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/context_tracking.h>
31 #include <linux/cpufreq.h>
32 #include <linux/cpumask_api.h>
33 #include <linux/ctype.h>
34 #include <linux/file.h>
35 #include <linux/fs_api.h>
36 #include <linux/hrtimer_api.h>
37 #include <linux/interrupt.h>
38 #include <linux/irq_work.h>
39 #include <linux/jiffies.h>
40 #include <linux/kref_api.h>
41 #include <linux/kthread.h>
42 #include <linux/ktime_api.h>
43 #include <linux/lockdep_api.h>
44 #include <linux/lockdep.h>
45 #include <linux/minmax.h>
46 #include <linux/mm.h>
47 #include <linux/module.h>
48 #include <linux/mutex_api.h>
49 #include <linux/plist.h>
50 #include <linux/poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/psi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/seq_file.h>
56 #include <linux/seqlock.h>
57 #include <linux/softirq.h>
58 #include <linux/spinlock_api.h>
59 #include <linux/static_key.h>
60 #include <linux/stop_machine.h>
61 #include <linux/syscalls_api.h>
62 #include <linux/syscalls.h>
63 #include <linux/tick.h>
64 #include <linux/topology.h>
65 #include <linux/types.h>
66 #include <linux/u64_stats_sync_api.h>
67 #include <linux/uaccess.h>
68 #include <linux/wait_api.h>
69 #include <linux/wait_bit.h>
70 #include <linux/workqueue_api.h>
71
72 #include <trace/events/power.h>
73 #include <trace/events/sched.h>
74
75 #include "../workqueue_internal.h"
76
77 struct rq;
78 struct cfs_rq;
79 struct rt_rq;
80 struct sched_group;
81 struct cpuidle_state;
82
83 #ifdef CONFIG_PARAVIRT
84 # include <asm/paravirt.h>
85 # include <asm/paravirt_api_clock.h>
86 #endif
87
88 #include <asm/barrier.h>
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 /* task_struct::on_rq states: */
100 #define TASK_ON_RQ_QUEUED 1
101 #define TASK_ON_RQ_MIGRATING 2
102
103 extern __read_mostly int scheduler_running;
104
105 extern unsigned long calc_load_update;
106 extern atomic_long_t calc_load_tasks;
107
108 extern void calc_global_load_tick(struct rq *this_rq);
109 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
110
111 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
112
113 extern int sysctl_sched_rt_period;
114 extern int sysctl_sched_rt_runtime;
115 extern int sched_rr_timeslice;
116
117 /*
118 * Asymmetric CPU capacity bits
119 */
120 struct asym_cap_data {
121 struct list_head link;
122 struct rcu_head rcu;
123 unsigned long capacity;
124 unsigned long cpus[];
125 };
126
127 extern struct list_head asym_cap_list;
128
129 #define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
130
131 /*
132 * Helpers for converting nanosecond timing to jiffy resolution
133 */
134 #define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
135
136 /*
137 * Increase resolution of nice-level calculations for 64-bit architectures.
138 * The extra resolution improves shares distribution and load balancing of
139 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
140 * hierarchies, especially on larger systems. This is not a user-visible change
141 * and does not change the user-interface for setting shares/weights.
142 *
143 * We increase resolution only if we have enough bits to allow this increased
144 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
145 * are pretty high and the returns do not justify the increased costs.
146 *
147 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
148 * increase coverage and consistency always enable it on 64-bit platforms.
149 */
150 #ifdef CONFIG_64BIT
151 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
152 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
153 # define scale_load_down(w) \
154 ({ \
155 unsigned long __w = (w); \
156 \
157 if (__w) \
158 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
159 __w; \
160 })
161 #else
162 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
163 # define scale_load(w) (w)
164 # define scale_load_down(w) (w)
165 #endif
166
167 /*
168 * Task weight (visible to users) and its load (invisible to users) have
169 * independent resolution, but they should be well calibrated. We use
170 * scale_load() and scale_load_down(w) to convert between them. The
171 * following must be true:
172 *
173 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
174 *
175 */
176 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
177
178 /*
179 * Single value that decides SCHED_DEADLINE internal math precision.
180 * 10 -> just above 1us
181 * 9 -> just above 0.5us
182 */
183 #define DL_SCALE 10
184
185 /*
186 * Single value that denotes runtime == period, ie unlimited time.
187 */
188 #define RUNTIME_INF ((u64)~0ULL)
189
190 static inline int idle_policy(int policy)
191 {
192 return policy == SCHED_IDLE;
193 }
194
195 static inline int fair_policy(int policy)
196 {
197 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
198 }
199
200 static inline int rt_policy(int policy)
201 {
202 return policy == SCHED_FIFO || policy == SCHED_RR;
203 }
204
205 static inline int dl_policy(int policy)
206 {
207 return policy == SCHED_DEADLINE;
208 }
209
210 static inline bool valid_policy(int policy)
211 {
212 return idle_policy(policy) || fair_policy(policy) ||
213 rt_policy(policy) || dl_policy(policy);
214 }
215
216 static inline int task_has_idle_policy(struct task_struct *p)
217 {
218 return idle_policy(p->policy);
219 }
220
221 static inline int task_has_rt_policy(struct task_struct *p)
222 {
223 return rt_policy(p->policy);
224 }
225
226 static inline int task_has_dl_policy(struct task_struct *p)
227 {
228 return dl_policy(p->policy);
229 }
230
231 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
232
233 static inline void update_avg(u64 *avg, u64 sample)
234 {
235 s64 diff = sample - *avg;
236
237 *avg += diff / 8;
238 }
239
240 /*
241 * Shifting a value by an exponent greater *or equal* to the size of said value
242 * is UB; cap at size-1.
243 */
244 #define shr_bound(val, shift) \
245 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
246
247 /*
248 * !! For sched_setattr_nocheck() (kernel) only !!
249 *
250 * This is actually gross. :(
251 *
252 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
253 * tasks, but still be able to sleep. We need this on platforms that cannot
254 * atomically change clock frequency. Remove once fast switching will be
255 * available on such platforms.
256 *
257 * SUGOV stands for SchedUtil GOVernor.
258 */
259 #define SCHED_FLAG_SUGOV 0x10000000
260
261 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
262
263 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
264 {
265 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
266 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
267 #else
268 return false;
269 #endif
270 }
271
272 /*
273 * Tells if entity @a should preempt entity @b.
274 */
275 static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
276 const struct sched_dl_entity *b)
277 {
278 return dl_entity_is_special(a) ||
279 dl_time_before(a->deadline, b->deadline);
280 }
281
282 /*
283 * This is the priority-queue data structure of the RT scheduling class:
284 */
285 struct rt_prio_array {
286 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
287 struct list_head queue[MAX_RT_PRIO];
288 };
289
290 struct rt_bandwidth {
291 /* nests inside the rq lock: */
292 raw_spinlock_t rt_runtime_lock;
293 ktime_t rt_period;
294 u64 rt_runtime;
295 struct hrtimer rt_period_timer;
296 unsigned int rt_period_active;
297 };
298
299 static inline int dl_bandwidth_enabled(void)
300 {
301 return sysctl_sched_rt_runtime >= 0;
302 }
303
304 /*
305 * To keep the bandwidth of -deadline tasks under control
306 * we need some place where:
307 * - store the maximum -deadline bandwidth of each cpu;
308 * - cache the fraction of bandwidth that is currently allocated in
309 * each root domain;
310 *
311 * This is all done in the data structure below. It is similar to the
312 * one used for RT-throttling (rt_bandwidth), with the main difference
313 * that, since here we are only interested in admission control, we
314 * do not decrease any runtime while the group "executes", neither we
315 * need a timer to replenish it.
316 *
317 * With respect to SMP, bandwidth is given on a per root domain basis,
318 * meaning that:
319 * - bw (< 100%) is the deadline bandwidth of each CPU;
320 * - total_bw is the currently allocated bandwidth in each root domain;
321 */
322 struct dl_bw {
323 raw_spinlock_t lock;
324 u64 bw;
325 u64 total_bw;
326 };
327
328 extern void init_dl_bw(struct dl_bw *dl_b);
329 extern int sched_dl_global_validate(void);
330 extern void sched_dl_do_global(void);
331 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
332 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
333 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
334 extern bool __checkparam_dl(const struct sched_attr *attr);
335 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
336 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
337 extern int dl_bw_check_overflow(int cpu);
338
339 /*
340 * SCHED_DEADLINE supports servers (nested scheduling) with the following
341 * interface:
342 *
343 * dl_se::rq -- runqueue we belong to.
344 *
345 * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
346 * server when it runs out of tasks to run.
347 *
348 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
349 * returns NULL.
350 *
351 * dl_server_update() -- called from update_curr_common(), propagates runtime
352 * to the server.
353 *
354 * dl_server_start()
355 * dl_server_stop() -- start/stop the server when it has (no) tasks.
356 *
357 * dl_server_init() -- initializes the server.
358 */
359 extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
360 extern void dl_server_start(struct sched_dl_entity *dl_se);
361 extern void dl_server_stop(struct sched_dl_entity *dl_se);
362 extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
363 dl_server_has_tasks_f has_tasks,
364 dl_server_pick_f pick);
365
366 #ifdef CONFIG_CGROUP_SCHED
367
368 extern struct list_head task_groups;
369
370 struct cfs_bandwidth {
371 #ifdef CONFIG_CFS_BANDWIDTH
372 raw_spinlock_t lock;
373 ktime_t period;
374 u64 quota;
375 u64 runtime;
376 u64 burst;
377 u64 runtime_snap;
378 s64 hierarchical_quota;
379
380 u8 idle;
381 u8 period_active;
382 u8 slack_started;
383 struct hrtimer period_timer;
384 struct hrtimer slack_timer;
385 struct list_head throttled_cfs_rq;
386
387 /* Statistics: */
388 int nr_periods;
389 int nr_throttled;
390 int nr_burst;
391 u64 throttled_time;
392 u64 burst_time;
393 #endif
394 };
395
396 /* Task group related information */
397 struct task_group {
398 struct cgroup_subsys_state css;
399
400 #ifdef CONFIG_FAIR_GROUP_SCHED
401 /* schedulable entities of this group on each CPU */
402 struct sched_entity **se;
403 /* runqueue "owned" by this group on each CPU */
404 struct cfs_rq **cfs_rq;
405 unsigned long shares;
406
407 /* A positive value indicates that this is a SCHED_IDLE group. */
408 int idle;
409
410 #ifdef CONFIG_SMP
411 /*
412 * load_avg can be heavily contended at clock tick time, so put
413 * it in its own cache-line separated from the fields above which
414 * will also be accessed at each tick.
415 */
416 atomic_long_t load_avg ____cacheline_aligned;
417 #endif
418 #endif
419
420 #ifdef CONFIG_RT_GROUP_SCHED
421 struct sched_rt_entity **rt_se;
422 struct rt_rq **rt_rq;
423
424 struct rt_bandwidth rt_bandwidth;
425 #endif
426
427 struct rcu_head rcu;
428 struct list_head list;
429
430 struct task_group *parent;
431 struct list_head siblings;
432 struct list_head children;
433
434 #ifdef CONFIG_SCHED_AUTOGROUP
435 struct autogroup *autogroup;
436 #endif
437
438 struct cfs_bandwidth cfs_bandwidth;
439
440 #ifdef CONFIG_UCLAMP_TASK_GROUP
441 /* The two decimal precision [%] value requested from user-space */
442 unsigned int uclamp_pct[UCLAMP_CNT];
443 /* Clamp values requested for a task group */
444 struct uclamp_se uclamp_req[UCLAMP_CNT];
445 /* Effective clamp values used for a task group */
446 struct uclamp_se uclamp[UCLAMP_CNT];
447 #endif
448
449 };
450
451 #ifdef CONFIG_FAIR_GROUP_SCHED
452 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
453
454 /*
455 * A weight of 0 or 1 can cause arithmetics problems.
456 * A weight of a cfs_rq is the sum of weights of which entities
457 * are queued on this cfs_rq, so a weight of a entity should not be
458 * too large, so as the shares value of a task group.
459 * (The default weight is 1024 - so there's no practical
460 * limitation from this.)
461 */
462 #define MIN_SHARES (1UL << 1)
463 #define MAX_SHARES (1UL << 18)
464 #endif
465
466 typedef int (*tg_visitor)(struct task_group *, void *);
467
468 extern int walk_tg_tree_from(struct task_group *from,
469 tg_visitor down, tg_visitor up, void *data);
470
471 /*
472 * Iterate the full tree, calling @down when first entering a node and @up when
473 * leaving it for the final time.
474 *
475 * Caller must hold rcu_lock or sufficient equivalent.
