Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : /*
3 : * Scheduler internal types and methods:
4 : */
5 : #ifndef _KERNEL_SCHED_SCHED_H
6 : #define _KERNEL_SCHED_SCHED_H
7 :
8 : #include <linux/sched/affinity.h>
9 : #include <linux/sched/autogroup.h>
10 : #include <linux/sched/cpufreq.h>
11 : #include <linux/sched/deadline.h>
12 : #include <linux/sched.h>
13 : #include <linux/sched/loadavg.h>
14 : #include <linux/sched/mm.h>
15 : #include <linux/sched/rseq_api.h>
16 : #include <linux/sched/signal.h>
17 : #include <linux/sched/smt.h>
18 : #include <linux/sched/stat.h>
19 : #include <linux/sched/sysctl.h>
20 : #include <linux/sched/task_flags.h>
21 : #include <linux/sched/task.h>
22 : #include <linux/sched/topology.h>
23 :
24 : #include <linux/atomic.h>
25 : #include <linux/bitmap.h>
26 : #include <linux/bug.h>
27 : #include <linux/capability.h>
28 : #include <linux/cgroup_api.h>
29 : #include <linux/cgroup.h>
30 : #include <linux/cpufreq.h>
31 : #include <linux/cpumask_api.h>
32 : #include <linux/ctype.h>
33 : #include <linux/file.h>
34 : #include <linux/fs_api.h>
35 : #include <linux/hrtimer_api.h>
36 : #include <linux/interrupt.h>
37 : #include <linux/irq_work.h>
38 : #include <linux/jiffies.h>
39 : #include <linux/kref_api.h>
40 : #include <linux/kthread.h>
41 : #include <linux/ktime_api.h>
42 : #include <linux/lockdep_api.h>
43 : #include <linux/lockdep.h>
44 : #include <linux/minmax.h>
45 : #include <linux/mm.h>
46 : #include <linux/module.h>
47 : #include <linux/mutex_api.h>
48 : #include <linux/plist.h>
49 : #include <linux/poll.h>
50 : #include <linux/proc_fs.h>
51 : #include <linux/profile.h>
52 : #include <linux/psi.h>
53 : #include <linux/rcupdate.h>
54 : #include <linux/seq_file.h>
55 : #include <linux/seqlock.h>
56 : #include <linux/softirq.h>
57 : #include <linux/spinlock_api.h>
58 : #include <linux/static_key.h>
59 : #include <linux/stop_machine.h>
60 : #include <linux/syscalls_api.h>
61 : #include <linux/syscalls.h>
62 : #include <linux/tick.h>
63 : #include <linux/topology.h>
64 : #include <linux/types.h>
65 : #include <linux/u64_stats_sync_api.h>
66 : #include <linux/uaccess.h>
67 : #include <linux/wait_api.h>
68 : #include <linux/wait_bit.h>
69 : #include <linux/workqueue_api.h>
70 :
71 : #include <trace/events/power.h>
72 : #include <trace/events/sched.h>
73 :
74 : #include "../workqueue_internal.h"
75 :
76 : #ifdef CONFIG_CGROUP_SCHED
77 : #include <linux/cgroup.h>
78 : #include <linux/psi.h>
79 : #endif
80 :
81 : #ifdef CONFIG_SCHED_DEBUG
82 : # include <linux/static_key.h>
83 : #endif
84 :
85 : #ifdef CONFIG_PARAVIRT
86 : # include <asm/paravirt.h>
87 : # include <asm/paravirt_api_clock.h>
88 : #endif
89 :
90 : #include "cpupri.h"
91 : #include "cpudeadline.h"
92 :
93 : #ifdef CONFIG_SCHED_DEBUG
94 : # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
95 : #else
96 : # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
97 : #endif
98 :
99 : struct rq;
100 : struct cpuidle_state;
101 :
102 : /* task_struct::on_rq states: */
103 : #define TASK_ON_RQ_QUEUED 1
104 : #define TASK_ON_RQ_MIGRATING 2
105 :
106 : extern __read_mostly int scheduler_running;
107 :
108 : extern unsigned long calc_load_update;
109 : extern atomic_long_t calc_load_tasks;
110 :
111 : extern void calc_global_load_tick(struct rq *this_rq);
112 : extern long calc_load_fold_active(struct rq *this_rq, long adjust);
113 :
114 : extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
115 : /*
116 : * Helpers for converting nanosecond timing to jiffy resolution
117 : */
118 : #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
119 :
120 : /*
121 : * Increase resolution of nice-level calculations for 64-bit architectures.
122 : * The extra resolution improves shares distribution and load balancing of
123 : * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
124 : * hierarchies, especially on larger systems. This is not a user-visible change
125 : * and does not change the user-interface for setting shares/weights.
126 : *
127 : * We increase resolution only if we have enough bits to allow this increased
128 : * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
129 : * are pretty high and the returns do not justify the increased costs.
130 : *
131 : * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
132 : * increase coverage and consistency always enable it on 64-bit platforms.
133 : */
134 : #ifdef CONFIG_64BIT
135 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
136 : # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
137 : # define scale_load_down(w) \
138 : ({ \
139 : unsigned long __w = (w); \
140 : if (__w) \
141 : __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
142 : __w; \
143 : })
144 : #else
145 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
146 : # define scale_load(w) (w)
147 : # define scale_load_down(w) (w)
148 : #endif
149 :
150 : /*
151 : * Task weight (visible to users) and its load (invisible to users) have
152 : * independent resolution, but they should be well calibrated. We use
153 : * scale_load() and scale_load_down(w) to convert between them. The
154 : * following must be true:
155 : *
156 : * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
157 : *
158 : */
159 : #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
160 :
161 : /*
162 : * Single value that decides SCHED_DEADLINE internal math precision.
163 : * 10 -> just above 1us
164 : * 9 -> just above 0.5us
165 : */
166 : #define DL_SCALE 10
167 :
168 : /*
169 : * Single value that denotes runtime == period, ie unlimited time.
170 : */
171 : #define RUNTIME_INF ((u64)~0ULL)
172 :
173 : static inline int idle_policy(int policy)
174 : {
175 1754 : return policy == SCHED_IDLE;
176 : }
177 : static inline int fair_policy(int policy)
178 : {
179 210 : return policy == SCHED_NORMAL || policy == SCHED_BATCH;
180 : }
181 :
182 : static inline int rt_policy(int policy)
183 : {
184 113 : return policy == SCHED_FIFO || policy == SCHED_RR;
185 : }
186 :
187 : static inline int dl_policy(int policy)
188 : {
189 : return policy == SCHED_DEADLINE;
190 : }
191 : static inline bool valid_policy(int policy)
192 : {
193 210 : return idle_policy(policy) || fair_policy(policy) ||
194 105 : rt_policy(policy) || dl_policy(policy);
195 : }
196 :
197 : static inline int task_has_idle_policy(struct task_struct *p)
198 : {
199 3513 : return idle_policy(p->policy);
200 : }
201 :
202 : static inline int task_has_rt_policy(struct task_struct *p)
203 : {
204 8 : return rt_policy(p->policy);
205 : }
206 :
207 : static inline int task_has_dl_policy(struct task_struct *p)
208 : {
209 4 : return dl_policy(p->policy);
210 : }
211 :
212 : #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
213 :
214 : static inline void update_avg(u64 *avg, u64 sample)
215 : {
216 : s64 diff = sample - *avg;
217 : *avg += diff / 8;
218 : }
219 :
220 : /*
221 : * Shifting a value by an exponent greater *or equal* to the size of said value
222 : * is UB; cap at size-1.
223 : */
224 : #define shr_bound(val, shift) \
225 : (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
226 :
227 : /*
228 : * !! For sched_setattr_nocheck() (kernel) only !!
229 : *
230 : * This is actually gross. :(
231 : *
232 : * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
233 : * tasks, but still be able to sleep. We need this on platforms that cannot
234 : * atomically change clock frequency. Remove once fast switching will be
235 : * available on such platforms.
236 : *
237 : * SUGOV stands for SchedUtil GOVernor.
238 : */
239 : #define SCHED_FLAG_SUGOV 0x10000000
240 :
241 : #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
242 :
243 : static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
244 : {
245 : #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
246 : return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
247 : #else
248 : return false;
249 : #endif
250 : }
251 :
252 : /*
253 : * Tells if entity @a should preempt entity @b.
254 : */
255 : static inline bool
256 : dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
257 : {
258 0 : return dl_entity_is_special(a) ||
259 0 : dl_time_before(a->deadline, b->deadline);
260 : }
261 :
262 : /*
263 : * This is the priority-queue data structure of the RT scheduling class:
264 : */
265 : struct rt_prio_array {
266 : DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
267 : struct list_head queue[MAX_RT_PRIO];
268 : };
269 :
270 : struct rt_bandwidth {
271 : /* nests inside the rq lock: */
272 : raw_spinlock_t rt_runtime_lock;
273 : ktime_t rt_period;
274 : u64 rt_runtime;
275 : struct hrtimer rt_period_timer;
276 : unsigned int rt_period_active;
277 : };
278 :
279 : void __dl_clear_params(struct task_struct *p);
280 :
281 : struct dl_bandwidth {
282 : raw_spinlock_t dl_runtime_lock;
283 : u64 dl_runtime;
284 : u64 dl_period;
285 : };
286 :
287 : static inline int dl_bandwidth_enabled(void)
288 : {
289 : return sysctl_sched_rt_runtime >= 0;
290 : }
291 :
292 : /*
293 : * To keep the bandwidth of -deadline tasks under control
294 : * we need some place where:
295 : * - store the maximum -deadline bandwidth of each cpu;
296 : * - cache the fraction of bandwidth that is currently allocated in
297 : * each root domain;
298 : *
299 : * This is all done in the data structure below. It is similar to the
300 : * one used for RT-throttling (rt_bandwidth), with the main difference
301 : * that, since here we are only interested in admission control, we
302 : * do not decrease any runtime while the group "executes", neither we
303 : * need a timer to replenish it.
304 : *
305 : * With respect to SMP, bandwidth is given on a per root domain basis,
306 : * meaning that:
307 : * - bw (< 100%) is the deadline bandwidth of each CPU;
308 : * - total_bw is the currently allocated bandwidth in each root domain;
309 : */
310 : struct dl_bw {
311 : raw_spinlock_t lock;
312 : u64 bw;
313 : u64 total_bw;
314 : };
315 :
316 : /*
317 : * Verify the fitness of task @p to run on @cpu taking into account the
318 : * CPU original capacity and the runtime/deadline ratio of the task.
319 : *
320 : * The function will return true if the CPU original capacity of the
321 : * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
322 : * task and false otherwise.
323 : */
324 : static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
325 : {
326 : unsigned long cap = arch_scale_cpu_capacity(cpu);
327 :
328 : return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
329 : }
330 :
331 : extern void init_dl_bw(struct dl_bw *dl_b);
332 : extern int sched_dl_global_validate(void);
333 : extern void sched_dl_do_global(void);
334 : extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
335 : extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
336 : extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
337 : extern bool __checkparam_dl(const struct sched_attr *attr);
338 : extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
339 : extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
340 : extern int dl_cpu_busy(int cpu, struct task_struct *p);
341 :
342 : #ifdef CONFIG_CGROUP_SCHED
343 :
344 : struct cfs_rq;
345 : struct rt_rq;
346 :
347 : extern struct list_head task_groups;
348 :
349 : struct cfs_bandwidth {
350 : #ifdef CONFIG_CFS_BANDWIDTH
351 : raw_spinlock_t lock;
352 : ktime_t period;
353 : u64 quota;
354 : u64 runtime;
355 : u64 burst;
356 : u64 runtime_snap;
357 : s64 hierarchical_quota;
358 :
359 : u8 idle;
360 : u8 period_active;
361 : u8 slack_started;
362 : struct hrtimer period_timer;
363 : struct hrtimer slack_timer;
364 : struct list_head throttled_cfs_rq;
365 :
366 : /* Statistics: */
367 : int nr_periods;
368 : int nr_throttled;
369 : int nr_burst;
370 : u64 throttled_time;
371 : u64 burst_time;
372 : #endif
373 : };
374 :
375 : /* Task group related information */
376 : struct task_group {
377 : struct cgroup_subsys_state css;
378 :
379 : #ifdef CONFIG_FAIR_GROUP_SCHED
380 : /* schedulable entities of this group on each CPU */
381 : struct sched_entity **se;
382 : /* runqueue "owned" by this group on each CPU */
383 : struct cfs_rq **cfs_rq;
384 : unsigned long shares;
385 :
386 : /* A positive value indicates that this is a SCHED_IDLE group. */
387 : int idle;
388 :
389 : #ifdef CONFIG_SMP
390 : /*
391 : * load_avg can be heavily contended at clock tick time, so put
392 : * it in its own cacheline separated from the fields above which
393 : * will also be accessed at each tick.
