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/prandom.h>
9#include <linux/sched/affinity.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/cpufreq.h>
12#include <linux/sched/deadline.h>
13#include <linux/sched.h>
14#include <linux/sched/loadavg.h>
15#include <linux/sched/mm.h>
16#include <linux/sched/rseq_api.h>
17#include <linux/sched/signal.h>
18#include <linux/sched/smt.h>
19#include <linux/sched/stat.h>
20#include <linux/sched/sysctl.h>
21#include <linux/sched/task_flags.h>
22#include <linux/sched/task.h>
23#include <linux/sched/topology.h>
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#include <linux/delayacct.h>
72#include <linux/mmu_context.h>
73
74#include <trace/events/power.h>
75#include <trace/events/sched.h>
76
77#include "../workqueue_internal.h"
78
79struct rq;
80struct cfs_rq;
81struct rt_rq;
82struct sched_group;
83struct cpuidle_state;
84
85#ifdef CONFIG_PARAVIRT
86# include <asm/paravirt.h>
87# include <asm/paravirt_api_clock.h>
88#endif
89
90#include <asm/barrier.h>
91
92#include "cpupri.h"
93#include "cpudeadline.h"
94
95/* task_struct::on_rq states: */
96#define TASK_ON_RQ_QUEUED 1
97#define TASK_ON_RQ_MIGRATING 2
98
99extern __read_mostly int scheduler_running;
100
101extern unsigned long calc_load_update;
102extern atomic_long_t calc_load_tasks;
103
104extern void calc_global_load_tick(struct rq *this_rq);
105extern long calc_load_fold_active(struct rq *this_rq, long adjust);
106
107extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
108
109extern int sysctl_sched_rt_period;
110extern int sysctl_sched_rt_runtime;
111extern int sched_rr_timeslice;
112
113/*
114 * Asymmetric CPU capacity bits
115 */
116struct asym_cap_data {
117 struct list_head link;
118 struct rcu_head rcu;
119 unsigned long capacity;
120 unsigned long cpus[];
121};
122
123extern struct list_head asym_cap_list;
124
125#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
126
127/*
128 * Helpers for converting nanosecond timing to jiffy resolution
129 */
130#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
131
132/*
133 * Increase resolution of nice-level calculations for 64-bit architectures.
134 * The extra resolution improves shares distribution and load balancing of
135 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
136 * hierarchies, especially on larger systems. This is not a user-visible change
137 * and does not change the user-interface for setting shares/weights.
138 *
139 * We increase resolution only if we have enough bits to allow this increased
140 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
141 * are pretty high and the returns do not justify the increased costs.
142 *
143 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
144 * increase coverage and consistency always enable it on 64-bit platforms.
145 */
146#ifdef CONFIG_64BIT
147# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
148# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
149# define scale_load_down(w) \
150({ \
151 unsigned long __w = (w); \
152 \
153 if (__w) \
154 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
155 __w; \
156})
157#else
158# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
159# define scale_load(w) (w)
160# define scale_load_down(w) (w)
161#endif
162
163/*
164 * Task weight (visible to users) and its load (invisible to users) have
165 * independent resolution, but they should be well calibrated. We use
166 * scale_load() and scale_load_down(w) to convert between them. The
167 * following must be true:
168 *
169 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
170 *
171 */
172#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
173
174/*
175 * Single value that decides SCHED_DEADLINE internal math precision.
176 * 10 -> just above 1us
177 * 9 -> just above 0.5us
178 */
179#define DL_SCALE 10
180
181/*
182 * Single value that denotes runtime == period, ie unlimited time.
183 */
184#define RUNTIME_INF ((u64)~0ULL)
185
186static inline int idle_policy(int policy)
187{
188 return policy == SCHED_IDLE;
189}
190
191static inline int normal_policy(int policy)
192{
193#ifdef CONFIG_SCHED_CLASS_EXT
194 if (policy == SCHED_EXT)
195 return true;
196#endif
197 return policy == SCHED_NORMAL;
198}
199
200static inline int fair_policy(int policy)
201{
202 return normal_policy(policy) || policy == SCHED_BATCH;
203}
204
205static inline int rt_policy(int policy)
206{
207 return policy == SCHED_FIFO || policy == SCHED_RR;
208}
209
210static inline int dl_policy(int policy)
211{
212 return policy == SCHED_DEADLINE;
213}
214
215static inline bool valid_policy(int policy)
216{
217 return idle_policy(policy) || fair_policy(policy) ||
218 rt_policy(policy) || dl_policy(policy);
219}
220
221static inline int task_has_idle_policy(struct task_struct *p)
222{
223 return idle_policy(policy: p->policy);
224}
225
226static inline int task_has_rt_policy(struct task_struct *p)
227{
228 return rt_policy(policy: p->policy);
229}
230
231static inline int task_has_dl_policy(struct task_struct *p)
232{
233 return dl_policy(policy: p->policy);
234}
235
236#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
237
238static inline void update_avg(u64 *avg, u64 sample)
239{
240 s64 diff = sample - *avg;
241
242 *avg += diff / 8;
243}
244
245/*
246 * Shifting a value by an exponent greater *or equal* to the size of said value
247 * is UB; cap at size-1.
248 */
249#define shr_bound(val, shift) \
250 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
251
252/*
253 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
254 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
255 * maps pretty well onto the shares value used by scheduler and the round-trip
256 * conversions preserve the original value over the entire range.
257 */
258static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
259{
260 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
261}
262
263static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
264{
265 return clamp_t(unsigned long,
266 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
267 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
268}
269
270/*
271 * !! For sched_setattr_nocheck() (kernel) only !!
272 *
273 * This is actually gross. :(
274 *
275 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
276 * tasks, but still be able to sleep. We need this on platforms that cannot
277 * atomically change clock frequency. Remove once fast switching will be
278 * available on such platforms.
279 *
280 * SUGOV stands for SchedUtil GOVernor.
281 */
282#define SCHED_FLAG_SUGOV 0x10000000
283
284#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
285
286static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
287{
288#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
289 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
290#else
291 return false;
292#endif
293}
294
295/*
296 * Tells if entity @a should preempt entity @b.
297 */
298static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
299 const struct sched_dl_entity *b)
300{
301 return dl_entity_is_special(dl_se: a) ||
302 dl_time_before(a: a->deadline, b: b->deadline);
303}
304
305/*
306 * This is the priority-queue data structure of the RT scheduling class:
307 */
308struct rt_prio_array {
309 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
310 struct list_head queue[MAX_RT_PRIO];
311};
312
313struct rt_bandwidth {
314 /* nests inside the rq lock: */
315 raw_spinlock_t rt_runtime_lock;
316 ktime_t rt_period;
317 u64 rt_runtime;
318 struct hrtimer rt_period_timer;
319 unsigned int rt_period_active;
320};
321
322static inline int dl_bandwidth_enabled(void)
323{
324 return sysctl_sched_rt_runtime >= 0;
325}
326
327/*
328 * To keep the bandwidth of -deadline tasks under control
329 * we need some place where:
330 * - store the maximum -deadline bandwidth of each cpu;
331 * - cache the fraction of bandwidth that is currently allocated in
332 * each root domain;
333 *
334 * This is all done in the data structure below. It is similar to the
335 * one used for RT-throttling (rt_bandwidth), with the main difference
336 * that, since here we are only interested in admission control, we
337 * do not decrease any runtime while the group "executes", neither we
338 * need a timer to replenish it.
339 *
340 * With respect to SMP, bandwidth is given on a per root domain basis,
341 * meaning that:
342 * - bw (< 100%) is the deadline bandwidth of each CPU;
343 * - total_bw is the currently allocated bandwidth in each root domain;
344 */
345struct dl_bw {
346 raw_spinlock_t lock;
347 u64 bw;
348 u64 total_bw;
349};
350
351extern void init_dl_bw(struct dl_bw *dl_b);
352extern int sched_dl_global_validate(void);
353extern void sched_dl_do_global(void);
354extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
355extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
356extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
357extern bool __checkparam_dl(const struct sched_attr *attr);
358extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
359extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
360extern int dl_bw_deactivate(int cpu);
361extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
362/*
363 * SCHED_DEADLINE supports servers (nested scheduling) with the following
364 * interface:
365 *
366 * dl_se::rq -- runqueue we belong to.
367 *
368 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
369 * returns NULL.
370 *
371 * dl_server_update() -- called from update_curr_common(), propagates runtime
372 * to the server.
373 *
374 * dl_server_start() -- start the server when it has tasks; it will stop
375 * automatically when there are no more tasks, per
376 * dl_se::server_pick() returning NULL.
377 *
378 * dl_server_stop() -- (force) stop the server; use when updating
379 * parameters.
380 *
381 * dl_server_init() -- initializes the server.
382 *
383 * When started the dl_server will (per dl_defer) schedule a timer for its
384 * zero-laxity point -- that is, unlike regular EDF tasks which run ASAP, a
385 * server will run at the very end of its period.
386 *
387 * This is done such that any runtime from the target class can be accounted
388 * against the server -- through dl_server_update() above -- such that when it
389 * becomes time to run, it might already be out of runtime and get deferred
390 * until the next period. In this case dl_server_timer() will alternate
391 * between defer and replenish but never actually enqueue the server.
392 *
393 * Only when the target class does not manage to exhaust the server's runtime
394 * (there's actualy starvation in the given period), will the dl_server get on
395 * the runqueue. Once queued it will pick tasks from the target class and run
396 * them until either its runtime is exhaused, at which point its back to
397 * dl_server_timer, or until there are no more tasks to run, at which point
398 * the dl_server stops itself.
399 *
400 * By stopping at this point the dl_server retains bandwidth, which, if a new
401 * task wakes up imminently (starting the server again), can be used --
402 * subject to CBS wakeup rules -- without having to wait for the next period.
403 *
404 * Additionally, because of the dl_defer behaviour the start/stop behaviour is
405 * naturally thottled to once per period, avoiding high context switch
406 * workloads from spamming the hrtimer program/cancel paths.
407 */
408extern void dl_server_update_idle(struct sched_dl_entity *dl_se, s64 delta_exec);
409extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
410extern void dl_server_start(struct sched_dl_entity *dl_se);
411extern void dl_server_stop(struct sched_dl_entity *dl_se);
412extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
413 dl_server_pick_f pick_task);
414extern void sched_init_dl_servers(void);
415
416extern void fair_server_init(struct rq *rq);
417extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
418extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
419 u64 runtime, u64 period, bool init);
420
421static inline bool dl_server_active(struct sched_dl_entity *dl_se)
422{
423 return dl_se->dl_server_active;
424}
425
426#ifdef CONFIG_CGROUP_SCHED
427
428extern struct list_head task_groups;
429
430#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
431extern const u64 max_bw_quota_period_us;
432
433/*
434 * default period for group bandwidth.
435 * default: 0.1s, units: microseconds
436 */
437static inline u64 default_bw_period_us(void)
438{
439 return 100000ULL;
440}
441#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */
442
443struct cfs_bandwidth {
444#ifdef CONFIG_CFS_BANDWIDTH
445 raw_spinlock_t lock;
446 ktime_t period;
447 u64 quota;
448 u64 runtime;
449 u64 burst;
450 u64 runtime_snap;
451 s64 hierarchical_quota;
452
453 u8 idle;
454 u8 period_active;
455 u8 slack_started;
456 struct hrtimer period_timer;
457 struct hrtimer slack_timer;
458 struct list_head throttled_cfs_rq;
459
460 /* Statistics: */
461 int nr_periods;
462 int nr_throttled;
463 int nr_burst;
464 u64 throttled_time;
465 u64 burst_time;
466#endif /* CONFIG_CFS_BANDWIDTH */
467};
468
469/* Task group related information */
470struct task_group {
471 struct cgroup_subsys_state css;
472
473#ifdef CONFIG_GROUP_SCHED_WEIGHT
474 /* A positive value indicates that this is a SCHED_IDLE group. */
475 int idle;
476#endif
477
478#ifdef CONFIG_FAIR_GROUP_SCHED
479 /* schedulable entities of this group on each CPU */
480 struct sched_entity **se;
481 /* runqueue "owned" by this group on each CPU */
482 struct cfs_rq **cfs_rq;
483 unsigned long shares;
484 /*
485 * load_avg can be heavily contended at clock tick time, so put
486 * it in its own cache-line separated from the fields above which
487 * will also be accessed at each tick.
488 */
489 atomic_long_t load_avg ____cacheline_aligned;
490#endif /* CONFIG_FAIR_GROUP_SCHED */
491
492#ifdef CONFIG_RT_GROUP_SCHED
493 struct sched_rt_entity **rt_se;
494 struct rt_rq **rt_rq;
495
496 struct rt_bandwidth rt_bandwidth;
497#endif
498
499 struct scx_task_group scx;
500
501 struct rcu_head rcu;
502 struct list_head list;
503
504 struct task_group *parent;
505 struct list_head siblings;
506 struct list_head children;
507
508#ifdef CONFIG_SCHED_AUTOGROUP
509 struct autogroup *autogroup;
510#endif
511
512 struct cfs_bandwidth cfs_bandwidth;
513
514#ifdef CONFIG_UCLAMP_TASK_GROUP
515 /* The two decimal precision [%] value requested from user-space */
516 unsigned int uclamp_pct[UCLAMP_CNT];
517 /* Clamp values requested for a task group */
518 struct uclamp_se uclamp_req[UCLAMP_CNT];
519 /* Effective clamp values used for a task group */
520 struct uclamp_se uclamp[UCLAMP_CNT];
521#endif
522
523};
524
525#ifdef CONFIG_GROUP_SCHED_WEIGHT
526#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
527
528/*
529 * A weight of 0 or 1 can cause arithmetics problems.
530 * A weight of a cfs_rq is the sum of weights of which entities
531 * are queued on this cfs_rq, so a weight of a entity should not be
532 * too large, so as the shares value of a task group.
533 * (The default weight is 1024 - so there's no practical
534 * limitation from this.)
535 */
536#define MIN_SHARES (1UL << 1)
537#define MAX_SHARES (1UL << 18)
538#endif
539
540typedef int (*tg_visitor)(struct task_group *, void *);
541
542extern int walk_tg_tree_from(struct task_group *from,
543 tg_visitor down, tg_visitor up, void *data);
544
545/*
546 * Iterate the full tree, calling @down when first entering a node and @up when
547 * leaving it for the final time.
548 *
549 * Caller must hold rcu_lock or sufficient equivalent.