476 */
477 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
478 {
479 return walk_tg_tree_from(&root_task_group, down, up, data);
480 }
481
482 extern int tg_nop(struct task_group *tg, void *data);
483
484 #ifdef CONFIG_FAIR_GROUP_SCHED
485 extern void free_fair_sched_group(struct task_group *tg);
486 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
487 extern void online_fair_sched_group(struct task_group *tg);
488 extern void unregister_fair_sched_group(struct task_group *tg);
489 #else
490 static inline void free_fair_sched_group(struct task_group *tg) { }
491 static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
492 {
493 return 1;
494 }
495 static inline void online_fair_sched_group(struct task_group *tg) { }
496 static inline void unregister_fair_sched_group(struct task_group *tg) { }
497 #endif
498
499 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
500 struct sched_entity *se, int cpu,
501 struct sched_entity *parent);
502 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
503
504 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
505 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
506 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
507 extern bool cfs_task_bw_constrained(struct task_struct *p);
508
509 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
510 struct sched_rt_entity *rt_se, int cpu,
511 struct sched_rt_entity *parent);
512 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
513 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
514 extern long sched_group_rt_runtime(struct task_group *tg);
515 extern long sched_group_rt_period(struct task_group *tg);
516 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
517
518 extern struct task_group *sched_create_group(struct task_group *parent);
519 extern void sched_online_group(struct task_group *tg,
520 struct task_group *parent);
521 extern void sched_destroy_group(struct task_group *tg);
522 extern void sched_release_group(struct task_group *tg);
523
524 extern void sched_move_task(struct task_struct *tsk);
525
526 #ifdef CONFIG_FAIR_GROUP_SCHED
527 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
528
529 extern int sched_group_set_idle(struct task_group *tg, long idle);
530
531 #ifdef CONFIG_SMP
532 extern void set_task_rq_fair(struct sched_entity *se,
533 struct cfs_rq *prev, struct cfs_rq *next);
534 #else /* !CONFIG_SMP */
535 static inline void set_task_rq_fair(struct sched_entity *se,
536 struct cfs_rq *prev, struct cfs_rq *next) { }
537 #endif /* CONFIG_SMP */
538 #endif /* CONFIG_FAIR_GROUP_SCHED */
539
540 #else /* CONFIG_CGROUP_SCHED */
541
542 struct cfs_bandwidth { };
543
544 static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
545
546 #endif /* CONFIG_CGROUP_SCHED */
547
548 extern void unregister_rt_sched_group(struct task_group *tg);
549 extern void free_rt_sched_group(struct task_group *tg);
550 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
551
552 /*
553 * u64_u32_load/u64_u32_store
554 *
555 * Use a copy of a u64 value to protect against data race. This is only
556 * applicable for 32-bits architectures.
557 */
558 #ifdef CONFIG_64BIT
559 # define u64_u32_load_copy(var, copy) var
560 # define u64_u32_store_copy(var, copy, val) (var = val)
561 #else
562 # define u64_u32_load_copy(var, copy) \
563 ({ \
564 u64 __val, __val_copy; \
565 do { \
566 __val_copy = copy; \
567 /* \
568 * paired with u64_u32_store_copy(), ordering access \
569 * to var and copy. \
570 */ \
571 smp_rmb(); \
572 __val = var; \
573 } while (__val != __val_copy); \
574 __val; \
575 })
576 # define u64_u32_store_copy(var, copy, val) \
577 do { \
578 typeof(val) __val = (val); \
579 var = __val; \
580 /* \
581 * paired with u64_u32_load_copy(), ordering access to var and \
582 * copy. \
583 */ \
584 smp_wmb(); \
585 copy = __val; \
586 } while (0)
587 #endif
588 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
589 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
590
591 /* CFS-related fields in a runqueue */
592 struct cfs_rq {
593 struct load_weight load;
594 unsigned int nr_running;
595 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
596 unsigned int idle_nr_running; /* SCHED_IDLE */
597 unsigned int idle_h_nr_running; /* SCHED_IDLE */
598
599 s64 avg_vruntime;
600 u64 avg_load;
601
602 u64 exec_clock;
603 u64 min_vruntime;
604 #ifdef CONFIG_SCHED_CORE
605 unsigned int forceidle_seq;
606 u64 min_vruntime_fi;
607 #endif
608
609 #ifndef CONFIG_64BIT
610 u64 min_vruntime_copy;
611 #endif
612
613 struct rb_root_cached tasks_timeline;
614
615 /*
616 * 'curr' points to currently running entity on this cfs_rq.
617 * It is set to NULL otherwise (i.e when none are currently running).
618 */
619 struct sched_entity *curr;
620 struct sched_entity *next;
621
622 #ifdef CONFIG_SCHED_DEBUG
623 unsigned int nr_spread_over;
624 #endif
625
626 #ifdef CONFIG_SMP
627 /*
628 * CFS load tracking
629 */
630 struct sched_avg avg;
631 #ifndef CONFIG_64BIT
632 u64 last_update_time_copy;
633 #endif
634 struct {
635 raw_spinlock_t lock ____cacheline_aligned;
636 int nr;
637 unsigned long load_avg;
638 unsigned long util_avg;
639 unsigned long runnable_avg;
640 } removed;
641
642 #ifdef CONFIG_FAIR_GROUP_SCHED
643 u64 last_update_tg_load_avg;
644 unsigned long tg_load_avg_contrib;
645 long propagate;
646 long prop_runnable_sum;
647
648 /*
649 * h_load = weight * f(tg)
650 *
651 * Where f(tg) is the recursive weight fraction assigned to
652 * this group.
653 */
654 unsigned long h_load;
655 u64 last_h_load_update;
656 struct sched_entity *h_load_next;
657 #endif /* CONFIG_FAIR_GROUP_SCHED */
658 #endif /* CONFIG_SMP */
659
660 #ifdef CONFIG_FAIR_GROUP_SCHED
661 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
662
663 /*
664 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
665 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
666 * (like users, containers etc.)
667 *
668 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
669 * This list is used during load balance.
670 */
671 int on_list;
672 struct list_head leaf_cfs_rq_list;
673 struct task_group *tg; /* group that "owns" this runqueue */
674
675 /* Locally cached copy of our task_group's idle value */
676 int idle;
677
678 #ifdef CONFIG_CFS_BANDWIDTH
679 int runtime_enabled;
680 s64 runtime_remaining;
681
682 u64 throttled_pelt_idle;
683 #ifndef CONFIG_64BIT
684 u64 throttled_pelt_idle_copy;
685 #endif
686 u64 throttled_clock;
687 u64 throttled_clock_pelt;
688 u64 throttled_clock_pelt_time;
689 u64 throttled_clock_self;
690 u64 throttled_clock_self_time;
691 int throttled;
692 int throttle_count;
693 struct list_head throttled_list;
694 struct list_head throttled_csd_list;
695 #endif /* CONFIG_CFS_BANDWIDTH */
696 #endif /* CONFIG_FAIR_GROUP_SCHED */
697 };
698
699 static inline int rt_bandwidth_enabled(void)
700 {
701 return sysctl_sched_rt_runtime >= 0;
702 }
703
704 /* RT IPI pull logic requires IRQ_WORK */
705 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
706 # define HAVE_RT_PUSH_IPI
707 #endif
708
709 /* Real-Time classes' related field in a runqueue: */
710 struct rt_rq {
711 struct rt_prio_array active;
712 unsigned int rt_nr_running;
713 unsigned int rr_nr_running;
714 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
715 struct {
716 int curr; /* highest queued rt task prio */
717 #ifdef CONFIG_SMP
718 int next; /* next highest */
719 #endif
720 } highest_prio;
721 #endif
722 #ifdef CONFIG_SMP
723 bool overloaded;
724 struct plist_head pushable_tasks;
725
726 #endif /* CONFIG_SMP */
727 int rt_queued;
728
729 int rt_throttled;
730 u64 rt_time;
731 u64 rt_runtime;
732 /* Nests inside the rq lock: */
733 raw_spinlock_t rt_runtime_lock;
734
735 #ifdef CONFIG_RT_GROUP_SCHED
736 unsigned int rt_nr_boosted;
737
738 struct rq *rq;
739 struct task_group *tg;
740 #endif
741 };
742
743 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
744 {
745 return rt_rq->rt_queued && rt_rq->rt_nr_running;
746 }
747
748 /* Deadline class' related fields in a runqueue */
749 struct dl_rq {
750 /* runqueue is an rbtree, ordered by deadline */
751 struct rb_root_cached root;
752
753 unsigned int dl_nr_running;
754
755 #ifdef CONFIG_SMP
756 /*
757 * Deadline values of the currently executing and the
758 * earliest ready task on this rq. Caching these facilitates
759 * the decision whether or not a ready but not running task
760 * should migrate somewhere else.
761 */
762 struct {
763 u64 curr;
764 u64 next;
765 } earliest_dl;
766
767 bool overloaded;
768
769 /*
770 * Tasks on this rq that can be pushed away. They are kept in
771 * an rb-tree, ordered by tasks' deadlines, with caching
772 * of the leftmost (earliest deadline) element.
773 */
774 struct rb_root_cached pushable_dl_tasks_root;
775 #else
776 struct dl_bw dl_bw;
777 #endif
778 /*
779 * "Active utilization" for this runqueue: increased when a
780 * task wakes up (becomes TASK_RUNNING) and decreased when a
781 * task blocks
782 */
783 u64 running_bw;
784
785 /*
786 * Utilization of the tasks "assigned" to this runqueue (including
787 * the tasks that are in runqueue and the tasks that executed on this
788 * CPU and blocked). Increased when a task moves to this runqueue, and
789 * decreased when the task moves away (migrates, changes scheduling
790 * policy, or terminates).
791 * This is needed to compute the "inactive utilization" for the
792 * runqueue (inactive utilization = this_bw - running_bw).
793 */
794 u64 this_bw;
795 u64 extra_bw;
796
797 /*
798 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
799 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
800 */
801 u64 max_bw;
802
803 /*
804 * Inverse of the fraction of CPU utilization that can be reclaimed
805 * by the GRUB algorithm.
806 */
807 u64 bw_ratio;
808 };
809
810 #ifdef CONFIG_FAIR_GROUP_SCHED
811
812 /* An entity is a task if it doesn't "own" a runqueue */
813 #define entity_is_task(se) (!se->my_q)
814
815 static inline void se_update_runnable(struct sched_entity *se)
816 {
817 if (!entity_is_task(se))
818 se->runnable_weight = se->my_q->h_nr_running;
819 }
820
821 static inline long se_runnable(struct sched_entity *se)
822 {
823 if (entity_is_task(se))
824 return !!se->on_rq;
825 else
826 return se->runnable_weight;
827 }
828
829 #else /* !CONFIG_FAIR_GROUP_SCHED: */
830
831 #define entity_is_task(se) 1
832
833 static inline void se_update_runnable(struct sched_entity *se) { }
834
835 static inline long se_runnable(struct sched_entity *se)
836 {
837 return !!se->on_rq;
838 }
839
840 #endif /* !CONFIG_FAIR_GROUP_SCHED */
841
842 #ifdef CONFIG_SMP
843 /*
844 * XXX we want to get rid of these helpers and use the full load resolution.
845 */
846 static inline long se_weight(struct sched_entity *se)
847 {
848 return scale_load_down(se->load.weight);
849 }
850
851
852 static inline bool sched_asym_prefer(int a, int b)
853 {
854 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
855 }
856
857 struct perf_domain {
858 struct em_perf_domain *em_pd;
859 struct perf_domain *next;
860 struct rcu_head rcu;
861 };
862
863 /*
864 * We add the notion of a root-domain which will be used to define per-domain
865 * variables. Each exclusive cpuset essentially defines an island domain by
866 * fully partitioning the member CPUs from any other cpuset. Whenever a new
867 * exclusive cpuset is created, we also create and attach a new root-domain
868 * object.
869 *
870 */
871 struct root_domain {
872 atomic_t refcount;
873 atomic_t rto_count;
874 struct rcu_head rcu;
875 cpumask_var_t span;
876 cpumask_var_t online;
877
878 /*
879 * Indicate pullable load on at least one CPU, e.g:
880 * - More than one runnable task
881 * - Running task is misfit
882 */
883 bool overloaded;
884
885 /* Indicate one or more CPUs over-utilized (tipping point) */
886 bool overutilized;
887
888 /*
889 * The bit corresponding to a CPU gets set here if such CPU has more
890 * than one runnable -deadline task (as it is below for RT tasks).
891 */
892 cpumask_var_t dlo_mask;
893 atomic_t dlo_count;
894 struct dl_bw dl_bw;
895 struct cpudl cpudl;
896
897 /*
898 * Indicate whether a root_domain's dl_bw has been checked or
899 * updated. It's monotonously increasing value.
900 *
901 * Also, some corner cases, like 'wrap around' is dangerous, but given
902 * that u64 is 'big enough'. So that shouldn't be a concern.
903 */
904 u64 visit_gen;
905
906 #ifdef HAVE_RT_PUSH_IPI
907 /*
908 * For IPI pull requests, loop across the rto_mask.
