projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'mm-hotfixes-stable-2025-07-11-16-16' of git://git.kernel.org/pub/scm/linux...
[linux-block.git] / kernel / sched / sched.h
... / ...
CommitLineData
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Scheduler internal types and methods:
4 */
5#ifndef _KERNEL_SCHED_SCHED_H
6#define _KERNEL_SCHED_SCHED_H
7
8#include <linux/sched/affinity.h>
9#include <linux/sched/autogroup.h>
10#include <linux/sched/cpufreq.h>
11#include <linux/sched/deadline.h>
12#include <linux/sched.h>
13#include <linux/sched/loadavg.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/rseq_api.h>
16#include <linux/sched/signal.h>
17#include <linux/sched/smt.h>
18#include <linux/sched/stat.h>
19#include <linux/sched/sysctl.h>
20#include <linux/sched/task_flags.h>
21#include <linux/sched/task.h>
22#include <linux/sched/topology.h>
23
24#include <linux/atomic.h>
25#include <linux/bitmap.h>
26#include <linux/bug.h>
27#include <linux/capability.h>
28#include <linux/cgroup_api.h>
29#include <linux/cgroup.h>
30#include <linux/context_tracking.h>
31#include <linux/cpufreq.h>
32#include <linux/cpumask_api.h>
33#include <linux/ctype.h>
34#include <linux/file.h>
35#include <linux/fs_api.h>
36#include <linux/hrtimer_api.h>
37#include <linux/interrupt.h>
38#include <linux/irq_work.h>
39#include <linux/jiffies.h>
40#include <linux/kref_api.h>
41#include <linux/kthread.h>
42#include <linux/ktime_api.h>
43#include <linux/lockdep_api.h>
44#include <linux/lockdep.h>
45#include <linux/minmax.h>
46#include <linux/mm.h>
47#include <linux/module.h>
48#include <linux/mutex_api.h>
49#include <linux/plist.h>
50#include <linux/poll.h>
51#include <linux/proc_fs.h>
52#include <linux/profile.h>
53#include <linux/psi.h>
54#include <linux/rcupdate.h>
55#include <linux/seq_file.h>
56#include <linux/seqlock.h>
57#include <linux/softirq.h>
58#include <linux/spinlock_api.h>
59#include <linux/static_key.h>
60#include <linux/stop_machine.h>
61#include <linux/syscalls_api.h>
62#include <linux/syscalls.h>
63#include <linux/tick.h>
64#include <linux/topology.h>
65#include <linux/types.h>
66#include <linux/u64_stats_sync_api.h>
67#include <linux/uaccess.h>
68#include <linux/wait_api.h>
69#include <linux/wait_bit.h>
70#include <linux/workqueue_api.h>
71#include <linux/delayacct.h>
72
73#include <trace/events/power.h>
74#include <trace/events/sched.h>
75
76#include "../workqueue_internal.h"
77
78struct rq;
79struct cfs_rq;
80struct rt_rq;
81struct sched_group;
82struct cpuidle_state;
83
84#ifdef CONFIG_PARAVIRT
85# include <asm/paravirt.h>
86# include <asm/paravirt_api_clock.h>
87#endif
88
89#include <asm/barrier.h>
90
91#include "cpupri.h"
92#include "cpudeadline.h"
93
94/* task_struct::on_rq states: */
95#define TASK_ON_RQ_QUEUED 1
96#define TASK_ON_RQ_MIGRATING 2
97
98extern __read_mostly int scheduler_running;
99
100extern unsigned long calc_load_update;
101extern atomic_long_t calc_load_tasks;
102
103extern void calc_global_load_tick(struct rq *this_rq);
104extern long calc_load_fold_active(struct rq *this_rq, long adjust);
105
106extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
107
108extern int sysctl_sched_rt_period;
109extern int sysctl_sched_rt_runtime;
110extern int sched_rr_timeslice;
111
112/*
113 * Asymmetric CPU capacity bits
114 */
115struct asym_cap_data {
116 struct list_head link;
117 struct rcu_head rcu;
118 unsigned long capacity;
119 unsigned long cpus[];
120};
121
122extern struct list_head asym_cap_list;
123
124#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
125
126/*
127 * Helpers for converting nanosecond timing to jiffy resolution
128 */
129#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
130
131/*
132 * Increase resolution of nice-level calculations for 64-bit architectures.
133 * The extra resolution improves shares distribution and load balancing of
134 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
135 * hierarchies, especially on larger systems. This is not a user-visible change
136 * and does not change the user-interface for setting shares/weights.
137 *
138 * We increase resolution only if we have enough bits to allow this increased
139 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
140 * are pretty high and the returns do not justify the increased costs.
141 *
142 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
143 * increase coverage and consistency always enable it on 64-bit platforms.
144 */
145#ifdef CONFIG_64BIT
146# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
147# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
148# define scale_load_down(w) \
149({ \
150 unsigned long __w = (w); \
151 \
152 if (__w) \
153 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
154 __w; \
155})
156#else
157# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
158# define scale_load(w) (w)
159# define scale_load_down(w) (w)
160#endif
161
162/*
163 * Task weight (visible to users) and its load (invisible to users) have
164 * independent resolution, but they should be well calibrated. We use
165 * scale_load() and scale_load_down(w) to convert between them. The
166 * following must be true:
167 *
168 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
169 *
170 */
171#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
172
173/*
174 * Single value that decides SCHED_DEADLINE internal math precision.
175 * 10 -> just above 1us
176 * 9 -> just above 0.5us
177 */
178#define DL_SCALE 10
179
180/*
181 * Single value that denotes runtime == period, ie unlimited time.
182 */
183#define RUNTIME_INF ((u64)~0ULL)
184
185static inline int idle_policy(int policy)
186{
187 return policy == SCHED_IDLE;
188}
189
190static inline int normal_policy(int policy)
191{
192#ifdef CONFIG_SCHED_CLASS_EXT
193 if (policy == SCHED_EXT)
194 return true;
195#endif
196 return policy == SCHED_NORMAL;
197}
198
199static inline int fair_policy(int policy)
200{
201 return normal_policy(policy) || policy == SCHED_BATCH;
202}
203
204static inline int rt_policy(int policy)
205{
206 return policy == SCHED_FIFO || policy == SCHED_RR;
207}
208
209static inline int dl_policy(int policy)
210{
211 return policy == SCHED_DEADLINE;
212}
213
214static inline bool valid_policy(int policy)
215{
216 return idle_policy(policy) || fair_policy(policy) ||
217 rt_policy(policy) || dl_policy(policy);
218}
219
220static inline int task_has_idle_policy(struct task_struct *p)
221{
222 return idle_policy(p->policy);
223}
224
225static inline int task_has_rt_policy(struct task_struct *p)
226{
227 return rt_policy(p->policy);
228}
229
230static inline int task_has_dl_policy(struct task_struct *p)
231{
232 return dl_policy(p->policy);
233}
234
235#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
236
237static inline void update_avg(u64 *avg, u64 sample)
238{
239 s64 diff = sample - *avg;
240
241 *avg += diff / 8;
242}
243
244/*
245 * Shifting a value by an exponent greater *or equal* to the size of said value
246 * is UB; cap at size-1.
247 */
248#define shr_bound(val, shift) \
249 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
250
251/*
252 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
253 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
254 * maps pretty well onto the shares value used by scheduler and the round-trip
255 * conversions preserve the original value over the entire range.
256 */
257static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
258{
259 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
260}
261
262static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
263{
264 return clamp_t(unsigned long,
265 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
266 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
267}
268
269/*
270 * !! For sched_setattr_nocheck() (kernel) only !!
271 *
272 * This is actually gross. :(
273 *
274 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
275 * tasks, but still be able to sleep. We need this on platforms that cannot
276 * atomically change clock frequency. Remove once fast switching will be
277 * available on such platforms.
278 *
279 * SUGOV stands for SchedUtil GOVernor.
280 */
281#define SCHED_FLAG_SUGOV 0x10000000
282
283#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
284
285static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
286{
287#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
288 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
289#else
290 return false;
291#endif
292}
293
294/*
295 * Tells if entity @a should preempt entity @b.
296 */
297static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
298 const struct sched_dl_entity *b)
299{
300 return dl_entity_is_special(a) ||
301 dl_time_before(a->deadline, b->deadline);
302}
303
304/*
305 * This is the priority-queue data structure of the RT scheduling class:
306 */
307struct rt_prio_array {
308 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
309 struct list_head queue[MAX_RT_PRIO];
310};
311
312struct rt_bandwidth {
313 /* nests inside the rq lock: */
314 raw_spinlock_t rt_runtime_lock;
315 ktime_t rt_period;
316 u64 rt_runtime;
317 struct hrtimer rt_period_timer;
318 unsigned int rt_period_active;
319};
320
321static inline int dl_bandwidth_enabled(void)
322{
323 return sysctl_sched_rt_runtime >= 0;
324}
325
326/*
327 * To keep the bandwidth of -deadline tasks under control
328 * we need some place where:
329 * - store the maximum -deadline bandwidth of each cpu;
330 * - cache the fraction of bandwidth that is currently allocated in
331 * each root domain;
332 *
333 * This is all done in the data structure below. It is similar to the
334 * one used for RT-throttling (rt_bandwidth), with the main difference
335 * that, since here we are only interested in admission control, we
336 * do not decrease any runtime while the group "executes", neither we
337 * need a timer to replenish it.
338 *
339 * With respect to SMP, bandwidth is given on a per root domain basis,
340 * meaning that:
341 * - bw (< 100%) is the deadline bandwidth of each CPU;
342 * - total_bw is the currently allocated bandwidth in each root domain;
343 */
344struct dl_bw {
345 raw_spinlock_t lock;
346 u64 bw;
347 u64 total_bw;
348};
349
350extern void init_dl_bw(struct dl_bw *dl_b);
351extern int sched_dl_global_validate(void);
352extern void sched_dl_do_global(void);
353extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
354extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
355extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
356extern bool __checkparam_dl(const struct sched_attr *attr);
357extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
358extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
359extern int dl_bw_deactivate(int cpu);
360extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
361/*
362 * SCHED_DEADLINE supports servers (nested scheduling) with the following
363 * interface:
364 *
365 * dl_se::rq -- runqueue we belong to.
366 *
367 * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
368 * server when it runs out of tasks to run.
369 *
370 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
371 * returns NULL.
372 *
373 * dl_server_update() -- called from update_curr_common(), propagates runtime
374 * to the server.
375 *
376 * dl_server_start()
377 * dl_server_stop() -- start/stop the server when it has (no) tasks.
378 *
379 * dl_server_init() -- initializes the server.
380 */
381extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
382extern void dl_server_start(struct sched_dl_entity *dl_se);
383extern void dl_server_stop(struct sched_dl_entity *dl_se);
384extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
385 dl_server_has_tasks_f has_tasks,
386 dl_server_pick_f pick_task);
387
388extern void dl_server_update_idle_time(struct rq *rq,
389 struct task_struct *p);
390extern void fair_server_init(struct rq *rq);
391extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
392extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
393 u64 runtime, u64 period, bool init);
394
395static inline bool dl_server_active(struct sched_dl_entity *dl_se)
396{
397 return dl_se->dl_server_active;
398}
399
400#ifdef CONFIG_CGROUP_SCHED
401
402extern struct list_head task_groups;
403
404struct cfs_bandwidth {
405#ifdef CONFIG_CFS_BANDWIDTH
406 raw_spinlock_t lock;
407 ktime_t period;
408 u64 quota;
409 u64 runtime;
410 u64 burst;
411 u64 runtime_snap;
412 s64 hierarchical_quota;
413
414 u8 idle;
415 u8 period_active;
416 u8 slack_started;
417 struct hrtimer period_timer;
418 struct hrtimer slack_timer;
419 struct list_head throttled_cfs_rq;
420
421 /* Statistics: */
422 int nr_periods;
423 int nr_throttled;
424 int nr_burst;
425 u64 throttled_time;
426 u64 burst_time;
427#endif
428};
429
430/* Task group related information */
431struct task_group {
432 struct cgroup_subsys_state css;
433
434#ifdef CONFIG_GROUP_SCHED_WEIGHT
435 /* A positive value indicates that this is a SCHED_IDLE group. */
436 int idle;
437#endif
438
439#ifdef CONFIG_FAIR_GROUP_SCHED
440 /* schedulable entities of this group on each CPU */
441 struct sched_entity **se;
442 /* runqueue "owned" by this group on each CPU */
443 struct cfs_rq **cfs_rq;
444 unsigned long shares;
445#ifdef CONFIG_SMP
446 /*
447 * load_avg can be heavily contended at clock tick time, so put
448 * it in its own cache-line separated from the fields above which
449 * will also be accessed at each tick.
450 */
451 atomic_long_t load_avg ____cacheline_aligned;
452#endif
453#endif
454
455#ifdef CONFIG_RT_GROUP_SCHED
456 struct sched_rt_entity **rt_se;
457 struct rt_rq **rt_rq;
458
459 struct rt_bandwidth rt_bandwidth;
460#endif
461
462#ifdef CONFIG_EXT_GROUP_SCHED
463 u32 scx_flags; /* SCX_TG_* */
464 u32 scx_weight;
465#endif
466
467 struct rcu_head rcu;
468 struct list_head list;
469
470 struct task_group *parent;
471 struct list_head siblings;
472 struct list_head children;
473
474#ifdef CONFIG_SCHED_AUTOGROUP
475 struct autogroup *autogroup;
476#endif
477
478 struct cfs_bandwidth cfs_bandwidth;
479
480#ifdef CONFIG_UCLAMP_TASK_GROUP
481 /* The two decimal precision [%] value requested from user-space */
482 unsigned int uclamp_pct[UCLAMP_CNT];
483 /* Clamp values requested for a task group */
484 struct uclamp_se uclamp_req[UCLAMP_CNT];
485 /* Effective clamp values used for a task group */
486 struct uclamp_se uclamp[UCLAMP_CNT];
487#endif
488
489};
490
491#ifdef CONFIG_GROUP_SCHED_WEIGHT
492#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
493
494/*
495 * A weight of 0 or 1 can cause arithmetics problems.
496 * A weight of a cfs_rq is the sum of weights of which entities
497 * are queued on this cfs_rq, so a weight of a entity should not be
498 * too large, so as the shares value of a task group.