550 */
551static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
552{
553 return walk_tg_tree_from(from: &root_task_group, down, up, data);
554}
555
556static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
557{
558 return css ? container_of(css, struct task_group, css) : NULL;
559}
560
561extern int tg_nop(struct task_group *tg, void *data);
562
563#ifdef CONFIG_FAIR_GROUP_SCHED
564extern void free_fair_sched_group(struct task_group *tg);
565extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
566extern void online_fair_sched_group(struct task_group *tg);
567extern void unregister_fair_sched_group(struct task_group *tg);
568#else /* !CONFIG_FAIR_GROUP_SCHED: */
569static inline void free_fair_sched_group(struct task_group *tg) { }
570static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
571{
572 return 1;
573}
574static inline void online_fair_sched_group(struct task_group *tg) { }
575static inline void unregister_fair_sched_group(struct task_group *tg) { }
576#endif /* !CONFIG_FAIR_GROUP_SCHED */
577
578extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
579 struct sched_entity *se, int cpu,
580 struct sched_entity *parent);
581extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
582
583extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
584extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
585extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
586extern bool cfs_task_bw_constrained(struct task_struct *p);
587
588extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
589 struct sched_rt_entity *rt_se, int cpu,
590 struct sched_rt_entity *parent);
591extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
592extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
593extern long sched_group_rt_runtime(struct task_group *tg);
594extern long sched_group_rt_period(struct task_group *tg);
595extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
596
597extern struct task_group *sched_create_group(struct task_group *parent);
598extern void sched_online_group(struct task_group *tg,
599 struct task_group *parent);
600extern void sched_destroy_group(struct task_group *tg);
601extern void sched_release_group(struct task_group *tg);
602
603extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
604
605#ifdef CONFIG_FAIR_GROUP_SCHED
606extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
607
608extern int sched_group_set_idle(struct task_group *tg, long idle);
609
610extern void set_task_rq_fair(struct sched_entity *se,
611 struct cfs_rq *prev, struct cfs_rq *next);
612#else /* !CONFIG_FAIR_GROUP_SCHED: */
613static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
614static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
615#endif /* !CONFIG_FAIR_GROUP_SCHED */
616
617#else /* !CONFIG_CGROUP_SCHED: */
618
619struct cfs_bandwidth { };
620
621static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
622
623#endif /* !CONFIG_CGROUP_SCHED */
624
625extern void unregister_rt_sched_group(struct task_group *tg);
626extern void free_rt_sched_group(struct task_group *tg);
627extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
628
629/*
630 * u64_u32_load/u64_u32_store
631 *
632 * Use a copy of a u64 value to protect against data race. This is only
633 * applicable for 32-bits architectures.
634 */
635#ifdef CONFIG_64BIT
636# define u64_u32_load_copy(var, copy) var
637# define u64_u32_store_copy(var, copy, val) (var = val)
638#else
639# define u64_u32_load_copy(var, copy) \
640({ \
641 u64 __val, __val_copy; \
642 do { \
643 __val_copy = copy; \
644 /* \
645 * paired with u64_u32_store_copy(), ordering access \
646 * to var and copy. \
647 */ \
648 smp_rmb(); \
649 __val = var; \
650 } while (__val != __val_copy); \
651 __val; \
652})
653# define u64_u32_store_copy(var, copy, val) \
654do { \
655 typeof(val) __val = (val); \
656 var = __val; \
657 /* \
658 * paired with u64_u32_load_copy(), ordering access to var and \
659 * copy. \
660 */ \
661 smp_wmb(); \
662 copy = __val; \
663} while (0)
664#endif
665# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
666# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
667
668struct balance_callback {
669 struct balance_callback *next;
670 void (*func)(struct rq *rq);
671};
672
673/* CFS-related fields in a runqueue */
674struct cfs_rq {
675 struct load_weight load;
676 unsigned int nr_queued;
677 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */
678 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */
679 unsigned int h_nr_idle; /* SCHED_IDLE */
680
681 s64 avg_vruntime;
682 u64 avg_load;
683
684 u64 zero_vruntime;
685#ifdef CONFIG_SCHED_CORE
686 unsigned int forceidle_seq;
687 u64 zero_vruntime_fi;
688#endif
689
690 struct rb_root_cached tasks_timeline;
691
692 /*
693 * 'curr' points to currently running entity on this cfs_rq.
694 * It is set to NULL otherwise (i.e when none are currently running).
695 */
696 struct sched_entity *curr;
697 struct sched_entity *next;
698
699 /*
700 * CFS load tracking
701 */
702 struct sched_avg avg;
703#ifndef CONFIG_64BIT
704 u64 last_update_time_copy;
705#endif
706 struct {
707 raw_spinlock_t lock ____cacheline_aligned;
708 int nr;
709 unsigned long load_avg;
710 unsigned long util_avg;
711 unsigned long runnable_avg;
712 } removed;
713
714#ifdef CONFIG_FAIR_GROUP_SCHED
715 u64 last_update_tg_load_avg;
716 unsigned long tg_load_avg_contrib;
717 long propagate;
718 long prop_runnable_sum;
719
720 /*
721 * h_load = weight * f(tg)
722 *
723 * Where f(tg) is the recursive weight fraction assigned to
724 * this group.
725 */
726 unsigned long h_load;
727 u64 last_h_load_update;
728 struct sched_entity *h_load_next;
729#endif /* CONFIG_FAIR_GROUP_SCHED */
730
731#ifdef CONFIG_FAIR_GROUP_SCHED
732 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
733
734 /*
735 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
736 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
737 * (like users, containers etc.)
738 *
739 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
740 * This list is used during load balance.
741 */
742 int on_list;
743 struct list_head leaf_cfs_rq_list;
744 struct task_group *tg; /* group that "owns" this runqueue */
745
746 /* Locally cached copy of our task_group's idle value */
747 int idle;
748
749#ifdef CONFIG_CFS_BANDWIDTH
750 int runtime_enabled;
751 s64 runtime_remaining;
752
753 u64 throttled_pelt_idle;
754#ifndef CONFIG_64BIT
755 u64 throttled_pelt_idle_copy;
756#endif
757 u64 throttled_clock;
758 u64 throttled_clock_pelt;
759 u64 throttled_clock_pelt_time;
760 u64 throttled_clock_self;
761 u64 throttled_clock_self_time;
762 bool throttled:1;
763 bool pelt_clock_throttled:1;
764 int throttle_count;
765 struct list_head throttled_list;
766 struct list_head throttled_csd_list;
767 struct list_head throttled_limbo_list;
768#endif /* CONFIG_CFS_BANDWIDTH */
769#endif /* CONFIG_FAIR_GROUP_SCHED */
770};
771
772#ifdef CONFIG_SCHED_CLASS_EXT
773/* scx_rq->flags, protected by the rq lock */
774enum scx_rq_flags {
775 /*
776 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
777 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
778 * only while the BPF scheduler considers the CPU to be online.
779 */
780 SCX_RQ_ONLINE = 1 << 0,
781 SCX_RQ_CAN_STOP_TICK = 1 << 1,
782 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
783 SCX_RQ_BYPASSING = 1 << 4,
784 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */
785 SCX_RQ_BAL_CB_PENDING = 1 << 6, /* must queue a cb after dispatching */
786
787 SCX_RQ_IN_WAKEUP = 1 << 16,
788 SCX_RQ_IN_BALANCE = 1 << 17,
789};
790
791struct scx_rq {
792 struct scx_dispatch_q local_dsq;
793 struct list_head runnable_list; /* runnable tasks on this rq */
794 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
795 unsigned long ops_qseq;
796 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
797 u32 nr_running;
798 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
799 bool cpu_released;
800 u32 flags;
801 u64 clock; /* current per-rq clock -- see scx_bpf_now() */
802 cpumask_var_t cpus_to_kick;
803 cpumask_var_t cpus_to_kick_if_idle;
804 cpumask_var_t cpus_to_preempt;
805 cpumask_var_t cpus_to_wait;
806 unsigned long kick_sync;
807 local_t reenq_local_deferred;
808 struct balance_callback deferred_bal_cb;
809 struct irq_work deferred_irq_work;
810 struct irq_work kick_cpus_irq_work;
811 struct scx_dispatch_q bypass_dsq;
812};
813#endif /* CONFIG_SCHED_CLASS_EXT */
814
815static inline int rt_bandwidth_enabled(void)
816{
817 return sysctl_sched_rt_runtime >= 0;
818}
819
820/* RT IPI pull logic requires IRQ_WORK */
821#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
822# define HAVE_RT_PUSH_IPI
823#endif
824
825/* Real-Time classes' related field in a runqueue: */
826struct rt_rq {
827 struct rt_prio_array active;
828 unsigned int rt_nr_running;
829 unsigned int rr_nr_running;
830 struct {
831 int curr; /* highest queued rt task prio */
832 int next; /* next highest */
833 } highest_prio;
834 bool overloaded;
835 struct plist_head pushable_tasks;
836
837 int rt_queued;
838
839#ifdef CONFIG_RT_GROUP_SCHED
840 int rt_throttled;
841 u64 rt_time; /* consumed RT time, goes up in update_curr_rt */
842 u64 rt_runtime; /* allotted RT time, "slice" from rt_bandwidth, RT sharing/balancing */
843 /* Nests inside the rq lock: */
844 raw_spinlock_t rt_runtime_lock;
845
846 unsigned int rt_nr_boosted;
847
848 struct rq *rq; /* this is always top-level rq, cache? */
849#endif
850#ifdef CONFIG_CGROUP_SCHED
851 struct task_group *tg; /* this tg has "this" rt_rq on given CPU for runnable entities */
852#endif
853};
854
855static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
856{
857 return rt_rq->rt_queued && rt_rq->rt_nr_running;
858}
859
860/* Deadline class' related fields in a runqueue */
861struct dl_rq {
862 /* runqueue is an rbtree, ordered by deadline */
863 struct rb_root_cached root;
864
865 unsigned int dl_nr_running;
866
867 /*
868 * Deadline values of the currently executing and the
869 * earliest ready task on this rq. Caching these facilitates
870 * the decision whether or not a ready but not running task
871 * should migrate somewhere else.
872 */
873 struct {
874 u64 curr;
875 u64 next;
876 } earliest_dl;
877
878 bool overloaded;
879
880 /*
881 * Tasks on this rq that can be pushed away. They are kept in
882 * an rb-tree, ordered by tasks' deadlines, with caching
883 * of the leftmost (earliest deadline) element.
884 */
885 struct rb_root_cached pushable_dl_tasks_root;
886
887 /*
888 * "Active utilization" for this runqueue: increased when a
889 * task wakes up (becomes TASK_RUNNING) and decreased when a
890 * task blocks
891 */
892 u64 running_bw;
893
894 /*
895 * Utilization of the tasks "assigned" to this runqueue (including
896 * the tasks that are in runqueue and the tasks that executed on this
897 * CPU and blocked). Increased when a task moves to this runqueue, and
898 * decreased when the task moves away (migrates, changes scheduling
899 * policy, or terminates).
900 * This is needed to compute the "inactive utilization" for the
901 * runqueue (inactive utilization = this_bw - running_bw).
902 */
903 u64 this_bw;
904 u64 extra_bw;
905
906 /*
907 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
908 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
909 */
910 u64 max_bw;
911
912 /*
913 * Inverse of the fraction of CPU utilization that can be reclaimed
914 * by the GRUB algorithm.
915 */
916 u64 bw_ratio;
917};
918
919#ifdef CONFIG_FAIR_GROUP_SCHED
920
921/* An entity is a task if it doesn't "own" a runqueue */
922#define entity_is_task(se) (!se->my_q)
923
924static inline void se_update_runnable(struct sched_entity *se)
925{
926 if (!entity_is_task(se))
927 se->runnable_weight = se->my_q->h_nr_runnable;
928}
929
930static inline long se_runnable(struct sched_entity *se)
931{
932 if (se->sched_delayed)
933 return false;
934
935 if (entity_is_task(se))
936 return !!se->on_rq;
937 else
938 return se->runnable_weight;
939}
940
941#else /* !CONFIG_FAIR_GROUP_SCHED: */
942
943#define entity_is_task(se) 1
944
945static inline void se_update_runnable(struct sched_entity *se) { }
946
947static inline long se_runnable(struct sched_entity *se)
948{
949 if (se->sched_delayed)
950 return false;
951
952 return !!se->on_rq;
953}
954
955#endif /* !CONFIG_FAIR_GROUP_SCHED */
956
957/*
958 * XXX we want to get rid of these helpers and use the full load resolution.
959 */
960static inline long se_weight(struct sched_entity *se)
961{
962 return scale_load_down(se->load.weight);
963}
964
965
966static inline bool sched_asym_prefer(int a, int b)
967{
968 return arch_asym_cpu_priority(cpu: a) > arch_asym_cpu_priority(cpu: b);
969}
970
971struct perf_domain {
972 struct em_perf_domain *em_pd;
973 struct perf_domain *next;
974 struct rcu_head rcu;
975};
976
977/*
978 * We add the notion of a root-domain which will be used to define per-domain
979 * variables. Each exclusive cpuset essentially defines an island domain by
980 * fully partitioning the member CPUs from any other cpuset. Whenever a new
981 * exclusive cpuset is created, we also create and attach a new root-domain
982 * object.
983 *
984 */
985struct root_domain {
986 atomic_t refcount;
987 atomic_t rto_count;
988 struct rcu_head rcu;
989 cpumask_var_t span;
990 cpumask_var_t online;
991
992 /*
993 * Indicate pullable load on at least one CPU, e.g:
994 * - More than one runnable task
995 * - Running task is misfit
996 */
997 bool overloaded;
998
999 /* Indicate one or more CPUs over-utilized (tipping point) */
1000 bool overutilized;
1001
1002 /*
1003 * The bit corresponding to a CPU gets set here if such CPU has more
1004 * than one runnable -deadline task (as it is below for RT tasks).
1005 */
1006 cpumask_var_t dlo_mask;
1007 atomic_t dlo_count;
1008 struct dl_bw dl_bw;
1009 struct cpudl cpudl;
1010
1011 /*
1012 * Indicate whether a root_domain's dl_bw has been checked or
1013 * updated. It's monotonously increasing value.
1014 *
1015 * Also, some corner cases, like 'wrap around' is dangerous, but given
1016 * that u64 is 'big enough'. So that shouldn't be a concern.
1017 */
1018 u64 visit_cookie;
1019
1020#ifdef HAVE_RT_PUSH_IPI
1021 /*
1022 * For IPI pull requests, loop across the rto_mask.
1023 */
1024 struct irq_work rto_push_work;
1025 raw_spinlock_t rto_lock;
1026 /* These are only updated and read within rto_lock */
1027 int rto_loop;
1028 int rto_cpu;
1029 /* These atomics are updated outside of a lock */
1030 atomic_t rto_loop_next;
1031 atomic_t rto_loop_start;
1032#endif /* HAVE_RT_PUSH_IPI */
1033 /*
1034 * The "RT overload" flag: it gets set if a CPU has more than
1035 * one runnable RT task.
1036 */
1037 cpumask_var_t rto_mask;
1038 struct cpupri cpupri;
1039
1040 /*
1041 * NULL-terminated list of performance domains intersecting with the
1042 * CPUs of the rd. Protected by RCU.
