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 #ifdef CONFIG_PARAVIRT
78 # include <asm/paravirt.h>
79 # include <asm/paravirt_api_clock.h>
80 #endif
81
82 #include <asm/barrier.h>
83
84 #include "cpupri.h"
85 #include "cpudeadline.h"
86
87 #ifdef CONFIG_SCHED_DEBUG
88 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
89 #else
90 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
91 #endif
92
93 struct rq;
94 struct cpuidle_state;
95
96 /* task_struct::on_rq states: */
97 #define TASK_ON_RQ_QUEUED 1
98 #define TASK_ON_RQ_MIGRATING 2
99
100 extern __read_mostly int scheduler_running;
101
102 extern unsigned long calc_load_update;
103 extern atomic_long_t calc_load_tasks;
104
105 extern void calc_global_load_tick(struct rq *this_rq);
106 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
107
108 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
109
110 extern int sysctl_sched_rt_period;
111 extern int sysctl_sched_rt_runtime;
112 extern int sched_rr_timeslice;
113
114 /*
115 * Asymmetric CPU capacity bits
116 */
117 struct asym_cap_data {
118 struct list_head link;
119 struct rcu_head rcu;
120 unsigned long capacity;
121 unsigned long cpus[];
122 };
123
124 extern struct list_head asym_cap_list;
125
126 #define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
127
128 /*
129 * Helpers for converting nanosecond timing to jiffy resolution
130 */
131 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
132
133 /*
134 * Increase resolution of nice-level calculations for 64-bit architectures.
135 * The extra resolution improves shares distribution and load balancing of
136 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
137 * hierarchies, especially on larger systems. This is not a user-visible change
138 * and does not change the user-interface for setting shares/weights.
139 *
140 * We increase resolution only if we have enough bits to allow this increased
141 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
142 * are pretty high and the returns do not justify the increased costs.
143 *
144 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
145 * increase coverage and consistency always enable it on 64-bit platforms.
146 */
147 #ifdef CONFIG_64BIT
148 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
149 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
150 # define scale_load_down(w) \
151 ({ \
152 unsigned long __w = (w); \
153 if (__w) \
154 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
155 __w; \
156 })
157 #else
158 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
159 # define scale_load(w) (w)
160 # define scale_load_down(w) (w)
161 #endif
162
163 /*
164 * Task weight (visible to users) and its load (invisible to users) have
165 * independent resolution, but they should be well calibrated. We use
166 * scale_load() and scale_load_down(w) to convert between them. The
167 * following must be true:
168 *
169 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
170 *
171 */
172 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
173
174 /*
175 * Single value that decides SCHED_DEADLINE internal math precision.
176 * 10 -> just above 1us
177 * 9 -> just above 0.5us
178 */
179 #define DL_SCALE 10
180
181 /*
182 * Single value that denotes runtime == period, ie unlimited time.
183 */
184 #define RUNTIME_INF ((u64)~0ULL)
185
idle_policy(int policy)186 static inline int idle_policy(int policy)
187 {
188 return policy == SCHED_IDLE;
189 }
fair_policy(int policy)190 static inline int fair_policy(int policy)
191 {
192 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
193 }
194
rt_policy(int policy)195 static inline int rt_policy(int policy)
196 {
197 return policy == SCHED_FIFO || policy == SCHED_RR;
198 }
199
dl_policy(int policy)200 static inline int dl_policy(int policy)
201 {
202 return policy == SCHED_DEADLINE;
203 }
valid_policy(int policy)204 static inline bool valid_policy(int policy)
205 {
206 return idle_policy(policy) || fair_policy(policy) ||
207 rt_policy(policy) || dl_policy(policy);
208 }
209
task_has_idle_policy(struct task_struct * p)210 static inline int task_has_idle_policy(struct task_struct *p)
211 {
212 return idle_policy(p->policy);
213 }
214
task_has_rt_policy(struct task_struct * p)215 static inline int task_has_rt_policy(struct task_struct *p)
216 {
217 return rt_policy(p->policy);
218 }
219
task_has_dl_policy(struct task_struct * p)220 static inline int task_has_dl_policy(struct task_struct *p)
221 {
222 return dl_policy(p->policy);
223 }
224
225 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
226
update_avg(u64 * avg,u64 sample)227 static inline void update_avg(u64 *avg, u64 sample)
228 {
229 s64 diff = sample - *avg;
230 *avg += diff / 8;
231 }
232
233 /*
234 * Shifting a value by an exponent greater *or equal* to the size of said value
235 * is UB; cap at size-1.
236 */
237 #define shr_bound(val, shift) \
238 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
239
240 /*
241 * !! For sched_setattr_nocheck() (kernel) only !!
242 *
243 * This is actually gross. :(
244 *
245 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
246 * tasks, but still be able to sleep. We need this on platforms that cannot
247 * atomically change clock frequency. Remove once fast switching will be
248 * available on such platforms.
249 *
250 * SUGOV stands for SchedUtil GOVernor.
251 */
252 #define SCHED_FLAG_SUGOV 0x10000000
253
254 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
255
dl_entity_is_special(const struct sched_dl_entity * dl_se)256 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
257 {
258 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
259 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
260 #else
261 return false;
262 #endif
263 }
264
265 /*
266 * Tells if entity @a should preempt entity @b.
267 */
dl_entity_preempt(const struct sched_dl_entity * a,const struct sched_dl_entity * b)268 static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
269 const struct sched_dl_entity *b)
270 {
271 return dl_entity_is_special(a) ||
272 dl_time_before(a->deadline, b->deadline);
273 }
274
275 /*
276 * This is the priority-queue data structure of the RT scheduling class:
277 */
278 struct rt_prio_array {
279 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
280 struct list_head queue[MAX_RT_PRIO];
281 };
282
283 struct rt_bandwidth {
284 /* nests inside the rq lock: */
285 raw_spinlock_t rt_runtime_lock;
286 ktime_t rt_period;
287 u64 rt_runtime;
288 struct hrtimer rt_period_timer;
289 unsigned int rt_period_active;
290 };
291
dl_bandwidth_enabled(void)292 static inline int dl_bandwidth_enabled(void)
293 {
294 return sysctl_sched_rt_runtime >= 0;
295 }
296
297 /*
298 * To keep the bandwidth of -deadline tasks under control
299 * we need some place where:
300 * - store the maximum -deadline bandwidth of each cpu;
301 * - cache the fraction of bandwidth that is currently allocated in
302 * each root domain;
303 *
304 * This is all done in the data structure below. It is similar to the
305 * one used for RT-throttling (rt_bandwidth), with the main difference
306 * that, since here we are only interested in admission control, we
307 * do not decrease any runtime while the group "executes", neither we
308 * need a timer to replenish it.
309 *
310 * With respect to SMP, bandwidth is given on a per root domain basis,
311 * meaning that:
312 * - bw (< 100%) is the deadline bandwidth of each CPU;
313 * - total_bw is the currently allocated bandwidth in each root domain;
314 */
315 struct dl_bw {
316 raw_spinlock_t lock;
317 u64 bw;
318 u64 total_bw;
319 };
320
321 extern void init_dl_bw(struct dl_bw *dl_b);
322 extern int sched_dl_global_validate(void);
323 extern void sched_dl_do_global(void);
324 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
325 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
326 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
327 extern bool __checkparam_dl(const struct sched_attr *attr);
328 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
329 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
330 extern int dl_bw_check_overflow(int cpu);
331
332 /*
333 * SCHED_DEADLINE supports servers (nested scheduling) with the following
334 * interface:
335 *
336 * dl_se::rq -- runqueue we belong to.
337 *
338 * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
339 * server when it runs out of tasks to run.
340 *
341 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
342 * returns NULL.
343 *
344 * dl_server_update() -- called from update_curr_common(), propagates runtime
345 * to the server.
346 *
347 * dl_server_start()
348 * dl_server_stop() -- start/stop the server when it has (no) tasks.
349 *
350 * dl_server_init() -- initializes the server.
351 */
352 extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
353 extern void dl_server_start(struct sched_dl_entity *dl_se);
354 extern void dl_server_stop(struct sched_dl_entity *dl_se);
355 extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
356 dl_server_has_tasks_f has_tasks,
357 dl_server_pick_f pick);
358
359 #ifdef CONFIG_CGROUP_SCHED
360
361 struct cfs_rq;
362 struct rt_rq;
363
364 extern struct list_head task_groups;
365
366 struct cfs_bandwidth {
367 #ifdef CONFIG_CFS_BANDWIDTH
368 raw_spinlock_t lock;
369 ktime_t period;
370 u64 quota;
371 u64 runtime;
372 u64 burst;
373 u64 runtime_snap;
374 s64 hierarchical_quota;
375
376 u8 idle;
377 u8 period_active;
378 u8 slack_started;
379 struct hrtimer period_timer;
380 struct hrtimer slack_timer;
381 struct list_head throttled_cfs_rq;
382
383 /* Statistics: */
384 int nr_periods;
385 int nr_throttled;
386 int nr_burst;
387 u64 throttled_time;
388 u64 burst_time;
389 #endif
390 };
391
392 /* Task group related information */
393 struct task_group {
394 struct cgroup_subsys_state css;
395
396 #ifdef CONFIG_FAIR_GROUP_SCHED
397 /* schedulable entities of this group on each CPU */
398 struct sched_entity **se;
399 /* runqueue "owned" by this group on each CPU */
400 struct cfs_rq **cfs_rq;
401 unsigned long shares;
402
403 /* A positive value indicates that this is a SCHED_IDLE group. */
404 int idle;
405
406 #ifdef CONFIG_SMP
407 /*
408 * load_avg can be heavily contended at clock tick time, so put
409 * it in its own cacheline separated from the fields above which
410 * will also be accessed at each tick.
411 */
412 atomic_long_t load_avg ____cacheline_aligned;
413 #endif
414 #endif
415
416 #ifdef CONFIG_RT_GROUP_SCHED
417 struct sched_rt_entity **rt_se;
418 struct rt_rq **rt_rq;
419
420 struct rt_bandwidth rt_bandwidth;
421 #endif
422
423 struct rcu_head rcu;
424 struct list_head list;
425
426 struct task_group *parent;
427 struct list_head siblings;
428 struct list_head children;
429
430 #ifdef CONFIG_SCHED_AUTOGROUP
431 struct autogroup *autogroup;
432 #endif
433
434 struct cfs_bandwidth cfs_bandwidth;
435
436 #ifdef CONFIG_UCLAMP_TASK_GROUP
437 /* The two decimal precision [%] value requested from user-space */
438 unsigned int uclamp_pct[UCLAMP_CNT];
439 /* Clamp values requested for a task group */
440 struct uclamp_se uclamp_req[UCLAMP_CNT];
441 /* Effective clamp values used for a task group */
442 struct uclamp_se uclamp[UCLAMP_CNT];
443 #endif
444
445 };
446
447 #ifdef CONFIG_FAIR_GROUP_SCHED
448 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
449
450 /*
451 * A weight of 0 or 1 can cause arithmetics problems.
452 * A weight of a cfs_rq is the sum of weights of which entities
453 * are queued on this cfs_rq, so a weight of a entity should not be
454 * too large, so as the shares value of a task group.
455 * (The default weight is 1024 - so there's no practical
456 * limitation from this.)
457 */
458 #define MIN_SHARES (1UL << 1)
459 #define MAX_SHARES (1UL << 18)
460 #endif
461
462 typedef int (*tg_visitor)(struct task_group *, void *);
463
464 extern int walk_tg_tree_from(struct task_group *from,
465 tg_visitor down, tg_visitor up, void *data);
466
467 /*
468 * Iterate the full tree, calling @down when first entering a node and @up when
469 * leaving it for the final time.
470 *
471 * Caller must hold rcu_lock or sufficient equivalent.
