projects/linux·blob master · refs

4311 lines · 117.7 KiB

1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Scheduler internal types and methods:
4 */
5#ifndef _KERNEL_SCHED_SCHED_H
6#define _KERNEL_SCHED_SCHED_H
7
8#include <linux/prandom.h>
9#include <linux/sched/affinity.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/cpufreq.h>
12#include <linux/sched/deadline.h>
13#include <linux/sched.h>
14#include <linux/sched/loadavg.h>
15#include <linux/sched/mm.h>
16#include <linux/sched/rseq_api.h>
17#include <linux/sched/signal.h>
18#include <linux/sched/smt.h>
19#include <linux/sched/stat.h>
20#include <linux/sched/sysctl.h>
21#include <linux/sched/task_flags.h>
22#include <linux/sched/task.h>
23#include <linux/sched/topology.h>
24#include <linux/atomic.h>
25#include <linux/bitmap.h>
26#include <linux/bug.h>
27#include <linux/capability.h>
28#include <linux/cgroup_api.h>
29#include <linux/cgroup.h>
30#include <linux/context_tracking.h>
31#include <linux/cpufreq.h>
32#include <linux/cpumask_api.h>
33#include <linux/cpuset.h>
34#include <linux/ctype.h>
35#include <linux/file.h>
36#include <linux/fs_api.h>
37#include <linux/hrtimer_api.h>
38#include <linux/interrupt.h>
39#include <linux/irq_work.h>
40#include <linux/jiffies.h>
41#include <linux/kref_api.h>
42#include <linux/kthread.h>
43#include <linux/ktime_api.h>
44#include <linux/lockdep_api.h>
45#include <linux/lockdep.h>
46#include <linux/memblock.h>
47#include <linux/memcontrol.h>
48#include <linux/minmax.h>
49#include <linux/mm.h>
50#include <linux/module.h>
51#include <linux/mutex_api.h>
52#include <linux/plist.h>
53#include <linux/poll.h>
54#include <linux/proc_fs.h>
55#include <linux/profile.h>
56#include <linux/psi.h>
57#include <linux/rcupdate.h>
58#include <linux/seq_file.h>
59#include <linux/seqlock.h>
60#include <linux/softirq.h>
61#include <linux/spinlock_api.h>
62#include <linux/static_key.h>
63#include <linux/stop_machine.h>
64#include <linux/syscalls_api.h>
65#include <linux/syscalls.h>
66#include <linux/tick.h>
67#include <linux/topology.h>
68#include <linux/types.h>
69#include <linux/u64_stats_sync_api.h>
70#include <linux/uaccess.h>
71#include <linux/vmstat.h>
72#include <linux/wait_api.h>
73#include <linux/wait_bit.h>
74#include <linux/workqueue_api.h>
75#include <linux/delayacct.h>
76#include <linux/mmu_context.h>
77
78#include <trace/events/power.h>
79#include <trace/events/sched.h>
80
81#include "../workqueue_internal.h"
82
83struct rq;
84struct cfs_rq;
85struct rt_rq;
86struct sched_group;
87struct cpuidle_state;
88
89#if defined(CONFIG_PARAVIRT) && !defined(CONFIG_HAVE_PV_STEAL_CLOCK_GEN)
90# include <asm/paravirt.h>
91#endif
92
93#include <asm/barrier.h>
94
95#include "cpupri.h"
96#include "cpudeadline.h"
97
98/* task_struct::on_rq states: */
99#define TASK_ON_RQ_QUEUED 1
100#define TASK_ON_RQ_MIGRATING 2
101
102extern __read_mostly int scheduler_running;
103
104extern unsigned long calc_load_update;
105extern atomic_long_t calc_load_tasks;
106
107extern void calc_global_load_tick(struct rq *this_rq);
108extern long calc_load_fold_active(struct rq *this_rq, long adjust);
109
110extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
111
112extern int sysctl_sched_rt_period;
113extern int sysctl_sched_rt_runtime;
114extern int sched_rr_timeslice;
115
116/*
117 * Asymmetric CPU capacity bits
118 */
119struct asym_cap_data {
120 struct list_head link;
121 struct rcu_head rcu;
122 unsigned long capacity;
123 unsigned long cpus[];
124};
125
126extern struct list_head asym_cap_list;
127
128#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
129
130/*
131 * Helpers for converting nanosecond timing to jiffy resolution
132 */
133#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
134
135/*
136 * Increase resolution of nice-level calculations for 64-bit architectures.
137 * The extra resolution improves shares distribution and load balancing of
138 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
139 * hierarchies, especially on larger systems. This is not a user-visible change
140 * and does not change the user-interface for setting shares/weights.
141 *
142 * We increase resolution only if we have enough bits to allow this increased
143 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
144 * are pretty high and the returns do not justify the increased costs.
145 *
146 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
147 * increase coverage and consistency always enable it on 64-bit platforms.
148 */
149#ifdef CONFIG_64BIT
150# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
151# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
152# define scale_load_down(w) \
153({ \
154 unsigned long __w = (w); \
155 \
156 if (__w) \
157 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
158 __w; \
159})
160#else
161# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
162# define scale_load(w) (w)
163# define scale_load_down(w) (w)
164#endif
165
166/*
167 * Task weight (visible to users) and its load (invisible to users) have
168 * independent resolution, but they should be well calibrated. We use
169 * scale_load() and scale_load_down(w) to convert between them. The
170 * following must be true:
171 *
172 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
173 *
174 */
175#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
176
177/*
178 * Single value that decides SCHED_DEADLINE internal math precision.
179 * 10 -> just above 1us
180 * 9 -> just above 0.5us
181 */
182#define DL_SCALE 10
183
184/*
185 * Single value that denotes runtime == period, ie unlimited time.
186 */
187#define RUNTIME_INF ((u64)~0ULL)
188
189static inline int idle_policy(int policy)
190{
191 return policy == SCHED_IDLE;
192}
193
194static inline int normal_policy(int policy)
195{
196#ifdef CONFIG_SCHED_CLASS_EXT
197 if (policy == SCHED_EXT)
198 return true;
199#endif
200 return policy == SCHED_NORMAL;
201}
202
203static inline int fair_policy(int policy)
204{
205 return normal_policy(policy) || policy == SCHED_BATCH;
206}
207
208static inline int rt_policy(int policy)
209{
210 return policy == SCHED_FIFO || policy == SCHED_RR;
211}
212
213static inline int dl_policy(int policy)
214{
215 return policy == SCHED_DEADLINE;
216}
217
218static inline bool valid_policy(int policy)
219{
220 return idle_policy(policy) || fair_policy(policy) ||
221 rt_policy(policy) || dl_policy(policy);
222}
223
224static inline int task_has_idle_policy(struct task_struct *p)
225{
226 return idle_policy(p->policy);
227}
228
229static inline int task_has_rt_policy(struct task_struct *p)
230{
231 return rt_policy(p->policy);
232}
233
234static inline int task_has_dl_policy(struct task_struct *p)
235{
236 return dl_policy(p->policy);
237}
238
239#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
240
241static inline void update_avg(u64 *avg, u64 sample)
242{
243 s64 diff = sample - *avg;
244
245 *avg += diff / 8;
246}
247
248/*
249 * Shifting a value by an exponent greater *or equal* to the size of said value
250 * is UB; cap at size-1.
251 */
252#define shr_bound(val, shift) \
253 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
254
255/*
256 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
257 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
258 * maps pretty well onto the shares value used by scheduler and the round-trip
259 * conversions preserve the original value over the entire range.
260 */
261static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
262{
263 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
264}
265
266static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
267{
268 return clamp_t(unsigned long,
269 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
270 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
271}
272
273/*
274 * !! For sched_setattr_nocheck() (kernel) only !!
275 *
276 * This is actually gross. :(
277 *
278 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
279 * tasks, but still be able to sleep. We need this on platforms that cannot
280 * atomically change clock frequency. Remove once fast switching will be
281 * available on such platforms.
282 *
283 * SUGOV stands for SchedUtil GOVernor.
284 */
285#define SCHED_FLAG_SUGOV 0x10000000
286
287#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
288
289static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
290{
291#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
292 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
293#else
294 return false;
295#endif
296}
297
298/*
299 * Tells if entity @a should preempt entity @b.
300 */
301static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
302 const struct sched_dl_entity *b)
303{
304 return dl_entity_is_special(a) ||
305 dl_time_before(a->deadline, b->deadline);
306}
307
308/*
309 * This is the priority-queue data structure of the RT scheduling class:
310 */
311struct rt_prio_array {
312 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
313 struct list_head queue[MAX_RT_PRIO];
314};
315
316struct rt_bandwidth {
317 /* nests inside the rq lock: */
318 raw_spinlock_t rt_runtime_lock;
319 ktime_t rt_period;
320 u64 rt_runtime;
321 struct hrtimer rt_period_timer;
322 unsigned int rt_period_active;
323};
324
325static inline int dl_bandwidth_enabled(void)
326{
327 return sysctl_sched_rt_runtime >= 0;
328}
329
330/*
331 * To keep the bandwidth of -deadline tasks under control
332 * we need some place where:
333 * - store the maximum -deadline bandwidth of each cpu;
334 * - cache the fraction of bandwidth that is currently allocated in
335 * each root domain;
336 *
337 * This is all done in the data structure below. It is similar to the
338 * one used for RT-throttling (rt_bandwidth), with the main difference
339 * that, since here we are only interested in admission control, we
340 * do not decrease any runtime while the group "executes", neither we
341 * need a timer to replenish it.
342 *
343 * With respect to SMP, bandwidth is given on a per root domain basis,
344 * meaning that:
345 * - bw (< 100%) is the deadline bandwidth of each CPU;
346 * - total_bw is the currently allocated bandwidth in each root domain;
347 */
348struct dl_bw {
349 raw_spinlock_t lock;
350 u64 bw;
351 u64 total_bw;
352};
353
354extern void init_dl_bw(struct dl_bw *dl_b);
355extern int sched_dl_global_validate(void);
356extern void sched_dl_do_global(void);
357extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
358extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
359extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr, unsigned int flags);
360extern bool __checkparam_dl(const struct sched_attr *attr);
361extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
362extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
363extern int dl_bw_deactivate(int cpu);
364extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
365/*
366 * SCHED_DEADLINE supports servers (nested scheduling) with the following
367 * interface:
368 *
369 * dl_se::rq -- runqueue we belong to.
370 *
371 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
372 * returns NULL.
373 *
374 * dl_server_update() -- called from update_curr_common(), propagates runtime
375 * to the server.
376 *
377 * dl_server_start() -- start the server when it has tasks; it will stop
378 * automatically when there are no more tasks, per
379 * dl_se::server_pick() returning NULL.
380 *
381 * dl_server_stop() -- (force) stop the server; use when updating
382 * parameters.
383 *
384 * dl_server_init() -- initializes the server.
385 *
386 * When started the dl_server will (per dl_defer) schedule a timer for its
387 * zero-laxity point -- that is, unlike regular EDF tasks which run ASAP, a
388 * server will run at the very end of its period.
389 *
390 * This is done such that any runtime from the target class can be accounted
391 * against the server -- through dl_server_update() above -- such that when it
392 * becomes time to run, it might already be out of runtime and get deferred
393 * until the next period. In this case dl_server_timer() will alternate
394 * between defer and replenish but never actually enqueue the server.
395 *
396 * Only when the target class does not manage to exhaust the server's runtime
397 * (there's actualy starvation in the given period), will the dl_server get on
398 * the runqueue. Once queued it will pick tasks from the target class and run
399 * them until either its runtime is exhaused, at which point its back to
400 * dl_server_timer, or until there are no more tasks to run, at which point
401 * the dl_server stops itself.
402 *
403 * By stopping at this point the dl_server retains bandwidth, which, if a new
404 * task wakes up imminently (starting the server again), can be used --
405 * subject to CBS wakeup rules -- without having to wait for the next period.
406 *
407 * Additionally, because of the dl_defer behaviour the start/stop behaviour is
408 * naturally thottled to once per period, avoiding high context switch
409 * workloads from spamming the hrtimer program/cancel paths.
410 */
411extern void dl_server_update_idle(struct sched_dl_entity *dl_se, s64 delta_exec);
412extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
413extern void dl_server_start(struct sched_dl_entity *dl_se);
414extern void dl_server_stop(struct sched_dl_entity *dl_se);
415extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
416 dl_server_pick_f pick_task);
417extern void sched_init_dl_servers(void);
418
419extern void fair_server_init(struct rq *rq);
420extern void ext_server_init(struct rq *rq);
421extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
422extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
423 u64 runtime, u64 period, bool init);
424extern int dl_server_attach_bw(struct sched_dl_entity *dl_se);
425extern void dl_server_detach_bw(struct sched_dl_entity *dl_se);
426extern int dl_server_swap_bw(struct sched_dl_entity *detach_se,
427 struct sched_dl_entity *attach_se);
428
429static inline bool dl_server_active(struct sched_dl_entity *dl_se)
430{
431 return dl_se->dl_server_active;
432}
433
434#ifdef CONFIG_CGROUP_SCHED
435
436extern struct list_head task_groups;
437
438#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
439extern const u64 max_bw_quota_period_us;
440
441/*
442 * default period for group bandwidth.
443 * default: 0.1s, units: microseconds
444 */
445static inline u64 default_bw_period_us(void)
446{
447 return 100000ULL;
448}
449#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */
450
451struct cfs_bandwidth {
452#ifdef CONFIG_CFS_BANDWIDTH
453 raw_spinlock_t lock;
454 ktime_t period;
455 u64 quota;
456 u64 runtime;
457 u64 burst;
458 u64 runtime_snap;
459 s64 hierarchical_quota;
460
461 u8 idle;
462 u8 period_active;
463 u8 slack_started;
464 struct hrtimer period_timer;
465 struct hrtimer slack_timer;
466 struct list_head throttled_cfs_rq;
467
468 /* Statistics: */
469 int nr_periods;
470 int nr_throttled;
471 int nr_burst;
472 u64 throttled_time;
473 u64 burst_time;
474#endif /* CONFIG_CFS_BANDWIDTH */
475};
476
477/* Task group related information */
478struct task_group {
479 struct cgroup_subsys_state css;
480
481#ifdef CONFIG_GROUP_SCHED_WEIGHT
482 /* A positive value indicates that this is a SCHED_IDLE group. */
483 int idle;
484#endif
485
486#ifdef CONFIG_FAIR_GROUP_SCHED
487 /* runqueue "owned" by this group on each CPU */
488 struct cfs_rq __percpu *cfs_rq;
489 unsigned long shares;
490 /*
491 * load_avg can be heavily contended at clock tick time, so put
492 * it in its own cache-line separated from the fields above which
493 * will also be accessed at each tick.
494 */
495 atomic_long_t load_avg ____cacheline_aligned;
496 atomic_long_t runnable_avg;
497
498#endif /* CONFIG_FAIR_GROUP_SCHED */
499
500#ifdef CONFIG_RT_GROUP_SCHED
501 struct sched_rt_entity **rt_se;
502 struct rt_rq **rt_rq;
503
504 struct rt_bandwidth rt_bandwidth;
505#endif
506
507 struct scx_task_group scx;
508
509 struct rcu_head rcu;
510 struct list_head list;
511
512 struct task_group *parent;
513 struct list_head siblings;
514 struct list_head children;
515
516#ifdef CONFIG_SCHED_AUTOGROUP
517 struct autogroup *autogroup;
518#endif
519
520 struct cfs_bandwidth cfs_bandwidth;
521
522#ifdef CONFIG_UCLAMP_TASK_GROUP
523 /* The two decimal precision [%] value requested from user-space */
524 unsigned int uclamp_pct[UCLAMP_CNT];
525 /* Clamp values requested for a task group */
526 struct uclamp_se uclamp_req[UCLAMP_CNT];
527 /* Effective clamp values used for a task group */
528 struct uclamp_se uclamp[UCLAMP_CNT];
529#endif
530
531};
532
533#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
534
535typedef int (*tg_visitor)(struct task_group *, void *);
536
537extern int walk_tg_tree_from(struct task_group *from,
538 tg_visitor down, tg_visitor up, void *data);
539
540/*
541 * Iterate the full tree, calling @down when first entering a node and @up when
542 * leaving it for the final time.
543 *
544 * Caller must hold rcu_lock or sufficient equivalent.
