Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : /*
3 : * Scheduler internal types and methods:
4 : */
5 : #ifndef _KERNEL_SCHED_SCHED_H
6 : #define _KERNEL_SCHED_SCHED_H
7 :
8 : #include <linux/sched/affinity.h>
9 : #include <linux/sched/autogroup.h>
10 : #include <linux/sched/cpufreq.h>
11 : #include <linux/sched/deadline.h>
12 : #include <linux/sched.h>
13 : #include <linux/sched/loadavg.h>
14 : #include <linux/sched/mm.h>
15 : #include <linux/sched/rseq_api.h>
16 : #include <linux/sched/signal.h>
17 : #include <linux/sched/smt.h>
18 : #include <linux/sched/stat.h>
19 : #include <linux/sched/sysctl.h>
20 : #include <linux/sched/task_flags.h>
21 : #include <linux/sched/task.h>
22 : #include <linux/sched/topology.h>
23 :
24 : #include <linux/atomic.h>
25 : #include <linux/bitmap.h>
26 : #include <linux/bug.h>
27 : #include <linux/capability.h>
28 : #include <linux/cgroup_api.h>
29 : #include <linux/cgroup.h>
30 : #include <linux/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_CGROUP_SCHED
78 : #include <linux/cgroup.h>
79 : #include <linux/psi.h>
80 : #endif
81 :
82 : #ifdef CONFIG_SCHED_DEBUG
83 : # include <linux/static_key.h>
84 : #endif
85 :
86 : #ifdef CONFIG_PARAVIRT
87 : # include <asm/paravirt.h>
88 : # include <asm/paravirt_api_clock.h>
89 : #endif
90 :
91 : #include "cpupri.h"
92 : #include "cpudeadline.h"
93 :
94 : #ifdef CONFIG_SCHED_DEBUG
95 : # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
96 : #else
97 : # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
98 : #endif
99 :
100 : struct rq;
101 : struct cpuidle_state;
102 :
103 : /* task_struct::on_rq states: */
104 : #define TASK_ON_RQ_QUEUED 1
105 : #define TASK_ON_RQ_MIGRATING 2
106 :
107 : extern __read_mostly int scheduler_running;
108 :
109 : extern unsigned long calc_load_update;
110 : extern atomic_long_t calc_load_tasks;
111 :
112 : extern unsigned int sysctl_sched_child_runs_first;
113 :
114 : extern void calc_global_load_tick(struct rq *this_rq);
115 : extern long calc_load_fold_active(struct rq *this_rq, long adjust);
116 :
117 : extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
118 :
119 : extern unsigned int sysctl_sched_rt_period;
120 : extern int sysctl_sched_rt_runtime;
121 : extern int sched_rr_timeslice;
122 :
123 : /*
124 : * Helpers for converting nanosecond timing to jiffy resolution
125 : */
126 : #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
127 :
128 : /*
129 : * Increase resolution of nice-level calculations for 64-bit architectures.
130 : * The extra resolution improves shares distribution and load balancing of
131 : * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
132 : * hierarchies, especially on larger systems. This is not a user-visible change
133 : * and does not change the user-interface for setting shares/weights.
134 : *
135 : * We increase resolution only if we have enough bits to allow this increased
136 : * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
137 : * are pretty high and the returns do not justify the increased costs.
138 : *
139 : * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
140 : * increase coverage and consistency always enable it on 64-bit platforms.
141 : */
142 : #ifdef CONFIG_64BIT
143 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
144 : # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
145 : # define scale_load_down(w) \
146 : ({ \
147 : unsigned long __w = (w); \
148 : if (__w) \
149 : __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
150 : __w; \
151 : })
152 : #else
153 : # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
154 : # define scale_load(w) (w)
155 : # define scale_load_down(w) (w)
156 : #endif
157 :
158 : /*
159 : * Task weight (visible to users) and its load (invisible to users) have
160 : * independent resolution, but they should be well calibrated. We use
161 : * scale_load() and scale_load_down(w) to convert between them. The
162 : * following must be true:
163 : *
164 : * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
165 : *
166 : */
167 : #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
168 :
169 : /*
170 : * Single value that decides SCHED_DEADLINE internal math precision.
171 : * 10 -> just above 1us
172 : * 9 -> just above 0.5us
173 : */
174 : #define DL_SCALE 10
175 :
176 : /*
177 : * Single value that denotes runtime == period, ie unlimited time.
178 : */
179 : #define RUNTIME_INF ((u64)~0ULL)
180 :
181 : static inline int idle_policy(int policy)
182 : {
183 6917 : return policy == SCHED_IDLE;
184 : }
185 : static inline int fair_policy(int policy)
186 : {
187 760 : return policy == SCHED_NORMAL || policy == SCHED_BATCH;
188 : }
189 :
190 : static inline int rt_policy(int policy)
191 : {
192 390 : return policy == SCHED_FIFO || policy == SCHED_RR;
193 : }
194 :
195 : static inline int dl_policy(int policy)
196 : {
197 : return policy == SCHED_DEADLINE;
198 : }
199 : static inline bool valid_policy(int policy)
200 : {
201 760 : return idle_policy(policy) || fair_policy(policy) ||
202 380 : rt_policy(policy) || dl_policy(policy);
203 : }
204 :
205 : static inline int task_has_idle_policy(struct task_struct *p)
206 : {
207 13840 : return idle_policy(p->policy);
208 : }
209 :
210 : static inline int task_has_rt_policy(struct task_struct *p)
211 : {
212 10 : return rt_policy(p->policy);
213 : }
214 :
215 : static inline int task_has_dl_policy(struct task_struct *p)
216 : {
217 5 : return dl_policy(p->policy);
218 : }
219 :
220 : #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
221 :
222 : static inline void update_avg(u64 *avg, u64 sample)
223 : {
224 : s64 diff = sample - *avg;
225 : *avg += diff / 8;
226 : }
227 :
228 : /*
229 : * Shifting a value by an exponent greater *or equal* to the size of said value
230 : * is UB; cap at size-1.
231 : */
232 : #define shr_bound(val, shift) \
233 : (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
234 :
235 : /*
236 : * !! For sched_setattr_nocheck() (kernel) only !!
237 : *
238 : * This is actually gross. :(
239 : *
240 : * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
241 : * tasks, but still be able to sleep. We need this on platforms that cannot
242 : * atomically change clock frequency. Remove once fast switching will be
243 : * available on such platforms.
244 : *
245 : * SUGOV stands for SchedUtil GOVernor.
246 : */
247 : #define SCHED_FLAG_SUGOV 0x10000000
248 :
249 : #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
250 :
251 : static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
252 : {
253 : #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
254 : return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
255 : #else
256 : return false;
257 : #endif
258 : }
259 :
260 : /*
261 : * Tells if entity @a should preempt entity @b.
262 : */
263 : static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
264 : const struct sched_dl_entity *b)
265 : {
266 0 : return dl_entity_is_special(a) ||
267 0 : dl_time_before(a->deadline, b->deadline);
268 : }
269 :
270 : /*
271 : * This is the priority-queue data structure of the RT scheduling class:
272 : */
273 : struct rt_prio_array {
274 : DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
275 : struct list_head queue[MAX_RT_PRIO];
276 : };
277 :
278 : struct rt_bandwidth {
279 : /* nests inside the rq lock: */
280 : raw_spinlock_t rt_runtime_lock;
281 : ktime_t rt_period;
282 : u64 rt_runtime;
283 : struct hrtimer rt_period_timer;
284 : unsigned int rt_period_active;
285 : };
286 :
287 : void __dl_clear_params(struct task_struct *p);
288 :
289 : struct dl_bandwidth {
290 : raw_spinlock_t dl_runtime_lock;
291 : u64 dl_runtime;
292 : u64 dl_period;
293 : };
294 :
295 : static inline int dl_bandwidth_enabled(void)
296 : {
297 : return sysctl_sched_rt_runtime >= 0;
298 : }
299 :
300 : /*
301 : * To keep the bandwidth of -deadline tasks under control
302 : * we need some place where:
303 : * - store the maximum -deadline bandwidth of each cpu;
304 : * - cache the fraction of bandwidth that is currently allocated in
305 : * each root domain;
306 : *
307 : * This is all done in the data structure below. It is similar to the
308 : * one used for RT-throttling (rt_bandwidth), with the main difference
309 : * that, since here we are only interested in admission control, we
310 : * do not decrease any runtime while the group "executes", neither we
311 : * need a timer to replenish it.
312 : *
313 : * With respect to SMP, bandwidth is given on a per root domain basis,
314 : * meaning that:
315 : * - bw (< 100%) is the deadline bandwidth of each CPU;
316 : * - total_bw is the currently allocated bandwidth in each root domain;
317 : */
318 : struct dl_bw {
319 : raw_spinlock_t lock;
320 : u64 bw;
321 : u64 total_bw;
322 : };
323 :
324 : extern void init_dl_bw(struct dl_bw *dl_b);
325 : extern int sched_dl_global_validate(void);
326 : extern void sched_dl_do_global(void);
327 : extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
328 : extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
329 : extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
330 : extern bool __checkparam_dl(const struct sched_attr *attr);
331 : extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
332 : extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
333 : extern int dl_cpu_busy(int cpu, struct task_struct *p);
334 :
335 : #ifdef CONFIG_CGROUP_SCHED
336 :
337 : struct cfs_rq;
338 : struct rt_rq;
339 :
340 : extern struct list_head task_groups;
341 :
342 : struct cfs_bandwidth {
343 : #ifdef CONFIG_CFS_BANDWIDTH
344 : raw_spinlock_t lock;
345 : ktime_t period;
346 : u64 quota;
347 : u64 runtime;
348 : u64 burst;
349 : u64 runtime_snap;
350 : s64 hierarchical_quota;
351 :
352 : u8 idle;
353 : u8 period_active;
354 : u8 slack_started;
355 : struct hrtimer period_timer;
356 : struct hrtimer slack_timer;
357 : struct list_head throttled_cfs_rq;
358 :
359 : /* Statistics: */
360 : int nr_periods;
361 : int nr_throttled;
362 : int nr_burst;
363 : u64 throttled_time;
364 : u64 burst_time;
365 : #endif
366 : };
367 :
368 : /* Task group related information */
369 : struct task_group {
370 : struct cgroup_subsys_state css;
371 :
372 : #ifdef CONFIG_FAIR_GROUP_SCHED
373 : /* schedulable entities of this group on each CPU */
374 : struct sched_entity **se;
375 : /* runqueue "owned" by this group on each CPU */
376 : struct cfs_rq **cfs_rq;
377 : unsigned long shares;
378 :
379 : /* A positive value indicates that this is a SCHED_IDLE group. */
380 : int idle;
381 :
382 : #ifdef CONFIG_SMP
383 : /*
384 : * load_avg can be heavily contended at clock tick time, so put
385 : * it in its own cacheline separated from the fields above which
386 : * will also be accessed at each tick.
387 : */
388 : atomic_long_t load_avg ____cacheline_aligned;
389 : #endif
390 : #endif
391 :
392 : #ifdef CONFIG_RT_GROUP_SCHED
393 : struct sched_rt_entity **rt_se;
394 : struct rt_rq **rt_rq;
395 :
396 : struct rt_bandwidth rt_bandwidth;
397 : #endif
398 :
399 : struct rcu_head rcu;
400 : struct list_head list;
401 :
402 : struct task_group *parent;
403 : struct list_head siblings;
404 : struct list_head children;
405 :
406 : #ifdef CONFIG_SCHED_AUTOGROUP
407 : struct autogroup *autogroup;
408 : #endif
409 :
410 : struct cfs_bandwidth cfs_bandwidth;
411 :
412 : #ifdef CONFIG_UCLAMP_TASK_GROUP
413 : /* The two decimal precision [%] value requested from user-space */
414 : unsigned int uclamp_pct[UCLAMP_CNT];
415 : /* Clamp values requested for a task group */
416 : struct uclamp_se uclamp_req[UCLAMP_CNT];
417 : /* Effective clamp values used for a task group */
418 : struct uclamp_se uclamp[UCLAMP_CNT];
419 : #endif
420 :
421 : };
422 :
423 : #ifdef CONFIG_FAIR_GROUP_SCHED
424 : #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
425 :
426 : /*
427 : * A weight of 0 or 1 can cause arithmetics problems.
428 : * A weight of a cfs_rq is the sum of weights of which entities
429 : * are queued on this cfs_rq, so a weight of a entity should not be
430 : * too large, so as the shares value of a task group.
431 : * (The default weight is 1024 - so there's no practical
432 : * limitation from this.)
433 : */
434 : #define MIN_SHARES (1UL << 1)
435 : #define MAX_SHARES (1UL << 18)
436 : #endif
437 :
438 : typedef int (*tg_visitor)(struct task_group *, void *);
439 :
440 : extern int walk_tg_tree_from(struct task_group *from,
441 : tg_visitor down, tg_visitor up, void *data);
442 :
443 : /*
444 : * Iterate the full tree, calling @down when first entering a node and @up when
445 : * leaving it for the final time.
446 : *
447 : * Caller must hold rcu_lock or sufficient equivalent.