472 */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)473 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
474 {
475 return walk_tg_tree_from(&root_task_group, down, up, data);
476 }
477
478 extern int tg_nop(struct task_group *tg, void *data);
479
480 #ifdef CONFIG_FAIR_GROUP_SCHED
481 extern void free_fair_sched_group(struct task_group *tg);
482 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
483 extern void online_fair_sched_group(struct task_group *tg);
484 extern void unregister_fair_sched_group(struct task_group *tg);
485 #else
free_fair_sched_group(struct task_group * tg)486 static inline void free_fair_sched_group(struct task_group *tg) { }
alloc_fair_sched_group(struct task_group * tg,struct task_group * parent)487 static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
488 {
489 return 1;
490 }
online_fair_sched_group(struct task_group * tg)491 static inline void online_fair_sched_group(struct task_group *tg) { }
unregister_fair_sched_group(struct task_group * tg)492 static inline void unregister_fair_sched_group(struct task_group *tg) { }
493 #endif
494
495 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
496 struct sched_entity *se, int cpu,
497 struct sched_entity *parent);
498 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
499
500 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
501 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
502 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
503 extern bool cfs_task_bw_constrained(struct task_struct *p);
504
505 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
506 struct sched_rt_entity *rt_se, int cpu,
507 struct sched_rt_entity *parent);
508 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
509 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
510 extern long sched_group_rt_runtime(struct task_group *tg);
511 extern long sched_group_rt_period(struct task_group *tg);
512 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
513
514 extern struct task_group *sched_create_group(struct task_group *parent);
515 extern void sched_online_group(struct task_group *tg,
516 struct task_group *parent);
517 extern void sched_destroy_group(struct task_group *tg);
518 extern void sched_release_group(struct task_group *tg);
519
520 extern void sched_move_task(struct task_struct *tsk);
521
522 #ifdef CONFIG_FAIR_GROUP_SCHED
523 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
524
525 extern int sched_group_set_idle(struct task_group *tg, long idle);
526
527 #ifdef CONFIG_SMP
528 extern void set_task_rq_fair(struct sched_entity *se,
529 struct cfs_rq *prev, struct cfs_rq *next);
530 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)531 static inline void set_task_rq_fair(struct sched_entity *se,
532 struct cfs_rq *prev, struct cfs_rq *next) { }
533 #endif /* CONFIG_SMP */
534 #endif /* CONFIG_FAIR_GROUP_SCHED */
535
536 #else /* CONFIG_CGROUP_SCHED */
537
538 struct cfs_bandwidth { };
cfs_task_bw_constrained(struct task_struct * p)539 static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
540
541 #endif /* CONFIG_CGROUP_SCHED */
542
543 extern void unregister_rt_sched_group(struct task_group *tg);
544 extern void free_rt_sched_group(struct task_group *tg);
545 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
546
547 /*
548 * u64_u32_load/u64_u32_store
549 *
550 * Use a copy of a u64 value to protect against data race. This is only
551 * applicable for 32-bits architectures.
552 */
553 #ifdef CONFIG_64BIT
554 # define u64_u32_load_copy(var, copy) var
555 # define u64_u32_store_copy(var, copy, val) (var = val)
556 #else
557 # define u64_u32_load_copy(var, copy) \
558 ({ \
559 u64 __val, __val_copy; \
560 do { \
561 __val_copy = copy; \
562 /* \
563 * paired with u64_u32_store_copy(), ordering access \
564 * to var and copy. \
565 */ \
566 smp_rmb(); \
567 __val = var; \
568 } while (__val != __val_copy); \
569 __val; \
570 })
571 # define u64_u32_store_copy(var, copy, val) \
572 do { \
573 typeof(val) __val = (val); \
574 var = __val; \
575 /* \
576 * paired with u64_u32_load_copy(), ordering access to var and \
577 * copy. \
578 */ \
579 smp_wmb(); \
580 copy = __val; \
581 } while (0)
582 #endif
583 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
584 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
585
586 /* CFS-related fields in a runqueue */
587 struct cfs_rq {
588 struct load_weight load;
589 unsigned int nr_running;
590 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
591 unsigned int idle_nr_running; /* SCHED_IDLE */
592 unsigned int idle_h_nr_running; /* SCHED_IDLE */
593
594 s64 avg_vruntime;
595 u64 avg_load;
596
597 u64 exec_clock;
598 u64 min_vruntime;
599 #ifdef CONFIG_SCHED_CORE
600 unsigned int forceidle_seq;
601 u64 min_vruntime_fi;
602 #endif
603
604 #ifndef CONFIG_64BIT
605 u64 min_vruntime_copy;
606 #endif
607
608 struct rb_root_cached tasks_timeline;
609
610 /*
611 * 'curr' points to currently running entity on this cfs_rq.
612 * It is set to NULL otherwise (i.e when none are currently running).
613 */
614 struct sched_entity *curr;
615 struct sched_entity *next;
616
617 #ifdef CONFIG_SCHED_DEBUG
618 unsigned int nr_spread_over;
619 #endif
620
621 #ifdef CONFIG_SMP
622 /*
623 * CFS load tracking
624 */
625 struct sched_avg avg;
626 #ifndef CONFIG_64BIT
627 u64 last_update_time_copy;
628 #endif
629 struct {
630 raw_spinlock_t lock ____cacheline_aligned;
631 int nr;
632 unsigned long load_avg;
633 unsigned long util_avg;
634 unsigned long runnable_avg;
635 } removed;
636
637 #ifdef CONFIG_FAIR_GROUP_SCHED
638 u64 last_update_tg_load_avg;
639 unsigned long tg_load_avg_contrib;
640 long propagate;
641 long prop_runnable_sum;
642
643 /*
644 * h_load = weight * f(tg)
645 *
646 * Where f(tg) is the recursive weight fraction assigned to
647 * this group.
648 */
649 unsigned long h_load;
650 u64 last_h_load_update;
651 struct sched_entity *h_load_next;
652 #endif /* CONFIG_FAIR_GROUP_SCHED */
653 #endif /* CONFIG_SMP */
654
655 #ifdef CONFIG_FAIR_GROUP_SCHED
656 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
657
658 /*
659 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
660 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
661 * (like users, containers etc.)
662 *
663 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
664 * This list is used during load balance.
665 */
666 int on_list;
667 struct list_head leaf_cfs_rq_list;
668 struct task_group *tg; /* group that "owns" this runqueue */
669
670 /* Locally cached copy of our task_group's idle value */
671 int idle;
672
673 #ifdef CONFIG_CFS_BANDWIDTH
674 int runtime_enabled;
675 s64 runtime_remaining;
676
677 u64 throttled_pelt_idle;
678 #ifndef CONFIG_64BIT
679 u64 throttled_pelt_idle_copy;
680 #endif
681 u64 throttled_clock;
682 u64 throttled_clock_pelt;
683 u64 throttled_clock_pelt_time;
684 u64 throttled_clock_self;
685 u64 throttled_clock_self_time;
686 int throttled;
687 int throttle_count;
688 struct list_head throttled_list;
689 struct list_head throttled_csd_list;
690 #endif /* CONFIG_CFS_BANDWIDTH */
691 #endif /* CONFIG_FAIR_GROUP_SCHED */
692 };
693
rt_bandwidth_enabled(void)694 static inline int rt_bandwidth_enabled(void)
695 {
696 return sysctl_sched_rt_runtime >= 0;
697 }
698
699 /* RT IPI pull logic requires IRQ_WORK */
700 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
701 # define HAVE_RT_PUSH_IPI
702 #endif
703
704 /* Real-Time classes' related field in a runqueue: */
705 struct rt_rq {
706 struct rt_prio_array active;
707 unsigned int rt_nr_running;
708 unsigned int rr_nr_running;
709 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
710 struct {
711 int curr; /* highest queued rt task prio */
712 #ifdef CONFIG_SMP
713 int next; /* next highest */
714 #endif
715 } highest_prio;
716 #endif
717 #ifdef CONFIG_SMP
718 bool overloaded;
719 struct plist_head pushable_tasks;
720
721 #endif /* CONFIG_SMP */
722 int rt_queued;
723
724 int rt_throttled;
725 u64 rt_time;
726 u64 rt_runtime;
727 /* Nests inside the rq lock: */
728 raw_spinlock_t rt_runtime_lock;
729
730 #ifdef CONFIG_RT_GROUP_SCHED
731 unsigned int rt_nr_boosted;
732
733 struct rq *rq;
734 struct task_group *tg;
735 #endif
736 };
737
rt_rq_is_runnable(struct rt_rq * rt_rq)738 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
739 {
740 return rt_rq->rt_queued && rt_rq->rt_nr_running;
741 }
742
743 /* Deadline class' related fields in a runqueue */
744 struct dl_rq {
745 /* runqueue is an rbtree, ordered by deadline */
746 struct rb_root_cached root;
747
748 unsigned int dl_nr_running;
749
750 #ifdef CONFIG_SMP
751 /*
752 * Deadline values of the currently executing and the
753 * earliest ready task on this rq. Caching these facilitates
754 * the decision whether or not a ready but not running task
755 * should migrate somewhere else.
756 */
757 struct {
758 u64 curr;
759 u64 next;
760 } earliest_dl;
761
762 bool overloaded;
763
764 /*
765 * Tasks on this rq that can be pushed away. They are kept in
766 * an rb-tree, ordered by tasks' deadlines, with caching
767 * of the leftmost (earliest deadline) element.
768 */
769 struct rb_root_cached pushable_dl_tasks_root;
770 #else
771 struct dl_bw dl_bw;
772 #endif
773 /*
774 * "Active utilization" for this runqueue: increased when a
775 * task wakes up (becomes TASK_RUNNING) and decreased when a
776 * task blocks
777 */
778 u64 running_bw;
779
780 /*
781 * Utilization of the tasks "assigned" to this runqueue (including
782 * the tasks that are in runqueue and the tasks that executed on this
783 * CPU and blocked). Increased when a task moves to this runqueue, and
784 * decreased when the task moves away (migrates, changes scheduling
785 * policy, or terminates).
786 * This is needed to compute the "inactive utilization" for the
787 * runqueue (inactive utilization = this_bw - running_bw).
788 */
789 u64 this_bw;
790 u64 extra_bw;
791
792 /*
793 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
794 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
795 */
796 u64 max_bw;
797
798 /*
799 * Inverse of the fraction of CPU utilization that can be reclaimed
800 * by the GRUB algorithm.
801 */
802 u64 bw_ratio;
803 };
804
805 #ifdef CONFIG_FAIR_GROUP_SCHED
806 /* An entity is a task if it doesn't "own" a runqueue */
807 #define entity_is_task(se) (!se->my_q)
808
se_update_runnable(struct sched_entity * se)809 static inline void se_update_runnable(struct sched_entity *se)
810 {
811 if (!entity_is_task(se))
812 se->runnable_weight = se->my_q->h_nr_running;
813 }
814
se_runnable(struct sched_entity * se)815 static inline long se_runnable(struct sched_entity *se)
816 {
817 if (entity_is_task(se))
818 return !!se->on_rq;
819 else
820 return se->runnable_weight;
821 }
822
823 #else
824 #define entity_is_task(se) 1
825
se_update_runnable(struct sched_entity * se)826 static inline void se_update_runnable(struct sched_entity *se) {}
827
se_runnable(struct sched_entity * se)828 static inline long se_runnable(struct sched_entity *se)
829 {
830 return !!se->on_rq;
831 }
832 #endif
833
834 #ifdef CONFIG_SMP
835 /*
836 * XXX we want to get rid of these helpers and use the full load resolution.
837 */
se_weight(struct sched_entity * se)838 static inline long se_weight(struct sched_entity *se)
839 {
840 return scale_load_down(se->load.weight);
841 }
842
843
sched_asym_prefer(int a,int b)844 static inline bool sched_asym_prefer(int a, int b)
845 {
846 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
847 }
848
849 struct perf_domain {
850 struct em_perf_domain *em_pd;
851 struct perf_domain *next;
852 struct rcu_head rcu;
853 };
854
855 /*
856 * We add the notion of a root-domain which will be used to define per-domain
857 * variables. Each exclusive cpuset essentially defines an island domain by
858 * fully partitioning the member CPUs from any other cpuset. Whenever a new
859 * exclusive cpuset is created, we also create and attach a new root-domain
860 * object.
861 *
862 */
863 struct root_domain {
864 atomic_t refcount;
865 atomic_t rto_count;
866 struct rcu_head rcu;
867 cpumask_var_t span;
868 cpumask_var_t online;
869
870 /*
871 * Indicate pullable load on at least one CPU, e.g:
872 * - More than one runnable task
873 * - Running task is misfit
874 */
875 bool overloaded;
876
877 /* Indicate one or more cpus over-utilized (tipping point) */
878 bool overutilized;
879
880 /*
881 * The bit corresponding to a CPU gets set here if such CPU has more
882 * than one runnable -deadline task (as it is below for RT tasks).
883 */
884 cpumask_var_t dlo_mask;
885 atomic_t dlo_count;
886 struct dl_bw dl_bw;
887 struct cpudl cpudl;
888
889 /*
890 * Indicate whether a root_domain's dl_bw has been checked or
891 * updated. It's monotonously increasing value.
892 *
893 * Also, some corner cases, like 'wrap around' is dangerous, but given
894 * that u64 is 'big enough'. So that shouldn't be a concern.
895 */
896 u64 visit_gen;
897
898 #ifdef HAVE_RT_PUSH_IPI
899 /*
900 * For IPI pull requests, loop across the rto_mask.
901 */
902 struct irq_work rto_push_work;
903 raw_spinlock_t rto_lock;
904 /* These are only updated and read within rto_lock */
905 int rto_loop;
906 int rto_cpu;
907 /* These atomics are updated outside of a lock */
908 atomic_t rto_loop_next;
909 atomic_t rto_loop_start;
910 #endif
911 /*
912 * The "RT overload" flag: it gets set if a CPU has more than
913 * one runnable RT task.
914 */
915 cpumask_var_t rto_mask;
916 struct cpupri cpupri;
917
918 /*
919 * NULL-terminated list of performance domains intersecting with the
920 * CPUs of the rd. Protected by RCU.