545 */
546static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
547{
548 return walk_tg_tree_from(&root_task_group, down, up, data);
549}
550
551static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
552{
553 return css ? container_of(css, struct task_group, css) : NULL;
554}
555
556extern int tg_nop(struct task_group *tg, void *data);
557
558#ifdef CONFIG_FAIR_GROUP_SCHED
559extern void free_fair_sched_group(struct task_group *tg);
560extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
561extern void online_fair_sched_group(struct task_group *tg);
562extern void unregister_fair_sched_group(struct task_group *tg);
563extern void __sched_cgroup_mode_update(int mode);
564#else /* !CONFIG_FAIR_GROUP_SCHED: */
565static inline void free_fair_sched_group(struct task_group *tg) { }
566static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
567{
568 return 1;
569}
570static inline void online_fair_sched_group(struct task_group *tg) { }
571static inline void unregister_fair_sched_group(struct task_group *tg) { }
572#endif /* !CONFIG_FAIR_GROUP_SCHED */
573
574extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
575 struct sched_entity *se, int cpu,
576 struct sched_entity *parent);
577extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
578
579extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
580extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
581extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
582extern bool cfs_task_bw_constrained(struct task_struct *p);
583
584extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
585 struct sched_rt_entity *rt_se, int cpu,
586 struct sched_rt_entity *parent);
587extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
588extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
589extern long sched_group_rt_runtime(struct task_group *tg);
590extern long sched_group_rt_period(struct task_group *tg);
591extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
592
593extern struct task_group *sched_create_group(struct task_group *parent);
594extern void sched_online_group(struct task_group *tg,
595 struct task_group *parent);
596extern void sched_destroy_group(struct task_group *tg);
597extern void sched_release_group(struct task_group *tg);
598
599extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
600
601#ifdef CONFIG_FAIR_GROUP_SCHED
602extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
603
604extern int sched_group_set_idle(struct task_group *tg, long idle);
605
606extern void set_task_rq_fair(struct sched_entity *se,
607 struct cfs_rq *prev, struct cfs_rq *next);
608#else /* !CONFIG_FAIR_GROUP_SCHED: */
609static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
610static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
611#endif /* !CONFIG_FAIR_GROUP_SCHED */
612
613#else /* !CONFIG_CGROUP_SCHED: */
614
615struct cfs_bandwidth { };
616
617static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
618
619#endif /* !CONFIG_CGROUP_SCHED */
620
621/*
622 * A weight of 0 or 1 can cause arithmetics problems.
623 * A weight of a cfs_rq is the sum of weights of which entities
624 * are queued on this cfs_rq, so a weight of a entity should not be
625 * too large, so as the shares value of a task group.
626 * (The default weight is 1024 - so there's no practical
627 * limitation from this.)
628 */
629#define MIN_SHARES (1UL << 1)
630#define MAX_SHARES (1UL << 18)
631
632extern void unregister_rt_sched_group(struct task_group *tg);
633extern void free_rt_sched_group(struct task_group *tg);
634extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
635
636/*
637 * u64_u32_load/u64_u32_store
638 *
639 * Use a copy of a u64 value to protect against data race. This is only
640 * applicable for 32-bits architectures.
641 */
642#ifdef CONFIG_64BIT
643# define u64_u32_load_copy(var, copy) var
644# define u64_u32_store_copy(var, copy, val) (var = val)
645#else
646# define u64_u32_load_copy(var, copy) \
647({ \
648 u64 __val, __val_copy; \
649 do { \
650 __val_copy = copy; \
651 /* \
652 * paired with u64_u32_store_copy(), ordering access \
653 * to var and copy. \
654 */ \
655 smp_rmb(); \
656 __val = var; \
657 } while (__val != __val_copy); \
658 __val; \
659})
660# define u64_u32_store_copy(var, copy, val) \
661do { \
662 typeof(val) __val = (val); \
663 var = __val; \
664 /* \
665 * paired with u64_u32_load_copy(), ordering access to var and \
666 * copy. \
667 */ \
668 smp_wmb(); \
669 copy = __val; \
670} while (0)
671#endif
672# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
673# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
674
675struct balance_callback {
676 struct balance_callback *next;
677 void (*func)(struct rq *rq);
678};
679
680/* Fair scheduling SCHED_{NORMAL,BATCH,IDLE} related fields in a runqueue: */
681struct cfs_rq {
682 struct load_weight load;
683 unsigned int nr_queued;
684 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */
685 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */
686 unsigned int h_nr_idle; /* SCHED_IDLE */
687
688 s64 sum_w_vruntime;
689 u64 sum_weight;
690 u64 zero_vruntime;
691 unsigned int sum_shift;
692
693#ifdef CONFIG_SCHED_CORE
694 unsigned int forceidle_seq;
695 u64 zero_vruntime_fi;
696#endif
697
698 struct rb_root_cached tasks_timeline;
699
700 /*
701 * 'curr' points to the currently running entity on this cfs_rq.
702 * It is set to NULL otherwise (i.e when none are currently running).
703 */
704 struct sched_entity *curr;
705 struct sched_entity *next;
706
707 /*
708 * CFS load tracking
709 */
710 struct sched_entity *h_curr;
711 struct sched_avg avg;
712#ifndef CONFIG_64BIT
713 u64 last_update_time_copy;
714#endif
715 struct {
716 raw_spinlock_t lock ____cacheline_aligned;
717 int nr;
718 unsigned long load_avg;
719 unsigned long util_avg;
720 unsigned long runnable_avg;
721 } removed;
722
723#ifdef CONFIG_FAIR_GROUP_SCHED
724 u64 last_update_tg_load_avg;
725 unsigned long tg_load_avg_contrib;
726 unsigned long tg_runnable_avg_contrib;
727 long propagate;
728 long prop_runnable_sum;
729
730 /*
731 * h_load = weight * f(tg)
732 *
733 * Where f(tg) is the recursive weight fraction assigned to
734 * this group.
735 */
736 unsigned long h_load;
737 u64 last_h_load_update;
738 struct sched_entity *h_load_next;
739
740 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
741
742 /*
743 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
744 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
745 * (like users, containers etc.)
746 *
747 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
748 * This list is used during load balance.
749 */
750 int on_list;
751 struct list_head leaf_cfs_rq_list;
752 struct task_group *tg; /* Group that "owns" this runqueue */
753
754 /* Locally cached copy of our task_group's idle value */
755 int idle;
756
757# ifdef CONFIG_CFS_BANDWIDTH
758 int runtime_enabled;
759 s64 runtime_remaining;
760
761 u64 throttled_pelt_idle;
762# ifndef CONFIG_64BIT
763 u64 throttled_pelt_idle_copy;
764# endif
765 u64 throttled_clock;
766 u64 throttled_clock_pelt;
767 u64 throttled_clock_pelt_time;
768 u64 throttled_clock_self;
769 u64 throttled_clock_self_time;
770 bool throttled:1;
771 bool pelt_clock_throttled:1;
772 int throttle_count;
773 struct list_head throttled_list;
774 struct list_head throttled_csd_list;
775 struct list_head throttled_limbo_list;
776# endif /* CONFIG_CFS_BANDWIDTH */
777#endif /* CONFIG_FAIR_GROUP_SCHED */
778};
779
780#ifdef CONFIG_SCHED_CLASS_EXT
781/* scx_rq->flags, protected by the rq lock */
782enum scx_rq_flags {
783 /*
784 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
785 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
786 * only while the BPF scheduler considers the CPU to be online.
787 */
788 SCX_RQ_ONLINE = 1 << 0,
789 SCX_RQ_CAN_STOP_TICK = 1 << 1,
790 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */
791 SCX_RQ_BAL_CB_PENDING = 1 << 6, /* must queue a cb after dispatching */
792 SCX_RQ_SUB_IDLE_RENOTIFY = 1 << 7, /* sub-scheds are owed update_idle() */
793 SCX_RQ_ROOT_IDLE_RENOTIFY = 1 << 8, /* the root is owed update_idle() */
794
795 SCX_RQ_IN_WAKEUP = 1 << 16,
796 SCX_RQ_IN_DISPATCH = 1 << 17,
797};
798
799/* per-rq rescue execution state, see scx_rescue_timerfn() */
800struct scx_rq_rescue {
801 struct scx_dispatch_q dsq; /* stranded tasks awaiting rescue */
802 s64 budget; /* execution token bucket, ns */
803 u64 clock; /* last budget accrual timestamp */
804 struct task_struct *curr; /* task being rescued, one at a time */
805 s64 slice; /* curr's admitted slice */
806 u64 exec_snap; /* sum_exec_runtime at admission */
807 struct timer_list timer; /* paces admission and escalation */
808 u64 kill_at; /* last ejection, init before any */
809};
810
811struct scx_rq {
812 struct scx_dispatch_q local_dsq;
813#ifdef CONFIG_EXT_SUB_SCHED
814 struct scx_dispatch_q reject_dsq; /* staging for cap-rejected tasks */
815 struct scx_rq_rescue rescue;
816#endif
817 struct list_head runnable_list; /* runnable tasks on this rq */
818 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
819 /* both stashed across the activate_task() in move_remote_task_to_local_dsq() */
820 u64 remote_activate_enq_flags;
821 struct scx_sched *remote_activate_sch;
822 u32 nr_running;
823 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
824 bool in_select_cpu;
825 bool cpu_released;
826 u32 flags;
827 u32 nr_immed; /* ENQ_IMMED tasks on local_dsq */
828#ifdef CONFIG_SCHED_CORE
829 u32 lock_drop_seq; /* nr dispatch lock releases */
830#endif
831 u64 clock; /* current per-rq clock -- see scx_bpf_now() */
832#ifdef CONFIG_EXT_SUB_SCHED
833 struct llist_head ecaps_to_sync; /* pending ecaps syncs */
834 struct task_struct *sub_dispatch_prev;
835#endif
836 cpumask_var_t cpus_to_sync;
837 bool kick_sync_pending;
838 unsigned long kick_sync;
839
840 struct list_head sched_pcpus_to_kick; /* see kick_cpus_irq_workfn() */
841
842 raw_spinlock_t deferred_reenq_lock;
843 struct list_head deferred_reenq_locals; /* scheds requesting reenq of local DSQ */
844 struct list_head deferred_reenq_users; /* user DSQs requesting reenq */
845 struct balance_callback deferred_bal_cb;
846 struct balance_callback kick_sync_bal_cb;
847 struct irq_work deferred_irq_work;
848 struct irq_work kick_cpus_irq_work;
849};
850#endif /* CONFIG_SCHED_CLASS_EXT */
851
852static inline int rt_bandwidth_enabled(void)
853{
854 return sysctl_sched_rt_runtime >= 0;
855}
856
857/* RT IPI pull logic requires IRQ_WORK */
858#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
859# define HAVE_RT_PUSH_IPI
860#endif
861
862/* Real-Time classes' related field in a runqueue: */
863struct rt_rq {
864 struct rt_prio_array active;
865 unsigned int rt_nr_running;
866 unsigned int rr_nr_running;
867 struct {
868 int curr; /* highest queued rt task prio */
869 int next; /* next highest */
870 } highest_prio;
871 bool overloaded;
872 struct plist_head pushable_tasks;
873
874 int rt_queued;
875
876#ifdef CONFIG_RT_GROUP_SCHED
877 int rt_throttled;
878 u64 rt_time; /* consumed RT time, goes up in update_curr_rt */
879 u64 rt_runtime; /* allotted RT time, "slice" from rt_bandwidth, RT sharing/balancing */
880 /* Nests inside the rq lock: */
881 raw_spinlock_t rt_runtime_lock;
882
883 unsigned int rt_nr_boosted;
884
885 struct rq *rq; /* this is always top-level rq, cache? */
886#endif
887#ifdef CONFIG_CGROUP_SCHED
888 struct task_group *tg; /* this tg has "this" rt_rq on given CPU for runnable entities */
889#endif
890};
891
892static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
893{
894 return rt_rq->rt_queued && rt_rq->rt_nr_running;
895}
896
897/* Deadline class' related fields in a runqueue */
898struct dl_rq {
899 /* runqueue is an rbtree, ordered by deadline */
900 struct rb_root_cached root;
901
902 unsigned int dl_nr_running;
903
904 /*
905 * Deadline values of the currently executing and the
906 * earliest ready task on this rq. Caching these facilitates
907 * the decision whether or not a ready but not running task
908 * should migrate somewhere else.
909 */
910 struct {
911 u64 curr;
912 u64 next;
913 } earliest_dl;
914
915 bool overloaded;
916
917 struct sched_dl_entity *curr;
918 /*
919 * Tasks on this rq that can be pushed away. They are kept in
920 * an rb-tree, ordered by tasks' deadlines, with caching
921 * of the leftmost (earliest deadline) element.
922 */
923 struct rb_root_cached pushable_dl_tasks_root;
924
925 /*
926 * "Active utilization" for this runqueue: increased when a
927 * task wakes up (becomes TASK_RUNNING) and decreased when a
928 * task blocks
929 */
930 u64 running_bw;
931
932 /*
933 * Utilization of the tasks "assigned" to this runqueue (including
934 * the tasks that are in runqueue and the tasks that executed on this
935 * CPU and blocked). Increased when a task moves to this runqueue, and
936 * decreased when the task moves away (migrates, changes scheduling
937 * policy, or terminates).
938 * This is needed to compute the "inactive utilization" for the
939 * runqueue (inactive utilization = this_bw - running_bw).
940 */
941 u64 this_bw;
942 u64 extra_bw;
943
944 /*
945 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
946 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
947 */
948 u64 max_bw;
949
950 /*
951 * Inverse of the fraction of CPU utilization that can be reclaimed
952 * by the GRUB algorithm.
953 */
954 u64 bw_ratio;
955};
956
957#ifdef CONFIG_FAIR_GROUP_SCHED
958/* Check whether a task group is root tg */
959#define is_root_task_group(tg) ((tg) == &root_task_group)
960/* An entity is a task if it doesn't "own" a runqueue */
961#define entity_is_task(se) (!se->my_q)
962
963static inline void se_update_runnable(struct sched_entity *se)
964{
965 if (!entity_is_task(se))
966 se->runnable_weight = se->my_q->h_nr_runnable;
967}
968
969static inline long se_runnable(struct sched_entity *se)
970{
971 if (se->sched_delayed)
972 return false;
973
974 if (entity_is_task(se))
975 return !!se->on_rq;
976 else
977 return se->runnable_weight;
978}
979
980#else /* !CONFIG_FAIR_GROUP_SCHED: */
981
982#define entity_is_task(se) 1
983
984static inline void se_update_runnable(struct sched_entity *se) { }
985
986static inline long se_runnable(struct sched_entity *se)
987{
988 if (se->sched_delayed)
989 return false;
990
991 return !!se->on_rq;
992}
993
994#endif /* !CONFIG_FAIR_GROUP_SCHED */
995
996/*
997 * XXX we want to get rid of these helpers and use the full load resolution.
998 */
999static inline long se_weight(struct sched_entity *se)
1000{
1001 return scale_load_down(se->load.weight);
1002}
1003
1004
1005static inline bool sched_asym_prefer(int a, int b)
1006{
1007 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
1008}
1009
1010struct perf_domain {
1011 struct em_perf_domain *em_pd;
1012 struct perf_domain *next;
1013 struct rcu_head rcu;
1014};
1015
1016/*
1017 * We add the notion of a root-domain which will be used to define per-domain
1018 * variables. Each exclusive cpuset essentially defines an island domain by
1019 * fully partitioning the member CPUs from any other cpuset. Whenever a new
1020 * exclusive cpuset is created, we also create and attach a new root-domain
1021 * object.
1022 *
1023 */
1024struct root_domain {
1025 atomic_t refcount;
1026 atomic_t rto_count;
1027 struct rcu_head rcu;
1028 cpumask_var_t span;
1029 cpumask_var_t online;
1030
1031 /*
1032 * Indicate pullable load on at least one CPU, e.g:
1033 * - More than one runnable task
1034 * - Running task is misfit
1035 */
1036 bool overloaded;
1037
1038 /* Indicate one or more CPUs over-utilized (tipping point) */
1039 bool overutilized;
1040
1041 /*
1042 * The bit corresponding to a CPU gets set here if such CPU has more
1043 * than one runnable -deadline task (as it is below for RT tasks).
1044 */
1045 cpumask_var_t dlo_mask;
1046 atomic_t dlo_count;
1047 struct dl_bw dl_bw;
1048 struct cpudl cpudl;
1049
1050 /*
1051 * Indicate whether a root_domain's dl_bw has been checked or
1052 * updated. It's monotonously increasing value.
1053 *
1054 * Also, some corner cases, like 'wrap around' is dangerous, but given
1055 * that u64 is 'big enough'. So that shouldn't be a concern.
1056 */
1057 u64 visit_cookie;
1058
1059#ifdef HAVE_RT_PUSH_IPI
1060 /*
1061 * For IPI pull requests, loop across the rto_mask.
1062 */
1063 struct irq_work rto_push_work;
1064 raw_spinlock_t rto_lock;
1065 /* These are only updated and read within rto_lock */
1066 int rto_loop;
1067 int rto_cpu;
1068 /* These atomics are updated outside of a lock */
1069 atomic_t rto_loop_next;
1070 atomic_t rto_loop_start;
1071#endif /* HAVE_RT_PUSH_IPI */
1072 /*
1073 * The "RT overload" flag: it gets set if a CPU has more than
1074 * one runnable RT task.
1075 */
1076 cpumask_var_t rto_mask;
1077 struct cpupri cpupri;
1078
1079 /*
1080 * NULL-terminated list of performance domains intersecting with the
1081 * CPUs of the rd. Protected by RCU.
