1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Scheduler internal types and methods:
4 */
5#ifndef _KERNEL_SCHED_SCHED_H
6#define _KERNEL_SCHED_SCHED_H
7
8#include <linux/sched/affinity.h>
9#include <linux/sched/autogroup.h>
10#include <linux/sched/cpufreq.h>
11#include <linux/sched/deadline.h>
12#include <linux/sched.h>
13#include <linux/sched/loadavg.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/rseq_api.h>
16#include <linux/sched/signal.h>
17#include <linux/sched/smt.h>
18#include <linux/sched/stat.h>
19#include <linux/sched/sysctl.h>
20#include <linux/sched/task_flags.h>
21#include <linux/sched/task.h>
22#include <linux/sched/topology.h>
23
24#include <linux/atomic.h>
25#include <linux/bitmap.h>
26#include <linux/bug.h>
27#include <linux/capability.h>
28#include <linux/cgroup_api.h>
29#include <linux/cgroup.h>
30#include <linux/context_tracking.h>
31#include <linux/cpufreq.h>
32#include <linux/cpumask_api.h>
33#include <linux/ctype.h>
34#include <linux/file.h>
35#include <linux/fs_api.h>
36#include <linux/hrtimer_api.h>
37#include <linux/interrupt.h>
38#include <linux/irq_work.h>
39#include <linux/jiffies.h>
40#include <linux/kref_api.h>
41#include <linux/kthread.h>
42#include <linux/ktime_api.h>
43#include <linux/lockdep_api.h>
44#include <linux/lockdep.h>
45#include <linux/minmax.h>
46#include <linux/mm.h>
47#include <linux/module.h>
48#include <linux/mutex_api.h>
49#include <linux/plist.h>
50#include <linux/poll.h>
51#include <linux/proc_fs.h>
52#include <linux/profile.h>
53#include <linux/psi.h>
54#include <linux/rcupdate.h>
55#include <linux/seq_file.h>
56#include <linux/seqlock.h>
57#include <linux/softirq.h>
58#include <linux/spinlock_api.h>
59#include <linux/static_key.h>
60#include <linux/stop_machine.h>
61#include <linux/syscalls_api.h>
62#include <linux/syscalls.h>
63#include <linux/tick.h>
64#include <linux/topology.h>
65#include <linux/types.h>
66#include <linux/u64_stats_sync_api.h>
67#include <linux/uaccess.h>
68#include <linux/wait_api.h>
69#include <linux/wait_bit.h>
70#include <linux/workqueue_api.h>
71
72#include <trace/events/power.h>
73#include <trace/events/sched.h>
74
75#include "../workqueue_internal.h"
76
77#ifdef CONFIG_PARAVIRT
78# include <asm/paravirt.h>
79# include <asm/paravirt_api_clock.h>
80#endif
81
82#include "cpupri.h"
83#include "cpudeadline.h"
84
85#ifdef CONFIG_SCHED_DEBUG
86# define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
87#else
88# define SCHED_WARN_ON(x) ({ (void)(x), 0; })
89#endif
90
91struct rq;
92struct cpuidle_state;
93
94/* task_struct::on_rq states: */
95#define TASK_ON_RQ_QUEUED 1
96#define TASK_ON_RQ_MIGRATING 2
97
98extern __read_mostly int scheduler_running;
99
100extern unsigned long calc_load_update;
101extern atomic_long_t calc_load_tasks;
102
103extern void calc_global_load_tick(struct rq *this_rq);
104extern long calc_load_fold_active(struct rq *this_rq, long adjust);
105
106extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
107
108extern int sysctl_sched_rt_period;
109extern int sysctl_sched_rt_runtime;
110extern int sched_rr_timeslice;
111
112/*
113 * Helpers for converting nanosecond timing to jiffy resolution
114 */
115#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
116
117/*
118 * Increase resolution of nice-level calculations for 64-bit architectures.
119 * The extra resolution improves shares distribution and load balancing of
120 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
121 * hierarchies, especially on larger systems. This is not a user-visible change
122 * and does not change the user-interface for setting shares/weights.
123 *
124 * We increase resolution only if we have enough bits to allow this increased
125 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
126 * are pretty high and the returns do not justify the increased costs.
127 *
128 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
129 * increase coverage and consistency always enable it on 64-bit platforms.
130 */
131#ifdef CONFIG_64BIT
132# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
133# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
134# define scale_load_down(w) \
135({ \
136 unsigned long __w = (w); \
137 if (__w) \
138 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
139 __w; \
140})
141#else
142# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
143# define scale_load(w) (w)
144# define scale_load_down(w) (w)
145#endif
146
147/*
148 * Task weight (visible to users) and its load (invisible to users) have
149 * independent resolution, but they should be well calibrated. We use
150 * scale_load() and scale_load_down(w) to convert between them. The
151 * following must be true:
152 *
153 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
154 *
155 */
156#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
157
158/*
159 * Single value that decides SCHED_DEADLINE internal math precision.
160 * 10 -> just above 1us
161 * 9 -> just above 0.5us
162 */
163#define DL_SCALE 10
164
165/*
166 * Single value that denotes runtime == period, ie unlimited time.
167 */
168#define RUNTIME_INF ((u64)~0ULL)
169
170static inline int idle_policy(int policy)
171{
172 return policy == SCHED_IDLE;
173}
174static inline int fair_policy(int policy)
175{
176 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
177}
178
179static inline int rt_policy(int policy)
180{
181 return policy == SCHED_FIFO || policy == SCHED_RR;
182}
183
184static inline int dl_policy(int policy)
185{
186 return policy == SCHED_DEADLINE;
187}
188static inline bool valid_policy(int policy)
189{
190 return idle_policy(policy) || fair_policy(policy) ||
191 rt_policy(policy) || dl_policy(policy);
192}
193
194static inline int task_has_idle_policy(struct task_struct *p)
195{
196 return idle_policy(policy: p->policy);
197}
198
199static inline int task_has_rt_policy(struct task_struct *p)
200{
201 return rt_policy(policy: p->policy);
202}
203
204static inline int task_has_dl_policy(struct task_struct *p)
205{
206 return dl_policy(policy: p->policy);
207}
208
209#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
210
211static inline void update_avg(u64 *avg, u64 sample)
212{
213 s64 diff = sample - *avg;
214 *avg += diff / 8;
215}
216
217/*
218 * Shifting a value by an exponent greater *or equal* to the size of said value
219 * is UB; cap at size-1.
220 */
221#define shr_bound(val, shift) \
222 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
223
224/*
225 * !! For sched_setattr_nocheck() (kernel) only !!
226 *
227 * This is actually gross. :(
228 *
229 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
230 * tasks, but still be able to sleep. We need this on platforms that cannot
231 * atomically change clock frequency. Remove once fast switching will be
232 * available on such platforms.
233 *
234 * SUGOV stands for SchedUtil GOVernor.
235 */
236#define SCHED_FLAG_SUGOV 0x10000000
237
238#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
239
240static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
241{
242#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
243 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
244#else
245 return false;
246#endif
247}
248
249/*
250 * Tells if entity @a should preempt entity @b.
251 */
252static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
253 const struct sched_dl_entity *b)
254{
255 return dl_entity_is_special(dl_se: a) ||
256 dl_time_before(a: a->deadline, b: b->deadline);
257}
258
259/*
260 * This is the priority-queue data structure of the RT scheduling class:
261 */
262struct rt_prio_array {
263 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
264 struct list_head queue[MAX_RT_PRIO];
265};
266
267struct rt_bandwidth {
268 /* nests inside the rq lock: */
269 raw_spinlock_t rt_runtime_lock;
270 ktime_t rt_period;
271 u64 rt_runtime;
272 struct hrtimer rt_period_timer;
273 unsigned int rt_period_active;
274};
275
276void __dl_clear_params(struct task_struct *p);
277
278static inline int dl_bandwidth_enabled(void)
279{
280 return sysctl_sched_rt_runtime >= 0;
281}
282
283/*
284 * To keep the bandwidth of -deadline tasks under control
285 * we need some place where:
286 * - store the maximum -deadline bandwidth of each cpu;
287 * - cache the fraction of bandwidth that is currently allocated in
288 * each root domain;
289 *
290 * This is all done in the data structure below. It is similar to the
291 * one used for RT-throttling (rt_bandwidth), with the main difference
292 * that, since here we are only interested in admission control, we
293 * do not decrease any runtime while the group "executes", neither we
294 * need a timer to replenish it.
295 *
296 * With respect to SMP, bandwidth is given on a per root domain basis,
297 * meaning that:
298 * - bw (< 100%) is the deadline bandwidth of each CPU;
299 * - total_bw is the currently allocated bandwidth in each root domain;
300 */
301struct dl_bw {
302 raw_spinlock_t lock;
303 u64 bw;
304 u64 total_bw;
305};
306
307extern void init_dl_bw(struct dl_bw *dl_b);
308extern int sched_dl_global_validate(void);
309extern void sched_dl_do_global(void);
310extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
311extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
312extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
313extern bool __checkparam_dl(const struct sched_attr *attr);
314extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
315extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
316extern int dl_bw_check_overflow(int cpu);
317
318#ifdef CONFIG_CGROUP_SCHED
319
320struct cfs_rq;
321struct rt_rq;
322
323extern struct list_head task_groups;
324
325struct cfs_bandwidth {
326#ifdef CONFIG_CFS_BANDWIDTH
327 raw_spinlock_t lock;
328 ktime_t period;
329 u64 quota;
330 u64 runtime;
331 u64 burst;
332 u64 runtime_snap;
333 s64 hierarchical_quota;
334
335 u8 idle;
336 u8 period_active;
337 u8 slack_started;
338 struct hrtimer period_timer;
339 struct hrtimer slack_timer;
340 struct list_head throttled_cfs_rq;
341
342 /* Statistics: */
343 int nr_periods;
344 int nr_throttled;
345 int nr_burst;
346 u64 throttled_time;
347 u64 burst_time;
348#endif
349};
350
351/* Task group related information */
352struct task_group {
353 struct cgroup_subsys_state css;
354
355#ifdef CONFIG_FAIR_GROUP_SCHED
356 /* schedulable entities of this group on each CPU */
357 struct sched_entity **se;
358 /* runqueue "owned" by this group on each CPU */
359 struct cfs_rq **cfs_rq;
360 unsigned long shares;
361
362 /* A positive value indicates that this is a SCHED_IDLE group. */
363 int idle;
364
365#ifdef CONFIG_SMP
366 /*
367 * load_avg can be heavily contended at clock tick time, so put
368 * it in its own cacheline separated from the fields above which
369 * will also be accessed at each tick.
370 */
371 atomic_long_t load_avg ____cacheline_aligned;
372#endif
373#endif
374
375#ifdef CONFIG_RT_GROUP_SCHED
376 struct sched_rt_entity **rt_se;
377 struct rt_rq **rt_rq;
378
379 struct rt_bandwidth rt_bandwidth;
380#endif
381
382 struct rcu_head rcu;
383 struct list_head list;
384
385 struct task_group *parent;
386 struct list_head siblings;
387 struct list_head children;
388
389#ifdef CONFIG_SCHED_AUTOGROUP
390 struct autogroup *autogroup;
391#endif
392
393 struct cfs_bandwidth cfs_bandwidth;
394
395#ifdef CONFIG_UCLAMP_TASK_GROUP
396 /* The two decimal precision [%] value requested from user-space */
397 unsigned int uclamp_pct[UCLAMP_CNT];
398 /* Clamp values requested for a task group */
399 struct uclamp_se uclamp_req[UCLAMP_CNT];
400 /* Effective clamp values used for a task group */
401 struct uclamp_se uclamp[UCLAMP_CNT];
402#endif
403
404};
405
406#ifdef CONFIG_FAIR_GROUP_SCHED
407#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
408
409/*
410 * A weight of 0 or 1 can cause arithmetics problems.
411 * A weight of a cfs_rq is the sum of weights of which entities
412 * are queued on this cfs_rq, so a weight of a entity should not be
413 * too large, so as the shares value of a task group.
414 * (The default weight is 1024 - so there's no practical
415 * limitation from this.)
416 */
417#define MIN_SHARES (1UL << 1)
418#define MAX_SHARES (1UL << 18)
419#endif
420
421typedef int (*tg_visitor)(struct task_group *, void *);
422
423extern int walk_tg_tree_from(struct task_group *from,
424 tg_visitor down, tg_visitor up, void *data);
425
426/*
427 * Iterate the full tree, calling @down when first entering a node and @up when
428 * leaving it for the final time.
429 *
430 * Caller must hold rcu_lock or sufficient equivalent.
