projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'x86-splitlock-2021-06-28' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-block.git] / kernel / sched / sched.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
97fb7a0a
IM		 2/*
3 * Scheduler internal types and methods:
4 */
029632fb 5#include <linux/sched.h>
325ea10c 6
dfc3401a 7#include <linux/sched/autogroup.h>
e6017571 8#include <linux/sched/clock.h>
325ea10c 9#include <linux/sched/coredump.h>
55687da1 10#include <linux/sched/cpufreq.h>
325ea10c
IM		 11#include <linux/sched/cputime.h>
12#include <linux/sched/deadline.h>
b17b0153 13#include <linux/sched/debug.h>
ef8bd77f 14#include <linux/sched/hotplug.h>
325ea10c
IM		 15#include <linux/sched/idle.h>
16#include <linux/sched/init.h>
17#include <linux/sched/isolation.h>
18#include <linux/sched/jobctl.h>
19#include <linux/sched/loadavg.h>
20#include <linux/sched/mm.h>
21#include <linux/sched/nohz.h>
22#include <linux/sched/numa_balancing.h>
23#include <linux/sched/prio.h>
24#include <linux/sched/rt.h>
25#include <linux/sched/signal.h>
321a874a 26#include <linux/sched/smt.h>
325ea10c
IM		 27#include <linux/sched/stat.h>
28#include <linux/sched/sysctl.h>
29930025 29#include <linux/sched/task.h>
68db0cf1 30#include <linux/sched/task_stack.h>
325ea10c
IM		 31#include <linux/sched/topology.h>
32#include <linux/sched/user.h>
33#include <linux/sched/wake_q.h>
34#include <linux/sched/xacct.h>
35
36#include <uapi/linux/sched/types.h>
ef8bd77f 37
3866e845 38#include <linux/binfmts.h>
1e17fb8e 39#include <linux/bitops.h>
325ea10c
IM		 40#include <linux/blkdev.h>
41#include <linux/compat.h>
42#include <linux/context_tracking.h>
43#include <linux/cpufreq.h>
44#include <linux/cpuidle.h>
45#include <linux/cpuset.h>
46#include <linux/ctype.h>
47#include <linux/debugfs.h>
48#include <linux/delayacct.h>
6aa140fa 49#include <linux/energy_model.h>
325ea10c
IM		 50#include <linux/init_task.h>
51#include <linux/kprobes.h>
52#include <linux/kthread.h>
53#include <linux/membarrier.h>
54#include <linux/migrate.h>
55#include <linux/mmu_context.h>
56#include <linux/nmi.h>
57#include <linux/proc_fs.h>
58#include <linux/prefetch.h>
59#include <linux/profile.h>
eb414681 60#include <linux/psi.h>
c006fac5 61#include <linux/ratelimit.h>
325ea10c
IM		 62#include <linux/rcupdate_wait.h>
63#include <linux/security.h>
029632fb 64#include <linux/stop_machine.h>
325ea10c
IM		 65#include <linux/suspend.h>
66#include <linux/swait.h>
67#include <linux/syscalls.h>
68#include <linux/task_work.h>
69#include <linux/tsacct_kern.h>
70
71#include <asm/tlb.h>
029632fb 72
7fce777c 73#ifdef CONFIG_PARAVIRT
325ea10c 74# include <asm/paravirt.h>
7fce777c
IM		 75#endif
76
391e43da 77#include "cpupri.h"
6bfd6d72 78#include "cpudeadline.h"
029632fb 79
9d246053
PA		 80#include <trace/events/sched.h>
81
9148a3a1 82#ifdef CONFIG_SCHED_DEBUG
6d3aed3d 83# define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
9148a3a1 84#else
6d3aed3d 85# define SCHED_WARN_ON(x) ({ (void)(x), 0; })
9148a3a1
PZ		 86#endif
87
45ceebf7 88struct rq;
442bf3aa 89struct cpuidle_state;
45ceebf7 90
da0c1e65
KT		 91/* task_struct::on_rq states: */
92#define TASK_ON_RQ_QUEUED 1
cca26e80 93#define TASK_ON_RQ_MIGRATING 2
da0c1e65 94
029632fb
PZ		 95extern __read_mostly int scheduler_running;
96
45ceebf7
PG		 97extern unsigned long calc_load_update;
98extern atomic_long_t calc_load_tasks;
99
3289bdb4 100extern void calc_global_load_tick(struct rq *this_rq);
d60585c5 101extern long calc_load_fold_active(struct rq *this_rq, long adjust);
3289bdb4 102
9d246053 103extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
029632fb
PZ		 104/*
105 * Helpers for converting nanosecond timing to jiffy resolution
106 */
107#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108
cc1f4b1f
LZ		 109/*
110 * Increase resolution of nice-level calculations for 64-bit architectures.
111 * The extra resolution improves shares distribution and load balancing of
112 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113 * hierarchies, especially on larger systems. This is not a user-visible change
114 * and does not change the user-interface for setting shares/weights.
115 *
116 * We increase resolution only if we have enough bits to allow this increased
97fb7a0a
IM		 117 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118 * are pretty high and the returns do not justify the increased costs.
2159197d 119 *
97fb7a0a
IM		 120 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121 * increase coverage and consistency always enable it on 64-bit platforms.
cc1f4b1f 122 */
2159197d 123#ifdef CONFIG_64BIT
172895e6 124# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
6ecdd749 125# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
26cf5222
MW		 126# define scale_load_down(w) \
127({ \
128 unsigned long __w = (w); \
129 if (__w) \
130 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
131 __w; \
132})
cc1f4b1f 133#else
172895e6 134# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
cc1f4b1f
LZ		 135# define scale_load(w) (w)
136# define scale_load_down(w) (w)
137#endif
138
6ecdd749 139/*
172895e6
YD		 140 * Task weight (visible to users) and its load (invisible to users) have
141 * independent resolution, but they should be well calibrated. We use
142 * scale_load() and scale_load_down(w) to convert between them. The
143 * following must be true:
144 *
9d061ba6 145 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
172895e6 146 *
6ecdd749 147 */
172895e6 148#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
029632fb 149
332ac17e
DF		 150/*
151 * Single value that decides SCHED_DEADLINE internal math precision.
152 * 10 -> just above 1us
153 * 9 -> just above 0.5us
154 */
97fb7a0a 155#define DL_SCALE 10
029632fb
PZ		 156
157/*
97fb7a0a 158 * Single value that denotes runtime == period, ie unlimited time.
029632fb 159 */
97fb7a0a 160#define RUNTIME_INF ((u64)~0ULL)
029632fb 161
20f9cd2a
HA		 162static inline int idle_policy(int policy)
163{
164 return policy == SCHED_IDLE;
165}
d50dde5a
DF		 166static inline int fair_policy(int policy)
167{
168 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
169}
170
029632fb
PZ		 171static inline int rt_policy(int policy)
172{
d50dde5a 173 return policy == SCHED_FIFO || policy == SCHED_RR;
029632fb
PZ		 174}
175
aab03e05
DF		 176static inline int dl_policy(int policy)
177{
178 return policy == SCHED_DEADLINE;
179}
20f9cd2a
HA		 180static inline bool valid_policy(int policy)
181{
182 return idle_policy(policy) || fair_policy(policy) ||
183 rt_policy(policy) || dl_policy(policy);
184}
aab03e05 185
1da1843f
VK		 186static inline int task_has_idle_policy(struct task_struct *p)
187{
188 return idle_policy(p->policy);
189}
190
029632fb
PZ		 191static inline int task_has_rt_policy(struct task_struct *p)
192{
193 return rt_policy(p->policy);
194}
195
aab03e05
DF		 196static inline int task_has_dl_policy(struct task_struct *p)
197{
198 return dl_policy(p->policy);
199}
200
07881166
JL		 201#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
202
d76343c6
VS		 203static inline void update_avg(u64 *avg, u64 sample)
204{
205 s64 diff = sample - *avg;
206 *avg += diff / 8;
207}
208
39a2a6eb
VS		 209/*
210 * Shifting a value by an exponent greater *or equal* to the size of said value
211 * is UB; cap at size-1.
212 */
213#define shr_bound(val, shift) \
214 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
215
794a56eb
JL		 216/*
217 * !! For sched_setattr_nocheck() (kernel) only !!
218 *
219 * This is actually gross. :(
220 *
221 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
222 * tasks, but still be able to sleep. We need this on platforms that cannot
223 * atomically change clock frequency. Remove once fast switching will be
224 * available on such platforms.
225 *
226 * SUGOV stands for SchedUtil GOVernor.
227 */
228#define SCHED_FLAG_SUGOV 0x10000000
229
230static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
231{
232#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
233 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
234#else
235 return false;
236#endif
237}
238
2d3d891d
DF		 239/*
240 * Tells if entity @a should preempt entity @b.
241 */
332ac17e
DF		 242static inline bool
243dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
2d3d891d 244{
794a56eb
JL		 245 return dl_entity_is_special(a) ||
246 dl_time_before(a->deadline, b->deadline);
2d3d891d
DF		 247}
248
029632fb
PZ		 249/*
250 * This is the priority-queue data structure of the RT scheduling class:
251 */
252struct rt_prio_array {
253 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
254 struct list_head queue[MAX_RT_PRIO];
255};
256
257struct rt_bandwidth {
258 /* nests inside the rq lock: */
259 raw_spinlock_t rt_runtime_lock;
260 ktime_t rt_period;
261 u64 rt_runtime;
262 struct hrtimer rt_period_timer;
4cfafd30 263 unsigned int rt_period_active;
029632fb 264};
a5e7be3b
JL		 265
266void __dl_clear_params(struct task_struct *p);
267
332ac17e 268struct dl_bandwidth {
97fb7a0a
IM		 269 raw_spinlock_t dl_runtime_lock;
270 u64 dl_runtime;
271 u64 dl_period;
332ac17e
DF		 272};
273
274static inline int dl_bandwidth_enabled(void)
275{
1724813d 276 return sysctl_sched_rt_runtime >= 0;
332ac17e
DF		 277}
278
a57415f5
PL		 279/*
280 * To keep the bandwidth of -deadline tasks under control
281 * we need some place where:
282 * - store the maximum -deadline bandwidth of each cpu;
283 * - cache the fraction of bandwidth that is currently allocated in
284 * each root domain;
285 *
286 * This is all done in the data structure below. It is similar to the
287 * one used for RT-throttling (rt_bandwidth), with the main difference
288 * that, since here we are only interested in admission control, we
289 * do not decrease any runtime while the group "executes", neither we
290 * need a timer to replenish it.
291 *
292 * With respect to SMP, bandwidth is given on a per root domain basis,
293 * meaning that:
294 * - bw (< 100%) is the deadline bandwidth of each CPU;
295 * - total_bw is the currently allocated bandwidth in each root domain;
296 */
332ac17e 297struct dl_bw {
97fb7a0a
IM		 298 raw_spinlock_t lock;
299 u64 bw;
300 u64 total_bw;
332ac17e
DF		 301};
302
daec5798
LA		 303static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
304
7f51412a 305static inline
8c0944ce 306void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
7f51412a
JL		 307{
308 dl_b->total_bw -= tsk_bw;
daec5798 309 __dl_update(dl_b, (s32)tsk_bw / cpus);
7f51412a
JL		 310}
311
312static inline
daec5798 313void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
7f51412a
JL		 314{
315 dl_b->total_bw += tsk_bw;
daec5798 316 __dl_update(dl_b, -((s32)tsk_bw / cpus));
7f51412a
JL		 317}
318
60ffd5ed
LA		 319static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
320 u64 old_bw, u64 new_bw)
7f51412a
JL		 321{
322 return dl_b->bw != -1 &&
60ffd5ed 323 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
7f51412a
JL		 324}
325
b4118988
LA		 326/*
327 * Verify the fitness of task @p to run on @cpu taking into account the
328 * CPU original capacity and the runtime/deadline ratio of the task.
329 *
330 * The function will return true if the CPU original capacity of the
331 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
332 * task and false otherwise.
333 */
334static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
335{
336 unsigned long cap = arch_scale_cpu_capacity(cpu);
337
338 return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
339}
340
f2cb1360 341extern void init_dl_bw(struct dl_bw *dl_b);
97fb7a0a 342extern int sched_dl_global_validate(void);
06a76fe0 343extern void sched_dl_do_global(void);
97fb7a0a 344extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
06a76fe0
NP		 345extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
346extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
347extern bool __checkparam_dl(const struct sched_attr *attr);
06a76fe0 348extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
97fb7a0a
IM		 349extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
350extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
06a76fe0 351extern bool dl_cpu_busy(unsigned int cpu);
029632fb
PZ		 352
353#ifdef CONFIG_CGROUP_SCHED
354
355#include <linux/cgroup.h>
eb414681 356#include <linux/psi.h>
029632fb
PZ		 357
358struct cfs_rq;
359struct rt_rq;
360
35cf4e50 361extern struct list_head task_groups;
029632fb
PZ		 362
363struct cfs_bandwidth {
364#ifdef CONFIG_CFS_BANDWIDTH
97fb7a0a
IM		 365 raw_spinlock_t lock;
366 ktime_t period;
367 u64 quota;
368 u64 runtime;
f4183717 369 u64 burst;
97fb7a0a 370 s64 hierarchical_quota;
97fb7a0a 371
66567fcb 372 u8 idle;
373 u8 period_active;
66567fcb 374 u8 slack_started;
97fb7a0a
IM		 375 struct hrtimer period_timer;
376 struct hrtimer slack_timer;
377 struct list_head throttled_cfs_rq;
378
379 /* Statistics: */
380 int nr_periods;
381 int nr_throttled;
382 u64 throttled_time;
029632fb
PZ		 383#endif
384};
385
97fb7a0a 386/* Task group related information */
029632fb
PZ		 387struct task_group {
388 struct cgroup_subsys_state css;
389
390#ifdef CONFIG_FAIR_GROUP_SCHED
97fb7a0a
IM		 391 /* schedulable entities of this group on each CPU */
392 struct sched_entity **se;
393 /* runqueue "owned" by this group on each CPU */
394 struct cfs_rq **cfs_rq;
395 unsigned long shares;
029632fb 396
fa6bddeb 397#ifdef CONFIG_SMP
b0367629
WL		 398 /*
399 * load_avg can be heavily contended at clock tick time, so put
400 * it in its own cacheline separated from the fields above which
401 * will also be accessed at each tick.
