Loading...
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#ifdef CONFIG_SCHED_DEBUG
95# define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
96#else
97# define SCHED_WARN_ON(x) ({ (void)(x), 0; })
98#endif
99
100/* task_struct::on_rq states: */
101#define TASK_ON_RQ_QUEUED 1
102#define TASK_ON_RQ_MIGRATING 2
103
104extern __read_mostly int scheduler_running;
105
106extern unsigned long calc_load_update;
107extern atomic_long_t calc_load_tasks;
108
109extern void calc_global_load_tick(struct rq *this_rq);
110extern long calc_load_fold_active(struct rq *this_rq, long adjust);
111
112extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
113
114extern int sysctl_sched_rt_period;
115extern int sysctl_sched_rt_runtime;
116extern int sched_rr_timeslice;
117
118/*
119 * Asymmetric CPU capacity bits
120 */
121struct asym_cap_data {
122 struct list_head link;
123 struct rcu_head rcu;
124 unsigned long capacity;
125 unsigned long cpus[];
126};
127
128extern struct list_head asym_cap_list;
129
130#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
131
132/*
133 * Helpers for converting nanosecond timing to jiffy resolution
134 */
135#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
136
137/*
138 * Increase resolution of nice-level calculations for 64-bit architectures.
139 * The extra resolution improves shares distribution and load balancing of
140 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
141 * hierarchies, especially on larger systems. This is not a user-visible change
142 * and does not change the user-interface for setting shares/weights.
143 *
144 * We increase resolution only if we have enough bits to allow this increased
145 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
146 * are pretty high and the returns do not justify the increased costs.
147 *
148 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
149 * increase coverage and consistency always enable it on 64-bit platforms.
150 */
151#ifdef CONFIG_64BIT
152# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
153# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
154# define scale_load_down(w) \
155({ \
156 unsigned long __w = (w); \
157 \
158 if (__w) \
159 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
160 __w; \
161})
162#else
163# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
164# define scale_load(w) (w)
165# define scale_load_down(w) (w)
166#endif
167
168/*
169 * Task weight (visible to users) and its load (invisible to users) have
170 * independent resolution, but they should be well calibrated. We use
171 * scale_load() and scale_load_down(w) to convert between them. The
172 * following must be true:
173 *
174 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
175 *
176 */
177#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
178
179/*
180 * Single value that decides SCHED_DEADLINE internal math precision.
181 * 10 -> just above 1us
182 * 9 -> just above 0.5us
183 */
184#define DL_SCALE 10
185
186/*
187 * Single value that denotes runtime == period, ie unlimited time.
188 */
189#define RUNTIME_INF ((u64)~0ULL)
190
191static inline int idle_policy(int policy)
192{
193 return policy == SCHED_IDLE;
194}
195
196static inline int normal_policy(int policy)
197{
198#ifdef CONFIG_SCHED_CLASS_EXT
199 if (policy == SCHED_EXT)
200 return true;
201#endif
202 return policy == SCHED_NORMAL;
203}
204
205static inline int fair_policy(int policy)
206{
207 return normal_policy(policy) || policy == SCHED_BATCH;
208}
209
210static inline int rt_policy(int policy)
211{
212 return policy == SCHED_FIFO || policy == SCHED_RR;
213}
214
215static inline int dl_policy(int policy)
216{
217 return policy == SCHED_DEADLINE;
218}
219
220static inline bool valid_policy(int policy)
221{
222 return idle_policy(policy) || fair_policy(policy) ||
223 rt_policy(policy) || dl_policy(policy);
224}
225
226static inline int task_has_idle_policy(struct task_struct *p)
227{
228 return idle_policy(p->policy);
229}
230
231static inline int task_has_rt_policy(struct task_struct *p)
232{
233 return rt_policy(p->policy);
234}
235
236static inline int task_has_dl_policy(struct task_struct *p)
237{
238 return dl_policy(p->policy);
239}
240
241#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
242
243static inline void update_avg(u64 *avg, u64 sample)
244{
245 s64 diff = sample - *avg;
246
247 *avg += diff / 8;
248}
249
250/*
251 * Shifting a value by an exponent greater *or equal* to the size of said value
252 * is UB; cap at size-1.
253 */
254#define shr_bound(val, shift) \
255 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
256
257/*
258 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
259 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
260 * maps pretty well onto the shares value used by scheduler and the round-trip
261 * conversions preserve the original value over the entire range.
262 */
263static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
264{
265 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
266}
267
268static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
269{
270 return clamp_t(unsigned long,
271 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
272 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
273}
274
275/*
276 * !! For sched_setattr_nocheck() (kernel) only !!
277 *
278 * This is actually gross. :(
279 *
280 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
281 * tasks, but still be able to sleep. We need this on platforms that cannot
282 * atomically change clock frequency. Remove once fast switching will be
283 * available on such platforms.
284 *
285 * SUGOV stands for SchedUtil GOVernor.
286 */
287#define SCHED_FLAG_SUGOV 0x10000000
288
289#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
290
291static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
292{
293#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
294 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
295#else
296 return false;
297#endif
298}
299
300/*
301 * Tells if entity @a should preempt entity @b.
302 */
303static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
304 const struct sched_dl_entity *b)
305{
306 return dl_entity_is_special(a) ||
307 dl_time_before(a->deadline, b->deadline);
308}
309
310/*
311 * This is the priority-queue data structure of the RT scheduling class:
312 */
313struct rt_prio_array {
314 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
315 struct list_head queue[MAX_RT_PRIO];
316};
317
318struct rt_bandwidth {
319 /* nests inside the rq lock: */
320 raw_spinlock_t rt_runtime_lock;
321 ktime_t rt_period;
322 u64 rt_runtime;
323 struct hrtimer rt_period_timer;
324 unsigned int rt_period_active;
325};
326
327static inline int dl_bandwidth_enabled(void)
328{
329 return sysctl_sched_rt_runtime >= 0;
330}
331
332/*
333 * To keep the bandwidth of -deadline tasks under control
334 * we need some place where:
335 * - store the maximum -deadline bandwidth of each cpu;
336 * - cache the fraction of bandwidth that is currently allocated in
337 * each root domain;
338 *
339 * This is all done in the data structure below. It is similar to the
340 * one used for RT-throttling (rt_bandwidth), with the main difference
341 * that, since here we are only interested in admission control, we
342 * do not decrease any runtime while the group "executes", neither we
343 * need a timer to replenish it.
344 *
345 * With respect to SMP, bandwidth is given on a per root domain basis,
346 * meaning that:
347 * - bw (< 100%) is the deadline bandwidth of each CPU;
348 * - total_bw is the currently allocated bandwidth in each root domain;
349 */
350struct dl_bw {
351 raw_spinlock_t lock;
352 u64 bw;
353 u64 total_bw;
354};
355
356extern void init_dl_bw(struct dl_bw *dl_b);
357extern int sched_dl_global_validate(void);
358extern void sched_dl_do_global(void);
359extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
360extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
361extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
362extern bool __checkparam_dl(const struct sched_attr *attr);
363extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
364extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
365extern int dl_bw_check_overflow(int cpu);
366extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
367/*
368 * SCHED_DEADLINE supports servers (nested scheduling) with the following
369 * interface:
370 *
371 * dl_se::rq -- runqueue we belong to.
372 *
373 * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
374 * server when it runs out of tasks to run.
375 *
376 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
377 * returns NULL.
378 *
379 * dl_server_update() -- called from update_curr_common(), propagates runtime
380 * to the server.
381 *
382 * dl_server_start()
383 * dl_server_stop() -- start/stop the server when it has (no) tasks.
384 *
385 * dl_server_init() -- initializes the server.
386 */
387extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
388extern void dl_server_start(struct sched_dl_entity *dl_se);
389extern void dl_server_stop(struct sched_dl_entity *dl_se);
390extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
391 dl_server_has_tasks_f has_tasks,
392 dl_server_pick_f pick_task);
393
394extern void dl_server_update_idle_time(struct rq *rq,
395 struct task_struct *p);
396extern void fair_server_init(struct rq *rq);
397extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
398extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
399 u64 runtime, u64 period, bool init);
400
401static inline bool dl_server_active(struct sched_dl_entity *dl_se)
402{
403 return dl_se->dl_server_active;
404}
405
406#ifdef CONFIG_CGROUP_SCHED
407
408extern struct list_head task_groups;
409
410struct cfs_bandwidth {
411#ifdef CONFIG_CFS_BANDWIDTH
412 raw_spinlock_t lock;
413 ktime_t period;
414 u64 quota;
415 u64 runtime;
416 u64 burst;
417 u64 runtime_snap;
418 s64 hierarchical_quota;
419
420 u8 idle;
421 u8 period_active;
422 u8 slack_started;
423 struct hrtimer period_timer;
424 struct hrtimer slack_timer;
425 struct list_head throttled_cfs_rq;
426
427 /* Statistics: */
428 int nr_periods;
429 int nr_throttled;
430 int nr_burst;
431 u64 throttled_time;
432 u64 burst_time;
433#endif
434};
435
436/* Task group related information */
437struct task_group {
438 struct cgroup_subsys_state css;
439
440#ifdef CONFIG_GROUP_SCHED_WEIGHT
441 /* A positive value indicates that this is a SCHED_IDLE group. */
442 int idle;
443#endif
444
445#ifdef CONFIG_FAIR_GROUP_SCHED
446 /* schedulable entities of this group on each CPU */
447 struct sched_entity **se;
448 /* runqueue "owned" by this group on each CPU */
449 struct cfs_rq **cfs_rq;
450 unsigned long shares;
451#ifdef CONFIG_SMP
452 /*
453 * load_avg can be heavily contended at clock tick time, so put
454 * it in its own cache-line separated from the fields above which
455 * will also be accessed at each tick.
456 */
457 atomic_long_t load_avg ____cacheline_aligned;
458#endif
459#endif
460
461#ifdef CONFIG_RT_GROUP_SCHED
462 struct sched_rt_entity **rt_se;
463 struct rt_rq **rt_rq;
464
465 struct rt_bandwidth rt_bandwidth;
466#endif
467
468#ifdef CONFIG_EXT_GROUP_SCHED
469 u32 scx_flags; /* SCX_TG_* */
470 u32 scx_weight;
471#endif
472
473 struct rcu_head rcu;
474 struct list_head list;
475
476 struct task_group *parent;
477 struct list_head siblings;
478 struct list_head children;
479
480#ifdef CONFIG_SCHED_AUTOGROUP
481 struct autogroup *autogroup;
482#endif
483
484 struct cfs_bandwidth cfs_bandwidth;
485
486#ifdef CONFIG_UCLAMP_TASK_GROUP
487 /* The two decimal precision [%] value requested from user-space */
488 unsigned int uclamp_pct[UCLAMP_CNT];
489 /* Clamp values requested for a task group */
490 struct uclamp_se uclamp_req[UCLAMP_CNT];
491 /* Effective clamp values used for a task group */
492 struct uclamp_se uclamp[UCLAMP_CNT];
493#endif
494
495};
496
497#ifdef CONFIG_GROUP_SCHED_WEIGHT
498#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
499
500/*
501 * A weight of 0 or 1 can cause arithmetics problems.
502 * A weight of a cfs_rq is the sum of weights of which entities
503 * are queued on this cfs_rq, so a weight of a entity should not be
504 * too large, so as the shares value of a task group.
505 * (The default weight is 1024 - so there's no practical
506 * limitation from this.)
507 */
508#define MIN_SHARES (1UL << 1)
509#define MAX_SHARES (1UL << 18)
510#endif
511
512typedef int (*tg_visitor)(struct task_group *, void *);
513
514extern int walk_tg_tree_from(struct task_group *from,
515 tg_visitor down, tg_visitor up, void *data);
516
517/*
518 * Iterate the full tree, calling @down when first entering a node and @up when
519 * leaving it for the final time.
520 *
521 * Caller must hold rcu_lock or sufficient equivalent.
522 */
523static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
524{
525 return walk_tg_tree_from(&root_task_group, down, up, data);
526}
527
528static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
529{
530 return css ? container_of(css, struct task_group, css) : NULL;
531}
532
533extern int tg_nop(struct task_group *tg, void *data);
534
535#ifdef CONFIG_FAIR_GROUP_SCHED
536extern void free_fair_sched_group(struct task_group *tg);
537extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
538extern void online_fair_sched_group(struct task_group *tg);
539extern void unregister_fair_sched_group(struct task_group *tg);
540#else
541static inline void free_fair_sched_group(struct task_group *tg) { }
542static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
543{
544 return 1;
545}
546static inline void online_fair_sched_group(struct task_group *tg) { }
547static inline void unregister_fair_sched_group(struct task_group *tg) { }
548#endif
549
550extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
551 struct sched_entity *se, int cpu,
552 struct sched_entity *parent);
553extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
554
555extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
556extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
557extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
558extern bool cfs_task_bw_constrained(struct task_struct *p);
559
560extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
561 struct sched_rt_entity *rt_se, int cpu,
562 struct sched_rt_entity *parent);
563extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
564extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
565extern long sched_group_rt_runtime(struct task_group *tg);
566extern long sched_group_rt_period(struct task_group *tg);
567extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
568
569extern struct task_group *sched_create_group(struct task_group *parent);
570extern void sched_online_group(struct task_group *tg,
571 struct task_group *parent);
572extern void sched_destroy_group(struct task_group *tg);
573extern void sched_release_group(struct task_group *tg);
574
575extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
576
577#ifdef CONFIG_FAIR_GROUP_SCHED
578extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
579
580extern int sched_group_set_idle(struct task_group *tg, long idle);
581
582#ifdef CONFIG_SMP
583extern void set_task_rq_fair(struct sched_entity *se,
584 struct cfs_rq *prev, struct cfs_rq *next);
585#else /* !CONFIG_SMP */
586static inline void set_task_rq_fair(struct sched_entity *se,
587 struct cfs_rq *prev, struct cfs_rq *next) { }
588#endif /* CONFIG_SMP */
589#else /* !CONFIG_FAIR_GROUP_SCHED */
590static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
591static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
592#endif /* CONFIG_FAIR_GROUP_SCHED */
593
594#else /* CONFIG_CGROUP_SCHED */
595
596struct cfs_bandwidth { };
597
598static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
599
600#endif /* CONFIG_CGROUP_SCHED */
601
602extern void unregister_rt_sched_group(struct task_group *tg);
603extern void free_rt_sched_group(struct task_group *tg);
604extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
605
606/*
607 * u64_u32_load/u64_u32_store
608 *
609 * Use a copy of a u64 value to protect against data race. This is only
610 * applicable for 32-bits architectures.
611 */
612#ifdef CONFIG_64BIT
613# define u64_u32_load_copy(var, copy) var
614# define u64_u32_store_copy(var, copy, val) (var = val)
615#else
616# define u64_u32_load_copy(var, copy) \
617({ \
618 u64 __val, __val_copy; \
619 do { \
620 __val_copy = copy; \
621 /* \
622 * paired with u64_u32_store_copy(), ordering access \
623 * to var and copy. \
624 */ \
625 smp_rmb(); \
626 __val = var; \
627 } while (__val != __val_copy); \
628 __val; \
629})
630# define u64_u32_store_copy(var, copy, val) \
631do { \
632 typeof(val) __val = (val); \
633 var = __val; \
634 /* \
635 * paired with u64_u32_load_copy(), ordering access to var and \
636 * copy. \
637 */ \
638 smp_wmb(); \
639 copy = __val; \
640} while (0)
641#endif
642# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
643# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
644
645struct balance_callback {
646 struct balance_callback *next;
647 void (*func)(struct rq *rq);
648};
649
650/* CFS-related fields in a runqueue */
651struct cfs_rq {
652 struct load_weight load;
653 unsigned int nr_running;
654 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
655 unsigned int idle_nr_running; /* SCHED_IDLE */
656 unsigned int idle_h_nr_running; /* SCHED_IDLE */
657 unsigned int h_nr_delayed;
658
659 s64 avg_vruntime;
660 u64 avg_load;
661
662 u64 min_vruntime;
663#ifdef CONFIG_SCHED_CORE
664 unsigned int forceidle_seq;
665 u64 min_vruntime_fi;
666#endif
667
668 struct rb_root_cached tasks_timeline;
669
670 /*
671 * 'curr' points to currently running entity on this cfs_rq.
672 * It is set to NULL otherwise (i.e when none are currently running).
673 */
674 struct sched_entity *curr;
675 struct sched_entity *next;
676
677#ifdef CONFIG_SMP
678 /*
679 * CFS load tracking
680 */
681 struct sched_avg avg;
682#ifndef CONFIG_64BIT
683 u64 last_update_time_copy;
684#endif
685 struct {
686 raw_spinlock_t lock ____cacheline_aligned;
687 int nr;
688 unsigned long load_avg;
689 unsigned long util_avg;
690 unsigned long runnable_avg;
691 } removed;
692
693#ifdef CONFIG_FAIR_GROUP_SCHED
694 u64 last_update_tg_load_avg;
695 unsigned long tg_load_avg_contrib;
696 long propagate;
697 long prop_runnable_sum;
698
699 /*
700 * h_load = weight * f(tg)
701 *
702 * Where f(tg) is the recursive weight fraction assigned to
703 * this group.
704 */
705 unsigned long h_load;
706 u64 last_h_load_update;
707 struct sched_entity *h_load_next;
708#endif /* CONFIG_FAIR_GROUP_SCHED */
709#endif /* CONFIG_SMP */
710
711#ifdef CONFIG_FAIR_GROUP_SCHED
712 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
713
714 /*
715 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
716 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
717 * (like users, containers etc.)
718 *
719 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
720 * This list is used during load balance.
721 */
722 int on_list;
723 struct list_head leaf_cfs_rq_list;
724 struct task_group *tg; /* group that "owns" this runqueue */
725
726 /* Locally cached copy of our task_group's idle value */
727 int idle;
728
729#ifdef CONFIG_CFS_BANDWIDTH
730 int runtime_enabled;
731 s64 runtime_remaining;
732
733 u64 throttled_pelt_idle;
734#ifndef CONFIG_64BIT
735 u64 throttled_pelt_idle_copy;
736#endif
737 u64 throttled_clock;
738 u64 throttled_clock_pelt;
739 u64 throttled_clock_pelt_time;
740 u64 throttled_clock_self;
741 u64 throttled_clock_self_time;
742 int throttled;
743 int throttle_count;
744 struct list_head throttled_list;
745 struct list_head throttled_csd_list;
746#endif /* CONFIG_CFS_BANDWIDTH */
747#endif /* CONFIG_FAIR_GROUP_SCHED */
748};
749
750#ifdef CONFIG_SCHED_CLASS_EXT
751/* scx_rq->flags, protected by the rq lock */
752enum scx_rq_flags {
753 /*
754 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
755 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
756 * only while the BPF scheduler considers the CPU to be online.
757 */
758 SCX_RQ_ONLINE = 1 << 0,
759 SCX_RQ_CAN_STOP_TICK = 1 << 1,
760 SCX_RQ_BAL_PENDING = 1 << 2, /* balance hasn't run yet */
761 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
762 SCX_RQ_BYPASSING = 1 << 4,
763
764 SCX_RQ_IN_WAKEUP = 1 << 16,
765 SCX_RQ_IN_BALANCE = 1 << 17,
766};
767
768struct scx_rq {
769 struct scx_dispatch_q local_dsq;
770 struct list_head runnable_list; /* runnable tasks on this rq */
771 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
772 unsigned long ops_qseq;
773 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
774 u32 nr_running;
775 u32 flags;
776 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
777 bool cpu_released;
778 cpumask_var_t cpus_to_kick;
779 cpumask_var_t cpus_to_kick_if_idle;
780 cpumask_var_t cpus_to_preempt;
781 cpumask_var_t cpus_to_wait;
782 unsigned long pnt_seq;
783 struct balance_callback deferred_bal_cb;
784 struct irq_work deferred_irq_work;
785 struct irq_work kick_cpus_irq_work;
786};
787#endif /* CONFIG_SCHED_CLASS_EXT */
788
789static inline int rt_bandwidth_enabled(void)
790{
791 return sysctl_sched_rt_runtime >= 0;
792}
793
794/* RT IPI pull logic requires IRQ_WORK */
795#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
796# define HAVE_RT_PUSH_IPI
797#endif
798
799/* Real-Time classes' related field in a runqueue: */
800struct rt_rq {
801 struct rt_prio_array active;
802 unsigned int rt_nr_running;
803 unsigned int rr_nr_running;
804#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
805 struct {
806 int curr; /* highest queued rt task prio */
807#ifdef CONFIG_SMP
808 int next; /* next highest */
809#endif
810 } highest_prio;
811#endif
812#ifdef CONFIG_SMP
813 bool overloaded;
814 struct plist_head pushable_tasks;
815
816#endif /* CONFIG_SMP */
817 int rt_queued;
818
819#ifdef CONFIG_RT_GROUP_SCHED
820 int rt_throttled;
821 u64 rt_time;
822 u64 rt_runtime;
823 /* Nests inside the rq lock: */
824 raw_spinlock_t rt_runtime_lock;
825
826 unsigned int rt_nr_boosted;
827
828 struct rq *rq;
829 struct task_group *tg;
830#endif
831};
832
833static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
834{
835 return rt_rq->rt_queued && rt_rq->rt_nr_running;
836}
837
838/* Deadline class' related fields in a runqueue */
839struct dl_rq {
840 /* runqueue is an rbtree, ordered by deadline */
841 struct rb_root_cached root;
842
843 unsigned int dl_nr_running;
844
845#ifdef CONFIG_SMP
846 /*
847 * Deadline values of the currently executing and the
848 * earliest ready task on this rq. Caching these facilitates
849 * the decision whether or not a ready but not running task
850 * should migrate somewhere else.
851 */
852 struct {
853 u64 curr;
854 u64 next;
855 } earliest_dl;
856
857 bool overloaded;
858
859 /*
860 * Tasks on this rq that can be pushed away. They are kept in
861 * an rb-tree, ordered by tasks' deadlines, with caching
862 * of the leftmost (earliest deadline) element.
863 */
864 struct rb_root_cached pushable_dl_tasks_root;
865#else
866 struct dl_bw dl_bw;
867#endif
868 /*
869 * "Active utilization" for this runqueue: increased when a
870 * task wakes up (becomes TASK_RUNNING) and decreased when a
871 * task blocks
872 */
873 u64 running_bw;
874
875 /*
876 * Utilization of the tasks "assigned" to this runqueue (including
877 * the tasks that are in runqueue and the tasks that executed on this
878 * CPU and blocked). Increased when a task moves to this runqueue, and
879 * decreased when the task moves away (migrates, changes scheduling
880 * policy, or terminates).
881 * This is needed to compute the "inactive utilization" for the
882 * runqueue (inactive utilization = this_bw - running_bw).
883 */
884 u64 this_bw;
885 u64 extra_bw;
886
887 /*
888 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
889 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
890 */
891 u64 max_bw;
892
893 /*
894 * Inverse of the fraction of CPU utilization that can be reclaimed
895 * by the GRUB algorithm.
896 */
897 u64 bw_ratio;
898};
899
900#ifdef CONFIG_FAIR_GROUP_SCHED
901
902/* An entity is a task if it doesn't "own" a runqueue */
903#define entity_is_task(se) (!se->my_q)
904
905static inline void se_update_runnable(struct sched_entity *se)
906{
907 if (!entity_is_task(se)) {
908 struct cfs_rq *cfs_rq = se->my_q;
909
910 se->runnable_weight = cfs_rq->h_nr_running - cfs_rq->h_nr_delayed;
911 }
912}
913
914static inline long se_runnable(struct sched_entity *se)
915{
916 if (se->sched_delayed)
917 return false;
918
919 if (entity_is_task(se))
920 return !!se->on_rq;
921 else
922 return se->runnable_weight;
923}
924
925#else /* !CONFIG_FAIR_GROUP_SCHED: */
926
927#define entity_is_task(se) 1
928
929static inline void se_update_runnable(struct sched_entity *se) { }
930
931static inline long se_runnable(struct sched_entity *se)
932{
933 if (se->sched_delayed)
934 return false;
935
936 return !!se->on_rq;
937}
938
939#endif /* !CONFIG_FAIR_GROUP_SCHED */
940
941#ifdef CONFIG_SMP
942/*
943 * XXX we want to get rid of these helpers and use the full load resolution.