1043 */
1044 struct perf_domain __rcu *pd;
1045};
1046
1047extern void init_defrootdomain(void);
1048extern int sched_init_domains(const struct cpumask *cpu_map);
1049extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1050extern void sched_get_rd(struct root_domain *rd);
1051extern void sched_put_rd(struct root_domain *rd);
1052
1053static inline int get_rd_overloaded(struct root_domain *rd)
1054{
1055 return READ_ONCE(rd->overloaded);
1056}
1057
1058static inline void set_rd_overloaded(struct root_domain *rd, int status)
1059{
1060 if (get_rd_overloaded(rd) != status)
1061 WRITE_ONCE(rd->overloaded, status);
1062}
1063
1064#ifdef HAVE_RT_PUSH_IPI
1065extern void rto_push_irq_work_func(struct irq_work *work);
1066#endif
1067
1068#ifdef CONFIG_UCLAMP_TASK
1069/*
1070 * struct uclamp_bucket - Utilization clamp bucket
1071 * @value: utilization clamp value for tasks on this clamp bucket
1072 * @tasks: number of RUNNABLE tasks on this clamp bucket
1073 *
1074 * Keep track of how many tasks are RUNNABLE for a given utilization
1075 * clamp value.
1076 */
1077struct uclamp_bucket {
1078 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1079 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1080};
1081
1082/*
1083 * struct uclamp_rq - rq's utilization clamp
1084 * @value: currently active clamp values for a rq
1085 * @bucket: utilization clamp buckets affecting a rq
1086 *
1087 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1088 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1089 * (or actually running) with that value.
1090 *
1091 * There are up to UCLAMP_CNT possible different clamp values, currently there
1092 * are only two: minimum utilization and maximum utilization.
1093 *
1094 * All utilization clamping values are MAX aggregated, since:
1095 * - for util_min: we want to run the CPU at least at the max of the minimum
1096 * utilization required by its currently RUNNABLE tasks.
1097 * - for util_max: we want to allow the CPU to run up to the max of the
1098 * maximum utilization allowed by its currently RUNNABLE tasks.
1099 *
1100 * Since on each system we expect only a limited number of different
1101 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1102 * the metrics required to compute all the per-rq utilization clamp values.
1103 */
1104struct uclamp_rq {
1105 unsigned int value;
1106 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1107};
1108
1109DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1110#endif /* CONFIG_UCLAMP_TASK */
1111
1112/*
1113 * This is the main, per-CPU runqueue data structure.
1114 *
1115 * Locking rule: those places that want to lock multiple runqueues
1116 * (such as the load balancing or the thread migration code), lock
1117 * acquire operations must be ordered by ascending &runqueue.
1118 */
1119struct rq {
1120 /* runqueue lock: */
1121 raw_spinlock_t __lock;
1122
1123 /* Per class runqueue modification mask; bits in class order. */
1124 unsigned int queue_mask;
1125 unsigned int nr_running;
1126#ifdef CONFIG_NUMA_BALANCING
1127 unsigned int nr_numa_running;
1128 unsigned int nr_preferred_running;
1129 unsigned int numa_migrate_on;
1130#endif
1131#ifdef CONFIG_NO_HZ_COMMON
1132 unsigned long last_blocked_load_update_tick;
1133 unsigned int has_blocked_load;
1134 call_single_data_t nohz_csd;
1135 unsigned int nohz_tick_stopped;
1136 atomic_t nohz_flags;
1137#endif /* CONFIG_NO_HZ_COMMON */
1138
1139 unsigned int ttwu_pending;
1140 u64 nr_switches;
1141
1142#ifdef CONFIG_UCLAMP_TASK
1143 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1144 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1145 unsigned int uclamp_flags;
1146#define UCLAMP_FLAG_IDLE 0x01
1147#endif
1148
1149 struct cfs_rq cfs;
1150 struct rt_rq rt;
1151 struct dl_rq dl;
1152#ifdef CONFIG_SCHED_CLASS_EXT
1153 struct scx_rq scx;
1154#endif
1155
1156 struct sched_dl_entity fair_server;
1157
1158#ifdef CONFIG_FAIR_GROUP_SCHED
1159 /* list of leaf cfs_rq on this CPU: */
1160 struct list_head leaf_cfs_rq_list;
1161 struct list_head *tmp_alone_branch;
1162#endif /* CONFIG_FAIR_GROUP_SCHED */
1163
1164 /*
1165 * This is part of a global counter where only the total sum
1166 * over all CPUs matters. A task can increase this counter on
1167 * one CPU and if it got migrated afterwards it may decrease
1168 * it on another CPU. Always updated under the runqueue lock:
1169 */
1170 unsigned long nr_uninterruptible;
1171
1172#ifdef CONFIG_SCHED_PROXY_EXEC
1173 struct task_struct __rcu *donor; /* Scheduling context */
1174 struct task_struct __rcu *curr; /* Execution context */
1175#else
1176 union {
1177 struct task_struct __rcu *donor; /* Scheduler context */
1178 struct task_struct __rcu *curr; /* Execution context */
1179 };
1180#endif
1181 struct sched_dl_entity *dl_server;
1182 struct task_struct *idle;
1183 struct task_struct *stop;
1184 unsigned long next_balance;
1185 struct mm_struct *prev_mm;
1186
1187 unsigned int clock_update_flags;
1188 u64 clock;
1189 /* Ensure that all clocks are in the same cache line */
1190 u64 clock_task ____cacheline_aligned;
1191 u64 clock_pelt;
1192 unsigned long lost_idle_time;
1193 u64 clock_pelt_idle;
1194 u64 clock_idle;
1195#ifndef CONFIG_64BIT
1196 u64 clock_pelt_idle_copy;
1197 u64 clock_idle_copy;
1198#endif
1199
1200 atomic_t nr_iowait;
1201
1202 u64 last_seen_need_resched_ns;
1203 int ticks_without_resched;
1204
1205#ifdef CONFIG_MEMBARRIER
1206 int membarrier_state;
1207#endif
1208
1209 struct root_domain *rd;
1210 struct sched_domain __rcu *sd;
1211
1212 unsigned long cpu_capacity;
1213
1214 struct balance_callback *balance_callback;
1215
1216 unsigned char nohz_idle_balance;
1217 unsigned char idle_balance;
1218
1219 unsigned long misfit_task_load;
1220
1221 /* For active balancing */
1222 int active_balance;
1223 int push_cpu;
1224 struct cpu_stop_work active_balance_work;
1225
1226 /* CPU of this runqueue: */
1227 int cpu;
1228 int online;
1229
1230 struct list_head cfs_tasks;
1231
1232 struct sched_avg avg_rt;
1233 struct sched_avg avg_dl;
1234#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1235 struct sched_avg avg_irq;
1236#endif
1237#ifdef CONFIG_SCHED_HW_PRESSURE
1238 struct sched_avg avg_hw;
1239#endif
1240 u64 idle_stamp;
1241 u64 avg_idle;
1242
1243 /* This is used to determine avg_idle's max value */
1244 u64 max_idle_balance_cost;
1245
1246#ifdef CONFIG_HOTPLUG_CPU
1247 struct rcuwait hotplug_wait;
1248#endif
1249
1250#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1251 u64 prev_irq_time;
1252 u64 psi_irq_time;
1253#endif
1254#ifdef CONFIG_PARAVIRT
1255 u64 prev_steal_time;
1256#endif
1257#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1258 u64 prev_steal_time_rq;
1259#endif
1260
1261 /* calc_load related fields */
1262 unsigned long calc_load_update;
1263 long calc_load_active;
1264
1265#ifdef CONFIG_SCHED_HRTICK
1266 call_single_data_t hrtick_csd;
1267 struct hrtimer hrtick_timer;
1268 ktime_t hrtick_time;
1269#endif
1270
1271#ifdef CONFIG_SCHEDSTATS
1272 /* latency stats */
1273 struct sched_info rq_sched_info;
1274 unsigned long long rq_cpu_time;
1275
1276 /* sys_sched_yield() stats */
1277 unsigned int yld_count;
1278
1279 /* schedule() stats */
1280 unsigned int sched_count;
1281 unsigned int sched_goidle;
1282
1283 /* try_to_wake_up() stats */
1284 unsigned int ttwu_count;
1285 unsigned int ttwu_local;
1286#endif
1287
1288#ifdef CONFIG_CPU_IDLE
1289 /* Must be inspected within a RCU lock section */
1290 struct cpuidle_state *idle_state;
1291#endif
1292
1293 unsigned int nr_pinned;
1294 unsigned int push_busy;
1295 struct cpu_stop_work push_work;
1296
1297#ifdef CONFIG_SCHED_CORE
1298 /* per rq */
1299 struct rq *core;
1300 struct task_struct *core_pick;
1301 struct sched_dl_entity *core_dl_server;
1302 unsigned int core_enabled;
1303 unsigned int core_sched_seq;
1304 struct rb_root core_tree;
1305
1306 /* shared state -- careful with sched_core_cpu_deactivate() */
1307 unsigned int core_task_seq;
1308 unsigned int core_pick_seq;
1309 unsigned long core_cookie;
1310 unsigned int core_forceidle_count;
1311 unsigned int core_forceidle_seq;
1312 unsigned int core_forceidle_occupation;
1313 u64 core_forceidle_start;
1314#endif /* CONFIG_SCHED_CORE */
1315
1316 /* Scratch cpumask to be temporarily used under rq_lock */
1317 cpumask_var_t scratch_mask;
1318
1319#ifdef CONFIG_CFS_BANDWIDTH
1320 call_single_data_t cfsb_csd;
1321 struct list_head cfsb_csd_list;
1322#endif
1323};
1324
1325#ifdef CONFIG_FAIR_GROUP_SCHED
1326
1327/* CPU runqueue to which this cfs_rq is attached */
1328static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1329{
1330 return cfs_rq->rq;
1331}
1332
1333#else /* !CONFIG_FAIR_GROUP_SCHED: */
1334
1335static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1336{
1337 return container_of(cfs_rq, struct rq, cfs);
1338}
1339#endif /* !CONFIG_FAIR_GROUP_SCHED */
1340
1341static inline int cpu_of(struct rq *rq)
1342{
1343 return rq->cpu;
1344}
1345
1346#define MDF_PUSH 0x01
1347
1348static inline bool is_migration_disabled(struct task_struct *p)
1349{
1350 return p->migration_disabled;
1351}
1352
1353DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1354DECLARE_PER_CPU(struct rnd_state, sched_rnd_state);
1355
1356static inline u32 sched_rng(void)
1357{
1358 return prandom_u32_state(this_cpu_ptr(&sched_rnd_state));
1359}
1360
1361#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1362#define this_rq() this_cpu_ptr(&runqueues)
1363#define task_rq(p) cpu_rq(task_cpu(p))
1364#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1365#define raw_rq() raw_cpu_ptr(&runqueues)
1366
1367static inline bool idle_rq(struct rq *rq)
1368{
1369 return rq->curr == rq->idle && !rq->nr_running && !rq->ttwu_pending;
1370}
1371
1372/**
1373 * available_idle_cpu - is a given CPU idle for enqueuing work.
1374 * @cpu: the CPU in question.
1375 *
1376 * Return: 1 if the CPU is currently idle. 0 otherwise.
1377 */
1378static inline bool available_idle_cpu(int cpu)
1379{
1380 if (!idle_rq(cpu_rq(cpu)))
1381 return 0;
1382
1383 if (vcpu_is_preempted(cpu))
1384 return 0;
1385
1386 return 1;
1387}
1388
1389#ifdef CONFIG_SCHED_PROXY_EXEC
1390static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1391{
1392 rcu_assign_pointer(rq->donor, t);
1393}
1394#else
1395static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1396{
1397 /* Do nothing */
1398}
1399#endif
1400
1401#ifdef CONFIG_SCHED_CORE
1402static inline struct cpumask *sched_group_span(struct sched_group *sg);
1403
1404DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1405
1406static inline bool sched_core_enabled(struct rq *rq)
1407{
1408 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1409}
1410
1411static inline bool sched_core_disabled(void)
1412{
1413 return !static_branch_unlikely(&__sched_core_enabled);
1414}
1415
1416/*
1417 * Be careful with this function; not for general use. The return value isn't
1418 * stable unless you actually hold a relevant rq->__lock.
1419 */
1420static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1421{
1422 if (sched_core_enabled(rq))
1423 return &rq->core->__lock;
1424
1425 return &rq->__lock;
1426}
1427
1428static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1429{
1430 if (rq->core_enabled)
1431 return &rq->core->__lock;
1432
1433 return &rq->__lock;
1434}
1435
1436extern bool
1437cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1438
1439extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1440
1441/*
1442 * Helpers to check if the CPU's core cookie matches with the task's cookie
1443 * when core scheduling is enabled.
1444 * A special case is that the task's cookie always matches with CPU's core
1445 * cookie if the CPU is in an idle core.
1446 */
1447static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1448{
1449 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1450 if (!sched_core_enabled(rq))
1451 return true;
1452
1453 return rq->core->core_cookie == p->core_cookie;
1454}
1455
1456static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1457{
1458 bool idle_core = true;
1459 int cpu;
1460
1461 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1462 if (!sched_core_enabled(rq))
1463 return true;
1464
1465 if (rq->core->core_cookie == p->core_cookie)
1466 return true;
1467
1468 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1469 if (!available_idle_cpu(cpu)) {
1470 idle_core = false;
1471 break;
1472 }
1473 }
1474
1475 /*
1476 * A CPU in an idle core is always the best choice for tasks with
1477 * cookies.