402 */
97fb7a0a 403 atomic_long_t load_avg ____cacheline_aligned;
029632fb 404#endif
fa6bddeb 405#endif
029632fb
PZ		 406
407#ifdef CONFIG_RT_GROUP_SCHED
97fb7a0a
IM		 408 struct sched_rt_entity **rt_se;
409 struct rt_rq **rt_rq;
029632fb 410
97fb7a0a 411 struct rt_bandwidth rt_bandwidth;
029632fb
PZ		 412#endif
413
97fb7a0a
IM		 414 struct rcu_head rcu;
415 struct list_head list;
029632fb 416
97fb7a0a
IM		 417 struct task_group *parent;
418 struct list_head siblings;
419 struct list_head children;
029632fb
PZ		 420
421#ifdef CONFIG_SCHED_AUTOGROUP
97fb7a0a 422 struct autogroup *autogroup;
029632fb
PZ		 423#endif
424
97fb7a0a 425 struct cfs_bandwidth cfs_bandwidth;
2480c093
PB		 426
427#ifdef CONFIG_UCLAMP_TASK_GROUP
428 /* The two decimal precision [%] value requested from user-space */
429 unsigned int uclamp_pct[UCLAMP_CNT];
430 /* Clamp values requested for a task group */
431 struct uclamp_se uclamp_req[UCLAMP_CNT];
0b60ba2d
PB		 432 /* Effective clamp values used for a task group */
433 struct uclamp_se uclamp[UCLAMP_CNT];
2480c093
PB		 434#endif
435
029632fb
PZ		 436};
437
438#ifdef CONFIG_FAIR_GROUP_SCHED
439#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
440
441/*
442 * A weight of 0 or 1 can cause arithmetics problems.
443 * A weight of a cfs_rq is the sum of weights of which entities
444 * are queued on this cfs_rq, so a weight of a entity should not be
445 * too large, so as the shares value of a task group.
446 * (The default weight is 1024 - so there's no practical
447 * limitation from this.)
448 */
97fb7a0a
IM		 449#define MIN_SHARES (1UL << 1)
450#define MAX_SHARES (1UL << 18)
029632fb
PZ		 451#endif
452
029632fb
PZ		 453typedef int (*tg_visitor)(struct task_group *, void *);
454
455extern int walk_tg_tree_from(struct task_group *from,
456 tg_visitor down, tg_visitor up, void *data);
457
458/*
459 * Iterate the full tree, calling @down when first entering a node and @up when
460 * leaving it for the final time.
461 *
462 * Caller must hold rcu_lock or sufficient equivalent.
463 */
464static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
465{
466 return walk_tg_tree_from(&root_task_group, down, up, data);
467}
468
469extern int tg_nop(struct task_group *tg, void *data);
470
471extern void free_fair_sched_group(struct task_group *tg);
472extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
8663e24d 473extern void online_fair_sched_group(struct task_group *tg);
6fe1f348 474extern void unregister_fair_sched_group(struct task_group *tg);
029632fb
PZ		 475extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
476 struct sched_entity *se, int cpu,
477 struct sched_entity *parent);
478extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
029632fb
PZ		 479
480extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
77a4d1a1 481extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
029632fb
PZ		 482extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
483
484extern void free_rt_sched_group(struct task_group *tg);
485extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
486extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
487 struct sched_rt_entity *rt_se, int cpu,
488 struct sched_rt_entity *parent);
8887cd99
NP		 489extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
490extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
491extern long sched_group_rt_runtime(struct task_group *tg);
492extern long sched_group_rt_period(struct task_group *tg);
493extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
029632fb 494
25cc7da7
LZ		 495extern struct task_group *sched_create_group(struct task_group *parent);
496extern void sched_online_group(struct task_group *tg,
497 struct task_group *parent);
498extern void sched_destroy_group(struct task_group *tg);
499extern void sched_offline_group(struct task_group *tg);
500
501extern void sched_move_task(struct task_struct *tsk);
502
503#ifdef CONFIG_FAIR_GROUP_SCHED
504extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
ad936d86
BP		 505
506#ifdef CONFIG_SMP
507extern void set_task_rq_fair(struct sched_entity *se,
508 struct cfs_rq *prev, struct cfs_rq *next);
509#else /* !CONFIG_SMP */
510static inline void set_task_rq_fair(struct sched_entity *se,
511 struct cfs_rq *prev, struct cfs_rq *next) { }
512#endif /* CONFIG_SMP */
513#endif /* CONFIG_FAIR_GROUP_SCHED */
25cc7da7 514
029632fb
PZ		 515#else /* CONFIG_CGROUP_SCHED */
516
517struct cfs_bandwidth { };
518
519#endif /* CONFIG_CGROUP_SCHED */
520
521/* CFS-related fields in a runqueue */
522struct cfs_rq {
97fb7a0a 523 struct load_weight load;
97fb7a0a 524 unsigned int nr_running;
43e9f7f2
VK		 525 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */
526 unsigned int idle_h_nr_running; /* SCHED_IDLE */
029632fb 527
97fb7a0a
IM		 528 u64 exec_clock;
529 u64 min_vruntime;
c6047c2e
JFG		 530#ifdef CONFIG_SCHED_CORE
531 unsigned int forceidle_seq;
532 u64 min_vruntime_fi;
533#endif
534
029632fb 535#ifndef CONFIG_64BIT
97fb7a0a 536 u64 min_vruntime_copy;
029632fb
PZ		 537#endif
538
97fb7a0a 539 struct rb_root_cached tasks_timeline;
029632fb 540
029632fb
PZ		 541 /*
542 * 'curr' points to currently running entity on this cfs_rq.
543 * It is set to NULL otherwise (i.e when none are currently running).
544 */
97fb7a0a
IM		 545 struct sched_entity *curr;
546 struct sched_entity *next;
547 struct sched_entity *last;
548 struct sched_entity *skip;
029632fb
PZ		 549
550#ifdef CONFIG_SCHED_DEBUG
97fb7a0a 551 unsigned int nr_spread_over;
029632fb
PZ		 552#endif
553
2dac754e
PT		 554#ifdef CONFIG_SMP
555 /*
9d89c257 556 * CFS load tracking
2dac754e 557 */
97fb7a0a 558 struct sched_avg avg;
2a2f5d4e 559#ifndef CONFIG_64BIT
97fb7a0a 560 u64 load_last_update_time_copy;
9d89c257 561#endif
2a2f5d4e
PZ		 562 struct {
563 raw_spinlock_t lock ____cacheline_aligned;
564 int nr;
565 unsigned long load_avg;
566 unsigned long util_avg;
9f683953 567 unsigned long runnable_avg;
2a2f5d4e 568 } removed;
82958366 569
9d89c257 570#ifdef CONFIG_FAIR_GROUP_SCHED
97fb7a0a
IM		 571 unsigned long tg_load_avg_contrib;
572 long propagate;
573 long prop_runnable_sum;
0e2d2aaa 574
82958366
PT		 575 /*
576 * h_load = weight * f(tg)
577 *
578 * Where f(tg) is the recursive weight fraction assigned to
579 * this group.
580 */
97fb7a0a
IM		 581 unsigned long h_load;
582 u64 last_h_load_update;
583 struct sched_entity *h_load_next;
68520796 584#endif /* CONFIG_FAIR_GROUP_SCHED */
82958366
PT		 585#endif /* CONFIG_SMP */
586
029632fb 587#ifdef CONFIG_FAIR_GROUP_SCHED
97fb7a0a 588 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
029632fb
PZ		 589
590 /*
591 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
592 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
593 * (like users, containers etc.)
594 *
97fb7a0a
IM		 595 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
596 * This list is used during load balance.
029632fb 597 */
97fb7a0a
IM		 598 int on_list;
599 struct list_head leaf_cfs_rq_list;
600 struct task_group *tg; /* group that "owns" this runqueue */
029632fb 601
029632fb 602#ifdef CONFIG_CFS_BANDWIDTH
97fb7a0a 603 int runtime_enabled;
97fb7a0a
IM		 604 s64 runtime_remaining;
605
606 u64 throttled_clock;
607 u64 throttled_clock_task;
608 u64 throttled_clock_task_time;
609 int throttled;
610 int throttle_count;
611 struct list_head throttled_list;
029632fb
PZ		 612#endif /* CONFIG_CFS_BANDWIDTH */
613#endif /* CONFIG_FAIR_GROUP_SCHED */
614};
615
616static inline int rt_bandwidth_enabled(void)
617{
618 return sysctl_sched_rt_runtime >= 0;
619}
620
b6366f04 621/* RT IPI pull logic requires IRQ_WORK */
4bdced5c 622#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
b6366f04
SR		 623# define HAVE_RT_PUSH_IPI
624#endif
625
029632fb
PZ		 626/* Real-Time classes' related field in a runqueue: */
627struct rt_rq {
97fb7a0a
IM		 628 struct rt_prio_array active;
629 unsigned int rt_nr_running;
630 unsigned int rr_nr_running;
029632fb
PZ		 631#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
632 struct {
97fb7a0a 633 int curr; /* highest queued rt task prio */
029632fb 634#ifdef CONFIG_SMP
97fb7a0a 635 int next; /* next highest */
029632fb
PZ		 636#endif
637 } highest_prio;
638#endif
639#ifdef CONFIG_SMP
e6fe3f42
AD		 640 unsigned int rt_nr_migratory;
641 unsigned int rt_nr_total;
97fb7a0a
IM		 642 int overloaded;
643 struct plist_head pushable_tasks;
371bf427 644
b6366f04 645#endif /* CONFIG_SMP */
97fb7a0a 646 int rt_queued;
f4ebcbc0 647
97fb7a0a
IM		 648 int rt_throttled;
649 u64 rt_time;
650 u64 rt_runtime;
029632fb 651 /* Nests inside the rq lock: */
97fb7a0a 652 raw_spinlock_t rt_runtime_lock;
029632fb
PZ		 653
654#ifdef CONFIG_RT_GROUP_SCHED
e6fe3f42 655 unsigned int rt_nr_boosted;
029632fb 656
97fb7a0a
IM		 657 struct rq *rq;
658 struct task_group *tg;
029632fb
PZ		 659#endif
660};
661
296b2ffe
VG		 662static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
663{
664 return rt_rq->rt_queued && rt_rq->rt_nr_running;
665}
666
aab03e05
DF		 667/* Deadline class' related fields in a runqueue */
668struct dl_rq {
669 /* runqueue is an rbtree, ordered by deadline */
97fb7a0a 670 struct rb_root_cached root;
aab03e05 671
e6fe3f42 672 unsigned int dl_nr_running;
1baca4ce
JL		 673
674#ifdef CONFIG_SMP
675 /*
676 * Deadline values of the currently executing and the
677 * earliest ready task on this rq. Caching these facilitates
dfcb245e 678 * the decision whether or not a ready but not running task
1baca4ce
JL		 679 * should migrate somewhere else.
680 */
681 struct {
97fb7a0a
IM		 682 u64 curr;
683 u64 next;
1baca4ce
JL		 684 } earliest_dl;
685
e6fe3f42 686 unsigned int dl_nr_migratory;
97fb7a0a 687 int overloaded;
1baca4ce
JL		 688
689 /*
690 * Tasks on this rq that can be pushed away. They are kept in
691 * an rb-tree, ordered by tasks' deadlines, with caching
692 * of the leftmost (earliest deadline) element.
693 */
97fb7a0a 694 struct rb_root_cached pushable_dl_tasks_root;
332ac17e 695#else
97fb7a0a 696 struct dl_bw dl_bw;
1baca4ce 697#endif
e36d8677
LA		 698 /*
699 * "Active utilization" for this runqueue: increased when a
700 * task wakes up (becomes TASK_RUNNING) and decreased when a
701 * task blocks
702 */
97fb7a0a 703 u64 running_bw;
4da3abce 704
8fd27231
LA		 705 /*
706 * Utilization of the tasks "assigned" to this runqueue (including
707 * the tasks that are in runqueue and the tasks that executed on this
708 * CPU and blocked). Increased when a task moves to this runqueue, and
709 * decreased when the task moves away (migrates, changes scheduling
710 * policy, or terminates).
711 * This is needed to compute the "inactive utilization" for the
712 * runqueue (inactive utilization = this_bw - running_bw).
713 */
97fb7a0a
IM		 714 u64 this_bw;
715 u64 extra_bw;
8fd27231 716
4da3abce
LA		 717 /*
718 * Inverse of the fraction of CPU utilization that can be reclaimed
719 * by the GRUB algorithm.
