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#include <linux/mmu_context.h>
73
74#include <trace/events/power.h>
75#include <trace/events/sched.h>
76
77#include "../workqueue_internal.h"
78
79struct rq;
80struct cfs_rq;
81struct rt_rq;
82struct sched_group;
83struct cpuidle_state;
84
85#ifdef CONFIG_PARAVIRT
86# include <asm/paravirt.h>
87# include <asm/paravirt_api_clock.h>
88#endif
89
90#include <asm/barrier.h>
91
92#include "cpupri.h"
93#include "cpudeadline.h"
94
95/* task_struct::on_rq states: */
96#define TASK_ON_RQ_QUEUED 1
97#define TASK_ON_RQ_MIGRATING 2
98
99extern __read_mostly int scheduler_running;
100
101extern unsigned long calc_load_update;
102extern atomic_long_t calc_load_tasks;
103
104extern void calc_global_load_tick(struct rq *this_rq);
105extern long calc_load_fold_active(struct rq *this_rq, long adjust);
106
107extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
108
109extern int sysctl_sched_rt_period;
110extern int sysctl_sched_rt_runtime;
111extern int sched_rr_timeslice;
112
113/*
114 * Asymmetric CPU capacity bits
115 */
116struct asym_cap_data {
117 struct list_head link;
118 struct rcu_head rcu;
119 unsigned long capacity;
120 unsigned long cpus[];
121};
122
123extern struct list_head asym_cap_list;
124
125#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
126
127/*
128 * Helpers for converting nanosecond timing to jiffy resolution
129 */
130#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
131
132/*
133 * Increase resolution of nice-level calculations for 64-bit architectures.
134 * The extra resolution improves shares distribution and load balancing of
135 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
136 * hierarchies, especially on larger systems. This is not a user-visible change
137 * and does not change the user-interface for setting shares/weights.
138 *
139 * We increase resolution only if we have enough bits to allow this increased
140 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
141 * are pretty high and the returns do not justify the increased costs.
142 *
143 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
144 * increase coverage and consistency always enable it on 64-bit platforms.
145 */
146#ifdef CONFIG_64BIT
147# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
148# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
149# define scale_load_down(w) \
150({ \
151 unsigned long __w = (w); \
152 \
153 if (__w) \
154 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
155 __w; \
156})
157#else
158# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
159# define scale_load(w) (w)
160# define scale_load_down(w) (w)
161#endif
162
163/*
164 * Task weight (visible to users) and its load (invisible to users) have
165 * independent resolution, but they should be well calibrated. We use
166 * scale_load() and scale_load_down(w) to convert between them. The
167 * following must be true:
168 *
169 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
170 *
171 */
172#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
173
174/*
175 * Single value that decides SCHED_DEADLINE internal math precision.
176 * 10 -> just above 1us
177 * 9 -> just above 0.5us
178 */
179#define DL_SCALE 10
180
181/*
182 * Single value that denotes runtime == period, ie unlimited time.
183 */
184#define RUNTIME_INF ((u64)~0ULL)
185
186static inline int idle_policy(int policy)
187{
188 return policy == SCHED_IDLE;
189}
190
191static inline int normal_policy(int policy)
192{
193#ifdef CONFIG_SCHED_CLASS_EXT
194 if (policy == SCHED_EXT)
195 return true;
196#endif
197 return policy == SCHED_NORMAL;
198}
199
200static inline int fair_policy(int policy)
201{
202 return normal_policy(policy) || policy == SCHED_BATCH;
203}
204
205static inline int rt_policy(int policy)
206{
207 return policy == SCHED_FIFO || policy == SCHED_RR;
208}
209
210static inline int dl_policy(int policy)
211{
212 return policy == SCHED_DEADLINE;
213}
214
215static inline bool valid_policy(int policy)
216{
217 return idle_policy(policy) || fair_policy(policy) ||
218 rt_policy(policy) || dl_policy(policy);
219}
220
221static inline int task_has_idle_policy(struct task_struct *p)
222{
223 return idle_policy(policy: p->policy);
224}
225
226static inline int task_has_rt_policy(struct task_struct *p)
227{
228 return rt_policy(policy: p->policy);
229}
230
231static inline int task_has_dl_policy(struct task_struct *p)
232{
233 return dl_policy(policy: p->policy);
234}
235
236#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
237
238static inline void update_avg(u64 *avg, u64 sample)
239{
240 s64 diff = sample - *avg;
241
242 *avg += diff / 8;
243}
244
245/*
246 * Shifting a value by an exponent greater *or equal* to the size of said value
247 * is UB; cap at size-1.
248 */
249#define shr_bound(val, shift) \
250 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
251
252/*
253 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
254 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
255 * maps pretty well onto the shares value used by scheduler and the round-trip
256 * conversions preserve the original value over the entire range.
257 */
258static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
259{
260 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
261}
262
263static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
264{
265 return clamp_t(unsigned long,
266 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
267 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
268}
269
270/*
271 * !! For sched_setattr_nocheck() (kernel) only !!
272 *
273 * This is actually gross. :(
274 *
275 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
276 * tasks, but still be able to sleep. We need this on platforms that cannot
277 * atomically change clock frequency. Remove once fast switching will be
278 * available on such platforms.
279 *
280 * SUGOV stands for SchedUtil GOVernor.
281 */
282#define SCHED_FLAG_SUGOV 0x10000000
283
284#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
285
286static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
287{
288#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
289 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
290#else
291 return false;
292#endif
293}
294
295/*
296 * Tells if entity @a should preempt entity @b.
297 */
298static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
299 const struct sched_dl_entity *b)
300{
301 return dl_entity_is_special(dl_se: a) ||
302 dl_time_before(a: a->deadline, b: b->deadline);
303}
304
305/*
306 * This is the priority-queue data structure of the RT scheduling class:
307 */
308struct rt_prio_array {
309 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
310 struct list_head queue[MAX_RT_PRIO];
311};
312
313struct rt_bandwidth {
314 /* nests inside the rq lock: */
315 raw_spinlock_t rt_runtime_lock;
316 ktime_t rt_period;
317 u64 rt_runtime;
318 struct hrtimer rt_period_timer;
319 unsigned int rt_period_active;
320};
321
322static inline int dl_bandwidth_enabled(void)
323{
324 return sysctl_sched_rt_runtime >= 0;
325}
326
327/*
328 * To keep the bandwidth of -deadline tasks under control
329 * we need some place where:
330 * - store the maximum -deadline bandwidth of each cpu;
331 * - cache the fraction of bandwidth that is currently allocated in
332 * each root domain;
333 *
334 * This is all done in the data structure below. It is similar to the
335 * one used for RT-throttling (rt_bandwidth), with the main difference
336 * that, since here we are only interested in admission control, we
337 * do not decrease any runtime while the group "executes", neither we
338 * need a timer to replenish it.
339 *
340 * With respect to SMP, bandwidth is given on a per root domain basis,
341 * meaning that:
342 * - bw (< 100%) is the deadline bandwidth of each CPU;
343 * - total_bw is the currently allocated bandwidth in each root domain;
344 */
345struct dl_bw {
346 raw_spinlock_t lock;
347 u64 bw;
348 u64 total_bw;
349};
350
351extern void init_dl_bw(struct dl_bw *dl_b);
352extern int sched_dl_global_validate(void);
353extern void sched_dl_do_global(void);
354extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
355extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
356extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
357extern bool __checkparam_dl(const struct sched_attr *attr);
358extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
359extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
360extern int dl_bw_deactivate(int cpu);
361extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
362/*
363 * SCHED_DEADLINE supports servers (nested scheduling) with the following
364 * interface:
365 *
366 * dl_se::rq -- runqueue we belong to.
367 *
368 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
369 * returns NULL.
370 *
371 * dl_server_update() -- called from update_curr_common(), propagates runtime
372 * to the server.
373 *
374 * dl_server_start() -- start the server when it has tasks; it will stop
375 * automatically when there are no more tasks, per
376 * dl_se::server_pick() returning NULL.
377 *
378 * dl_server_stop() -- (force) stop the server; use when updating
379 * parameters.
380 *
381 * dl_server_init() -- initializes the server.
382 *
383 * When started the dl_server will (per dl_defer) schedule a timer for its
384 * zero-laxity point -- that is, unlike regular EDF tasks which run ASAP, a
385 * server will run at the very end of its period.
386 *
387 * This is done such that any runtime from the target class can be accounted
388 * against the server -- through dl_server_update() above -- such that when it
389 * becomes time to run, it might already be out of runtime and get deferred
390 * until the next period. In this case dl_server_timer() will alternate
391 * between defer and replenish but never actually enqueue the server.
392 *
393 * Only when the target class does not manage to exhaust the server's runtime
394 * (there's actualy starvation in the given period), will the dl_server get on
395 * the runqueue. Once queued it will pick tasks from the target class and run
396 * them until either its runtime is exhaused, at which point its back to
397 * dl_server_timer, or until there are no more tasks to run, at which point
398 * the dl_server stops itself.
399 *
400 * By stopping at this point the dl_server retains bandwidth, which, if a new
401 * task wakes up imminently (starting the server again), can be used --
402 * subject to CBS wakeup rules -- without having to wait for the next period.
403 *
404 * Additionally, because of the dl_defer behaviour the start/stop behaviour is
405 * naturally thottled to once per period, avoiding high context switch
406 * workloads from spamming the hrtimer program/cancel paths.
407 */
408extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
409extern void dl_server_start(struct sched_dl_entity *dl_se);
410extern void dl_server_stop(struct sched_dl_entity *dl_se);
411extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
412 dl_server_pick_f pick_task);
413extern void sched_init_dl_servers(void);
414
415extern void dl_server_update_idle_time(struct rq *rq,
416 struct task_struct *p);
417extern void fair_server_init(struct rq *rq);
418extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
419extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
420 u64 runtime, u64 period, bool init);
421
422static inline bool dl_server_active(struct sched_dl_entity *dl_se)
423{
424 return dl_se->dl_server_active;
425}
426
427#ifdef CONFIG_CGROUP_SCHED
428
429extern struct list_head task_groups;
430
431#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
432extern const u64 max_bw_quota_period_us;
433
434/*
435 * default period for group bandwidth.
436 * default: 0.1s, units: microseconds
437 */
438static inline u64 default_bw_period_us(void)
439{
440 return 100000ULL;
441}
442#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */
443
444struct cfs_bandwidth {
445#ifdef CONFIG_CFS_BANDWIDTH
446 raw_spinlock_t lock;
447 ktime_t period;
448 u64 quota;
449 u64 runtime;
450 u64 burst;
451 u64 runtime_snap;
452 s64 hierarchical_quota;
453
454 u8 idle;
455 u8 period_active;
456 u8 slack_started;
457 struct hrtimer period_timer;
458 struct hrtimer slack_timer;
459 struct list_head throttled_cfs_rq;
460
461 /* Statistics: */
462 int nr_periods;
463 int nr_throttled;
464 int nr_burst;
465 u64 throttled_time;
466 u64 burst_time;
467#endif /* CONFIG_CFS_BANDWIDTH */
468};
469
470/* Task group related information */
471struct task_group {
472 struct cgroup_subsys_state css;
473
474#ifdef CONFIG_GROUP_SCHED_WEIGHT
475 /* A positive value indicates that this is a SCHED_IDLE group. */
476 int idle;
477#endif
478
479#ifdef CONFIG_FAIR_GROUP_SCHED
480 /* schedulable entities of this group on each CPU */
481 struct sched_entity **se;
482 /* runqueue "owned" by this group on each CPU */
483 struct cfs_rq **cfs_rq;
484 unsigned long shares;
485 /*
486 * load_avg can be heavily contended at clock tick time, so put
487 * it in its own cache-line separated from the fields above which
488 * will also be accessed at each tick.
489 */
490 atomic_long_t load_avg ____cacheline_aligned;
491#endif /* CONFIG_FAIR_GROUP_SCHED */
492
493#ifdef CONFIG_RT_GROUP_SCHED
494 struct sched_rt_entity **rt_se;
495 struct rt_rq **rt_rq;
496
497 struct rt_bandwidth rt_bandwidth;
498#endif
499
500 struct scx_task_group scx;
501
502 struct rcu_head rcu;
503 struct list_head list;
504
505 struct task_group *parent;
506 struct list_head siblings;
507 struct list_head children;
508
509#ifdef CONFIG_SCHED_AUTOGROUP
510 struct autogroup *autogroup;
511#endif
512
513 struct cfs_bandwidth cfs_bandwidth;
514
515#ifdef CONFIG_UCLAMP_TASK_GROUP
516 /* The two decimal precision [%] value requested from user-space */
517 unsigned int uclamp_pct[UCLAMP_CNT];
518 /* Clamp values requested for a task group */
519 struct uclamp_se uclamp_req[UCLAMP_CNT];
520 /* Effective clamp values used for a task group */
521 struct uclamp_se uclamp[UCLAMP_CNT];
522#endif
523
524};
525
526#ifdef CONFIG_GROUP_SCHED_WEIGHT
527#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
528
529/*
530 * A weight of 0 or 1 can cause arithmetics problems.
531 * A weight of a cfs_rq is the sum of weights of which entities
532 * are queued on this cfs_rq, so a weight of a entity should not be
533 * too large, so as the shares value of a task group.
534 * (The default weight is 1024 - so there's no practical
535 * limitation from this.)
536 */
537#define MIN_SHARES (1UL << 1)
538#define MAX_SHARES (1UL << 18)
539#endif
540
541typedef int (*tg_visitor)(struct task_group *, void *);
542
543extern int walk_tg_tree_from(struct task_group *from,
544 tg_visitor down, tg_visitor up, void *data);
545
546/*
547 * Iterate the full tree, calling @down when first entering a node and @up when
548 * leaving it for the final time.
549 *
550 * Caller must hold rcu_lock or sufficient equivalent.