1082 */
1083 struct perf_domain __rcu *pd;
1084};
1085
1086extern void init_defrootdomain(void);
1087extern int sched_init_domains(const struct cpumask *cpu_map);
1088extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1089extern void sched_get_rd(struct root_domain *rd);
1090extern void sched_put_rd(struct root_domain *rd);
1091
1092static inline int get_rd_overloaded(struct root_domain *rd)
1093{
1094 return READ_ONCE(rd->overloaded);
1095}
1096
1097static inline void set_rd_overloaded(struct root_domain *rd, int status)
1098{
1099 if (get_rd_overloaded(rd) != status)
1100 WRITE_ONCE(rd->overloaded, status);
1101}
1102
1103#ifdef HAVE_RT_PUSH_IPI
1104extern void rto_push_irq_work_func(struct irq_work *work);
1105#endif
1106
1107#ifdef CONFIG_UCLAMP_TASK
1108/*
1109 * struct uclamp_bucket - Utilization clamp bucket
1110 * @value: utilization clamp value for tasks on this clamp bucket
1111 * @tasks: number of RUNNABLE tasks on this clamp bucket
1112 *
1113 * Keep track of how many tasks are RUNNABLE for a given utilization
1114 * clamp value.
1115 */
1116struct uclamp_bucket {
1117 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1118 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1119};
1120
1121/*
1122 * struct uclamp_rq - rq's utilization clamp
1123 * @value: currently active clamp values for a rq
1124 * @bucket: utilization clamp buckets affecting a rq
1125 *
1126 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1127 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1128 * (or actually running) with that value.
1129 *
1130 * There are up to UCLAMP_CNT possible different clamp values, currently there
1131 * are only two: minimum utilization and maximum utilization.
1132 *
1133 * All utilization clamping values are MAX aggregated, since:
1134 * - for util_min: we want to run the CPU at least at the max of the minimum
1135 * utilization required by its currently RUNNABLE tasks.
1136 * - for util_max: we want to allow the CPU to run up to the max of the
1137 * maximum utilization allowed by its currently RUNNABLE tasks.
1138 *
1139 * Since on each system we expect only a limited number of different
1140 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1141 * the metrics required to compute all the per-rq utilization clamp values.
1142 */
1143struct uclamp_rq {
1144 unsigned int value;
1145 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1146};
1147
1148DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1149#endif /* CONFIG_UCLAMP_TASK */
1150
1151/*
1152 * This is the main, per-CPU runqueue data structure.
1153 *
1154 * Locking rule: those places that want to lock multiple runqueues
1155 * (such as the load balancing or the thread migration code), lock
1156 * acquire operations must be ordered by ascending &runqueue.
1157 */
1158struct rq {
1159 /*
1160 * The following members are loaded together, without holding the
1161 * rq->lock, in an extremely hot loop in update_sg_lb_stats()
1162 * (called from pick_next_task()). To reduce cache pollution from
1163 * this operation, they are placed together on this dedicated cache
1164 * line. Even though some of them are frequently modified, they are
1165 * loaded much more frequently than they are stored.
1166 */
1167 unsigned int nr_running;
1168#ifdef CONFIG_NUMA_BALANCING
1169 unsigned int nr_numa_running;
1170 unsigned int nr_preferred_running;
1171#endif
1172 unsigned int ttwu_pending;
1173 unsigned long cpu_capacity;
1174#ifdef CONFIG_SCHED_PROXY_EXEC
1175 struct task_struct __rcu *donor; /* Scheduling context */
1176 struct task_struct __rcu *curr; /* Execution context */
1177#else
1178 union {
1179 struct task_struct __rcu *donor; /* Scheduler context */
1180 struct task_struct __rcu *curr; /* Execution context */
1181 };
1182#endif
1183 struct task_struct *idle;
1184 /* padding left here deliberately */
1185
1186 /*
1187 * The next cacheline holds the (hot) runqueue lock, as well as
1188 * some other less performance-critical fields.
1189 */
1190 u64 nr_switches ____cacheline_aligned;
1191
1192 /* runqueue lock: */
1193 raw_spinlock_t __lock;
1194
1195#ifdef CONFIG_NO_HZ_COMMON
1196 unsigned int nohz_tick_stopped;
1197 atomic_t nohz_flags;
1198 unsigned int has_blocked_load;
1199 unsigned long last_blocked_load_update_tick;
1200 call_single_data_t nohz_csd;
1201#endif /* CONFIG_NO_HZ_COMMON */
1202
1203#ifdef CONFIG_UCLAMP_TASK
1204 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1205 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1206 unsigned int uclamp_flags;
1207#define UCLAMP_FLAG_IDLE 0x01
1208#endif
1209
1210 struct cfs_rq cfs;
1211 struct rt_rq rt;
1212 struct dl_rq dl;
1213#ifdef CONFIG_SCHED_CLASS_EXT
1214 struct scx_rq scx;
1215 struct sched_dl_entity ext_server;
1216#endif
1217#ifdef CONFIG_SCHED_CACHE
1218 raw_spinlock_t cpu_epoch_lock ____cacheline_aligned;
1219 u64 cpu_runtime;
1220 unsigned long cpu_epoch;
1221 unsigned long cpu_epoch_next;
1222#endif
1223
1224 struct sched_dl_entity fair_server;
1225
1226#ifdef CONFIG_FAIR_GROUP_SCHED
1227 /* list of leaf cfs_rq on this CPU: */
1228 struct list_head leaf_cfs_rq_list;
1229 struct list_head *tmp_alone_branch;
1230#endif /* CONFIG_FAIR_GROUP_SCHED */
1231
1232#ifdef CONFIG_NUMA_BALANCING
1233 unsigned int numa_migrate_on;
1234#endif
1235
1236#ifdef CONFIG_SCHED_CACHE
1237 unsigned int nr_pref_llc_running;
1238 unsigned int nr_llc_running;
1239#endif
1240
1241 /*
1242 * This is part of a global counter where only the total sum
1243 * over all CPUs matters. A task can increase this counter on
1244 * one CPU and if it got migrated afterwards it may decrease
1245 * it on another CPU. Always updated under the runqueue lock:
1246 */
1247 unsigned long nr_uninterruptible;
1248
1249 struct sched_dl_entity *dl_server;
1250 struct task_struct *stop;
1251 const struct sched_class *next_class;
1252 unsigned long next_balance;
1253 struct mm_struct *prev_mm;
1254
1255 /*
1256 * The following fields of clock data are frequently referenced
1257 * and updated together, and should go on their own cache line.
1258 */
1259 u64 clock_task ____cacheline_aligned;
1260 u64 clock_pelt;
1261 u64 clock;
1262 unsigned long lost_idle_time;
1263 unsigned int clock_update_flags;
1264 u64 clock_pelt_idle;
1265 u64 clock_idle;
1266
1267#ifndef CONFIG_64BIT
1268 u64 clock_pelt_idle_copy;
1269 u64 clock_idle_copy;
1270#endif
1271
1272 u64 last_seen_need_resched_ns;
1273 int ticks_without_resched;
1274
1275#ifdef CONFIG_MEMBARRIER
1276 int membarrier_state;
1277#endif
1278
1279 struct root_domain *rd;
1280 struct sched_domain __rcu *sd;
1281
1282 struct balance_callback *balance_callback;
1283
1284 unsigned char nohz_idle_balance;
1285 unsigned char idle_balance;
1286
1287 unsigned long misfit_task_load;
1288
1289 /* For active balancing */
1290 int active_balance;
1291 int push_cpu;
1292 struct cpu_stop_work active_balance_work;
1293
1294 /* CPU of this runqueue: */
1295 int cpu;
1296 int online;
1297
1298 struct list_head cfs_tasks;
1299
1300 struct sched_avg avg_rt;
1301 struct sched_avg avg_dl;
1302#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1303 struct sched_avg avg_irq;
1304#endif
1305#ifdef CONFIG_SCHED_HW_PRESSURE
1306 struct sched_avg avg_hw;
1307#endif
1308 u64 idle_stamp;
1309 u64 avg_idle;
1310
1311 /* This is used to determine avg_idle's max value */
1312 u64 max_idle_balance_cost;
1313
1314#ifdef CONFIG_HOTPLUG_CPU
1315 struct rcuwait hotplug_wait;
1316#endif
1317
1318#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1319 u64 prev_irq_time;
1320 u64 psi_irq_time;
1321#endif
1322#ifdef CONFIG_PARAVIRT
1323 u64 prev_steal_time;
1324#endif
1325#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1326 u64 prev_steal_time_rq;
1327#endif
1328#ifdef CONFIG_PREFERRED_CPU
1329 bool npc_push_work_pending;
1330#endif
1331
1332 /* calc_load related fields */
1333 unsigned long calc_load_update;
1334 long calc_load_active;
1335
1336#ifdef CONFIG_SCHED_HRTICK
1337 call_single_data_t hrtick_csd;
1338 struct hrtimer hrtick_timer;
1339 ktime_t hrtick_time;
1340 ktime_t hrtick_delay;
1341 unsigned int hrtick_sched;
1342#endif
1343
1344#ifdef CONFIG_SCHEDSTATS
1345 /* latency stats */
1346 struct sched_info rq_sched_info;
1347 unsigned long long rq_cpu_time;
1348
1349 /* sys_sched_yield() stats */
1350 unsigned int yld_count;
1351
1352 /* schedule() stats */
1353 unsigned int sched_count;
1354 unsigned int sched_goidle;
1355
1356 /* try_to_wake_up() stats */
1357 unsigned int ttwu_count;
1358 unsigned int ttwu_local;
1359#endif
1360
1361#ifdef CONFIG_CPU_IDLE
1362 /* Must be inspected within a RCU lock section */
1363 struct cpuidle_state *idle_state;
1364#endif
1365
1366 unsigned int nr_pinned;
1367 unsigned int push_busy;
1368 struct cpu_stop_work push_work;
1369
1370#ifdef CONFIG_SCHED_CORE
1371 /* per rq */
1372 struct rq *core;
1373 struct task_struct *core_pick;
1374 struct sched_dl_entity *core_dl_server;
1375 unsigned int core_enabled;
1376 struct rb_root core_tree;
1377
1378 /* shared state -- careful with sched_core_cpu_deactivate() */
1379 unsigned int core_task_seq;
1380 unsigned int core_pick_seq;
1381 unsigned long core_cookie;
1382 unsigned int core_forceidle_count;
1383 unsigned int core_forceidle_seq;
1384 unsigned int core_forceidle_occupation;
1385 u64 core_forceidle_start;
1386 unsigned int core_pick_in_flight;
1387#endif /* CONFIG_SCHED_CORE */
1388
1389 /* Scratch cpumask to be temporarily used under rq_lock */
1390 cpumask_var_t scratch_mask;
1391
1392#ifdef CONFIG_CFS_BANDWIDTH
1393 call_single_data_t cfsb_csd;
1394 struct list_head cfsb_csd_list;
1395#endif
1396
1397 atomic_t nr_iowait;
1398} __no_randomize_layout;
1399
1400#ifdef CONFIG_FAIR_GROUP_SCHED
1401
1402/* CPU runqueue to which this cfs_rq is attached */
1403static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1404{
1405 return cfs_rq->rq;
1406}
1407
1408#else /* !CONFIG_FAIR_GROUP_SCHED: */
1409
1410static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1411{
1412 return container_of(cfs_rq, struct rq, cfs);
1413}
1414#endif /* !CONFIG_FAIR_GROUP_SCHED */
1415
1416static inline int cpu_of(struct rq *rq)
1417{
1418 return rq->cpu;
1419}
1420
1421#define MDF_PUSH 0x01
1422
1423static inline bool is_migration_disabled(struct task_struct *p)
1424{
1425 return p->migration_disabled;
1426}
1427
1428DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1429DECLARE_PER_CPU(struct rnd_state, sched_rnd_state);
1430
1431static inline u32 sched_rng(void)
1432{
1433 return prandom_u32_state(this_cpu_ptr(&sched_rnd_state));
1434}
1435
1436static __always_inline struct rq *__this_rq(void)
1437{
1438 return this_cpu_ptr(&runqueues);
1439}
1440
1441#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1442#define this_rq() __this_rq()
1443#define task_rq(p) cpu_rq(task_cpu(p))
1444#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1445#define raw_rq() raw_cpu_ptr(&runqueues)
1446
1447static inline bool idle_rq(struct rq *rq)
1448{
1449 return rq->curr == rq->idle && !rq->nr_running && !rq->ttwu_pending;
1450}
1451
1452/**
1453 * available_idle_cpu - is a given CPU idle for enqueuing work.
1454 * @cpu: the CPU in question.
1455 *
1456 * Return: 1 if the CPU is currently idle. 0 otherwise.
1457 */
1458static inline bool available_idle_cpu(int cpu)
1459{
1460 if (!idle_rq(cpu_rq(cpu)))
1461 return 0;
1462
1463 if (vcpu_is_preempted(cpu))
1464 return 0;
1465
1466 return 1;
1467}
1468
1469#ifdef CONFIG_SCHED_PROXY_EXEC
1470static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1471{
1472 rcu_assign_pointer(rq->donor, t);
1473}
1474#else
1475static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1476{
1477 /* Do nothing */
1478}
1479#endif
1480
1481#ifdef CONFIG_SCHED_CORE
1482static inline struct cpumask *sched_group_span(struct sched_group *sg);
1483
1484DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1485
1486static inline bool sched_core_enabled(struct rq *rq)
1487{
1488 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1489}
1490
1491static inline bool sched_core_disabled(void)
1492{
1493 return !static_branch_unlikely(&__sched_core_enabled);
1494}
1495
1496/*
1497 * Be careful with this function; not for general use. The return value isn't
1498 * stable unless you actually hold a relevant rq->__lock.
1499 */
1500static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1501{
1502 if (sched_core_enabled(rq))
1503 return &rq->core->__lock;
1504
1505 return &rq->__lock;
1506}
1507
1508static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1509 __returns_ctx_lock(rq_lockp(rq)) /* alias them */
1510{
1511 if (rq->core_enabled)
1512 return &rq->core->__lock;
1513
1514 return &rq->__lock;
1515}
1516
1517extern bool
1518cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1519
1520extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1521
1522/*
1523 * Helpers to check if the CPU's core cookie matches with the task's cookie
1524 * when core scheduling is enabled.
1525 * A special case is that the task's cookie always matches with CPU's core
1526 * cookie if the CPU is in an idle core.
1527 */
1528static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1529{
1530 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1531 if (!sched_core_enabled(rq))
1532 return true;
1533
1534 return rq->core->core_cookie == p->core_cookie;
1535}
1536
1537static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1538{
1539 bool idle_core = true;
1540 int cpu;
1541
1542 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1543 if (!sched_core_enabled(rq))
1544 return true;
1545
1546 if (rq->core->core_cookie == p->core_cookie)
1547 return true;
1548
1549 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1550 if (!available_idle_cpu(cpu)) {
1551 idle_core = false;
1552 break;
1553 }
1554 }
1555
1556 /*
1557 * A CPU in an idle core is always the best choice for tasks with
1558 * cookies.