394 : */
395 : atomic_long_t load_avg ____cacheline_aligned;
396 : #endif
397 : #endif
398 :
399 : #ifdef CONFIG_RT_GROUP_SCHED
400 : struct sched_rt_entity **rt_se;
401 : struct rt_rq **rt_rq;
402 :
403 : struct rt_bandwidth rt_bandwidth;
404 : #endif
405 :
406 : struct rcu_head rcu;
407 : struct list_head list;
408 :
409 : struct task_group *parent;
410 : struct list_head siblings;
411 : struct list_head children;
412 :
413 : #ifdef CONFIG_SCHED_AUTOGROUP
414 : struct autogroup *autogroup;
415 : #endif
416 :
417 : struct cfs_bandwidth cfs_bandwidth;
418 :
419 : #ifdef CONFIG_UCLAMP_TASK_GROUP
420 : /* The two decimal precision [%] value requested from user-space */
421 : unsigned int uclamp_pct[UCLAMP_CNT];
422 : /* Clamp values requested for a task group */
423 : struct uclamp_se uclamp_req[UCLAMP_CNT];
424 : /* Effective clamp values used for a task group */
425 : struct uclamp_se uclamp[UCLAMP_CNT];
426 : #endif
427 :
428 : };
429 :
430 : #ifdef CONFIG_FAIR_GROUP_SCHED
431 : #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
432 :
433 : /*
434 : * A weight of 0 or 1 can cause arithmetics problems.
435 : * A weight of a cfs_rq is the sum of weights of which entities
436 : * are queued on this cfs_rq, so a weight of a entity should not be
437 : * too large, so as the shares value of a task group.
438 : * (The default weight is 1024 - so there's no practical
439 : * limitation from this.)
440 : */
441 : #define MIN_SHARES (1UL << 1)
442 : #define MAX_SHARES (1UL << 18)
443 : #endif
444 :
445 : typedef int (*tg_visitor)(struct task_group *, void *);
446 :
447 : extern int walk_tg_tree_from(struct task_group *from,
448 : tg_visitor down, tg_visitor up, void *data);
449 :
450 : /*
451 : * Iterate the full tree, calling @down when first entering a node and @up when
452 : * leaving it for the final time.
453 : *
454 : * Caller must hold rcu_lock or sufficient equivalent.
455 : */
456 : static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
457 : {
458 : return walk_tg_tree_from(&root_task_group, down, up, data);
459 : }
460 :
461 : extern int tg_nop(struct task_group *tg, void *data);
462 :
463 : extern void free_fair_sched_group(struct task_group *tg);
464 : extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
465 : extern void online_fair_sched_group(struct task_group *tg);
466 : extern void unregister_fair_sched_group(struct task_group *tg);
467 : extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
468 : struct sched_entity *se, int cpu,
469 : struct sched_entity *parent);
470 : extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
471 :
472 : extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
473 : extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
474 : extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
475 :
476 : extern void unregister_rt_sched_group(struct task_group *tg);
477 : extern void free_rt_sched_group(struct task_group *tg);
478 : extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
479 : extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
480 : struct sched_rt_entity *rt_se, int cpu,
481 : struct sched_rt_entity *parent);
482 : extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
483 : extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
484 : extern long sched_group_rt_runtime(struct task_group *tg);
485 : extern long sched_group_rt_period(struct task_group *tg);
486 : extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
487 :
488 : extern struct task_group *sched_create_group(struct task_group *parent);
489 : extern void sched_online_group(struct task_group *tg,
490 : struct task_group *parent);
491 : extern void sched_destroy_group(struct task_group *tg);
492 : extern void sched_release_group(struct task_group *tg);
493 :
494 : extern void sched_move_task(struct task_struct *tsk);
495 :
496 : #ifdef CONFIG_FAIR_GROUP_SCHED
497 : extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
498 :
499 : extern int sched_group_set_idle(struct task_group *tg, long idle);
500 :
501 : #ifdef CONFIG_SMP
502 : extern void set_task_rq_fair(struct sched_entity *se,
503 : struct cfs_rq *prev, struct cfs_rq *next);
504 : #else /* !CONFIG_SMP */
505 : static inline void set_task_rq_fair(struct sched_entity *se,
506 : struct cfs_rq *prev, struct cfs_rq *next) { }
507 : #endif /* CONFIG_SMP */
508 : #endif /* CONFIG_FAIR_GROUP_SCHED */
509 :
510 : #else /* CONFIG_CGROUP_SCHED */
511 :
512 : struct cfs_bandwidth { };
513 :
514 : #endif /* CONFIG_CGROUP_SCHED */
515 :
516 : /* CFS-related fields in a runqueue */
517 : struct cfs_rq {
518 : struct load_weight load;
519 : unsigned int nr_running;
520 : unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
521 : unsigned int idle_nr_running; /* SCHED_IDLE */
522 : unsigned int idle_h_nr_running; /* SCHED_IDLE */
523 :
524 : u64 exec_clock;
525 : u64 min_vruntime;
526 : #ifdef CONFIG_SCHED_CORE
527 : unsigned int forceidle_seq;
528 : u64 min_vruntime_fi;
529 : #endif
530 :
531 : #ifndef CONFIG_64BIT
532 : u64 min_vruntime_copy;
533 : #endif
534 :
535 : struct rb_root_cached tasks_timeline;
536 :
537 : /*
538 : * 'curr' points to currently running entity on this cfs_rq.
539 : * It is set to NULL otherwise (i.e when none are currently running).
540 : */
541 : struct sched_entity *curr;
542 : struct sched_entity *next;
543 : struct sched_entity *last;
544 : struct sched_entity *skip;
545 :
546 : #ifdef CONFIG_SCHED_DEBUG
547 : unsigned int nr_spread_over;
548 : #endif
549 :
550 : #ifdef CONFIG_SMP
551 : /*
552 : * CFS load tracking
553 : */
554 : struct sched_avg avg;
555 : #ifndef CONFIG_64BIT
556 : u64 load_last_update_time_copy;
557 : #endif
558 : struct {
559 : raw_spinlock_t lock ____cacheline_aligned;
560 : int nr;
561 : unsigned long load_avg;
562 : unsigned long util_avg;
563 : unsigned long runnable_avg;
564 : } removed;
565 :
566 : #ifdef CONFIG_FAIR_GROUP_SCHED
567 : unsigned long tg_load_avg_contrib;
568 : long propagate;
569 : long prop_runnable_sum;
570 :
571 : /*
572 : * h_load = weight * f(tg)
573 : *
574 : * Where f(tg) is the recursive weight fraction assigned to
575 : * this group.
576 : */
577 : unsigned long h_load;
578 : u64 last_h_load_update;
579 : struct sched_entity *h_load_next;
580 : #endif /* CONFIG_FAIR_GROUP_SCHED */
581 : #endif /* CONFIG_SMP */
582 :
583 : #ifdef CONFIG_FAIR_GROUP_SCHED
584 : struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
585 :
586 : /*
587 : * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
588 : * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
589 : * (like users, containers etc.)
590 : *
591 : * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
592 : * This list is used during load balance.
593 : */
594 : int on_list;
595 : struct list_head leaf_cfs_rq_list;
596 : struct task_group *tg; /* group that "owns" this runqueue */
597 :
598 : /* Locally cached copy of our task_group's idle value */
599 : int idle;
600 :
601 : #ifdef CONFIG_CFS_BANDWIDTH
602 : int runtime_enabled;
603 : s64 runtime_remaining;
604 :
605 : u64 throttled_clock;
606 : u64 throttled_clock_task;
607 : u64 throttled_clock_task_time;
608 : int throttled;
609 : int throttle_count;
610 : struct list_head throttled_list;
611 : #endif /* CONFIG_CFS_BANDWIDTH */
612 : #endif /* CONFIG_FAIR_GROUP_SCHED */
613 : };
614 :
615 : static inline int rt_bandwidth_enabled(void)
616 : {
617 0 : return sysctl_sched_rt_runtime >= 0;
618 : }
619 :
620 : /* RT IPI pull logic requires IRQ_WORK */
621 : #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
622 : # define HAVE_RT_PUSH_IPI
623 : #endif
624 :
625 : /* Real-Time classes' related field in a runqueue: */
626 : struct rt_rq {
627 : struct rt_prio_array active;
628 : unsigned int rt_nr_running;
629 : unsigned int rr_nr_running;
630 : #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
631 : struct {
632 : int curr; /* highest queued rt task prio */
633 : #ifdef CONFIG_SMP
634 : int next; /* next highest */
635 : #endif
636 : } highest_prio;
637 : #endif
638 : #ifdef CONFIG_SMP
639 : unsigned int rt_nr_migratory;
640 : unsigned int rt_nr_total;
641 : int overloaded;
642 : struct plist_head pushable_tasks;
643 :
644 : #endif /* CONFIG_SMP */
645 : int rt_queued;
646 :
647 : int rt_throttled;
648 : u64 rt_time;
649 : u64 rt_runtime;
650 : /* Nests inside the rq lock: */
651 : raw_spinlock_t rt_runtime_lock;
652 :
653 : #ifdef CONFIG_RT_GROUP_SCHED
654 : unsigned int rt_nr_boosted;
655 :
656 : struct rq *rq;
657 : struct task_group *tg;
658 : #endif
659 : };
660 :
661 : static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
662 : {
663 : return rt_rq->rt_queued && rt_rq->rt_nr_running;
664 : }
665 :
666 : /* Deadline class' related fields in a runqueue */
667 : struct dl_rq {
668 : /* runqueue is an rbtree, ordered by deadline */
669 : struct rb_root_cached root;
670 :
671 : unsigned int dl_nr_running;
672 :
673 : #ifdef CONFIG_SMP
674 : /*
675 : * Deadline values of the currently executing and the
676 : * earliest ready task on this rq. Caching these facilitates
677 : * the decision whether or not a ready but not running task
678 : * should migrate somewhere else.
679 : */
680 : struct {
681 : u64 curr;
682 : u64 next;
683 : } earliest_dl;
684 :
685 : unsigned int dl_nr_migratory;
686 : int overloaded;
687 :
688 : /*
689 : * Tasks on this rq that can be pushed away. They are kept in
690 : * an rb-tree, ordered by tasks' deadlines, with caching
691 : * of the leftmost (earliest deadline) element.
692 : */
693 : struct rb_root_cached pushable_dl_tasks_root;
694 : #else
695 : struct dl_bw dl_bw;
696 : #endif
697 : /*
698 : * "Active utilization" for this runqueue: increased when a
699 : * task wakes up (becomes TASK_RUNNING) and decreased when a
700 : * task blocks
701 : */
702 : u64 running_bw;
703 :
704 : /*
705 : * Utilization of the tasks "assigned" to this runqueue (including
706 : * the tasks that are in runqueue and the tasks that executed on this
707 : * CPU and blocked). Increased when a task moves to this runqueue, and
708 : * decreased when the task moves away (migrates, changes scheduling
709 : * policy, or terminates).
710 : * This is needed to compute the "inactive utilization" for the
711 : * runqueue (inactive utilization = this_bw - running_bw).
712 : */
713 : u64 this_bw;
714 : u64 extra_bw;
715 :
716 : /*
717 : * Inverse of the fraction of CPU utilization that can be reclaimed
718 : * by the GRUB algorithm.
719 : */
720 : u64 bw_ratio;
721 : };
722 :
723 : #ifdef CONFIG_FAIR_GROUP_SCHED
724 : /* An entity is a task if it doesn't "own" a runqueue */
725 : #define entity_is_task(se) (!se->my_q)
726 :
727 : static inline void se_update_runnable(struct sched_entity *se)
728 : {
729 : if (!entity_is_task(se))
730 : se->runnable_weight = se->my_q->h_nr_running;
731 : }
732 :
733 : static inline long se_runnable(struct sched_entity *se)
734 : {
735 : if (entity_is_task(se))
736 : return !!se->on_rq;
737 : else
738 : return se->runnable_weight;
739 : }
740 :
741 : #else
742 : #define entity_is_task(se) 1
743 :
744 : static inline void se_update_runnable(struct sched_entity *se) {}
745 :
746 : static inline long se_runnable(struct sched_entity *se)
747 : {
748 : return !!se->on_rq;
749 : }
750 : #endif
751 :
752 : #ifdef CONFIG_SMP
753 : /*
754 : * XXX we want to get rid of these helpers and use the full load resolution.