909 */
910 struct irq_work rto_push_work;
911 raw_spinlock_t rto_lock;
912 /* These are only updated and read within rto_lock */
913 int rto_loop;
914 int rto_cpu;
915 /* These atomics are updated outside of a lock */
916 atomic_t rto_loop_next;
917 atomic_t rto_loop_start;
918 #endif
919 /*
920 * The "RT overload" flag: it gets set if a CPU has more than
921 * one runnable RT task.
922 */
923 cpumask_var_t rto_mask;
924 struct cpupri cpupri;
925
926 /*
927 * NULL-terminated list of performance domains intersecting with the
928 * CPUs of the rd. Protected by RCU.
929 */
930 struct perf_domain __rcu *pd;
931 };
932
933 extern void init_defrootdomain(void);
934 extern int sched_init_domains(const struct cpumask *cpu_map);
935 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
936 extern void sched_get_rd(struct root_domain *rd);
937 extern void sched_put_rd(struct root_domain *rd);
938
939 static inline int get_rd_overloaded(struct root_domain *rd)
940 {
941 return READ_ONCE(rd->overloaded);
942 }
943
944 static inline void set_rd_overloaded(struct root_domain *rd, int status)
945 {
946 if (get_rd_overloaded(rd) != status)
947 WRITE_ONCE(rd->overloaded, status);
948 }
949
950 #ifdef HAVE_RT_PUSH_IPI
951 extern void rto_push_irq_work_func(struct irq_work *work);
952 #endif
953 #endif /* CONFIG_SMP */
954
955 #ifdef CONFIG_UCLAMP_TASK
956 /*
957 * struct uclamp_bucket - Utilization clamp bucket
958 * @value: utilization clamp value for tasks on this clamp bucket
959 * @tasks: number of RUNNABLE tasks on this clamp bucket
960 *
961 * Keep track of how many tasks are RUNNABLE for a given utilization
962 * clamp value.
963 */
964 struct uclamp_bucket {
965 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
966 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
967 };
968
969 /*
970 * struct uclamp_rq - rq's utilization clamp
971 * @value: currently active clamp values for a rq
972 * @bucket: utilization clamp buckets affecting a rq
973 *
974 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
975 * A clamp value is affecting a rq when there is at least one task RUNNABLE
976 * (or actually running) with that value.
977 *
978 * There are up to UCLAMP_CNT possible different clamp values, currently there
979 * are only two: minimum utilization and maximum utilization.
980 *
981 * All utilization clamping values are MAX aggregated, since:
982 * - for util_min: we want to run the CPU at least at the max of the minimum
983 * utilization required by its currently RUNNABLE tasks.
984 * - for util_max: we want to allow the CPU to run up to the max of the
985 * maximum utilization allowed by its currently RUNNABLE tasks.
986 *
987 * Since on each system we expect only a limited number of different
988 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
989 * the metrics required to compute all the per-rq utilization clamp values.
990 */
991 struct uclamp_rq {
992 unsigned int value;
993 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
994 };
995
996 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
997 #endif /* CONFIG_UCLAMP_TASK */
998
999 struct balance_callback {
1000 struct balance_callback *next;
1001 void (*func)(struct rq *rq);
1002 };
1003
1004 /*
1005 * This is the main, per-CPU runqueue data structure.
1006 *
1007 * Locking rule: those places that want to lock multiple runqueues
1008 * (such as the load balancing or the thread migration code), lock
1009 * acquire operations must be ordered by ascending &runqueue.
1010 */
1011 struct rq {
1012 /* runqueue lock: */
1013 raw_spinlock_t __lock;
1014
1015 unsigned int nr_running;
1016 #ifdef CONFIG_NUMA_BALANCING
1017 unsigned int nr_numa_running;
1018 unsigned int nr_preferred_running;
1019 unsigned int numa_migrate_on;
1020 #endif
1021 #ifdef CONFIG_NO_HZ_COMMON
1022 #ifdef CONFIG_SMP
1023 unsigned long last_blocked_load_update_tick;
1024 unsigned int has_blocked_load;
1025 call_single_data_t nohz_csd;
1026 #endif /* CONFIG_SMP */
1027 unsigned int nohz_tick_stopped;
1028 atomic_t nohz_flags;
1029 #endif /* CONFIG_NO_HZ_COMMON */
1030
1031 #ifdef CONFIG_SMP
1032 unsigned int ttwu_pending;
1033 #endif
1034 u64 nr_switches;
1035
1036 #ifdef CONFIG_UCLAMP_TASK
1037 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1038 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1039 unsigned int uclamp_flags;
1040 #define UCLAMP_FLAG_IDLE 0x01
1041 #endif
1042
1043 struct cfs_rq cfs;
1044 struct rt_rq rt;
1045 struct dl_rq dl;
1046
1047 #ifdef CONFIG_FAIR_GROUP_SCHED
1048 /* list of leaf cfs_rq on this CPU: */
1049 struct list_head leaf_cfs_rq_list;
1050 struct list_head *tmp_alone_branch;
1051 #endif /* CONFIG_FAIR_GROUP_SCHED */
1052
1053 /*
1054 * This is part of a global counter where only the total sum
1055 * over all CPUs matters. A task can increase this counter on
1056 * one CPU and if it got migrated afterwards it may decrease
1057 * it on another CPU. Always updated under the runqueue lock:
1058 */
1059 unsigned int nr_uninterruptible;
1060
1061 struct task_struct __rcu *curr;
1062 struct task_struct *idle;
1063 struct task_struct *stop;
1064 unsigned long next_balance;
1065 struct mm_struct *prev_mm;
1066
1067 unsigned int clock_update_flags;
1068 u64 clock;
1069 /* Ensure that all clocks are in the same cache line */
1070 u64 clock_task ____cacheline_aligned;
1071 u64 clock_pelt;
1072 unsigned long lost_idle_time;
1073 u64 clock_pelt_idle;
1074 u64 clock_idle;
1075 #ifndef CONFIG_64BIT
1076 u64 clock_pelt_idle_copy;
1077 u64 clock_idle_copy;
1078 #endif
1079
1080 atomic_t nr_iowait;
1081
1082 #ifdef CONFIG_SCHED_DEBUG
1083 u64 last_seen_need_resched_ns;
1084 int ticks_without_resched;
1085 #endif
1086
1087 #ifdef CONFIG_MEMBARRIER
1088 int membarrier_state;
1089 #endif
1090
1091 #ifdef CONFIG_SMP
1092 struct root_domain *rd;
1093 struct sched_domain __rcu *sd;
1094
1095 unsigned long cpu_capacity;
1096
1097 struct balance_callback *balance_callback;
1098
1099 unsigned char nohz_idle_balance;
1100 unsigned char idle_balance;
1101
1102 unsigned long misfit_task_load;
1103
1104 /* For active balancing */
1105 int active_balance;
1106 int push_cpu;
1107 struct cpu_stop_work active_balance_work;
1108
1109 /* CPU of this runqueue: */
1110 int cpu;
1111 int online;
1112
1113 struct list_head cfs_tasks;
1114
1115 struct sched_avg avg_rt;
1116 struct sched_avg avg_dl;
1117 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1118 struct sched_avg avg_irq;
1119 #endif
1120 #ifdef CONFIG_SCHED_HW_PRESSURE
1121 struct sched_avg avg_hw;
1122 #endif
1123 u64 idle_stamp;
1124 u64 avg_idle;
1125
1126 /* This is used to determine avg_idle's max value */
1127 u64 max_idle_balance_cost;
1128
1129 #ifdef CONFIG_HOTPLUG_CPU
1130 struct rcuwait hotplug_wait;
1131 #endif
1132 #endif /* CONFIG_SMP */
1133
1134 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1135 u64 prev_irq_time;
1136 u64 psi_irq_time;
1137 #endif
1138 #ifdef CONFIG_PARAVIRT
1139 u64 prev_steal_time;
1140 #endif
1141 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1142 u64 prev_steal_time_rq;
1143 #endif
1144
1145 /* calc_load related fields */
1146 unsigned long calc_load_update;
1147 long calc_load_active;
1148
1149 #ifdef CONFIG_SCHED_HRTICK
1150 #ifdef CONFIG_SMP
1151 call_single_data_t hrtick_csd;
1152 #endif
1153 struct hrtimer hrtick_timer;
1154 ktime_t hrtick_time;
1155 #endif
1156
1157 #ifdef CONFIG_SCHEDSTATS
1158 /* latency stats */
1159 struct sched_info rq_sched_info;
1160 unsigned long long rq_cpu_time;
1161 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1162
1163 /* sys_sched_yield() stats */
1164 unsigned int yld_count;
1165
1166 /* schedule() stats */
1167 unsigned int sched_count;
1168 unsigned int sched_goidle;
1169
1170 /* try_to_wake_up() stats */
1171 unsigned int ttwu_count;
1172 unsigned int ttwu_local;
1173 #endif
1174
1175 #ifdef CONFIG_CPU_IDLE
1176 /* Must be inspected within a RCU lock section */
1177 struct cpuidle_state *idle_state;
1178 #endif
1179
1180 #ifdef CONFIG_SMP
1181 unsigned int nr_pinned;
1182 #endif
1183 unsigned int push_busy;
1184 struct cpu_stop_work push_work;
1185
1186 #ifdef CONFIG_SCHED_CORE
1187 /* per rq */
1188 struct rq *core;
1189 struct task_struct *core_pick;
1190 unsigned int core_enabled;
1191 unsigned int core_sched_seq;
1192 struct rb_root core_tree;
1193
1194 /* shared state -- careful with sched_core_cpu_deactivate() */
1195 unsigned int core_task_seq;
1196 unsigned int core_pick_seq;
1197 unsigned long core_cookie;
1198 unsigned int core_forceidle_count;
1199 unsigned int core_forceidle_seq;
1200 unsigned int core_forceidle_occupation;
1201 u64 core_forceidle_start;
1202 #endif
1203
1204 /* Scratch cpumask to be temporarily used under rq_lock */
1205 cpumask_var_t scratch_mask;
1206
1207 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1208 call_single_data_t cfsb_csd;
1209 struct list_head cfsb_csd_list;
1210 #endif
1211 };
1212
1213 #ifdef CONFIG_FAIR_GROUP_SCHED
1214
1215 /* CPU runqueue to which this cfs_rq is attached */
1216 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1217 {
1218 return cfs_rq->rq;
1219 }
1220
1221 #else
1222
1223 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1224 {
1225 return container_of(cfs_rq, struct rq, cfs);
1226 }
1227 #endif
1228
1229 static inline int cpu_of(struct rq *rq)
1230 {
1231 #ifdef CONFIG_SMP
1232 return rq->cpu;
1233 #else
1234 return 0;
1235 #endif
1236 }
1237
1238 #define MDF_PUSH 0x01
1239
1240 static inline bool is_migration_disabled(struct task_struct *p)
1241 {
1242 #ifdef CONFIG_SMP
1243 return p->migration_disabled;
1244 #else
1245 return false;
1246 #endif
1247 }
1248
1249 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1250
1251 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1252 #define this_rq() this_cpu_ptr(&runqueues)
1253 #define task_rq(p) cpu_rq(task_cpu(p))
1254 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1255 #define raw_rq() raw_cpu_ptr(&runqueues)
1256
1257 #ifdef CONFIG_SCHED_CORE
1258 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1259
1260 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1261
1262 static inline bool sched_core_enabled(struct rq *rq)
1263 {
1264 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1265 }
1266
1267 static inline bool sched_core_disabled(void)
1268 {
1269 return !static_branch_unlikely(&__sched_core_enabled);
1270 }
1271
1272 /*
1273 * Be careful with this function; not for general use. The return value isn't
1274 * stable unless you actually hold a relevant rq->__lock.
1275 */
1276 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1277 {
1278 if (sched_core_enabled(rq))
1279 return &rq->core->__lock;
1280
1281 return &rq->__lock;
1282 }
1283
1284 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1285 {
1286 if (rq->core_enabled)
1287 return &rq->core->__lock;
1288
1289 return &rq->__lock;
1290 }
1291
1292 extern bool
1293 cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1294
1295 extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1296
1297 /*
1298 * Helpers to check if the CPU's core cookie matches with the task's cookie
1299 * when core scheduling is enabled.
1300 * A special case is that the task's cookie always matches with CPU's core
1301 * cookie if the CPU is in an idle core.
1302 */
1303 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1304 {
1305 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1306 if (!sched_core_enabled(rq))
1307 return true;
1308
1309 return rq->core->core_cookie == p->core_cookie;
1310 }
1311
1312 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1313 {
1314 bool idle_core = true;
1315 int cpu;
1316
1317 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1318 if (!sched_core_enabled(rq))
1319 return true;
1320
1321 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1322 if (!available_idle_cpu(cpu)) {
1323 idle_core = false;
1324 break;
1325 }
1326 }
1327
1328 /*
1329 * A CPU in an idle core is always the best choice for tasks with
1330 * cookies.