1478 */
1479 return idle_core;
1480}
1481
1482static inline bool sched_group_cookie_match(struct rq *rq,
1483 struct task_struct *p,
1484 struct sched_group *group)
1485{
1486 int cpu;
1487
1488 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1489 if (!sched_core_enabled(rq))
1490 return true;
1491
1492 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1493 if (sched_core_cookie_match(cpu_rq(cpu), p))
1494 return true;
1495 }
1496 return false;
1497}
1498
1499static inline bool sched_core_enqueued(struct task_struct *p)
1500{
1501 return !RB_EMPTY_NODE(&p->core_node);
1502}
1503
1504extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1505extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1506
1507extern void sched_core_get(void);
1508extern void sched_core_put(void);
1509
1510#else /* !CONFIG_SCHED_CORE: */
1511
1512static inline bool sched_core_enabled(struct rq *rq)
1513{
1514 return false;
1515}
1516
1517static inline bool sched_core_disabled(void)
1518{
1519 return true;
1520}
1521
1522static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1523{
1524 return &rq->__lock;
1525}
1526
1527static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1528{
1529 return &rq->__lock;
1530}
1531
1532static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1533{
1534 return true;
1535}
1536
1537static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1538{
1539 return true;
1540}
1541
1542static inline bool sched_group_cookie_match(struct rq *rq,
1543 struct task_struct *p,
1544 struct sched_group *group)
1545{
1546 return true;
1547}
1548
1549#endif /* !CONFIG_SCHED_CORE */
1550
1551#ifdef CONFIG_RT_GROUP_SCHED
1552# ifdef CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED
1553DECLARE_STATIC_KEY_FALSE(rt_group_sched);
1554static inline bool rt_group_sched_enabled(void)
1555{
1556 return static_branch_unlikely(&rt_group_sched);
1557}
1558# else /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED: */
1559DECLARE_STATIC_KEY_TRUE(rt_group_sched);
1560static inline bool rt_group_sched_enabled(void)
1561{
1562 return static_branch_likely(&rt_group_sched);
1563}
1564# endif /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED */
1565#else /* !CONFIG_RT_GROUP_SCHED: */
1566# define rt_group_sched_enabled() false
1567#endif /* !CONFIG_RT_GROUP_SCHED */
1568
1569static inline void lockdep_assert_rq_held(struct rq *rq)
1570{
1571 lockdep_assert_held(__rq_lockp(rq));
1572}
1573
1574extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1575extern bool raw_spin_rq_trylock(struct rq *rq);
1576extern void raw_spin_rq_unlock(struct rq *rq);
1577
1578static inline void raw_spin_rq_lock(struct rq *rq)
1579{
1580 raw_spin_rq_lock_nested(rq, subclass: 0);
1581}
1582
1583static inline void raw_spin_rq_lock_irq(struct rq *rq)
1584{
1585 local_irq_disable();
1586 raw_spin_rq_lock(rq);
1587}
1588
1589static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1590{
1591 raw_spin_rq_unlock(rq);
1592 local_irq_enable();
1593}
1594
1595static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1596{
1597 unsigned long flags;
1598
1599 local_irq_save(flags);
1600 raw_spin_rq_lock(rq);
1601
1602 return flags;
1603}
1604
1605static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1606{
1607 raw_spin_rq_unlock(rq);
1608 local_irq_restore(flags);
1609}
1610
1611#define raw_spin_rq_lock_irqsave(rq, flags) \
1612do { \
1613 flags = _raw_spin_rq_lock_irqsave(rq); \
1614} while (0)
1615
1616#ifdef CONFIG_SCHED_SMT
1617extern void __update_idle_core(struct rq *rq);
1618
1619static inline void update_idle_core(struct rq *rq)
1620{
1621 if (static_branch_unlikely(&sched_smt_present))
1622 __update_idle_core(rq);
1623}
1624
1625#else /* !CONFIG_SCHED_SMT: */
1626static inline void update_idle_core(struct rq *rq) { }
1627#endif /* !CONFIG_SCHED_SMT */
1628
1629#ifdef CONFIG_FAIR_GROUP_SCHED
1630
1631static inline struct task_struct *task_of(struct sched_entity *se)
1632{
1633 WARN_ON_ONCE(!entity_is_task(se));
1634 return container_of(se, struct task_struct, se);
1635}
1636
1637static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1638{
1639 return p->se.cfs_rq;
1640}
1641
1642/* runqueue on which this entity is (to be) queued */
1643static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1644{
1645 return se->cfs_rq;
1646}
1647
1648/* runqueue "owned" by this group */
1649static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1650{
1651 return grp->my_q;
1652}
1653
1654#else /* !CONFIG_FAIR_GROUP_SCHED: */
1655
1656#define task_of(_se) container_of(_se, struct task_struct, se)
1657
1658static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1659{
1660 return &task_rq(p)->cfs;
1661}
1662
1663static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1664{
1665 const struct task_struct *p = task_of(se);
1666 struct rq *rq = task_rq(p);
1667
1668 return &rq->cfs;
1669}
1670
1671/* runqueue "owned" by this group */
1672static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1673{
1674 return NULL;
1675}
1676
1677#endif /* !CONFIG_FAIR_GROUP_SCHED */
1678
1679extern void update_rq_clock(struct rq *rq);
1680
1681/*
1682 * rq::clock_update_flags bits
1683 *
1684 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1685 * call to __schedule(). This is an optimisation to avoid
1686 * neighbouring rq clock updates.
1687 *
1688 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1689 * in effect and calls to update_rq_clock() are being ignored.
1690 *
1691 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1692 * made to update_rq_clock() since the last time rq::lock was pinned.
1693 *
1694 * If inside of __schedule(), clock_update_flags will have been
1695 * shifted left (a left shift is a cheap operation for the fast path
1696 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1697 *
1698 * if (rq-clock_update_flags >= RQCF_UPDATED)
1699 *
1700 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1701 * one position though, because the next rq_unpin_lock() will shift it
1702 * back.
1703 */
1704#define RQCF_REQ_SKIP 0x01
1705#define RQCF_ACT_SKIP 0x02
1706#define RQCF_UPDATED 0x04
1707
1708static inline void assert_clock_updated(struct rq *rq)
1709{
1710 /*
1711 * The only reason for not seeing a clock update since the
1712 * last rq_pin_lock() is if we're currently skipping updates.
1713 */
1714 WARN_ON_ONCE(rq->clock_update_flags < RQCF_ACT_SKIP);
1715}
1716
1717static inline u64 rq_clock(struct rq *rq)
1718{
1719 lockdep_assert_rq_held(rq);
1720 assert_clock_updated(rq);
1721
1722 return rq->clock;
1723}
1724
1725static inline u64 rq_clock_task(struct rq *rq)
1726{
1727 lockdep_assert_rq_held(rq);
1728 assert_clock_updated(rq);
1729
1730 return rq->clock_task;
1731}
1732
1733static inline void rq_clock_skip_update(struct rq *rq)
1734{
1735 lockdep_assert_rq_held(rq);
1736 rq->clock_update_flags |= RQCF_REQ_SKIP;
1737}
1738
1739/*
1740 * See rt task throttling, which is the only time a skip
1741 * request is canceled.
1742 */
1743static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1744{
1745 lockdep_assert_rq_held(rq);
1746 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1747}
1748
1749/*
1750 * During cpu offlining and rq wide unthrottling, we can trigger
1751 * an update_rq_clock() for several cfs and rt runqueues (Typically
1752 * when using list_for_each_entry_*)
1753 * rq_clock_start_loop_update() can be called after updating the clock
1754 * once and before iterating over the list to prevent multiple update.
1755 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1756 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1757 */
1758static inline void rq_clock_start_loop_update(struct rq *rq)
1759{
1760 lockdep_assert_rq_held(rq);
1761 WARN_ON_ONCE(rq->clock_update_flags & RQCF_ACT_SKIP);
1762 rq->clock_update_flags |= RQCF_ACT_SKIP;
1763}
1764
1765static inline void rq_clock_stop_loop_update(struct rq *rq)
1766{
1767 lockdep_assert_rq_held(rq);
1768 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1769}
1770
1771struct rq_flags {
1772 unsigned long flags;
1773 struct pin_cookie cookie;
1774 /*
1775 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1776 * current pin context is stashed here in case it needs to be
1777 * restored in rq_repin_lock().
1778 */
1779 unsigned int clock_update_flags;
1780};
1781
1782extern struct balance_callback balance_push_callback;
1783
1784#ifdef CONFIG_SCHED_CLASS_EXT
1785extern const struct sched_class ext_sched_class;
1786
1787DECLARE_STATIC_KEY_FALSE(__scx_enabled); /* SCX BPF scheduler loaded */
1788DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
1789
1790#define scx_enabled() static_branch_unlikely(&__scx_enabled)
1791#define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
1792
1793static inline void scx_rq_clock_update(struct rq *rq, u64 clock)
1794{
1795 if (!scx_enabled())
1796 return;
1797 WRITE_ONCE(rq->scx.clock, clock);
1798 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID);
1799}
1800
1801static inline void scx_rq_clock_invalidate(struct rq *rq)
1802{
1803 if (!scx_enabled())
1804 return;
1805 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID);
1806}
1807
1808#else /* !CONFIG_SCHED_CLASS_EXT: */
1809#define scx_enabled() false
1810#define scx_switched_all() false
1811
1812static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {}
1813static inline void scx_rq_clock_invalidate(struct rq *rq) {}
1814#endif /* !CONFIG_SCHED_CLASS_EXT */
1815
1816/*
1817 * Lockdep annotation that avoids accidental unlocks; it's like a
1818 * sticky/continuous lockdep_assert_held().
1819 *
1820 * This avoids code that has access to 'struct rq *rq' (basically everything in
1821 * the scheduler) from accidentally unlocking the rq if they do not also have a
1822 * copy of the (on-stack) 'struct rq_flags rf'.
1823 *
1824 * Also see Documentation/locking/lockdep-design.rst.
1825 */
1826static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1827{
1828 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1829
1830 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1831 rf->clock_update_flags = 0;
1832 WARN_ON_ONCE(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1833}
1834
1835static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1836{
1837 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1838 rf->clock_update_flags = RQCF_UPDATED;
1839
1840 scx_rq_clock_invalidate(rq);
1841 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1842}
1843
1844static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1845{
1846 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1847
1848 /*
1849 * Restore the value we stashed in @rf for this pin context.
1850 */
1851 rq->clock_update_flags |= rf->clock_update_flags;
1852}
1853
1854extern
1855struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1856 __acquires(rq->lock);
1857
1858extern
1859struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1860 __acquires(p->pi_lock)
1861 __acquires(rq->lock);
1862
1863static inline void
1864__task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1865 __releases(rq->lock)
1866{
1867 rq_unpin_lock(rq, rf);
1868 raw_spin_rq_unlock(rq);
1869}
1870
1871static inline void
1872task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1873 __releases(rq->lock)
1874 __releases(p->pi_lock)
1875{
1876 __task_rq_unlock(rq, p, rf);
1877 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1878}
1879
1880DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1881 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1882 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1883 struct rq *rq; struct rq_flags rf)
1884
1885DEFINE_LOCK_GUARD_1(__task_rq_lock, struct task_struct,
1886 _T->rq = __task_rq_lock(_T->lock, &_T->rf),
1887 __task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1888 struct rq *rq; struct rq_flags rf)
1889
1890static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1891 __acquires(rq->lock)
1892{
1893 raw_spin_rq_lock_irqsave(rq, rf->flags);
1894 rq_pin_lock(rq, rf);
1895}
1896
1897static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1898 __acquires(rq->lock)
1899{
1900 raw_spin_rq_lock_irq(rq);
1901 rq_pin_lock(rq, rf);
1902}
1903
1904static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1905 __acquires(rq->lock)
1906{
1907 raw_spin_rq_lock(rq);
1908 rq_pin_lock(rq, rf);
1909}
1910
1911static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1912 __releases(rq->lock)
1913{
1914 rq_unpin_lock(rq, rf);
1915 raw_spin_rq_unlock_irqrestore(rq, flags: rf->flags);
1916}
1917
1918static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1919 __releases(rq->lock)
1920{
1921 rq_unpin_lock(rq, rf);
1922 raw_spin_rq_unlock_irq(rq);
1923}
1924
1925static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1926 __releases(rq->lock)
1927{
1928 rq_unpin_lock(rq, rf);
1929 raw_spin_rq_unlock(rq);
1930}
1931
1932DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1933 rq_lock(_T->lock, &_T->rf),
1934 rq_unlock(_T->lock, &_T->rf),
1935 struct rq_flags rf)
1936
1937DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1938 rq_lock_irq(_T->lock, &_T->rf),
1939 rq_unlock_irq(_T->lock, &_T->rf),
1940 struct rq_flags rf)
1941
1942DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1943 rq_lock_irqsave(_T->lock, &_T->rf),
1944 rq_unlock_irqrestore(_T->lock, &_T->rf),
1945 struct rq_flags rf)
1946
1947static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1948 __acquires(rq->lock)
1949{
1950 struct rq *rq;
1951
1952 local_irq_disable();
1953 rq = this_rq();
1954 rq_lock(rq, rf);
1955
1956 return rq;
1957}
1958
1959#ifdef CONFIG_NUMA
1960
1961enum numa_topology_type {
1962 NUMA_DIRECT,
1963 NUMA_GLUELESS_MESH,
1964 NUMA_BACKPLANE,
1965};
1966
1967extern enum numa_topology_type sched_numa_topology_type;
1968extern int sched_max_numa_distance;
1969extern bool find_numa_distance(int distance);
1970extern void sched_init_numa(int offline_node);
1971extern void sched_update_numa(int cpu, bool online);
1972extern void sched_domains_numa_masks_set(unsigned int cpu);
1973extern void sched_domains_numa_masks_clear(unsigned int cpu);
1974extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1975
1976#else /* !CONFIG_NUMA: */
1977
1978static inline void sched_init_numa(int offline_node) { }
1979static inline void sched_update_numa(int cpu, bool online) { }
1980static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1981static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1982
1983static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1984{
1985 return nr_cpu_ids;
1986}
1987
1988#endif /* !CONFIG_NUMA */
1989
1990#ifdef CONFIG_NUMA_BALANCING
1991
1992/* The regions in numa_faults array from task_struct */
1993enum numa_faults_stats {
1994 NUMA_MEM = 0,
1995 NUMA_CPU,
1996 NUMA_MEMBUF,
1997 NUMA_CPUBUF
1998};
1999
2000extern void sched_setnuma(struct task_struct *p, int node);
2001extern int migrate_task_to(struct task_struct *p, int cpu);
2002extern int migrate_swap(struct task_struct *p, struct task_struct *t,
2003 int cpu, int scpu);
2004extern void init_numa_balancing(u64 clone_flags, struct task_struct *p);
2005
2006#else /* !CONFIG_NUMA_BALANCING: */
2007
2008static inline void
2009init_numa_balancing(u64 clone_flags, struct task_struct *p)
2010{
2011}
2012
2013#endif /* !CONFIG_NUMA_BALANCING */
2014
2015static inline void
2016queue_balance_callback(struct rq *rq,
2017 struct balance_callback *head,
2018 void (*func)(struct rq *rq))
2019{
2020 lockdep_assert_rq_held(rq);
2021
2022 /*
2023 * Don't (re)queue an already queued item; nor queue anything when
2024 * balance_push() is active, see the comment with
2025 * balance_push_callback.
2026 */
2027 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
2028 return;
2029
2030 head->func = func;
2031 head->next = rq->balance_callback;
2032 rq->balance_callback = head;
2033}
2034
2035#define rcu_dereference_check_sched_domain(p) \
2036 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
2037
2038/*
2039 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
2040 * See destroy_sched_domains: call_rcu for details.
2041 *
2042 * The domain tree of any CPU may only be accessed from within
2043 * preempt-disabled sections.
2044 */
2045#define for_each_domain(cpu, __sd) \
2046 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
2047 __sd; __sd = __sd->parent)
2048
2049/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
2050#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
2051static const unsigned int SD_SHARED_CHILD_MASK =
2052#include <linux/sched/sd_flags.h>
20530;
2054#undef SD_FLAG
2055
2056/**
2057 * highest_flag_domain - Return highest sched_domain containing flag.