551 */
552static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
553{
554 return walk_tg_tree_from(from: &root_task_group, down, up, data);
555}
556
557static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
558{
559 return css ? container_of(css, struct task_group, css) : NULL;
560}
561
562extern int tg_nop(struct task_group *tg, void *data);
563
564#ifdef CONFIG_FAIR_GROUP_SCHED
565extern void free_fair_sched_group(struct task_group *tg);
566extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
567extern void online_fair_sched_group(struct task_group *tg);
568extern void unregister_fair_sched_group(struct task_group *tg);
569#else /* !CONFIG_FAIR_GROUP_SCHED: */
570static inline void free_fair_sched_group(struct task_group *tg) { }
571static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
572{
573 return 1;
574}
575static inline void online_fair_sched_group(struct task_group *tg) { }
576static inline void unregister_fair_sched_group(struct task_group *tg) { }
577#endif /* !CONFIG_FAIR_GROUP_SCHED */
578
579extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
580 struct sched_entity *se, int cpu,
581 struct sched_entity *parent);
582extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
583
584extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
585extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
586extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
587extern bool cfs_task_bw_constrained(struct task_struct *p);
588
589extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
590 struct sched_rt_entity *rt_se, int cpu,
591 struct sched_rt_entity *parent);
592extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
593extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
594extern long sched_group_rt_runtime(struct task_group *tg);
595extern long sched_group_rt_period(struct task_group *tg);
596extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
597
598extern struct task_group *sched_create_group(struct task_group *parent);
599extern void sched_online_group(struct task_group *tg,
600 struct task_group *parent);
601extern void sched_destroy_group(struct task_group *tg);
602extern void sched_release_group(struct task_group *tg);
603
604extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
605
606#ifdef CONFIG_FAIR_GROUP_SCHED
607extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
608
609extern int sched_group_set_idle(struct task_group *tg, long idle);
610
611extern void set_task_rq_fair(struct sched_entity *se,
612 struct cfs_rq *prev, struct cfs_rq *next);
613#else /* !CONFIG_FAIR_GROUP_SCHED: */
614static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
615static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
616#endif /* !CONFIG_FAIR_GROUP_SCHED */
617
618#else /* !CONFIG_CGROUP_SCHED: */
619
620struct cfs_bandwidth { };
621
622static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
623
624#endif /* !CONFIG_CGROUP_SCHED */
625
626extern void unregister_rt_sched_group(struct task_group *tg);
627extern void free_rt_sched_group(struct task_group *tg);
628extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
629
630/*
631 * u64_u32_load/u64_u32_store
632 *
633 * Use a copy of a u64 value to protect against data race. This is only
634 * applicable for 32-bits architectures.
635 */
636#ifdef CONFIG_64BIT
637# define u64_u32_load_copy(var, copy) var
638# define u64_u32_store_copy(var, copy, val) (var = val)
639#else
640# define u64_u32_load_copy(var, copy) \
641({ \
642 u64 __val, __val_copy; \
643 do { \
644 __val_copy = copy; \
645 /* \
646 * paired with u64_u32_store_copy(), ordering access \
647 * to var and copy. \
648 */ \
649 smp_rmb(); \
650 __val = var; \
651 } while (__val != __val_copy); \
652 __val; \
653})
654# define u64_u32_store_copy(var, copy, val) \
655do { \
656 typeof(val) __val = (val); \
657 var = __val; \
658 /* \
659 * paired with u64_u32_load_copy(), ordering access to var and \
660 * copy. \
661 */ \
662 smp_wmb(); \
663 copy = __val; \
664} while (0)
665#endif
666# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
667# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
668
669struct balance_callback {
670 struct balance_callback *next;
671 void (*func)(struct rq *rq);
672};
673
674/* CFS-related fields in a runqueue */
675struct cfs_rq {
676 struct load_weight load;
677 unsigned int nr_queued;
678 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */
679 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */
680 unsigned int h_nr_idle; /* SCHED_IDLE */
681
682 s64 avg_vruntime;
683 u64 avg_load;
684
685 u64 min_vruntime;
686#ifdef CONFIG_SCHED_CORE
687 unsigned int forceidle_seq;
688 u64 min_vruntime_fi;
689#endif
690
691 struct rb_root_cached tasks_timeline;
692
693 /*
694 * 'curr' points to currently running entity on this cfs_rq.
695 * It is set to NULL otherwise (i.e when none are currently running).
696 */
697 struct sched_entity *curr;
698 struct sched_entity *next;
699
700 /*
701 * CFS load tracking
702 */
703 struct sched_avg avg;
704#ifndef CONFIG_64BIT
705 u64 last_update_time_copy;
706#endif
707 struct {
708 raw_spinlock_t lock ____cacheline_aligned;
709 int nr;
710 unsigned long load_avg;
711 unsigned long util_avg;
712 unsigned long runnable_avg;
713 } removed;
714
715#ifdef CONFIG_FAIR_GROUP_SCHED
716 u64 last_update_tg_load_avg;
717 unsigned long tg_load_avg_contrib;
718 long propagate;
719 long prop_runnable_sum;
720
721 /*
722 * h_load = weight * f(tg)
723 *
724 * Where f(tg) is the recursive weight fraction assigned to
725 * this group.
726 */
727 unsigned long h_load;
728 u64 last_h_load_update;
729 struct sched_entity *h_load_next;
730#endif /* CONFIG_FAIR_GROUP_SCHED */
731
732#ifdef CONFIG_FAIR_GROUP_SCHED
733 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
734
735 /*
736 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
737 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
738 * (like users, containers etc.)
739 *
740 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
741 * This list is used during load balance.
742 */
743 int on_list;
744 struct list_head leaf_cfs_rq_list;
745 struct task_group *tg; /* group that "owns" this runqueue */
746
747 /* Locally cached copy of our task_group's idle value */
748 int idle;
749
750#ifdef CONFIG_CFS_BANDWIDTH
751 int runtime_enabled;
752 s64 runtime_remaining;
753
754 u64 throttled_pelt_idle;
755#ifndef CONFIG_64BIT
756 u64 throttled_pelt_idle_copy;
757#endif
758 u64 throttled_clock;
759 u64 throttled_clock_pelt;
760 u64 throttled_clock_pelt_time;
761 u64 throttled_clock_self;
762 u64 throttled_clock_self_time;
763 bool throttled:1;
764 bool pelt_clock_throttled:1;
765 int throttle_count;
766 struct list_head throttled_list;
767 struct list_head throttled_csd_list;
768 struct list_head throttled_limbo_list;
769#endif /* CONFIG_CFS_BANDWIDTH */
770#endif /* CONFIG_FAIR_GROUP_SCHED */
771};
772
773#ifdef CONFIG_SCHED_CLASS_EXT
774/* scx_rq->flags, protected by the rq lock */
775enum scx_rq_flags {
776 /*
777 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
778 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
779 * only while the BPF scheduler considers the CPU to be online.
780 */
781 SCX_RQ_ONLINE = 1 << 0,
782 SCX_RQ_CAN_STOP_TICK = 1 << 1,
783 SCX_RQ_BAL_PENDING = 1 << 2, /* balance hasn't run yet */
784 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
785 SCX_RQ_BYPASSING = 1 << 4,
786 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */
787
788 SCX_RQ_IN_WAKEUP = 1 << 16,
789 SCX_RQ_IN_BALANCE = 1 << 17,
790};
791
792struct scx_rq {
793 struct scx_dispatch_q local_dsq;
794 struct list_head runnable_list; /* runnable tasks on this rq */
795 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
796 unsigned long ops_qseq;
797 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
798 u32 nr_running;
799 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
800 bool cpu_released;
801 u32 flags;
802 u64 clock; /* current per-rq clock -- see scx_bpf_now() */
803 cpumask_var_t cpus_to_kick;
804 cpumask_var_t cpus_to_kick_if_idle;
805 cpumask_var_t cpus_to_preempt;
806 cpumask_var_t cpus_to_wait;
807 unsigned long pnt_seq;
808 struct balance_callback deferred_bal_cb;
809 struct irq_work deferred_irq_work;
810 struct irq_work kick_cpus_irq_work;
811};
812#endif /* CONFIG_SCHED_CLASS_EXT */
813
814static inline int rt_bandwidth_enabled(void)
815{
816 return sysctl_sched_rt_runtime >= 0;
817}
818
819/* RT IPI pull logic requires IRQ_WORK */
820#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
821# define HAVE_RT_PUSH_IPI
822#endif
823
824/* Real-Time classes' related field in a runqueue: */
825struct rt_rq {
826 struct rt_prio_array active;
827 unsigned int rt_nr_running;
828 unsigned int rr_nr_running;
829 struct {
830 int curr; /* highest queued rt task prio */
831 int next; /* next highest */
832 } highest_prio;
833 bool overloaded;
834 struct plist_head pushable_tasks;
835
836 int rt_queued;
837
838#ifdef CONFIG_RT_GROUP_SCHED
839 int rt_throttled;
840 u64 rt_time; /* consumed RT time, goes up in update_curr_rt */
841 u64 rt_runtime; /* allotted RT time, "slice" from rt_bandwidth, RT sharing/balancing */
842 /* Nests inside the rq lock: */
843 raw_spinlock_t rt_runtime_lock;
844
845 unsigned int rt_nr_boosted;
846
847 struct rq *rq; /* this is always top-level rq, cache? */
848#endif
849#ifdef CONFIG_CGROUP_SCHED
850 struct task_group *tg; /* this tg has "this" rt_rq on given CPU for runnable entities */
851#endif
852};
853
854static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
855{
856 return rt_rq->rt_queued && rt_rq->rt_nr_running;
857}
858
859/* Deadline class' related fields in a runqueue */
860struct dl_rq {
861 /* runqueue is an rbtree, ordered by deadline */
862 struct rb_root_cached root;
863
864 unsigned int dl_nr_running;
865
866 /*
867 * Deadline values of the currently executing and the
868 * earliest ready task on this rq. Caching these facilitates
869 * the decision whether or not a ready but not running task
870 * should migrate somewhere else.
871 */
872 struct {
873 u64 curr;
874 u64 next;
875 } earliest_dl;
876
877 bool overloaded;
878
879 /*
880 * Tasks on this rq that can be pushed away. They are kept in
881 * an rb-tree, ordered by tasks' deadlines, with caching
882 * of the leftmost (earliest deadline) element.
883 */
884 struct rb_root_cached pushable_dl_tasks_root;
885
886 /*
887 * "Active utilization" for this runqueue: increased when a
888 * task wakes up (becomes TASK_RUNNING) and decreased when a
889 * task blocks
890 */
891 u64 running_bw;
892
893 /*
894 * Utilization of the tasks "assigned" to this runqueue (including
895 * the tasks that are in runqueue and the tasks that executed on this
896 * CPU and blocked). Increased when a task moves to this runqueue, and
897 * decreased when the task moves away (migrates, changes scheduling
898 * policy, or terminates).
899 * This is needed to compute the "inactive utilization" for the
900 * runqueue (inactive utilization = this_bw - running_bw).
901 */
902 u64 this_bw;
903 u64 extra_bw;
904
905 /*
906 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
907 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
908 */
909 u64 max_bw;
910
911 /*
912 * Inverse of the fraction of CPU utilization that can be reclaimed
913 * by the GRUB algorithm.
914 */
915 u64 bw_ratio;
916};
917
918#ifdef CONFIG_FAIR_GROUP_SCHED
919
920/* An entity is a task if it doesn't "own" a runqueue */
921#define entity_is_task(se) (!se->my_q)
922
923static inline void se_update_runnable(struct sched_entity *se)
924{
925 if (!entity_is_task(se))
926 se->runnable_weight = se->my_q->h_nr_runnable;
927}
928
929static inline long se_runnable(struct sched_entity *se)
930{
931 if (se->sched_delayed)
932 return false;
933
934 if (entity_is_task(se))
935 return !!se->on_rq;
936 else
937 return se->runnable_weight;
938}
939
940#else /* !CONFIG_FAIR_GROUP_SCHED: */
941
942#define entity_is_task(se) 1
943
944static inline void se_update_runnable(struct sched_entity *se) { }
945
946static inline long se_runnable(struct sched_entity *se)
947{
948 if (se->sched_delayed)
949 return false;
950
951 return !!se->on_rq;
952}
953
954#endif /* !CONFIG_FAIR_GROUP_SCHED */
955
956/*
957 * XXX we want to get rid of these helpers and use the full load resolution.
958 */
959static inline long se_weight(struct sched_entity *se)
960{
961 return scale_load_down(se->load.weight);
962}
963
964
965static inline bool sched_asym_prefer(int a, int b)
966{
967 return arch_asym_cpu_priority(cpu: a) > arch_asym_cpu_priority(cpu: b);
968}
969
970struct perf_domain {
971 struct em_perf_domain *em_pd;
972 struct perf_domain *next;
973 struct rcu_head rcu;
974};
975
976/*
977 * We add the notion of a root-domain which will be used to define per-domain
978 * variables. Each exclusive cpuset essentially defines an island domain by
979 * fully partitioning the member CPUs from any other cpuset. Whenever a new
980 * exclusive cpuset is created, we also create and attach a new root-domain
981 * object.
982 *
983 */
984struct root_domain {
985 atomic_t refcount;
986 atomic_t rto_count;
987 struct rcu_head rcu;
988 cpumask_var_t span;
989 cpumask_var_t online;
990
991 /*
992 * Indicate pullable load on at least one CPU, e.g:
993 * - More than one runnable task
994 * - Running task is misfit
995 */
996 bool overloaded;
997
998 /* Indicate one or more CPUs over-utilized (tipping point) */
999 bool overutilized;
1000
1001 /*
1002 * The bit corresponding to a CPU gets set here if such CPU has more
1003 * than one runnable -deadline task (as it is below for RT tasks).
1004 */
1005 cpumask_var_t dlo_mask;
1006 atomic_t dlo_count;
1007 struct dl_bw dl_bw;
1008 struct cpudl cpudl;
1009
1010 /*
1011 * Indicate whether a root_domain's dl_bw has been checked or
1012 * updated. It's monotonously increasing value.