1331 */
1332 return idle_core || rq->core->core_cookie == p->core_cookie;
1333 }
1334
1335 static inline bool sched_group_cookie_match(struct rq *rq,
1336 struct task_struct *p,
1337 struct sched_group *group)
1338 {
1339 int cpu;
1340
1341 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1342 if (!sched_core_enabled(rq))
1343 return true;
1344
1345 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1346 if (sched_core_cookie_match(cpu_rq(cpu), p))
1347 return true;
1348 }
1349 return false;
1350 }
1351
1352 static inline bool sched_core_enqueued(struct task_struct *p)
1353 {
1354 return !RB_EMPTY_NODE(&p->core_node);
1355 }
1356
1357 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1358 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1359
1360 extern void sched_core_get(void);
1361 extern void sched_core_put(void);
1362
1363 #else /* !CONFIG_SCHED_CORE: */
1364
1365 static inline bool sched_core_enabled(struct rq *rq)
1366 {
1367 return false;
1368 }
1369
1370 static inline bool sched_core_disabled(void)
1371 {
1372 return true;
1373 }
1374
1375 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1376 {
1377 return &rq->__lock;
1378 }
1379
1380 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1381 {
1382 return &rq->__lock;
1383 }
1384
1385 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1386 {
1387 return true;
1388 }
1389
1390 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1391 {
1392 return true;
1393 }
1394
1395 static inline bool sched_group_cookie_match(struct rq *rq,
1396 struct task_struct *p,
1397 struct sched_group *group)
1398 {
1399 return true;
1400 }
1401
1402 #endif /* !CONFIG_SCHED_CORE */
1403
1404 static inline void lockdep_assert_rq_held(struct rq *rq)
1405 {
1406 lockdep_assert_held(__rq_lockp(rq));
1407 }
1408
1409 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1410 extern bool raw_spin_rq_trylock(struct rq *rq);
1411 extern void raw_spin_rq_unlock(struct rq *rq);
1412
1413 static inline void raw_spin_rq_lock(struct rq *rq)
1414 {
1415 raw_spin_rq_lock_nested(rq, 0);
1416 }
1417
1418 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1419 {
1420 local_irq_disable();
1421 raw_spin_rq_lock(rq);
1422 }
1423
1424 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1425 {
1426 raw_spin_rq_unlock(rq);
1427 local_irq_enable();
1428 }
1429
1430 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1431 {
1432 unsigned long flags;
1433
1434 local_irq_save(flags);
1435 raw_spin_rq_lock(rq);
1436
1437 return flags;
1438 }
1439
1440 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1441 {
1442 raw_spin_rq_unlock(rq);
1443 local_irq_restore(flags);
1444 }
1445
1446 #define raw_spin_rq_lock_irqsave(rq, flags) \
1447 do { \
1448 flags = _raw_spin_rq_lock_irqsave(rq); \
1449 } while (0)
1450
1451 #ifdef CONFIG_SCHED_SMT
1452 extern void __update_idle_core(struct rq *rq);
1453
1454 static inline void update_idle_core(struct rq *rq)
1455 {
1456 if (static_branch_unlikely(&sched_smt_present))
1457 __update_idle_core(rq);
1458 }
1459
1460 #else
1461 static inline void update_idle_core(struct rq *rq) { }
1462 #endif
1463
1464 #ifdef CONFIG_FAIR_GROUP_SCHED
1465
1466 static inline struct task_struct *task_of(struct sched_entity *se)
1467 {
1468 SCHED_WARN_ON(!entity_is_task(se));
1469 return container_of(se, struct task_struct, se);
1470 }
1471
1472 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1473 {
1474 return p->se.cfs_rq;
1475 }
1476
1477 /* runqueue on which this entity is (to be) queued */
1478 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1479 {
1480 return se->cfs_rq;
1481 }
1482
1483 /* runqueue "owned" by this group */
1484 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1485 {
1486 return grp->my_q;
1487 }
1488
1489 #else /* !CONFIG_FAIR_GROUP_SCHED: */
1490
1491 #define task_of(_se) container_of(_se, struct task_struct, se)
1492
1493 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1494 {
1495 return &task_rq(p)->cfs;
1496 }
1497
1498 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1499 {
1500 const struct task_struct *p = task_of(se);
1501 struct rq *rq = task_rq(p);
1502
1503 return &rq->cfs;
1504 }
1505
1506 /* runqueue "owned" by this group */
1507 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1508 {
1509 return NULL;
1510 }
1511
1512 #endif /* !CONFIG_FAIR_GROUP_SCHED */
1513
1514 extern void update_rq_clock(struct rq *rq);
1515
1516 /*
1517 * rq::clock_update_flags bits
1518 *
1519 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1520 * call to __schedule(). This is an optimisation to avoid
1521 * neighbouring rq clock updates.
1522 *
1523 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1524 * in effect and calls to update_rq_clock() are being ignored.
1525 *
1526 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1527 * made to update_rq_clock() since the last time rq::lock was pinned.
1528 *
1529 * If inside of __schedule(), clock_update_flags will have been
1530 * shifted left (a left shift is a cheap operation for the fast path
1531 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1532 *
1533 * if (rq-clock_update_flags >= RQCF_UPDATED)
1534 *
1535 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1536 * one position though, because the next rq_unpin_lock() will shift it
1537 * back.
1538 */
1539 #define RQCF_REQ_SKIP 0x01
1540 #define RQCF_ACT_SKIP 0x02
1541 #define RQCF_UPDATED 0x04
1542
1543 static inline void assert_clock_updated(struct rq *rq)
1544 {
1545 /*
1546 * The only reason for not seeing a clock update since the
1547 * last rq_pin_lock() is if we're currently skipping updates.
1548 */
1549 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1550 }
1551
1552 static inline u64 rq_clock(struct rq *rq)
1553 {
1554 lockdep_assert_rq_held(rq);
1555 assert_clock_updated(rq);
1556
1557 return rq->clock;
1558 }
1559
1560 static inline u64 rq_clock_task(struct rq *rq)
1561 {
1562 lockdep_assert_rq_held(rq);
1563 assert_clock_updated(rq);
1564
1565 return rq->clock_task;
1566 }
1567
1568 static inline void rq_clock_skip_update(struct rq *rq)
1569 {
1570 lockdep_assert_rq_held(rq);
1571 rq->clock_update_flags |= RQCF_REQ_SKIP;
1572 }
1573
1574 /*
1575 * See rt task throttling, which is the only time a skip
1576 * request is canceled.
1577 */
1578 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1579 {
1580 lockdep_assert_rq_held(rq);
1581 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1582 }
1583
1584 /*
1585 * During cpu offlining and rq wide unthrottling, we can trigger
1586 * an update_rq_clock() for several cfs and rt runqueues (Typically
1587 * when using list_for_each_entry_*)
1588 * rq_clock_start_loop_update() can be called after updating the clock
1589 * once and before iterating over the list to prevent multiple update.
1590 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1591 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1592 */
1593 static inline void rq_clock_start_loop_update(struct rq *rq)
1594 {
1595 lockdep_assert_rq_held(rq);
1596 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1597 rq->clock_update_flags |= RQCF_ACT_SKIP;
1598 }
1599
1600 static inline void rq_clock_stop_loop_update(struct rq *rq)
1601 {
1602 lockdep_assert_rq_held(rq);
1603 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1604 }
1605
1606 struct rq_flags {
1607 unsigned long flags;
1608 struct pin_cookie cookie;
1609 #ifdef CONFIG_SCHED_DEBUG
1610 /*
1611 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1612 * current pin context is stashed here in case it needs to be
1613 * restored in rq_repin_lock().
1614 */
1615 unsigned int clock_update_flags;
1616 #endif
1617 };
1618
1619 extern struct balance_callback balance_push_callback;
1620
1621 /*
1622 * Lockdep annotation that avoids accidental unlocks; it's like a
1623 * sticky/continuous lockdep_assert_held().
1624 *
1625 * This avoids code that has access to 'struct rq *rq' (basically everything in
1626 * the scheduler) from accidentally unlocking the rq if they do not also have a
1627 * copy of the (on-stack) 'struct rq_flags rf'.
1628 *
1629 * Also see Documentation/locking/lockdep-design.rst.
1630 */
1631 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1632 {
1633 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1634
1635 #ifdef CONFIG_SCHED_DEBUG
1636 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1637 rf->clock_update_flags = 0;
1638 # ifdef CONFIG_SMP
1639 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1640 # endif
1641 #endif
1642 }
1643
1644 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1645 {
1646 #ifdef CONFIG_SCHED_DEBUG
1647 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1648 rf->clock_update_flags = RQCF_UPDATED;
1649 #endif
1650
1651 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1652 }
1653
1654 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1655 {
1656 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1657
1658 #ifdef CONFIG_SCHED_DEBUG
1659 /*
1660 * Restore the value we stashed in @rf for this pin context.
1661 */
1662 rq->clock_update_flags |= rf->clock_update_flags;
1663 #endif
1664 }
1665
1666 extern
1667 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1668 __acquires(rq->lock);
1669
1670 extern
1671 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1672 __acquires(p->pi_lock)
1673 __acquires(rq->lock);
1674
1675 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1676 __releases(rq->lock)
1677 {
1678 rq_unpin_lock(rq, rf);
1679 raw_spin_rq_unlock(rq);
1680 }
1681
1682 static inline void
1683 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1684 __releases(rq->lock)
1685 __releases(p->pi_lock)
1686 {
1687 rq_unpin_lock(rq, rf);
1688 raw_spin_rq_unlock(rq);
1689 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1690 }
1691
1692 DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1693 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1694 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1695 struct rq *rq; struct rq_flags rf)
1696
1697 static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1698 __acquires(rq->lock)
1699 {
1700 raw_spin_rq_lock_irqsave(rq, rf->flags);
1701 rq_pin_lock(rq, rf);
1702 }
1703
1704 static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1705 __acquires(rq->lock)
1706 {
1707 raw_spin_rq_lock_irq(rq);
1708 rq_pin_lock(rq, rf);
1709 }
1710
1711 static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1712 __acquires(rq->lock)
1713 {
1714 raw_spin_rq_lock(rq);
1715 rq_pin_lock(rq, rf);
1716 }
1717
1718 static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1719 __releases(rq->lock)
1720 {
1721 rq_unpin_lock(rq, rf);
1722 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1723 }
1724
1725 static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1726 __releases(rq->lock)
1727 {
1728 rq_unpin_lock(rq, rf);
1729 raw_spin_rq_unlock_irq(rq);
1730 }
1731
1732 static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1733 __releases(rq->lock)
1734 {
1735 rq_unpin_lock(rq, rf);
1736 raw_spin_rq_unlock(rq);
1737 }
1738
1739 DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1740 rq_lock(_T->lock, &_T->rf),
1741 rq_unlock(_T->lock, &_T->rf),
1742 struct rq_flags rf)
1743
1744 DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1745 rq_lock_irq(_T->lock, &_T->rf),
1746 rq_unlock_irq(_T->lock, &_T->rf),
1747 struct rq_flags rf)
1748
1749 DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1750 rq_lock_irqsave(_T->lock, &_T->rf),
1751 rq_unlock_irqrestore(_T->lock, &_T->rf),
1752 struct rq_flags rf)
1753
1754 static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1755 __acquires(rq->lock)
1756 {
1757 struct rq *rq;
1758
1759 local_irq_disable();
1760 rq = this_rq();
1761 rq_lock(rq, rf);
1762
1763 return rq;
1764 }
1765
1766 #ifdef CONFIG_NUMA
1767
1768 enum numa_topology_type {
1769 NUMA_DIRECT,
1770 NUMA_GLUELESS_MESH,
1771 NUMA_BACKPLANE,
1772 };
1773
1774 extern enum numa_topology_type sched_numa_topology_type;
1775 extern int sched_max_numa_distance;
1776 extern bool find_numa_distance(int distance);
1777 extern void sched_init_numa(int offline_node);
1778 extern void sched_update_numa(int cpu, bool online);
1779 extern void sched_domains_numa_masks_set(unsigned int cpu);
1780 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1781 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1782
1783 #else /* !CONFIG_NUMA: */
1784
1785 static inline void sched_init_numa(int offline_node) { }
1786 static inline void sched_update_numa(int cpu, bool online) { }
1787 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1788 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1789
1790 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1791 {
1792 return nr_cpu_ids;
1793 }
1794
1795 #endif /* !CONFIG_NUMA */
1796
1797 #ifdef CONFIG_NUMA_BALANCING
1798
1799 /* The regions in numa_faults array from task_struct */
1800 enum numa_faults_stats {
1801 NUMA_MEM = 0,
1802 NUMA_CPU,
1803 NUMA_MEMBUF,
1804 NUMA_CPUBUF
1805 };
1806
1807 extern void sched_setnuma(struct task_struct *p, int node);
1808 extern int migrate_task_to(struct task_struct *p, int cpu);
1809 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1810 int cpu, int scpu);
1811 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1812
1813 #else /* !CONFIG_NUMA_BALANCING: */
1814
1815 static inline void
1816 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1817 {
1818 }
1819
1820 #endif /* !CONFIG_NUMA_BALANCING */
1821
1822 #ifdef CONFIG_SMP
1823
1824 static inline void
1825 queue_balance_callback(struct rq *rq,
1826 struct balance_callback *head,
1827 void (*func)(struct rq *rq))
1828 {
1829 lockdep_assert_rq_held(rq);
1830
1831 /*
1832 * Don't (re)queue an already queued item; nor queue anything when
1833 * balance_push() is active, see the comment with
1834 * balance_push_callback.