431 */
432static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
433{
434 return walk_tg_tree_from(from: &root_task_group, down, up, data);
435}
436
437extern int tg_nop(struct task_group *tg, void *data);
438
439extern void free_fair_sched_group(struct task_group *tg);
440extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
441extern void online_fair_sched_group(struct task_group *tg);
442extern void unregister_fair_sched_group(struct task_group *tg);
443extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
444 struct sched_entity *se, int cpu,
445 struct sched_entity *parent);
446extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
447
448extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
449extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
450extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
451extern bool cfs_task_bw_constrained(struct task_struct *p);
452
453extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
454 struct sched_rt_entity *rt_se, int cpu,
455 struct sched_rt_entity *parent);
456extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
457extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
458extern long sched_group_rt_runtime(struct task_group *tg);
459extern long sched_group_rt_period(struct task_group *tg);
460extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
461
462extern struct task_group *sched_create_group(struct task_group *parent);
463extern void sched_online_group(struct task_group *tg,
464 struct task_group *parent);
465extern void sched_destroy_group(struct task_group *tg);
466extern void sched_release_group(struct task_group *tg);
467
468extern void sched_move_task(struct task_struct *tsk);
469
470#ifdef CONFIG_FAIR_GROUP_SCHED
471extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
472
473extern int sched_group_set_idle(struct task_group *tg, long idle);
474
475#ifdef CONFIG_SMP
476extern void set_task_rq_fair(struct sched_entity *se,
477 struct cfs_rq *prev, struct cfs_rq *next);
478#else /* !CONFIG_SMP */
479static inline void set_task_rq_fair(struct sched_entity *se,
480 struct cfs_rq *prev, struct cfs_rq *next) { }
481#endif /* CONFIG_SMP */
482#endif /* CONFIG_FAIR_GROUP_SCHED */
483
484#else /* CONFIG_CGROUP_SCHED */
485
486struct cfs_bandwidth { };
487static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
488
489#endif /* CONFIG_CGROUP_SCHED */
490
491extern void unregister_rt_sched_group(struct task_group *tg);
492extern void free_rt_sched_group(struct task_group *tg);
493extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
494
495/*
496 * u64_u32_load/u64_u32_store
497 *
498 * Use a copy of a u64 value to protect against data race. This is only
499 * applicable for 32-bits architectures.
500 */
501#ifdef CONFIG_64BIT
502# define u64_u32_load_copy(var, copy) var
503# define u64_u32_store_copy(var, copy, val) (var = val)
504#else
505# define u64_u32_load_copy(var, copy) \
506({ \
507 u64 __val, __val_copy; \
508 do { \
509 __val_copy = copy; \
510 /* \
511 * paired with u64_u32_store_copy(), ordering access \
512 * to var and copy. \
513 */ \
514 smp_rmb(); \
515 __val = var; \
516 } while (__val != __val_copy); \
517 __val; \
518})
519# define u64_u32_store_copy(var, copy, val) \
520do { \
521 typeof(val) __val = (val); \
522 var = __val; \
523 /* \
524 * paired with u64_u32_load_copy(), ordering access to var and \
525 * copy. \
526 */ \
527 smp_wmb(); \
528 copy = __val; \
529} while (0)
530#endif
531# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
532# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
533
534/* CFS-related fields in a runqueue */
535struct cfs_rq {
536 struct load_weight load;
537 unsigned int nr_running;
538 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
539 unsigned int idle_nr_running; /* SCHED_IDLE */
540 unsigned int idle_h_nr_running; /* SCHED_IDLE */
541
542 s64 avg_vruntime;
543 u64 avg_load;
544
545 u64 exec_clock;
546 u64 min_vruntime;
547#ifdef CONFIG_SCHED_CORE
548 unsigned int forceidle_seq;
549 u64 min_vruntime_fi;
550#endif
551
552#ifndef CONFIG_64BIT
553 u64 min_vruntime_copy;
554#endif
555
556 struct rb_root_cached tasks_timeline;
557
558 /*
559 * 'curr' points to currently running entity on this cfs_rq.
560 * It is set to NULL otherwise (i.e when none are currently running).
561 */
562 struct sched_entity *curr;
563 struct sched_entity *next;
564
565#ifdef CONFIG_SCHED_DEBUG
566 unsigned int nr_spread_over;
567#endif
568
569#ifdef CONFIG_SMP
570 /*
571 * CFS load tracking
572 */
573 struct sched_avg avg;
574#ifndef CONFIG_64BIT
575 u64 last_update_time_copy;
576#endif
577 struct {
578 raw_spinlock_t lock ____cacheline_aligned;
579 int nr;
580 unsigned long load_avg;
581 unsigned long util_avg;
582 unsigned long runnable_avg;
583 } removed;
584
585#ifdef CONFIG_FAIR_GROUP_SCHED
586 u64 last_update_tg_load_avg;
587 unsigned long tg_load_avg_contrib;
588 long propagate;
589 long prop_runnable_sum;
590
591 /*
592 * h_load = weight * f(tg)
593 *
594 * Where f(tg) is the recursive weight fraction assigned to
595 * this group.
596 */
597 unsigned long h_load;
598 u64 last_h_load_update;
599 struct sched_entity *h_load_next;
600#endif /* CONFIG_FAIR_GROUP_SCHED */
601#endif /* CONFIG_SMP */
602
603#ifdef CONFIG_FAIR_GROUP_SCHED
604 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
605
606 /*
607 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
608 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
609 * (like users, containers etc.)
610 *
611 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
612 * This list is used during load balance.
613 */
614 int on_list;
615 struct list_head leaf_cfs_rq_list;
616 struct task_group *tg; /* group that "owns" this runqueue */
617
618 /* Locally cached copy of our task_group's idle value */
619 int idle;
620
621#ifdef CONFIG_CFS_BANDWIDTH
622 int runtime_enabled;
623 s64 runtime_remaining;
624
625 u64 throttled_pelt_idle;
626#ifndef CONFIG_64BIT
627 u64 throttled_pelt_idle_copy;
628#endif
629 u64 throttled_clock;
630 u64 throttled_clock_pelt;
631 u64 throttled_clock_pelt_time;
632 u64 throttled_clock_self;
633 u64 throttled_clock_self_time;
634 int throttled;
635 int throttle_count;
636 struct list_head throttled_list;
637 struct list_head throttled_csd_list;
638#endif /* CONFIG_CFS_BANDWIDTH */
639#endif /* CONFIG_FAIR_GROUP_SCHED */
640};
641
642static inline int rt_bandwidth_enabled(void)
643{
644 return sysctl_sched_rt_runtime >= 0;
645}
646
647/* RT IPI pull logic requires IRQ_WORK */
648#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
649# define HAVE_RT_PUSH_IPI
650#endif
651
652/* Real-Time classes' related field in a runqueue: */
653struct rt_rq {
654 struct rt_prio_array active;
655 unsigned int rt_nr_running;
656 unsigned int rr_nr_running;
657#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
658 struct {
659 int curr; /* highest queued rt task prio */
660#ifdef CONFIG_SMP
661 int next; /* next highest */
662#endif
663 } highest_prio;
664#endif
665#ifdef CONFIG_SMP
666 int overloaded;
667 struct plist_head pushable_tasks;
668
669#endif /* CONFIG_SMP */
670 int rt_queued;
671
672 int rt_throttled;
673 u64 rt_time;
674 u64 rt_runtime;
675 /* Nests inside the rq lock: */
676 raw_spinlock_t rt_runtime_lock;
677
678#ifdef CONFIG_RT_GROUP_SCHED
679 unsigned int rt_nr_boosted;
680
681 struct rq *rq;
682 struct task_group *tg;
683#endif
684};
685
686static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
687{
688 return rt_rq->rt_queued && rt_rq->rt_nr_running;
689}
690
691/* Deadline class' related fields in a runqueue */
692struct dl_rq {
693 /* runqueue is an rbtree, ordered by deadline */
694 struct rb_root_cached root;
695
696 unsigned int dl_nr_running;
697
698#ifdef CONFIG_SMP
699 /*
700 * Deadline values of the currently executing and the
701 * earliest ready task on this rq. Caching these facilitates
702 * the decision whether or not a ready but not running task
703 * should migrate somewhere else.
704 */
705 struct {
706 u64 curr;
707 u64 next;
708 } earliest_dl;
709
710 int overloaded;
711
712 /*
713 * Tasks on this rq that can be pushed away. They are kept in
714 * an rb-tree, ordered by tasks' deadlines, with caching
715 * of the leftmost (earliest deadline) element.
716 */
717 struct rb_root_cached pushable_dl_tasks_root;
718#else
719 struct dl_bw dl_bw;
720#endif
721 /*
722 * "Active utilization" for this runqueue: increased when a
723 * task wakes up (becomes TASK_RUNNING) and decreased when a
724 * task blocks
725 */
726 u64 running_bw;
727
728 /*
729 * Utilization of the tasks "assigned" to this runqueue (including
730 * the tasks that are in runqueue and the tasks that executed on this
731 * CPU and blocked). Increased when a task moves to this runqueue, and
732 * decreased when the task moves away (migrates, changes scheduling
733 * policy, or terminates).
734 * This is needed to compute the "inactive utilization" for the
735 * runqueue (inactive utilization = this_bw - running_bw).
736 */
737 u64 this_bw;
738 u64 extra_bw;
739
740 /*
741 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
742 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
743 */
744 u64 max_bw;
745
746 /*
747 * Inverse of the fraction of CPU utilization that can be reclaimed
748 * by the GRUB algorithm.
749 */
750 u64 bw_ratio;
751};
752
753#ifdef CONFIG_FAIR_GROUP_SCHED
754/* An entity is a task if it doesn't "own" a runqueue */
755#define entity_is_task(se) (!se->my_q)
756
757static inline void se_update_runnable(struct sched_entity *se)
758{
759 if (!entity_is_task(se))
760 se->runnable_weight = se->my_q->h_nr_running;
761}
762
763static inline long se_runnable(struct sched_entity *se)
764{
765 if (entity_is_task(se))
766 return !!se->on_rq;
767 else
768 return se->runnable_weight;
769}
770
771#else
772#define entity_is_task(se) 1
773
774static inline void se_update_runnable(struct sched_entity *se) {}
775
776static inline long se_runnable(struct sched_entity *se)
777{
778 return !!se->on_rq;
779}
780#endif
781
782#ifdef CONFIG_SMP
783/*
784 * XXX we want to get rid of these helpers and use the full load resolution.
785 */
786static inline long se_weight(struct sched_entity *se)
787{
788 return scale_load_down(se->load.weight);
789}
790
791
792static inline bool sched_asym_prefer(int a, int b)
793{
794 return arch_asym_cpu_priority(cpu: a) > arch_asym_cpu_priority(cpu: b);
795}
796
797struct perf_domain {
798 struct em_perf_domain *em_pd;
799 struct perf_domain *next;
800 struct rcu_head rcu;
801};
802
803/* Scheduling group status flags */
804#define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
805#define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
806
807/*
808 * We add the notion of a root-domain which will be used to define per-domain
809 * variables. Each exclusive cpuset essentially defines an island domain by
810 * fully partitioning the member CPUs from any other cpuset. Whenever a new
811 * exclusive cpuset is created, we also create and attach a new root-domain
812 * object.
813 *
814 */
815struct root_domain {
816 atomic_t refcount;
817 atomic_t rto_count;
818 struct rcu_head rcu;
819 cpumask_var_t span;
820 cpumask_var_t online;
821
822 /*
823 * Indicate pullable load on at least one CPU, e.g:
824 * - More than one runnable task
825 * - Running task is misfit
826 */
827 int overload;
828
829 /* Indicate one or more cpus over-utilized (tipping point) */
830 int overutilized;
831
832 /*
833 * The bit corresponding to a CPU gets set here if such CPU has more
834 * than one runnable -deadline task (as it is below for RT tasks).
835 */
836 cpumask_var_t dlo_mask;
837 atomic_t dlo_count;
838 struct dl_bw dl_bw;
839 struct cpudl cpudl;
840
841 /*
842 * Indicate whether a root_domain's dl_bw has been checked or
843 * updated. It's monotonously increasing value.
844 *
845 * Also, some corner cases, like 'wrap around' is dangerous, but given
846 * that u64 is 'big enough'. So that shouldn't be a concern.
847 */
848 u64 visit_gen;
849
850#ifdef HAVE_RT_PUSH_IPI
851 /*
852 * For IPI pull requests, loop across the rto_mask.