499 * (The default weight is 1024 - so there's no practical
500 * limitation from this.)
501 */
502#define MIN_SHARES (1UL << 1)
503#define MAX_SHARES (1UL << 18)
504#endif
505
506typedef int (*tg_visitor)(struct task_group *, void *);
507
508extern int walk_tg_tree_from(struct task_group *from,
509 tg_visitor down, tg_visitor up, void *data);
510
511/*
512 * Iterate the full tree, calling @down when first entering a node and @up when
513 * leaving it for the final time.
514 *
515 * Caller must hold rcu_lock or sufficient equivalent.
516 */
517static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
518{
519 return walk_tg_tree_from(&root_task_group, down, up, data);
520}
521
522static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
523{
524 return css ? container_of(css, struct task_group, css) : NULL;
525}
526
527extern int tg_nop(struct task_group *tg, void *data);
528
529#ifdef CONFIG_FAIR_GROUP_SCHED
530extern void free_fair_sched_group(struct task_group *tg);
531extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
532extern void online_fair_sched_group(struct task_group *tg);
533extern void unregister_fair_sched_group(struct task_group *tg);
534#else
535static inline void free_fair_sched_group(struct task_group *tg) { }
536static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
537{
538 return 1;
539}
540static inline void online_fair_sched_group(struct task_group *tg) { }
541static inline void unregister_fair_sched_group(struct task_group *tg) { }
542#endif
543
544extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
545 struct sched_entity *se, int cpu,
546 struct sched_entity *parent);
547extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
548
549extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
550extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
551extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
552extern bool cfs_task_bw_constrained(struct task_struct *p);
553
554extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
555 struct sched_rt_entity *rt_se, int cpu,
556 struct sched_rt_entity *parent);
557extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
558extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
559extern long sched_group_rt_runtime(struct task_group *tg);
560extern long sched_group_rt_period(struct task_group *tg);
561extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
562
563extern struct task_group *sched_create_group(struct task_group *parent);
564extern void sched_online_group(struct task_group *tg,
565 struct task_group *parent);
566extern void sched_destroy_group(struct task_group *tg);
567extern void sched_release_group(struct task_group *tg);
568
569extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
570
571#ifdef CONFIG_FAIR_GROUP_SCHED
572extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
573
574extern int sched_group_set_idle(struct task_group *tg, long idle);
575
576#ifdef CONFIG_SMP
577extern void set_task_rq_fair(struct sched_entity *se,
578 struct cfs_rq *prev, struct cfs_rq *next);
579#else /* !CONFIG_SMP */
580static inline void set_task_rq_fair(struct sched_entity *se,
581 struct cfs_rq *prev, struct cfs_rq *next) { }
582#endif /* CONFIG_SMP */
583#else /* !CONFIG_FAIR_GROUP_SCHED */
584static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
585static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
586#endif /* CONFIG_FAIR_GROUP_SCHED */
587
588#else /* CONFIG_CGROUP_SCHED */
589
590struct cfs_bandwidth { };
591
592static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
593
594#endif /* CONFIG_CGROUP_SCHED */
595
596extern void unregister_rt_sched_group(struct task_group *tg);
597extern void free_rt_sched_group(struct task_group *tg);
598extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
599
600/*
601 * u64_u32_load/u64_u32_store
602 *
603 * Use a copy of a u64 value to protect against data race. This is only
604 * applicable for 32-bits architectures.
605 */
606#ifdef CONFIG_64BIT
607# define u64_u32_load_copy(var, copy) var
608# define u64_u32_store_copy(var, copy, val) (var = val)
609#else
610# define u64_u32_load_copy(var, copy) \
611({ \
612 u64 __val, __val_copy; \
613 do { \
614 __val_copy = copy; \
615 /* \
616 * paired with u64_u32_store_copy(), ordering access \
617 * to var and copy. \
618 */ \
619 smp_rmb(); \
620 __val = var; \
621 } while (__val != __val_copy); \
622 __val; \
623})
624# define u64_u32_store_copy(var, copy, val) \
625do { \
626 typeof(val) __val = (val); \
627 var = __val; \
628 /* \
629 * paired with u64_u32_load_copy(), ordering access to var and \
630 * copy. \
631 */ \
632 smp_wmb(); \
633 copy = __val; \
634} while (0)
635#endif
636# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
637# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
638
639struct balance_callback {
640 struct balance_callback *next;
641 void (*func)(struct rq *rq);
642};
643
644/* CFS-related fields in a runqueue */
645struct cfs_rq {
646 struct load_weight load;
647 unsigned int nr_queued;
648 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */
649 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */
650 unsigned int h_nr_idle; /* SCHED_IDLE */
651
652 s64 avg_vruntime;
653 u64 avg_load;
654
655 u64 min_vruntime;
656#ifdef CONFIG_SCHED_CORE
657 unsigned int forceidle_seq;
658 u64 min_vruntime_fi;
659#endif
660
661 struct rb_root_cached tasks_timeline;
662
663 /*
664 * 'curr' points to currently running entity on this cfs_rq.
665 * It is set to NULL otherwise (i.e when none are currently running).
666 */
667 struct sched_entity *curr;
668 struct sched_entity *next;
669
670#ifdef CONFIG_SMP
671 /*
672 * CFS load tracking
673 */
674 struct sched_avg avg;
675#ifndef CONFIG_64BIT
676 u64 last_update_time_copy;
677#endif
678 struct {
679 raw_spinlock_t lock ____cacheline_aligned;
680 int nr;
681 unsigned long load_avg;
682 unsigned long util_avg;
683 unsigned long runnable_avg;
684 } removed;
685
686#ifdef CONFIG_FAIR_GROUP_SCHED
687 u64 last_update_tg_load_avg;
688 unsigned long tg_load_avg_contrib;
689 long propagate;
690 long prop_runnable_sum;
691
692 /*
693 * h_load = weight * f(tg)
694 *
695 * Where f(tg) is the recursive weight fraction assigned to
696 * this group.
697 */
698 unsigned long h_load;
699 u64 last_h_load_update;
700 struct sched_entity *h_load_next;
701#endif /* CONFIG_FAIR_GROUP_SCHED */
702#endif /* CONFIG_SMP */
703
704#ifdef CONFIG_FAIR_GROUP_SCHED
705 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
706
707 /*
708 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
709 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
710 * (like users, containers etc.)
711 *
712 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
713 * This list is used during load balance.
714 */
715 int on_list;
716 struct list_head leaf_cfs_rq_list;
717 struct task_group *tg; /* group that "owns" this runqueue */
718
719 /* Locally cached copy of our task_group's idle value */
720 int idle;
721
722#ifdef CONFIG_CFS_BANDWIDTH
723 int runtime_enabled;
724 s64 runtime_remaining;
725
726 u64 throttled_pelt_idle;
727#ifndef CONFIG_64BIT
728 u64 throttled_pelt_idle_copy;
729#endif
730 u64 throttled_clock;
731 u64 throttled_clock_pelt;
732 u64 throttled_clock_pelt_time;
733 u64 throttled_clock_self;
734 u64 throttled_clock_self_time;
735 int throttled;
736 int throttle_count;
737 struct list_head throttled_list;
738 struct list_head throttled_csd_list;
739#endif /* CONFIG_CFS_BANDWIDTH */
740#endif /* CONFIG_FAIR_GROUP_SCHED */
741};
742
743#ifdef CONFIG_SCHED_CLASS_EXT
744/* scx_rq->flags, protected by the rq lock */
745enum scx_rq_flags {
746 /*
747 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
748 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
749 * only while the BPF scheduler considers the CPU to be online.
750 */
751 SCX_RQ_ONLINE = 1 << 0,
752 SCX_RQ_CAN_STOP_TICK = 1 << 1,
753 SCX_RQ_BAL_PENDING = 1 << 2, /* balance hasn't run yet */
754 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
755 SCX_RQ_BYPASSING = 1 << 4,
756 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */
757
758 SCX_RQ_IN_WAKEUP = 1 << 16,
759 SCX_RQ_IN_BALANCE = 1 << 17,
760};
761
762struct scx_rq {
763 struct scx_dispatch_q local_dsq;
764 struct list_head runnable_list; /* runnable tasks on this rq */
765 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
766 unsigned long ops_qseq;
767 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
768 u32 nr_running;
769 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
770 bool cpu_released;
771 u32 flags;
772 u64 clock; /* current per-rq clock -- see scx_bpf_now() */
773 cpumask_var_t cpus_to_kick;
774 cpumask_var_t cpus_to_kick_if_idle;
775 cpumask_var_t cpus_to_preempt;
776 cpumask_var_t cpus_to_wait;
777 unsigned long pnt_seq;
778 struct balance_callback deferred_bal_cb;
779 struct irq_work deferred_irq_work;
780 struct irq_work kick_cpus_irq_work;
781};
782#endif /* CONFIG_SCHED_CLASS_EXT */
783
784static inline int rt_bandwidth_enabled(void)
785{
786 return sysctl_sched_rt_runtime >= 0;
787}
788
789/* RT IPI pull logic requires IRQ_WORK */
790#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
791# define HAVE_RT_PUSH_IPI
792#endif
793
794/* Real-Time classes' related field in a runqueue: */
795struct rt_rq {
796 struct rt_prio_array active;
797 unsigned int rt_nr_running;
798 unsigned int rr_nr_running;
799#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
800 struct {
801 int curr; /* highest queued rt task prio */
802#ifdef CONFIG_SMP
803 int next; /* next highest */
804#endif
805 } highest_prio;
806#endif
807#ifdef CONFIG_SMP
808 bool overloaded;
809 struct plist_head pushable_tasks;
810
811#endif /* CONFIG_SMP */
812 int rt_queued;
813
814#ifdef CONFIG_RT_GROUP_SCHED
815 int rt_throttled;
816 u64 rt_time; /* consumed RT time, goes up in update_curr_rt */
817 u64 rt_runtime; /* allotted RT time, "slice" from rt_bandwidth, RT sharing/balancing */
818 /* Nests inside the rq lock: */
819 raw_spinlock_t rt_runtime_lock;
820
821 unsigned int rt_nr_boosted;
822
823 struct rq *rq; /* this is always top-level rq, cache? */
824#endif
825#ifdef CONFIG_CGROUP_SCHED
826 struct task_group *tg; /* this tg has "this" rt_rq on given CPU for runnable entities */
827#endif
828};
829
830static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
831{
832 return rt_rq->rt_queued && rt_rq->rt_nr_running;
833}
834
835/* Deadline class' related fields in a runqueue */
836struct dl_rq {
837 /* runqueue is an rbtree, ordered by deadline */
838 struct rb_root_cached root;
839
840 unsigned int dl_nr_running;
841
842#ifdef CONFIG_SMP
843 /*
844 * Deadline values of the currently executing and the
845 * earliest ready task on this rq. Caching these facilitates
846 * the decision whether or not a ready but not running task
847 * should migrate somewhere else.
848 */
849 struct {
850 u64 curr;
851 u64 next;
852 } earliest_dl;
853
854 bool overloaded;
855
856 /*
857 * Tasks on this rq that can be pushed away. They are kept in
858 * an rb-tree, ordered by tasks' deadlines, with caching
859 * of the leftmost (earliest deadline) element.
860 */
861 struct rb_root_cached pushable_dl_tasks_root;
862#else
863 struct dl_bw dl_bw;
864#endif
865 /*
866 * "Active utilization" for this runqueue: increased when a
867 * task wakes up (becomes TASK_RUNNING) and decreased when a
868 * task blocks
869 */
870 u64 running_bw;
871
872 /*
873 * Utilization of the tasks "assigned" to this runqueue (including
874 * the tasks that are in runqueue and the tasks that executed on this
875 * CPU and blocked). Increased when a task moves to this runqueue, and
876 * decreased when the task moves away (migrates, changes scheduling
877 * policy, or terminates).
878 * This is needed to compute the "inactive utilization" for the
879 * runqueue (inactive utilization = this_bw - running_bw).
880 */
881 u64 this_bw;
882 u64 extra_bw;
883
884 /*
885 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
886 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
887 */
888 u64 max_bw;
889
890 /*
891 * Inverse of the fraction of CPU utilization that can be reclaimed
892 * by the GRUB algorithm.
893 */
894 u64 bw_ratio;
895};
896
897#ifdef CONFIG_FAIR_GROUP_SCHED
898
899/* An entity is a task if it doesn't "own" a runqueue */
900#define entity_is_task(se) (!se->my_q)
901
902static inline void se_update_runnable(struct sched_entity *se)
903{
904 if (!entity_is_task(se))
905 se->runnable_weight = se->my_q->h_nr_runnable;
906}
907
908static inline long se_runnable(struct sched_entity *se)
909{
910 if (se->sched_delayed)
911 return false;
912
913 if (entity_is_task(se))
914 return !!se->on_rq;
915 else
916 return se->runnable_weight;
917}
918
919#else /* !CONFIG_FAIR_GROUP_SCHED: */
920
921#define entity_is_task(se) 1
922
923static inline void se_update_runnable(struct sched_entity *se) { }
924
925static inline long se_runnable(struct sched_entity *se)
926{
927 if (se->sched_delayed)
928 return false;
929
930 return !!se->on_rq;
931}
932
933#endif /* !CONFIG_FAIR_GROUP_SCHED */
934
935#ifdef CONFIG_SMP
936/*
937 * XXX we want to get rid of these helpers and use the full load resolution.
938 */
939static inline long se_weight(struct sched_entity *se)
940{
941 return scale_load_down(se->load.weight);
942}
943
944
945static inline bool sched_asym_prefer(int a, int b)
946{
947 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
948}
949
950struct perf_domain {
951 struct em_perf_domain *em_pd;
952 struct perf_domain *next;
953 struct rcu_head rcu;
954};
955
956/*
957 * We add the notion of a root-domain which will be used to define per-domain
958 * variables. Each exclusive cpuset essentially defines an island domain by
959 * fully partitioning the member CPUs from any other cpuset. Whenever a new
960 * exclusive cpuset is created, we also create and attach a new root-domain
961 * object.