921 */
922 struct perf_domain __rcu *pd;
923 };
924
925 extern void init_defrootdomain(void);
926 extern int sched_init_domains(const struct cpumask *cpu_map);
927 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
928 extern void sched_get_rd(struct root_domain *rd);
929 extern void sched_put_rd(struct root_domain *rd);
930
get_rd_overloaded(struct root_domain * rd)931 static inline int get_rd_overloaded(struct root_domain *rd)
932 {
933 return READ_ONCE(rd->overloaded);
934 }
935
set_rd_overloaded(struct root_domain * rd,int status)936 static inline void set_rd_overloaded(struct root_domain *rd, int status)
937 {
938 if (get_rd_overloaded(rd) != status)
939 WRITE_ONCE(rd->overloaded, status);
940 }
941
942 #ifdef HAVE_RT_PUSH_IPI
943 extern void rto_push_irq_work_func(struct irq_work *work);
944 #endif
945 #endif /* CONFIG_SMP */
946
947 #ifdef CONFIG_UCLAMP_TASK
948 /*
949 * struct uclamp_bucket - Utilization clamp bucket
950 * @value: utilization clamp value for tasks on this clamp bucket
951 * @tasks: number of RUNNABLE tasks on this clamp bucket
952 *
953 * Keep track of how many tasks are RUNNABLE for a given utilization
954 * clamp value.
955 */
956 struct uclamp_bucket {
957 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
958 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
959 };
960
961 /*
962 * struct uclamp_rq - rq's utilization clamp
963 * @value: currently active clamp values for a rq
964 * @bucket: utilization clamp buckets affecting a rq
965 *
966 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
967 * A clamp value is affecting a rq when there is at least one task RUNNABLE
968 * (or actually running) with that value.
969 *
970 * There are up to UCLAMP_CNT possible different clamp values, currently there
971 * are only two: minimum utilization and maximum utilization.
972 *
973 * All utilization clamping values are MAX aggregated, since:
974 * - for util_min: we want to run the CPU at least at the max of the minimum
975 * utilization required by its currently RUNNABLE tasks.
976 * - for util_max: we want to allow the CPU to run up to the max of the
977 * maximum utilization allowed by its currently RUNNABLE tasks.
978 *
979 * Since on each system we expect only a limited number of different
980 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
981 * the metrics required to compute all the per-rq utilization clamp values.
982 */
983 struct uclamp_rq {
984 unsigned int value;
985 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
986 };
987
988 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
989 #endif /* CONFIG_UCLAMP_TASK */
990
991 struct rq;
992 struct balance_callback {
993 struct balance_callback *next;
994 void (*func)(struct rq *rq);
995 };
996
997 /*
998 * This is the main, per-CPU runqueue data structure.
999 *
1000 * Locking rule: those places that want to lock multiple runqueues
1001 * (such as the load balancing or the thread migration code), lock
1002 * acquire operations must be ordered by ascending &runqueue.
1003 */
1004 struct rq {
1005 /* runqueue lock: */
1006 raw_spinlock_t __lock;
1007
1008 unsigned int nr_running;
1009 #ifdef CONFIG_NUMA_BALANCING
1010 unsigned int nr_numa_running;
1011 unsigned int nr_preferred_running;
1012 unsigned int numa_migrate_on;
1013 #endif
1014 #ifdef CONFIG_NO_HZ_COMMON
1015 #ifdef CONFIG_SMP
1016 unsigned long last_blocked_load_update_tick;
1017 unsigned int has_blocked_load;
1018 call_single_data_t nohz_csd;
1019 #endif /* CONFIG_SMP */
1020 unsigned int nohz_tick_stopped;
1021 atomic_t nohz_flags;
1022 #endif /* CONFIG_NO_HZ_COMMON */
1023
1024 #ifdef CONFIG_SMP
1025 unsigned int ttwu_pending;
1026 #endif
1027 u64 nr_switches;
1028
1029 #ifdef CONFIG_UCLAMP_TASK
1030 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1031 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1032 unsigned int uclamp_flags;
1033 #define UCLAMP_FLAG_IDLE 0x01
1034 #endif
1035
1036 struct cfs_rq cfs;
1037 struct rt_rq rt;
1038 struct dl_rq dl;
1039
1040 #ifdef CONFIG_FAIR_GROUP_SCHED
1041 /* list of leaf cfs_rq on this CPU: */
1042 struct list_head leaf_cfs_rq_list;
1043 struct list_head *tmp_alone_branch;
1044 #endif /* CONFIG_FAIR_GROUP_SCHED */
1045
1046 /*
1047 * This is part of a global counter where only the total sum
1048 * over all CPUs matters. A task can increase this counter on
1049 * one CPU and if it got migrated afterwards it may decrease
1050 * it on another CPU. Always updated under the runqueue lock:
1051 */
1052 unsigned int nr_uninterruptible;
1053
1054 struct task_struct __rcu *curr;
1055 struct task_struct *idle;
1056 struct task_struct *stop;
1057 unsigned long next_balance;
1058 struct mm_struct *prev_mm;
1059
1060 unsigned int clock_update_flags;
1061 u64 clock;
1062 /* Ensure that all clocks are in the same cache line */
1063 u64 clock_task ____cacheline_aligned;
1064 u64 clock_pelt;
1065 unsigned long lost_idle_time;
1066 u64 clock_pelt_idle;
1067 u64 clock_idle;
1068 #ifndef CONFIG_64BIT
1069 u64 clock_pelt_idle_copy;
1070 u64 clock_idle_copy;
1071 #endif
1072
1073 atomic_t nr_iowait;
1074
1075 #ifdef CONFIG_SCHED_DEBUG
1076 u64 last_seen_need_resched_ns;
1077 int ticks_without_resched;
1078 #endif
1079
1080 #ifdef CONFIG_MEMBARRIER
1081 int membarrier_state;
1082 #endif
1083
1084 #ifdef CONFIG_SMP
1085 struct root_domain *rd;
1086 struct sched_domain __rcu *sd;
1087
1088 unsigned long cpu_capacity;
1089
1090 struct balance_callback *balance_callback;
1091
1092 unsigned char nohz_idle_balance;
1093 unsigned char idle_balance;
1094
1095 unsigned long misfit_task_load;
1096
1097 /* For active balancing */
1098 int active_balance;
1099 int push_cpu;
1100 struct cpu_stop_work active_balance_work;
1101
1102 /* CPU of this runqueue: */
1103 int cpu;
1104 int online;
1105
1106 struct list_head cfs_tasks;
1107
1108 struct sched_avg avg_rt;
1109 struct sched_avg avg_dl;
1110 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1111 struct sched_avg avg_irq;
1112 #endif
1113 #ifdef CONFIG_SCHED_HW_PRESSURE
1114 struct sched_avg avg_hw;
1115 #endif
1116 u64 idle_stamp;
1117 u64 avg_idle;
1118
1119 /* This is used to determine avg_idle's max value */
1120 u64 max_idle_balance_cost;
1121
1122 #ifdef CONFIG_HOTPLUG_CPU
1123 struct rcuwait hotplug_wait;
1124 #endif
1125 #endif /* CONFIG_SMP */
1126
1127 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1128 u64 prev_irq_time;
1129 #endif
1130 #ifdef CONFIG_PARAVIRT
1131 u64 prev_steal_time;
1132 #endif
1133 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1134 u64 prev_steal_time_rq;
1135 #endif
1136
1137 /* calc_load related fields */
1138 unsigned long calc_load_update;
1139 long calc_load_active;
1140
1141 #ifdef CONFIG_SCHED_HRTICK
1142 #ifdef CONFIG_SMP
1143 call_single_data_t hrtick_csd;
1144 #endif
1145 struct hrtimer hrtick_timer;
1146 ktime_t hrtick_time;
1147 #endif
1148
1149 #ifdef CONFIG_SCHEDSTATS
1150 /* latency stats */
1151 struct sched_info rq_sched_info;
1152 unsigned long long rq_cpu_time;
1153 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1154
1155 /* sys_sched_yield() stats */
1156 unsigned int yld_count;
1157
1158 /* schedule() stats */
1159 unsigned int sched_count;
1160 unsigned int sched_goidle;
1161
1162 /* try_to_wake_up() stats */
1163 unsigned int ttwu_count;
1164 unsigned int ttwu_local;
1165 #endif
1166
1167 #ifdef CONFIG_CPU_IDLE
1168 /* Must be inspected within a rcu lock section */
1169 struct cpuidle_state *idle_state;
1170 #endif
1171
1172 #ifdef CONFIG_SMP
1173 unsigned int nr_pinned;
1174 #endif
1175 unsigned int push_busy;
1176 struct cpu_stop_work push_work;
1177
1178 #ifdef CONFIG_SCHED_CORE
1179 /* per rq */
1180 struct rq *core;
1181 struct task_struct *core_pick;
1182 unsigned int core_enabled;
1183 unsigned int core_sched_seq;
1184 struct rb_root core_tree;
1185
1186 /* shared state -- careful with sched_core_cpu_deactivate() */
1187 unsigned int core_task_seq;
1188 unsigned int core_pick_seq;
1189 unsigned long core_cookie;
1190 unsigned int core_forceidle_count;
1191 unsigned int core_forceidle_seq;
1192 unsigned int core_forceidle_occupation;
1193 u64 core_forceidle_start;
1194 #endif
1195
1196 /* Scratch cpumask to be temporarily used under rq_lock */
1197 cpumask_var_t scratch_mask;
1198
1199 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1200 call_single_data_t cfsb_csd;
1201 struct list_head cfsb_csd_list;
1202 #endif
1203 };
1204
1205 #ifdef CONFIG_FAIR_GROUP_SCHED
1206
1207 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1208 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1209 {
1210 return cfs_rq->rq;
1211 }
1212
1213 #else
1214
rq_of(struct cfs_rq * cfs_rq)1215 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1216 {
1217 return container_of(cfs_rq, struct rq, cfs);
1218 }
1219 #endif
1220
cpu_of(struct rq * rq)1221 static inline int cpu_of(struct rq *rq)
1222 {
1223 #ifdef CONFIG_SMP
1224 return rq->cpu;
1225 #else
1226 return 0;
1227 #endif
1228 }
1229
1230 #define MDF_PUSH 0x01
1231
is_migration_disabled(struct task_struct * p)1232 static inline bool is_migration_disabled(struct task_struct *p)
1233 {
1234 #ifdef CONFIG_SMP
1235 return p->migration_disabled;
1236 #else
1237 return false;
1238 #endif
1239 }
1240
1241 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1242
1243 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1244 #define this_rq() this_cpu_ptr(&runqueues)
1245 #define task_rq(p) cpu_rq(task_cpu(p))
1246 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1247 #define raw_rq() raw_cpu_ptr(&runqueues)
1248
1249 struct sched_group;
1250 #ifdef CONFIG_SCHED_CORE
1251 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1252
1253 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1254
sched_core_enabled(struct rq * rq)1255 static inline bool sched_core_enabled(struct rq *rq)
1256 {
1257 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1258 }
1259
sched_core_disabled(void)1260 static inline bool sched_core_disabled(void)
1261 {
1262 return !static_branch_unlikely(&__sched_core_enabled);
1263 }
1264
1265 /*
1266 * Be careful with this function; not for general use. The return value isn't
1267 * stable unless you actually hold a relevant rq->__lock.
1268 */
rq_lockp(struct rq * rq)1269 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1270 {
1271 if (sched_core_enabled(rq))
1272 return &rq->core->__lock;
1273
1274 return &rq->__lock;
1275 }
1276
__rq_lockp(struct rq * rq)1277 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1278 {
1279 if (rq->core_enabled)
1280 return &rq->core->__lock;
1281
1282 return &rq->__lock;
1283 }
1284
1285 bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1286 bool fi);
1287 void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1288
1289 /*
1290 * Helpers to check if the CPU's core cookie matches with the task's cookie
1291 * when core scheduling is enabled.
1292 * A special case is that the task's cookie always matches with CPU's core
1293 * cookie if the CPU is in an idle core.
1294 */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1295 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1296 {
1297 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1298 if (!sched_core_enabled(rq))
1299 return true;
1300
1301 return rq->core->core_cookie == p->core_cookie;
1302 }
1303
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1304 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1305 {
1306 bool idle_core = true;
1307 int cpu;
1308
1309 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1310 if (!sched_core_enabled(rq))
1311 return true;
1312
1313 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1314 if (!available_idle_cpu(cpu)) {
1315 idle_core = false;
1316 break;
1317 }
1318 }
1319
1320 /*
1321 * A CPU in an idle core is always the best choice for tasks with
1322 * cookies.