944 */
945static inline long se_weight(struct sched_entity *se)
946{
947 return scale_load_down(se->load.weight);
948}
949
950
951static inline bool sched_asym_prefer(int a, int b)
952{
953 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
954}
955
956struct perf_domain {
957 struct em_perf_domain *em_pd;
958 struct perf_domain *next;
959 struct rcu_head rcu;
960};
961
962/*
963 * We add the notion of a root-domain which will be used to define per-domain
964 * variables. Each exclusive cpuset essentially defines an island domain by
965 * fully partitioning the member CPUs from any other cpuset. Whenever a new
966 * exclusive cpuset is created, we also create and attach a new root-domain
967 * object.
968 *
969 */
970struct root_domain {
971 atomic_t refcount;
972 atomic_t rto_count;
973 struct rcu_head rcu;
974 cpumask_var_t span;
975 cpumask_var_t online;
976
977 /*
978 * Indicate pullable load on at least one CPU, e.g:
979 * - More than one runnable task
980 * - Running task is misfit
981 */
982 bool overloaded;
983
984 /* Indicate one or more CPUs over-utilized (tipping point) */
985 bool overutilized;
986
987 /*
988 * The bit corresponding to a CPU gets set here if such CPU has more
989 * than one runnable -deadline task (as it is below for RT tasks).
990 */
991 cpumask_var_t dlo_mask;
992 atomic_t dlo_count;
993 struct dl_bw dl_bw;
994 struct cpudl cpudl;
995
996 /*
997 * Indicate whether a root_domain's dl_bw has been checked or
998 * updated. It's monotonously increasing value.
999 *
1000 * Also, some corner cases, like 'wrap around' is dangerous, but given
1001 * that u64 is 'big enough'. So that shouldn't be a concern.
1002 */
1003 u64 visit_gen;
1004
1005#ifdef HAVE_RT_PUSH_IPI
1006 /*
1007 * For IPI pull requests, loop across the rto_mask.
1008 */
1009 struct irq_work rto_push_work;
1010 raw_spinlock_t rto_lock;
1011 /* These are only updated and read within rto_lock */
1012 int rto_loop;
1013 int rto_cpu;
1014 /* These atomics are updated outside of a lock */
1015 atomic_t rto_loop_next;
1016 atomic_t rto_loop_start;
1017#endif
1018 /*
1019 * The "RT overload" flag: it gets set if a CPU has more than
1020 * one runnable RT task.
1021 */
1022 cpumask_var_t rto_mask;
1023 struct cpupri cpupri;
1024
1025 /*
1026 * NULL-terminated list of performance domains intersecting with the
1027 * CPUs of the rd. Protected by RCU.
1028 */
1029 struct perf_domain __rcu *pd;
1030};
1031
1032extern void init_defrootdomain(void);
1033extern int sched_init_domains(const struct cpumask *cpu_map);
1034extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1035extern void sched_get_rd(struct root_domain *rd);
1036extern void sched_put_rd(struct root_domain *rd);
1037
1038static inline int get_rd_overloaded(struct root_domain *rd)
1039{
1040 return READ_ONCE(rd->overloaded);
1041}
1042
1043static inline void set_rd_overloaded(struct root_domain *rd, int status)
1044{
1045 if (get_rd_overloaded(rd) != status)
1046 WRITE_ONCE(rd->overloaded, status);
1047}
1048
1049#ifdef HAVE_RT_PUSH_IPI
1050extern void rto_push_irq_work_func(struct irq_work *work);
1051#endif
1052#endif /* CONFIG_SMP */
1053
1054#ifdef CONFIG_UCLAMP_TASK
1055/*
1056 * struct uclamp_bucket - Utilization clamp bucket
1057 * @value: utilization clamp value for tasks on this clamp bucket
1058 * @tasks: number of RUNNABLE tasks on this clamp bucket
1059 *
1060 * Keep track of how many tasks are RUNNABLE for a given utilization
1061 * clamp value.
1062 */
1063struct uclamp_bucket {
1064 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1065 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1066};
1067
1068/*
1069 * struct uclamp_rq - rq's utilization clamp
1070 * @value: currently active clamp values for a rq
1071 * @bucket: utilization clamp buckets affecting a rq
1072 *
1073 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1074 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1075 * (or actually running) with that value.
1076 *
1077 * There are up to UCLAMP_CNT possible different clamp values, currently there
1078 * are only two: minimum utilization and maximum utilization.
1079 *
1080 * All utilization clamping values are MAX aggregated, since:
1081 * - for util_min: we want to run the CPU at least at the max of the minimum
1082 * utilization required by its currently RUNNABLE tasks.
1083 * - for util_max: we want to allow the CPU to run up to the max of the
1084 * maximum utilization allowed by its currently RUNNABLE tasks.
1085 *
1086 * Since on each system we expect only a limited number of different
1087 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1088 * the metrics required to compute all the per-rq utilization clamp values.
1089 */
1090struct uclamp_rq {
1091 unsigned int value;
1092 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1093};
1094
1095DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1096#endif /* CONFIG_UCLAMP_TASK */
1097
1098/*
1099 * This is the main, per-CPU runqueue data structure.
1100 *
1101 * Locking rule: those places that want to lock multiple runqueues
1102 * (such as the load balancing or the thread migration code), lock
1103 * acquire operations must be ordered by ascending &runqueue.
1104 */
1105struct rq {
1106 /* runqueue lock: */
1107 raw_spinlock_t __lock;
1108
1109 unsigned int nr_running;
1110#ifdef CONFIG_NUMA_BALANCING
1111 unsigned int nr_numa_running;
1112 unsigned int nr_preferred_running;
1113 unsigned int numa_migrate_on;
1114#endif
1115#ifdef CONFIG_NO_HZ_COMMON
1116#ifdef CONFIG_SMP
1117 unsigned long last_blocked_load_update_tick;
1118 unsigned int has_blocked_load;
1119 call_single_data_t nohz_csd;
1120#endif /* CONFIG_SMP */
1121 unsigned int nohz_tick_stopped;
1122 atomic_t nohz_flags;
1123#endif /* CONFIG_NO_HZ_COMMON */
1124
1125#ifdef CONFIG_SMP
1126 unsigned int ttwu_pending;
1127#endif
1128 u64 nr_switches;
1129
1130#ifdef CONFIG_UCLAMP_TASK
1131 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1132 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1133 unsigned int uclamp_flags;
1134#define UCLAMP_FLAG_IDLE 0x01
1135#endif
1136
1137 struct cfs_rq cfs;
1138 struct rt_rq rt;
1139 struct dl_rq dl;
1140#ifdef CONFIG_SCHED_CLASS_EXT
1141 struct scx_rq scx;
1142#endif
1143
1144 struct sched_dl_entity fair_server;
1145
1146#ifdef CONFIG_FAIR_GROUP_SCHED
1147 /* list of leaf cfs_rq on this CPU: */
1148 struct list_head leaf_cfs_rq_list;
1149 struct list_head *tmp_alone_branch;
1150#endif /* CONFIG_FAIR_GROUP_SCHED */
1151
1152 /*
1153 * This is part of a global counter where only the total sum
1154 * over all CPUs matters. A task can increase this counter on
1155 * one CPU and if it got migrated afterwards it may decrease
1156 * it on another CPU. Always updated under the runqueue lock:
1157 */
1158 unsigned int nr_uninterruptible;
1159
1160 union {
1161 struct task_struct __rcu *donor; /* Scheduler context */
1162 struct task_struct __rcu *curr; /* Execution context */
1163 };
1164 struct sched_dl_entity *dl_server;
1165 struct task_struct *idle;
1166 struct task_struct *stop;
1167 unsigned long next_balance;
1168 struct mm_struct *prev_mm;
1169
1170 unsigned int clock_update_flags;
1171 u64 clock;
1172 /* Ensure that all clocks are in the same cache line */
1173 u64 clock_task ____cacheline_aligned;
1174 u64 clock_pelt;
1175 unsigned long lost_idle_time;
1176 u64 clock_pelt_idle;
1177 u64 clock_idle;
1178#ifndef CONFIG_64BIT
1179 u64 clock_pelt_idle_copy;
1180 u64 clock_idle_copy;
1181#endif
1182
1183 atomic_t nr_iowait;
1184
1185#ifdef CONFIG_SCHED_DEBUG
1186 u64 last_seen_need_resched_ns;
1187 int ticks_without_resched;
1188#endif
1189
1190#ifdef CONFIG_MEMBARRIER
1191 int membarrier_state;
1192#endif
1193
1194#ifdef CONFIG_SMP
1195 struct root_domain *rd;
1196 struct sched_domain __rcu *sd;
1197
1198 unsigned long cpu_capacity;
1199
1200 struct balance_callback *balance_callback;
1201
1202 unsigned char nohz_idle_balance;
1203 unsigned char idle_balance;
1204
1205 unsigned long misfit_task_load;
1206
1207 /* For active balancing */
1208 int active_balance;
1209 int push_cpu;
1210 struct cpu_stop_work active_balance_work;
1211
1212 /* CPU of this runqueue: */
1213 int cpu;
1214 int online;
1215
1216 struct list_head cfs_tasks;
1217
1218 struct sched_avg avg_rt;
1219 struct sched_avg avg_dl;
1220#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1221 struct sched_avg avg_irq;
1222#endif
1223#ifdef CONFIG_SCHED_HW_PRESSURE
1224 struct sched_avg avg_hw;
1225#endif
1226 u64 idle_stamp;
1227 u64 avg_idle;
1228
1229 /* This is used to determine avg_idle's max value */
1230 u64 max_idle_balance_cost;
1231
1232#ifdef CONFIG_HOTPLUG_CPU
1233 struct rcuwait hotplug_wait;
1234#endif
1235#endif /* CONFIG_SMP */
1236
1237#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1238 u64 prev_irq_time;
1239 u64 psi_irq_time;
1240#endif
1241#ifdef CONFIG_PARAVIRT
1242 u64 prev_steal_time;
1243#endif
1244#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1245 u64 prev_steal_time_rq;
1246#endif
1247
1248 /* calc_load related fields */
1249 unsigned long calc_load_update;
1250 long calc_load_active;
1251
1252#ifdef CONFIG_SCHED_HRTICK
1253#ifdef CONFIG_SMP
1254 call_single_data_t hrtick_csd;
1255#endif
1256 struct hrtimer hrtick_timer;
1257 ktime_t hrtick_time;
1258#endif
1259
1260#ifdef CONFIG_SCHEDSTATS
1261 /* latency stats */
1262 struct sched_info rq_sched_info;
1263 unsigned long long rq_cpu_time;
1264
1265 /* sys_sched_yield() stats */
1266 unsigned int yld_count;
1267
1268 /* schedule() stats */
1269 unsigned int sched_count;
1270 unsigned int sched_goidle;
1271
1272 /* try_to_wake_up() stats */
1273 unsigned int ttwu_count;
1274 unsigned int ttwu_local;
1275#endif
1276
1277#ifdef CONFIG_CPU_IDLE
1278 /* Must be inspected within a RCU lock section */
1279 struct cpuidle_state *idle_state;
1280#endif
1281
1282#ifdef CONFIG_SMP
1283 unsigned int nr_pinned;
1284#endif
1285 unsigned int push_busy;
1286 struct cpu_stop_work push_work;
1287
1288#ifdef CONFIG_SCHED_CORE
1289 /* per rq */
1290 struct rq *core;
1291 struct task_struct *core_pick;
1292 struct sched_dl_entity *core_dl_server;
1293 unsigned int core_enabled;
1294 unsigned int core_sched_seq;
1295 struct rb_root core_tree;
1296
1297 /* shared state -- careful with sched_core_cpu_deactivate() */
1298 unsigned int core_task_seq;
1299 unsigned int core_pick_seq;
1300 unsigned long core_cookie;
1301 unsigned int core_forceidle_count;
1302 unsigned int core_forceidle_seq;
1303 unsigned int core_forceidle_occupation;
1304 u64 core_forceidle_start;
1305#endif
1306
1307 /* Scratch cpumask to be temporarily used under rq_lock */
1308 cpumask_var_t scratch_mask;
1309
1310#if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1311 call_single_data_t cfsb_csd;
1312 struct list_head cfsb_csd_list;
1313#endif
1314};
1315
1316#ifdef CONFIG_FAIR_GROUP_SCHED
1317
1318/* CPU runqueue to which this cfs_rq is attached */
1319static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1320{
1321 return cfs_rq->rq;
1322}
1323
1324#else
1325
1326static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1327{
1328 return container_of(cfs_rq, struct rq, cfs);
1329}
1330#endif
1331
1332static inline int cpu_of(struct rq *rq)
1333{
1334#ifdef CONFIG_SMP
1335 return rq->cpu;
1336#else
1337 return 0;
1338#endif
1339}
1340
1341#define MDF_PUSH 0x01
1342
1343static inline bool is_migration_disabled(struct task_struct *p)
1344{
1345#ifdef CONFIG_SMP
1346 return p->migration_disabled;
1347#else
1348 return false;
1349#endif
1350}
1351
1352DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1353
1354#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1355#define this_rq() this_cpu_ptr(&runqueues)
1356#define task_rq(p) cpu_rq(task_cpu(p))
1357#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1358#define raw_rq() raw_cpu_ptr(&runqueues)
1359
1360static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1361{
1362 /* Do nothing */
1363}
1364
1365#ifdef CONFIG_SCHED_CORE
1366static inline struct cpumask *sched_group_span(struct sched_group *sg);
1367
1368DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1369
1370static inline bool sched_core_enabled(struct rq *rq)
1371{
1372 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1373}
1374
1375static inline bool sched_core_disabled(void)
1376{
1377 return !static_branch_unlikely(&__sched_core_enabled);
1378}
1379
1380/*
1381 * Be careful with this function; not for general use. The return value isn't
1382 * stable unless you actually hold a relevant rq->__lock.
1383 */
1384static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1385{
1386 if (sched_core_enabled(rq))
1387 return &rq->core->__lock;
1388
1389 return &rq->__lock;
1390}
1391
1392static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1393{
1394 if (rq->core_enabled)
1395 return &rq->core->__lock;
1396
1397 return &rq->__lock;
1398}
1399
1400extern bool
1401cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1402
1403extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1404
1405/*
1406 * Helpers to check if the CPU's core cookie matches with the task's cookie
1407 * when core scheduling is enabled.
1408 * A special case is that the task's cookie always matches with CPU's core
1409 * cookie if the CPU is in an idle core.
1410 */
1411static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1412{
1413 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1414 if (!sched_core_enabled(rq))
1415 return true;
1416
1417 return rq->core->core_cookie == p->core_cookie;
1418}
1419
1420static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1421{
1422 bool idle_core = true;
1423 int cpu;
1424
1425 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1426 if (!sched_core_enabled(rq))
1427 return true;
1428
1429 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1430 if (!available_idle_cpu(cpu)) {
1431 idle_core = false;
1432 break;
1433 }
1434 }
1435
1436 /*
1437 * A CPU in an idle core is always the best choice for tasks with
1438 * cookies.
1439 */
1440 return idle_core || rq->core->core_cookie == p->core_cookie;
1441}
1442
1443static inline bool sched_group_cookie_match(struct rq *rq,
1444 struct task_struct *p,
1445 struct sched_group *group)
1446{
1447 int cpu;
1448
1449 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1450 if (!sched_core_enabled(rq))
1451 return true;
1452
1453 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1454 if (sched_core_cookie_match(cpu_rq(cpu), p))
1455 return true;
1456 }
1457 return false;
1458}
1459
1460static inline bool sched_core_enqueued(struct task_struct *p)
1461{
1462 return !RB_EMPTY_NODE(&p->core_node);
1463}
1464
1465extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1466extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1467
1468extern void sched_core_get(void);
1469extern void sched_core_put(void);
1470
1471#else /* !CONFIG_SCHED_CORE: */
1472
1473static inline bool sched_core_enabled(struct rq *rq)
1474{
1475 return false;
1476}
1477
1478static inline bool sched_core_disabled(void)
1479{
1480 return true;
1481}
1482
1483static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1484{
1485 return &rq->__lock;
1486}
1487
1488static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1489{
1490 return &rq->__lock;
1491}
1492
1493static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1494{
1495 return true;
1496}
1497
1498static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1499{
1500 return true;
1501}
1502
1503static inline bool sched_group_cookie_match(struct rq *rq,
1504 struct task_struct *p,
1505 struct sched_group *group)
1506{
1507 return true;
1508}
1509
1510#endif /* !CONFIG_SCHED_CORE */
1511
1512static inline void lockdep_assert_rq_held(struct rq *rq)
1513{
1514 lockdep_assert_held(__rq_lockp(rq));
1515}
1516
1517extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1518extern bool raw_spin_rq_trylock(struct rq *rq);
1519extern void raw_spin_rq_unlock(struct rq *rq);
1520
1521static inline void raw_spin_rq_lock(struct rq *rq)
1522{
1523 raw_spin_rq_lock_nested(rq, 0);
1524}
1525
1526static inline void raw_spin_rq_lock_irq(struct rq *rq)
1527{
1528 local_irq_disable();
1529 raw_spin_rq_lock(rq);
1530}
1531
1532static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1533{
1534 raw_spin_rq_unlock(rq);
1535 local_irq_enable();
1536}
1537
1538static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1539{
1540 unsigned long flags;
1541
1542 local_irq_save(flags);
1543 raw_spin_rq_lock(rq);
1544
1545 return flags;
1546}
1547
1548static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1549{
1550 raw_spin_rq_unlock(rq);
1551 local_irq_restore(flags);
1552}
1553
1554#define raw_spin_rq_lock_irqsave(rq, flags) \
1555do { \
1556 flags = _raw_spin_rq_lock_irqsave(rq); \
1557} while (0)
1558
1559#ifdef CONFIG_SCHED_SMT
1560extern void __update_idle_core(struct rq *rq);
1561
1562static inline void update_idle_core(struct rq *rq)
1563{
1564 if (static_branch_unlikely(&sched_smt_present))
1565 __update_idle_core(rq);
1566}
1567
1568#else
1569static inline void update_idle_core(struct rq *rq) { }
1570#endif
1571
1572#ifdef CONFIG_FAIR_GROUP_SCHED
1573
1574static inline struct task_struct *task_of(struct sched_entity *se)
1575{
1576 SCHED_WARN_ON(!entity_is_task(se));
1577 return container_of(se, struct task_struct, se);
1578}
1579
1580static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1581{
1582 return p->se.cfs_rq;
1583}
1584
1585/* runqueue on which this entity is (to be) queued */
1586static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1587{
1588 return se->cfs_rq;
1589}
1590
1591/* runqueue "owned" by this group */
1592static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1593{
1594 return grp->my_q;
1595}
1596
1597#else /* !CONFIG_FAIR_GROUP_SCHED: */
1598
1599#define task_of(_se) container_of(_se, struct task_struct, se)
1600
1601static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1602{
1603 return &task_rq(p)->cfs;
1604}
1605
1606static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1607{
1608 const struct task_struct *p = task_of(se);
1609 struct rq *rq = task_rq(p);
1610
1611 return &rq->cfs;
1612}
1613
1614/* runqueue "owned" by this group */
1615static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1616{
1617 return NULL;
1618}
1619
1620#endif /* !CONFIG_FAIR_GROUP_SCHED */
1621
1622extern void update_rq_clock(struct rq *rq);
1623
1624/*
1625 * rq::clock_update_flags bits
1626 *
1627 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1628 * call to __schedule(). This is an optimisation to avoid
1629 * neighbouring rq clock updates.
1630 *
1631 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1632 * in effect and calls to update_rq_clock() are being ignored.
1633 *
1634 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1635 * made to update_rq_clock() since the last time rq::lock was pinned.
1636 *
1637 * If inside of __schedule(), clock_update_flags will have been
1638 * shifted left (a left shift is a cheap operation for the fast path
1639 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1640 *
1641 * if (rq-clock_update_flags >= RQCF_UPDATED)
1642 *
1643 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1644 * one position though, because the next rq_unpin_lock() will shift it
1645 * back.
1646 */
1647#define RQCF_REQ_SKIP 0x01
1648#define RQCF_ACT_SKIP 0x02
1649#define RQCF_UPDATED 0x04
1650
1651static inline void assert_clock_updated(struct rq *rq)
1652{
1653 /*
1654 * The only reason for not seeing a clock update since the
1655 * last rq_pin_lock() is if we're currently skipping updates.
1656 */
1657 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1658}
1659
1660static inline u64 rq_clock(struct rq *rq)
1661{
1662 lockdep_assert_rq_held(rq);
1663 assert_clock_updated(rq);
1664
1665 return rq->clock;
1666}
1667
1668static inline u64 rq_clock_task(struct rq *rq)
1669{
1670 lockdep_assert_rq_held(rq);
1671 assert_clock_updated(rq);
1672
1673 return rq->clock_task;
1674}
1675
1676static inline void rq_clock_skip_update(struct rq *rq)
1677{
1678 lockdep_assert_rq_held(rq);
1679 rq->clock_update_flags |= RQCF_REQ_SKIP;
1680}
1681
1682/*
1683 * See rt task throttling, which is the only time a skip
1684 * request is canceled.
1685 */
1686static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1687{
1688 lockdep_assert_rq_held(rq);
1689 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1690}
1691
1692/*
1693 * During cpu offlining and rq wide unthrottling, we can trigger
1694 * an update_rq_clock() for several cfs and rt runqueues (Typically
1695 * when using list_for_each_entry_*)
1696 * rq_clock_start_loop_update() can be called after updating the clock
1697 * once and before iterating over the list to prevent multiple update.
1698 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1699 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1700 */
1701static inline void rq_clock_start_loop_update(struct rq *rq)
1702{
1703 lockdep_assert_rq_held(rq);
1704 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1705 rq->clock_update_flags |= RQCF_ACT_SKIP;
1706}
1707
1708static inline void rq_clock_stop_loop_update(struct rq *rq)
1709{
1710 lockdep_assert_rq_held(rq);
1711 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1712}
1713
1714struct rq_flags {
1715 unsigned long flags;
1716 struct pin_cookie cookie;
1717#ifdef CONFIG_SCHED_DEBUG
1718 /*
1719 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1720 * current pin context is stashed here in case it needs to be
1721 * restored in rq_repin_lock().
1722 */
1723 unsigned int clock_update_flags;
1724#endif
1725};
1726
1727extern struct balance_callback balance_push_callback;
1728
1729/*
1730 * Lockdep annotation that avoids accidental unlocks; it's like a
1731 * sticky/continuous lockdep_assert_held().
1732 *
1733 * This avoids code that has access to 'struct rq *rq' (basically everything in
1734 * the scheduler) from accidentally unlocking the rq if they do not also have a
1735 * copy of the (on-stack) 'struct rq_flags rf'.
1736 *
1737 * Also see Documentation/locking/lockdep-design.rst.
1738 */
1739static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1740{
1741 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1742
1743#ifdef CONFIG_SCHED_DEBUG
1744 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1745 rf->clock_update_flags = 0;
1746# ifdef CONFIG_SMP
1747 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1748# endif
1749#endif
1750}
1751
1752static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1753{
1754#ifdef CONFIG_SCHED_DEBUG
1755 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1756 rf->clock_update_flags = RQCF_UPDATED;
1757#endif
1758
1759 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1760}
1761
1762static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1763{
1764 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1765
1766#ifdef CONFIG_SCHED_DEBUG
1767 /*
1768 * Restore the value we stashed in @rf for this pin context.