962 *
963 */
964struct root_domain {
965 atomic_t refcount;
966 atomic_t rto_count;
967 struct rcu_head rcu;
968 cpumask_var_t span;
969 cpumask_var_t online;
970
971 /*
972 * Indicate pullable load on at least one CPU, e.g:
973 * - More than one runnable task
974 * - Running task is misfit
975 */
976 bool overloaded;
977
978 /* Indicate one or more CPUs over-utilized (tipping point) */
979 bool overutilized;
980
981 /*
982 * The bit corresponding to a CPU gets set here if such CPU has more
983 * than one runnable -deadline task (as it is below for RT tasks).
984 */
985 cpumask_var_t dlo_mask;
986 atomic_t dlo_count;
987 struct dl_bw dl_bw;
988 struct cpudl cpudl;
989
990 /*
991 * Indicate whether a root_domain's dl_bw has been checked or
992 * updated. It's monotonously increasing value.
993 *
994 * Also, some corner cases, like 'wrap around' is dangerous, but given
995 * that u64 is 'big enough'. So that shouldn't be a concern.
996 */
997 u64 visit_cookie;
998
999#ifdef HAVE_RT_PUSH_IPI
1000 /*
1001 * For IPI pull requests, loop across the rto_mask.
1002 */
1003 struct irq_work rto_push_work;
1004 raw_spinlock_t rto_lock;
1005 /* These are only updated and read within rto_lock */
1006 int rto_loop;
1007 int rto_cpu;
1008 /* These atomics are updated outside of a lock */
1009 atomic_t rto_loop_next;
1010 atomic_t rto_loop_start;
1011#endif
1012 /*
1013 * The "RT overload" flag: it gets set if a CPU has more than
1014 * one runnable RT task.
1015 */
1016 cpumask_var_t rto_mask;
1017 struct cpupri cpupri;
1018
1019 /*
1020 * NULL-terminated list of performance domains intersecting with the
1021 * CPUs of the rd. Protected by RCU.
1022 */
1023 struct perf_domain __rcu *pd;
1024};
1025
1026extern void init_defrootdomain(void);
1027extern int sched_init_domains(const struct cpumask *cpu_map);
1028extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1029extern void sched_get_rd(struct root_domain *rd);
1030extern void sched_put_rd(struct root_domain *rd);
1031
1032static inline int get_rd_overloaded(struct root_domain *rd)
1033{
1034 return READ_ONCE(rd->overloaded);
1035}
1036
1037static inline void set_rd_overloaded(struct root_domain *rd, int status)
1038{
1039 if (get_rd_overloaded(rd) != status)
1040 WRITE_ONCE(rd->overloaded, status);
1041}
1042
1043#ifdef HAVE_RT_PUSH_IPI
1044extern void rto_push_irq_work_func(struct irq_work *work);
1045#endif
1046#endif /* CONFIG_SMP */
1047
1048#ifdef CONFIG_UCLAMP_TASK
1049/*
1050 * struct uclamp_bucket - Utilization clamp bucket
1051 * @value: utilization clamp value for tasks on this clamp bucket
1052 * @tasks: number of RUNNABLE tasks on this clamp bucket
1053 *
1054 * Keep track of how many tasks are RUNNABLE for a given utilization
1055 * clamp value.
1056 */
1057struct uclamp_bucket {
1058 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1059 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1060};
1061
1062/*
1063 * struct uclamp_rq - rq's utilization clamp
1064 * @value: currently active clamp values for a rq
1065 * @bucket: utilization clamp buckets affecting a rq
1066 *
1067 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1068 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1069 * (or actually running) with that value.
1070 *
1071 * There are up to UCLAMP_CNT possible different clamp values, currently there
1072 * are only two: minimum utilization and maximum utilization.
1073 *
1074 * All utilization clamping values are MAX aggregated, since:
1075 * - for util_min: we want to run the CPU at least at the max of the minimum
1076 * utilization required by its currently RUNNABLE tasks.
1077 * - for util_max: we want to allow the CPU to run up to the max of the
1078 * maximum utilization allowed by its currently RUNNABLE tasks.
1079 *
1080 * Since on each system we expect only a limited number of different
1081 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1082 * the metrics required to compute all the per-rq utilization clamp values.
1083 */
1084struct uclamp_rq {
1085 unsigned int value;
1086 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1087};
1088
1089DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1090#endif /* CONFIG_UCLAMP_TASK */
1091
1092/*
1093 * This is the main, per-CPU runqueue data structure.
1094 *
1095 * Locking rule: those places that want to lock multiple runqueues
1096 * (such as the load balancing or the thread migration code), lock
1097 * acquire operations must be ordered by ascending &runqueue.
1098 */
1099struct rq {
1100 /* runqueue lock: */
1101 raw_spinlock_t __lock;
1102
1103 unsigned int nr_running;
1104#ifdef CONFIG_NUMA_BALANCING
1105 unsigned int nr_numa_running;
1106 unsigned int nr_preferred_running;
1107 unsigned int numa_migrate_on;
1108#endif
1109#ifdef CONFIG_NO_HZ_COMMON
1110#ifdef CONFIG_SMP
1111 unsigned long last_blocked_load_update_tick;
1112 unsigned int has_blocked_load;
1113 call_single_data_t nohz_csd;
1114#endif /* CONFIG_SMP */
1115 unsigned int nohz_tick_stopped;
1116 atomic_t nohz_flags;
1117#endif /* CONFIG_NO_HZ_COMMON */
1118
1119#ifdef CONFIG_SMP
1120 unsigned int ttwu_pending;
1121#endif
1122 u64 nr_switches;
1123
1124#ifdef CONFIG_UCLAMP_TASK
1125 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1126 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1127 unsigned int uclamp_flags;
1128#define UCLAMP_FLAG_IDLE 0x01
1129#endif
1130
1131 struct cfs_rq cfs;
1132 struct rt_rq rt;
1133 struct dl_rq dl;
1134#ifdef CONFIG_SCHED_CLASS_EXT
1135 struct scx_rq scx;
1136#endif
1137
1138 struct sched_dl_entity fair_server;
1139
1140#ifdef CONFIG_FAIR_GROUP_SCHED
1141 /* list of leaf cfs_rq on this CPU: */
1142 struct list_head leaf_cfs_rq_list;
1143 struct list_head *tmp_alone_branch;
1144#endif /* CONFIG_FAIR_GROUP_SCHED */
1145
1146 /*
1147 * This is part of a global counter where only the total sum
1148 * over all CPUs matters. A task can increase this counter on
1149 * one CPU and if it got migrated afterwards it may decrease
1150 * it on another CPU. Always updated under the runqueue lock:
1151 */
1152 unsigned int nr_uninterruptible;
1153
1154 union {
1155 struct task_struct __rcu *donor; /* Scheduler context */
1156 struct task_struct __rcu *curr; /* Execution context */
1157 };
1158 struct sched_dl_entity *dl_server;
1159 struct task_struct *idle;
1160 struct task_struct *stop;
1161 unsigned long next_balance;
1162 struct mm_struct *prev_mm;
1163
1164 unsigned int clock_update_flags;
1165 u64 clock;
1166 /* Ensure that all clocks are in the same cache line */
1167 u64 clock_task ____cacheline_aligned;
1168 u64 clock_pelt;
1169 unsigned long lost_idle_time;
1170 u64 clock_pelt_idle;
1171 u64 clock_idle;
1172#ifndef CONFIG_64BIT
1173 u64 clock_pelt_idle_copy;
1174 u64 clock_idle_copy;
1175#endif
1176
1177 atomic_t nr_iowait;
1178
1179 u64 last_seen_need_resched_ns;
1180 int ticks_without_resched;
1181
1182#ifdef CONFIG_MEMBARRIER
1183 int membarrier_state;
1184#endif
1185
1186#ifdef CONFIG_SMP
1187 struct root_domain *rd;
1188 struct sched_domain __rcu *sd;
1189
1190 unsigned long cpu_capacity;
1191
1192 struct balance_callback *balance_callback;
1193
1194 unsigned char nohz_idle_balance;
1195 unsigned char idle_balance;
1196
1197 unsigned long misfit_task_load;
1198
1199 /* For active balancing */
1200 int active_balance;
1201 int push_cpu;
1202 struct cpu_stop_work active_balance_work;
1203
1204 /* CPU of this runqueue: */
1205 int cpu;
1206 int online;
1207
1208 struct list_head cfs_tasks;
1209
1210 struct sched_avg avg_rt;
1211 struct sched_avg avg_dl;
1212#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1213 struct sched_avg avg_irq;
1214#endif
1215#ifdef CONFIG_SCHED_HW_PRESSURE
1216 struct sched_avg avg_hw;
1217#endif
1218 u64 idle_stamp;
1219 u64 avg_idle;
1220
1221 /* This is used to determine avg_idle's max value */
1222 u64 max_idle_balance_cost;
1223
1224#ifdef CONFIG_HOTPLUG_CPU
1225 struct rcuwait hotplug_wait;
1226#endif
1227#endif /* CONFIG_SMP */
1228
1229#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1230 u64 prev_irq_time;
1231 u64 psi_irq_time;
1232#endif
1233#ifdef CONFIG_PARAVIRT
1234 u64 prev_steal_time;
1235#endif
1236#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1237 u64 prev_steal_time_rq;
1238#endif
1239
1240 /* calc_load related fields */
1241 unsigned long calc_load_update;
1242 long calc_load_active;
1243
1244#ifdef CONFIG_SCHED_HRTICK
1245#ifdef CONFIG_SMP
1246 call_single_data_t hrtick_csd;
1247#endif
1248 struct hrtimer hrtick_timer;
1249 ktime_t hrtick_time;
1250#endif
1251
1252#ifdef CONFIG_SCHEDSTATS
1253 /* latency stats */
1254 struct sched_info rq_sched_info;
1255 unsigned long long rq_cpu_time;
1256
1257 /* sys_sched_yield() stats */
1258 unsigned int yld_count;
1259
1260 /* schedule() stats */
1261 unsigned int sched_count;
1262 unsigned int sched_goidle;
1263
1264 /* try_to_wake_up() stats */
1265 unsigned int ttwu_count;
1266 unsigned int ttwu_local;
1267#endif
1268
1269#ifdef CONFIG_CPU_IDLE
1270 /* Must be inspected within a RCU lock section */
1271 struct cpuidle_state *idle_state;
1272#endif
1273
1274#ifdef CONFIG_SMP
1275 unsigned int nr_pinned;
1276#endif
1277 unsigned int push_busy;
1278 struct cpu_stop_work push_work;
1279
1280#ifdef CONFIG_SCHED_CORE
1281 /* per rq */
1282 struct rq *core;
1283 struct task_struct *core_pick;
1284 struct sched_dl_entity *core_dl_server;
1285 unsigned int core_enabled;
1286 unsigned int core_sched_seq;
1287 struct rb_root core_tree;
1288
1289 /* shared state -- careful with sched_core_cpu_deactivate() */
1290 unsigned int core_task_seq;
1291 unsigned int core_pick_seq;
1292 unsigned long core_cookie;
1293 unsigned int core_forceidle_count;
1294 unsigned int core_forceidle_seq;
1295 unsigned int core_forceidle_occupation;
1296 u64 core_forceidle_start;
1297#endif
1298
1299 /* Scratch cpumask to be temporarily used under rq_lock */
1300 cpumask_var_t scratch_mask;
1301
1302#if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1303 call_single_data_t cfsb_csd;
1304 struct list_head cfsb_csd_list;
1305#endif
1306};
1307
1308#ifdef CONFIG_FAIR_GROUP_SCHED
1309
1310/* CPU runqueue to which this cfs_rq is attached */
1311static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1312{
1313 return cfs_rq->rq;
1314}
1315
1316#else
1317
1318static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1319{
1320 return container_of(cfs_rq, struct rq, cfs);
1321}
1322#endif
1323
1324static inline int cpu_of(struct rq *rq)
1325{
1326#ifdef CONFIG_SMP
1327 return rq->cpu;
1328#else
1329 return 0;
1330#endif
1331}
1332
1333#define MDF_PUSH 0x01
1334
1335static inline bool is_migration_disabled(struct task_struct *p)
1336{
1337#ifdef CONFIG_SMP
1338 return p->migration_disabled;
1339#else
1340 return false;
1341#endif
1342}
1343
1344DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1345
1346#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1347#define this_rq() this_cpu_ptr(&runqueues)
1348#define task_rq(p) cpu_rq(task_cpu(p))
1349#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1350#define raw_rq() raw_cpu_ptr(&runqueues)
1351
1352static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1353{
1354 /* Do nothing */
1355}
1356
1357#ifdef CONFIG_SCHED_CORE
1358static inline struct cpumask *sched_group_span(struct sched_group *sg);
1359
1360DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1361
1362static inline bool sched_core_enabled(struct rq *rq)
1363{
1364 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1365}
1366
1367static inline bool sched_core_disabled(void)
1368{
1369 return !static_branch_unlikely(&__sched_core_enabled);
1370}
1371
1372/*
1373 * Be careful with this function; not for general use. The return value isn't
1374 * stable unless you actually hold a relevant rq->__lock.
1375 */
1376static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1377{
1378 if (sched_core_enabled(rq))
1379 return &rq->core->__lock;
1380
1381 return &rq->__lock;
1382}
1383
1384static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1385{
1386 if (rq->core_enabled)
1387 return &rq->core->__lock;
1388
1389 return &rq->__lock;
1390}
1391
1392extern bool
1393cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1394
1395extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1396
1397/*
1398 * Helpers to check if the CPU's core cookie matches with the task's cookie
1399 * when core scheduling is enabled.
1400 * A special case is that the task's cookie always matches with CPU's core
1401 * cookie if the CPU is in an idle core.
1402 */
1403static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1404{
1405 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1406 if (!sched_core_enabled(rq))
1407 return true;
1408
1409 return rq->core->core_cookie == p->core_cookie;
1410}
1411
1412static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1413{
1414 bool idle_core = true;
1415 int cpu;
1416
1417 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1418 if (!sched_core_enabled(rq))
1419 return true;
1420
1421 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1422 if (!available_idle_cpu(cpu)) {
1423 idle_core = false;
1424 break;
1425 }
1426 }
1427
1428 /*
1429 * A CPU in an idle core is always the best choice for tasks with
1430 * cookies.
