1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Scheduler internal types and methods:
4 */
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
7
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
23
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/context_tracking.h>
31 #include <linux/cpufreq.h>
32 #include <linux/cpumask_api.h>
33 #include <linux/ctype.h>
34 #include <linux/file.h>
35 #include <linux/fs_api.h>
36 #include <linux/hrtimer_api.h>
37 #include <linux/interrupt.h>
38 #include <linux/irq_work.h>
39 #include <linux/jiffies.h>
40 #include <linux/kref_api.h>
41 #include <linux/kthread.h>
42 #include <linux/ktime_api.h>
43 #include <linux/lockdep_api.h>
44 #include <linux/lockdep.h>
45 #include <linux/minmax.h>
46 #include <linux/mm.h>
47 #include <linux/module.h>
48 #include <linux/mutex_api.h>
49 #include <linux/plist.h>
50 #include <linux/poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/psi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/seq_file.h>
56 #include <linux/seqlock.h>
57 #include <linux/softirq.h>
58 #include <linux/spinlock_api.h>
59 #include <linux/static_key.h>
60 #include <linux/stop_machine.h>
61 #include <linux/syscalls_api.h>
62 #include <linux/syscalls.h>
63 #include <linux/tick.h>
64 #include <linux/topology.h>
65 #include <linux/types.h>
66 #include <linux/u64_stats_sync_api.h>
67 #include <linux/uaccess.h>
68 #include <linux/wait_api.h>
69 #include <linux/wait_bit.h>
70 #include <linux/workqueue_api.h>
71 #include <linux/delayacct.h>
72
73 #include <trace/events/power.h>
74 #include <trace/events/sched.h>
75
76 #include "../workqueue_internal.h"
77
78 struct rq;
79 struct cfs_rq;
80 struct rt_rq;
81 struct sched_group;
82 struct cpuidle_state;
83
84 #ifdef CONFIG_PARAVIRT
85 # include <asm/paravirt.h>
86 # include <asm/paravirt_api_clock.h>
87 #endif
88
89 #include <asm/barrier.h>
90
91 #include "cpupri.h"
92 #include "cpudeadline.h"
93
94 #ifdef CONFIG_SCHED_DEBUG
95 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
96 #else
97 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
98 #endif
99
100 /* task_struct::on_rq states: */
101 #define TASK_ON_RQ_QUEUED 1
102 #define TASK_ON_RQ_MIGRATING 2
103
104 extern __read_mostly int scheduler_running;
105
106 extern unsigned long calc_load_update;
107 extern atomic_long_t calc_load_tasks;
108
109 extern void calc_global_load_tick(struct rq *this_rq);
110 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
111
112 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
113
114 extern int sysctl_sched_rt_period;
115 extern int sysctl_sched_rt_runtime;
116 extern int sched_rr_timeslice;
117
118 /*
119 * Asymmetric CPU capacity bits
120 */
121 struct asym_cap_data {
122 struct list_head link;
123 struct rcu_head rcu;
124 unsigned long capacity;
125 unsigned long cpus[];
126 };
127
128 extern struct list_head asym_cap_list;
129
130 #define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus)
131
132 /*
133 * Helpers for converting nanosecond timing to jiffy resolution
134 */
135 #define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ))
136
137 /*
138 * Increase resolution of nice-level calculations for 64-bit architectures.
139 * The extra resolution improves shares distribution and load balancing of
140 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group
141 * hierarchies, especially on larger systems. This is not a user-visible change
142 * and does not change the user-interface for setting shares/weights.
143 *
144 * We increase resolution only if we have enough bits to allow this increased
145 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
146 * are pretty high and the returns do not justify the increased costs.
147 *
148 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
149 * increase coverage and consistency always enable it on 64-bit platforms.
150 */
151 #ifdef CONFIG_64BIT
152 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
153 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
154 # define scale_load_down(w) \
155 ({ \
156 unsigned long __w = (w); \
157 \
158 if (__w) \
159 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
160 __w; \
161 })
162 #else
163 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
164 # define scale_load(w) (w)
165 # define scale_load_down(w) (w)
166 #endif
167
168 /*
169 * Task weight (visible to users) and its load (invisible to users) have
170 * independent resolution, but they should be well calibrated. We use
171 * scale_load() and scale_load_down(w) to convert between them. The
172 * following must be true:
173 *
174 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
175 *
176 */
177 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
178
179 /*
180 * Single value that decides SCHED_DEADLINE internal math precision.
181 * 10 -> just above 1us
182 * 9 -> just above 0.5us
183 */
184 #define DL_SCALE 10
185
186 /*
187 * Single value that denotes runtime == period, ie unlimited time.
188 */
189 #define RUNTIME_INF ((u64)~0ULL)
190
idle_policy(int policy)191 static inline int idle_policy(int policy)
192 {
193 return policy == SCHED_IDLE;
194 }
195
normal_policy(int policy)196 static inline int normal_policy(int policy)
197 {
198 #ifdef CONFIG_SCHED_CLASS_EXT
199 if (policy == SCHED_EXT)
200 return true;
201 #endif
202 return policy == SCHED_NORMAL;
203 }
204
fair_policy(int policy)205 static inline int fair_policy(int policy)
206 {
207 return normal_policy(policy) || policy == SCHED_BATCH;
208 }
209
rt_policy(int policy)210 static inline int rt_policy(int policy)
211 {
212 return policy == SCHED_FIFO || policy == SCHED_RR;
213 }
214
dl_policy(int policy)215 static inline int dl_policy(int policy)
216 {
217 return policy == SCHED_DEADLINE;
218 }
219
valid_policy(int policy)220 static inline bool valid_policy(int policy)
221 {
222 return idle_policy(policy) || fair_policy(policy) ||
223 rt_policy(policy) || dl_policy(policy);
224 }
225
task_has_idle_policy(struct task_struct * p)226 static inline int task_has_idle_policy(struct task_struct *p)
227 {
228 return idle_policy(p->policy);
229 }
230
task_has_rt_policy(struct task_struct * p)231 static inline int task_has_rt_policy(struct task_struct *p)
232 {
233 return rt_policy(p->policy);
234 }
235
task_has_dl_policy(struct task_struct * p)236 static inline int task_has_dl_policy(struct task_struct *p)
237 {
238 return dl_policy(p->policy);
239 }
240
241 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
242
update_avg(u64 * avg,u64 sample)243 static inline void update_avg(u64 *avg, u64 sample)
244 {
245 s64 diff = sample - *avg;
246
247 *avg += diff / 8;
248 }
249
250 /*
251 * Shifting a value by an exponent greater *or equal* to the size of said value
252 * is UB; cap at size-1.
253 */
254 #define shr_bound(val, shift) \
255 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
256
257 /*
258 * cgroup weight knobs should use the common MIN, DFL and MAX values which are
259 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it
260 * maps pretty well onto the shares value used by scheduler and the round-trip
261 * conversions preserve the original value over the entire range.
262 */
sched_weight_from_cgroup(unsigned long cgrp_weight)263 static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight)
264 {
265 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL);
266 }
267
sched_weight_to_cgroup(unsigned long weight)268 static inline unsigned long sched_weight_to_cgroup(unsigned long weight)
269 {
270 return clamp_t(unsigned long,
271 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024),
272 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
273 }
274
275 /*
276 * !! For sched_setattr_nocheck() (kernel) only !!
277 *
278 * This is actually gross. :(
279 *
280 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
281 * tasks, but still be able to sleep. We need this on platforms that cannot
282 * atomically change clock frequency. Remove once fast switching will be
283 * available on such platforms.
284 *
285 * SUGOV stands for SchedUtil GOVernor.
286 */
287 #define SCHED_FLAG_SUGOV 0x10000000
288
289 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
290
dl_entity_is_special(const struct sched_dl_entity * dl_se)291 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
292 {
293 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
294 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
295 #else
296 return false;
297 #endif
298 }
299
300 /*
301 * Tells if entity @a should preempt entity @b.
302 */
dl_entity_preempt(const struct sched_dl_entity * a,const struct sched_dl_entity * b)303 static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
304 const struct sched_dl_entity *b)
305 {
306 return dl_entity_is_special(a) ||
307 dl_time_before(a->deadline, b->deadline);
308 }
309
310 /*
311 * This is the priority-queue data structure of the RT scheduling class:
312 */
313 struct rt_prio_array {
314 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
315 struct list_head queue[MAX_RT_PRIO];
316 };
317
318 struct rt_bandwidth {
319 /* nests inside the rq lock: */
320 raw_spinlock_t rt_runtime_lock;
321 ktime_t rt_period;
322 u64 rt_runtime;
323 struct hrtimer rt_period_timer;
324 unsigned int rt_period_active;
325 };
326
dl_bandwidth_enabled(void)327 static inline int dl_bandwidth_enabled(void)
328 {
329 return sysctl_sched_rt_runtime >= 0;
330 }
331
332 /*
333 * To keep the bandwidth of -deadline tasks under control
334 * we need some place where:
335 * - store the maximum -deadline bandwidth of each cpu;
336 * - cache the fraction of bandwidth that is currently allocated in
337 * each root domain;
338 *
339 * This is all done in the data structure below. It is similar to the
340 * one used for RT-throttling (rt_bandwidth), with the main difference
341 * that, since here we are only interested in admission control, we
342 * do not decrease any runtime while the group "executes", neither we
343 * need a timer to replenish it.
344 *
345 * With respect to SMP, bandwidth is given on a per root domain basis,
346 * meaning that:
347 * - bw (< 100%) is the deadline bandwidth of each CPU;
348 * - total_bw is the currently allocated bandwidth in each root domain;
349 */
350 struct dl_bw {
351 raw_spinlock_t lock;
352 u64 bw;
353 u64 total_bw;
354 };
355
356 extern void init_dl_bw(struct dl_bw *dl_b);
357 extern int sched_dl_global_validate(void);
358 extern void sched_dl_do_global(void);
359 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
360 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
361 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
362 extern bool __checkparam_dl(const struct sched_attr *attr);
363 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
364 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
365 extern int dl_bw_deactivate(int cpu);
366 extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec);
367 /*
368 * SCHED_DEADLINE supports servers (nested scheduling) with the following
369 * interface:
370 *
371 * dl_se::rq -- runqueue we belong to.
372 *
373 * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
374 * server when it runs out of tasks to run.
375 *
376 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
377 * returns NULL.
378 *
379 * dl_server_update() -- called from update_curr_common(), propagates runtime
380 * to the server.
381 *
382 * dl_server_start()
383 * dl_server_stop() -- start/stop the server when it has (no) tasks.
384 *
385 * dl_server_init() -- initializes the server.
386 */
387 extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
388 extern void dl_server_start(struct sched_dl_entity *dl_se);
389 extern void dl_server_stop(struct sched_dl_entity *dl_se);
390 extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
391 dl_server_has_tasks_f has_tasks,
392 dl_server_pick_f pick_task);
393
394 extern void dl_server_update_idle_time(struct rq *rq,
395 struct task_struct *p);
396 extern void fair_server_init(struct rq *rq);
397 extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq);
398 extern int dl_server_apply_params(struct sched_dl_entity *dl_se,
399 u64 runtime, u64 period, bool init);
400
dl_server_active(struct sched_dl_entity * dl_se)401 static inline bool dl_server_active(struct sched_dl_entity *dl_se)
402 {
403 return dl_se->dl_server_active;
404 }
405
406 #ifdef CONFIG_CGROUP_SCHED
407
408 extern struct list_head task_groups;
409
410 struct cfs_bandwidth {
411 #ifdef CONFIG_CFS_BANDWIDTH
412 raw_spinlock_t lock;
413 ktime_t period;
414 u64 quota;
415 u64 runtime;
416 u64 burst;
417 u64 runtime_snap;
418 s64 hierarchical_quota;
419
420 u8 idle;
421 u8 period_active;
422 u8 slack_started;
423 struct hrtimer period_timer;
424 struct hrtimer slack_timer;
425 struct list_head throttled_cfs_rq;
426
427 /* Statistics: */
428 int nr_periods;
429 int nr_throttled;
430 int nr_burst;
431 u64 throttled_time;
432 u64 burst_time;
433 #endif
434 };
435
436 /* Task group related information */
437 struct task_group {
438 struct cgroup_subsys_state css;
439
440 #ifdef CONFIG_GROUP_SCHED_WEIGHT
441 /* A positive value indicates that this is a SCHED_IDLE group. */
442 int idle;
443 #endif
444
445 #ifdef CONFIG_FAIR_GROUP_SCHED
446 /* schedulable entities of this group on each CPU */
447 struct sched_entity **se;
448 /* runqueue "owned" by this group on each CPU */
449 struct cfs_rq **cfs_rq;
450 unsigned long shares;
451 #ifdef CONFIG_SMP
452 /*
453 * load_avg can be heavily contended at clock tick time, so put
454 * it in its own cache-line separated from the fields above which
455 * will also be accessed at each tick.