2058 * @cpu: The CPU whose highest level of sched domain is to
2059 * be returned.
2060 * @flag: The flag to check for the highest sched_domain
2061 * for the given CPU.
2062 *
2063 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
2064 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
2065 */
2066static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
2067{
2068 struct sched_domain *sd, *hsd = NULL;
2069
2070 for_each_domain(cpu, sd) {
2071 if (sd->flags & flag) {
2072 hsd = sd;
2073 continue;
2074 }
2075
2076 /*
2077 * Stop the search if @flag is known to be shared at lower
2078 * levels. It will not be found further up.
2079 */
2080 if (flag & SD_SHARED_CHILD_MASK)
2081 break;
2082 }
2083
2084 return hsd;
2085}
2086
2087static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2088{
2089 struct sched_domain *sd;
2090
2091 for_each_domain(cpu, sd) {
2092 if (sd->flags & flag)
2093 break;
2094 }
2095
2096 return sd;
2097}
2098
2099DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2100DECLARE_PER_CPU(int, sd_llc_size);
2101DECLARE_PER_CPU(int, sd_llc_id);
2102DECLARE_PER_CPU(int, sd_share_id);
2103DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2104DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2105DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2106DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2107
2108extern struct static_key_false sched_asym_cpucapacity;
2109extern struct static_key_false sched_cluster_active;
2110
2111static __always_inline bool sched_asym_cpucap_active(void)
2112{
2113 return static_branch_unlikely(&sched_asym_cpucapacity);
2114}
2115
2116struct sched_group_capacity {
2117 atomic_t ref;
2118 /*
2119 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2120 * for a single CPU.
2121 */
2122 unsigned long capacity;
2123 unsigned long min_capacity; /* Min per-CPU capacity in group */
2124 unsigned long max_capacity; /* Max per-CPU capacity in group */
2125 unsigned long next_update;
2126 int imbalance; /* XXX unrelated to capacity but shared group state */
2127
2128 int id;
2129
2130 unsigned long cpumask[]; /* Balance mask */
2131};
2132
2133struct sched_group {
2134 struct sched_group *next; /* Must be a circular list */
2135 atomic_t ref;
2136
2137 unsigned int group_weight;
2138 unsigned int cores;
2139 struct sched_group_capacity *sgc;
2140 int asym_prefer_cpu; /* CPU of highest priority in group */
2141 int flags;
2142
2143 /*
2144 * The CPUs this group covers.
2145 *
2146 * NOTE: this field is variable length. (Allocated dynamically
2147 * by attaching extra space to the end of the structure,
2148 * depending on how many CPUs the kernel has booted up with)
2149 */
2150 unsigned long cpumask[];
2151};
2152
2153static inline struct cpumask *sched_group_span(struct sched_group *sg)
2154{
2155 return to_cpumask(sg->cpumask);
2156}
2157
2158/*
2159 * See build_balance_mask().
2160 */
2161static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2162{
2163 return to_cpumask(sg->sgc->cpumask);
2164}
2165
2166extern int group_balance_cpu(struct sched_group *sg);
2167
2168extern void update_sched_domain_debugfs(void);
2169extern void dirty_sched_domain_sysctl(int cpu);
2170
2171extern int sched_update_scaling(void);
2172
2173static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2174{
2175 if (!p->user_cpus_ptr)
2176 return cpu_possible_mask; /* &init_task.cpus_mask */
2177 return p->user_cpus_ptr;
2178}
2179
2180#ifdef CONFIG_CGROUP_SCHED
2181
2182/*
2183 * Return the group to which this tasks belongs.
2184 *
2185 * We cannot use task_css() and friends because the cgroup subsystem
2186 * changes that value before the cgroup_subsys::attach() method is called,
2187 * therefore we cannot pin it and might observe the wrong value.
2188 *
2189 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2190 * core changes this before calling sched_move_task().
2191 *
2192 * Instead we use a 'copy' which is updated from sched_move_task() while
2193 * holding both task_struct::pi_lock and rq::lock.
2194 */
2195static inline struct task_group *task_group(struct task_struct *p)
2196{
2197 return p->sched_task_group;
2198}
2199
2200/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2201static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2202{
2203#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2204 struct task_group *tg = task_group(p);
2205#endif
2206
2207#ifdef CONFIG_FAIR_GROUP_SCHED
2208 set_task_rq_fair(se: &p->se, prev: p->se.cfs_rq, next: tg->cfs_rq[cpu]);
2209 p->se.cfs_rq = tg->cfs_rq[cpu];
2210 p->se.parent = tg->se[cpu];
2211 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2212#endif
2213
2214#ifdef CONFIG_RT_GROUP_SCHED
2215 /*
2216 * p->rt.rt_rq is NULL initially and it is easier to assign
2217 * root_task_group's rt_rq than switching in rt_rq_of_se()
2218 * Clobbers tg(!)
2219 */
2220 if (!rt_group_sched_enabled())
2221 tg = &root_task_group;
2222 p->rt.rt_rq = tg->rt_rq[cpu];
2223 p->rt.parent = tg->rt_se[cpu];
2224#endif /* CONFIG_RT_GROUP_SCHED */
2225}
2226
2227#else /* !CONFIG_CGROUP_SCHED: */
2228
2229static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2230
2231static inline struct task_group *task_group(struct task_struct *p)
2232{
2233 return NULL;
2234}
2235
2236#endif /* !CONFIG_CGROUP_SCHED */
2237
2238static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2239{
2240 set_task_rq(p, cpu);
2241#ifdef CONFIG_SMP
2242 /*
2243 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2244 * successfully executed on another CPU. We must ensure that updates of
2245 * per-task data have been completed by this moment.
2246 */
2247 smp_wmb();
2248 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2249 p->wake_cpu = cpu;
2250 rseq_sched_set_ids_changed(t: p);
2251#endif /* CONFIG_SMP */
2252}
2253
2254/*
2255 * Tunables:
2256 */
2257
2258#define SCHED_FEAT(name, enabled) \
2259 __SCHED_FEAT_##name ,
2260
2261enum {
2262#include "features.h"
2263 __SCHED_FEAT_NR,
2264};
2265
2266#undef SCHED_FEAT
2267
2268/*
2269 * To support run-time toggling of sched features, all the translation units
2270 * (but core.c) reference the sysctl_sched_features defined in core.c.
2271 */
2272extern __read_mostly unsigned int sysctl_sched_features;
2273
2274#ifdef CONFIG_JUMP_LABEL
2275
2276#define SCHED_FEAT(name, enabled) \
2277static __always_inline bool static_branch_##name(struct static_key *key) \
2278{ \
2279 return static_key_##enabled(key); \
2280}
2281
2282#include "features.h"
2283#undef SCHED_FEAT
2284
2285extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2286#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2287
2288#else /* !CONFIG_JUMP_LABEL: */
2289
2290#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2291
2292#endif /* !CONFIG_JUMP_LABEL */
2293
2294extern struct static_key_false sched_numa_balancing;
2295extern struct static_key_false sched_schedstats;
2296
2297static inline u64 global_rt_period(void)
2298{
2299 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2300}
2301
2302static inline u64 global_rt_runtime(void)
2303{
2304 if (sysctl_sched_rt_runtime < 0)
2305 return RUNTIME_INF;
2306
2307 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2308}
2309
2310/*
2311 * Is p the current execution context?
2312 */
2313static inline int task_current(struct rq *rq, struct task_struct *p)
2314{
2315 return rq->curr == p;
2316}
2317
2318/*
2319 * Is p the current scheduling context?
2320 *
2321 * Note that it might be the current execution context at the same time if
2322 * rq->curr == rq->donor == p.
2323 */
2324static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2325{
2326 return rq->donor == p;
2327}
2328
2329static inline bool task_is_blocked(struct task_struct *p)
2330{
2331 if (!sched_proxy_exec())
2332 return false;
2333
2334 return !!p->blocked_on;
2335}
2336
2337static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2338{
2339 return p->on_cpu;
2340}
2341
2342static inline int task_on_rq_queued(struct task_struct *p)
2343{
2344 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED;
2345}
2346
2347static inline int task_on_rq_migrating(struct task_struct *p)
2348{
2349 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2350}
2351
2352/* Wake flags. The first three directly map to some SD flag value */
2353#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2354#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2355#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2356
2357#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2358#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2359#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2360#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2361
2362static_assert(WF_EXEC == SD_BALANCE_EXEC);
2363static_assert(WF_FORK == SD_BALANCE_FORK);
2364static_assert(WF_TTWU == SD_BALANCE_WAKE);
2365
2366/*
2367 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2368 * of tasks with abnormal "nice" values across CPUs the contribution that
2369 * each task makes to its run queue's load is weighted according to its
2370 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2371 * scaled version of the new time slice allocation that they receive on time
2372 * slice expiry etc.
2373 */
2374
2375#define WEIGHT_IDLEPRIO 3
2376#define WMULT_IDLEPRIO 1431655765
2377
2378extern const int sched_prio_to_weight[40];
2379extern const u32 sched_prio_to_wmult[40];
2380
2381/*
2382 * {de,en}queue flags:
2383 *
2384 * SLEEP/WAKEUP - task is no-longer/just-became runnable
2385 *
2386 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2387 * are in a known state which allows modification. Such pairs
2388 * should preserve as much state as possible.
2389 *
2390 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2391 * in the runqueue. IOW the priority is allowed to change. Callers
2392 * must expect to deal with balance callbacks.
2393 *
2394 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2395 *
2396 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2397 *
2398 * DELAYED - de/re-queue a sched_delayed task
2399 *
2400 * CLASS - going to update p->sched_class; makes sched_change call the
2401 * various switch methods.
2402 *
2403 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2404 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2405 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2406 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2407 *
2408 * XXX SAVE/RESTORE in combination with CLASS doesn't really make sense, but
2409 * SCHED_DEADLINE seems to rely on this for now.
2410 */
2411
2412#define DEQUEUE_SLEEP 0x0001 /* Matches ENQUEUE_WAKEUP */
2413#define DEQUEUE_SAVE 0x0002 /* Matches ENQUEUE_RESTORE */
2414#define DEQUEUE_MOVE 0x0004 /* Matches ENQUEUE_MOVE */
2415#define DEQUEUE_NOCLOCK 0x0008 /* Matches ENQUEUE_NOCLOCK */
2416
2417#define DEQUEUE_MIGRATING 0x0010 /* Matches ENQUEUE_MIGRATING */
2418#define DEQUEUE_DELAYED 0x0020 /* Matches ENQUEUE_DELAYED */
2419#define DEQUEUE_CLASS 0x0040 /* Matches ENQUEUE_CLASS */
2420
2421#define DEQUEUE_SPECIAL 0x00010000
2422#define DEQUEUE_THROTTLE 0x00020000
2423
2424#define ENQUEUE_WAKEUP 0x0001
2425#define ENQUEUE_RESTORE 0x0002
2426#define ENQUEUE_MOVE 0x0004
2427#define ENQUEUE_NOCLOCK 0x0008
2428
2429#define ENQUEUE_MIGRATING 0x0010
2430#define ENQUEUE_DELAYED 0x0020
2431#define ENQUEUE_CLASS 0x0040
2432
2433#define ENQUEUE_HEAD 0x00010000
2434#define ENQUEUE_REPLENISH 0x00020000
2435#define ENQUEUE_MIGRATED 0x00040000
2436#define ENQUEUE_INITIAL 0x00080000
2437#define ENQUEUE_RQ_SELECTED 0x00100000
2438
2439#define RETRY_TASK ((void *)-1UL)
2440
2441struct affinity_context {
2442 const struct cpumask *new_mask;
2443 struct cpumask *user_mask;
2444 unsigned int flags;
2445};
2446
2447extern s64 update_curr_common(struct rq *rq);
2448
2449struct sched_class {
2450
2451#ifdef CONFIG_UCLAMP_TASK
2452 int uclamp_enabled;
2453#endif
2454 /*
2455 * idle: 0
2456 * ext: 1
2457 * fair: 2
2458 * rt: 4
2459 * dl: 8
2460 * stop: 16
2461 */
2462 unsigned int queue_mask;
2463
2464 /*
2465 * move_queued_task/activate_task/enqueue_task: rq->lock
2466 * ttwu_do_activate/activate_task/enqueue_task: rq->lock
2467 * wake_up_new_task/activate_task/enqueue_task: task_rq_lock
2468 * ttwu_runnable/enqueue_task: task_rq_lock
2469 * proxy_task_current: rq->lock
2470 * sched_change_end
2471 */
2472 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2473 /*
2474 * move_queued_task/deactivate_task/dequeue_task: rq->lock
2475 * __schedule/block_task/dequeue_task: rq->lock
2476 * proxy_task_current: rq->lock
2477 * wait_task_inactive: task_rq_lock
2478 * sched_change_begin
2479 */
2480 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2481
2482 /*
2483 * do_sched_yield: rq->lock
2484 */
2485 void (*yield_task) (struct rq *rq);
2486 /*
2487 * yield_to: rq->lock (double)
2488 */
2489 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2490
2491 /*
2492 * move_queued_task: rq->lock
2493 * __migrate_swap_task: rq->lock
2494 * ttwu_do_activate: rq->lock
2495 * ttwu_runnable: task_rq_lock
2496 * wake_up_new_task: task_rq_lock
2497 */
2498 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2499
2500 /*
2501 * schedule/pick_next_task/prev_balance: rq->lock
2502 */
2503 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2504
2505 /*
2506 * schedule/pick_next_task: rq->lock
2507 */
2508 struct task_struct *(*pick_task)(struct rq *rq, struct rq_flags *rf);
2509 /*
2510 * Optional! When implemented pick_next_task() should be equivalent to:
2511 *
2512 * next = pick_task();
2513 * if (next) {
2514 * put_prev_task(prev);
2515 * set_next_task_first(next);
2516 * }
2517 */
2518 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev,
2519 struct rq_flags *rf);
2520
2521 /*
2522 * sched_change:
2523 * __schedule: rq->lock
2524 */
2525 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2526 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2527
2528 /*
2529 * select_task_rq: p->pi_lock
2530 * sched_exec: p->pi_lock
2531 */
2532 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2533
2534 /*
2535 * set_task_cpu: p->pi_lock || rq->lock (ttwu like)
2536 */
2537 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2538
2539 /*
2540 * ttwu_do_activate: rq->lock
2541 * wake_up_new_task: task_rq_lock
2542 */
2543 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2544
2545 /*
2546 * do_set_cpus_allowed: task_rq_lock + sched_change
2547 */
2548 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2549
2550 /*
2551 * sched_set_rq_{on,off}line: rq->lock
2552 */
2553 void (*rq_online)(struct rq *rq);
2554 void (*rq_offline)(struct rq *rq);
2555
2556 /*
2557 * push_cpu_stop: p->pi_lock && rq->lock
2558 */
2559 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2560
2561 /*
2562 * hrtick: rq->lock
2563 * sched_tick: rq->lock
2564 * sched_tick_remote: rq->lock
2565 */
2566 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2567 /*
2568 * sched_cgroup_fork: p->pi_lock
2569 */
2570 void (*task_fork)(struct task_struct *p);
2571 /*
2572 * finish_task_switch: no locks
2573 */
2574 void (*task_dead)(struct task_struct *p);
2575
2576 /*
2577 * sched_change
2578 */
2579 void (*switching_from)(struct rq *this_rq, struct task_struct *task);
2580 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
2581 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2582 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2583 u64 (*get_prio) (struct rq *this_rq, struct task_struct *task);
2584 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2585 u64 oldprio);
2586
2587 /*
2588 * set_load_weight: task_rq_lock + sched_change
2589 * __setscheduler_parms: task_rq_lock + sched_change
2590 */
2591 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2592 const struct load_weight *lw);
2593
2594 /*
2595 * sched_rr_get_interval: task_rq_lock
2596 */
2597 unsigned int (*get_rr_interval)(struct rq *rq,
2598 struct task_struct *task);
2599
2600 /*
2601 * task_sched_runtime: task_rq_lock
2602 */
2603 void (*update_curr)(struct rq *rq);
2604
2605#ifdef CONFIG_FAIR_GROUP_SCHED
2606 /*
2607 * sched_change_group: task_rq_lock + sched_change
2608 */
2609 void (*task_change_group)(struct task_struct *p);
2610#endif
2611
2612#ifdef CONFIG_SCHED_CORE
2613 /*
2614 * pick_next_task: rq->lock
2615 * try_steal_cookie: rq->lock (double)
2616 */
2617 int (*task_is_throttled)(struct task_struct *p, int cpu);
2618#endif
2619};
2620
2621/*
2622 * Does not nest; only used around sched_class::pick_task() rq-lock-breaks.