1013 *
1014 * Also, some corner cases, like 'wrap around' is dangerous, but given
1015 * that u64 is 'big enough'. So that shouldn't be a concern.
1016 */
1017 u64 visit_cookie;
1018
1019#ifdef HAVE_RT_PUSH_IPI
1020 /*
1021 * For IPI pull requests, loop across the rto_mask.
1022 */
1023 struct irq_work rto_push_work;
1024 raw_spinlock_t rto_lock;
1025 /* These are only updated and read within rto_lock */
1026 int rto_loop;
1027 int rto_cpu;
1028 /* These atomics are updated outside of a lock */
1029 atomic_t rto_loop_next;
1030 atomic_t rto_loop_start;
1031#endif /* HAVE_RT_PUSH_IPI */
1032 /*
1033 * The "RT overload" flag: it gets set if a CPU has more than
1034 * one runnable RT task.
1035 */
1036 cpumask_var_t rto_mask;
1037 struct cpupri cpupri;
1038
1039 /*
1040 * NULL-terminated list of performance domains intersecting with the
1041 * CPUs of the rd. Protected by RCU.
1042 */
1043 struct perf_domain __rcu *pd;
1044};
1045
1046extern void init_defrootdomain(void);
1047extern int sched_init_domains(const struct cpumask *cpu_map);
1048extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1049extern void sched_get_rd(struct root_domain *rd);
1050extern void sched_put_rd(struct root_domain *rd);
1051
1052static inline int get_rd_overloaded(struct root_domain *rd)
1053{
1054 return READ_ONCE(rd->overloaded);
1055}
1056
1057static inline void set_rd_overloaded(struct root_domain *rd, int status)
1058{
1059 if (get_rd_overloaded(rd) != status)
1060 WRITE_ONCE(rd->overloaded, status);
1061}
1062
1063#ifdef HAVE_RT_PUSH_IPI
1064extern void rto_push_irq_work_func(struct irq_work *work);
1065#endif
1066
1067#ifdef CONFIG_UCLAMP_TASK
1068/*
1069 * struct uclamp_bucket - Utilization clamp bucket
1070 * @value: utilization clamp value for tasks on this clamp bucket
1071 * @tasks: number of RUNNABLE tasks on this clamp bucket
1072 *
1073 * Keep track of how many tasks are RUNNABLE for a given utilization
1074 * clamp value.
1075 */
1076struct uclamp_bucket {
1077 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1078 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1079};
1080
1081/*
1082 * struct uclamp_rq - rq's utilization clamp
1083 * @value: currently active clamp values for a rq
1084 * @bucket: utilization clamp buckets affecting a rq
1085 *
1086 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1087 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1088 * (or actually running) with that value.
1089 *
1090 * There are up to UCLAMP_CNT possible different clamp values, currently there
1091 * are only two: minimum utilization and maximum utilization.
1092 *
1093 * All utilization clamping values are MAX aggregated, since:
1094 * - for util_min: we want to run the CPU at least at the max of the minimum
1095 * utilization required by its currently RUNNABLE tasks.
1096 * - for util_max: we want to allow the CPU to run up to the max of the
1097 * maximum utilization allowed by its currently RUNNABLE tasks.
1098 *
1099 * Since on each system we expect only a limited number of different
1100 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1101 * the metrics required to compute all the per-rq utilization clamp values.
1102 */
1103struct uclamp_rq {
1104 unsigned int value;
1105 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1106};
1107
1108DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1109#endif /* CONFIG_UCLAMP_TASK */
1110
1111/*
1112 * This is the main, per-CPU runqueue data structure.
1113 *
1114 * Locking rule: those places that want to lock multiple runqueues
1115 * (such as the load balancing or the thread migration code), lock
1116 * acquire operations must be ordered by ascending &runqueue.
1117 */
1118struct rq {
1119 /* runqueue lock: */
1120 raw_spinlock_t __lock;
1121
1122 unsigned int nr_running;
1123#ifdef CONFIG_NUMA_BALANCING
1124 unsigned int nr_numa_running;
1125 unsigned int nr_preferred_running;
1126 unsigned int numa_migrate_on;
1127#endif
1128#ifdef CONFIG_NO_HZ_COMMON
1129 unsigned long last_blocked_load_update_tick;
1130 unsigned int has_blocked_load;
1131 call_single_data_t nohz_csd;
1132 unsigned int nohz_tick_stopped;
1133 atomic_t nohz_flags;
1134#endif /* CONFIG_NO_HZ_COMMON */
1135
1136 unsigned int ttwu_pending;
1137 u64 nr_switches;
1138
1139#ifdef CONFIG_UCLAMP_TASK
1140 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1141 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1142 unsigned int uclamp_flags;
1143#define UCLAMP_FLAG_IDLE 0x01
1144#endif
1145
1146 struct cfs_rq cfs;
1147 struct rt_rq rt;
1148 struct dl_rq dl;
1149#ifdef CONFIG_SCHED_CLASS_EXT
1150 struct scx_rq scx;
1151#endif
1152
1153 struct sched_dl_entity fair_server;
1154
1155#ifdef CONFIG_FAIR_GROUP_SCHED
1156 /* list of leaf cfs_rq on this CPU: */
1157 struct list_head leaf_cfs_rq_list;
1158 struct list_head *tmp_alone_branch;
1159#endif /* CONFIG_FAIR_GROUP_SCHED */
1160
1161 /*
1162 * This is part of a global counter where only the total sum
1163 * over all CPUs matters. A task can increase this counter on
1164 * one CPU and if it got migrated afterwards it may decrease
1165 * it on another CPU. Always updated under the runqueue lock:
1166 */
1167 unsigned long nr_uninterruptible;
1168
1169#ifdef CONFIG_SCHED_PROXY_EXEC
1170 struct task_struct __rcu *donor; /* Scheduling context */
1171 struct task_struct __rcu *curr; /* Execution context */
1172#else
1173 union {
1174 struct task_struct __rcu *donor; /* Scheduler context */
1175 struct task_struct __rcu *curr; /* Execution context */
1176 };
1177#endif
1178 struct sched_dl_entity *dl_server;
1179 struct task_struct *idle;
1180 struct task_struct *stop;
1181 unsigned long next_balance;
1182 struct mm_struct *prev_mm;
1183
1184 unsigned int clock_update_flags;
1185 u64 clock;
1186 /* Ensure that all clocks are in the same cache line */
1187 u64 clock_task ____cacheline_aligned;
1188 u64 clock_pelt;
1189 unsigned long lost_idle_time;
1190 u64 clock_pelt_idle;
1191 u64 clock_idle;
1192#ifndef CONFIG_64BIT
1193 u64 clock_pelt_idle_copy;
1194 u64 clock_idle_copy;
1195#endif
1196
1197 atomic_t nr_iowait;
1198
1199 u64 last_seen_need_resched_ns;
1200 int ticks_without_resched;
1201
1202#ifdef CONFIG_MEMBARRIER
1203 int membarrier_state;
1204#endif
1205
1206 struct root_domain *rd;
1207 struct sched_domain __rcu *sd;
1208
1209 unsigned long cpu_capacity;
1210
1211 struct balance_callback *balance_callback;
1212
1213 unsigned char nohz_idle_balance;
1214 unsigned char idle_balance;
1215
1216 unsigned long misfit_task_load;
1217
1218 /* For active balancing */
1219 int active_balance;
1220 int push_cpu;
1221 struct cpu_stop_work active_balance_work;
1222
1223 /* CPU of this runqueue: */
1224 int cpu;
1225 int online;
1226
1227 struct list_head cfs_tasks;
1228
1229 struct sched_avg avg_rt;
1230 struct sched_avg avg_dl;
1231#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1232 struct sched_avg avg_irq;
1233#endif
1234#ifdef CONFIG_SCHED_HW_PRESSURE
1235 struct sched_avg avg_hw;
1236#endif
1237 u64 idle_stamp;
1238 u64 avg_idle;
1239
1240 /* This is used to determine avg_idle's max value */
1241 u64 max_idle_balance_cost;
1242
1243#ifdef CONFIG_HOTPLUG_CPU
1244 struct rcuwait hotplug_wait;
1245#endif
1246
1247#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1248 u64 prev_irq_time;
1249 u64 psi_irq_time;
1250#endif
1251#ifdef CONFIG_PARAVIRT
1252 u64 prev_steal_time;
1253#endif
1254#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1255 u64 prev_steal_time_rq;
1256#endif
1257
1258 /* calc_load related fields */
1259 unsigned long calc_load_update;
1260 long calc_load_active;
1261
1262#ifdef CONFIG_SCHED_HRTICK
1263 call_single_data_t hrtick_csd;
1264 struct hrtimer hrtick_timer;
1265 ktime_t hrtick_time;
1266#endif
1267
1268#ifdef CONFIG_SCHEDSTATS
1269 /* latency stats */
1270 struct sched_info rq_sched_info;
1271 unsigned long long rq_cpu_time;
1272
1273 /* sys_sched_yield() stats */
1274 unsigned int yld_count;
1275
1276 /* schedule() stats */
1277 unsigned int sched_count;
1278 unsigned int sched_goidle;
1279
1280 /* try_to_wake_up() stats */
1281 unsigned int ttwu_count;
1282 unsigned int ttwu_local;
1283#endif
1284
1285#ifdef CONFIG_CPU_IDLE
1286 /* Must be inspected within a RCU lock section */
1287 struct cpuidle_state *idle_state;
1288#endif
1289
1290 unsigned int nr_pinned;
1291 unsigned int push_busy;
1292 struct cpu_stop_work push_work;
1293
1294#ifdef CONFIG_SCHED_CORE
1295 /* per rq */
1296 struct rq *core;
1297 struct task_struct *core_pick;
1298 struct sched_dl_entity *core_dl_server;
1299 unsigned int core_enabled;
1300 unsigned int core_sched_seq;
1301 struct rb_root core_tree;
1302
1303 /* shared state -- careful with sched_core_cpu_deactivate() */
1304 unsigned int core_task_seq;
1305 unsigned int core_pick_seq;
1306 unsigned long core_cookie;
1307 unsigned int core_forceidle_count;
1308 unsigned int core_forceidle_seq;
1309 unsigned int core_forceidle_occupation;
1310 u64 core_forceidle_start;
1311#endif /* CONFIG_SCHED_CORE */
1312
1313 /* Scratch cpumask to be temporarily used under rq_lock */
1314 cpumask_var_t scratch_mask;
1315
1316#ifdef CONFIG_CFS_BANDWIDTH
1317 call_single_data_t cfsb_csd;
1318 struct list_head cfsb_csd_list;
1319#endif
1320};
1321
1322#ifdef CONFIG_FAIR_GROUP_SCHED
1323
1324/* CPU runqueue to which this cfs_rq is attached */
1325static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1326{
1327 return cfs_rq->rq;
1328}
1329
1330#else /* !CONFIG_FAIR_GROUP_SCHED: */
1331
1332static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1333{
1334 return container_of(cfs_rq, struct rq, cfs);
1335}
1336#endif /* !CONFIG_FAIR_GROUP_SCHED */
1337
1338static inline int cpu_of(struct rq *rq)
1339{
1340 return rq->cpu;
1341}
1342
1343#define MDF_PUSH 0x01
1344
1345static inline bool is_migration_disabled(struct task_struct *p)
1346{
1347 return p->migration_disabled;
1348}
1349
1350DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1351
1352#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1353#define this_rq() this_cpu_ptr(&runqueues)
1354#define task_rq(p) cpu_rq(task_cpu(p))
1355#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1356#define raw_rq() raw_cpu_ptr(&runqueues)
1357
1358#ifdef CONFIG_SCHED_PROXY_EXEC
1359static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1360{
1361 rcu_assign_pointer(rq->donor, t);
1362}
1363#else
1364static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1365{
1366 /* Do nothing */
1367}
1368#endif
1369
1370#ifdef CONFIG_SCHED_CORE
1371static inline struct cpumask *sched_group_span(struct sched_group *sg);
1372
1373DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1374
1375static inline bool sched_core_enabled(struct rq *rq)
1376{
1377 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1378}
1379
1380static inline bool sched_core_disabled(void)
1381{
1382 return !static_branch_unlikely(&__sched_core_enabled);
1383}
1384
1385/*
1386 * Be careful with this function; not for general use. The return value isn't
1387 * stable unless you actually hold a relevant rq->__lock.
1388 */
1389static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1390{
1391 if (sched_core_enabled(rq))
1392 return &rq->core->__lock;
1393
1394 return &rq->__lock;
1395}
1396
1397static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1398{
1399 if (rq->core_enabled)
1400 return &rq->core->__lock;
1401
1402 return &rq->__lock;
1403}
1404
1405extern bool
1406cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1407
1408extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1409
1410/*
1411 * Helpers to check if the CPU's core cookie matches with the task's cookie
1412 * when core scheduling is enabled.
1413 * A special case is that the task's cookie always matches with CPU's core
1414 * cookie if the CPU is in an idle core.
1415 */
1416static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1417{
1418 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1419 if (!sched_core_enabled(rq))
1420 return true;
1421
1422 return rq->core->core_cookie == p->core_cookie;
1423}
1424
1425static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1426{
1427 bool idle_core = true;
1428 int cpu;
1429
1430 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1431 if (!sched_core_enabled(rq))
1432 return true;
1433
1434 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1435 if (!available_idle_cpu(cpu)) {
1436 idle_core = false;
1437 break;
1438 }
1439 }
1440
1441 /*
1442 * A CPU in an idle core is always the best choice for tasks with
1443 * cookies.