1431 */
1432 return idle_core || rq->core->core_cookie == p->core_cookie;
1433}
1434
1435static inline bool sched_group_cookie_match(struct rq *rq,
1436 struct task_struct *p,
1437 struct sched_group *group)
1438{
1439 int cpu;
1440
1441 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1442 if (!sched_core_enabled(rq))
1443 return true;
1444
1445 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1446 if (sched_core_cookie_match(cpu_rq(cpu), p))
1447 return true;
1448 }
1449 return false;
1450}
1451
1452static inline bool sched_core_enqueued(struct task_struct *p)
1453{
1454 return !RB_EMPTY_NODE(&p->core_node);
1455}
1456
1457extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1458extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1459
1460extern void sched_core_get(void);
1461extern void sched_core_put(void);
1462
1463#else /* !CONFIG_SCHED_CORE: */
1464
1465static inline bool sched_core_enabled(struct rq *rq)
1466{
1467 return false;
1468}
1469
1470static inline bool sched_core_disabled(void)
1471{
1472 return true;
1473}
1474
1475static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1476{
1477 return &rq->__lock;
1478}
1479
1480static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1481{
1482 return &rq->__lock;
1483}
1484
1485static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1486{
1487 return true;
1488}
1489
1490static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1491{
1492 return true;
1493}
1494
1495static inline bool sched_group_cookie_match(struct rq *rq,
1496 struct task_struct *p,
1497 struct sched_group *group)
1498{
1499 return true;
1500}
1501
1502#endif /* !CONFIG_SCHED_CORE */
1503#ifdef CONFIG_RT_GROUP_SCHED
1504# ifdef CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED
1505DECLARE_STATIC_KEY_FALSE(rt_group_sched);
1506static inline bool rt_group_sched_enabled(void)
1507{
1508 return static_branch_unlikely(&rt_group_sched);
1509}
1510# else
1511DECLARE_STATIC_KEY_TRUE(rt_group_sched);
1512static inline bool rt_group_sched_enabled(void)
1513{
1514 return static_branch_likely(&rt_group_sched);
1515}
1516# endif /* CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED */
1517#else
1518# define rt_group_sched_enabled() false
1519#endif /* CONFIG_RT_GROUP_SCHED */
1520
1521static inline void lockdep_assert_rq_held(struct rq *rq)
1522{
1523 lockdep_assert_held(__rq_lockp(rq));
1524}
1525
1526extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1527extern bool raw_spin_rq_trylock(struct rq *rq);
1528extern void raw_spin_rq_unlock(struct rq *rq);
1529
1530static inline void raw_spin_rq_lock(struct rq *rq)
1531{
1532 raw_spin_rq_lock_nested(rq, 0);
1533}
1534
1535static inline void raw_spin_rq_lock_irq(struct rq *rq)
1536{
1537 local_irq_disable();
1538 raw_spin_rq_lock(rq);
1539}
1540
1541static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1542{
1543 raw_spin_rq_unlock(rq);
1544 local_irq_enable();
1545}
1546
1547static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1548{
1549 unsigned long flags;
1550
1551 local_irq_save(flags);
1552 raw_spin_rq_lock(rq);
1553
1554 return flags;
1555}
1556
1557static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1558{
1559 raw_spin_rq_unlock(rq);
1560 local_irq_restore(flags);
1561}
1562
1563#define raw_spin_rq_lock_irqsave(rq, flags) \
1564do { \
1565 flags = _raw_spin_rq_lock_irqsave(rq); \
1566} while (0)
1567
1568#ifdef CONFIG_SCHED_SMT
1569extern void __update_idle_core(struct rq *rq);
1570
1571static inline void update_idle_core(struct rq *rq)
1572{
1573 if (static_branch_unlikely(&sched_smt_present))
1574 __update_idle_core(rq);
1575}
1576
1577#else
1578static inline void update_idle_core(struct rq *rq) { }
1579#endif
1580
1581#ifdef CONFIG_FAIR_GROUP_SCHED
1582
1583static inline struct task_struct *task_of(struct sched_entity *se)
1584{
1585 WARN_ON_ONCE(!entity_is_task(se));
1586 return container_of(se, struct task_struct, se);
1587}
1588
1589static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1590{
1591 return p->se.cfs_rq;
1592}
1593
1594/* runqueue on which this entity is (to be) queued */
1595static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1596{
1597 return se->cfs_rq;
1598}
1599
1600/* runqueue "owned" by this group */
1601static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1602{
1603 return grp->my_q;
1604}
1605
1606#else /* !CONFIG_FAIR_GROUP_SCHED: */
1607
1608#define task_of(_se) container_of(_se, struct task_struct, se)
1609
1610static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1611{
1612 return &task_rq(p)->cfs;
1613}
1614
1615static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1616{
1617 const struct task_struct *p = task_of(se);
1618 struct rq *rq = task_rq(p);
1619
1620 return &rq->cfs;
1621}
1622
1623/* runqueue "owned" by this group */
1624static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1625{
1626 return NULL;
1627}
1628
1629#endif /* !CONFIG_FAIR_GROUP_SCHED */
1630
1631extern void update_rq_clock(struct rq *rq);
1632
1633/*
1634 * rq::clock_update_flags bits
1635 *
1636 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1637 * call to __schedule(). This is an optimisation to avoid
1638 * neighbouring rq clock updates.
1639 *
1640 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1641 * in effect and calls to update_rq_clock() are being ignored.
1642 *
1643 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1644 * made to update_rq_clock() since the last time rq::lock was pinned.
1645 *
1646 * If inside of __schedule(), clock_update_flags will have been
1647 * shifted left (a left shift is a cheap operation for the fast path
1648 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1649 *
1650 * if (rq-clock_update_flags >= RQCF_UPDATED)
1651 *
1652 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1653 * one position though, because the next rq_unpin_lock() will shift it
1654 * back.
1655 */
1656#define RQCF_REQ_SKIP 0x01
1657#define RQCF_ACT_SKIP 0x02
1658#define RQCF_UPDATED 0x04
1659
1660static inline void assert_clock_updated(struct rq *rq)
1661{
1662 /*
1663 * The only reason for not seeing a clock update since the
1664 * last rq_pin_lock() is if we're currently skipping updates.
1665 */
1666 WARN_ON_ONCE(rq->clock_update_flags < RQCF_ACT_SKIP);
1667}
1668
1669static inline u64 rq_clock(struct rq *rq)
1670{
1671 lockdep_assert_rq_held(rq);
1672 assert_clock_updated(rq);
1673
1674 return rq->clock;
1675}
1676
1677static inline u64 rq_clock_task(struct rq *rq)
1678{
1679 lockdep_assert_rq_held(rq);
1680 assert_clock_updated(rq);
1681
1682 return rq->clock_task;
1683}
1684
1685static inline void rq_clock_skip_update(struct rq *rq)
1686{
1687 lockdep_assert_rq_held(rq);
1688 rq->clock_update_flags |= RQCF_REQ_SKIP;
1689}
1690
1691/*
1692 * See rt task throttling, which is the only time a skip
1693 * request is canceled.
1694 */
1695static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1696{
1697 lockdep_assert_rq_held(rq);
1698 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1699}
1700
1701/*
1702 * During cpu offlining and rq wide unthrottling, we can trigger
1703 * an update_rq_clock() for several cfs and rt runqueues (Typically
1704 * when using list_for_each_entry_*)
1705 * rq_clock_start_loop_update() can be called after updating the clock
1706 * once and before iterating over the list to prevent multiple update.
1707 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1708 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1709 */
1710static inline void rq_clock_start_loop_update(struct rq *rq)
1711{
1712 lockdep_assert_rq_held(rq);
1713 WARN_ON_ONCE(rq->clock_update_flags & RQCF_ACT_SKIP);
1714 rq->clock_update_flags |= RQCF_ACT_SKIP;
1715}
1716
1717static inline void rq_clock_stop_loop_update(struct rq *rq)
1718{
1719 lockdep_assert_rq_held(rq);
1720 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1721}
1722
1723struct rq_flags {
1724 unsigned long flags;
1725 struct pin_cookie cookie;
1726 /*
1727 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1728 * current pin context is stashed here in case it needs to be
1729 * restored in rq_repin_lock().
1730 */
1731 unsigned int clock_update_flags;
1732};
1733
1734extern struct balance_callback balance_push_callback;
1735
1736#ifdef CONFIG_SCHED_CLASS_EXT
1737extern const struct sched_class ext_sched_class;
1738
1739DECLARE_STATIC_KEY_FALSE(__scx_enabled); /* SCX BPF scheduler loaded */
1740DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
1741
1742#define scx_enabled() static_branch_unlikely(&__scx_enabled)
1743#define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
1744
1745static inline void scx_rq_clock_update(struct rq *rq, u64 clock)
1746{
1747 if (!scx_enabled())
1748 return;
1749 WRITE_ONCE(rq->scx.clock, clock);
1750 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID);
1751}
1752
1753static inline void scx_rq_clock_invalidate(struct rq *rq)
1754{
1755 if (!scx_enabled())
1756 return;
1757 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID);
1758}
1759
1760#else /* !CONFIG_SCHED_CLASS_EXT */
1761#define scx_enabled() false
1762#define scx_switched_all() false
1763
1764static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {}
1765static inline void scx_rq_clock_invalidate(struct rq *rq) {}
1766#endif /* !CONFIG_SCHED_CLASS_EXT */
1767
1768/*
1769 * Lockdep annotation that avoids accidental unlocks; it's like a
1770 * sticky/continuous lockdep_assert_held().
1771 *
1772 * This avoids code that has access to 'struct rq *rq' (basically everything in
1773 * the scheduler) from accidentally unlocking the rq if they do not also have a
1774 * copy of the (on-stack) 'struct rq_flags rf'.
1775 *
1776 * Also see Documentation/locking/lockdep-design.rst.
1777 */
1778static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1779{
1780 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1781
1782 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1783 rf->clock_update_flags = 0;
1784#ifdef CONFIG_SMP
1785 WARN_ON_ONCE(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1786#endif
1787}
1788
1789static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1790{
1791 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1792 rf->clock_update_flags = RQCF_UPDATED;
1793
1794 scx_rq_clock_invalidate(rq);
1795 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1796}
1797
1798static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1799{
1800 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1801
1802 /*
1803 * Restore the value we stashed in @rf for this pin context.
1804 */
1805 rq->clock_update_flags |= rf->clock_update_flags;
1806}
1807
1808extern
1809struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1810 __acquires(rq->lock);
1811
1812extern
1813struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1814 __acquires(p->pi_lock)
1815 __acquires(rq->lock);
1816
1817static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1818 __releases(rq->lock)
1819{
1820 rq_unpin_lock(rq, rf);
1821 raw_spin_rq_unlock(rq);
1822}
1823
1824static inline void
1825task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1826 __releases(rq->lock)
1827 __releases(p->pi_lock)
1828{
1829 rq_unpin_lock(rq, rf);
1830 raw_spin_rq_unlock(rq);
1831 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1832}
1833
1834DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1835 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1836 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1837 struct rq *rq; struct rq_flags rf)
1838
1839static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1840 __acquires(rq->lock)
1841{
1842 raw_spin_rq_lock_irqsave(rq, rf->flags);
1843 rq_pin_lock(rq, rf);
1844}
1845
1846static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1847 __acquires(rq->lock)
1848{
1849 raw_spin_rq_lock_irq(rq);
1850 rq_pin_lock(rq, rf);
1851}
1852
1853static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1854 __acquires(rq->lock)
1855{
1856 raw_spin_rq_lock(rq);
1857 rq_pin_lock(rq, rf);
1858}
1859
1860static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1861 __releases(rq->lock)
1862{
1863 rq_unpin_lock(rq, rf);
1864 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1865}
1866
1867static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1868 __releases(rq->lock)
1869{
1870 rq_unpin_lock(rq, rf);
1871 raw_spin_rq_unlock_irq(rq);
1872}
1873
1874static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1875 __releases(rq->lock)
1876{
1877 rq_unpin_lock(rq, rf);
1878 raw_spin_rq_unlock(rq);
1879}
1880
1881DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1882 rq_lock(_T->lock, &_T->rf),
1883 rq_unlock(_T->lock, &_T->rf),
1884 struct rq_flags rf)
1885
1886DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1887 rq_lock_irq(_T->lock, &_T->rf),
1888 rq_unlock_irq(_T->lock, &_T->rf),
1889 struct rq_flags rf)
1890
1891DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1892 rq_lock_irqsave(_T->lock, &_T->rf),
1893 rq_unlock_irqrestore(_T->lock, &_T->rf),
1894 struct rq_flags rf)
1895
1896static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1897 __acquires(rq->lock)
1898{
1899 struct rq *rq;
1900
1901 local_irq_disable();
1902 rq = this_rq();
1903 rq_lock(rq, rf);
1904
1905 return rq;
1906}
1907
1908#ifdef CONFIG_NUMA
1909
1910enum numa_topology_type {
1911 NUMA_DIRECT,
1912 NUMA_GLUELESS_MESH,
1913 NUMA_BACKPLANE,
1914};
1915
1916extern enum numa_topology_type sched_numa_topology_type;
1917extern int sched_max_numa_distance;
1918extern bool find_numa_distance(int distance);
1919extern void sched_init_numa(int offline_node);
1920extern void sched_update_numa(int cpu, bool online);
1921extern void sched_domains_numa_masks_set(unsigned int cpu);
1922extern void sched_domains_numa_masks_clear(unsigned int cpu);
1923extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1924
1925#else /* !CONFIG_NUMA: */
1926
1927static inline void sched_init_numa(int offline_node) { }
1928static inline void sched_update_numa(int cpu, bool online) { }
1929static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1930static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1931
1932static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1933{
1934 return nr_cpu_ids;
1935}
1936
1937#endif /* !CONFIG_NUMA */
1938
1939#ifdef CONFIG_NUMA_BALANCING
1940
1941/* The regions in numa_faults array from task_struct */
1942enum numa_faults_stats {
1943 NUMA_MEM = 0,
1944 NUMA_CPU,
1945 NUMA_MEMBUF,
1946 NUMA_CPUBUF
1947};
1948
1949extern void sched_setnuma(struct task_struct *p, int node);
1950extern int migrate_task_to(struct task_struct *p, int cpu);
1951extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1952 int cpu, int scpu);
1953extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1954
1955#else /* !CONFIG_NUMA_BALANCING: */
1956
1957static inline void
1958init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1959{
1960}
1961
1962#endif /* !CONFIG_NUMA_BALANCING */
1963
1964#ifdef CONFIG_SMP
1965
1966static inline void
1967queue_balance_callback(struct rq *rq,
1968 struct balance_callback *head,
1969 void (*func)(struct rq *rq))
1970{
1971 lockdep_assert_rq_held(rq);
1972
1973 /*
1974 * Don't (re)queue an already queued item; nor queue anything when
1975 * balance_push() is active, see the comment with
1976 * balance_push_callback.