1559 */
1560 return idle_core;
1561}
1562
1563static inline bool sched_group_cookie_match(struct rq *rq,
1564 struct task_struct *p,
1565 struct sched_group *group)
1566{
1567 int cpu;
1568
1569 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1570 if (!sched_core_enabled(rq))
1571 return true;
1572
1573 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1574 if (sched_core_cookie_match(cpu_rq(cpu), p))
1575 return true;
1576 }
1577 return false;
1578}
1579
1580static inline bool sched_core_enqueued(struct task_struct *p)
1581{
1582 return !RB_EMPTY_NODE(&p->core_node);
1583}
1584
1585extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1586extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1587
1588extern void sched_core_get(void);
1589extern void sched_core_put(void);
1590
1591static inline bool task_has_sched_core(struct task_struct *p)
1592{
1593 if (sched_core_disabled())
1594 return false;
1595
1596 return !!p->core_cookie;
1597}
1598
1599#else /* !CONFIG_SCHED_CORE: */
1600
1601static inline bool sched_core_enabled(struct rq *rq)
1602{
1603 return false;
1604}
1605
1606static inline bool sched_core_disabled(void)
1607{
1608 return true;
1609}
1610
1611static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1612{
1613 return &rq->__lock;
1614}
1615
1616static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1617 __returns_ctx_lock(rq_lockp(rq)) /* alias them */
1618{
1619 return &rq->__lock;
1620}
1621
1622static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1623{
1624 return true;
1625}
1626
1627static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1628{
1629 return true;
1630}
1631
1632static inline bool sched_group_cookie_match(struct rq *rq,
1633 struct task_struct *p,
1634 struct sched_group *group)
1635{
1636 return true;
1637}
1638
1639static inline bool task_has_sched_core(struct task_struct *p)
1640{
1641 return false;
1642}
1643
1644#endif /* !CONFIG_SCHED_CORE */
1645
1646#ifdef CONFIG_RT_GROUP_SCHED
1647# ifdef CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED
1648DECLARE_STATIC_KEY_FALSE(rt_group_sched);
1649static inline bool rt_group_sched_enabled(void)
1650{
1651 return static_branch_unlikely(&rt_group_sched);
1652}
1653# else /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED: */
1654DECLARE_STATIC_KEY_TRUE(rt_group_sched);
1655static inline bool rt_group_sched_enabled(void)
1656{
1657 return static_branch_likely(&rt_group_sched);
1658}
1659# endif /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED */
1660#else /* !CONFIG_RT_GROUP_SCHED: */
1661# define rt_group_sched_enabled() false
1662#endif /* !CONFIG_RT_GROUP_SCHED */
1663
1664static inline void lockdep_assert_rq_held(struct rq *rq)
1665 __assumes_ctx_lock(__rq_lockp(rq))
1666{
1667 lockdep_assert_held(__rq_lockp(rq));
1668}
1669
1670extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass)
1671 __acquires(__rq_lockp(rq));
1672
1673extern bool raw_spin_rq_trylock(struct rq *rq)
1674 __cond_acquires(true, __rq_lockp(rq));
1675
1676static inline void raw_spin_rq_lock(struct rq *rq)
1677 __acquires(__rq_lockp(rq))
1678{
1679 raw_spin_rq_lock_nested(rq, 0);
1680}
1681
1682static inline void raw_spin_rq_unlock(struct rq *rq)
1683 __releases(__rq_lockp(rq))
1684{
1685 raw_spin_unlock(rq_lockp(rq));
1686}
1687
1688static inline void raw_spin_rq_lock_irq(struct rq *rq)
1689 __acquires(__rq_lockp(rq))
1690{
1691 local_irq_disable();
1692 raw_spin_rq_lock(rq);
1693}
1694
1695static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1696 __releases(__rq_lockp(rq))
1697{
1698 raw_spin_rq_unlock(rq);
1699 local_irq_enable();
1700}
1701
1702static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1703 __acquires(__rq_lockp(rq))
1704{
1705 unsigned long flags;
1706
1707 local_irq_save(flags);
1708 raw_spin_rq_lock(rq);
1709
1710 return flags;
1711}
1712
1713static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1714 __releases(__rq_lockp(rq))
1715{
1716 raw_spin_rq_unlock(rq);
1717 local_irq_restore(flags);
1718}
1719
1720#define raw_spin_rq_lock_irqsave(rq, flags) \
1721do { \
1722 flags = _raw_spin_rq_lock_irqsave(rq); \
1723} while (0)
1724
1725extern void __update_idle_core(struct rq *rq);
1726
1727static inline void update_idle_core(struct rq *rq)
1728{
1729 if (sched_smt_active())
1730 __update_idle_core(rq);
1731}
1732
1733#ifdef CONFIG_FAIR_GROUP_SCHED
1734static inline struct task_struct *task_of(struct sched_entity *se)
1735{
1736 WARN_ON_ONCE(!entity_is_task(se));
1737 return container_of(se, struct task_struct, se);
1738}
1739
1740static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1741{
1742 return p->se.cfs_rq;
1743}
1744
1745/* runqueue on which this entity is (to be) queued */
1746static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1747{
1748 return se->cfs_rq;
1749}
1750
1751/* runqueue "owned" by this group */
1752static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1753{
1754 return grp->my_q;
1755}
1756
1757#else /* !CONFIG_FAIR_GROUP_SCHED: */
1758
1759#define task_of(_se) container_of(_se, struct task_struct, se)
1760
1761static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1762{
1763 return &task_rq(p)->cfs;
1764}
1765
1766static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1767{
1768 const struct task_struct *p = task_of(se);
1769 struct rq *rq = task_rq(p);
1770
1771 return &rq->cfs;
1772}
1773
1774/* runqueue "owned" by this group */
1775static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1776{
1777 return NULL;
1778}
1779
1780#endif /* !CONFIG_FAIR_GROUP_SCHED */
1781
1782extern void update_rq_avg_idle(struct rq *rq);
1783extern void update_rq_clock(struct rq *rq);
1784
1785/*
1786 * rq::clock_update_flags bits
1787 *
1788 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1789 * call to __schedule(). This is an optimisation to avoid
1790 * neighbouring rq clock updates.
1791 *
1792 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1793 * in effect and calls to update_rq_clock() are being ignored.
1794 *
1795 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1796 * made to update_rq_clock() since the last time rq::lock was pinned.
1797 *
1798 * If inside of __schedule(), clock_update_flags will have been
1799 * shifted left (a left shift is a cheap operation for the fast path
1800 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1801 *
1802 * if (rq-clock_update_flags >= RQCF_UPDATED)
1803 *
1804 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1805 * one position though, because the next rq_unpin_lock() will shift it
1806 * back.
1807 */
1808#define RQCF_REQ_SKIP 0x01
1809#define RQCF_ACT_SKIP 0x02
1810#define RQCF_UPDATED 0x04
1811
1812static inline void assert_clock_updated(struct rq *rq)
1813{
1814 /*
1815 * The only reason for not seeing a clock update since the
1816 * last rq_pin_lock() is if we're currently skipping updates.
1817 */
1818 WARN_ON_ONCE(rq->clock_update_flags < RQCF_ACT_SKIP);
1819}
1820
1821static inline u64 rq_clock(struct rq *rq)
1822{
1823 lockdep_assert_rq_held(rq);
1824 assert_clock_updated(rq);
1825
1826 return rq->clock;
1827}
1828
1829static inline u64 rq_clock_task(struct rq *rq)
1830{
1831 lockdep_assert_rq_held(rq);
1832 assert_clock_updated(rq);
1833
1834 return rq->clock_task;
1835}
1836
1837static inline void rq_clock_skip_update(struct rq *rq)
1838{
1839 lockdep_assert_rq_held(rq);
1840 rq->clock_update_flags |= RQCF_REQ_SKIP;
1841}
1842
1843/*
1844 * See rt task throttling, which is the only time a skip
1845 * request is canceled.
1846 */
1847static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1848{
1849 lockdep_assert_rq_held(rq);
1850 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1851}
1852
1853/*
1854 * During cpu offlining and rq wide unthrottling, we can trigger
1855 * an update_rq_clock() for several cfs and rt runqueues (Typically
1856 * when using list_for_each_entry_*)
1857 * rq_clock_start_loop_update() can be called after updating the clock
1858 * once and before iterating over the list to prevent multiple update.
1859 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1860 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1861 */
1862static inline void rq_clock_start_loop_update(struct rq *rq)
1863{
1864 lockdep_assert_rq_held(rq);
1865 WARN_ON_ONCE(rq->clock_update_flags & RQCF_ACT_SKIP);
1866 rq->clock_update_flags |= RQCF_ACT_SKIP;
1867}
1868
1869static inline void rq_clock_stop_loop_update(struct rq *rq)
1870{
1871 lockdep_assert_rq_held(rq);
1872 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1873}
1874
1875struct rq_flags {
1876 unsigned long flags;
1877 struct pin_cookie cookie;
1878 /*
1879 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1880 * current pin context is stashed here in case it needs to be
1881 * restored in rq_repin_lock().
1882 */
1883 unsigned int clock_update_flags;
1884};
1885
1886extern struct balance_callback balance_push_callback;
1887
1888#ifdef CONFIG_SCHED_CLASS_EXT
1889extern const struct sched_class ext_sched_class;
1890
1891DECLARE_STATIC_KEY_FALSE(__scx_enabled); /* SCX BPF scheduler loaded */
1892DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
1893
1894#define scx_enabled() static_branch_unlikely(&__scx_enabled)
1895#define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
1896
1897static inline void scx_rq_clock_update(struct rq *rq, u64 clock)
1898{
1899 if (!scx_enabled())
1900 return;
1901 WRITE_ONCE(rq->scx.clock, clock);
1902 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID);
1903}
1904
1905static inline void scx_rq_clock_invalidate(struct rq *rq)
1906{
1907 if (!scx_enabled())
1908 return;
1909 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID);
1910}
1911
1912#else /* !CONFIG_SCHED_CLASS_EXT: */
1913#define scx_enabled() false
1914#define scx_switched_all() false
1915
1916static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {}
1917static inline void scx_rq_clock_invalidate(struct rq *rq) {}
1918#endif /* !CONFIG_SCHED_CLASS_EXT */
1919
1920static inline void assert_balance_callbacks_empty(struct rq *rq)
1921{
1922 WARN_ON_ONCE(IS_ENABLED(CONFIG_PROVE_LOCKING) &&
1923 rq->balance_callback &&
1924 rq->balance_callback != &balance_push_callback);
1925}
1926
1927/*
1928 * Lockdep annotation that avoids accidental unlocks; it's like a
1929 * sticky/continuous lockdep_assert_held().
1930 *
1931 * This avoids code that has access to 'struct rq *rq' (basically everything in
1932 * the scheduler) from accidentally unlocking the rq if they do not also have a
1933 * copy of the (on-stack) 'struct rq_flags rf'.
1934 *
1935 * Also see Documentation/locking/lockdep-design.rst.
1936 */
1937static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1938{
1939 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1940
1941 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1942 rf->clock_update_flags = 0;
1943 assert_balance_callbacks_empty(rq);
1944}
1945
1946static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1947{
1948 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1949 rf->clock_update_flags = RQCF_UPDATED;
1950
1951 scx_rq_clock_invalidate(rq);
1952 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1953}
1954
1955static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1956{
1957 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1958
1959 /*
1960 * Restore the value we stashed in @rf for this pin context.
1961 */
1962 rq->clock_update_flags |= rf->clock_update_flags;
1963}
1964
1965#define __task_rq_lock(...) __acquire_ret(___task_rq_lock(__VA_ARGS__), __rq_lockp(__ret))
1966extern struct rq *___task_rq_lock(struct task_struct *p, struct rq_flags *rf) __acquires_ret;
1967
1968#define task_rq_lock(...) __acquire_ret(_task_rq_lock(__VA_ARGS__), __rq_lockp(__ret))
1969extern struct rq *_task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1970 __acquires(&p->pi_lock) __acquires_ret;
1971
1972static inline void
1973__task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1974 __releases(__rq_lockp(rq))
1975{
1976 rq_unpin_lock(rq, rf);
1977 raw_spin_rq_unlock(rq);
1978}
1979
1980static inline void
1981task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1982 __releases(__rq_lockp(rq), &p->pi_lock)
1983{
1984 __task_rq_unlock(rq, p, rf);
1985 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1986}
1987
1988DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1989 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1990 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1991 struct rq *rq; struct rq_flags rf)
1992DECLARE_LOCK_GUARD_1_ATTRS(task_rq_lock, __acquires(_T->pi_lock), __releases((*(struct task_struct **)_T)->pi_lock))
1993#define class_task_rq_lock_constructor(_T) WITH_LOCK_GUARD_1_ATTRS(task_rq_lock, _T)
1994
1995DEFINE_LOCK_GUARD_1(__task_rq_lock, struct task_struct,
1996 _T->rq = __task_rq_lock(_T->lock, &_T->rf),
1997 __task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1998 struct rq *rq; struct rq_flags rf)
1999
2000static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
2001 __acquires(__rq_lockp(rq))
2002{
2003 raw_spin_rq_lock_irqsave(rq, rf->flags);
2004 rq_pin_lock(rq, rf);
2005}
2006
2007static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
2008 __acquires(__rq_lockp(rq))
2009{
2010 raw_spin_rq_lock_irq(rq);
2011 rq_pin_lock(rq, rf);
2012}
2013
2014static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
2015 __acquires(__rq_lockp(rq))
2016{
2017 raw_spin_rq_lock(rq);
2018 rq_pin_lock(rq, rf);
2019}
2020
2021static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
2022 __releases(__rq_lockp(rq))
2023{
2024 rq_unpin_lock(rq, rf);
2025 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
2026}
2027
2028static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
2029 __releases(__rq_lockp(rq))
2030{
2031 rq_unpin_lock(rq, rf);
2032 raw_spin_rq_unlock_irq(rq);
2033}
2034
2035static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
2036 __releases(__rq_lockp(rq))
2037{
2038 rq_unpin_lock(rq, rf);
2039 raw_spin_rq_unlock(rq);
2040}
2041
2042DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
2043 rq_lock(_T->lock, &_T->rf),
2044 rq_unlock(_T->lock, &_T->rf),
2045 struct rq_flags rf)
2046
2047DECLARE_LOCK_GUARD_1_ATTRS(rq_lock, __acquires(__rq_lockp(_T)),
2048 __releases(__rq_lockp(*(struct rq **)_T)));
2049#define class_rq_lock_constructor(_T) WITH_LOCK_GUARD_1_ATTRS(rq_lock, _T)
2050
2051DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
2052 rq_lock_irq(_T->lock, &_T->rf),
2053 rq_unlock_irq(_T->lock, &_T->rf),
2054 struct rq_flags rf)
2055
2056DECLARE_LOCK_GUARD_1_ATTRS(rq_lock_irq, __acquires(__rq_lockp(_T)),
2057 __releases(__rq_lockp(*(struct rq **)_T)));
2058#define class_rq_lock_irq_constructor(_T) WITH_LOCK_GUARD_1_ATTRS(rq_lock_irq, _T)
2059
2060DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
2061 rq_lock_irqsave(_T->lock, &_T->rf),
2062 rq_unlock_irqrestore(_T->lock, &_T->rf),
2063 struct rq_flags rf)
2064
2065DECLARE_LOCK_GUARD_1_ATTRS(rq_lock_irqsave, __acquires(__rq_lockp(_T)),
2066 __releases(__rq_lockp(*(struct rq **)_T)));
2067#define class_rq_lock_irqsave_constructor(_T) WITH_LOCK_GUARD_1_ATTRS(rq_lock_irqsave, _T)
2068
2069DEFINE_LOCK_GUARD_1(raw_spin_rq_lock_irqsave, struct rq,
2070 raw_spin_rq_lock_irqsave(_T->lock, _T->flags),
2071 raw_spin_rq_unlock_irqrestore(_T->lock, _T->flags),
2072 unsigned long flags)
2073
2074DECLARE_LOCK_GUARD_1_ATTRS(raw_spin_rq_lock_irqsave, __acquires(__rq_lockp(_T)),
2075 __releases(__rq_lockp(*(struct rq **)_T)));
2076#define class_raw_spin_rq_lock_irqsave_constructor(_T) \
2077 WITH_LOCK_GUARD_1_ATTRS(raw_spin_rq_lock_irqsave, _T)
2078
2079#define this_rq_lock_irq(...) __acquire_ret(_this_rq_lock_irq(__VA_ARGS__), __rq_lockp(__ret))
2080static inline struct rq *_this_rq_lock_irq(struct rq_flags *rf) __acquires_ret
2081{
2082 struct rq *rq;
2083
2084 local_irq_disable();
2085 rq = this_rq();
2086 rq_lock(rq, rf);
2087
2088 return rq;
2089}
2090
2091#ifdef CONFIG_NUMA
2092
2093enum numa_topology_type {
2094 NUMA_DIRECT,
2095 NUMA_GLUELESS_MESH,
2096 NUMA_BACKPLANE,
2097};
2098
2099extern enum numa_topology_type sched_numa_topology_type;
2100extern int sched_max_numa_distance;
2101extern bool find_numa_distance(int distance);
2102extern void sched_init_numa(int offline_node);
2103extern void sched_update_numa(int cpu, bool online);
2104extern void sched_domains_numa_masks_set(unsigned int cpu);
2105extern void sched_domains_numa_masks_clear(unsigned int cpu);
2106extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
2107
2108#else /* !CONFIG_NUMA: */
2109
2110static inline void sched_init_numa(int offline_node) { }
2111static inline void sched_update_numa(int cpu, bool online) { }
2112static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
2113static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
2114
2115static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
2116{
2117 return nr_cpu_ids;
2118}
2119
2120#endif /* !CONFIG_NUMA */
2121
2122#ifdef CONFIG_NUMA_BALANCING
2123
2124/* The regions in numa_faults array from task_struct */
2125enum numa_faults_stats {
2126 NUMA_MEM = 0,
2127 NUMA_CPU,
2128 NUMA_MEMBUF,
2129 NUMA_CPUBUF
2130};
2131
2132extern void sched_setnuma(struct task_struct *p, int node);
2133extern int migrate_task_to(struct task_struct *p, int cpu);
2134extern int migrate_swap(struct task_struct *p, struct task_struct *t,
2135 int cpu, int scpu);
2136extern void init_numa_balancing(u64 clone_flags, struct task_struct *p);
2137
2138#else /* !CONFIG_NUMA_BALANCING: */
2139
2140static inline void
2141init_numa_balancing(u64 clone_flags, struct task_struct *p)
2142{
2143}
2144
2145#endif /* !CONFIG_NUMA_BALANCING */
2146
2147int task_llc(const struct task_struct *p);
2148
2149static inline void
2150queue_balance_callback(struct rq *rq,
2151 struct balance_callback *head,
2152 void (*func)(struct rq *rq))
2153{
2154 lockdep_assert_rq_held(rq);
2155
2156 /*
2157 * Don't (re)queue an already queued item; nor queue anything when
2158 * balance_push() is active, see the comment with
2159 * balance_push_callback.