1444 */
1445 return idle_core || rq->core->core_cookie == p->core_cookie;
1446}
1447
1448static inline bool sched_group_cookie_match(struct rq *rq,
1449 struct task_struct *p,
1450 struct sched_group *group)
1451{
1452 int cpu;
1453
1454 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1455 if (!sched_core_enabled(rq))
1456 return true;
1457
1458 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1459 if (sched_core_cookie_match(cpu_rq(cpu), p))
1460 return true;
1461 }
1462 return false;
1463}
1464
1465static inline bool sched_core_enqueued(struct task_struct *p)
1466{
1467 return !RB_EMPTY_NODE(&p->core_node);
1468}
1469
1470extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1471extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1472
1473extern void sched_core_get(void);
1474extern void sched_core_put(void);
1475
1476#else /* !CONFIG_SCHED_CORE: */
1477
1478static inline bool sched_core_enabled(struct rq *rq)
1479{
1480 return false;
1481}
1482
1483static inline bool sched_core_disabled(void)
1484{
1485 return true;
1486}
1487
1488static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1489{
1490 return &rq->__lock;
1491}
1492
1493static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1494{
1495 return &rq->__lock;
1496}
1497
1498static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1499{
1500 return true;
1501}
1502
1503static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1504{
1505 return true;
1506}
1507
1508static inline bool sched_group_cookie_match(struct rq *rq,
1509 struct task_struct *p,
1510 struct sched_group *group)
1511{
1512 return true;
1513}
1514
1515#endif /* !CONFIG_SCHED_CORE */
1516
1517#ifdef CONFIG_RT_GROUP_SCHED
1518# ifdef CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED
1519DECLARE_STATIC_KEY_FALSE(rt_group_sched);
1520static inline bool rt_group_sched_enabled(void)
1521{
1522 return static_branch_unlikely(&rt_group_sched);
1523}
1524# else /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED: */
1525DECLARE_STATIC_KEY_TRUE(rt_group_sched);
1526static inline bool rt_group_sched_enabled(void)
1527{
1528 return static_branch_likely(&rt_group_sched);
1529}
1530# endif /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED */
1531#else /* !CONFIG_RT_GROUP_SCHED: */
1532# define rt_group_sched_enabled() false
1533#endif /* !CONFIG_RT_GROUP_SCHED */
1534
1535static inline void lockdep_assert_rq_held(struct rq *rq)
1536{
1537 lockdep_assert_held(__rq_lockp(rq));
1538}
1539
1540extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1541extern bool raw_spin_rq_trylock(struct rq *rq);
1542extern void raw_spin_rq_unlock(struct rq *rq);
1543
1544static inline void raw_spin_rq_lock(struct rq *rq)
1545{
1546 raw_spin_rq_lock_nested(rq, subclass: 0);
1547}
1548
1549static inline void raw_spin_rq_lock_irq(struct rq *rq)
1550{
1551 local_irq_disable();
1552 raw_spin_rq_lock(rq);
1553}
1554
1555static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1556{
1557 raw_spin_rq_unlock(rq);
1558 local_irq_enable();
1559}
1560
1561static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1562{
1563 unsigned long flags;
1564
1565 local_irq_save(flags);
1566 raw_spin_rq_lock(rq);
1567
1568 return flags;
1569}
1570
1571static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1572{
1573 raw_spin_rq_unlock(rq);
1574 local_irq_restore(flags);
1575}
1576
1577#define raw_spin_rq_lock_irqsave(rq, flags) \
1578do { \
1579 flags = _raw_spin_rq_lock_irqsave(rq); \
1580} while (0)
1581
1582#ifdef CONFIG_SCHED_SMT
1583extern void __update_idle_core(struct rq *rq);
1584
1585static inline void update_idle_core(struct rq *rq)
1586{
1587 if (static_branch_unlikely(&sched_smt_present))
1588 __update_idle_core(rq);
1589}
1590
1591#else /* !CONFIG_SCHED_SMT: */
1592static inline void update_idle_core(struct rq *rq) { }
1593#endif /* !CONFIG_SCHED_SMT */
1594
1595#ifdef CONFIG_FAIR_GROUP_SCHED
1596
1597static inline struct task_struct *task_of(struct sched_entity *se)
1598{
1599 WARN_ON_ONCE(!entity_is_task(se));
1600 return container_of(se, struct task_struct, se);
1601}
1602
1603static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1604{
1605 return p->se.cfs_rq;
1606}
1607
1608/* runqueue on which this entity is (to be) queued */
1609static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1610{
1611 return se->cfs_rq;
1612}
1613
1614/* runqueue "owned" by this group */
1615static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1616{
1617 return grp->my_q;
1618}
1619
1620#else /* !CONFIG_FAIR_GROUP_SCHED: */
1621
1622#define task_of(_se) container_of(_se, struct task_struct, se)
1623
1624static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1625{
1626 return &task_rq(p)->cfs;
1627}
1628
1629static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1630{
1631 const struct task_struct *p = task_of(se);
1632 struct rq *rq = task_rq(p);
1633
1634 return &rq->cfs;
1635}
1636
1637/* runqueue "owned" by this group */
1638static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1639{
1640 return NULL;
1641}
1642
1643#endif /* !CONFIG_FAIR_GROUP_SCHED */
1644
1645extern void update_rq_clock(struct rq *rq);
1646
1647/*
1648 * rq::clock_update_flags bits
1649 *
1650 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1651 * call to __schedule(). This is an optimisation to avoid
1652 * neighbouring rq clock updates.
1653 *
1654 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1655 * in effect and calls to update_rq_clock() are being ignored.
1656 *
1657 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1658 * made to update_rq_clock() since the last time rq::lock was pinned.
1659 *
1660 * If inside of __schedule(), clock_update_flags will have been
1661 * shifted left (a left shift is a cheap operation for the fast path
1662 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1663 *
1664 * if (rq-clock_update_flags >= RQCF_UPDATED)
1665 *
1666 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1667 * one position though, because the next rq_unpin_lock() will shift it
1668 * back.
1669 */
1670#define RQCF_REQ_SKIP 0x01
1671#define RQCF_ACT_SKIP 0x02
1672#define RQCF_UPDATED 0x04
1673
1674static inline void assert_clock_updated(struct rq *rq)
1675{
1676 /*
1677 * The only reason for not seeing a clock update since the
1678 * last rq_pin_lock() is if we're currently skipping updates.
1679 */
1680 WARN_ON_ONCE(rq->clock_update_flags < RQCF_ACT_SKIP);
1681}
1682
1683static inline u64 rq_clock(struct rq *rq)
1684{
1685 lockdep_assert_rq_held(rq);
1686 assert_clock_updated(rq);
1687
1688 return rq->clock;
1689}
1690
1691static inline u64 rq_clock_task(struct rq *rq)
1692{
1693 lockdep_assert_rq_held(rq);
1694 assert_clock_updated(rq);
1695
1696 return rq->clock_task;
1697}
1698
1699static inline void rq_clock_skip_update(struct rq *rq)
1700{
1701 lockdep_assert_rq_held(rq);
1702 rq->clock_update_flags |= RQCF_REQ_SKIP;
1703}
1704
1705/*
1706 * See rt task throttling, which is the only time a skip
1707 * request is canceled.
1708 */
1709static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1710{
1711 lockdep_assert_rq_held(rq);
1712 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1713}
1714
1715/*
1716 * During cpu offlining and rq wide unthrottling, we can trigger
1717 * an update_rq_clock() for several cfs and rt runqueues (Typically
1718 * when using list_for_each_entry_*)
1719 * rq_clock_start_loop_update() can be called after updating the clock
1720 * once and before iterating over the list to prevent multiple update.
1721 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1722 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1723 */
1724static inline void rq_clock_start_loop_update(struct rq *rq)
1725{
1726 lockdep_assert_rq_held(rq);
1727 WARN_ON_ONCE(rq->clock_update_flags & RQCF_ACT_SKIP);
1728 rq->clock_update_flags |= RQCF_ACT_SKIP;
1729}
1730
1731static inline void rq_clock_stop_loop_update(struct rq *rq)
1732{
1733 lockdep_assert_rq_held(rq);
1734 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1735}
1736
1737struct rq_flags {
1738 unsigned long flags;
1739 struct pin_cookie cookie;
1740 /*
1741 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1742 * current pin context is stashed here in case it needs to be
1743 * restored in rq_repin_lock().
1744 */
1745 unsigned int clock_update_flags;
1746};
1747
1748extern struct balance_callback balance_push_callback;
1749
1750#ifdef CONFIG_SCHED_CLASS_EXT
1751extern const struct sched_class ext_sched_class;
1752
1753DECLARE_STATIC_KEY_FALSE(__scx_enabled); /* SCX BPF scheduler loaded */
1754DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
1755
1756#define scx_enabled() static_branch_unlikely(&__scx_enabled)
1757#define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
1758
1759static inline void scx_rq_clock_update(struct rq *rq, u64 clock)
1760{
1761 if (!scx_enabled())
1762 return;
1763 WRITE_ONCE(rq->scx.clock, clock);
1764 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID);
1765}
1766
1767static inline void scx_rq_clock_invalidate(struct rq *rq)
1768{
1769 if (!scx_enabled())
1770 return;
1771 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID);
1772}
1773
1774#else /* !CONFIG_SCHED_CLASS_EXT: */
1775#define scx_enabled() false
1776#define scx_switched_all() false
1777
1778static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {}
1779static inline void scx_rq_clock_invalidate(struct rq *rq) {}
1780#endif /* !CONFIG_SCHED_CLASS_EXT */
1781
1782/*
1783 * Lockdep annotation that avoids accidental unlocks; it's like a
1784 * sticky/continuous lockdep_assert_held().
1785 *
1786 * This avoids code that has access to 'struct rq *rq' (basically everything in
1787 * the scheduler) from accidentally unlocking the rq if they do not also have a
1788 * copy of the (on-stack) 'struct rq_flags rf'.
1789 *
1790 * Also see Documentation/locking/lockdep-design.rst.
1791 */
1792static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1793{
1794 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1795
1796 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1797 rf->clock_update_flags = 0;
1798 WARN_ON_ONCE(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1799}
1800
1801static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1802{
1803 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1804 rf->clock_update_flags = RQCF_UPDATED;
1805
1806 scx_rq_clock_invalidate(rq);
1807 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1808}
1809
1810static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1811{
1812 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1813
1814 /*
1815 * Restore the value we stashed in @rf for this pin context.
1816 */
1817 rq->clock_update_flags |= rf->clock_update_flags;
1818}
1819
1820extern
1821struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1822 __acquires(rq->lock);
1823
1824extern
1825struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1826 __acquires(p->pi_lock)
1827 __acquires(rq->lock);
1828
1829static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1830 __releases(rq->lock)
1831{
1832 rq_unpin_lock(rq, rf);
1833 raw_spin_rq_unlock(rq);
1834}
1835
1836static inline void
1837task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1838 __releases(rq->lock)
1839 __releases(p->pi_lock)
1840{
1841 rq_unpin_lock(rq, rf);
1842 raw_spin_rq_unlock(rq);
1843 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1844}
1845
1846DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1847 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1848 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1849 struct rq *rq; struct rq_flags rf)
1850
1851static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1852 __acquires(rq->lock)
1853{
1854 raw_spin_rq_lock_irqsave(rq, rf->flags);
1855 rq_pin_lock(rq, rf);
1856}
1857
1858static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1859 __acquires(rq->lock)
1860{
1861 raw_spin_rq_lock_irq(rq);
1862 rq_pin_lock(rq, rf);
1863}
1864
1865static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1866 __acquires(rq->lock)
1867{
1868 raw_spin_rq_lock(rq);
1869 rq_pin_lock(rq, rf);
1870}
1871
1872static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1873 __releases(rq->lock)
1874{
1875 rq_unpin_lock(rq, rf);
1876 raw_spin_rq_unlock_irqrestore(rq, flags: rf->flags);
1877}
1878
1879static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1880 __releases(rq->lock)
1881{
1882 rq_unpin_lock(rq, rf);
1883 raw_spin_rq_unlock_irq(rq);
1884}
1885
1886static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1887 __releases(rq->lock)
1888{
1889 rq_unpin_lock(rq, rf);
1890 raw_spin_rq_unlock(rq);
1891}
1892
1893DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1894 rq_lock(_T->lock, &_T->rf),
1895 rq_unlock(_T->lock, &_T->rf),
1896 struct rq_flags rf)
1897
1898DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1899 rq_lock_irq(_T->lock, &_T->rf),
1900 rq_unlock_irq(_T->lock, &_T->rf),
1901 struct rq_flags rf)
1902
1903DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1904 rq_lock_irqsave(_T->lock, &_T->rf),
1905 rq_unlock_irqrestore(_T->lock, &_T->rf),
1906 struct rq_flags rf)
1907
1908static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1909 __acquires(rq->lock)
1910{
1911 struct rq *rq;
1912
1913 local_irq_disable();
1914 rq = this_rq();
1915 rq_lock(rq, rf);
1916
1917 return rq;
1918}
1919
1920#ifdef CONFIG_NUMA
1921
1922enum numa_topology_type {
1923 NUMA_DIRECT,
1924 NUMA_GLUELESS_MESH,
1925 NUMA_BACKPLANE,
1926};
1927
1928extern enum numa_topology_type sched_numa_topology_type;
1929extern int sched_max_numa_distance;
1930extern bool find_numa_distance(int distance);
1931extern void sched_init_numa(int offline_node);
1932extern void sched_update_numa(int cpu, bool online);
1933extern void sched_domains_numa_masks_set(unsigned int cpu);
1934extern void sched_domains_numa_masks_clear(unsigned int cpu);
1935extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1936
1937#else /* !CONFIG_NUMA: */
1938
1939static inline void sched_init_numa(int offline_node) { }
1940static inline void sched_update_numa(int cpu, bool online) { }
1941static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1942static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1943
1944static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1945{
1946 return nr_cpu_ids;
1947}
1948
1949#endif /* !CONFIG_NUMA */
1950
1951#ifdef CONFIG_NUMA_BALANCING
1952
1953/* The regions in numa_faults array from task_struct */
1954enum numa_faults_stats {
1955 NUMA_MEM = 0,
1956 NUMA_CPU,
1957 NUMA_MEMBUF,
1958 NUMA_CPUBUF
1959};
1960
1961extern void sched_setnuma(struct task_struct *p, int node);
1962extern int migrate_task_to(struct task_struct *p, int cpu);
1963extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1964 int cpu, int scpu);
1965extern void init_numa_balancing(u64 clone_flags, struct task_struct *p);
1966
1967#else /* !CONFIG_NUMA_BALANCING: */
1968
1969static inline void
1970init_numa_balancing(u64 clone_flags, struct task_struct *p)
1971{
1972}
1973
1974#endif /* !CONFIG_NUMA_BALANCING */
1975
1976static inline void
1977queue_balance_callback(struct rq *rq,
1978 struct balance_callback *head,
1979 void (*func)(struct rq *rq))
1980{
1981 lockdep_assert_rq_held(rq);
1982
1983 /*
1984 * Don't (re)queue an already queued item; nor queue anything when
1985 * balance_push() is active, see the comment with
1986 * balance_push_callback.