720 */
97fb7a0a 721 u64 bw_ratio;
aab03e05
DF		 722};
723
c0796298
VG		 724#ifdef CONFIG_FAIR_GROUP_SCHED
725/* An entity is a task if it doesn't "own" a runqueue */
726#define entity_is_task(se) (!se->my_q)
0dacee1b 727
9f683953
VG		 728static inline void se_update_runnable(struct sched_entity *se)
729{
730 if (!entity_is_task(se))
731 se->runnable_weight = se->my_q->h_nr_running;
732}
733
734static inline long se_runnable(struct sched_entity *se)
735{
736 if (entity_is_task(se))
737 return !!se->on_rq;
738 else
739 return se->runnable_weight;
740}
741
c0796298
VG		 742#else
743#define entity_is_task(se) 1
0dacee1b 744
9f683953
VG		 745static inline void se_update_runnable(struct sched_entity *se) {}
746
747static inline long se_runnable(struct sched_entity *se)
748{
749 return !!se->on_rq;
750}
c0796298
VG		 751#endif
752
029632fb 753#ifdef CONFIG_SMP
c0796298
VG		 754/*
755 * XXX we want to get rid of these helpers and use the full load resolution.
756 */
757static inline long se_weight(struct sched_entity *se)
758{
759 return scale_load_down(se->load.weight);
760}
761
029632fb 762
afe06efd
TC		 763static inline bool sched_asym_prefer(int a, int b)
764{
765 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
766}
767
6aa140fa
QP		 768struct perf_domain {
769 struct em_perf_domain *em_pd;
770 struct perf_domain *next;
771 struct rcu_head rcu;
772};
773
630246a0
QP		 774/* Scheduling group status flags */
775#define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
2802bf3c 776#define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
630246a0 777
029632fb
PZ		 778/*
779 * We add the notion of a root-domain which will be used to define per-domain
780 * variables. Each exclusive cpuset essentially defines an island domain by
97fb7a0a 781 * fully partitioning the member CPUs from any other cpuset. Whenever a new
029632fb
PZ		 782 * exclusive cpuset is created, we also create and attach a new root-domain
783 * object.
784 *
785 */
786struct root_domain {
97fb7a0a
IM		 787 atomic_t refcount;
788 atomic_t rto_count;
789 struct rcu_head rcu;
790 cpumask_var_t span;
791 cpumask_var_t online;
029632fb 792
757ffdd7
VS		 793 /*
794 * Indicate pullable load on at least one CPU, e.g:
795 * - More than one runnable task
796 * - Running task is misfit
797 */
575638d1 798 int overload;
4486edd1 799
2802bf3c
MR		 800 /* Indicate one or more cpus over-utilized (tipping point) */
801 int overutilized;
802
1baca4ce
JL		 803 /*
804 * The bit corresponding to a CPU gets set here if such CPU has more
805 * than one runnable -deadline task (as it is below for RT tasks).
806 */
97fb7a0a
IM		 807 cpumask_var_t dlo_mask;
808 atomic_t dlo_count;
809 struct dl_bw dl_bw;
810 struct cpudl cpudl;
1baca4ce 811
26762423
PL		 812 /*
813 * Indicate whether a root_domain's dl_bw has been checked or
814 * updated. It's monotonously increasing value.
815 *
816 * Also, some corner cases, like 'wrap around' is dangerous, but given
817 * that u64 is 'big enough'. So that shouldn't be a concern.
818 */
819 u64 visit_gen;
820
4bdced5c
SRRH		 821#ifdef HAVE_RT_PUSH_IPI
822 /*
823 * For IPI pull requests, loop across the rto_mask.
824 */
97fb7a0a
IM		 825 struct irq_work rto_push_work;
826 raw_spinlock_t rto_lock;
4bdced5c 827 /* These are only updated and read within rto_lock */
97fb7a0a
IM		 828 int rto_loop;
829 int rto_cpu;
4bdced5c 830 /* These atomics are updated outside of a lock */
97fb7a0a
IM		 831 atomic_t rto_loop_next;
832 atomic_t rto_loop_start;
4bdced5c 833#endif
029632fb
PZ		 834 /*
835 * The "RT overload" flag: it gets set if a CPU has more than
836 * one runnable RT task.
837 */
97fb7a0a
IM		 838 cpumask_var_t rto_mask;
839 struct cpupri cpupri;
cd92bfd3 840
97fb7a0a 841 unsigned long max_cpu_capacity;
6aa140fa
QP		 842
843 /*
844 * NULL-terminated list of performance domains intersecting with the
845 * CPUs of the rd. Protected by RCU.
846 */
7ba7319f 847 struct perf_domain __rcu *pd;
029632fb
PZ		 848};
849
f2cb1360 850extern void init_defrootdomain(void);
8d5dc512 851extern int sched_init_domains(const struct cpumask *cpu_map);
f2cb1360 852extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
364f5665
SRV		 853extern void sched_get_rd(struct root_domain *rd);
854extern void sched_put_rd(struct root_domain *rd);
029632fb 855
4bdced5c
SRRH		 856#ifdef HAVE_RT_PUSH_IPI
857extern void rto_push_irq_work_func(struct irq_work *work);
858#endif
029632fb
PZ		 859#endif /* CONFIG_SMP */
860
69842cba
PB		 861#ifdef CONFIG_UCLAMP_TASK
862/*
863 * struct uclamp_bucket - Utilization clamp bucket
864 * @value: utilization clamp value for tasks on this clamp bucket
865 * @tasks: number of RUNNABLE tasks on this clamp bucket
866 *
867 * Keep track of how many tasks are RUNNABLE for a given utilization
868 * clamp value.
869 */
870struct uclamp_bucket {
871 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
872 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
873};
874
875/*
876 * struct uclamp_rq - rq's utilization clamp
877 * @value: currently active clamp values for a rq
878 * @bucket: utilization clamp buckets affecting a rq
879 *
880 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
881 * A clamp value is affecting a rq when there is at least one task RUNNABLE
882 * (or actually running) with that value.
883 *
884 * There are up to UCLAMP_CNT possible different clamp values, currently there
885 * are only two: minimum utilization and maximum utilization.
886 *
887 * All utilization clamping values are MAX aggregated, since:
888 * - for util_min: we want to run the CPU at least at the max of the minimum
889 * utilization required by its currently RUNNABLE tasks.
890 * - for util_max: we want to allow the CPU to run up to the max of the
891 * maximum utilization allowed by its currently RUNNABLE tasks.
892 *
893 * Since on each system we expect only a limited number of different
894 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
895 * the metrics required to compute all the per-rq utilization clamp values.
896 */
897struct uclamp_rq {
898 unsigned int value;
899 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
900};
46609ce2
QY		 901
902DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
69842cba
PB		 903#endif /* CONFIG_UCLAMP_TASK */
904
029632fb
PZ		 905/*
906 * This is the main, per-CPU runqueue data structure.
907 *
908 * Locking rule: those places that want to lock multiple runqueues
909 * (such as the load balancing or the thread migration code), lock
910 * acquire operations must be ordered by ascending &runqueue.
911 */
912struct rq {
913 /* runqueue lock: */
5cb9eaa3 914 raw_spinlock_t __lock;
029632fb
PZ		 915
916 /*
917 * nr_running and cpu_load should be in the same cacheline because
918 * remote CPUs use both these fields when doing load calculation.
919 */
97fb7a0a 920 unsigned int nr_running;
0ec8aa00 921#ifdef CONFIG_NUMA_BALANCING
97fb7a0a
IM		 922 unsigned int nr_numa_running;
923 unsigned int nr_preferred_running;
a4739eca 924 unsigned int numa_migrate_on;
0ec8aa00 925#endif
3451d024 926#ifdef CONFIG_NO_HZ_COMMON
9fd81dd5 927#ifdef CONFIG_SMP
e022e0d3 928 unsigned long last_blocked_load_update_tick;
f643ea22 929 unsigned int has_blocked_load;
90b5363a 930 call_single_data_t nohz_csd;
9fd81dd5 931#endif /* CONFIG_SMP */
00357f5e 932 unsigned int nohz_tick_stopped;
90b5363a 933 atomic_t nohz_flags;
9fd81dd5 934#endif /* CONFIG_NO_HZ_COMMON */
dcdedb24 935
126c2092
PZ		 936#ifdef CONFIG_SMP
937 unsigned int ttwu_pending;
938#endif
97fb7a0a 939 u64 nr_switches;
029632fb 940
69842cba
PB		 941#ifdef CONFIG_UCLAMP_TASK
942 /* Utilization clamp values based on CPU's RUNNABLE tasks */
943 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
e496187d
PB		 944 unsigned int uclamp_flags;
945#define UCLAMP_FLAG_IDLE 0x01
69842cba
PB		 946#endif
947
97fb7a0a
IM		 948 struct cfs_rq cfs;
949 struct rt_rq rt;
950 struct dl_rq dl;
029632fb
PZ		 951
952#ifdef CONFIG_FAIR_GROUP_SCHED
97fb7a0a
IM		 953 /* list of leaf cfs_rq on this CPU: */
954 struct list_head leaf_cfs_rq_list;
955 struct list_head *tmp_alone_branch;
a35b6466
PZ		 956#endif /* CONFIG_FAIR_GROUP_SCHED */
957
029632fb
PZ		 958 /*
959 * This is part of a global counter where only the total sum
960 * over all CPUs matters. A task can increase this counter on
961 * one CPU and if it got migrated afterwards it may decrease
962 * it on another CPU. Always updated under the runqueue lock:
963 */
e6fe3f42 964 unsigned int nr_uninterruptible;
029632fb 965
4104a562 966 struct task_struct __rcu *curr;
97fb7a0a
IM		 967 struct task_struct *idle;
968 struct task_struct *stop;
969 unsigned long next_balance;
970 struct mm_struct *prev_mm;
029632fb 971
97fb7a0a
IM		 972 unsigned int clock_update_flags;
973 u64 clock;
23127296
VG		 974 /* Ensure that all clocks are in the same cache line */
975 u64 clock_task ____cacheline_aligned;
976 u64 clock_pelt;
977 unsigned long lost_idle_time;
029632fb 978
97fb7a0a 979 atomic_t nr_iowait;
029632fb 980
c006fac5
PT		 981#ifdef CONFIG_SCHED_DEBUG
982 u64 last_seen_need_resched_ns;
983 int ticks_without_resched;
984#endif
985
227a4aad
MD		 986#ifdef CONFIG_MEMBARRIER
987 int membarrier_state;
988#endif
989
029632fb 990#ifdef CONFIG_SMP
994aeb7a
JFG		 991 struct root_domain *rd;
992 struct sched_domain __rcu *sd;
97fb7a0a
IM		 993
994 unsigned long cpu_capacity;
995 unsigned long cpu_capacity_orig;
029632fb 996
97fb7a0a 997 struct callback_head *balance_callback;
029632fb 998
19a1f5ec 999 unsigned char nohz_idle_balance;
97fb7a0a 1000 unsigned char idle_balance;
e3fca9e7 1001
3b1baa64
MR		 1002 unsigned long misfit_task_load;
1003
029632fb 1004 /* For active balancing */
97fb7a0a
IM		 1005 int active_balance;
1006 int push_cpu;
1007 struct cpu_stop_work active_balance_work;
1008
1009 /* CPU of this runqueue: */
1010 int cpu;
1011 int online;
029632fb 1012
367456c7
PZ		 1013 struct list_head cfs_tasks;
1014
371bf427 1015 struct sched_avg avg_rt;
3727e0e1 1016 struct sched_avg avg_dl;
11d4afd4 1017#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
91c27493 1018 struct sched_avg avg_irq;
76504793
TG		 1019#endif
1020#ifdef CONFIG_SCHED_THERMAL_PRESSURE
1021 struct sched_avg avg_thermal;
91c27493 1022#endif
97fb7a0a
IM		 1023 u64 idle_stamp;
1024 u64 avg_idle;
9bd721c5 1025
94aafc3e
PZ		 1026 unsigned long wake_stamp;
1027 u64 wake_avg_idle;
1028
9bd721c5 1029 /* This is used to determine avg_idle's max value */
97fb7a0a 1030 u64 max_idle_balance_cost;
f2469a1f
TG		 1031
1032#ifdef CONFIG_HOTPLUG_CPU
1033 struct rcuwait hotplug_wait;
1034#endif
90b5363a 1035#endif /* CONFIG_SMP */
029632fb
PZ		 1036
1037#ifdef CONFIG_IRQ_TIME_ACCOUNTING
97fb7a0a 1038 u64 prev_irq_time;
029632fb
PZ		 1039#endif
1040#ifdef CONFIG_PARAVIRT
97fb7a0a 1041 u64 prev_steal_time;
029632fb
PZ		 1042#endif
1043#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
97fb7a0a 1044 u64 prev_steal_time_rq;
029632fb
PZ		 1045#endif
1046
1047 /* calc_load related fields */
97fb7a0a
IM		 1048 unsigned long calc_load_update;
1049 long calc_load_active;
029632fb
PZ		 1050
1051#ifdef CONFIG_SCHED_HRTICK
1052#ifdef CONFIG_SMP
97fb7a0a 1053 call_single_data_t hrtick_csd;
029632fb 1054#endif
97fb7a0a 1055 struct hrtimer hrtick_timer;
156ec6f4 1056 ktime_t hrtick_time;
029632fb
PZ		 1057#endif
1058
1059#ifdef CONFIG_SCHEDSTATS
1060 /* latency stats */
97fb7a0a
IM		 1061 struct sched_info rq_sched_info;
1062 unsigned long long rq_cpu_time;
029632fb
PZ		 1063 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1064
1065 /* sys_sched_yield() stats */
97fb7a0a 1066 unsigned int yld_count;
029632fb
PZ		 1067
1068 /* schedule() stats */
97fb7a0a
IM		 1069 unsigned int sched_count;
1070 unsigned int sched_goidle;
029632fb
PZ		 1071
1072 /* try_to_wake_up() stats */
97fb7a0a
IM		 1073 unsigned int ttwu_count;
1074 unsigned int ttwu_local;
029632fb
PZ		 1075#endif
1076
442bf3aa
DL		 1077#ifdef CONFIG_CPU_IDLE
1078 /* Must be inspected within a rcu lock section */
97fb7a0a 1079 struct cpuidle_state *idle_state;
442bf3aa 1080#endif
3015ef4b 1081
74d862b6 1082#ifdef CONFIG_SMP
3015ef4b
TG		 1083 unsigned int nr_pinned;
1084#endif
a7c81556
PZ		 1085 unsigned int push_busy;
1086 struct cpu_stop_work push_work;
9edeaea1
PZ		 1087
1088#ifdef CONFIG_SCHED_CORE
1089 /* per rq */
1090 struct rq *core;
539f6512 1091 struct task_struct *core_pick;
9edeaea1 1092 unsigned int core_enabled;
539f6512 1093 unsigned int core_sched_seq;
8a311c74
PZ		 1094 struct rb_root core_tree;
1095
1096 /* shared state */
1097 unsigned int core_task_seq;
539f6512
PZ		 1098 unsigned int core_pick_seq;
1099 unsigned long core_cookie;
8039e96f 1100 unsigned char core_forceidle;
c6047c2e 1101 unsigned int core_forceidle_seq;
9edeaea1 1102#endif
029632fb
PZ		 1103};
1104
62478d99
VG		 1105#ifdef CONFIG_FAIR_GROUP_SCHED
1106
1107/* CPU runqueue to which this cfs_rq is attached */
1108static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1109{
1110 return cfs_rq->rq;
1111}
1112
1113#else
1114
1115static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1116{
1117 return container_of(cfs_rq, struct rq, cfs);
1118}
1119#endif
1120
029632fb
PZ		 1121static inline int cpu_of(struct rq *rq)
1122{
1123#ifdef CONFIG_SMP
1124 return rq->cpu;
1125#else
1126 return 0;
1127#endif
1128}
1129
a7c81556
PZ		 1130#define MDF_PUSH 0x01
1131
1132static inline bool is_migration_disabled(struct task_struct *p)
1133{
74d862b6 1134#ifdef CONFIG_SMP
a7c81556
PZ		 1135 return p->migration_disabled;
1136#else
1137 return false;
1138#endif
1139}
1b568f0a 1140
97886d9d 1141struct sched_group;
9edeaea1 1142#ifdef CONFIG_SCHED_CORE
97886d9d 1143static inline struct cpumask *sched_group_span(struct sched_group *sg);
9edeaea1
PZ		 1144
1145DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1146
1147static inline bool sched_core_enabled(struct rq *rq)
1148{
1149 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1150}
1151
1152static inline bool sched_core_disabled(void)
1153{
1154 return !static_branch_unlikely(&__sched_core_enabled);
1155}
1156
9ef7e7e3
PZ		 1157/*
1158 * Be careful with this function; not for general use. The return value isn't
1159 * stable unless you actually hold a relevant rq->__lock.