2160 */
2161 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
2162 return;
2163
2164 head->func = func;
2165 head->next = rq->balance_callback;
2166 rq->balance_callback = head;
2167}
2168
2169#define rcu_dereference_sched_domain(p) \
2170 rcu_dereference_all_check((p), lockdep_is_held(&sched_domains_mutex))
2171
2172/*
2173 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
2174 * See destroy_sched_domains: call_rcu for details.
2175 *
2176 * The domain tree of any CPU may only be accessed from within
2177 * preempt-disabled sections.
2178 */
2179#define for_each_domain(cpu, __sd) \
2180 for (__sd = rcu_dereference_sched_domain(cpu_rq(cpu)->sd); \
2181 __sd; __sd = __sd->parent)
2182
2183/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
2184#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
2185static const unsigned int SD_SHARED_CHILD_MASK =
2186#include <linux/sched/sd_flags.h>
21870;
2188#undef SD_FLAG
2189
2190/**
2191 * highest_flag_domain - Return highest sched_domain containing flag.
2192 * @cpu: The CPU whose highest level of sched domain is to
2193 * be returned.
2194 * @flag: The flag to check for the highest sched_domain
2195 * for the given CPU.
2196 *
2197 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
2198 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
2199 */
2200static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
2201{
2202 struct sched_domain *sd, *hsd = NULL;
2203
2204 for_each_domain(cpu, sd) {
2205 if (sd->flags & flag) {
2206 hsd = sd;
2207 continue;
2208 }
2209
2210 /*
2211 * Stop the search if @flag is known to be shared at lower
2212 * levels. It will not be found further up.
2213 */
2214 if (flag & SD_SHARED_CHILD_MASK)
2215 break;
2216 }
2217
2218 return hsd;
2219}
2220
2221static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2222{
2223 struct sched_domain *sd;
2224
2225 for_each_domain(cpu, sd) {
2226 if (sd->flags & flag)
2227 break;
2228 }
2229
2230 return sd;
2231}
2232
2233DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2234DECLARE_PER_CPU(int, sd_llc_size);
2235DECLARE_PER_CPU(int, sd_llc_id);
2236DECLARE_PER_CPU(int, sd_share_id);
2237DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2238DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_balance_shared);
2239DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2240DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2241DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2242
2243extern struct static_key_false sched_asym_cpucapacity;
2244extern struct static_key_false sched_cluster_active;
2245
2246static __always_inline bool sched_asym_cpucap_active(void)
2247{
2248 return static_branch_unlikely(&sched_asym_cpucapacity);
2249}
2250
2251struct sched_group_capacity {
2252 atomic_t ref;
2253 /*
2254 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2255 * for a single CPU.
2256 */
2257 unsigned long capacity;
2258 unsigned long min_capacity; /* Min per-CPU capacity in group */
2259 unsigned long max_capacity; /* Max per-CPU capacity in group */
2260 unsigned long next_update;
2261 int imbalance; /* XXX unrelated to capacity but shared group state */
2262
2263 int id;
2264
2265 unsigned long cpumask[]; /* Balance mask */
2266};
2267
2268struct sched_group {
2269 struct sched_group *next; /* Must be a circular list */
2270 atomic_t ref;
2271
2272 unsigned int group_weight;
2273 unsigned int cores;
2274 struct sched_group_capacity *sgc;
2275 int asym_prefer_cpu; /* CPU of highest priority in group */
2276 int flags;
2277
2278 /*
2279 * The CPUs this group covers.
2280 *
2281 * NOTE: this field is variable length. (Allocated dynamically
2282 * by attaching extra space to the end of the structure,
2283 * depending on how many CPUs the kernel has booted up with)
2284 */
2285 unsigned long cpumask[];
2286};
2287
2288static inline struct cpumask *sched_group_span(struct sched_group *sg)
2289{
2290 return to_cpumask(sg->cpumask);
2291}
2292
2293/*
2294 * See build_balance_mask().
2295 */
2296static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2297{
2298 return to_cpumask(sg->sgc->cpumask);
2299}
2300
2301extern int group_balance_cpu(struct sched_group *sg);
2302
2303extern void update_sched_domain_debugfs(void);
2304extern void dirty_sched_domain_sysctl(int cpu);
2305
2306extern int sched_update_scaling(void);
2307
2308static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2309{
2310 if (!p->user_cpus_ptr)
2311 return cpu_possible_mask; /* &init_task.cpus_mask */
2312 return p->user_cpus_ptr;
2313}
2314
2315#ifdef CONFIG_CGROUP_SCHED
2316
2317/*
2318 * Return the group to which this tasks belongs.
2319 *
2320 * We cannot use task_css() and friends because the cgroup subsystem
2321 * changes that value before the cgroup_subsys::attach() method is called,
2322 * therefore we cannot pin it and might observe the wrong value.
2323 *
2324 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2325 * core changes this before calling sched_move_task().
2326 *
2327 * Instead we use a 'copy' which is updated from sched_move_task() while
2328 * holding both task_struct::pi_lock and rq::lock.
2329 */
2330static inline struct task_group *task_group(struct task_struct *p)
2331{
2332 return p->sched_task_group;
2333}
2334
2335#ifdef CONFIG_FAIR_GROUP_SCHED
2336/*
2337 * Defined here to be available before stats.h is included, since
2338 * stats.h has dependencies on things defined later in this file.
2339 */
2340struct cfs_tg_state {
2341 struct cfs_rq cfs_rq;
2342 struct sched_entity se;
2343 struct sched_statistics stats;
2344} __no_randomize_layout;
2345
2346/* Access a specific CPU's cfs_rq from a task group */
2347static inline struct cfs_rq *tg_cfs_rq(struct task_group *tg, int cpu)
2348{
2349 return per_cpu_ptr(tg->cfs_rq, cpu);
2350}
2351
2352static inline struct sched_entity *tg_se(struct task_group *tg, int cpu)
2353{
2354 struct cfs_tg_state *state;
2355
2356 if (is_root_task_group(tg))
2357 return NULL;
2358
2359 state = container_of(tg_cfs_rq(tg, cpu), struct cfs_tg_state, cfs_rq);
2360 return &state->se;
2361}
2362
2363static inline struct sched_entity *cfs_rq_se(struct cfs_rq *cfs_rq)
2364{
2365 struct cfs_tg_state *state;
2366
2367 if (is_root_task_group(cfs_rq->tg))
2368 return NULL;
2369
2370 state = container_of(cfs_rq, struct cfs_tg_state, cfs_rq);
2371 return &state->se;
2372}
2373#endif
2374
2375/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2376static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2377{
2378#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2379 struct task_group *tg = task_group(p);
2380#endif
2381
2382#ifdef CONFIG_FAIR_GROUP_SCHED
2383 set_task_rq_fair(&p->se, p->se.cfs_rq, tg_cfs_rq(tg, cpu));
2384 p->se.cfs_rq = tg_cfs_rq(tg, cpu);
2385 p->se.parent = tg_se(tg, cpu);
2386 p->se.depth = p->se.parent ? p->se.parent->depth + 1 : 0;
2387#endif
2388
2389#ifdef CONFIG_RT_GROUP_SCHED
2390 /*
2391 * p->rt.rt_rq is NULL initially and it is easier to assign
2392 * root_task_group's rt_rq than switching in rt_rq_of_se()
2393 * Clobbers tg(!)
2394 */
2395 if (!rt_group_sched_enabled())
2396 tg = &root_task_group;
2397 p->rt.rt_rq = tg->rt_rq[cpu];
2398 p->rt.parent = tg->rt_se[cpu];
2399#endif /* CONFIG_RT_GROUP_SCHED */
2400}
2401
2402#else /* !CONFIG_CGROUP_SCHED: */
2403
2404static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2405
2406static inline struct task_group *task_group(struct task_struct *p)
2407{
2408 return NULL;
2409}
2410
2411#endif /* !CONFIG_CGROUP_SCHED */
2412
2413static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2414{
2415 set_task_rq(p, cpu);
2416#ifdef CONFIG_SMP
2417 /*
2418 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2419 * successfully executed on another CPU. We must ensure that updates of
2420 * per-task data have been completed by this moment.
2421 */
2422 smp_wmb();
2423 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2424 p->wake_cpu = cpu;
2425 rseq_sched_set_ids_changed(p);
2426#endif /* CONFIG_SMP */
2427}
2428
2429/*
2430 * Tunables:
2431 */
2432
2433#define SCHED_FEAT(name, enabled) \
2434 __SCHED_FEAT_##name ,
2435
2436enum {
2437#include "features.h"
2438 __SCHED_FEAT_NR,
2439};
2440
2441#undef SCHED_FEAT
2442
2443/*
2444 * To support run-time toggling of sched features, all the translation units
2445 * (but core.c) reference the sysctl_sched_features defined in core.c.
2446 */
2447extern __read_mostly unsigned int sysctl_sched_features;
2448
2449#ifdef CONFIG_JUMP_LABEL
2450
2451union sched_feat_key {
2452 struct static_key_true key_true;
2453 struct static_key_false key_false;
2454};
2455
2456#define sched_feat_branch_true(key) static_branch_likely(&(key)->key_true)
2457#define sched_feat_branch_false(key) static_branch_unlikely(&(key)->key_false)
2458
2459#define SCHED_FEAT(name, enabled) \
2460static __always_inline bool \
2461static_branch_##name(union sched_feat_key *key) \
2462{ \
2463 return sched_feat_branch_##enabled(key); \
2464}
2465
2466#include "features.h"
2467#undef SCHED_FEAT
2468
2469extern union sched_feat_key sched_feat_keys[__SCHED_FEAT_NR];
2470#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2471
2472#else /* !CONFIG_JUMP_LABEL: */
2473
2474#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2475
2476#endif /* !CONFIG_JUMP_LABEL */
2477
2478extern struct static_key_false sched_numa_balancing;
2479extern struct static_key_false sched_schedstats;
2480
2481static inline u64 global_rt_period(void)
2482{
2483 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2484}
2485
2486static inline u64 global_rt_runtime(void)
2487{
2488 if (sysctl_sched_rt_runtime < 0)
2489 return RUNTIME_INF;
2490
2491 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2492}
2493
2494/*
2495 * Is p the current execution context?
2496 */
2497static inline int task_current(struct rq *rq, struct task_struct *p)
2498{
2499 return rq->curr == p;
2500}
2501
2502/*
2503 * Is p the current scheduling context?
2504 *
2505 * Note that it might be the current execution context at the same time if
2506 * rq->curr == rq->donor == p.
2507 */
2508static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2509{
2510 return rq->donor == p;
2511}
2512
2513static inline bool task_is_blocked(struct task_struct *p)
2514{
2515 if (!sched_proxy_exec())
2516 return false;
2517
2518 return !!p->blocked_on;
2519}
2520
2521#ifdef CONFIG_SCHED_PROXY_EXEC
2522void sched_proxy_block_task(struct rq *rq, struct task_struct *p);
2523#else
2524static inline void sched_proxy_block_task(struct rq *rq, struct task_struct *p) {}
2525#endif
2526
2527static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2528{
2529 return p->on_cpu;
2530}
2531
2532static inline int task_on_rq_queued(struct task_struct *p)
2533{
2534 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED;
2535}
2536
2537static inline int task_on_rq_migrating(struct task_struct *p)
2538{
2539 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2540}
2541
2542/* Wake flags. The first three directly map to some SD flag value */
2543#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2544#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2545#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2546/*
2547 * Hint that the caller expects the waker to sleep soon.
2548 * Scheduler classes may use it for placement or preemption.
2549 * Callers must not rely on it to prevent migration,
2550 * preserve CPU locality or make the wakee run next.
2551 */
2552#define WF_SYNC 0x10
2553#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2554#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2555#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2556#define WF_TTWU_RQ 0x100 /* Wakeup completed through ttwu_runnable() */
2557
2558static_assert(WF_EXEC == SD_BALANCE_EXEC);
2559static_assert(WF_FORK == SD_BALANCE_FORK);
2560static_assert(WF_TTWU == SD_BALANCE_WAKE);
2561
2562/*
2563 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2564 * of tasks with abnormal "nice" values across CPUs the contribution that
2565 * each task makes to its run queue's load is weighted according to its
2566 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2567 * scaled version of the new time slice allocation that they receive on time
2568 * slice expiry etc.
2569 */
2570
2571#define WEIGHT_IDLEPRIO 3
2572#define WMULT_IDLEPRIO 1431655765
2573
2574extern const int sched_prio_to_weight[40];
2575extern const u32 sched_prio_to_wmult[40];
2576
2577/*
2578 * {de,en}queue flags:
2579 *
2580 * SLEEP/WAKEUP - task is no-longer/just-became runnable
2581 *
2582 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2583 * are in a known state which allows modification. Such pairs
2584 * should preserve as much state as possible.
2585 *
2586 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2587 * in the runqueue. IOW the priority is allowed to change. Callers
2588 * must expect to deal with balance callbacks.
2589 *
2590 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2591 *
2592 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2593 *
2594 * DELAYED - de/re-queue a sched_delayed task
2595 *
2596 * CLASS - going to update p->sched_class; makes sched_change call the
2597 * various switch methods.
2598 *
2599 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2600 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2601 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2602 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2603 *
2604 * XXX SAVE/RESTORE in combination with CLASS doesn't really make sense, but
2605 * SCHED_DEADLINE seems to rely on this for now.