1977 */
1978 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1979 return;
1980
1981 head->func = func;
1982 head->next = rq->balance_callback;
1983 rq->balance_callback = head;
1984}
1985
1986#define rcu_dereference_check_sched_domain(p) \
1987 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1988
1989/*
1990 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1991 * See destroy_sched_domains: call_rcu for details.
1992 *
1993 * The domain tree of any CPU may only be accessed from within
1994 * preempt-disabled sections.
1995 */
1996#define for_each_domain(cpu, __sd) \
1997 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1998 __sd; __sd = __sd->parent)
1999
2000/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
2001#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
2002static const unsigned int SD_SHARED_CHILD_MASK =
2003#include <linux/sched/sd_flags.h>
20040;
2005#undef SD_FLAG
2006
2007/**
2008 * highest_flag_domain - Return highest sched_domain containing flag.
2009 * @cpu: The CPU whose highest level of sched domain is to
2010 * be returned.
2011 * @flag: The flag to check for the highest sched_domain
2012 * for the given CPU.
2013 *
2014 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
2015 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
2016 */
2017static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
2018{
2019 struct sched_domain *sd, *hsd = NULL;
2020
2021 for_each_domain(cpu, sd) {
2022 if (sd->flags & flag) {
2023 hsd = sd;
2024 continue;
2025 }
2026
2027 /*
2028 * Stop the search if @flag is known to be shared at lower
2029 * levels. It will not be found further up.
2030 */
2031 if (flag & SD_SHARED_CHILD_MASK)
2032 break;
2033 }
2034
2035 return hsd;
2036}
2037
2038static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2039{
2040 struct sched_domain *sd;
2041
2042 for_each_domain(cpu, sd) {
2043 if (sd->flags & flag)
2044 break;
2045 }
2046
2047 return sd;
2048}
2049
2050DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2051DECLARE_PER_CPU(int, sd_llc_size);
2052DECLARE_PER_CPU(int, sd_llc_id);
2053DECLARE_PER_CPU(int, sd_share_id);
2054DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2055DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2056DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2057DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2058
2059extern struct static_key_false sched_asym_cpucapacity;
2060extern struct static_key_false sched_cluster_active;
2061
2062static __always_inline bool sched_asym_cpucap_active(void)
2063{
2064 return static_branch_unlikely(&sched_asym_cpucapacity);
2065}
2066
2067struct sched_group_capacity {
2068 atomic_t ref;
2069 /*
2070 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2071 * for a single CPU.
2072 */
2073 unsigned long capacity;
2074 unsigned long min_capacity; /* Min per-CPU capacity in group */
2075 unsigned long max_capacity; /* Max per-CPU capacity in group */
2076 unsigned long next_update;
2077 int imbalance; /* XXX unrelated to capacity but shared group state */
2078
2079 int id;
2080
2081 unsigned long cpumask[]; /* Balance mask */
2082};
2083
2084struct sched_group {
2085 struct sched_group *next; /* Must be a circular list */
2086 atomic_t ref;
2087
2088 unsigned int group_weight;
2089 unsigned int cores;
2090 struct sched_group_capacity *sgc;
2091 int asym_prefer_cpu; /* CPU of highest priority in group */
2092 int flags;
2093
2094 /*
2095 * The CPUs this group covers.
2096 *
2097 * NOTE: this field is variable length. (Allocated dynamically
2098 * by attaching extra space to the end of the structure,
2099 * depending on how many CPUs the kernel has booted up with)
2100 */
2101 unsigned long cpumask[];
2102};
2103
2104static inline struct cpumask *sched_group_span(struct sched_group *sg)
2105{
2106 return to_cpumask(sg->cpumask);
2107}
2108
2109/*
2110 * See build_balance_mask().
2111 */
2112static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2113{
2114 return to_cpumask(sg->sgc->cpumask);
2115}
2116
2117extern int group_balance_cpu(struct sched_group *sg);
2118
2119extern void update_sched_domain_debugfs(void);
2120extern void dirty_sched_domain_sysctl(int cpu);
2121
2122extern int sched_update_scaling(void);
2123
2124static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2125{
2126 if (!p->user_cpus_ptr)
2127 return cpu_possible_mask; /* &init_task.cpus_mask */
2128 return p->user_cpus_ptr;
2129}
2130
2131#endif /* CONFIG_SMP */
2132
2133#ifdef CONFIG_CGROUP_SCHED
2134
2135/*
2136 * Return the group to which this tasks belongs.
2137 *
2138 * We cannot use task_css() and friends because the cgroup subsystem
2139 * changes that value before the cgroup_subsys::attach() method is called,
2140 * therefore we cannot pin it and might observe the wrong value.
2141 *
2142 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2143 * core changes this before calling sched_move_task().
2144 *
2145 * Instead we use a 'copy' which is updated from sched_move_task() while
2146 * holding both task_struct::pi_lock and rq::lock.
2147 */
2148static inline struct task_group *task_group(struct task_struct *p)
2149{
2150 return p->sched_task_group;
2151}
2152
2153/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2154static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2155{
2156#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2157 struct task_group *tg = task_group(p);
2158#endif
2159
2160#ifdef CONFIG_FAIR_GROUP_SCHED
2161 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2162 p->se.cfs_rq = tg->cfs_rq[cpu];
2163 p->se.parent = tg->se[cpu];
2164 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2165#endif
2166
2167#ifdef CONFIG_RT_GROUP_SCHED
2168 /*
2169 * p->rt.rt_rq is NULL initially and it is easier to assign
2170 * root_task_group's rt_rq than switching in rt_rq_of_se()
2171 * Clobbers tg(!)
2172 */
2173 if (!rt_group_sched_enabled())
2174 tg = &root_task_group;
2175 p->rt.rt_rq = tg->rt_rq[cpu];
2176 p->rt.parent = tg->rt_se[cpu];
2177#endif
2178}
2179
2180#else /* !CONFIG_CGROUP_SCHED: */
2181
2182static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2183
2184static inline struct task_group *task_group(struct task_struct *p)
2185{
2186 return NULL;
2187}
2188
2189#endif /* !CONFIG_CGROUP_SCHED */
2190
2191static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2192{
2193 set_task_rq(p, cpu);
2194#ifdef CONFIG_SMP
2195 /*
2196 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2197 * successfully executed on another CPU. We must ensure that updates of
2198 * per-task data have been completed by this moment.
2199 */
2200 smp_wmb();
2201 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2202 p->wake_cpu = cpu;
2203#endif
2204}
2205
2206/*
2207 * Tunables:
2208 */
2209
2210#define SCHED_FEAT(name, enabled) \
2211 __SCHED_FEAT_##name ,
2212
2213enum {
2214#include "features.h"
2215 __SCHED_FEAT_NR,
2216};
2217
2218#undef SCHED_FEAT
2219
2220/*
2221 * To support run-time toggling of sched features, all the translation units
2222 * (but core.c) reference the sysctl_sched_features defined in core.c.
2223 */
2224extern __read_mostly unsigned int sysctl_sched_features;
2225
2226#ifdef CONFIG_JUMP_LABEL
2227
2228#define SCHED_FEAT(name, enabled) \
2229static __always_inline bool static_branch_##name(struct static_key *key) \
2230{ \
2231 return static_key_##enabled(key); \
2232}
2233
2234#include "features.h"
2235#undef SCHED_FEAT
2236
2237extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2238#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2239
2240#else /* !CONFIG_JUMP_LABEL: */
2241
2242#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2243
2244#endif /* !CONFIG_JUMP_LABEL */
2245
2246extern struct static_key_false sched_numa_balancing;
2247extern struct static_key_false sched_schedstats;
2248
2249static inline u64 global_rt_period(void)
2250{
2251 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2252}
2253
2254static inline u64 global_rt_runtime(void)
2255{
2256 if (sysctl_sched_rt_runtime < 0)
2257 return RUNTIME_INF;
2258
2259 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2260}
2261
2262/*
2263 * Is p the current execution context?
2264 */
2265static inline int task_current(struct rq *rq, struct task_struct *p)
2266{
2267 return rq->curr == p;
2268}
2269
2270/*
2271 * Is p the current scheduling context?
2272 *
2273 * Note that it might be the current execution context at the same time if
2274 * rq->curr == rq->donor == p.
2275 */
2276static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2277{
2278 return rq->donor == p;
2279}
2280
2281static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2282{
2283#ifdef CONFIG_SMP
2284 return p->on_cpu;
2285#else
2286 return task_current(rq, p);
2287#endif
2288}
2289
2290static inline int task_on_rq_queued(struct task_struct *p)
2291{
2292 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED;
2293}
2294
2295static inline int task_on_rq_migrating(struct task_struct *p)
2296{
2297 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2298}
2299
2300/* Wake flags. The first three directly map to some SD flag value */
2301#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2302#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2303#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2304
2305#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2306#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2307#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2308#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2309
2310#ifdef CONFIG_SMP
2311static_assert(WF_EXEC == SD_BALANCE_EXEC);
2312static_assert(WF_FORK == SD_BALANCE_FORK);
2313static_assert(WF_TTWU == SD_BALANCE_WAKE);
2314#endif
2315
2316/*
2317 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2318 * of tasks with abnormal "nice" values across CPUs the contribution that
2319 * each task makes to its run queue's load is weighted according to its
2320 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2321 * scaled version of the new time slice allocation that they receive on time
2322 * slice expiry etc.
2323 */
2324
2325#define WEIGHT_IDLEPRIO 3
2326#define WMULT_IDLEPRIO 1431655765
2327
2328extern const int sched_prio_to_weight[40];
2329extern const u32 sched_prio_to_wmult[40];
2330
2331/*
2332 * {de,en}queue flags:
2333 *
2334 * DEQUEUE_SLEEP - task is no longer runnable
2335 * ENQUEUE_WAKEUP - task just became runnable
2336 *
2337 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2338 * are in a known state which allows modification. Such pairs
2339 * should preserve as much state as possible.
2340 *
2341 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2342 * in the runqueue.
2343 *
2344 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2345 *
2346 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2347 *
2348 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2349 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2350 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2351 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2352 *
2353 */
2354
2355#define DEQUEUE_SLEEP 0x01 /* Matches ENQUEUE_WAKEUP */
2356#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2357#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2358#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2359#define DEQUEUE_SPECIAL 0x10
2360#define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2361#define DEQUEUE_DELAYED 0x200 /* Matches ENQUEUE_DELAYED */
2362
2363#define ENQUEUE_WAKEUP 0x01
2364#define ENQUEUE_RESTORE 0x02
2365#define ENQUEUE_MOVE 0x04
2366#define ENQUEUE_NOCLOCK 0x08
2367
2368#define ENQUEUE_HEAD 0x10
2369#define ENQUEUE_REPLENISH 0x20
2370#ifdef CONFIG_SMP
2371#define ENQUEUE_MIGRATED 0x40
2372#else
2373#define ENQUEUE_MIGRATED 0x00
2374#endif
2375#define ENQUEUE_INITIAL 0x80
2376#define ENQUEUE_MIGRATING 0x100
2377#define ENQUEUE_DELAYED 0x200
2378#define ENQUEUE_RQ_SELECTED 0x400
2379
2380#define RETRY_TASK ((void *)-1UL)
2381
2382struct affinity_context {
2383 const struct cpumask *new_mask;
2384 struct cpumask *user_mask;
2385 unsigned int flags;
2386};
2387
2388extern s64 update_curr_common(struct rq *rq);
2389
2390struct sched_class {
2391
2392#ifdef CONFIG_UCLAMP_TASK
2393 int uclamp_enabled;
2394#endif
2395
2396 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2397 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2398 void (*yield_task) (struct rq *rq);
2399 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2400
2401 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2402
2403 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2404 struct task_struct *(*pick_task)(struct rq *rq);
2405 /*
2406 * Optional! When implemented pick_next_task() should be equivalent to:
2407 *
2408 * next = pick_task();
2409 * if (next) {
2410 * put_prev_task(prev);
2411 * set_next_task_first(next);
2412 * }
2413 */
2414 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev);
2415
2416 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2417 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2418
2419#ifdef CONFIG_SMP
2420 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2421
2422 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2423
2424 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2425
2426 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2427
2428 void (*rq_online)(struct rq *rq);
2429 void (*rq_offline)(struct rq *rq);
2430
2431 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2432#endif
2433
2434 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2435 void (*task_fork)(struct task_struct *p);
2436 void (*task_dead)(struct task_struct *p);
2437
2438 /*
2439 * The switched_from() call is allowed to drop rq->lock, therefore we
2440 * cannot assume the switched_from/switched_to pair is serialized by
2441 * rq->lock. They are however serialized by p->pi_lock.
2442 */
2443 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2444 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2445 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2446 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2447 const struct load_weight *lw);
2448 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2449 int oldprio);
2450
2451 unsigned int (*get_rr_interval)(struct rq *rq,
2452 struct task_struct *task);
2453
2454 void (*update_curr)(struct rq *rq);
2455
2456#ifdef CONFIG_FAIR_GROUP_SCHED
2457 void (*task_change_group)(struct task_struct *p);
2458#endif
2459
2460#ifdef CONFIG_SCHED_CORE
2461 int (*task_is_throttled)(struct task_struct *p, int cpu);
2462#endif
2463};
2464
2465static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2466{
2467 WARN_ON_ONCE(rq->donor != prev);
2468 prev->sched_class->put_prev_task(rq, prev, NULL);
2469}
2470
2471static inline void set_next_task(struct rq *rq, struct task_struct *next)
2472{
2473 next->sched_class->set_next_task(rq, next, false);
2474}
2475
2476static inline void
2477__put_prev_set_next_dl_server(struct rq *rq,
2478 struct task_struct *prev,
2479 struct task_struct *next)
2480{
2481 prev->dl_server = NULL;
2482 next->dl_server = rq->dl_server;
2483 rq->dl_server = NULL;
2484}
2485
2486static inline void put_prev_set_next_task(struct rq *rq,
2487 struct task_struct *prev,
2488 struct task_struct *next)
2489{
2490 WARN_ON_ONCE(rq->curr != prev);
2491
2492 __put_prev_set_next_dl_server(rq, prev, next);
2493
2494 if (next == prev)
2495 return;
2496
2497 prev->sched_class->put_prev_task(rq, prev, next);
2498 next->sched_class->set_next_task(rq, next, true);
2499}
2500
2501/*
2502 * Helper to define a sched_class instance; each one is placed in a separate
2503 * section which is ordered by the linker script:
2504 *
2505 * include/asm-generic/vmlinux.lds.h
2506 *
2507 * *CAREFUL* they are laid out in *REVERSE* order!!!
