
2592 : * enum is used within effective_cpu_util() to differentiate the types of
2593 : * utilization expected by the callers, and adjust the aggregation accordingly.
2594 : */
2595 : enum cpu_util_type {
2596 : FREQUENCY_UTIL,
2597 : ENERGY_UTIL,
2598 : };
2599 :
2600 : unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2601 : unsigned long max, enum cpu_util_type type,
2602 : struct task_struct *p);
2603 :
2604 0 : static inline unsigned long cpu_bw_dl(struct rq *rq)
2605 : {
2606 0 : return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2607 : }
2608 :
2609 0 : static inline unsigned long cpu_util_dl(struct rq *rq)
2610 : {
2611 0 : return READ_ONCE(rq->avg_dl.util_avg);
2612 : }
2613 :
2614 0 : static inline unsigned long cpu_util_cfs(struct rq *rq)
2615 : {
2616 0 : unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2617 :
2618 0 : if (sched_feat(UTIL_EST)) {
2619 0 : util = max_t(unsigned long, util,
2620 : READ_ONCE(rq->cfs.avg.util_est.enqueued));
2621 : }
2622 :
2623 0 : return util;
2624 : }
2625 :
2626 0 : static inline unsigned long cpu_util_rt(struct rq *rq)
2627 : {
2628 0 : return READ_ONCE(rq->avg_rt.util_avg);
2629 : }
2630 : #endif
2631 :
2632 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2633 7892 : static inline unsigned long cpu_util_irq(struct rq *rq)
2634 : {
2635 7892 : return rq->avg_irq.util_avg;
2636 : }
2637 :
2638 : static inline
2639 7892 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2640 : {
2641 7892 : util *= (max - irq);
2642 7892 : util /= max;
2643 :
2644 7892 : return util;
2645 :
2646 : }
2647 : #else
2648 : static inline unsigned long cpu_util_irq(struct rq *rq)
2649 : {
2650 : return 0;
2651 : }
2652 :
2653 : static inline
2654 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2655 : {
2656 : return util;
2657 : }
2658 : #endif
2659 :
2660 : #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2661 :
2662 : #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2663 :
2664 : DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2665 :
2666 : static inline bool sched_energy_enabled(void)
2667 : {
2668 : return static_branch_unlikely(&sched_energy_present);
2669 : }
2670 :
2671 : #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2672 :
2673 : #define perf_domain_span(pd) NULL
2674 19509 : static inline bool sched_energy_enabled(void) { return false; }
2675 :
2676 : #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2677 :
2678 : #ifdef CONFIG_MEMBARRIER
2679 : /*
2680 : * The scheduler provides memory barriers required by membarrier between:
2681 : * - prior user-space memory accesses and store to rq->membarrier_state,
2682 : * - store to rq->membarrier_state and following user-space memory accesses.
2683 : * In the same way it provides those guarantees around store to rq->curr.
2684 : */
2685 : static inline void membarrier_switch_mm(struct rq *rq,
2686 : struct mm_struct *prev_mm,
2687 : struct mm_struct *next_mm)
2688 : {
2689 : int membarrier_state;
2690 :
2691 : if (prev_mm == next_mm)
2692 : return;
2693 :
2694 : membarrier_state = atomic_read(&next_mm->membarrier_state);
2695 : if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2696 : return;
2697 :
2698 : WRITE_ONCE(rq->membarrier_state, membarrier_state);
2699 : }
2700 : #else
2701 : static inline void membarrier_switch_mm(struct rq *rq,
2702 : struct mm_struct *prev_mm,
2703 : struct mm_struct *next_mm)
2704 : {
2705 : }
2706 : #endif
2707 :
2708 : #ifdef CONFIG_SMP
2709 6787 : static inline bool is_per_cpu_kthread(struct task_struct *p)
2710 : {
2711 6787 : if (!(p->flags & PF_KTHREAD))
2712 : return false;
2713 :
2714 1710 : if (p->nr_cpus_allowed != 1)
2715 : return false;
2716 :
2717 : return true;
2718 : }
2719 : #endif
2720 :
2721 : void swake_up_all_locked(struct swait_queue_head *q);
2722 : void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
|