2623 */
2624static inline void rq_modified_clear(struct rq *rq)
2625{
2626 rq->queue_mask = 0;
2627}
2628
2629static inline bool rq_modified_above(struct rq *rq, const struct sched_class * class)
2630{
2631 unsigned int mask = class->queue_mask;
2632 return rq->queue_mask & ~((mask << 1) - 1);
2633}
2634
2635static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2636{
2637 WARN_ON_ONCE(rq->donor != prev);
2638 prev->sched_class->put_prev_task(rq, prev, NULL);
2639}
2640
2641static inline void set_next_task(struct rq *rq, struct task_struct *next)
2642{
2643 next->sched_class->set_next_task(rq, next, false);
2644}
2645
2646static inline void
2647__put_prev_set_next_dl_server(struct rq *rq,
2648 struct task_struct *prev,
2649 struct task_struct *next)
2650{
2651 prev->dl_server = NULL;
2652 next->dl_server = rq->dl_server;
2653 rq->dl_server = NULL;
2654}
2655
2656static inline void put_prev_set_next_task(struct rq *rq,
2657 struct task_struct *prev,
2658 struct task_struct *next)
2659{
2660 WARN_ON_ONCE(rq->donor != prev);
2661
2662 __put_prev_set_next_dl_server(rq, prev, next);
2663
2664 if (next == prev)
2665 return;
2666
2667 prev->sched_class->put_prev_task(rq, prev, next);
2668 next->sched_class->set_next_task(rq, next, true);
2669}
2670
2671/*
2672 * Helper to define a sched_class instance; each one is placed in a separate
2673 * section which is ordered by the linker script:
2674 *
2675 * include/asm-generic/vmlinux.lds.h
2676 *
2677 * *CAREFUL* they are laid out in *REVERSE* order!!!
2678 *
2679 * Also enforce alignment on the instance, not the type, to guarantee layout.
2680 */
2681#define DEFINE_SCHED_CLASS(name) \
2682const struct sched_class name##_sched_class \
2683 __aligned(__alignof__(struct sched_class)) \
2684 __section("__" #name "_sched_class")
2685
2686/* Defined in include/asm-generic/vmlinux.lds.h */
2687extern struct sched_class __sched_class_highest[];
2688extern struct sched_class __sched_class_lowest[];
2689
2690extern const struct sched_class stop_sched_class;
2691extern const struct sched_class dl_sched_class;
2692extern const struct sched_class rt_sched_class;
2693extern const struct sched_class fair_sched_class;
2694extern const struct sched_class idle_sched_class;
2695
2696/*
2697 * Iterate only active classes. SCX can take over all fair tasks or be
2698 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2699 */
2700static inline const struct sched_class *next_active_class(const struct sched_class *class)
2701{
2702 class++;
2703#ifdef CONFIG_SCHED_CLASS_EXT
2704 if (scx_switched_all() && class == &fair_sched_class)
2705 class++;
2706 if (!scx_enabled() && class == &ext_sched_class)
2707 class++;
2708#endif
2709 return class;
2710}
2711
2712#define for_class_range(class, _from, _to) \
2713 for (class = (_from); class < (_to); class++)
2714
2715#define for_each_class(class) \
2716 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2717
2718#define for_active_class_range(class, _from, _to) \
2719 for (class = (_from); class != (_to); class = next_active_class(class))
2720
2721#define for_each_active_class(class) \
2722 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2723
2724#define sched_class_above(_a, _b) ((_a) < (_b))
2725
2726static inline bool sched_stop_runnable(struct rq *rq)
2727{
2728 return rq->stop && task_on_rq_queued(p: rq->stop);
2729}
2730
2731static inline bool sched_dl_runnable(struct rq *rq)
2732{
2733 return rq->dl.dl_nr_running > 0;
2734}
2735
2736static inline bool sched_rt_runnable(struct rq *rq)
2737{
2738 return rq->rt.rt_queued > 0;
2739}
2740
2741static inline bool sched_fair_runnable(struct rq *rq)
2742{
2743 return rq->cfs.nr_queued > 0;
2744}
2745
2746extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev,
2747 struct rq_flags *rf);
2748extern struct task_struct *pick_task_idle(struct rq *rq, struct rq_flags *rf);
2749
2750#define SCA_CHECK 0x01
2751#define SCA_MIGRATE_DISABLE 0x02
2752#define SCA_MIGRATE_ENABLE 0x04
2753#define SCA_USER 0x08
2754
2755extern void update_group_capacity(struct sched_domain *sd, int cpu);
2756
2757extern void sched_balance_trigger(struct rq *rq);
2758
2759extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2760extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2761
2762static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2763{
2764 /* When not in the task's cpumask, no point in looking further. */
2765 if (!cpumask_test_cpu(cpu, cpumask: p->cpus_ptr))
2766 return false;
2767
2768 /* Can @cpu run a user thread? */
2769 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2770 return false;
2771
2772 return true;
2773}
2774
2775static inline cpumask_t *alloc_user_cpus_ptr(int node)
2776{
2777 /*
2778 * See set_cpus_allowed_force() above for the rcu_head usage.
2779 */
2780 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2781
2782 return kmalloc_node(size, GFP_KERNEL, node);
2783}
2784
2785static inline struct task_struct *get_push_task(struct rq *rq)
2786{
2787 struct task_struct *p = rq->donor;
2788
2789 lockdep_assert_rq_held(rq);
2790
2791 if (rq->push_busy)
2792 return NULL;
2793
2794 if (p->nr_cpus_allowed == 1)
2795 return NULL;
2796
2797 if (p->migration_disabled)
2798 return NULL;
2799
2800 rq->push_busy = true;
2801 return get_task_struct(t: p);
2802}
2803
2804extern int push_cpu_stop(void *arg);
2805
2806#ifdef CONFIG_CPU_IDLE
2807
2808static inline void idle_set_state(struct rq *rq,
2809 struct cpuidle_state *idle_state)
2810{
2811 rq->idle_state = idle_state;
2812}
2813
2814static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2815{
2816 WARN_ON_ONCE(!rcu_read_lock_held());
2817
2818 return rq->idle_state;
2819}
2820
2821#else /* !CONFIG_CPU_IDLE: */
2822
2823static inline void idle_set_state(struct rq *rq,
2824 struct cpuidle_state *idle_state)
2825{
2826}
2827
2828static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2829{
2830 return NULL;
2831}
2832
2833#endif /* !CONFIG_CPU_IDLE */
2834
2835extern void schedule_idle(void);
2836asmlinkage void schedule_user(void);
2837
2838extern void sysrq_sched_debug_show(void);
2839extern void sched_init_granularity(void);
2840extern void update_max_interval(void);
2841
2842extern void init_sched_dl_class(void);
2843extern void init_sched_rt_class(void);
2844extern void init_sched_fair_class(void);
2845
2846extern void resched_curr(struct rq *rq);
2847extern void resched_curr_lazy(struct rq *rq);
2848extern void resched_cpu(int cpu);
2849
2850extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2851extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2852
2853extern void init_dl_entity(struct sched_dl_entity *dl_se);
2854
2855extern void init_cfs_throttle_work(struct task_struct *p);
2856
2857#define BW_SHIFT 20
2858#define BW_UNIT (1 << BW_SHIFT)
2859#define RATIO_SHIFT 8
2860#define MAX_BW_BITS (64 - BW_SHIFT)
2861#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2862
2863extern unsigned long to_ratio(u64 period, u64 runtime);
2864
2865extern void init_entity_runnable_average(struct sched_entity *se);
2866extern void post_init_entity_util_avg(struct task_struct *p);
2867
2868#ifdef CONFIG_NO_HZ_FULL
2869extern bool sched_can_stop_tick(struct rq *rq);
2870extern int __init sched_tick_offload_init(void);
2871
2872/*
2873 * Tick may be needed by tasks in the runqueue depending on their policy and
2874 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2875 * nohz mode if necessary.
2876 */
2877static inline void sched_update_tick_dependency(struct rq *rq)
2878{
2879 int cpu = cpu_of(rq);
2880
2881 if (!tick_nohz_full_cpu(cpu))
2882 return;
2883
2884 if (sched_can_stop_tick(rq))
2885 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2886 else
2887 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2888}
2889#else /* !CONFIG_NO_HZ_FULL: */
2890static inline int sched_tick_offload_init(void) { return 0; }
2891static inline void sched_update_tick_dependency(struct rq *rq) { }
2892#endif /* !CONFIG_NO_HZ_FULL */
2893
2894static inline void add_nr_running(struct rq *rq, unsigned count)
2895{
2896 unsigned prev_nr = rq->nr_running;
2897
2898 rq->nr_running = prev_nr + count;
2899 if (trace_sched_update_nr_running_tp_enabled()) {
2900 call_trace_sched_update_nr_running(rq, count);
2901 }
2902
2903 if (prev_nr < 2 && rq->nr_running >= 2)
2904 set_rd_overloaded(rd: rq->rd, status: 1);
2905
2906 sched_update_tick_dependency(rq);
2907}
2908
2909static inline void sub_nr_running(struct rq *rq, unsigned count)
2910{
2911 rq->nr_running -= count;
2912 if (trace_sched_update_nr_running_tp_enabled()) {
2913 call_trace_sched_update_nr_running(rq, count: -count);
2914 }
2915
2916 /* Check if we still need preemption */
2917 sched_update_tick_dependency(rq);
2918}
2919
2920static inline void __block_task(struct rq *rq, struct task_struct *p)
2921{
2922 if (p->sched_contributes_to_load)
2923 rq->nr_uninterruptible++;
2924
2925 if (p->in_iowait) {
2926 atomic_inc(v: &rq->nr_iowait);
2927 delayacct_blkio_start();
2928 }
2929
2930 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2931
2932 /*
2933 * The moment this write goes through, ttwu() can swoop in and migrate
2934 * this task, rendering our rq->__lock ineffective.
2935 *
2936 * __schedule() try_to_wake_up()
2937 * LOCK rq->__lock LOCK p->pi_lock
2938 * pick_next_task()
2939 * pick_next_task_fair()
2940 * pick_next_entity()
2941 * dequeue_entities()
2942 * __block_task()
2943 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2944 * break;
2945 *
2946 * ACQUIRE (after ctrl-dep)
2947 *
2948 * cpu = select_task_rq();
2949 * set_task_cpu(p, cpu);
2950 * ttwu_queue()
2951 * ttwu_do_activate()
2952 * LOCK rq->__lock
2953 * activate_task()
2954 * STORE p->on_rq = 1
2955 * UNLOCK rq->__lock
2956 *
2957 * Callers must ensure to not reference @p after this -- we no longer
2958 * own it.
2959 */
2960 smp_store_release(&p->on_rq, 0);
2961}
2962
2963extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2964extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2965
2966extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2967
2968#ifdef CONFIG_PREEMPT_RT
2969# define SCHED_NR_MIGRATE_BREAK 8
2970#else
2971# define SCHED_NR_MIGRATE_BREAK 32
2972#endif
2973
2974extern __read_mostly unsigned int sysctl_sched_nr_migrate;
2975extern __read_mostly unsigned int sysctl_sched_migration_cost;
2976
2977extern unsigned int sysctl_sched_base_slice;
2978
2979extern int sysctl_resched_latency_warn_ms;
2980extern int sysctl_resched_latency_warn_once;
2981
2982extern unsigned int sysctl_sched_tunable_scaling;
2983
2984extern unsigned int sysctl_numa_balancing_scan_delay;
2985extern unsigned int sysctl_numa_balancing_scan_period_min;
2986extern unsigned int sysctl_numa_balancing_scan_period_max;
2987extern unsigned int sysctl_numa_balancing_scan_size;
2988extern unsigned int sysctl_numa_balancing_hot_threshold;
2989
2990#ifdef CONFIG_SCHED_HRTICK
2991
2992/*
2993 * Use hrtick when:
2994 * - enabled by features
2995 * - hrtimer is actually high res
2996 */
2997static inline int hrtick_enabled(struct rq *rq)
2998{
2999 if (!cpu_active(cpu: cpu_of(rq)))
3000 return 0;
3001 return hrtimer_is_hres_active(timer: &rq->hrtick_timer);
3002}
3003
3004static inline int hrtick_enabled_fair(struct rq *rq)
3005{
3006 if (!sched_feat(HRTICK))
3007 return 0;
3008 return hrtick_enabled(rq);
3009}
3010
3011static inline int hrtick_enabled_dl(struct rq *rq)
3012{
3013 if (!sched_feat(HRTICK_DL))
3014 return 0;
3015 return hrtick_enabled(rq);
3016}
3017
3018extern void hrtick_start(struct rq *rq, u64 delay);
3019
3020#else /* !CONFIG_SCHED_HRTICK: */
3021
3022static inline int hrtick_enabled_fair(struct rq *rq)
3023{
3024 return 0;
3025}
3026
3027static inline int hrtick_enabled_dl(struct rq *rq)
3028{
3029 return 0;
3030}
3031
3032static inline int hrtick_enabled(struct rq *rq)
3033{
3034 return 0;
3035}
3036
3037#endif /* !CONFIG_SCHED_HRTICK */
3038
3039#ifndef arch_scale_freq_tick
3040static __always_inline void arch_scale_freq_tick(void) { }
3041#endif
3042
3043#ifndef arch_scale_freq_capacity
3044/**
3045 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
3046 * @cpu: the CPU in question.