456 */
457 atomic_long_t load_avg ____cacheline_aligned;
458 #endif
459 #endif
460
461 #ifdef CONFIG_RT_GROUP_SCHED
462 struct sched_rt_entity **rt_se;
463 struct rt_rq **rt_rq;
464
465 struct rt_bandwidth rt_bandwidth;
466 #endif
467
468 #ifdef CONFIG_EXT_GROUP_SCHED
469 u32 scx_flags; /* SCX_TG_* */
470 u32 scx_weight;
471 #endif
472
473 struct rcu_head rcu;
474 struct list_head list;
475
476 struct task_group *parent;
477 struct list_head siblings;
478 struct list_head children;
479
480 #ifdef CONFIG_SCHED_AUTOGROUP
481 struct autogroup *autogroup;
482 #endif
483
484 struct cfs_bandwidth cfs_bandwidth;
485
486 #ifdef CONFIG_UCLAMP_TASK_GROUP
487 /* The two decimal precision [%] value requested from user-space */
488 unsigned int uclamp_pct[UCLAMP_CNT];
489 /* Clamp values requested for a task group */
490 struct uclamp_se uclamp_req[UCLAMP_CNT];
491 /* Effective clamp values used for a task group */
492 struct uclamp_se uclamp[UCLAMP_CNT];
493 #endif
494
495 };
496
497 #ifdef CONFIG_GROUP_SCHED_WEIGHT
498 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
499
500 /*
501 * A weight of 0 or 1 can cause arithmetics problems.
502 * A weight of a cfs_rq is the sum of weights of which entities
503 * are queued on this cfs_rq, so a weight of a entity should not be
504 * too large, so as the shares value of a task group.
505 * (The default weight is 1024 - so there's no practical
506 * limitation from this.)
507 */
508 #define MIN_SHARES (1UL << 1)
509 #define MAX_SHARES (1UL << 18)
510 #endif
511
512 typedef int (*tg_visitor)(struct task_group *, void *);
513
514 extern int walk_tg_tree_from(struct task_group *from,
515 tg_visitor down, tg_visitor up, void *data);
516
517 /*
518 * Iterate the full tree, calling @down when first entering a node and @up when
519 * leaving it for the final time.
520 *
521 * Caller must hold rcu_lock or sufficient equivalent.
522 */
walk_tg_tree(tg_visitor down,tg_visitor up,void * data)523 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
524 {
525 return walk_tg_tree_from(&root_task_group, down, up, data);
526 }
527
css_tg(struct cgroup_subsys_state * css)528 static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
529 {
530 return css ? container_of(css, struct task_group, css) : NULL;
531 }
532
533 extern int tg_nop(struct task_group *tg, void *data);
534
535 #ifdef CONFIG_FAIR_GROUP_SCHED
536 extern void free_fair_sched_group(struct task_group *tg);
537 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
538 extern void online_fair_sched_group(struct task_group *tg);
539 extern void unregister_fair_sched_group(struct task_group *tg);
540 #else
free_fair_sched_group(struct task_group * tg)541 static inline void free_fair_sched_group(struct task_group *tg) { }
alloc_fair_sched_group(struct task_group * tg,struct task_group * parent)542 static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
543 {
544 return 1;
545 }
online_fair_sched_group(struct task_group * tg)546 static inline void online_fair_sched_group(struct task_group *tg) { }
unregister_fair_sched_group(struct task_group * tg)547 static inline void unregister_fair_sched_group(struct task_group *tg) { }
548 #endif
549
550 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
551 struct sched_entity *se, int cpu,
552 struct sched_entity *parent);
553 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
554
555 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
556 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
557 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
558 extern bool cfs_task_bw_constrained(struct task_struct *p);
559
560 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
561 struct sched_rt_entity *rt_se, int cpu,
562 struct sched_rt_entity *parent);
563 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
564 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
565 extern long sched_group_rt_runtime(struct task_group *tg);
566 extern long sched_group_rt_period(struct task_group *tg);
567 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
568
569 extern struct task_group *sched_create_group(struct task_group *parent);
570 extern void sched_online_group(struct task_group *tg,
571 struct task_group *parent);
572 extern void sched_destroy_group(struct task_group *tg);
573 extern void sched_release_group(struct task_group *tg);
574
575 extern void sched_move_task(struct task_struct *tsk, bool for_autogroup);
576
577 #ifdef CONFIG_FAIR_GROUP_SCHED
578 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
579
580 extern int sched_group_set_idle(struct task_group *tg, long idle);
581
582 #ifdef CONFIG_SMP
583 extern void set_task_rq_fair(struct sched_entity *se,
584 struct cfs_rq *prev, struct cfs_rq *next);
585 #else /* !CONFIG_SMP */
set_task_rq_fair(struct sched_entity * se,struct cfs_rq * prev,struct cfs_rq * next)586 static inline void set_task_rq_fair(struct sched_entity *se,
587 struct cfs_rq *prev, struct cfs_rq *next) { }
588 #endif /* CONFIG_SMP */
589 #else /* !CONFIG_FAIR_GROUP_SCHED */
sched_group_set_shares(struct task_group * tg,unsigned long shares)590 static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; }
sched_group_set_idle(struct task_group * tg,long idle)591 static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; }
592 #endif /* CONFIG_FAIR_GROUP_SCHED */
593
594 #else /* CONFIG_CGROUP_SCHED */
595
596 struct cfs_bandwidth { };
597
cfs_task_bw_constrained(struct task_struct * p)598 static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
599
600 #endif /* CONFIG_CGROUP_SCHED */
601
602 extern void unregister_rt_sched_group(struct task_group *tg);
603 extern void free_rt_sched_group(struct task_group *tg);
604 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
605
606 /*
607 * u64_u32_load/u64_u32_store
608 *
609 * Use a copy of a u64 value to protect against data race. This is only
610 * applicable for 32-bits architectures.
611 */
612 #ifdef CONFIG_64BIT
613 # define u64_u32_load_copy(var, copy) var
614 # define u64_u32_store_copy(var, copy, val) (var = val)
615 #else
616 # define u64_u32_load_copy(var, copy) \
617 ({ \
618 u64 __val, __val_copy; \
619 do { \
620 __val_copy = copy; \
621 /* \
622 * paired with u64_u32_store_copy(), ordering access \
623 * to var and copy. \
624 */ \
625 smp_rmb(); \
626 __val = var; \
627 } while (__val != __val_copy); \
628 __val; \
629 })
630 # define u64_u32_store_copy(var, copy, val) \
631 do { \
632 typeof(val) __val = (val); \
633 var = __val; \
634 /* \
635 * paired with u64_u32_load_copy(), ordering access to var and \
636 * copy. \
637 */ \
638 smp_wmb(); \
639 copy = __val; \
640 } while (0)
641 #endif
642 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy)
643 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
644
645 struct balance_callback {
646 struct balance_callback *next;
647 void (*func)(struct rq *rq);
648 };
649
650 /* CFS-related fields in a runqueue */
651 struct cfs_rq {
652 struct load_weight load;
653 unsigned int nr_queued;
654 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */
655 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */
656 unsigned int h_nr_idle; /* SCHED_IDLE */
657
658 s64 avg_vruntime;
659 u64 avg_load;
660
661 u64 min_vruntime;
662 #ifdef CONFIG_SCHED_CORE
663 unsigned int forceidle_seq;
664 u64 min_vruntime_fi;
665 #endif
666
667 struct rb_root_cached tasks_timeline;
668
669 /*
670 * 'curr' points to currently running entity on this cfs_rq.
671 * It is set to NULL otherwise (i.e when none are currently running).
672 */
673 struct sched_entity *curr;
674 struct sched_entity *next;
675
676 #ifdef CONFIG_SMP
677 /*
678 * CFS load tracking
679 */
680 struct sched_avg avg;
681 #ifndef CONFIG_64BIT
682 u64 last_update_time_copy;
683 #endif
684 struct {
685 raw_spinlock_t lock ____cacheline_aligned;
686 int nr;
687 unsigned long load_avg;
688 unsigned long util_avg;
689 unsigned long runnable_avg;
690 } removed;
691
692 #ifdef CONFIG_FAIR_GROUP_SCHED
693 u64 last_update_tg_load_avg;
694 unsigned long tg_load_avg_contrib;
695 long propagate;
696 long prop_runnable_sum;
697
698 /*
699 * h_load = weight * f(tg)
700 *
701 * Where f(tg) is the recursive weight fraction assigned to
702 * this group.
703 */
704 unsigned long h_load;
705 u64 last_h_load_update;
706 struct sched_entity *h_load_next;
707 #endif /* CONFIG_FAIR_GROUP_SCHED */
708 #endif /* CONFIG_SMP */
709
710 #ifdef CONFIG_FAIR_GROUP_SCHED
711 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */
712
713 /*
714 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
715 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
716 * (like users, containers etc.)
717 *
718 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
719 * This list is used during load balance.
720 */
721 int on_list;
722 struct list_head leaf_cfs_rq_list;
723 struct task_group *tg; /* group that "owns" this runqueue */
724
725 /* Locally cached copy of our task_group's idle value */
726 int idle;
727
728 #ifdef CONFIG_CFS_BANDWIDTH
729 int runtime_enabled;
730 s64 runtime_remaining;
731
732 u64 throttled_pelt_idle;
733 #ifndef CONFIG_64BIT
734 u64 throttled_pelt_idle_copy;
735 #endif
736 u64 throttled_clock;
737 u64 throttled_clock_pelt;
738 u64 throttled_clock_pelt_time;
739 u64 throttled_clock_self;
740 u64 throttled_clock_self_time;
741 int throttled;
742 int throttle_count;
743 struct list_head throttled_list;
744 struct list_head throttled_csd_list;
745 #endif /* CONFIG_CFS_BANDWIDTH */
746 #endif /* CONFIG_FAIR_GROUP_SCHED */
747 };
748
749 #ifdef CONFIG_SCHED_CLASS_EXT
750 /* scx_rq->flags, protected by the rq lock */
751 enum scx_rq_flags {
752 /*
753 * A hotplugged CPU starts scheduling before rq_online_scx(). Track
754 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called
755 * only while the BPF scheduler considers the CPU to be online.