2606 */
2607
2608#define DEQUEUE_SLEEP 0x0001 /* Matches ENQUEUE_WAKEUP */
2609#define DEQUEUE_SAVE 0x0002 /* Matches ENQUEUE_RESTORE */
2610#define DEQUEUE_MOVE 0x0004 /* Matches ENQUEUE_MOVE */
2611#define DEQUEUE_NOCLOCK 0x0008 /* Matches ENQUEUE_NOCLOCK */
2612
2613#define DEQUEUE_MIGRATING 0x0010 /* Matches ENQUEUE_MIGRATING */
2614#define DEQUEUE_DELAYED 0x0020 /* Matches ENQUEUE_DELAYED */
2615#define DEQUEUE_CLASS 0x0040 /* Matches ENQUEUE_CLASS */
2616
2617#define DEQUEUE_SPECIAL 0x00010000
2618#define DEQUEUE_THROTTLE 0x00020000
2619
2620#define ENQUEUE_WAKEUP 0x0001
2621#define ENQUEUE_RESTORE 0x0002
2622#define ENQUEUE_MOVE 0x0004
2623#define ENQUEUE_NOCLOCK 0x0008
2624
2625#define ENQUEUE_MIGRATING 0x0010
2626#define ENQUEUE_DELAYED 0x0020
2627#define ENQUEUE_CLASS 0x0040
2628
2629#define ENQUEUE_HEAD 0x00010000
2630#define ENQUEUE_REPLENISH 0x00020000
2631#define ENQUEUE_MIGRATED 0x00040000
2632#define ENQUEUE_INITIAL 0x00080000
2633#define ENQUEUE_RQ_SELECTED 0x00100000
2634#define ENQUEUE_QUEUED 0x00200000
2635
2636#define RETRY_TASK ((void *)-1UL)
2637
2638struct affinity_context {
2639 const struct cpumask *new_mask;
2640 struct cpumask *user_mask;
2641 unsigned int flags;
2642};
2643
2644extern s64 update_curr_common(struct rq *rq);
2645
2646enum snt_e {
2647 SNT_NORMAL, /* set_next_task() */
2648 SNT_PICK, /* put_prev_set_next_task(): prev != next */
2649 SNT_REPICK, /* put_prev_set_next_task(): prev == next */
2650};
2651
2652struct sched_class {
2653
2654#ifdef CONFIG_UCLAMP_TASK
2655 int uclamp_enabled;
2656#endif
2657
2658 /*
2659 * move_queued_task/activate_task/enqueue_task: rq->lock
2660 * ttwu_do_activate/activate_task/enqueue_task: rq->lock
2661 * wake_up_new_task/activate_task/enqueue_task: task_rq_lock
2662 * ttwu_runnable/enqueue_task: task_rq_lock
2663 * proxy_task_current: rq->lock
2664 * sched_change_end
2665 */
2666 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2667 /*
2668 * move_queued_task/deactivate_task/dequeue_task: rq->lock
2669 * __schedule/block_task/dequeue_task: rq->lock
2670 * proxy_task_current: rq->lock
2671 * wait_task_inactive: task_rq_lock
2672 * sched_change_begin
2673 */
2674 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2675
2676 /*
2677 * do_sched_yield: rq->lock
2678 */
2679 void (*yield_task) (struct rq *rq);
2680 /*
2681 * yield_to: rq->lock (double)
2682 */
2683 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2684
2685 /*
2686 * move_queued_task: rq->lock
2687 * __migrate_swap_task: rq->lock
2688 * ttwu_do_activate: rq->lock
2689 * ttwu_runnable: task_rq_lock
2690 * wake_up_new_task: task_rq_lock
2691 */
2692 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2693
2694 /*
2695 * schedule/pick_next_task/prev_balance: rq->lock
2696 */
2697 int (*balance)(struct rq *rq, struct rq_flags *rf);
2698
2699 /*
2700 * schedule/pick_next_task: rq->lock
2701 */
2702 struct task_struct *(*pick_task)(struct rq *rq, struct rq_flags *rf);
2703
2704 /*
2705 * sched_change:
2706 * __schedule: rq->lock
2707 */
2708 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2709 void (*set_next_task)(struct rq *rq, struct task_struct *p, enum snt_e type);
2710
2711 /*
2712 * select_task_rq: p->pi_lock
2713 * sched_exec: p->pi_lock
2714 */
2715 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2716
2717 /*
2718 * set_task_cpu: p->pi_lock || rq->lock (ttwu like)
2719 */
2720 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2721
2722 /*
2723 * ttwu_do_activate: rq->lock
2724 * wake_up_new_task: task_rq_lock
2725 */
2726 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2727
2728 /*
2729 * do_set_cpus_allowed: task_rq_lock + sched_change
2730 */
2731 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2732
2733 /*
2734 * sched_set_rq_{on,off}line: rq->lock
2735 */
2736 void (*rq_online)(struct rq *rq);
2737 void (*rq_offline)(struct rq *rq);
2738
2739 /*
2740 * push_cpu_stop: p->pi_lock && rq->lock
2741 */
2742 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2743
2744 /*
2745 * hrtick: rq->lock
2746 * sched_tick: rq->lock
2747 * sched_tick_remote: rq->lock
2748 */
2749 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2750 /*
2751 * sched_cgroup_fork: p->pi_lock
2752 */
2753 void (*task_fork)(struct task_struct *p);
2754 /*
2755 * finish_task_switch: no locks
2756 */
2757 void (*task_dead)(struct task_struct *p);
2758
2759 /*
2760 * sched_change
2761 */
2762 void (*switching_from)(struct rq *this_rq, struct task_struct *task);
2763 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
2764 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2765 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2766 u64 (*get_prio) (struct rq *this_rq, struct task_struct *task);
2767 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2768 u64 oldprio);
2769
2770 /*
2771 * set_load_weight: task_rq_lock + sched_change
2772 * __setscheduler_parms: task_rq_lock + sched_change
2773 */
2774 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2775 const struct load_weight *lw);
2776
2777 /*
2778 * sched_rr_get_interval: task_rq_lock
2779 */
2780 unsigned int (*get_rr_interval)(struct rq *rq,
2781 struct task_struct *task);
2782
2783 /*
2784 * task_sched_runtime: task_rq_lock
2785 */
2786 void (*update_curr)(struct rq *rq);
2787
2788#ifdef CONFIG_FAIR_GROUP_SCHED
2789 /*
2790 * sched_change_group: task_rq_lock + sched_change
2791 */
2792 void (*task_change_group)(struct task_struct *p);
2793#endif
2794
2795#ifdef CONFIG_SCHED_CORE
2796 /*
2797 * pick_next_task: rq->lock
2798 * try_steal_cookie: rq->lock (double)
2799 */
2800 int (*task_is_throttled)(struct task_struct *p, int cpu);
2801#endif
2802};
2803
2804static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2805{
2806 WARN_ON_ONCE(rq->donor != prev);
2807 prev->sched_class->put_prev_task(rq, prev, NULL);
2808}
2809
2810static inline void set_next_task(struct rq *rq, struct task_struct *next)
2811{
2812 next->sched_class->set_next_task(rq, next, SNT_NORMAL);
2813}
2814
2815static inline void
2816__put_prev_set_next_dl_server(struct rq *rq,
2817 struct task_struct *prev,
2818 struct task_struct *next)
2819{
2820 prev->dl_server = NULL;
2821 next->dl_server = rq->dl_server;
2822 rq->dl_server = NULL;
2823}
2824
2825static inline void put_prev_set_next_task(struct rq *rq,
2826 struct task_struct *prev,
2827 struct task_struct *next)
2828{
2829 WARN_ON_ONCE(rq->donor != prev);
2830
2831 __put_prev_set_next_dl_server(rq, prev, next);
2832
2833 if (next == prev) {
2834 next->sched_class->set_next_task(rq, next, SNT_REPICK);
2835 return;
2836 }
2837
2838 prev->sched_class->put_prev_task(rq, prev, next);
2839 next->sched_class->set_next_task(rq, next, SNT_PICK);
2840}
2841
2842/*
2843 * Helper to define a sched_class instance; each one is placed in a separate
2844 * section which is ordered by the linker script:
2845 *
2846 * include/asm-generic/vmlinux.lds.h
2847 *
2848 * *CAREFUL* they are laid out in *REVERSE* order!!!
2849 *
2850 * Also enforce alignment on the instance, not the type, to guarantee layout.
2851 */
2852#define DEFINE_SCHED_CLASS(name) \
2853const struct sched_class name##_sched_class \
2854 __aligned(__alignof__(struct sched_class)) \
2855 __section("__" #name "_sched_class")
2856
2857/* Defined in include/asm-generic/vmlinux.lds.h */
2858extern struct sched_class __sched_class_highest[];
2859extern struct sched_class __sched_class_lowest[];
2860
2861extern const struct sched_class stop_sched_class;
2862extern const struct sched_class dl_sched_class;
2863extern const struct sched_class rt_sched_class;
2864extern const struct sched_class fair_sched_class;
2865extern const struct sched_class idle_sched_class;
2866
2867/*
2868 * Iterate only active classes. SCX can take over all fair tasks or be
2869 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2870 */
2871static inline const struct sched_class *next_active_class(const struct sched_class *class)
2872{
2873 class++;
2874#ifdef CONFIG_SCHED_CLASS_EXT
2875 if (scx_switched_all() && class == &fair_sched_class)
2876 class++;
2877 if (!scx_enabled() && class == &ext_sched_class)
2878 class++;
2879#endif
2880 return class;
2881}
2882
2883#define for_class_range(class, _from, _to) \
2884 for (class = (_from); class < (_to); class++)
2885
2886#define for_each_class(class) \
2887 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2888
2889#define for_active_class_range(class, _from, _to) \
2890 for (class = (_from); class != (_to); class = next_active_class(class))
2891
2892#define for_each_active_class(class) \
2893 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2894
2895#define sched_class_above(_a, _b) ((_a) < (_b))
2896
2897static inline void rq_modified_begin(struct rq *rq, const struct sched_class *class)
2898{
2899 if (sched_class_above(rq->next_class, class))
2900 rq->next_class = class;
2901}
2902
2903static inline bool rq_modified_above(struct rq *rq, const struct sched_class *class)
2904{
2905 return sched_class_above(rq->next_class, class);
2906}
2907
2908static inline bool sched_stop_runnable(struct rq *rq)
2909{
2910 return rq->stop && task_on_rq_queued(rq->stop);
2911}
2912
2913static inline bool sched_dl_runnable(struct rq *rq)
2914{
2915 return rq->dl.dl_nr_running > 0;
2916}
2917
2918static inline bool sched_rt_runnable(struct rq *rq)
2919{
2920 return rq->rt.rt_queued > 0;
2921}
2922
2923static inline bool sched_fair_runnable(struct rq *rq)
2924{
2925 return rq->cfs.nr_queued > 0;
2926}
2927
2928extern struct task_struct *pick_task_fair(struct rq *rq, struct rq_flags *rf);
2929extern struct task_struct *pick_task_idle(struct rq *rq, struct rq_flags *rf);
2930
2931#define SCA_CHECK 0x01
2932#define SCA_MIGRATE_DISABLE 0x02
2933#define SCA_MIGRATE_ENABLE 0x04
2934#define SCA_USER 0x08
2935
2936extern void update_group_capacity(struct sched_domain *sd, int cpu);
2937
2938extern void sched_balance_trigger(struct rq *rq);
2939
2940extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2941extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2942
2943static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2944{
2945 /* When not in the task's cpumask, no point in looking further. */
2946 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2947 return false;
2948
2949 /* Can @cpu run a user thread? */
2950 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2951 return false;
2952
2953 return true;
2954}
2955
2956static inline cpumask_t *alloc_user_cpus_ptr(int node)
2957{
2958 /*
2959 * See set_cpus_allowed_force() above for the rcu_head usage.
2960 */
2961 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2962
2963 return kmalloc_node(size, GFP_KERNEL, node);
2964}
2965
2966static inline struct task_struct *get_push_task(struct rq *rq)
2967{
2968 struct task_struct *p = rq->donor;
2969
2970 lockdep_assert_rq_held(rq);
2971
2972 if (rq->push_busy)
2973 return NULL;
2974
2975 if (p->nr_cpus_allowed == 1)
2976 return NULL;
2977
2978 if (p->migration_disabled)
2979 return NULL;
2980
2981 rq->push_busy = true;
2982 return get_task_struct(p);
2983}
2984
2985extern int push_cpu_stop(void *arg);
2986
2987#ifdef CONFIG_CPU_IDLE
2988
2989static inline void idle_set_state(struct rq *rq,
2990 struct cpuidle_state *idle_state)
2991{
2992 rq->idle_state = idle_state;
2993}
2994
2995static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2996{
2997 lockdep_assert(rcu_read_lock_any_held());
2998
2999 return rq->idle_state;
3000}
3001
3002#else /* !CONFIG_CPU_IDLE: */
3003
3004static inline void idle_set_state(struct rq *rq,
3005 struct cpuidle_state *idle_state)
3006{
3007}
3008
3009static inline struct cpuidle_state *idle_get_state(struct rq *rq)
3010{
3011 return NULL;
3012}
3013
3014#endif /* !CONFIG_CPU_IDLE */
3015
3016extern void schedule_idle(void);
3017asmlinkage void schedule_user(void);
3018
3019extern void sysrq_sched_debug_show(void);
3020extern void sched_init_granularity(void);
3021extern void update_max_interval(void);
3022
3023extern void init_sched_dl_class(void);
3024extern void init_sched_rt_class(void);
3025extern void init_sched_fair_class(void);
3026
3027extern void resched_curr(struct rq *rq);
3028extern void resched_curr_lazy(struct rq *rq);
3029extern void resched_cpu(int cpu);
3030
3031extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
3032extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
3033
3034extern void init_dl_entity(struct sched_dl_entity *dl_se);
3035
3036extern void init_cfs_throttle_work(struct task_struct *p);
3037
3038#define BW_SHIFT 20
3039#define BW_UNIT (1 << BW_SHIFT)
3040#define RATIO_SHIFT 8
3041#define MAX_BW_BITS (64 - BW_SHIFT)
3042#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
3043
3044extern u64 to_ratio(u64 period, u64 runtime);
3045
3046extern void init_entity_runnable_average(struct sched_entity *se);
3047extern void post_init_entity_util_avg(struct task_struct *p);
3048
3049#ifdef CONFIG_NO_HZ_FULL
3050extern bool sched_can_stop_tick(struct rq *rq);
3051extern int __init sched_tick_offload_init(void);
3052
3053/*
3054 * Tick may be needed by tasks in the runqueue depending on their policy and
3055 * requirements. If tick is needed, lets send the target an IPI to kick it out of
3056 * nohz mode if necessary.
3057 */
3058static inline void sched_update_tick_dependency(struct rq *rq)
3059{
3060 int cpu = cpu_of(rq);
3061
3062 if (!tick_nohz_full_cpu(cpu))
3063 return;
3064
3065 if (sched_can_stop_tick(rq))
3066 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
3067 else
3068 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
3069}
3070#else /* !CONFIG_NO_HZ_FULL: */
3071static inline int sched_tick_offload_init(void) { return 0; }
3072static inline void sched_update_tick_dependency(struct rq *rq) { }
3073#endif /* !CONFIG_NO_HZ_FULL */
3074
3075static inline void add_nr_running(struct rq *rq, unsigned count)
3076{
3077 unsigned prev_nr = rq->nr_running;
3078
3079 rq->nr_running = prev_nr + count;
3080 if (trace_sched_update_nr_running_tp_enabled()) {
3081 call_trace_sched_update_nr_running(rq, count);
3082 }
3083
3084 if (prev_nr < 2 && rq->nr_running >= 2)
3085 set_rd_overloaded(rq->rd, 1);
3086
3087 sched_update_tick_dependency(rq);
3088}
3089
3090static inline void sub_nr_running(struct rq *rq, unsigned count)
3091{
3092 rq->nr_running -= count;
3093 if (trace_sched_update_nr_running_tp_enabled()) {
3094 call_trace_sched_update_nr_running(rq, -count);
3095 }
3096
3097 /* Check if we still need preemption */
3098 sched_update_tick_dependency(rq);
3099}
3100
3101static inline void __block_task(struct rq *rq, struct task_struct *p)
3102{
3103 if (p->sched_contributes_to_load)
3104 rq->nr_uninterruptible++;
3105
3106 if (p->in_iowait) {
3107 atomic_inc(&rq->nr_iowait);
3108 delayacct_blkio_start();
3109 }
3110
3111 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
3112
3113 /*
3114 * The moment this write goes through, ttwu() can swoop in and migrate
3115 * this task, rendering our rq->__lock ineffective.
3116 *
3117 * __schedule() try_to_wake_up()
3118 * LOCK rq->__lock LOCK p->pi_lock
3119 * pick_next_task()
3120 * pick_next_task_fair()
3121 * pick_next_entity()
3122 * dequeue_entities()
3123 * __block_task()
3124 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
3125 * break;
3126 *
3127 * ACQUIRE (after ctrl-dep)
3128 *
3129 * cpu = select_task_rq();
3130 * set_task_cpu(p, cpu);
3131 * ttwu_queue()
3132 * ttwu_do_activate()
3133 * LOCK rq->__lock
3134 * activate_task()
3135 * STORE p->on_rq = 1
3136 * UNLOCK rq->__lock
3137 *
3138 * Callers must ensure to not reference @p after this -- we no longer
3139 * own it.
3140 */
3141 smp_store_release(&p->on_rq, 0);
3142}
3143
3144extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
3145extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
3146
3147extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
3148
3149/*
3150 * attach_task() -- attach the task detached by detach_task() to its new rq.
3151 */
3152static inline void attach_task(struct rq *rq, struct task_struct *p)
3153{
3154 lockdep_assert_rq_held(rq);
3155
3156 WARN_ON_ONCE(task_rq(p) != rq);
3157 activate_task(rq, p, ENQUEUE_NOCLOCK);
3158 wakeup_preempt(rq, p, 0);
3159}
3160
3161/*
3162 * attach_one_task() -- attaches the task returned from detach_one_task() to
3163 * its new rq.
3164 */
3165static inline void attach_one_task(struct rq *rq, struct task_struct *p)
3166{
3167 guard(rq_lock)(rq);
3168 update_rq_clock(rq);
3169 attach_task(rq, p);
3170}
3171
3172/*
3173 * __attach_tasks() - attaches a list of tasks (using se.group_node) to
3174 * the new rq
3175 */
3176static inline void __attach_tasks(struct rq *rq, struct list_head *tasks)
3177{
3178 guard(rq_lock)(rq);
3179 update_rq_clock(rq);
3180
3181 while (!list_empty(tasks)) {
3182 struct task_struct *p;
3183
3184 p = list_first_entry(tasks, struct task_struct, se.group_node);
3185 list_del_init(&p->se.group_node);
3186
3187 attach_task(rq, p);
3188 }
3189}
3190
3191#ifdef CONFIG_PREEMPT_RT
3192# define SCHED_NR_MIGRATE_BREAK 8
3193#else
3194# define SCHED_NR_MIGRATE_BREAK 32
3195#endif
3196
3197extern __read_mostly unsigned int sysctl_sched_nr_migrate;
3198extern __read_mostly unsigned int sysctl_sched_migration_cost;
3199
3200extern unsigned int sysctl_sched_base_slice;
3201
3202extern int sysctl_resched_latency_warn_ms;
3203extern int sysctl_resched_latency_warn_once;
3204
3205extern unsigned int sysctl_sched_tunable_scaling;
3206
3207extern unsigned int sysctl_numa_balancing_scan_delay;
3208extern unsigned int sysctl_numa_balancing_scan_period_min;
3209extern unsigned int sysctl_numa_balancing_scan_period_max;
3210extern unsigned int sysctl_numa_balancing_scan_size;
3211extern unsigned int sysctl_numa_balancing_hot_threshold;
3212
3213#ifdef CONFIG_SCHED_HRTICK
3214
3215/*
3216 * Use hrtick when:
3217 * - enabled by features
3218 * - hrtimer is actually high res
3219 */
3220static inline bool hrtick_enabled(struct rq *rq)
3221{
3222 return cpu_active(cpu_of(rq)) && hrtimer_highres_enabled();
3223}
3224
3225static inline bool hrtick_enabled_fair(struct rq *rq)
3226{
3227 return sched_feat(HRTICK) && hrtick_enabled(rq);
3228}
3229
3230static inline bool hrtick_enabled_dl(struct rq *rq)
3231{
3232 return sched_feat(HRTICK_DL) && hrtick_enabled(rq);
3233}
3234
3235extern void hrtick_start(struct rq *rq, u64 delay);
3236static inline bool hrtick_active(struct rq *rq)
3237{
3238 return hrtimer_active(&rq->hrtick_timer);
3239}
3240
3241#else /* !CONFIG_SCHED_HRTICK: */
3242static inline bool hrtick_enabled_fair(struct rq *rq) { return false; }
3243static inline bool hrtick_enabled_dl(struct rq *rq) { return false; }
3244static inline bool hrtick_enabled(struct rq *rq) { return false; }
3245#endif /* !CONFIG_SCHED_HRTICK */
3246
3247#ifndef arch_scale_freq_tick
3248static __always_inline void arch_scale_freq_tick(void) { }
3249#endif
3250
3251#ifndef arch_scale_freq_capacity
3252/**
3253 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
3254 * @cpu: the CPU in question.