853 */
854 struct irq_work rto_push_work;
855 raw_spinlock_t rto_lock;
856 /* These are only updated and read within rto_lock */
857 int rto_loop;
858 int rto_cpu;
859 /* These atomics are updated outside of a lock */
860 atomic_t rto_loop_next;
861 atomic_t rto_loop_start;
862#endif
863 /*
864 * The "RT overload" flag: it gets set if a CPU has more than
865 * one runnable RT task.
866 */
867 cpumask_var_t rto_mask;
868 struct cpupri cpupri;
869
870 unsigned long max_cpu_capacity;
871
872 /*
873 * NULL-terminated list of performance domains intersecting with the
874 * CPUs of the rd. Protected by RCU.
875 */
876 struct perf_domain __rcu *pd;
877};
878
879extern void init_defrootdomain(void);
880extern int sched_init_domains(const struct cpumask *cpu_map);
881extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
882extern void sched_get_rd(struct root_domain *rd);
883extern void sched_put_rd(struct root_domain *rd);
884
885#ifdef HAVE_RT_PUSH_IPI
886extern void rto_push_irq_work_func(struct irq_work *work);
887#endif
888#endif /* CONFIG_SMP */
889
890#ifdef CONFIG_UCLAMP_TASK
891/*
892 * struct uclamp_bucket - Utilization clamp bucket
893 * @value: utilization clamp value for tasks on this clamp bucket
894 * @tasks: number of RUNNABLE tasks on this clamp bucket
895 *
896 * Keep track of how many tasks are RUNNABLE for a given utilization
897 * clamp value.
898 */
899struct uclamp_bucket {
900 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
901 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
902};
903
904/*
905 * struct uclamp_rq - rq's utilization clamp
906 * @value: currently active clamp values for a rq
907 * @bucket: utilization clamp buckets affecting a rq
908 *
909 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
910 * A clamp value is affecting a rq when there is at least one task RUNNABLE
911 * (or actually running) with that value.
912 *
913 * There are up to UCLAMP_CNT possible different clamp values, currently there
914 * are only two: minimum utilization and maximum utilization.
915 *
916 * All utilization clamping values are MAX aggregated, since:
917 * - for util_min: we want to run the CPU at least at the max of the minimum
918 * utilization required by its currently RUNNABLE tasks.
919 * - for util_max: we want to allow the CPU to run up to the max of the
920 * maximum utilization allowed by its currently RUNNABLE tasks.
921 *
922 * Since on each system we expect only a limited number of different
923 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
924 * the metrics required to compute all the per-rq utilization clamp values.
925 */
926struct uclamp_rq {
927 unsigned int value;
928 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
929};
930
931DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
932#endif /* CONFIG_UCLAMP_TASK */
933
934struct rq;
935struct balance_callback {
936 struct balance_callback *next;
937 void (*func)(struct rq *rq);
938};
939
940/*
941 * This is the main, per-CPU runqueue data structure.
942 *
943 * Locking rule: those places that want to lock multiple runqueues
944 * (such as the load balancing or the thread migration code), lock
945 * acquire operations must be ordered by ascending &runqueue.
946 */
947struct rq {
948 /* runqueue lock: */
949 raw_spinlock_t __lock;
950
951 unsigned int nr_running;
952#ifdef CONFIG_NUMA_BALANCING
953 unsigned int nr_numa_running;
954 unsigned int nr_preferred_running;
955 unsigned int numa_migrate_on;
956#endif
957#ifdef CONFIG_NO_HZ_COMMON
958#ifdef CONFIG_SMP
959 unsigned long last_blocked_load_update_tick;
960 unsigned int has_blocked_load;
961 call_single_data_t nohz_csd;
962#endif /* CONFIG_SMP */
963 unsigned int nohz_tick_stopped;
964 atomic_t nohz_flags;
965#endif /* CONFIG_NO_HZ_COMMON */
966
967#ifdef CONFIG_SMP
968 unsigned int ttwu_pending;
969#endif
970 u64 nr_switches;
971
972#ifdef CONFIG_UCLAMP_TASK
973 /* Utilization clamp values based on CPU's RUNNABLE tasks */
974 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
975 unsigned int uclamp_flags;
976#define UCLAMP_FLAG_IDLE 0x01
977#endif
978
979 struct cfs_rq cfs;
980 struct rt_rq rt;
981 struct dl_rq dl;
982
983#ifdef CONFIG_FAIR_GROUP_SCHED
984 /* list of leaf cfs_rq on this CPU: */
985 struct list_head leaf_cfs_rq_list;
986 struct list_head *tmp_alone_branch;
987#endif /* CONFIG_FAIR_GROUP_SCHED */
988
989 /*
990 * This is part of a global counter where only the total sum
991 * over all CPUs matters. A task can increase this counter on
992 * one CPU and if it got migrated afterwards it may decrease
993 * it on another CPU. Always updated under the runqueue lock:
994 */
995 unsigned int nr_uninterruptible;
996
997 struct task_struct __rcu *curr;
998 struct task_struct *idle;
999 struct task_struct *stop;
1000 unsigned long next_balance;
1001 struct mm_struct *prev_mm;
1002
1003 unsigned int clock_update_flags;
1004 u64 clock;
1005 /* Ensure that all clocks are in the same cache line */
1006 u64 clock_task ____cacheline_aligned;
1007 u64 clock_pelt;
1008 unsigned long lost_idle_time;
1009 u64 clock_pelt_idle;
1010 u64 clock_idle;
1011#ifndef CONFIG_64BIT
1012 u64 clock_pelt_idle_copy;
1013 u64 clock_idle_copy;
1014#endif
1015
1016 atomic_t nr_iowait;
1017
1018#ifdef CONFIG_SCHED_DEBUG
1019 u64 last_seen_need_resched_ns;
1020 int ticks_without_resched;
1021#endif
1022
1023#ifdef CONFIG_MEMBARRIER
1024 int membarrier_state;
1025#endif
1026
1027#ifdef CONFIG_SMP
1028 struct root_domain *rd;
1029 struct sched_domain __rcu *sd;
1030
1031 unsigned long cpu_capacity;
1032
1033 struct balance_callback *balance_callback;
1034
1035 unsigned char nohz_idle_balance;
1036 unsigned char idle_balance;
1037
1038 unsigned long misfit_task_load;
1039
1040 /* For active balancing */
1041 int active_balance;
1042 int push_cpu;
1043 struct cpu_stop_work active_balance_work;
1044
1045 /* CPU of this runqueue: */
1046 int cpu;
1047 int online;
1048
1049 struct list_head cfs_tasks;
1050
1051 struct sched_avg avg_rt;
1052 struct sched_avg avg_dl;
1053#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1054 struct sched_avg avg_irq;
1055#endif
1056#ifdef CONFIG_SCHED_THERMAL_PRESSURE
1057 struct sched_avg avg_thermal;
1058#endif
1059 u64 idle_stamp;
1060 u64 avg_idle;
1061
1062 /* This is used to determine avg_idle's max value */
1063 u64 max_idle_balance_cost;
1064
1065#ifdef CONFIG_HOTPLUG_CPU
1066 struct rcuwait hotplug_wait;
1067#endif
1068#endif /* CONFIG_SMP */
1069
1070#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1071 u64 prev_irq_time;
1072#endif
1073#ifdef CONFIG_PARAVIRT
1074 u64 prev_steal_time;
1075#endif
1076#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1077 u64 prev_steal_time_rq;
1078#endif
1079
1080 /* calc_load related fields */
1081 unsigned long calc_load_update;
1082 long calc_load_active;
1083
1084#ifdef CONFIG_SCHED_HRTICK
1085#ifdef CONFIG_SMP
1086 call_single_data_t hrtick_csd;
1087#endif
1088 struct hrtimer hrtick_timer;
1089 ktime_t hrtick_time;
1090#endif
1091
1092#ifdef CONFIG_SCHEDSTATS
1093 /* latency stats */
1094 struct sched_info rq_sched_info;
1095 unsigned long long rq_cpu_time;
1096 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1097
1098 /* sys_sched_yield() stats */
1099 unsigned int yld_count;
1100
1101 /* schedule() stats */
1102 unsigned int sched_count;
1103 unsigned int sched_goidle;
1104
1105 /* try_to_wake_up() stats */
1106 unsigned int ttwu_count;
1107 unsigned int ttwu_local;
1108#endif
1109
1110#ifdef CONFIG_CPU_IDLE
1111 /* Must be inspected within a rcu lock section */
1112 struct cpuidle_state *idle_state;
1113#endif
1114
1115#ifdef CONFIG_SMP
1116 unsigned int nr_pinned;
1117#endif
1118 unsigned int push_busy;
1119 struct cpu_stop_work push_work;
1120
1121#ifdef CONFIG_SCHED_CORE
1122 /* per rq */
1123 struct rq *core;
1124 struct task_struct *core_pick;
1125 unsigned int core_enabled;
1126 unsigned int core_sched_seq;
1127 struct rb_root core_tree;
1128
1129 /* shared state -- careful with sched_core_cpu_deactivate() */
1130 unsigned int core_task_seq;
1131 unsigned int core_pick_seq;
1132 unsigned long core_cookie;
1133 unsigned int core_forceidle_count;
1134 unsigned int core_forceidle_seq;
1135 unsigned int core_forceidle_occupation;
1136 u64 core_forceidle_start;
1137#endif
1138
1139 /* Scratch cpumask to be temporarily used under rq_lock */
1140 cpumask_var_t scratch_mask;
1141
1142#if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1143 call_single_data_t cfsb_csd;
1144 struct list_head cfsb_csd_list;
1145#endif
1146};
1147
1148#ifdef CONFIG_FAIR_GROUP_SCHED
1149
1150/* CPU runqueue to which this cfs_rq is attached */
1151static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1152{
1153 return cfs_rq->rq;
1154}
1155
1156#else
1157
1158static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1159{
1160 return container_of(cfs_rq, struct rq, cfs);
1161}
1162#endif
1163
1164static inline int cpu_of(struct rq *rq)
1165{
1166#ifdef CONFIG_SMP
1167 return rq->cpu;
1168#else
1169 return 0;
1170#endif
1171}
1172
1173#define MDF_PUSH 0x01
1174
1175static inline bool is_migration_disabled(struct task_struct *p)
1176{
1177#ifdef CONFIG_SMP
1178 return p->migration_disabled;
1179#else
1180 return false;
1181#endif
1182}
1183
1184DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1185
1186#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1187#define this_rq() this_cpu_ptr(&runqueues)
1188#define task_rq(p) cpu_rq(task_cpu(p))
1189#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1190#define raw_rq() raw_cpu_ptr(&runqueues)
1191
1192struct sched_group;
1193#ifdef CONFIG_SCHED_CORE
1194static inline struct cpumask *sched_group_span(struct sched_group *sg);
1195
1196DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1197
1198static inline bool sched_core_enabled(struct rq *rq)
1199{
1200 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1201}
1202
1203static inline bool sched_core_disabled(void)
1204{
1205 return !static_branch_unlikely(&__sched_core_enabled);
1206}
1207
1208/*
1209 * Be careful with this function; not for general use. The return value isn't
1210 * stable unless you actually hold a relevant rq->__lock.
1211 */
1212static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1213{
1214 if (sched_core_enabled(rq))
1215 return &rq->core->__lock;
1216
1217 return &rq->__lock;
1218}
1219
1220static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1221{
1222 if (rq->core_enabled)
1223 return &rq->core->__lock;
1224
1225 return &rq->__lock;
1226}
1227
1228bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1229 bool fi);
1230void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1231
1232/*
1233 * Helpers to check if the CPU's core cookie matches with the task's cookie
1234 * when core scheduling is enabled.
1235 * A special case is that the task's cookie always matches with CPU's core
1236 * cookie if the CPU is in an idle core.
1237 */
1238static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1239{
1240 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1241 if (!sched_core_enabled(rq))
1242 return true;
1243
1244 return rq->core->core_cookie == p->core_cookie;
1245}
1246
1247static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1248{
1249 bool idle_core = true;
1250 int cpu;
1251
1252 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1253 if (!sched_core_enabled(rq))
1254 return true;
1255
1256 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1257 if (!available_idle_cpu(cpu)) {
1258 idle_core = false;
1259 break;
1260 }
1261 }
1262
1263 /*
1264 * A CPU in an idle core is always the best choice for tasks with
1265 * cookies.