756 */
757 SCX_RQ_ONLINE = 1 << 0,
758 SCX_RQ_CAN_STOP_TICK = 1 << 1,
759 SCX_RQ_BAL_PENDING = 1 << 2, /* balance hasn't run yet */
760 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
761 SCX_RQ_BYPASSING = 1 << 4,
762 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */
763
764 SCX_RQ_IN_WAKEUP = 1 << 16,
765 SCX_RQ_IN_BALANCE = 1 << 17,
766 };
767
768 struct scx_rq {
769 struct scx_dispatch_q local_dsq;
770 struct list_head runnable_list; /* runnable tasks on this rq */
771 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
772 unsigned long ops_qseq;
773 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
774 u32 nr_running;
775 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
776 bool cpu_released;
777 u32 flags;
778 u64 clock; /* current per-rq clock -- see scx_bpf_now() */
779 cpumask_var_t cpus_to_kick;
780 cpumask_var_t cpus_to_kick_if_idle;
781 cpumask_var_t cpus_to_preempt;
782 cpumask_var_t cpus_to_wait;
783 unsigned long pnt_seq;
784 struct balance_callback deferred_bal_cb;
785 struct irq_work deferred_irq_work;
786 struct irq_work kick_cpus_irq_work;
787 };
788 #endif /* CONFIG_SCHED_CLASS_EXT */
789
rt_bandwidth_enabled(void)790 static inline int rt_bandwidth_enabled(void)
791 {
792 return sysctl_sched_rt_runtime >= 0;
793 }
794
795 /* RT IPI pull logic requires IRQ_WORK */
796 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
797 # define HAVE_RT_PUSH_IPI
798 #endif
799
800 /* Real-Time classes' related field in a runqueue: */
801 struct rt_rq {
802 struct rt_prio_array active;
803 unsigned int rt_nr_running;
804 unsigned int rr_nr_running;
805 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
806 struct {
807 int curr; /* highest queued rt task prio */
808 #ifdef CONFIG_SMP
809 int next; /* next highest */
810 #endif
811 } highest_prio;
812 #endif
813 #ifdef CONFIG_SMP
814 bool overloaded;
815 struct plist_head pushable_tasks;
816
817 #endif /* CONFIG_SMP */
818 int rt_queued;
819
820 #ifdef CONFIG_RT_GROUP_SCHED
821 int rt_throttled;
822 u64 rt_time;
823 u64 rt_runtime;
824 /* Nests inside the rq lock: */
825 raw_spinlock_t rt_runtime_lock;
826
827 unsigned int rt_nr_boosted;
828
829 struct rq *rq;
830 struct task_group *tg;
831 #endif
832 };
833
rt_rq_is_runnable(struct rt_rq * rt_rq)834 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
835 {
836 return rt_rq->rt_queued && rt_rq->rt_nr_running;
837 }
838
839 /* Deadline class' related fields in a runqueue */
840 struct dl_rq {
841 /* runqueue is an rbtree, ordered by deadline */
842 struct rb_root_cached root;
843
844 unsigned int dl_nr_running;
845
846 #ifdef CONFIG_SMP
847 /*
848 * Deadline values of the currently executing and the
849 * earliest ready task on this rq. Caching these facilitates
850 * the decision whether or not a ready but not running task
851 * should migrate somewhere else.
852 */
853 struct {
854 u64 curr;
855 u64 next;
856 } earliest_dl;
857
858 bool overloaded;
859
860 /*
861 * Tasks on this rq that can be pushed away. They are kept in
862 * an rb-tree, ordered by tasks' deadlines, with caching
863 * of the leftmost (earliest deadline) element.
864 */
865 struct rb_root_cached pushable_dl_tasks_root;
866 #else
867 struct dl_bw dl_bw;
868 #endif
869 /*
870 * "Active utilization" for this runqueue: increased when a
871 * task wakes up (becomes TASK_RUNNING) and decreased when a
872 * task blocks
873 */
874 u64 running_bw;
875
876 /*
877 * Utilization of the tasks "assigned" to this runqueue (including
878 * the tasks that are in runqueue and the tasks that executed on this
879 * CPU and blocked). Increased when a task moves to this runqueue, and
880 * decreased when the task moves away (migrates, changes scheduling
881 * policy, or terminates).
882 * This is needed to compute the "inactive utilization" for the
883 * runqueue (inactive utilization = this_bw - running_bw).
884 */
885 u64 this_bw;
886 u64 extra_bw;
887
888 /*
889 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
890 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
891 */
892 u64 max_bw;
893
894 /*
895 * Inverse of the fraction of CPU utilization that can be reclaimed
896 * by the GRUB algorithm.
897 */
898 u64 bw_ratio;
899 };
900
901 #ifdef CONFIG_FAIR_GROUP_SCHED
902
903 /* An entity is a task if it doesn't "own" a runqueue */
904 #define entity_is_task(se) (!se->my_q)
905
se_update_runnable(struct sched_entity * se)906 static inline void se_update_runnable(struct sched_entity *se)
907 {
908 if (!entity_is_task(se))
909 se->runnable_weight = se->my_q->h_nr_runnable;
910 }
911
se_runnable(struct sched_entity * se)912 static inline long se_runnable(struct sched_entity *se)
913 {
914 if (se->sched_delayed)
915 return false;
916
917 if (entity_is_task(se))
918 return !!se->on_rq;
919 else
920 return se->runnable_weight;
921 }
922
923 #else /* !CONFIG_FAIR_GROUP_SCHED: */
924
925 #define entity_is_task(se) 1
926
se_update_runnable(struct sched_entity * se)927 static inline void se_update_runnable(struct sched_entity *se) { }
928
se_runnable(struct sched_entity * se)929 static inline long se_runnable(struct sched_entity *se)
930 {
931 if (se->sched_delayed)
932 return false;
933
934 return !!se->on_rq;
935 }
936
937 #endif /* !CONFIG_FAIR_GROUP_SCHED */
938
939 #ifdef CONFIG_SMP
940 /*
941 * XXX we want to get rid of these helpers and use the full load resolution.
942 */
se_weight(struct sched_entity * se)943 static inline long se_weight(struct sched_entity *se)
944 {
945 return scale_load_down(se->load.weight);
946 }
947
948
sched_asym_prefer(int a,int b)949 static inline bool sched_asym_prefer(int a, int b)
950 {
951 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
952 }
953
954 struct perf_domain {
955 struct em_perf_domain *em_pd;
956 struct perf_domain *next;
957 struct rcu_head rcu;
958 };
959
960 /*
961 * We add the notion of a root-domain which will be used to define per-domain
962 * variables. Each exclusive cpuset essentially defines an island domain by
963 * fully partitioning the member CPUs from any other cpuset. Whenever a new
964 * exclusive cpuset is created, we also create and attach a new root-domain
965 * object.
966 *
967 */
968 struct root_domain {
969 atomic_t refcount;
970 atomic_t rto_count;
971 struct rcu_head rcu;
972 cpumask_var_t span;
973 cpumask_var_t online;
974
975 /*
976 * Indicate pullable load on at least one CPU, e.g:
977 * - More than one runnable task
978 * - Running task is misfit
979 */
980 bool overloaded;
981
982 /* Indicate one or more CPUs over-utilized (tipping point) */
983 bool overutilized;
984
985 /*
986 * The bit corresponding to a CPU gets set here if such CPU has more
987 * than one runnable -deadline task (as it is below for RT tasks).
988 */
989 cpumask_var_t dlo_mask;
990 atomic_t dlo_count;
991 struct dl_bw dl_bw;
992 struct cpudl cpudl;
993
994 /*
995 * Indicate whether a root_domain's dl_bw has been checked or
996 * updated. It's monotonously increasing value.
997 *
998 * Also, some corner cases, like 'wrap around' is dangerous, but given
999 * that u64 is 'big enough'. So that shouldn't be a concern.
1000 */
1001 u64 visit_gen;
1002
1003 #ifdef HAVE_RT_PUSH_IPI
1004 /*
1005 * For IPI pull requests, loop across the rto_mask.
1006 */
1007 struct irq_work rto_push_work;
1008 raw_spinlock_t rto_lock;
1009 /* These are only updated and read within rto_lock */
1010 int rto_loop;
1011 int rto_cpu;
1012 /* These atomics are updated outside of a lock */
1013 atomic_t rto_loop_next;
1014 atomic_t rto_loop_start;
1015 #endif
1016 /*
1017 * The "RT overload" flag: it gets set if a CPU has more than
1018 * one runnable RT task.
1019 */
1020 cpumask_var_t rto_mask;
1021 struct cpupri cpupri;
1022
1023 /*
1024 * NULL-terminated list of performance domains intersecting with the
1025 * CPUs of the rd. Protected by RCU.
1026 */
1027 struct perf_domain __rcu *pd;
1028 };
1029
1030 extern void init_defrootdomain(void);
1031 extern int sched_init_domains(const struct cpumask *cpu_map);
1032 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1033 extern void sched_get_rd(struct root_domain *rd);
1034 extern void sched_put_rd(struct root_domain *rd);
1035
get_rd_overloaded(struct root_domain * rd)1036 static inline int get_rd_overloaded(struct root_domain *rd)
1037 {
1038 return READ_ONCE(rd->overloaded);
1039 }
1040
set_rd_overloaded(struct root_domain * rd,int status)1041 static inline void set_rd_overloaded(struct root_domain *rd, int status)
1042 {
1043 if (get_rd_overloaded(rd) != status)
1044 WRITE_ONCE(rd->overloaded, status);
1045 }
1046
1047 #ifdef HAVE_RT_PUSH_IPI
1048 extern void rto_push_irq_work_func(struct irq_work *work);
1049 #endif
1050 #endif /* CONFIG_SMP */
1051
1052 #ifdef CONFIG_UCLAMP_TASK
1053 /*
1054 * struct uclamp_bucket - Utilization clamp bucket
1055 * @value: utilization clamp value for tasks on this clamp bucket
1056 * @tasks: number of RUNNABLE tasks on this clamp bucket
1057 *
1058 * Keep track of how many tasks are RUNNABLE for a given utilization
1059 * clamp value.
1060 */
1061 struct uclamp_bucket {
1062 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1063 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1064 };
1065
1066 /*
1067 * struct uclamp_rq - rq's utilization clamp
1068 * @value: currently active clamp values for a rq
1069 * @bucket: utilization clamp buckets affecting a rq
1070 *
1071 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1072 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1073 * (or actually running) with that value.
1074 *
1075 * There are up to UCLAMP_CNT possible different clamp values, currently there
1076 * are only two: minimum utilization and maximum utilization.
1077 *
1078 * All utilization clamping values are MAX aggregated, since:
1079 * - for util_min: we want to run the CPU at least at the max of the minimum
1080 * utilization required by its currently RUNNABLE tasks.
1081 * - for util_max: we want to allow the CPU to run up to the max of the
1082 * maximum utilization allowed by its currently RUNNABLE tasks.
1083 *
1084 * Since on each system we expect only a limited number of different
1085 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1086 * the metrics required to compute all the per-rq utilization clamp values.
1087 */
1088 struct uclamp_rq {
1089 unsigned int value;
1090 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1091 };
1092
1093 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1094 #endif /* CONFIG_UCLAMP_TASK */
1095
1096 /*
1097 * This is the main, per-CPU runqueue data structure.
1098 *
1099 * Locking rule: those places that want to lock multiple runqueues
1100 * (such as the load balancing or the thread migration code), lock
1101 * acquire operations must be ordered by ascending &runqueue.