1160 */
9edeaea1
PZ		 1161static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1162{
1163 if (sched_core_enabled(rq))
1164 return &rq->core->__lock;
1165
1166 return &rq->__lock;
1167}
1168
9ef7e7e3
PZ		 1169static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1170{
1171 if (rq->core_enabled)
1172 return &rq->core->__lock;
1173
1174 return &rq->__lock;
1175}
1176
c6047c2e
JFG		 1177bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1178
97886d9d
AL		 1179/*
1180 * Helpers to check if the CPU's core cookie matches with the task's cookie
1181 * when core scheduling is enabled.
1182 * A special case is that the task's cookie always matches with CPU's core
1183 * cookie if the CPU is in an idle core.
1184 */
1185static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1186{
1187 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1188 if (!sched_core_enabled(rq))
1189 return true;
1190
1191 return rq->core->core_cookie == p->core_cookie;
1192}
1193
1194static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1195{
1196 bool idle_core = true;
1197 int cpu;
1198
1199 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1200 if (!sched_core_enabled(rq))
1201 return true;
1202
1203 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1204 if (!available_idle_cpu(cpu)) {
1205 idle_core = false;
1206 break;
1207 }
1208 }
1209
1210 /*
1211 * A CPU in an idle core is always the best choice for tasks with
1212 * cookies.
1213 */
1214 return idle_core || rq->core->core_cookie == p->core_cookie;
1215}
1216
1217static inline bool sched_group_cookie_match(struct rq *rq,
1218 struct task_struct *p,
1219 struct sched_group *group)
1220{
1221 int cpu;
1222
1223 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1224 if (!sched_core_enabled(rq))
1225 return true;
1226
1227 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1228 if (sched_core_cookie_match(rq, p))
1229 return true;
1230 }
1231 return false;
1232}
1233
d2dfa17b
PZ		 1234extern void queue_core_balance(struct rq *rq);
1235
6e33cad0
PZ		 1236static inline bool sched_core_enqueued(struct task_struct *p)
1237{
1238 return !RB_EMPTY_NODE(&p->core_node);
1239}
1240
1241extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1242extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1243
1244extern void sched_core_get(void);
1245extern void sched_core_put(void);
1246
1247extern unsigned long sched_core_alloc_cookie(void);
1248extern void sched_core_put_cookie(unsigned long cookie);
1249extern unsigned long sched_core_get_cookie(unsigned long cookie);
1250extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1251
9edeaea1
PZ		 1252#else /* !CONFIG_SCHED_CORE */
1253
1254static inline bool sched_core_enabled(struct rq *rq)
1255{
1256 return false;
1257}
1258
d66f1b06
PZ		 1259static inline bool sched_core_disabled(void)
1260{
1261 return true;
1262}
1263
39d371b7
PZ		 1264static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1265{
5cb9eaa3 1266 return &rq->__lock;
39d371b7
PZ		 1267}
1268
9ef7e7e3
PZ		 1269static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1270{
1271 return &rq->__lock;
1272}
1273
d2dfa17b
PZ		 1274static inline void queue_core_balance(struct rq *rq)
1275{
1276}
1277
97886d9d
AL		 1278static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1279{
1280 return true;
1281}
1282
1283static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1284{
1285 return true;
1286}
1287
1288static inline bool sched_group_cookie_match(struct rq *rq,
1289 struct task_struct *p,
1290 struct sched_group *group)
1291{
1292 return true;
1293}
9edeaea1
PZ		 1294#endif /* CONFIG_SCHED_CORE */
1295
39d371b7
PZ		 1296static inline void lockdep_assert_rq_held(struct rq *rq)
1297{
9ef7e7e3 1298 lockdep_assert_held(__rq_lockp(rq));
39d371b7
PZ		 1299}
1300
1301extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1302extern bool raw_spin_rq_trylock(struct rq *rq);
1303extern void raw_spin_rq_unlock(struct rq *rq);
1304
1305static inline void raw_spin_rq_lock(struct rq *rq)
1306{
1307 raw_spin_rq_lock_nested(rq, 0);
1308}
1309
1310static inline void raw_spin_rq_lock_irq(struct rq *rq)
1311{
1312 local_irq_disable();
1313 raw_spin_rq_lock(rq);
1314}
1315
1316static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1317{
1318 raw_spin_rq_unlock(rq);
1319 local_irq_enable();
1320}
1321
1322static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1323{
1324 unsigned long flags;
1325 local_irq_save(flags);
1326 raw_spin_rq_lock(rq);
1327 return flags;
1328}
1329
1330static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1331{
1332 raw_spin_rq_unlock(rq);
1333 local_irq_restore(flags);
1334}
1335
1336#define raw_spin_rq_lock_irqsave(rq, flags) \
1337do { \
1338 flags = _raw_spin_rq_lock_irqsave(rq); \
1339} while (0)
1340
1b568f0a 1341#ifdef CONFIG_SCHED_SMT
1b568f0a
PZ		 1342extern void __update_idle_core(struct rq *rq);
1343
1344static inline void update_idle_core(struct rq *rq)
1345{
1346 if (static_branch_unlikely(&sched_smt_present))
1347 __update_idle_core(rq);
1348}
1349
1350#else
1351static inline void update_idle_core(struct rq *rq) { }
1352#endif
1353
8b06c55b 1354DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
029632fb 1355
518cd623 1356#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
4a32fea9 1357#define this_rq() this_cpu_ptr(&runqueues)
518cd623
PZ		 1358#define task_rq(p) cpu_rq(task_cpu(p))
1359#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
4a32fea9 1360#define raw_rq() raw_cpu_ptr(&runqueues)
518cd623 1361
8a311c74
PZ		 1362#ifdef CONFIG_FAIR_GROUP_SCHED
1363static inline struct task_struct *task_of(struct sched_entity *se)
1364{
1365 SCHED_WARN_ON(!entity_is_task(se));
1366 return container_of(se, struct task_struct, se);
1367}
1368
1369static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1370{
1371 return p->se.cfs_rq;
1372}
1373
1374/* runqueue on which this entity is (to be) queued */
1375static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1376{
1377 return se->cfs_rq;
1378}
1379
1380/* runqueue "owned" by this group */
1381static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1382{
1383 return grp->my_q;
1384}
1385
1386#else
1387
1388static inline struct task_struct *task_of(struct sched_entity *se)
1389{
1390 return container_of(se, struct task_struct, se);
1391}
1392
1393static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1394{
1395 return &task_rq(p)->cfs;
1396}
1397
1398static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1399{
1400 struct task_struct *p = task_of(se);
1401 struct rq *rq = task_rq(p);
1402
1403 return &rq->cfs;
1404}
1405
1406/* runqueue "owned" by this group */
1407static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1408{
1409 return NULL;
1410}
1411#endif
1412
1f351d7f
JW		 1413extern void update_rq_clock(struct rq *rq);
1414
cebde6d6
PZ		 1415static inline u64 __rq_clock_broken(struct rq *rq)
1416{
316c1608 1417 return READ_ONCE(rq->clock);
cebde6d6
PZ		 1418}
1419
cb42c9a3
MF		 1420/*
1421 * rq::clock_update_flags bits
1422 *
1423 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1424 * call to __schedule(). This is an optimisation to avoid
1425 * neighbouring rq clock updates.
1426 *
1427 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1428 * in effect and calls to update_rq_clock() are being ignored.
1429 *
1430 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1431 * made to update_rq_clock() since the last time rq::lock was pinned.
1432 *
1433 * If inside of __schedule(), clock_update_flags will have been
1434 * shifted left (a left shift is a cheap operation for the fast path
1435 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1436 *
1437 * if (rq-clock_update_flags >= RQCF_UPDATED)
1438 *
3b03706f 1439 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
cb42c9a3
MF		 1440 * one position though, because the next rq_unpin_lock() will shift it
1441 * back.
1442 */
97fb7a0a
IM		 1443#define RQCF_REQ_SKIP 0x01
1444#define RQCF_ACT_SKIP 0x02
1445#define RQCF_UPDATED 0x04
cb42c9a3
MF		 1446
1447static inline void assert_clock_updated(struct rq *rq)
1448{
1449 /*
1450 * The only reason for not seeing a clock update since the
1451 * last rq_pin_lock() is if we're currently skipping updates.
1452 */
1453 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1454}
1455
78becc27
FW		 1456static inline u64 rq_clock(struct rq *rq)
1457{
5cb9eaa3 1458 lockdep_assert_rq_held(rq);
cb42c9a3
MF		 1459 assert_clock_updated(rq);
1460
78becc27
FW		 1461 return rq->clock;
1462}
1463
1464static inline u64 rq_clock_task(struct rq *rq)
1465{
5cb9eaa3 1466 lockdep_assert_rq_held(rq);
cb42c9a3
MF		 1467 assert_clock_updated(rq);
1468
78becc27
FW		 1469 return rq->clock_task;
1470}
1471
05289b90
TG		 1472/**
1473 * By default the decay is the default pelt decay period.
1474 * The decay shift can change the decay period in
1475 * multiples of 32.
1476 * Decay shift Decay period(ms)
1477 * 0 32
1478 * 1 64
1479 * 2 128
1480 * 3 256
1481 * 4 512
1482 */
1483extern int sched_thermal_decay_shift;
1484
1485static inline u64 rq_clock_thermal(struct rq *rq)
1486{
1487 return rq_clock_task(rq) >> sched_thermal_decay_shift;
1488}
1489
adcc8da8 1490static inline void rq_clock_skip_update(struct rq *rq)
9edfbfed 1491{
5cb9eaa3 1492 lockdep_assert_rq_held(rq);
adcc8da8
DB		 1493 rq->clock_update_flags |= RQCF_REQ_SKIP;
1494}
1495
1496/*
595058b6 1497 * See rt task throttling, which is the only time a skip
3b03706f 1498 * request is canceled.
adcc8da8
DB		 1499 */
1500static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1501{
5cb9eaa3 1502 lockdep_assert_rq_held(rq);
adcc8da8 1503 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
9edfbfed
PZ		 1504}
1505
d8ac8971
MF		 1506struct rq_flags {
1507 unsigned long flags;
1508 struct pin_cookie cookie;
cb42c9a3
MF		 1509#ifdef CONFIG_SCHED_DEBUG
1510 /*
1511 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1512 * current pin context is stashed here in case it needs to be
1513 * restored in rq_repin_lock().