1987 */
1988 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1989 return;
1990
1991 head->func = func;
1992 head->next = rq->balance_callback;
1993 rq->balance_callback = head;
1994}
1995
1996#define rcu_dereference_check_sched_domain(p) \
1997 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1998
1999/*
2000 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
2001 * See destroy_sched_domains: call_rcu for details.
2002 *
2003 * The domain tree of any CPU may only be accessed from within
2004 * preempt-disabled sections.
2005 */
2006#define for_each_domain(cpu, __sd) \
2007 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
2008 __sd; __sd = __sd->parent)
2009
2010/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
2011#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
2012static const unsigned int SD_SHARED_CHILD_MASK =
2013#include <linux/sched/sd_flags.h>
20140;
2015#undef SD_FLAG
2016
2017/**
2018 * highest_flag_domain - Return highest sched_domain containing flag.
2019 * @cpu: The CPU whose highest level of sched domain is to
2020 * be returned.
2021 * @flag: The flag to check for the highest sched_domain
2022 * for the given CPU.
2023 *
2024 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
2025 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
2026 */
2027static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
2028{
2029 struct sched_domain *sd, *hsd = NULL;
2030
2031 for_each_domain(cpu, sd) {
2032 if (sd->flags & flag) {
2033 hsd = sd;
2034 continue;
2035 }
2036
2037 /*
2038 * Stop the search if @flag is known to be shared at lower
2039 * levels. It will not be found further up.
2040 */
2041 if (flag & SD_SHARED_CHILD_MASK)
2042 break;
2043 }
2044
2045 return hsd;
2046}
2047
2048static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2049{
2050 struct sched_domain *sd;
2051
2052 for_each_domain(cpu, sd) {
2053 if (sd->flags & flag)
2054 break;
2055 }
2056
2057 return sd;
2058}
2059
2060DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2061DECLARE_PER_CPU(int, sd_llc_size);
2062DECLARE_PER_CPU(int, sd_llc_id);
2063DECLARE_PER_CPU(int, sd_share_id);
2064DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2065DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2066DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2067DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2068
2069extern struct static_key_false sched_asym_cpucapacity;
2070extern struct static_key_false sched_cluster_active;
2071
2072static __always_inline bool sched_asym_cpucap_active(void)
2073{
2074 return static_branch_unlikely(&sched_asym_cpucapacity);
2075}
2076
2077struct sched_group_capacity {
2078 atomic_t ref;
2079 /*
2080 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2081 * for a single CPU.
2082 */
2083 unsigned long capacity;
2084 unsigned long min_capacity; /* Min per-CPU capacity in group */
2085 unsigned long max_capacity; /* Max per-CPU capacity in group */
2086 unsigned long next_update;
2087 int imbalance; /* XXX unrelated to capacity but shared group state */
2088
2089 int id;
2090
2091 unsigned long cpumask[]; /* Balance mask */
2092};
2093
2094struct sched_group {
2095 struct sched_group *next; /* Must be a circular list */
2096 atomic_t ref;
2097
2098 unsigned int group_weight;
2099 unsigned int cores;
2100 struct sched_group_capacity *sgc;
2101 int asym_prefer_cpu; /* CPU of highest priority in group */
2102 int flags;
2103
2104 /*
2105 * The CPUs this group covers.
2106 *
2107 * NOTE: this field is variable length. (Allocated dynamically
2108 * by attaching extra space to the end of the structure,
2109 * depending on how many CPUs the kernel has booted up with)
2110 */
2111 unsigned long cpumask[];
2112};
2113
2114static inline struct cpumask *sched_group_span(struct sched_group *sg)
2115{
2116 return to_cpumask(sg->cpumask);
2117}
2118
2119/*
2120 * See build_balance_mask().
2121 */
2122static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2123{
2124 return to_cpumask(sg->sgc->cpumask);
2125}
2126
2127extern int group_balance_cpu(struct sched_group *sg);
2128
2129extern void update_sched_domain_debugfs(void);
2130extern void dirty_sched_domain_sysctl(int cpu);
2131
2132extern int sched_update_scaling(void);
2133
2134static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2135{
2136 if (!p->user_cpus_ptr)
2137 return cpu_possible_mask; /* &init_task.cpus_mask */
2138 return p->user_cpus_ptr;
2139}
2140
2141#ifdef CONFIG_CGROUP_SCHED
2142
2143/*
2144 * Return the group to which this tasks belongs.
2145 *
2146 * We cannot use task_css() and friends because the cgroup subsystem
2147 * changes that value before the cgroup_subsys::attach() method is called,
2148 * therefore we cannot pin it and might observe the wrong value.
2149 *
2150 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2151 * core changes this before calling sched_move_task().
2152 *
2153 * Instead we use a 'copy' which is updated from sched_move_task() while
2154 * holding both task_struct::pi_lock and rq::lock.
2155 */
2156static inline struct task_group *task_group(struct task_struct *p)
2157{
2158 return p->sched_task_group;
2159}
2160
2161/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2162static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2163{
2164#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2165 struct task_group *tg = task_group(p);
2166#endif
2167
2168#ifdef CONFIG_FAIR_GROUP_SCHED
2169 set_task_rq_fair(se: &p->se, prev: p->se.cfs_rq, next: tg->cfs_rq[cpu]);
2170 p->se.cfs_rq = tg->cfs_rq[cpu];
2171 p->se.parent = tg->se[cpu];
2172 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2173#endif
2174
2175#ifdef CONFIG_RT_GROUP_SCHED
2176 /*
2177 * p->rt.rt_rq is NULL initially and it is easier to assign
2178 * root_task_group's rt_rq than switching in rt_rq_of_se()
2179 * Clobbers tg(!)
2180 */
2181 if (!rt_group_sched_enabled())
2182 tg = &root_task_group;
2183 p->rt.rt_rq = tg->rt_rq[cpu];
2184 p->rt.parent = tg->rt_se[cpu];
2185#endif /* CONFIG_RT_GROUP_SCHED */
2186}
2187
2188#else /* !CONFIG_CGROUP_SCHED: */
2189
2190static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2191
2192static inline struct task_group *task_group(struct task_struct *p)
2193{
2194 return NULL;
2195}
2196
2197#endif /* !CONFIG_CGROUP_SCHED */
2198
2199static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2200{
2201 set_task_rq(p, cpu);
2202#ifdef CONFIG_SMP
2203 /*
2204 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2205 * successfully executed on another CPU. We must ensure that updates of
2206 * per-task data have been completed by this moment.
2207 */
2208 smp_wmb();
2209 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2210 p->wake_cpu = cpu;
2211#endif /* CONFIG_SMP */
2212}
2213
2214/*
2215 * Tunables:
2216 */
2217
2218#define SCHED_FEAT(name, enabled) \
2219 __SCHED_FEAT_##name ,
2220
2221enum {
2222#include "features.h"
2223 __SCHED_FEAT_NR,
2224};
2225
2226#undef SCHED_FEAT
2227
2228/*
2229 * To support run-time toggling of sched features, all the translation units
2230 * (but core.c) reference the sysctl_sched_features defined in core.c.
2231 */
2232extern __read_mostly unsigned int sysctl_sched_features;
2233
2234#ifdef CONFIG_JUMP_LABEL
2235
2236#define SCHED_FEAT(name, enabled) \
2237static __always_inline bool static_branch_##name(struct static_key *key) \
2238{ \
2239 return static_key_##enabled(key); \
2240}
2241
2242#include "features.h"
2243#undef SCHED_FEAT
2244
2245extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2246#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2247
2248#else /* !CONFIG_JUMP_LABEL: */
2249
2250#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2251
2252#endif /* !CONFIG_JUMP_LABEL */
2253
2254extern struct static_key_false sched_numa_balancing;
2255extern struct static_key_false sched_schedstats;
2256
2257static inline u64 global_rt_period(void)
2258{
2259 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2260}
2261
2262static inline u64 global_rt_runtime(void)
2263{
2264 if (sysctl_sched_rt_runtime < 0)
2265 return RUNTIME_INF;
2266
2267 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2268}
2269
2270/*
2271 * Is p the current execution context?
2272 */
2273static inline int task_current(struct rq *rq, struct task_struct *p)
2274{
2275 return rq->curr == p;
2276}
2277
2278/*
2279 * Is p the current scheduling context?
2280 *
2281 * Note that it might be the current execution context at the same time if
2282 * rq->curr == rq->donor == p.
2283 */
2284static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2285{
2286 return rq->donor == p;
2287}
2288
2289static inline bool task_is_blocked(struct task_struct *p)
2290{
2291 if (!sched_proxy_exec())
2292 return false;
2293
2294 return !!p->blocked_on;
2295}
2296
2297static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2298{
2299 return p->on_cpu;
2300}
2301
2302static inline int task_on_rq_queued(struct task_struct *p)
2303{
2304 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED;
2305}
2306
2307static inline int task_on_rq_migrating(struct task_struct *p)
2308{
2309 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2310}
2311
2312/* Wake flags. The first three directly map to some SD flag value */
2313#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2314#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2315#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2316
2317#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2318#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2319#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2320#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2321
2322static_assert(WF_EXEC == SD_BALANCE_EXEC);
2323static_assert(WF_FORK == SD_BALANCE_FORK);
2324static_assert(WF_TTWU == SD_BALANCE_WAKE);
2325
2326/*
2327 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2328 * of tasks with abnormal "nice" values across CPUs the contribution that
2329 * each task makes to its run queue's load is weighted according to its
2330 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2331 * scaled version of the new time slice allocation that they receive on time
2332 * slice expiry etc.
2333 */
2334
2335#define WEIGHT_IDLEPRIO 3
2336#define WMULT_IDLEPRIO 1431655765
2337
2338extern const int sched_prio_to_weight[40];
2339extern const u32 sched_prio_to_wmult[40];
2340
2341/*
2342 * {de,en}queue flags:
2343 *
2344 * DEQUEUE_SLEEP - task is no longer runnable
2345 * ENQUEUE_WAKEUP - task just became runnable
2346 *
2347 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2348 * are in a known state which allows modification. Such pairs
2349 * should preserve as much state as possible.
2350 *
2351 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2352 * in the runqueue.
2353 *
2354 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2355 *
2356 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2357 *
2358 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2359 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2360 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2361 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2362 *
2363 */
2364
2365#define DEQUEUE_SLEEP 0x01 /* Matches ENQUEUE_WAKEUP */
2366#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2367#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2368#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2369#define DEQUEUE_SPECIAL 0x10
2370#define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2371#define DEQUEUE_DELAYED 0x200 /* Matches ENQUEUE_DELAYED */
2372#define DEQUEUE_THROTTLE 0x800
2373
2374#define ENQUEUE_WAKEUP 0x01
2375#define ENQUEUE_RESTORE 0x02
2376#define ENQUEUE_MOVE 0x04
2377#define ENQUEUE_NOCLOCK 0x08
2378
2379#define ENQUEUE_HEAD 0x10
2380#define ENQUEUE_REPLENISH 0x20
2381#define ENQUEUE_MIGRATED 0x40
2382#define ENQUEUE_INITIAL 0x80
2383#define ENQUEUE_MIGRATING 0x100
2384#define ENQUEUE_DELAYED 0x200
2385#define ENQUEUE_RQ_SELECTED 0x400
2386
2387#define RETRY_TASK ((void *)-1UL)
2388
2389struct affinity_context {
2390 const struct cpumask *new_mask;
2391 struct cpumask *user_mask;
2392 unsigned int flags;
2393};
2394
2395extern s64 update_curr_common(struct rq *rq);
2396
2397struct sched_class {
2398
2399#ifdef CONFIG_UCLAMP_TASK
2400 int uclamp_enabled;
2401#endif
2402
2403 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2404 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2405 void (*yield_task) (struct rq *rq);
2406 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2407
2408 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2409
2410 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2411 struct task_struct *(*pick_task)(struct rq *rq);
2412 /*
2413 * Optional! When implemented pick_next_task() should be equivalent to:
2414 *
2415 * next = pick_task();
2416 * if (next) {
2417 * put_prev_task(prev);
2418 * set_next_task_first(next);
2419 * }
2420 */
2421 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev);
2422
2423 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2424 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2425
2426 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2427
2428 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2429
2430 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2431
2432 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2433
2434 void (*rq_online)(struct rq *rq);
2435 void (*rq_offline)(struct rq *rq);
2436
2437 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2438
2439 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2440 void (*task_fork)(struct task_struct *p);
2441 void (*task_dead)(struct task_struct *p);
2442
2443 /*
2444 * The switched_from() call is allowed to drop rq->lock, therefore we
2445 * cannot assume the switched_from/switched_to pair is serialized by
2446 * rq->lock. They are however serialized by p->pi_lock.