448 : */
449 : static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
450 : {
451 : return walk_tg_tree_from(&root_task_group, down, up, data);
452 : }
453 :
454 : extern int tg_nop(struct task_group *tg, void *data);
455 :
456 : extern void free_fair_sched_group(struct task_group *tg);
457 : extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
458 : extern void online_fair_sched_group(struct task_group *tg);
459 : extern void unregister_fair_sched_group(struct task_group *tg);
460 : extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
461 : struct sched_entity *se, int cpu,
462 : struct sched_entity *parent);
463 : extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
464 :
465 : extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
466 : extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
467 : extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
468 :
469 : extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
470 : struct sched_rt_entity *rt_se, int cpu,
471 : struct sched_rt_entity *parent);
472 : extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
473 : extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
474 : extern long sched_group_rt_runtime(struct task_group *tg);
475 : extern long sched_group_rt_period(struct task_group *tg);
476 : extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
477 :
478 : extern struct task_group *sched_create_group(struct task_group *parent);
479 : extern void sched_online_group(struct task_group *tg,
480 : struct task_group *parent);
481 : extern void sched_destroy_group(struct task_group *tg);
482 : extern void sched_release_group(struct task_group *tg);
483 :
484 : extern void sched_move_task(struct task_struct *tsk);
485 :
486 : #ifdef CONFIG_FAIR_GROUP_SCHED
487 : extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
488 :
489 : extern int sched_group_set_idle(struct task_group *tg, long idle);
490 :
491 : #ifdef CONFIG_SMP
492 : extern void set_task_rq_fair(struct sched_entity *se,
493 : struct cfs_rq *prev, struct cfs_rq *next);
494 : #else /* !CONFIG_SMP */
495 : static inline void set_task_rq_fair(struct sched_entity *se,
496 : struct cfs_rq *prev, struct cfs_rq *next) { }
497 : #endif /* CONFIG_SMP */
498 : #endif /* CONFIG_FAIR_GROUP_SCHED */
499 :
500 : #else /* CONFIG_CGROUP_SCHED */
501 :
502 : struct cfs_bandwidth { };
503 :
504 : #endif /* CONFIG_CGROUP_SCHED */
505 :
506 : extern void unregister_rt_sched_group(struct task_group *tg);
507 : extern void free_rt_sched_group(struct task_group *tg);
508 : extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
509 :
510 : /*
511 : * u64_u32_load/u64_u32_store
512 : *
513 : * Use a copy of a u64 value to protect against data race. This is only
514 : * applicable for 32-bits architectures.
515 : */
516 : #ifdef CONFIG_64BIT
517 : # define u64_u32_load_copy(var, copy) var
518 : # define u64_u32_store_copy(var, copy, val) (var = val)
519 : #else
520 : # define u64_u32_load_copy(var, copy) \
521 : ({ \
522 : u64 __val, __val_copy; \
523 : do { \
524 : __val_copy = copy; \
525 : /* \
526 : * paired with u64_u32_store_copy(), ordering access \
527 : * to var and copy. \
528 : */ \
529 : smp_rmb(); \
530 : __val = var; \
531 : } while (__val != __val_copy); \
532 : __val; \
533 : })
534 : # define u64_u32_store_copy(var, copy, val) \
535 : do { \
536 : typeof(val) __val = (val); \
537 : var = __val; \
538 : /* \
539 : * paired with u64_u32_load_copy(), ordering access to var and \
540 : * copy. \
541 : */ \
542 : smp_wmb(); \
543 : copy = __val; \
544 : } while (0)
545 : #endif
546 : # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
547 : # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
548 :
549 : /* CFS-related fields in a runqueue */
550 : struct cfs_rq {
551 : struct load_weight load;
552 : unsigned int nr_running;
553 : unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
554 : unsigned int idle_nr_running; /* SCHED_IDLE */
555 : unsigned int idle_h_nr_running; /* SCHED_IDLE */
556 :
557 : u64 exec_clock;
558 : u64 min_vruntime;
559 : #ifdef CONFIG_SCHED_CORE
560 : unsigned int forceidle_seq;
561 : u64 min_vruntime_fi;
562 : #endif
563 :
564 : #ifndef CONFIG_64BIT
565 : u64 min_vruntime_copy;
566 : #endif
567 :
568 : struct rb_root_cached tasks_timeline;
569 :
570 : /*
571 : * 'curr' points to currently running entity on this cfs_rq.
572 : * It is set to NULL otherwise (i.e when none are currently running).
573 : */
574 : struct sched_entity *curr;
575 : struct sched_entity *next;
576 : struct sched_entity *last;
577 : struct sched_entity *skip;
578 :
579 : #ifdef CONFIG_SCHED_DEBUG
580 : unsigned int nr_spread_over;
581 : #endif
582 :
583 : #ifdef CONFIG_SMP
584 : /*
585 : * CFS load tracking
586 : */
587 : struct sched_avg avg;
588 : #ifndef CONFIG_64BIT
589 : u64 last_update_time_copy;
590 : #endif
591 : struct {
592 : raw_spinlock_t lock ____cacheline_aligned;
593 : int nr;
594 : unsigned long load_avg;
595 : unsigned long util_avg;
596 : unsigned long runnable_avg;
597 : } removed;
598 :
599 : #ifdef CONFIG_FAIR_GROUP_SCHED
600 : unsigned long tg_load_avg_contrib;
601 : long propagate;
602 : long prop_runnable_sum;
603 :
604 : /*
605 : * h_load = weight * f(tg)
606 : *
607 : * Where f(tg) is the recursive weight fraction assigned to
608 : * this group.
609 : */
610 : unsigned long h_load;
611 : u64 last_h_load_update;
612 : struct sched_entity *h_load_next;
613 : #endif /* CONFIG_FAIR_GROUP_SCHED */
614 : #endif /* CONFIG_SMP */
615 :
616 : #ifdef CONFIG_FAIR_GROUP_SCHED
617 : struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
618 :
619 : /*
620 : * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
621 : * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
622 : * (like users, containers etc.)
623 : *
624 : * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
625 : * This list is used during load balance.
626 : */
627 : int on_list;
628 : struct list_head leaf_cfs_rq_list;
629 : struct task_group *tg; /* group that "owns" this runqueue */
630 :
631 : /* Locally cached copy of our task_group's idle value */
632 : int idle;
633 :
634 : #ifdef CONFIG_CFS_BANDWIDTH
635 : int runtime_enabled;
636 : s64 runtime_remaining;
637 :
638 : u64 throttled_pelt_idle;
639 : #ifndef CONFIG_64BIT
640 : u64 throttled_pelt_idle_copy;
641 : #endif
642 : u64 throttled_clock;
643 : u64 throttled_clock_pelt;
644 : u64 throttled_clock_pelt_time;
645 : int throttled;
646 : int throttle_count;
647 : struct list_head throttled_list;
648 : #ifdef CONFIG_SMP
649 : struct list_head throttled_csd_list;
650 : #endif
651 : #endif /* CONFIG_CFS_BANDWIDTH */
652 : #endif /* CONFIG_FAIR_GROUP_SCHED */
653 : };
654 :
655 : static inline int rt_bandwidth_enabled(void)
656 : {
657 0 : return sysctl_sched_rt_runtime >= 0;
658 : }
659 :
660 : /* RT IPI pull logic requires IRQ_WORK */
661 : #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
662 : # define HAVE_RT_PUSH_IPI
663 : #endif
664 :
665 : /* Real-Time classes' related field in a runqueue: */
666 : struct rt_rq {
667 : struct rt_prio_array active;
668 : unsigned int rt_nr_running;
669 : unsigned int rr_nr_running;
670 : #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
671 : struct {
672 : int curr; /* highest queued rt task prio */
673 : #ifdef CONFIG_SMP
674 : int next; /* next highest */
675 : #endif
676 : } highest_prio;
677 : #endif
678 : #ifdef CONFIG_SMP
679 : unsigned int rt_nr_migratory;
680 : unsigned int rt_nr_total;
681 : int overloaded;
682 : struct plist_head pushable_tasks;
683 :
684 : #endif /* CONFIG_SMP */
685 : int rt_queued;
686 :
687 : int rt_throttled;
688 : u64 rt_time;
689 : u64 rt_runtime;
690 : /* Nests inside the rq lock: */
691 : raw_spinlock_t rt_runtime_lock;
692 :
693 : #ifdef CONFIG_RT_GROUP_SCHED
694 : unsigned int rt_nr_boosted;
695 :
696 : struct rq *rq;
697 : struct task_group *tg;
698 : #endif
699 : };
700 :
701 : static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
702 : {
703 : return rt_rq->rt_queued && rt_rq->rt_nr_running;
704 : }
705 :
706 : /* Deadline class' related fields in a runqueue */
707 : struct dl_rq {
708 : /* runqueue is an rbtree, ordered by deadline */
709 : struct rb_root_cached root;
710 :
711 : unsigned int dl_nr_running;
712 :
713 : #ifdef CONFIG_SMP
714 : /*
715 : * Deadline values of the currently executing and the
716 : * earliest ready task on this rq. Caching these facilitates
717 : * the decision whether or not a ready but not running task
718 : * should migrate somewhere else.
719 : */
720 : struct {
721 : u64 curr;
722 : u64 next;
723 : } earliest_dl;
724 :
725 : unsigned int dl_nr_migratory;
726 : int overloaded;
727 :
728 : /*
729 : * Tasks on this rq that can be pushed away. They are kept in
730 : * an rb-tree, ordered by tasks' deadlines, with caching
731 : * of the leftmost (earliest deadline) element.
732 : */
733 : struct rb_root_cached pushable_dl_tasks_root;
734 : #else
735 : struct dl_bw dl_bw;
736 : #endif
737 : /*
738 : * "Active utilization" for this runqueue: increased when a
739 : * task wakes up (becomes TASK_RUNNING) and decreased when a
740 : * task blocks
741 : */
742 : u64 running_bw;
743 :
744 : /*
745 : * Utilization of the tasks "assigned" to this runqueue (including
746 : * the tasks that are in runqueue and the tasks that executed on this
747 : * CPU and blocked). Increased when a task moves to this runqueue, and
748 : * decreased when the task moves away (migrates, changes scheduling
749 : * policy, or terminates).
750 : * This is needed to compute the "inactive utilization" for the
751 : * runqueue (inactive utilization = this_bw - running_bw).
752 : */
753 : u64 this_bw;
754 : u64 extra_bw;
755 :
756 : /*
757 : * Inverse of the fraction of CPU utilization that can be reclaimed
758 : * by the GRUB algorithm.
759 : */
760 : u64 bw_ratio;
761 : };
762 :
763 : #ifdef CONFIG_FAIR_GROUP_SCHED
764 : /* An entity is a task if it doesn't "own" a runqueue */
765 : #define entity_is_task(se) (!se->my_q)
766 :
767 : static inline void se_update_runnable(struct sched_entity *se)
768 : {
769 : if (!entity_is_task(se))
770 : se->runnable_weight = se->my_q->h_nr_running;
771 : }
772 :
773 : static inline long se_runnable(struct sched_entity *se)
774 : {
775 : if (entity_is_task(se))
776 : return !!se->on_rq;
777 : else
778 : return se->runnable_weight;
779 : }
780 :
781 : #else
782 : #define entity_is_task(se) 1
783 :
784 : static inline void se_update_runnable(struct sched_entity *se) {}
785 :
786 : static inline long se_runnable(struct sched_entity *se)
787 : {
788 : return !!se->on_rq;
789 : }
790 : #endif
791 :
792 : #ifdef CONFIG_SMP
793 : /*
794 : * XXX we want to get rid of these helpers and use the full load resolution.
795 : */
796 : static inline long se_weight(struct sched_entity *se)
797 : {
798 : return scale_load_down(se->load.weight);
799 : }
800 :
801 :
802 : static inline bool sched_asym_prefer(int a, int b)
803 : {
804 : return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
805 : }
806 :
807 : struct perf_domain {
808 : struct em_perf_domain *em_pd;
809 : struct perf_domain *next;
810 : struct rcu_head rcu;
811 : };
812 :
813 : /* Scheduling group status flags */
814 : #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
815 : #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
816 :
817 : /*
818 : * We add the notion of a root-domain which will be used to define per-domain
819 : * variables. Each exclusive cpuset essentially defines an island domain by
820 : * fully partitioning the member CPUs from any other cpuset. Whenever a new
821 : * exclusive cpuset is created, we also create and attach a new root-domain
822 : * object.
823 : *
824 : */
825 : struct root_domain {
826 : atomic_t refcount;
827 : atomic_t rto_count;
828 : struct rcu_head rcu;
829 : cpumask_var_t span;
830 : cpumask_var_t online;
831 :
832 : /*
833 : * Indicate pullable load on at least one CPU, e.g:
834 : * - More than one runnable task
835 : * - Running task is misfit
836 : */
837 : int overload;
838 :
839 : /* Indicate one or more cpus over-utilized (tipping point) */
840 : int overutilized;
841 :
842 : /*
843 : * The bit corresponding to a CPU gets set here if such CPU has more
844 : * than one runnable -deadline task (as it is below for RT tasks).
845 : */
846 : cpumask_var_t dlo_mask;
847 : atomic_t dlo_count;
848 : struct dl_bw dl_bw;
849 : struct cpudl cpudl;
850 :
851 : /*
852 : * Indicate whether a root_domain's dl_bw has been checked or
853 : * updated. It's monotonously increasing value.
854 : *
855 : * Also, some corner cases, like 'wrap around' is dangerous, but given
856 : * that u64 is 'big enough'. So that shouldn't be a concern.
857 : */
858 : u64 visit_gen;
859 :
860 : #ifdef HAVE_RT_PUSH_IPI
861 : /*
862 : * For IPI pull requests, loop across the rto_mask.
863 : */
864 : struct irq_work rto_push_work;
865 : raw_spinlock_t rto_lock;
866 : /* These are only updated and read within rto_lock */
867 : int rto_loop;
868 : int rto_cpu;
869 : /* These atomics are updated outside of a lock */
870 : atomic_t rto_loop_next;
871 : atomic_t rto_loop_start;
872 : #endif
873 : /*
874 : * The "RT overload" flag: it gets set if a CPU has more than
875 : * one runnable RT task.
876 : */
877 : cpumask_var_t rto_mask;
878 : struct cpupri cpupri;
879 :
880 : unsigned long max_cpu_capacity;
881 :
882 : /*
883 : * NULL-terminated list of performance domains intersecting with the
884 : * CPUs of the rd. Protected by RCU.
885 : */
886 : struct perf_domain __rcu *pd;
887 : };
888 :
889 : extern void init_defrootdomain(void);
890 : extern int sched_init_domains(const struct cpumask *cpu_map);
891 : extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
892 : extern void sched_get_rd(struct root_domain *rd);
893 : extern void sched_put_rd(struct root_domain *rd);
894 :
895 : #ifdef HAVE_RT_PUSH_IPI
896 : extern void rto_push_irq_work_func(struct irq_work *work);
897 : #endif
898 : #endif /* CONFIG_SMP */
899 :
900 : #ifdef CONFIG_UCLAMP_TASK
901 : /*
902 : * struct uclamp_bucket - Utilization clamp bucket
903 : * @value: utilization clamp value for tasks on this clamp bucket
904 : * @tasks: number of RUNNABLE tasks on this clamp bucket
905 : *
906 : * Keep track of how many tasks are RUNNABLE for a given utilization
907 : * clamp value.