755 : */
756 : static inline long se_weight(struct sched_entity *se)
757 : {
758 : return scale_load_down(se->load.weight);
759 : }
760 :
761 :
762 : static inline bool sched_asym_prefer(int a, int b)
763 : {
764 : return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
765 : }
766 :
767 : struct perf_domain {
768 : struct em_perf_domain *em_pd;
769 : struct perf_domain *next;
770 : struct rcu_head rcu;
771 : };
772 :
773 : /* Scheduling group status flags */
774 : #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
775 : #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
776 :
777 : /*
778 : * We add the notion of a root-domain which will be used to define per-domain
779 : * variables. Each exclusive cpuset essentially defines an island domain by
780 : * fully partitioning the member CPUs from any other cpuset. Whenever a new
781 : * exclusive cpuset is created, we also create and attach a new root-domain
782 : * object.
783 : *
784 : */
785 : struct root_domain {
786 : atomic_t refcount;
787 : atomic_t rto_count;
788 : struct rcu_head rcu;
789 : cpumask_var_t span;
790 : cpumask_var_t online;
791 :
792 : /*
793 : * Indicate pullable load on at least one CPU, e.g:
794 : * - More than one runnable task
795 : * - Running task is misfit
796 : */
797 : int overload;
798 :
799 : /* Indicate one or more cpus over-utilized (tipping point) */
800 : int overutilized;
801 :
802 : /*
803 : * The bit corresponding to a CPU gets set here if such CPU has more
804 : * than one runnable -deadline task (as it is below for RT tasks).
805 : */
806 : cpumask_var_t dlo_mask;
807 : atomic_t dlo_count;
808 : struct dl_bw dl_bw;
809 : struct cpudl cpudl;
810 :
811 : /*
812 : * Indicate whether a root_domain's dl_bw has been checked or
813 : * updated. It's monotonously increasing value.
814 : *
815 : * Also, some corner cases, like 'wrap around' is dangerous, but given
816 : * that u64 is 'big enough'. So that shouldn't be a concern.
817 : */
818 : u64 visit_gen;
819 :
820 : #ifdef HAVE_RT_PUSH_IPI
821 : /*
822 : * For IPI pull requests, loop across the rto_mask.
823 : */
824 : struct irq_work rto_push_work;
825 : raw_spinlock_t rto_lock;
826 : /* These are only updated and read within rto_lock */
827 : int rto_loop;
828 : int rto_cpu;
829 : /* These atomics are updated outside of a lock */
830 : atomic_t rto_loop_next;
831 : atomic_t rto_loop_start;
832 : #endif
833 : /*
834 : * The "RT overload" flag: it gets set if a CPU has more than
835 : * one runnable RT task.
836 : */
837 : cpumask_var_t rto_mask;
838 : struct cpupri cpupri;
839 :
840 : unsigned long max_cpu_capacity;
841 :
842 : /*
843 : * NULL-terminated list of performance domains intersecting with the
844 : * CPUs of the rd. Protected by RCU.
845 : */
846 : struct perf_domain __rcu *pd;
847 : };
848 :
849 : extern void init_defrootdomain(void);
850 : extern int sched_init_domains(const struct cpumask *cpu_map);
851 : extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
852 : extern void sched_get_rd(struct root_domain *rd);
853 : extern void sched_put_rd(struct root_domain *rd);
854 :
855 : #ifdef HAVE_RT_PUSH_IPI
856 : extern void rto_push_irq_work_func(struct irq_work *work);
857 : #endif
858 : #endif /* CONFIG_SMP */
859 :
860 : #ifdef CONFIG_UCLAMP_TASK
861 : /*
862 : * struct uclamp_bucket - Utilization clamp bucket
863 : * @value: utilization clamp value for tasks on this clamp bucket
864 : * @tasks: number of RUNNABLE tasks on this clamp bucket
865 : *
866 : * Keep track of how many tasks are RUNNABLE for a given utilization
867 : * clamp value.
868 : */
869 : struct uclamp_bucket {
870 : unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
871 : unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
872 : };
873 :
874 : /*
875 : * struct uclamp_rq - rq's utilization clamp
876 : * @value: currently active clamp values for a rq
877 : * @bucket: utilization clamp buckets affecting a rq
878 : *
879 : * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
880 : * A clamp value is affecting a rq when there is at least one task RUNNABLE
881 : * (or actually running) with that value.
882 : *
883 : * There are up to UCLAMP_CNT possible different clamp values, currently there
884 : * are only two: minimum utilization and maximum utilization.
885 : *
886 : * All utilization clamping values are MAX aggregated, since:
887 : * - for util_min: we want to run the CPU at least at the max of the minimum
888 : * utilization required by its currently RUNNABLE tasks.
889 : * - for util_max: we want to allow the CPU to run up to the max of the
890 : * maximum utilization allowed by its currently RUNNABLE tasks.
891 : *
892 : * Since on each system we expect only a limited number of different
893 : * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
894 : * the metrics required to compute all the per-rq utilization clamp values.
895 : */
896 : struct uclamp_rq {
897 : unsigned int value;
898 : struct uclamp_bucket bucket[UCLAMP_BUCKETS];
899 : };
900 :
901 : DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
902 : #endif /* CONFIG_UCLAMP_TASK */
903 :
904 : /*
905 : * This is the main, per-CPU runqueue data structure.
906 : *
907 : * Locking rule: those places that want to lock multiple runqueues
908 : * (such as the load balancing or the thread migration code), lock
909 : * acquire operations must be ordered by ascending &runqueue.
910 : */
911 : struct rq {
912 : /* runqueue lock: */
913 : raw_spinlock_t __lock;
914 :
915 : /*
916 : * nr_running and cpu_load should be in the same cacheline because
917 : * remote CPUs use both these fields when doing load calculation.
918 : */
919 : unsigned int nr_running;
920 : #ifdef CONFIG_NUMA_BALANCING
921 : unsigned int nr_numa_running;
922 : unsigned int nr_preferred_running;
923 : unsigned int numa_migrate_on;
924 : #endif
925 : #ifdef CONFIG_NO_HZ_COMMON
926 : #ifdef CONFIG_SMP
927 : unsigned long last_blocked_load_update_tick;
928 : unsigned int has_blocked_load;
929 : call_single_data_t nohz_csd;
930 : #endif /* CONFIG_SMP */
931 : unsigned int nohz_tick_stopped;
932 : atomic_t nohz_flags;
933 : #endif /* CONFIG_NO_HZ_COMMON */
934 :
935 : #ifdef CONFIG_SMP
936 : unsigned int ttwu_pending;
937 : #endif
938 : u64 nr_switches;
939 :
940 : #ifdef CONFIG_UCLAMP_TASK
941 : /* Utilization clamp values based on CPU's RUNNABLE tasks */
942 : struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
943 : unsigned int uclamp_flags;
944 : #define UCLAMP_FLAG_IDLE 0x01
945 : #endif
946 :
947 : struct cfs_rq cfs;
948 : struct rt_rq rt;
949 : struct dl_rq dl;
950 :
951 : #ifdef CONFIG_FAIR_GROUP_SCHED
952 : /* list of leaf cfs_rq on this CPU: */
953 : struct list_head leaf_cfs_rq_list;
954 : struct list_head *tmp_alone_branch;
955 : #endif /* CONFIG_FAIR_GROUP_SCHED */
956 :
957 : /*
958 : * This is part of a global counter where only the total sum
959 : * over all CPUs matters. A task can increase this counter on
960 : * one CPU and if it got migrated afterwards it may decrease
961 : * it on another CPU. Always updated under the runqueue lock:
962 : */
963 : unsigned int nr_uninterruptible;
964 :
965 : struct task_struct __rcu *curr;
966 : struct task_struct *idle;
967 : struct task_struct *stop;
968 : unsigned long next_balance;
969 : struct mm_struct *prev_mm;
970 :
971 : unsigned int clock_update_flags;
972 : u64 clock;
973 : /* Ensure that all clocks are in the same cache line */
974 : u64 clock_task ____cacheline_aligned;
975 : u64 clock_pelt;
976 : unsigned long lost_idle_time;
977 :
978 : atomic_t nr_iowait;
979 :
980 : #ifdef CONFIG_SCHED_DEBUG
981 : u64 last_seen_need_resched_ns;
982 : int ticks_without_resched;
983 : #endif
984 :
985 : #ifdef CONFIG_MEMBARRIER
986 : int membarrier_state;
987 : #endif
988 :
989 : #ifdef CONFIG_SMP
990 : struct root_domain *rd;
991 : struct sched_domain __rcu *sd;
992 :
993 : unsigned long cpu_capacity;
994 : unsigned long cpu_capacity_orig;
995 :
996 : struct callback_head *balance_callback;
997 :
998 : unsigned char nohz_idle_balance;
999 : unsigned char idle_balance;
1000 :
1001 : unsigned long misfit_task_load;
1002 :
1003 : /* For active balancing */
1004 : int active_balance;
1005 : int push_cpu;
1006 : struct cpu_stop_work active_balance_work;
1007 :
1008 : /* CPU of this runqueue: */
1009 : int cpu;
1010 : int online;
1011 :
1012 : struct list_head cfs_tasks;
1013 :
1014 : struct sched_avg avg_rt;
1015 : struct sched_avg avg_dl;
1016 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1017 : struct sched_avg avg_irq;
1018 : #endif
1019 : #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1020 : struct sched_avg avg_thermal;
1021 : #endif
1022 : u64 idle_stamp;
1023 : u64 avg_idle;
1024 :
1025 : unsigned long wake_stamp;
1026 : u64 wake_avg_idle;
1027 :
1028 : /* This is used to determine avg_idle's max value */
1029 : u64 max_idle_balance_cost;
1030 :
1031 : #ifdef CONFIG_HOTPLUG_CPU
1032 : struct rcuwait hotplug_wait;
1033 : #endif
1034 : #endif /* CONFIG_SMP */
1035 :
1036 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1037 : u64 prev_irq_time;
1038 : #endif
1039 : #ifdef CONFIG_PARAVIRT
1040 : u64 prev_steal_time;
1041 : #endif
1042 : #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1043 : u64 prev_steal_time_rq;
1044 : #endif
1045 :
1046 : /* calc_load related fields */
1047 : unsigned long calc_load_update;
1048 : long calc_load_active;
1049 :
1050 : #ifdef CONFIG_SCHED_HRTICK
1051 : #ifdef CONFIG_SMP
1052 : call_single_data_t hrtick_csd;
1053 : #endif
1054 : struct hrtimer hrtick_timer;
1055 : ktime_t hrtick_time;
1056 : #endif
1057 :
1058 : #ifdef CONFIG_SCHEDSTATS
1059 : /* latency stats */
1060 : struct sched_info rq_sched_info;
1061 : unsigned long long rq_cpu_time;
1062 : /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1063 :
1064 : /* sys_sched_yield() stats */
1065 : unsigned int yld_count;
1066 :
1067 : /* schedule() stats */
1068 : unsigned int sched_count;
1069 : unsigned int sched_goidle;
1070 :
1071 : /* try_to_wake_up() stats */
1072 : unsigned int ttwu_count;
1073 : unsigned int ttwu_local;
1074 : #endif
1075 :
1076 : #ifdef CONFIG_CPU_IDLE
1077 : /* Must be inspected within a rcu lock section */
1078 : struct cpuidle_state *idle_state;
1079 : #endif
1080 :
1081 : #ifdef CONFIG_SMP
1082 : unsigned int nr_pinned;
1083 : #endif
1084 : unsigned int push_busy;
1085 : struct cpu_stop_work push_work;
1086 :
1087 : #ifdef CONFIG_SCHED_CORE
1088 : /* per rq */
1089 : struct rq *core;
1090 : struct task_struct *core_pick;
1091 : unsigned int core_enabled;
1092 : unsigned int core_sched_seq;
1093 : struct rb_root core_tree;
1094 :
1095 : /* shared state -- careful with sched_core_cpu_deactivate() */
1096 : unsigned int core_task_seq;
1097 : unsigned int core_pick_seq;
1098 : unsigned long core_cookie;
1099 : unsigned int core_forceidle_count;
1100 : unsigned int core_forceidle_seq;
1101 : unsigned int core_forceidle_occupation;
1102 : u64 core_forceidle_start;
1103 : #endif
1104 : };
1105 :
1106 : #ifdef CONFIG_FAIR_GROUP_SCHED
1107 :
1108 : /* CPU runqueue to which this cfs_rq is attached */
1109 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1110 : {
1111 : return cfs_rq->rq;
1112 : }
1113 :
1114 : #else
1115 :
1116 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1117 : {
1118 3837 : return container_of(cfs_rq, struct rq, cfs);
1119 : }
1120 : #endif
1121 :
1122 : static inline int cpu_of(struct rq *rq)
1123 : {
1124 : #ifdef CONFIG_SMP
1125 : return rq->cpu;
1126 : #else
1127 : return 0;
1128 : #endif
1129 : }
1130 :
1131 : #define MDF_PUSH 0x01
1132 :
1133 : static inline bool is_migration_disabled(struct task_struct *p)
1134 : {
1135 : #ifdef CONFIG_SMP
1136 : return p->migration_disabled;
1137 : #else
1138 : return false;
1139 : #endif
1140 : }
1141 :
1142 : struct sched_group;
1143 : #ifdef CONFIG_SCHED_CORE
1144 : static inline struct cpumask *sched_group_span(struct sched_group *sg);
1145 :
1146 : DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1147 :
1148 : static inline bool sched_core_enabled(struct rq *rq)
1149 : {
1150 : return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1151 : }
1152 :
1153 : static inline bool sched_core_disabled(void)
1154 : {
1155 : return !static_branch_unlikely(&__sched_core_enabled);
1156 : }
1157 :
1158 : /*
1159 : * Be careful with this function; not for general use. The return value isn't
1160 : * stable unless you actually hold a relevant rq->__lock.