1769 */
1770 rq->clock_update_flags |= rf->clock_update_flags;
1771#endif
1772}
1773
1774extern
1775struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1776 __acquires(rq->lock);
1777
1778extern
1779struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1780 __acquires(p->pi_lock)
1781 __acquires(rq->lock);
1782
1783static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1784 __releases(rq->lock)
1785{
1786 rq_unpin_lock(rq, rf);
1787 raw_spin_rq_unlock(rq);
1788}
1789
1790static inline void
1791task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1792 __releases(rq->lock)
1793 __releases(p->pi_lock)
1794{
1795 rq_unpin_lock(rq, rf);
1796 raw_spin_rq_unlock(rq);
1797 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1798}
1799
1800DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1801 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1802 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1803 struct rq *rq; struct rq_flags rf)
1804
1805static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1806 __acquires(rq->lock)
1807{
1808 raw_spin_rq_lock_irqsave(rq, rf->flags);
1809 rq_pin_lock(rq, rf);
1810}
1811
1812static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1813 __acquires(rq->lock)
1814{
1815 raw_spin_rq_lock_irq(rq);
1816 rq_pin_lock(rq, rf);
1817}
1818
1819static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1820 __acquires(rq->lock)
1821{
1822 raw_spin_rq_lock(rq);
1823 rq_pin_lock(rq, rf);
1824}
1825
1826static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1827 __releases(rq->lock)
1828{
1829 rq_unpin_lock(rq, rf);
1830 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1831}
1832
1833static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1834 __releases(rq->lock)
1835{
1836 rq_unpin_lock(rq, rf);
1837 raw_spin_rq_unlock_irq(rq);
1838}
1839
1840static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1841 __releases(rq->lock)
1842{
1843 rq_unpin_lock(rq, rf);
1844 raw_spin_rq_unlock(rq);
1845}
1846
1847DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1848 rq_lock(_T->lock, &_T->rf),
1849 rq_unlock(_T->lock, &_T->rf),
1850 struct rq_flags rf)
1851
1852DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1853 rq_lock_irq(_T->lock, &_T->rf),
1854 rq_unlock_irq(_T->lock, &_T->rf),
1855 struct rq_flags rf)
1856
1857DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1858 rq_lock_irqsave(_T->lock, &_T->rf),
1859 rq_unlock_irqrestore(_T->lock, &_T->rf),
1860 struct rq_flags rf)
1861
1862static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1863 __acquires(rq->lock)
1864{
1865 struct rq *rq;
1866
1867 local_irq_disable();
1868 rq = this_rq();
1869 rq_lock(rq, rf);
1870
1871 return rq;
1872}
1873
1874#ifdef CONFIG_NUMA
1875
1876enum numa_topology_type {
1877 NUMA_DIRECT,
1878 NUMA_GLUELESS_MESH,
1879 NUMA_BACKPLANE,
1880};
1881
1882extern enum numa_topology_type sched_numa_topology_type;
1883extern int sched_max_numa_distance;
1884extern bool find_numa_distance(int distance);
1885extern void sched_init_numa(int offline_node);
1886extern void sched_update_numa(int cpu, bool online);
1887extern void sched_domains_numa_masks_set(unsigned int cpu);
1888extern void sched_domains_numa_masks_clear(unsigned int cpu);
1889extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1890
1891#else /* !CONFIG_NUMA: */
1892
1893static inline void sched_init_numa(int offline_node) { }
1894static inline void sched_update_numa(int cpu, bool online) { }
1895static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1896static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1897
1898static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1899{
1900 return nr_cpu_ids;
1901}
1902
1903#endif /* !CONFIG_NUMA */
1904
1905#ifdef CONFIG_NUMA_BALANCING
1906
1907/* The regions in numa_faults array from task_struct */
1908enum numa_faults_stats {
1909 NUMA_MEM = 0,
1910 NUMA_CPU,
1911 NUMA_MEMBUF,
1912 NUMA_CPUBUF
1913};
1914
1915extern void sched_setnuma(struct task_struct *p, int node);
1916extern int migrate_task_to(struct task_struct *p, int cpu);
1917extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1918 int cpu, int scpu);
1919extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1920
1921#else /* !CONFIG_NUMA_BALANCING: */
1922
1923static inline void
1924init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1925{
1926}
1927
1928#endif /* !CONFIG_NUMA_BALANCING */
1929
1930#ifdef CONFIG_SMP
1931
1932static inline void
1933queue_balance_callback(struct rq *rq,
1934 struct balance_callback *head,
1935 void (*func)(struct rq *rq))
1936{
1937 lockdep_assert_rq_held(rq);
1938
1939 /*
1940 * Don't (re)queue an already queued item; nor queue anything when
1941 * balance_push() is active, see the comment with
1942 * balance_push_callback.
1943 */
1944 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1945 return;
1946
1947 head->func = func;
1948 head->next = rq->balance_callback;
1949 rq->balance_callback = head;
1950}
1951
1952#define rcu_dereference_check_sched_domain(p) \
1953 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1954
1955/*
1956 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1957 * See destroy_sched_domains: call_rcu for details.
1958 *
1959 * The domain tree of any CPU may only be accessed from within
1960 * preempt-disabled sections.
1961 */
1962#define for_each_domain(cpu, __sd) \
1963 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1964 __sd; __sd = __sd->parent)
1965
1966/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1967#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1968static const unsigned int SD_SHARED_CHILD_MASK =
1969#include <linux/sched/sd_flags.h>
19700;
1971#undef SD_FLAG
1972
1973/**
1974 * highest_flag_domain - Return highest sched_domain containing flag.
1975 * @cpu: The CPU whose highest level of sched domain is to
1976 * be returned.
1977 * @flag: The flag to check for the highest sched_domain
1978 * for the given CPU.
1979 *
1980 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1981 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1982 */
1983static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1984{
1985 struct sched_domain *sd, *hsd = NULL;
1986
1987 for_each_domain(cpu, sd) {
1988 if (sd->flags & flag) {
1989 hsd = sd;
1990 continue;
1991 }
1992
1993 /*
1994 * Stop the search if @flag is known to be shared at lower
1995 * levels. It will not be found further up.
1996 */
1997 if (flag & SD_SHARED_CHILD_MASK)
1998 break;
1999 }
2000
2001 return hsd;
2002}
2003
2004static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2005{
2006 struct sched_domain *sd;
2007
2008 for_each_domain(cpu, sd) {
2009 if (sd->flags & flag)
2010 break;
2011 }
2012
2013 return sd;
2014}
2015
2016DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2017DECLARE_PER_CPU(int, sd_llc_size);
2018DECLARE_PER_CPU(int, sd_llc_id);
2019DECLARE_PER_CPU(int, sd_share_id);
2020DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2021DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2022DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2023DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2024
2025extern struct static_key_false sched_asym_cpucapacity;
2026extern struct static_key_false sched_cluster_active;
2027
2028static __always_inline bool sched_asym_cpucap_active(void)
2029{
2030 return static_branch_unlikely(&sched_asym_cpucapacity);
2031}
2032
2033struct sched_group_capacity {
2034 atomic_t ref;
2035 /*
2036 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2037 * for a single CPU.
2038 */
2039 unsigned long capacity;
2040 unsigned long min_capacity; /* Min per-CPU capacity in group */
2041 unsigned long max_capacity; /* Max per-CPU capacity in group */
2042 unsigned long next_update;
2043 int imbalance; /* XXX unrelated to capacity but shared group state */
2044
2045#ifdef CONFIG_SCHED_DEBUG
2046 int id;
2047#endif
2048
2049 unsigned long cpumask[]; /* Balance mask */
2050};
2051
2052struct sched_group {
2053 struct sched_group *next; /* Must be a circular list */
2054 atomic_t ref;
2055
2056 unsigned int group_weight;
2057 unsigned int cores;
2058 struct sched_group_capacity *sgc;
2059 int asym_prefer_cpu; /* CPU of highest priority in group */
2060 int flags;
2061
2062 /*
2063 * The CPUs this group covers.
2064 *
2065 * NOTE: this field is variable length. (Allocated dynamically
2066 * by attaching extra space to the end of the structure,
2067 * depending on how many CPUs the kernel has booted up with)
2068 */
2069 unsigned long cpumask[];
2070};
2071
2072static inline struct cpumask *sched_group_span(struct sched_group *sg)
2073{
2074 return to_cpumask(sg->cpumask);
2075}
2076
2077/*
2078 * See build_balance_mask().
2079 */
2080static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2081{
2082 return to_cpumask(sg->sgc->cpumask);
2083}
2084
2085extern int group_balance_cpu(struct sched_group *sg);
2086
2087#ifdef CONFIG_SCHED_DEBUG
2088extern void update_sched_domain_debugfs(void);
2089extern void dirty_sched_domain_sysctl(int cpu);
2090#else
2091static inline void update_sched_domain_debugfs(void) { }
2092static inline void dirty_sched_domain_sysctl(int cpu) { }
2093#endif
2094
2095extern int sched_update_scaling(void);
2096
2097static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2098{
2099 if (!p->user_cpus_ptr)
2100 return cpu_possible_mask; /* &init_task.cpus_mask */
2101 return p->user_cpus_ptr;
2102}
2103
2104#endif /* CONFIG_SMP */
2105
2106#ifdef CONFIG_CGROUP_SCHED
2107
2108/*
2109 * Return the group to which this tasks belongs.
2110 *
2111 * We cannot use task_css() and friends because the cgroup subsystem
2112 * changes that value before the cgroup_subsys::attach() method is called,
2113 * therefore we cannot pin it and might observe the wrong value.
2114 *
2115 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2116 * core changes this before calling sched_move_task().
2117 *
2118 * Instead we use a 'copy' which is updated from sched_move_task() while
2119 * holding both task_struct::pi_lock and rq::lock.
2120 */
2121static inline struct task_group *task_group(struct task_struct *p)
2122{
2123 return p->sched_task_group;
2124}
2125
2126/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2127static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2128{
2129#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2130 struct task_group *tg = task_group(p);
2131#endif
2132
2133#ifdef CONFIG_FAIR_GROUP_SCHED
2134 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2135 p->se.cfs_rq = tg->cfs_rq[cpu];
2136 p->se.parent = tg->se[cpu];
2137 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2138#endif
2139
2140#ifdef CONFIG_RT_GROUP_SCHED
2141 p->rt.rt_rq = tg->rt_rq[cpu];
2142 p->rt.parent = tg->rt_se[cpu];
2143#endif
2144}
2145
2146#else /* !CONFIG_CGROUP_SCHED: */
2147
2148static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2149
2150static inline struct task_group *task_group(struct task_struct *p)
2151{
2152 return NULL;
2153}
2154
2155#endif /* !CONFIG_CGROUP_SCHED */
2156
2157static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2158{
2159 set_task_rq(p, cpu);
2160#ifdef CONFIG_SMP
2161 /*
2162 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2163 * successfully executed on another CPU. We must ensure that updates of
2164 * per-task data have been completed by this moment.
2165 */
2166 smp_wmb();
2167 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2168 p->wake_cpu = cpu;
2169#endif
2170}
2171
2172/*
2173 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2174 */
2175#ifdef CONFIG_SCHED_DEBUG
2176# define const_debug __read_mostly
2177#else
2178# define const_debug const
2179#endif
2180
2181#define SCHED_FEAT(name, enabled) \
2182 __SCHED_FEAT_##name ,
2183
2184enum {
2185#include "features.h"
2186 __SCHED_FEAT_NR,
2187};
2188
2189#undef SCHED_FEAT
2190
2191#ifdef CONFIG_SCHED_DEBUG
2192
2193/*
2194 * To support run-time toggling of sched features, all the translation units
2195 * (but core.c) reference the sysctl_sched_features defined in core.c.
2196 */
2197extern const_debug unsigned int sysctl_sched_features;
2198
2199#ifdef CONFIG_JUMP_LABEL
2200
2201#define SCHED_FEAT(name, enabled) \
2202static __always_inline bool static_branch_##name(struct static_key *key) \
2203{ \
2204 return static_key_##enabled(key); \
2205}
2206
2207#include "features.h"
2208#undef SCHED_FEAT
2209
2210extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2211#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2212
2213#else /* !CONFIG_JUMP_LABEL: */
2214
2215#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2216
2217#endif /* !CONFIG_JUMP_LABEL */
2218
2219#else /* !SCHED_DEBUG: */
2220
2221/*
2222 * Each translation unit has its own copy of sysctl_sched_features to allow
2223 * constants propagation at compile time and compiler optimization based on
2224 * features default.
2225 */
2226#define SCHED_FEAT(name, enabled) \
2227 (1UL << __SCHED_FEAT_##name) * enabled |
2228static const_debug __maybe_unused unsigned int sysctl_sched_features =
2229#include "features.h"
2230 0;
2231#undef SCHED_FEAT
2232
2233#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2234
2235#endif /* !SCHED_DEBUG */
2236
2237extern struct static_key_false sched_numa_balancing;
2238extern struct static_key_false sched_schedstats;
2239
2240static inline u64 global_rt_period(void)
2241{
2242 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2243}
2244
2245static inline u64 global_rt_runtime(void)
2246{
2247 if (sysctl_sched_rt_runtime < 0)
2248 return RUNTIME_INF;
2249
2250 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2251}
2252
2253/*
2254 * Is p the current execution context?
2255 */
2256static inline int task_current(struct rq *rq, struct task_struct *p)
2257{
2258 return rq->curr == p;
2259}
2260
2261/*
2262 * Is p the current scheduling context?
2263 *
2264 * Note that it might be the current execution context at the same time if
2265 * rq->curr == rq->donor == p.
2266 */
2267static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2268{
2269 return rq->donor == p;
2270}
2271
2272static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2273{
2274#ifdef CONFIG_SMP
2275 return p->on_cpu;
2276#else
2277 return task_current(rq, p);
2278#endif
2279}
2280
2281static inline int task_on_rq_queued(struct task_struct *p)
2282{
2283 return p->on_rq == TASK_ON_RQ_QUEUED;
2284}
2285
2286static inline int task_on_rq_migrating(struct task_struct *p)
2287{
2288 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2289}
2290
2291/* Wake flags. The first three directly map to some SD flag value */
2292#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2293#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2294#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2295
2296#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2297#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2298#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2299#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2300
2301#ifdef CONFIG_SMP
2302static_assert(WF_EXEC == SD_BALANCE_EXEC);
2303static_assert(WF_FORK == SD_BALANCE_FORK);
2304static_assert(WF_TTWU == SD_BALANCE_WAKE);
2305#endif
2306
2307/*
2308 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2309 * of tasks with abnormal "nice" values across CPUs the contribution that
2310 * each task makes to its run queue's load is weighted according to its
2311 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2312 * scaled version of the new time slice allocation that they receive on time
2313 * slice expiry etc.
2314 */
2315
2316#define WEIGHT_IDLEPRIO 3
2317#define WMULT_IDLEPRIO 1431655765
2318
2319extern const int sched_prio_to_weight[40];
2320extern const u32 sched_prio_to_wmult[40];
2321
2322/*
2323 * {de,en}queue flags:
2324 *
2325 * DEQUEUE_SLEEP - task is no longer runnable
2326 * ENQUEUE_WAKEUP - task just became runnable
2327 *
2328 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2329 * are in a known state which allows modification. Such pairs
2330 * should preserve as much state as possible.
2331 *
2332 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2333 * in the runqueue.
2334 *
2335 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2336 *
2337 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2338 *
2339 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2340 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2341 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2342 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2343 *
2344 */
2345
2346#define DEQUEUE_SLEEP 0x01 /* Matches ENQUEUE_WAKEUP */
2347#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2348#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2349#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2350#define DEQUEUE_SPECIAL 0x10
2351#define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2352#define DEQUEUE_DELAYED 0x200 /* Matches ENQUEUE_DELAYED */
2353
2354#define ENQUEUE_WAKEUP 0x01
2355#define ENQUEUE_RESTORE 0x02
2356#define ENQUEUE_MOVE 0x04
2357#define ENQUEUE_NOCLOCK 0x08
2358
2359#define ENQUEUE_HEAD 0x10
2360#define ENQUEUE_REPLENISH 0x20
2361#ifdef CONFIG_SMP
2362#define ENQUEUE_MIGRATED 0x40
2363#else
2364#define ENQUEUE_MIGRATED 0x00
2365#endif
2366#define ENQUEUE_INITIAL 0x80
2367#define ENQUEUE_MIGRATING 0x100
2368#define ENQUEUE_DELAYED 0x200
2369#define ENQUEUE_RQ_SELECTED 0x400
2370
2371#define RETRY_TASK ((void *)-1UL)
2372
2373struct affinity_context {
2374 const struct cpumask *new_mask;
2375 struct cpumask *user_mask;
2376 unsigned int flags;
2377};
2378
2379extern s64 update_curr_common(struct rq *rq);
2380
2381struct sched_class {
2382
2383#ifdef CONFIG_UCLAMP_TASK
2384 int uclamp_enabled;
2385#endif
2386
2387 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2388 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2389 void (*yield_task) (struct rq *rq);
2390 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2391
2392 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2393
2394 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2395 struct task_struct *(*pick_task)(struct rq *rq);
2396 /*
2397 * Optional! When implemented pick_next_task() should be equivalent to:
2398 *
2399 * next = pick_task();
2400 * if (next) {
2401 * put_prev_task(prev);
2402 * set_next_task_first(next);
2403 * }
2404 */
2405 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev);
2406
2407 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2408 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2409
2410#ifdef CONFIG_SMP
2411 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2412
2413 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2414
2415 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2416
2417 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2418
2419 void (*rq_online)(struct rq *rq);
2420 void (*rq_offline)(struct rq *rq);
2421
2422 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2423#endif
2424
2425 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2426 void (*task_fork)(struct task_struct *p);
2427 void (*task_dead)(struct task_struct *p);
2428
2429 /*
2430 * The switched_from() call is allowed to drop rq->lock, therefore we
2431 * cannot assume the switched_from/switched_to pair is serialized by
2432 * rq->lock. They are however serialized by p->pi_lock.
2433 */
2434 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2435 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2436 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2437 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2438 const struct load_weight *lw);
2439 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2440 int oldprio);
2441
2442 unsigned int (*get_rr_interval)(struct rq *rq,
2443 struct task_struct *task);
2444
2445 void (*update_curr)(struct rq *rq);
2446
2447#ifdef CONFIG_FAIR_GROUP_SCHED
2448 void (*task_change_group)(struct task_struct *p);
2449#endif
2450
2451#ifdef CONFIG_SCHED_CORE
2452 int (*task_is_throttled)(struct task_struct *p, int cpu);
2453#endif
2454};
2455
2456static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2457{
2458 WARN_ON_ONCE(rq->donor != prev);
2459 prev->sched_class->put_prev_task(rq, prev, NULL);
2460}
2461
2462static inline void set_next_task(struct rq *rq, struct task_struct *next)
2463{
2464 next->sched_class->set_next_task(rq, next, false);
2465}
2466
2467static inline void
2468__put_prev_set_next_dl_server(struct rq *rq,
2469 struct task_struct *prev,
2470 struct task_struct *next)
2471{
2472 prev->dl_server = NULL;
2473 next->dl_server = rq->dl_server;
2474 rq->dl_server = NULL;
2475}
2476
2477static inline void put_prev_set_next_task(struct rq *rq,
2478 struct task_struct *prev,
2479 struct task_struct *next)
2480{
2481 WARN_ON_ONCE(rq->curr != prev);
2482
2483 __put_prev_set_next_dl_server(rq, prev, next);
2484
2485 if (next == prev)
2486 return;
2487
2488 prev->sched_class->put_prev_task(rq, prev, next);
2489 next->sched_class->set_next_task(rq, next, true);
2490}
2491
2492/*
2493 * Helper to define a sched_class instance; each one is placed in a separate
2494 * section which is ordered by the linker script:
2495 *
2496 * include/asm-generic/vmlinux.lds.h
2497 *
2498 * *CAREFUL* they are laid out in *REVERSE* order!!!
2499 *
2500 * Also enforce alignment on the instance, not the type, to guarantee layout.
2501 */
2502#define DEFINE_SCHED_CLASS(name) \
2503const struct sched_class name##_sched_class \
2504 __aligned(__alignof__(struct sched_class)) \
2505 __section("__" #name "_sched_class")
2506
2507/* Defined in include/asm-generic/vmlinux.lds.h */
2508extern struct sched_class __sched_class_highest[];
2509extern struct sched_class __sched_class_lowest[];
2510
2511extern const struct sched_class stop_sched_class;
2512extern const struct sched_class dl_sched_class;
2513extern const struct sched_class rt_sched_class;
2514extern const struct sched_class fair_sched_class;
2515extern const struct sched_class idle_sched_class;
2516
2517#ifdef CONFIG_SCHED_CLASS_EXT
2518extern const struct sched_class ext_sched_class;
2519
2520DECLARE_STATIC_KEY_FALSE(__scx_ops_enabled); /* SCX BPF scheduler loaded */
2521DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
2522
2523#define scx_enabled() static_branch_unlikely(&__scx_ops_enabled)
2524#define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
2525#else /* !CONFIG_SCHED_CLASS_EXT */
2526#define scx_enabled() false
2527#define scx_switched_all() false
2528#endif /* !CONFIG_SCHED_CLASS_EXT */
2529
2530/*
2531 * Iterate only active classes. SCX can take over all fair tasks or be
2532 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2533 */
2534static inline const struct sched_class *next_active_class(const struct sched_class *class)
2535{
2536 class++;
2537#ifdef CONFIG_SCHED_CLASS_EXT
2538 if (scx_switched_all() && class == &fair_sched_class)
2539 class++;
2540 if (!scx_enabled() && class == &ext_sched_class)
2541 class++;
2542#endif
2543 return class;
2544}
2545
2546#define for_class_range(class, _from, _to) \
2547 for (class = (_from); class < (_to); class++)
2548
2549#define for_each_class(class) \
2550 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2551
2552#define for_active_class_range(class, _from, _to) \
2553 for (class = (_from); class != (_to); class = next_active_class(class))
2554
2555#define for_each_active_class(class) \
2556 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2557
2558#define sched_class_above(_a, _b) ((_a) < (_b))
2559
2560static inline bool sched_stop_runnable(struct rq *rq)
2561{
2562 return rq->stop && task_on_rq_queued(rq->stop);
2563}
2564
2565static inline bool sched_dl_runnable(struct rq *rq)
2566{
2567 return rq->dl.dl_nr_running > 0;
2568}
2569
2570static inline bool sched_rt_runnable(struct rq *rq)
2571{
2572 return rq->rt.rt_queued > 0;
2573}
2574
2575static inline bool sched_fair_runnable(struct rq *rq)
2576{
2577 return rq->cfs.nr_running > 0;
2578}
2579
2580extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2581extern struct task_struct *pick_task_idle(struct rq *rq);
2582
2583#define SCA_CHECK 0x01
2584#define SCA_MIGRATE_DISABLE 0x02
2585#define SCA_MIGRATE_ENABLE 0x04
2586#define SCA_USER 0x08
2587
2588#ifdef CONFIG_SMP
2589
2590extern void update_group_capacity(struct sched_domain *sd, int cpu);
2591
2592extern void sched_balance_trigger(struct rq *rq);
2593
2594extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2595extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2596
2597static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2598{
2599 /* When not in the task's cpumask, no point in looking further. */
2600 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2601 return false;
2602
2603 /* Can @cpu run a user thread? */
2604 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2605 return false;
2606
2607 return true;
2608}
2609
2610static inline cpumask_t *alloc_user_cpus_ptr(int node)
2611{
2612 /*
2613 * See do_set_cpus_allowed() above for the rcu_head usage.