908 : */
909 : struct uclamp_bucket {
910 : unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
911 : unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
912 : };
913 :
914 : /*
915 : * struct uclamp_rq - rq's utilization clamp
916 : * @value: currently active clamp values for a rq
917 : * @bucket: utilization clamp buckets affecting a rq
918 : *
919 : * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
920 : * A clamp value is affecting a rq when there is at least one task RUNNABLE
921 : * (or actually running) with that value.
922 : *
923 : * There are up to UCLAMP_CNT possible different clamp values, currently there
924 : * are only two: minimum utilization and maximum utilization.
925 : *
926 : * All utilization clamping values are MAX aggregated, since:
927 : * - for util_min: we want to run the CPU at least at the max of the minimum
928 : * utilization required by its currently RUNNABLE tasks.
929 : * - for util_max: we want to allow the CPU to run up to the max of the
930 : * maximum utilization allowed by its currently RUNNABLE tasks.
931 : *
932 : * Since on each system we expect only a limited number of different
933 : * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
934 : * the metrics required to compute all the per-rq utilization clamp values.
935 : */
936 : struct uclamp_rq {
937 : unsigned int value;
938 : struct uclamp_bucket bucket[UCLAMP_BUCKETS];
939 : };
940 :
941 : DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
942 : #endif /* CONFIG_UCLAMP_TASK */
943 :
944 : struct rq;
945 : struct balance_callback {
946 : struct balance_callback *next;
947 : void (*func)(struct rq *rq);
948 : };
949 :
950 : /*
951 : * This is the main, per-CPU runqueue data structure.
952 : *
953 : * Locking rule: those places that want to lock multiple runqueues
954 : * (such as the load balancing or the thread migration code), lock
955 : * acquire operations must be ordered by ascending &runqueue.
956 : */
957 : struct rq {
958 : /* runqueue lock: */
959 : raw_spinlock_t __lock;
960 :
961 : /*
962 : * nr_running and cpu_load should be in the same cacheline because
963 : * remote CPUs use both these fields when doing load calculation.
964 : */
965 : unsigned int nr_running;
966 : #ifdef CONFIG_NUMA_BALANCING
967 : unsigned int nr_numa_running;
968 : unsigned int nr_preferred_running;
969 : unsigned int numa_migrate_on;
970 : #endif
971 : #ifdef CONFIG_NO_HZ_COMMON
972 : #ifdef CONFIG_SMP
973 : unsigned long last_blocked_load_update_tick;
974 : unsigned int has_blocked_load;
975 : call_single_data_t nohz_csd;
976 : #endif /* CONFIG_SMP */
977 : unsigned int nohz_tick_stopped;
978 : atomic_t nohz_flags;
979 : #endif /* CONFIG_NO_HZ_COMMON */
980 :
981 : #ifdef CONFIG_SMP
982 : unsigned int ttwu_pending;
983 : #endif
984 : u64 nr_switches;
985 :
986 : #ifdef CONFIG_UCLAMP_TASK
987 : /* Utilization clamp values based on CPU's RUNNABLE tasks */
988 : struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
989 : unsigned int uclamp_flags;
990 : #define UCLAMP_FLAG_IDLE 0x01
991 : #endif
992 :
993 : struct cfs_rq cfs;
994 : struct rt_rq rt;
995 : struct dl_rq dl;
996 :
997 : #ifdef CONFIG_FAIR_GROUP_SCHED
998 : /* list of leaf cfs_rq on this CPU: */
999 : struct list_head leaf_cfs_rq_list;
1000 : struct list_head *tmp_alone_branch;
1001 : #endif /* CONFIG_FAIR_GROUP_SCHED */
1002 :
1003 : /*
1004 : * This is part of a global counter where only the total sum
1005 : * over all CPUs matters. A task can increase this counter on
1006 : * one CPU and if it got migrated afterwards it may decrease
1007 : * it on another CPU. Always updated under the runqueue lock:
1008 : */
1009 : unsigned int nr_uninterruptible;
1010 :
1011 : struct task_struct __rcu *curr;
1012 : struct task_struct *idle;
1013 : struct task_struct *stop;
1014 : unsigned long next_balance;
1015 : struct mm_struct *prev_mm;
1016 :
1017 : unsigned int clock_update_flags;
1018 : u64 clock;
1019 : /* Ensure that all clocks are in the same cache line */
1020 : u64 clock_task ____cacheline_aligned;
1021 : u64 clock_pelt;
1022 : unsigned long lost_idle_time;
1023 : u64 clock_pelt_idle;
1024 : u64 clock_idle;
1025 : #ifndef CONFIG_64BIT
1026 : u64 clock_pelt_idle_copy;
1027 : u64 clock_idle_copy;
1028 : #endif
1029 :
1030 : atomic_t nr_iowait;
1031 :
1032 : #ifdef CONFIG_SCHED_DEBUG
1033 : u64 last_seen_need_resched_ns;
1034 : int ticks_without_resched;
1035 : #endif
1036 :
1037 : #ifdef CONFIG_MEMBARRIER
1038 : int membarrier_state;
1039 : #endif
1040 :
1041 : #ifdef CONFIG_SMP
1042 : struct root_domain *rd;
1043 : struct sched_domain __rcu *sd;
1044 :
1045 : unsigned long cpu_capacity;
1046 : unsigned long cpu_capacity_orig;
1047 :
1048 : struct balance_callback *balance_callback;
1049 :
1050 : unsigned char nohz_idle_balance;
1051 : unsigned char idle_balance;
1052 :
1053 : unsigned long misfit_task_load;
1054 :
1055 : /* For active balancing */
1056 : int active_balance;
1057 : int push_cpu;
1058 : struct cpu_stop_work active_balance_work;
1059 :
1060 : /* CPU of this runqueue: */
1061 : int cpu;
1062 : int online;
1063 :
1064 : struct list_head cfs_tasks;
1065 :
1066 : struct sched_avg avg_rt;
1067 : struct sched_avg avg_dl;
1068 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1069 : struct sched_avg avg_irq;
1070 : #endif
1071 : #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1072 : struct sched_avg avg_thermal;
1073 : #endif
1074 : u64 idle_stamp;
1075 : u64 avg_idle;
1076 :
1077 : unsigned long wake_stamp;
1078 : u64 wake_avg_idle;
1079 :
1080 : /* This is used to determine avg_idle's max value */
1081 : u64 max_idle_balance_cost;
1082 :
1083 : #ifdef CONFIG_HOTPLUG_CPU
1084 : struct rcuwait hotplug_wait;
1085 : #endif
1086 : #endif /* CONFIG_SMP */
1087 :
1088 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1089 : u64 prev_irq_time;
1090 : #endif
1091 : #ifdef CONFIG_PARAVIRT
1092 : u64 prev_steal_time;
1093 : #endif
1094 : #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1095 : u64 prev_steal_time_rq;
1096 : #endif
1097 :
1098 : /* calc_load related fields */
1099 : unsigned long calc_load_update;
1100 : long calc_load_active;
1101 :
1102 : #ifdef CONFIG_SCHED_HRTICK
1103 : #ifdef CONFIG_SMP
1104 : call_single_data_t hrtick_csd;
1105 : #endif
1106 : struct hrtimer hrtick_timer;
1107 : ktime_t hrtick_time;
1108 : #endif
1109 :
1110 : #ifdef CONFIG_SCHEDSTATS
1111 : /* latency stats */
1112 : struct sched_info rq_sched_info;
1113 : unsigned long long rq_cpu_time;
1114 : /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1115 :
1116 : /* sys_sched_yield() stats */
1117 : unsigned int yld_count;
1118 :
1119 : /* schedule() stats */
1120 : unsigned int sched_count;
1121 : unsigned int sched_goidle;
1122 :
1123 : /* try_to_wake_up() stats */
1124 : unsigned int ttwu_count;
1125 : unsigned int ttwu_local;
1126 : #endif
1127 :
1128 : #ifdef CONFIG_CPU_IDLE
1129 : /* Must be inspected within a rcu lock section */
1130 : struct cpuidle_state *idle_state;
1131 : #endif
1132 :
1133 : #ifdef CONFIG_SMP
1134 : unsigned int nr_pinned;
1135 : #endif
1136 : unsigned int push_busy;
1137 : struct cpu_stop_work push_work;
1138 :
1139 : #ifdef CONFIG_SCHED_CORE
1140 : /* per rq */
1141 : struct rq *core;
1142 : struct task_struct *core_pick;
1143 : unsigned int core_enabled;
1144 : unsigned int core_sched_seq;
1145 : struct rb_root core_tree;
1146 :
1147 : /* shared state -- careful with sched_core_cpu_deactivate() */
1148 : unsigned int core_task_seq;
1149 : unsigned int core_pick_seq;
1150 : unsigned long core_cookie;
1151 : unsigned int core_forceidle_count;
1152 : unsigned int core_forceidle_seq;
1153 : unsigned int core_forceidle_occupation;
1154 : u64 core_forceidle_start;
1155 : #endif
1156 :
1157 : /* Scratch cpumask to be temporarily used under rq_lock */
1158 : cpumask_var_t scratch_mask;
1159 :
1160 : #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1161 : call_single_data_t cfsb_csd;
1162 : struct list_head cfsb_csd_list;
1163 : #endif
1164 : };
1165 :
1166 : #ifdef CONFIG_FAIR_GROUP_SCHED
1167 :
1168 : /* CPU runqueue to which this cfs_rq is attached */
1169 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1170 : {
1171 : return cfs_rq->rq;
1172 : }
1173 :
1174 : #else
1175 :
1176 : static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1177 : {
1178 23546 : return container_of(cfs_rq, struct rq, cfs);
1179 : }
1180 : #endif
1181 :
1182 : static inline int cpu_of(struct rq *rq)
1183 : {
1184 : #ifdef CONFIG_SMP
1185 : return rq->cpu;
1186 : #else
1187 : return 0;
1188 : #endif
1189 : }
1190 :
1191 : #define MDF_PUSH 0x01
1192 :
1193 : static inline bool is_migration_disabled(struct task_struct *p)
1194 : {
1195 : #ifdef CONFIG_SMP
1196 : return p->migration_disabled;
1197 : #else
1198 : return false;
1199 : #endif
1200 : }
1201 :
1202 : DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1203 :
1204 : #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1205 : #define this_rq() this_cpu_ptr(&runqueues)
1206 : #define task_rq(p) cpu_rq(task_cpu(p))
1207 : #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1208 : #define raw_rq() raw_cpu_ptr(&runqueues)
1209 :
1210 : struct sched_group;
1211 : #ifdef CONFIG_SCHED_CORE
1212 : static inline struct cpumask *sched_group_span(struct sched_group *sg);
1213 :
1214 : DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1215 :
1216 : static inline bool sched_core_enabled(struct rq *rq)
1217 : {
1218 : return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1219 : }
1220 :
1221 : static inline bool sched_core_disabled(void)
1222 : {
1223 : return !static_branch_unlikely(&__sched_core_enabled);
1224 : }
1225 :
1226 : /*
1227 : * Be careful with this function; not for general use. The return value isn't
1228 : * stable unless you actually hold a relevant rq->__lock.
1229 : */
1230 : static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1231 : {
1232 : if (sched_core_enabled(rq))
1233 : return &rq->core->__lock;
1234 :
1235 : return &rq->__lock;
1236 : }
1237 :
1238 : static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1239 : {
1240 : if (rq->core_enabled)
1241 : return &rq->core->__lock;
1242 :
1243 : return &rq->__lock;
1244 : }
1245 :
1246 : bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1247 : bool fi);
1248 :
1249 : /*
1250 : * Helpers to check if the CPU's core cookie matches with the task's cookie
1251 : * when core scheduling is enabled.
1252 : * A special case is that the task's cookie always matches with CPU's core
1253 : * cookie if the CPU is in an idle core.
1254 : */
1255 : static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1256 : {
1257 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1258 : if (!sched_core_enabled(rq))
1259 : return true;
1260 :
1261 : return rq->core->core_cookie == p->core_cookie;
1262 : }
1263 :
1264 : static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1265 : {
1266 : bool idle_core = true;
1267 : int cpu;
1268 :
1269 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1270 : if (!sched_core_enabled(rq))
1271 : return true;
1272 :
1273 : for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1274 : if (!available_idle_cpu(cpu)) {
1275 : idle_core = false;
1276 : break;
1277 : }
1278 : }
1279 :
1280 : /*
1281 : * A CPU in an idle core is always the best choice for tasks with
1282 : * cookies.