1161 : */
1162 : static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1163 : {
1164 : if (sched_core_enabled(rq))
1165 : return &rq->core->__lock;
1166 :
1167 : return &rq->__lock;
1168 : }
1169 :
1170 : static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1171 : {
1172 : if (rq->core_enabled)
1173 : return &rq->core->__lock;
1174 :
1175 : return &rq->__lock;
1176 : }
1177 :
1178 : bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1179 :
1180 : /*
1181 : * Helpers to check if the CPU's core cookie matches with the task's cookie
1182 : * when core scheduling is enabled.
1183 : * A special case is that the task's cookie always matches with CPU's core
1184 : * cookie if the CPU is in an idle core.
1185 : */
1186 : static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1187 : {
1188 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1189 : if (!sched_core_enabled(rq))
1190 : return true;
1191 :
1192 : return rq->core->core_cookie == p->core_cookie;
1193 : }
1194 :
1195 : static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1196 : {
1197 : bool idle_core = true;
1198 : int cpu;
1199 :
1200 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1201 : if (!sched_core_enabled(rq))
1202 : return true;
1203 :
1204 : for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1205 : if (!available_idle_cpu(cpu)) {
1206 : idle_core = false;
1207 : break;
1208 : }
1209 : }
1210 :
1211 : /*
1212 : * A CPU in an idle core is always the best choice for tasks with
1213 : * cookies.
1214 : */
1215 : return idle_core || rq->core->core_cookie == p->core_cookie;
1216 : }
1217 :
1218 : static inline bool sched_group_cookie_match(struct rq *rq,
1219 : struct task_struct *p,
1220 : struct sched_group *group)
1221 : {
1222 : int cpu;
1223 :
1224 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1225 : if (!sched_core_enabled(rq))
1226 : return true;
1227 :
1228 : for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1229 : if (sched_core_cookie_match(rq, p))
1230 : return true;
1231 : }
1232 : return false;
1233 : }
1234 :
1235 : static inline bool sched_core_enqueued(struct task_struct *p)
1236 : {
1237 : return !RB_EMPTY_NODE(&p->core_node);
1238 : }
1239 :
1240 : extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1241 : extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1242 :
1243 : extern void sched_core_get(void);
1244 : extern void sched_core_put(void);
1245 :
1246 : #else /* !CONFIG_SCHED_CORE */
1247 :
1248 : static inline bool sched_core_enabled(struct rq *rq)
1249 : {
1250 : return false;
1251 : }
1252 :
1253 : static inline bool sched_core_disabled(void)
1254 : {
1255 : return true;
1256 : }
1257 :
1258 : static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1259 : {
1260 : return &rq->__lock;
1261 : }
1262 :
1263 : static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1264 : {
1265 : return &rq->__lock;
1266 : }
1267 :
1268 : static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1269 : {
1270 : return true;
1271 : }
1272 :
1273 : static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1274 : {
1275 : return true;
1276 : }
1277 :
1278 : static inline bool sched_group_cookie_match(struct rq *rq,
1279 : struct task_struct *p,
1280 : struct sched_group *group)
1281 : {
1282 : return true;
1283 : }
1284 : #endif /* CONFIG_SCHED_CORE */
1285 :
1286 : static inline void lockdep_assert_rq_held(struct rq *rq)
1287 : {
1288 4977 : lockdep_assert_held(__rq_lockp(rq));
1289 : }
1290 :
1291 : extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1292 : extern bool raw_spin_rq_trylock(struct rq *rq);
1293 : extern void raw_spin_rq_unlock(struct rq *rq);
1294 :
1295 : static inline void raw_spin_rq_lock(struct rq *rq)
1296 : {
1297 1465 : raw_spin_rq_lock_nested(rq, 0);
1298 : }
1299 :
1300 : static inline void raw_spin_rq_lock_irq(struct rq *rq)
1301 : {
1302 0 : local_irq_disable();
1303 0 : raw_spin_rq_lock(rq);
1304 : }
1305 :
1306 : static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1307 : {
1308 618 : raw_spin_rq_unlock(rq);
1309 : local_irq_enable();
1310 : }
1311 :
1312 : static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1313 : {
1314 : unsigned long flags;
1315 0 : local_irq_save(flags);
1316 0 : raw_spin_rq_lock(rq);
1317 : return flags;
1318 : }
1319 :
1320 : static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1321 : {
1322 0 : raw_spin_rq_unlock(rq);
1323 0 : local_irq_restore(flags);
1324 : }
1325 :
1326 : #define raw_spin_rq_lock_irqsave(rq, flags) \
1327 : do { \
1328 : flags = _raw_spin_rq_lock_irqsave(rq); \
1329 : } while (0)
1330 :
1331 : #ifdef CONFIG_SCHED_SMT
1332 : extern void __update_idle_core(struct rq *rq);
1333 :
1334 : static inline void update_idle_core(struct rq *rq)
1335 : {
1336 : if (static_branch_unlikely(&sched_smt_present))
1337 : __update_idle_core(rq);
1338 : }
1339 :
1340 : #else
1341 : static inline void update_idle_core(struct rq *rq) { }
1342 : #endif
1343 :
1344 : DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1345 :
1346 : #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1347 : #define this_rq() this_cpu_ptr(&runqueues)
1348 : #define task_rq(p) cpu_rq(task_cpu(p))
1349 : #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1350 : #define raw_rq() raw_cpu_ptr(&runqueues)
1351 :
1352 : #ifdef CONFIG_FAIR_GROUP_SCHED
1353 : static inline struct task_struct *task_of(struct sched_entity *se)
1354 : {
1355 : SCHED_WARN_ON(!entity_is_task(se));
1356 : return container_of(se, struct task_struct, se);
1357 : }
1358 :
1359 : static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1360 : {
1361 : return p->se.cfs_rq;
1362 : }
1363 :
1364 : /* runqueue on which this entity is (to be) queued */
1365 : static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1366 : {
1367 : return se->cfs_rq;
1368 : }
1369 :
1370 : /* runqueue "owned" by this group */
1371 : static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1372 : {
1373 : return grp->my_q;
1374 : }
1375 :
1376 : #else
1377 :
1378 : static inline struct task_struct *task_of(struct sched_entity *se)
1379 : {
1380 4127 : return container_of(se, struct task_struct, se);
1381 : }
1382 :
1383 : static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1384 : {
1385 721 : return &task_rq(p)->cfs;
1386 : }
1387 :
1388 : static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1389 : {
1390 3507 : struct task_struct *p = task_of(se);
1391 3507 : struct rq *rq = task_rq(p);
1392 :
1393 : return &rq->cfs;
1394 : }
1395 :
1396 : /* runqueue "owned" by this group */
1397 : static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1398 : {
1399 : return NULL;
1400 : }
1401 : #endif
1402 :
1403 : extern void update_rq_clock(struct rq *rq);
1404 :
1405 : /*
1406 : * rq::clock_update_flags bits
1407 : *
1408 : * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1409 : * call to __schedule(). This is an optimisation to avoid
1410 : * neighbouring rq clock updates.
1411 : *
1412 : * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1413 : * in effect and calls to update_rq_clock() are being ignored.
1414 : *
1415 : * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1416 : * made to update_rq_clock() since the last time rq::lock was pinned.
1417 : *
1418 : * If inside of __schedule(), clock_update_flags will have been
1419 : * shifted left (a left shift is a cheap operation for the fast path
1420 : * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1421 : *
1422 : * if (rq-clock_update_flags >= RQCF_UPDATED)
1423 : *
1424 : * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1425 : * one position though, because the next rq_unpin_lock() will shift it
1426 : * back.
1427 : */
1428 : #define RQCF_REQ_SKIP 0x01
1429 : #define RQCF_ACT_SKIP 0x02
1430 : #define RQCF_UPDATED 0x04
1431 :
1432 2496 : static inline void assert_clock_updated(struct rq *rq)
1433 : {
1434 : /*
1435 : * The only reason for not seeing a clock update since the
1436 : * last rq_pin_lock() is if we're currently skipping updates.
1437 : */
1438 2496 : SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1439 2496 : }
1440 :
1441 : static inline u64 rq_clock(struct rq *rq)
1442 : {
1443 0 : lockdep_assert_rq_held(rq);
1444 0 : assert_clock_updated(rq);
1445 :
1446 0 : return rq->clock;
1447 : }
1448 :
1449 : static inline u64 rq_clock_task(struct rq *rq)
1450 : {
1451 13 : lockdep_assert_rq_held(rq);
1452 2496 : assert_clock_updated(rq);
1453 :
1454 2483 : return rq->clock_task;
1455 : }
1456 :
1457 : /**
1458 : * By default the decay is the default pelt decay period.
1459 : * The decay shift can change the decay period in
1460 : * multiples of 32.
1461 : * Decay shift Decay period(ms)
1462 : * 0 32
1463 : * 1 64
1464 : * 2 128
1465 : * 3 256
1466 : * 4 512
1467 : */
1468 : extern int sched_thermal_decay_shift;
1469 :
1470 : static inline u64 rq_clock_thermal(struct rq *rq)
1471 : {
1472 13 : return rq_clock_task(rq) >> sched_thermal_decay_shift;
1473 : }
1474 :
1475 : static inline void rq_clock_skip_update(struct rq *rq)
1476 : {
1477 301 : lockdep_assert_rq_held(rq);
1478 301 : rq->clock_update_flags |= RQCF_REQ_SKIP;
1479 : }
1480 :
1481 : /*
1482 : * See rt task throttling, which is the only time a skip
1483 : * request is canceled.
1484 : */
1485 : static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1486 : {
1487 0 : lockdep_assert_rq_held(rq);
1488 0 : rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1489 : }
1490 :
1491 : struct rq_flags {
1492 : unsigned long flags;
1493 : struct pin_cookie cookie;
1494 : #ifdef CONFIG_SCHED_DEBUG
1495 : /*
1496 : * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1497 : * current pin context is stashed here in case it needs to be
1498 : * restored in rq_repin_lock().
1499 : */
1500 : unsigned int clock_update_flags;
1501 : #endif
1502 : };
1503 :
1504 : extern struct callback_head balance_push_callback;
1505 :
1506 : /*
1507 : * Lockdep annotation that avoids accidental unlocks; it's like a
1508 : * sticky/continuous lockdep_assert_held().
1509 : *
1510 : * This avoids code that has access to 'struct rq *rq' (basically everything in
1511 : * the scheduler) from accidentally unlocking the rq if they do not also have a
1512 : * copy of the (on-stack) 'struct rq_flags rf'.
1513 : *
1514 : * Also see Documentation/locking/lockdep-design.rst.
1515 : */
1516 : static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1517 : {
1518 : rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1519 :
1520 : #ifdef CONFIG_SCHED_DEBUG
1521 1464 : rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1522 1464 : rf->clock_update_flags = 0;
1523 : #ifdef CONFIG_SMP
1524 : SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1525 : #endif
1526 : #endif
1527 : }
1528 :
1529 : static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1530 : {
1531 : #ifdef CONFIG_SCHED_DEBUG
1532 727 : if (rq->clock_update_flags > RQCF_ACT_SKIP)
1533 522 : rf->clock_update_flags = RQCF_UPDATED;
1534 : #endif
1535 :
1536 1464 : lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1537 : }
1538 :
1539 : static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1540 : {
1541 : lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1542 :
1543 : #ifdef CONFIG_SCHED_DEBUG
1544 : /*
1545 : * Restore the value we stashed in @rf for this pin context.