1514 */
1515 unsigned int clock_update_flags;
1516#endif
d8ac8971
MF		 1517};
1518
ae792702
PZ		 1519extern struct callback_head balance_push_callback;
1520
58877d34
PZ		 1521/*
1522 * Lockdep annotation that avoids accidental unlocks; it's like a
1523 * sticky/continuous lockdep_assert_held().
1524 *
1525 * This avoids code that has access to 'struct rq *rq' (basically everything in
1526 * the scheduler) from accidentally unlocking the rq if they do not also have a
1527 * copy of the (on-stack) 'struct rq_flags rf'.
1528 *
1529 * Also see Documentation/locking/lockdep-design.rst.
1530 */
d8ac8971
MF		 1531static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1532{
9ef7e7e3 1533 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
cb42c9a3
MF		 1534
1535#ifdef CONFIG_SCHED_DEBUG
1536 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1537 rf->clock_update_flags = 0;
565790d2 1538#ifdef CONFIG_SMP
ae792702
PZ		 1539 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1540#endif
565790d2 1541#endif
d8ac8971
MF		 1542}
1543
1544static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1545{
cb42c9a3
MF		 1546#ifdef CONFIG_SCHED_DEBUG
1547 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1548 rf->clock_update_flags = RQCF_UPDATED;
1549#endif
1550
9ef7e7e3 1551 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
d8ac8971
MF		 1552}
1553
1554static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1555{
9ef7e7e3 1556 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
cb42c9a3
MF		 1557
1558#ifdef CONFIG_SCHED_DEBUG
1559 /*
1560 * Restore the value we stashed in @rf for this pin context.
1561 */
1562 rq->clock_update_flags |= rf->clock_update_flags;
1563#endif
d8ac8971
MF		 1564}
1565
1f351d7f
JW		 1566struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1567 __acquires(rq->lock);
1568
1569struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1570 __acquires(p->pi_lock)
1571 __acquires(rq->lock);
1572
1573static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1574 __releases(rq->lock)
1575{
1576 rq_unpin_lock(rq, rf);
5cb9eaa3 1577 raw_spin_rq_unlock(rq);
1f351d7f
JW		 1578}
1579
1580static inline void
1581task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1582 __releases(rq->lock)
1583 __releases(p->pi_lock)
1584{
1585 rq_unpin_lock(rq, rf);
5cb9eaa3 1586 raw_spin_rq_unlock(rq);
1f351d7f
JW		 1587 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1588}
1589
1590static inline void
1591rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1592 __acquires(rq->lock)
1593{
5cb9eaa3 1594 raw_spin_rq_lock_irqsave(rq, rf->flags);
1f351d7f
JW		 1595 rq_pin_lock(rq, rf);
1596}
1597
1598static inline void
1599rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1600 __acquires(rq->lock)
1601{
5cb9eaa3 1602 raw_spin_rq_lock_irq(rq);
1f351d7f
JW		 1603 rq_pin_lock(rq, rf);
1604}
1605
1606static inline void
1607rq_lock(struct rq *rq, struct rq_flags *rf)
1608 __acquires(rq->lock)
1609{
5cb9eaa3 1610 raw_spin_rq_lock(rq);
1f351d7f
JW		 1611 rq_pin_lock(rq, rf);
1612}
1613
1614static inline void
1615rq_relock(struct rq *rq, struct rq_flags *rf)
1616 __acquires(rq->lock)
1617{
5cb9eaa3 1618 raw_spin_rq_lock(rq);
1f351d7f
JW		 1619 rq_repin_lock(rq, rf);
1620}
1621
1622static inline void
1623rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1624 __releases(rq->lock)
1625{
1626 rq_unpin_lock(rq, rf);
5cb9eaa3 1627 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1f351d7f
JW		 1628}
1629
1630static inline void
1631rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1632 __releases(rq->lock)
1633{
1634 rq_unpin_lock(rq, rf);
5cb9eaa3 1635 raw_spin_rq_unlock_irq(rq);
1f351d7f
JW		 1636}
1637
1638static inline void
1639rq_unlock(struct rq *rq, struct rq_flags *rf)
1640 __releases(rq->lock)
1641{
1642 rq_unpin_lock(rq, rf);
5cb9eaa3 1643 raw_spin_rq_unlock(rq);
1f351d7f
JW		 1644}
1645
246b3b33
JW		 1646static inline struct rq *
1647this_rq_lock_irq(struct rq_flags *rf)
1648 __acquires(rq->lock)
1649{
1650 struct rq *rq;
1651
1652 local_irq_disable();
1653 rq = this_rq();
1654 rq_lock(rq, rf);
1655 return rq;
1656}
1657
9942f79b 1658#ifdef CONFIG_NUMA
e3fe70b1
RR		 1659enum numa_topology_type {
1660 NUMA_DIRECT,
1661 NUMA_GLUELESS_MESH,
1662 NUMA_BACKPLANE,
1663};
1664extern enum numa_topology_type sched_numa_topology_type;
9942f79b
RR		 1665extern int sched_max_numa_distance;
1666extern bool find_numa_distance(int distance);
f2cb1360
IM		 1667extern void sched_init_numa(void);
1668extern void sched_domains_numa_masks_set(unsigned int cpu);
1669extern void sched_domains_numa_masks_clear(unsigned int cpu);
e0e8d491 1670extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
f2cb1360
IM		 1671#else
1672static inline void sched_init_numa(void) { }
1673static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1674static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
e0e8d491
WL		 1675static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1676{
1677 return nr_cpu_ids;
1678}
f2cb1360
IM		 1679#endif
1680
f809ca9a 1681#ifdef CONFIG_NUMA_BALANCING
44dba3d5
IM		 1682/* The regions in numa_faults array from task_struct */
1683enum numa_faults_stats {
1684 NUMA_MEM = 0,
1685 NUMA_CPU,
1686 NUMA_MEMBUF,
1687 NUMA_CPUBUF
1688};
0ec8aa00 1689extern void sched_setnuma(struct task_struct *p, int node);
e6628d5b 1690extern int migrate_task_to(struct task_struct *p, int cpu);
0ad4e3df
SD		 1691extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1692 int cpu, int scpu);
13784475
MG		 1693extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1694#else
1695static inline void
1696init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1697{
1698}
f809ca9a
MG		 1699#endif /* CONFIG_NUMA_BALANCING */
1700
518cd623
PZ		 1701#ifdef CONFIG_SMP
1702
e3fca9e7
PZ		 1703static inline void
1704queue_balance_callback(struct rq *rq,
1705 struct callback_head *head,
1706 void (*func)(struct rq *rq))
1707{
5cb9eaa3 1708 lockdep_assert_rq_held(rq);
e3fca9e7 1709
ae792702 1710 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
e3fca9e7
PZ		 1711 return;
1712
1713 head->func = (void (*)(struct callback_head *))func;
1714 head->next = rq->balance_callback;
1715 rq->balance_callback = head;
1716}
1717
029632fb
PZ		 1718#define rcu_dereference_check_sched_domain(p) \
1719 rcu_dereference_check((p), \
1720 lockdep_is_held(&sched_domains_mutex))
1721
1722/*
1723 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
337e9b07 1724 * See destroy_sched_domains: call_rcu for details.
029632fb
PZ		 1725 *
1726 * The domain tree of any CPU may only be accessed from within
1727 * preempt-disabled sections.
1728 */
1729#define for_each_domain(cpu, __sd) \
518cd623
PZ		 1730 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1731 __sd; __sd = __sd->parent)
029632fb 1732
518cd623
PZ		 1733/**
1734 * highest_flag_domain - Return highest sched_domain containing flag.
97fb7a0a 1735 * @cpu: The CPU whose highest level of sched domain is to
518cd623
PZ		 1736 * be returned.
1737 * @flag: The flag to check for the highest sched_domain
97fb7a0a 1738 * for the given CPU.
518cd623 1739 *
97fb7a0a 1740 * Returns the highest sched_domain of a CPU which contains the given flag.
518cd623
PZ		 1741 */
1742static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1743{
1744 struct sched_domain *sd, *hsd = NULL;
1745
1746 for_each_domain(cpu, sd) {
1747 if (!(sd->flags & flag))
1748 break;
1749 hsd = sd;
1750 }
1751
1752 return hsd;
1753}
1754
fb13c7ee
MG		 1755static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1756{
1757 struct sched_domain *sd;
1758
1759 for_each_domain(cpu, sd) {
1760 if (sd->flags & flag)
1761 break;
1762 }
1763
1764 return sd;
1765}
1766
994aeb7a 1767DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
7d9ffa89 1768DECLARE_PER_CPU(int, sd_llc_size);
518cd623 1769DECLARE_PER_CPU(int, sd_llc_id);
994aeb7a
JFG		 1770DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1771DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1772DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1773DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
df054e84 1774extern struct static_key_false sched_asym_cpucapacity;
518cd623 1775
63b2ca30 1776struct sched_group_capacity {
97fb7a0a 1777 atomic_t ref;
5e6521ea 1778 /*
172895e6 1779 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
63b2ca30 1780 * for a single CPU.
5e6521ea 1781 */
97fb7a0a
IM		 1782 unsigned long capacity;
1783 unsigned long min_capacity; /* Min per-CPU capacity in group */
e3d6d0cb 1784 unsigned long max_capacity; /* Max per-CPU capacity in group */
97fb7a0a
IM		 1785 unsigned long next_update;
1786 int imbalance; /* XXX unrelated to capacity but shared group state */
5e6521ea 1787
005f874d 1788#ifdef CONFIG_SCHED_DEBUG
97fb7a0a 1789 int id;
005f874d
PZ		 1790#endif
1791
eba9f082 1792 unsigned long cpumask[]; /* Balance mask */
5e6521ea
LZ		 1793};
1794
1795struct sched_group {
97fb7a0a
IM		 1796 struct sched_group *next; /* Must be a circular list */
1797 atomic_t ref;
5e6521ea 1798
97fb7a0a 1799 unsigned int group_weight;
63b2ca30 1800 struct sched_group_capacity *sgc;
97fb7a0a 1801 int asym_prefer_cpu; /* CPU of highest priority in group */
5e6521ea
LZ		 1802
1803 /*
1804 * The CPUs this group covers.
1805 *
1806 * NOTE: this field is variable length. (Allocated dynamically
1807 * by attaching extra space to the end of the structure,
1808 * depending on how many CPUs the kernel has booted up with)
1809 */
04f5c362 1810 unsigned long cpumask[];
5e6521ea
LZ		 1811};
1812
ae4df9d6 1813static inline struct cpumask *sched_group_span(struct sched_group *sg)
5e6521ea
LZ		 1814{
1815 return to_cpumask(sg->cpumask);
1816}
1817
1818/*
e5c14b1f 1819 * See build_balance_mask().
5e6521ea 1820 */
e5c14b1f 1821static inline struct cpumask *group_balance_mask(struct sched_group *sg)
5e6521ea 1822{
63b2ca30 1823 return to_cpumask(sg->sgc->cpumask);
5e6521ea
LZ		 1824}
1825
1826/**
97fb7a0a
IM		 1827 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1828 * @group: The group whose first CPU is to be returned.
5e6521ea
LZ		 1829 */
1830static inline unsigned int group_first_cpu(struct sched_group *group)
1831{
ae4df9d6 1832 return cpumask_first(sched_group_span(group));
5e6521ea
LZ		 1833}
1834
c1174876
PZ		 1835extern int group_balance_cpu(struct sched_group *sg);
1836
3b87f136
PZ		 1837#ifdef CONFIG_SCHED_DEBUG
1838void update_sched_domain_debugfs(void);
bbdacdfe 1839void dirty_sched_domain_sysctl(int cpu);
3866e845 1840#else
3b87f136 1841static inline void update_sched_domain_debugfs(void)
3866e845
SRRH		 1842{
1843}
bbdacdfe
PZ		 1844static inline void dirty_sched_domain_sysctl(int cpu)
1845{
1846}
3866e845
SRRH		 1847#endif
1848
8a99b683
PZ		 1849extern int sched_update_scaling(void);
1850
b2a02fc4 1851extern void flush_smp_call_function_from_idle(void);
e3baac47 1852
b2a02fc4
PZ		 1853#else /* !CONFIG_SMP: */
1854static inline void flush_smp_call_function_from_idle(void) { }
b2a02fc4 1855#endif
029632fb 1856
391e43da 1857#include "stats.h"
1051408f 1858#include "autogroup.h"
029632fb
PZ		 1859
1860#ifdef CONFIG_CGROUP_SCHED
1861
1862/*
1863 * Return the group to which this tasks belongs.
1864 *
8af01f56
TH		 1865 * We cannot use task_css() and friends because the cgroup subsystem
1866 * changes that value before the cgroup_subsys::attach() method is called,
1867 * therefore we cannot pin it and might observe the wrong value.
8323f26c
PZ		 1868 *
1869 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1870 * core changes this before calling sched_move_task().
1871 *
1872 * Instead we use a 'copy' which is updated from sched_move_task() while
1873 * holding both task_struct::pi_lock and rq::lock.