1283 : */
1284 : return idle_core || rq->core->core_cookie == p->core_cookie;
1285 : }
1286 :
1287 : static inline bool sched_group_cookie_match(struct rq *rq,
1288 : struct task_struct *p,
1289 : struct sched_group *group)
1290 : {
1291 : int cpu;
1292 :
1293 : /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1294 : if (!sched_core_enabled(rq))
1295 : return true;
1296 :
1297 : for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1298 : if (sched_core_cookie_match(cpu_rq(cpu), p))
1299 : return true;
1300 : }
1301 : return false;
1302 : }
1303 :
1304 : static inline bool sched_core_enqueued(struct task_struct *p)
1305 : {
1306 : return !RB_EMPTY_NODE(&p->core_node);
1307 : }
1308 :
1309 : extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1310 : extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1311 :
1312 : extern void sched_core_get(void);
1313 : extern void sched_core_put(void);
1314 :
1315 : #else /* !CONFIG_SCHED_CORE */
1316 :
1317 : static inline bool sched_core_enabled(struct rq *rq)
1318 : {
1319 : return false;
1320 : }
1321 :
1322 : static inline bool sched_core_disabled(void)
1323 : {
1324 : return true;
1325 : }
1326 :
1327 : static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1328 : {
1329 : return &rq->__lock;
1330 : }
1331 :
1332 : static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1333 : {
1334 : return &rq->__lock;
1335 : }
1336 :
1337 : static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1338 : {
1339 : return true;
1340 : }
1341 :
1342 : static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1343 : {
1344 : return true;
1345 : }
1346 :
1347 : static inline bool sched_group_cookie_match(struct rq *rq,
1348 : struct task_struct *p,
1349 : struct sched_group *group)
1350 : {
1351 : return true;
1352 : }
1353 : #endif /* CONFIG_SCHED_CORE */
1354 :
1355 : static inline void lockdep_assert_rq_held(struct rq *rq)
1356 : {
1357 30869 : lockdep_assert_held(__rq_lockp(rq));
1358 : }
1359 :
1360 : extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1361 : extern bool raw_spin_rq_trylock(struct rq *rq);
1362 : extern void raw_spin_rq_unlock(struct rq *rq);
1363 :
1364 : static inline void raw_spin_rq_lock(struct rq *rq)
1365 : {
1366 8666 : raw_spin_rq_lock_nested(rq, 0);
1367 : }
1368 :
1369 : static inline void raw_spin_rq_lock_irq(struct rq *rq)
1370 : {
1371 0 : local_irq_disable();
1372 0 : raw_spin_rq_lock(rq);
1373 : }
1374 :
1375 : static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1376 : {
1377 2513 : raw_spin_rq_unlock(rq);
1378 : local_irq_enable();
1379 : }
1380 :
1381 : static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1382 : {
1383 : unsigned long flags;
1384 0 : local_irq_save(flags);
1385 0 : raw_spin_rq_lock(rq);
1386 : return flags;
1387 : }
1388 :
1389 : static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1390 : {
1391 0 : raw_spin_rq_unlock(rq);
1392 0 : local_irq_restore(flags);
1393 : }
1394 :
1395 : #define raw_spin_rq_lock_irqsave(rq, flags) \
1396 : do { \
1397 : flags = _raw_spin_rq_lock_irqsave(rq); \
1398 : } while (0)
1399 :
1400 : #ifdef CONFIG_SCHED_SMT
1401 : extern void __update_idle_core(struct rq *rq);
1402 :
1403 : static inline void update_idle_core(struct rq *rq)
1404 : {
1405 : if (static_branch_unlikely(&sched_smt_present))
1406 : __update_idle_core(rq);
1407 : }
1408 :
1409 : #else
1410 : static inline void update_idle_core(struct rq *rq) { }
1411 : #endif
1412 :
1413 : #ifdef CONFIG_FAIR_GROUP_SCHED
1414 : static inline struct task_struct *task_of(struct sched_entity *se)
1415 : {
1416 : SCHED_WARN_ON(!entity_is_task(se));
1417 : return container_of(se, struct task_struct, se);
1418 : }
1419 :
1420 : static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1421 : {
1422 : return p->se.cfs_rq;
1423 : }
1424 :
1425 : /* runqueue on which this entity is (to be) queued */
1426 : static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1427 : {
1428 : return se->cfs_rq;
1429 : }
1430 :
1431 : /* runqueue "owned" by this group */
1432 : static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1433 : {
1434 : return grp->my_q;
1435 : }
1436 :
1437 : #else
1438 :
1439 : #define task_of(_se) container_of(_se, struct task_struct, se)
1440 :
1441 : static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1442 : {
1443 2821 : return &task_rq(p)->cfs;
1444 : }
1445 :
1446 : static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1447 : {
1448 19161 : const struct task_struct *p = task_of(se);
1449 19161 : struct rq *rq = task_rq(p);
1450 :
1451 : return &rq->cfs;
1452 : }
1453 :
1454 : /* runqueue "owned" by this group */
1455 : static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1456 : {
1457 : return NULL;
1458 : }
1459 : #endif
1460 :
1461 : extern void update_rq_clock(struct rq *rq);
1462 :
1463 : /*
1464 : * rq::clock_update_flags bits
1465 : *
1466 : * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1467 : * call to __schedule(). This is an optimisation to avoid
1468 : * neighbouring rq clock updates.
1469 : *
1470 : * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1471 : * in effect and calls to update_rq_clock() are being ignored.
1472 : *
1473 : * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1474 : * made to update_rq_clock() since the last time rq::lock was pinned.
1475 : *
1476 : * If inside of __schedule(), clock_update_flags will have been
1477 : * shifted left (a left shift is a cheap operation for the fast path
1478 : * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1479 : *
1480 : * if (rq-clock_update_flags >= RQCF_UPDATED)
1481 : *
1482 : * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1483 : * one position though, because the next rq_unpin_lock() will shift it
1484 : * back.
1485 : */
1486 : #define RQCF_REQ_SKIP 0x01
1487 : #define RQCF_ACT_SKIP 0x02
1488 : #define RQCF_UPDATED 0x04
1489 :
1490 : static inline void assert_clock_updated(struct rq *rq)
1491 : {
1492 : /*
1493 : * The only reason for not seeing a clock update since the
1494 : * last rq_pin_lock() is if we're currently skipping updates.
1495 : */
1496 : SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1497 : }
1498 :
1499 : static inline u64 rq_clock(struct rq *rq)
1500 : {
1501 0 : lockdep_assert_rq_held(rq);
1502 0 : assert_clock_updated(rq);
1503 :
1504 : return rq->clock;
1505 : }
1506 :
1507 : static inline u64 rq_clock_task(struct rq *rq)
1508 : {
1509 17817 : lockdep_assert_rq_held(rq);
1510 17817 : assert_clock_updated(rq);
1511 :
1512 : return rq->clock_task;
1513 : }
1514 :
1515 : /**
1516 : * By default the decay is the default pelt decay period.
1517 : * The decay shift can change the decay period in
1518 : * multiples of 32.
1519 : * Decay shift Decay period(ms)
1520 : * 0 32
1521 : * 1 64
1522 : * 2 128
1523 : * 3 256
1524 : * 4 512
1525 : */
1526 : extern int sched_thermal_decay_shift;
1527 :
1528 : static inline u64 rq_clock_thermal(struct rq *rq)
1529 : {
1530 2943 : return rq_clock_task(rq) >> sched_thermal_decay_shift;
1531 : }
1532 :
1533 : static inline void rq_clock_skip_update(struct rq *rq)
1534 : {
1535 1252 : lockdep_assert_rq_held(rq);
1536 1252 : rq->clock_update_flags |= RQCF_REQ_SKIP;
1537 : }
1538 :
1539 : /*
1540 : * See rt task throttling, which is the only time a skip
1541 : * request is canceled.
1542 : */
1543 : static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1544 : {
1545 0 : lockdep_assert_rq_held(rq);
1546 0 : rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1547 : }
1548 :
1549 : struct rq_flags {
1550 : unsigned long flags;
1551 : struct pin_cookie cookie;
1552 : #ifdef CONFIG_SCHED_DEBUG
1553 : /*
1554 : * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1555 : * current pin context is stashed here in case it needs to be
1556 : * restored in rq_repin_lock().
1557 : */
1558 : unsigned int clock_update_flags;
1559 : #endif
1560 : };
1561 :
1562 : extern struct balance_callback balance_push_callback;
1563 :
1564 : /*
1565 : * Lockdep annotation that avoids accidental unlocks; it's like a
1566 : * sticky/continuous lockdep_assert_held().
1567 : *
1568 : * This avoids code that has access to 'struct rq *rq' (basically everything in
1569 : * the scheduler) from accidentally unlocking the rq if they do not also have a
1570 : * copy of the (on-stack) 'struct rq_flags rf'.
1571 : *
1572 : * Also see Documentation/locking/lockdep-design.rst.
1573 : */
1574 : static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1575 : {
1576 : rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1577 :
1578 : #ifdef CONFIG_SCHED_DEBUG
1579 : rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1580 : rf->clock_update_flags = 0;
1581 : #ifdef CONFIG_SMP
1582 : SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1583 : #endif
1584 : #endif
1585 : }
1586 :
1587 : static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1588 : {
1589 : #ifdef CONFIG_SCHED_DEBUG
1590 : if (rq->clock_update_flags > RQCF_ACT_SKIP)
1591 : rf->clock_update_flags = RQCF_UPDATED;
1592 : #endif
1593 :
1594 6152 : lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1595 : }
1596 :
1597 : static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1598 : {
1599 : lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1600 :
1601 : #ifdef CONFIG_SCHED_DEBUG
1602 : /*
1603 : * Restore the value we stashed in @rf for this pin context.
1604 : */
1605 : rq->clock_update_flags |= rf->clock_update_flags;
1606 : #endif
1607 : }
1608 :
1609 : struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1610 : __acquires(rq->lock);
1611 :
1612 : struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1613 : __acquires(p->pi_lock)
1614 : __acquires(rq->lock);
1615 :
1616 : static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1617 : __releases(rq->lock)
1618 : {
1619 0 : rq_unpin_lock(rq, rf);
1620 0 : raw_spin_rq_unlock(rq);
1621 : }
1622 :
1623 : static inline void
1624 : task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1625 : __releases(rq->lock)
1626 : __releases(p->pi_lock)
1627 : {
1628 1534 : rq_unpin_lock(rq, rf);
1629 767 : raw_spin_rq_unlock(rq);
1630 1534 : raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1631 : }
1632 :
1633 : static inline void
1634 : rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1635 : __acquires(rq->lock)
1636 : {
1637 : raw_spin_rq_lock_irqsave(rq, rf->flags);
1638 : rq_pin_lock(rq, rf);
1639 : }
1640 :
1641 : static inline void
1642 : rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1643 : __acquires(rq->lock)
1644 : {
1645 : raw_spin_rq_lock_irq(rq);
1646 : rq_pin_lock(rq, rf);
1647 : }
1648 :
1649 : static inline void
1650 : rq_lock(struct rq *rq, struct rq_flags *rf)
1651 : __acquires(rq->lock)
1652 : {
1653 7898 : raw_spin_rq_lock(rq);
1654 7898 : rq_pin_lock(rq, rf);
1655 : }
1656 :
1657 : static inline void
1658 : rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1659 : __releases(rq->lock)
1660 : {
1661 : rq_unpin_lock(rq, rf);
1662 : raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1663 : }
1664 :
1665 : static inline void
1666 : rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1667 : __releases(rq->lock)
1668 : {
1669 0 : rq_unpin_lock(rq, rf);
1670 0 : raw_spin_rq_unlock_irq(rq);
1671 : }
1672 :
1673 : static inline void
1674 : rq_unlock(struct rq *rq, struct rq_flags *rf)
1675 : __releases(rq->lock)
1676 : {
1677 10388 : rq_unpin_lock(rq, rf);
1678 5385 : raw_spin_rq_unlock(rq);
1679 : }
1680 :
1681 : static inline struct rq *
1682 : this_rq_lock_irq(struct rq_flags *rf)
1683 : __acquires(rq->lock)
1684 : {
1685 : struct rq *rq;
1686 :
1687 : local_irq_disable();
1688 0 : rq = this_rq();
1689 0 : rq_lock(rq, rf);
1690 : return rq;
1691 : }
1692 :
1693 : #ifdef CONFIG_NUMA
1694 : enum numa_topology_type {
1695 : NUMA_DIRECT,
1696 : NUMA_GLUELESS_MESH,
1697 : NUMA_BACKPLANE,
1698 : };
1699 : extern enum numa_topology_type sched_numa_topology_type;
1700 : extern int sched_max_numa_distance;
1701 : extern bool find_numa_distance(int distance);
1702 : extern void sched_init_numa(int offline_node);
1703 : extern void sched_update_numa(int cpu, bool online);
1704 : extern void sched_domains_numa_masks_set(unsigned int cpu);
1705 : extern void sched_domains_numa_masks_clear(unsigned int cpu);
1706 : extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1707 : #else
1708 : static inline void sched_init_numa(int offline_node) { }
1709 : static inline void sched_update_numa(int cpu, bool online) { }
1710 : static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1711 : static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1712 : static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1713 : {
1714 : return nr_cpu_ids;
1715 : }
1716 : #endif
1717 :
1718 : #ifdef CONFIG_NUMA_BALANCING
1719 : /* The regions in numa_faults array from task_struct */
1720 : enum numa_faults_stats {
1721 : NUMA_MEM = 0,
1722 : NUMA_CPU,
1723 : NUMA_MEMBUF,
1724 : NUMA_CPUBUF
1725 : };
1726 : extern void sched_setnuma(struct task_struct *p, int node);
1727 : extern int migrate_task_to(struct task_struct *p, int cpu);
1728 : extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1729 : int cpu, int scpu);
1730 : extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1731 : #else
1732 : static inline void
1733 : init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1734 : {
1735 : }
1736 : #endif /* CONFIG_NUMA_BALANCING */
1737 :
1738 : #ifdef CONFIG_SMP
1739 :
1740 : static inline void
1741 : queue_balance_callback(struct rq *rq,
1742 : struct balance_callback *head,
1743 : void (*func)(struct rq *rq))
1744 : {
1745 : lockdep_assert_rq_held(rq);
1746 :
1747 : /*
1748 : * Don't (re)queue an already queued item; nor queue anything when
1749 : * balance_push() is active, see the comment with
1750 : * balance_push_callback.
1751 : */
1752 : if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1753 : return;
1754 :
1755 : head->func = func;
1756 : head->next = rq->balance_callback;
1757 : rq->balance_callback = head;
1758 : }
1759 :
1760 : #define rcu_dereference_check_sched_domain(p) \
1761 : rcu_dereference_check((p), \
1762 : lockdep_is_held(&sched_domains_mutex))
1763 :
1764 : /*
1765 : * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1766 : * See destroy_sched_domains: call_rcu for details.
1767 : *
1768 : * The domain tree of any CPU may only be accessed from within
1769 : * preempt-disabled sections.
1770 : */
1771 : #define for_each_domain(cpu, __sd) \
1772 : for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1773 : __sd; __sd = __sd->parent)
1774 :
1775 : /**
1776 : * highest_flag_domain - Return highest sched_domain containing flag.
1777 : * @cpu: The CPU whose highest level of sched domain is to
1778 : * be returned.
1779 : * @flag: The flag to check for the highest sched_domain
1780 : * for the given CPU.