2447 */
2448 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2449 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2450 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2451 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2452 const struct load_weight *lw);
2453 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2454 int oldprio);
2455
2456 unsigned int (*get_rr_interval)(struct rq *rq,
2457 struct task_struct *task);
2458
2459 void (*update_curr)(struct rq *rq);
2460
2461#ifdef CONFIG_FAIR_GROUP_SCHED
2462 void (*task_change_group)(struct task_struct *p);
2463#endif
2464
2465#ifdef CONFIG_SCHED_CORE
2466 int (*task_is_throttled)(struct task_struct *p, int cpu);
2467#endif
2468};
2469
2470static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2471{
2472 WARN_ON_ONCE(rq->donor != prev);
2473 prev->sched_class->put_prev_task(rq, prev, NULL);
2474}
2475
2476static inline void set_next_task(struct rq *rq, struct task_struct *next)
2477{
2478 next->sched_class->set_next_task(rq, next, false);
2479}
2480
2481static inline void
2482__put_prev_set_next_dl_server(struct rq *rq,
2483 struct task_struct *prev,
2484 struct task_struct *next)
2485{
2486 prev->dl_server = NULL;
2487 next->dl_server = rq->dl_server;
2488 rq->dl_server = NULL;
2489}
2490
2491static inline void put_prev_set_next_task(struct rq *rq,
2492 struct task_struct *prev,
2493 struct task_struct *next)
2494{
2495 WARN_ON_ONCE(rq->donor != prev);
2496
2497 __put_prev_set_next_dl_server(rq, prev, next);
2498
2499 if (next == prev)
2500 return;
2501
2502 prev->sched_class->put_prev_task(rq, prev, next);
2503 next->sched_class->set_next_task(rq, next, true);
2504}
2505
2506/*
2507 * Helper to define a sched_class instance; each one is placed in a separate
2508 * section which is ordered by the linker script:
2509 *
2510 * include/asm-generic/vmlinux.lds.h
2511 *
2512 * *CAREFUL* they are laid out in *REVERSE* order!!!
2513 *
2514 * Also enforce alignment on the instance, not the type, to guarantee layout.
2515 */
2516#define DEFINE_SCHED_CLASS(name) \
2517const struct sched_class name##_sched_class \
2518 __aligned(__alignof__(struct sched_class)) \
2519 __section("__" #name "_sched_class")
2520
2521/* Defined in include/asm-generic/vmlinux.lds.h */
2522extern struct sched_class __sched_class_highest[];
2523extern struct sched_class __sched_class_lowest[];
2524
2525extern const struct sched_class stop_sched_class;
2526extern const struct sched_class dl_sched_class;
2527extern const struct sched_class rt_sched_class;
2528extern const struct sched_class fair_sched_class;
2529extern const struct sched_class idle_sched_class;
2530
2531/*
2532 * Iterate only active classes. SCX can take over all fair tasks or be
2533 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2534 */
2535static inline const struct sched_class *next_active_class(const struct sched_class *class)
2536{
2537 class++;
2538#ifdef CONFIG_SCHED_CLASS_EXT
2539 if (scx_switched_all() && class == &fair_sched_class)
2540 class++;
2541 if (!scx_enabled() && class == &ext_sched_class)
2542 class++;
2543#endif
2544 return class;
2545}
2546
2547#define for_class_range(class, _from, _to) \
2548 for (class = (_from); class < (_to); class++)
2549
2550#define for_each_class(class) \
2551 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2552
2553#define for_active_class_range(class, _from, _to) \
2554 for (class = (_from); class != (_to); class = next_active_class(class))
2555
2556#define for_each_active_class(class) \
2557 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2558
2559#define sched_class_above(_a, _b) ((_a) < (_b))
2560
2561static inline bool sched_stop_runnable(struct rq *rq)
2562{
2563 return rq->stop && task_on_rq_queued(p: rq->stop);
2564}
2565
2566static inline bool sched_dl_runnable(struct rq *rq)
2567{
2568 return rq->dl.dl_nr_running > 0;
2569}
2570
2571static inline bool sched_rt_runnable(struct rq *rq)
2572{
2573 return rq->rt.rt_queued > 0;
2574}
2575
2576static inline bool sched_fair_runnable(struct rq *rq)
2577{
2578 return rq->cfs.nr_queued > 0;
2579}
2580
2581extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2582extern struct task_struct *pick_task_idle(struct rq *rq);
2583
2584#define SCA_CHECK 0x01
2585#define SCA_MIGRATE_DISABLE 0x02
2586#define SCA_MIGRATE_ENABLE 0x04
2587#define SCA_USER 0x08
2588
2589extern void update_group_capacity(struct sched_domain *sd, int cpu);
2590
2591extern void sched_balance_trigger(struct rq *rq);
2592
2593extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2594extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2595
2596static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2597{
2598 /* When not in the task's cpumask, no point in looking further. */
2599 if (!cpumask_test_cpu(cpu, cpumask: p->cpus_ptr))
2600 return false;
2601
2602 /* Can @cpu run a user thread? */
2603 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2604 return false;
2605
2606 return true;
2607}
2608
2609static inline cpumask_t *alloc_user_cpus_ptr(int node)
2610{
2611 /*
2612 * See do_set_cpus_allowed() above for the rcu_head usage.
2613 */
2614 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2615
2616 return kmalloc_node(size, GFP_KERNEL, node);
2617}
2618
2619static inline struct task_struct *get_push_task(struct rq *rq)
2620{
2621 struct task_struct *p = rq->donor;
2622
2623 lockdep_assert_rq_held(rq);
2624
2625 if (rq->push_busy)
2626 return NULL;
2627
2628 if (p->nr_cpus_allowed == 1)
2629 return NULL;
2630
2631 if (p->migration_disabled)
2632 return NULL;
2633
2634 rq->push_busy = true;
2635 return get_task_struct(t: p);
2636}
2637
2638extern int push_cpu_stop(void *arg);
2639
2640#ifdef CONFIG_CPU_IDLE
2641
2642static inline void idle_set_state(struct rq *rq,
2643 struct cpuidle_state *idle_state)
2644{
2645 rq->idle_state = idle_state;
2646}
2647
2648static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2649{
2650 WARN_ON_ONCE(!rcu_read_lock_held());
2651
2652 return rq->idle_state;
2653}
2654
2655#else /* !CONFIG_CPU_IDLE: */
2656
2657static inline void idle_set_state(struct rq *rq,
2658 struct cpuidle_state *idle_state)
2659{
2660}
2661
2662static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2663{
2664 return NULL;
2665}
2666
2667#endif /* !CONFIG_CPU_IDLE */
2668
2669extern void schedule_idle(void);
2670asmlinkage void schedule_user(void);
2671
2672extern void sysrq_sched_debug_show(void);
2673extern void sched_init_granularity(void);
2674extern void update_max_interval(void);
2675
2676extern void init_sched_dl_class(void);
2677extern void init_sched_rt_class(void);
2678extern void init_sched_fair_class(void);
2679
2680extern void resched_curr(struct rq *rq);
2681extern void resched_curr_lazy(struct rq *rq);
2682extern void resched_cpu(int cpu);
2683
2684extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2685extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2686
2687extern void init_dl_entity(struct sched_dl_entity *dl_se);
2688
2689extern void init_cfs_throttle_work(struct task_struct *p);
2690
2691#define BW_SHIFT 20
2692#define BW_UNIT (1 << BW_SHIFT)
2693#define RATIO_SHIFT 8
2694#define MAX_BW_BITS (64 - BW_SHIFT)
2695#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2696
2697extern unsigned long to_ratio(u64 period, u64 runtime);
2698
2699extern void init_entity_runnable_average(struct sched_entity *se);
2700extern void post_init_entity_util_avg(struct task_struct *p);
2701
2702#ifdef CONFIG_NO_HZ_FULL
2703extern bool sched_can_stop_tick(struct rq *rq);
2704extern int __init sched_tick_offload_init(void);
2705
2706/*
2707 * Tick may be needed by tasks in the runqueue depending on their policy and
2708 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2709 * nohz mode if necessary.
2710 */
2711static inline void sched_update_tick_dependency(struct rq *rq)
2712{
2713 int cpu = cpu_of(rq);
2714
2715 if (!tick_nohz_full_cpu(cpu))
2716 return;
2717
2718 if (sched_can_stop_tick(rq))
2719 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2720 else
2721 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2722}
2723#else /* !CONFIG_NO_HZ_FULL: */
2724static inline int sched_tick_offload_init(void) { return 0; }
2725static inline void sched_update_tick_dependency(struct rq *rq) { }
2726#endif /* !CONFIG_NO_HZ_FULL */
2727
2728static inline void add_nr_running(struct rq *rq, unsigned count)
2729{
2730 unsigned prev_nr = rq->nr_running;
2731
2732 rq->nr_running = prev_nr + count;
2733 if (trace_sched_update_nr_running_tp_enabled()) {
2734 call_trace_sched_update_nr_running(rq, count);
2735 }
2736
2737 if (prev_nr < 2 && rq->nr_running >= 2)
2738 set_rd_overloaded(rd: rq->rd, status: 1);
2739
2740 sched_update_tick_dependency(rq);
2741}
2742
2743static inline void sub_nr_running(struct rq *rq, unsigned count)
2744{
2745 rq->nr_running -= count;
2746 if (trace_sched_update_nr_running_tp_enabled()) {
2747 call_trace_sched_update_nr_running(rq, count: -count);
2748 }
2749
2750 /* Check if we still need preemption */
2751 sched_update_tick_dependency(rq);
2752}
2753
2754static inline void __block_task(struct rq *rq, struct task_struct *p)
2755{
2756 if (p->sched_contributes_to_load)
2757 rq->nr_uninterruptible++;
2758
2759 if (p->in_iowait) {
2760 atomic_inc(v: &rq->nr_iowait);
2761 delayacct_blkio_start();
2762 }
2763
2764 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2765
2766 /*
2767 * The moment this write goes through, ttwu() can swoop in and migrate
2768 * this task, rendering our rq->__lock ineffective.
2769 *
2770 * __schedule() try_to_wake_up()
2771 * LOCK rq->__lock LOCK p->pi_lock
2772 * pick_next_task()
2773 * pick_next_task_fair()
2774 * pick_next_entity()
2775 * dequeue_entities()
2776 * __block_task()
2777 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2778 * break;
2779 *
2780 * ACQUIRE (after ctrl-dep)
2781 *
2782 * cpu = select_task_rq();
2783 * set_task_cpu(p, cpu);
2784 * ttwu_queue()
2785 * ttwu_do_activate()
2786 * LOCK rq->__lock
2787 * activate_task()
2788 * STORE p->on_rq = 1
2789 * UNLOCK rq->__lock
2790 *
2791 * Callers must ensure to not reference @p after this -- we no longer
2792 * own it.
2793 */
2794 smp_store_release(&p->on_rq, 0);
2795}
2796
2797extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2798extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2799
2800extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2801
2802#ifdef CONFIG_PREEMPT_RT
2803# define SCHED_NR_MIGRATE_BREAK 8
2804#else
2805# define SCHED_NR_MIGRATE_BREAK 32
2806#endif
2807
2808extern __read_mostly unsigned int sysctl_sched_nr_migrate;
2809extern __read_mostly unsigned int sysctl_sched_migration_cost;
2810
2811extern unsigned int sysctl_sched_base_slice;
2812
2813extern int sysctl_resched_latency_warn_ms;
2814extern int sysctl_resched_latency_warn_once;
2815
2816extern unsigned int sysctl_sched_tunable_scaling;
2817
2818extern unsigned int sysctl_numa_balancing_scan_delay;
2819extern unsigned int sysctl_numa_balancing_scan_period_min;
2820extern unsigned int sysctl_numa_balancing_scan_period_max;
2821extern unsigned int sysctl_numa_balancing_scan_size;
2822extern unsigned int sysctl_numa_balancing_hot_threshold;
2823
2824#ifdef CONFIG_SCHED_HRTICK
2825
2826/*
2827 * Use hrtick when:
2828 * - enabled by features
2829 * - hrtimer is actually high res
2830 */
2831static inline int hrtick_enabled(struct rq *rq)
2832{
2833 if (!cpu_active(cpu: cpu_of(rq)))
2834 return 0;
2835 return hrtimer_is_hres_active(timer: &rq->hrtick_timer);
2836}
2837
2838static inline int hrtick_enabled_fair(struct rq *rq)
2839{
2840 if (!sched_feat(HRTICK))
2841 return 0;
2842 return hrtick_enabled(rq);
2843}
2844
2845static inline int hrtick_enabled_dl(struct rq *rq)
2846{
2847 if (!sched_feat(HRTICK_DL))
2848 return 0;
2849 return hrtick_enabled(rq);
2850}
2851
2852extern void hrtick_start(struct rq *rq, u64 delay);
2853
2854#else /* !CONFIG_SCHED_HRTICK: */
2855
2856static inline int hrtick_enabled_fair(struct rq *rq)
2857{
2858 return 0;
2859}
2860
2861static inline int hrtick_enabled_dl(struct rq *rq)
2862{
2863 return 0;
2864}
2865
2866static inline int hrtick_enabled(struct rq *rq)
2867{
2868 return 0;
2869}
2870
2871#endif /* !CONFIG_SCHED_HRTICK */
2872
2873#ifndef arch_scale_freq_tick
2874static __always_inline void arch_scale_freq_tick(void) { }
2875#endif
2876
2877#ifndef arch_scale_freq_capacity
2878/**
2879 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2880 * @cpu: the CPU in question.
2881 *
2882 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2883 *
2884 * f_curr
2885 * ------ * SCHED_CAPACITY_SCALE
2886 * f_max
2887 */
2888static __always_inline
2889unsigned long arch_scale_freq_capacity(int cpu)
2890{
2891 return SCHED_CAPACITY_SCALE;
2892}
2893#endif
2894
2895/*
2896 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2897 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2898 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2899 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2900 */
2901static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2902{
2903 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2904 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2905}
2906
2907#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2908__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2909static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2910{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2911 _lock; return _t; }
2912
2913static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2914{
2915#ifdef CONFIG_SCHED_CORE
2916 /*
2917 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2918 * order by core-id first and cpu-id second.
2919 *
2920 * Notably:
2921 *
2922 * double_rq_lock(0,3); will take core-0, core-1 lock
2923 * double_rq_lock(1,2); will take core-1, core-0 lock
2924 *
2925 * when only cpu-id is considered.
2926 */
2927 if (rq1->core->cpu < rq2->core->cpu)
2928 return true;
2929 if (rq1->core->cpu > rq2->core->cpu)
2930 return false;
2931
2932 /*
2933 * __sched_core_flip() relies on SMT having cpu-id lock order.