2614 */
2615 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2616
2617 return kmalloc_node(size, GFP_KERNEL, node);
2618}
2619
2620static inline struct task_struct *get_push_task(struct rq *rq)
2621{
2622 struct task_struct *p = rq->donor;
2623
2624 lockdep_assert_rq_held(rq);
2625
2626 if (rq->push_busy)
2627 return NULL;
2628
2629 if (p->nr_cpus_allowed == 1)
2630 return NULL;
2631
2632 if (p->migration_disabled)
2633 return NULL;
2634
2635 rq->push_busy = true;
2636 return get_task_struct(p);
2637}
2638
2639extern int push_cpu_stop(void *arg);
2640
2641#else /* !CONFIG_SMP: */
2642
2643static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2644{
2645 return true;
2646}
2647
2648static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2649 struct affinity_context *ctx)
2650{
2651 return set_cpus_allowed_ptr(p, ctx->new_mask);
2652}
2653
2654static inline cpumask_t *alloc_user_cpus_ptr(int node)
2655{
2656 return NULL;
2657}
2658
2659#endif /* !CONFIG_SMP */
2660
2661#ifdef CONFIG_CPU_IDLE
2662
2663static inline void idle_set_state(struct rq *rq,
2664 struct cpuidle_state *idle_state)
2665{
2666 rq->idle_state = idle_state;
2667}
2668
2669static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2670{
2671 SCHED_WARN_ON(!rcu_read_lock_held());
2672
2673 return rq->idle_state;
2674}
2675
2676#else /* !CONFIG_CPU_IDLE: */
2677
2678static inline void idle_set_state(struct rq *rq,
2679 struct cpuidle_state *idle_state)
2680{
2681}
2682
2683static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2684{
2685 return NULL;
2686}
2687
2688#endif /* !CONFIG_CPU_IDLE */
2689
2690extern void schedule_idle(void);
2691asmlinkage void schedule_user(void);
2692
2693extern void sysrq_sched_debug_show(void);
2694extern void sched_init_granularity(void);
2695extern void update_max_interval(void);
2696
2697extern void init_sched_dl_class(void);
2698extern void init_sched_rt_class(void);
2699extern void init_sched_fair_class(void);
2700
2701extern void resched_curr(struct rq *rq);
2702extern void resched_curr_lazy(struct rq *rq);
2703extern void resched_cpu(int cpu);
2704
2705extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2706extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2707
2708extern void init_dl_entity(struct sched_dl_entity *dl_se);
2709
2710#define BW_SHIFT 20
2711#define BW_UNIT (1 << BW_SHIFT)
2712#define RATIO_SHIFT 8
2713#define MAX_BW_BITS (64 - BW_SHIFT)
2714#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2715
2716extern unsigned long to_ratio(u64 period, u64 runtime);
2717
2718extern void init_entity_runnable_average(struct sched_entity *se);
2719extern void post_init_entity_util_avg(struct task_struct *p);
2720
2721#ifdef CONFIG_NO_HZ_FULL
2722extern bool sched_can_stop_tick(struct rq *rq);
2723extern int __init sched_tick_offload_init(void);
2724
2725/*
2726 * Tick may be needed by tasks in the runqueue depending on their policy and
2727 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2728 * nohz mode if necessary.
2729 */
2730static inline void sched_update_tick_dependency(struct rq *rq)
2731{
2732 int cpu = cpu_of(rq);
2733
2734 if (!tick_nohz_full_cpu(cpu))
2735 return;
2736
2737 if (sched_can_stop_tick(rq))
2738 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2739 else
2740 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2741}
2742#else /* !CONFIG_NO_HZ_FULL: */
2743static inline int sched_tick_offload_init(void) { return 0; }
2744static inline void sched_update_tick_dependency(struct rq *rq) { }
2745#endif /* !CONFIG_NO_HZ_FULL */
2746
2747static inline void add_nr_running(struct rq *rq, unsigned count)
2748{
2749 unsigned prev_nr = rq->nr_running;
2750
2751 rq->nr_running = prev_nr + count;
2752 if (trace_sched_update_nr_running_tp_enabled()) {
2753 call_trace_sched_update_nr_running(rq, count);
2754 }
2755
2756#ifdef CONFIG_SMP
2757 if (prev_nr < 2 && rq->nr_running >= 2)
2758 set_rd_overloaded(rq->rd, 1);
2759#endif
2760
2761 sched_update_tick_dependency(rq);
2762}
2763
2764static inline void sub_nr_running(struct rq *rq, unsigned count)
2765{
2766 rq->nr_running -= count;
2767 if (trace_sched_update_nr_running_tp_enabled()) {
2768 call_trace_sched_update_nr_running(rq, -count);
2769 }
2770
2771 /* Check if we still need preemption */
2772 sched_update_tick_dependency(rq);
2773}
2774
2775static inline void __block_task(struct rq *rq, struct task_struct *p)
2776{
2777 if (p->sched_contributes_to_load)
2778 rq->nr_uninterruptible++;
2779
2780 if (p->in_iowait) {
2781 atomic_inc(&rq->nr_iowait);
2782 delayacct_blkio_start();
2783 }
2784
2785 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2786
2787 /*
2788 * The moment this write goes through, ttwu() can swoop in and migrate
2789 * this task, rendering our rq->__lock ineffective.
2790 *
2791 * __schedule() try_to_wake_up()
2792 * LOCK rq->__lock LOCK p->pi_lock
2793 * pick_next_task()
2794 * pick_next_task_fair()
2795 * pick_next_entity()
2796 * dequeue_entities()
2797 * __block_task()
2798 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2799 * break;
2800 *
2801 * ACQUIRE (after ctrl-dep)
2802 *
2803 * cpu = select_task_rq();
2804 * set_task_cpu(p, cpu);
2805 * ttwu_queue()
2806 * ttwu_do_activate()
2807 * LOCK rq->__lock
2808 * activate_task()
2809 * STORE p->on_rq = 1
2810 * UNLOCK rq->__lock
2811 *
2812 * Callers must ensure to not reference @p after this -- we no longer
2813 * own it.
2814 */
2815 smp_store_release(&p->on_rq, 0);
2816}
2817
2818extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2819extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2820
2821extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2822
2823#ifdef CONFIG_PREEMPT_RT
2824# define SCHED_NR_MIGRATE_BREAK 8
2825#else
2826# define SCHED_NR_MIGRATE_BREAK 32
2827#endif
2828
2829extern const_debug unsigned int sysctl_sched_nr_migrate;
2830extern const_debug unsigned int sysctl_sched_migration_cost;
2831
2832extern unsigned int sysctl_sched_base_slice;
2833
2834#ifdef CONFIG_SCHED_DEBUG
2835extern int sysctl_resched_latency_warn_ms;
2836extern int sysctl_resched_latency_warn_once;
2837
2838extern unsigned int sysctl_sched_tunable_scaling;
2839
2840extern unsigned int sysctl_numa_balancing_scan_delay;
2841extern unsigned int sysctl_numa_balancing_scan_period_min;
2842extern unsigned int sysctl_numa_balancing_scan_period_max;
2843extern unsigned int sysctl_numa_balancing_scan_size;
2844extern unsigned int sysctl_numa_balancing_hot_threshold;
2845#endif
2846
2847#ifdef CONFIG_SCHED_HRTICK
2848
2849/*
2850 * Use hrtick when:
2851 * - enabled by features
2852 * - hrtimer is actually high res
2853 */
2854static inline int hrtick_enabled(struct rq *rq)
2855{
2856 if (!cpu_active(cpu_of(rq)))
2857 return 0;
2858 return hrtimer_is_hres_active(&rq->hrtick_timer);
2859}
2860
2861static inline int hrtick_enabled_fair(struct rq *rq)
2862{
2863 if (!sched_feat(HRTICK))
2864 return 0;
2865 return hrtick_enabled(rq);
2866}
2867
2868static inline int hrtick_enabled_dl(struct rq *rq)
2869{
2870 if (!sched_feat(HRTICK_DL))
2871 return 0;
2872 return hrtick_enabled(rq);
2873}
2874
2875extern void hrtick_start(struct rq *rq, u64 delay);
2876
2877#else /* !CONFIG_SCHED_HRTICK: */
2878
2879static inline int hrtick_enabled_fair(struct rq *rq)
2880{
2881 return 0;
2882}
2883
2884static inline int hrtick_enabled_dl(struct rq *rq)
2885{
2886 return 0;
2887}
2888
2889static inline int hrtick_enabled(struct rq *rq)
2890{
2891 return 0;
2892}
2893
2894#endif /* !CONFIG_SCHED_HRTICK */
2895
2896#ifndef arch_scale_freq_tick
2897static __always_inline void arch_scale_freq_tick(void) { }
2898#endif
2899
2900#ifndef arch_scale_freq_capacity
2901/**
2902 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2903 * @cpu: the CPU in question.
2904 *
2905 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2906 *
2907 * f_curr
2908 * ------ * SCHED_CAPACITY_SCALE
2909 * f_max
2910 */
2911static __always_inline
2912unsigned long arch_scale_freq_capacity(int cpu)
2913{
2914 return SCHED_CAPACITY_SCALE;
2915}
2916#endif
2917
2918#ifdef CONFIG_SCHED_DEBUG
2919/*
2920 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2921 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2922 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2923 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2924 */
2925static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2926{
2927 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2928 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2929#ifdef CONFIG_SMP
2930 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2931#endif
2932}
2933#else
2934static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) { }
2935#endif
2936
2937#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2938__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2939static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2940{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2941 _lock; return _t; }
2942
2943#ifdef CONFIG_SMP
2944
2945static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2946{
2947#ifdef CONFIG_SCHED_CORE
2948 /*
2949 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2950 * order by core-id first and cpu-id second.
2951 *
2952 * Notably:
2953 *
2954 * double_rq_lock(0,3); will take core-0, core-1 lock
2955 * double_rq_lock(1,2); will take core-1, core-0 lock
2956 *
2957 * when only cpu-id is considered.
2958 */
2959 if (rq1->core->cpu < rq2->core->cpu)
2960 return true;
2961 if (rq1->core->cpu > rq2->core->cpu)
2962 return false;
2963
2964 /*
2965 * __sched_core_flip() relies on SMT having cpu-id lock order.
2966 */
2967#endif
2968 return rq1->cpu < rq2->cpu;
2969}
2970
2971extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2972
2973#ifdef CONFIG_PREEMPTION
2974
2975/*
2976 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2977 * way at the expense of forcing extra atomic operations in all
2978 * invocations. This assures that the double_lock is acquired using the
2979 * same underlying policy as the spinlock_t on this architecture, which
2980 * reduces latency compared to the unfair variant below. However, it
2981 * also adds more overhead and therefore may reduce throughput.
2982 */
2983static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2984 __releases(this_rq->lock)
2985 __acquires(busiest->lock)
2986 __acquires(this_rq->lock)
2987{
2988 raw_spin_rq_unlock(this_rq);
2989 double_rq_lock(this_rq, busiest);
2990
2991 return 1;
2992}
2993
2994#else /* !CONFIG_PREEMPTION: */
2995/*
2996 * Unfair double_lock_balance: Optimizes throughput at the expense of
2997 * latency by eliminating extra atomic operations when the locks are
2998 * already in proper order on entry. This favors lower CPU-ids and will
2999 * grant the double lock to lower CPUs over higher ids under contention,
3000 * regardless of entry order into the function.
3001 */
3002static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3003 __releases(this_rq->lock)
3004 __acquires(busiest->lock)
3005 __acquires(this_rq->lock)
3006{
3007 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
3008 likely(raw_spin_rq_trylock(busiest))) {
3009 double_rq_clock_clear_update(this_rq, busiest);
3010 return 0;
3011 }
3012
3013 if (rq_order_less(this_rq, busiest)) {
3014 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3015 double_rq_clock_clear_update(this_rq, busiest);
3016 return 0;
3017 }
3018
3019 raw_spin_rq_unlock(this_rq);
3020 double_rq_lock(this_rq, busiest);
3021
3022 return 1;
3023}
3024
3025#endif /* !CONFIG_PREEMPTION */
3026
3027/*
3028 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3029 */
3030static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3031{
3032 lockdep_assert_irqs_disabled();
3033
3034 return _double_lock_balance(this_rq, busiest);
3035}
3036
3037static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3038 __releases(busiest->lock)
3039{
3040 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
3041 raw_spin_rq_unlock(busiest);
3042 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3043}
3044
3045static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3046{
3047 if (l1 > l2)
3048 swap(l1, l2);
3049
3050 spin_lock(l1);
3051 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3052}
3053
3054static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3055{
3056 if (l1 > l2)
3057 swap(l1, l2);
3058
3059 spin_lock_irq(l1);
3060 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3061}
3062
3063static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3064{
3065 if (l1 > l2)
3066 swap(l1, l2);
3067
3068 raw_spin_lock(l1);
3069 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3070}
3071
3072static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3073{
3074 raw_spin_unlock(l1);
3075 raw_spin_unlock(l2);
3076}
3077
3078DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3079 double_raw_lock(_T->lock, _T->lock2),
3080 double_raw_unlock(_T->lock, _T->lock2))
3081
3082/*
3083 * double_rq_unlock - safely unlock two runqueues
3084 *
3085 * Note this does not restore interrupts like task_rq_unlock,
3086 * you need to do so manually after calling.
3087 */
3088static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3089 __releases(rq1->lock)
3090 __releases(rq2->lock)
3091{
3092 if (__rq_lockp(rq1) != __rq_lockp(rq2))
3093 raw_spin_rq_unlock(rq2);
3094 else
3095 __release(rq2->lock);
3096 raw_spin_rq_unlock(rq1);
3097}
3098
3099extern void set_rq_online (struct rq *rq);
3100extern void set_rq_offline(struct rq *rq);
3101
3102extern bool sched_smp_initialized;
3103
3104#else /* !CONFIG_SMP: */
3105
3106/*
3107 * double_rq_lock - safely lock two runqueues
3108 *
3109 * Note this does not disable interrupts like task_rq_lock,
3110 * you need to do so manually before calling.
3111 */
3112static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
3113 __acquires(rq1->lock)
3114 __acquires(rq2->lock)
3115{
3116 WARN_ON_ONCE(!irqs_disabled());
3117 WARN_ON_ONCE(rq1 != rq2);
3118 raw_spin_rq_lock(rq1);
3119 __acquire(rq2->lock); /* Fake it out ;) */
3120 double_rq_clock_clear_update(rq1, rq2);
3121}
3122
3123/*
3124 * double_rq_unlock - safely unlock two runqueues
3125 *
3126 * Note this does not restore interrupts like task_rq_unlock,
3127 * you need to do so manually after calling.
3128 */
3129static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3130 __releases(rq1->lock)
3131 __releases(rq2->lock)
3132{
3133 WARN_ON_ONCE(rq1 != rq2);
3134 raw_spin_rq_unlock(rq1);
3135 __release(rq2->lock);
3136}
3137
3138#endif /* !CONFIG_SMP */
3139
3140DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3141 double_rq_lock(_T->lock, _T->lock2),
3142 double_rq_unlock(_T->lock, _T->lock2))
3143
3144extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3145extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3146extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3147
3148#ifdef CONFIG_SCHED_DEBUG
3149extern bool sched_debug_verbose;
3150
3151extern void print_cfs_stats(struct seq_file *m, int cpu);
3152extern void print_rt_stats(struct seq_file *m, int cpu);
3153extern void print_dl_stats(struct seq_file *m, int cpu);
3154extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3155extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3156extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3157
3158extern void resched_latency_warn(int cpu, u64 latency);
3159# ifdef CONFIG_NUMA_BALANCING
3160extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3161extern void
3162print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3163 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3164# endif /* CONFIG_NUMA_BALANCING */
3165#else /* !CONFIG_SCHED_DEBUG: */
3166static inline void resched_latency_warn(int cpu, u64 latency) { }
3167#endif /* !CONFIG_SCHED_DEBUG */
3168
3169extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3170extern void init_rt_rq(struct rt_rq *rt_rq);
3171extern void init_dl_rq(struct dl_rq *dl_rq);
3172
3173extern void cfs_bandwidth_usage_inc(void);
3174extern void cfs_bandwidth_usage_dec(void);
3175
3176#ifdef CONFIG_NO_HZ_COMMON
3177
3178#define NOHZ_BALANCE_KICK_BIT 0
3179#define NOHZ_STATS_KICK_BIT 1
3180#define NOHZ_NEWILB_KICK_BIT 2
3181#define NOHZ_NEXT_KICK_BIT 3
3182
3183/* Run sched_balance_domains() */
3184#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3185/* Update blocked load */
3186#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3187/* Update blocked load when entering idle */
3188#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3189/* Update nohz.next_balance */
3190#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3191
3192#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3193
3194#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3195
3196extern void nohz_balance_exit_idle(struct rq *rq);
3197#else /* !CONFIG_NO_HZ_COMMON: */
3198static inline void nohz_balance_exit_idle(struct rq *rq) { }
3199#endif /* !CONFIG_NO_HZ_COMMON */
3200
3201#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
3202extern void nohz_run_idle_balance(int cpu);
3203#else
3204static inline void nohz_run_idle_balance(int cpu) { }
3205#endif
3206
3207#include "stats.h"
3208
3209#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3210
3211extern void __sched_core_account_forceidle(struct rq *rq);
3212
3213static inline void sched_core_account_forceidle(struct rq *rq)
3214{
3215 if (schedstat_enabled())
3216 __sched_core_account_forceidle(rq);
3217}
3218
3219extern void __sched_core_tick(struct rq *rq);
3220
3221static inline void sched_core_tick(struct rq *rq)
3222{
3223 if (sched_core_enabled(rq) && schedstat_enabled())
3224 __sched_core_tick(rq);
3225}
3226
3227#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3228
3229static inline void sched_core_account_forceidle(struct rq *rq) { }
3230
3231static inline void sched_core_tick(struct rq *rq) { }
3232
3233#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3234
3235#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3236
3237struct irqtime {
3238 u64 total;
3239 u64 tick_delta;
3240 u64 irq_start_time;
3241 struct u64_stats_sync sync;
3242};
3243
3244DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3245
3246/*
3247 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3248 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3249 * and never move forward.
3250 */
3251static inline u64 irq_time_read(int cpu)
3252{
3253 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3254 unsigned int seq;
3255 u64 total;
3256
3257 do {
3258 seq = __u64_stats_fetch_begin(&irqtime->sync);
3259 total = irqtime->total;
3260 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3261
3262 return total;
3263}
3264
3265#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3266
3267#ifdef CONFIG_CPU_FREQ
3268
3269DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3270
3271/**
3272 * cpufreq_update_util - Take a note about CPU utilization changes.
3273 * @rq: Runqueue to carry out the update for.
3274 * @flags: Update reason flags.
3275 *
3276 * This function is called by the scheduler on the CPU whose utilization is
3277 * being updated.
3278 *
3279 * It can only be called from RCU-sched read-side critical sections.
3280 *
3281 * The way cpufreq is currently arranged requires it to evaluate the CPU
3282 * performance state (frequency/voltage) on a regular basis to prevent it from
3283 * being stuck in a completely inadequate performance level for too long.
3284 * That is not guaranteed to happen if the updates are only triggered from CFS
3285 * and DL, though, because they may not be coming in if only RT tasks are
3286 * active all the time (or there are RT tasks only).
3287 *
3288 * As a workaround for that issue, this function is called periodically by the
3289 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3290 * but that really is a band-aid. Going forward it should be replaced with
3291 * solutions targeted more specifically at RT tasks.
3292 */
3293static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3294{
3295 struct update_util_data *data;
3296
3297 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3298 cpu_of(rq)));
3299 if (data)
3300 data->func(data, rq_clock(rq), flags);
3301}
3302#else /* !CONFIG_CPU_FREQ: */
3303static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3304#endif /* !CONFIG_CPU_FREQ */
3305
3306#ifdef arch_scale_freq_capacity
3307# ifndef arch_scale_freq_invariant
3308# define arch_scale_freq_invariant() true
3309# endif
3310#else
3311# define arch_scale_freq_invariant() false
3312#endif
3313
3314#ifdef CONFIG_SMP
3315
3316unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3317 unsigned long *min,
3318 unsigned long *max);
3319
3320unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3321 unsigned long min,
3322 unsigned long max);
3323
3324
3325/*
3326 * Verify the fitness of task @p to run on @cpu taking into account the
3327 * CPU original capacity and the runtime/deadline ratio of the task.
3328 *
3329 * The function will return true if the original capacity of @cpu is
3330 * greater than or equal to task's deadline density right shifted by
3331 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3332 */
3333static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3334{
3335 unsigned long cap = arch_scale_cpu_capacity(cpu);
3336
3337 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3338}
3339
3340static inline unsigned long cpu_bw_dl(struct rq *rq)
3341{
3342 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3343}
3344
3345static inline unsigned long cpu_util_dl(struct rq *rq)
3346{
3347 return READ_ONCE(rq->avg_dl.util_avg);
3348}
3349
3350
3351extern unsigned long cpu_util_cfs(int cpu);
3352extern unsigned long cpu_util_cfs_boost(int cpu);
3353
3354static inline unsigned long cpu_util_rt(struct rq *rq)
3355{
3356 return READ_ONCE(rq->avg_rt.util_avg);
3357}
3358
3359#else /* !CONFIG_SMP */
3360static inline bool update_other_load_avgs(struct rq *rq) { return false; }
3361#endif /* CONFIG_SMP */
3362
3363#ifdef CONFIG_UCLAMP_TASK
3364
3365unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3366
3367static inline unsigned long uclamp_rq_get(struct rq *rq,
3368 enum uclamp_id clamp_id)
3369{
3370 return READ_ONCE(rq->uclamp[clamp_id].value);
3371}
3372
3373static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3374 unsigned int value)
3375{
3376 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3377}
3378
3379static inline bool uclamp_rq_is_idle(struct rq *rq)
3380{
3381 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3382}
3383
3384/* Is the rq being capped/throttled by uclamp_max? */
3385static inline bool uclamp_rq_is_capped(struct rq *rq)
3386{
3387 unsigned long rq_util;
3388 unsigned long max_util;
3389
3390 if (!static_branch_likely(&sched_uclamp_used))
3391 return false;
3392
3393 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3394 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3395
3396 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3397}
3398
3399/*
3400 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3401 * by default in the fast path and only gets turned on once userspace performs
3402 * an operation that requires it.
3403 *
3404 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3405 * hence is active.
3406 */
3407static inline bool uclamp_is_used(void)
3408{
3409 return static_branch_likely(&sched_uclamp_used);
3410}
3411
3412#define for_each_clamp_id(clamp_id) \
3413 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3414
3415extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3416
3417
3418static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3419{
3420 if (clamp_id == UCLAMP_MIN)
3421 return 0;
3422 return SCHED_CAPACITY_SCALE;
3423}
3424
3425/* Integer rounded range for each bucket */
3426#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3427
3428static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3429{
3430 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3431}
3432
3433static inline void
3434uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3435{
3436 uc_se->value = value;
3437 uc_se->bucket_id = uclamp_bucket_id(value);
3438 uc_se->user_defined = user_defined;
3439}
3440
3441#else /* !CONFIG_UCLAMP_TASK: */
3442
3443static inline unsigned long
3444uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3445{
3446 if (clamp_id == UCLAMP_MIN)
3447 return 0;
3448
3449 return SCHED_CAPACITY_SCALE;
3450}
3451
3452static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3453
3454static inline bool uclamp_is_used(void)
3455{
3456 return false;
3457}
3458
3459static inline unsigned long
3460uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3461{
3462 if (clamp_id == UCLAMP_MIN)
3463 return 0;
3464
3465 return SCHED_CAPACITY_SCALE;
3466}
3467
3468static inline void
3469uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3470{
3471}
3472
3473static inline bool uclamp_rq_is_idle(struct rq *rq)
3474{
3475 return false;
3476}
3477
3478#endif /* !CONFIG_UCLAMP_TASK */
3479
3480#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3481
3482static inline unsigned long cpu_util_irq(struct rq *rq)
3483{
3484 return READ_ONCE(rq->avg_irq.util_avg);
3485}
3486
3487static inline
3488unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3489{
3490 util *= (max - irq);
3491 util /= max;
3492
3493 return util;
3494
3495}
3496
3497#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3498
3499static inline unsigned long cpu_util_irq(struct rq *rq)
3500{
3501 return 0;
3502}
3503
3504static inline
3505unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3506{
3507 return util;
3508}
3509
3510#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3511
3512#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3513
3514#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3515
3516DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3517
3518static inline bool sched_energy_enabled(void)
3519{
3520 return static_branch_unlikely(&sched_energy_present);
3521}
3522
3523extern struct cpufreq_governor schedutil_gov;
3524
3525#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3526
3527#define perf_domain_span(pd) NULL
3528
3529static inline bool sched_energy_enabled(void) { return false; }
3530
3531#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3532
3533#ifdef CONFIG_MEMBARRIER
3534
3535/*
3536 * The scheduler provides memory barriers required by membarrier between:
3537 * - prior user-space memory accesses and store to rq->membarrier_state,
3538 * - store to rq->membarrier_state and following user-space memory accesses.
3539 * In the same way it provides those guarantees around store to rq->curr.