1835 */
1836 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1837 return;
1838
1839 head->func = func;
1840 head->next = rq->balance_callback;
1841 rq->balance_callback = head;
1842 }
1843
1844 #define rcu_dereference_check_sched_domain(p) \
1845 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1846
1847 /*
1848 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1849 * See destroy_sched_domains: call_rcu for details.
1850 *
1851 * The domain tree of any CPU may only be accessed from within
1852 * preempt-disabled sections.
1853 */
1854 #define for_each_domain(cpu, __sd) \
1855 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1856 __sd; __sd = __sd->parent)
1857
1858 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1859 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1860 static const unsigned int SD_SHARED_CHILD_MASK =
1861 #include <linux/sched/sd_flags.h>
1862 0;
1863 #undef SD_FLAG
1864
1865 /**
1866 * highest_flag_domain - Return highest sched_domain containing flag.
1867 * @cpu: The CPU whose highest level of sched domain is to
1868 * be returned.
1869 * @flag: The flag to check for the highest sched_domain
1870 * for the given CPU.
1871 *
1872 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1873 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1874 */
1875 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1876 {
1877 struct sched_domain *sd, *hsd = NULL;
1878
1879 for_each_domain(cpu, sd) {
1880 if (sd->flags & flag) {
1881 hsd = sd;
1882 continue;
1883 }
1884
1885 /*
1886 * Stop the search if @flag is known to be shared at lower
1887 * levels. It will not be found further up.
1888 */
1889 if (flag & SD_SHARED_CHILD_MASK)
1890 break;
1891 }
1892
1893 return hsd;
1894 }
1895
1896 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1897 {
1898 struct sched_domain *sd;
1899
1900 for_each_domain(cpu, sd) {
1901 if (sd->flags & flag)
1902 break;
1903 }
1904
1905 return sd;
1906 }
1907
1908 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1909 DECLARE_PER_CPU(int, sd_llc_size);
1910 DECLARE_PER_CPU(int, sd_llc_id);
1911 DECLARE_PER_CPU(int, sd_share_id);
1912 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1913 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1914 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1915 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1916
1917 extern struct static_key_false sched_asym_cpucapacity;
1918 extern struct static_key_false sched_cluster_active;
1919
1920 static __always_inline bool sched_asym_cpucap_active(void)
1921 {
1922 return static_branch_unlikely(&sched_asym_cpucapacity);
1923 }
1924
1925 struct sched_group_capacity {
1926 atomic_t ref;
1927 /*
1928 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1929 * for a single CPU.
1930 */
1931 unsigned long capacity;
1932 unsigned long min_capacity; /* Min per-CPU capacity in group */
1933 unsigned long max_capacity; /* Max per-CPU capacity in group */
1934 unsigned long next_update;
1935 int imbalance; /* XXX unrelated to capacity but shared group state */
1936
1937 #ifdef CONFIG_SCHED_DEBUG
1938 int id;
1939 #endif
1940
1941 unsigned long cpumask[]; /* Balance mask */
1942 };
1943
1944 struct sched_group {
1945 struct sched_group *next; /* Must be a circular list */
1946 atomic_t ref;
1947
1948 unsigned int group_weight;
1949 unsigned int cores;
1950 struct sched_group_capacity *sgc;
1951 int asym_prefer_cpu; /* CPU of highest priority in group */
1952 int flags;
1953
1954 /*
1955 * The CPUs this group covers.
1956 *
1957 * NOTE: this field is variable length. (Allocated dynamically
1958 * by attaching extra space to the end of the structure,
1959 * depending on how many CPUs the kernel has booted up with)
1960 */
1961 unsigned long cpumask[];
1962 };
1963
1964 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1965 {
1966 return to_cpumask(sg->cpumask);
1967 }
1968
1969 /*
1970 * See build_balance_mask().
1971 */
1972 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1973 {
1974 return to_cpumask(sg->sgc->cpumask);
1975 }
1976
1977 extern int group_balance_cpu(struct sched_group *sg);
1978
1979 #ifdef CONFIG_SCHED_DEBUG
1980 extern void update_sched_domain_debugfs(void);
1981 extern void dirty_sched_domain_sysctl(int cpu);
1982 #else
1983 static inline void update_sched_domain_debugfs(void) { }
1984 static inline void dirty_sched_domain_sysctl(int cpu) { }
1985 #endif
1986
1987 extern int sched_update_scaling(void);
1988
1989 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1990 {
1991 if (!p->user_cpus_ptr)
1992 return cpu_possible_mask; /* &init_task.cpus_mask */
1993 return p->user_cpus_ptr;
1994 }
1995
1996 #endif /* CONFIG_SMP */
1997
1998 #include "stats.h"
1999
2000 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
2001
2002 extern void __sched_core_account_forceidle(struct rq *rq);
2003
2004 static inline void sched_core_account_forceidle(struct rq *rq)
2005 {
2006 if (schedstat_enabled())
2007 __sched_core_account_forceidle(rq);
2008 }
2009
2010 extern void __sched_core_tick(struct rq *rq);
2011
2012 static inline void sched_core_tick(struct rq *rq)
2013 {
2014 if (sched_core_enabled(rq) && schedstat_enabled())
2015 __sched_core_tick(rq);
2016 }
2017
2018 #else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
2019
2020 static inline void sched_core_account_forceidle(struct rq *rq) { }
2021
2022 static inline void sched_core_tick(struct rq *rq) { }
2023
2024 #endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
2025
2026 #ifdef CONFIG_CGROUP_SCHED
2027
2028 /*
2029 * Return the group to which this tasks belongs.
2030 *
2031 * We cannot use task_css() and friends because the cgroup subsystem
2032 * changes that value before the cgroup_subsys::attach() method is called,
2033 * therefore we cannot pin it and might observe the wrong value.
2034 *
2035 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2036 * core changes this before calling sched_move_task().
2037 *
2038 * Instead we use a 'copy' which is updated from sched_move_task() while
2039 * holding both task_struct::pi_lock and rq::lock.
2040 */
2041 static inline struct task_group *task_group(struct task_struct *p)
2042 {
2043 return p->sched_task_group;
2044 }
2045
2046 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2047 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2048 {
2049 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2050 struct task_group *tg = task_group(p);
2051 #endif
2052
2053 #ifdef CONFIG_FAIR_GROUP_SCHED
2054 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2055 p->se.cfs_rq = tg->cfs_rq[cpu];
2056 p->se.parent = tg->se[cpu];
2057 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2058 #endif
2059
2060 #ifdef CONFIG_RT_GROUP_SCHED
2061 p->rt.rt_rq = tg->rt_rq[cpu];
2062 p->rt.parent = tg->rt_se[cpu];
2063 #endif
2064 }
2065
2066 #else /* !CONFIG_CGROUP_SCHED: */
2067
2068 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2069
2070 static inline struct task_group *task_group(struct task_struct *p)
2071 {
2072 return NULL;
2073 }
2074
2075 #endif /* !CONFIG_CGROUP_SCHED */
2076
2077 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2078 {
2079 set_task_rq(p, cpu);
2080 #ifdef CONFIG_SMP
2081 /*
2082 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2083 * successfully executed on another CPU. We must ensure that updates of
2084 * per-task data have been completed by this moment.
2085 */
2086 smp_wmb();
2087 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2088 p->wake_cpu = cpu;
2089 #endif
2090 }
2091
2092 /*
2093 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2094 */
2095 #ifdef CONFIG_SCHED_DEBUG
2096 # define const_debug __read_mostly
2097 #else
2098 # define const_debug const
2099 #endif
2100
2101 #define SCHED_FEAT(name, enabled) \
2102 __SCHED_FEAT_##name ,
2103
2104 enum {
2105 #include "features.h"
2106 __SCHED_FEAT_NR,
2107 };
2108
2109 #undef SCHED_FEAT
2110
2111 #ifdef CONFIG_SCHED_DEBUG
2112
2113 /*
2114 * To support run-time toggling of sched features, all the translation units
2115 * (but core.c) reference the sysctl_sched_features defined in core.c.
2116 */
2117 extern const_debug unsigned int sysctl_sched_features;
2118
2119 #ifdef CONFIG_JUMP_LABEL
2120
2121 #define SCHED_FEAT(name, enabled) \
2122 static __always_inline bool static_branch_##name(struct static_key *key) \
2123 { \
2124 return static_key_##enabled(key); \
2125 }
2126
2127 #include "features.h"
2128 #undef SCHED_FEAT
2129
2130 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2131 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2132
2133 #else /* !CONFIG_JUMP_LABEL: */
2134
2135 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2136
2137 #endif /* !CONFIG_JUMP_LABEL */
2138
2139 #else /* !SCHED_DEBUG: */
2140
2141 /*
2142 * Each translation unit has its own copy of sysctl_sched_features to allow
2143 * constants propagation at compile time and compiler optimization based on
2144 * features default.
2145 */
2146 #define SCHED_FEAT(name, enabled) \
2147 (1UL << __SCHED_FEAT_##name) * enabled |
2148 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2149 #include "features.h"
2150 0;
2151 #undef SCHED_FEAT
2152
2153 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2154
2155 #endif /* !SCHED_DEBUG */
2156
2157 extern struct static_key_false sched_numa_balancing;
2158 extern struct static_key_false sched_schedstats;
2159
2160 static inline u64 global_rt_period(void)
2161 {
2162 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2163 }
2164
2165 static inline u64 global_rt_runtime(void)
2166 {
2167 if (sysctl_sched_rt_runtime < 0)
2168 return RUNTIME_INF;
2169
2170 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2171 }
2172
2173 static inline int task_current(struct rq *rq, struct task_struct *p)
2174 {
2175 return rq->curr == p;
2176 }
2177
2178 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2179 {
2180 #ifdef CONFIG_SMP
2181 return p->on_cpu;
2182 #else
2183 return task_current(rq, p);
2184 #endif
2185 }
2186
2187 static inline int task_on_rq_queued(struct task_struct *p)
2188 {
2189 return p->on_rq == TASK_ON_RQ_QUEUED;
2190 }
2191
2192 static inline int task_on_rq_migrating(struct task_struct *p)
2193 {
2194 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2195 }
2196
2197 /* Wake flags. The first three directly map to some SD flag value */
2198 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2199 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2200 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2201
2202 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2203 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2204 #define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2205
2206 #ifdef CONFIG_SMP
2207 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2208 static_assert(WF_FORK == SD_BALANCE_FORK);
2209 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2210 #endif
2211
2212 /*
2213 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2214 * of tasks with abnormal "nice" values across CPUs the contribution that
2215 * each task makes to its run queue's load is weighted according to its
2216 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2217 * scaled version of the new time slice allocation that they receive on time
2218 * slice expiry etc.
2219 */
2220
2221 #define WEIGHT_IDLEPRIO 3
2222 #define WMULT_IDLEPRIO 1431655765
2223
2224 extern const int sched_prio_to_weight[40];
2225 extern const u32 sched_prio_to_wmult[40];
2226
2227 /*
2228 * {de,en}queue flags:
2229 *
2230 * DEQUEUE_SLEEP - task is no longer runnable
2231 * ENQUEUE_WAKEUP - task just became runnable
2232 *
2233 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2234 * are in a known state which allows modification. Such pairs
2235 * should preserve as much state as possible.
2236 *
2237 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2238 * in the runqueue.