1102 */
1103 struct rq {
1104 /* runqueue lock: */
1105 raw_spinlock_t __lock;
1106
1107 unsigned int nr_running;
1108 #ifdef CONFIG_NUMA_BALANCING
1109 unsigned int nr_numa_running;
1110 unsigned int nr_preferred_running;
1111 unsigned int numa_migrate_on;
1112 #endif
1113 #ifdef CONFIG_NO_HZ_COMMON
1114 #ifdef CONFIG_SMP
1115 unsigned long last_blocked_load_update_tick;
1116 unsigned int has_blocked_load;
1117 call_single_data_t nohz_csd;
1118 #endif /* CONFIG_SMP */
1119 unsigned int nohz_tick_stopped;
1120 atomic_t nohz_flags;
1121 #endif /* CONFIG_NO_HZ_COMMON */
1122
1123 #ifdef CONFIG_SMP
1124 unsigned int ttwu_pending;
1125 #endif
1126 u64 nr_switches;
1127
1128 #ifdef CONFIG_UCLAMP_TASK
1129 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1130 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1131 unsigned int uclamp_flags;
1132 #define UCLAMP_FLAG_IDLE 0x01
1133 #endif
1134
1135 struct cfs_rq cfs;
1136 struct rt_rq rt;
1137 struct dl_rq dl;
1138 #ifdef CONFIG_SCHED_CLASS_EXT
1139 struct scx_rq scx;
1140 #endif
1141
1142 struct sched_dl_entity fair_server;
1143
1144 #ifdef CONFIG_FAIR_GROUP_SCHED
1145 /* list of leaf cfs_rq on this CPU: */
1146 struct list_head leaf_cfs_rq_list;
1147 struct list_head *tmp_alone_branch;
1148 #endif /* CONFIG_FAIR_GROUP_SCHED */
1149
1150 /*
1151 * This is part of a global counter where only the total sum
1152 * over all CPUs matters. A task can increase this counter on
1153 * one CPU and if it got migrated afterwards it may decrease
1154 * it on another CPU. Always updated under the runqueue lock:
1155 */
1156 unsigned int nr_uninterruptible;
1157
1158 union {
1159 struct task_struct __rcu *donor; /* Scheduler context */
1160 struct task_struct __rcu *curr; /* Execution context */
1161 };
1162 struct sched_dl_entity *dl_server;
1163 struct task_struct *idle;
1164 struct task_struct *stop;
1165 unsigned long next_balance;
1166 struct mm_struct *prev_mm;
1167
1168 unsigned int clock_update_flags;
1169 u64 clock;
1170 /* Ensure that all clocks are in the same cache line */
1171 u64 clock_task ____cacheline_aligned;
1172 u64 clock_pelt;
1173 unsigned long lost_idle_time;
1174 u64 clock_pelt_idle;
1175 u64 clock_idle;
1176 #ifndef CONFIG_64BIT
1177 u64 clock_pelt_idle_copy;
1178 u64 clock_idle_copy;
1179 #endif
1180
1181 atomic_t nr_iowait;
1182
1183 #ifdef CONFIG_SCHED_DEBUG
1184 u64 last_seen_need_resched_ns;
1185 int ticks_without_resched;
1186 #endif
1187
1188 #ifdef CONFIG_MEMBARRIER
1189 int membarrier_state;
1190 #endif
1191
1192 #ifdef CONFIG_SMP
1193 struct root_domain *rd;
1194 struct sched_domain __rcu *sd;
1195
1196 unsigned long cpu_capacity;
1197
1198 struct balance_callback *balance_callback;
1199
1200 unsigned char nohz_idle_balance;
1201 unsigned char idle_balance;
1202
1203 unsigned long misfit_task_load;
1204
1205 /* For active balancing */
1206 int active_balance;
1207 int push_cpu;
1208 struct cpu_stop_work active_balance_work;
1209
1210 /* CPU of this runqueue: */
1211 int cpu;
1212 int online;
1213
1214 struct list_head cfs_tasks;
1215
1216 struct sched_avg avg_rt;
1217 struct sched_avg avg_dl;
1218 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1219 struct sched_avg avg_irq;
1220 #endif
1221 #ifdef CONFIG_SCHED_HW_PRESSURE
1222 struct sched_avg avg_hw;
1223 #endif
1224 u64 idle_stamp;
1225 u64 avg_idle;
1226
1227 /* This is used to determine avg_idle's max value */
1228 u64 max_idle_balance_cost;
1229
1230 #ifdef CONFIG_HOTPLUG_CPU
1231 struct rcuwait hotplug_wait;
1232 #endif
1233 #endif /* CONFIG_SMP */
1234
1235 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1236 u64 prev_irq_time;
1237 u64 psi_irq_time;
1238 #endif
1239 #ifdef CONFIG_PARAVIRT
1240 u64 prev_steal_time;
1241 #endif
1242 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1243 u64 prev_steal_time_rq;
1244 #endif
1245
1246 /* calc_load related fields */
1247 unsigned long calc_load_update;
1248 long calc_load_active;
1249
1250 #ifdef CONFIG_SCHED_HRTICK
1251 #ifdef CONFIG_SMP
1252 call_single_data_t hrtick_csd;
1253 #endif
1254 struct hrtimer hrtick_timer;
1255 ktime_t hrtick_time;
1256 #endif
1257
1258 #ifdef CONFIG_SCHEDSTATS
1259 /* latency stats */
1260 struct sched_info rq_sched_info;
1261 unsigned long long rq_cpu_time;
1262
1263 /* sys_sched_yield() stats */
1264 unsigned int yld_count;
1265
1266 /* schedule() stats */
1267 unsigned int sched_count;
1268 unsigned int sched_goidle;
1269
1270 /* try_to_wake_up() stats */
1271 unsigned int ttwu_count;
1272 unsigned int ttwu_local;
1273 #endif
1274
1275 #ifdef CONFIG_CPU_IDLE
1276 /* Must be inspected within a RCU lock section */
1277 struct cpuidle_state *idle_state;
1278 #endif
1279
1280 #ifdef CONFIG_SMP
1281 unsigned int nr_pinned;
1282 #endif
1283 unsigned int push_busy;
1284 struct cpu_stop_work push_work;
1285
1286 #ifdef CONFIG_SCHED_CORE
1287 /* per rq */
1288 struct rq *core;
1289 struct task_struct *core_pick;
1290 struct sched_dl_entity *core_dl_server;
1291 unsigned int core_enabled;
1292 unsigned int core_sched_seq;
1293 struct rb_root core_tree;
1294
1295 /* shared state -- careful with sched_core_cpu_deactivate() */
1296 unsigned int core_task_seq;
1297 unsigned int core_pick_seq;
1298 unsigned long core_cookie;
1299 unsigned int core_forceidle_count;
1300 unsigned int core_forceidle_seq;
1301 unsigned int core_forceidle_occupation;
1302 u64 core_forceidle_start;
1303 #endif
1304
1305 /* Scratch cpumask to be temporarily used under rq_lock */
1306 cpumask_var_t scratch_mask;
1307
1308 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1309 call_single_data_t cfsb_csd;
1310 struct list_head cfsb_csd_list;
1311 #endif
1312 };
1313
1314 #ifdef CONFIG_FAIR_GROUP_SCHED
1315
1316 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1317 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1318 {
1319 return cfs_rq->rq;
1320 }
1321
1322 #else
1323
rq_of(struct cfs_rq * cfs_rq)1324 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1325 {
1326 return container_of(cfs_rq, struct rq, cfs);
1327 }
1328 #endif
1329
cpu_of(struct rq * rq)1330 static inline int cpu_of(struct rq *rq)
1331 {
1332 #ifdef CONFIG_SMP
1333 return rq->cpu;
1334 #else
1335 return 0;
1336 #endif
1337 }
1338
1339 #define MDF_PUSH 0x01
1340
is_migration_disabled(struct task_struct * p)1341 static inline bool is_migration_disabled(struct task_struct *p)
1342 {
1343 #ifdef CONFIG_SMP
1344 return p->migration_disabled;
1345 #else
1346 return false;
1347 #endif
1348 }
1349
1350 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1351
1352 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1353 #define this_rq() this_cpu_ptr(&runqueues)
1354 #define task_rq(p) cpu_rq(task_cpu(p))
1355 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1356 #define raw_rq() raw_cpu_ptr(&runqueues)
1357
rq_set_donor(struct rq * rq,struct task_struct * t)1358 static inline void rq_set_donor(struct rq *rq, struct task_struct *t)
1359 {
1360 /* Do nothing */
1361 }
1362
1363 #ifdef CONFIG_SCHED_CORE
1364 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1365
1366 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1367
sched_core_enabled(struct rq * rq)1368 static inline bool sched_core_enabled(struct rq *rq)
1369 {
1370 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1371 }
1372
sched_core_disabled(void)1373 static inline bool sched_core_disabled(void)
1374 {
1375 return !static_branch_unlikely(&__sched_core_enabled);
1376 }
1377
1378 /*
1379 * Be careful with this function; not for general use. The return value isn't
1380 * stable unless you actually hold a relevant rq->__lock.
1381 */
rq_lockp(struct rq * rq)1382 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1383 {
1384 if (sched_core_enabled(rq))
1385 return &rq->core->__lock;
1386
1387 return &rq->__lock;
1388 }
1389
__rq_lockp(struct rq * rq)1390 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1391 {
1392 if (rq->core_enabled)
1393 return &rq->core->__lock;
1394
1395 return &rq->__lock;
1396 }
1397
1398 extern bool
1399 cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1400
1401 extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1402
1403 /*
1404 * Helpers to check if the CPU's core cookie matches with the task's cookie
1405 * when core scheduling is enabled.
1406 * A special case is that the task's cookie always matches with CPU's core
1407 * cookie if the CPU is in an idle core.
1408 */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1409 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1410 {
1411 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1412 if (!sched_core_enabled(rq))
1413 return true;
1414
1415 return rq->core->core_cookie == p->core_cookie;
1416 }
1417
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1418 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1419 {
1420 bool idle_core = true;
1421 int cpu;
1422
1423 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1424 if (!sched_core_enabled(rq))
1425 return true;
1426
1427 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1428 if (!available_idle_cpu(cpu)) {
1429 idle_core = false;
1430 break;
1431 }
1432 }
1433
1434 /*
1435 * A CPU in an idle core is always the best choice for tasks with
1436 * cookies.