2508 *
2509 * Also enforce alignment on the instance, not the type, to guarantee layout.
2510 */
2511#define DEFINE_SCHED_CLASS(name) \
2512const struct sched_class name##_sched_class \
2513 __aligned(__alignof__(struct sched_class)) \
2514 __section("__" #name "_sched_class")
2515
2516/* Defined in include/asm-generic/vmlinux.lds.h */
2517extern struct sched_class __sched_class_highest[];
2518extern struct sched_class __sched_class_lowest[];
2519
2520extern const struct sched_class stop_sched_class;
2521extern const struct sched_class dl_sched_class;
2522extern const struct sched_class rt_sched_class;
2523extern const struct sched_class fair_sched_class;
2524extern const struct sched_class idle_sched_class;
2525
2526/*
2527 * Iterate only active classes. SCX can take over all fair tasks or be
2528 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2529 */
2530static inline const struct sched_class *next_active_class(const struct sched_class *class)
2531{
2532 class++;
2533#ifdef CONFIG_SCHED_CLASS_EXT
2534 if (scx_switched_all() && class == &fair_sched_class)
2535 class++;
2536 if (!scx_enabled() && class == &ext_sched_class)
2537 class++;
2538#endif
2539 return class;
2540}
2541
2542#define for_class_range(class, _from, _to) \
2543 for (class = (_from); class < (_to); class++)
2544
2545#define for_each_class(class) \
2546 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2547
2548#define for_active_class_range(class, _from, _to) \
2549 for (class = (_from); class != (_to); class = next_active_class(class))
2550
2551#define for_each_active_class(class) \
2552 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2553
2554#define sched_class_above(_a, _b) ((_a) < (_b))
2555
2556static inline bool sched_stop_runnable(struct rq *rq)
2557{
2558 return rq->stop && task_on_rq_queued(rq->stop);
2559}
2560
2561static inline bool sched_dl_runnable(struct rq *rq)
2562{
2563 return rq->dl.dl_nr_running > 0;
2564}
2565
2566static inline bool sched_rt_runnable(struct rq *rq)
2567{
2568 return rq->rt.rt_queued > 0;
2569}
2570
2571static inline bool sched_fair_runnable(struct rq *rq)
2572{
2573 return rq->cfs.nr_queued > 0;
2574}
2575
2576extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2577extern struct task_struct *pick_task_idle(struct rq *rq);
2578
2579#define SCA_CHECK 0x01
2580#define SCA_MIGRATE_DISABLE 0x02
2581#define SCA_MIGRATE_ENABLE 0x04
2582#define SCA_USER 0x08
2583
2584#ifdef CONFIG_SMP
2585
2586extern void update_group_capacity(struct sched_domain *sd, int cpu);
2587
2588extern void sched_balance_trigger(struct rq *rq);
2589
2590extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2591extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2592
2593static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2594{
2595 /* When not in the task's cpumask, no point in looking further. */
2596 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2597 return false;
2598
2599 /* Can @cpu run a user thread? */
2600 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2601 return false;
2602
2603 return true;
2604}
2605
2606static inline cpumask_t *alloc_user_cpus_ptr(int node)
2607{
2608 /*
2609 * See do_set_cpus_allowed() above for the rcu_head usage.
2610 */
2611 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2612
2613 return kmalloc_node(size, GFP_KERNEL, node);
2614}
2615
2616static inline struct task_struct *get_push_task(struct rq *rq)
2617{
2618 struct task_struct *p = rq->donor;
2619
2620 lockdep_assert_rq_held(rq);
2621
2622 if (rq->push_busy)
2623 return NULL;
2624
2625 if (p->nr_cpus_allowed == 1)
2626 return NULL;
2627
2628 if (p->migration_disabled)
2629 return NULL;
2630
2631 rq->push_busy = true;
2632 return get_task_struct(p);
2633}
2634
2635extern int push_cpu_stop(void *arg);
2636
2637#else /* !CONFIG_SMP: */
2638
2639static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2640{
2641 return true;
2642}
2643
2644static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2645 struct affinity_context *ctx)
2646{
2647 return set_cpus_allowed_ptr(p, ctx->new_mask);
2648}
2649
2650static inline cpumask_t *alloc_user_cpus_ptr(int node)
2651{
2652 return NULL;
2653}
2654
2655#endif /* !CONFIG_SMP */
2656
2657#ifdef CONFIG_CPU_IDLE
2658
2659static inline void idle_set_state(struct rq *rq,
2660 struct cpuidle_state *idle_state)
2661{
2662 rq->idle_state = idle_state;
2663}
2664
2665static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2666{
2667 WARN_ON_ONCE(!rcu_read_lock_held());
2668
2669 return rq->idle_state;
2670}
2671
2672#else /* !CONFIG_CPU_IDLE: */
2673
2674static inline void idle_set_state(struct rq *rq,
2675 struct cpuidle_state *idle_state)
2676{
2677}
2678
2679static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2680{
2681 return NULL;
2682}
2683
2684#endif /* !CONFIG_CPU_IDLE */
2685
2686extern void schedule_idle(void);
2687asmlinkage void schedule_user(void);
2688
2689extern void sysrq_sched_debug_show(void);
2690extern void sched_init_granularity(void);
2691extern void update_max_interval(void);
2692
2693extern void init_sched_dl_class(void);
2694extern void init_sched_rt_class(void);
2695extern void init_sched_fair_class(void);
2696
2697extern void resched_curr(struct rq *rq);
2698extern void resched_curr_lazy(struct rq *rq);
2699extern void resched_cpu(int cpu);
2700
2701extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2702extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2703
2704extern void init_dl_entity(struct sched_dl_entity *dl_se);
2705
2706#define BW_SHIFT 20
2707#define BW_UNIT (1 << BW_SHIFT)
2708#define RATIO_SHIFT 8
2709#define MAX_BW_BITS (64 - BW_SHIFT)
2710#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2711
2712extern unsigned long to_ratio(u64 period, u64 runtime);
2713
2714extern void init_entity_runnable_average(struct sched_entity *se);
2715extern void post_init_entity_util_avg(struct task_struct *p);
2716
2717#ifdef CONFIG_NO_HZ_FULL
2718extern bool sched_can_stop_tick(struct rq *rq);
2719extern int __init sched_tick_offload_init(void);
2720
2721/*
2722 * Tick may be needed by tasks in the runqueue depending on their policy and
2723 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2724 * nohz mode if necessary.
2725 */
2726static inline void sched_update_tick_dependency(struct rq *rq)
2727{
2728 int cpu = cpu_of(rq);
2729
2730 if (!tick_nohz_full_cpu(cpu))
2731 return;
2732
2733 if (sched_can_stop_tick(rq))
2734 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2735 else
2736 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2737}
2738#else /* !CONFIG_NO_HZ_FULL: */
2739static inline int sched_tick_offload_init(void) { return 0; }
2740static inline void sched_update_tick_dependency(struct rq *rq) { }
2741#endif /* !CONFIG_NO_HZ_FULL */
2742
2743static inline void add_nr_running(struct rq *rq, unsigned count)
2744{
2745 unsigned prev_nr = rq->nr_running;
2746
2747 rq->nr_running = prev_nr + count;
2748 if (trace_sched_update_nr_running_tp_enabled()) {
2749 call_trace_sched_update_nr_running(rq, count);
2750 }
2751
2752#ifdef CONFIG_SMP
2753 if (prev_nr < 2 && rq->nr_running >= 2)
2754 set_rd_overloaded(rq->rd, 1);
2755#endif
2756
2757 sched_update_tick_dependency(rq);
2758}
2759
2760static inline void sub_nr_running(struct rq *rq, unsigned count)
2761{
2762 rq->nr_running -= count;
2763 if (trace_sched_update_nr_running_tp_enabled()) {
2764 call_trace_sched_update_nr_running(rq, -count);
2765 }
2766
2767 /* Check if we still need preemption */
2768 sched_update_tick_dependency(rq);
2769}
2770
2771static inline void __block_task(struct rq *rq, struct task_struct *p)
2772{
2773 if (p->sched_contributes_to_load)
2774 rq->nr_uninterruptible++;
2775
2776 if (p->in_iowait) {
2777 atomic_inc(&rq->nr_iowait);
2778 delayacct_blkio_start();
2779 }
2780
2781 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2782
2783 /*
2784 * The moment this write goes through, ttwu() can swoop in and migrate
2785 * this task, rendering our rq->__lock ineffective.
2786 *
2787 * __schedule() try_to_wake_up()
2788 * LOCK rq->__lock LOCK p->pi_lock
2789 * pick_next_task()
2790 * pick_next_task_fair()
2791 * pick_next_entity()
2792 * dequeue_entities()
2793 * __block_task()
2794 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2795 * break;
2796 *
2797 * ACQUIRE (after ctrl-dep)
2798 *
2799 * cpu = select_task_rq();
2800 * set_task_cpu(p, cpu);
2801 * ttwu_queue()
2802 * ttwu_do_activate()
2803 * LOCK rq->__lock
2804 * activate_task()
2805 * STORE p->on_rq = 1
2806 * UNLOCK rq->__lock
2807 *
2808 * Callers must ensure to not reference @p after this -- we no longer
2809 * own it.
2810 */
2811 smp_store_release(&p->on_rq, 0);
2812}
2813
2814extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2815extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2816
2817extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2818
2819#ifdef CONFIG_PREEMPT_RT
2820# define SCHED_NR_MIGRATE_BREAK 8
2821#else
2822# define SCHED_NR_MIGRATE_BREAK 32
2823#endif
2824
2825extern __read_mostly unsigned int sysctl_sched_nr_migrate;
2826extern __read_mostly unsigned int sysctl_sched_migration_cost;
2827
2828extern unsigned int sysctl_sched_base_slice;
2829
2830extern int sysctl_resched_latency_warn_ms;
2831extern int sysctl_resched_latency_warn_once;
2832
2833extern unsigned int sysctl_sched_tunable_scaling;
2834
2835extern unsigned int sysctl_numa_balancing_scan_delay;
2836extern unsigned int sysctl_numa_balancing_scan_period_min;
2837extern unsigned int sysctl_numa_balancing_scan_period_max;
2838extern unsigned int sysctl_numa_balancing_scan_size;
2839extern unsigned int sysctl_numa_balancing_hot_threshold;
2840
2841#ifdef CONFIG_SCHED_HRTICK
2842
2843/*
2844 * Use hrtick when:
2845 * - enabled by features
2846 * - hrtimer is actually high res
2847 */
2848static inline int hrtick_enabled(struct rq *rq)
2849{
2850 if (!cpu_active(cpu_of(rq)))
2851 return 0;
2852 return hrtimer_is_hres_active(&rq->hrtick_timer);
2853}
2854
2855static inline int hrtick_enabled_fair(struct rq *rq)
2856{
2857 if (!sched_feat(HRTICK))
2858 return 0;
2859 return hrtick_enabled(rq);
2860}
2861
2862static inline int hrtick_enabled_dl(struct rq *rq)
2863{
2864 if (!sched_feat(HRTICK_DL))
2865 return 0;
2866 return hrtick_enabled(rq);
2867}
2868
2869extern void hrtick_start(struct rq *rq, u64 delay);
2870
2871#else /* !CONFIG_SCHED_HRTICK: */
2872
2873static inline int hrtick_enabled_fair(struct rq *rq)
2874{
2875 return 0;
2876}
2877
2878static inline int hrtick_enabled_dl(struct rq *rq)
2879{
2880 return 0;
2881}
2882
2883static inline int hrtick_enabled(struct rq *rq)
2884{
2885 return 0;
2886}
2887
2888#endif /* !CONFIG_SCHED_HRTICK */
2889
2890#ifndef arch_scale_freq_tick
2891static __always_inline void arch_scale_freq_tick(void) { }
2892#endif
2893
2894#ifndef arch_scale_freq_capacity
2895/**
2896 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2897 * @cpu: the CPU in question.
2898 *
2899 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2900 *
2901 * f_curr
2902 * ------ * SCHED_CAPACITY_SCALE
2903 * f_max
2904 */
2905static __always_inline
2906unsigned long arch_scale_freq_capacity(int cpu)
2907{
2908 return SCHED_CAPACITY_SCALE;
2909}
2910#endif
2911
2912/*
2913 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2914 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2915 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2916 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2917 */
2918static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2919{
2920 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2921 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2922#ifdef CONFIG_SMP
2923 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2924#endif
2925}
2926
2927#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2928__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2929static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2930{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2931 _lock; return _t; }
2932
2933#ifdef CONFIG_SMP
2934
2935static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2936{
2937#ifdef CONFIG_SCHED_CORE
2938 /*
2939 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2940 * order by core-id first and cpu-id second.
2941 *
2942 * Notably:
2943 *
2944 * double_rq_lock(0,3); will take core-0, core-1 lock
2945 * double_rq_lock(1,2); will take core-1, core-0 lock
2946 *
2947 * when only cpu-id is considered.
2948 */
2949 if (rq1->core->cpu < rq2->core->cpu)
2950 return true;
2951 if (rq1->core->cpu > rq2->core->cpu)
2952 return false;
2953
2954 /*
2955 * __sched_core_flip() relies on SMT having cpu-id lock order.
2956 */
2957#endif
2958 return rq1->cpu < rq2->cpu;
2959}
2960
2961extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2962
2963#ifdef CONFIG_PREEMPTION
2964
2965/*
2966 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2967 * way at the expense of forcing extra atomic operations in all
2968 * invocations. This assures that the double_lock is acquired using the
2969 * same underlying policy as the spinlock_t on this architecture, which
2970 * reduces latency compared to the unfair variant below. However, it
2971 * also adds more overhead and therefore may reduce throughput.