1323 */
1324 return idle_core || rq->core->core_cookie == p->core_cookie;
1325 }
1326
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1327 static inline bool sched_group_cookie_match(struct rq *rq,
1328 struct task_struct *p,
1329 struct sched_group *group)
1330 {
1331 int cpu;
1332
1333 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1334 if (!sched_core_enabled(rq))
1335 return true;
1336
1337 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1338 if (sched_core_cookie_match(cpu_rq(cpu), p))
1339 return true;
1340 }
1341 return false;
1342 }
1343
sched_core_enqueued(struct task_struct * p)1344 static inline bool sched_core_enqueued(struct task_struct *p)
1345 {
1346 return !RB_EMPTY_NODE(&p->core_node);
1347 }
1348
1349 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1350 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1351
1352 extern void sched_core_get(void);
1353 extern void sched_core_put(void);
1354
1355 #else /* !CONFIG_SCHED_CORE */
1356
sched_core_enabled(struct rq * rq)1357 static inline bool sched_core_enabled(struct rq *rq)
1358 {
1359 return false;
1360 }
1361
sched_core_disabled(void)1362 static inline bool sched_core_disabled(void)
1363 {
1364 return true;
1365 }
1366
rq_lockp(struct rq * rq)1367 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1368 {
1369 return &rq->__lock;
1370 }
1371
__rq_lockp(struct rq * rq)1372 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1373 {
1374 return &rq->__lock;
1375 }
1376
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1377 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1378 {
1379 return true;
1380 }
1381
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1382 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1383 {
1384 return true;
1385 }
1386
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1387 static inline bool sched_group_cookie_match(struct rq *rq,
1388 struct task_struct *p,
1389 struct sched_group *group)
1390 {
1391 return true;
1392 }
1393 #endif /* CONFIG_SCHED_CORE */
1394
lockdep_assert_rq_held(struct rq * rq)1395 static inline void lockdep_assert_rq_held(struct rq *rq)
1396 {
1397 lockdep_assert_held(__rq_lockp(rq));
1398 }
1399
1400 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1401 extern bool raw_spin_rq_trylock(struct rq *rq);
1402 extern void raw_spin_rq_unlock(struct rq *rq);
1403
raw_spin_rq_lock(struct rq * rq)1404 static inline void raw_spin_rq_lock(struct rq *rq)
1405 {
1406 raw_spin_rq_lock_nested(rq, 0);
1407 }
1408
raw_spin_rq_lock_irq(struct rq * rq)1409 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1410 {
1411 local_irq_disable();
1412 raw_spin_rq_lock(rq);
1413 }
1414
raw_spin_rq_unlock_irq(struct rq * rq)1415 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1416 {
1417 raw_spin_rq_unlock(rq);
1418 local_irq_enable();
1419 }
1420
_raw_spin_rq_lock_irqsave(struct rq * rq)1421 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1422 {
1423 unsigned long flags;
1424 local_irq_save(flags);
1425 raw_spin_rq_lock(rq);
1426 return flags;
1427 }
1428
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1429 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1430 {
1431 raw_spin_rq_unlock(rq);
1432 local_irq_restore(flags);
1433 }
1434
1435 #define raw_spin_rq_lock_irqsave(rq, flags) \
1436 do { \
1437 flags = _raw_spin_rq_lock_irqsave(rq); \
1438 } while (0)
1439
1440 #ifdef CONFIG_SCHED_SMT
1441 extern void __update_idle_core(struct rq *rq);
1442
update_idle_core(struct rq * rq)1443 static inline void update_idle_core(struct rq *rq)
1444 {
1445 if (static_branch_unlikely(&sched_smt_present))
1446 __update_idle_core(rq);
1447 }
1448
1449 #else
update_idle_core(struct rq * rq)1450 static inline void update_idle_core(struct rq *rq) { }
1451 #endif
1452
1453 #ifdef CONFIG_FAIR_GROUP_SCHED
task_of(struct sched_entity * se)1454 static inline struct task_struct *task_of(struct sched_entity *se)
1455 {
1456 SCHED_WARN_ON(!entity_is_task(se));
1457 return container_of(se, struct task_struct, se);
1458 }
1459
task_cfs_rq(struct task_struct * p)1460 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1461 {
1462 return p->se.cfs_rq;
1463 }
1464
1465 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(const struct sched_entity * se)1466 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1467 {
1468 return se->cfs_rq;
1469 }
1470
1471 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1472 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1473 {
1474 return grp->my_q;
1475 }
1476
1477 #else
1478
1479 #define task_of(_se) container_of(_se, struct task_struct, se)
1480
task_cfs_rq(const struct task_struct * p)1481 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1482 {
1483 return &task_rq(p)->cfs;
1484 }
1485
cfs_rq_of(const struct sched_entity * se)1486 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1487 {
1488 const struct task_struct *p = task_of(se);
1489 struct rq *rq = task_rq(p);
1490
1491 return &rq->cfs;
1492 }
1493
1494 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1495 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1496 {
1497 return NULL;
1498 }
1499 #endif
1500
1501 extern void update_rq_clock(struct rq *rq);
1502
1503 /*
1504 * rq::clock_update_flags bits
1505 *
1506 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1507 * call to __schedule(). This is an optimisation to avoid
1508 * neighbouring rq clock updates.
1509 *
1510 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1511 * in effect and calls to update_rq_clock() are being ignored.
1512 *
1513 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1514 * made to update_rq_clock() since the last time rq::lock was pinned.
1515 *
1516 * If inside of __schedule(), clock_update_flags will have been
1517 * shifted left (a left shift is a cheap operation for the fast path
1518 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1519 *
1520 * if (rq-clock_update_flags >= RQCF_UPDATED)
1521 *
1522 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1523 * one position though, because the next rq_unpin_lock() will shift it
1524 * back.
1525 */
1526 #define RQCF_REQ_SKIP 0x01
1527 #define RQCF_ACT_SKIP 0x02
1528 #define RQCF_UPDATED 0x04
1529
assert_clock_updated(struct rq * rq)1530 static inline void assert_clock_updated(struct rq *rq)
1531 {
1532 /*
1533 * The only reason for not seeing a clock update since the
1534 * last rq_pin_lock() is if we're currently skipping updates.
1535 */
1536 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1537 }
1538
rq_clock(struct rq * rq)1539 static inline u64 rq_clock(struct rq *rq)
1540 {
1541 lockdep_assert_rq_held(rq);
1542 assert_clock_updated(rq);
1543
1544 return rq->clock;
1545 }
1546
rq_clock_task(struct rq * rq)1547 static inline u64 rq_clock_task(struct rq *rq)
1548 {
1549 lockdep_assert_rq_held(rq);
1550 assert_clock_updated(rq);
1551
1552 return rq->clock_task;
1553 }
1554
rq_clock_skip_update(struct rq * rq)1555 static inline void rq_clock_skip_update(struct rq *rq)
1556 {
1557 lockdep_assert_rq_held(rq);
1558 rq->clock_update_flags |= RQCF_REQ_SKIP;
1559 }
1560
1561 /*
1562 * See rt task throttling, which is the only time a skip
1563 * request is canceled.
1564 */
rq_clock_cancel_skipupdate(struct rq * rq)1565 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1566 {
1567 lockdep_assert_rq_held(rq);
1568 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1569 }
1570
1571 /*
1572 * During cpu offlining and rq wide unthrottling, we can trigger
1573 * an update_rq_clock() for several cfs and rt runqueues (Typically
1574 * when using list_for_each_entry_*)
1575 * rq_clock_start_loop_update() can be called after updating the clock
1576 * once and before iterating over the list to prevent multiple update.
1577 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1578 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1579 */
rq_clock_start_loop_update(struct rq * rq)1580 static inline void rq_clock_start_loop_update(struct rq *rq)
1581 {
1582 lockdep_assert_rq_held(rq);
1583 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1584 rq->clock_update_flags |= RQCF_ACT_SKIP;
1585 }
1586
rq_clock_stop_loop_update(struct rq * rq)1587 static inline void rq_clock_stop_loop_update(struct rq *rq)
1588 {
1589 lockdep_assert_rq_held(rq);
1590 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1591 }
1592
1593 struct rq_flags {
1594 unsigned long flags;
1595 struct pin_cookie cookie;
1596 #ifdef CONFIG_SCHED_DEBUG
1597 /*
1598 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1599 * current pin context is stashed here in case it needs to be
1600 * restored in rq_repin_lock().
1601 */
1602 unsigned int clock_update_flags;
1603 #endif
1604 };
1605
1606 extern struct balance_callback balance_push_callback;
1607
1608 /*
1609 * Lockdep annotation that avoids accidental unlocks; it's like a
1610 * sticky/continuous lockdep_assert_held().
1611 *
1612 * This avoids code that has access to 'struct rq *rq' (basically everything in
1613 * the scheduler) from accidentally unlocking the rq if they do not also have a
1614 * copy of the (on-stack) 'struct rq_flags rf'.
1615 *
1616 * Also see Documentation/locking/lockdep-design.rst.
1617 */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1618 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1619 {
1620 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1621
1622 #ifdef CONFIG_SCHED_DEBUG
1623 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1624 rf->clock_update_flags = 0;
1625 #ifdef CONFIG_SMP
1626 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1627 #endif
1628 #endif
1629 }
1630
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1631 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1632 {
1633 #ifdef CONFIG_SCHED_DEBUG
1634 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1635 rf->clock_update_flags = RQCF_UPDATED;
1636 #endif
1637
1638 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1639 }
1640
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1641 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1642 {
1643 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1644
1645 #ifdef CONFIG_SCHED_DEBUG
1646 /*
1647 * Restore the value we stashed in @rf for this pin context.
1648 */
1649 rq->clock_update_flags |= rf->clock_update_flags;
1650 #endif
1651 }
1652
1653 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1654 __acquires(rq->lock);
1655
1656 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1657 __acquires(p->pi_lock)
1658 __acquires(rq->lock);
1659
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1660 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1661 __releases(rq->lock)
1662 {
1663 rq_unpin_lock(rq, rf);
1664 raw_spin_rq_unlock(rq);
1665 }
1666
1667 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1668 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1669 __releases(rq->lock)
1670 __releases(p->pi_lock)
1671 {
1672 rq_unpin_lock(rq, rf);
1673 raw_spin_rq_unlock(rq);
1674 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1675 }
1676
1677 DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1678 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1679 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1680 struct rq *rq; struct rq_flags rf)
1681
1682 static inline void
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1683 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1684 __acquires(rq->lock)
1685 {
1686 raw_spin_rq_lock_irqsave(rq, rf->flags);
1687 rq_pin_lock(rq, rf);
1688 }
1689
1690 static inline void
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1691 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1692 __acquires(rq->lock)
1693 {
1694 raw_spin_rq_lock_irq(rq);
1695 rq_pin_lock(rq, rf);
1696 }
1697
1698 static inline void
rq_lock(struct rq * rq,struct rq_flags * rf)1699 rq_lock(struct rq *rq, struct rq_flags *rf)
1700 __acquires(rq->lock)
1701 {
1702 raw_spin_rq_lock(rq);
1703 rq_pin_lock(rq, rf);
1704 }
1705
1706 static inline void
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1707 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1708 __releases(rq->lock)
1709 {
1710 rq_unpin_lock(rq, rf);
1711 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1712 }
1713
1714 static inline void
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1715 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1716 __releases(rq->lock)
1717 {
1718 rq_unpin_lock(rq, rf);
1719 raw_spin_rq_unlock_irq(rq);
1720 }
1721
1722 static inline void
rq_unlock(struct rq * rq,struct rq_flags * rf)1723 rq_unlock(struct rq *rq, struct rq_flags *rf)
1724 __releases(rq->lock)
1725 {
1726 rq_unpin_lock(rq, rf);
1727 raw_spin_rq_unlock(rq);
1728 }
1729
1730 DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1731 rq_lock(_T->lock, &_T->rf),
1732 rq_unlock(_T->lock, &_T->rf),
1733 struct rq_flags rf)
1734
1735 DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1736 rq_lock_irq(_T->lock, &_T->rf),
1737 rq_unlock_irq(_T->lock, &_T->rf),
1738 struct rq_flags rf)
1739
1740 DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1741 rq_lock_irqsave(_T->lock, &_T->rf),
1742 rq_unlock_irqrestore(_T->lock, &_T->rf),
1743 struct rq_flags rf)
1744
1745 static inline struct rq *
this_rq_lock_irq(struct rq_flags * rf)1746 this_rq_lock_irq(struct rq_flags *rf)
1747 __acquires(rq->lock)
1748 {
1749 struct rq *rq;
1750
1751 local_irq_disable();
1752 rq = this_rq();
1753 rq_lock(rq, rf);
1754 return rq;
1755 }
1756
1757 #ifdef CONFIG_NUMA
1758 enum numa_topology_type {
1759 NUMA_DIRECT,
1760 NUMA_GLUELESS_MESH,
1761 NUMA_BACKPLANE,
1762 };
1763 extern enum numa_topology_type sched_numa_topology_type;
1764 extern int sched_max_numa_distance;
1765 extern bool find_numa_distance(int distance);
1766 extern void sched_init_numa(int offline_node);
1767 extern void sched_update_numa(int cpu, bool online);
1768 extern void sched_domains_numa_masks_set(unsigned int cpu);
1769 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1770 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1771 #else
sched_init_numa(int offline_node)1772 static inline void sched_init_numa(int offline_node) { }
sched_update_numa(int cpu,bool online)1773 static inline void sched_update_numa(int cpu, bool online) { }
sched_domains_numa_masks_set(unsigned int cpu)1774 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1775 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1776 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1777 {
1778 return nr_cpu_ids;
1779 }
1780 #endif
1781
1782 #ifdef CONFIG_NUMA_BALANCING
1783 /* The regions in numa_faults array from task_struct */
1784 enum numa_faults_stats {
1785 NUMA_MEM = 0,
1786 NUMA_CPU,
1787 NUMA_MEMBUF,
1788 NUMA_CPUBUF
1789 };
1790 extern void sched_setnuma(struct task_struct *p, int node);
1791 extern int migrate_task_to(struct task_struct *p, int cpu);
1792 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1793 int cpu, int scpu);
1794 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1795 #else
1796 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1797 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1798 {
1799 }
1800 #endif /* CONFIG_NUMA_BALANCING */
1801
1802 #ifdef CONFIG_SMP
1803
1804 static inline void
queue_balance_callback(struct rq * rq,struct balance_callback * head,void (* func)(struct rq * rq))1805 queue_balance_callback(struct rq *rq,
1806 struct balance_callback *head,
1807 void (*func)(struct rq *rq))
1808 {
1809 lockdep_assert_rq_held(rq);
1810
1811 /*
1812 * Don't (re)queue an already queued item; nor queue anything when
1813 * balance_push() is active, see the comment with
1814 * balance_push_callback.