2239 *
2240 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2241 *
2242 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2243 *
2244 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2245 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2246 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2247 *
2248 */
2249
2250 #define DEQUEUE_SLEEP 0x01
2251 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2252 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2253 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2254 #define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2255
2256 #define ENQUEUE_WAKEUP 0x01
2257 #define ENQUEUE_RESTORE 0x02
2258 #define ENQUEUE_MOVE 0x04
2259 #define ENQUEUE_NOCLOCK 0x08
2260
2261 #define ENQUEUE_HEAD 0x10
2262 #define ENQUEUE_REPLENISH 0x20
2263 #ifdef CONFIG_SMP
2264 #define ENQUEUE_MIGRATED 0x40
2265 #else
2266 #define ENQUEUE_MIGRATED 0x00
2267 #endif
2268 #define ENQUEUE_INITIAL 0x80
2269 #define ENQUEUE_MIGRATING 0x100
2270
2271 #define RETRY_TASK ((void *)-1UL)
2272
2273 struct affinity_context {
2274 const struct cpumask *new_mask;
2275 struct cpumask *user_mask;
2276 unsigned int flags;
2277 };
2278
2279 extern s64 update_curr_common(struct rq *rq);
2280
2281 struct sched_class {
2282
2283 #ifdef CONFIG_UCLAMP_TASK
2284 int uclamp_enabled;
2285 #endif
2286
2287 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2288 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2289 void (*yield_task) (struct rq *rq);
2290 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2291
2292 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2293
2294 struct task_struct *(*pick_next_task)(struct rq *rq);
2295
2296 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2297 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2298
2299 #ifdef CONFIG_SMP
2300 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2301 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2302
2303 struct task_struct * (*pick_task)(struct rq *rq);
2304
2305 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2306
2307 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2308
2309 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2310
2311 void (*rq_online)(struct rq *rq);
2312 void (*rq_offline)(struct rq *rq);
2313
2314 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2315 #endif
2316
2317 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2318 void (*task_fork)(struct task_struct *p);
2319 void (*task_dead)(struct task_struct *p);
2320
2321 /*
2322 * The switched_from() call is allowed to drop rq->lock, therefore we
2323 * cannot assume the switched_from/switched_to pair is serialized by
2324 * rq->lock. They are however serialized by p->pi_lock.
2325 */
2326 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2327 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2328 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2329 int oldprio);
2330
2331 unsigned int (*get_rr_interval)(struct rq *rq,
2332 struct task_struct *task);
2333
2334 void (*update_curr)(struct rq *rq);
2335
2336 #ifdef CONFIG_FAIR_GROUP_SCHED
2337 void (*task_change_group)(struct task_struct *p);
2338 #endif
2339
2340 #ifdef CONFIG_SCHED_CORE
2341 int (*task_is_throttled)(struct task_struct *p, int cpu);
2342 #endif
2343 };
2344
2345 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2346 {
2347 WARN_ON_ONCE(rq->curr != prev);
2348 prev->sched_class->put_prev_task(rq, prev);
2349 }
2350
2351 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2352 {
2353 next->sched_class->set_next_task(rq, next, false);
2354 }
2355
2356
2357 /*
2358 * Helper to define a sched_class instance; each one is placed in a separate
2359 * section which is ordered by the linker script:
2360 *
2361 * include/asm-generic/vmlinux.lds.h
2362 *
2363 * *CAREFUL* they are laid out in *REVERSE* order!!!
2364 *
2365 * Also enforce alignment on the instance, not the type, to guarantee layout.
2366 */
2367 #define DEFINE_SCHED_CLASS(name) \
2368 const struct sched_class name##_sched_class \
2369 __aligned(__alignof__(struct sched_class)) \
2370 __section("__" #name "_sched_class")
2371
2372 /* Defined in include/asm-generic/vmlinux.lds.h */
2373 extern struct sched_class __sched_class_highest[];
2374 extern struct sched_class __sched_class_lowest[];
2375
2376 #define for_class_range(class, _from, _to) \
2377 for (class = (_from); class < (_to); class++)
2378
2379 #define for_each_class(class) \
2380 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2381
2382 #define sched_class_above(_a, _b) ((_a) < (_b))
2383
2384 extern const struct sched_class stop_sched_class;
2385 extern const struct sched_class dl_sched_class;
2386 extern const struct sched_class rt_sched_class;
2387 extern const struct sched_class fair_sched_class;
2388 extern const struct sched_class idle_sched_class;
2389
2390 static inline bool sched_stop_runnable(struct rq *rq)
2391 {
2392 return rq->stop && task_on_rq_queued(rq->stop);
2393 }
2394
2395 static inline bool sched_dl_runnable(struct rq *rq)
2396 {
2397 return rq->dl.dl_nr_running > 0;
2398 }
2399
2400 static inline bool sched_rt_runnable(struct rq *rq)
2401 {
2402 return rq->rt.rt_queued > 0;
2403 }
2404
2405 static inline bool sched_fair_runnable(struct rq *rq)
2406 {
2407 return rq->cfs.nr_running > 0;
2408 }
2409
2410 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2411 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2412
2413 #define SCA_CHECK 0x01
2414 #define SCA_MIGRATE_DISABLE 0x02
2415 #define SCA_MIGRATE_ENABLE 0x04
2416 #define SCA_USER 0x08
2417
2418 #ifdef CONFIG_SMP
2419
2420 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2421
2422 extern void sched_balance_trigger(struct rq *rq);
2423
2424 extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2425 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2426
2427 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2428 {
2429 /*
2430 * See do_set_cpus_allowed() above for the rcu_head usage.
2431 */
2432 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2433
2434 return kmalloc_node(size, GFP_KERNEL, node);
2435 }
2436
2437 static inline struct task_struct *get_push_task(struct rq *rq)
2438 {
2439 struct task_struct *p = rq->curr;
2440
2441 lockdep_assert_rq_held(rq);
2442
2443 if (rq->push_busy)
2444 return NULL;
2445
2446 if (p->nr_cpus_allowed == 1)
2447 return NULL;
2448
2449 if (p->migration_disabled)
2450 return NULL;
2451
2452 rq->push_busy = true;
2453 return get_task_struct(p);
2454 }
2455
2456 extern int push_cpu_stop(void *arg);
2457
2458 #else /* !CONFIG_SMP: */
2459
2460 static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2461 struct affinity_context *ctx)
2462 {
2463 return set_cpus_allowed_ptr(p, ctx->new_mask);
2464 }
2465
2466 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2467 {
2468 return NULL;
2469 }
2470
2471 #endif /* !CONFIG_SMP */
2472
2473 #ifdef CONFIG_CPU_IDLE
2474
2475 static inline void idle_set_state(struct rq *rq,
2476 struct cpuidle_state *idle_state)
2477 {
2478 rq->idle_state = idle_state;
2479 }
2480
2481 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2482 {
2483 SCHED_WARN_ON(!rcu_read_lock_held());
2484
2485 return rq->idle_state;
2486 }
2487
2488 #else /* !CONFIG_CPU_IDLE: */
2489
2490 static inline void idle_set_state(struct rq *rq,
2491 struct cpuidle_state *idle_state)
2492 {
2493 }
2494
2495 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2496 {
2497 return NULL;
2498 }
2499
2500 #endif /* !CONFIG_CPU_IDLE */
2501
2502 extern void schedule_idle(void);
2503 asmlinkage void schedule_user(void);
2504
2505 extern void sysrq_sched_debug_show(void);
2506 extern void sched_init_granularity(void);
2507 extern void update_max_interval(void);
2508
2509 extern void init_sched_dl_class(void);
2510 extern void init_sched_rt_class(void);
2511 extern void init_sched_fair_class(void);
2512
2513 extern void reweight_task(struct task_struct *p, const struct load_weight *lw);
2514
2515 extern void resched_curr(struct rq *rq);
2516 extern void resched_cpu(int cpu);
2517
2518 extern struct rt_bandwidth def_rt_bandwidth;
2519 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2520 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2521
2522 extern void init_dl_entity(struct sched_dl_entity *dl_se);
2523
2524 #define BW_SHIFT 20
2525 #define BW_UNIT (1 << BW_SHIFT)
2526 #define RATIO_SHIFT 8
2527 #define MAX_BW_BITS (64 - BW_SHIFT)
2528 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2529
2530 extern unsigned long to_ratio(u64 period, u64 runtime);
2531
2532 extern void init_entity_runnable_average(struct sched_entity *se);
2533 extern void post_init_entity_util_avg(struct task_struct *p);
2534
2535 #ifdef CONFIG_NO_HZ_FULL
2536 extern bool sched_can_stop_tick(struct rq *rq);
2537 extern int __init sched_tick_offload_init(void);
2538
2539 /*
2540 * Tick may be needed by tasks in the runqueue depending on their policy and
2541 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2542 * nohz mode if necessary.
2543 */
2544 static inline void sched_update_tick_dependency(struct rq *rq)
2545 {
2546 int cpu = cpu_of(rq);
2547
2548 if (!tick_nohz_full_cpu(cpu))
2549 return;
2550
2551 if (sched_can_stop_tick(rq))
2552 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2553 else
2554 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2555 }
2556 #else /* !CONFIG_NO_HZ_FULL: */
2557 static inline int sched_tick_offload_init(void) { return 0; }
2558 static inline void sched_update_tick_dependency(struct rq *rq) { }
2559 #endif /* !CONFIG_NO_HZ_FULL */
2560
2561 static inline void add_nr_running(struct rq *rq, unsigned count)
2562 {
2563 unsigned prev_nr = rq->nr_running;
2564
2565 rq->nr_running = prev_nr + count;
2566 if (trace_sched_update_nr_running_tp_enabled()) {
2567 call_trace_sched_update_nr_running(rq, count);
2568 }
2569
2570 #ifdef CONFIG_SMP
2571 if (prev_nr < 2 && rq->nr_running >= 2)
2572 set_rd_overloaded(rq->rd, 1);
2573 #endif
2574
2575 sched_update_tick_dependency(rq);
2576 }
2577
2578 static inline void sub_nr_running(struct rq *rq, unsigned count)
2579 {
2580 rq->nr_running -= count;
2581 if (trace_sched_update_nr_running_tp_enabled()) {
2582 call_trace_sched_update_nr_running(rq, -count);
2583 }
2584
2585 /* Check if we still need preemption */
2586 sched_update_tick_dependency(rq);
2587 }
2588
2589 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2590 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2591
2592 extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2593
2594 #ifdef CONFIG_PREEMPT_RT
2595 # define SCHED_NR_MIGRATE_BREAK 8
2596 #else
2597 # define SCHED_NR_MIGRATE_BREAK 32
2598 #endif
2599
2600 extern const_debug unsigned int sysctl_sched_nr_migrate;
2601 extern const_debug unsigned int sysctl_sched_migration_cost;
2602
2603 extern unsigned int sysctl_sched_base_slice;
2604
2605 #ifdef CONFIG_SCHED_DEBUG
2606 extern int sysctl_resched_latency_warn_ms;
2607 extern int sysctl_resched_latency_warn_once;
2608
2609 extern unsigned int sysctl_sched_tunable_scaling;
2610
2611 extern unsigned int sysctl_numa_balancing_scan_delay;
2612 extern unsigned int sysctl_numa_balancing_scan_period_min;
2613 extern unsigned int sysctl_numa_balancing_scan_period_max;
2614 extern unsigned int sysctl_numa_balancing_scan_size;
2615 extern unsigned int sysctl_numa_balancing_hot_threshold;
2616 #endif
2617
2618 #ifdef CONFIG_SCHED_HRTICK
2619
2620 /*
2621 * Use hrtick when:
2622 * - enabled by features
2623 * - hrtimer is actually high res
2624 */
2625 static inline int hrtick_enabled(struct rq *rq)
2626 {
2627 if (!cpu_active(cpu_of(rq)))
2628 return 0;
2629 return hrtimer_is_hres_active(&rq->hrtick_timer);
2630 }
2631
2632 static inline int hrtick_enabled_fair(struct rq *rq)
2633 {
2634 if (!sched_feat(HRTICK))
2635 return 0;
2636 return hrtick_enabled(rq);
2637 }
2638
2639 static inline int hrtick_enabled_dl(struct rq *rq)
2640 {
2641 if (!sched_feat(HRTICK_DL))
2642 return 0;
2643 return hrtick_enabled(rq);
2644 }
2645
2646 extern void hrtick_start(struct rq *rq, u64 delay);
2647
2648 #else /* !CONFIG_SCHED_HRTICK: */
2649
2650 static inline int hrtick_enabled_fair(struct rq *rq)
2651 {
2652 return 0;
2653 }
2654
2655 static inline int hrtick_enabled_dl(struct rq *rq)
2656 {
2657 return 0;
2658 }
2659
2660 static inline int hrtick_enabled(struct rq *rq)
2661 {
2662 return 0;
2663 }
2664
2665 #endif /* !CONFIG_SCHED_HRTICK */
2666
2667 #ifndef arch_scale_freq_tick
2668 static __always_inline void arch_scale_freq_tick(void) { }
2669 #endif
2670
2671 #ifndef arch_scale_freq_capacity
2672 /**
2673 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2674 * @cpu: the CPU in question.