3047 *
3048 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
3049 *
3050 * f_curr
3051 * ------ * SCHED_CAPACITY_SCALE
3052 * f_max
3053 */
3054static __always_inline
3055unsigned long arch_scale_freq_capacity(int cpu)
3056{
3057 return SCHED_CAPACITY_SCALE;
3058}
3059#endif
3060
3061/*
3062 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
3063 * acquire rq lock instead of rq_lock(). So at the end of these two functions
3064 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
3065 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
3066 */
3067static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
3068{
3069 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
3070 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
3071}
3072
3073#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
3074__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
3075static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
3076{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
3077 _lock; return _t; }
3078
3079static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
3080{
3081#ifdef CONFIG_SCHED_CORE
3082 /*
3083 * In order to not have {0,2},{1,3} turn into into an AB-BA,
3084 * order by core-id first and cpu-id second.
3085 *
3086 * Notably:
3087 *
3088 * double_rq_lock(0,3); will take core-0, core-1 lock
3089 * double_rq_lock(1,2); will take core-1, core-0 lock
3090 *
3091 * when only cpu-id is considered.
3092 */
3093 if (rq1->core->cpu < rq2->core->cpu)
3094 return true;
3095 if (rq1->core->cpu > rq2->core->cpu)
3096 return false;
3097
3098 /*
3099 * __sched_core_flip() relies on SMT having cpu-id lock order.
3100 */
3101#endif /* CONFIG_SCHED_CORE */
3102 return rq1->cpu < rq2->cpu;
3103}
3104
3105extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
3106
3107#ifdef CONFIG_PREEMPTION
3108
3109/*
3110 * fair double_lock_balance: Safely acquires both rq->locks in a fair
3111 * way at the expense of forcing extra atomic operations in all
3112 * invocations. This assures that the double_lock is acquired using the
3113 * same underlying policy as the spinlock_t on this architecture, which
3114 * reduces latency compared to the unfair variant below. However, it
3115 * also adds more overhead and therefore may reduce throughput.
3116 */
3117static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3118 __releases(this_rq->lock)
3119 __acquires(busiest->lock)
3120 __acquires(this_rq->lock)
3121{
3122 raw_spin_rq_unlock(rq: this_rq);
3123 double_rq_lock(rq1: this_rq, rq2: busiest);
3124
3125 return 1;
3126}
3127
3128#else /* !CONFIG_PREEMPTION: */
3129/*
3130 * Unfair double_lock_balance: Optimizes throughput at the expense of
3131 * latency by eliminating extra atomic operations when the locks are
3132 * already in proper order on entry. This favors lower CPU-ids and will
3133 * grant the double lock to lower CPUs over higher ids under contention,
3134 * regardless of entry order into the function.
3135 */
3136static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3137 __releases(this_rq->lock)
3138 __acquires(busiest->lock)
3139 __acquires(this_rq->lock)
3140{
3141 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
3142 likely(raw_spin_rq_trylock(busiest))) {
3143 double_rq_clock_clear_update(this_rq, busiest);
3144 return 0;
3145 }
3146
3147 if (rq_order_less(this_rq, busiest)) {
3148 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3149 double_rq_clock_clear_update(this_rq, busiest);
3150 return 0;
3151 }
3152
3153 raw_spin_rq_unlock(this_rq);
3154 double_rq_lock(this_rq, busiest);
3155
3156 return 1;
3157}
3158
3159#endif /* !CONFIG_PREEMPTION */
3160
3161/*
3162 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3163 */
3164static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3165{
3166 lockdep_assert_irqs_disabled();
3167
3168 return _double_lock_balance(this_rq, busiest);
3169}
3170
3171static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3172 __releases(busiest->lock)
3173{
3174 if (__rq_lockp(rq: this_rq) != __rq_lockp(rq: busiest))
3175 raw_spin_rq_unlock(rq: busiest);
3176 lock_set_subclass(lock: &__rq_lockp(rq: this_rq)->dep_map, subclass: 0, _RET_IP_);
3177}
3178
3179static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3180{
3181 if (l1 > l2)
3182 swap(l1, l2);
3183
3184 spin_lock(lock: l1);
3185 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3186}
3187
3188static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3189{
3190 if (l1 > l2)
3191 swap(l1, l2);
3192
3193 spin_lock_irq(lock: l1);
3194 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3195}
3196
3197static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3198{
3199 if (l1 > l2)
3200 swap(l1, l2);
3201
3202 raw_spin_lock(l1);
3203 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3204}
3205
3206static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3207{
3208 raw_spin_unlock(l1);
3209 raw_spin_unlock(l2);
3210}
3211
3212DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3213 double_raw_lock(_T->lock, _T->lock2),
3214 double_raw_unlock(_T->lock, _T->lock2))
3215
3216/*
3217 * double_rq_unlock - safely unlock two runqueues
3218 *
3219 * Note this does not restore interrupts like task_rq_unlock,
3220 * you need to do so manually after calling.
3221 */
3222static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3223 __releases(rq1->lock)
3224 __releases(rq2->lock)
3225{
3226 if (__rq_lockp(rq: rq1) != __rq_lockp(rq: rq2))
3227 raw_spin_rq_unlock(rq: rq2);
3228 else
3229 __release(rq2->lock);
3230 raw_spin_rq_unlock(rq: rq1);
3231}
3232
3233extern void set_rq_online (struct rq *rq);
3234extern void set_rq_offline(struct rq *rq);
3235
3236extern bool sched_smp_initialized;
3237
3238DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3239 double_rq_lock(_T->lock, _T->lock2),
3240 double_rq_unlock(_T->lock, _T->lock2))
3241
3242extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3243extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3244extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3245
3246extern bool sched_debug_verbose;
3247
3248extern void print_cfs_stats(struct seq_file *m, int cpu);
3249extern void print_rt_stats(struct seq_file *m, int cpu);
3250extern void print_dl_stats(struct seq_file *m, int cpu);
3251extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3252extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3253extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3254
3255extern void resched_latency_warn(int cpu, u64 latency);
3256
3257#ifdef CONFIG_NUMA_BALANCING
3258extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3259extern void
3260print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3261 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3262#endif /* CONFIG_NUMA_BALANCING */
3263
3264extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3265extern void init_rt_rq(struct rt_rq *rt_rq);
3266extern void init_dl_rq(struct dl_rq *dl_rq);
3267
3268extern void cfs_bandwidth_usage_inc(void);
3269extern void cfs_bandwidth_usage_dec(void);
3270
3271#ifdef CONFIG_NO_HZ_COMMON
3272
3273#define NOHZ_BALANCE_KICK_BIT 0
3274#define NOHZ_STATS_KICK_BIT 1
3275#define NOHZ_NEWILB_KICK_BIT 2
3276#define NOHZ_NEXT_KICK_BIT 3
3277
3278/* Run sched_balance_domains() */
3279#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3280/* Update blocked load */
3281#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3282/* Update blocked load when entering idle */
3283#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3284/* Update nohz.next_balance */
3285#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3286
3287#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3288
3289#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3290
3291extern void nohz_balance_exit_idle(struct rq *rq);
3292#else /* !CONFIG_NO_HZ_COMMON: */
3293static inline void nohz_balance_exit_idle(struct rq *rq) { }
3294#endif /* !CONFIG_NO_HZ_COMMON */
3295
3296#ifdef CONFIG_NO_HZ_COMMON
3297extern void nohz_run_idle_balance(int cpu);
3298#else
3299static inline void nohz_run_idle_balance(int cpu) { }
3300#endif
3301
3302#include "stats.h"
3303
3304#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3305
3306extern void __sched_core_account_forceidle(struct rq *rq);
3307
3308static inline void sched_core_account_forceidle(struct rq *rq)
3309{
3310 if (schedstat_enabled())
3311 __sched_core_account_forceidle(rq);
3312}
3313
3314extern void __sched_core_tick(struct rq *rq);
3315
3316static inline void sched_core_tick(struct rq *rq)
3317{
3318 if (sched_core_enabled(rq) && schedstat_enabled())
3319 __sched_core_tick(rq);
3320}
3321
3322#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3323
3324static inline void sched_core_account_forceidle(struct rq *rq) { }
3325
3326static inline void sched_core_tick(struct rq *rq) { }
3327
3328#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3329
3330#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3331
3332struct irqtime {
3333 u64 total;
3334 u64 tick_delta;
3335 u64 irq_start_time;
3336 struct u64_stats_sync sync;
3337};
3338
3339DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3340extern int sched_clock_irqtime;
3341
3342static inline int irqtime_enabled(void)
3343{
3344 return sched_clock_irqtime;
3345}
3346
3347/*
3348 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3349 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3350 * and never move forward.
3351 */
3352static inline u64 irq_time_read(int cpu)
3353{
3354 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3355 unsigned int seq;
3356 u64 total;
3357
3358 do {
3359 seq = __u64_stats_fetch_begin(syncp: &irqtime->sync);
3360 total = irqtime->total;
3361 } while (__u64_stats_fetch_retry(syncp: &irqtime->sync, start: seq));
3362
3363 return total;
3364}
3365
3366#else /* !CONFIG_IRQ_TIME_ACCOUNTING: */
3367
3368static inline int irqtime_enabled(void)
3369{
3370 return 0;
3371}
3372
3373#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
3374
3375#ifdef CONFIG_CPU_FREQ
3376
3377DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3378
3379/**
3380 * cpufreq_update_util - Take a note about CPU utilization changes.
3381 * @rq: Runqueue to carry out the update for.
3382 * @flags: Update reason flags.
3383 *
3384 * This function is called by the scheduler on the CPU whose utilization is
3385 * being updated.
3386 *
3387 * It can only be called from RCU-sched read-side critical sections.
3388 *
3389 * The way cpufreq is currently arranged requires it to evaluate the CPU
3390 * performance state (frequency/voltage) on a regular basis to prevent it from
3391 * being stuck in a completely inadequate performance level for too long.
3392 * That is not guaranteed to happen if the updates are only triggered from CFS
3393 * and DL, though, because they may not be coming in if only RT tasks are
3394 * active all the time (or there are RT tasks only).
3395 *
3396 * As a workaround for that issue, this function is called periodically by the
3397 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3398 * but that really is a band-aid. Going forward it should be replaced with
3399 * solutions targeted more specifically at RT tasks.
3400 */
3401static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3402{
3403 struct update_util_data *data;
3404
3405 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3406 cpu_of(rq)));
3407 if (data)
3408 data->func(data, rq_clock(rq), flags);
3409}
3410#else /* !CONFIG_CPU_FREQ: */
3411static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3412#endif /* !CONFIG_CPU_FREQ */
3413
3414#ifdef arch_scale_freq_capacity
3415# ifndef arch_scale_freq_invariant
3416# define arch_scale_freq_invariant() true
3417# endif
3418#else
3419# define arch_scale_freq_invariant() false
3420#endif
3421
3422unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3423 unsigned long *min,
3424 unsigned long *max);
3425
3426unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3427 unsigned long min,
3428 unsigned long max);
3429
3430
3431/*
3432 * Verify the fitness of task @p to run on @cpu taking into account the
3433 * CPU original capacity and the runtime/deadline ratio of the task.
3434 *
3435 * The function will return true if the original capacity of @cpu is
3436 * greater than or equal to task's deadline density right shifted by
3437 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3438 */
3439static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3440{
3441 unsigned long cap = arch_scale_cpu_capacity(cpu);
3442
3443 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3444}
3445
3446static inline unsigned long cpu_bw_dl(struct rq *rq)
3447{
3448 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3449}
3450
3451static inline unsigned long cpu_util_dl(struct rq *rq)
3452{
3453 return READ_ONCE(rq->avg_dl.util_avg);
3454}
3455
3456
3457extern unsigned long cpu_util_cfs(int cpu);
3458extern unsigned long cpu_util_cfs_boost(int cpu);
3459
3460static inline unsigned long cpu_util_rt(struct rq *rq)
3461{
3462 return READ_ONCE(rq->avg_rt.util_avg);
3463}
3464
3465#ifdef CONFIG_UCLAMP_TASK
3466
3467unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3468
3469/*
3470 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3471 * by default in the fast path and only gets turned on once userspace performs
3472 * an operation that requires it.
3473 *
3474 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3475 * hence is active.
3476 */
3477static inline bool uclamp_is_used(void)
3478{
3479 return static_branch_likely(&sched_uclamp_used);
3480}
3481
3482/*
3483 * Enabling static branches would get the cpus_read_lock(),
3484 * check whether uclamp_is_used before enable it to avoid always
3485 * calling cpus_read_lock(). Because we never disable this
3486 * static key once enable it.
3487 */
3488static inline void sched_uclamp_enable(void)
3489{
3490 if (!uclamp_is_used())
3491 static_branch_enable(&sched_uclamp_used);
3492}
3493
3494static inline unsigned long uclamp_rq_get(struct rq *rq,
3495 enum uclamp_id clamp_id)
3496{
3497 return READ_ONCE(rq->uclamp[clamp_id].value);
3498}
3499
3500static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3501 unsigned int value)
3502{
3503 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3504}
3505
3506static inline bool uclamp_rq_is_idle(struct rq *rq)
3507{
3508 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3509}
3510
3511/* Is the rq being capped/throttled by uclamp_max? */
3512static inline bool uclamp_rq_is_capped(struct rq *rq)
3513{
3514 unsigned long rq_util;
3515 unsigned long max_util;
3516
3517 if (!uclamp_is_used())
3518 return false;
3519
3520 rq_util = cpu_util_cfs(cpu: cpu_of(rq)) + cpu_util_rt(rq);
3521 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3522
3523 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3524}
3525
3526#define for_each_clamp_id(clamp_id) \
3527 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3528
3529extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3530
3531
3532static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3533{
3534 if (clamp_id == UCLAMP_MIN)
3535 return 0;
3536 return SCHED_CAPACITY_SCALE;
3537}
3538
3539/* Integer rounded range for each bucket */
3540#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3541
3542static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3543{
3544 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3545}
3546
3547static inline void
3548uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3549{
3550 uc_se->value = value;
3551 uc_se->bucket_id = uclamp_bucket_id(clamp_value: value);
3552 uc_se->user_defined = user_defined;
3553}
3554
3555#else /* !CONFIG_UCLAMP_TASK: */
3556
3557static inline unsigned long
3558uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3559{
3560 if (clamp_id == UCLAMP_MIN)
3561 return 0;
3562
3563 return SCHED_CAPACITY_SCALE;
3564}
3565
3566static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3567
3568static inline bool uclamp_is_used(void)
3569{
3570 return false;
3571}
3572
3573static inline void sched_uclamp_enable(void) {}
3574
3575static inline unsigned long
3576uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3577{
3578 if (clamp_id == UCLAMP_MIN)
3579 return 0;
3580
3581 return SCHED_CAPACITY_SCALE;
3582}
3583
3584static inline void
3585uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3586{
3587}
3588
3589static inline bool uclamp_rq_is_idle(struct rq *rq)
3590{
3591 return false;
3592}
3593
3594#endif /* !CONFIG_UCLAMP_TASK */
3595
3596#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3597
3598static inline unsigned long cpu_util_irq(struct rq *rq)
3599{
3600 return READ_ONCE(rq->avg_irq.util_avg);
3601}
3602
3603static inline
3604unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3605{
3606 util *= (max - irq);
3607 util /= max;
3608
3609 return util;
3610
3611}
3612
3613#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3614
3615static inline unsigned long cpu_util_irq(struct rq *rq)
3616{
3617 return 0;
3618}
3619
3620static inline
3621unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3622{
3623 return util;
3624}
3625
3626#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3627
3628extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr);
3629
3630#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3631
3632#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3633
3634DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3635
3636static inline bool sched_energy_enabled(void)
3637{
3638 return static_branch_unlikely(&sched_energy_present);
3639}
3640
3641#else /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL): */
3642
3643#define perf_domain_span(pd) NULL
3644
3645static inline bool sched_energy_enabled(void) { return false; }
3646
3647#endif /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3648
3649#ifdef CONFIG_MEMBARRIER
3650
3651/*
3652 * The scheduler provides memory barriers required by membarrier between:
3653 * - prior user-space memory accesses and store to rq->membarrier_state,
3654 * - store to rq->membarrier_state and following user-space memory accesses.