1437 */
1438 return idle_core || rq->core->core_cookie == p->core_cookie;
1439 }
1440
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1441 static inline bool sched_group_cookie_match(struct rq *rq,
1442 struct task_struct *p,
1443 struct sched_group *group)
1444 {
1445 int cpu;
1446
1447 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1448 if (!sched_core_enabled(rq))
1449 return true;
1450
1451 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1452 if (sched_core_cookie_match(cpu_rq(cpu), p))
1453 return true;
1454 }
1455 return false;
1456 }
1457
sched_core_enqueued(struct task_struct * p)1458 static inline bool sched_core_enqueued(struct task_struct *p)
1459 {
1460 return !RB_EMPTY_NODE(&p->core_node);
1461 }
1462
1463 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1464 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1465
1466 extern void sched_core_get(void);
1467 extern void sched_core_put(void);
1468
1469 #else /* !CONFIG_SCHED_CORE: */
1470
sched_core_enabled(struct rq * rq)1471 static inline bool sched_core_enabled(struct rq *rq)
1472 {
1473 return false;
1474 }
1475
sched_core_disabled(void)1476 static inline bool sched_core_disabled(void)
1477 {
1478 return true;
1479 }
1480
rq_lockp(struct rq * rq)1481 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1482 {
1483 return &rq->__lock;
1484 }
1485
__rq_lockp(struct rq * rq)1486 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1487 {
1488 return &rq->__lock;
1489 }
1490
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1491 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1492 {
1493 return true;
1494 }
1495
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1496 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1497 {
1498 return true;
1499 }
1500
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1501 static inline bool sched_group_cookie_match(struct rq *rq,
1502 struct task_struct *p,
1503 struct sched_group *group)
1504 {
1505 return true;
1506 }
1507
1508 #endif /* !CONFIG_SCHED_CORE */
1509
lockdep_assert_rq_held(struct rq * rq)1510 static inline void lockdep_assert_rq_held(struct rq *rq)
1511 {
1512 lockdep_assert_held(__rq_lockp(rq));
1513 }
1514
1515 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1516 extern bool raw_spin_rq_trylock(struct rq *rq);
1517 extern void raw_spin_rq_unlock(struct rq *rq);
1518
raw_spin_rq_lock(struct rq * rq)1519 static inline void raw_spin_rq_lock(struct rq *rq)
1520 {
1521 raw_spin_rq_lock_nested(rq, 0);
1522 }
1523
raw_spin_rq_lock_irq(struct rq * rq)1524 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1525 {
1526 local_irq_disable();
1527 raw_spin_rq_lock(rq);
1528 }
1529
raw_spin_rq_unlock_irq(struct rq * rq)1530 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1531 {
1532 raw_spin_rq_unlock(rq);
1533 local_irq_enable();
1534 }
1535
_raw_spin_rq_lock_irqsave(struct rq * rq)1536 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1537 {
1538 unsigned long flags;
1539
1540 local_irq_save(flags);
1541 raw_spin_rq_lock(rq);
1542
1543 return flags;
1544 }
1545
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1546 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1547 {
1548 raw_spin_rq_unlock(rq);
1549 local_irq_restore(flags);
1550 }
1551
1552 #define raw_spin_rq_lock_irqsave(rq, flags) \
1553 do { \
1554 flags = _raw_spin_rq_lock_irqsave(rq); \
1555 } while (0)
1556
1557 #ifdef CONFIG_SCHED_SMT
1558 extern void __update_idle_core(struct rq *rq);
1559
update_idle_core(struct rq * rq)1560 static inline void update_idle_core(struct rq *rq)
1561 {
1562 if (static_branch_unlikely(&sched_smt_present))
1563 __update_idle_core(rq);
1564 }
1565
1566 #else
update_idle_core(struct rq * rq)1567 static inline void update_idle_core(struct rq *rq) { }
1568 #endif
1569
1570 #ifdef CONFIG_FAIR_GROUP_SCHED
1571
task_of(struct sched_entity * se)1572 static inline struct task_struct *task_of(struct sched_entity *se)
1573 {
1574 SCHED_WARN_ON(!entity_is_task(se));
1575 return container_of(se, struct task_struct, se);
1576 }
1577
task_cfs_rq(struct task_struct * p)1578 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1579 {
1580 return p->se.cfs_rq;
1581 }
1582
1583 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(const struct sched_entity * se)1584 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1585 {
1586 return se->cfs_rq;
1587 }
1588
1589 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1590 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1591 {
1592 return grp->my_q;
1593 }
1594
1595 #else /* !CONFIG_FAIR_GROUP_SCHED: */
1596
1597 #define task_of(_se) container_of(_se, struct task_struct, se)
1598
task_cfs_rq(const struct task_struct * p)1599 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1600 {
1601 return &task_rq(p)->cfs;
1602 }
1603
cfs_rq_of(const struct sched_entity * se)1604 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1605 {
1606 const struct task_struct *p = task_of(se);
1607 struct rq *rq = task_rq(p);
1608
1609 return &rq->cfs;
1610 }
1611
1612 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1613 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1614 {
1615 return NULL;
1616 }
1617
1618 #endif /* !CONFIG_FAIR_GROUP_SCHED */
1619
1620 extern void update_rq_clock(struct rq *rq);
1621
1622 /*
1623 * rq::clock_update_flags bits
1624 *
1625 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1626 * call to __schedule(). This is an optimisation to avoid
1627 * neighbouring rq clock updates.
1628 *
1629 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1630 * in effect and calls to update_rq_clock() are being ignored.
1631 *
1632 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1633 * made to update_rq_clock() since the last time rq::lock was pinned.
1634 *
1635 * If inside of __schedule(), clock_update_flags will have been
1636 * shifted left (a left shift is a cheap operation for the fast path
1637 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1638 *
1639 * if (rq-clock_update_flags >= RQCF_UPDATED)
1640 *
1641 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1642 * one position though, because the next rq_unpin_lock() will shift it
1643 * back.
1644 */
1645 #define RQCF_REQ_SKIP 0x01
1646 #define RQCF_ACT_SKIP 0x02
1647 #define RQCF_UPDATED 0x04
1648
assert_clock_updated(struct rq * rq)1649 static inline void assert_clock_updated(struct rq *rq)
1650 {
1651 /*
1652 * The only reason for not seeing a clock update since the
1653 * last rq_pin_lock() is if we're currently skipping updates.
1654 */
1655 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1656 }
1657
rq_clock(struct rq * rq)1658 static inline u64 rq_clock(struct rq *rq)
1659 {
1660 lockdep_assert_rq_held(rq);
1661 assert_clock_updated(rq);
1662
1663 return rq->clock;
1664 }
1665
rq_clock_task(struct rq * rq)1666 static inline u64 rq_clock_task(struct rq *rq)
1667 {
1668 lockdep_assert_rq_held(rq);
1669 assert_clock_updated(rq);
1670
1671 return rq->clock_task;
1672 }
1673
rq_clock_skip_update(struct rq * rq)1674 static inline void rq_clock_skip_update(struct rq *rq)
1675 {
1676 lockdep_assert_rq_held(rq);
1677 rq->clock_update_flags |= RQCF_REQ_SKIP;
1678 }
1679
1680 /*
1681 * See rt task throttling, which is the only time a skip
1682 * request is canceled.
1683 */
rq_clock_cancel_skipupdate(struct rq * rq)1684 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1685 {
1686 lockdep_assert_rq_held(rq);
1687 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1688 }
1689
1690 /*
1691 * During cpu offlining and rq wide unthrottling, we can trigger
1692 * an update_rq_clock() for several cfs and rt runqueues (Typically
1693 * when using list_for_each_entry_*)
1694 * rq_clock_start_loop_update() can be called after updating the clock
1695 * once and before iterating over the list to prevent multiple update.
1696 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1697 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1698 */
rq_clock_start_loop_update(struct rq * rq)1699 static inline void rq_clock_start_loop_update(struct rq *rq)
1700 {
1701 lockdep_assert_rq_held(rq);
1702 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1703 rq->clock_update_flags |= RQCF_ACT_SKIP;
1704 }
1705
rq_clock_stop_loop_update(struct rq * rq)1706 static inline void rq_clock_stop_loop_update(struct rq *rq)
1707 {
1708 lockdep_assert_rq_held(rq);
1709 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1710 }
1711
1712 struct rq_flags {
1713 unsigned long flags;
1714 struct pin_cookie cookie;
1715 #ifdef CONFIG_SCHED_DEBUG
1716 /*
1717 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1718 * current pin context is stashed here in case it needs to be
1719 * restored in rq_repin_lock().
1720 */
1721 unsigned int clock_update_flags;
1722 #endif
1723 };
1724
1725 extern struct balance_callback balance_push_callback;
1726
1727 #ifdef CONFIG_SCHED_CLASS_EXT
1728 extern const struct sched_class ext_sched_class;
1729
1730 DECLARE_STATIC_KEY_FALSE(__scx_ops_enabled); /* SCX BPF scheduler loaded */
1731 DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
1732
1733 #define scx_enabled() static_branch_unlikely(&__scx_ops_enabled)
1734 #define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
1735
scx_rq_clock_update(struct rq * rq,u64 clock)1736 static inline void scx_rq_clock_update(struct rq *rq, u64 clock)
1737 {
1738 if (!scx_enabled())
1739 return;
1740 WRITE_ONCE(rq->scx.clock, clock);
1741 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID);
1742 }
1743
scx_rq_clock_invalidate(struct rq * rq)1744 static inline void scx_rq_clock_invalidate(struct rq *rq)
1745 {
1746 if (!scx_enabled())
1747 return;
1748 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID);
1749 }
1750
1751 #else /* !CONFIG_SCHED_CLASS_EXT */
1752 #define scx_enabled() false
1753 #define scx_switched_all() false
1754
scx_rq_clock_update(struct rq * rq,u64 clock)1755 static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {}
scx_rq_clock_invalidate(struct rq * rq)1756 static inline void scx_rq_clock_invalidate(struct rq *rq) {}
1757 #endif /* !CONFIG_SCHED_CLASS_EXT */
1758
1759 /*
1760 * Lockdep annotation that avoids accidental unlocks; it's like a
1761 * sticky/continuous lockdep_assert_held().
1762 *
1763 * This avoids code that has access to 'struct rq *rq' (basically everything in
1764 * the scheduler) from accidentally unlocking the rq if they do not also have a
1765 * copy of the (on-stack) 'struct rq_flags rf'.
1766 *
1767 * Also see Documentation/locking/lockdep-design.rst.
1768 */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1769 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1770 {
1771 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1772
1773 #ifdef CONFIG_SCHED_DEBUG
1774 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1775 rf->clock_update_flags = 0;
1776 # ifdef CONFIG_SMP
1777 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1778 # endif
1779 #endif
1780 }
1781
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1782 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1783 {
1784 #ifdef CONFIG_SCHED_DEBUG
1785 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1786 rf->clock_update_flags = RQCF_UPDATED;
1787 #endif
1788 scx_rq_clock_invalidate(rq);
1789 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1790 }
1791
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1792 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1793 {
1794 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1795
1796 #ifdef CONFIG_SCHED_DEBUG
1797 /*
1798 * Restore the value we stashed in @rf for this pin context.
1799 */
1800 rq->clock_update_flags |= rf->clock_update_flags;
1801 #endif
1802 }
1803
1804 extern
1805 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1806 __acquires(rq->lock);
1807
1808 extern
1809 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1810 __acquires(p->pi_lock)
1811 __acquires(rq->lock);
1812
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1813 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1814 __releases(rq->lock)
1815 {
1816 rq_unpin_lock(rq, rf);
1817 raw_spin_rq_unlock(rq);
1818 }
1819
1820 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1821 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1822 __releases(rq->lock)
1823 __releases(p->pi_lock)
1824 {
1825 rq_unpin_lock(rq, rf);
1826 raw_spin_rq_unlock(rq);
1827 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1828 }
1829
1830 DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1831 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1832 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1833 struct rq *rq; struct rq_flags rf)
1834
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1835 static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1836 __acquires(rq->lock)
1837 {
1838 raw_spin_rq_lock_irqsave(rq, rf->flags);
1839 rq_pin_lock(rq, rf);
1840 }
1841
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1842 static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1843 __acquires(rq->lock)
1844 {
1845 raw_spin_rq_lock_irq(rq);
1846 rq_pin_lock(rq, rf);
1847 }
1848
rq_lock(struct rq * rq,struct rq_flags * rf)1849 static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1850 __acquires(rq->lock)
1851 {
1852 raw_spin_rq_lock(rq);
1853 rq_pin_lock(rq, rf);
1854 }
1855
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1856 static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1857 __releases(rq->lock)
1858 {
1859 rq_unpin_lock(rq, rf);
1860 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1861 }
1862
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1863 static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1864 __releases(rq->lock)
1865 {
1866 rq_unpin_lock(rq, rf);
1867 raw_spin_rq_unlock_irq(rq);
1868 }
1869
rq_unlock(struct rq * rq,struct rq_flags * rf)1870 static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1871 __releases(rq->lock)
1872 {
1873 rq_unpin_lock(rq, rf);
1874 raw_spin_rq_unlock(rq);
1875 }
1876
1877 DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1878 rq_lock(_T->lock, &_T->rf),
1879 rq_unlock(_T->lock, &_T->rf),
1880 struct rq_flags rf)
1881
1882 DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1883 rq_lock_irq(_T->lock, &_T->rf),
1884 rq_unlock_irq(_T->lock, &_T->rf),
1885 struct rq_flags rf)
1886
1887 DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1888 rq_lock_irqsave(_T->lock, &_T->rf),
1889 rq_unlock_irqrestore(_T->lock, &_T->rf),
1890 struct rq_flags rf)
1891
this_rq_lock_irq(struct rq_flags * rf)1892 static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1893 __acquires(rq->lock)
1894 {
1895 struct rq *rq;
1896
1897 local_irq_disable();
1898 rq = this_rq();
1899 rq_lock(rq, rf);
1900
1901 return rq;
1902 }
1903
1904 #ifdef CONFIG_NUMA
1905
1906 enum numa_topology_type {
1907 NUMA_DIRECT,
1908 NUMA_GLUELESS_MESH,
1909 NUMA_BACKPLANE,
1910 };
1911
1912 extern enum numa_topology_type sched_numa_topology_type;
1913 extern int sched_max_numa_distance;
1914 extern bool find_numa_distance(int distance);
1915 extern void sched_init_numa(int offline_node);
1916 extern void sched_update_numa(int cpu, bool online);
1917 extern void sched_domains_numa_masks_set(unsigned int cpu);
1918 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1919 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1920
1921 #else /* !CONFIG_NUMA: */
1922
sched_init_numa(int offline_node)1923 static inline void sched_init_numa(int offline_node) { }
sched_update_numa(int cpu,bool online)1924 static inline void sched_update_numa(int cpu, bool online) { }
sched_domains_numa_masks_set(unsigned int cpu)1925 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1926 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1927
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1928 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1929 {
1930 return nr_cpu_ids;
1931 }
1932
1933 #endif /* !CONFIG_NUMA */
1934
1935 #ifdef CONFIG_NUMA_BALANCING
1936
1937 /* The regions in numa_faults array from task_struct */
1938 enum numa_faults_stats {
1939 NUMA_MEM = 0,
1940 NUMA_CPU,
1941 NUMA_MEMBUF,
1942 NUMA_CPUBUF
1943 };
1944
1945 extern void sched_setnuma(struct task_struct *p, int node);
1946 extern int migrate_task_to(struct task_struct *p, int cpu);
1947 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1948 int cpu, int scpu);
1949 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1950
1951 #else /* !CONFIG_NUMA_BALANCING: */
1952
1953 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1954 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1955 {
1956 }
1957
1958 #endif /* !CONFIG_NUMA_BALANCING */
1959
1960 #ifdef CONFIG_SMP
1961
1962 static inline void
queue_balance_callback(struct rq * rq,struct balance_callback * head,void (* func)(struct rq * rq))1963 queue_balance_callback(struct rq *rq,
1964 struct balance_callback *head,
1965 void (*func)(struct rq *rq))
1966 {
1967 lockdep_assert_rq_held(rq);
1968
1969 /*
1970 * Don't (re)queue an already queued item; nor queue anything when
1971 * balance_push() is active, see the comment with
1972 * balance_push_callback.