1266 */
1267 return idle_core || rq->core->core_cookie == p->core_cookie;
1268}
1269
1270static inline bool sched_group_cookie_match(struct rq *rq,
1271 struct task_struct *p,
1272 struct sched_group *group)
1273{
1274 int cpu;
1275
1276 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1277 if (!sched_core_enabled(rq))
1278 return true;
1279
1280 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1281 if (sched_core_cookie_match(cpu_rq(cpu), p))
1282 return true;
1283 }
1284 return false;
1285}
1286
1287static inline bool sched_core_enqueued(struct task_struct *p)
1288{
1289 return !RB_EMPTY_NODE(&p->core_node);
1290}
1291
1292extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1293extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1294
1295extern void sched_core_get(void);
1296extern void sched_core_put(void);
1297
1298#else /* !CONFIG_SCHED_CORE */
1299
1300static inline bool sched_core_enabled(struct rq *rq)
1301{
1302 return false;
1303}
1304
1305static inline bool sched_core_disabled(void)
1306{
1307 return true;
1308}
1309
1310static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1311{
1312 return &rq->__lock;
1313}
1314
1315static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1316{
1317 return &rq->__lock;
1318}
1319
1320static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1321{
1322 return true;
1323}
1324
1325static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1326{
1327 return true;
1328}
1329
1330static inline bool sched_group_cookie_match(struct rq *rq,
1331 struct task_struct *p,
1332 struct sched_group *group)
1333{
1334 return true;
1335}
1336#endif /* CONFIG_SCHED_CORE */
1337
1338static inline void lockdep_assert_rq_held(struct rq *rq)
1339{
1340 lockdep_assert_held(__rq_lockp(rq));
1341}
1342
1343extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1344extern bool raw_spin_rq_trylock(struct rq *rq);
1345extern void raw_spin_rq_unlock(struct rq *rq);
1346
1347static inline void raw_spin_rq_lock(struct rq *rq)
1348{
1349 raw_spin_rq_lock_nested(rq, subclass: 0);
1350}
1351
1352static inline void raw_spin_rq_lock_irq(struct rq *rq)
1353{
1354 local_irq_disable();
1355 raw_spin_rq_lock(rq);
1356}
1357
1358static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1359{
1360 raw_spin_rq_unlock(rq);
1361 local_irq_enable();
1362}
1363
1364static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1365{
1366 unsigned long flags;
1367 local_irq_save(flags);
1368 raw_spin_rq_lock(rq);
1369 return flags;
1370}
1371
1372static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1373{
1374 raw_spin_rq_unlock(rq);
1375 local_irq_restore(flags);
1376}
1377
1378#define raw_spin_rq_lock_irqsave(rq, flags) \
1379do { \
1380 flags = _raw_spin_rq_lock_irqsave(rq); \
1381} while (0)
1382
1383#ifdef CONFIG_SCHED_SMT
1384extern void __update_idle_core(struct rq *rq);
1385
1386static inline void update_idle_core(struct rq *rq)
1387{
1388 if (static_branch_unlikely(&sched_smt_present))
1389 __update_idle_core(rq);
1390}
1391
1392#else
1393static inline void update_idle_core(struct rq *rq) { }
1394#endif
1395
1396#ifdef CONFIG_FAIR_GROUP_SCHED
1397static inline struct task_struct *task_of(struct sched_entity *se)
1398{
1399 SCHED_WARN_ON(!entity_is_task(se));
1400 return container_of(se, struct task_struct, se);
1401}
1402
1403static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1404{
1405 return p->se.cfs_rq;
1406}
1407
1408/* runqueue on which this entity is (to be) queued */
1409static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1410{
1411 return se->cfs_rq;
1412}
1413
1414/* runqueue "owned" by this group */
1415static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1416{
1417 return grp->my_q;
1418}
1419
1420#else
1421
1422#define task_of(_se) container_of(_se, struct task_struct, se)
1423
1424static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1425{
1426 return &task_rq(p)->cfs;
1427}
1428
1429static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1430{
1431 const struct task_struct *p = task_of(se);
1432 struct rq *rq = task_rq(p);
1433
1434 return &rq->cfs;
1435}
1436
1437/* runqueue "owned" by this group */
1438static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1439{
1440 return NULL;
1441}
1442#endif
1443
1444extern void update_rq_clock(struct rq *rq);
1445
1446/*
1447 * rq::clock_update_flags bits
1448 *
1449 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1450 * call to __schedule(). This is an optimisation to avoid
1451 * neighbouring rq clock updates.
1452 *
1453 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1454 * in effect and calls to update_rq_clock() are being ignored.
1455 *
1456 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1457 * made to update_rq_clock() since the last time rq::lock was pinned.
1458 *
1459 * If inside of __schedule(), clock_update_flags will have been
1460 * shifted left (a left shift is a cheap operation for the fast path
1461 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1462 *
1463 * if (rq-clock_update_flags >= RQCF_UPDATED)
1464 *
1465 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1466 * one position though, because the next rq_unpin_lock() will shift it
1467 * back.
1468 */
1469#define RQCF_REQ_SKIP 0x01
1470#define RQCF_ACT_SKIP 0x02
1471#define RQCF_UPDATED 0x04
1472
1473static inline void assert_clock_updated(struct rq *rq)
1474{
1475 /*
1476 * The only reason for not seeing a clock update since the
1477 * last rq_pin_lock() is if we're currently skipping updates.
1478 */
1479 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1480}
1481
1482static inline u64 rq_clock(struct rq *rq)
1483{
1484 lockdep_assert_rq_held(rq);
1485 assert_clock_updated(rq);
1486
1487 return rq->clock;
1488}
1489
1490static inline u64 rq_clock_task(struct rq *rq)
1491{
1492 lockdep_assert_rq_held(rq);
1493 assert_clock_updated(rq);
1494
1495 return rq->clock_task;
1496}
1497
1498/**
1499 * By default the decay is the default pelt decay period.
1500 * The decay shift can change the decay period in
1501 * multiples of 32.
1502 * Decay shift Decay period(ms)
1503 * 0 32
1504 * 1 64
1505 * 2 128
1506 * 3 256
1507 * 4 512
1508 */
1509extern int sched_thermal_decay_shift;
1510
1511static inline u64 rq_clock_thermal(struct rq *rq)
1512{
1513 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1514}
1515
1516static inline void rq_clock_skip_update(struct rq *rq)
1517{
1518 lockdep_assert_rq_held(rq);
1519 rq->clock_update_flags |= RQCF_REQ_SKIP;
1520}
1521
1522/*
1523 * See rt task throttling, which is the only time a skip
1524 * request is canceled.
1525 */
1526static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1527{
1528 lockdep_assert_rq_held(rq);
1529 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1530}
1531
1532/*
1533 * During cpu offlining and rq wide unthrottling, we can trigger
1534 * an update_rq_clock() for several cfs and rt runqueues (Typically
1535 * when using list_for_each_entry_*)
1536 * rq_clock_start_loop_update() can be called after updating the clock
1537 * once and before iterating over the list to prevent multiple update.
1538 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1539 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1540 */
1541static inline void rq_clock_start_loop_update(struct rq *rq)
1542{
1543 lockdep_assert_rq_held(rq);
1544 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1545 rq->clock_update_flags |= RQCF_ACT_SKIP;
1546}
1547
1548static inline void rq_clock_stop_loop_update(struct rq *rq)
1549{
1550 lockdep_assert_rq_held(rq);
1551 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1552}
1553
1554struct rq_flags {
1555 unsigned long flags;
1556 struct pin_cookie cookie;
1557#ifdef CONFIG_SCHED_DEBUG
1558 /*
1559 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1560 * current pin context is stashed here in case it needs to be
1561 * restored in rq_repin_lock().
1562 */
1563 unsigned int clock_update_flags;
1564#endif
1565};
1566
1567extern struct balance_callback balance_push_callback;
1568
1569/*
1570 * Lockdep annotation that avoids accidental unlocks; it's like a
1571 * sticky/continuous lockdep_assert_held().
1572 *
1573 * This avoids code that has access to 'struct rq *rq' (basically everything in
1574 * the scheduler) from accidentally unlocking the rq if they do not also have a
1575 * copy of the (on-stack) 'struct rq_flags rf'.
1576 *
1577 * Also see Documentation/locking/lockdep-design.rst.
1578 */
1579static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1580{
1581 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1582
1583#ifdef CONFIG_SCHED_DEBUG
1584 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1585 rf->clock_update_flags = 0;
1586#ifdef CONFIG_SMP
1587 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1588#endif
1589#endif
1590}
1591
1592static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1593{
1594#ifdef CONFIG_SCHED_DEBUG
1595 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1596 rf->clock_update_flags = RQCF_UPDATED;
1597#endif
1598
1599 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1600}
1601
1602static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1603{
1604 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1605
1606#ifdef CONFIG_SCHED_DEBUG
1607 /*
1608 * Restore the value we stashed in @rf for this pin context.
1609 */
1610 rq->clock_update_flags |= rf->clock_update_flags;
1611#endif
1612}
1613
1614struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1615 __acquires(rq->lock);
1616
1617struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1618 __acquires(p->pi_lock)
1619 __acquires(rq->lock);
1620
1621static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1622 __releases(rq->lock)
1623{
1624 rq_unpin_lock(rq, rf);
1625 raw_spin_rq_unlock(rq);
1626}
1627
1628static inline void
1629task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1630 __releases(rq->lock)
1631 __releases(p->pi_lock)
1632{
1633 rq_unpin_lock(rq, rf);
1634 raw_spin_rq_unlock(rq);
1635 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1636}
1637
1638DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1639 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1640 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1641 struct rq *rq; struct rq_flags rf)
1642
1643static inline void
1644rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1645 __acquires(rq->lock)
1646{
1647 raw_spin_rq_lock_irqsave(rq, rf->flags);
1648 rq_pin_lock(rq, rf);
1649}
1650
1651static inline void
1652rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1653 __acquires(rq->lock)
1654{
1655 raw_spin_rq_lock_irq(rq);
1656 rq_pin_lock(rq, rf);
1657}
1658
1659static inline void
1660rq_lock(struct rq *rq, struct rq_flags *rf)
1661 __acquires(rq->lock)
1662{
1663 raw_spin_rq_lock(rq);
1664 rq_pin_lock(rq, rf);
1665}
1666
1667static inline void
1668rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1669 __releases(rq->lock)
1670{
1671 rq_unpin_lock(rq, rf);
1672 raw_spin_rq_unlock_irqrestore(rq, flags: rf->flags);
1673}
1674
1675static inline void
1676rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1677 __releases(rq->lock)
1678{
1679 rq_unpin_lock(rq, rf);
1680 raw_spin_rq_unlock_irq(rq);
1681}
1682
1683static inline void
1684rq_unlock(struct rq *rq, struct rq_flags *rf)
1685 __releases(rq->lock)
1686{
1687 rq_unpin_lock(rq, rf);
1688 raw_spin_rq_unlock(rq);
1689}
1690
1691DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1692 rq_lock(_T->lock, &_T->rf),
1693 rq_unlock(_T->lock, &_T->rf),
1694 struct rq_flags rf)
1695
1696DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1697 rq_lock_irq(_T->lock, &_T->rf),
1698 rq_unlock_irq(_T->lock, &_T->rf),
1699 struct rq_flags rf)
1700
1701DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1702 rq_lock_irqsave(_T->lock, &_T->rf),
1703 rq_unlock_irqrestore(_T->lock, &_T->rf),
1704 struct rq_flags rf)
1705
1706static inline struct rq *
1707this_rq_lock_irq(struct rq_flags *rf)
1708 __acquires(rq->lock)
1709{
1710 struct rq *rq;
1711
1712 local_irq_disable();
1713 rq = this_rq();
1714 rq_lock(rq, rf);
1715 return rq;
1716}
1717
1718#ifdef CONFIG_NUMA
1719enum numa_topology_type {
1720 NUMA_DIRECT,
1721 NUMA_GLUELESS_MESH,
1722 NUMA_BACKPLANE,
1723};
1724extern enum numa_topology_type sched_numa_topology_type;
1725extern int sched_max_numa_distance;
1726extern bool find_numa_distance(int distance);
1727extern void sched_init_numa(int offline_node);
1728extern void sched_update_numa(int cpu, bool online);
1729extern void sched_domains_numa_masks_set(unsigned int cpu);
1730extern void sched_domains_numa_masks_clear(unsigned int cpu);
1731extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1732#else
1733static inline void sched_init_numa(int offline_node) { }
1734static inline void sched_update_numa(int cpu, bool online) { }
1735static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1736static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1737static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1738{
1739 return nr_cpu_ids;
1740}
1741#endif
1742
1743#ifdef CONFIG_NUMA_BALANCING
1744/* The regions in numa_faults array from task_struct */
1745enum numa_faults_stats {
1746 NUMA_MEM = 0,
1747 NUMA_CPU,
1748 NUMA_MEMBUF,
1749 NUMA_CPUBUF
1750};
1751extern void sched_setnuma(struct task_struct *p, int node);
1752extern int migrate_task_to(struct task_struct *p, int cpu);
1753extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1754 int cpu, int scpu);
1755extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1756#else
1757static inline void
1758init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1759{
1760}
1761#endif /* CONFIG_NUMA_BALANCING */
1762
1763#ifdef CONFIG_SMP
1764
1765static inline void
1766queue_balance_callback(struct rq *rq,
1767 struct balance_callback *head,
1768 void (*func)(struct rq *rq))
1769{
1770 lockdep_assert_rq_held(rq);
1771
1772 /*
1773 * Don't (re)queue an already queued item; nor queue anything when
1774 * balance_push() is active, see the comment with
1775 * balance_push_callback.