1815 */
1816 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1817 return;
1818
1819 head->func = func;
1820 head->next = rq->balance_callback;
1821 rq->balance_callback = head;
1822 }
1823
1824 #define rcu_dereference_check_sched_domain(p) \
1825 rcu_dereference_check((p), \
1826 lockdep_is_held(&sched_domains_mutex))
1827
1828 /*
1829 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1830 * See destroy_sched_domains: call_rcu for details.
1831 *
1832 * The domain tree of any CPU may only be accessed from within
1833 * preempt-disabled sections.
1834 */
1835 #define for_each_domain(cpu, __sd) \
1836 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1837 __sd; __sd = __sd->parent)
1838
1839 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1840 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1841 static const unsigned int SD_SHARED_CHILD_MASK =
1842 #include <linux/sched/sd_flags.h>
1843 0;
1844 #undef SD_FLAG
1845
1846 /**
1847 * highest_flag_domain - Return highest sched_domain containing flag.
1848 * @cpu: The CPU whose highest level of sched domain is to
1849 * be returned.
1850 * @flag: The flag to check for the highest sched_domain
1851 * for the given CPU.
1852 *
1853 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1854 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1855 */
highest_flag_domain(int cpu,int flag)1856 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1857 {
1858 struct sched_domain *sd, *hsd = NULL;
1859
1860 for_each_domain(cpu, sd) {
1861 if (sd->flags & flag) {
1862 hsd = sd;
1863 continue;
1864 }
1865
1866 /*
1867 * Stop the search if @flag is known to be shared at lower
1868 * levels. It will not be found further up.
1869 */
1870 if (flag & SD_SHARED_CHILD_MASK)
1871 break;
1872 }
1873
1874 return hsd;
1875 }
1876
lowest_flag_domain(int cpu,int flag)1877 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1878 {
1879 struct sched_domain *sd;
1880
1881 for_each_domain(cpu, sd) {
1882 if (sd->flags & flag)
1883 break;
1884 }
1885
1886 return sd;
1887 }
1888
1889 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1890 DECLARE_PER_CPU(int, sd_llc_size);
1891 DECLARE_PER_CPU(int, sd_llc_id);
1892 DECLARE_PER_CPU(int, sd_share_id);
1893 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1894 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1895 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1896 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1897 extern struct static_key_false sched_asym_cpucapacity;
1898 extern struct static_key_false sched_cluster_active;
1899
sched_asym_cpucap_active(void)1900 static __always_inline bool sched_asym_cpucap_active(void)
1901 {
1902 return static_branch_unlikely(&sched_asym_cpucapacity);
1903 }
1904
1905 struct sched_group_capacity {
1906 atomic_t ref;
1907 /*
1908 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1909 * for a single CPU.
1910 */
1911 unsigned long capacity;
1912 unsigned long min_capacity; /* Min per-CPU capacity in group */
1913 unsigned long max_capacity; /* Max per-CPU capacity in group */
1914 unsigned long next_update;
1915 int imbalance; /* XXX unrelated to capacity but shared group state */
1916
1917 #ifdef CONFIG_SCHED_DEBUG
1918 int id;
1919 #endif
1920
1921 unsigned long cpumask[]; /* Balance mask */
1922 };
1923
1924 struct sched_group {
1925 struct sched_group *next; /* Must be a circular list */
1926 atomic_t ref;
1927
1928 unsigned int group_weight;
1929 unsigned int cores;
1930 struct sched_group_capacity *sgc;
1931 int asym_prefer_cpu; /* CPU of highest priority in group */
1932 int flags;
1933
1934 /*
1935 * The CPUs this group covers.
1936 *
1937 * NOTE: this field is variable length. (Allocated dynamically
1938 * by attaching extra space to the end of the structure,
1939 * depending on how many CPUs the kernel has booted up with)
1940 */
1941 unsigned long cpumask[];
1942 };
1943
sched_group_span(struct sched_group * sg)1944 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1945 {
1946 return to_cpumask(sg->cpumask);
1947 }
1948
1949 /*
1950 * See build_balance_mask().
1951 */
group_balance_mask(struct sched_group * sg)1952 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1953 {
1954 return to_cpumask(sg->sgc->cpumask);
1955 }
1956
1957 extern int group_balance_cpu(struct sched_group *sg);
1958
1959 #ifdef CONFIG_SCHED_DEBUG
1960 void update_sched_domain_debugfs(void);
1961 void dirty_sched_domain_sysctl(int cpu);
1962 #else
update_sched_domain_debugfs(void)1963 static inline void update_sched_domain_debugfs(void)
1964 {
1965 }
dirty_sched_domain_sysctl(int cpu)1966 static inline void dirty_sched_domain_sysctl(int cpu)
1967 {
1968 }
1969 #endif
1970
1971 extern int sched_update_scaling(void);
1972
task_user_cpus(struct task_struct * p)1973 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1974 {
1975 if (!p->user_cpus_ptr)
1976 return cpu_possible_mask; /* &init_task.cpus_mask */
1977 return p->user_cpus_ptr;
1978 }
1979 #endif /* CONFIG_SMP */
1980
1981 #include "stats.h"
1982
1983 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1984
1985 extern void __sched_core_account_forceidle(struct rq *rq);
1986
sched_core_account_forceidle(struct rq * rq)1987 static inline void sched_core_account_forceidle(struct rq *rq)
1988 {
1989 if (schedstat_enabled())
1990 __sched_core_account_forceidle(rq);
1991 }
1992
1993 extern void __sched_core_tick(struct rq *rq);
1994
sched_core_tick(struct rq * rq)1995 static inline void sched_core_tick(struct rq *rq)
1996 {
1997 if (sched_core_enabled(rq) && schedstat_enabled())
1998 __sched_core_tick(rq);
1999 }
2000
2001 #else
2002
sched_core_account_forceidle(struct rq * rq)2003 static inline void sched_core_account_forceidle(struct rq *rq) {}
2004
sched_core_tick(struct rq * rq)2005 static inline void sched_core_tick(struct rq *rq) {}
2006
2007 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
2008
2009 #ifdef CONFIG_CGROUP_SCHED
2010
2011 /*
2012 * Return the group to which this tasks belongs.
2013 *
2014 * We cannot use task_css() and friends because the cgroup subsystem
2015 * changes that value before the cgroup_subsys::attach() method is called,
2016 * therefore we cannot pin it and might observe the wrong value.
2017 *
2018 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2019 * core changes this before calling sched_move_task().
2020 *
2021 * Instead we use a 'copy' which is updated from sched_move_task() while
2022 * holding both task_struct::pi_lock and rq::lock.
2023 */
task_group(struct task_struct * p)2024 static inline struct task_group *task_group(struct task_struct *p)
2025 {
2026 return p->sched_task_group;
2027 }
2028
2029 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
set_task_rq(struct task_struct * p,unsigned int cpu)2030 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2031 {
2032 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2033 struct task_group *tg = task_group(p);
2034 #endif
2035
2036 #ifdef CONFIG_FAIR_GROUP_SCHED
2037 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2038 p->se.cfs_rq = tg->cfs_rq[cpu];
2039 p->se.parent = tg->se[cpu];
2040 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2041 #endif
2042
2043 #ifdef CONFIG_RT_GROUP_SCHED
2044 p->rt.rt_rq = tg->rt_rq[cpu];
2045 p->rt.parent = tg->rt_se[cpu];
2046 #endif
2047 }
2048
2049 #else /* CONFIG_CGROUP_SCHED */
2050
set_task_rq(struct task_struct * p,unsigned int cpu)2051 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
task_group(struct task_struct * p)2052 static inline struct task_group *task_group(struct task_struct *p)
2053 {
2054 return NULL;
2055 }
2056
2057 #endif /* CONFIG_CGROUP_SCHED */
2058
__set_task_cpu(struct task_struct * p,unsigned int cpu)2059 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2060 {
2061 set_task_rq(p, cpu);
2062 #ifdef CONFIG_SMP
2063 /*
2064 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2065 * successfully executed on another CPU. We must ensure that updates of
2066 * per-task data have been completed by this moment.
2067 */
2068 smp_wmb();
2069 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2070 p->wake_cpu = cpu;
2071 #endif
2072 }
2073
2074 /*
2075 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2076 */
2077 #ifdef CONFIG_SCHED_DEBUG
2078 # define const_debug __read_mostly
2079 #else
2080 # define const_debug const
2081 #endif
2082
2083 #define SCHED_FEAT(name, enabled) \
2084 __SCHED_FEAT_##name ,
2085
2086 enum {
2087 #include "features.h"
2088 __SCHED_FEAT_NR,
2089 };
2090
2091 #undef SCHED_FEAT
2092
2093 #ifdef CONFIG_SCHED_DEBUG
2094
2095 /*
2096 * To support run-time toggling of sched features, all the translation units
2097 * (but core.c) reference the sysctl_sched_features defined in core.c.
2098 */
2099 extern const_debug unsigned int sysctl_sched_features;
2100
2101 #ifdef CONFIG_JUMP_LABEL
2102 #define SCHED_FEAT(name, enabled) \
2103 static __always_inline bool static_branch_##name(struct static_key *key) \
2104 { \
2105 return static_key_##enabled(key); \
2106 }
2107
2108 #include "features.h"
2109 #undef SCHED_FEAT
2110
2111 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2112 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2113
2114 #else /* !CONFIG_JUMP_LABEL */
2115
2116 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2117
2118 #endif /* CONFIG_JUMP_LABEL */
2119
2120 #else /* !SCHED_DEBUG */
2121
2122 /*
2123 * Each translation unit has its own copy of sysctl_sched_features to allow
2124 * constants propagation at compile time and compiler optimization based on
2125 * features default.
2126 */
2127 #define SCHED_FEAT(name, enabled) \
2128 (1UL << __SCHED_FEAT_##name) * enabled |
2129 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2130 #include "features.h"
2131 0;
2132 #undef SCHED_FEAT
2133
2134 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2135
2136 #endif /* SCHED_DEBUG */
2137
2138 extern struct static_key_false sched_numa_balancing;
2139 extern struct static_key_false sched_schedstats;
2140
global_rt_period(void)2141 static inline u64 global_rt_period(void)
2142 {
2143 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2144 }
2145
global_rt_runtime(void)2146 static inline u64 global_rt_runtime(void)
2147 {
2148 if (sysctl_sched_rt_runtime < 0)
2149 return RUNTIME_INF;
2150
2151 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2152 }
2153
task_current(struct rq * rq,struct task_struct * p)2154 static inline int task_current(struct rq *rq, struct task_struct *p)
2155 {
2156 return rq->curr == p;
2157 }
2158
task_on_cpu(struct rq * rq,struct task_struct * p)2159 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2160 {
2161 #ifdef CONFIG_SMP
2162 return p->on_cpu;
2163 #else
2164 return task_current(rq, p);
2165 #endif
2166 }
2167
task_on_rq_queued(struct task_struct * p)2168 static inline int task_on_rq_queued(struct task_struct *p)
2169 {
2170 return p->on_rq == TASK_ON_RQ_QUEUED;
2171 }
2172
task_on_rq_migrating(struct task_struct * p)2173 static inline int task_on_rq_migrating(struct task_struct *p)
2174 {
2175 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2176 }
2177
2178 /* Wake flags. The first three directly map to some SD flag value */
2179 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2180 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2181 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2182
2183 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2184 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2185 #define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2186
2187 #ifdef CONFIG_SMP
2188 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2189 static_assert(WF_FORK == SD_BALANCE_FORK);
2190 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2191 #endif
2192
2193 /*
2194 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2195 * of tasks with abnormal "nice" values across CPUs the contribution that
2196 * each task makes to its run queue's load is weighted according to its
2197 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2198 * scaled version of the new time slice allocation that they receive on time
2199 * slice expiry etc.
2200 */
2201
2202 #define WEIGHT_IDLEPRIO 3
2203 #define WMULT_IDLEPRIO 1431655765
2204
2205 extern const int sched_prio_to_weight[40];
2206 extern const u32 sched_prio_to_wmult[40];
2207
2208 /*
2209 * {de,en}queue flags:
2210 *
2211 * DEQUEUE_SLEEP - task is no longer runnable
2212 * ENQUEUE_WAKEUP - task just became runnable
2213 *
2214 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2215 * are in a known state which allows modification. Such pairs
2216 * should preserve as much state as possible.