3540 */
3541static inline void membarrier_switch_mm(struct rq *rq,
3542 struct mm_struct *prev_mm,
3543 struct mm_struct *next_mm)
3544{
3545 int membarrier_state;
3546
3547 if (prev_mm == next_mm)
3548 return;
3549
3550 membarrier_state = atomic_read(&next_mm->membarrier_state);
3551 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3552 return;
3553
3554 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3555}
3556
3557#else /* !CONFIG_MEMBARRIER :*/
3558
3559static inline void membarrier_switch_mm(struct rq *rq,
3560 struct mm_struct *prev_mm,
3561 struct mm_struct *next_mm)
3562{
3563}
3564
3565#endif /* !CONFIG_MEMBARRIER */
3566
3567#ifdef CONFIG_SMP
3568static inline bool is_per_cpu_kthread(struct task_struct *p)
3569{
3570 if (!(p->flags & PF_KTHREAD))
3571 return false;
3572
3573 if (p->nr_cpus_allowed != 1)
3574 return false;
3575
3576 return true;
3577}
3578#endif
3579
3580extern void swake_up_all_locked(struct swait_queue_head *q);
3581extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3582
3583extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3584
3585#ifdef CONFIG_PREEMPT_DYNAMIC
3586extern int preempt_dynamic_mode;
3587extern int sched_dynamic_mode(const char *str);
3588extern void sched_dynamic_update(int mode);
3589#endif
3590
3591#ifdef CONFIG_SCHED_MM_CID
3592
3593#define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3594#define MM_CID_SCAN_DELAY 100 /* 100ms */
3595
3596extern raw_spinlock_t cid_lock;
3597extern int use_cid_lock;
3598
3599extern void sched_mm_cid_migrate_from(struct task_struct *t);
3600extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3601extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3602extern void init_sched_mm_cid(struct task_struct *t);
3603
3604static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3605{
3606 if (cid < 0)
3607 return;
3608 cpumask_clear_cpu(cid, mm_cidmask(mm));
3609}
3610
3611/*
3612 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3613 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3614 * be held to transition to other states.
3615 *
3616 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3617 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3618 */
3619static inline void mm_cid_put_lazy(struct task_struct *t)
3620{
3621 struct mm_struct *mm = t->mm;
3622 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3623 int cid;
3624
3625 lockdep_assert_irqs_disabled();
3626 cid = __this_cpu_read(pcpu_cid->cid);
3627 if (!mm_cid_is_lazy_put(cid) ||
3628 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3629 return;
3630 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3631}
3632
3633static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3634{
3635 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3636 int cid, res;
3637
3638 lockdep_assert_irqs_disabled();
3639 cid = __this_cpu_read(pcpu_cid->cid);
3640 for (;;) {
3641 if (mm_cid_is_unset(cid))
3642 return MM_CID_UNSET;
3643 /*
3644 * Attempt transition from valid or lazy-put to unset.
3645 */
3646 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3647 if (res == cid)
3648 break;
3649 cid = res;
3650 }
3651 return cid;
3652}
3653
3654static inline void mm_cid_put(struct mm_struct *mm)
3655{
3656 int cid;
3657
3658 lockdep_assert_irqs_disabled();
3659 cid = mm_cid_pcpu_unset(mm);
3660 if (cid == MM_CID_UNSET)
3661 return;
3662 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3663}
3664
3665static inline int __mm_cid_try_get(struct task_struct *t, struct mm_struct *mm)
3666{
3667 struct cpumask *cidmask = mm_cidmask(mm);
3668 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3669 int cid, max_nr_cid, allowed_max_nr_cid;
3670
3671 /*
3672 * After shrinking the number of threads or reducing the number
3673 * of allowed cpus, reduce the value of max_nr_cid so expansion
3674 * of cid allocation will preserve cache locality if the number
3675 * of threads or allowed cpus increase again.
3676 */
3677 max_nr_cid = atomic_read(&mm->max_nr_cid);
3678 while ((allowed_max_nr_cid = min_t(int, READ_ONCE(mm->nr_cpus_allowed),
3679 atomic_read(&mm->mm_users))),
3680 max_nr_cid > allowed_max_nr_cid) {
3681 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into max_nr_cid. */
3682 if (atomic_try_cmpxchg(&mm->max_nr_cid, &max_nr_cid, allowed_max_nr_cid)) {
3683 max_nr_cid = allowed_max_nr_cid;
3684 break;
3685 }
3686 }
3687 /* Try to re-use recent cid. This improves cache locality. */
3688 cid = __this_cpu_read(pcpu_cid->recent_cid);
3689 if (!mm_cid_is_unset(cid) && cid < max_nr_cid &&
3690 !cpumask_test_and_set_cpu(cid, cidmask))
3691 return cid;
3692 /*
3693 * Expand cid allocation if the maximum number of concurrency
3694 * IDs allocated (max_nr_cid) is below the number cpus allowed
3695 * and number of threads. Expanding cid allocation as much as
3696 * possible improves cache locality.
3697 */
3698 cid = max_nr_cid;
3699 while (cid < READ_ONCE(mm->nr_cpus_allowed) && cid < atomic_read(&mm->mm_users)) {
3700 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into cid. */
3701 if (!atomic_try_cmpxchg(&mm->max_nr_cid, &cid, cid + 1))
3702 continue;
3703 if (!cpumask_test_and_set_cpu(cid, cidmask))
3704 return cid;
3705 }
3706 /*
3707 * Find the first available concurrency id.
3708 * Retry finding first zero bit if the mask is temporarily
3709 * filled. This only happens during concurrent remote-clear
3710 * which owns a cid without holding a rq lock.
3711 */
3712 for (;;) {
3713 cid = cpumask_first_zero(cidmask);
3714 if (cid < READ_ONCE(mm->nr_cpus_allowed))
3715 break;
3716 cpu_relax();
3717 }
3718 if (cpumask_test_and_set_cpu(cid, cidmask))
3719 return -1;
3720
3721 return cid;
3722}
3723
3724/*
3725 * Save a snapshot of the current runqueue time of this cpu
3726 * with the per-cpu cid value, allowing to estimate how recently it was used.
3727 */
3728static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3729{
3730 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3731
3732 lockdep_assert_rq_held(rq);
3733 WRITE_ONCE(pcpu_cid->time, rq->clock);
3734}
3735
3736static inline int __mm_cid_get(struct rq *rq, struct task_struct *t,
3737 struct mm_struct *mm)
3738{
3739 int cid;
3740
3741 /*
3742 * All allocations (even those using the cid_lock) are lock-free. If
3743 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3744 * guarantee forward progress.
3745 */
3746 if (!READ_ONCE(use_cid_lock)) {
3747 cid = __mm_cid_try_get(t, mm);
3748 if (cid >= 0)
3749 goto end;
3750 raw_spin_lock(&cid_lock);
3751 } else {
3752 raw_spin_lock(&cid_lock);
3753 cid = __mm_cid_try_get(t, mm);
3754 if (cid >= 0)
3755 goto unlock;
3756 }
3757
3758 /*
3759 * cid concurrently allocated. Retry while forcing following
3760 * allocations to use the cid_lock to ensure forward progress.
3761 */
3762 WRITE_ONCE(use_cid_lock, 1);
3763 /*
3764 * Set use_cid_lock before allocation. Only care about program order
3765 * because this is only required for forward progress.
3766 */
3767 barrier();
3768 /*
3769 * Retry until it succeeds. It is guaranteed to eventually succeed once
3770 * all newcoming allocations observe the use_cid_lock flag set.
3771 */
3772 do {
3773 cid = __mm_cid_try_get(t, mm);
3774 cpu_relax();
3775 } while (cid < 0);
3776 /*
3777 * Allocate before clearing use_cid_lock. Only care about
3778 * program order because this is for forward progress.
3779 */
3780 barrier();
3781 WRITE_ONCE(use_cid_lock, 0);
3782unlock:
3783 raw_spin_unlock(&cid_lock);
3784end:
3785 mm_cid_snapshot_time(rq, mm);
3786
3787 return cid;
3788}
3789
3790static inline int mm_cid_get(struct rq *rq, struct task_struct *t,
3791 struct mm_struct *mm)
3792{
3793 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3794 struct cpumask *cpumask;
3795 int cid;
3796
3797 lockdep_assert_rq_held(rq);
3798 cpumask = mm_cidmask(mm);
3799 cid = __this_cpu_read(pcpu_cid->cid);
3800 if (mm_cid_is_valid(cid)) {
3801 mm_cid_snapshot_time(rq, mm);
3802 return cid;
3803 }
3804 if (mm_cid_is_lazy_put(cid)) {
3805 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3806 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3807 }
3808 cid = __mm_cid_get(rq, t, mm);
3809 __this_cpu_write(pcpu_cid->cid, cid);
3810 __this_cpu_write(pcpu_cid->recent_cid, cid);
3811
3812 return cid;
3813}
3814
3815static inline void switch_mm_cid(struct rq *rq,
3816 struct task_struct *prev,
3817 struct task_struct *next)
3818{
3819 /*
3820 * Provide a memory barrier between rq->curr store and load of
3821 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3822 *
3823 * Should be adapted if context_switch() is modified.
3824 */
3825 if (!next->mm) { // to kernel
3826 /*
3827 * user -> kernel transition does not guarantee a barrier, but
3828 * we can use the fact that it performs an atomic operation in
3829 * mmgrab().
3830 */
3831 if (prev->mm) // from user
3832 smp_mb__after_mmgrab();
3833 /*
3834 * kernel -> kernel transition does not change rq->curr->mm
3835 * state. It stays NULL.
3836 */
3837 } else { // to user
3838 /*
3839 * kernel -> user transition does not provide a barrier
3840 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3841 * Provide it here.
3842 */
3843 if (!prev->mm) { // from kernel
3844 smp_mb();
3845 } else { // from user
3846 /*
3847 * user->user transition relies on an implicit
3848 * memory barrier in switch_mm() when
3849 * current->mm changes. If the architecture
3850 * switch_mm() does not have an implicit memory
3851 * barrier, it is emitted here. If current->mm
3852 * is unchanged, no barrier is needed.
3853 */
3854 smp_mb__after_switch_mm();
3855 }
3856 }
3857 if (prev->mm_cid_active) {
3858 mm_cid_snapshot_time(rq, prev->mm);
3859 mm_cid_put_lazy(prev);
3860 prev->mm_cid = -1;
3861 }
3862 if (next->mm_cid_active)
3863 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next, next->mm);
3864}
3865
3866#else /* !CONFIG_SCHED_MM_CID: */
3867static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3868static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3869static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3870static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3871static inline void init_sched_mm_cid(struct task_struct *t) { }
3872#endif /* !CONFIG_SCHED_MM_CID */
3873
3874extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3875extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3876#ifdef CONFIG_SMP
3877static inline
3878void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
3879{
3880 lockdep_assert_rq_held(src_rq);
3881 lockdep_assert_rq_held(dst_rq);
3882
3883 deactivate_task(src_rq, task, 0);
3884 set_task_cpu(task, dst_rq->cpu);
3885 activate_task(dst_rq, task, 0);
3886}
3887
3888static inline
3889bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
3890{
3891 if (!task_on_cpu(rq, p) &&
3892 cpumask_test_cpu(cpu, &p->cpus_mask))
3893 return true;
3894
3895 return false;
3896}
3897#endif
3898
3899#ifdef CONFIG_RT_MUTEXES
3900
3901static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3902{
3903 if (pi_task)
3904 prio = min(prio, pi_task->prio);
3905
3906 return prio;
3907}
3908
3909static inline int rt_effective_prio(struct task_struct *p, int prio)
3910{
3911 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3912
3913 return __rt_effective_prio(pi_task, prio);
3914}
3915
3916#else /* !CONFIG_RT_MUTEXES: */
3917
3918static inline int rt_effective_prio(struct task_struct *p, int prio)
3919{
3920 return prio;
3921}
3922
3923#endif /* !CONFIG_RT_MUTEXES */
3924
3925extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3926extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3927extern const struct sched_class *__setscheduler_class(int policy, int prio);
3928extern void set_load_weight(struct task_struct *p, bool update_load);
3929extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3930extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3931
3932extern void check_class_changing(struct rq *rq, struct task_struct *p,
3933 const struct sched_class *prev_class);
3934extern void check_class_changed(struct rq *rq, struct task_struct *p,
3935 const struct sched_class *prev_class,
3936 int oldprio);
3937
3938#ifdef CONFIG_SMP
3939extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3940extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3941#else
3942
3943static inline struct balance_callback *splice_balance_callbacks(struct rq *rq)
3944{
3945 return NULL;
3946}
3947
3948static inline void balance_callbacks(struct rq *rq, struct balance_callback *head)
3949{
3950}
3951
3952#endif
3953
3954#ifdef CONFIG_SCHED_CLASS_EXT
3955/*
3956 * Used by SCX in the enable/disable paths to move tasks between sched_classes
3957 * and establish invariants.
3958 */
3959struct sched_enq_and_set_ctx {
3960 struct task_struct *p;
3961 int queue_flags;
3962 bool queued;
3963 bool running;
3964};
3965
3966void sched_deq_and_put_task(struct task_struct *p, int queue_flags,
3967 struct sched_enq_and_set_ctx *ctx);
3968void sched_enq_and_set_task(struct sched_enq_and_set_ctx *ctx);
3969
3970#endif /* CONFIG_SCHED_CLASS_EXT */
3971
3972#include "ext.h"
3973
3974#endif /* _KERNEL_SCHED_SCHED_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Scheduler internal types and methods:
4 */
5#include <linux/sched.h>
6
7#include <linux/sched/autogroup.h>
8#include <linux/sched/clock.h>
9#include <linux/sched/coredump.h>
10#include <linux/sched/cpufreq.h>
11#include <linux/sched/cputime.h>
12#include <linux/sched/deadline.h>
13#include <linux/sched/debug.h>
14#include <linux/sched/hotplug.h>
15#include <linux/sched/idle.h>
16#include <linux/sched/init.h>
17#include <linux/sched/isolation.h>
18#include <linux/sched/jobctl.h>
19#include <linux/sched/loadavg.h>
20#include <linux/sched/mm.h>
21#include <linux/sched/nohz.h>
22#include <linux/sched/numa_balancing.h>
23#include <linux/sched/prio.h>
24#include <linux/sched/rt.h>
25#include <linux/sched/signal.h>
26#include <linux/sched/smt.h>
27#include <linux/sched/stat.h>
28#include <linux/sched/sysctl.h>
29#include <linux/sched/task.h>
30#include <linux/sched/task_stack.h>
31#include <linux/sched/topology.h>
32#include <linux/sched/user.h>
33#include <linux/sched/wake_q.h>
34#include <linux/sched/xacct.h>
35
36#include <uapi/linux/sched/types.h>
37
38#include <linux/binfmts.h>
39#include <linux/bitops.h>
40#include <linux/blkdev.h>
41#include <linux/compat.h>
42#include <linux/context_tracking.h>
43#include <linux/cpufreq.h>
44#include <linux/cpuidle.h>
45#include <linux/cpuset.h>
46#include <linux/ctype.h>
47#include <linux/debugfs.h>
48#include <linux/delayacct.h>
49#include <linux/energy_model.h>
50#include <linux/init_task.h>
51#include <linux/kprobes.h>
52#include <linux/kthread.h>
53#include <linux/membarrier.h>
54#include <linux/migrate.h>
55#include <linux/mmu_context.h>
56#include <linux/nmi.h>
57#include <linux/proc_fs.h>
58#include <linux/prefetch.h>
59#include <linux/profile.h>
60#include <linux/psi.h>
61#include <linux/ratelimit.h>
62#include <linux/rcupdate_wait.h>
63#include <linux/security.h>
64#include <linux/stop_machine.h>
65#include <linux/suspend.h>
66#include <linux/swait.h>
67#include <linux/syscalls.h>
68#include <linux/task_work.h>
69#include <linux/tsacct_kern.h>
70
71#include <asm/tlb.h>
72
73#ifdef CONFIG_PARAVIRT
74# include <asm/paravirt.h>
75#endif
76
77#include "cpupri.h"
78#include "cpudeadline.h"
79
80#include <trace/events/sched.h>
81
82#ifdef CONFIG_SCHED_DEBUG
83# define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
84#else
85# define SCHED_WARN_ON(x) ({ (void)(x), 0; })
86#endif
87
88struct rq;
89struct cpuidle_state;
90
91/* task_struct::on_rq states: */
92#define TASK_ON_RQ_QUEUED 1
93#define TASK_ON_RQ_MIGRATING 2
94
95extern __read_mostly int scheduler_running;
96
97extern unsigned long calc_load_update;
98extern atomic_long_t calc_load_tasks;
99
100extern void calc_global_load_tick(struct rq *this_rq);
101extern long calc_load_fold_active(struct rq *this_rq, long adjust);
102
103extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
104/*
105 * Helpers for converting nanosecond timing to jiffy resolution
106 */
107#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108
109/*
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
115 *
116 * We increase resolution only if we have enough bits to allow this increased
117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
119 *
120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
122 */
123#ifdef CONFIG_64BIT
124# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
126# define scale_load_down(w) \
127({ \
128 unsigned long __w = (w); \
129 if (__w) \
130 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
131 __w; \
132})
133#else
134# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
135# define scale_load(w) (w)
136# define scale_load_down(w) (w)
137#endif
138
139/*
140 * Task weight (visible to users) and its load (invisible to users) have
141 * independent resolution, but they should be well calibrated. We use
142 * scale_load() and scale_load_down(w) to convert between them. The
143 * following must be true:
144 *
145 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
146 *
147 */
148#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
149
150/*
151 * Single value that decides SCHED_DEADLINE internal math precision.
152 * 10 -> just above 1us
153 * 9 -> just above 0.5us
154 */
155#define DL_SCALE 10
156
157/*
158 * Single value that denotes runtime == period, ie unlimited time.
159 */
160#define RUNTIME_INF ((u64)~0ULL)
161
162static inline int idle_policy(int policy)
163{
164 return policy == SCHED_IDLE;
165}
166static inline int fair_policy(int policy)
167{
168 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
169}
170
171static inline int rt_policy(int policy)
172{
173 return policy == SCHED_FIFO || policy == SCHED_RR;
174}
175
176static inline int dl_policy(int policy)
177{
178 return policy == SCHED_DEADLINE;
179}
180static inline bool valid_policy(int policy)
181{
182 return idle_policy(policy) || fair_policy(policy) ||
183 rt_policy(policy) || dl_policy(policy);
184}
185
186static inline int task_has_idle_policy(struct task_struct *p)
187{
188 return idle_policy(p->policy);
189}
190
191static inline int task_has_rt_policy(struct task_struct *p)
192{
193 return rt_policy(p->policy);
194}
195
196static inline int task_has_dl_policy(struct task_struct *p)
197{
198 return dl_policy(p->policy);
199}
200
201#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
202
203static inline void update_avg(u64 *avg, u64 sample)
204{
205 s64 diff = sample - *avg;
206 *avg += diff / 8;
207}
208
209/*
210 * Shifting a value by an exponent greater *or equal* to the size of said value
211 * is UB; cap at size-1.
212 */
213#define shr_bound(val, shift) \
214 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
215
216/*
217 * !! For sched_setattr_nocheck() (kernel) only !!
218 *
219 * This is actually gross. :(
220 *
221 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
222 * tasks, but still be able to sleep. We need this on platforms that cannot
223 * atomically change clock frequency. Remove once fast switching will be
224 * available on such platforms.
225 *
226 * SUGOV stands for SchedUtil GOVernor.
227 */
228#define SCHED_FLAG_SUGOV 0x10000000
229
230#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
231
232static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
233{
234#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
235 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
236#else
237 return false;
238#endif
239}
240
241/*
242 * Tells if entity @a should preempt entity @b.
243 */
244static inline bool
245dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
246{
247 return dl_entity_is_special(a) ||
248 dl_time_before(a->deadline, b->deadline);
249}
250
251/*
252 * This is the priority-queue data structure of the RT scheduling class:
253 */
254struct rt_prio_array {
255 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
256 struct list_head queue[MAX_RT_PRIO];
257};
258
259struct rt_bandwidth {
260 /* nests inside the rq lock: */
261 raw_spinlock_t rt_runtime_lock;
262 ktime_t rt_period;
263 u64 rt_runtime;
264 struct hrtimer rt_period_timer;
265 unsigned int rt_period_active;
266};
267
268void __dl_clear_params(struct task_struct *p);
269
270struct dl_bandwidth {
271 raw_spinlock_t dl_runtime_lock;
272 u64 dl_runtime;
273 u64 dl_period;
274};
275
276static inline int dl_bandwidth_enabled(void)
277{
278 return sysctl_sched_rt_runtime >= 0;
279}
280
281/*
282 * To keep the bandwidth of -deadline tasks under control
283 * we need some place where:
284 * - store the maximum -deadline bandwidth of each cpu;
285 * - cache the fraction of bandwidth that is currently allocated in
286 * each root domain;
287 *
288 * This is all done in the data structure below. It is similar to the
289 * one used for RT-throttling (rt_bandwidth), with the main difference
290 * that, since here we are only interested in admission control, we
291 * do not decrease any runtime while the group "executes", neither we
292 * need a timer to replenish it.
293 *
294 * With respect to SMP, bandwidth is given on a per root domain basis,
295 * meaning that:
296 * - bw (< 100%) is the deadline bandwidth of each CPU;
297 * - total_bw is the currently allocated bandwidth in each root domain;
298 */
299struct dl_bw {
300 raw_spinlock_t lock;
301 u64 bw;
302 u64 total_bw;
303};
304
305static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
306
307static inline
308void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
309{
310 dl_b->total_bw -= tsk_bw;
311 __dl_update(dl_b, (s32)tsk_bw / cpus);
312}
313
314static inline
315void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
316{
317 dl_b->total_bw += tsk_bw;
318 __dl_update(dl_b, -((s32)tsk_bw / cpus));
319}
320
321static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
322 u64 old_bw, u64 new_bw)
323{
324 return dl_b->bw != -1 &&
325 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
326}
327
328/*
329 * Verify the fitness of task @p to run on @cpu taking into account the
330 * CPU original capacity and the runtime/deadline ratio of the task.
331 *
332 * The function will return true if the CPU original capacity of the
333 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
334 * task and false otherwise.
335 */
336static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
337{
338 unsigned long cap = arch_scale_cpu_capacity(cpu);
339
340 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
341}
342
343extern void init_dl_bw(struct dl_bw *dl_b);
344extern int sched_dl_global_validate(void);
345extern void sched_dl_do_global(void);
346extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
347extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
348extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
349extern bool __checkparam_dl(const struct sched_attr *attr);
350extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
351extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
352extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
353extern bool dl_cpu_busy(unsigned int cpu);
354
355#ifdef CONFIG_CGROUP_SCHED
356
357#include <linux/cgroup.h>
358#include <linux/psi.h>
359
360struct cfs_rq;
361struct rt_rq;
362
363extern struct list_head task_groups;
364
365struct cfs_bandwidth {
366#ifdef CONFIG_CFS_BANDWIDTH
367 raw_spinlock_t lock;
368 ktime_t period;
369 u64 quota;
370 u64 runtime;
371 u64 burst;
372 s64 hierarchical_quota;
373
374 u8 idle;
375 u8 period_active;
376 u8 slack_started;
377 struct hrtimer period_timer;
378 struct hrtimer slack_timer;
379 struct list_head throttled_cfs_rq;
380
381 /* Statistics: */
382 int nr_periods;
383 int nr_throttled;
384 u64 throttled_time;
385#endif
386};
387
388/* Task group related information */
389struct task_group {
390 struct cgroup_subsys_state css;
391
392#ifdef CONFIG_FAIR_GROUP_SCHED
393 /* schedulable entities of this group on each CPU */
394 struct sched_entity **se;
395 /* runqueue "owned" by this group on each CPU */
396 struct cfs_rq **cfs_rq;
397 unsigned long shares;
398
399#ifdef CONFIG_SMP
400 /*
401 * load_avg can be heavily contended at clock tick time, so put
402 * it in its own cacheline separated from the fields above which
403 * will also be accessed at each tick.
404 */
405 atomic_long_t load_avg ____cacheline_aligned;
406#endif
407#endif
408
409#ifdef CONFIG_RT_GROUP_SCHED
410 struct sched_rt_entity **rt_se;
411 struct rt_rq **rt_rq;
412
413 struct rt_bandwidth rt_bandwidth;
414#endif
415
416 struct rcu_head rcu;
417 struct list_head list;
418
419 struct task_group *parent;
420 struct list_head siblings;
421 struct list_head children;
422
423#ifdef CONFIG_SCHED_AUTOGROUP
424 struct autogroup *autogroup;
425#endif
426
427 struct cfs_bandwidth cfs_bandwidth;
428
429#ifdef CONFIG_UCLAMP_TASK_GROUP
430 /* The two decimal precision [%] value requested from user-space */
431 unsigned int uclamp_pct[UCLAMP_CNT];
432 /* Clamp values requested for a task group */
433 struct uclamp_se uclamp_req[UCLAMP_CNT];
434 /* Effective clamp values used for a task group */
435 struct uclamp_se uclamp[UCLAMP_CNT];
436#endif
437
438};
439
440#ifdef CONFIG_FAIR_GROUP_SCHED
441#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
442
443/*
444 * A weight of 0 or 1 can cause arithmetics problems.