2675 *
2676 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2677 *
2678 * f_curr
2679 * ------ * SCHED_CAPACITY_SCALE
2680 * f_max
2681 */
2682 static __always_inline
2683 unsigned long arch_scale_freq_capacity(int cpu)
2684 {
2685 return SCHED_CAPACITY_SCALE;
2686 }
2687 #endif
2688
2689 #ifdef CONFIG_SCHED_DEBUG
2690 /*
2691 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2692 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2693 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2694 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2695 */
2696 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2697 {
2698 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2699 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2700 #ifdef CONFIG_SMP
2701 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2702 #endif
2703 }
2704 #else
2705 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) { }
2706 #endif
2707
2708 #define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2709 __DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2710 static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2711 { class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2712 _lock; return _t; }
2713
2714 #ifdef CONFIG_SMP
2715
2716 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2717 {
2718 #ifdef CONFIG_SCHED_CORE
2719 /*
2720 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2721 * order by core-id first and cpu-id second.
2722 *
2723 * Notably:
2724 *
2725 * double_rq_lock(0,3); will take core-0, core-1 lock
2726 * double_rq_lock(1,2); will take core-1, core-0 lock
2727 *
2728 * when only cpu-id is considered.
2729 */
2730 if (rq1->core->cpu < rq2->core->cpu)
2731 return true;
2732 if (rq1->core->cpu > rq2->core->cpu)
2733 return false;
2734
2735 /*
2736 * __sched_core_flip() relies on SMT having cpu-id lock order.
2737 */
2738 #endif
2739 return rq1->cpu < rq2->cpu;
2740 }
2741
2742 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2743
2744 #ifdef CONFIG_PREEMPTION
2745
2746 /*
2747 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2748 * way at the expense of forcing extra atomic operations in all
2749 * invocations. This assures that the double_lock is acquired using the
2750 * same underlying policy as the spinlock_t on this architecture, which
2751 * reduces latency compared to the unfair variant below. However, it
2752 * also adds more overhead and therefore may reduce throughput.
2753 */
2754 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2755 __releases(this_rq->lock)
2756 __acquires(busiest->lock)
2757 __acquires(this_rq->lock)
2758 {
2759 raw_spin_rq_unlock(this_rq);
2760 double_rq_lock(this_rq, busiest);
2761
2762 return 1;
2763 }
2764
2765 #else /* !CONFIG_PREEMPTION: */
2766 /*
2767 * Unfair double_lock_balance: Optimizes throughput at the expense of
2768 * latency by eliminating extra atomic operations when the locks are
2769 * already in proper order on entry. This favors lower CPU-ids and will
2770 * grant the double lock to lower CPUs over higher ids under contention,
2771 * regardless of entry order into the function.
2772 */
2773 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2774 __releases(this_rq->lock)
2775 __acquires(busiest->lock)
2776 __acquires(this_rq->lock)
2777 {
2778 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2779 likely(raw_spin_rq_trylock(busiest))) {
2780 double_rq_clock_clear_update(this_rq, busiest);
2781 return 0;
2782 }
2783
2784 if (rq_order_less(this_rq, busiest)) {
2785 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2786 double_rq_clock_clear_update(this_rq, busiest);
2787 return 0;
2788 }
2789
2790 raw_spin_rq_unlock(this_rq);
2791 double_rq_lock(this_rq, busiest);
2792
2793 return 1;
2794 }
2795
2796 #endif /* !CONFIG_PREEMPTION */
2797
2798 /*
2799 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2800 */
2801 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2802 {
2803 lockdep_assert_irqs_disabled();
2804
2805 return _double_lock_balance(this_rq, busiest);
2806 }
2807
2808 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2809 __releases(busiest->lock)
2810 {
2811 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2812 raw_spin_rq_unlock(busiest);
2813 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2814 }
2815
2816 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2817 {
2818 if (l1 > l2)
2819 swap(l1, l2);
2820
2821 spin_lock(l1);
2822 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2823 }
2824
2825 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2826 {
2827 if (l1 > l2)
2828 swap(l1, l2);
2829
2830 spin_lock_irq(l1);
2831 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2832 }
2833
2834 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2835 {
2836 if (l1 > l2)
2837 swap(l1, l2);
2838
2839 raw_spin_lock(l1);
2840 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2841 }
2842
2843 static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2844 {
2845 raw_spin_unlock(l1);
2846 raw_spin_unlock(l2);
2847 }
2848
2849 DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
2850 double_raw_lock(_T->lock, _T->lock2),
2851 double_raw_unlock(_T->lock, _T->lock2))
2852
2853 /*
2854 * double_rq_unlock - safely unlock two runqueues
2855 *
2856 * Note this does not restore interrupts like task_rq_unlock,
2857 * you need to do so manually after calling.
2858 */
2859 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2860 __releases(rq1->lock)
2861 __releases(rq2->lock)
2862 {
2863 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2864 raw_spin_rq_unlock(rq2);
2865 else
2866 __release(rq2->lock);
2867 raw_spin_rq_unlock(rq1);
2868 }
2869
2870 extern void set_rq_online (struct rq *rq);
2871 extern void set_rq_offline(struct rq *rq);
2872
2873 extern bool sched_smp_initialized;
2874
2875 #else /* !CONFIG_SMP: */
2876
2877 /*
2878 * double_rq_lock - safely lock two runqueues
2879 *
2880 * Note this does not disable interrupts like task_rq_lock,
2881 * you need to do so manually before calling.
2882 */
2883 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2884 __acquires(rq1->lock)
2885 __acquires(rq2->lock)
2886 {
2887 WARN_ON_ONCE(!irqs_disabled());
2888 WARN_ON_ONCE(rq1 != rq2);
2889 raw_spin_rq_lock(rq1);
2890 __acquire(rq2->lock); /* Fake it out ;) */
2891 double_rq_clock_clear_update(rq1, rq2);
2892 }
2893
2894 /*
2895 * double_rq_unlock - safely unlock two runqueues
2896 *
2897 * Note this does not restore interrupts like task_rq_unlock,
2898 * you need to do so manually after calling.
2899 */
2900 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2901 __releases(rq1->lock)
2902 __releases(rq2->lock)
2903 {
2904 WARN_ON_ONCE(rq1 != rq2);
2905 raw_spin_rq_unlock(rq1);
2906 __release(rq2->lock);
2907 }
2908
2909 #endif /* !CONFIG_SMP */
2910
2911 DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
2912 double_rq_lock(_T->lock, _T->lock2),
2913 double_rq_unlock(_T->lock, _T->lock2))
2914
2915 extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
2916 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2917 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2918
2919 #ifdef CONFIG_SCHED_DEBUG
2920 extern bool sched_debug_verbose;
2921
2922 extern void print_cfs_stats(struct seq_file *m, int cpu);
2923 extern void print_rt_stats(struct seq_file *m, int cpu);
2924 extern void print_dl_stats(struct seq_file *m, int cpu);
2925 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2926 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2927 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2928
2929 extern void resched_latency_warn(int cpu, u64 latency);
2930 # ifdef CONFIG_NUMA_BALANCING
2931 extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
2932 extern void
2933 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2934 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2935 # endif /* CONFIG_NUMA_BALANCING */
2936 #else /* !CONFIG_SCHED_DEBUG: */
2937 static inline void resched_latency_warn(int cpu, u64 latency) { }
2938 #endif /* !CONFIG_SCHED_DEBUG */
2939
2940 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2941 extern void init_rt_rq(struct rt_rq *rt_rq);
2942 extern void init_dl_rq(struct dl_rq *dl_rq);
2943
2944 extern void cfs_bandwidth_usage_inc(void);
2945 extern void cfs_bandwidth_usage_dec(void);
2946
2947 #ifdef CONFIG_NO_HZ_COMMON
2948
2949 #define NOHZ_BALANCE_KICK_BIT 0
2950 #define NOHZ_STATS_KICK_BIT 1
2951 #define NOHZ_NEWILB_KICK_BIT 2
2952 #define NOHZ_NEXT_KICK_BIT 3
2953
2954 /* Run sched_balance_domains() */
2955 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2956 /* Update blocked load */
2957 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2958 /* Update blocked load when entering idle */
2959 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2960 /* Update nohz.next_balance */
2961 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2962
2963 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2964
2965 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2966
2967 extern void nohz_balance_exit_idle(struct rq *rq);
2968 #else /* !CONFIG_NO_HZ_COMMON: */
2969 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2970 #endif /* !CONFIG_NO_HZ_COMMON */
2971
2972 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2973 extern void nohz_run_idle_balance(int cpu);
2974 #else
2975 static inline void nohz_run_idle_balance(int cpu) { }
2976 #endif
2977
2978 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2979
2980 struct irqtime {
2981 u64 total;
2982 u64 tick_delta;
2983 u64 irq_start_time;
2984 struct u64_stats_sync sync;
2985 };
2986
2987 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2988
2989 /*
2990 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2991 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2992 * and never move forward.
2993 */
2994 static inline u64 irq_time_read(int cpu)
2995 {
2996 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2997 unsigned int seq;
2998 u64 total;
2999
3000 do {
3001 seq = __u64_stats_fetch_begin(&irqtime->sync);
3002 total = irqtime->total;
3003 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3004
3005 return total;
3006 }
3007
3008 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3009
3010 #ifdef CONFIG_CPU_FREQ
3011
3012 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3013
3014 /**
3015 * cpufreq_update_util - Take a note about CPU utilization changes.
3016 * @rq: Runqueue to carry out the update for.
3017 * @flags: Update reason flags.
3018 *
3019 * This function is called by the scheduler on the CPU whose utilization is
3020 * being updated.
3021 *
3022 * It can only be called from RCU-sched read-side critical sections.
3023 *
3024 * The way cpufreq is currently arranged requires it to evaluate the CPU
3025 * performance state (frequency/voltage) on a regular basis to prevent it from
3026 * being stuck in a completely inadequate performance level for too long.
3027 * That is not guaranteed to happen if the updates are only triggered from CFS
3028 * and DL, though, because they may not be coming in if only RT tasks are
3029 * active all the time (or there are RT tasks only).
3030 *
3031 * As a workaround for that issue, this function is called periodically by the
3032 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3033 * but that really is a band-aid. Going forward it should be replaced with
3034 * solutions targeted more specifically at RT tasks.
3035 */
3036 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3037 {
3038 struct update_util_data *data;
3039
3040 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3041 cpu_of(rq)));
3042 if (data)
3043 data->func(data, rq_clock(rq), flags);
3044 }
3045 #else /* !CONFIG_CPU_FREQ: */
3046 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3047 #endif /* !CONFIG_CPU_FREQ */
3048
3049 #ifdef arch_scale_freq_capacity
3050 # ifndef arch_scale_freq_invariant
3051 # define arch_scale_freq_invariant() true
3052 # endif
3053 #else
3054 # define arch_scale_freq_invariant() false
3055 #endif
3056
3057 #ifdef CONFIG_SMP
3058
3059 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3060 unsigned long *min,
3061 unsigned long *max);
3062
3063 unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3064 unsigned long min,
3065 unsigned long max);
3066
3067
3068 /*
3069 * Verify the fitness of task @p to run on @cpu taking into account the
3070 * CPU original capacity and the runtime/deadline ratio of the task.
3071 *
3072 * The function will return true if the original capacity of @cpu is
3073 * greater than or equal to task's deadline density right shifted by
3074 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3075 */
3076 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3077 {
3078 unsigned long cap = arch_scale_cpu_capacity(cpu);
3079
3080 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3081 }
3082
3083 static inline unsigned long cpu_bw_dl(struct rq *rq)
3084 {
3085 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3086 }
3087
3088 static inline unsigned long cpu_util_dl(struct rq *rq)
3089 {
3090 return READ_ONCE(rq->avg_dl.util_avg);
3091 }
3092
3093
3094 extern unsigned long cpu_util_cfs(int cpu);
3095 extern unsigned long cpu_util_cfs_boost(int cpu);
3096
3097 static inline unsigned long cpu_util_rt(struct rq *rq)
3098 {
3099 return READ_ONCE(rq->avg_rt.util_avg);
3100 }
3101
3102 #endif /* CONFIG_SMP */
3103
3104 #ifdef CONFIG_UCLAMP_TASK
3105
3106 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3107
3108 static inline unsigned long uclamp_rq_get(struct rq *rq,
3109 enum uclamp_id clamp_id)
3110 {
3111 return READ_ONCE(rq->uclamp[clamp_id].value);
3112 }
3113
3114 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3115 unsigned int value)
3116 {
3117 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3118 }
3119
3120 static inline bool uclamp_rq_is_idle(struct rq *rq)
3121 {
3122 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3123 }
3124
3125 /* Is the rq being capped/throttled by uclamp_max? */
3126 static inline bool uclamp_rq_is_capped(struct rq *rq)
3127 {
3128 unsigned long rq_util;
3129 unsigned long max_util;
3130
3131 if (!static_branch_likely(&sched_uclamp_used))
3132 return false;
3133
3134 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3135 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3136
3137 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3138 }
3139
3140 /*
3141 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3142 * by default in the fast path and only gets turned on once userspace performs
3143 * an operation that requires it.
3144 *
3145 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3146 * hence is active.