1776 */
1777 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1778 return;
1779
1780 head->func = func;
1781 head->next = rq->balance_callback;
1782 rq->balance_callback = head;
1783}
1784
1785#define rcu_dereference_check_sched_domain(p) \
1786 rcu_dereference_check((p), \
1787 lockdep_is_held(&sched_domains_mutex))
1788
1789/*
1790 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1791 * See destroy_sched_domains: call_rcu for details.
1792 *
1793 * The domain tree of any CPU may only be accessed from within
1794 * preempt-disabled sections.
1795 */
1796#define for_each_domain(cpu, __sd) \
1797 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1798 __sd; __sd = __sd->parent)
1799
1800/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1801#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1802static const unsigned int SD_SHARED_CHILD_MASK =
1803#include <linux/sched/sd_flags.h>
18040;
1805#undef SD_FLAG
1806
1807/**
1808 * highest_flag_domain - Return highest sched_domain containing flag.
1809 * @cpu: The CPU whose highest level of sched domain is to
1810 * be returned.
1811 * @flag: The flag to check for the highest sched_domain
1812 * for the given CPU.
1813 *
1814 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1815 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1816 */
1817static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1818{
1819 struct sched_domain *sd, *hsd = NULL;
1820
1821 for_each_domain(cpu, sd) {
1822 if (sd->flags & flag) {
1823 hsd = sd;
1824 continue;
1825 }
1826
1827 /*
1828 * Stop the search if @flag is known to be shared at lower
1829 * levels. It will not be found further up.
1830 */
1831 if (flag & SD_SHARED_CHILD_MASK)
1832 break;
1833 }
1834
1835 return hsd;
1836}
1837
1838static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1839{
1840 struct sched_domain *sd;
1841
1842 for_each_domain(cpu, sd) {
1843 if (sd->flags & flag)
1844 break;
1845 }
1846
1847 return sd;
1848}
1849
1850DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1851DECLARE_PER_CPU(int, sd_llc_size);
1852DECLARE_PER_CPU(int, sd_llc_id);
1853DECLARE_PER_CPU(int, sd_share_id);
1854DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1855DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1856DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1857DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1858extern struct static_key_false sched_asym_cpucapacity;
1859extern struct static_key_false sched_cluster_active;
1860
1861static __always_inline bool sched_asym_cpucap_active(void)
1862{
1863 return static_branch_unlikely(&sched_asym_cpucapacity);
1864}
1865
1866struct sched_group_capacity {
1867 atomic_t ref;
1868 /*
1869 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1870 * for a single CPU.
1871 */
1872 unsigned long capacity;
1873 unsigned long min_capacity; /* Min per-CPU capacity in group */
1874 unsigned long max_capacity; /* Max per-CPU capacity in group */
1875 unsigned long next_update;
1876 int imbalance; /* XXX unrelated to capacity but shared group state */
1877
1878#ifdef CONFIG_SCHED_DEBUG
1879 int id;
1880#endif
1881
1882 unsigned long cpumask[]; /* Balance mask */
1883};
1884
1885struct sched_group {
1886 struct sched_group *next; /* Must be a circular list */
1887 atomic_t ref;
1888
1889 unsigned int group_weight;
1890 unsigned int cores;
1891 struct sched_group_capacity *sgc;
1892 int asym_prefer_cpu; /* CPU of highest priority in group */
1893 int flags;
1894
1895 /*
1896 * The CPUs this group covers.
1897 *
1898 * NOTE: this field is variable length. (Allocated dynamically
1899 * by attaching extra space to the end of the structure,
1900 * depending on how many CPUs the kernel has booted up with)
1901 */
1902 unsigned long cpumask[];
1903};
1904
1905static inline struct cpumask *sched_group_span(struct sched_group *sg)
1906{
1907 return to_cpumask(sg->cpumask);
1908}
1909
1910/*
1911 * See build_balance_mask().
1912 */
1913static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1914{
1915 return to_cpumask(sg->sgc->cpumask);
1916}
1917
1918extern int group_balance_cpu(struct sched_group *sg);
1919
1920#ifdef CONFIG_SCHED_DEBUG
1921void update_sched_domain_debugfs(void);
1922void dirty_sched_domain_sysctl(int cpu);
1923#else
1924static inline void update_sched_domain_debugfs(void)
1925{
1926}
1927static inline void dirty_sched_domain_sysctl(int cpu)
1928{
1929}
1930#endif
1931
1932extern int sched_update_scaling(void);
1933
1934static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1935{
1936 if (!p->user_cpus_ptr)
1937 return cpu_possible_mask; /* &init_task.cpus_mask */
1938 return p->user_cpus_ptr;
1939}
1940#endif /* CONFIG_SMP */
1941
1942#include "stats.h"
1943
1944#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1945
1946extern void __sched_core_account_forceidle(struct rq *rq);
1947
1948static inline void sched_core_account_forceidle(struct rq *rq)
1949{
1950 if (schedstat_enabled())
1951 __sched_core_account_forceidle(rq);
1952}
1953
1954extern void __sched_core_tick(struct rq *rq);
1955
1956static inline void sched_core_tick(struct rq *rq)
1957{
1958 if (sched_core_enabled(rq) && schedstat_enabled())
1959 __sched_core_tick(rq);
1960}
1961
1962#else
1963
1964static inline void sched_core_account_forceidle(struct rq *rq) {}
1965
1966static inline void sched_core_tick(struct rq *rq) {}
1967
1968#endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1969
1970#ifdef CONFIG_CGROUP_SCHED
1971
1972/*
1973 * Return the group to which this tasks belongs.
1974 *
1975 * We cannot use task_css() and friends because the cgroup subsystem
1976 * changes that value before the cgroup_subsys::attach() method is called,
1977 * therefore we cannot pin it and might observe the wrong value.
1978 *
1979 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1980 * core changes this before calling sched_move_task().
1981 *
1982 * Instead we use a 'copy' which is updated from sched_move_task() while
1983 * holding both task_struct::pi_lock and rq::lock.
1984 */
1985static inline struct task_group *task_group(struct task_struct *p)
1986{
1987 return p->sched_task_group;
1988}
1989
1990/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1991static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1992{
1993#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1994 struct task_group *tg = task_group(p);
1995#endif
1996
1997#ifdef CONFIG_FAIR_GROUP_SCHED
1998 set_task_rq_fair(se: &p->se, prev: p->se.cfs_rq, next: tg->cfs_rq[cpu]);
1999 p->se.cfs_rq = tg->cfs_rq[cpu];
2000 p->se.parent = tg->se[cpu];
2001 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2002#endif
2003
2004#ifdef CONFIG_RT_GROUP_SCHED
2005 p->rt.rt_rq = tg->rt_rq[cpu];
2006 p->rt.parent = tg->rt_se[cpu];
2007#endif
2008}
2009
2010#else /* CONFIG_CGROUP_SCHED */
2011
2012static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2013static inline struct task_group *task_group(struct task_struct *p)
2014{
2015 return NULL;
2016}
2017
2018#endif /* CONFIG_CGROUP_SCHED */
2019
2020static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2021{
2022 set_task_rq(p, cpu);
2023#ifdef CONFIG_SMP
2024 /*
2025 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2026 * successfully executed on another CPU. We must ensure that updates of
2027 * per-task data have been completed by this moment.
2028 */
2029 smp_wmb();
2030 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2031 p->wake_cpu = cpu;
2032#endif
2033}
2034
2035/*
2036 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2037 */
2038#ifdef CONFIG_SCHED_DEBUG
2039# define const_debug __read_mostly
2040#else
2041# define const_debug const
2042#endif
2043
2044#define SCHED_FEAT(name, enabled) \
2045 __SCHED_FEAT_##name ,
2046
2047enum {
2048#include "features.h"
2049 __SCHED_FEAT_NR,
2050};
2051
2052#undef SCHED_FEAT
2053
2054#ifdef CONFIG_SCHED_DEBUG
2055
2056/*
2057 * To support run-time toggling of sched features, all the translation units
2058 * (but core.c) reference the sysctl_sched_features defined in core.c.
2059 */
2060extern const_debug unsigned int sysctl_sched_features;
2061
2062#ifdef CONFIG_JUMP_LABEL
2063#define SCHED_FEAT(name, enabled) \
2064static __always_inline bool static_branch_##name(struct static_key *key) \
2065{ \
2066 return static_key_##enabled(key); \
2067}
2068
2069#include "features.h"
2070#undef SCHED_FEAT
2071
2072extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2073#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2074
2075#else /* !CONFIG_JUMP_LABEL */
2076
2077#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2078
2079#endif /* CONFIG_JUMP_LABEL */
2080
2081#else /* !SCHED_DEBUG */
2082
2083/*
2084 * Each translation unit has its own copy of sysctl_sched_features to allow
2085 * constants propagation at compile time and compiler optimization based on
2086 * features default.
2087 */
2088#define SCHED_FEAT(name, enabled) \
2089 (1UL << __SCHED_FEAT_##name) * enabled |
2090static const_debug __maybe_unused unsigned int sysctl_sched_features =
2091#include "features.h"
2092 0;
2093#undef SCHED_FEAT
2094
2095#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2096
2097#endif /* SCHED_DEBUG */
2098
2099extern struct static_key_false sched_numa_balancing;
2100extern struct static_key_false sched_schedstats;
2101
2102static inline u64 global_rt_period(void)
2103{
2104 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2105}
2106
2107static inline u64 global_rt_runtime(void)
2108{
2109 if (sysctl_sched_rt_runtime < 0)
2110 return RUNTIME_INF;
2111
2112 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2113}
2114
2115static inline int task_current(struct rq *rq, struct task_struct *p)
2116{
2117 return rq->curr == p;
2118}
2119
2120static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2121{
2122#ifdef CONFIG_SMP
2123 return p->on_cpu;
2124#else
2125 return task_current(rq, p);
2126#endif
2127}
2128
2129static inline int task_on_rq_queued(struct task_struct *p)
2130{
2131 return p->on_rq == TASK_ON_RQ_QUEUED;
2132}
2133
2134static inline int task_on_rq_migrating(struct task_struct *p)
2135{
2136 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2137}
2138
2139/* Wake flags. The first three directly map to some SD flag value */
2140#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2141#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2142#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2143
2144#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2145#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2146#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2147
2148#ifdef CONFIG_SMP
2149static_assert(WF_EXEC == SD_BALANCE_EXEC);
2150static_assert(WF_FORK == SD_BALANCE_FORK);
2151static_assert(WF_TTWU == SD_BALANCE_WAKE);
2152#endif
2153
2154/*
2155 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2156 * of tasks with abnormal "nice" values across CPUs the contribution that
2157 * each task makes to its run queue's load is weighted according to its
2158 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2159 * scaled version of the new time slice allocation that they receive on time
2160 * slice expiry etc.
2161 */
2162
2163#define WEIGHT_IDLEPRIO 3
2164#define WMULT_IDLEPRIO 1431655765
2165
2166extern const int sched_prio_to_weight[40];
2167extern const u32 sched_prio_to_wmult[40];
2168
2169/*
2170 * {de,en}queue flags:
2171 *
2172 * DEQUEUE_SLEEP - task is no longer runnable
2173 * ENQUEUE_WAKEUP - task just became runnable
2174 *
2175 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2176 * are in a known state which allows modification. Such pairs
2177 * should preserve as much state as possible.
2178 *
2179 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2180 * in the runqueue.