2217 *
2218 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2219 * in the runqueue.
2220 *
2221 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2222 *
2223 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2224 *
2225 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2226 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2227 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2228 *
2229 */
2230
2231 #define DEQUEUE_SLEEP 0x01
2232 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2233 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2234 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2235 #define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2236
2237 #define ENQUEUE_WAKEUP 0x01
2238 #define ENQUEUE_RESTORE 0x02
2239 #define ENQUEUE_MOVE 0x04
2240 #define ENQUEUE_NOCLOCK 0x08
2241
2242 #define ENQUEUE_HEAD 0x10
2243 #define ENQUEUE_REPLENISH 0x20
2244 #ifdef CONFIG_SMP
2245 #define ENQUEUE_MIGRATED 0x40
2246 #else
2247 #define ENQUEUE_MIGRATED 0x00
2248 #endif
2249 #define ENQUEUE_INITIAL 0x80
2250 #define ENQUEUE_MIGRATING 0x100
2251
2252 #define RETRY_TASK ((void *)-1UL)
2253
2254 struct affinity_context {
2255 const struct cpumask *new_mask;
2256 struct cpumask *user_mask;
2257 unsigned int flags;
2258 };
2259
2260 extern s64 update_curr_common(struct rq *rq);
2261
2262 struct sched_class {
2263
2264 #ifdef CONFIG_UCLAMP_TASK
2265 int uclamp_enabled;
2266 #endif
2267
2268 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2269 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2270 void (*yield_task) (struct rq *rq);
2271 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2272
2273 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2274
2275 struct task_struct *(*pick_next_task)(struct rq *rq);
2276
2277 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2278 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2279
2280 #ifdef CONFIG_SMP
2281 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2282 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2283
2284 struct task_struct * (*pick_task)(struct rq *rq);
2285
2286 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2287
2288 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2289
2290 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2291
2292 void (*rq_online)(struct rq *rq);
2293 void (*rq_offline)(struct rq *rq);
2294
2295 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2296 #endif
2297
2298 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2299 void (*task_fork)(struct task_struct *p);
2300 void (*task_dead)(struct task_struct *p);
2301
2302 /*
2303 * The switched_from() call is allowed to drop rq->lock, therefore we
2304 * cannot assume the switched_from/switched_to pair is serialized by
2305 * rq->lock. They are however serialized by p->pi_lock.
2306 */
2307 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2308 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2309 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2310 int oldprio);
2311
2312 unsigned int (*get_rr_interval)(struct rq *rq,
2313 struct task_struct *task);
2314
2315 void (*update_curr)(struct rq *rq);
2316
2317 #ifdef CONFIG_FAIR_GROUP_SCHED
2318 void (*task_change_group)(struct task_struct *p);
2319 #endif
2320
2321 #ifdef CONFIG_SCHED_CORE
2322 int (*task_is_throttled)(struct task_struct *p, int cpu);
2323 #endif
2324 };
2325
put_prev_task(struct rq * rq,struct task_struct * prev)2326 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2327 {
2328 WARN_ON_ONCE(rq->curr != prev);
2329 prev->sched_class->put_prev_task(rq, prev);
2330 }
2331
set_next_task(struct rq * rq,struct task_struct * next)2332 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2333 {
2334 next->sched_class->set_next_task(rq, next, false);
2335 }
2336
2337
2338 /*
2339 * Helper to define a sched_class instance; each one is placed in a separate
2340 * section which is ordered by the linker script:
2341 *
2342 * include/asm-generic/vmlinux.lds.h
2343 *
2344 * *CAREFUL* they are laid out in *REVERSE* order!!!
2345 *
2346 * Also enforce alignment on the instance, not the type, to guarantee layout.
2347 */
2348 #define DEFINE_SCHED_CLASS(name) \
2349 const struct sched_class name##_sched_class \
2350 __aligned(__alignof__(struct sched_class)) \
2351 __section("__" #name "_sched_class")
2352
2353 /* Defined in include/asm-generic/vmlinux.lds.h */
2354 extern struct sched_class __sched_class_highest[];
2355 extern struct sched_class __sched_class_lowest[];
2356
2357 #define for_class_range(class, _from, _to) \
2358 for (class = (_from); class < (_to); class++)
2359
2360 #define for_each_class(class) \
2361 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2362
2363 #define sched_class_above(_a, _b) ((_a) < (_b))
2364
2365 extern const struct sched_class stop_sched_class;
2366 extern const struct sched_class dl_sched_class;
2367 extern const struct sched_class rt_sched_class;
2368 extern const struct sched_class fair_sched_class;
2369 extern const struct sched_class idle_sched_class;
2370
sched_stop_runnable(struct rq * rq)2371 static inline bool sched_stop_runnable(struct rq *rq)
2372 {
2373 return rq->stop && task_on_rq_queued(rq->stop);
2374 }
2375
sched_dl_runnable(struct rq * rq)2376 static inline bool sched_dl_runnable(struct rq *rq)
2377 {
2378 return rq->dl.dl_nr_running > 0;
2379 }
2380
sched_rt_runnable(struct rq * rq)2381 static inline bool sched_rt_runnable(struct rq *rq)
2382 {
2383 return rq->rt.rt_queued > 0;
2384 }
2385
sched_fair_runnable(struct rq * rq)2386 static inline bool sched_fair_runnable(struct rq *rq)
2387 {
2388 return rq->cfs.nr_running > 0;
2389 }
2390
2391 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2392 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2393
2394 #define SCA_CHECK 0x01
2395 #define SCA_MIGRATE_DISABLE 0x02
2396 #define SCA_MIGRATE_ENABLE 0x04
2397 #define SCA_USER 0x08
2398
2399 #ifdef CONFIG_SMP
2400
2401 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2402
2403 extern void sched_balance_trigger(struct rq *rq);
2404
2405 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2406
get_push_task(struct rq * rq)2407 static inline struct task_struct *get_push_task(struct rq *rq)
2408 {
2409 struct task_struct *p = rq->curr;
2410
2411 lockdep_assert_rq_held(rq);
2412
2413 if (rq->push_busy)
2414 return NULL;
2415
2416 if (p->nr_cpus_allowed == 1)
2417 return NULL;
2418
2419 if (p->migration_disabled)
2420 return NULL;
2421
2422 rq->push_busy = true;
2423 return get_task_struct(p);
2424 }
2425
2426 extern int push_cpu_stop(void *arg);
2427
2428 #endif
2429
2430 #ifdef CONFIG_CPU_IDLE
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2431 static inline void idle_set_state(struct rq *rq,
2432 struct cpuidle_state *idle_state)
2433 {
2434 rq->idle_state = idle_state;
2435 }
2436
idle_get_state(struct rq * rq)2437 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2438 {
2439 SCHED_WARN_ON(!rcu_read_lock_held());
2440
2441 return rq->idle_state;
2442 }
2443 #else
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2444 static inline void idle_set_state(struct rq *rq,
2445 struct cpuidle_state *idle_state)
2446 {
2447 }
2448
idle_get_state(struct rq * rq)2449 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2450 {
2451 return NULL;
2452 }
2453 #endif
2454
2455 extern void schedule_idle(void);
2456 asmlinkage void schedule_user(void);
2457
2458 extern void sysrq_sched_debug_show(void);
2459 extern void sched_init_granularity(void);
2460 extern void update_max_interval(void);
2461
2462 extern void init_sched_dl_class(void);
2463 extern void init_sched_rt_class(void);
2464 extern void init_sched_fair_class(void);
2465
2466 extern void reweight_task(struct task_struct *p, int prio);
2467
2468 extern void resched_curr(struct rq *rq);
2469 extern void resched_cpu(int cpu);
2470
2471 extern struct rt_bandwidth def_rt_bandwidth;
2472 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2473 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2474
2475 extern void init_dl_entity(struct sched_dl_entity *dl_se);
2476
2477 #define BW_SHIFT 20
2478 #define BW_UNIT (1 << BW_SHIFT)
2479 #define RATIO_SHIFT 8
2480 #define MAX_BW_BITS (64 - BW_SHIFT)
2481 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2482 unsigned long to_ratio(u64 period, u64 runtime);
2483
2484 extern void init_entity_runnable_average(struct sched_entity *se);
2485 extern void post_init_entity_util_avg(struct task_struct *p);
2486
2487 #ifdef CONFIG_NO_HZ_FULL
2488 extern bool sched_can_stop_tick(struct rq *rq);
2489 extern int __init sched_tick_offload_init(void);
2490
2491 /*
2492 * Tick may be needed by tasks in the runqueue depending on their policy and
2493 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2494 * nohz mode if necessary.
2495 */
sched_update_tick_dependency(struct rq * rq)2496 static inline void sched_update_tick_dependency(struct rq *rq)
2497 {
2498 int cpu = cpu_of(rq);
2499
2500 if (!tick_nohz_full_cpu(cpu))
2501 return;
2502
2503 if (sched_can_stop_tick(rq))
2504 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2505 else
2506 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2507 }
2508 #else
sched_tick_offload_init(void)2509 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2510 static inline void sched_update_tick_dependency(struct rq *rq) { }
2511 #endif
2512
add_nr_running(struct rq * rq,unsigned count)2513 static inline void add_nr_running(struct rq *rq, unsigned count)
2514 {
2515 unsigned prev_nr = rq->nr_running;
2516
2517 rq->nr_running = prev_nr + count;
2518 if (trace_sched_update_nr_running_tp_enabled()) {
2519 call_trace_sched_update_nr_running(rq, count);
2520 }
2521
2522 #ifdef CONFIG_SMP
2523 if (prev_nr < 2 && rq->nr_running >= 2)
2524 set_rd_overloaded(rq->rd, 1);
2525 #endif
2526
2527 sched_update_tick_dependency(rq);
2528 }
2529
sub_nr_running(struct rq * rq,unsigned count)2530 static inline void sub_nr_running(struct rq *rq, unsigned count)
2531 {
2532 rq->nr_running -= count;
2533 if (trace_sched_update_nr_running_tp_enabled()) {
2534 call_trace_sched_update_nr_running(rq, -count);
2535 }
2536
2537 /* Check if we still need preemption */
2538 sched_update_tick_dependency(rq);
2539 }
2540
2541 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2542 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2543
2544 extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2545
2546 #ifdef CONFIG_PREEMPT_RT
2547 #define SCHED_NR_MIGRATE_BREAK 8
2548 #else
2549 #define SCHED_NR_MIGRATE_BREAK 32
2550 #endif
2551
2552 extern const_debug unsigned int sysctl_sched_nr_migrate;
2553 extern const_debug unsigned int sysctl_sched_migration_cost;
2554
2555 extern unsigned int sysctl_sched_base_slice;
2556
2557 #ifdef CONFIG_SCHED_DEBUG
2558 extern int sysctl_resched_latency_warn_ms;
2559 extern int sysctl_resched_latency_warn_once;
2560
2561 extern unsigned int sysctl_sched_tunable_scaling;
2562
2563 extern unsigned int sysctl_numa_balancing_scan_delay;
2564 extern unsigned int sysctl_numa_balancing_scan_period_min;
2565 extern unsigned int sysctl_numa_balancing_scan_period_max;
2566 extern unsigned int sysctl_numa_balancing_scan_size;
2567 extern unsigned int sysctl_numa_balancing_hot_threshold;
2568 #endif
2569
2570 #ifdef CONFIG_SCHED_HRTICK
2571
2572 /*
2573 * Use hrtick when:
2574 * - enabled by features
2575 * - hrtimer is actually high res
2576 */
hrtick_enabled(struct rq * rq)2577 static inline int hrtick_enabled(struct rq *rq)
2578 {
2579 if (!cpu_active(cpu_of(rq)))
2580 return 0;
2581 return hrtimer_is_hres_active(&rq->hrtick_timer);
2582 }
2583
hrtick_enabled_fair(struct rq * rq)2584 static inline int hrtick_enabled_fair(struct rq *rq)
2585 {
2586 if (!sched_feat(HRTICK))
2587 return 0;
2588 return hrtick_enabled(rq);
2589 }
2590
hrtick_enabled_dl(struct rq * rq)2591 static inline int hrtick_enabled_dl(struct rq *rq)
2592 {
2593 if (!sched_feat(HRTICK_DL))
2594 return 0;
2595 return hrtick_enabled(rq);
2596 }
2597
2598 void hrtick_start(struct rq *rq, u64 delay);
2599
2600 #else
2601
hrtick_enabled_fair(struct rq * rq)2602 static inline int hrtick_enabled_fair(struct rq *rq)
2603 {
2604 return 0;
2605 }
2606
hrtick_enabled_dl(struct rq * rq)2607 static inline int hrtick_enabled_dl(struct rq *rq)
2608 {
2609 return 0;
2610 }
2611
hrtick_enabled(struct rq * rq)2612 static inline int hrtick_enabled(struct rq *rq)
2613 {
2614 return 0;
2615 }
2616
2617 #endif /* CONFIG_SCHED_HRTICK */
2618
2619 #ifndef arch_scale_freq_tick
2620 static __always_inline
arch_scale_freq_tick(void)2621 void arch_scale_freq_tick(void)
2622 {
2623 }
2624 #endif
2625
2626 #ifndef arch_scale_freq_capacity
2627 /**
2628 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2629 * @cpu: the CPU in question.