1973 */
1974 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1975 return;
1976
1977 head->func = func;
1978 head->next = rq->balance_callback;
1979 rq->balance_callback = head;
1980 }
1981
1982 #define rcu_dereference_check_sched_domain(p) \
1983 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1984
1985 /*
1986 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1987 * See destroy_sched_domains: call_rcu for details.
1988 *
1989 * The domain tree of any CPU may only be accessed from within
1990 * preempt-disabled sections.
1991 */
1992 #define for_each_domain(cpu, __sd) \
1993 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1994 __sd; __sd = __sd->parent)
1995
1996 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1997 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1998 static const unsigned int SD_SHARED_CHILD_MASK =
1999 #include <linux/sched/sd_flags.h>
2000 0;
2001 #undef SD_FLAG
2002
2003 /**
2004 * highest_flag_domain - Return highest sched_domain containing flag.
2005 * @cpu: The CPU whose highest level of sched domain is to
2006 * be returned.
2007 * @flag: The flag to check for the highest sched_domain
2008 * for the given CPU.
2009 *
2010 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
2011 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
2012 */
highest_flag_domain(int cpu,int flag)2013 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
2014 {
2015 struct sched_domain *sd, *hsd = NULL;
2016
2017 for_each_domain(cpu, sd) {
2018 if (sd->flags & flag) {
2019 hsd = sd;
2020 continue;
2021 }
2022
2023 /*
2024 * Stop the search if @flag is known to be shared at lower
2025 * levels. It will not be found further up.
2026 */
2027 if (flag & SD_SHARED_CHILD_MASK)
2028 break;
2029 }
2030
2031 return hsd;
2032 }
2033
lowest_flag_domain(int cpu,int flag)2034 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
2035 {
2036 struct sched_domain *sd;
2037
2038 for_each_domain(cpu, sd) {
2039 if (sd->flags & flag)
2040 break;
2041 }
2042
2043 return sd;
2044 }
2045
2046 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2047 DECLARE_PER_CPU(int, sd_llc_size);
2048 DECLARE_PER_CPU(int, sd_llc_id);
2049 DECLARE_PER_CPU(int, sd_share_id);
2050 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2051 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2052 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2053 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2054
2055 extern struct static_key_false sched_asym_cpucapacity;
2056 extern struct static_key_false sched_cluster_active;
2057
sched_asym_cpucap_active(void)2058 static __always_inline bool sched_asym_cpucap_active(void)
2059 {
2060 return static_branch_unlikely(&sched_asym_cpucapacity);
2061 }
2062
2063 struct sched_group_capacity {
2064 atomic_t ref;
2065 /*
2066 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2067 * for a single CPU.
2068 */
2069 unsigned long capacity;
2070 unsigned long min_capacity; /* Min per-CPU capacity in group */
2071 unsigned long max_capacity; /* Max per-CPU capacity in group */
2072 unsigned long next_update;
2073 int imbalance; /* XXX unrelated to capacity but shared group state */
2074
2075 #ifdef CONFIG_SCHED_DEBUG
2076 int id;
2077 #endif
2078
2079 unsigned long cpumask[]; /* Balance mask */
2080 };
2081
2082 struct sched_group {
2083 struct sched_group *next; /* Must be a circular list */
2084 atomic_t ref;
2085
2086 unsigned int group_weight;
2087 unsigned int cores;
2088 struct sched_group_capacity *sgc;
2089 int asym_prefer_cpu; /* CPU of highest priority in group */
2090 int flags;
2091
2092 /*
2093 * The CPUs this group covers.
2094 *
2095 * NOTE: this field is variable length. (Allocated dynamically
2096 * by attaching extra space to the end of the structure,
2097 * depending on how many CPUs the kernel has booted up with)
2098 */
2099 unsigned long cpumask[];
2100 };
2101
sched_group_span(struct sched_group * sg)2102 static inline struct cpumask *sched_group_span(struct sched_group *sg)
2103 {
2104 return to_cpumask(sg->cpumask);
2105 }
2106
2107 /*
2108 * See build_balance_mask().
2109 */
group_balance_mask(struct sched_group * sg)2110 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2111 {
2112 return to_cpumask(sg->sgc->cpumask);
2113 }
2114
2115 extern int group_balance_cpu(struct sched_group *sg);
2116
2117 #ifdef CONFIG_SCHED_DEBUG
2118 extern void update_sched_domain_debugfs(void);
2119 extern void dirty_sched_domain_sysctl(int cpu);
2120 #else
update_sched_domain_debugfs(void)2121 static inline void update_sched_domain_debugfs(void) { }
dirty_sched_domain_sysctl(int cpu)2122 static inline void dirty_sched_domain_sysctl(int cpu) { }
2123 #endif
2124
2125 extern int sched_update_scaling(void);
2126
task_user_cpus(struct task_struct * p)2127 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2128 {
2129 if (!p->user_cpus_ptr)
2130 return cpu_possible_mask; /* &init_task.cpus_mask */
2131 return p->user_cpus_ptr;
2132 }
2133
2134 #endif /* CONFIG_SMP */
2135
2136 #ifdef CONFIG_CGROUP_SCHED
2137
2138 /*
2139 * Return the group to which this tasks belongs.
2140 *
2141 * We cannot use task_css() and friends because the cgroup subsystem
2142 * changes that value before the cgroup_subsys::attach() method is called,
2143 * therefore we cannot pin it and might observe the wrong value.
2144 *
2145 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2146 * core changes this before calling sched_move_task().
2147 *
2148 * Instead we use a 'copy' which is updated from sched_move_task() while
2149 * holding both task_struct::pi_lock and rq::lock.
2150 */
task_group(struct task_struct * p)2151 static inline struct task_group *task_group(struct task_struct *p)
2152 {
2153 return p->sched_task_group;
2154 }
2155
2156 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
set_task_rq(struct task_struct * p,unsigned int cpu)2157 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2158 {
2159 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2160 struct task_group *tg = task_group(p);
2161 #endif
2162
2163 #ifdef CONFIG_FAIR_GROUP_SCHED
2164 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2165 p->se.cfs_rq = tg->cfs_rq[cpu];
2166 p->se.parent = tg->se[cpu];
2167 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2168 #endif
2169
2170 #ifdef CONFIG_RT_GROUP_SCHED
2171 p->rt.rt_rq = tg->rt_rq[cpu];
2172 p->rt.parent = tg->rt_se[cpu];
2173 #endif
2174 }
2175
2176 #else /* !CONFIG_CGROUP_SCHED: */
2177
set_task_rq(struct task_struct * p,unsigned int cpu)2178 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2179
task_group(struct task_struct * p)2180 static inline struct task_group *task_group(struct task_struct *p)
2181 {
2182 return NULL;
2183 }
2184
2185 #endif /* !CONFIG_CGROUP_SCHED */
2186
__set_task_cpu(struct task_struct * p,unsigned int cpu)2187 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2188 {
2189 set_task_rq(p, cpu);
2190 #ifdef CONFIG_SMP
2191 /*
2192 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2193 * successfully executed on another CPU. We must ensure that updates of
2194 * per-task data have been completed by this moment.
2195 */
2196 smp_wmb();
2197 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2198 p->wake_cpu = cpu;
2199 #endif
2200 }
2201
2202 /*
2203 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2204 */
2205 #ifdef CONFIG_SCHED_DEBUG
2206 # define const_debug __read_mostly
2207 #else
2208 # define const_debug const
2209 #endif
2210
2211 #define SCHED_FEAT(name, enabled) \
2212 __SCHED_FEAT_##name ,
2213
2214 enum {
2215 #include "features.h"
2216 __SCHED_FEAT_NR,
2217 };
2218
2219 #undef SCHED_FEAT
2220
2221 #ifdef CONFIG_SCHED_DEBUG
2222
2223 /*
2224 * To support run-time toggling of sched features, all the translation units
2225 * (but core.c) reference the sysctl_sched_features defined in core.c.
2226 */
2227 extern const_debug unsigned int sysctl_sched_features;
2228
2229 #ifdef CONFIG_JUMP_LABEL
2230
2231 #define SCHED_FEAT(name, enabled) \
2232 static __always_inline bool static_branch_##name(struct static_key *key) \
2233 { \
2234 return static_key_##enabled(key); \
2235 }
2236
2237 #include "features.h"
2238 #undef SCHED_FEAT
2239
2240 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2241 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2242
2243 #else /* !CONFIG_JUMP_LABEL: */
2244
2245 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2246
2247 #endif /* !CONFIG_JUMP_LABEL */
2248
2249 #else /* !SCHED_DEBUG: */
2250
2251 /*
2252 * Each translation unit has its own copy of sysctl_sched_features to allow
2253 * constants propagation at compile time and compiler optimization based on
2254 * features default.
2255 */
2256 #define SCHED_FEAT(name, enabled) \
2257 (1UL << __SCHED_FEAT_##name) * enabled |
2258 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2259 #include "features.h"
2260 0;
2261 #undef SCHED_FEAT
2262
2263 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2264
2265 #endif /* !SCHED_DEBUG */
2266
2267 extern struct static_key_false sched_numa_balancing;
2268 extern struct static_key_false sched_schedstats;
2269
global_rt_period(void)2270 static inline u64 global_rt_period(void)
2271 {
2272 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2273 }
2274
global_rt_runtime(void)2275 static inline u64 global_rt_runtime(void)
2276 {
2277 if (sysctl_sched_rt_runtime < 0)
2278 return RUNTIME_INF;
2279
2280 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2281 }
2282
2283 /*
2284 * Is p the current execution context?
2285 */
task_current(struct rq * rq,struct task_struct * p)2286 static inline int task_current(struct rq *rq, struct task_struct *p)
2287 {
2288 return rq->curr == p;
2289 }
2290
2291 /*
2292 * Is p the current scheduling context?