1781 : *
1782 : * Returns the highest sched_domain of a CPU which contains the given flag.
1783 : */
1784 : static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1785 : {
1786 : struct sched_domain *sd, *hsd = NULL;
1787 :
1788 : for_each_domain(cpu, sd) {
1789 : if (!(sd->flags & flag))
1790 : break;
1791 : hsd = sd;
1792 : }
1793 :
1794 : return hsd;
1795 : }
1796 :
1797 : static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1798 : {
1799 : struct sched_domain *sd;
1800 :
1801 : for_each_domain(cpu, sd) {
1802 : if (sd->flags & flag)
1803 : break;
1804 : }
1805 :
1806 : return sd;
1807 : }
1808 :
1809 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1810 : DECLARE_PER_CPU(int, sd_llc_size);
1811 : DECLARE_PER_CPU(int, sd_llc_id);
1812 : DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1813 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1814 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1815 : DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1816 : extern struct static_key_false sched_asym_cpucapacity;
1817 :
1818 : static __always_inline bool sched_asym_cpucap_active(void)
1819 : {
1820 : return static_branch_unlikely(&sched_asym_cpucapacity);
1821 : }
1822 :
1823 : struct sched_group_capacity {
1824 : atomic_t ref;
1825 : /*
1826 : * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1827 : * for a single CPU.
1828 : */
1829 : unsigned long capacity;
1830 : unsigned long min_capacity; /* Min per-CPU capacity in group */
1831 : unsigned long max_capacity; /* Max per-CPU capacity in group */
1832 : unsigned long next_update;
1833 : int imbalance; /* XXX unrelated to capacity but shared group state */
1834 :
1835 : #ifdef CONFIG_SCHED_DEBUG
1836 : int id;
1837 : #endif
1838 :
1839 : unsigned long cpumask[]; /* Balance mask */
1840 : };
1841 :
1842 : struct sched_group {
1843 : struct sched_group *next; /* Must be a circular list */
1844 : atomic_t ref;
1845 :
1846 : unsigned int group_weight;
1847 : struct sched_group_capacity *sgc;
1848 : int asym_prefer_cpu; /* CPU of highest priority in group */
1849 : int flags;
1850 :
1851 : /*
1852 : * The CPUs this group covers.
1853 : *
1854 : * NOTE: this field is variable length. (Allocated dynamically
1855 : * by attaching extra space to the end of the structure,
1856 : * depending on how many CPUs the kernel has booted up with)
1857 : */
1858 : unsigned long cpumask[];
1859 : };
1860 :
1861 : static inline struct cpumask *sched_group_span(struct sched_group *sg)
1862 : {
1863 : return to_cpumask(sg->cpumask);
1864 : }
1865 :
1866 : /*
1867 : * See build_balance_mask().
1868 : */
1869 : static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1870 : {
1871 : return to_cpumask(sg->sgc->cpumask);
1872 : }
1873 :
1874 : extern int group_balance_cpu(struct sched_group *sg);
1875 :
1876 : #ifdef CONFIG_SCHED_DEBUG
1877 : void update_sched_domain_debugfs(void);
1878 : void dirty_sched_domain_sysctl(int cpu);
1879 : #else
1880 : static inline void update_sched_domain_debugfs(void)
1881 : {
1882 : }
1883 : static inline void dirty_sched_domain_sysctl(int cpu)
1884 : {
1885 : }
1886 : #endif
1887 :
1888 : extern int sched_update_scaling(void);
1889 :
1890 : static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1891 : {
1892 : if (!p->user_cpus_ptr)
1893 : return cpu_possible_mask; /* &init_task.cpus_mask */
1894 : return p->user_cpus_ptr;
1895 : }
1896 : #endif /* CONFIG_SMP */
1897 :
1898 : #include "stats.h"
1899 :
1900 : #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1901 :
1902 : extern void __sched_core_account_forceidle(struct rq *rq);
1903 :
1904 : static inline void sched_core_account_forceidle(struct rq *rq)
1905 : {
1906 : if (schedstat_enabled())
1907 : __sched_core_account_forceidle(rq);
1908 : }
1909 :
1910 : extern void __sched_core_tick(struct rq *rq);
1911 :
1912 : static inline void sched_core_tick(struct rq *rq)
1913 : {
1914 : if (sched_core_enabled(rq) && schedstat_enabled())
1915 : __sched_core_tick(rq);
1916 : }
1917 :
1918 : #else
1919 :
1920 : static inline void sched_core_account_forceidle(struct rq *rq) {}
1921 :
1922 : static inline void sched_core_tick(struct rq *rq) {}
1923 :
1924 : #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1925 :
1926 : #ifdef CONFIG_CGROUP_SCHED
1927 :
1928 : /*
1929 : * Return the group to which this tasks belongs.
1930 : *
1931 : * We cannot use task_css() and friends because the cgroup subsystem
1932 : * changes that value before the cgroup_subsys::attach() method is called,
1933 : * therefore we cannot pin it and might observe the wrong value.
1934 : *
1935 : * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1936 : * core changes this before calling sched_move_task().
1937 : *
1938 : * Instead we use a 'copy' which is updated from sched_move_task() while
1939 : * holding both task_struct::pi_lock and rq::lock.
1940 : */
1941 : static inline struct task_group *task_group(struct task_struct *p)
1942 : {
1943 : return p->sched_task_group;
1944 : }
1945 :
1946 : /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1947 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1948 : {
1949 : #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1950 : struct task_group *tg = task_group(p);
1951 : #endif
1952 :
1953 : #ifdef CONFIG_FAIR_GROUP_SCHED
1954 : set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1955 : p->se.cfs_rq = tg->cfs_rq[cpu];
1956 : p->se.parent = tg->se[cpu];
1957 : p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
1958 : #endif
1959 :
1960 : #ifdef CONFIG_RT_GROUP_SCHED
1961 : p->rt.rt_rq = tg->rt_rq[cpu];
1962 : p->rt.parent = tg->rt_se[cpu];
1963 : #endif
1964 : }
1965 :
1966 : #else /* CONFIG_CGROUP_SCHED */
1967 :
1968 : static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1969 : static inline struct task_group *task_group(struct task_struct *p)
1970 : {
1971 : return NULL;
1972 : }
1973 :
1974 : #endif /* CONFIG_CGROUP_SCHED */
1975 :
1976 : static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1977 : {
1978 383 : set_task_rq(p, cpu);
1979 : #ifdef CONFIG_SMP
1980 : /*
1981 : * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1982 : * successfully executed on another CPU. We must ensure that updates of
1983 : * per-task data have been completed by this moment.
1984 : */
1985 : smp_wmb();
1986 : WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1987 : p->wake_cpu = cpu;
1988 : #endif
1989 : }
1990 :
1991 : /*
1992 : * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1993 : */
1994 : #ifdef CONFIG_SCHED_DEBUG
1995 : # define const_debug __read_mostly
1996 : #else
1997 : # define const_debug const
1998 : #endif
1999 :
2000 : #define SCHED_FEAT(name, enabled) \
2001 : __SCHED_FEAT_##name ,
2002 :
2003 : enum {
2004 : #include "features.h"
2005 : __SCHED_FEAT_NR,
2006 : };
2007 :
2008 : #undef SCHED_FEAT
2009 :
2010 : #ifdef CONFIG_SCHED_DEBUG
2011 :
2012 : /*
2013 : * To support run-time toggling of sched features, all the translation units
2014 : * (but core.c) reference the sysctl_sched_features defined in core.c.
2015 : */
2016 : extern const_debug unsigned int sysctl_sched_features;
2017 :
2018 : #ifdef CONFIG_JUMP_LABEL
2019 : #define SCHED_FEAT(name, enabled) \
2020 : static __always_inline bool static_branch_##name(struct static_key *key) \
2021 : { \
2022 : return static_key_##enabled(key); \
2023 : }
2024 :
2025 : #include "features.h"
2026 : #undef SCHED_FEAT
2027 :
2028 : extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2029 : #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2030 :
2031 : #else /* !CONFIG_JUMP_LABEL */
2032 :
2033 : #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2034 :
2035 : #endif /* CONFIG_JUMP_LABEL */
2036 :
2037 : #else /* !SCHED_DEBUG */
2038 :
2039 : /*
2040 : * Each translation unit has its own copy of sysctl_sched_features to allow
2041 : * constants propagation at compile time and compiler optimization based on
2042 : * features default.
2043 : */
2044 : #define SCHED_FEAT(name, enabled) \
2045 : (1UL << __SCHED_FEAT_##name) * enabled |
2046 : static const_debug __maybe_unused unsigned int sysctl_sched_features =
2047 : #include "features.h"
2048 : 0;
2049 : #undef SCHED_FEAT
2050 :
2051 : #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2052 :
2053 : #endif /* SCHED_DEBUG */
2054 :
2055 : extern struct static_key_false sched_numa_balancing;
2056 : extern struct static_key_false sched_schedstats;
2057 :
2058 : static inline u64 global_rt_period(void)
2059 : {
2060 4 : return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2061 : }
2062 :
2063 : static inline u64 global_rt_runtime(void)
2064 : {
2065 4 : if (sysctl_sched_rt_runtime < 0)
2066 : return RUNTIME_INF;
2067 :
2068 4 : return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2069 : }
2070 :
2071 : static inline int task_current(struct rq *rq, struct task_struct *p)
2072 : {
2073 : return rq->curr == p;
2074 : }
2075 :
2076 : static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2077 : {
2078 : #ifdef CONFIG_SMP
2079 : return p->on_cpu;
2080 : #else
2081 0 : return task_current(rq, p);
2082 : #endif
2083 : }
2084 :
2085 : static inline int task_on_rq_queued(struct task_struct *p)
2086 : {
2087 0 : return p->on_rq == TASK_ON_RQ_QUEUED;
2088 : }
2089 :
2090 : static inline int task_on_rq_migrating(struct task_struct *p)
2091 : {
2092 5653 : return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2093 : }
2094 :
2095 : /* Wake flags. The first three directly map to some SD flag value */
2096 : #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2097 : #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2098 : #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2099 :
2100 : #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2101 : #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2102 :
2103 : #ifdef CONFIG_SMP
2104 : static_assert(WF_EXEC == SD_BALANCE_EXEC);
2105 : static_assert(WF_FORK == SD_BALANCE_FORK);
2106 : static_assert(WF_TTWU == SD_BALANCE_WAKE);
2107 : #endif
2108 :
2109 : /*
2110 : * To aid in avoiding the subversion of "niceness" due to uneven distribution
2111 : * of tasks with abnormal "nice" values across CPUs the contribution that
2112 : * each task makes to its run queue's load is weighted according to its
2113 : * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2114 : * scaled version of the new time slice allocation that they receive on time
2115 : * slice expiry etc.
2116 : */
2117 :
2118 : #define WEIGHT_IDLEPRIO 3
2119 : #define WMULT_IDLEPRIO 1431655765
2120 :
2121 : extern const int sched_prio_to_weight[40];
2122 : extern const u32 sched_prio_to_wmult[40];
2123 :
2124 : /*
2125 : * {de,en}queue flags:
2126 : *
2127 : * DEQUEUE_SLEEP - task is no longer runnable
2128 : * ENQUEUE_WAKEUP - task just became runnable
2129 : *
2130 : * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2131 : * are in a known state which allows modification. Such pairs
2132 : * should preserve as much state as possible.
2133 : *
2134 : * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2135 : * in the runqueue.
2136 : *
2137 : * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2138 : * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2139 : * ENQUEUE_MIGRATED - the task was migrated during wakeup
2140 : *
2141 : */
2142 :
2143 : #define DEQUEUE_SLEEP 0x01
2144 : #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2145 : #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2146 : #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2147 :
2148 : #define ENQUEUE_WAKEUP 0x01
2149 : #define ENQUEUE_RESTORE 0x02
2150 : #define ENQUEUE_MOVE 0x04
2151 : #define ENQUEUE_NOCLOCK 0x08
2152 :
2153 : #define ENQUEUE_HEAD 0x10
2154 : #define ENQUEUE_REPLENISH 0x20
2155 : #ifdef CONFIG_SMP
2156 : #define ENQUEUE_MIGRATED 0x40
2157 : #else
2158 : #define ENQUEUE_MIGRATED 0x00
2159 : #endif
2160 :
2161 : #define RETRY_TASK ((void *)-1UL)
2162 :
2163 : struct affinity_context {
2164 : const struct cpumask *new_mask;
2165 : struct cpumask *user_mask;
2166 : unsigned int flags;
2167 : };
2168 :
2169 : struct sched_class {
2170 :
2171 : #ifdef CONFIG_UCLAMP_TASK
2172 : int uclamp_enabled;
2173 : #endif
2174 :
2175 : void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2176 : void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2177 : void (*yield_task) (struct rq *rq);
2178 : bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2179 :
2180 : void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2181 :
2182 : struct task_struct *(*pick_next_task)(struct rq *rq);
2183 :
2184 : void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2185 : void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2186 :
2187 : #ifdef CONFIG_SMP
2188 : int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2189 : int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2190 :
2191 : struct task_struct * (*pick_task)(struct rq *rq);
2192 :
2193 : void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2194 :
2195 : void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2196 :
2197 : void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2198 :
2199 : void (*rq_online)(struct rq *rq);
2200 : void (*rq_offline)(struct rq *rq);
2201 :
2202 : struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2203 : #endif
2204 :
2205 : void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2206 : void (*task_fork)(struct task_struct *p);
2207 : void (*task_dead)(struct task_struct *p);
2208 :
2209 : /*
2210 : * The switched_from() call is allowed to drop rq->lock, therefore we
2211 : * cannot assume the switched_from/switched_to pair is serialized by
2212 : * rq->lock. They are however serialized by p->pi_lock.