3255 *
3256 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
3257 *
3258 * f_curr
3259 * ------ * SCHED_CAPACITY_SCALE
3260 * f_max
3261 */
3262static __always_inline
3263unsigned long arch_scale_freq_capacity(int cpu)
3264{
3265 return SCHED_CAPACITY_SCALE;
3266}
3267#endif
3268
3269/*
3270 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
3271 * acquire rq lock instead of rq_lock(). So at the end of these two functions
3272 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
3273 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
3274 */
3275static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
3276{
3277 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
3278 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
3279}
3280
3281#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
3282__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
3283static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
3284 __no_context_analysis \
3285{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
3286 _lock; return _t; }
3287#define DECLARE_LOCK_GUARD_2_ATTRS(_name, _lock, _unlock1, _unlock2) \
3288static inline class_##_name##_t class_##_name##_constructor(lock_##_name##_t *_T1, \
3289 lock_##_name##_t *_T2) _lock; \
3290static __always_inline void __class_##_name##_cleanup_ctx1(class_##_name##_t **_T1) \
3291 __no_context_analysis _unlock1 { } \
3292static __always_inline void __class_##_name##_cleanup_ctx2(class_##_name##_t **_T2) \
3293 __no_context_analysis _unlock2 { }
3294#define WITH_LOCK_GUARD_2_ATTRS(_name, _T1, _T2) \
3295 class_##_name##_constructor(_T1, _T2), \
3296 *__UNIQUE_ID(unlock1) __cleanup(__class_##_name##_cleanup_ctx1) = (void *)(_T1),\
3297 *__UNIQUE_ID(unlock2) __cleanup(__class_##_name##_cleanup_ctx2) = (void *)(_T2)
3298
3299static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
3300{
3301#ifdef CONFIG_SCHED_CORE
3302 /*
3303 * In order to not have {0,2},{1,3} turn into into an AB-BA,
3304 * order by core-id first and cpu-id second.
3305 *
3306 * Notably:
3307 *
3308 * double_rq_lock(0,3); will take core-0, core-1 lock
3309 * double_rq_lock(1,2); will take core-1, core-0 lock
3310 *
3311 * when only cpu-id is considered.
3312 */
3313 if (rq1->core->cpu < rq2->core->cpu)
3314 return true;
3315 if (rq1->core->cpu > rq2->core->cpu)
3316 return false;
3317
3318 /*
3319 * __sched_core_flip() relies on SMT having cpu-id lock order.
3320 */
3321#endif /* CONFIG_SCHED_CORE */
3322 return rq1->cpu < rq2->cpu;
3323}
3324
3325extern void double_rq_lock(struct rq *rq1, struct rq *rq2)
3326 __acquires(__rq_lockp(rq1), __rq_lockp(rq2));
3327
3328#ifdef CONFIG_PREEMPTION
3329
3330/*
3331 * fair double_lock_balance: Safely acquires both rq->locks in a fair
3332 * way at the expense of forcing extra atomic operations in all
3333 * invocations. This assures that the double_lock is acquired using the
3334 * same underlying policy as the spinlock_t on this architecture, which
3335 * reduces latency compared to the unfair variant below. However, it
3336 * also adds more overhead and therefore may reduce throughput.
3337 */
3338static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3339 __must_hold(__rq_lockp(this_rq))
3340 __acquires(__rq_lockp(busiest))
3341{
3342 raw_spin_rq_unlock(this_rq);
3343 double_rq_lock(this_rq, busiest);
3344
3345 return 1;
3346}
3347
3348#else /* !CONFIG_PREEMPTION: */
3349/*
3350 * Unfair double_lock_balance: Optimizes throughput at the expense of
3351 * latency by eliminating extra atomic operations when the locks are
3352 * already in proper order on entry. This favors lower CPU-ids and will
3353 * grant the double lock to lower CPUs over higher ids under contention,
3354 * regardless of entry order into the function.
3355 */
3356static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3357 __must_hold(__rq_lockp(this_rq))
3358 __acquires(__rq_lockp(busiest))
3359{
3360 if (__rq_lockp(this_rq) == __rq_lockp(busiest)) {
3361 __acquire(__rq_lockp(busiest)); /* already held */
3362 double_rq_clock_clear_update(this_rq, busiest);
3363 return 0;
3364 }
3365
3366 if (likely(raw_spin_rq_trylock(busiest))) {
3367 double_rq_clock_clear_update(this_rq, busiest);
3368 return 0;
3369 }
3370
3371 if (rq_order_less(this_rq, busiest)) {
3372 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3373 double_rq_clock_clear_update(this_rq, busiest);
3374 return 0;
3375 }
3376
3377 raw_spin_rq_unlock(this_rq);
3378 double_rq_lock(this_rq, busiest);
3379
3380 return 1;
3381}
3382
3383#endif /* !CONFIG_PREEMPTION */
3384
3385/*
3386 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3387 */
3388static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3389 __must_hold(__rq_lockp(this_rq))
3390 __acquires(__rq_lockp(busiest))
3391{
3392 lockdep_assert_irqs_disabled();
3393
3394 return _double_lock_balance(this_rq, busiest);
3395}
3396
3397static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3398 __releases(__rq_lockp(busiest))
3399{
3400 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
3401 raw_spin_rq_unlock(busiest);
3402 else
3403 __release(__rq_lockp(busiest)); /* fake release */
3404 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3405}
3406
3407static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3408 __acquires(l1, l2)
3409{
3410 if (l1 > l2)
3411 swap(l1, l2);
3412
3413 spin_lock(l1);
3414 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3415}
3416
3417static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3418 __acquires(l1, l2)
3419{
3420 if (l1 > l2)
3421 swap(l1, l2);
3422
3423 spin_lock_irq(l1);
3424 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3425}
3426
3427static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3428 __acquires(l1, l2)
3429{
3430 if (l1 > l2)
3431 swap(l1, l2);
3432
3433 raw_spin_lock(l1);
3434 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3435}
3436
3437static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3438 __releases(l1, l2)
3439{
3440 raw_spin_unlock(l1);
3441 raw_spin_unlock(l2);
3442}
3443
3444DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3445 double_raw_lock(_T->lock, _T->lock2),
3446 double_raw_unlock(_T->lock, _T->lock2))
3447
3448DECLARE_LOCK_GUARD_2_ATTRS(double_raw_spinlock,
3449 __acquires(_T1, _T2),
3450 __releases(*(raw_spinlock_t **)_T1),
3451 __releases(*(raw_spinlock_t **)_T2));
3452#define class_double_raw_spinlock_constructor(_T1, _T2) \
3453 WITH_LOCK_GUARD_2_ATTRS(double_raw_spinlock, _T1, _T2)
3454
3455/*
3456 * double_rq_unlock - safely unlock two runqueues
3457 *
3458 * Note this does not restore interrupts like task_rq_unlock,
3459 * you need to do so manually after calling.
3460 */
3461static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3462 __releases(__rq_lockp(rq1), __rq_lockp(rq2))
3463{
3464 if (__rq_lockp(rq1) != __rq_lockp(rq2))
3465 raw_spin_rq_unlock(rq2);
3466 else
3467 __release(__rq_lockp(rq2)); /* fake release */
3468 raw_spin_rq_unlock(rq1);
3469}
3470
3471extern void set_rq_online (struct rq *rq);
3472extern void set_rq_offline(struct rq *rq);
3473
3474extern bool sched_smp_initialized;
3475
3476DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3477 double_rq_lock(_T->lock, _T->lock2),
3478 double_rq_unlock(_T->lock, _T->lock2))
3479
3480extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3481extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3482extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3483
3484extern bool sched_debug_verbose;
3485
3486extern void print_cfs_stats(struct seq_file *m, int cpu);
3487extern void print_rt_stats(struct seq_file *m, int cpu);
3488extern void print_dl_stats(struct seq_file *m, int cpu);
3489extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3490extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3491extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3492
3493extern void resched_latency_warn(int cpu, u64 latency);
3494
3495#ifdef CONFIG_NUMA_BALANCING
3496extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3497extern void
3498print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3499 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3500#endif /* CONFIG_NUMA_BALANCING */
3501
3502extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3503extern void init_rt_rq(struct rt_rq *rt_rq);
3504extern void init_dl_rq(struct dl_rq *dl_rq);
3505
3506extern void cfs_bandwidth_usage_inc(void);
3507extern void cfs_bandwidth_usage_dec(void);
3508
3509#ifdef CONFIG_NO_HZ_COMMON
3510
3511#define NOHZ_BALANCE_KICK_BIT 0
3512#define NOHZ_STATS_KICK_BIT 1
3513#define NOHZ_NEWILB_KICK_BIT 2
3514#define NOHZ_NEXT_KICK_BIT 3
3515
3516/* Run sched_balance_domains() */
3517#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3518/* Update blocked load */
3519#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3520/* Update blocked load when entering idle */
3521#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3522/* Update nohz.next_balance */
3523#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3524
3525#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3526
3527#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3528
3529extern void nohz_balance_exit_idle(struct rq *rq);
3530#else /* !CONFIG_NO_HZ_COMMON: */
3531static inline void nohz_balance_exit_idle(struct rq *rq) { }
3532#endif /* !CONFIG_NO_HZ_COMMON */
3533
3534#ifdef CONFIG_NO_HZ_COMMON
3535extern void nohz_run_idle_balance(int cpu);
3536#else
3537static inline void nohz_run_idle_balance(int cpu) { }
3538#endif
3539
3540#include "stats.h"
3541
3542#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3543
3544extern void __sched_core_account_forceidle(struct rq *rq);
3545
3546static inline void sched_core_account_forceidle(struct rq *rq)
3547{
3548 if (schedstat_enabled())
3549 __sched_core_account_forceidle(rq);
3550}
3551
3552extern void __sched_core_tick(struct rq *rq);
3553
3554static inline void sched_core_tick(struct rq *rq)
3555{
3556 if (sched_core_enabled(rq) && schedstat_enabled())
3557 __sched_core_tick(rq);
3558}
3559
3560#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3561
3562static inline void sched_core_account_forceidle(struct rq *rq) { }
3563
3564static inline void sched_core_tick(struct rq *rq) { }
3565
3566#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3567
3568#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3569
3570struct irqtime {
3571 u64 total;
3572 u64 tick_delta;
3573 u64 irq_start_time;
3574 struct u64_stats_sync sync;
3575};
3576
3577DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3578DECLARE_STATIC_KEY_FALSE(sched_clock_irqtime);
3579
3580static inline int irqtime_enabled(void)
3581{
3582 return static_branch_likely(&sched_clock_irqtime);
3583}
3584
3585/*
3586 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3587 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3588 * and never move forward.
3589 */
3590static inline u64 irq_time_read(int cpu)
3591{
3592 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3593 unsigned int seq;
3594 u64 total;
3595
3596 do {
3597 seq = __u64_stats_fetch_begin(&irqtime->sync);
3598 total = irqtime->total;
3599 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3600
3601 return total;
3602}
3603
3604#else /* !CONFIG_IRQ_TIME_ACCOUNTING: */
3605
3606static inline int irqtime_enabled(void)
3607{
3608 return 0;
3609}
3610
3611#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
3612
3613#ifdef CONFIG_CPU_FREQ
3614
3615DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3616
3617/**
3618 * cpufreq_update_util - Take a note about CPU utilization changes.
3619 * @rq: Runqueue to carry out the update for.
3620 * @flags: Update reason flags.
3621 *
3622 * This function is called by the scheduler on the CPU whose utilization is
3623 * being updated.
3624 *
3625 * It can only be called from RCU-sched read-side critical sections.
3626 *
3627 * The way cpufreq is currently arranged requires it to evaluate the CPU
3628 * performance state (frequency/voltage) on a regular basis to prevent it from
3629 * being stuck in a completely inadequate performance level for too long.
3630 * That is not guaranteed to happen if the updates are only triggered from CFS
3631 * and DL, though, because they may not be coming in if only RT tasks are
3632 * active all the time (or there are RT tasks only).
3633 *
3634 * As a workaround for that issue, this function is called periodically by the
3635 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3636 * but that really is a band-aid. Going forward it should be replaced with
3637 * solutions targeted more specifically at RT tasks.
3638 */
3639static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3640{
3641 struct update_util_data *data;
3642
3643 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3644 cpu_of(rq)));
3645 if (data)
3646 data->func(data, rq_clock(rq), flags);
3647}
3648#else /* !CONFIG_CPU_FREQ: */
3649static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3650#endif /* !CONFIG_CPU_FREQ */
3651
3652#ifdef arch_scale_freq_capacity
3653# ifndef arch_scale_freq_invariant
3654# define arch_scale_freq_invariant() true
3655# endif
3656#else
3657# define arch_scale_freq_invariant() false
3658#endif
3659
3660unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3661 unsigned long *min,
3662 unsigned long *max);
3663
3664unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3665 unsigned long min,
3666 unsigned long max);
3667
3668
3669/*
3670 * Verify the fitness of task @p to run on @cpu taking into account the
3671 * CPU original capacity and the runtime/deadline ratio of the task.
3672 *
3673 * The function will return true if the original capacity of @cpu is
3674 * greater than or equal to task's deadline density right shifted by
3675 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3676 */
3677static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3678{
3679 unsigned long cap = arch_scale_cpu_capacity(cpu);
3680
3681 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3682}
3683
3684static inline unsigned long cpu_bw_dl(struct rq *rq)
3685{
3686 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3687}
3688
3689static inline unsigned long cpu_util_dl(struct rq *rq)
3690{
3691 return READ_ONCE(rq->avg_dl.util_avg);
3692}
3693
3694
3695extern unsigned long cpu_util_cfs(int cpu);
3696extern unsigned long cpu_util_cfs_boost(int cpu);
3697
3698static inline unsigned long cpu_util_rt(struct rq *rq)
3699{
3700 return READ_ONCE(rq->avg_rt.util_avg);
3701}
3702
3703#ifdef CONFIG_UCLAMP_TASK
3704
3705unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3706
3707/*
3708 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3709 * by default in the fast path and only gets turned on once userspace performs
3710 * an operation that requires it.
3711 *
3712 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3713 * hence is active.
3714 */
3715static inline bool uclamp_is_used(void)
3716{
3717 return static_branch_likely(&sched_uclamp_used);
3718}
3719
3720/*
3721 * Enabling static branches would get the cpus_read_lock(),
3722 * check whether uclamp_is_used before enable it to avoid always
3723 * calling cpus_read_lock(). Because we never disable this
3724 * static key once enable it.
3725 */
3726static inline void sched_uclamp_enable(void)
3727{
3728 if (!uclamp_is_used())
3729 static_branch_enable(&sched_uclamp_used);
3730}
3731
3732static inline unsigned long uclamp_rq_get(struct rq *rq,
3733 enum uclamp_id clamp_id)
3734{
3735 return READ_ONCE(rq->uclamp[clamp_id].value);
3736}
3737
3738static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3739 unsigned int value)
3740{
3741 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3742}
3743
3744static inline bool uclamp_rq_is_idle(struct rq *rq)
3745{
3746 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3747}
3748
3749/* Is the rq being capped/throttled by uclamp_max? */
3750static inline bool uclamp_rq_is_capped(struct rq *rq)
3751{
3752 unsigned long rq_util;
3753 unsigned long max_util;
3754
3755 if (!uclamp_is_used())
3756 return false;
3757
3758 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3759 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3760
3761 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3762}
3763
3764#define for_each_clamp_id(clamp_id) \
3765 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3766
3767extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3768
3769
3770static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3771{
3772 if (clamp_id == UCLAMP_MIN)
3773 return 0;
3774 return SCHED_CAPACITY_SCALE;
3775}
3776
3777/* Integer rounded range for each bucket */
3778#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3779
3780static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3781{
3782 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3783}
3784
3785static inline void
3786uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3787{
3788 uc_se->value = value;
3789 uc_se->bucket_id = uclamp_bucket_id(value);
3790 uc_se->user_defined = user_defined;
3791}
3792
3793#else /* !CONFIG_UCLAMP_TASK: */
3794
3795static inline unsigned long
3796uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3797{
3798 if (clamp_id == UCLAMP_MIN)
3799 return 0;
3800
3801 return SCHED_CAPACITY_SCALE;
3802}
3803
3804static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3805
3806static inline bool uclamp_is_used(void)
3807{
3808 return false;
3809}
3810
3811static inline void sched_uclamp_enable(void) {}
3812
3813static inline unsigned long
3814uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3815{
3816 if (clamp_id == UCLAMP_MIN)
3817 return 0;
3818
3819 return SCHED_CAPACITY_SCALE;
3820}
3821
3822static inline void
3823uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3824{
3825}
3826
3827static inline bool uclamp_rq_is_idle(struct rq *rq)
3828{
3829 return false;
3830}
3831
3832#endif /* !CONFIG_UCLAMP_TASK */
3833
3834#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3835
3836static inline unsigned long cpu_util_irq(struct rq *rq)
3837{
3838 return READ_ONCE(rq->avg_irq.util_avg);
3839}
3840
3841static inline
3842unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3843{
3844 util *= (max - irq);
3845 util /= max;
3846
3847 return util;
3848
3849}
3850
3851#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3852
3853static inline unsigned long cpu_util_irq(struct rq *rq)
3854{
3855 return 0;
3856}
3857
3858static inline
3859unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3860{
3861 return util;
3862}
3863
3864#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3865
3866extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr);
3867
3868#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3869
3870#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3871
3872DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3873
3874static inline bool sched_energy_enabled(void)
3875{
3876 return static_branch_unlikely(&sched_energy_present);
3877}
3878
3879#else /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL): */
3880
3881#define perf_domain_span(pd) NULL
3882
3883static inline bool sched_energy_enabled(void) { return false; }
3884
3885#endif /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3886
3887#ifdef CONFIG_MEMBARRIER
3888
3889/*
3890 * The scheduler provides memory barriers required by membarrier between:
3891 * - prior user-space memory accesses and store to rq->membarrier_state,
3892 * - store to rq->membarrier_state and following user-space memory accesses.