2934 */
2935#endif /* CONFIG_SCHED_CORE */
2936 return rq1->cpu < rq2->cpu;
2937}
2938
2939extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2940
2941#ifdef CONFIG_PREEMPTION
2942
2943/*
2944 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2945 * way at the expense of forcing extra atomic operations in all
2946 * invocations. This assures that the double_lock is acquired using the
2947 * same underlying policy as the spinlock_t on this architecture, which
2948 * reduces latency compared to the unfair variant below. However, it
2949 * also adds more overhead and therefore may reduce throughput.
2950 */
2951static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2952 __releases(this_rq->lock)
2953 __acquires(busiest->lock)
2954 __acquires(this_rq->lock)
2955{
2956 raw_spin_rq_unlock(rq: this_rq);
2957 double_rq_lock(rq1: this_rq, rq2: busiest);
2958
2959 return 1;
2960}
2961
2962#else /* !CONFIG_PREEMPTION: */
2963/*
2964 * Unfair double_lock_balance: Optimizes throughput at the expense of
2965 * latency by eliminating extra atomic operations when the locks are
2966 * already in proper order on entry. This favors lower CPU-ids and will
2967 * grant the double lock to lower CPUs over higher ids under contention,
2968 * regardless of entry order into the function.
2969 */
2970static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2971 __releases(this_rq->lock)
2972 __acquires(busiest->lock)
2973 __acquires(this_rq->lock)
2974{
2975 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2976 likely(raw_spin_rq_trylock(busiest))) {
2977 double_rq_clock_clear_update(this_rq, busiest);
2978 return 0;
2979 }
2980
2981 if (rq_order_less(this_rq, busiest)) {
2982 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2983 double_rq_clock_clear_update(this_rq, busiest);
2984 return 0;
2985 }
2986
2987 raw_spin_rq_unlock(this_rq);
2988 double_rq_lock(this_rq, busiest);
2989
2990 return 1;
2991}
2992
2993#endif /* !CONFIG_PREEMPTION */
2994
2995/*
2996 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2997 */
2998static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2999{
3000 lockdep_assert_irqs_disabled();
3001
3002 return _double_lock_balance(this_rq, busiest);
3003}
3004
3005static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3006 __releases(busiest->lock)
3007{
3008 if (__rq_lockp(rq: this_rq) != __rq_lockp(rq: busiest))
3009 raw_spin_rq_unlock(rq: busiest);
3010 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3011}
3012
3013static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3014{
3015 if (l1 > l2)
3016 swap(l1, l2);
3017
3018 spin_lock(lock: l1);
3019 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3020}
3021
3022static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3023{
3024 if (l1 > l2)
3025 swap(l1, l2);
3026
3027 spin_lock_irq(lock: l1);
3028 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3029}
3030
3031static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3032{
3033 if (l1 > l2)
3034 swap(l1, l2);
3035
3036 raw_spin_lock(l1);
3037 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3038}
3039
3040static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3041{
3042 raw_spin_unlock(l1);
3043 raw_spin_unlock(l2);
3044}
3045
3046DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3047 double_raw_lock(_T->lock, _T->lock2),
3048 double_raw_unlock(_T->lock, _T->lock2))
3049
3050/*
3051 * double_rq_unlock - safely unlock two runqueues
3052 *
3053 * Note this does not restore interrupts like task_rq_unlock,
3054 * you need to do so manually after calling.
3055 */
3056static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3057 __releases(rq1->lock)
3058 __releases(rq2->lock)
3059{
3060 if (__rq_lockp(rq: rq1) != __rq_lockp(rq: rq2))
3061 raw_spin_rq_unlock(rq: rq2);
3062 else
3063 __release(rq2->lock);
3064 raw_spin_rq_unlock(rq: rq1);
3065}
3066
3067extern void set_rq_online (struct rq *rq);
3068extern void set_rq_offline(struct rq *rq);
3069
3070extern bool sched_smp_initialized;
3071
3072DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3073 double_rq_lock(_T->lock, _T->lock2),
3074 double_rq_unlock(_T->lock, _T->lock2))
3075
3076extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3077extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3078extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3079
3080extern bool sched_debug_verbose;
3081
3082extern void print_cfs_stats(struct seq_file *m, int cpu);
3083extern void print_rt_stats(struct seq_file *m, int cpu);
3084extern void print_dl_stats(struct seq_file *m, int cpu);
3085extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3086extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3087extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3088
3089extern void resched_latency_warn(int cpu, u64 latency);
3090
3091#ifdef CONFIG_NUMA_BALANCING
3092extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3093extern void
3094print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3095 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3096#endif /* CONFIG_NUMA_BALANCING */
3097
3098extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3099extern void init_rt_rq(struct rt_rq *rt_rq);
3100extern void init_dl_rq(struct dl_rq *dl_rq);
3101
3102extern void cfs_bandwidth_usage_inc(void);
3103extern void cfs_bandwidth_usage_dec(void);
3104
3105#ifdef CONFIG_NO_HZ_COMMON
3106
3107#define NOHZ_BALANCE_KICK_BIT 0
3108#define NOHZ_STATS_KICK_BIT 1
3109#define NOHZ_NEWILB_KICK_BIT 2
3110#define NOHZ_NEXT_KICK_BIT 3
3111
3112/* Run sched_balance_domains() */
3113#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3114/* Update blocked load */
3115#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3116/* Update blocked load when entering idle */
3117#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3118/* Update nohz.next_balance */
3119#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3120
3121#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3122
3123#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3124
3125extern void nohz_balance_exit_idle(struct rq *rq);
3126#else /* !CONFIG_NO_HZ_COMMON: */
3127static inline void nohz_balance_exit_idle(struct rq *rq) { }
3128#endif /* !CONFIG_NO_HZ_COMMON */
3129
3130#ifdef CONFIG_NO_HZ_COMMON
3131extern void nohz_run_idle_balance(int cpu);
3132#else
3133static inline void nohz_run_idle_balance(int cpu) { }
3134#endif
3135
3136#include "stats.h"
3137
3138#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3139
3140extern void __sched_core_account_forceidle(struct rq *rq);
3141
3142static inline void sched_core_account_forceidle(struct rq *rq)
3143{
3144 if (schedstat_enabled())
3145 __sched_core_account_forceidle(rq);
3146}
3147
3148extern void __sched_core_tick(struct rq *rq);
3149
3150static inline void sched_core_tick(struct rq *rq)
3151{
3152 if (sched_core_enabled(rq) && schedstat_enabled())
3153 __sched_core_tick(rq);
3154}
3155
3156#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3157
3158static inline void sched_core_account_forceidle(struct rq *rq) { }
3159
3160static inline void sched_core_tick(struct rq *rq) { }
3161
3162#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3163
3164#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3165
3166struct irqtime {
3167 u64 total;
3168 u64 tick_delta;
3169 u64 irq_start_time;
3170 struct u64_stats_sync sync;
3171};
3172
3173DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3174extern int sched_clock_irqtime;
3175
3176static inline int irqtime_enabled(void)
3177{
3178 return sched_clock_irqtime;
3179}
3180
3181/*
3182 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3183 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3184 * and never move forward.
3185 */
3186static inline u64 irq_time_read(int cpu)
3187{
3188 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3189 unsigned int seq;
3190 u64 total;
3191
3192 do {
3193 seq = __u64_stats_fetch_begin(&irqtime->sync);
3194 total = irqtime->total;
3195 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3196
3197 return total;
3198}
3199
3200#else /* !CONFIG_IRQ_TIME_ACCOUNTING: */
3201
3202static inline int irqtime_enabled(void)
3203{
3204 return 0;
3205}
3206
3207#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
3208
3209#ifdef CONFIG_CPU_FREQ
3210
3211DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3212
3213/**
3214 * cpufreq_update_util - Take a note about CPU utilization changes.
3215 * @rq: Runqueue to carry out the update for.
3216 * @flags: Update reason flags.
3217 *
3218 * This function is called by the scheduler on the CPU whose utilization is
3219 * being updated.
3220 *
3221 * It can only be called from RCU-sched read-side critical sections.
3222 *
3223 * The way cpufreq is currently arranged requires it to evaluate the CPU
3224 * performance state (frequency/voltage) on a regular basis to prevent it from
3225 * being stuck in a completely inadequate performance level for too long.
3226 * That is not guaranteed to happen if the updates are only triggered from CFS
3227 * and DL, though, because they may not be coming in if only RT tasks are
3228 * active all the time (or there are RT tasks only).
3229 *
3230 * As a workaround for that issue, this function is called periodically by the
3231 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3232 * but that really is a band-aid. Going forward it should be replaced with
3233 * solutions targeted more specifically at RT tasks.
3234 */
3235static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3236{
3237 struct update_util_data *data;
3238
3239 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3240 cpu_of(rq)));
3241 if (data)
3242 data->func(data, rq_clock(rq), flags);
3243}
3244#else /* !CONFIG_CPU_FREQ: */
3245static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3246#endif /* !CONFIG_CPU_FREQ */
3247
3248#ifdef arch_scale_freq_capacity
3249# ifndef arch_scale_freq_invariant
3250# define arch_scale_freq_invariant() true
3251# endif
3252#else
3253# define arch_scale_freq_invariant() false
3254#endif
3255
3256unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3257 unsigned long *min,
3258 unsigned long *max);
3259
3260unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3261 unsigned long min,
3262 unsigned long max);
3263
3264
3265/*
3266 * Verify the fitness of task @p to run on @cpu taking into account the
3267 * CPU original capacity and the runtime/deadline ratio of the task.
3268 *
3269 * The function will return true if the original capacity of @cpu is
3270 * greater than or equal to task's deadline density right shifted by
3271 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3272 */
3273static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3274{
3275 unsigned long cap = arch_scale_cpu_capacity(cpu);
3276
3277 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3278}
3279
3280static inline unsigned long cpu_bw_dl(struct rq *rq)
3281{
3282 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3283}
3284
3285static inline unsigned long cpu_util_dl(struct rq *rq)
3286{
3287 return READ_ONCE(rq->avg_dl.util_avg);
3288}
3289
3290
3291extern unsigned long cpu_util_cfs(int cpu);
3292extern unsigned long cpu_util_cfs_boost(int cpu);
3293
3294static inline unsigned long cpu_util_rt(struct rq *rq)
3295{
3296 return READ_ONCE(rq->avg_rt.util_avg);
3297}
3298
3299#ifdef CONFIG_UCLAMP_TASK
3300
3301unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3302
3303/*
3304 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3305 * by default in the fast path and only gets turned on once userspace performs
3306 * an operation that requires it.
3307 *
3308 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3309 * hence is active.
3310 */
3311static inline bool uclamp_is_used(void)
3312{
3313 return static_branch_likely(&sched_uclamp_used);
3314}
3315
3316/*
3317 * Enabling static branches would get the cpus_read_lock(),
3318 * check whether uclamp_is_used before enable it to avoid always
3319 * calling cpus_read_lock(). Because we never disable this
3320 * static key once enable it.
3321 */
3322static inline void sched_uclamp_enable(void)
3323{
3324 if (!uclamp_is_used())
3325 static_branch_enable(&sched_uclamp_used);
3326}
3327
3328static inline unsigned long uclamp_rq_get(struct rq *rq,
3329 enum uclamp_id clamp_id)
3330{
3331 return READ_ONCE(rq->uclamp[clamp_id].value);
3332}
3333
3334static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3335 unsigned int value)
3336{
3337 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3338}
3339
3340static inline bool uclamp_rq_is_idle(struct rq *rq)
3341{
3342 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3343}
3344
3345/* Is the rq being capped/throttled by uclamp_max? */
3346static inline bool uclamp_rq_is_capped(struct rq *rq)
3347{
3348 unsigned long rq_util;
3349 unsigned long max_util;
3350
3351 if (!uclamp_is_used())
3352 return false;
3353
3354 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3355 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3356
3357 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3358}
3359
3360#define for_each_clamp_id(clamp_id) \
3361 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3362
3363extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3364
3365
3366static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3367{
3368 if (clamp_id == UCLAMP_MIN)
3369 return 0;
3370 return SCHED_CAPACITY_SCALE;
3371}
3372
3373/* Integer rounded range for each bucket */
3374#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3375
3376static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3377{
3378 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3379}
3380
3381static inline void
3382uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3383{
3384 uc_se->value = value;
3385 uc_se->bucket_id = uclamp_bucket_id(value);
3386 uc_se->user_defined = user_defined;
3387}
3388
3389#else /* !CONFIG_UCLAMP_TASK: */
3390
3391static inline unsigned long
3392uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3393{
3394 if (clamp_id == UCLAMP_MIN)
3395 return 0;
3396
3397 return SCHED_CAPACITY_SCALE;
3398}
3399
3400static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3401
3402static inline bool uclamp_is_used(void)
3403{
3404 return false;
3405}
3406
3407static inline void sched_uclamp_enable(void) {}
3408
3409static inline unsigned long
3410uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3411{
3412 if (clamp_id == UCLAMP_MIN)
3413 return 0;
3414
3415 return SCHED_CAPACITY_SCALE;
3416}
3417
3418static inline void
3419uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3420{
3421}
3422
3423static inline bool uclamp_rq_is_idle(struct rq *rq)
3424{
3425 return false;
3426}
3427
3428#endif /* !CONFIG_UCLAMP_TASK */
3429
3430#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3431
3432static inline unsigned long cpu_util_irq(struct rq *rq)
3433{
3434 return READ_ONCE(rq->avg_irq.util_avg);
3435}
3436
3437static inline
3438unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3439{
3440 util *= (max - irq);
3441 util /= max;
3442
3443 return util;
3444
3445}
3446
3447#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3448
3449static inline unsigned long cpu_util_irq(struct rq *rq)
3450{
3451 return 0;
3452}
3453
3454static inline
3455unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3456{
3457 return util;
3458}
3459
3460#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3461
3462extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr);
3463
3464#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3465
3466#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3467
3468DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3469
3470static inline bool sched_energy_enabled(void)
3471{
3472 return static_branch_unlikely(&sched_energy_present);
3473}
3474
3475#else /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL): */
3476
3477#define perf_domain_span(pd) NULL
3478
3479static inline bool sched_energy_enabled(void) { return false; }
3480
3481#endif /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3482
3483#ifdef CONFIG_MEMBARRIER
3484
3485/*
3486 * The scheduler provides memory barriers required by membarrier between:
3487 * - prior user-space memory accesses and store to rq->membarrier_state,
3488 * - store to rq->membarrier_state and following user-space memory accesses.