2213 : */
2214 : void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2215 : void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2216 : void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2217 : int oldprio);
2218 :
2219 : unsigned int (*get_rr_interval)(struct rq *rq,
2220 : struct task_struct *task);
2221 :
2222 : void (*update_curr)(struct rq *rq);
2223 :
2224 : #ifdef CONFIG_FAIR_GROUP_SCHED
2225 : void (*task_change_group)(struct task_struct *p);
2226 : #endif
2227 :
2228 : #ifdef CONFIG_SCHED_CORE
2229 : int (*task_is_throttled)(struct task_struct *p, int cpu);
2230 : #endif
2231 : };
2232 :
2233 2517 : static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2234 : {
2235 2517 : WARN_ON_ONCE(rq->curr != prev);
2236 2517 : prev->sched_class->put_prev_task(rq, prev);
2237 2517 : }
2238 :
2239 : static inline void set_next_task(struct rq *rq, struct task_struct *next)
2240 : {
2241 4 : next->sched_class->set_next_task(rq, next, false);
2242 : }
2243 :
2244 :
2245 : /*
2246 : * Helper to define a sched_class instance; each one is placed in a separate
2247 : * section which is ordered by the linker script:
2248 : *
2249 : * include/asm-generic/vmlinux.lds.h
2250 : *
2251 : * *CAREFUL* they are laid out in *REVERSE* order!!!
2252 : *
2253 : * Also enforce alignment on the instance, not the type, to guarantee layout.
2254 : */
2255 : #define DEFINE_SCHED_CLASS(name) \
2256 : const struct sched_class name##_sched_class \
2257 : __aligned(__alignof__(struct sched_class)) \
2258 : __section("__" #name "_sched_class")
2259 :
2260 : /* Defined in include/asm-generic/vmlinux.lds.h */
2261 : extern struct sched_class __sched_class_highest[];
2262 : extern struct sched_class __sched_class_lowest[];
2263 :
2264 : #define for_class_range(class, _from, _to) \
2265 : for (class = (_from); class < (_to); class++)
2266 :
2267 : #define for_each_class(class) \
2268 : for_class_range(class, __sched_class_highest, __sched_class_lowest)
2269 :
2270 : #define sched_class_above(_a, _b) ((_a) < (_b))
2271 :
2272 : extern const struct sched_class stop_sched_class;
2273 : extern const struct sched_class dl_sched_class;
2274 : extern const struct sched_class rt_sched_class;
2275 : extern const struct sched_class fair_sched_class;
2276 : extern const struct sched_class idle_sched_class;
2277 :
2278 : static inline bool sched_stop_runnable(struct rq *rq)
2279 : {
2280 : return rq->stop && task_on_rq_queued(rq->stop);
2281 : }
2282 :
2283 : static inline bool sched_dl_runnable(struct rq *rq)
2284 : {
2285 : return rq->dl.dl_nr_running > 0;
2286 : }
2287 :
2288 : static inline bool sched_rt_runnable(struct rq *rq)
2289 : {
2290 : return rq->rt.rt_queued > 0;
2291 : }
2292 :
2293 : static inline bool sched_fair_runnable(struct rq *rq)
2294 : {
2295 : return rq->cfs.nr_running > 0;
2296 : }
2297 :
2298 : extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2299 : extern struct task_struct *pick_next_task_idle(struct rq *rq);
2300 :
2301 : #define SCA_CHECK 0x01
2302 : #define SCA_MIGRATE_DISABLE 0x02
2303 : #define SCA_MIGRATE_ENABLE 0x04
2304 : #define SCA_USER 0x08
2305 :
2306 : #ifdef CONFIG_SMP
2307 :
2308 : extern void update_group_capacity(struct sched_domain *sd, int cpu);
2309 :
2310 : extern void trigger_load_balance(struct rq *rq);
2311 :
2312 : extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2313 :
2314 : static inline struct task_struct *get_push_task(struct rq *rq)
2315 : {
2316 : struct task_struct *p = rq->curr;
2317 :
2318 : lockdep_assert_rq_held(rq);
2319 :
2320 : if (rq->push_busy)
2321 : return NULL;
2322 :
2323 : if (p->nr_cpus_allowed == 1)
2324 : return NULL;
2325 :
2326 : if (p->migration_disabled)
2327 : return NULL;
2328 :
2329 : rq->push_busy = true;
2330 : return get_task_struct(p);
2331 : }
2332 :
2333 : extern int push_cpu_stop(void *arg);
2334 :
2335 : #endif
2336 :
2337 : #ifdef CONFIG_CPU_IDLE
2338 : static inline void idle_set_state(struct rq *rq,
2339 : struct cpuidle_state *idle_state)
2340 : {
2341 : rq->idle_state = idle_state;
2342 : }
2343 :
2344 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2345 : {
2346 : SCHED_WARN_ON(!rcu_read_lock_held());
2347 :
2348 : return rq->idle_state;
2349 : }
2350 : #else
2351 : static inline void idle_set_state(struct rq *rq,
2352 : struct cpuidle_state *idle_state)
2353 : {
2354 : }
2355 :
2356 : static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2357 : {
2358 : return NULL;
2359 : }
2360 : #endif
2361 :
2362 : extern void schedule_idle(void);
2363 :
2364 : extern void sysrq_sched_debug_show(void);
2365 : extern void sched_init_granularity(void);
2366 : extern void update_max_interval(void);
2367 :
2368 : extern void init_sched_dl_class(void);
2369 : extern void init_sched_rt_class(void);
2370 : extern void init_sched_fair_class(void);
2371 :
2372 : extern void reweight_task(struct task_struct *p, int prio);
2373 :
2374 : extern void resched_curr(struct rq *rq);
2375 : extern void resched_cpu(int cpu);
2376 :
2377 : extern struct rt_bandwidth def_rt_bandwidth;
2378 : extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2379 : extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2380 :
2381 : extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2382 : extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2383 : extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2384 :
2385 : #define BW_SHIFT 20
2386 : #define BW_UNIT (1 << BW_SHIFT)
2387 : #define RATIO_SHIFT 8
2388 : #define MAX_BW_BITS (64 - BW_SHIFT)
2389 : #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2390 : unsigned long to_ratio(u64 period, u64 runtime);
2391 :
2392 : extern void init_entity_runnable_average(struct sched_entity *se);
2393 : extern void post_init_entity_util_avg(struct task_struct *p);
2394 :
2395 : #ifdef CONFIG_NO_HZ_FULL
2396 : extern bool sched_can_stop_tick(struct rq *rq);
2397 : extern int __init sched_tick_offload_init(void);
2398 :
2399 : /*
2400 : * Tick may be needed by tasks in the runqueue depending on their policy and
2401 : * requirements. If tick is needed, lets send the target an IPI to kick it out of
2402 : * nohz mode if necessary.
2403 : */
2404 : static inline void sched_update_tick_dependency(struct rq *rq)
2405 : {
2406 : int cpu = cpu_of(rq);
2407 :
2408 : if (!tick_nohz_full_cpu(cpu))
2409 : return;
2410 :
2411 : if (sched_can_stop_tick(rq))
2412 : tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2413 : else
2414 : tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2415 : }
2416 : #else
2417 : static inline int sched_tick_offload_init(void) { return 0; }
2418 : static inline void sched_update_tick_dependency(struct rq *rq) { }
2419 : #endif
2420 :
2421 : static inline void add_nr_running(struct rq *rq, unsigned count)
2422 : {
2423 2446 : unsigned prev_nr = rq->nr_running;
2424 :
2425 2446 : rq->nr_running = prev_nr + count;
2426 : if (trace_sched_update_nr_running_tp_enabled()) {
2427 : call_trace_sched_update_nr_running(rq, count);
2428 : }
2429 :
2430 : #ifdef CONFIG_SMP
2431 : if (prev_nr < 2 && rq->nr_running >= 2) {
2432 : if (!READ_ONCE(rq->rd->overload))
2433 : WRITE_ONCE(rq->rd->overload, 1);
2434 : }
2435 : #endif
2436 :
2437 2446 : sched_update_tick_dependency(rq);
2438 : }
2439 :
2440 : static inline void sub_nr_running(struct rq *rq, unsigned count)
2441 : {
2442 2444 : rq->nr_running -= count;
2443 : if (trace_sched_update_nr_running_tp_enabled()) {
2444 : call_trace_sched_update_nr_running(rq, -count);
2445 : }
2446 :
2447 : /* Check if we still need preemption */
2448 2444 : sched_update_tick_dependency(rq);
2449 : }
2450 :
2451 : extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2452 : extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2453 :
2454 : extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2455 :
2456 : #ifdef CONFIG_PREEMPT_RT
2457 : #define SCHED_NR_MIGRATE_BREAK 8
2458 : #else
2459 : #define SCHED_NR_MIGRATE_BREAK 32
2460 : #endif
2461 :
2462 : extern const_debug unsigned int sysctl_sched_nr_migrate;
2463 : extern const_debug unsigned int sysctl_sched_migration_cost;
2464 :
2465 : #ifdef CONFIG_SCHED_DEBUG
2466 : extern unsigned int sysctl_sched_latency;
2467 : extern unsigned int sysctl_sched_min_granularity;
2468 : extern unsigned int sysctl_sched_idle_min_granularity;
2469 : extern unsigned int sysctl_sched_wakeup_granularity;
2470 : extern int sysctl_resched_latency_warn_ms;
2471 : extern int sysctl_resched_latency_warn_once;
2472 :
2473 : extern unsigned int sysctl_sched_tunable_scaling;
2474 :
2475 : extern unsigned int sysctl_numa_balancing_scan_delay;
2476 : extern unsigned int sysctl_numa_balancing_scan_period_min;
2477 : extern unsigned int sysctl_numa_balancing_scan_period_max;
2478 : extern unsigned int sysctl_numa_balancing_scan_size;
2479 : extern unsigned int sysctl_numa_balancing_hot_threshold;
2480 : #endif
2481 :
2482 : #ifdef CONFIG_SCHED_HRTICK
2483 :
2484 : /*
2485 : * Use hrtick when:
2486 : * - enabled by features
2487 : * - hrtimer is actually high res
2488 : */
2489 : static inline int hrtick_enabled(struct rq *rq)
2490 : {
2491 : if (!cpu_active(cpu_of(rq)))
2492 : return 0;
2493 : return hrtimer_is_hres_active(&rq->hrtick_timer);
2494 : }
2495 :
2496 : static inline int hrtick_enabled_fair(struct rq *rq)
2497 : {
2498 : if (!sched_feat(HRTICK))
2499 : return 0;
2500 : return hrtick_enabled(rq);
2501 : }
2502 :
2503 : static inline int hrtick_enabled_dl(struct rq *rq)
2504 : {
2505 : if (!sched_feat(HRTICK_DL))
2506 : return 0;
2507 : return hrtick_enabled(rq);
2508 : }
2509 :
2510 : void hrtick_start(struct rq *rq, u64 delay);
2511 :
2512 : #else
2513 :
2514 : static inline int hrtick_enabled_fair(struct rq *rq)
2515 : {
2516 : return 0;
2517 : }
2518 :
2519 : static inline int hrtick_enabled_dl(struct rq *rq)
2520 : {
2521 : return 0;
2522 : }
2523 :
2524 : static inline int hrtick_enabled(struct rq *rq)
2525 : {
2526 : return 0;
2527 : }
2528 :
2529 : #endif /* CONFIG_SCHED_HRTICK */
2530 :
2531 : #ifndef arch_scale_freq_tick
2532 : static __always_inline
2533 : void arch_scale_freq_tick(void)
2534 : {
2535 : }
2536 : #endif
2537 :
2538 : #ifndef arch_scale_freq_capacity
2539 : /**
2540 : * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2541 : * @cpu: the CPU in question.
2542 : *
2543 : * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2544 : *
2545 : * f_curr
2546 : * ------ * SCHED_CAPACITY_SCALE
2547 : * f_max
2548 : */
2549 : static __always_inline
2550 : unsigned long arch_scale_freq_capacity(int cpu)
2551 : {
2552 : return SCHED_CAPACITY_SCALE;
2553 : }
2554 : #endif
2555 :
2556 : #ifdef CONFIG_SCHED_DEBUG
2557 : /*
2558 : * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2559 : * acquire rq lock instead of rq_lock(). So at the end of these two functions
2560 : * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2561 : * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2562 : */
2563 : static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2564 : {
2565 : rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2566 : /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2567 : #ifdef CONFIG_SMP
2568 : rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2569 : #endif
2570 : }
2571 : #else
2572 : static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2573 : #endif
2574 :
2575 : #ifdef CONFIG_SMP
2576 :
2577 : static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2578 : {
2579 : #ifdef CONFIG_SCHED_CORE
2580 : /*
2581 : * In order to not have {0,2},{1,3} turn into into an AB-BA,
2582 : * order by core-id first and cpu-id second.
2583 : *
2584 : * Notably:
2585 : *
2586 : * double_rq_lock(0,3); will take core-0, core-1 lock
2587 : * double_rq_lock(1,2); will take core-1, core-0 lock
2588 : *
2589 : * when only cpu-id is considered.
2590 : */
2591 : if (rq1->core->cpu < rq2->core->cpu)
2592 : return true;
2593 : if (rq1->core->cpu > rq2->core->cpu)
2594 : return false;
2595 :
2596 : /*
2597 : * __sched_core_flip() relies on SMT having cpu-id lock order.
2598 : */
2599 : #endif
2600 : return rq1->cpu < rq2->cpu;
2601 : }
2602 :
2603 : extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2604 :
2605 : #ifdef CONFIG_PREEMPTION
2606 :
2607 : /*
2608 : * fair double_lock_balance: Safely acquires both rq->locks in a fair
2609 : * way at the expense of forcing extra atomic operations in all
2610 : * invocations. This assures that the double_lock is acquired using the
2611 : * same underlying policy as the spinlock_t on this architecture, which
2612 : * reduces latency compared to the unfair variant below. However, it
2613 : * also adds more overhead and therefore may reduce throughput.
2614 : */
2615 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2616 : __releases(this_rq->lock)
2617 : __acquires(busiest->lock)
2618 : __acquires(this_rq->lock)
2619 : {
2620 : raw_spin_rq_unlock(this_rq);
2621 : double_rq_lock(this_rq, busiest);
2622 :
2623 : return 1;
2624 : }
2625 :
2626 : #else
2627 : /*
2628 : * Unfair double_lock_balance: Optimizes throughput at the expense of
2629 : * latency by eliminating extra atomic operations when the locks are
2630 : * already in proper order on entry. This favors lower CPU-ids and will
2631 : * grant the double lock to lower CPUs over higher ids under contention,
2632 : * regardless of entry order into the function.