1546 : */
1547 : rq->clock_update_flags |= rf->clock_update_flags;
1548 : #endif
1549 : }
1550 :
1551 : struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1552 : __acquires(rq->lock);
1553 :
1554 : struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1555 : __acquires(p->pi_lock)
1556 : __acquires(rq->lock);
1557 :
1558 : static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1559 : __releases(rq->lock)
1560 : {
1561 0 : rq_unpin_lock(rq, rf);
1562 0 : raw_spin_rq_unlock(rq);
1563 : }
1564 :
1565 : static inline void
1566 0 : task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1567 : __releases(rq->lock)
1568 : __releases(p->pi_lock)
1569 : {
1570 432 : rq_unpin_lock(rq, rf);
1571 216 : raw_spin_rq_unlock(rq);
1572 432 : raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1573 0 : }
1574 :
1575 : static inline void
1576 : rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1577 : __acquires(rq->lock)
1578 : {
1579 : raw_spin_rq_lock_irqsave(rq, rf->flags);
1580 : rq_pin_lock(rq, rf);
1581 : }
1582 :
1583 : static inline void
1584 : rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1585 : __acquires(rq->lock)
1586 : {
1587 : raw_spin_rq_lock_irq(rq);
1588 : rq_pin_lock(rq, rf);
1589 : }
1590 :
1591 : static inline void
1592 : rq_lock(struct rq *rq, struct rq_flags *rf)
1593 : __acquires(rq->lock)
1594 : {
1595 1248 : raw_spin_rq_lock(rq);
1596 1248 : rq_pin_lock(rq, rf);
1597 : }
1598 :
1599 : static inline void
1600 : rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1601 : __releases(rq->lock)
1602 : {
1603 : rq_unpin_lock(rq, rf);
1604 : raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1605 : }
1606 :
1607 : static inline void
1608 : rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1609 : __releases(rq->lock)
1610 : {
1611 0 : rq_unpin_lock(rq, rf);
1612 : raw_spin_rq_unlock_irq(rq);
1613 : }
1614 :
1615 : static inline void
1616 : rq_unlock(struct rq *rq, struct rq_flags *rf)
1617 : __releases(rq->lock)
1618 : {
1619 1153 : rq_unpin_lock(rq, rf);
1620 630 : raw_spin_rq_unlock(rq);
1621 : }
1622 :
1623 : static inline struct rq *
1624 : this_rq_lock_irq(struct rq_flags *rf)
1625 : __acquires(rq->lock)
1626 : {
1627 : struct rq *rq;
1628 :
1629 : local_irq_disable();
1630 0 : rq = this_rq();
1631 0 : rq_lock(rq, rf);
1632 : return rq;
1633 : }
1634 :
1635 : #ifdef CONFIG_NUMA
1636 : enum numa_topology_type {
1637 : NUMA_DIRECT,
1638 : NUMA_GLUELESS_MESH,
1639 : NUMA_BACKPLANE,
1640 : };
1641 : extern enum numa_topology_type sched_numa_topology_type;
1642 : extern int sched_max_numa_distance;
1643 : extern bool find_numa_distance(int distance);
1644 : extern void sched_init_numa(int offline_node);
1645 : extern void sched_update_numa(int cpu, bool online);
1646 : extern void sched_domains_numa_masks_set(unsigned int cpu);
1647 : extern void sched_domains_numa_masks_clear(unsigned int cpu);
1648 : extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1649 : #else
1650 : static inline void sched_init_numa(int offline_node) { }
1651 : static inline void sched_update_numa(int cpu, bool online) { }
1652 : static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1653 : static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1654 : static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1655 : {
1656 : return nr_cpu_ids;
1657 : }
1658 : #endif
1659 :
1660 : #ifdef CONFIG_NUMA_BALANCING
1661 : /* The regions in numa_faults array from task_struct */
1662 : enum numa_faults_stats {
1663 : NUMA_MEM = 0,
1664 : NUMA_CPU,
1665 : NUMA_MEMBUF,
1666 : NUMA_CPUBUF
1667 : };
1668 : extern void sched_setnuma(struct task_struct *p, int node);
1669 : extern int migrate_task_to(struct task_struct *p, int cpu);
1670 : extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1671 : int cpu, int scpu);
1672 : extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1673 : #else
1674 : static inline void
1675 : init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1676 : {
1677 : }
1678 : #endif /* CONFIG_NUMA_BALANCING */
1679 :
1680 : #ifdef CONFIG_SMP
1681 :
1682 : static inline void
1683 : queue_balance_callback(struct rq *rq,
1684 : struct callback_head *head,
1685 : void (*func)(struct rq *rq))
1686 : {
1687 : lockdep_assert_rq_held(rq);
1688 :
1689 : if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1690 : return;
1691 :
1692 : head->func = (void (*)(struct callback_head *))func;
1693 : head->next = rq->balance_callback;
1694 : rq->balance_callback = head;
1695 : }
1696 :
1697 : #define rcu_dereference_check_sched_domain(p) \
1698 : rcu_dereference_check((p), \
1699 : lockdep_is_held(&sched_domains_mutex))
1700 :
1701 : /*
1702 : * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1703 : * See destroy_sched_domains: call_rcu for details.
1704 : *
1705 : * The domain tree of any CPU may only be accessed from within
1706 : * preempt-disabled sections.
1707 : */
1708 : #define for_each_domain(cpu, __sd) \
1709 : for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1710 : __sd; __sd = __sd->parent)
1711 :
1712 : /**
1713 : * highest_flag_domain - Return highest sched_domain containing flag.
1714 : * @cpu: The CPU whose highest level of sched domain is to
1715 : * be returned.
1716 : * @flag: The flag to check for the highest sched_domain
1717 : * for the given CPU.
1718 : *
1719 : * Returns the highest sched_domain of a CPU which contains the given flag.
1720 : */
1721 : static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1722 : {
1723 : struct sched_domain *sd, *hsd = NULL;
1724 :
1725 : for_each_domain(cpu, sd) {
1726 : if (!(sd->flags & flag))
1727 : break;
1728 : hsd = sd;
1729 : }
1730 :
1731 : return hsd;
1732 : }
1733 :
1734 : static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1735 : {
1736 : struct sched_domain *sd;
1737 :
1738 : for_each_domain(cpu, sd) {
1739 : if (sd->flags & flag)
1740 : break;
1741 : }
1742 :
1743 : return sd;
1744 : }
1745 :
1746 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1747 : DECLARE_PER_CPU(int, sd_llc_size);
1748 : DECLARE_PER_CPU(int, sd_llc_id);
1749 : DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1750 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1751 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1752 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1753 : extern struct static_key_false sched_asym_cpucapacity;
1754 :
1755 : struct sched_group_capacity {
1756 : atomic_t ref;
1757 : /*
1758 : * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1759 : * for a single CPU.
1760 : */
1761 : unsigned long capacity;
1762 : unsigned long min_capacity; /* Min per-CPU capacity in group */
1763 : unsigned long max_capacity; /* Max per-CPU capacity in group */
1764 : unsigned long next_update;
1765 : int imbalance; /* XXX unrelated to capacity but shared group state */
1766 :
1767 : #ifdef CONFIG_SCHED_DEBUG
1768 : int id;
1769 : #endif
1770 :
1771 : unsigned long cpumask[]; /* Balance mask */
1772 : };
1773 :
1774 : struct sched_group {
1775 : struct sched_group *next; /* Must be a circular list */
1776 : atomic_t ref;
1777 :
1778 : unsigned int group_weight;
1779 : struct sched_group_capacity *sgc;
1780 : int asym_prefer_cpu; /* CPU of highest priority in group */
1781 : int flags;
1782 :
1783 : /*
1784 : * The CPUs this group covers.
1785 : *
1786 : * NOTE: this field is variable length. (Allocated dynamically
1787 : * by attaching extra space to the end of the structure,
1788 : * depending on how many CPUs the kernel has booted up with)
1789 : */
1790 : unsigned long cpumask[];
1791 : };
1792 :
1793 : static inline struct cpumask *sched_group_span(struct sched_group *sg)
1794 : {
1795 : return to_cpumask(sg->cpumask);
1796 : }
1797 :
1798 : /*
1799 : * See build_balance_mask().
1800 : */
1801 : static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1802 : {
1803 : return to_cpumask(sg->sgc->cpumask);
1804 : }
1805 :
1806 : /**
1807 : * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1808 : * @group: The group whose first CPU is to be returned.
1809 : */
1810 : static inline unsigned int group_first_cpu(struct sched_group *group)
1811 : {
1812 : return cpumask_first(sched_group_span(group));
1813 : }
1814 :
1815 : extern int group_balance_cpu(struct sched_group *sg);
1816 :
1817 : #ifdef CONFIG_SCHED_DEBUG
1818 : void update_sched_domain_debugfs(void);
1819 : void dirty_sched_domain_sysctl(int cpu);
1820 : #else
1821 : static inline void update_sched_domain_debugfs(void)
1822 : {
1823 : }
1824 : static inline void dirty_sched_domain_sysctl(int cpu)
1825 : {
1826 : }
1827 : #endif
1828 :
1829 : extern int sched_update_scaling(void);
1830 :
1831 : extern void flush_smp_call_function_from_idle(void);
1832 :
1833 : #else /* !CONFIG_SMP: */
1834 : static inline void flush_smp_call_function_from_idle(void) { }
1835 : #endif
1836 :
1837 : #include "stats.h"
1838 :
1839 : #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1840 :
1841 : extern void __sched_core_account_forceidle(struct rq *rq);
1842 :
1843 : static inline void sched_core_account_forceidle(struct rq *rq)
1844 : {
1845 : if (schedstat_enabled())
1846 : __sched_core_account_forceidle(rq);
1847 : }
1848 :
1849 : extern void __sched_core_tick(struct rq *rq);
1850 :
1851 : static inline void sched_core_tick(struct rq *rq)
1852 : {
1853 : if (sched_core_enabled(rq) && schedstat_enabled())
1854 : __sched_core_tick(rq);
1855 : }
1856 :
1857 : #else
1858 :
1859 : static inline void sched_core_account_forceidle(struct rq *rq) {}
1860 :
1861 : static inline void sched_core_tick(struct rq *rq) {}
1862 :
1863 : #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1864 :
1865 : #ifdef CONFIG_CGROUP_SCHED
1866 :
1867 : /*
1868 : * Return the group to which this tasks belongs.
1869 : *
1870 : * We cannot use task_css() and friends because the cgroup subsystem
1871 : * changes that value before the cgroup_subsys::attach() method is called,
1872 : * therefore we cannot pin it and might observe the wrong value.
1873 : *
1874 : * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1875 : * core changes this before calling sched_move_task().
1876 : *
1877 : * Instead we use a 'copy' which is updated from sched_move_task() while
1878 : * holding both task_struct::pi_lock and rq::lock.
1879 : */
1880 : static inline struct task_group *task_group(struct task_struct *p)
1881 : {
1882 : return p->sched_task_group;
1883 : }
1884 :
1885 : /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1886 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1887 : {
1888 : #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1889 : struct task_group *tg = task_group(p);
1890 : #endif
1891 :
1892 : #ifdef CONFIG_FAIR_GROUP_SCHED
1893 : set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1894 : p->se.cfs_rq = tg->cfs_rq[cpu];
1895 : p->se.parent = tg->se[cpu];
1896 : #endif
1897 :
1898 : #ifdef CONFIG_RT_GROUP_SCHED
1899 : p->rt.rt_rq = tg->rt_rq[cpu];
1900 : p->rt.parent = tg->rt_se[cpu];
1901 : #endif
1902 : }
1903 :
1904 : #else /* CONFIG_CGROUP_SCHED */
1905 :
1906 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1907 : static inline struct task_group *task_group(struct task_struct *p)
1908 : {
1909 : return NULL;
1910 : }
1911 :
1912 : #endif /* CONFIG_CGROUP_SCHED */
1913 :
1914 : static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1915 : {
1916 108 : set_task_rq(p, cpu);
1917 : #ifdef CONFIG_SMP
1918 : /*
1919 : * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1920 : * successfully executed on another CPU. We must ensure that updates of
1921 : * per-task data have been completed by this moment.
1922 : */
1923 : smp_wmb();
1924 : WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1925 : p->wake_cpu = cpu;
1926 : #endif
1927 : }
1928 :
1929 : /*
1930 : * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1931 : */
1932 : #ifdef CONFIG_SCHED_DEBUG
1933 : # define const_debug __read_mostly
1934 : #else
1935 : # define const_debug const
1936 : #endif
1937 :
1938 : #define SCHED_FEAT(name, enabled) \
1939 : __SCHED_FEAT_##name ,
1940 :
1941 : enum {
1942 : #include "features.h"
1943 : __SCHED_FEAT_NR,
1944 : };
1945 :
1946 : #undef SCHED_FEAT
1947 :
1948 : #ifdef CONFIG_SCHED_DEBUG
1949 :
1950 : /*
1951 : * To support run-time toggling of sched features, all the translation units
1952 : * (but core.c) reference the sysctl_sched_features defined in core.c.