3893 * In the same way it provides those guarantees around store to rq->curr.
3894 */
3895static inline void membarrier_switch_mm(struct rq *rq,
3896 struct mm_struct *prev_mm,
3897 struct mm_struct *next_mm)
3898{
3899 int membarrier_state;
3900
3901 if (prev_mm == next_mm)
3902 return;
3903
3904 membarrier_state = atomic_read(&next_mm->membarrier_state);
3905 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3906 return;
3907
3908 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3909}
3910
3911#else /* !CONFIG_MEMBARRIER: */
3912
3913static inline void membarrier_switch_mm(struct rq *rq,
3914 struct mm_struct *prev_mm,
3915 struct mm_struct *next_mm)
3916{
3917}
3918
3919#endif /* !CONFIG_MEMBARRIER */
3920
3921static inline bool is_per_cpu_kthread(struct task_struct *p)
3922{
3923 if (!(p->flags & PF_KTHREAD))
3924 return false;
3925
3926 if (p->nr_cpus_allowed != 1)
3927 return false;
3928
3929 return true;
3930}
3931
3932extern void swake_up_all_locked(struct swait_queue_head *q);
3933extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3934
3935extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3936
3937#ifdef CONFIG_PREEMPT_DYNAMIC
3938extern int preempt_dynamic_mode;
3939extern int sched_dynamic_mode(const char *str);
3940extern void sched_dynamic_update(int mode);
3941#endif
3942extern const char *preempt_modes[];
3943
3944#ifdef CONFIG_SCHED_MM_CID
3945
3946static __always_inline bool cid_on_cpu(unsigned int cid)
3947{
3948 return cid & MM_CID_ONCPU;
3949}
3950
3951static __always_inline bool cid_in_transit(unsigned int cid)
3952{
3953 return cid & MM_CID_TRANSIT;
3954}
3955
3956static __always_inline unsigned int cpu_cid_to_cid(unsigned int cid)
3957{
3958 return cid & ~MM_CID_ONCPU;
3959}
3960
3961static __always_inline unsigned int cid_to_cpu_cid(unsigned int cid)
3962{
3963 return cid | MM_CID_ONCPU;
3964}
3965
3966static __always_inline unsigned int cid_to_transit_cid(unsigned int cid)
3967{
3968 return cid | MM_CID_TRANSIT;
3969}
3970
3971static __always_inline unsigned int cid_from_transit_cid(unsigned int cid)
3972{
3973 return cid & ~MM_CID_TRANSIT;
3974}
3975
3976static __always_inline bool cid_on_task(unsigned int cid)
3977{
3978 /* True if none of the MM_CID_ONCPU, MM_CID_TRANSIT, MM_CID_UNSET bits is set */
3979 return cid < MM_CID_TRANSIT;
3980}
3981
3982static __always_inline void mm_drop_cid(struct mm_struct *mm, unsigned int cid)
3983{
3984 clear_bit(cid, mm_cidmask(mm));
3985}
3986
3987static __always_inline void mm_unset_cid_on_task(struct task_struct *t)
3988{
3989 unsigned int cid = t->mm_cid.cid;
3990
3991 t->mm_cid.cid = MM_CID_UNSET;
3992 if (cid_on_task(cid))
3993 mm_drop_cid(t->mm, cid);
3994}
3995
3996static __always_inline void mm_drop_cid_on_cpu(struct mm_struct *mm, struct mm_cid_pcpu *pcp)
3997{
3998 /* Clear the ONCPU bit, but do not set UNSET in the per CPU storage */
3999 if (cid_on_cpu(pcp->cid)) {
4000 pcp->cid = cpu_cid_to_cid(pcp->cid);
4001 mm_drop_cid(mm, pcp->cid);
4002 }
4003}
4004
4005static inline unsigned int __mm_get_cid(struct mm_struct *mm, unsigned int max_cids)
4006{
4007 unsigned int cid = find_first_zero_bit(mm_cidmask(mm), max_cids);
4008
4009 if (cid >= max_cids)
4010 return MM_CID_UNSET;
4011 if (test_and_set_bit(cid, mm_cidmask(mm)))
4012 return MM_CID_UNSET;
4013 return cid;
4014}
4015
4016static inline unsigned int mm_get_cid(struct mm_struct *mm)
4017{
4018 unsigned int cid = __mm_get_cid(mm, READ_ONCE(mm->mm_cid.max_cids));
4019
4020 while (cid == MM_CID_UNSET) {
4021 cpu_relax();
4022 cid = __mm_get_cid(mm, num_possible_cpus());
4023 }
4024 return cid;
4025}
4026
4027static inline unsigned int mm_cid_converge(struct mm_struct *mm, unsigned int orig_cid,
4028 unsigned int max_cids)
4029{
4030 unsigned int new_cid, cid = cpu_cid_to_cid(orig_cid);
4031
4032 /* Is it in the optimal CID space? */
4033 if (likely(cid < max_cids))
4034 return orig_cid;
4035
4036 /* Try to find one in the optimal space. Otherwise keep the provided. */
4037 new_cid = __mm_get_cid(mm, max_cids);
4038 if (new_cid != MM_CID_UNSET) {
4039 mm_drop_cid(mm, cid);
4040 /* Preserve the ONCPU mode of the original CID */
4041 return new_cid | (orig_cid & MM_CID_ONCPU);
4042 }
4043 return orig_cid;
4044}
4045
4046static __always_inline void mm_cid_update_task_cid(struct task_struct *t, unsigned int cid)
4047{
4048 if (t->mm_cid.cid != cid) {
4049 t->mm_cid.cid = cid;
4050 rseq_sched_set_ids_changed(t);
4051 }
4052}
4053
4054static __always_inline void mm_cid_update_pcpu_cid(struct mm_struct *mm, unsigned int cid)
4055{
4056 __this_cpu_write(mm->mm_cid.pcpu->cid, cid);
4057}
4058
4059static __always_inline void mm_cid_from_cpu(struct task_struct *t, unsigned int cpu_cid,
4060 unsigned int mode)
4061{
4062 unsigned int max_cids, tcid = t->mm_cid.cid;
4063 struct mm_struct *mm = t->mm;
4064
4065 max_cids = READ_ONCE(mm->mm_cid.max_cids);
4066 /* Optimize for the common case where both have the ONCPU bit set */
4067 if (likely(cid_on_cpu(cpu_cid & tcid))) {
4068 if (likely(cpu_cid_to_cid(cpu_cid) < max_cids)) {
4069 mm_cid_update_task_cid(t, cpu_cid);
4070 return;
4071 }
4072 /* Try to converge into the optimal CID space */
4073 cpu_cid = mm_cid_converge(mm, cpu_cid, max_cids);
4074 } else {
4075 /* Hand over or drop the task owned CID */
4076 if (cid_on_task(tcid)) {
4077 if (cid_on_cpu(cpu_cid))
4078 mm_unset_cid_on_task(t);
4079 else
4080 cpu_cid = cid_to_cpu_cid(tcid);
4081 }
4082 /* Still nothing, allocate a new one */
4083 if (!cid_on_cpu(cpu_cid))
4084 cpu_cid = cid_to_cpu_cid(mm_get_cid(mm));
4085
4086 /* Handle the transition mode flag if required */
4087 if (mode & MM_CID_TRANSIT)
4088 cpu_cid = cpu_cid_to_cid(cpu_cid) | MM_CID_TRANSIT;
4089 }
4090 mm_cid_update_pcpu_cid(mm, cpu_cid);
4091 mm_cid_update_task_cid(t, cpu_cid);
4092}
4093
4094static __always_inline void mm_cid_from_task(struct task_struct *t, unsigned int cpu_cid,
4095 unsigned int mode)
4096{
4097 unsigned int max_cids, tcid = t->mm_cid.cid;
4098 struct mm_struct *mm = t->mm;
4099
4100 max_cids = READ_ONCE(mm->mm_cid.max_cids);
4101 /* Optimize for the common case, where both have the ONCPU bit clear */
4102 if (likely(cid_on_task(tcid | cpu_cid))) {
4103 if (likely(tcid < max_cids)) {
4104 mm_cid_update_pcpu_cid(mm, tcid);
4105 return;
4106 }
4107 /* Try to converge into the optimal CID space */
4108 tcid = mm_cid_converge(mm, tcid, max_cids);
4109 } else {
4110 /* Hand over or drop the CPU owned CID */
4111 if (cid_on_cpu(cpu_cid)) {
4112 if (cid_on_task(tcid))
4113 mm_drop_cid_on_cpu(mm, this_cpu_ptr(mm->mm_cid.pcpu));
4114 else
4115 tcid = cpu_cid_to_cid(cpu_cid);
4116 }
4117 /* Still nothing, allocate a new one */
4118 if (!cid_on_task(tcid))
4119 tcid = mm_get_cid(mm);
4120 /* Set the transition mode flag if required */
4121 tcid |= mode & MM_CID_TRANSIT;
4122 }
4123 mm_cid_update_pcpu_cid(mm, tcid);
4124 mm_cid_update_task_cid(t, tcid);
4125}
4126
4127static __always_inline void mm_cid_schedin(struct task_struct *next)
4128{
4129 struct mm_struct *mm = next->mm;
4130 unsigned int cpu_cid, mode;
4131
4132 if (!next->mm_cid.active)
4133 return;
4134
4135 cpu_cid = __this_cpu_read(mm->mm_cid.pcpu->cid);
4136 mode = READ_ONCE(mm->mm_cid.mode);
4137 if (likely(!cid_on_cpu(mode)))
4138 mm_cid_from_task(next, cpu_cid, mode);
4139 else
4140 mm_cid_from_cpu(next, cpu_cid, mode);
4141}
4142
4143static __always_inline void mm_cid_schedout(struct task_struct *prev)
4144{
4145 struct mm_struct *mm = prev->mm;
4146 unsigned int mode, cid;
4147
4148 /* During mode transitions CIDs are temporary and need to be dropped */
4149 if (likely(!cid_in_transit(prev->mm_cid.cid)))
4150 return;
4151
4152 mode = READ_ONCE(mm->mm_cid.mode);
4153 cid = cid_from_transit_cid(prev->mm_cid.cid);
4154
4155 /*
4156 * If transition mode is done, transfer ownership when the CID is
4157 * within the convergence range to optimize the next schedule in.
4158 */
4159 if (!cid_in_transit(mode) && cid < READ_ONCE(mm->mm_cid.max_cids)) {
4160 if (cid_on_cpu(mode))
4161 cid = cid_to_cpu_cid(cid);
4162
4163 /* Update both so that the next schedule in goes into the fast path */
4164 mm_cid_update_pcpu_cid(mm, cid);
4165 prev->mm_cid.cid = cid;
4166 } else {
4167 mm_drop_cid(mm, cid);
4168 prev->mm_cid.cid = MM_CID_UNSET;
4169 }
4170}
4171
4172static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next)
4173{
4174 mm_cid_schedout(prev);
4175 mm_cid_schedin(next);
4176}
4177
4178#else /* !CONFIG_SCHED_MM_CID: */
4179static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next) { }
4180#endif /* !CONFIG_SCHED_MM_CID */
4181
4182#ifdef CONFIG_SCHED_CACHE
4183DECLARE_STATIC_KEY_FALSE(sched_cache_present);
4184DECLARE_STATIC_KEY_FALSE(sched_cache_active);
4185extern int sysctl_sched_cache_user;
4186extern unsigned int llc_aggr_tolerance;
4187extern unsigned int llc_epoch_period;
4188extern unsigned int llc_epoch_affinity_timeout;
4189extern unsigned int llc_imb_pct;
4190extern unsigned int llc_overaggr_pct;
4191
4192static inline bool sched_cache_enabled(void)
4193{
4194 return static_branch_unlikely(&sched_cache_active);
4195}
4196
4197extern void sched_cache_active_set(void);
4198
4199#endif
4200
4201void sched_domains_free_llc_id(int cpu);
4202
4203extern void init_sched_mm(struct task_struct *p);
4204
4205extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
4206extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
4207static inline
4208void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
4209{
4210 lockdep_assert_rq_held(src_rq);
4211 lockdep_assert_rq_held(dst_rq);
4212
4213 deactivate_task(src_rq, task, 0);
4214 set_task_cpu(task, dst_rq->cpu);
4215 activate_task(dst_rq, task, 0);
4216 wakeup_preempt(dst_rq, task, 0);
4217}
4218
4219static inline
4220bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
4221{
4222 if (!task_on_cpu(rq, p) &&
4223 cpumask_test_cpu(cpu, &p->cpus_mask))
4224 return true;
4225
4226 return false;
4227}
4228
4229#ifdef CONFIG_RT_MUTEXES
4230
4231static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
4232{
4233 if (pi_task)
4234 prio = min(prio, pi_task->prio);
4235
4236 return prio;
4237}
4238
4239static inline int rt_effective_prio(struct task_struct *p, int prio)
4240{
4241 struct task_struct *pi_task = rt_mutex_get_top_task(p);
4242
4243 return __rt_effective_prio(pi_task, prio);
4244}
4245
4246#else /* !CONFIG_RT_MUTEXES: */
4247
4248static inline int rt_effective_prio(struct task_struct *p, int prio)
4249{
4250 return prio;
4251}
4252
4253#endif /* !CONFIG_RT_MUTEXES */
4254
4255extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
4256extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
4257extern const struct sched_class *__setscheduler_class(int policy, int prio);
4258extern void set_load_weight(struct task_struct *p, bool update_load);
4259extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
4260extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
4261
4262extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
4263
4264extern void __balance_callbacks(struct rq *rq, struct rq_flags *rf);
4265extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
4266
4267/*
4268 * The 'sched_change' pattern is the safe, easy and slow way of changing a
4269 * task's scheduling properties. It dequeues a task, such that the scheduler
4270 * is fully unaware of it; at which point its properties can be modified;
4271 * after which it is enqueued again.
4272 *
4273 * Typically this must be called while holding task_rq_lock, since most/all
4274 * properties are serialized under those locks. There are currently two
4275 * exceptions to this rule in sched/ext which only hold rq->lock: scx_bypass()
4276 * and rq_offline_scx().
4277 */
4278
4279/*
4280 * This structure is a temporary, used to preserve/convey the queueing state
4281 * of the task between sched_change_begin() and sched_change_end(). Ensuring
4282 * the task's queueing state is idempotent across the operation.
4283 */
4284struct sched_change_ctx {
4285 u64 prio;
4286 struct task_struct *p;
4287 const struct sched_class *class;
4288 int flags;
4289 bool queued;
4290 bool running;
4291};
4292
4293struct sched_change_ctx *sched_change_begin(struct task_struct *p, unsigned int flags);
4294void sched_change_end(struct sched_change_ctx *ctx);
4295
4296DEFINE_CLASS(sched_change, struct sched_change_ctx *,
4297 sched_change_end(_T),
4298 sched_change_begin(p, flags),
4299 struct task_struct *p, unsigned int flags)
4300
4301DEFINE_CLASS_IS_UNCONDITIONAL(sched_change)
4302
4303#include "ext/ext.h"
4304
4305#ifdef CONFIG_PREFERRED_CPU
4306void sched_push_current_non_preferred_cpu(struct rq *rq);
4307#else /* !CONFIG_PREFERRED_CPU */
4308static inline void sched_push_current_non_preferred_cpu(struct rq *rq) { }
4309#endif
4310
4311#endif /* _KERNEL_SCHED_SCHED_H */