3655 * In the same way it provides those guarantees around store to rq->curr.
3656 */
3657static inline void membarrier_switch_mm(struct rq *rq,
3658 struct mm_struct *prev_mm,
3659 struct mm_struct *next_mm)
3660{
3661 int membarrier_state;
3662
3663 if (prev_mm == next_mm)
3664 return;
3665
3666 membarrier_state = atomic_read(v: &next_mm->membarrier_state);
3667 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3668 return;
3669
3670 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3671}
3672
3673#else /* !CONFIG_MEMBARRIER: */
3674
3675static inline void membarrier_switch_mm(struct rq *rq,
3676 struct mm_struct *prev_mm,
3677 struct mm_struct *next_mm)
3678{
3679}
3680
3681#endif /* !CONFIG_MEMBARRIER */
3682
3683static inline bool is_per_cpu_kthread(struct task_struct *p)
3684{
3685 if (!(p->flags & PF_KTHREAD))
3686 return false;
3687
3688 if (p->nr_cpus_allowed != 1)
3689 return false;
3690
3691 return true;
3692}
3693
3694extern void swake_up_all_locked(struct swait_queue_head *q);
3695extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3696
3697extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3698
3699#ifdef CONFIG_PREEMPT_DYNAMIC
3700extern int preempt_dynamic_mode;
3701extern int sched_dynamic_mode(const char *str);
3702extern void sched_dynamic_update(int mode);
3703#endif
3704extern const char *preempt_modes[];
3705
3706#ifdef CONFIG_SCHED_MM_CID
3707
3708static __always_inline bool cid_on_cpu(unsigned int cid)
3709{
3710 return cid & MM_CID_ONCPU;
3711}
3712
3713static __always_inline bool cid_in_transit(unsigned int cid)
3714{
3715 return cid & MM_CID_TRANSIT;
3716}
3717
3718static __always_inline unsigned int cpu_cid_to_cid(unsigned int cid)
3719{
3720 return cid & ~MM_CID_ONCPU;
3721}
3722
3723static __always_inline unsigned int cid_to_cpu_cid(unsigned int cid)
3724{
3725 return cid | MM_CID_ONCPU;
3726}
3727
3728static __always_inline unsigned int cid_to_transit_cid(unsigned int cid)
3729{
3730 return cid | MM_CID_TRANSIT;
3731}
3732
3733static __always_inline unsigned int cid_from_transit_cid(unsigned int cid)
3734{
3735 return cid & ~MM_CID_TRANSIT;
3736}
3737
3738static __always_inline bool cid_on_task(unsigned int cid)
3739{
3740 /* True if none of the MM_CID_ONCPU, MM_CID_TRANSIT, MM_CID_UNSET bits is set */
3741 return cid < MM_CID_TRANSIT;
3742}
3743
3744static __always_inline void mm_drop_cid(struct mm_struct *mm, unsigned int cid)
3745{
3746 clear_bit(nr: cid, addr: mm_cidmask(mm));
3747}
3748
3749static __always_inline void mm_unset_cid_on_task(struct task_struct *t)
3750{
3751 unsigned int cid = t->mm_cid.cid;
3752
3753 t->mm_cid.cid = MM_CID_UNSET;
3754 if (cid_on_task(cid))
3755 mm_drop_cid(mm: t->mm, cid);
3756}
3757
3758static __always_inline void mm_drop_cid_on_cpu(struct mm_struct *mm, struct mm_cid_pcpu *pcp)
3759{
3760 /* Clear the ONCPU bit, but do not set UNSET in the per CPU storage */
3761 pcp->cid = cpu_cid_to_cid(cid: pcp->cid);
3762 mm_drop_cid(mm, cid: pcp->cid);
3763}
3764
3765static inline unsigned int __mm_get_cid(struct mm_struct *mm, unsigned int max_cids)
3766{
3767 unsigned int cid = find_first_zero_bit(addr: mm_cidmask(mm), size: max_cids);
3768
3769 if (cid >= max_cids)
3770 return MM_CID_UNSET;
3771 if (test_and_set_bit(nr: cid, addr: mm_cidmask(mm)))
3772 return MM_CID_UNSET;
3773 return cid;
3774}
3775
3776static inline unsigned int mm_get_cid(struct mm_struct *mm)
3777{
3778 unsigned int cid = __mm_get_cid(mm, READ_ONCE(mm->mm_cid.max_cids));
3779
3780 while (cid == MM_CID_UNSET) {
3781 cpu_relax();
3782 cid = __mm_get_cid(mm, max_cids: num_possible_cpus());
3783 }
3784 return cid;
3785}
3786
3787static inline unsigned int mm_cid_converge(struct mm_struct *mm, unsigned int orig_cid,
3788 unsigned int max_cids)
3789{
3790 unsigned int new_cid, cid = cpu_cid_to_cid(cid: orig_cid);
3791
3792 /* Is it in the optimal CID space? */
3793 if (likely(cid < max_cids))
3794 return orig_cid;
3795
3796 /* Try to find one in the optimal space. Otherwise keep the provided. */
3797 new_cid = __mm_get_cid(mm, max_cids);
3798 if (new_cid != MM_CID_UNSET) {
3799 mm_drop_cid(mm, cid);
3800 /* Preserve the ONCPU mode of the original CID */
3801 return new_cid | (orig_cid & MM_CID_ONCPU);
3802 }
3803 return orig_cid;
3804}
3805
3806static __always_inline void mm_cid_update_task_cid(struct task_struct *t, unsigned int cid)
3807{
3808 if (t->mm_cid.cid != cid) {
3809 t->mm_cid.cid = cid;
3810 rseq_sched_set_ids_changed(t);
3811 }
3812}
3813
3814static __always_inline void mm_cid_update_pcpu_cid(struct mm_struct *mm, unsigned int cid)
3815{
3816 __this_cpu_write(mm->mm_cid.pcpu->cid, cid);
3817}
3818
3819static __always_inline void mm_cid_from_cpu(struct task_struct *t, unsigned int cpu_cid)
3820{
3821 unsigned int max_cids, tcid = t->mm_cid.cid;
3822 struct mm_struct *mm = t->mm;
3823
3824 max_cids = READ_ONCE(mm->mm_cid.max_cids);
3825 /* Optimize for the common case where both have the ONCPU bit set */
3826 if (likely(cid_on_cpu(cpu_cid & tcid))) {
3827 if (likely(cpu_cid_to_cid(cpu_cid) < max_cids)) {
3828 mm_cid_update_task_cid(t, cid: cpu_cid);
3829 return;
3830 }
3831 /* Try to converge into the optimal CID space */
3832 cpu_cid = mm_cid_converge(mm, orig_cid: cpu_cid, max_cids);
3833 } else {
3834 /* Hand over or drop the task owned CID */
3835 if (cid_on_task(cid: tcid)) {
3836 if (cid_on_cpu(cid: cpu_cid))
3837 mm_unset_cid_on_task(t);
3838 else
3839 cpu_cid = cid_to_cpu_cid(cid: tcid);
3840 }
3841 /* Still nothing, allocate a new one */
3842 if (!cid_on_cpu(cid: cpu_cid))
3843 cpu_cid = cid_to_cpu_cid(cid: mm_get_cid(mm));
3844 }
3845 mm_cid_update_pcpu_cid(mm, cid: cpu_cid);
3846 mm_cid_update_task_cid(t, cid: cpu_cid);
3847}
3848
3849static __always_inline void mm_cid_from_task(struct task_struct *t, unsigned int cpu_cid)
3850{
3851 unsigned int max_cids, tcid = t->mm_cid.cid;
3852 struct mm_struct *mm = t->mm;
3853
3854 max_cids = READ_ONCE(mm->mm_cid.max_cids);
3855 /* Optimize for the common case, where both have the ONCPU bit clear */
3856 if (likely(cid_on_task(tcid | cpu_cid))) {
3857 if (likely(tcid < max_cids)) {
3858 mm_cid_update_pcpu_cid(mm, cid: tcid);
3859 return;
3860 }
3861 /* Try to converge into the optimal CID space */
3862 tcid = mm_cid_converge(mm, orig_cid: tcid, max_cids);
3863 } else {
3864 /* Hand over or drop the CPU owned CID */
3865 if (cid_on_cpu(cid: cpu_cid)) {
3866 if (cid_on_task(cid: tcid))
3867 mm_drop_cid_on_cpu(mm, this_cpu_ptr(mm->mm_cid.pcpu));
3868 else
3869 tcid = cpu_cid_to_cid(cid: cpu_cid);
3870 }
3871 /* Still nothing, allocate a new one */
3872 if (!cid_on_task(cid: tcid))
3873 tcid = mm_get_cid(mm);
3874 /* Set the transition mode flag if required */
3875 tcid |= READ_ONCE(mm->mm_cid.transit);
3876 }
3877 mm_cid_update_pcpu_cid(mm, cid: tcid);
3878 mm_cid_update_task_cid(t, cid: tcid);
3879}
3880
3881static __always_inline void mm_cid_schedin(struct task_struct *next)
3882{
3883 struct mm_struct *mm = next->mm;
3884 unsigned int cpu_cid;
3885
3886 if (!next->mm_cid.active)
3887 return;
3888
3889 cpu_cid = __this_cpu_read(mm->mm_cid.pcpu->cid);
3890 if (likely(!READ_ONCE(mm->mm_cid.percpu)))
3891 mm_cid_from_task(t: next, cpu_cid);
3892 else
3893 mm_cid_from_cpu(t: next, cpu_cid);
3894}
3895
3896static __always_inline void mm_cid_schedout(struct task_struct *prev)
3897{
3898 /* During mode transitions CIDs are temporary and need to be dropped */
3899 if (likely(!cid_in_transit(prev->mm_cid.cid)))
3900 return;
3901
3902 mm_drop_cid(mm: prev->mm, cid: cid_from_transit_cid(cid: prev->mm_cid.cid));
3903 prev->mm_cid.cid = MM_CID_UNSET;
3904}
3905
3906static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next)
3907{
3908 mm_cid_schedout(prev);
3909 mm_cid_schedin(next);
3910}
3911
3912#else /* !CONFIG_SCHED_MM_CID: */
3913static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next) { }
3914#endif /* !CONFIG_SCHED_MM_CID */
3915
3916extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3917extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3918static inline
3919void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
3920{
3921 lockdep_assert_rq_held(rq: src_rq);
3922 lockdep_assert_rq_held(rq: dst_rq);
3923
3924 deactivate_task(rq: src_rq, p: task, flags: 0);
3925 set_task_cpu(p: task, cpu: dst_rq->cpu);
3926 activate_task(rq: dst_rq, p: task, flags: 0);
3927}
3928
3929static inline
3930bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
3931{
3932 if (!task_on_cpu(rq, p) &&
3933 cpumask_test_cpu(cpu, cpumask: &p->cpus_mask))
3934 return true;
3935
3936 return false;
3937}
3938
3939#ifdef CONFIG_RT_MUTEXES
3940
3941static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3942{
3943 if (pi_task)
3944 prio = min(prio, pi_task->prio);
3945
3946 return prio;
3947}
3948
3949static inline int rt_effective_prio(struct task_struct *p, int prio)
3950{
3951 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3952
3953 return __rt_effective_prio(pi_task, prio);
3954}
3955
3956#else /* !CONFIG_RT_MUTEXES: */
3957
3958static inline int rt_effective_prio(struct task_struct *p, int prio)
3959{
3960 return prio;
3961}
3962
3963#endif /* !CONFIG_RT_MUTEXES */
3964
3965extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3966extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3967extern const struct sched_class *__setscheduler_class(int policy, int prio);
3968extern void set_load_weight(struct task_struct *p, bool update_load);
3969extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3970extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3971
3972extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3973
3974extern void __balance_callbacks(struct rq *rq, struct rq_flags *rf);
3975extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3976
3977/*
3978 * The 'sched_change' pattern is the safe, easy and slow way of changing a
3979 * task's scheduling properties. It dequeues a task, such that the scheduler
3980 * is fully unaware of it; at which point its properties can be modified;
3981 * after which it is enqueued again.
3982 *
3983 * Typically this must be called while holding task_rq_lock, since most/all
3984 * properties are serialized under those locks. There is currently one
3985 * exception to this rule in sched/ext which only holds rq->lock.
3986 */
3987
3988/*
3989 * This structure is a temporary, used to preserve/convey the queueing state
3990 * of the task between sched_change_begin() and sched_change_end(). Ensuring
3991 * the task's queueing state is idempotent across the operation.
3992 */
3993struct sched_change_ctx {
3994 u64 prio;
3995 struct task_struct *p;
3996 int flags;
3997 bool queued;
3998 bool running;
3999};
4000
4001struct sched_change_ctx *sched_change_begin(struct task_struct *p, unsigned int flags);
4002void sched_change_end(struct sched_change_ctx *ctx);
4003
4004DEFINE_CLASS(sched_change, struct sched_change_ctx *,
4005 sched_change_end(_T),
4006 sched_change_begin(p, flags),
4007 struct task_struct *p, unsigned int flags)
4008
4009DEFINE_CLASS_IS_UNCONDITIONAL(sched_change)
4010
4011#include "ext.h"
4012
4013#endif /* _KERNEL_SCHED_SCHED_H */
4014

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