1953 : */
1954 : extern const_debug unsigned int sysctl_sched_features;
1955 :
1956 : #ifdef CONFIG_JUMP_LABEL
1957 : #define SCHED_FEAT(name, enabled) \
1958 : static __always_inline bool static_branch_##name(struct static_key *key) \
1959 : { \
1960 : return static_key_##enabled(key); \
1961 : }
1962 :
1963 : #include "features.h"
1964 : #undef SCHED_FEAT
1965 :
1966 : extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1967 : #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1968 :
1969 : #else /* !CONFIG_JUMP_LABEL */
1970 :
1971 : #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1972 :
1973 : #endif /* CONFIG_JUMP_LABEL */
1974 :
1975 : #else /* !SCHED_DEBUG */
1976 :
1977 : /*
1978 : * Each translation unit has its own copy of sysctl_sched_features to allow
1979 : * constants propagation at compile time and compiler optimization based on
1980 : * features default.
1981 : */
1982 : #define SCHED_FEAT(name, enabled) \
1983 : (1UL << __SCHED_FEAT_##name) * enabled |
1984 : static const_debug __maybe_unused unsigned int sysctl_sched_features =
1985 : #include "features.h"
1986 : 0;
1987 : #undef SCHED_FEAT
1988 :
1989 : #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1990 :
1991 : #endif /* SCHED_DEBUG */
1992 :
1993 : extern struct static_key_false sched_numa_balancing;
1994 : extern struct static_key_false sched_schedstats;
1995 :
1996 : static inline u64 global_rt_period(void)
1997 : {
1998 4 : return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1999 : }
2000 :
2001 : static inline u64 global_rt_runtime(void)
2002 : {
2003 4 : if (sysctl_sched_rt_runtime < 0)
2004 : return RUNTIME_INF;
2005 :
2006 4 : return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2007 : }
2008 :
2009 : static inline int task_current(struct rq *rq, struct task_struct *p)
2010 : {
2011 : return rq->curr == p;
2012 : }
2013 :
2014 : static inline int task_running(struct rq *rq, struct task_struct *p)
2015 : {
2016 : #ifdef CONFIG_SMP
2017 : return p->on_cpu;
2018 : #else
2019 0 : return task_current(rq, p);
2020 : #endif
2021 : }
2022 :
2023 : static inline int task_on_rq_queued(struct task_struct *p)
2024 : {
2025 0 : return p->on_rq == TASK_ON_RQ_QUEUED;
2026 : }
2027 :
2028 : static inline int task_on_rq_migrating(struct task_struct *p)
2029 : {
2030 216 : return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2031 : }
2032 :
2033 : /* Wake flags. The first three directly map to some SD flag value */
2034 : #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2035 : #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2036 : #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2037 :
2038 : #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2039 : #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2040 : #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2041 :
2042 : #ifdef CONFIG_SMP
2043 : static_assert(WF_EXEC == SD_BALANCE_EXEC);
2044 : static_assert(WF_FORK == SD_BALANCE_FORK);
2045 : static_assert(WF_TTWU == SD_BALANCE_WAKE);
2046 : #endif
2047 :
2048 : /*
2049 : * To aid in avoiding the subversion of "niceness" due to uneven distribution
2050 : * of tasks with abnormal "nice" values across CPUs the contribution that
2051 : * each task makes to its run queue's load is weighted according to its
2052 : * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2053 : * scaled version of the new time slice allocation that they receive on time
2054 : * slice expiry etc.
2055 : */
2056 :
2057 : #define WEIGHT_IDLEPRIO 3
2058 : #define WMULT_IDLEPRIO 1431655765
2059 :
2060 : extern const int sched_prio_to_weight[40];
2061 : extern const u32 sched_prio_to_wmult[40];
2062 :
2063 : /*
2064 : * {de,en}queue flags:
2065 : *
2066 : * DEQUEUE_SLEEP - task is no longer runnable
2067 : * ENQUEUE_WAKEUP - task just became runnable
2068 : *
2069 : * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2070 : * are in a known state which allows modification. Such pairs
2071 : * should preserve as much state as possible.
2072 : *
2073 : * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2074 : * in the runqueue.
2075 : *
2076 : * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2077 : * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2078 : * ENQUEUE_MIGRATED - the task was migrated during wakeup
2079 : *
2080 : */
2081 :
2082 : #define DEQUEUE_SLEEP 0x01
2083 : #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2084 : #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2085 : #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2086 :
2087 : #define ENQUEUE_WAKEUP 0x01
2088 : #define ENQUEUE_RESTORE 0x02
2089 : #define ENQUEUE_MOVE 0x04
2090 : #define ENQUEUE_NOCLOCK 0x08
2091 :
2092 : #define ENQUEUE_HEAD 0x10
2093 : #define ENQUEUE_REPLENISH 0x20
2094 : #ifdef CONFIG_SMP
2095 : #define ENQUEUE_MIGRATED 0x40
2096 : #else
2097 : #define ENQUEUE_MIGRATED 0x00
2098 : #endif
2099 :
2100 : #define RETRY_TASK ((void *)-1UL)
2101 :
2102 : struct sched_class {
2103 :
2104 : #ifdef CONFIG_UCLAMP_TASK
2105 : int uclamp_enabled;
2106 : #endif
2107 :
2108 : void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2109 : void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2110 : void (*yield_task) (struct rq *rq);
2111 : bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2112 :
2113 : void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2114 :
2115 : struct task_struct *(*pick_next_task)(struct rq *rq);
2116 :
2117 : void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2118 : void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2119 :
2120 : #ifdef CONFIG_SMP
2121 : int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2122 : int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2123 :
2124 : struct task_struct * (*pick_task)(struct rq *rq);
2125 :
2126 : void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2127 :
2128 : void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2129 :
2130 : void (*set_cpus_allowed)(struct task_struct *p,
2131 : const struct cpumask *newmask,
2132 : u32 flags);
2133 :
2134 : void (*rq_online)(struct rq *rq);
2135 : void (*rq_offline)(struct rq *rq);
2136 :
2137 : struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2138 : #endif
2139 :
2140 : void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2141 : void (*task_fork)(struct task_struct *p);
2142 : void (*task_dead)(struct task_struct *p);
2143 :
2144 : /*
2145 : * The switched_from() call is allowed to drop rq->lock, therefore we
2146 : * cannot assume the switched_from/switched_to pair is serialized by
2147 : * rq->lock. They are however serialized by p->pi_lock.
2148 : */
2149 : void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2150 : void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2151 : void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2152 : int oldprio);
2153 :
2154 : unsigned int (*get_rr_interval)(struct rq *rq,
2155 : struct task_struct *task);
2156 :
2157 : void (*update_curr)(struct rq *rq);
2158 :
2159 : #define TASK_SET_GROUP 0
2160 : #define TASK_MOVE_GROUP 1
2161 :
2162 : #ifdef CONFIG_FAIR_GROUP_SCHED
2163 : void (*task_change_group)(struct task_struct *p, int type);
2164 : #endif
2165 : };
2166 :
2167 621 : static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2168 : {
2169 621 : WARN_ON_ONCE(rq->curr != prev);
2170 621 : prev->sched_class->put_prev_task(rq, prev);
2171 621 : }
2172 :
2173 : static inline void set_next_task(struct rq *rq, struct task_struct *next)
2174 : {
2175 3 : next->sched_class->set_next_task(rq, next, false);
2176 : }
2177 :
2178 :
2179 : /*
2180 : * Helper to define a sched_class instance; each one is placed in a separate
2181 : * section which is ordered by the linker script:
2182 : *
2183 : * include/asm-generic/vmlinux.lds.h
2184 : *
2185 : * Also enforce alignment on the instance, not the type, to guarantee layout.
2186 : */
2187 : #define DEFINE_SCHED_CLASS(name) \
2188 : const struct sched_class name##_sched_class \
2189 : __aligned(__alignof__(struct sched_class)) \
2190 : __section("__" #name "_sched_class")
2191 :
2192 : /* Defined in include/asm-generic/vmlinux.lds.h */
2193 : extern struct sched_class __begin_sched_classes[];
2194 : extern struct sched_class __end_sched_classes[];
2195 :
2196 : #define sched_class_highest (__end_sched_classes - 1)
2197 : #define sched_class_lowest (__begin_sched_classes - 1)
2198 :
2199 : #define for_class_range(class, _from, _to) \
2200 : for (class = (_from); class != (_to); class--)
2201 :
2202 : #define for_each_class(class) \
2203 : for_class_range(class, sched_class_highest, sched_class_lowest)
2204 :
2205 : extern const struct sched_class stop_sched_class;
2206 : extern const struct sched_class dl_sched_class;
2207 : extern const struct sched_class rt_sched_class;
2208 : extern const struct sched_class fair_sched_class;
2209 : extern const struct sched_class idle_sched_class;
2210 :
2211 : static inline bool sched_stop_runnable(struct rq *rq)
2212 : {
2213 : return rq->stop && task_on_rq_queued(rq->stop);
2214 : }
2215 :
2216 : static inline bool sched_dl_runnable(struct rq *rq)
2217 : {
2218 : return rq->dl.dl_nr_running > 0;
2219 : }
2220 :
2221 : static inline bool sched_rt_runnable(struct rq *rq)
2222 : {
2223 : return rq->rt.rt_queued > 0;
2224 : }
2225 :
2226 : static inline bool sched_fair_runnable(struct rq *rq)
2227 : {
2228 : return rq->cfs.nr_running > 0;
2229 : }
2230 :
2231 : extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2232 : extern struct task_struct *pick_next_task_idle(struct rq *rq);
2233 :
2234 : #define SCA_CHECK 0x01
2235 : #define SCA_MIGRATE_DISABLE 0x02
2236 : #define SCA_MIGRATE_ENABLE 0x04
2237 : #define SCA_USER 0x08
2238 :
2239 : #ifdef CONFIG_SMP
2240 :
2241 : extern void update_group_capacity(struct sched_domain *sd, int cpu);
2242 :
2243 : extern void trigger_load_balance(struct rq *rq);
2244 :
2245 : extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2246 :
2247 : static inline struct task_struct *get_push_task(struct rq *rq)
2248 : {
2249 : struct task_struct *p = rq->curr;
2250 :
2251 : lockdep_assert_rq_held(rq);
2252 :
2253 : if (rq->push_busy)
2254 : return NULL;
2255 :
2256 : if (p->nr_cpus_allowed == 1)
2257 : return NULL;
2258 :
2259 : if (p->migration_disabled)
2260 : return NULL;
2261 :
2262 : rq->push_busy = true;
2263 : return get_task_struct(p);
2264 : }
2265 :
2266 : extern int push_cpu_stop(void *arg);
2267 :
2268 : #endif
2269 :
2270 : #ifdef CONFIG_CPU_IDLE
2271 : static inline void idle_set_state(struct rq *rq,
2272 : struct cpuidle_state *idle_state)
2273 : {
2274 : rq->idle_state = idle_state;
2275 : }
2276 :
2277 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2278 : {
2279 : SCHED_WARN_ON(!rcu_read_lock_held());
2280 :
2281 : return rq->idle_state;
2282 : }
2283 : #else
2284 : static inline void idle_set_state(struct rq *rq,
2285 : struct cpuidle_state *idle_state)
2286 : {
2287 : }
2288 :
2289 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2290 : {
2291 : return NULL;
2292 : }
2293 : #endif
2294 :
2295 : extern void schedule_idle(void);
2296 :
2297 : extern void sysrq_sched_debug_show(void);
2298 : extern void sched_init_granularity(void);
2299 : extern void update_max_interval(void);
2300 :
2301 : extern void init_sched_dl_class(void);
2302 : extern void init_sched_rt_class(void);
2303 : extern void init_sched_fair_class(void);
2304 :
2305 : extern void reweight_task(struct task_struct *p, int prio);
2306 :
2307 : extern void resched_curr(struct rq *rq);
2308 : extern void resched_cpu(int cpu);
2309 :
2310 : extern struct rt_bandwidth def_rt_bandwidth;
2311 : extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2312 :
2313 : extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2314 : extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2315 : extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2316 :
2317 : #define BW_SHIFT 20
2318 : #define BW_UNIT (1 << BW_SHIFT)
2319 : #define RATIO_SHIFT 8
2320 : #define MAX_BW_BITS (64 - BW_SHIFT)
2321 : #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2322 : unsigned long to_ratio(u64 period, u64 runtime);
2323 :
2324 : extern void init_entity_runnable_average(struct sched_entity *se);
2325 : extern void post_init_entity_util_avg(struct task_struct *p);
2326 :
2327 : #ifdef CONFIG_NO_HZ_FULL
2328 : extern bool sched_can_stop_tick(struct rq *rq);
2329 : extern int __init sched_tick_offload_init(void);
2330 :
2331 : /*
2332 : * Tick may be needed by tasks in the runqueue depending on their policy and
2333 : * requirements. If tick is needed, lets send the target an IPI to kick it out of
2334 : * nohz mode if necessary.