029632fb
PZ		 1874 */
1875static inline struct task_group *task_group(struct task_struct *p)
1876{
8323f26c 1877 return p->sched_task_group;
029632fb
PZ		 1878}
1879
1880/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1881static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1882{
1883#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1884 struct task_group *tg = task_group(p);
1885#endif
1886
1887#ifdef CONFIG_FAIR_GROUP_SCHED
ad936d86 1888 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
029632fb
PZ		 1889 p->se.cfs_rq = tg->cfs_rq[cpu];
1890 p->se.parent = tg->se[cpu];
1891#endif
1892
1893#ifdef CONFIG_RT_GROUP_SCHED
1894 p->rt.rt_rq = tg->rt_rq[cpu];
1895 p->rt.parent = tg->rt_se[cpu];
1896#endif
1897}
1898
1899#else /* CONFIG_CGROUP_SCHED */
1900
1901static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1902static inline struct task_group *task_group(struct task_struct *p)
1903{
1904 return NULL;
1905}
1906
1907#endif /* CONFIG_CGROUP_SCHED */
1908
1909static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1910{
1911 set_task_rq(p, cpu);
1912#ifdef CONFIG_SMP
1913 /*
1914 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
dfcb245e 1915 * successfully executed on another CPU. We must ensure that updates of
029632fb
PZ		 1916 * per-task data have been completed by this moment.
1917 */
1918 smp_wmb();
c65eacbe 1919#ifdef CONFIG_THREAD_INFO_IN_TASK
c546951d 1920 WRITE_ONCE(p->cpu, cpu);
c65eacbe 1921#else
c546951d 1922 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
c65eacbe 1923#endif
ac66f547 1924 p->wake_cpu = cpu;
029632fb
PZ		 1925#endif
1926}
1927
1928/*
1929 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1930 */
1931#ifdef CONFIG_SCHED_DEBUG
c5905afb 1932# include <linux/static_key.h>
029632fb
PZ		 1933# define const_debug __read_mostly
1934#else
1935# define const_debug const
1936#endif
1937
029632fb
PZ		 1938#define SCHED_FEAT(name, enabled) \
1939 __SCHED_FEAT_##name ,
1940
1941enum {
391e43da 1942#include "features.h"
f8b6d1cc 1943 __SCHED_FEAT_NR,
029632fb
PZ		 1944};
1945
1946#undef SCHED_FEAT
1947
a73f863a 1948#ifdef CONFIG_SCHED_DEBUG
765cc3a4
PB		 1949
1950/*
1951 * To support run-time toggling of sched features, all the translation units
1952 * (but core.c) reference the sysctl_sched_features defined in core.c.
1953 */
1954extern const_debug unsigned int sysctl_sched_features;
1955
a73f863a 1956#ifdef CONFIG_JUMP_LABEL
f8b6d1cc 1957#define SCHED_FEAT(name, enabled) \
c5905afb 1958static __always_inline bool static_branch_##name(struct static_key *key) \
f8b6d1cc 1959{ \
6e76ea8a 1960 return static_key_##enabled(key); \
f8b6d1cc
PZ		 1961}
1962
1963#include "features.h"
f8b6d1cc
PZ		 1964#undef SCHED_FEAT
1965
c5905afb 1966extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
f8b6d1cc 1967#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
765cc3a4 1968
a73f863a
JL		 1969#else /* !CONFIG_JUMP_LABEL */
1970
1971#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1972
1973#endif /* CONFIG_JUMP_LABEL */
1974
1975#else /* !SCHED_DEBUG */
765cc3a4
PB		 1976
1977/*
1978 * Each translation unit has its own copy of sysctl_sched_features to allow
1979 * constants propagation at compile time and compiler optimization based on
1980 * features default.
1981 */
1982#define SCHED_FEAT(name, enabled) \
1983 (1UL << __SCHED_FEAT_##name) * enabled |
1984static const_debug __maybe_unused unsigned int sysctl_sched_features =
1985#include "features.h"
1986 0;
1987#undef SCHED_FEAT
1988
7e6f4c5d 1989#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
765cc3a4 1990
a73f863a 1991#endif /* SCHED_DEBUG */
029632fb 1992
2a595721 1993extern struct static_key_false sched_numa_balancing;
cb251765 1994extern struct static_key_false sched_schedstats;
cbee9f88 1995
029632fb
PZ		 1996static inline u64 global_rt_period(void)
1997{
1998 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1999}
2000
2001static inline u64 global_rt_runtime(void)
2002{
2003 if (sysctl_sched_rt_runtime < 0)
2004 return RUNTIME_INF;
2005
2006 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2007}
2008
029632fb
PZ		 2009static inline int task_current(struct rq *rq, struct task_struct *p)
2010{
2011 return rq->curr == p;
2012}
2013
2014static inline int task_running(struct rq *rq, struct task_struct *p)
2015{
2016#ifdef CONFIG_SMP
2017 return p->on_cpu;
2018#else
2019 return task_current(rq, p);
2020#endif
2021}
2022
da0c1e65
KT		 2023static inline int task_on_rq_queued(struct task_struct *p)
2024{
2025 return p->on_rq == TASK_ON_RQ_QUEUED;
2026}
029632fb 2027
cca26e80
KT		 2028static inline int task_on_rq_migrating(struct task_struct *p)
2029{
c546951d 2030 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
cca26e80
KT		 2031}
2032
17770579
VS		 2033/* Wake flags. The first three directly map to some SD flag value */
2034#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2035#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2036#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2037
2038#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2039#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2040#define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2041
2042#ifdef CONFIG_SMP
2043static_assert(WF_EXEC == SD_BALANCE_EXEC);
2044static_assert(WF_FORK == SD_BALANCE_FORK);
2045static_assert(WF_TTWU == SD_BALANCE_WAKE);
2046#endif
b13095f0 2047
029632fb
PZ		 2048/*
2049 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2050 * of tasks with abnormal "nice" values across CPUs the contribution that
2051 * each task makes to its run queue's load is weighted according to its
2052 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2053 * scaled version of the new time slice allocation that they receive on time
2054 * slice expiry etc.
2055 */
2056
97fb7a0a
IM		 2057#define WEIGHT_IDLEPRIO 3
2058#define WMULT_IDLEPRIO 1431655765
029632fb 2059
97fb7a0a
IM		 2060extern const int sched_prio_to_weight[40];
2061extern const u32 sched_prio_to_wmult[40];
029632fb 2062
ff77e468
PZ		 2063/*
2064 * {de,en}queue flags:
2065 *
2066 * DEQUEUE_SLEEP - task is no longer runnable
2067 * ENQUEUE_WAKEUP - task just became runnable
2068 *
2069 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2070 * are in a known state which allows modification. Such pairs
2071 * should preserve as much state as possible.
2072 *
2073 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2074 * in the runqueue.
2075 *
2076 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2077 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
59efa0ba 2078 * ENQUEUE_MIGRATED - the task was migrated during wakeup
ff77e468
PZ		 2079 *
2080 */
2081
2082#define DEQUEUE_SLEEP 0x01
97fb7a0a
IM		 2083#define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2084#define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2085#define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
ff77e468 2086
1de64443 2087#define ENQUEUE_WAKEUP 0x01
ff77e468
PZ		 2088#define ENQUEUE_RESTORE 0x02
2089#define ENQUEUE_MOVE 0x04
0a67d1ee 2090#define ENQUEUE_NOCLOCK 0x08
ff77e468 2091
0a67d1ee
PZ		 2092#define ENQUEUE_HEAD 0x10
2093#define ENQUEUE_REPLENISH 0x20
c82ba9fa 2094#ifdef CONFIG_SMP
0a67d1ee 2095#define ENQUEUE_MIGRATED 0x40
c82ba9fa 2096#else
59efa0ba 2097#define ENQUEUE_MIGRATED 0x00
c82ba9fa 2098#endif
c82ba9fa 2099
37e117c0
PZ		 2100#define RETRY_TASK ((void *)-1UL)
2101
c82ba9fa 2102struct sched_class {
c82ba9fa 2103
69842cba
PB		 2104#ifdef CONFIG_UCLAMP_TASK
2105 int uclamp_enabled;
2106#endif
2107
c82ba9fa
LZ		 2108 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2109 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
97fb7a0a 2110 void (*yield_task) (struct rq *rq);
0900acf2 2111 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
c82ba9fa 2112
97fb7a0a 2113 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
c82ba9fa 2114
98c2f700
PZ		 2115 struct task_struct *(*pick_next_task)(struct rq *rq);
2116
6e2df058 2117 void (*put_prev_task)(struct rq *rq, struct task_struct *p);
a0e813f2 2118 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
c82ba9fa
LZ		 2119
2120#ifdef CONFIG_SMP
6e2df058 2121 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
3aef1551 2122 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
21f56ffe
PZ		 2123
2124 struct task_struct * (*pick_task)(struct rq *rq);
2125
1327237a 2126 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
c82ba9fa 2127
97fb7a0a 2128 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
c82ba9fa
LZ		 2129
2130 void (*set_cpus_allowed)(struct task_struct *p,
9cfc3e18
PZ		 2131 const struct cpumask *newmask,
2132 u32 flags);
c82ba9fa
LZ		 2133
2134 void (*rq_online)(struct rq *rq);
2135 void (*rq_offline)(struct rq *rq);
a7c81556
PZ		 2136
2137 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
c82ba9fa
LZ		 2138#endif
2139
97fb7a0a
IM		 2140 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2141 void (*task_fork)(struct task_struct *p);
2142 void (*task_dead)(struct task_struct *p);
c82ba9fa 2143
67dfa1b7
KT		 2144 /*
2145 * The switched_from() call is allowed to drop rq->lock, therefore we
3b03706f 2146 * cannot assume the switched_from/switched_to pair is serialized by
67dfa1b7
KT		 2147 * rq->lock. They are however serialized by p->pi_lock.
2148 */
97fb7a0a
IM		 2149 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2150 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
c82ba9fa 2151 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
97fb7a0a 2152 int oldprio);
c82ba9fa 2153
97fb7a0a
IM		 2154 unsigned int (*get_rr_interval)(struct rq *rq,
2155 struct task_struct *task);
c82ba9fa 2156
97fb7a0a 2157 void (*update_curr)(struct rq *rq);
6e998916 2158
97fb7a0a
IM		 2159#define TASK_SET_GROUP 0
2160#define TASK_MOVE_GROUP 1
ea86cb4b 2161
c82ba9fa 2162#ifdef CONFIG_FAIR_GROUP_SCHED
97fb7a0a 2163 void (*task_change_group)(struct task_struct *p, int type);
c82ba9fa 2164#endif
43c31ac0 2165};
029632fb 2166
3f1d2a31
PZ		 2167static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2168{
10e7071b 2169 WARN_ON_ONCE(rq->curr != prev);
6e2df058 2170 prev->sched_class->put_prev_task(rq, prev);
3f1d2a31
PZ		 2171}
2172
03b7fad1 2173static inline void set_next_task(struct rq *rq, struct task_struct *next)
b2bf6c31 2174{
a0e813f2 2175 next->sched_class->set_next_task(rq, next, false);
b2bf6c31
PZ		 2176}
2177
43c31ac0
PZ		 2178
2179/*
2180 * Helper to define a sched_class instance; each one is placed in a separate
2181 * section which is ordered by the linker script:
2182 *
2183 * include/asm-generic/vmlinux.lds.h
2184 *
2185 * Also enforce alignment on the instance, not the type, to guarantee layout.