445 * A weight of a cfs_rq is the sum of weights of which entities
446 * are queued on this cfs_rq, so a weight of a entity should not be
447 * too large, so as the shares value of a task group.
448 * (The default weight is 1024 - so there's no practical
449 * limitation from this.)
450 */
451#define MIN_SHARES (1UL << 1)
452#define MAX_SHARES (1UL << 18)
453#endif
454
455typedef int (*tg_visitor)(struct task_group *, void *);
456
457extern int walk_tg_tree_from(struct task_group *from,
458 tg_visitor down, tg_visitor up, void *data);
459
460/*
461 * Iterate the full tree, calling @down when first entering a node and @up when
462 * leaving it for the final time.
463 *
464 * Caller must hold rcu_lock or sufficient equivalent.
465 */
466static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
467{
468 return walk_tg_tree_from(&root_task_group, down, up, data);
469}
470
471extern int tg_nop(struct task_group *tg, void *data);
472
473extern void free_fair_sched_group(struct task_group *tg);
474extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
475extern void online_fair_sched_group(struct task_group *tg);
476extern void unregister_fair_sched_group(struct task_group *tg);
477extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
478 struct sched_entity *se, int cpu,
479 struct sched_entity *parent);
480extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
481
482extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
483extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
484extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
485
486extern void free_rt_sched_group(struct task_group *tg);
487extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
488extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
489 struct sched_rt_entity *rt_se, int cpu,
490 struct sched_rt_entity *parent);
491extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
492extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
493extern long sched_group_rt_runtime(struct task_group *tg);
494extern long sched_group_rt_period(struct task_group *tg);
495extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
496
497extern struct task_group *sched_create_group(struct task_group *parent);
498extern void sched_online_group(struct task_group *tg,
499 struct task_group *parent);
500extern void sched_destroy_group(struct task_group *tg);
501extern void sched_offline_group(struct task_group *tg);
502
503extern void sched_move_task(struct task_struct *tsk);
504
505#ifdef CONFIG_FAIR_GROUP_SCHED
506extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
507
508#ifdef CONFIG_SMP
509extern void set_task_rq_fair(struct sched_entity *se,
510 struct cfs_rq *prev, struct cfs_rq *next);
511#else /* !CONFIG_SMP */
512static inline void set_task_rq_fair(struct sched_entity *se,
513 struct cfs_rq *prev, struct cfs_rq *next) { }
514#endif /* CONFIG_SMP */
515#endif /* CONFIG_FAIR_GROUP_SCHED */
516
517#else /* CONFIG_CGROUP_SCHED */
518
519struct cfs_bandwidth { };
520
521#endif /* CONFIG_CGROUP_SCHED */
522
523/* CFS-related fields in a runqueue */
524struct cfs_rq {
525 struct load_weight load;
526 unsigned int nr_running;
527 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
528 unsigned int idle_h_nr_running; /* SCHED_IDLE */
529
530 u64 exec_clock;
531 u64 min_vruntime;
532#ifdef CONFIG_SCHED_CORE
533 unsigned int forceidle_seq;
534 u64 min_vruntime_fi;
535#endif
536
537#ifndef CONFIG_64BIT
538 u64 min_vruntime_copy;
539#endif
540
541 struct rb_root_cached tasks_timeline;
542
543 /*
544 * 'curr' points to currently running entity on this cfs_rq.
545 * It is set to NULL otherwise (i.e when none are currently running).
546 */
547 struct sched_entity *curr;
548 struct sched_entity *next;
549 struct sched_entity *last;
550 struct sched_entity *skip;
551
552#ifdef CONFIG_SCHED_DEBUG
553 unsigned int nr_spread_over;
554#endif
555
556#ifdef CONFIG_SMP
557 /*
558 * CFS load tracking
559 */
560 struct sched_avg avg;
561#ifndef CONFIG_64BIT
562 u64 load_last_update_time_copy;
563#endif
564 struct {
565 raw_spinlock_t lock ____cacheline_aligned;
566 int nr;
567 unsigned long load_avg;
568 unsigned long util_avg;
569 unsigned long runnable_avg;
570 } removed;
571
572#ifdef CONFIG_FAIR_GROUP_SCHED
573 unsigned long tg_load_avg_contrib;
574 long propagate;
575 long prop_runnable_sum;
576
577 /*
578 * h_load = weight * f(tg)
579 *
580 * Where f(tg) is the recursive weight fraction assigned to
581 * this group.
582 */
583 unsigned long h_load;
584 u64 last_h_load_update;
585 struct sched_entity *h_load_next;
586#endif /* CONFIG_FAIR_GROUP_SCHED */
587#endif /* CONFIG_SMP */
588
589#ifdef CONFIG_FAIR_GROUP_SCHED
590 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
591
592 /*
593 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
594 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
595 * (like users, containers etc.)
596 *
597 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
598 * This list is used during load balance.
599 */
600 int on_list;
601 struct list_head leaf_cfs_rq_list;
602 struct task_group *tg; /* group that "owns" this runqueue */
603
604#ifdef CONFIG_CFS_BANDWIDTH
605 int runtime_enabled;
606 s64 runtime_remaining;
607
608 u64 throttled_clock;
609 u64 throttled_clock_task;
610 u64 throttled_clock_task_time;
611 int throttled;
612 int throttle_count;
613 struct list_head throttled_list;
614#endif /* CONFIG_CFS_BANDWIDTH */
615#endif /* CONFIG_FAIR_GROUP_SCHED */
616};
617
618static inline int rt_bandwidth_enabled(void)
619{
620 return sysctl_sched_rt_runtime >= 0;
621}
622
623/* RT IPI pull logic requires IRQ_WORK */
624#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
625# define HAVE_RT_PUSH_IPI
626#endif
627
628/* Real-Time classes' related field in a runqueue: */
629struct rt_rq {
630 struct rt_prio_array active;
631 unsigned int rt_nr_running;
632 unsigned int rr_nr_running;
633#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
634 struct {
635 int curr; /* highest queued rt task prio */
636#ifdef CONFIG_SMP
637 int next; /* next highest */
638#endif
639 } highest_prio;
640#endif
641#ifdef CONFIG_SMP
642 unsigned int rt_nr_migratory;
643 unsigned int rt_nr_total;
644 int overloaded;
645 struct plist_head pushable_tasks;
646
647#endif /* CONFIG_SMP */
648 int rt_queued;
649
650 int rt_throttled;
651 u64 rt_time;
652 u64 rt_runtime;
653 /* Nests inside the rq lock: */
654 raw_spinlock_t rt_runtime_lock;
655
656#ifdef CONFIG_RT_GROUP_SCHED
657 unsigned int rt_nr_boosted;
658
659 struct rq *rq;
660 struct task_group *tg;
661#endif
662};
663
664static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
665{
666 return rt_rq->rt_queued && rt_rq->rt_nr_running;
667}
668
669/* Deadline class' related fields in a runqueue */
670struct dl_rq {
671 /* runqueue is an rbtree, ordered by deadline */
672 struct rb_root_cached root;
673
674 unsigned int dl_nr_running;
675
676#ifdef CONFIG_SMP
677 /*
678 * Deadline values of the currently executing and the
679 * earliest ready task on this rq. Caching these facilitates
680 * the decision whether or not a ready but not running task
681 * should migrate somewhere else.
682 */
683 struct {
684 u64 curr;
685 u64 next;
686 } earliest_dl;
687
688 unsigned int dl_nr_migratory;
689 int overloaded;
690
691 /*
692 * Tasks on this rq that can be pushed away. They are kept in
693 * an rb-tree, ordered by tasks' deadlines, with caching
694 * of the leftmost (earliest deadline) element.
695 */
696 struct rb_root_cached pushable_dl_tasks_root;
697#else
698 struct dl_bw dl_bw;
699#endif
700 /*
701 * "Active utilization" for this runqueue: increased when a
702 * task wakes up (becomes TASK_RUNNING) and decreased when a
703 * task blocks
704 */
705 u64 running_bw;
706
707 /*
708 * Utilization of the tasks "assigned" to this runqueue (including
709 * the tasks that are in runqueue and the tasks that executed on this
710 * CPU and blocked). Increased when a task moves to this runqueue, and
711 * decreased when the task moves away (migrates, changes scheduling
712 * policy, or terminates).
713 * This is needed to compute the "inactive utilization" for the
714 * runqueue (inactive utilization = this_bw - running_bw).
715 */
716 u64 this_bw;
717 u64 extra_bw;
718
719 /*
720 * Inverse of the fraction of CPU utilization that can be reclaimed
721 * by the GRUB algorithm.
722 */
723 u64 bw_ratio;
724};
725
726#ifdef CONFIG_FAIR_GROUP_SCHED
727/* An entity is a task if it doesn't "own" a runqueue */
728#define entity_is_task(se) (!se->my_q)
729
730static inline void se_update_runnable(struct sched_entity *se)
731{
732 if (!entity_is_task(se))
733 se->runnable_weight = se->my_q->h_nr_running;
734}
735
736static inline long se_runnable(struct sched_entity *se)
737{
738 if (entity_is_task(se))
739 return !!se->on_rq;
740 else
741 return se->runnable_weight;
742}
743
744#else
745#define entity_is_task(se) 1
746
747static inline void se_update_runnable(struct sched_entity *se) {}
748
749static inline long se_runnable(struct sched_entity *se)
750{
751 return !!se->on_rq;
752}
753#endif
754
755#ifdef CONFIG_SMP
756/*
757 * XXX we want to get rid of these helpers and use the full load resolution.
758 */
759static inline long se_weight(struct sched_entity *se)
760{
761 return scale_load_down(se->load.weight);
762}
763
764
765static inline bool sched_asym_prefer(int a, int b)
766{
767 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
768}
769
770struct perf_domain {
771 struct em_perf_domain *em_pd;
772 struct perf_domain *next;
773 struct rcu_head rcu;
774};
775
776/* Scheduling group status flags */
777#define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
778#define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
779
780/*
781 * We add the notion of a root-domain which will be used to define per-domain
782 * variables. Each exclusive cpuset essentially defines an island domain by
783 * fully partitioning the member CPUs from any other cpuset. Whenever a new
784 * exclusive cpuset is created, we also create and attach a new root-domain
785 * object.
786 *
787 */
788struct root_domain {
789 atomic_t refcount;
790 atomic_t rto_count;
791 struct rcu_head rcu;
792 cpumask_var_t span;
793 cpumask_var_t online;
794
795 /*
796 * Indicate pullable load on at least one CPU, e.g:
797 * - More than one runnable task
798 * - Running task is misfit
799 */
800 int overload;
801
802 /* Indicate one or more cpus over-utilized (tipping point) */
803 int overutilized;
804
805 /*
806 * The bit corresponding to a CPU gets set here if such CPU has more
807 * than one runnable -deadline task (as it is below for RT tasks).
808 */
809 cpumask_var_t dlo_mask;
810 atomic_t dlo_count;
811 struct dl_bw dl_bw;
812 struct cpudl cpudl;
813
814 /*
815 * Indicate whether a root_domain's dl_bw has been checked or
816 * updated. It's monotonously increasing value.
817 *
818 * Also, some corner cases, like 'wrap around' is dangerous, but given
819 * that u64 is 'big enough'. So that shouldn't be a concern.
820 */
821 u64 visit_gen;
822
823#ifdef HAVE_RT_PUSH_IPI
824 /*
825 * For IPI pull requests, loop across the rto_mask.
826 */
827 struct irq_work rto_push_work;
828 raw_spinlock_t rto_lock;
829 /* These are only updated and read within rto_lock */
830 int rto_loop;
831 int rto_cpu;
832 /* These atomics are updated outside of a lock */
833 atomic_t rto_loop_next;
834 atomic_t rto_loop_start;
835#endif
836 /*
837 * The "RT overload" flag: it gets set if a CPU has more than
838 * one runnable RT task.
839 */
840 cpumask_var_t rto_mask;
841 struct cpupri cpupri;
842
843 unsigned long max_cpu_capacity;
844
845 /*
846 * NULL-terminated list of performance domains intersecting with the
847 * CPUs of the rd. Protected by RCU.
848 */
849 struct perf_domain __rcu *pd;
850};
851
852extern void init_defrootdomain(void);
853extern int sched_init_domains(const struct cpumask *cpu_map);
854extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
855extern void sched_get_rd(struct root_domain *rd);
856extern void sched_put_rd(struct root_domain *rd);
857
858#ifdef HAVE_RT_PUSH_IPI
859extern void rto_push_irq_work_func(struct irq_work *work);
860#endif
861#endif /* CONFIG_SMP */
862
863#ifdef CONFIG_UCLAMP_TASK
864/*
865 * struct uclamp_bucket - Utilization clamp bucket
866 * @value: utilization clamp value for tasks on this clamp bucket
867 * @tasks: number of RUNNABLE tasks on this clamp bucket
868 *
869 * Keep track of how many tasks are RUNNABLE for a given utilization
870 * clamp value.
871 */
872struct uclamp_bucket {
873 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
874 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
875};
876
877/*
878 * struct uclamp_rq - rq's utilization clamp
879 * @value: currently active clamp values for a rq
880 * @bucket: utilization clamp buckets affecting a rq
881 *
882 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
883 * A clamp value is affecting a rq when there is at least one task RUNNABLE
884 * (or actually running) with that value.
885 *
886 * There are up to UCLAMP_CNT possible different clamp values, currently there
887 * are only two: minimum utilization and maximum utilization.
888 *
889 * All utilization clamping values are MAX aggregated, since:
890 * - for util_min: we want to run the CPU at least at the max of the minimum
891 * utilization required by its currently RUNNABLE tasks.
892 * - for util_max: we want to allow the CPU to run up to the max of the
893 * maximum utilization allowed by its currently RUNNABLE tasks.
894 *
895 * Since on each system we expect only a limited number of different
896 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
897 * the metrics required to compute all the per-rq utilization clamp values.
898 */
899struct uclamp_rq {
900 unsigned int value;
901 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
902};
903
904DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
905#endif /* CONFIG_UCLAMP_TASK */
906
907/*
908 * This is the main, per-CPU runqueue data structure.
909 *
910 * Locking rule: those places that want to lock multiple runqueues
911 * (such as the load balancing or the thread migration code), lock
912 * acquire operations must be ordered by ascending &runqueue.
913 */
914struct rq {
915 /* runqueue lock: */
916 raw_spinlock_t __lock;
917
918 /*
919 * nr_running and cpu_load should be in the same cacheline because
920 * remote CPUs use both these fields when doing load calculation.
921 */
922 unsigned int nr_running;
923#ifdef CONFIG_NUMA_BALANCING
924 unsigned int nr_numa_running;
925 unsigned int nr_preferred_running;
926 unsigned int numa_migrate_on;
927#endif
928#ifdef CONFIG_NO_HZ_COMMON
929#ifdef CONFIG_SMP
930 unsigned long last_blocked_load_update_tick;
931 unsigned int has_blocked_load;
932 call_single_data_t nohz_csd;
933#endif /* CONFIG_SMP */
934 unsigned int nohz_tick_stopped;
935 atomic_t nohz_flags;
936#endif /* CONFIG_NO_HZ_COMMON */
937
938#ifdef CONFIG_SMP
939 unsigned int ttwu_pending;
940#endif
941 u64 nr_switches;
942
943#ifdef CONFIG_UCLAMP_TASK
944 /* Utilization clamp values based on CPU's RUNNABLE tasks */
945 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
946 unsigned int uclamp_flags;
947#define UCLAMP_FLAG_IDLE 0x01
948#endif
949
950 struct cfs_rq cfs;
951 struct rt_rq rt;
952 struct dl_rq dl;
953
954#ifdef CONFIG_FAIR_GROUP_SCHED
955 /* list of leaf cfs_rq on this CPU: */
956 struct list_head leaf_cfs_rq_list;
957 struct list_head *tmp_alone_branch;
958#endif /* CONFIG_FAIR_GROUP_SCHED */
959
960 /*
961 * This is part of a global counter where only the total sum
962 * over all CPUs matters. A task can increase this counter on
963 * one CPU and if it got migrated afterwards it may decrease
964 * it on another CPU. Always updated under the runqueue lock:
965 */
966 unsigned int nr_uninterruptible;
967
968 struct task_struct __rcu *curr;
969 struct task_struct *idle;
970 struct task_struct *stop;
971 unsigned long next_balance;
972 struct mm_struct *prev_mm;
973
974 unsigned int clock_update_flags;
975 u64 clock;
976 /* Ensure that all clocks are in the same cache line */
977 u64 clock_task ____cacheline_aligned;
978 u64 clock_pelt;
979 unsigned long lost_idle_time;
980
981 atomic_t nr_iowait;
982
983#ifdef CONFIG_SCHED_DEBUG
984 u64 last_seen_need_resched_ns;
985 int ticks_without_resched;
986#endif
987
988#ifdef CONFIG_MEMBARRIER
989 int membarrier_state;
990#endif
991
992#ifdef CONFIG_SMP
993 struct root_domain *rd;
994 struct sched_domain __rcu *sd;
995
996 unsigned long cpu_capacity;
997 unsigned long cpu_capacity_orig;
998
999 struct callback_head *balance_callback;
1000
1001 unsigned char nohz_idle_balance;
1002 unsigned char idle_balance;
1003
1004 unsigned long misfit_task_load;
1005
1006 /* For active balancing */
1007 int active_balance;
1008 int push_cpu;
1009 struct cpu_stop_work active_balance_work;
1010
1011 /* CPU of this runqueue: */
1012 int cpu;
1013 int online;
1014
1015 struct list_head cfs_tasks;
1016
1017 struct sched_avg avg_rt;
1018 struct sched_avg avg_dl;
1019#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1020 struct sched_avg avg_irq;
1021#endif
1022#ifdef CONFIG_SCHED_THERMAL_PRESSURE
1023 struct sched_avg avg_thermal;
1024#endif
1025 u64 idle_stamp;
1026 u64 avg_idle;
1027
1028 unsigned long wake_stamp;
1029 u64 wake_avg_idle;
1030
1031 /* This is used to determine avg_idle's max value */
1032 u64 max_idle_balance_cost;
1033
1034#ifdef CONFIG_HOTPLUG_CPU
1035 struct rcuwait hotplug_wait;
1036#endif
1037#endif /* CONFIG_SMP */
1038
1039#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1040 u64 prev_irq_time;
1041#endif
1042#ifdef CONFIG_PARAVIRT
1043 u64 prev_steal_time;
1044#endif
1045#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1046 u64 prev_steal_time_rq;
1047#endif
1048
1049 /* calc_load related fields */
1050 unsigned long calc_load_update;
1051 long calc_load_active;
1052
1053#ifdef CONFIG_SCHED_HRTICK
1054#ifdef CONFIG_SMP
1055 call_single_data_t hrtick_csd;
1056#endif
1057 struct hrtimer hrtick_timer;
1058 ktime_t hrtick_time;
1059#endif
1060
1061#ifdef CONFIG_SCHEDSTATS
1062 /* latency stats */
1063 struct sched_info rq_sched_info;
1064 unsigned long long rq_cpu_time;
1065 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1066
1067 /* sys_sched_yield() stats */
1068 unsigned int yld_count;
1069
1070 /* schedule() stats */
1071 unsigned int sched_count;
1072 unsigned int sched_goidle;
1073
1074 /* try_to_wake_up() stats */
1075 unsigned int ttwu_count;
1076 unsigned int ttwu_local;
1077#endif
1078
1079#ifdef CONFIG_CPU_IDLE
1080 /* Must be inspected within a rcu lock section */
1081 struct cpuidle_state *idle_state;
1082#endif
1083
1084#ifdef CONFIG_SMP
1085 unsigned int nr_pinned;
1086#endif
1087 unsigned int push_busy;
1088 struct cpu_stop_work push_work;
1089
1090#ifdef CONFIG_SCHED_CORE
1091 /* per rq */
1092 struct rq *core;
1093 struct task_struct *core_pick;
1094 unsigned int core_enabled;
1095 unsigned int core_sched_seq;
1096 struct rb_root core_tree;
1097
1098 /* shared state -- careful with sched_core_cpu_deactivate() */
1099 unsigned int core_task_seq;
1100 unsigned int core_pick_seq;
1101 unsigned long core_cookie;
1102 unsigned char core_forceidle;
1103 unsigned int core_forceidle_seq;
1104#endif
1105};
1106
1107#ifdef CONFIG_FAIR_GROUP_SCHED
1108
1109/* CPU runqueue to which this cfs_rq is attached */
1110static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1111{
1112 return cfs_rq->rq;
1113}
1114
1115#else
1116
1117static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1118{
1119 return container_of(cfs_rq, struct rq, cfs);
1120}
1121#endif
1122
1123static inline int cpu_of(struct rq *rq)
1124{
1125#ifdef CONFIG_SMP
1126 return rq->cpu;
1127#else
1128 return 0;
1129#endif
1130}
1131
1132#define MDF_PUSH 0x01
1133
1134static inline bool is_migration_disabled(struct task_struct *p)
1135{
1136#ifdef CONFIG_SMP
1137 return p->migration_disabled;
1138#else
1139 return false;
1140#endif
1141}
1142
1143struct sched_group;
1144#ifdef CONFIG_SCHED_CORE
1145static inline struct cpumask *sched_group_span(struct sched_group *sg);
1146
1147DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1148
1149static inline bool sched_core_enabled(struct rq *rq)
1150{
1151 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1152}
1153
1154static inline bool sched_core_disabled(void)
1155{
1156 return !static_branch_unlikely(&__sched_core_enabled);
1157}
1158
1159/*
1160 * Be careful with this function; not for general use. The return value isn't
1161 * stable unless you actually hold a relevant rq->__lock.
1162 */
1163static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1164{
1165 if (sched_core_enabled(rq))
1166 return &rq->core->__lock;
1167
1168 return &rq->__lock;
1169}
1170
1171static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1172{
1173 if (rq->core_enabled)
1174 return &rq->core->__lock;
1175
1176 return &rq->__lock;
1177}
1178
1179bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1180
1181/*
1182 * Helpers to check if the CPU's core cookie matches with the task's cookie
1183 * when core scheduling is enabled.
1184 * A special case is that the task's cookie always matches with CPU's core
1185 * cookie if the CPU is in an idle core.
1186 */
1187static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1188{
1189 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1190 if (!sched_core_enabled(rq))
1191 return true;
1192
1193 return rq->core->core_cookie == p->core_cookie;
1194}
1195
1196static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1197{
1198 bool idle_core = true;
1199 int cpu;
1200
1201 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1202 if (!sched_core_enabled(rq))
1203 return true;
1204
1205 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1206 if (!available_idle_cpu(cpu)) {
1207 idle_core = false;
1208 break;
1209 }
1210 }
1211
1212 /*
1213 * A CPU in an idle core is always the best choice for tasks with
1214 * cookies.