2335 : */
2336 : static inline void sched_update_tick_dependency(struct rq *rq)
2337 : {
2338 : int cpu = cpu_of(rq);
2339 :
2340 : if (!tick_nohz_full_cpu(cpu))
2341 : return;
2342 :
2343 : if (sched_can_stop_tick(rq))
2344 : tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2345 : else
2346 : tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2347 : }
2348 : #else
2349 : static inline int sched_tick_offload_init(void) { return 0; }
2350 : static inline void sched_update_tick_dependency(struct rq *rq) { }
2351 : #endif
2352 :
2353 : static inline void add_nr_running(struct rq *rq, unsigned count)
2354 : {
2355 620 : unsigned prev_nr = rq->nr_running;
2356 :
2357 620 : rq->nr_running = prev_nr + count;
2358 : if (trace_sched_update_nr_running_tp_enabled()) {
2359 : call_trace_sched_update_nr_running(rq, count);
2360 : }
2361 :
2362 : #ifdef CONFIG_SMP
2363 : if (prev_nr < 2 && rq->nr_running >= 2) {
2364 : if (!READ_ONCE(rq->rd->overload))
2365 : WRITE_ONCE(rq->rd->overload, 1);
2366 : }
2367 : #endif
2368 :
2369 620 : sched_update_tick_dependency(rq);
2370 : }
2371 :
2372 : static inline void sub_nr_running(struct rq *rq, unsigned count)
2373 : {
2374 618 : rq->nr_running -= count;
2375 : if (trace_sched_update_nr_running_tp_enabled()) {
2376 : call_trace_sched_update_nr_running(rq, -count);
2377 : }
2378 :
2379 : /* Check if we still need preemption */
2380 618 : sched_update_tick_dependency(rq);
2381 : }
2382 :
2383 : extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2384 : extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2385 :
2386 : extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2387 :
2388 : extern const_debug unsigned int sysctl_sched_nr_migrate;
2389 : extern const_debug unsigned int sysctl_sched_migration_cost;
2390 :
2391 : #ifdef CONFIG_SCHED_DEBUG
2392 : extern unsigned int sysctl_sched_latency;
2393 : extern unsigned int sysctl_sched_min_granularity;
2394 : extern unsigned int sysctl_sched_idle_min_granularity;
2395 : extern unsigned int sysctl_sched_wakeup_granularity;
2396 : extern int sysctl_resched_latency_warn_ms;
2397 : extern int sysctl_resched_latency_warn_once;
2398 :
2399 : extern unsigned int sysctl_sched_tunable_scaling;
2400 :
2401 : extern unsigned int sysctl_numa_balancing_scan_delay;
2402 : extern unsigned int sysctl_numa_balancing_scan_period_min;
2403 : extern unsigned int sysctl_numa_balancing_scan_period_max;
2404 : extern unsigned int sysctl_numa_balancing_scan_size;
2405 : #endif
2406 :
2407 : #ifdef CONFIG_SCHED_HRTICK
2408 :
2409 : /*
2410 : * Use hrtick when:
2411 : * - enabled by features
2412 : * - hrtimer is actually high res
2413 : */
2414 : static inline int hrtick_enabled(struct rq *rq)
2415 : {
2416 : if (!cpu_active(cpu_of(rq)))
2417 : return 0;
2418 : return hrtimer_is_hres_active(&rq->hrtick_timer);
2419 : }
2420 :
2421 : static inline int hrtick_enabled_fair(struct rq *rq)
2422 : {
2423 : if (!sched_feat(HRTICK))
2424 : return 0;
2425 : return hrtick_enabled(rq);
2426 : }
2427 :
2428 : static inline int hrtick_enabled_dl(struct rq *rq)
2429 : {
2430 : if (!sched_feat(HRTICK_DL))
2431 : return 0;
2432 : return hrtick_enabled(rq);
2433 : }
2434 :
2435 : void hrtick_start(struct rq *rq, u64 delay);
2436 :
2437 : #else
2438 :
2439 : static inline int hrtick_enabled_fair(struct rq *rq)
2440 : {
2441 : return 0;
2442 : }
2443 :
2444 : static inline int hrtick_enabled_dl(struct rq *rq)
2445 : {
2446 : return 0;
2447 : }
2448 :
2449 : static inline int hrtick_enabled(struct rq *rq)
2450 : {
2451 : return 0;
2452 : }
2453 :
2454 : #endif /* CONFIG_SCHED_HRTICK */
2455 :
2456 : #ifndef arch_scale_freq_tick
2457 : static __always_inline
2458 : void arch_scale_freq_tick(void)
2459 : {
2460 : }
2461 : #endif
2462 :
2463 : #ifndef arch_scale_freq_capacity
2464 : /**
2465 : * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2466 : * @cpu: the CPU in question.
2467 : *
2468 : * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2469 : *
2470 : * f_curr
2471 : * ------ * SCHED_CAPACITY_SCALE
2472 : * f_max
2473 : */
2474 : static __always_inline
2475 : unsigned long arch_scale_freq_capacity(int cpu)
2476 : {
2477 : return SCHED_CAPACITY_SCALE;
2478 : }
2479 : #endif
2480 :
2481 :
2482 : #ifdef CONFIG_SMP
2483 :
2484 : static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2485 : {
2486 : #ifdef CONFIG_SCHED_CORE
2487 : /*
2488 : * In order to not have {0,2},{1,3} turn into into an AB-BA,
2489 : * order by core-id first and cpu-id second.
2490 : *
2491 : * Notably:
2492 : *
2493 : * double_rq_lock(0,3); will take core-0, core-1 lock
2494 : * double_rq_lock(1,2); will take core-1, core-0 lock
2495 : *
2496 : * when only cpu-id is considered.
2497 : */
2498 : if (rq1->core->cpu < rq2->core->cpu)
2499 : return true;
2500 : if (rq1->core->cpu > rq2->core->cpu)
2501 : return false;
2502 :
2503 : /*
2504 : * __sched_core_flip() relies on SMT having cpu-id lock order.
2505 : */
2506 : #endif
2507 : return rq1->cpu < rq2->cpu;
2508 : }
2509 :
2510 : extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2511 :
2512 : #ifdef CONFIG_PREEMPTION
2513 :
2514 : /*
2515 : * fair double_lock_balance: Safely acquires both rq->locks in a fair
2516 : * way at the expense of forcing extra atomic operations in all
2517 : * invocations. This assures that the double_lock is acquired using the
2518 : * same underlying policy as the spinlock_t on this architecture, which
2519 : * reduces latency compared to the unfair variant below. However, it
2520 : * also adds more overhead and therefore may reduce throughput.
2521 : */
2522 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2523 : __releases(this_rq->lock)
2524 : __acquires(busiest->lock)
2525 : __acquires(this_rq->lock)
2526 : {
2527 : raw_spin_rq_unlock(this_rq);
2528 : double_rq_lock(this_rq, busiest);
2529 :
2530 : return 1;
2531 : }
2532 :
2533 : #else
2534 : /*
2535 : * Unfair double_lock_balance: Optimizes throughput at the expense of
2536 : * latency by eliminating extra atomic operations when the locks are
2537 : * already in proper order on entry. This favors lower CPU-ids and will
2538 : * grant the double lock to lower CPUs over higher ids under contention,
2539 : * regardless of entry order into the function.
2540 : */
2541 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2542 : __releases(this_rq->lock)
2543 : __acquires(busiest->lock)
2544 : __acquires(this_rq->lock)
2545 : {
2546 : if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2547 : return 0;
2548 :
2549 : if (likely(raw_spin_rq_trylock(busiest)))
2550 : return 0;
2551 :
2552 : if (rq_order_less(this_rq, busiest)) {
2553 : raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2554 : return 0;
2555 : }
2556 :
2557 : raw_spin_rq_unlock(this_rq);
2558 : double_rq_lock(this_rq, busiest);
2559 :
2560 : return 1;
2561 : }
2562 :
2563 : #endif /* CONFIG_PREEMPTION */
2564 :
2565 : /*
2566 : * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2567 : */
2568 : static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2569 : {
2570 : lockdep_assert_irqs_disabled();
2571 :
2572 : return _double_lock_balance(this_rq, busiest);
2573 : }
2574 :
2575 : static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2576 : __releases(busiest->lock)
2577 : {
2578 : if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2579 : raw_spin_rq_unlock(busiest);
2580 : lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2581 : }
2582 :
2583 : static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2584 : {
2585 : if (l1 > l2)
2586 : swap(l1, l2);
2587 :
2588 : spin_lock(l1);
2589 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2590 : }
2591 :
2592 : static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2593 : {
2594 : if (l1 > l2)
2595 : swap(l1, l2);
2596 :
2597 : spin_lock_irq(l1);
2598 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2599 : }
2600 :
2601 : static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2602 : {
2603 : if (l1 > l2)
2604 : swap(l1, l2);
2605 :
2606 : raw_spin_lock(l1);
2607 : raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2608 : }
2609 :
2610 : /*
2611 : * double_rq_unlock - safely unlock two runqueues
2612 : *
2613 : * Note this does not restore interrupts like task_rq_unlock,
2614 : * you need to do so manually after calling.
2615 : */
2616 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2617 : __releases(rq1->lock)
2618 : __releases(rq2->lock)
2619 : {
2620 : if (__rq_lockp(rq1) != __rq_lockp(rq2))
2621 : raw_spin_rq_unlock(rq2);
2622 : else
2623 : __release(rq2->lock);
2624 : raw_spin_rq_unlock(rq1);
2625 : }
2626 :
2627 : extern void set_rq_online (struct rq *rq);
2628 : extern void set_rq_offline(struct rq *rq);
2629 : extern bool sched_smp_initialized;
2630 :
2631 : #else /* CONFIG_SMP */
2632 :
2633 : /*
2634 : * double_rq_lock - safely lock two runqueues
2635 : *
2636 : * Note this does not disable interrupts like task_rq_lock,
2637 : * you need to do so manually before calling.
2638 : */
2639 0 : static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2640 : __acquires(rq1->lock)
2641 : __acquires(rq2->lock)
2642 : {
2643 0 : BUG_ON(!irqs_disabled());
2644 0 : BUG_ON(rq1 != rq2);
2645 0 : raw_spin_rq_lock(rq1);
2646 : __acquire(rq2->lock); /* Fake it out ;) */
2647 0 : }
2648 :
2649 : /*
2650 : * double_rq_unlock - safely unlock two runqueues
2651 : *
2652 : * Note this does not restore interrupts like task_rq_unlock,
2653 : * you need to do so manually after calling.
2654 : */
2655 0 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2656 : __releases(rq1->lock)
2657 : __releases(rq2->lock)
2658 : {
2659 0 : BUG_ON(rq1 != rq2);
2660 0 : raw_spin_rq_unlock(rq1);
2661 : __release(rq2->lock);
2662 0 : }
2663 :
2664 : #endif
2665 :
2666 : extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2667 : extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2668 :
2669 : #ifdef CONFIG_SCHED_DEBUG
2670 : extern bool sched_debug_verbose;
2671 :
2672 : extern void print_cfs_stats(struct seq_file *m, int cpu);
2673 : extern void print_rt_stats(struct seq_file *m, int cpu);
2674 : extern void print_dl_stats(struct seq_file *m, int cpu);
2675 : extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2676 : extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2677 : extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2678 :
2679 : extern void resched_latency_warn(int cpu, u64 latency);
2680 : #ifdef CONFIG_NUMA_BALANCING
2681 : extern void
2682 : show_numa_stats(struct task_struct *p, struct seq_file *m);
2683 : extern void
2684 : print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2685 : unsigned long tpf, unsigned long gsf, unsigned long gpf);
2686 : #endif /* CONFIG_NUMA_BALANCING */
2687 : #else
2688 : static inline void resched_latency_warn(int cpu, u64 latency) {}
2689 : #endif /* CONFIG_SCHED_DEBUG */
2690 :
2691 : extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2692 : extern void init_rt_rq(struct rt_rq *rt_rq);
2693 : extern void init_dl_rq(struct dl_rq *dl_rq);
2694 :
2695 : extern void cfs_bandwidth_usage_inc(void);
2696 : extern void cfs_bandwidth_usage_dec(void);
2697 :
2698 : #ifdef CONFIG_NO_HZ_COMMON
2699 : #define NOHZ_BALANCE_KICK_BIT 0
2700 : #define NOHZ_STATS_KICK_BIT 1
2701 : #define NOHZ_NEWILB_KICK_BIT 2
2702 : #define NOHZ_NEXT_KICK_BIT 3
2703 :
2704 : /* Run rebalance_domains() */
2705 : #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2706 : /* Update blocked load */
2707 : #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2708 : /* Update blocked load when entering idle */
2709 : #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2710 : /* Update nohz.next_balance */
2711 : #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2712 :
2713 : #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2714 :
2715 : #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2716 :
2717 : extern void nohz_balance_exit_idle(struct rq *rq);
2718 : #else
2719 : static inline void nohz_balance_exit_idle(struct rq *rq) { }
2720 : #endif
2721 :
2722 : #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2723 : extern void nohz_run_idle_balance(int cpu);
2724 : #else
2725 : static inline void nohz_run_idle_balance(int cpu) { }
2726 : #endif
2727 :
2728 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2729 : struct irqtime {
2730 : u64 total;
2731 : u64 tick_delta;
2732 : u64 irq_start_time;
2733 : struct u64_stats_sync sync;
2734 : };
2735 :
2736 : DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2737 :
2738 : /*
2739 : * Returns the irqtime minus the softirq time computed by ksoftirqd.