2293 *
2294 * Note that it might be the current execution context at the same time if
2295 * rq->curr == rq->donor == p.
2296 */
task_current_donor(struct rq * rq,struct task_struct * p)2297 static inline int task_current_donor(struct rq *rq, struct task_struct *p)
2298 {
2299 return rq->donor == p;
2300 }
2301
task_on_cpu(struct rq * rq,struct task_struct * p)2302 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2303 {
2304 #ifdef CONFIG_SMP
2305 return p->on_cpu;
2306 #else
2307 return task_current(rq, p);
2308 #endif
2309 }
2310
task_on_rq_queued(struct task_struct * p)2311 static inline int task_on_rq_queued(struct task_struct *p)
2312 {
2313 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED;
2314 }
2315
task_on_rq_migrating(struct task_struct * p)2316 static inline int task_on_rq_migrating(struct task_struct *p)
2317 {
2318 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2319 }
2320
2321 /* Wake flags. The first three directly map to some SD flag value */
2322 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2323 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2324 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2325
2326 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2327 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2328 #define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2329 #define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2330
2331 #ifdef CONFIG_SMP
2332 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2333 static_assert(WF_FORK == SD_BALANCE_FORK);
2334 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2335 #endif
2336
2337 /*
2338 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2339 * of tasks with abnormal "nice" values across CPUs the contribution that
2340 * each task makes to its run queue's load is weighted according to its
2341 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2342 * scaled version of the new time slice allocation that they receive on time
2343 * slice expiry etc.
2344 */
2345
2346 #define WEIGHT_IDLEPRIO 3
2347 #define WMULT_IDLEPRIO 1431655765
2348
2349 extern const int sched_prio_to_weight[40];
2350 extern const u32 sched_prio_to_wmult[40];
2351
2352 /*
2353 * {de,en}queue flags:
2354 *
2355 * DEQUEUE_SLEEP - task is no longer runnable
2356 * ENQUEUE_WAKEUP - task just became runnable
2357 *
2358 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2359 * are in a known state which allows modification. Such pairs
2360 * should preserve as much state as possible.
2361 *
2362 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2363 * in the runqueue.
2364 *
2365 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2366 *
2367 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2368 *
2369 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2370 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2371 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2372 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2373 *
2374 */
2375
2376 #define DEQUEUE_SLEEP 0x01 /* Matches ENQUEUE_WAKEUP */
2377 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2378 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2379 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2380 #define DEQUEUE_SPECIAL 0x10
2381 #define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2382 #define DEQUEUE_DELAYED 0x200 /* Matches ENQUEUE_DELAYED */
2383
2384 #define ENQUEUE_WAKEUP 0x01
2385 #define ENQUEUE_RESTORE 0x02
2386 #define ENQUEUE_MOVE 0x04
2387 #define ENQUEUE_NOCLOCK 0x08
2388
2389 #define ENQUEUE_HEAD 0x10
2390 #define ENQUEUE_REPLENISH 0x20
2391 #ifdef CONFIG_SMP
2392 #define ENQUEUE_MIGRATED 0x40
2393 #else
2394 #define ENQUEUE_MIGRATED 0x00
2395 #endif
2396 #define ENQUEUE_INITIAL 0x80
2397 #define ENQUEUE_MIGRATING 0x100
2398 #define ENQUEUE_DELAYED 0x200
2399 #define ENQUEUE_RQ_SELECTED 0x400
2400
2401 #define RETRY_TASK ((void *)-1UL)
2402
2403 struct affinity_context {
2404 const struct cpumask *new_mask;
2405 struct cpumask *user_mask;
2406 unsigned int flags;
2407 };
2408
2409 extern s64 update_curr_common(struct rq *rq);
2410
2411 struct sched_class {
2412
2413 #ifdef CONFIG_UCLAMP_TASK
2414 int uclamp_enabled;
2415 #endif
2416
2417 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2418 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2419 void (*yield_task) (struct rq *rq);
2420 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2421
2422 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2423
2424 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2425 struct task_struct *(*pick_task)(struct rq *rq);
2426 /*
2427 * Optional! When implemented pick_next_task() should be equivalent to:
2428 *
2429 * next = pick_task();
2430 * if (next) {
2431 * put_prev_task(prev);
2432 * set_next_task_first(next);
2433 * }
2434 */
2435 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev);
2436
2437 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2438 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2439
2440 #ifdef CONFIG_SMP
2441 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2442
2443 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2444
2445 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2446
2447 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2448
2449 void (*rq_online)(struct rq *rq);
2450 void (*rq_offline)(struct rq *rq);
2451
2452 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2453 #endif
2454
2455 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2456 void (*task_fork)(struct task_struct *p);
2457 void (*task_dead)(struct task_struct *p);
2458
2459 /*
2460 * The switched_from() call is allowed to drop rq->lock, therefore we
2461 * cannot assume the switched_from/switched_to pair is serialized by
2462 * rq->lock. They are however serialized by p->pi_lock.
2463 */
2464 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2465 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2466 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2467 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2468 const struct load_weight *lw);
2469 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2470 int oldprio);
2471
2472 unsigned int (*get_rr_interval)(struct rq *rq,
2473 struct task_struct *task);
2474
2475 void (*update_curr)(struct rq *rq);
2476
2477 #ifdef CONFIG_FAIR_GROUP_SCHED
2478 void (*task_change_group)(struct task_struct *p);
2479 #endif
2480
2481 #ifdef CONFIG_SCHED_CORE
2482 int (*task_is_throttled)(struct task_struct *p, int cpu);
2483 #endif
2484 };
2485
put_prev_task(struct rq * rq,struct task_struct * prev)2486 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2487 {
2488 WARN_ON_ONCE(rq->donor != prev);
2489 prev->sched_class->put_prev_task(rq, prev, NULL);
2490 }
2491
set_next_task(struct rq * rq,struct task_struct * next)2492 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2493 {
2494 next->sched_class->set_next_task(rq, next, false);
2495 }
2496
2497 static inline void
__put_prev_set_next_dl_server(struct rq * rq,struct task_struct * prev,struct task_struct * next)2498 __put_prev_set_next_dl_server(struct rq *rq,
2499 struct task_struct *prev,
2500 struct task_struct *next)
2501 {
2502 prev->dl_server = NULL;
2503 next->dl_server = rq->dl_server;
2504 rq->dl_server = NULL;
2505 }
2506
put_prev_set_next_task(struct rq * rq,struct task_struct * prev,struct task_struct * next)2507 static inline void put_prev_set_next_task(struct rq *rq,
2508 struct task_struct *prev,
2509 struct task_struct *next)
2510 {
2511 WARN_ON_ONCE(rq->curr != prev);
2512
2513 __put_prev_set_next_dl_server(rq, prev, next);
2514
2515 if (next == prev)
2516 return;
2517
2518 prev->sched_class->put_prev_task(rq, prev, next);
2519 next->sched_class->set_next_task(rq, next, true);
2520 }
2521
2522 /*
2523 * Helper to define a sched_class instance; each one is placed in a separate
2524 * section which is ordered by the linker script:
2525 *
2526 * include/asm-generic/vmlinux.lds.h
2527 *
2528 * *CAREFUL* they are laid out in *REVERSE* order!!!
2529 *
2530 * Also enforce alignment on the instance, not the type, to guarantee layout.
2531 */
2532 #define DEFINE_SCHED_CLASS(name) \
2533 const struct sched_class name##_sched_class \
2534 __aligned(__alignof__(struct sched_class)) \
2535 __section("__" #name "_sched_class")
2536
2537 /* Defined in include/asm-generic/vmlinux.lds.h */
2538 extern struct sched_class __sched_class_highest[];
2539 extern struct sched_class __sched_class_lowest[];
2540
2541 extern const struct sched_class stop_sched_class;
2542 extern const struct sched_class dl_sched_class;
2543 extern const struct sched_class rt_sched_class;
2544 extern const struct sched_class fair_sched_class;
2545 extern const struct sched_class idle_sched_class;
2546
2547 /*
2548 * Iterate only active classes. SCX can take over all fair tasks or be
2549 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2550 */
next_active_class(const struct sched_class * class)2551 static inline const struct sched_class *next_active_class(const struct sched_class *class)
2552 {
2553 class++;
2554 #ifdef CONFIG_SCHED_CLASS_EXT
2555 if (scx_switched_all() && class == &fair_sched_class)
2556 class++;
2557 if (!scx_enabled() && class == &ext_sched_class)
2558 class++;
2559 #endif
2560 return class;
2561 }
2562
2563 #define for_class_range(class, _from, _to) \
2564 for (class = (_from); class < (_to); class++)
2565
2566 #define for_each_class(class) \
2567 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2568
2569 #define for_active_class_range(class, _from, _to) \
2570 for (class = (_from); class != (_to); class = next_active_class(class))
2571
2572 #define for_each_active_class(class) \
2573 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2574
2575 #define sched_class_above(_a, _b) ((_a) < (_b))
2576
sched_stop_runnable(struct rq * rq)2577 static inline bool sched_stop_runnable(struct rq *rq)
2578 {
2579 return rq->stop && task_on_rq_queued(rq->stop);
2580 }
2581
sched_dl_runnable(struct rq * rq)2582 static inline bool sched_dl_runnable(struct rq *rq)
2583 {
2584 return rq->dl.dl_nr_running > 0;
2585 }
2586
sched_rt_runnable(struct rq * rq)2587 static inline bool sched_rt_runnable(struct rq *rq)
2588 {
2589 return rq->rt.rt_queued > 0;
2590 }
2591
sched_fair_runnable(struct rq * rq)2592 static inline bool sched_fair_runnable(struct rq *rq)
2593 {
2594 return rq->cfs.nr_queued > 0;
2595 }
2596
2597 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2598 extern struct task_struct *pick_task_idle(struct rq *rq);
2599
2600 #define SCA_CHECK 0x01
2601 #define SCA_MIGRATE_DISABLE 0x02
2602 #define SCA_MIGRATE_ENABLE 0x04
2603 #define SCA_USER 0x08
2604
2605 #ifdef CONFIG_SMP
2606
2607 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2608
2609 extern void sched_balance_trigger(struct rq *rq);
2610
2611 extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2612 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2613
task_allowed_on_cpu(struct task_struct * p,int cpu)2614 static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2615 {
2616 /* When not in the task's cpumask, no point in looking further. */
2617 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2618 return false;
2619
2620 /* Can @cpu run a user thread? */
2621 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2622 return false;
2623
2624 return true;
2625 }
2626
alloc_user_cpus_ptr(int node)2627 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2628 {
2629 /*
2630 * See do_set_cpus_allowed() above for the rcu_head usage.