2181 *
2182 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2183 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2184 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2185 *
2186 */
2187
2188#define DEQUEUE_SLEEP 0x01
2189#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2190#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2191#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2192
2193#define ENQUEUE_WAKEUP 0x01
2194#define ENQUEUE_RESTORE 0x02
2195#define ENQUEUE_MOVE 0x04
2196#define ENQUEUE_NOCLOCK 0x08
2197
2198#define ENQUEUE_HEAD 0x10
2199#define ENQUEUE_REPLENISH 0x20
2200#ifdef CONFIG_SMP
2201#define ENQUEUE_MIGRATED 0x40
2202#else
2203#define ENQUEUE_MIGRATED 0x00
2204#endif
2205#define ENQUEUE_INITIAL 0x80
2206
2207#define RETRY_TASK ((void *)-1UL)
2208
2209struct affinity_context {
2210 const struct cpumask *new_mask;
2211 struct cpumask *user_mask;
2212 unsigned int flags;
2213};
2214
2215struct sched_class {
2216
2217#ifdef CONFIG_UCLAMP_TASK
2218 int uclamp_enabled;
2219#endif
2220
2221 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2222 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2223 void (*yield_task) (struct rq *rq);
2224 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2225
2226 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2227
2228 struct task_struct *(*pick_next_task)(struct rq *rq);
2229
2230 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2231 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2232
2233#ifdef CONFIG_SMP
2234 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2235 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2236
2237 struct task_struct * (*pick_task)(struct rq *rq);
2238
2239 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2240
2241 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2242
2243 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2244
2245 void (*rq_online)(struct rq *rq);
2246 void (*rq_offline)(struct rq *rq);
2247
2248 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2249#endif
2250
2251 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2252 void (*task_fork)(struct task_struct *p);
2253 void (*task_dead)(struct task_struct *p);
2254
2255 /*
2256 * The switched_from() call is allowed to drop rq->lock, therefore we
2257 * cannot assume the switched_from/switched_to pair is serialized by
2258 * rq->lock. They are however serialized by p->pi_lock.
2259 */
2260 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2261 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2262 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2263 int oldprio);
2264
2265 unsigned int (*get_rr_interval)(struct rq *rq,
2266 struct task_struct *task);
2267
2268 void (*update_curr)(struct rq *rq);
2269
2270#ifdef CONFIG_FAIR_GROUP_SCHED
2271 void (*task_change_group)(struct task_struct *p);
2272#endif
2273
2274#ifdef CONFIG_SCHED_CORE
2275 int (*task_is_throttled)(struct task_struct *p, int cpu);
2276#endif
2277};
2278
2279static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2280{
2281 WARN_ON_ONCE(rq->curr != prev);
2282 prev->sched_class->put_prev_task(rq, prev);
2283}
2284
2285static inline void set_next_task(struct rq *rq, struct task_struct *next)
2286{
2287 next->sched_class->set_next_task(rq, next, false);
2288}
2289
2290
2291/*
2292 * Helper to define a sched_class instance; each one is placed in a separate
2293 * section which is ordered by the linker script:
2294 *
2295 * include/asm-generic/vmlinux.lds.h
2296 *
2297 * *CAREFUL* they are laid out in *REVERSE* order!!!
2298 *
2299 * Also enforce alignment on the instance, not the type, to guarantee layout.
2300 */
2301#define DEFINE_SCHED_CLASS(name) \
2302const struct sched_class name##_sched_class \
2303 __aligned(__alignof__(struct sched_class)) \
2304 __section("__" #name "_sched_class")
2305
2306/* Defined in include/asm-generic/vmlinux.lds.h */
2307extern struct sched_class __sched_class_highest[];
2308extern struct sched_class __sched_class_lowest[];
2309
2310#define for_class_range(class, _from, _to) \
2311 for (class = (_from); class < (_to); class++)
2312
2313#define for_each_class(class) \
2314 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2315
2316#define sched_class_above(_a, _b) ((_a) < (_b))
2317
2318extern const struct sched_class stop_sched_class;
2319extern const struct sched_class dl_sched_class;
2320extern const struct sched_class rt_sched_class;
2321extern const struct sched_class fair_sched_class;
2322extern const struct sched_class idle_sched_class;
2323
2324static inline bool sched_stop_runnable(struct rq *rq)
2325{
2326 return rq->stop && task_on_rq_queued(p: rq->stop);
2327}
2328
2329static inline bool sched_dl_runnable(struct rq *rq)
2330{
2331 return rq->dl.dl_nr_running > 0;
2332}
2333
2334static inline bool sched_rt_runnable(struct rq *rq)
2335{
2336 return rq->rt.rt_queued > 0;
2337}
2338
2339static inline bool sched_fair_runnable(struct rq *rq)
2340{
2341 return rq->cfs.nr_running > 0;
2342}
2343
2344extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2345extern struct task_struct *pick_next_task_idle(struct rq *rq);
2346
2347#define SCA_CHECK 0x01
2348#define SCA_MIGRATE_DISABLE 0x02
2349#define SCA_MIGRATE_ENABLE 0x04
2350#define SCA_USER 0x08
2351
2352#ifdef CONFIG_SMP
2353
2354extern void update_group_capacity(struct sched_domain *sd, int cpu);
2355
2356extern void trigger_load_balance(struct rq *rq);
2357
2358extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2359
2360static inline struct task_struct *get_push_task(struct rq *rq)
2361{
2362 struct task_struct *p = rq->curr;
2363
2364 lockdep_assert_rq_held(rq);
2365
2366 if (rq->push_busy)
2367 return NULL;
2368
2369 if (p->nr_cpus_allowed == 1)
2370 return NULL;
2371
2372 if (p->migration_disabled)
2373 return NULL;
2374
2375 rq->push_busy = true;
2376 return get_task_struct(t: p);
2377}
2378
2379extern int push_cpu_stop(void *arg);
2380
2381#endif
2382
2383#ifdef CONFIG_CPU_IDLE
2384static inline void idle_set_state(struct rq *rq,
2385 struct cpuidle_state *idle_state)
2386{
2387 rq->idle_state = idle_state;
2388}
2389
2390static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2391{
2392 SCHED_WARN_ON(!rcu_read_lock_held());
2393
2394 return rq->idle_state;
2395}
2396#else
2397static inline void idle_set_state(struct rq *rq,
2398 struct cpuidle_state *idle_state)
2399{
2400}
2401
2402static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2403{
2404 return NULL;
2405}
2406#endif
2407
2408extern void schedule_idle(void);
2409asmlinkage void schedule_user(void);
2410
2411extern void sysrq_sched_debug_show(void);
2412extern void sched_init_granularity(void);
2413extern void update_max_interval(void);
2414
2415extern void init_sched_dl_class(void);
2416extern void init_sched_rt_class(void);
2417extern void init_sched_fair_class(void);
2418
2419extern void reweight_task(struct task_struct *p, int prio);
2420
2421extern void resched_curr(struct rq *rq);
2422extern void resched_cpu(int cpu);
2423
2424extern struct rt_bandwidth def_rt_bandwidth;
2425extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2426extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2427
2428extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2429extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2430
2431#define BW_SHIFT 20
2432#define BW_UNIT (1 << BW_SHIFT)
2433#define RATIO_SHIFT 8
2434#define MAX_BW_BITS (64 - BW_SHIFT)
2435#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2436unsigned long to_ratio(u64 period, u64 runtime);
2437
2438extern void init_entity_runnable_average(struct sched_entity *se);
2439extern void post_init_entity_util_avg(struct task_struct *p);
2440
2441#ifdef CONFIG_NO_HZ_FULL
2442extern bool sched_can_stop_tick(struct rq *rq);
2443extern int __init sched_tick_offload_init(void);
2444
2445/*
2446 * Tick may be needed by tasks in the runqueue depending on their policy and
2447 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2448 * nohz mode if necessary.
2449 */
2450static inline void sched_update_tick_dependency(struct rq *rq)
2451{
2452 int cpu = cpu_of(rq);
2453
2454 if (!tick_nohz_full_cpu(cpu))
2455 return;
2456
2457 if (sched_can_stop_tick(rq))
2458 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2459 else
2460 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2461}
2462#else
2463static inline int sched_tick_offload_init(void) { return 0; }
2464static inline void sched_update_tick_dependency(struct rq *rq) { }
2465#endif
2466
2467static inline void add_nr_running(struct rq *rq, unsigned count)
2468{
2469 unsigned prev_nr = rq->nr_running;
2470
2471 rq->nr_running = prev_nr + count;
2472 if (trace_sched_update_nr_running_tp_enabled()) {
2473 call_trace_sched_update_nr_running(rq, count);
2474 }
2475
2476#ifdef CONFIG_SMP
2477 if (prev_nr < 2 && rq->nr_running >= 2) {
2478 if (!READ_ONCE(rq->rd->overload))
2479 WRITE_ONCE(rq->rd->overload, 1);
2480 }
2481#endif
2482
2483 sched_update_tick_dependency(rq);
2484}
2485
2486static inline void sub_nr_running(struct rq *rq, unsigned count)
2487{
2488 rq->nr_running -= count;
2489 if (trace_sched_update_nr_running_tp_enabled()) {
2490 call_trace_sched_update_nr_running(rq, count: -count);
2491 }
2492
2493 /* Check if we still need preemption */
2494 sched_update_tick_dependency(rq);
2495}
2496
2497extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2498extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2499
2500extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2501
2502#ifdef CONFIG_PREEMPT_RT
2503#define SCHED_NR_MIGRATE_BREAK 8
2504#else
2505#define SCHED_NR_MIGRATE_BREAK 32
2506#endif
2507
2508extern const_debug unsigned int sysctl_sched_nr_migrate;
2509extern const_debug unsigned int sysctl_sched_migration_cost;
2510
2511extern unsigned int sysctl_sched_base_slice;
2512
2513#ifdef CONFIG_SCHED_DEBUG
2514extern int sysctl_resched_latency_warn_ms;
2515extern int sysctl_resched_latency_warn_once;
2516
2517extern unsigned int sysctl_sched_tunable_scaling;
2518
2519extern unsigned int sysctl_numa_balancing_scan_delay;
2520extern unsigned int sysctl_numa_balancing_scan_period_min;
2521extern unsigned int sysctl_numa_balancing_scan_period_max;
2522extern unsigned int sysctl_numa_balancing_scan_size;
2523extern unsigned int sysctl_numa_balancing_hot_threshold;
2524#endif
2525
2526#ifdef CONFIG_SCHED_HRTICK
2527
2528/*
2529 * Use hrtick when:
2530 * - enabled by features
2531 * - hrtimer is actually high res
2532 */
2533static inline int hrtick_enabled(struct rq *rq)
2534{
2535 if (!cpu_active(cpu: cpu_of(rq)))
2536 return 0;
2537 return hrtimer_is_hres_active(timer: &rq->hrtick_timer);
2538}
2539
2540static inline int hrtick_enabled_fair(struct rq *rq)
2541{
2542 if (!sched_feat(HRTICK))
2543 return 0;
2544 return hrtick_enabled(rq);
2545}
2546
2547static inline int hrtick_enabled_dl(struct rq *rq)
2548{
2549 if (!sched_feat(HRTICK_DL))
2550 return 0;
2551 return hrtick_enabled(rq);
2552}
2553
2554void hrtick_start(struct rq *rq, u64 delay);
2555
2556#else
2557
2558static inline int hrtick_enabled_fair(struct rq *rq)
2559{
2560 return 0;
2561}
2562
2563static inline int hrtick_enabled_dl(struct rq *rq)
2564{
2565 return 0;
2566}
2567
2568static inline int hrtick_enabled(struct rq *rq)
2569{
2570 return 0;
2571}
2572
2573#endif /* CONFIG_SCHED_HRTICK */
2574
2575#ifndef arch_scale_freq_tick
2576static __always_inline
2577void arch_scale_freq_tick(void)
2578{
2579}
2580#endif
2581
2582#ifndef arch_scale_freq_capacity
2583/**
2584 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2585 * @cpu: the CPU in question.
2586 *
2587 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2588 *
2589 * f_curr
2590 * ------ * SCHED_CAPACITY_SCALE
2591 * f_max
2592 */
2593static __always_inline
2594unsigned long arch_scale_freq_capacity(int cpu)
2595{
2596 return SCHED_CAPACITY_SCALE;
2597}
2598#endif
2599
2600#ifdef CONFIG_SCHED_DEBUG
2601/*
2602 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2603 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2604 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2605 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2606 */
2607static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2608{
2609 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2610 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2611#ifdef CONFIG_SMP
2612 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2613#endif
2614}
2615#else
2616static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2617#endif
2618
2619#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2620__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2621static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2622{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2623 _lock; return _t; }
2624
2625#ifdef CONFIG_SMP
2626
2627static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2628{
2629#ifdef CONFIG_SCHED_CORE
2630 /*
2631 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2632 * order by core-id first and cpu-id second.
2633 *
2634 * Notably:
2635 *
2636 * double_rq_lock(0,3); will take core-0, core-1 lock
2637 * double_rq_lock(1,2); will take core-1, core-0 lock
2638 *
2639 * when only cpu-id is considered.