2630 *
2631 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2632 *
2633 * f_curr
2634 * ------ * SCHED_CAPACITY_SCALE
2635 * f_max
2636 */
2637 static __always_inline
arch_scale_freq_capacity(int cpu)2638 unsigned long arch_scale_freq_capacity(int cpu)
2639 {
2640 return SCHED_CAPACITY_SCALE;
2641 }
2642 #endif
2643
2644 #ifdef CONFIG_SCHED_DEBUG
2645 /*
2646 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2647 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2648 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2649 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2650 */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2651 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2652 {
2653 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2654 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2655 #ifdef CONFIG_SMP
2656 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2657 #endif
2658 }
2659 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2660 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2661 #endif
2662
2663 #define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2664 __DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2665 static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2666 { class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2667 _lock; return _t; }
2668
2669 #ifdef CONFIG_SMP
2670
rq_order_less(struct rq * rq1,struct rq * rq2)2671 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2672 {
2673 #ifdef CONFIG_SCHED_CORE
2674 /*
2675 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2676 * order by core-id first and cpu-id second.
2677 *
2678 * Notably:
2679 *
2680 * double_rq_lock(0,3); will take core-0, core-1 lock
2681 * double_rq_lock(1,2); will take core-1, core-0 lock
2682 *
2683 * when only cpu-id is considered.
2684 */
2685 if (rq1->core->cpu < rq2->core->cpu)
2686 return true;
2687 if (rq1->core->cpu > rq2->core->cpu)
2688 return false;
2689
2690 /*
2691 * __sched_core_flip() relies on SMT having cpu-id lock order.
2692 */
2693 #endif
2694 return rq1->cpu < rq2->cpu;
2695 }
2696
2697 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2698
2699 #ifdef CONFIG_PREEMPTION
2700
2701 /*
2702 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2703 * way at the expense of forcing extra atomic operations in all
2704 * invocations. This assures that the double_lock is acquired using the
2705 * same underlying policy as the spinlock_t on this architecture, which
2706 * reduces latency compared to the unfair variant below. However, it
2707 * also adds more overhead and therefore may reduce throughput.
2708 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2709 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2710 __releases(this_rq->lock)
2711 __acquires(busiest->lock)
2712 __acquires(this_rq->lock)
2713 {
2714 raw_spin_rq_unlock(this_rq);
2715 double_rq_lock(this_rq, busiest);
2716
2717 return 1;
2718 }
2719
2720 #else
2721 /*
2722 * Unfair double_lock_balance: Optimizes throughput at the expense of
2723 * latency by eliminating extra atomic operations when the locks are
2724 * already in proper order on entry. This favors lower CPU-ids and will
2725 * grant the double lock to lower CPUs over higher ids under contention,
2726 * regardless of entry order into the function.
2727 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2728 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2729 __releases(this_rq->lock)
2730 __acquires(busiest->lock)
2731 __acquires(this_rq->lock)
2732 {
2733 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2734 likely(raw_spin_rq_trylock(busiest))) {
2735 double_rq_clock_clear_update(this_rq, busiest);
2736 return 0;
2737 }
2738
2739 if (rq_order_less(this_rq, busiest)) {
2740 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2741 double_rq_clock_clear_update(this_rq, busiest);
2742 return 0;
2743 }
2744
2745 raw_spin_rq_unlock(this_rq);
2746 double_rq_lock(this_rq, busiest);
2747
2748 return 1;
2749 }
2750
2751 #endif /* CONFIG_PREEMPTION */
2752
2753 /*
2754 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2755 */
double_lock_balance(struct rq * this_rq,struct rq * busiest)2756 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2757 {
2758 lockdep_assert_irqs_disabled();
2759
2760 return _double_lock_balance(this_rq, busiest);
2761 }
2762
double_unlock_balance(struct rq * this_rq,struct rq * busiest)2763 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2764 __releases(busiest->lock)
2765 {
2766 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2767 raw_spin_rq_unlock(busiest);
2768 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2769 }
2770
double_lock(spinlock_t * l1,spinlock_t * l2)2771 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2772 {
2773 if (l1 > l2)
2774 swap(l1, l2);
2775
2776 spin_lock(l1);
2777 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2778 }
2779
double_lock_irq(spinlock_t * l1,spinlock_t * l2)2780 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2781 {
2782 if (l1 > l2)
2783 swap(l1, l2);
2784
2785 spin_lock_irq(l1);
2786 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2787 }
2788
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)2789 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2790 {
2791 if (l1 > l2)
2792 swap(l1, l2);
2793
2794 raw_spin_lock(l1);
2795 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2796 }
2797
double_raw_unlock(raw_spinlock_t * l1,raw_spinlock_t * l2)2798 static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2799 {
2800 raw_spin_unlock(l1);
2801 raw_spin_unlock(l2);
2802 }
2803
2804 DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
2805 double_raw_lock(_T->lock, _T->lock2),
2806 double_raw_unlock(_T->lock, _T->lock2))
2807
2808 /*
2809 * double_rq_unlock - safely unlock two runqueues
2810 *
2811 * Note this does not restore interrupts like task_rq_unlock,
2812 * you need to do so manually after calling.
2813 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2814 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2815 __releases(rq1->lock)
2816 __releases(rq2->lock)
2817 {
2818 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2819 raw_spin_rq_unlock(rq2);
2820 else
2821 __release(rq2->lock);
2822 raw_spin_rq_unlock(rq1);
2823 }
2824
2825 extern void set_rq_online (struct rq *rq);
2826 extern void set_rq_offline(struct rq *rq);
2827 extern bool sched_smp_initialized;
2828
2829 #else /* CONFIG_SMP */
2830
2831 /*
2832 * double_rq_lock - safely lock two runqueues
2833 *
2834 * Note this does not disable interrupts like task_rq_lock,
2835 * you need to do so manually before calling.
2836 */
double_rq_lock(struct rq * rq1,struct rq * rq2)2837 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2838 __acquires(rq1->lock)
2839 __acquires(rq2->lock)
2840 {
2841 WARN_ON_ONCE(!irqs_disabled());
2842 WARN_ON_ONCE(rq1 != rq2);
2843 raw_spin_rq_lock(rq1);
2844 __acquire(rq2->lock); /* Fake it out ;) */
2845 double_rq_clock_clear_update(rq1, rq2);
2846 }
2847
2848 /*
2849 * double_rq_unlock - safely unlock two runqueues
2850 *
2851 * Note this does not restore interrupts like task_rq_unlock,
2852 * you need to do so manually after calling.
2853 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)2854 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2855 __releases(rq1->lock)
2856 __releases(rq2->lock)
2857 {
2858 WARN_ON_ONCE(rq1 != rq2);
2859 raw_spin_rq_unlock(rq1);
2860 __release(rq2->lock);
2861 }
2862
2863 #endif
2864
2865 DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
2866 double_rq_lock(_T->lock, _T->lock2),
2867 double_rq_unlock(_T->lock, _T->lock2))
2868
2869 extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
2870 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2871 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2872
2873 #ifdef CONFIG_SCHED_DEBUG
2874 extern bool sched_debug_verbose;
2875
2876 extern void print_cfs_stats(struct seq_file *m, int cpu);
2877 extern void print_rt_stats(struct seq_file *m, int cpu);
2878 extern void print_dl_stats(struct seq_file *m, int cpu);
2879 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2880 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2881 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2882
2883 extern void resched_latency_warn(int cpu, u64 latency);
2884 #ifdef CONFIG_NUMA_BALANCING
2885 extern void
2886 show_numa_stats(struct task_struct *p, struct seq_file *m);
2887 extern void
2888 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2889 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2890 #endif /* CONFIG_NUMA_BALANCING */
2891 #else
resched_latency_warn(int cpu,u64 latency)2892 static inline void resched_latency_warn(int cpu, u64 latency) {}
2893 #endif /* CONFIG_SCHED_DEBUG */
2894
2895 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2896 extern void init_rt_rq(struct rt_rq *rt_rq);
2897 extern void init_dl_rq(struct dl_rq *dl_rq);
2898
2899 extern void cfs_bandwidth_usage_inc(void);
2900 extern void cfs_bandwidth_usage_dec(void);
2901
2902 #ifdef CONFIG_NO_HZ_COMMON
2903 #define NOHZ_BALANCE_KICK_BIT 0
2904 #define NOHZ_STATS_KICK_BIT 1
2905 #define NOHZ_NEWILB_KICK_BIT 2
2906 #define NOHZ_NEXT_KICK_BIT 3
2907
2908 /* Run sched_balance_domains() */
2909 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2910 /* Update blocked load */
2911 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2912 /* Update blocked load when entering idle */
2913 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2914 /* Update nohz.next_balance */
2915 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2916
2917 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2918
2919 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2920
2921 extern void nohz_balance_exit_idle(struct rq *rq);
2922 #else
nohz_balance_exit_idle(struct rq * rq)2923 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2924 #endif
2925
2926 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2927 extern void nohz_run_idle_balance(int cpu);
2928 #else
nohz_run_idle_balance(int cpu)2929 static inline void nohz_run_idle_balance(int cpu) { }
2930 #endif
2931
2932 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2933 struct irqtime {
2934 u64 total;
2935 u64 tick_delta;
2936 u64 irq_start_time;
2937 struct u64_stats_sync sync;
2938 };
2939
2940 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2941
2942 /*
2943 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2944 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2945 * and never move forward.
2946 */
irq_time_read(int cpu)2947 static inline u64 irq_time_read(int cpu)
2948 {
2949 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2950 unsigned int seq;
2951 u64 total;
2952
2953 do {
2954 seq = __u64_stats_fetch_begin(&irqtime->sync);
2955 total = irqtime->total;
2956 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2957
2958 return total;
2959 }
2960 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2961
2962 #ifdef CONFIG_CPU_FREQ
2963 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2964
2965 /**
2966 * cpufreq_update_util - Take a note about CPU utilization changes.
2967 * @rq: Runqueue to carry out the update for.
2968 * @flags: Update reason flags.
2969 *
2970 * This function is called by the scheduler on the CPU whose utilization is
2971 * being updated.
2972 *
2973 * It can only be called from RCU-sched read-side critical sections.
2974 *
2975 * The way cpufreq is currently arranged requires it to evaluate the CPU
2976 * performance state (frequency/voltage) on a regular basis to prevent it from
2977 * being stuck in a completely inadequate performance level for too long.
2978 * That is not guaranteed to happen if the updates are only triggered from CFS
2979 * and DL, though, because they may not be coming in if only RT tasks are
2980 * active all the time (or there are RT tasks only).
2981 *
2982 * As a workaround for that issue, this function is called periodically by the
2983 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2984 * but that really is a band-aid. Going forward it should be replaced with
2985 * solutions targeted more specifically at RT tasks.
2986 */
cpufreq_update_util(struct rq * rq,unsigned int flags)2987 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2988 {
2989 struct update_util_data *data;
2990
2991 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2992 cpu_of(rq)));
2993 if (data)
2994 data->func(data, rq_clock(rq), flags);
2995 }
2996 #else
cpufreq_update_util(struct rq * rq,unsigned int flags)2997 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2998 #endif /* CONFIG_CPU_FREQ */
2999
3000 #ifdef arch_scale_freq_capacity
3001 # ifndef arch_scale_freq_invariant
3002 # define arch_scale_freq_invariant() true
3003 # endif
3004 #else
3005 # define arch_scale_freq_invariant() false
3006 #endif
3007
3008 #ifdef CONFIG_SMP
3009 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3010 unsigned long *min,
3011 unsigned long *max);
3012
3013 unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3014 unsigned long min,
3015 unsigned long max);
3016
3017
3018 /*
3019 * Verify the fitness of task @p to run on @cpu taking into account the
3020 * CPU original capacity and the runtime/deadline ratio of the task.