2972 */
2973static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2974 __releases(this_rq->lock)
2975 __acquires(busiest->lock)
2976 __acquires(this_rq->lock)
2977{
2978 raw_spin_rq_unlock(this_rq);
2979 double_rq_lock(this_rq, busiest);
2980
2981 return 1;
2982}
2983
2984#else /* !CONFIG_PREEMPTION: */
2985/*
2986 * Unfair double_lock_balance: Optimizes throughput at the expense of
2987 * latency by eliminating extra atomic operations when the locks are
2988 * already in proper order on entry. This favors lower CPU-ids and will
2989 * grant the double lock to lower CPUs over higher ids under contention,
2990 * regardless of entry order into the function.
2991 */
2992static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2993 __releases(this_rq->lock)
2994 __acquires(busiest->lock)
2995 __acquires(this_rq->lock)
2996{
2997 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2998 likely(raw_spin_rq_trylock(busiest))) {
2999 double_rq_clock_clear_update(this_rq, busiest);
3000 return 0;
3001 }
3002
3003 if (rq_order_less(this_rq, busiest)) {
3004 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3005 double_rq_clock_clear_update(this_rq, busiest);
3006 return 0;
3007 }
3008
3009 raw_spin_rq_unlock(this_rq);
3010 double_rq_lock(this_rq, busiest);
3011
3012 return 1;
3013}
3014
3015#endif /* !CONFIG_PREEMPTION */
3016
3017/*
3018 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3019 */
3020static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3021{
3022 lockdep_assert_irqs_disabled();
3023
3024 return _double_lock_balance(this_rq, busiest);
3025}
3026
3027static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3028 __releases(busiest->lock)
3029{
3030 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
3031 raw_spin_rq_unlock(busiest);
3032 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3033}
3034
3035static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3036{
3037 if (l1 > l2)
3038 swap(l1, l2);
3039
3040 spin_lock(l1);
3041 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3042}
3043
3044static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3045{
3046 if (l1 > l2)
3047 swap(l1, l2);
3048
3049 spin_lock_irq(l1);
3050 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3051}
3052
3053static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3054{
3055 if (l1 > l2)
3056 swap(l1, l2);
3057
3058 raw_spin_lock(l1);
3059 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3060}
3061
3062static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3063{
3064 raw_spin_unlock(l1);
3065 raw_spin_unlock(l2);
3066}
3067
3068DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3069 double_raw_lock(_T->lock, _T->lock2),
3070 double_raw_unlock(_T->lock, _T->lock2))
3071
3072/*
3073 * double_rq_unlock - safely unlock two runqueues
3074 *
3075 * Note this does not restore interrupts like task_rq_unlock,
3076 * you need to do so manually after calling.
3077 */
3078static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3079 __releases(rq1->lock)
3080 __releases(rq2->lock)
3081{
3082 if (__rq_lockp(rq1) != __rq_lockp(rq2))
3083 raw_spin_rq_unlock(rq2);
3084 else
3085 __release(rq2->lock);
3086 raw_spin_rq_unlock(rq1);
3087}
3088
3089extern void set_rq_online (struct rq *rq);
3090extern void set_rq_offline(struct rq *rq);
3091
3092extern bool sched_smp_initialized;
3093
3094#else /* !CONFIG_SMP: */
3095
3096/*
3097 * double_rq_lock - safely lock two runqueues
3098 *
3099 * Note this does not disable interrupts like task_rq_lock,
3100 * you need to do so manually before calling.
3101 */
3102static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
3103 __acquires(rq1->lock)
3104 __acquires(rq2->lock)
3105{
3106 WARN_ON_ONCE(!irqs_disabled());
3107 WARN_ON_ONCE(rq1 != rq2);
3108 raw_spin_rq_lock(rq1);
3109 __acquire(rq2->lock); /* Fake it out ;) */
3110 double_rq_clock_clear_update(rq1, rq2);
3111}
3112
3113/*
3114 * double_rq_unlock - safely unlock two runqueues
3115 *
3116 * Note this does not restore interrupts like task_rq_unlock,
3117 * you need to do so manually after calling.
3118 */
3119static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3120 __releases(rq1->lock)
3121 __releases(rq2->lock)
3122{
3123 WARN_ON_ONCE(rq1 != rq2);
3124 raw_spin_rq_unlock(rq1);
3125 __release(rq2->lock);
3126}
3127
3128#endif /* !CONFIG_SMP */
3129
3130DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3131 double_rq_lock(_T->lock, _T->lock2),
3132 double_rq_unlock(_T->lock, _T->lock2))
3133
3134extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3135extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3136extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3137
3138extern bool sched_debug_verbose;
3139
3140extern void print_cfs_stats(struct seq_file *m, int cpu);
3141extern void print_rt_stats(struct seq_file *m, int cpu);
3142extern void print_dl_stats(struct seq_file *m, int cpu);
3143extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3144extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3145extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3146
3147extern void resched_latency_warn(int cpu, u64 latency);
3148#ifdef CONFIG_NUMA_BALANCING
3149extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3150extern void
3151print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3152 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3153#endif /* CONFIG_NUMA_BALANCING */
3154
3155extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3156extern void init_rt_rq(struct rt_rq *rt_rq);
3157extern void init_dl_rq(struct dl_rq *dl_rq);
3158
3159extern void cfs_bandwidth_usage_inc(void);
3160extern void cfs_bandwidth_usage_dec(void);
3161
3162#ifdef CONFIG_NO_HZ_COMMON
3163
3164#define NOHZ_BALANCE_KICK_BIT 0
3165#define NOHZ_STATS_KICK_BIT 1
3166#define NOHZ_NEWILB_KICK_BIT 2
3167#define NOHZ_NEXT_KICK_BIT 3
3168
3169/* Run sched_balance_domains() */
3170#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3171/* Update blocked load */
3172#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3173/* Update blocked load when entering idle */
3174#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3175/* Update nohz.next_balance */
3176#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3177
3178#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3179
3180#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3181
3182extern void nohz_balance_exit_idle(struct rq *rq);
3183#else /* !CONFIG_NO_HZ_COMMON: */
3184static inline void nohz_balance_exit_idle(struct rq *rq) { }
3185#endif /* !CONFIG_NO_HZ_COMMON */
3186
3187#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
3188extern void nohz_run_idle_balance(int cpu);
3189#else
3190static inline void nohz_run_idle_balance(int cpu) { }
3191#endif
3192
3193#include "stats.h"
3194
3195#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3196
3197extern void __sched_core_account_forceidle(struct rq *rq);
3198
3199static inline void sched_core_account_forceidle(struct rq *rq)
3200{
3201 if (schedstat_enabled())
3202 __sched_core_account_forceidle(rq);
3203}
3204
3205extern void __sched_core_tick(struct rq *rq);
3206
3207static inline void sched_core_tick(struct rq *rq)
3208{
3209 if (sched_core_enabled(rq) && schedstat_enabled())
3210 __sched_core_tick(rq);
3211}
3212
3213#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3214
3215static inline void sched_core_account_forceidle(struct rq *rq) { }
3216
3217static inline void sched_core_tick(struct rq *rq) { }
3218
3219#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3220
3221#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3222
3223struct irqtime {
3224 u64 total;
3225 u64 tick_delta;
3226 u64 irq_start_time;
3227 struct u64_stats_sync sync;
3228};
3229
3230DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3231extern int sched_clock_irqtime;
3232
3233static inline int irqtime_enabled(void)
3234{
3235 return sched_clock_irqtime;
3236}
3237
3238/*
3239 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3240 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3241 * and never move forward.
3242 */
3243static inline u64 irq_time_read(int cpu)
3244{
3245 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3246 unsigned int seq;
3247 u64 total;
3248
3249 do {
3250 seq = __u64_stats_fetch_begin(&irqtime->sync);
3251 total = irqtime->total;
3252 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3253
3254 return total;
3255}
3256
3257#else
3258
3259static inline int irqtime_enabled(void)
3260{
3261 return 0;
3262}
3263
3264#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3265
3266#ifdef CONFIG_CPU_FREQ
3267
3268DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3269
3270/**
3271 * cpufreq_update_util - Take a note about CPU utilization changes.
3272 * @rq: Runqueue to carry out the update for.
3273 * @flags: Update reason flags.
3274 *
3275 * This function is called by the scheduler on the CPU whose utilization is
3276 * being updated.
3277 *
3278 * It can only be called from RCU-sched read-side critical sections.
3279 *
3280 * The way cpufreq is currently arranged requires it to evaluate the CPU
3281 * performance state (frequency/voltage) on a regular basis to prevent it from
3282 * being stuck in a completely inadequate performance level for too long.
3283 * That is not guaranteed to happen if the updates are only triggered from CFS
3284 * and DL, though, because they may not be coming in if only RT tasks are
3285 * active all the time (or there are RT tasks only).
3286 *
3287 * As a workaround for that issue, this function is called periodically by the
3288 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3289 * but that really is a band-aid. Going forward it should be replaced with
3290 * solutions targeted more specifically at RT tasks.
3291 */
3292static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3293{
3294 struct update_util_data *data;
3295
3296 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3297 cpu_of(rq)));
3298 if (data)
3299 data->func(data, rq_clock(rq), flags);
3300}
3301#else /* !CONFIG_CPU_FREQ: */
3302static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3303#endif /* !CONFIG_CPU_FREQ */
3304
3305#ifdef arch_scale_freq_capacity
3306# ifndef arch_scale_freq_invariant
3307# define arch_scale_freq_invariant() true
3308# endif
3309#else
3310# define arch_scale_freq_invariant() false
3311#endif
3312
3313#ifdef CONFIG_SMP
3314
3315unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3316 unsigned long *min,
3317 unsigned long *max);
3318
3319unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3320 unsigned long min,
3321 unsigned long max);
3322
3323
3324/*
3325 * Verify the fitness of task @p to run on @cpu taking into account the
3326 * CPU original capacity and the runtime/deadline ratio of the task.
3327 *
3328 * The function will return true if the original capacity of @cpu is
3329 * greater than or equal to task's deadline density right shifted by
3330 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3331 */
3332static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3333{
3334 unsigned long cap = arch_scale_cpu_capacity(cpu);
3335
3336 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3337}
3338
3339static inline unsigned long cpu_bw_dl(struct rq *rq)
3340{
3341 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3342}
3343
3344static inline unsigned long cpu_util_dl(struct rq *rq)
3345{
3346 return READ_ONCE(rq->avg_dl.util_avg);
3347}
3348
3349
3350extern unsigned long cpu_util_cfs(int cpu);
3351extern unsigned long cpu_util_cfs_boost(int cpu);
3352
3353static inline unsigned long cpu_util_rt(struct rq *rq)
3354{
3355 return READ_ONCE(rq->avg_rt.util_avg);
3356}
3357
3358#else /* !CONFIG_SMP */
3359static inline bool update_other_load_avgs(struct rq *rq) { return false; }
3360#endif /* CONFIG_SMP */
3361
3362#ifdef CONFIG_UCLAMP_TASK
3363
3364unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3365
3366/*
3367 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3368 * by default in the fast path and only gets turned on once userspace performs
3369 * an operation that requires it.
3370 *
3371 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3372 * hence is active.
3373 */
3374static inline bool uclamp_is_used(void)
3375{
3376 return static_branch_likely(&sched_uclamp_used);
3377}
3378
3379/*
3380 * Enabling static branches would get the cpus_read_lock(),
3381 * check whether uclamp_is_used before enable it to avoid always
3382 * calling cpus_read_lock(). Because we never disable this
3383 * static key once enable it.
3384 */
3385static inline void sched_uclamp_enable(void)
3386{
3387 if (!uclamp_is_used())
3388 static_branch_enable(&sched_uclamp_used);
3389}
3390
3391static inline unsigned long uclamp_rq_get(struct rq *rq,
3392 enum uclamp_id clamp_id)
3393{
3394 return READ_ONCE(rq->uclamp[clamp_id].value);
3395}
3396
3397static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3398 unsigned int value)
3399{
3400 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3401}
3402
3403static inline bool uclamp_rq_is_idle(struct rq *rq)
3404{
3405 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3406}
3407
3408/* Is the rq being capped/throttled by uclamp_max? */
3409static inline bool uclamp_rq_is_capped(struct rq *rq)
3410{
3411 unsigned long rq_util;
3412 unsigned long max_util;
3413
3414 if (!uclamp_is_used())
3415 return false;
3416
3417 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3418 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3419
3420 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3421}
3422
3423#define for_each_clamp_id(clamp_id) \
3424 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3425
3426extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3427
3428
3429static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3430{
3431 if (clamp_id == UCLAMP_MIN)
3432 return 0;
3433 return SCHED_CAPACITY_SCALE;
3434}
3435
3436/* Integer rounded range for each bucket */
3437#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3438
3439static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3440{
3441 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3442}
3443
3444static inline void
3445uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3446{
3447 uc_se->value = value;
3448 uc_se->bucket_id = uclamp_bucket_id(value);
3449 uc_se->user_defined = user_defined;
3450}
3451
3452#else /* !CONFIG_UCLAMP_TASK: */
3453
3454static inline unsigned long
3455uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3456{
3457 if (clamp_id == UCLAMP_MIN)
3458 return 0;
3459
3460 return SCHED_CAPACITY_SCALE;
3461}
3462
3463static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3464
3465static inline bool uclamp_is_used(void)
3466{
3467 return false;
3468}
3469
3470static inline void sched_uclamp_enable(void) {}
3471
3472static inline unsigned long
3473uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3474{
3475 if (clamp_id == UCLAMP_MIN)
3476 return 0;
3477
3478 return SCHED_CAPACITY_SCALE;
3479}
3480
3481static inline void
3482uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3483{
3484}
3485
3486static inline bool uclamp_rq_is_idle(struct rq *rq)
3487{
3488 return false;
3489}
3490
3491#endif /* !CONFIG_UCLAMP_TASK */
3492
3493#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3494
3495static inline unsigned long cpu_util_irq(struct rq *rq)
3496{
3497 return READ_ONCE(rq->avg_irq.util_avg);
3498}
3499
3500static inline
3501unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3502{
3503 util *= (max - irq);
3504 util /= max;
3505
3506 return util;
3507
3508}
3509
3510#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3511
3512static inline unsigned long cpu_util_irq(struct rq *rq)
3513{
3514 return 0;
3515}
3516
3517static inline
3518unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3519{
3520 return util;
3521}
3522
3523#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3524
3525extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr);
3526
3527#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3528
3529#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3530
3531DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3532
3533static inline bool sched_energy_enabled(void)
3534{
3535 return static_branch_unlikely(&sched_energy_present);
3536}
3537
3538#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3539
3540#define perf_domain_span(pd) NULL
3541
3542static inline bool sched_energy_enabled(void) { return false; }
3543
3544#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3545
3546#ifdef CONFIG_MEMBARRIER
3547
3548/*
3549 * The scheduler provides memory barriers required by membarrier between:
3550 * - prior user-space memory accesses and store to rq->membarrier_state,
3551 * - store to rq->membarrier_state and following user-space memory accesses.