2633 : */
2634 : static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2635 : __releases(this_rq->lock)
2636 : __acquires(busiest->lock)
2637 : __acquires(this_rq->lock)
2638 : {
2639 : if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2640 : likely(raw_spin_rq_trylock(busiest))) {
2641 : double_rq_clock_clear_update(this_rq, busiest);
2642 : return 0;
2643 : }
2644 :
2645 : if (rq_order_less(this_rq, busiest)) {
2646 : raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2647 : double_rq_clock_clear_update(this_rq, busiest);
2648 : return 0;
2649 : }
2650 :
2651 : raw_spin_rq_unlock(this_rq);
2652 : double_rq_lock(this_rq, busiest);
2653 :
2654 : return 1;
2655 : }
2656 :
2657 : #endif /* CONFIG_PREEMPTION */
2658 :
2659 : /*
2660 : * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2661 : */
2662 : static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2663 : {
2664 : lockdep_assert_irqs_disabled();
2665 :
2666 : return _double_lock_balance(this_rq, busiest);
2667 : }
2668 :
2669 : static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2670 : __releases(busiest->lock)
2671 : {
2672 : if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2673 : raw_spin_rq_unlock(busiest);
2674 : lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2675 : }
2676 :
2677 : static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2678 : {
2679 : if (l1 > l2)
2680 : swap(l1, l2);
2681 :
2682 : spin_lock(l1);
2683 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2684 : }
2685 :
2686 : static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2687 : {
2688 : if (l1 > l2)
2689 : swap(l1, l2);
2690 :
2691 : spin_lock_irq(l1);
2692 : spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2693 : }
2694 :
2695 : static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2696 : {
2697 : if (l1 > l2)
2698 : swap(l1, l2);
2699 :
2700 : raw_spin_lock(l1);
2701 : raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2702 : }
2703 :
2704 : /*
2705 : * double_rq_unlock - safely unlock two runqueues
2706 : *
2707 : * Note this does not restore interrupts like task_rq_unlock,
2708 : * you need to do so manually after calling.
2709 : */
2710 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2711 : __releases(rq1->lock)
2712 : __releases(rq2->lock)
2713 : {
2714 : if (__rq_lockp(rq1) != __rq_lockp(rq2))
2715 : raw_spin_rq_unlock(rq2);
2716 : else
2717 : __release(rq2->lock);
2718 : raw_spin_rq_unlock(rq1);
2719 : }
2720 :
2721 : extern void set_rq_online (struct rq *rq);
2722 : extern void set_rq_offline(struct rq *rq);
2723 : extern bool sched_smp_initialized;
2724 :
2725 : #else /* CONFIG_SMP */
2726 :
2727 : /*
2728 : * double_rq_lock - safely lock two runqueues
2729 : *
2730 : * Note this does not disable interrupts like task_rq_lock,
2731 : * you need to do so manually before calling.
2732 : */
2733 0 : static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2734 : __acquires(rq1->lock)
2735 : __acquires(rq2->lock)
2736 : {
2737 0 : WARN_ON_ONCE(!irqs_disabled());
2738 0 : WARN_ON_ONCE(rq1 != rq2);
2739 0 : raw_spin_rq_lock(rq1);
2740 : __acquire(rq2->lock); /* Fake it out ;) */
2741 0 : double_rq_clock_clear_update(rq1, rq2);
2742 0 : }
2743 :
2744 : /*
2745 : * double_rq_unlock - safely unlock two runqueues
2746 : *
2747 : * Note this does not restore interrupts like task_rq_unlock,
2748 : * you need to do so manually after calling.
2749 : */
2750 0 : static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2751 : __releases(rq1->lock)
2752 : __releases(rq2->lock)
2753 : {
2754 0 : WARN_ON_ONCE(rq1 != rq2);
2755 0 : raw_spin_rq_unlock(rq1);
2756 : __release(rq2->lock);
2757 0 : }
2758 :
2759 : #endif
2760 :
2761 : extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2762 : extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2763 :
2764 : #ifdef CONFIG_SCHED_DEBUG
2765 : extern bool sched_debug_verbose;
2766 :
2767 : extern void print_cfs_stats(struct seq_file *m, int cpu);
2768 : extern void print_rt_stats(struct seq_file *m, int cpu);
2769 : extern void print_dl_stats(struct seq_file *m, int cpu);
2770 : extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2771 : extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2772 : extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2773 :
2774 : extern void resched_latency_warn(int cpu, u64 latency);
2775 : #ifdef CONFIG_NUMA_BALANCING
2776 : extern void
2777 : show_numa_stats(struct task_struct *p, struct seq_file *m);
2778 : extern void
2779 : print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2780 : unsigned long tpf, unsigned long gsf, unsigned long gpf);
2781 : #endif /* CONFIG_NUMA_BALANCING */
2782 : #else
2783 : static inline void resched_latency_warn(int cpu, u64 latency) {}
2784 : #endif /* CONFIG_SCHED_DEBUG */
2785 :
2786 : extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2787 : extern void init_rt_rq(struct rt_rq *rt_rq);
2788 : extern void init_dl_rq(struct dl_rq *dl_rq);
2789 :
2790 : extern void cfs_bandwidth_usage_inc(void);
2791 : extern void cfs_bandwidth_usage_dec(void);
2792 :
2793 : #ifdef CONFIG_NO_HZ_COMMON
2794 : #define NOHZ_BALANCE_KICK_BIT 0
2795 : #define NOHZ_STATS_KICK_BIT 1
2796 : #define NOHZ_NEWILB_KICK_BIT 2
2797 : #define NOHZ_NEXT_KICK_BIT 3
2798 :
2799 : /* Run rebalance_domains() */
2800 : #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2801 : /* Update blocked load */
2802 : #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2803 : /* Update blocked load when entering idle */
2804 : #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2805 : /* Update nohz.next_balance */
2806 : #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2807 :
2808 : #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2809 :
2810 : #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2811 :
2812 : extern void nohz_balance_exit_idle(struct rq *rq);
2813 : #else
2814 : static inline void nohz_balance_exit_idle(struct rq *rq) { }
2815 : #endif
2816 :
2817 : #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2818 : extern void nohz_run_idle_balance(int cpu);
2819 : #else
2820 : static inline void nohz_run_idle_balance(int cpu) { }
2821 : #endif
2822 :
2823 : #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2824 : struct irqtime {
2825 : u64 total;
2826 : u64 tick_delta;
2827 : u64 irq_start_time;
2828 : struct u64_stats_sync sync;
2829 : };
2830 :
2831 : DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2832 :
2833 : /*
2834 : * Returns the irqtime minus the softirq time computed by ksoftirqd.
2835 : * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2836 : * and never move forward.
2837 : */
2838 : static inline u64 irq_time_read(int cpu)
2839 : {
2840 : struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2841 : unsigned int seq;
2842 : u64 total;
2843 :
2844 : do {
2845 : seq = __u64_stats_fetch_begin(&irqtime->sync);
2846 : total = irqtime->total;
2847 : } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2848 :
2849 : return total;
2850 : }
2851 : #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2852 :
2853 : #ifdef CONFIG_CPU_FREQ
2854 : DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2855 :
2856 : /**
2857 : * cpufreq_update_util - Take a note about CPU utilization changes.
2858 : * @rq: Runqueue to carry out the update for.
2859 : * @flags: Update reason flags.
2860 : *
2861 : * This function is called by the scheduler on the CPU whose utilization is
2862 : * being updated.
2863 : *
2864 : * It can only be called from RCU-sched read-side critical sections.
2865 : *
2866 : * The way cpufreq is currently arranged requires it to evaluate the CPU
2867 : * performance state (frequency/voltage) on a regular basis to prevent it from
2868 : * being stuck in a completely inadequate performance level for too long.
2869 : * That is not guaranteed to happen if the updates are only triggered from CFS
2870 : * and DL, though, because they may not be coming in if only RT tasks are
2871 : * active all the time (or there are RT tasks only).
2872 : *
2873 : * As a workaround for that issue, this function is called periodically by the
2874 : * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2875 : * but that really is a band-aid. Going forward it should be replaced with
2876 : * solutions targeted more specifically at RT tasks.
2877 : */
2878 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2879 : {
2880 : struct update_util_data *data;
2881 :
2882 : data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2883 : cpu_of(rq)));
2884 : if (data)
2885 : data->func(data, rq_clock(rq), flags);
2886 : }
2887 : #else
2888 : static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2889 : #endif /* CONFIG_CPU_FREQ */
2890 :
2891 : #ifdef arch_scale_freq_capacity
2892 : # ifndef arch_scale_freq_invariant
2893 : # define arch_scale_freq_invariant() true
2894 : # endif
2895 : #else
2896 : # define arch_scale_freq_invariant() false
2897 : #endif
2898 :
2899 : #ifdef CONFIG_SMP
2900 : static inline unsigned long capacity_orig_of(int cpu)
2901 : {
2902 : return cpu_rq(cpu)->cpu_capacity_orig;
2903 : }
2904 :
2905 : /**
2906 : * enum cpu_util_type - CPU utilization type
2907 : * @FREQUENCY_UTIL: Utilization used to select frequency
2908 : * @ENERGY_UTIL: Utilization used during energy calculation
2909 : *
2910 : * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2911 : * need to be aggregated differently depending on the usage made of them. This
2912 : * enum is used within effective_cpu_util() to differentiate the types of
2913 : * utilization expected by the callers, and adjust the aggregation accordingly.
2914 : */
2915 : enum cpu_util_type {
2916 : FREQUENCY_UTIL,
2917 : ENERGY_UTIL,
2918 : };
2919 :
2920 : unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2921 : enum cpu_util_type type,
2922 : struct task_struct *p);
2923 :
2924 : /*
2925 : * Verify the fitness of task @p to run on @cpu taking into account the
2926 : * CPU original capacity and the runtime/deadline ratio of the task.
2927 : *
2928 : * The function will return true if the original capacity of @cpu is
2929 : * greater than or equal to task's deadline density right shifted by
2930 : * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2931 : */
2932 : static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2933 : {
2934 : unsigned long cap = arch_scale_cpu_capacity(cpu);
2935 :
2936 : return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2937 : }
2938 :
2939 : static inline unsigned long cpu_bw_dl(struct rq *rq)
2940 : {
2941 : return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2942 : }
2943 :
2944 : static inline unsigned long cpu_util_dl(struct rq *rq)
2945 : {
2946 : return READ_ONCE(rq->avg_dl.util_avg);
2947 : }
2948 :
2949 : /**
2950 : * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2951 : * @cpu: the CPU to get the utilization for.
2952 : *
2953 : * The unit of the return value must be the same as the one of CPU capacity
2954 : * so that CPU utilization can be compared with CPU capacity.
2955 : *
2956 : * CPU utilization is the sum of running time of runnable tasks plus the
2957 : * recent utilization of currently non-runnable tasks on that CPU.
2958 : * It represents the amount of CPU capacity currently used by CFS tasks in
2959 : * the range [0..max CPU capacity] with max CPU capacity being the CPU
2960 : * capacity at f_max.
2961 : *
2962 : * The estimated CPU utilization is defined as the maximum between CPU
2963 : * utilization and sum of the estimated utilization of the currently
2964 : * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2965 : * previously-executed tasks, which helps better deduce how busy a CPU will
2966 : * be when a long-sleeping task wakes up. The contribution to CPU utilization
2967 : * of such a task would be significantly decayed at this point of time.
2968 : *
2969 : * CPU utilization can be higher than the current CPU capacity
2970 : * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2971 : * of rounding errors as well as task migrations or wakeups of new tasks.
2972 : * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2973 : * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2974 : * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2975 : * capacity. CPU utilization is allowed to overshoot current CPU capacity
2976 : * though since this is useful for predicting the CPU capacity required
2977 : * after task migrations (scheduler-driven DVFS).
2978 : *
2979 : * Return: (Estimated) utilization for the specified CPU.
2980 : */
2981 : static inline unsigned long cpu_util_cfs(int cpu)
2982 : {
2983 : struct cfs_rq *cfs_rq;
2984 : unsigned long util;
2985 :
2986 : cfs_rq = &cpu_rq(cpu)->cfs;
2987 : util = READ_ONCE(cfs_rq->avg.util_avg);
2988 :
2989 : if (sched_feat(UTIL_EST)) {
2990 : util = max_t(unsigned long, util,
2991 : READ_ONCE(cfs_rq->avg.util_est.enqueued));
2992 : }
2993 :
2994 : return min(util, capacity_orig_of(cpu));
2995 : }
2996 :
2997 : static inline unsigned long cpu_util_rt(struct rq *rq)
2998 : {
2999 : return READ_ONCE(rq->avg_rt.util_avg);
3000 : }
3001 : #endif
3002 :
3003 : #ifdef CONFIG_UCLAMP_TASK
3004 : unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3005 :
3006 : static inline unsigned long uclamp_rq_get(struct rq *rq,
3007 : enum uclamp_id clamp_id)
3008 : {
3009 : return READ_ONCE(rq->uclamp[clamp_id].value);
3010 : }
3011 :
3012 : static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3013 : unsigned int value)
3014 : {
3015 : WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3016 : }
3017 :
3018 : static inline bool uclamp_rq_is_idle(struct rq *rq)
3019 : {
3020 : return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3021 : }
3022 :
3023 : /**
3024 : * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3025 : * @rq: The rq to clamp against. Must not be NULL.
3026 : * @util: The util value to clamp.
3027 : * @p: The task to clamp against. Can be NULL if you want to clamp
3028 : * against @rq only.
3029 : *
3030 : * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3031 : *
3032 : * If sched_uclamp_used static key is disabled, then just return the util
3033 : * without any clamping since uclamp aggregation at the rq level in the fast
3034 : * path is disabled, rendering this operation a NOP.
3035 : *
3036 : * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3037 : * will return the correct effective uclamp value of the task even if the
3038 : * static key is disabled.