3147 */
3148 static inline bool uclamp_is_used(void)
3149 {
3150 return static_branch_likely(&sched_uclamp_used);
3151 }
3152
3153 #define for_each_clamp_id(clamp_id) \
3154 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3155
3156 extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3157
3158
3159 static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3160 {
3161 if (clamp_id == UCLAMP_MIN)
3162 return 0;
3163 return SCHED_CAPACITY_SCALE;
3164 }
3165
3166 /* Integer rounded range for each bucket */
3167 #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3168
3169 static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3170 {
3171 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3172 }
3173
3174 static inline void
3175 uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3176 {
3177 uc_se->value = value;
3178 uc_se->bucket_id = uclamp_bucket_id(value);
3179 uc_se->user_defined = user_defined;
3180 }
3181
3182 #else /* !CONFIG_UCLAMP_TASK: */
3183
3184 static inline unsigned long
3185 uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3186 {
3187 if (clamp_id == UCLAMP_MIN)
3188 return 0;
3189
3190 return SCHED_CAPACITY_SCALE;
3191 }
3192
3193 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3194
3195 static inline bool uclamp_is_used(void)
3196 {
3197 return false;
3198 }
3199
3200 static inline unsigned long
3201 uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3202 {
3203 if (clamp_id == UCLAMP_MIN)
3204 return 0;
3205
3206 return SCHED_CAPACITY_SCALE;
3207 }
3208
3209 static inline void
3210 uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3211 {
3212 }
3213
3214 static inline bool uclamp_rq_is_idle(struct rq *rq)
3215 {
3216 return false;
3217 }
3218
3219 #endif /* !CONFIG_UCLAMP_TASK */
3220
3221 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3222
3223 static inline unsigned long cpu_util_irq(struct rq *rq)
3224 {
3225 return READ_ONCE(rq->avg_irq.util_avg);
3226 }
3227
3228 static inline
3229 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3230 {
3231 util *= (max - irq);
3232 util /= max;
3233
3234 return util;
3235
3236 }
3237
3238 #else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3239
3240 static inline unsigned long cpu_util_irq(struct rq *rq)
3241 {
3242 return 0;
3243 }
3244
3245 static inline
3246 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3247 {
3248 return util;
3249 }
3250
3251 #endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3252
3253 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3254
3255 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3256
3257 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3258
3259 static inline bool sched_energy_enabled(void)
3260 {
3261 return static_branch_unlikely(&sched_energy_present);
3262 }
3263
3264 extern struct cpufreq_governor schedutil_gov;
3265
3266 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3267
3268 #define perf_domain_span(pd) NULL
3269
3270 static inline bool sched_energy_enabled(void) { return false; }
3271
3272 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3273
3274 #ifdef CONFIG_MEMBARRIER
3275
3276 /*
3277 * The scheduler provides memory barriers required by membarrier between:
3278 * - prior user-space memory accesses and store to rq->membarrier_state,
3279 * - store to rq->membarrier_state and following user-space memory accesses.
3280 * In the same way it provides those guarantees around store to rq->curr.
3281 */
3282 static inline void membarrier_switch_mm(struct rq *rq,
3283 struct mm_struct *prev_mm,
3284 struct mm_struct *next_mm)
3285 {
3286 int membarrier_state;
3287
3288 if (prev_mm == next_mm)
3289 return;
3290
3291 membarrier_state = atomic_read(&next_mm->membarrier_state);
3292 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3293 return;
3294
3295 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3296 }
3297
3298 #else /* !CONFIG_MEMBARRIER :*/
3299
3300 static inline void membarrier_switch_mm(struct rq *rq,
3301 struct mm_struct *prev_mm,
3302 struct mm_struct *next_mm)
3303 {
3304 }
3305
3306 #endif /* !CONFIG_MEMBARRIER */
3307
3308 #ifdef CONFIG_SMP
3309 static inline bool is_per_cpu_kthread(struct task_struct *p)
3310 {
3311 if (!(p->flags & PF_KTHREAD))
3312 return false;
3313
3314 if (p->nr_cpus_allowed != 1)
3315 return false;
3316
3317 return true;
3318 }
3319 #endif
3320
3321 extern void swake_up_all_locked(struct swait_queue_head *q);
3322 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3323
3324 extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3325
3326 #ifdef CONFIG_PREEMPT_DYNAMIC
3327 extern int preempt_dynamic_mode;
3328 extern int sched_dynamic_mode(const char *str);
3329 extern void sched_dynamic_update(int mode);
3330 #endif
3331
3332 #ifdef CONFIG_SCHED_MM_CID
3333
3334 #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3335 #define MM_CID_SCAN_DELAY 100 /* 100ms */
3336
3337 extern raw_spinlock_t cid_lock;
3338 extern int use_cid_lock;
3339
3340 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3341 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3342 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3343 extern void init_sched_mm_cid(struct task_struct *t);
3344
3345 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3346 {
3347 if (cid < 0)
3348 return;
3349 cpumask_clear_cpu(cid, mm_cidmask(mm));
3350 }
3351
3352 /*
3353 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3354 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3355 * be held to transition to other states.
3356 *
3357 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3358 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3359 */
3360 static inline void mm_cid_put_lazy(struct task_struct *t)
3361 {
3362 struct mm_struct *mm = t->mm;
3363 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3364 int cid;
3365
3366 lockdep_assert_irqs_disabled();
3367 cid = __this_cpu_read(pcpu_cid->cid);
3368 if (!mm_cid_is_lazy_put(cid) ||
3369 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3370 return;
3371 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3372 }
3373
3374 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3375 {
3376 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3377 int cid, res;
3378
3379 lockdep_assert_irqs_disabled();
3380 cid = __this_cpu_read(pcpu_cid->cid);
3381 for (;;) {
3382 if (mm_cid_is_unset(cid))
3383 return MM_CID_UNSET;
3384 /*
3385 * Attempt transition from valid or lazy-put to unset.
3386 */
3387 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3388 if (res == cid)
3389 break;
3390 cid = res;
3391 }
3392 return cid;
3393 }
3394
3395 static inline void mm_cid_put(struct mm_struct *mm)
3396 {
3397 int cid;
3398
3399 lockdep_assert_irqs_disabled();
3400 cid = mm_cid_pcpu_unset(mm);
3401 if (cid == MM_CID_UNSET)
3402 return;
3403 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3404 }
3405
3406 static inline int __mm_cid_try_get(struct mm_struct *mm)
3407 {
3408 struct cpumask *cpumask;
3409 int cid;
3410
3411 cpumask = mm_cidmask(mm);
3412 /*
3413 * Retry finding first zero bit if the mask is temporarily
3414 * filled. This only happens during concurrent remote-clear
3415 * which owns a cid without holding a rq lock.
3416 */
3417 for (;;) {
3418 cid = cpumask_first_zero(cpumask);
3419 if (cid < nr_cpu_ids)
3420 break;
3421 cpu_relax();
3422 }
3423 if (cpumask_test_and_set_cpu(cid, cpumask))
3424 return -1;
3425
3426 return cid;
3427 }
3428
3429 /*
3430 * Save a snapshot of the current runqueue time of this cpu
3431 * with the per-cpu cid value, allowing to estimate how recently it was used.
3432 */
3433 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3434 {
3435 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3436
3437 lockdep_assert_rq_held(rq);
3438 WRITE_ONCE(pcpu_cid->time, rq->clock);
3439 }
3440
3441 static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3442 {
3443 int cid;
3444
3445 /*
3446 * All allocations (even those using the cid_lock) are lock-free. If
3447 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3448 * guarantee forward progress.
3449 */
3450 if (!READ_ONCE(use_cid_lock)) {
3451 cid = __mm_cid_try_get(mm);
3452 if (cid >= 0)
3453 goto end;
3454 raw_spin_lock(&cid_lock);
3455 } else {
3456 raw_spin_lock(&cid_lock);
3457 cid = __mm_cid_try_get(mm);
3458 if (cid >= 0)
3459 goto unlock;
3460 }
3461
3462 /*
3463 * cid concurrently allocated. Retry while forcing following
3464 * allocations to use the cid_lock to ensure forward progress.
3465 */
3466 WRITE_ONCE(use_cid_lock, 1);
3467 /*
3468 * Set use_cid_lock before allocation. Only care about program order
3469 * because this is only required for forward progress.
3470 */
3471 barrier();
3472 /*
3473 * Retry until it succeeds. It is guaranteed to eventually succeed once
3474 * all newcoming allocations observe the use_cid_lock flag set.
3475 */
3476 do {
3477 cid = __mm_cid_try_get(mm);
3478 cpu_relax();
3479 } while (cid < 0);
3480 /*
3481 * Allocate before clearing use_cid_lock. Only care about
3482 * program order because this is for forward progress.
3483 */
3484 barrier();
3485 WRITE_ONCE(use_cid_lock, 0);
3486 unlock:
3487 raw_spin_unlock(&cid_lock);
3488 end:
3489 mm_cid_snapshot_time(rq, mm);
3490
3491 return cid;
3492 }
3493
3494 static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3495 {
3496 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3497 struct cpumask *cpumask;
3498 int cid;
3499
3500 lockdep_assert_rq_held(rq);
3501 cpumask = mm_cidmask(mm);
3502 cid = __this_cpu_read(pcpu_cid->cid);
3503 if (mm_cid_is_valid(cid)) {
3504 mm_cid_snapshot_time(rq, mm);
3505 return cid;
3506 }
3507 if (mm_cid_is_lazy_put(cid)) {
3508 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3509 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3510 }
3511 cid = __mm_cid_get(rq, mm);
3512 __this_cpu_write(pcpu_cid->cid, cid);
3513
3514 return cid;
3515 }
3516
3517 static inline void switch_mm_cid(struct rq *rq,
3518 struct task_struct *prev,
3519 struct task_struct *next)
3520 {
3521 /*
3522 * Provide a memory barrier between rq->curr store and load of
3523 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3524 *
3525 * Should be adapted if context_switch() is modified.
3526 */
3527 if (!next->mm) { // to kernel
3528 /*
3529 * user -> kernel transition does not guarantee a barrier, but
3530 * we can use the fact that it performs an atomic operation in
3531 * mmgrab().
3532 */
3533 if (prev->mm) // from user
3534 smp_mb__after_mmgrab();
3535 /*
3536 * kernel -> kernel transition does not change rq->curr->mm
3537 * state. It stays NULL.
3538 */
3539 } else { // to user
3540 /*
3541 * kernel -> user transition does not provide a barrier
3542 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3543 * Provide it here.
3544 */
3545 if (!prev->mm) { // from kernel
3546 smp_mb();
3547 } else { // from user
3548 /*
3549 * user->user transition relies on an implicit
3550 * memory barrier in switch_mm() when
3551 * current->mm changes. If the architecture
3552 * switch_mm() does not have an implicit memory
3553 * barrier, it is emitted here. If current->mm
3554 * is unchanged, no barrier is needed.
3555 */
3556 smp_mb__after_switch_mm();
3557 }
3558 }
3559 if (prev->mm_cid_active) {
3560 mm_cid_snapshot_time(rq, prev->mm);
3561 mm_cid_put_lazy(prev);
3562 prev->mm_cid = -1;
3563 }
3564 if (next->mm_cid_active)
3565 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3566 }
3567
3568 #else /* !CONFIG_SCHED_MM_CID: */
3569 static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3570 static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3571 static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3572 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3573 static inline void init_sched_mm_cid(struct task_struct *t) { }
3574 #endif /* !CONFIG_SCHED_MM_CID */
3575
3576 extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3577 extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3578
3579 #ifdef CONFIG_RT_MUTEXES
3580
3581 static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3582 {
3583 if (pi_task)
3584 prio = min(prio, pi_task->prio);
3585
3586 return prio;
3587 }
3588
3589 static inline int rt_effective_prio(struct task_struct *p, int prio)
3590 {
3591 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3592
3593 return __rt_effective_prio(pi_task, prio);
3594 }
3595
3596 #else /* !CONFIG_RT_MUTEXES: */
3597
3598 static inline int rt_effective_prio(struct task_struct *p, int prio)
3599 {
3600 return prio;
3601 }
3602
3603 #endif /* !CONFIG_RT_MUTEXES */
3604
3605 extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3606 extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3607 extern void __setscheduler_prio(struct task_struct *p, int prio);
3608 extern void set_load_weight(struct task_struct *p, bool update_load);
3609 extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3610 extern void dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3611
3612 extern void check_class_changed(struct rq *rq, struct task_struct *p,
3613 const struct sched_class *prev_class,
3614 int oldprio);
3615
3616 #ifdef CONFIG_SMP
3617 extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3618 extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3619 #else
3620
3621 static inline struct balance_callback *splice_balance_callbacks(struct rq *rq)
3622 {
3623 return NULL;
3624 }
3625
3626 static inline void balance_callbacks(struct rq *rq, struct balance_callback *head)
3627 {
3628 }
3629
3630 #endif
3631
3632 #endif /* _KERNEL_SCHED_SCHED_H */
3633
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.