2640 */
2641 if (rq1->core->cpu < rq2->core->cpu)
2642 return true;
2643 if (rq1->core->cpu > rq2->core->cpu)
2644 return false;
2645
2646 /*
2647 * __sched_core_flip() relies on SMT having cpu-id lock order.
2648 */
2649#endif
2650 return rq1->cpu < rq2->cpu;
2651}
2652
2653extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2654
2655#ifdef CONFIG_PREEMPTION
2656
2657/*
2658 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2659 * way at the expense of forcing extra atomic operations in all
2660 * invocations. This assures that the double_lock is acquired using the
2661 * same underlying policy as the spinlock_t on this architecture, which
2662 * reduces latency compared to the unfair variant below. However, it
2663 * also adds more overhead and therefore may reduce throughput.
2664 */
2665static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2666 __releases(this_rq->lock)
2667 __acquires(busiest->lock)
2668 __acquires(this_rq->lock)
2669{
2670 raw_spin_rq_unlock(rq: this_rq);
2671 double_rq_lock(rq1: this_rq, rq2: busiest);
2672
2673 return 1;
2674}
2675
2676#else
2677/*
2678 * Unfair double_lock_balance: Optimizes throughput at the expense of
2679 * latency by eliminating extra atomic operations when the locks are
2680 * already in proper order on entry. This favors lower CPU-ids and will
2681 * grant the double lock to lower CPUs over higher ids under contention,
2682 * regardless of entry order into the function.
2683 */
2684static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2685 __releases(this_rq->lock)
2686 __acquires(busiest->lock)
2687 __acquires(this_rq->lock)
2688{
2689 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2690 likely(raw_spin_rq_trylock(busiest))) {
2691 double_rq_clock_clear_update(this_rq, busiest);
2692 return 0;
2693 }
2694
2695 if (rq_order_less(this_rq, busiest)) {
2696 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2697 double_rq_clock_clear_update(this_rq, busiest);
2698 return 0;
2699 }
2700
2701 raw_spin_rq_unlock(this_rq);
2702 double_rq_lock(this_rq, busiest);
2703
2704 return 1;
2705}
2706
2707#endif /* CONFIG_PREEMPTION */
2708
2709/*
2710 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2711 */
2712static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2713{
2714 lockdep_assert_irqs_disabled();
2715
2716 return _double_lock_balance(this_rq, busiest);
2717}
2718
2719static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2720 __releases(busiest->lock)
2721{
2722 if (__rq_lockp(rq: this_rq) != __rq_lockp(rq: busiest))
2723 raw_spin_rq_unlock(rq: busiest);
2724 lock_set_subclass(lock: &__rq_lockp(rq: this_rq)->dep_map, subclass: 0, _RET_IP_);
2725}
2726
2727static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2728{
2729 if (l1 > l2)
2730 swap(l1, l2);
2731
2732 spin_lock(lock: l1);
2733 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2734}
2735
2736static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2737{
2738 if (l1 > l2)
2739 swap(l1, l2);
2740
2741 spin_lock_irq(lock: l1);
2742 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2743}
2744
2745static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2746{
2747 if (l1 > l2)
2748 swap(l1, l2);
2749
2750 raw_spin_lock(l1);
2751 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2752}
2753
2754static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2755{
2756 raw_spin_unlock(l1);
2757 raw_spin_unlock(l2);
2758}
2759
2760DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
2761 double_raw_lock(_T->lock, _T->lock2),
2762 double_raw_unlock(_T->lock, _T->lock2))
2763
2764/*
2765 * double_rq_unlock - safely unlock two runqueues
2766 *
2767 * Note this does not restore interrupts like task_rq_unlock,
2768 * you need to do so manually after calling.
2769 */
2770static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2771 __releases(rq1->lock)
2772 __releases(rq2->lock)
2773{
2774 if (__rq_lockp(rq: rq1) != __rq_lockp(rq: rq2))
2775 raw_spin_rq_unlock(rq: rq2);
2776 else
2777 __release(rq2->lock);
2778 raw_spin_rq_unlock(rq: rq1);
2779}
2780
2781extern void set_rq_online (struct rq *rq);
2782extern void set_rq_offline(struct rq *rq);
2783extern bool sched_smp_initialized;
2784
2785#else /* CONFIG_SMP */
2786
2787/*
2788 * double_rq_lock - safely lock two runqueues
2789 *
2790 * Note this does not disable interrupts like task_rq_lock,
2791 * you need to do so manually before calling.
2792 */
2793static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2794 __acquires(rq1->lock)
2795 __acquires(rq2->lock)
2796{
2797 WARN_ON_ONCE(!irqs_disabled());
2798 WARN_ON_ONCE(rq1 != rq2);
2799 raw_spin_rq_lock(rq1);
2800 __acquire(rq2->lock); /* Fake it out ;) */
2801 double_rq_clock_clear_update(rq1, rq2);
2802}
2803
2804/*
2805 * double_rq_unlock - safely unlock two runqueues
2806 *
2807 * Note this does not restore interrupts like task_rq_unlock,
2808 * you need to do so manually after calling.
2809 */
2810static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2811 __releases(rq1->lock)
2812 __releases(rq2->lock)
2813{
2814 WARN_ON_ONCE(rq1 != rq2);
2815 raw_spin_rq_unlock(rq1);
2816 __release(rq2->lock);
2817}
2818
2819#endif
2820
2821DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
2822 double_rq_lock(_T->lock, _T->lock2),
2823 double_rq_unlock(_T->lock, _T->lock2))
2824
2825extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2826extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2827
2828#ifdef CONFIG_SCHED_DEBUG
2829extern bool sched_debug_verbose;
2830
2831extern void print_cfs_stats(struct seq_file *m, int cpu);
2832extern void print_rt_stats(struct seq_file *m, int cpu);
2833extern void print_dl_stats(struct seq_file *m, int cpu);
2834extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2835extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2836extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2837
2838extern void resched_latency_warn(int cpu, u64 latency);
2839#ifdef CONFIG_NUMA_BALANCING
2840extern void
2841show_numa_stats(struct task_struct *p, struct seq_file *m);
2842extern void
2843print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2844 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2845#endif /* CONFIG_NUMA_BALANCING */
2846#else
2847static inline void resched_latency_warn(int cpu, u64 latency) {}
2848#endif /* CONFIG_SCHED_DEBUG */
2849
2850extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2851extern void init_rt_rq(struct rt_rq *rt_rq);
2852extern void init_dl_rq(struct dl_rq *dl_rq);
2853
2854extern void cfs_bandwidth_usage_inc(void);
2855extern void cfs_bandwidth_usage_dec(void);
2856
2857#ifdef CONFIG_NO_HZ_COMMON
2858#define NOHZ_BALANCE_KICK_BIT 0
2859#define NOHZ_STATS_KICK_BIT 1
2860#define NOHZ_NEWILB_KICK_BIT 2
2861#define NOHZ_NEXT_KICK_BIT 3
2862
2863/* Run rebalance_domains() */
2864#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2865/* Update blocked load */
2866#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2867/* Update blocked load when entering idle */
2868#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2869/* Update nohz.next_balance */
2870#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2871
2872#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2873
2874#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2875
2876extern void nohz_balance_exit_idle(struct rq *rq);
2877#else
2878static inline void nohz_balance_exit_idle(struct rq *rq) { }
2879#endif
2880
2881#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2882extern void nohz_run_idle_balance(int cpu);
2883#else
2884static inline void nohz_run_idle_balance(int cpu) { }
2885#endif
2886
2887#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2888struct irqtime {
2889 u64 total;
2890 u64 tick_delta;
2891 u64 irq_start_time;
2892 struct u64_stats_sync sync;
2893};
2894
2895DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2896
2897/*
2898 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2899 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2900 * and never move forward.
2901 */
2902static inline u64 irq_time_read(int cpu)
2903{
2904 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2905 unsigned int seq;
2906 u64 total;
2907
2908 do {
2909 seq = __u64_stats_fetch_begin(syncp: &irqtime->sync);
2910 total = irqtime->total;
2911 } while (__u64_stats_fetch_retry(syncp: &irqtime->sync, start: seq));
2912
2913 return total;
2914}
2915#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2916
2917#ifdef CONFIG_CPU_FREQ
2918DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2919
2920/**
2921 * cpufreq_update_util - Take a note about CPU utilization changes.
2922 * @rq: Runqueue to carry out the update for.
2923 * @flags: Update reason flags.
2924 *
2925 * This function is called by the scheduler on the CPU whose utilization is
2926 * being updated.
2927 *
2928 * It can only be called from RCU-sched read-side critical sections.
2929 *
2930 * The way cpufreq is currently arranged requires it to evaluate the CPU
2931 * performance state (frequency/voltage) on a regular basis to prevent it from
2932 * being stuck in a completely inadequate performance level for too long.
2933 * That is not guaranteed to happen if the updates are only triggered from CFS
2934 * and DL, though, because they may not be coming in if only RT tasks are
2935 * active all the time (or there are RT tasks only).
2936 *
2937 * As a workaround for that issue, this function is called periodically by the
2938 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2939 * but that really is a band-aid. Going forward it should be replaced with
2940 * solutions targeted more specifically at RT tasks.
2941 */
2942static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2943{
2944 struct update_util_data *data;
2945
2946 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2947 cpu_of(rq)));
2948 if (data)
2949 data->func(data, rq_clock(rq), flags);
2950}
2951#else
2952static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2953#endif /* CONFIG_CPU_FREQ */
2954
2955#ifdef arch_scale_freq_capacity
2956# ifndef arch_scale_freq_invariant
2957# define arch_scale_freq_invariant() true
2958# endif
2959#else
2960# define arch_scale_freq_invariant() false
2961#endif
2962
2963#ifdef CONFIG_SMP
2964/**
2965 * enum cpu_util_type - CPU utilization type
2966 * @FREQUENCY_UTIL: Utilization used to select frequency
2967 * @ENERGY_UTIL: Utilization used during energy calculation
2968 *
2969 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2970 * need to be aggregated differently depending on the usage made of them. This
2971 * enum is used within effective_cpu_util() to differentiate the types of
2972 * utilization expected by the callers, and adjust the aggregation accordingly.
2973 */
2974enum cpu_util_type {
2975 FREQUENCY_UTIL,
2976 ENERGY_UTIL,
2977};
2978
2979unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2980 enum cpu_util_type type,
2981 struct task_struct *p);
2982
2983/*
2984 * Verify the fitness of task @p to run on @cpu taking into account the
2985 * CPU original capacity and the runtime/deadline ratio of the task.
2986 *
2987 * The function will return true if the original capacity of @cpu is
2988 * greater than or equal to task's deadline density right shifted by
2989 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2990 */
2991static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2992{
2993 unsigned long cap = arch_scale_cpu_capacity(cpu);
2994
2995 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2996}
2997
2998static inline unsigned long cpu_bw_dl(struct rq *rq)
2999{
3000 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3001}
3002
3003static inline unsigned long cpu_util_dl(struct rq *rq)
3004{
3005 return READ_ONCE(rq->avg_dl.util_avg);
3006}
3007
3008
3009extern unsigned long cpu_util_cfs(int cpu);
3010extern unsigned long cpu_util_cfs_boost(int cpu);
3011
3012static inline unsigned long cpu_util_rt(struct rq *rq)
3013{
3014 return READ_ONCE(rq->avg_rt.util_avg);
3015}
3016#endif
3017
3018#ifdef CONFIG_UCLAMP_TASK
3019unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3020
3021static inline unsigned long uclamp_rq_get(struct rq *rq,
3022 enum uclamp_id clamp_id)
3023{
3024 return READ_ONCE(rq->uclamp[clamp_id].value);
3025}
3026
3027static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3028 unsigned int value)
3029{
3030 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3031}
3032
3033static inline bool uclamp_rq_is_idle(struct rq *rq)
3034{
3035 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3036}
3037
3038/**
3039 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3040 * @rq: The rq to clamp against. Must not be NULL.
3041 * @util: The util value to clamp.
3042 * @p: The task to clamp against. Can be NULL if you want to clamp
3043 * against @rq only.
3044 *
3045 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3046 *
3047 * If sched_uclamp_used static key is disabled, then just return the util
3048 * without any clamping since uclamp aggregation at the rq level in the fast
3049 * path is disabled, rendering this operation a NOP.
3050 *
3051 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3052 * will return the correct effective uclamp value of the task even if the
3053 * static key is disabled.