1215 */
1216 return idle_core || rq->core->core_cookie == p->core_cookie;
1217}
1218
1219static inline bool sched_group_cookie_match(struct rq *rq,
1220 struct task_struct *p,
1221 struct sched_group *group)
1222{
1223 int cpu;
1224
1225 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1226 if (!sched_core_enabled(rq))
1227 return true;
1228
1229 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1230 if (sched_core_cookie_match(rq, p))
1231 return true;
1232 }
1233 return false;
1234}
1235
1236extern void queue_core_balance(struct rq *rq);
1237
1238static inline bool sched_core_enqueued(struct task_struct *p)
1239{
1240 return !RB_EMPTY_NODE(&p->core_node);
1241}
1242
1243extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1244extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1245
1246extern void sched_core_get(void);
1247extern void sched_core_put(void);
1248
1249extern unsigned long sched_core_alloc_cookie(void);
1250extern void sched_core_put_cookie(unsigned long cookie);
1251extern unsigned long sched_core_get_cookie(unsigned long cookie);
1252extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1253
1254#else /* !CONFIG_SCHED_CORE */
1255
1256static inline bool sched_core_enabled(struct rq *rq)
1257{
1258 return false;
1259}
1260
1261static inline bool sched_core_disabled(void)
1262{
1263 return true;
1264}
1265
1266static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1267{
1268 return &rq->__lock;
1269}
1270
1271static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1272{
1273 return &rq->__lock;
1274}
1275
1276static inline void queue_core_balance(struct rq *rq)
1277{
1278}
1279
1280static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1281{
1282 return true;
1283}
1284
1285static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1286{
1287 return true;
1288}
1289
1290static inline bool sched_group_cookie_match(struct rq *rq,
1291 struct task_struct *p,
1292 struct sched_group *group)
1293{
1294 return true;
1295}
1296#endif /* CONFIG_SCHED_CORE */
1297
1298static inline void lockdep_assert_rq_held(struct rq *rq)
1299{
1300 lockdep_assert_held(__rq_lockp(rq));
1301}
1302
1303extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1304extern bool raw_spin_rq_trylock(struct rq *rq);
1305extern void raw_spin_rq_unlock(struct rq *rq);
1306
1307static inline void raw_spin_rq_lock(struct rq *rq)
1308{
1309 raw_spin_rq_lock_nested(rq, 0);
1310}
1311
1312static inline void raw_spin_rq_lock_irq(struct rq *rq)
1313{
1314 local_irq_disable();
1315 raw_spin_rq_lock(rq);
1316}
1317
1318static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1319{
1320 raw_spin_rq_unlock(rq);
1321 local_irq_enable();
1322}
1323
1324static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1325{
1326 unsigned long flags;
1327 local_irq_save(flags);
1328 raw_spin_rq_lock(rq);
1329 return flags;
1330}
1331
1332static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1333{
1334 raw_spin_rq_unlock(rq);
1335 local_irq_restore(flags);
1336}
1337
1338#define raw_spin_rq_lock_irqsave(rq, flags) \
1339do { \
1340 flags = _raw_spin_rq_lock_irqsave(rq); \
1341} while (0)
1342
1343#ifdef CONFIG_SCHED_SMT
1344extern void __update_idle_core(struct rq *rq);
1345
1346static inline void update_idle_core(struct rq *rq)
1347{
1348 if (static_branch_unlikely(&sched_smt_present))
1349 __update_idle_core(rq);
1350}
1351
1352#else
1353static inline void update_idle_core(struct rq *rq) { }
1354#endif
1355
1356DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1357
1358#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1359#define this_rq() this_cpu_ptr(&runqueues)
1360#define task_rq(p) cpu_rq(task_cpu(p))
1361#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1362#define raw_rq() raw_cpu_ptr(&runqueues)
1363
1364#ifdef CONFIG_FAIR_GROUP_SCHED
1365static inline struct task_struct *task_of(struct sched_entity *se)
1366{
1367 SCHED_WARN_ON(!entity_is_task(se));
1368 return container_of(se, struct task_struct, se);
1369}
1370
1371static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1372{
1373 return p->se.cfs_rq;
1374}
1375
1376/* runqueue on which this entity is (to be) queued */
1377static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1378{
1379 return se->cfs_rq;
1380}
1381
1382/* runqueue "owned" by this group */
1383static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1384{
1385 return grp->my_q;
1386}
1387
1388#else
1389
1390static inline struct task_struct *task_of(struct sched_entity *se)
1391{
1392 return container_of(se, struct task_struct, se);
1393}
1394
1395static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1396{
1397 return &task_rq(p)->cfs;
1398}
1399
1400static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1401{
1402 struct task_struct *p = task_of(se);
1403 struct rq *rq = task_rq(p);
1404
1405 return &rq->cfs;
1406}
1407
1408/* runqueue "owned" by this group */
1409static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1410{
1411 return NULL;
1412}
1413#endif
1414
1415extern void update_rq_clock(struct rq *rq);
1416
1417static inline u64 __rq_clock_broken(struct rq *rq)
1418{
1419 return READ_ONCE(rq->clock);
1420}
1421
1422/*
1423 * rq::clock_update_flags bits
1424 *
1425 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1426 * call to __schedule(). This is an optimisation to avoid
1427 * neighbouring rq clock updates.
1428 *
1429 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1430 * in effect and calls to update_rq_clock() are being ignored.
1431 *
1432 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1433 * made to update_rq_clock() since the last time rq::lock was pinned.
1434 *
1435 * If inside of __schedule(), clock_update_flags will have been
1436 * shifted left (a left shift is a cheap operation for the fast path
1437 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1438 *
1439 * if (rq-clock_update_flags >= RQCF_UPDATED)
1440 *
1441 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1442 * one position though, because the next rq_unpin_lock() will shift it
1443 * back.
1444 */
1445#define RQCF_REQ_SKIP 0x01
1446#define RQCF_ACT_SKIP 0x02
1447#define RQCF_UPDATED 0x04
1448
1449static inline void assert_clock_updated(struct rq *rq)
1450{
1451 /*
1452 * The only reason for not seeing a clock update since the
1453 * last rq_pin_lock() is if we're currently skipping updates.
1454 */
1455 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1456}
1457
1458static inline u64 rq_clock(struct rq *rq)
1459{
1460 lockdep_assert_rq_held(rq);
1461 assert_clock_updated(rq);
1462
1463 return rq->clock;
1464}
1465
1466static inline u64 rq_clock_task(struct rq *rq)
1467{
1468 lockdep_assert_rq_held(rq);
1469 assert_clock_updated(rq);
1470
1471 return rq->clock_task;
1472}
1473
1474/**
1475 * By default the decay is the default pelt decay period.
1476 * The decay shift can change the decay period in
1477 * multiples of 32.
1478 * Decay shift Decay period(ms)
1479 * 0 32
1480 * 1 64
1481 * 2 128
1482 * 3 256
1483 * 4 512
1484 */
1485extern int sched_thermal_decay_shift;
1486
1487static inline u64 rq_clock_thermal(struct rq *rq)
1488{
1489 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1490}
1491
1492static inline void rq_clock_skip_update(struct rq *rq)
1493{
1494 lockdep_assert_rq_held(rq);
1495 rq->clock_update_flags |= RQCF_REQ_SKIP;
1496}
1497
1498/*
1499 * See rt task throttling, which is the only time a skip
1500 * request is canceled.
1501 */
1502static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1503{
1504 lockdep_assert_rq_held(rq);
1505 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1506}
1507
1508struct rq_flags {
1509 unsigned long flags;
1510 struct pin_cookie cookie;
1511#ifdef CONFIG_SCHED_DEBUG
1512 /*
1513 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1514 * current pin context is stashed here in case it needs to be
1515 * restored in rq_repin_lock().
1516 */
1517 unsigned int clock_update_flags;
1518#endif
1519};
1520
1521extern struct callback_head balance_push_callback;
1522
1523/*
1524 * Lockdep annotation that avoids accidental unlocks; it's like a
1525 * sticky/continuous lockdep_assert_held().
1526 *
1527 * This avoids code that has access to 'struct rq *rq' (basically everything in
1528 * the scheduler) from accidentally unlocking the rq if they do not also have a
1529 * copy of the (on-stack) 'struct rq_flags rf'.
1530 *
1531 * Also see Documentation/locking/lockdep-design.rst.
1532 */
1533static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1534{
1535 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1536
1537#ifdef CONFIG_SCHED_DEBUG
1538 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1539 rf->clock_update_flags = 0;
1540#ifdef CONFIG_SMP
1541 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1542#endif
1543#endif
1544}
1545
1546static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1547{
1548#ifdef CONFIG_SCHED_DEBUG
1549 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1550 rf->clock_update_flags = RQCF_UPDATED;
1551#endif
1552
1553 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1554}
1555
1556static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1557{
1558 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1559
1560#ifdef CONFIG_SCHED_DEBUG
1561 /*
1562 * Restore the value we stashed in @rf for this pin context.
1563 */
1564 rq->clock_update_flags |= rf->clock_update_flags;
1565#endif
1566}
1567
1568struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1569 __acquires(rq->lock);
1570
1571struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1572 __acquires(p->pi_lock)
1573 __acquires(rq->lock);
1574
1575static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1576 __releases(rq->lock)
1577{
1578 rq_unpin_lock(rq, rf);
1579 raw_spin_rq_unlock(rq);
1580}
1581
1582static inline void
1583task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1584 __releases(rq->lock)
1585 __releases(p->pi_lock)
1586{
1587 rq_unpin_lock(rq, rf);
1588 raw_spin_rq_unlock(rq);
1589 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1590}
1591
1592static inline void
1593rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1594 __acquires(rq->lock)
1595{
1596 raw_spin_rq_lock_irqsave(rq, rf->flags);
1597 rq_pin_lock(rq, rf);
1598}
1599
1600static inline void
1601rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1602 __acquires(rq->lock)
1603{
1604 raw_spin_rq_lock_irq(rq);
1605 rq_pin_lock(rq, rf);
1606}
1607
1608static inline void
1609rq_lock(struct rq *rq, struct rq_flags *rf)
1610 __acquires(rq->lock)
1611{
1612 raw_spin_rq_lock(rq);
1613 rq_pin_lock(rq, rf);
1614}
1615
1616static inline void
1617rq_relock(struct rq *rq, struct rq_flags *rf)
1618 __acquires(rq->lock)
1619{
1620 raw_spin_rq_lock(rq);
1621 rq_repin_lock(rq, rf);
1622}
1623
1624static inline void
1625rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1626 __releases(rq->lock)
1627{
1628 rq_unpin_lock(rq, rf);
1629 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1630}
1631
1632static inline void
1633rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1634 __releases(rq->lock)
1635{
1636 rq_unpin_lock(rq, rf);
1637 raw_spin_rq_unlock_irq(rq);
1638}
1639
1640static inline void
1641rq_unlock(struct rq *rq, struct rq_flags *rf)
1642 __releases(rq->lock)
1643{
1644 rq_unpin_lock(rq, rf);
1645 raw_spin_rq_unlock(rq);
1646}
1647
1648static inline struct rq *
1649this_rq_lock_irq(struct rq_flags *rf)
1650 __acquires(rq->lock)
1651{
1652 struct rq *rq;
1653
1654 local_irq_disable();
1655 rq = this_rq();
1656 rq_lock(rq, rf);
1657 return rq;
1658}
1659
1660#ifdef CONFIG_NUMA
1661enum numa_topology_type {
1662 NUMA_DIRECT,
1663 NUMA_GLUELESS_MESH,
1664 NUMA_BACKPLANE,
1665};
1666extern enum numa_topology_type sched_numa_topology_type;
1667extern int sched_max_numa_distance;
1668extern bool find_numa_distance(int distance);
1669extern void sched_init_numa(void);
1670extern void sched_domains_numa_masks_set(unsigned int cpu);
1671extern void sched_domains_numa_masks_clear(unsigned int cpu);
1672extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1673#else
1674static inline void sched_init_numa(void) { }
1675static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1676static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1677static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1678{
1679 return nr_cpu_ids;
1680}
1681#endif
1682
1683#ifdef CONFIG_NUMA_BALANCING
1684/* The regions in numa_faults array from task_struct */
1685enum numa_faults_stats {
1686 NUMA_MEM = 0,
1687 NUMA_CPU,
1688 NUMA_MEMBUF,
1689 NUMA_CPUBUF
1690};
1691extern void sched_setnuma(struct task_struct *p, int node);
1692extern int migrate_task_to(struct task_struct *p, int cpu);
1693extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1694 int cpu, int scpu);
1695extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1696#else
1697static inline void
1698init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1699{
1700}
1701#endif /* CONFIG_NUMA_BALANCING */
1702
1703#ifdef CONFIG_SMP
1704
1705static inline void
1706queue_balance_callback(struct rq *rq,
1707 struct callback_head *head,
1708 void (*func)(struct rq *rq))
1709{
1710 lockdep_assert_rq_held(rq);
1711
1712 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1713 return;
1714
1715 head->func = (void (*)(struct callback_head *))func;
1716 head->next = rq->balance_callback;
1717 rq->balance_callback = head;
1718}
1719
1720#define rcu_dereference_check_sched_domain(p) \
1721 rcu_dereference_check((p), \
1722 lockdep_is_held(&sched_domains_mutex))
1723
1724/*
1725 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1726 * See destroy_sched_domains: call_rcu for details.
1727 *
1728 * The domain tree of any CPU may only be accessed from within
1729 * preempt-disabled sections.
1730 */
1731#define for_each_domain(cpu, __sd) \
1732 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1733 __sd; __sd = __sd->parent)
1734
1735/**
1736 * highest_flag_domain - Return highest sched_domain containing flag.
1737 * @cpu: The CPU whose highest level of sched domain is to
1738 * be returned.
1739 * @flag: The flag to check for the highest sched_domain
1740 * for the given CPU.
1741 *
1742 * Returns the highest sched_domain of a CPU which contains the given flag.
1743 */
1744static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1745{
1746 struct sched_domain *sd, *hsd = NULL;
1747
1748 for_each_domain(cpu, sd) {
1749 if (!(sd->flags & flag))
1750 break;
1751 hsd = sd;
1752 }
1753
1754 return hsd;
1755}
1756
1757static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1758{
1759 struct sched_domain *sd;
1760
1761 for_each_domain(cpu, sd) {
1762 if (sd->flags & flag)
1763 break;
1764 }
1765
1766 return sd;
1767}
1768
1769DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1770DECLARE_PER_CPU(int, sd_llc_size);
1771DECLARE_PER_CPU(int, sd_llc_id);
1772DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1773DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1774DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1775DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1776extern struct static_key_false sched_asym_cpucapacity;
1777
1778struct sched_group_capacity {
1779 atomic_t ref;
1780 /*
1781 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1782 * for a single CPU.
1783 */
1784 unsigned long capacity;
1785 unsigned long min_capacity; /* Min per-CPU capacity in group */
1786 unsigned long max_capacity; /* Max per-CPU capacity in group */
1787 unsigned long next_update;
1788 int imbalance; /* XXX unrelated to capacity but shared group state */
1789
1790#ifdef CONFIG_SCHED_DEBUG
1791 int id;
1792#endif
1793
1794 unsigned long cpumask[]; /* Balance mask */
1795};
1796
1797struct sched_group {
1798 struct sched_group *next; /* Must be a circular list */
1799 atomic_t ref;
1800
1801 unsigned int group_weight;
1802 struct sched_group_capacity *sgc;
1803 int asym_prefer_cpu; /* CPU of highest priority in group */
1804
1805 /*
1806 * The CPUs this group covers.
1807 *
1808 * NOTE: this field is variable length. (Allocated dynamically
1809 * by attaching extra space to the end of the structure,
1810 * depending on how many CPUs the kernel has booted up with)
1811 */
1812 unsigned long cpumask[];
1813};
1814
1815static inline struct cpumask *sched_group_span(struct sched_group *sg)
1816{
1817 return to_cpumask(sg->cpumask);
1818}
1819
1820/*
1821 * See build_balance_mask().
1822 */
1823static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1824{
1825 return to_cpumask(sg->sgc->cpumask);
1826}
1827
1828/**
1829 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1830 * @group: The group whose first CPU is to be returned.
1831 */
1832static inline unsigned int group_first_cpu(struct sched_group *group)
1833{
1834 return cpumask_first(sched_group_span(group));
1835}
1836
1837extern int group_balance_cpu(struct sched_group *sg);
1838
1839#ifdef CONFIG_SCHED_DEBUG
1840void update_sched_domain_debugfs(void);
1841void dirty_sched_domain_sysctl(int cpu);
1842#else
1843static inline void update_sched_domain_debugfs(void)
1844{
1845}
1846static inline void dirty_sched_domain_sysctl(int cpu)
1847{
1848}
1849#endif
1850
1851extern int sched_update_scaling(void);
1852
1853extern void flush_smp_call_function_from_idle(void);
1854
1855#else /* !CONFIG_SMP: */
1856static inline void flush_smp_call_function_from_idle(void) { }
1857#endif
1858
1859#include "stats.h"
1860#include "autogroup.h"
1861
1862#ifdef CONFIG_CGROUP_SCHED
1863
1864/*
1865 * Return the group to which this tasks belongs.
1866 *
1867 * We cannot use task_css() and friends because the cgroup subsystem
1868 * changes that value before the cgroup_subsys::attach() method is called,
1869 * therefore we cannot pin it and might observe the wrong value.
1870 *
1871 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1872 * core changes this before calling sched_move_task().
1873 *
1874 * Instead we use a 'copy' which is updated from sched_move_task() while
1875 * holding both task_struct::pi_lock and rq::lock.
1876 */
1877static inline struct task_group *task_group(struct task_struct *p)
1878{
1879 return p->sched_task_group;
1880}
1881
1882/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1883static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1884{
1885#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1886 struct task_group *tg = task_group(p);
1887#endif
1888
1889#ifdef CONFIG_FAIR_GROUP_SCHED
1890 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1891 p->se.cfs_rq = tg->cfs_rq[cpu];
1892 p->se.parent = tg->se[cpu];
1893#endif
1894
1895#ifdef CONFIG_RT_GROUP_SCHED
1896 p->rt.rt_rq = tg->rt_rq[cpu];
1897 p->rt.parent = tg->rt_se[cpu];
1898#endif
1899}
1900
1901#else /* CONFIG_CGROUP_SCHED */
1902
1903static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1904static inline struct task_group *task_group(struct task_struct *p)
1905{
1906 return NULL;
1907}
1908
1909#endif /* CONFIG_CGROUP_SCHED */
1910
1911static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1912{
1913 set_task_rq(p, cpu);
1914#ifdef CONFIG_SMP
1915 /*
1916 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1917 * successfully executed on another CPU. We must ensure that updates of
1918 * per-task data have been completed by this moment.
1919 */
1920 smp_wmb();
1921#ifdef CONFIG_THREAD_INFO_IN_TASK
1922 WRITE_ONCE(p->cpu, cpu);
1923#else
1924 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1925#endif
1926 p->wake_cpu = cpu;
1927#endif
1928}
1929
1930/*
1931 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1932 */
1933#ifdef CONFIG_SCHED_DEBUG
1934# include <linux/static_key.h>
1935# define const_debug __read_mostly
1936#else
1937# define const_debug const
1938#endif
1939
1940#define SCHED_FEAT(name, enabled) \
1941 __SCHED_FEAT_##name ,
1942
1943enum {
1944#include "features.h"
1945 __SCHED_FEAT_NR,
1946};
1947
1948#undef SCHED_FEAT
1949
1950#ifdef CONFIG_SCHED_DEBUG
1951
1952/*
1953 * To support run-time toggling of sched features, all the translation units
1954 * (but core.c) reference the sysctl_sched_features defined in core.c.
1955 */
1956extern const_debug unsigned int sysctl_sched_features;
1957
1958#ifdef CONFIG_JUMP_LABEL
1959#define SCHED_FEAT(name, enabled) \
1960static __always_inline bool static_branch_##name(struct static_key *key) \
1961{ \
1962 return static_key_##enabled(key); \
1963}
1964
1965#include "features.h"
1966#undef SCHED_FEAT
1967
1968extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1969#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1970
1971#else /* !CONFIG_JUMP_LABEL */
1972
1973#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1974
1975#endif /* CONFIG_JUMP_LABEL */
1976
1977#else /* !SCHED_DEBUG */
1978
1979/*
1980 * Each translation unit has its own copy of sysctl_sched_features to allow
1981 * constants propagation at compile time and compiler optimization based on
1982 * features default.
1983 */
1984#define SCHED_FEAT(name, enabled) \
1985 (1UL << __SCHED_FEAT_##name) * enabled |
1986static const_debug __maybe_unused unsigned int sysctl_sched_features =
1987#include "features.h"
1988 0;
1989#undef SCHED_FEAT
1990
1991#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1992
1993#endif /* SCHED_DEBUG */
1994
1995extern struct static_key_false sched_numa_balancing;
1996extern struct static_key_false sched_schedstats;
1997
1998static inline u64 global_rt_period(void)
1999{
2000 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2001}
2002
2003static inline u64 global_rt_runtime(void)
2004{
2005 if (sysctl_sched_rt_runtime < 0)
2006 return RUNTIME_INF;
2007
2008 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2009}
2010
2011static inline int task_current(struct rq *rq, struct task_struct *p)
2012{
2013 return rq->curr == p;
2014}
2015
2016static inline int task_running(struct rq *rq, struct task_struct *p)
2017{
2018#ifdef CONFIG_SMP
2019 return p->on_cpu;
2020#else
2021 return task_current(rq, p);
2022#endif
2023}
2024
2025static inline int task_on_rq_queued(struct task_struct *p)
2026{
2027 return p->on_rq == TASK_ON_RQ_QUEUED;
2028}
2029
2030static inline int task_on_rq_migrating(struct task_struct *p)
2031{
2032 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2033}
2034
2035/* Wake flags. The first three directly map to some SD flag value */
2036#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2037#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2038#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2039
2040#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2041#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2042#define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2043
2044#ifdef CONFIG_SMP
2045static_assert(WF_EXEC == SD_BALANCE_EXEC);
2046static_assert(WF_FORK == SD_BALANCE_FORK);
2047static_assert(WF_TTWU == SD_BALANCE_WAKE);
2048#endif
2049
2050/*
2051 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2052 * of tasks with abnormal "nice" values across CPUs the contribution that
2053 * each task makes to its run queue's load is weighted according to its
2054 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2055 * scaled version of the new time slice allocation that they receive on time
2056 * slice expiry etc.
2057 */
2058
2059#define WEIGHT_IDLEPRIO 3
2060#define WMULT_IDLEPRIO 1431655765
2061
2062extern const int sched_prio_to_weight[40];
2063extern const u32 sched_prio_to_wmult[40];
2064
2065/*
2066 * {de,en}queue flags:
2067 *
2068 * DEQUEUE_SLEEP - task is no longer runnable
2069 * ENQUEUE_WAKEUP - task just became runnable
2070 *
2071 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2072 * are in a known state which allows modification. Such pairs
2073 * should preserve as much state as possible.
2074 *
2075 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2076 * in the runqueue.
2077 *
2078 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2079 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2080 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2081 *
2082 */
2083
2084#define DEQUEUE_SLEEP 0x01
2085#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2086#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2087#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2088
2089#define ENQUEUE_WAKEUP 0x01
2090#define ENQUEUE_RESTORE 0x02
2091#define ENQUEUE_MOVE 0x04
2092#define ENQUEUE_NOCLOCK 0x08
2093
2094#define ENQUEUE_HEAD 0x10
2095#define ENQUEUE_REPLENISH 0x20
2096#ifdef CONFIG_SMP
2097#define ENQUEUE_MIGRATED 0x40
2098#else
2099#define ENQUEUE_MIGRATED 0x00
2100#endif
2101
2102#define RETRY_TASK ((void *)-1UL)
2103
2104struct sched_class {
2105
2106#ifdef CONFIG_UCLAMP_TASK
2107 int uclamp_enabled;
2108#endif
2109
2110 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2111 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2112 void (*yield_task) (struct rq *rq);
2113 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2114
2115 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2116
2117 struct task_struct *(*pick_next_task)(struct rq *rq);
2118
2119 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2120 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2121
2122#ifdef CONFIG_SMP
2123 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2124 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2125
2126 struct task_struct * (*pick_task)(struct rq *rq);
2127
2128 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2129
2130 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2131
2132 void (*set_cpus_allowed)(struct task_struct *p,
2133 const struct cpumask *newmask,
2134 u32 flags);
2135
2136 void (*rq_online)(struct rq *rq);
2137 void (*rq_offline)(struct rq *rq);
2138
2139 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2140#endif
2141
2142 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2143 void (*task_fork)(struct task_struct *p);
2144 void (*task_dead)(struct task_struct *p);
2145
2146 /*
2147 * The switched_from() call is allowed to drop rq->lock, therefore we
2148 * cannot assume the switched_from/switched_to pair is serialized by
2149 * rq->lock. They are however serialized by p->pi_lock.