3552 * In the same way it provides those guarantees around store to rq->curr.
3553 */
3554static inline void membarrier_switch_mm(struct rq *rq,
3555 struct mm_struct *prev_mm,
3556 struct mm_struct *next_mm)
3557{
3558 int membarrier_state;
3559
3560 if (prev_mm == next_mm)
3561 return;
3562
3563 membarrier_state = atomic_read(&next_mm->membarrier_state);
3564 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3565 return;
3566
3567 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3568}
3569
3570#else /* !CONFIG_MEMBARRIER :*/
3571
3572static inline void membarrier_switch_mm(struct rq *rq,
3573 struct mm_struct *prev_mm,
3574 struct mm_struct *next_mm)
3575{
3576}
3577
3578#endif /* !CONFIG_MEMBARRIER */
3579
3580#ifdef CONFIG_SMP
3581static inline bool is_per_cpu_kthread(struct task_struct *p)
3582{
3583 if (!(p->flags & PF_KTHREAD))
3584 return false;
3585
3586 if (p->nr_cpus_allowed != 1)
3587 return false;
3588
3589 return true;
3590}
3591#endif
3592
3593extern void swake_up_all_locked(struct swait_queue_head *q);
3594extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3595
3596extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3597
3598#ifdef CONFIG_PREEMPT_DYNAMIC
3599extern int preempt_dynamic_mode;
3600extern int sched_dynamic_mode(const char *str);
3601extern void sched_dynamic_update(int mode);
3602#endif
3603extern const char *preempt_modes[];
3604
3605#ifdef CONFIG_SCHED_MM_CID
3606
3607#define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3608#define MM_CID_SCAN_DELAY 100 /* 100ms */
3609
3610extern raw_spinlock_t cid_lock;
3611extern int use_cid_lock;
3612
3613extern void sched_mm_cid_migrate_from(struct task_struct *t);
3614extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3615extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3616extern void init_sched_mm_cid(struct task_struct *t);
3617
3618static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3619{
3620 if (cid < 0)
3621 return;
3622 cpumask_clear_cpu(cid, mm_cidmask(mm));
3623}
3624
3625/*
3626 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3627 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3628 * be held to transition to other states.
3629 *
3630 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3631 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3632 */
3633static inline void mm_cid_put_lazy(struct task_struct *t)
3634{
3635 struct mm_struct *mm = t->mm;
3636 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3637 int cid;
3638
3639 lockdep_assert_irqs_disabled();
3640 cid = __this_cpu_read(pcpu_cid->cid);
3641 if (!mm_cid_is_lazy_put(cid) ||
3642 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3643 return;
3644 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3645}
3646
3647static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3648{
3649 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3650 int cid, res;
3651
3652 lockdep_assert_irqs_disabled();
3653 cid = __this_cpu_read(pcpu_cid->cid);
3654 for (;;) {
3655 if (mm_cid_is_unset(cid))
3656 return MM_CID_UNSET;
3657 /*
3658 * Attempt transition from valid or lazy-put to unset.
3659 */
3660 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3661 if (res == cid)
3662 break;
3663 cid = res;
3664 }
3665 return cid;
3666}
3667
3668static inline void mm_cid_put(struct mm_struct *mm)
3669{
3670 int cid;
3671
3672 lockdep_assert_irqs_disabled();
3673 cid = mm_cid_pcpu_unset(mm);
3674 if (cid == MM_CID_UNSET)
3675 return;
3676 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3677}
3678
3679static inline int __mm_cid_try_get(struct task_struct *t, struct mm_struct *mm)
3680{
3681 struct cpumask *cidmask = mm_cidmask(mm);
3682 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3683 int cid, max_nr_cid, allowed_max_nr_cid;
3684
3685 /*
3686 * After shrinking the number of threads or reducing the number
3687 * of allowed cpus, reduce the value of max_nr_cid so expansion
3688 * of cid allocation will preserve cache locality if the number
3689 * of threads or allowed cpus increase again.
3690 */
3691 max_nr_cid = atomic_read(&mm->max_nr_cid);
3692 while ((allowed_max_nr_cid = min_t(int, READ_ONCE(mm->nr_cpus_allowed),
3693 atomic_read(&mm->mm_users))),
3694 max_nr_cid > allowed_max_nr_cid) {
3695 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into max_nr_cid. */
3696 if (atomic_try_cmpxchg(&mm->max_nr_cid, &max_nr_cid, allowed_max_nr_cid)) {
3697 max_nr_cid = allowed_max_nr_cid;
3698 break;
3699 }
3700 }
3701 /* Try to re-use recent cid. This improves cache locality. */
3702 cid = __this_cpu_read(pcpu_cid->recent_cid);
3703 if (!mm_cid_is_unset(cid) && cid < max_nr_cid &&
3704 !cpumask_test_and_set_cpu(cid, cidmask))
3705 return cid;
3706 /*
3707 * Expand cid allocation if the maximum number of concurrency
3708 * IDs allocated (max_nr_cid) is below the number cpus allowed
3709 * and number of threads. Expanding cid allocation as much as
3710 * possible improves cache locality.
3711 */
3712 cid = max_nr_cid;
3713 while (cid < READ_ONCE(mm->nr_cpus_allowed) && cid < atomic_read(&mm->mm_users)) {
3714 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into cid. */
3715 if (!atomic_try_cmpxchg(&mm->max_nr_cid, &cid, cid + 1))
3716 continue;
3717 if (!cpumask_test_and_set_cpu(cid, cidmask))
3718 return cid;
3719 }
3720 /*
3721 * Find the first available concurrency id.
3722 * Retry finding first zero bit if the mask is temporarily
3723 * filled. This only happens during concurrent remote-clear
3724 * which owns a cid without holding a rq lock.
3725 */
3726 for (;;) {
3727 cid = cpumask_first_zero(cidmask);
3728 if (cid < READ_ONCE(mm->nr_cpus_allowed))
3729 break;
3730 cpu_relax();
3731 }
3732 if (cpumask_test_and_set_cpu(cid, cidmask))
3733 return -1;
3734
3735 return cid;
3736}
3737
3738/*
3739 * Save a snapshot of the current runqueue time of this cpu
3740 * with the per-cpu cid value, allowing to estimate how recently it was used.
3741 */
3742static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3743{
3744 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3745
3746 lockdep_assert_rq_held(rq);
3747 WRITE_ONCE(pcpu_cid->time, rq->clock);
3748}
3749
3750static inline int __mm_cid_get(struct rq *rq, struct task_struct *t,
3751 struct mm_struct *mm)
3752{
3753 int cid;
3754
3755 /*
3756 * All allocations (even those using the cid_lock) are lock-free. If
3757 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3758 * guarantee forward progress.
3759 */
3760 if (!READ_ONCE(use_cid_lock)) {
3761 cid = __mm_cid_try_get(t, mm);
3762 if (cid >= 0)
3763 goto end;
3764 raw_spin_lock(&cid_lock);
3765 } else {
3766 raw_spin_lock(&cid_lock);
3767 cid = __mm_cid_try_get(t, mm);
3768 if (cid >= 0)
3769 goto unlock;
3770 }
3771
3772 /*
3773 * cid concurrently allocated. Retry while forcing following
3774 * allocations to use the cid_lock to ensure forward progress.
3775 */
3776 WRITE_ONCE(use_cid_lock, 1);
3777 /*
3778 * Set use_cid_lock before allocation. Only care about program order
3779 * because this is only required for forward progress.
3780 */
3781 barrier();
3782 /*
3783 * Retry until it succeeds. It is guaranteed to eventually succeed once
3784 * all newcoming allocations observe the use_cid_lock flag set.
3785 */
3786 do {
3787 cid = __mm_cid_try_get(t, mm);
3788 cpu_relax();
3789 } while (cid < 0);
3790 /*
3791 * Allocate before clearing use_cid_lock. Only care about
3792 * program order because this is for forward progress.
3793 */
3794 barrier();
3795 WRITE_ONCE(use_cid_lock, 0);
3796unlock:
3797 raw_spin_unlock(&cid_lock);
3798end:
3799 mm_cid_snapshot_time(rq, mm);
3800
3801 return cid;
3802}
3803
3804static inline int mm_cid_get(struct rq *rq, struct task_struct *t,
3805 struct mm_struct *mm)
3806{
3807 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3808 struct cpumask *cpumask;
3809 int cid;
3810
3811 lockdep_assert_rq_held(rq);
3812 cpumask = mm_cidmask(mm);
3813 cid = __this_cpu_read(pcpu_cid->cid);
3814 if (mm_cid_is_valid(cid)) {
3815 mm_cid_snapshot_time(rq, mm);
3816 return cid;
3817 }
3818 if (mm_cid_is_lazy_put(cid)) {
3819 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3820 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3821 }
3822 cid = __mm_cid_get(rq, t, mm);
3823 __this_cpu_write(pcpu_cid->cid, cid);
3824 __this_cpu_write(pcpu_cid->recent_cid, cid);
3825
3826 return cid;
3827}
3828
3829static inline void switch_mm_cid(struct rq *rq,
3830 struct task_struct *prev,
3831 struct task_struct *next)
3832{
3833 /*
3834 * Provide a memory barrier between rq->curr store and load of
3835 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3836 *
3837 * Should be adapted if context_switch() is modified.
3838 */
3839 if (!next->mm) { // to kernel
3840 /*
3841 * user -> kernel transition does not guarantee a barrier, but
3842 * we can use the fact that it performs an atomic operation in
3843 * mmgrab().
3844 */
3845 if (prev->mm) // from user
3846 smp_mb__after_mmgrab();
3847 /*
3848 * kernel -> kernel transition does not change rq->curr->mm
3849 * state. It stays NULL.
3850 */
3851 } else { // to user
3852 /*
3853 * kernel -> user transition does not provide a barrier
3854 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3855 * Provide it here.
3856 */
3857 if (!prev->mm) { // from kernel
3858 smp_mb();
3859 } else { // from user
3860 /*
3861 * user->user transition relies on an implicit
3862 * memory barrier in switch_mm() when
3863 * current->mm changes. If the architecture
3864 * switch_mm() does not have an implicit memory
3865 * barrier, it is emitted here. If current->mm
3866 * is unchanged, no barrier is needed.
3867 */
3868 smp_mb__after_switch_mm();
3869 }
3870 }
3871 if (prev->mm_cid_active) {
3872 mm_cid_snapshot_time(rq, prev->mm);
3873 mm_cid_put_lazy(prev);
3874 prev->mm_cid = -1;
3875 }
3876 if (next->mm_cid_active)
3877 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next, next->mm);
3878}
3879
3880#else /* !CONFIG_SCHED_MM_CID: */
3881static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3882static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3883static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3884static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3885static inline void init_sched_mm_cid(struct task_struct *t) { }
3886#endif /* !CONFIG_SCHED_MM_CID */
3887
3888extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3889extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3890#ifdef CONFIG_SMP
3891static inline
3892void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
3893{
3894 lockdep_assert_rq_held(src_rq);
3895 lockdep_assert_rq_held(dst_rq);
3896
3897 deactivate_task(src_rq, task, 0);
3898 set_task_cpu(task, dst_rq->cpu);
3899 activate_task(dst_rq, task, 0);
3900}
3901
3902static inline
3903bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
3904{
3905 if (!task_on_cpu(rq, p) &&
3906 cpumask_test_cpu(cpu, &p->cpus_mask))
3907 return true;
3908
3909 return false;
3910}
3911#endif
3912
3913#ifdef CONFIG_RT_MUTEXES
3914
3915static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3916{
3917 if (pi_task)
3918 prio = min(prio, pi_task->prio);
3919
3920 return prio;
3921}
3922
3923static inline int rt_effective_prio(struct task_struct *p, int prio)
3924{
3925 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3926
3927 return __rt_effective_prio(pi_task, prio);
3928}
3929
3930#else /* !CONFIG_RT_MUTEXES: */
3931
3932static inline int rt_effective_prio(struct task_struct *p, int prio)
3933{
3934 return prio;
3935}
3936
3937#endif /* !CONFIG_RT_MUTEXES */
3938
3939extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3940extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3941extern const struct sched_class *__setscheduler_class(int policy, int prio);
3942extern void set_load_weight(struct task_struct *p, bool update_load);
3943extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3944extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3945
3946extern void check_class_changing(struct rq *rq, struct task_struct *p,
3947 const struct sched_class *prev_class);
3948extern void check_class_changed(struct rq *rq, struct task_struct *p,
3949 const struct sched_class *prev_class,
3950 int oldprio);
3951
3952#ifdef CONFIG_SMP
3953extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3954extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3955#else
3956
3957static inline struct balance_callback *splice_balance_callbacks(struct rq *rq)
3958{
3959 return NULL;
3960}
3961
3962static inline void balance_callbacks(struct rq *rq, struct balance_callback *head)
3963{
3964}
3965
3966#endif
3967
3968#ifdef CONFIG_SCHED_CLASS_EXT
3969/*
3970 * Used by SCX in the enable/disable paths to move tasks between sched_classes
3971 * and establish invariants.
3972 */
3973struct sched_enq_and_set_ctx {
3974 struct task_struct *p;
3975 int queue_flags;
3976 bool queued;
3977 bool running;
3978};
3979
3980void sched_deq_and_put_task(struct task_struct *p, int queue_flags,
3981 struct sched_enq_and_set_ctx *ctx);
3982void sched_enq_and_set_task(struct sched_enq_and_set_ctx *ctx);
3983
3984#endif /* CONFIG_SCHED_CLASS_EXT */
3985
3986#include "ext.h"
3987
3988#endif /* _KERNEL_SCHED_SCHED_H */
Linux 6.x block layer and io_uring tree(s)