3054 */
3055static __always_inline
3056unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3057 struct task_struct *p)
3058{
3059 unsigned long min_util = 0;
3060 unsigned long max_util = 0;
3061
3062 if (!static_branch_likely(&sched_uclamp_used))
3063 return util;
3064
3065 if (p) {
3066 min_util = uclamp_eff_value(p, clamp_id: UCLAMP_MIN);
3067 max_util = uclamp_eff_value(p, clamp_id: UCLAMP_MAX);
3068
3069 /*
3070 * Ignore last runnable task's max clamp, as this task will
3071 * reset it. Similarly, no need to read the rq's min clamp.
3072 */
3073 if (uclamp_rq_is_idle(rq))
3074 goto out;
3075 }
3076
3077 min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3078 max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3079out:
3080 /*
3081 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3082 * RUNNABLE tasks with _different_ clamps, we can end up with an
3083 * inversion. Fix it now when the clamps are applied.
3084 */
3085 if (unlikely(min_util >= max_util))
3086 return min_util;
3087
3088 return clamp(util, min_util, max_util);
3089}
3090
3091/* Is the rq being capped/throttled by uclamp_max? */
3092static inline bool uclamp_rq_is_capped(struct rq *rq)
3093{
3094 unsigned long rq_util;
3095 unsigned long max_util;
3096
3097 if (!static_branch_likely(&sched_uclamp_used))
3098 return false;
3099
3100 rq_util = cpu_util_cfs(cpu: cpu_of(rq)) + cpu_util_rt(rq);
3101 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3102
3103 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3104}
3105
3106/*
3107 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3108 * by default in the fast path and only gets turned on once userspace performs
3109 * an operation that requires it.
3110 *
3111 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3112 * hence is active.
3113 */
3114static inline bool uclamp_is_used(void)
3115{
3116 return static_branch_likely(&sched_uclamp_used);
3117}
3118#else /* CONFIG_UCLAMP_TASK */
3119static inline unsigned long uclamp_eff_value(struct task_struct *p,
3120 enum uclamp_id clamp_id)
3121{
3122 if (clamp_id == UCLAMP_MIN)
3123 return 0;
3124
3125 return SCHED_CAPACITY_SCALE;
3126}
3127
3128static inline
3129unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3130 struct task_struct *p)
3131{
3132 return util;
3133}
3134
3135static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3136
3137static inline bool uclamp_is_used(void)
3138{
3139 return false;
3140}
3141
3142static inline unsigned long uclamp_rq_get(struct rq *rq,
3143 enum uclamp_id clamp_id)
3144{
3145 if (clamp_id == UCLAMP_MIN)
3146 return 0;
3147
3148 return SCHED_CAPACITY_SCALE;
3149}
3150
3151static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3152 unsigned int value)
3153{
3154}
3155
3156static inline bool uclamp_rq_is_idle(struct rq *rq)
3157{
3158 return false;
3159}
3160#endif /* CONFIG_UCLAMP_TASK */
3161
3162#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3163static inline unsigned long cpu_util_irq(struct rq *rq)
3164{
3165 return rq->avg_irq.util_avg;
3166}
3167
3168static inline
3169unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3170{
3171 util *= (max - irq);
3172 util /= max;
3173
3174 return util;
3175
3176}
3177#else
3178static inline unsigned long cpu_util_irq(struct rq *rq)
3179{
3180 return 0;
3181}
3182
3183static inline
3184unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3185{
3186 return util;
3187}
3188#endif
3189
3190#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3191
3192#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3193
3194DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3195
3196static inline bool sched_energy_enabled(void)
3197{
3198 return static_branch_unlikely(&sched_energy_present);
3199}
3200
3201extern struct cpufreq_governor schedutil_gov;
3202
3203#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3204
3205#define perf_domain_span(pd) NULL
3206static inline bool sched_energy_enabled(void) { return false; }
3207
3208#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3209
3210#ifdef CONFIG_MEMBARRIER
3211/*
3212 * The scheduler provides memory barriers required by membarrier between:
3213 * - prior user-space memory accesses and store to rq->membarrier_state,
3214 * - store to rq->membarrier_state and following user-space memory accesses.
3215 * In the same way it provides those guarantees around store to rq->curr.
3216 */
3217static inline void membarrier_switch_mm(struct rq *rq,
3218 struct mm_struct *prev_mm,
3219 struct mm_struct *next_mm)
3220{
3221 int membarrier_state;
3222
3223 if (prev_mm == next_mm)
3224 return;
3225
3226 membarrier_state = atomic_read(v: &next_mm->membarrier_state);
3227 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3228 return;
3229
3230 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3231}
3232#else
3233static inline void membarrier_switch_mm(struct rq *rq,
3234 struct mm_struct *prev_mm,
3235 struct mm_struct *next_mm)
3236{
3237}
3238#endif
3239
3240#ifdef CONFIG_SMP
3241static inline bool is_per_cpu_kthread(struct task_struct *p)
3242{
3243 if (!(p->flags & PF_KTHREAD))
3244 return false;
3245
3246 if (p->nr_cpus_allowed != 1)
3247 return false;
3248
3249 return true;
3250}
3251#endif
3252
3253extern void swake_up_all_locked(struct swait_queue_head *q);
3254extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3255
3256extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3257
3258#ifdef CONFIG_PREEMPT_DYNAMIC
3259extern int preempt_dynamic_mode;
3260extern int sched_dynamic_mode(const char *str);
3261extern void sched_dynamic_update(int mode);
3262#endif
3263
3264static inline void update_current_exec_runtime(struct task_struct *curr,
3265 u64 now, u64 delta_exec)
3266{
3267 curr->se.sum_exec_runtime += delta_exec;
3268 account_group_exec_runtime(tsk: curr, ns: delta_exec);
3269
3270 curr->se.exec_start = now;
3271 cgroup_account_cputime(task: curr, delta_exec);
3272}
3273
3274#ifdef CONFIG_SCHED_MM_CID
3275
3276#define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3277#define MM_CID_SCAN_DELAY 100 /* 100ms */
3278
3279extern raw_spinlock_t cid_lock;
3280extern int use_cid_lock;
3281
3282extern void sched_mm_cid_migrate_from(struct task_struct *t);
3283extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3284extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3285extern void init_sched_mm_cid(struct task_struct *t);
3286
3287static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3288{
3289 if (cid < 0)
3290 return;
3291 cpumask_clear_cpu(cpu: cid, dstp: mm_cidmask(mm));
3292}
3293
3294/*
3295 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3296 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3297 * be held to transition to other states.
3298 *
3299 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3300 * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3301 */
3302static inline void mm_cid_put_lazy(struct task_struct *t)
3303{
3304 struct mm_struct *mm = t->mm;
3305 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3306 int cid;
3307
3308 lockdep_assert_irqs_disabled();
3309 cid = __this_cpu_read(pcpu_cid->cid);
3310 if (!mm_cid_is_lazy_put(cid) ||
3311 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3312 return;
3313 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3314}
3315
3316static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3317{
3318 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3319 int cid, res;
3320
3321 lockdep_assert_irqs_disabled();
3322 cid = __this_cpu_read(pcpu_cid->cid);
3323 for (;;) {
3324 if (mm_cid_is_unset(cid))
3325 return MM_CID_UNSET;
3326 /*
3327 * Attempt transition from valid or lazy-put to unset.
3328 */
3329 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3330 if (res == cid)
3331 break;
3332 cid = res;
3333 }
3334 return cid;
3335}
3336
3337static inline void mm_cid_put(struct mm_struct *mm)
3338{
3339 int cid;
3340
3341 lockdep_assert_irqs_disabled();
3342 cid = mm_cid_pcpu_unset(mm);
3343 if (cid == MM_CID_UNSET)
3344 return;
3345 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3346}
3347
3348static inline int __mm_cid_try_get(struct mm_struct *mm)
3349{
3350 struct cpumask *cpumask;
3351 int cid;
3352
3353 cpumask = mm_cidmask(mm);
3354 /*
3355 * Retry finding first zero bit if the mask is temporarily
3356 * filled. This only happens during concurrent remote-clear
3357 * which owns a cid without holding a rq lock.
3358 */
3359 for (;;) {
3360 cid = cpumask_first_zero(srcp: cpumask);
3361 if (cid < nr_cpu_ids)
3362 break;
3363 cpu_relax();
3364 }
3365 if (cpumask_test_and_set_cpu(cpu: cid, cpumask))
3366 return -1;
3367 return cid;
3368}
3369
3370/*
3371 * Save a snapshot of the current runqueue time of this cpu
3372 * with the per-cpu cid value, allowing to estimate how recently it was used.
3373 */
3374static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3375{
3376 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3377
3378 lockdep_assert_rq_held(rq);
3379 WRITE_ONCE(pcpu_cid->time, rq->clock);
3380}
3381
3382static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3383{
3384 int cid;
3385
3386 /*
3387 * All allocations (even those using the cid_lock) are lock-free. If
3388 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3389 * guarantee forward progress.
3390 */
3391 if (!READ_ONCE(use_cid_lock)) {
3392 cid = __mm_cid_try_get(mm);
3393 if (cid >= 0)
3394 goto end;
3395 raw_spin_lock(&cid_lock);
3396 } else {
3397 raw_spin_lock(&cid_lock);
3398 cid = __mm_cid_try_get(mm);
3399 if (cid >= 0)
3400 goto unlock;
3401 }
3402
3403 /*
3404 * cid concurrently allocated. Retry while forcing following
3405 * allocations to use the cid_lock to ensure forward progress.
3406 */
3407 WRITE_ONCE(use_cid_lock, 1);
3408 /*
3409 * Set use_cid_lock before allocation. Only care about program order
3410 * because this is only required for forward progress.
3411 */
3412 barrier();
3413 /*
3414 * Retry until it succeeds. It is guaranteed to eventually succeed once
3415 * all newcoming allocations observe the use_cid_lock flag set.
3416 */
3417 do {
3418 cid = __mm_cid_try_get(mm);
3419 cpu_relax();
3420 } while (cid < 0);
3421 /*
3422 * Allocate before clearing use_cid_lock. Only care about
3423 * program order because this is for forward progress.
3424 */
3425 barrier();
3426 WRITE_ONCE(use_cid_lock, 0);
3427unlock:
3428 raw_spin_unlock(&cid_lock);
3429end:
3430 mm_cid_snapshot_time(rq, mm);
3431 return cid;
3432}
3433
3434static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3435{
3436 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3437 struct cpumask *cpumask;
3438 int cid;
3439
3440 lockdep_assert_rq_held(rq);
3441 cpumask = mm_cidmask(mm);
3442 cid = __this_cpu_read(pcpu_cid->cid);
3443 if (mm_cid_is_valid(cid)) {
3444 mm_cid_snapshot_time(rq, mm);
3445 return cid;
3446 }
3447 if (mm_cid_is_lazy_put(cid)) {
3448 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3449 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3450 }
3451 cid = __mm_cid_get(rq, mm);
3452 __this_cpu_write(pcpu_cid->cid, cid);
3453 return cid;
3454}
3455
3456static inline void switch_mm_cid(struct rq *rq,
3457 struct task_struct *prev,
3458 struct task_struct *next)
3459{
3460 /*
3461 * Provide a memory barrier between rq->curr store and load of
3462 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3463 *
3464 * Should be adapted if context_switch() is modified.
3465 */
3466 if (!next->mm) { // to kernel
3467 /*
3468 * user -> kernel transition does not guarantee a barrier, but
3469 * we can use the fact that it performs an atomic operation in
3470 * mmgrab().
3471 */
3472 if (prev->mm) // from user
3473 smp_mb__after_mmgrab();
3474 /*
3475 * kernel -> kernel transition does not change rq->curr->mm
3476 * state. It stays NULL.
3477 */
3478 } else { // to user
3479 /*
3480 * kernel -> user transition does not provide a barrier
3481 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3482 * Provide it here.
3483 */
3484 if (!prev->mm) // from kernel
3485 smp_mb();
3486 /*
3487 * user -> user transition guarantees a memory barrier through
3488 * switch_mm() when current->mm changes. If current->mm is
3489 * unchanged, no barrier is needed.
3490 */
3491 }
3492 if (prev->mm_cid_active) {
3493 mm_cid_snapshot_time(rq, mm: prev->mm);
3494 mm_cid_put_lazy(t: prev);
3495 prev->mm_cid = -1;
3496 }
3497 if (next->mm_cid_active)
3498 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, mm: next->mm);
3499}
3500
3501#else
3502static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3503static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3504static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3505static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3506static inline void init_sched_mm_cid(struct task_struct *t) { }
3507#endif
3508
3509extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3510extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3511
3512#endif /* _KERNEL_SCHED_SCHED_H */
3513

source code of linux/kernel/sched/sched.h