2186 */
2187#define DEFINE_SCHED_CLASS(name) \
2188const struct sched_class name##_sched_class \
2189 __aligned(__alignof__(struct sched_class)) \
2190 __section("__" #name "_sched_class")
2191
c3a340f7
SRV		 2192/* Defined in include/asm-generic/vmlinux.lds.h */
2193extern struct sched_class __begin_sched_classes[];
2194extern struct sched_class __end_sched_classes[];
2195
2196#define sched_class_highest (__end_sched_classes - 1)
2197#define sched_class_lowest (__begin_sched_classes - 1)
6e2df058
PZ		 2198
2199#define for_class_range(class, _from, _to) \
c3a340f7 2200 for (class = (_from); class != (_to); class--)
6e2df058 2201
029632fb 2202#define for_each_class(class) \
c3a340f7 2203 for_class_range(class, sched_class_highest, sched_class_lowest)
029632fb
PZ		 2204
2205extern const struct sched_class stop_sched_class;
aab03e05 2206extern const struct sched_class dl_sched_class;
029632fb
PZ		 2207extern const struct sched_class rt_sched_class;
2208extern const struct sched_class fair_sched_class;
2209extern const struct sched_class idle_sched_class;
2210
6e2df058
PZ		 2211static inline bool sched_stop_runnable(struct rq *rq)
2212{
2213 return rq->stop && task_on_rq_queued(rq->stop);
2214}
2215
2216static inline bool sched_dl_runnable(struct rq *rq)
2217{
2218 return rq->dl.dl_nr_running > 0;
2219}
2220
2221static inline bool sched_rt_runnable(struct rq *rq)
2222{
2223 return rq->rt.rt_queued > 0;
2224}
2225
2226static inline bool sched_fair_runnable(struct rq *rq)
2227{
2228 return rq->cfs.nr_running > 0;
2229}
029632fb 2230
5d7d6056 2231extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
98c2f700 2232extern struct task_struct *pick_next_task_idle(struct rq *rq);
5d7d6056 2233
af449901
PZ		 2234#define SCA_CHECK 0x01
2235#define SCA_MIGRATE_DISABLE 0x02
2236#define SCA_MIGRATE_ENABLE 0x04
2237
029632fb
PZ		 2238#ifdef CONFIG_SMP
2239
63b2ca30 2240extern void update_group_capacity(struct sched_domain *sd, int cpu);
b719203b 2241
7caff66f 2242extern void trigger_load_balance(struct rq *rq);
029632fb 2243
9cfc3e18 2244extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
c5b28038 2245
a7c81556
PZ		 2246static inline struct task_struct *get_push_task(struct rq *rq)
2247{
2248 struct task_struct *p = rq->curr;
2249
5cb9eaa3 2250 lockdep_assert_rq_held(rq);
a7c81556
PZ		 2251
2252 if (rq->push_busy)
2253 return NULL;
2254
2255 if (p->nr_cpus_allowed == 1)
2256 return NULL;
2257
2258 rq->push_busy = true;
2259 return get_task_struct(p);
2260}
2261
2262extern int push_cpu_stop(void *arg);
c5b28038 2263
029632fb
PZ		 2264#endif
2265
442bf3aa
DL		 2266#ifdef CONFIG_CPU_IDLE
2267static inline void idle_set_state(struct rq *rq,
2268 struct cpuidle_state *idle_state)
2269{
2270 rq->idle_state = idle_state;
2271}
2272
2273static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2274{
9148a3a1 2275 SCHED_WARN_ON(!rcu_read_lock_held());
97fb7a0a 2276
442bf3aa
DL		 2277 return rq->idle_state;
2278}
2279#else
2280static inline void idle_set_state(struct rq *rq,
2281 struct cpuidle_state *idle_state)
2282{
2283}
2284
2285static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2286{
2287 return NULL;
2288}
2289#endif
2290
8663effb
SRV		 2291extern void schedule_idle(void);
2292
029632fb
PZ		 2293extern void sysrq_sched_debug_show(void);
2294extern void sched_init_granularity(void);
2295extern void update_max_interval(void);
1baca4ce
JL		 2296
2297extern void init_sched_dl_class(void);
029632fb
PZ		 2298extern void init_sched_rt_class(void);
2299extern void init_sched_fair_class(void);
2300
9059393e
VG		 2301extern void reweight_task(struct task_struct *p, int prio);
2302
8875125e 2303extern void resched_curr(struct rq *rq);
029632fb
PZ		 2304extern void resched_cpu(int cpu);
2305
2306extern struct rt_bandwidth def_rt_bandwidth;
2307extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2308
332ac17e
DF		 2309extern struct dl_bandwidth def_dl_bandwidth;
2310extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
aab03e05 2311extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
209a0cbd 2312extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
aab03e05 2313
97fb7a0a
IM		 2314#define BW_SHIFT 20
2315#define BW_UNIT (1 << BW_SHIFT)
2316#define RATIO_SHIFT 8
d505b8af
HC		 2317#define MAX_BW_BITS (64 - BW_SHIFT)
2318#define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
332ac17e
DF		 2319unsigned long to_ratio(u64 period, u64 runtime);
2320
540247fb 2321extern void init_entity_runnable_average(struct sched_entity *se);
d0fe0b9c 2322extern void post_init_entity_util_avg(struct task_struct *p);
a75cdaa9 2323
76d92ac3
FW		 2324#ifdef CONFIG_NO_HZ_FULL
2325extern bool sched_can_stop_tick(struct rq *rq);
d84b3131 2326extern int __init sched_tick_offload_init(void);
76d92ac3
FW		 2327
2328/*
2329 * Tick may be needed by tasks in the runqueue depending on their policy and
2330 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2331 * nohz mode if necessary.
2332 */
2333static inline void sched_update_tick_dependency(struct rq *rq)
2334{
21a6ee14 2335 int cpu = cpu_of(rq);
76d92ac3
FW		 2336
2337 if (!tick_nohz_full_cpu(cpu))
2338 return;
2339
2340 if (sched_can_stop_tick(rq))
2341 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2342 else
2343 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2344}
2345#else
d84b3131 2346static inline int sched_tick_offload_init(void) { return 0; }
76d92ac3
FW		 2347static inline void sched_update_tick_dependency(struct rq *rq) { }
2348#endif
2349
72465447 2350static inline void add_nr_running(struct rq *rq, unsigned count)
029632fb 2351{
72465447
KT		 2352 unsigned prev_nr = rq->nr_running;
2353
2354 rq->nr_running = prev_nr + count;
9d246053
PA		 2355 if (trace_sched_update_nr_running_tp_enabled()) {
2356 call_trace_sched_update_nr_running(rq, count);
2357 }
9f3660c2 2358
4486edd1 2359#ifdef CONFIG_SMP
3e184501 2360 if (prev_nr < 2 && rq->nr_running >= 2) {
e90c8fe1
VS		 2361 if (!READ_ONCE(rq->rd->overload))
2362 WRITE_ONCE(rq->rd->overload, 1);
4486edd1 2363 }
3e184501 2364#endif
76d92ac3
FW		 2365
2366 sched_update_tick_dependency(rq);
029632fb
PZ		 2367}
2368
72465447 2369static inline void sub_nr_running(struct rq *rq, unsigned count)
029632fb 2370{
72465447 2371 rq->nr_running -= count;
9d246053 2372 if (trace_sched_update_nr_running_tp_enabled()) {
a1bd0685 2373 call_trace_sched_update_nr_running(rq, -count);
9d246053
PA		 2374 }
2375
76d92ac3
FW		 2376 /* Check if we still need preemption */
2377 sched_update_tick_dependency(rq);
029632fb
PZ		 2378}
2379
029632fb
PZ		 2380extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2381extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2382
2383extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2384
029632fb
PZ		 2385extern const_debug unsigned int sysctl_sched_nr_migrate;
2386extern const_debug unsigned int sysctl_sched_migration_cost;
2387
029632fb
PZ		 2388#ifdef CONFIG_SCHED_HRTICK
2389
2390/*
2391 * Use hrtick when:
2392 * - enabled by features
2393 * - hrtimer is actually high res
2394 */
2395static inline int hrtick_enabled(struct rq *rq)
2396{
029632fb
PZ		 2397 if (!cpu_active(cpu_of(rq)))
2398 return 0;
2399 return hrtimer_is_hres_active(&rq->hrtick_timer);
2400}
2401
e0ee463c
JL		 2402static inline int hrtick_enabled_fair(struct rq *rq)
2403{
2404 if (!sched_feat(HRTICK))
2405 return 0;
2406 return hrtick_enabled(rq);
2407}
2408
2409static inline int hrtick_enabled_dl(struct rq *rq)
2410{
2411 if (!sched_feat(HRTICK_DL))
2412 return 0;
2413 return hrtick_enabled(rq);
2414}
2415
029632fb
PZ		 2416void hrtick_start(struct rq *rq, u64 delay);
2417
b39e66ea
MG		 2418#else
2419
e0ee463c
JL		 2420static inline int hrtick_enabled_fair(struct rq *rq)
2421{
2422 return 0;
2423}
2424
2425static inline int hrtick_enabled_dl(struct rq *rq)
2426{
2427 return 0;
2428}
2429
b39e66ea
MG		 2430static inline int hrtick_enabled(struct rq *rq)
2431{
2432 return 0;
2433}
2434
029632fb
PZ		 2435#endif /* CONFIG_SCHED_HRTICK */
2436
1567c3e3
GG		 2437#ifndef arch_scale_freq_tick
2438static __always_inline
2439void arch_scale_freq_tick(void)
2440{
2441}
2442#endif
2443
dfbca41f 2444#ifndef arch_scale_freq_capacity
f4470cdf
VS		 2445/**
2446 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2447 * @cpu: the CPU in question.
2448 *
2449 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2450 *
2451 * f_curr
2452 * ------ * SCHED_CAPACITY_SCALE
2453 * f_max
2454 */
dfbca41f 2455static __always_inline
7673c8a4 2456unsigned long arch_scale_freq_capacity(int cpu)
dfbca41f
PZ		 2457{
2458 return SCHED_CAPACITY_SCALE;
2459}
2460#endif
b5b4860d 2461
d66f1b06 2462
029632fb 2463#ifdef CONFIG_SMP
029632fb 2464
d66f1b06
PZ		 2465static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2466{
9edeaea1
PZ		 2467#ifdef CONFIG_SCHED_CORE
2468 /*
2469 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2470 * order by core-id first and cpu-id second.
2471 *
2472 * Notably:
2473 *
2474 * double_rq_lock(0,3); will take core-0, core-1 lock
2475 * double_rq_lock(1,2); will take core-1, core-0 lock
2476 *
2477 * when only cpu-id is considered.
2478 */
2479 if (rq1->core->cpu < rq2->core->cpu)
2480 return true;
2481 if (rq1->core->cpu > rq2->core->cpu)
2482 return false;
2483
2484 /*
2485 * __sched_core_flip() relies on SMT having cpu-id lock order.
2486 */
2487#endif
d66f1b06
PZ		 2488 return rq1->cpu < rq2->cpu;
2489}
2490
2491extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2492
2493#ifdef CONFIG_PREEMPTION
029632fb
PZ		 2494
2495/*
2496 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2497 * way at the expense of forcing extra atomic operations in all
2498 * invocations. This assures that the double_lock is acquired using the
2499 * same underlying policy as the spinlock_t on this architecture, which
2500 * reduces latency compared to the unfair variant below. However, it
2501 * also adds more overhead and therefore may reduce throughput.
2502 */
2503static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2504 __releases(this_rq->lock)
2505 __acquires(busiest->lock)
2506 __acquires(this_rq->lock)
2507{
5cb9eaa3 2508 raw_spin_rq_unlock(this_rq);
029632fb
PZ		 2509 double_rq_lock(this_rq, busiest);
2510
2511 return 1;
2512}
2513
2514#else
2515/*
2516 * Unfair double_lock_balance: Optimizes throughput at the expense of
2517 * latency by eliminating extra atomic operations when the locks are
97fb7a0a
IM		 2518 * already in proper order on entry. This favors lower CPU-ids and will
2519 * grant the double lock to lower CPUs over higher ids under contention,
029632fb
PZ		 2520 * regardless of entry order into the function.
2521 */
2522static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2523 __releases(this_rq->lock)
2524 __acquires(busiest->lock)
2525 __acquires(this_rq->lock)
2526{
9ef7e7e3 2527 if (__rq_lockp(this_rq) == __rq_lockp(busiest))
5cb9eaa3
PZ		 2528 return 0;
2529
2530 if (likely(raw_spin_rq_trylock(busiest)))
2531 return 0;
2532
d66f1b06 2533 if (rq_order_less(this_rq, busiest)) {
5cb9eaa3
PZ		 2534 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2535 return 0;
029632fb 2536 }
5cb9eaa3
PZ		 2537
2538 raw_spin_rq_unlock(this_rq);
d66f1b06 2539 double_rq_lock(this_rq, busiest);
5cb9eaa3
PZ		 2540
2541 return 1;
029632fb
PZ		 2542}
2543
c1a280b6 2544#endif /* CONFIG_PREEMPTION */
029632fb
PZ		 2545
2546/*
2547 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2548 */
2549static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2550{
5cb9eaa3 2551 lockdep_assert_irqs_disabled();
029632fb
PZ		 2552
2553 return _double_lock_balance(this_rq, busiest);
2554}
2555
2556static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2557 __releases(busiest->lock)
2558{
9ef7e7e3 2559 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
5cb9eaa3 2560 raw_spin_rq_unlock(busiest);
9ef7e7e3 2561 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
029632fb
PZ		 2562}
2563
74602315
PZ		 2564static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2565{
2566 if (l1 > l2)
2567 swap(l1, l2);
2568
2569 spin_lock(l1);
2570 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2571}
2572
60e69eed
MG		 2573static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2574{
2575 if (l1 > l2)
2576 swap(l1, l2);
2577
2578 spin_lock_irq(l1);
2579 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2580}
2581
74602315
PZ		 2582static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2583{
2584 if (l1 > l2)
2585 swap(l1, l2);
2586
2587 raw_spin_lock(l1);
2588 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2589}
2590
029632fb
PZ		 2591/*
2592 * double_rq_unlock - safely unlock two runqueues
2593 *
2594 * Note this does not restore interrupts like task_rq_unlock,
2595 * you need to do so manually after calling.
2596 */
2597static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2598 __releases(rq1->lock)
2599 __releases(rq2->lock)
2600{
9ef7e7e3 2601 if (__rq_lockp(rq1) != __rq_lockp(rq2))
5cb9eaa3 2602 raw_spin_rq_unlock(rq2);
029632fb
PZ		 2603 else
2604 __release(rq2->lock);
d66f1b06 2605 raw_spin_rq_unlock(rq1);
029632fb
PZ		 2606}
2607
f2cb1360
IM		 2608extern void set_rq_online (struct rq *rq);
2609extern void set_rq_offline(struct rq *rq);
2610extern bool sched_smp_initialized;
2611
029632fb
PZ		 2612#else /* CONFIG_SMP */
2613
2614/*
2615 * double_rq_lock - safely lock two runqueues
2616 *
2617 * Note this does not disable interrupts like task_rq_lock,
2618 * you need to do so manually before calling.
2619 */
2620static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2621 __acquires(rq1->lock)
2622 __acquires(rq2->lock)
2623{
2624 BUG_ON(!irqs_disabled());
2625 BUG_ON(rq1 != rq2);
5cb9eaa3 2626 raw_spin_rq_lock(rq1);
029632fb
PZ		 2627 __acquire(rq2->lock); /* Fake it out ;) */
2628}
2629
2630/*
2631 * double_rq_unlock - safely unlock two runqueues
2632 *
2633 * Note this does not restore interrupts like task_rq_unlock,
2634 * you need to do so manually after calling.