3039 : */
3040 : static __always_inline
3041 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3042 : struct task_struct *p)
3043 : {
3044 : unsigned long min_util = 0;
3045 : unsigned long max_util = 0;
3046 :
3047 : if (!static_branch_likely(&sched_uclamp_used))
3048 : return util;
3049 :
3050 : if (p) {
3051 : min_util = uclamp_eff_value(p, UCLAMP_MIN);
3052 : max_util = uclamp_eff_value(p, UCLAMP_MAX);
3053 :
3054 : /*
3055 : * Ignore last runnable task's max clamp, as this task will
3056 : * reset it. Similarly, no need to read the rq's min clamp.
3057 : */
3058 : if (uclamp_rq_is_idle(rq))
3059 : goto out;
3060 : }
3061 :
3062 : min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3063 : max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3064 : out:
3065 : /*
3066 : * Since CPU's {min,max}_util clamps are MAX aggregated considering
3067 : * RUNNABLE tasks with _different_ clamps, we can end up with an
3068 : * inversion. Fix it now when the clamps are applied.
3069 : */
3070 : if (unlikely(min_util >= max_util))
3071 : return min_util;
3072 :
3073 : return clamp(util, min_util, max_util);
3074 : }
3075 :
3076 : /* Is the rq being capped/throttled by uclamp_max? */
3077 : static inline bool uclamp_rq_is_capped(struct rq *rq)
3078 : {
3079 : unsigned long rq_util;
3080 : unsigned long max_util;
3081 :
3082 : if (!static_branch_likely(&sched_uclamp_used))
3083 : return false;
3084 :
3085 : rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3086 : max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3087 :
3088 : return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3089 : }
3090 :
3091 : /*
3092 : * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3093 : * by default in the fast path and only gets turned on once userspace performs
3094 : * an operation that requires it.
3095 : *
3096 : * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3097 : * hence is active.
3098 : */
3099 : static inline bool uclamp_is_used(void)
3100 : {
3101 : return static_branch_likely(&sched_uclamp_used);
3102 : }
3103 : #else /* CONFIG_UCLAMP_TASK */
3104 : static inline unsigned long uclamp_eff_value(struct task_struct *p,
3105 : enum uclamp_id clamp_id)
3106 : {
3107 : if (clamp_id == UCLAMP_MIN)
3108 : return 0;
3109 :
3110 : return SCHED_CAPACITY_SCALE;
3111 : }
3112 :
3113 : static inline
3114 : unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3115 : struct task_struct *p)
3116 : {
3117 : return util;
3118 : }
3119 :
3120 : static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3121 :
3122 : static inline bool uclamp_is_used(void)
3123 : {
3124 : return false;
3125 : }
3126 :
3127 : static inline unsigned long uclamp_rq_get(struct rq *rq,
3128 : enum uclamp_id clamp_id)
3129 : {
3130 : if (clamp_id == UCLAMP_MIN)
3131 : return 0;
3132 :
3133 : return SCHED_CAPACITY_SCALE;
3134 : }
3135 :
3136 : static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3137 : unsigned int value)
3138 : {
3139 : }
3140 :
3141 : static inline bool uclamp_rq_is_idle(struct rq *rq)
3142 : {
3143 : return false;
3144 : }
3145 : #endif /* CONFIG_UCLAMP_TASK */
3146 :
3147 : #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3148 : static inline unsigned long cpu_util_irq(struct rq *rq)
3149 : {
3150 : return rq->avg_irq.util_avg;
3151 : }
3152 :
3153 : static inline
3154 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3155 : {
3156 : util *= (max - irq);
3157 : util /= max;
3158 :
3159 : return util;
3160 :
3161 : }
3162 : #else
3163 : static inline unsigned long cpu_util_irq(struct rq *rq)
3164 : {
3165 : return 0;
3166 : }
3167 :
3168 : static inline
3169 : unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3170 : {
3171 : return util;
3172 : }
3173 : #endif
3174 :
3175 : #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3176 :
3177 : #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3178 :
3179 : DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3180 :
3181 : static inline bool sched_energy_enabled(void)
3182 : {
3183 : return static_branch_unlikely(&sched_energy_present);
3184 : }
3185 :
3186 : #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3187 :
3188 : #define perf_domain_span(pd) NULL
3189 : static inline bool sched_energy_enabled(void) { return false; }
3190 :
3191 : #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3192 :
3193 : #ifdef CONFIG_MEMBARRIER
3194 : /*
3195 : * The scheduler provides memory barriers required by membarrier between:
3196 : * - prior user-space memory accesses and store to rq->membarrier_state,
3197 : * - store to rq->membarrier_state and following user-space memory accesses.
3198 : * In the same way it provides those guarantees around store to rq->curr.
3199 : */
3200 : static inline void membarrier_switch_mm(struct rq *rq,
3201 : struct mm_struct *prev_mm,
3202 : struct mm_struct *next_mm)
3203 : {
3204 : int membarrier_state;
3205 :
3206 0 : if (prev_mm == next_mm)
3207 : return;
3208 :
3209 0 : membarrier_state = atomic_read(&next_mm->membarrier_state);
3210 0 : if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3211 : return;
3212 :
3213 0 : WRITE_ONCE(rq->membarrier_state, membarrier_state);
3214 : }
3215 : #else
3216 : static inline void membarrier_switch_mm(struct rq *rq,
3217 : struct mm_struct *prev_mm,
3218 : struct mm_struct *next_mm)
3219 : {
3220 : }
3221 : #endif
3222 :
3223 : #ifdef CONFIG_SMP
3224 : static inline bool is_per_cpu_kthread(struct task_struct *p)
3225 : {
3226 : if (!(p->flags & PF_KTHREAD))
3227 : return false;
3228 :
3229 : if (p->nr_cpus_allowed != 1)
3230 : return false;
3231 :
3232 : return true;
3233 : }
3234 : #endif
3235 :
3236 : extern void swake_up_all_locked(struct swait_queue_head *q);
3237 : extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3238 :
3239 : #ifdef CONFIG_PREEMPT_DYNAMIC
3240 : extern int preempt_dynamic_mode;
3241 : extern int sched_dynamic_mode(const char *str);
3242 : extern void sched_dynamic_update(int mode);
3243 : #endif
3244 :
3245 : static inline void update_current_exec_runtime(struct task_struct *curr,
3246 : u64 now, u64 delta_exec)
3247 : {
3248 0 : curr->se.sum_exec_runtime += delta_exec;
3249 0 : account_group_exec_runtime(curr, delta_exec);
3250 :
3251 0 : curr->se.exec_start = now;
3252 0 : cgroup_account_cputime(curr, delta_exec);
3253 : }
3254 :
3255 : #ifdef CONFIG_SCHED_MM_CID
3256 :
3257 : #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3258 : #define MM_CID_SCAN_DELAY 100 /* 100ms */
3259 :
3260 : extern raw_spinlock_t cid_lock;
3261 : extern int use_cid_lock;
3262 :
3263 : extern void sched_mm_cid_migrate_from(struct task_struct *t);
3264 : extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3265 : extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3266 : extern void init_sched_mm_cid(struct task_struct *t);
3267 :
3268 : static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3269 : {
3270 : if (cid < 0)
3271 : return;
3272 : cpumask_clear_cpu(cid, mm_cidmask(mm));
3273 : }
3274 :
3275 : /*
3276 : * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3277 : * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3278 : * be held to transition to other states.
3279 : *
3280 : * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3281 : * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3282 : */
3283 : static inline void mm_cid_put_lazy(struct task_struct *t)
3284 : {
3285 : struct mm_struct *mm = t->mm;
3286 : struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3287 : int cid;
3288 :
3289 : lockdep_assert_irqs_disabled();
3290 : cid = __this_cpu_read(pcpu_cid->cid);
3291 : if (!mm_cid_is_lazy_put(cid) ||
3292 : !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3293 : return;
3294 : __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3295 : }
3296 :
3297 : static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3298 : {
3299 : struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3300 : int cid, res;
3301 :
3302 : lockdep_assert_irqs_disabled();
3303 : cid = __this_cpu_read(pcpu_cid->cid);
3304 : for (;;) {
3305 : if (mm_cid_is_unset(cid))
3306 : return MM_CID_UNSET;
3307 : /*
3308 : * Attempt transition from valid or lazy-put to unset.
3309 : */
3310 : res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3311 : if (res == cid)
3312 : break;
3313 : cid = res;
3314 : }
3315 : return cid;
3316 : }
3317 :
3318 : static inline void mm_cid_put(struct mm_struct *mm)
3319 : {
3320 : int cid;
3321 :
3322 : lockdep_assert_irqs_disabled();
3323 : cid = mm_cid_pcpu_unset(mm);
3324 : if (cid == MM_CID_UNSET)
3325 : return;
3326 : __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3327 : }
3328 :
3329 : static inline int __mm_cid_try_get(struct mm_struct *mm)
3330 : {
3331 : struct cpumask *cpumask;
3332 : int cid;
3333 :
3334 : cpumask = mm_cidmask(mm);
3335 : /*
3336 : * Retry finding first zero bit if the mask is temporarily
3337 : * filled. This only happens during concurrent remote-clear
3338 : * which owns a cid without holding a rq lock.
3339 : */
3340 : for (;;) {
3341 : cid = cpumask_first_zero(cpumask);
3342 : if (cid < nr_cpu_ids)
3343 : break;
3344 : cpu_relax();
3345 : }
3346 : if (cpumask_test_and_set_cpu(cid, cpumask))
3347 : return -1;
3348 : return cid;
3349 : }
3350 :
3351 : /*
3352 : * Save a snapshot of the current runqueue time of this cpu
3353 : * with the per-cpu cid value, allowing to estimate how recently it was used.
3354 : */
3355 : static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3356 : {
3357 : struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3358 :
3359 : lockdep_assert_rq_held(rq);
3360 : WRITE_ONCE(pcpu_cid->time, rq->clock);
3361 : }
3362 :
3363 : static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3364 : {
3365 : int cid;
3366 :
3367 : /*
3368 : * All allocations (even those using the cid_lock) are lock-free. If
3369 : * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3370 : * guarantee forward progress.
3371 : */
3372 : if (!READ_ONCE(use_cid_lock)) {
3373 : cid = __mm_cid_try_get(mm);
3374 : if (cid >= 0)
3375 : goto end;
3376 : raw_spin_lock(&cid_lock);
3377 : } else {
3378 : raw_spin_lock(&cid_lock);
3379 : cid = __mm_cid_try_get(mm);
3380 : if (cid >= 0)
3381 : goto unlock;
3382 : }
3383 :
3384 : /*
3385 : * cid concurrently allocated. Retry while forcing following
3386 : * allocations to use the cid_lock to ensure forward progress.
3387 : */
3388 : WRITE_ONCE(use_cid_lock, 1);
3389 : /*
3390 : * Set use_cid_lock before allocation. Only care about program order
3391 : * because this is only required for forward progress.
3392 : */
3393 : barrier();
3394 : /*
3395 : * Retry until it succeeds. It is guaranteed to eventually succeed once
3396 : * all newcoming allocations observe the use_cid_lock flag set.
3397 : */
3398 : do {
3399 : cid = __mm_cid_try_get(mm);
3400 : cpu_relax();
3401 : } while (cid < 0);
3402 : /*
3403 : * Allocate before clearing use_cid_lock. Only care about
3404 : * program order because this is for forward progress.
3405 : */
3406 : barrier();
3407 : WRITE_ONCE(use_cid_lock, 0);
3408 : unlock:
3409 : raw_spin_unlock(&cid_lock);
3410 : end:
3411 : mm_cid_snapshot_time(rq, mm);
3412 : return cid;
3413 : }
3414 :
3415 : static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3416 : {
3417 : struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3418 : struct cpumask *cpumask;
3419 : int cid;
3420 :
3421 : lockdep_assert_rq_held(rq);
3422 : cpumask = mm_cidmask(mm);
3423 : cid = __this_cpu_read(pcpu_cid->cid);
3424 : if (mm_cid_is_valid(cid)) {
3425 : mm_cid_snapshot_time(rq, mm);
3426 : return cid;
3427 : }
3428 : if (mm_cid_is_lazy_put(cid)) {
3429 : if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3430 : __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3431 : }
3432 : cid = __mm_cid_get(rq, mm);
3433 : __this_cpu_write(pcpu_cid->cid, cid);
3434 : return cid;
3435 : }
3436 :
3437 : static inline void switch_mm_cid(struct rq *rq,
3438 : struct task_struct *prev,
3439 : struct task_struct *next)
3440 : {
3441 : /*
3442 : * Provide a memory barrier between rq->curr store and load of
3443 : * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3444 : *
3445 : * Should be adapted if context_switch() is modified.
3446 : */
3447 : if (!next->mm) { // to kernel
3448 : /*
3449 : * user -> kernel transition does not guarantee a barrier, but
3450 : * we can use the fact that it performs an atomic operation in
3451 : * mmgrab().
3452 : */
3453 : if (prev->mm) // from user
3454 : smp_mb__after_mmgrab();
3455 : /*
3456 : * kernel -> kernel transition does not change rq->curr->mm
3457 : * state. It stays NULL.
3458 : */
3459 : } else { // to user
3460 : /*
3461 : * kernel -> user transition does not provide a barrier
3462 : * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3463 : * Provide it here.
3464 : */
3465 : if (!prev->mm) // from kernel
3466 : smp_mb();
3467 : /*
3468 : * user -> user transition guarantees a memory barrier through
3469 : * switch_mm() when current->mm changes. If current->mm is
3470 : * unchanged, no barrier is needed.
3471 : */
3472 : }
3473 : if (prev->mm_cid_active) {
3474 : mm_cid_snapshot_time(rq, prev->mm);
3475 : mm_cid_put_lazy(prev);
3476 : prev->mm_cid = -1;
3477 : }
3478 : if (next->mm_cid_active)
3479 : next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3480 : }
3481 :
3482 : #else
3483 : static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3484 : static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3485 : static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3486 : static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3487 : static inline void init_sched_mm_cid(struct task_struct *t) { }
3488 : #endif
3489 :
3490 : #endif /* _KERNEL_SCHED_SCHED_H */
|