2740 : * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2741 : * and never move forward.
2742 : */
2743 : static inline u64 irq_time_read(int cpu)
2744 : {
2745 : struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2746 : unsigned int seq;
2747 : u64 total;
2748 :
2749 : do {
2750 : seq = __u64_stats_fetch_begin(&irqtime->sync);
2751 : total = irqtime->total;
2752 : } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2753 :
2754 : return total;
2755 : }
2756 : #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2757 :
2758 : #ifdef CONFIG_CPU_FREQ
2759 : DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2760 :
2761 : /**
2762 : * cpufreq_update_util - Take a note about CPU utilization changes.
2763 : * @rq: Runqueue to carry out the update for.
2764 : * @flags: Update reason flags.
2765 : *
2766 : * This function is called by the scheduler on the CPU whose utilization is
2767 : * being updated.
2768 : *
2769 : * It can only be called from RCU-sched read-side critical sections.
2770 : *
2771 : * The way cpufreq is currently arranged requires it to evaluate the CPU
2772 : * performance state (frequency/voltage) on a regular basis to prevent it from
2773 : * being stuck in a completely inadequate performance level for too long.
2774 : * That is not guaranteed to happen if the updates are only triggered from CFS
2775 : * and DL, though, because they may not be coming in if only RT tasks are
2776 : * active all the time (or there are RT tasks only).
2777 : *
2778 : * As a workaround for that issue, this function is called periodically by the
2779 : * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2780 : * but that really is a band-aid. Going forward it should be replaced with
2781 : * solutions targeted more specifically at RT tasks.
2782 : */
2783 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2784 : {
2785 : struct update_util_data *data;
2786 :
2787 : data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2788 : cpu_of(rq)));
2789 : if (data)
2790 : data->func(data, rq_clock(rq), flags);
2791 : }
2792 : #else
2793 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2794 : #endif /* CONFIG_CPU_FREQ */
2795 :
2796 : #ifdef arch_scale_freq_capacity
2797 : # ifndef arch_scale_freq_invariant
2798 : # define arch_scale_freq_invariant() true
2799 : # endif
2800 : #else
2801 : # define arch_scale_freq_invariant() false
2802 : #endif
2803 :
2804 : #ifdef CONFIG_SMP
2805 : static inline unsigned long capacity_orig_of(int cpu)
2806 : {
2807 : return cpu_rq(cpu)->cpu_capacity_orig;
2808 : }
2809 :
2810 : /**
2811 : * enum cpu_util_type - CPU utilization type
2812 : * @FREQUENCY_UTIL: Utilization used to select frequency
2813 : * @ENERGY_UTIL: Utilization used during energy calculation
2814 : *
2815 : * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2816 : * need to be aggregated differently depending on the usage made of them. This
2817 : * enum is used within effective_cpu_util() to differentiate the types of
2818 : * utilization expected by the callers, and adjust the aggregation accordingly.
2819 : */
2820 : enum cpu_util_type {
2821 : FREQUENCY_UTIL,
2822 : ENERGY_UTIL,
2823 : };
2824 :
2825 : unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2826 : unsigned long max, enum cpu_util_type type,
2827 : struct task_struct *p);
2828 :
2829 : static inline unsigned long cpu_bw_dl(struct rq *rq)
2830 : {
2831 : return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2832 : }
2833 :
2834 : static inline unsigned long cpu_util_dl(struct rq *rq)
2835 : {
2836 : return READ_ONCE(rq->avg_dl.util_avg);
2837 : }
2838 :
2839 : /**
2840 : * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2841 : * @cpu: the CPU to get the utilization for.
2842 : *
2843 : * The unit of the return value must be the same as the one of CPU capacity
2844 : * so that CPU utilization can be compared with CPU capacity.
2845 : *
2846 : * CPU utilization is the sum of running time of runnable tasks plus the
2847 : * recent utilization of currently non-runnable tasks on that CPU.
2848 : * It represents the amount of CPU capacity currently used by CFS tasks in
2849 : * the range [0..max CPU capacity] with max CPU capacity being the CPU
2850 : * capacity at f_max.
2851 : *
2852 : * The estimated CPU utilization is defined as the maximum between CPU
2853 : * utilization and sum of the estimated utilization of the currently
2854 : * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2855 : * previously-executed tasks, which helps better deduce how busy a CPU will
2856 : * be when a long-sleeping task wakes up. The contribution to CPU utilization
2857 : * of such a task would be significantly decayed at this point of time.
2858 : *
2859 : * CPU utilization can be higher than the current CPU capacity
2860 : * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2861 : * of rounding errors as well as task migrations or wakeups of new tasks.
2862 : * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2863 : * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2864 : * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2865 : * capacity. CPU utilization is allowed to overshoot current CPU capacity
2866 : * though since this is useful for predicting the CPU capacity required
2867 : * after task migrations (scheduler-driven DVFS).
2868 : *
2869 : * Return: (Estimated) utilization for the specified CPU.
2870 : */
2871 : static inline unsigned long cpu_util_cfs(int cpu)
2872 : {
2873 : struct cfs_rq *cfs_rq;
2874 : unsigned long util;
2875 :
2876 : cfs_rq = &cpu_rq(cpu)->cfs;
2877 : util = READ_ONCE(cfs_rq->avg.util_avg);
2878 :
2879 : if (sched_feat(UTIL_EST)) {
2880 : util = max_t(unsigned long, util,
2881 : READ_ONCE(cfs_rq->avg.util_est.enqueued));
2882 : }
2883 :
2884 : return min(util, capacity_orig_of(cpu));
2885 : }
2886 :
2887 : static inline unsigned long cpu_util_rt(struct rq *rq)
2888 : {
2889 : return READ_ONCE(rq->avg_rt.util_avg);
2890 : }
2891 : #endif
2892 :
2893 : #ifdef CONFIG_UCLAMP_TASK
2894 : unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2895 :
2896 : /**
2897 : * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2898 : * @rq: The rq to clamp against. Must not be NULL.
2899 : * @util: The util value to clamp.
2900 : * @p: The task to clamp against. Can be NULL if you want to clamp
2901 : * against @rq only.
2902 : *
2903 : * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2904 : *
2905 : * If sched_uclamp_used static key is disabled, then just return the util
2906 : * without any clamping since uclamp aggregation at the rq level in the fast
2907 : * path is disabled, rendering this operation a NOP.
2908 : *
2909 : * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2910 : * will return the correct effective uclamp value of the task even if the
2911 : * static key is disabled.
2912 : */
2913 : static __always_inline
2914 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2915 : struct task_struct *p)
2916 : {
2917 : unsigned long min_util = 0;
2918 : unsigned long max_util = 0;
2919 :
2920 : if (!static_branch_likely(&sched_uclamp_used))
2921 : return util;
2922 :
2923 : if (p) {
2924 : min_util = uclamp_eff_value(p, UCLAMP_MIN);
2925 : max_util = uclamp_eff_value(p, UCLAMP_MAX);
2926 :
2927 : /*
2928 : * Ignore last runnable task's max clamp, as this task will
2929 : * reset it. Similarly, no need to read the rq's min clamp.
2930 : */
2931 : if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2932 : goto out;
2933 : }
2934 :
2935 : min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2936 : max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2937 : out:
2938 : /*
2939 : * Since CPU's {min,max}_util clamps are MAX aggregated considering
2940 : * RUNNABLE tasks with _different_ clamps, we can end up with an
2941 : * inversion. Fix it now when the clamps are applied.
2942 : */
2943 : if (unlikely(min_util >= max_util))
2944 : return min_util;
2945 :
2946 : return clamp(util, min_util, max_util);
2947 : }
2948 :
2949 : /* Is the rq being capped/throttled by uclamp_max? */
2950 : static inline bool uclamp_rq_is_capped(struct rq *rq)
2951 : {
2952 : unsigned long rq_util;
2953 : unsigned long max_util;
2954 :
2955 : if (!static_branch_likely(&sched_uclamp_used))
2956 : return false;
2957 :
2958 : rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
2959 : max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2960 :
2961 : return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
2962 : }
2963 :
2964 : /*
2965 : * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2966 : * by default in the fast path and only gets turned on once userspace performs
2967 : * an operation that requires it.
2968 : *
2969 : * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2970 : * hence is active.
2971 : */
2972 : static inline bool uclamp_is_used(void)
2973 : {
2974 : return static_branch_likely(&sched_uclamp_used);
2975 : }
2976 : #else /* CONFIG_UCLAMP_TASK */
2977 : static inline
2978 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2979 : struct task_struct *p)
2980 : {
2981 : return util;
2982 : }
2983 :
2984 : static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
2985 :
2986 : static inline bool uclamp_is_used(void)
2987 : {
2988 : return false;
2989 : }
2990 : #endif /* CONFIG_UCLAMP_TASK */
2991 :
2992 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2993 : static inline unsigned long cpu_util_irq(struct rq *rq)
2994 : {
2995 : return rq->avg_irq.util_avg;
2996 : }
2997 :
2998 : static inline
2999 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3000 : {
3001 : util *= (max - irq);
3002 : util /= max;
3003 :
3004 : return util;
3005 :
3006 : }
3007 : #else
3008 : static inline unsigned long cpu_util_irq(struct rq *rq)
3009 : {
3010 : return 0;
3011 : }
3012 :
3013 : static inline
3014 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3015 : {
3016 : return util;
3017 : }
3018 : #endif
3019 :
3020 : #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3021 :
3022 : #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3023 :
3024 : DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3025 :
3026 : static inline bool sched_energy_enabled(void)
3027 : {
3028 : return static_branch_unlikely(&sched_energy_present);
3029 : }
3030 :
3031 : #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3032 :
3033 : #define perf_domain_span(pd) NULL
3034 : static inline bool sched_energy_enabled(void) { return false; }
3035 :
3036 : #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3037 :
3038 : #ifdef CONFIG_MEMBARRIER
3039 : /*
3040 : * The scheduler provides memory barriers required by membarrier between:
3041 : * - prior user-space memory accesses and store to rq->membarrier_state,
3042 : * - store to rq->membarrier_state and following user-space memory accesses.
3043 : * In the same way it provides those guarantees around store to rq->curr.
3044 : */
3045 : static inline void membarrier_switch_mm(struct rq *rq,
3046 : struct mm_struct *prev_mm,
3047 : struct mm_struct *next_mm)
3048 : {
3049 : int membarrier_state;
3050 :
3051 0 : if (prev_mm == next_mm)
3052 : return;
3053 :
3054 0 : membarrier_state = atomic_read(&next_mm->membarrier_state);
3055 0 : if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3056 : return;
3057 :
3058 0 : WRITE_ONCE(rq->membarrier_state, membarrier_state);
3059 : }
3060 : #else
3061 : static inline void membarrier_switch_mm(struct rq *rq,
3062 : struct mm_struct *prev_mm,
3063 : struct mm_struct *next_mm)
3064 : {
3065 : }
3066 : #endif
3067 :
3068 : #ifdef CONFIG_SMP
3069 : static inline bool is_per_cpu_kthread(struct task_struct *p)
3070 : {
3071 : if (!(p->flags & PF_KTHREAD))
3072 : return false;
3073 :
3074 : if (p->nr_cpus_allowed != 1)
3075 : return false;
3076 :
3077 : return true;
3078 : }
3079 : #endif
3080 :
3081 : extern void swake_up_all_locked(struct swait_queue_head *q);
3082 : extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3083 :
3084 : #ifdef CONFIG_PREEMPT_DYNAMIC
3085 : extern int preempt_dynamic_mode;
3086 : extern int sched_dynamic_mode(const char *str);
3087 : extern void sched_dynamic_update(int mode);
3088 : #endif
3089 :
3090 : #endif /* _KERNEL_SCHED_SCHED_H */
|