2150 */
2151 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2152 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2153 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2154 int oldprio);
2155
2156 unsigned int (*get_rr_interval)(struct rq *rq,
2157 struct task_struct *task);
2158
2159 void (*update_curr)(struct rq *rq);
2160
2161#define TASK_SET_GROUP 0
2162#define TASK_MOVE_GROUP 1
2163
2164#ifdef CONFIG_FAIR_GROUP_SCHED
2165 void (*task_change_group)(struct task_struct *p, int type);
2166#endif
2167};
2168
2169static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2170{
2171 WARN_ON_ONCE(rq->curr != prev);
2172 prev->sched_class->put_prev_task(rq, prev);
2173}
2174
2175static inline void set_next_task(struct rq *rq, struct task_struct *next)
2176{
2177 next->sched_class->set_next_task(rq, next, false);
2178}
2179
2180
2181/*
2182 * Helper to define a sched_class instance; each one is placed in a separate
2183 * section which is ordered by the linker script:
2184 *
2185 * include/asm-generic/vmlinux.lds.h
2186 *
2187 * Also enforce alignment on the instance, not the type, to guarantee layout.
2188 */
2189#define DEFINE_SCHED_CLASS(name) \
2190const struct sched_class name##_sched_class \
2191 __aligned(__alignof__(struct sched_class)) \
2192 __section("__" #name "_sched_class")
2193
2194/* Defined in include/asm-generic/vmlinux.lds.h */
2195extern struct sched_class __begin_sched_classes[];
2196extern struct sched_class __end_sched_classes[];
2197
2198#define sched_class_highest (__end_sched_classes - 1)
2199#define sched_class_lowest (__begin_sched_classes - 1)
2200
2201#define for_class_range(class, _from, _to) \
2202 for (class = (_from); class != (_to); class--)
2203
2204#define for_each_class(class) \
2205 for_class_range(class, sched_class_highest, sched_class_lowest)
2206
2207extern const struct sched_class stop_sched_class;
2208extern const struct sched_class dl_sched_class;
2209extern const struct sched_class rt_sched_class;
2210extern const struct sched_class fair_sched_class;
2211extern const struct sched_class idle_sched_class;
2212
2213static inline bool sched_stop_runnable(struct rq *rq)
2214{
2215 return rq->stop && task_on_rq_queued(rq->stop);
2216}
2217
2218static inline bool sched_dl_runnable(struct rq *rq)
2219{
2220 return rq->dl.dl_nr_running > 0;
2221}
2222
2223static inline bool sched_rt_runnable(struct rq *rq)
2224{
2225 return rq->rt.rt_queued > 0;
2226}
2227
2228static inline bool sched_fair_runnable(struct rq *rq)
2229{
2230 return rq->cfs.nr_running > 0;
2231}
2232
2233extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2234extern struct task_struct *pick_next_task_idle(struct rq *rq);
2235
2236#define SCA_CHECK 0x01
2237#define SCA_MIGRATE_DISABLE 0x02
2238#define SCA_MIGRATE_ENABLE 0x04
2239
2240#ifdef CONFIG_SMP
2241
2242extern void update_group_capacity(struct sched_domain *sd, int cpu);
2243
2244extern void trigger_load_balance(struct rq *rq);
2245
2246extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2247
2248static inline struct task_struct *get_push_task(struct rq *rq)
2249{
2250 struct task_struct *p = rq->curr;
2251
2252 lockdep_assert_rq_held(rq);
2253
2254 if (rq->push_busy)
2255 return NULL;
2256
2257 if (p->nr_cpus_allowed == 1)
2258 return NULL;
2259
2260 if (p->migration_disabled)
2261 return NULL;
2262
2263 rq->push_busy = true;
2264 return get_task_struct(p);
2265}
2266
2267extern int push_cpu_stop(void *arg);
2268
2269#endif
2270
2271#ifdef CONFIG_CPU_IDLE
2272static inline void idle_set_state(struct rq *rq,
2273 struct cpuidle_state *idle_state)
2274{
2275 rq->idle_state = idle_state;
2276}
2277
2278static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2279{
2280 SCHED_WARN_ON(!rcu_read_lock_held());
2281
2282 return rq->idle_state;
2283}
2284#else
2285static inline void idle_set_state(struct rq *rq,
2286 struct cpuidle_state *idle_state)
2287{
2288}
2289
2290static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2291{
2292 return NULL;
2293}
2294#endif
2295
2296extern void schedule_idle(void);
2297
2298extern void sysrq_sched_debug_show(void);
2299extern void sched_init_granularity(void);
2300extern void update_max_interval(void);
2301
2302extern void init_sched_dl_class(void);
2303extern void init_sched_rt_class(void);
2304extern void init_sched_fair_class(void);
2305
2306extern void reweight_task(struct task_struct *p, int prio);
2307
2308extern void resched_curr(struct rq *rq);
2309extern void resched_cpu(int cpu);
2310
2311extern struct rt_bandwidth def_rt_bandwidth;
2312extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2313
2314extern struct dl_bandwidth def_dl_bandwidth;
2315extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2316extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2317extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2318
2319#define BW_SHIFT 20
2320#define BW_UNIT (1 << BW_SHIFT)
2321#define RATIO_SHIFT 8
2322#define MAX_BW_BITS (64 - BW_SHIFT)
2323#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2324unsigned long to_ratio(u64 period, u64 runtime);
2325
2326extern void init_entity_runnable_average(struct sched_entity *se);
2327extern void post_init_entity_util_avg(struct task_struct *p);
2328
2329#ifdef CONFIG_NO_HZ_FULL
2330extern bool sched_can_stop_tick(struct rq *rq);
2331extern int __init sched_tick_offload_init(void);
2332
2333/*
2334 * Tick may be needed by tasks in the runqueue depending on their policy and
2335 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2336 * nohz mode if necessary.
2337 */
2338static inline void sched_update_tick_dependency(struct rq *rq)
2339{
2340 int cpu = cpu_of(rq);
2341
2342 if (!tick_nohz_full_cpu(cpu))
2343 return;
2344
2345 if (sched_can_stop_tick(rq))
2346 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2347 else
2348 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2349}
2350#else
2351static inline int sched_tick_offload_init(void) { return 0; }
2352static inline void sched_update_tick_dependency(struct rq *rq) { }
2353#endif
2354
2355static inline void add_nr_running(struct rq *rq, unsigned count)
2356{
2357 unsigned prev_nr = rq->nr_running;
2358
2359 rq->nr_running = prev_nr + count;
2360 if (trace_sched_update_nr_running_tp_enabled()) {
2361 call_trace_sched_update_nr_running(rq, count);
2362 }
2363
2364#ifdef CONFIG_SMP
2365 if (prev_nr < 2 && rq->nr_running >= 2) {
2366 if (!READ_ONCE(rq->rd->overload))
2367 WRITE_ONCE(rq->rd->overload, 1);
2368 }
2369#endif
2370
2371 sched_update_tick_dependency(rq);
2372}
2373
2374static inline void sub_nr_running(struct rq *rq, unsigned count)
2375{
2376 rq->nr_running -= count;
2377 if (trace_sched_update_nr_running_tp_enabled()) {
2378 call_trace_sched_update_nr_running(rq, -count);
2379 }
2380
2381 /* Check if we still need preemption */
2382 sched_update_tick_dependency(rq);
2383}
2384
2385extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2386extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2387
2388extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2389
2390extern const_debug unsigned int sysctl_sched_nr_migrate;
2391extern const_debug unsigned int sysctl_sched_migration_cost;
2392
2393#ifdef CONFIG_SCHED_HRTICK
2394
2395/*
2396 * Use hrtick when:
2397 * - enabled by features
2398 * - hrtimer is actually high res
2399 */
2400static inline int hrtick_enabled(struct rq *rq)
2401{
2402 if (!cpu_active(cpu_of(rq)))
2403 return 0;
2404 return hrtimer_is_hres_active(&rq->hrtick_timer);
2405}
2406
2407static inline int hrtick_enabled_fair(struct rq *rq)
2408{
2409 if (!sched_feat(HRTICK))
2410 return 0;
2411 return hrtick_enabled(rq);
2412}
2413
2414static inline int hrtick_enabled_dl(struct rq *rq)
2415{
2416 if (!sched_feat(HRTICK_DL))
2417 return 0;
2418 return hrtick_enabled(rq);
2419}
2420
2421void hrtick_start(struct rq *rq, u64 delay);
2422
2423#else
2424
2425static inline int hrtick_enabled_fair(struct rq *rq)
2426{
2427 return 0;
2428}
2429
2430static inline int hrtick_enabled_dl(struct rq *rq)
2431{
2432 return 0;
2433}
2434
2435static inline int hrtick_enabled(struct rq *rq)
2436{
2437 return 0;
2438}
2439
2440#endif /* CONFIG_SCHED_HRTICK */
2441
2442#ifndef arch_scale_freq_tick
2443static __always_inline
2444void arch_scale_freq_tick(void)
2445{
2446}
2447#endif
2448
2449#ifndef arch_scale_freq_capacity
2450/**
2451 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2452 * @cpu: the CPU in question.
2453 *
2454 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2455 *
2456 * f_curr
2457 * ------ * SCHED_CAPACITY_SCALE
2458 * f_max
2459 */
2460static __always_inline
2461unsigned long arch_scale_freq_capacity(int cpu)
2462{
2463 return SCHED_CAPACITY_SCALE;
2464}
2465#endif
2466
2467
2468#ifdef CONFIG_SMP
2469
2470static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2471{
2472#ifdef CONFIG_SCHED_CORE
2473 /*
2474 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2475 * order by core-id first and cpu-id second.
2476 *
2477 * Notably:
2478 *
2479 * double_rq_lock(0,3); will take core-0, core-1 lock
2480 * double_rq_lock(1,2); will take core-1, core-0 lock
2481 *
2482 * when only cpu-id is considered.
2483 */
2484 if (rq1->core->cpu < rq2->core->cpu)
2485 return true;
2486 if (rq1->core->cpu > rq2->core->cpu)
2487 return false;
2488
2489 /*
2490 * __sched_core_flip() relies on SMT having cpu-id lock order.
2491 */
2492#endif
2493 return rq1->cpu < rq2->cpu;
2494}
2495
2496extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2497
2498#ifdef CONFIG_PREEMPTION
2499
2500/*
2501 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2502 * way at the expense of forcing extra atomic operations in all
2503 * invocations. This assures that the double_lock is acquired using the
2504 * same underlying policy as the spinlock_t on this architecture, which
2505 * reduces latency compared to the unfair variant below. However, it
2506 * also adds more overhead and therefore may reduce throughput.
2507 */
2508static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2509 __releases(this_rq->lock)
2510 __acquires(busiest->lock)
2511 __acquires(this_rq->lock)
2512{
2513 raw_spin_rq_unlock(this_rq);
2514 double_rq_lock(this_rq, busiest);
2515
2516 return 1;
2517}
2518
2519#else
2520/*
2521 * Unfair double_lock_balance: Optimizes throughput at the expense of
2522 * latency by eliminating extra atomic operations when the locks are
2523 * already in proper order on entry. This favors lower CPU-ids and will
2524 * grant the double lock to lower CPUs over higher ids under contention,
2525 * regardless of entry order into the function.
2526 */
2527static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2528 __releases(this_rq->lock)
2529 __acquires(busiest->lock)
2530 __acquires(this_rq->lock)
2531{
2532 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2533 return 0;
2534
2535 if (likely(raw_spin_rq_trylock(busiest)))
2536 return 0;
2537
2538 if (rq_order_less(this_rq, busiest)) {
2539 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2540 return 0;
2541 }
2542
2543 raw_spin_rq_unlock(this_rq);
2544 double_rq_lock(this_rq, busiest);
2545
2546 return 1;
2547}
2548
2549#endif /* CONFIG_PREEMPTION */
2550
2551/*
2552 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2553 */
2554static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2555{
2556 lockdep_assert_irqs_disabled();
2557
2558 return _double_lock_balance(this_rq, busiest);
2559}
2560
2561static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2562 __releases(busiest->lock)
2563{
2564 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2565 raw_spin_rq_unlock(busiest);
2566 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2567}
2568
2569static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2570{
2571 if (l1 > l2)
2572 swap(l1, l2);
2573
2574 spin_lock(l1);
2575 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2576}
2577
2578static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2579{
2580 if (l1 > l2)
2581 swap(l1, l2);
2582
2583 spin_lock_irq(l1);
2584 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2585}
2586
2587static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2588{
2589 if (l1 > l2)
2590 swap(l1, l2);
2591
2592 raw_spin_lock(l1);
2593 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2594}
2595
2596/*
2597 * double_rq_unlock - safely unlock two runqueues
2598 *
2599 * Note this does not restore interrupts like task_rq_unlock,
2600 * you need to do so manually after calling.
2601 */
2602static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2603 __releases(rq1->lock)
2604 __releases(rq2->lock)
2605{
2606 if (__rq_lockp(rq1) != __rq_lockp(rq2))
2607 raw_spin_rq_unlock(rq2);
2608 else
2609 __release(rq2->lock);
2610 raw_spin_rq_unlock(rq1);
2611}
2612
2613extern void set_rq_online (struct rq *rq);
2614extern void set_rq_offline(struct rq *rq);
2615extern bool sched_smp_initialized;
2616
2617#else /* CONFIG_SMP */
2618
2619/*
2620 * double_rq_lock - safely lock two runqueues
2621 *
2622 * Note this does not disable interrupts like task_rq_lock,
2623 * you need to do so manually before calling.
2624 */
2625static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2626 __acquires(rq1->lock)
2627 __acquires(rq2->lock)
2628{
2629 BUG_ON(!irqs_disabled());
2630 BUG_ON(rq1 != rq2);
2631 raw_spin_rq_lock(rq1);
2632 __acquire(rq2->lock); /* Fake it out ;) */
2633}
2634
2635/*
2636 * double_rq_unlock - safely unlock two runqueues
2637 *
2638 * Note this does not restore interrupts like task_rq_unlock,
2639 * you need to do so manually after calling.
2640 */
2641static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2642 __releases(rq1->lock)
2643 __releases(rq2->lock)
2644{
2645 BUG_ON(rq1 != rq2);
2646 raw_spin_rq_unlock(rq1);
2647 __release(rq2->lock);
2648}
2649
2650#endif
2651
2652extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2653extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2654
2655#ifdef CONFIG_SCHED_DEBUG
2656extern bool sched_debug_verbose;
2657
2658extern void print_cfs_stats(struct seq_file *m, int cpu);
2659extern void print_rt_stats(struct seq_file *m, int cpu);
2660extern void print_dl_stats(struct seq_file *m, int cpu);
2661extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2662extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2663extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2664
2665extern void resched_latency_warn(int cpu, u64 latency);
2666#ifdef CONFIG_NUMA_BALANCING
2667extern void
2668show_numa_stats(struct task_struct *p, struct seq_file *m);
2669extern void
2670print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2671 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2672#endif /* CONFIG_NUMA_BALANCING */
2673#else
2674static inline void resched_latency_warn(int cpu, u64 latency) {}
2675#endif /* CONFIG_SCHED_DEBUG */
2676
2677extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2678extern void init_rt_rq(struct rt_rq *rt_rq);
2679extern void init_dl_rq(struct dl_rq *dl_rq);
2680
2681extern void cfs_bandwidth_usage_inc(void);
2682extern void cfs_bandwidth_usage_dec(void);
2683
2684#ifdef CONFIG_NO_HZ_COMMON
2685#define NOHZ_BALANCE_KICK_BIT 0
2686#define NOHZ_STATS_KICK_BIT 1
2687#define NOHZ_NEWILB_KICK_BIT 2
2688
2689#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2690#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2691#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2692
2693#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2694
2695#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2696
2697extern void nohz_balance_exit_idle(struct rq *rq);
2698#else
2699static inline void nohz_balance_exit_idle(struct rq *rq) { }
2700#endif
2701
2702#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2703extern void nohz_run_idle_balance(int cpu);
2704#else
2705static inline void nohz_run_idle_balance(int cpu) { }
2706#endif
2707
2708#ifdef CONFIG_SMP
2709static inline
2710void __dl_update(struct dl_bw *dl_b, s64 bw)
2711{
2712 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2713 int i;
2714
2715 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2716 "sched RCU must be held");
2717 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2718 struct rq *rq = cpu_rq(i);
2719
2720 rq->dl.extra_bw += bw;
2721 }
2722}
2723#else
2724static inline
2725void __dl_update(struct dl_bw *dl_b, s64 bw)
2726{
2727 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2728
2729 dl->extra_bw += bw;
2730}
2731#endif
2732
2733
2734#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2735struct irqtime {
2736 u64 total;
2737 u64 tick_delta;
2738 u64 irq_start_time;
2739 struct u64_stats_sync sync;
2740};
2741
2742DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2743
2744/*
2745 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2746 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2747 * and never move forward.
2748 */
2749static inline u64 irq_time_read(int cpu)
2750{
2751 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2752 unsigned int seq;
2753 u64 total;
2754
2755 do {
2756 seq = __u64_stats_fetch_begin(&irqtime->sync);
2757 total = irqtime->total;
2758 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2759
2760 return total;
2761}
2762#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2763
2764#ifdef CONFIG_CPU_FREQ
2765DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2766
2767/**
2768 * cpufreq_update_util - Take a note about CPU utilization changes.
2769 * @rq: Runqueue to carry out the update for.
2770 * @flags: Update reason flags.
2771 *
2772 * This function is called by the scheduler on the CPU whose utilization is
2773 * being updated.
2774 *
2775 * It can only be called from RCU-sched read-side critical sections.
2776 *
2777 * The way cpufreq is currently arranged requires it to evaluate the CPU
2778 * performance state (frequency/voltage) on a regular basis to prevent it from
2779 * being stuck in a completely inadequate performance level for too long.
2780 * That is not guaranteed to happen if the updates are only triggered from CFS
2781 * and DL, though, because they may not be coming in if only RT tasks are
2782 * active all the time (or there are RT tasks only).
2783 *
2784 * As a workaround for that issue, this function is called periodically by the
2785 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2786 * but that really is a band-aid. Going forward it should be replaced with
2787 * solutions targeted more specifically at RT tasks.
2788 */
2789static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2790{
2791 struct update_util_data *data;
2792
2793 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2794 cpu_of(rq)));
2795 if (data)
2796 data->func(data, rq_clock(rq), flags);
2797}
2798#else
2799static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2800#endif /* CONFIG_CPU_FREQ */
2801
2802#ifdef CONFIG_UCLAMP_TASK
2803unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2804
2805/**
2806 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2807 * @rq: The rq to clamp against. Must not be NULL.
2808 * @util: The util value to clamp.
2809 * @p: The task to clamp against. Can be NULL if you want to clamp
2810 * against @rq only.
2811 *
2812 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2813 *
2814 * If sched_uclamp_used static key is disabled, then just return the util
2815 * without any clamping since uclamp aggregation at the rq level in the fast
2816 * path is disabled, rendering this operation a NOP.
2817 *
2818 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2819 * will return the correct effective uclamp value of the task even if the
2820 * static key is disabled.
2821 */
2822static __always_inline
2823unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2824 struct task_struct *p)
2825{
2826 unsigned long min_util = 0;
2827 unsigned long max_util = 0;
2828
2829 if (!static_branch_likely(&sched_uclamp_used))
2830 return util;
2831
2832 if (p) {
2833 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2834 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2835
2836 /*
2837 * Ignore last runnable task's max clamp, as this task will
2838 * reset it. Similarly, no need to read the rq's min clamp.
2839 */
2840 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2841 goto out;
2842 }
2843
2844 min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2845 max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2846out:
2847 /*
2848 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2849 * RUNNABLE tasks with _different_ clamps, we can end up with an
2850 * inversion. Fix it now when the clamps are applied.
2851 */
2852 if (unlikely(min_util >= max_util))
2853 return min_util;
2854
2855 return clamp(util, min_util, max_util);
2856}
2857
2858/*
2859 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2860 * by default in the fast path and only gets turned on once userspace performs
2861 * an operation that requires it.
2862 *
2863 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2864 * hence is active.
2865 */
2866static inline bool uclamp_is_used(void)
2867{
2868 return static_branch_likely(&sched_uclamp_used);
2869}
2870#else /* CONFIG_UCLAMP_TASK */
2871static inline
2872unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2873 struct task_struct *p)
2874{
2875 return util;
2876}
2877
2878static inline bool uclamp_is_used(void)
2879{
2880 return false;
2881}
2882#endif /* CONFIG_UCLAMP_TASK */
2883
2884#ifdef arch_scale_freq_capacity
2885# ifndef arch_scale_freq_invariant
2886# define arch_scale_freq_invariant() true
2887# endif
2888#else
2889# define arch_scale_freq_invariant() false
2890#endif
2891
2892#ifdef CONFIG_SMP
2893static inline unsigned long capacity_orig_of(int cpu)
2894{
2895 return cpu_rq(cpu)->cpu_capacity_orig;
2896}
2897
2898/**
2899 * enum cpu_util_type - CPU utilization type
2900 * @FREQUENCY_UTIL: Utilization used to select frequency
2901 * @ENERGY_UTIL: Utilization used during energy calculation
2902 *
2903 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2904 * need to be aggregated differently depending on the usage made of them. This
2905 * enum is used within effective_cpu_util() to differentiate the types of
2906 * utilization expected by the callers, and adjust the aggregation accordingly.
2907 */
2908enum cpu_util_type {
2909 FREQUENCY_UTIL,
2910 ENERGY_UTIL,
2911};
2912
2913unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2914 unsigned long max, enum cpu_util_type type,
2915 struct task_struct *p);
2916
2917static inline unsigned long cpu_bw_dl(struct rq *rq)
2918{
2919 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2920}
2921
2922static inline unsigned long cpu_util_dl(struct rq *rq)
2923{
2924 return READ_ONCE(rq->avg_dl.util_avg);
2925}
2926
2927static inline unsigned long cpu_util_cfs(struct rq *rq)
2928{
2929 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2930
2931 if (sched_feat(UTIL_EST)) {
2932 util = max_t(unsigned long, util,
2933 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2934 }
2935
2936 return util;
2937}
2938
2939static inline unsigned long cpu_util_rt(struct rq *rq)
2940{
2941 return READ_ONCE(rq->avg_rt.util_avg);
2942}
2943#endif
2944
2945#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2946static inline unsigned long cpu_util_irq(struct rq *rq)
2947{
2948 return rq->avg_irq.util_avg;
2949}
2950
2951static inline
2952unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2953{
2954 util *= (max - irq);
2955 util /= max;
2956
2957 return util;
2958
2959}
2960#else
2961static inline unsigned long cpu_util_irq(struct rq *rq)
2962{
2963 return 0;
2964}
2965
2966static inline
2967unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2968{
2969 return util;
2970}
2971#endif
2972
2973#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2974
2975#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2976
2977DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2978
2979static inline bool sched_energy_enabled(void)
2980{
2981 return static_branch_unlikely(&sched_energy_present);
2982}
2983
2984#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2985
2986#define perf_domain_span(pd) NULL
2987static inline bool sched_energy_enabled(void) { return false; }
2988
2989#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2990
2991#ifdef CONFIG_MEMBARRIER
2992/*
2993 * The scheduler provides memory barriers required by membarrier between:
2994 * - prior user-space memory accesses and store to rq->membarrier_state,
2995 * - store to rq->membarrier_state and following user-space memory accesses.
2996 * In the same way it provides those guarantees around store to rq->curr.
2997 */
2998static inline void membarrier_switch_mm(struct rq *rq,
2999 struct mm_struct *prev_mm,
3000 struct mm_struct *next_mm)
3001{
3002 int membarrier_state;
3003
3004 if (prev_mm == next_mm)
3005 return;
3006
3007 membarrier_state = atomic_read(&next_mm->membarrier_state);
3008 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3009 return;
3010
3011 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3012}
3013#else
3014static inline void membarrier_switch_mm(struct rq *rq,
3015 struct mm_struct *prev_mm,
3016 struct mm_struct *next_mm)
3017{
3018}
3019#endif
3020
3021#ifdef CONFIG_SMP
3022static inline bool is_per_cpu_kthread(struct task_struct *p)
3023{
3024 if (!(p->flags & PF_KTHREAD))
3025 return false;
3026
3027 if (p->nr_cpus_allowed != 1)
3028 return false;
3029
3030 return true;
3031}
3032#endif
3033
3034extern void swake_up_all_locked(struct swait_queue_head *q);
3035extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3036
3037#ifdef CONFIG_PREEMPT_DYNAMIC
3038extern int preempt_dynamic_mode;
3039extern int sched_dynamic_mode(const char *str);
3040extern void sched_dynamic_update(int mode);
3041#endif
3042