3489 * In the same way it provides those guarantees around store to rq->curr.
3490 */
3491static inline void membarrier_switch_mm(struct rq *rq,
3492 struct mm_struct *prev_mm,
3493 struct mm_struct *next_mm)
3494{
3495 int membarrier_state;
3496
3497 if (prev_mm == next_mm)
3498 return;
3499
3500 membarrier_state = atomic_read(v: &next_mm->membarrier_state);
3501 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3502 return;
3503
3504 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3505}
3506
3507#else /* !CONFIG_MEMBARRIER: */
3508
3509static inline void membarrier_switch_mm(struct rq *rq,
3510 struct mm_struct *prev_mm,
3511 struct mm_struct *next_mm)
3512{
3513}
3514
3515#endif /* !CONFIG_MEMBARRIER */
3516
3517static inline bool is_per_cpu_kthread(struct task_struct *p)
3518{
3519 if (!(p->flags & PF_KTHREAD))
3520 return false;
3521
3522 if (p->nr_cpus_allowed != 1)
3523 return false;
3524
3525 return true;
3526}
3527
3528extern void swake_up_all_locked(struct swait_queue_head *q);
3529extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3530
3531extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3532
3533#ifdef CONFIG_PREEMPT_DYNAMIC
3534extern int preempt_dynamic_mode;
3535extern int sched_dynamic_mode(const char *str);
3536extern void sched_dynamic_update(int mode);
3537#endif
3538extern const char *preempt_modes[];
3539
3540#ifdef CONFIG_SCHED_MM_CID
3541
3542#define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3543#define MM_CID_SCAN_DELAY 100 /* 100ms */
3544
3545extern raw_spinlock_t cid_lock;
3546extern int use_cid_lock;
3547
3548extern void sched_mm_cid_migrate_from(struct task_struct *t);
3549extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3550extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3551extern void init_sched_mm_cid(struct task_struct *t);
3552
3553static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3554{
3555 if (cid < 0)
3556 return;
3557 cpumask_clear_cpu(cpu: cid, dstp: mm_cidmask(mm));
3558}
3559
3560/*
3561 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3562 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3563 * be held to transition to other states.
3564 *
3565 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3566 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3567 */
3568static inline void mm_cid_put_lazy(struct task_struct *t)
3569{
3570 struct mm_struct *mm = t->mm;
3571 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3572 int cid;
3573
3574 lockdep_assert_irqs_disabled();
3575 cid = __this_cpu_read(pcpu_cid->cid);
3576 if (!mm_cid_is_lazy_put(cid) ||
3577 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3578 return;
3579 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3580}
3581
3582static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3583{
3584 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3585 int cid, res;
3586
3587 lockdep_assert_irqs_disabled();
3588 cid = __this_cpu_read(pcpu_cid->cid);
3589 for (;;) {
3590 if (mm_cid_is_unset(cid))
3591 return MM_CID_UNSET;
3592 /*
3593 * Attempt transition from valid or lazy-put to unset.
3594 */
3595 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3596 if (res == cid)
3597 break;
3598 cid = res;
3599 }
3600 return cid;
3601}
3602
3603static inline void mm_cid_put(struct mm_struct *mm)
3604{
3605 int cid;
3606
3607 lockdep_assert_irqs_disabled();
3608 cid = mm_cid_pcpu_unset(mm);
3609 if (cid == MM_CID_UNSET)
3610 return;
3611 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3612}
3613
3614static inline int __mm_cid_try_get(struct task_struct *t, struct mm_struct *mm)
3615{
3616 struct cpumask *cidmask = mm_cidmask(mm);
3617 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3618 int cid, max_nr_cid, allowed_max_nr_cid;
3619
3620 /*
3621 * After shrinking the number of threads or reducing the number
3622 * of allowed cpus, reduce the value of max_nr_cid so expansion
3623 * of cid allocation will preserve cache locality if the number
3624 * of threads or allowed cpus increase again.
3625 */
3626 max_nr_cid = atomic_read(v: &mm->max_nr_cid);
3627 while ((allowed_max_nr_cid = min_t(int, READ_ONCE(mm->nr_cpus_allowed),
3628 atomic_read(&mm->mm_users))),
3629 max_nr_cid > allowed_max_nr_cid) {
3630 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into max_nr_cid. */
3631 if (atomic_try_cmpxchg(v: &mm->max_nr_cid, old: &max_nr_cid, new: allowed_max_nr_cid)) {
3632 max_nr_cid = allowed_max_nr_cid;
3633 break;
3634 }
3635 }
3636 /* Try to re-use recent cid. This improves cache locality. */
3637 cid = __this_cpu_read(pcpu_cid->recent_cid);
3638 if (!mm_cid_is_unset(cid) && cid < max_nr_cid &&
3639 !cpumask_test_and_set_cpu(cpu: cid, cpumask: cidmask))
3640 return cid;
3641 /*
3642 * Expand cid allocation if the maximum number of concurrency
3643 * IDs allocated (max_nr_cid) is below the number cpus allowed
3644 * and number of threads. Expanding cid allocation as much as
3645 * possible improves cache locality.
3646 */
3647 cid = max_nr_cid;
3648 while (cid < READ_ONCE(mm->nr_cpus_allowed) && cid < atomic_read(v: &mm->mm_users)) {
3649 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into cid. */
3650 if (!atomic_try_cmpxchg(v: &mm->max_nr_cid, old: &cid, new: cid + 1))
3651 continue;
3652 if (!cpumask_test_and_set_cpu(cpu: cid, cpumask: cidmask))
3653 return cid;
3654 }
3655 /*
3656 * Find the first available concurrency id.
3657 * Retry finding first zero bit if the mask is temporarily
3658 * filled. This only happens during concurrent remote-clear
3659 * which owns a cid without holding a rq lock.
3660 */
3661 for (;;) {
3662 cid = cpumask_first_zero(srcp: cidmask);
3663 if (cid < READ_ONCE(mm->nr_cpus_allowed))
3664 break;
3665 cpu_relax();
3666 }
3667 if (cpumask_test_and_set_cpu(cpu: cid, cpumask: cidmask))
3668 return -1;
3669
3670 return cid;
3671}
3672
3673/*
3674 * Save a snapshot of the current runqueue time of this cpu
3675 * with the per-cpu cid value, allowing to estimate how recently it was used.
3676 */
3677static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3678{
3679 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3680
3681 lockdep_assert_rq_held(rq);
3682 WRITE_ONCE(pcpu_cid->time, rq->clock);
3683}
3684
3685static inline int __mm_cid_get(struct rq *rq, struct task_struct *t,
3686 struct mm_struct *mm)
3687{
3688 int cid;
3689
3690 /*
3691 * All allocations (even those using the cid_lock) are lock-free. If
3692 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3693 * guarantee forward progress.
3694 */
3695 if (!READ_ONCE(use_cid_lock)) {
3696 cid = __mm_cid_try_get(t, mm);
3697 if (cid >= 0)
3698 goto end;
3699 raw_spin_lock(&cid_lock);
3700 } else {
3701 raw_spin_lock(&cid_lock);
3702 cid = __mm_cid_try_get(t, mm);
3703 if (cid >= 0)
3704 goto unlock;
3705 }
3706
3707 /*
3708 * cid concurrently allocated. Retry while forcing following
3709 * allocations to use the cid_lock to ensure forward progress.
3710 */
3711 WRITE_ONCE(use_cid_lock, 1);
3712 /*
3713 * Set use_cid_lock before allocation. Only care about program order
3714 * because this is only required for forward progress.
3715 */
3716 barrier();
3717 /*
3718 * Retry until it succeeds. It is guaranteed to eventually succeed once
3719 * all newcoming allocations observe the use_cid_lock flag set.
3720 */
3721 do {
3722 cid = __mm_cid_try_get(t, mm);
3723 cpu_relax();
3724 } while (cid < 0);
3725 /*
3726 * Allocate before clearing use_cid_lock. Only care about
3727 * program order because this is for forward progress.
3728 */
3729 barrier();
3730 WRITE_ONCE(use_cid_lock, 0);
3731unlock:
3732 raw_spin_unlock(&cid_lock);
3733end:
3734 mm_cid_snapshot_time(rq, mm);
3735
3736 return cid;
3737}
3738
3739static inline int mm_cid_get(struct rq *rq, struct task_struct *t,
3740 struct mm_struct *mm)
3741{
3742 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3743 struct cpumask *cpumask;
3744 int cid;
3745
3746 lockdep_assert_rq_held(rq);
3747 cpumask = mm_cidmask(mm);
3748 cid = __this_cpu_read(pcpu_cid->cid);
3749 if (mm_cid_is_valid(cid)) {
3750 mm_cid_snapshot_time(rq, mm);
3751 return cid;
3752 }
3753 if (mm_cid_is_lazy_put(cid)) {
3754 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3755 __mm_cid_put(mm, cid: mm_cid_clear_lazy_put(cid));
3756 }
3757 cid = __mm_cid_get(rq, t, mm);
3758 __this_cpu_write(pcpu_cid->cid, cid);
3759 __this_cpu_write(pcpu_cid->recent_cid, cid);
3760
3761 return cid;
3762}
3763
3764static inline void switch_mm_cid(struct rq *rq,
3765 struct task_struct *prev,
3766 struct task_struct *next)
3767{
3768 /*
3769 * Provide a memory barrier between rq->curr store and load of
3770 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3771 *
3772 * Should be adapted if context_switch() is modified.
3773 */
3774 if (!next->mm) { // to kernel
3775 /*
3776 * user -> kernel transition does not guarantee a barrier, but
3777 * we can use the fact that it performs an atomic operation in
3778 * mmgrab().
3779 */
3780 if (prev->mm) // from user
3781 smp_mb__after_mmgrab();
3782 /*
3783 * kernel -> kernel transition does not change rq->curr->mm
3784 * state. It stays NULL.
3785 */
3786 } else { // to user
3787 /*
3788 * kernel -> user transition does not provide a barrier
3789 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3790 * Provide it here.
3791 */
3792 if (!prev->mm) { // from kernel
3793 smp_mb();
3794 } else { // from user
3795 /*
3796 * user->user transition relies on an implicit
3797 * memory barrier in switch_mm() when
3798 * current->mm changes. If the architecture
3799 * switch_mm() does not have an implicit memory
3800 * barrier, it is emitted here. If current->mm
3801 * is unchanged, no barrier is needed.
3802 */
3803 smp_mb__after_switch_mm();
3804 }
3805 }
3806 if (prev->mm_cid_active) {
3807 mm_cid_snapshot_time(rq, mm: prev->mm);
3808 mm_cid_put_lazy(t: prev);
3809 prev->mm_cid = -1;
3810 }
3811 if (next->mm_cid_active)
3812 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, t: next, mm: next->mm);
3813}
3814
3815#else /* !CONFIG_SCHED_MM_CID: */
3816static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3817static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3818static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3819static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3820static inline void init_sched_mm_cid(struct task_struct *t) { }
3821#endif /* !CONFIG_SCHED_MM_CID */
3822
3823extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3824extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3825static inline
3826void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
3827{
3828 lockdep_assert_rq_held(rq: src_rq);
3829 lockdep_assert_rq_held(rq: dst_rq);
3830
3831 deactivate_task(rq: src_rq, p: task, flags: 0);
3832 set_task_cpu(p: task, cpu: dst_rq->cpu);
3833 activate_task(rq: dst_rq, p: task, flags: 0);
3834}
3835
3836static inline
3837bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
3838{
3839 if (!task_on_cpu(rq, p) &&
3840 cpumask_test_cpu(cpu, cpumask: &p->cpus_mask))
3841 return true;
3842
3843 return false;
3844}
3845
3846#ifdef CONFIG_RT_MUTEXES
3847
3848static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3849{
3850 if (pi_task)
3851 prio = min(prio, pi_task->prio);
3852
3853 return prio;
3854}
3855
3856static inline int rt_effective_prio(struct task_struct *p, int prio)
3857{
3858 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3859
3860 return __rt_effective_prio(pi_task, prio);
3861}
3862
3863#else /* !CONFIG_RT_MUTEXES: */
3864
3865static inline int rt_effective_prio(struct task_struct *p, int prio)
3866{
3867 return prio;
3868}
3869
3870#endif /* !CONFIG_RT_MUTEXES */
3871
3872extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3873extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3874extern const struct sched_class *__setscheduler_class(int policy, int prio);
3875extern void set_load_weight(struct task_struct *p, bool update_load);
3876extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3877extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3878
3879extern void check_class_changing(struct rq *rq, struct task_struct *p,
3880 const struct sched_class *prev_class);
3881extern void check_class_changed(struct rq *rq, struct task_struct *p,
3882 const struct sched_class *prev_class,
3883 int oldprio);
3884
3885extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3886extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3887
3888#ifdef CONFIG_SCHED_CLASS_EXT
3889/*
3890 * Used by SCX in the enable/disable paths to move tasks between sched_classes
3891 * and establish invariants.
3892 */
3893struct sched_enq_and_set_ctx {
3894 struct task_struct *p;
3895 int queue_flags;
3896 bool queued;
3897 bool running;
3898};
3899
3900void sched_deq_and_put_task(struct task_struct *p, int queue_flags,
3901 struct sched_enq_and_set_ctx *ctx);
3902void sched_enq_and_set_task(struct sched_enq_and_set_ctx *ctx);
3903
3904#endif /* CONFIG_SCHED_CLASS_EXT */
3905
3906#include "ext.h"
3907
3908#endif /* _KERNEL_SCHED_SCHED_H */
3909