2635 */
2636static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2637 __releases(rq1->lock)
2638 __releases(rq2->lock)
2639{
2640 BUG_ON(rq1 != rq2);
5cb9eaa3 2641 raw_spin_rq_unlock(rq1);
029632fb
PZ		 2642 __release(rq2->lock);
2643}
2644
2645#endif
2646
2647extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2648extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
6b55c965
SD		 2649
2650#ifdef CONFIG_SCHED_DEBUG
9406415f 2651extern bool sched_debug_verbose;
9469eb01 2652
029632fb
PZ		 2653extern void print_cfs_stats(struct seq_file *m, int cpu);
2654extern void print_rt_stats(struct seq_file *m, int cpu);
acb32132 2655extern void print_dl_stats(struct seq_file *m, int cpu);
f6a34630
MM		 2656extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2657extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2658extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
c006fac5
PT		 2659
2660extern void resched_latency_warn(int cpu, u64 latency);
397f2378
SD		 2661#ifdef CONFIG_NUMA_BALANCING
2662extern void
2663show_numa_stats(struct task_struct *p, struct seq_file *m);
2664extern void
2665print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2666 unsigned long tpf, unsigned long gsf, unsigned long gpf);
2667#endif /* CONFIG_NUMA_BALANCING */
c006fac5
PT		 2668#else
2669static inline void resched_latency_warn(int cpu, u64 latency) {}
397f2378 2670#endif /* CONFIG_SCHED_DEBUG */
029632fb
PZ		 2671
2672extern void init_cfs_rq(struct cfs_rq *cfs_rq);
07c54f7a
AV		 2673extern void init_rt_rq(struct rt_rq *rt_rq);
2674extern void init_dl_rq(struct dl_rq *dl_rq);
029632fb 2675
1ee14e6c
BS		 2676extern void cfs_bandwidth_usage_inc(void);
2677extern void cfs_bandwidth_usage_dec(void);
1c792db7 2678
3451d024 2679#ifdef CONFIG_NO_HZ_COMMON
00357f5e
PZ		 2680#define NOHZ_BALANCE_KICK_BIT 0
2681#define NOHZ_STATS_KICK_BIT 1
c6f88654 2682#define NOHZ_NEWILB_KICK_BIT 2
a22e47a4 2683
a22e47a4 2684#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
b7031a02 2685#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
c6f88654 2686#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
b7031a02
PZ		 2687
2688#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
1c792db7
SS		 2689
2690#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
20a5c8cc 2691
00357f5e 2692extern void nohz_balance_exit_idle(struct rq *rq);
20a5c8cc 2693#else
00357f5e 2694static inline void nohz_balance_exit_idle(struct rq *rq) { }
1c792db7 2695#endif
73fbec60 2696
c6f88654
VG		 2697#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2698extern void nohz_run_idle_balance(int cpu);
2699#else
2700static inline void nohz_run_idle_balance(int cpu) { }
2701#endif
daec5798
LA		 2702
2703#ifdef CONFIG_SMP
2704static inline
2705void __dl_update(struct dl_bw *dl_b, s64 bw)
2706{
2707 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2708 int i;
2709
2710 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2711 "sched RCU must be held");
2712 for_each_cpu_and(i, rd->span, cpu_active_mask) {
2713 struct rq *rq = cpu_rq(i);
2714
2715 rq->dl.extra_bw += bw;
2716 }
2717}
2718#else
2719static inline
2720void __dl_update(struct dl_bw *dl_b, s64 bw)
2721{
2722 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2723
2724 dl->extra_bw += bw;
2725}
2726#endif
2727
2728
73fbec60 2729#ifdef CONFIG_IRQ_TIME_ACCOUNTING
19d23dbf 2730struct irqtime {
25e2d8c1 2731 u64 total;
a499a5a1 2732 u64 tick_delta;
19d23dbf
FW		 2733 u64 irq_start_time;
2734 struct u64_stats_sync sync;
2735};
73fbec60 2736
19d23dbf 2737DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
73fbec60 2738
25e2d8c1
FW		 2739/*
2740 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3b03706f 2741 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
25e2d8c1
FW		 2742 * and never move forward.
2743 */
73fbec60
FW		 2744static inline u64 irq_time_read(int cpu)
2745{
19d23dbf
FW		 2746 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2747 unsigned int seq;
2748 u64 total;
73fbec60
FW		 2749
2750 do {
19d23dbf 2751 seq = __u64_stats_fetch_begin(&irqtime->sync);
25e2d8c1 2752 total = irqtime->total;
19d23dbf 2753 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
73fbec60 2754
19d23dbf 2755 return total;
73fbec60 2756}
73fbec60 2757#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
adaf9fcd
RW		 2758
2759#ifdef CONFIG_CPU_FREQ
b10abd0a 2760DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
adaf9fcd
RW		 2761
2762/**
2763 * cpufreq_update_util - Take a note about CPU utilization changes.
12bde33d 2764 * @rq: Runqueue to carry out the update for.
58919e83 2765 * @flags: Update reason flags.
adaf9fcd 2766 *
58919e83
RW		 2767 * This function is called by the scheduler on the CPU whose utilization is
2768 * being updated.
adaf9fcd
RW		 2769 *
2770 * It can only be called from RCU-sched read-side critical sections.
adaf9fcd
RW		 2771 *
2772 * The way cpufreq is currently arranged requires it to evaluate the CPU
2773 * performance state (frequency/voltage) on a regular basis to prevent it from
2774 * being stuck in a completely inadequate performance level for too long.
e0367b12
JL		 2775 * That is not guaranteed to happen if the updates are only triggered from CFS
2776 * and DL, though, because they may not be coming in if only RT tasks are
2777 * active all the time (or there are RT tasks only).
adaf9fcd 2778 *
e0367b12
JL		 2779 * As a workaround for that issue, this function is called periodically by the
2780 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
adaf9fcd 2781 * but that really is a band-aid. Going forward it should be replaced with
e0367b12 2782 * solutions targeted more specifically at RT tasks.
adaf9fcd 2783 */
12bde33d 2784static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
adaf9fcd 2785{
58919e83
RW		 2786 struct update_util_data *data;
2787
674e7541
VK		 2788 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2789 cpu_of(rq)));
58919e83 2790 if (data)
12bde33d
RW		 2791 data->func(data, rq_clock(rq), flags);
2792}
adaf9fcd 2793#else
12bde33d 2794static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
adaf9fcd 2795#endif /* CONFIG_CPU_FREQ */
be53f58f 2796
982d9cdc 2797#ifdef CONFIG_UCLAMP_TASK
686516b5 2798unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
9d20ad7d 2799
46609ce2
QY		 2800/**
2801 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2802 * @rq: The rq to clamp against. Must not be NULL.
2803 * @util: The util value to clamp.
2804 * @p: The task to clamp against. Can be NULL if you want to clamp
2805 * against @rq only.
2806 *
2807 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2808 *
2809 * If sched_uclamp_used static key is disabled, then just return the util
2810 * without any clamping since uclamp aggregation at the rq level in the fast
2811 * path is disabled, rendering this operation a NOP.
2812 *
2813 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2814 * will return the correct effective uclamp value of the task even if the
2815 * static key is disabled.
2816 */
9d20ad7d 2817static __always_inline
d2b58a28
VS		 2818unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2819 struct task_struct *p)
982d9cdc 2820{
46609ce2
QY		 2821 unsigned long min_util;
2822 unsigned long max_util;
2823
2824 if (!static_branch_likely(&sched_uclamp_used))
2825 return util;
2826
2827 min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2828 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
982d9cdc 2829
9d20ad7d
PB		 2830 if (p) {
2831 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2832 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2833 }
2834
982d9cdc
PB		 2835 /*
2836 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2837 * RUNNABLE tasks with _different_ clamps, we can end up with an
2838 * inversion. Fix it now when the clamps are applied.
2839 */
2840 if (unlikely(min_util >= max_util))
2841 return min_util;
2842
2843 return clamp(util, min_util, max_util);
2844}
46609ce2
QY		 2845
2846/*
2847 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2848 * by default in the fast path and only gets turned on once userspace performs
2849 * an operation that requires it.
2850 *
2851 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2852 * hence is active.
2853 */
2854static inline bool uclamp_is_used(void)
2855{
2856 return static_branch_likely(&sched_uclamp_used);
2857}
982d9cdc 2858#else /* CONFIG_UCLAMP_TASK */
d2b58a28
VS		 2859static inline
2860unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2861 struct task_struct *p)
9d20ad7d
PB		 2862{
2863 return util;
2864}
46609ce2
QY		 2865
2866static inline bool uclamp_is_used(void)
2867{
2868 return false;
2869}
982d9cdc
PB		 2870#endif /* CONFIG_UCLAMP_TASK */
2871
9bdcb44e 2872#ifdef arch_scale_freq_capacity
97fb7a0a
IM		 2873# ifndef arch_scale_freq_invariant
2874# define arch_scale_freq_invariant() true
2875# endif
2876#else
2877# define arch_scale_freq_invariant() false
9bdcb44e 2878#endif
d4edd662 2879
10a35e68
VG		 2880#ifdef CONFIG_SMP
2881static inline unsigned long capacity_orig_of(int cpu)
2882{
2883 return cpu_rq(cpu)->cpu_capacity_orig;
2884}
10a35e68 2885
938e5e4b 2886/**
a5418be9 2887 * enum cpu_util_type - CPU utilization type
938e5e4b
QP		 2888 * @FREQUENCY_UTIL: Utilization used to select frequency
2889 * @ENERGY_UTIL: Utilization used during energy calculation
2890 *
2891 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2892 * need to be aggregated differently depending on the usage made of them. This
a5418be9 2893 * enum is used within effective_cpu_util() to differentiate the types of
938e5e4b
QP		 2894 * utilization expected by the callers, and adjust the aggregation accordingly.
2895 */
a5418be9 2896enum cpu_util_type {
938e5e4b
QP		 2897 FREQUENCY_UTIL,
2898 ENERGY_UTIL,
2899};
2900
a5418be9
VK		 2901unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2902 unsigned long max, enum cpu_util_type type,
af24bde8 2903 struct task_struct *p);
938e5e4b 2904
8cc90515 2905static inline unsigned long cpu_bw_dl(struct rq *rq)
d4edd662
JL		 2906{
2907 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2908}
2909
8cc90515
VG		 2910static inline unsigned long cpu_util_dl(struct rq *rq)
2911{
2912 return READ_ONCE(rq->avg_dl.util_avg);
2913}
2914
d4edd662
JL		 2915static inline unsigned long cpu_util_cfs(struct rq *rq)
2916{
a07630b8
PB		 2917 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2918
2919 if (sched_feat(UTIL_EST)) {
2920 util = max_t(unsigned long, util,
2921 READ_ONCE(rq->cfs.avg.util_est.enqueued));
2922 }
2923
2924 return util;
d4edd662 2925}
371bf427
VG		 2926
2927static inline unsigned long cpu_util_rt(struct rq *rq)
2928{
dfa444dc 2929 return READ_ONCE(rq->avg_rt.util_avg);
371bf427 2930}
7d6a905f 2931#endif
9033ea11 2932
11d4afd4 2933#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
9033ea11
VG		 2934static inline unsigned long cpu_util_irq(struct rq *rq)
2935{
2936 return rq->avg_irq.util_avg;
2937}
2e62c474
VG		 2938
2939static inline
2940unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2941{
2942 util *= (max - irq);
2943 util /= max;
2944
2945 return util;
2946
2947}
9033ea11
VG		 2948#else
2949static inline unsigned long cpu_util_irq(struct rq *rq)
2950{
2951 return 0;
2952}
2953
2e62c474
VG		 2954static inline
2955unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2956{
2957 return util;
2958}
794a56eb 2959#endif
6aa140fa 2960
531b5c9f 2961#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
f8a696f2 2962
6aa140fa 2963#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
f8a696f2
PZ		 2964
2965DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2966
2967static inline bool sched_energy_enabled(void)
2968{
2969 return static_branch_unlikely(&sched_energy_present);
2970}
2971
2972#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2973
6aa140fa 2974#define perf_domain_span(pd) NULL
f8a696f2 2975static inline bool sched_energy_enabled(void) { return false; }
1f74de87 2976
f8a696f2 2977#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
227a4aad
MD		 2978
2979#ifdef CONFIG_MEMBARRIER
2980/*
2981 * The scheduler provides memory barriers required by membarrier between:
2982 * - prior user-space memory accesses and store to rq->membarrier_state,
2983 * - store to rq->membarrier_state and following user-space memory accesses.
2984 * In the same way it provides those guarantees around store to rq->curr.
2985 */
2986static inline void membarrier_switch_mm(struct rq *rq,
2987 struct mm_struct *prev_mm,
2988 struct mm_struct *next_mm)
2989{
2990 int membarrier_state;
2991
2992 if (prev_mm == next_mm)
2993 return;
2994
2995 membarrier_state = atomic_read(&next_mm->membarrier_state);
2996 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
2997 return;
2998
2999 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3000}
3001#else
3002static inline void membarrier_switch_mm(struct rq *rq,
3003 struct mm_struct *prev_mm,
3004 struct mm_struct *next_mm)
3005{
3006}
3007#endif
52262ee5
MG		 3008
3009#ifdef CONFIG_SMP
3010static inline bool is_per_cpu_kthread(struct task_struct *p)
3011{
3012 if (!(p->flags & PF_KTHREAD))
3013 return false;
3014
3015 if (p->nr_cpus_allowed != 1)
3016 return false;
3017
3018 return true;
3019}
3020#endif
b3212fe2 3021
1011dcce
PZ		 3022extern void swake_up_all_locked(struct swait_queue_head *q);
3023extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3024
3025#ifdef CONFIG_PREEMPT_DYNAMIC
3026extern int preempt_dynamic_mode;
3027extern int sched_dynamic_mode(const char *str);
3028extern void sched_dynamic_update(int mode);
3029#endif
3030
Linux 6.x block layer and io_uring tree(s)