3021 *
3022 * The function will return true if the original capacity of @cpu is
3023 * greater than or equal to task's deadline density right shifted by
3024 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3025 */
dl_task_fits_capacity(struct task_struct * p,int cpu)3026 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3027 {
3028 unsigned long cap = arch_scale_cpu_capacity(cpu);
3029
3030 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3031 }
3032
cpu_bw_dl(struct rq * rq)3033 static inline unsigned long cpu_bw_dl(struct rq *rq)
3034 {
3035 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3036 }
3037
cpu_util_dl(struct rq * rq)3038 static inline unsigned long cpu_util_dl(struct rq *rq)
3039 {
3040 return READ_ONCE(rq->avg_dl.util_avg);
3041 }
3042
3043
3044 extern unsigned long cpu_util_cfs(int cpu);
3045 extern unsigned long cpu_util_cfs_boost(int cpu);
3046
cpu_util_rt(struct rq * rq)3047 static inline unsigned long cpu_util_rt(struct rq *rq)
3048 {
3049 return READ_ONCE(rq->avg_rt.util_avg);
3050 }
3051 #endif
3052
3053 #ifdef CONFIG_UCLAMP_TASK
3054 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3055
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3056 static inline unsigned long uclamp_rq_get(struct rq *rq,
3057 enum uclamp_id clamp_id)
3058 {
3059 return READ_ONCE(rq->uclamp[clamp_id].value);
3060 }
3061
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3062 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3063 unsigned int value)
3064 {
3065 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3066 }
3067
uclamp_rq_is_idle(struct rq * rq)3068 static inline bool uclamp_rq_is_idle(struct rq *rq)
3069 {
3070 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3071 }
3072
3073 /* Is the rq being capped/throttled by uclamp_max? */
uclamp_rq_is_capped(struct rq * rq)3074 static inline bool uclamp_rq_is_capped(struct rq *rq)
3075 {
3076 unsigned long rq_util;
3077 unsigned long max_util;
3078
3079 if (!static_branch_likely(&sched_uclamp_used))
3080 return false;
3081
3082 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3083 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3084
3085 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3086 }
3087
3088 /*
3089 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3090 * by default in the fast path and only gets turned on once userspace performs
3091 * an operation that requires it.
3092 *
3093 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3094 * hence is active.
3095 */
uclamp_is_used(void)3096 static inline bool uclamp_is_used(void)
3097 {
3098 return static_branch_likely(&sched_uclamp_used);
3099 }
3100 #else /* CONFIG_UCLAMP_TASK */
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)3101 static inline unsigned long uclamp_eff_value(struct task_struct *p,
3102 enum uclamp_id clamp_id)
3103 {
3104 if (clamp_id == UCLAMP_MIN)
3105 return 0;
3106
3107 return SCHED_CAPACITY_SCALE;
3108 }
3109
uclamp_rq_is_capped(struct rq * rq)3110 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3111
uclamp_is_used(void)3112 static inline bool uclamp_is_used(void)
3113 {
3114 return false;
3115 }
3116
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3117 static inline unsigned long uclamp_rq_get(struct rq *rq,
3118 enum uclamp_id clamp_id)
3119 {
3120 if (clamp_id == UCLAMP_MIN)
3121 return 0;
3122
3123 return SCHED_CAPACITY_SCALE;
3124 }
3125
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3126 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3127 unsigned int value)
3128 {
3129 }
3130
uclamp_rq_is_idle(struct rq * rq)3131 static inline bool uclamp_rq_is_idle(struct rq *rq)
3132 {
3133 return false;
3134 }
3135 #endif /* CONFIG_UCLAMP_TASK */
3136
3137 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
cpu_util_irq(struct rq * rq)3138 static inline unsigned long cpu_util_irq(struct rq *rq)
3139 {
3140 return READ_ONCE(rq->avg_irq.util_avg);
3141 }
3142
3143 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3144 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3145 {
3146 util *= (max - irq);
3147 util /= max;
3148
3149 return util;
3150
3151 }
3152 #else
cpu_util_irq(struct rq * rq)3153 static inline unsigned long cpu_util_irq(struct rq *rq)
3154 {
3155 return 0;
3156 }
3157
3158 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3159 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3160 {
3161 return util;
3162 }
3163 #endif
3164
3165 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3166
3167 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3168
3169 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3170
sched_energy_enabled(void)3171 static inline bool sched_energy_enabled(void)
3172 {
3173 return static_branch_unlikely(&sched_energy_present);
3174 }
3175
3176 extern struct cpufreq_governor schedutil_gov;
3177
3178 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3179
3180 #define perf_domain_span(pd) NULL
sched_energy_enabled(void)3181 static inline bool sched_energy_enabled(void) { return false; }
3182
3183 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3184
3185 #ifdef CONFIG_MEMBARRIER
3186 /*
3187 * The scheduler provides memory barriers required by membarrier between:
3188 * - prior user-space memory accesses and store to rq->membarrier_state,
3189 * - store to rq->membarrier_state and following user-space memory accesses.
3190 * In the same way it provides those guarantees around store to rq->curr.
3191 */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3192 static inline void membarrier_switch_mm(struct rq *rq,
3193 struct mm_struct *prev_mm,
3194 struct mm_struct *next_mm)
3195 {
3196 int membarrier_state;
3197
3198 if (prev_mm == next_mm)
3199 return;
3200
3201 membarrier_state = atomic_read(&next_mm->membarrier_state);
3202 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3203 return;
3204
3205 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3206 }
3207 #else
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3208 static inline void membarrier_switch_mm(struct rq *rq,
3209 struct mm_struct *prev_mm,
3210 struct mm_struct *next_mm)
3211 {
3212 }
3213 #endif
3214
3215 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3216 static inline bool is_per_cpu_kthread(struct task_struct *p)
3217 {
3218 if (!(p->flags & PF_KTHREAD))
3219 return false;
3220
3221 if (p->nr_cpus_allowed != 1)
3222 return false;
3223
3224 return true;
3225 }
3226 #endif
3227
3228 extern void swake_up_all_locked(struct swait_queue_head *q);
3229 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3230
3231 extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3232
3233 #ifdef CONFIG_PREEMPT_DYNAMIC
3234 extern int preempt_dynamic_mode;
3235 extern int sched_dynamic_mode(const char *str);
3236 extern void sched_dynamic_update(int mode);
3237 #endif
3238
3239 #ifdef CONFIG_SCHED_MM_CID
3240
3241 #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3242 #define MM_CID_SCAN_DELAY 100 /* 100ms */
3243
3244 extern raw_spinlock_t cid_lock;
3245 extern int use_cid_lock;
3246
3247 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3248 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3249 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3250 extern void init_sched_mm_cid(struct task_struct *t);
3251
__mm_cid_put(struct mm_struct * mm,int cid)3252 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3253 {
3254 if (cid < 0)
3255 return;
3256 cpumask_clear_cpu(cid, mm_cidmask(mm));
3257 }
3258
3259 /*
3260 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3261 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3262 * be held to transition to other states.
3263 *
3264 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3265 * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3266 */
mm_cid_put_lazy(struct task_struct * t)3267 static inline void mm_cid_put_lazy(struct task_struct *t)
3268 {
3269 struct mm_struct *mm = t->mm;
3270 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3271 int cid;
3272
3273 lockdep_assert_irqs_disabled();
3274 cid = __this_cpu_read(pcpu_cid->cid);
3275 if (!mm_cid_is_lazy_put(cid) ||
3276 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3277 return;
3278 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3279 }
3280
mm_cid_pcpu_unset(struct mm_struct * mm)3281 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3282 {
3283 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3284 int cid, res;
3285
3286 lockdep_assert_irqs_disabled();
3287 cid = __this_cpu_read(pcpu_cid->cid);
3288 for (;;) {
3289 if (mm_cid_is_unset(cid))
3290 return MM_CID_UNSET;
3291 /*
3292 * Attempt transition from valid or lazy-put to unset.
3293 */
3294 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3295 if (res == cid)
3296 break;
3297 cid = res;
3298 }
3299 return cid;
3300 }
3301
mm_cid_put(struct mm_struct * mm)3302 static inline void mm_cid_put(struct mm_struct *mm)
3303 {
3304 int cid;
3305
3306 lockdep_assert_irqs_disabled();
3307 cid = mm_cid_pcpu_unset(mm);
3308 if (cid == MM_CID_UNSET)
3309 return;
3310 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3311 }
3312
__mm_cid_try_get(struct mm_struct * mm)3313 static inline int __mm_cid_try_get(struct mm_struct *mm)
3314 {
3315 struct cpumask *cpumask;
3316 int cid;
3317
3318 cpumask = mm_cidmask(mm);
3319 /*
3320 * Retry finding first zero bit if the mask is temporarily
3321 * filled. This only happens during concurrent remote-clear
3322 * which owns a cid without holding a rq lock.
3323 */
3324 for (;;) {
3325 cid = cpumask_first_zero(cpumask);
3326 if (cid < nr_cpu_ids)
3327 break;
3328 cpu_relax();
3329 }
3330 if (cpumask_test_and_set_cpu(cid, cpumask))
3331 return -1;
3332 return cid;
3333 }
3334
3335 /*
3336 * Save a snapshot of the current runqueue time of this cpu
3337 * with the per-cpu cid value, allowing to estimate how recently it was used.
3338 */
mm_cid_snapshot_time(struct rq * rq,struct mm_struct * mm)3339 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3340 {
3341 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3342
3343 lockdep_assert_rq_held(rq);
3344 WRITE_ONCE(pcpu_cid->time, rq->clock);
3345 }
3346
__mm_cid_get(struct rq * rq,struct mm_struct * mm)3347 static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3348 {
3349 int cid;
3350
3351 /*
3352 * All allocations (even those using the cid_lock) are lock-free. If
3353 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3354 * guarantee forward progress.
3355 */
3356 if (!READ_ONCE(use_cid_lock)) {
3357 cid = __mm_cid_try_get(mm);
3358 if (cid >= 0)
3359 goto end;
3360 raw_spin_lock(&cid_lock);
3361 } else {
3362 raw_spin_lock(&cid_lock);
3363 cid = __mm_cid_try_get(mm);
3364 if (cid >= 0)
3365 goto unlock;
3366 }
3367
3368 /*
3369 * cid concurrently allocated. Retry while forcing following
3370 * allocations to use the cid_lock to ensure forward progress.
3371 */
3372 WRITE_ONCE(use_cid_lock, 1);
3373 /*
3374 * Set use_cid_lock before allocation. Only care about program order
3375 * because this is only required for forward progress.
3376 */
3377 barrier();
3378 /*
3379 * Retry until it succeeds. It is guaranteed to eventually succeed once
3380 * all newcoming allocations observe the use_cid_lock flag set.
3381 */
3382 do {
3383 cid = __mm_cid_try_get(mm);
3384 cpu_relax();
3385 } while (cid < 0);
3386 /*
3387 * Allocate before clearing use_cid_lock. Only care about
3388 * program order because this is for forward progress.
3389 */
3390 barrier();
3391 WRITE_ONCE(use_cid_lock, 0);
3392 unlock:
3393 raw_spin_unlock(&cid_lock);
3394 end:
3395 mm_cid_snapshot_time(rq, mm);
3396 return cid;
3397 }
3398
mm_cid_get(struct rq * rq,struct mm_struct * mm)3399 static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3400 {
3401 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3402 struct cpumask *cpumask;
3403 int cid;
3404
3405 lockdep_assert_rq_held(rq);
3406 cpumask = mm_cidmask(mm);
3407 cid = __this_cpu_read(pcpu_cid->cid);
3408 if (mm_cid_is_valid(cid)) {
3409 mm_cid_snapshot_time(rq, mm);
3410 return cid;
3411 }
3412 if (mm_cid_is_lazy_put(cid)) {
3413 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3414 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3415 }
3416 cid = __mm_cid_get(rq, mm);
3417 __this_cpu_write(pcpu_cid->cid, cid);
3418 return cid;
3419 }
3420
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3421 static inline void switch_mm_cid(struct rq *rq,
3422 struct task_struct *prev,
3423 struct task_struct *next)
3424 {
3425 /*
3426 * Provide a memory barrier between rq->curr store and load of
3427 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3428 *
3429 * Should be adapted if context_switch() is modified.
3430 */
3431 if (!next->mm) { // to kernel
3432 /*
3433 * user -> kernel transition does not guarantee a barrier, but
3434 * we can use the fact that it performs an atomic operation in
3435 * mmgrab().
3436 */
3437 if (prev->mm) // from user
3438 smp_mb__after_mmgrab();
3439 /*
3440 * kernel -> kernel transition does not change rq->curr->mm
3441 * state. It stays NULL.
3442 */
3443 } else { // to user
3444 /*
3445 * kernel -> user transition does not provide a barrier
3446 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3447 * Provide it here.
3448 */
3449 if (!prev->mm) { // from kernel
3450 smp_mb();
3451 } else { // from user
3452 /*
3453 * user->user transition relies on an implicit
3454 * memory barrier in switch_mm() when
3455 * current->mm changes. If the architecture
3456 * switch_mm() does not have an implicit memory
3457 * barrier, it is emitted here. If current->mm
3458 * is unchanged, no barrier is needed.
3459 */
3460 smp_mb__after_switch_mm();
3461 }
3462 }
3463 if (prev->mm_cid_active) {
3464 mm_cid_snapshot_time(rq, prev->mm);
3465 mm_cid_put_lazy(prev);
3466 prev->mm_cid = -1;
3467 }
3468 if (next->mm_cid_active)
3469 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3470 }
3471
3472 #else
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3473 static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
sched_mm_cid_migrate_from(struct task_struct * t)3474 static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
sched_mm_cid_migrate_to(struct rq * dst_rq,struct task_struct * t)3475 static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
task_tick_mm_cid(struct rq * rq,struct task_struct * curr)3476 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
init_sched_mm_cid(struct task_struct * t)3477 static inline void init_sched_mm_cid(struct task_struct *t) { }
3478 #endif
3479
3480 extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3481 extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3482
3483 #endif /* _KERNEL_SCHED_SCHED_H */
3484