2631 */
2632 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2633
2634 return kmalloc_node(size, GFP_KERNEL, node);
2635 }
2636
get_push_task(struct rq * rq)2637 static inline struct task_struct *get_push_task(struct rq *rq)
2638 {
2639 struct task_struct *p = rq->donor;
2640
2641 lockdep_assert_rq_held(rq);
2642
2643 if (rq->push_busy)
2644 return NULL;
2645
2646 if (p->nr_cpus_allowed == 1)
2647 return NULL;
2648
2649 if (p->migration_disabled)
2650 return NULL;
2651
2652 rq->push_busy = true;
2653 return get_task_struct(p);
2654 }
2655
2656 extern int push_cpu_stop(void *arg);
2657
2658 #else /* !CONFIG_SMP: */
2659
task_allowed_on_cpu(struct task_struct * p,int cpu)2660 static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2661 {
2662 return true;
2663 }
2664
__set_cpus_allowed_ptr(struct task_struct * p,struct affinity_context * ctx)2665 static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2666 struct affinity_context *ctx)
2667 {
2668 return set_cpus_allowed_ptr(p, ctx->new_mask);
2669 }
2670
alloc_user_cpus_ptr(int node)2671 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2672 {
2673 return NULL;
2674 }
2675
2676 #endif /* !CONFIG_SMP */
2677
2678 #ifdef CONFIG_CPU_IDLE
2679
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2680 static inline void idle_set_state(struct rq *rq,
2681 struct cpuidle_state *idle_state)
2682 {
2683 rq->idle_state = idle_state;
2684 }
2685
idle_get_state(struct rq * rq)2686 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2687 {
2688 SCHED_WARN_ON(!rcu_read_lock_held());
2689
2690 return rq->idle_state;
2691 }
2692
2693 #else /* !CONFIG_CPU_IDLE: */
2694
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2695 static inline void idle_set_state(struct rq *rq,
2696 struct cpuidle_state *idle_state)
2697 {
2698 }
2699
idle_get_state(struct rq * rq)2700 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2701 {
2702 return NULL;
2703 }
2704
2705 #endif /* !CONFIG_CPU_IDLE */
2706
2707 extern void schedule_idle(void);
2708 asmlinkage void schedule_user(void);
2709
2710 extern void sysrq_sched_debug_show(void);
2711 extern void sched_init_granularity(void);
2712 extern void update_max_interval(void);
2713
2714 extern void init_sched_dl_class(void);
2715 extern void init_sched_rt_class(void);
2716 extern void init_sched_fair_class(void);
2717
2718 extern void resched_curr(struct rq *rq);
2719 extern void resched_curr_lazy(struct rq *rq);
2720 extern void resched_cpu(int cpu);
2721
2722 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2723 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2724
2725 extern void init_dl_entity(struct sched_dl_entity *dl_se);
2726
2727 #define BW_SHIFT 20
2728 #define BW_UNIT (1 << BW_SHIFT)
2729 #define RATIO_SHIFT 8
2730 #define MAX_BW_BITS (64 - BW_SHIFT)
2731 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2732
2733 extern unsigned long to_ratio(u64 period, u64 runtime);
2734
2735 extern void init_entity_runnable_average(struct sched_entity *se);
2736 extern void post_init_entity_util_avg(struct task_struct *p);
2737
2738 #ifdef CONFIG_NO_HZ_FULL
2739 extern bool sched_can_stop_tick(struct rq *rq);
2740 extern int __init sched_tick_offload_init(void);
2741
2742 /*
2743 * Tick may be needed by tasks in the runqueue depending on their policy and
2744 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2745 * nohz mode if necessary.
2746 */
sched_update_tick_dependency(struct rq * rq)2747 static inline void sched_update_tick_dependency(struct rq *rq)
2748 {
2749 int cpu = cpu_of(rq);
2750
2751 if (!tick_nohz_full_cpu(cpu))
2752 return;
2753
2754 if (sched_can_stop_tick(rq))
2755 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2756 else
2757 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2758 }
2759 #else /* !CONFIG_NO_HZ_FULL: */
sched_tick_offload_init(void)2760 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2761 static inline void sched_update_tick_dependency(struct rq *rq) { }
2762 #endif /* !CONFIG_NO_HZ_FULL */
2763
add_nr_running(struct rq * rq,unsigned count)2764 static inline void add_nr_running(struct rq *rq, unsigned count)
2765 {
2766 unsigned prev_nr = rq->nr_running;
2767
2768 rq->nr_running = prev_nr + count;
2769 if (trace_sched_update_nr_running_tp_enabled()) {
2770 call_trace_sched_update_nr_running(rq, count);
2771 }
2772
2773 #ifdef CONFIG_SMP
2774 if (prev_nr < 2 && rq->nr_running >= 2)
2775 set_rd_overloaded(rq->rd, 1);
2776 #endif
2777
2778 sched_update_tick_dependency(rq);
2779 }
2780
sub_nr_running(struct rq * rq,unsigned count)2781 static inline void sub_nr_running(struct rq *rq, unsigned count)
2782 {
2783 rq->nr_running -= count;
2784 if (trace_sched_update_nr_running_tp_enabled()) {
2785 call_trace_sched_update_nr_running(rq, -count);
2786 }
2787
2788 /* Check if we still need preemption */
2789 sched_update_tick_dependency(rq);
2790 }
2791
__block_task(struct rq * rq,struct task_struct * p)2792 static inline void __block_task(struct rq *rq, struct task_struct *p)
2793 {
2794 if (p->sched_contributes_to_load)
2795 rq->nr_uninterruptible++;
2796
2797 if (p->in_iowait) {
2798 atomic_inc(&rq->nr_iowait);
2799 delayacct_blkio_start();
2800 }
2801
2802 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2803
2804 /*
2805 * The moment this write goes through, ttwu() can swoop in and migrate
2806 * this task, rendering our rq->__lock ineffective.
2807 *
2808 * __schedule() try_to_wake_up()
2809 * LOCK rq->__lock LOCK p->pi_lock
2810 * pick_next_task()
2811 * pick_next_task_fair()
2812 * pick_next_entity()
2813 * dequeue_entities()
2814 * __block_task()
2815 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2816 * break;
2817 *
2818 * ACQUIRE (after ctrl-dep)
2819 *
2820 * cpu = select_task_rq();
2821 * set_task_cpu(p, cpu);
2822 * ttwu_queue()
2823 * ttwu_do_activate()
2824 * LOCK rq->__lock
2825 * activate_task()
2826 * STORE p->on_rq = 1
2827 * UNLOCK rq->__lock
2828 *
2829 * Callers must ensure to not reference @p after this -- we no longer
2830 * own it.
2831 */
2832 smp_store_release(&p->on_rq, 0);
2833 }
2834
2835 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2836 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2837
2838 extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2839
2840 #ifdef CONFIG_PREEMPT_RT
2841 # define SCHED_NR_MIGRATE_BREAK 8
2842 #else
2843 # define SCHED_NR_MIGRATE_BREAK 32
2844 #endif
2845
2846 extern const_debug unsigned int sysctl_sched_nr_migrate;
2847 extern const_debug unsigned int sysctl_sched_migration_cost;
2848
2849 extern unsigned int sysctl_sched_base_slice;
2850
2851 #ifdef CONFIG_SCHED_DEBUG
2852 extern int sysctl_resched_latency_warn_ms;
2853 extern int sysctl_resched_latency_warn_once;
2854
2855 extern unsigned int sysctl_sched_tunable_scaling;
2856
2857 extern unsigned int sysctl_numa_balancing_scan_delay;
2858 extern unsigned int sysctl_numa_balancing_scan_period_min;
2859 extern unsigned int sysctl_numa_balancing_scan_period_max;
2860 extern unsigned int sysctl_numa_balancing_scan_size;
2861 extern unsigned int sysctl_numa_balancing_hot_threshold;
2862 #endif
2863
2864 #ifdef CONFIG_SCHED_HRTICK
2865
2866 /*
2867 * Use hrtick when:
2868 * - enabled by features
2869 * - hrtimer is actually high res
2870 */
hrtick_enabled(struct rq * rq)2871 static inline int hrtick_enabled(struct rq *rq)
2872 {
2873 if (!cpu_active(cpu_of(rq)))
2874 return 0;
2875 return hrtimer_is_hres_active(&rq->hrtick_timer);
2876 }
2877
hrtick_enabled_fair(struct rq * rq)2878 static inline int hrtick_enabled_fair(struct rq *rq)
2879 {
2880 if (!sched_feat(HRTICK))
2881 return 0;
2882 return hrtick_enabled(rq);
2883 }
2884
hrtick_enabled_dl(struct rq * rq)2885 static inline int hrtick_enabled_dl(struct rq *rq)
2886 {
2887 if (!sched_feat(HRTICK_DL))
2888 return 0;
2889 return hrtick_enabled(rq);
2890 }
2891
2892 extern void hrtick_start(struct rq *rq, u64 delay);
2893
2894 #else /* !CONFIG_SCHED_HRTICK: */
2895
hrtick_enabled_fair(struct rq * rq)2896 static inline int hrtick_enabled_fair(struct rq *rq)
2897 {
2898 return 0;
2899 }
2900
hrtick_enabled_dl(struct rq * rq)2901 static inline int hrtick_enabled_dl(struct rq *rq)
2902 {
2903 return 0;
2904 }
2905
hrtick_enabled(struct rq * rq)2906 static inline int hrtick_enabled(struct rq *rq)
2907 {
2908 return 0;
2909 }
2910
2911 #endif /* !CONFIG_SCHED_HRTICK */
2912
2913 #ifndef arch_scale_freq_tick
arch_scale_freq_tick(void)2914 static __always_inline void arch_scale_freq_tick(void) { }
2915 #endif
2916
2917 #ifndef arch_scale_freq_capacity
2918 /**
2919 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2920 * @cpu: the CPU in question.
2921 *
2922 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2923 *
2924 * f_curr
2925 * ------ * SCHED_CAPACITY_SCALE
2926 * f_max
2927 */
2928 static __always_inline
arch_scale_freq_capacity(int cpu)2929 unsigned long arch_scale_freq_capacity(int cpu)
2930 {
2931 return SCHED_CAPACITY_SCALE;
2932 }
2933 #endif
2934
2935 #ifdef CONFIG_SCHED_DEBUG
2936 /*
2937 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2938 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2939 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2940 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2941 */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2942 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2943 {
2944 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2945 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2946 #ifdef CONFIG_SMP
2947 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2948 #endif
2949 }
2950 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2951 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) { }
2952 #endif
2953
2954 #define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2955 __DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2956 static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2957 { class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2958 _lock; return _t; }
2959
2960 #ifdef CONFIG_SMP
2961
rq_order_less(struct rq * rq1,struct rq * rq2)2962 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2963 {
2964 #ifdef CONFIG_SCHED_CORE
2965 /*
2966 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2967 * order by core-id first and cpu-id second.
2968 *
2969 * Notably:
2970 *
2971 * double_rq_lock(0,3); will take core-0, core-1 lock
2972 * double_rq_lock(1,2); will take core-1, core-0 lock
2973 *
2974 * when only cpu-id is considered.
2975 */
2976 if (rq1->core->cpu < rq2->core->cpu)
2977 return true;
2978 if (rq1->core->cpu > rq2->core->cpu)
2979 return false;
2980
2981 /*
2982 * __sched_core_flip() relies on SMT having cpu-id lock order.
2983 */
2984 #endif
2985 return rq1->cpu < rq2->cpu;
2986 }
2987
2988 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2989
2990 #ifdef CONFIG_PREEMPTION
2991
2992 /*
2993 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2994 * way at the expense of forcing extra atomic operations in all
2995 * invocations. This assures that the double_lock is acquired using the
2996 * same underlying policy as the spinlock_t on this architecture, which
2997 * reduces latency compared to the unfair variant below. However, it
2998 * also adds more overhead and therefore may reduce throughput.
2999 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)3000 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3001 __releases(this_rq->lock)
3002 __acquires(busiest->lock)
3003 __acquires(this_rq->lock)
3004 {
3005 raw_spin_rq_unlock(this_rq);
3006 double_rq_lock(this_rq, busiest);
3007
3008 return 1;
3009 }
3010
3011 #else /* !CONFIG_PREEMPTION: */
3012 /*
3013 * Unfair double_lock_balance: Optimizes throughput at the expense of
3014 * latency by eliminating extra atomic operations when the locks are
3015 * already in proper order on entry. This favors lower CPU-ids and will
3016 * grant the double lock to lower CPUs over higher ids under contention,
3017 * regardless of entry order into the function.
3018 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)3019 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
3020 __releases(this_rq->lock)
3021 __acquires(busiest->lock)
3022 __acquires(this_rq->lock)
3023 {
3024 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
3025 likely(raw_spin_rq_trylock(busiest))) {
3026 double_rq_clock_clear_update(this_rq, busiest);
3027 return 0;
3028 }
3029
3030 if (rq_order_less(this_rq, busiest)) {
3031 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3032 double_rq_clock_clear_update(this_rq, busiest);
3033 return 0;
3034 }
3035
3036 raw_spin_rq_unlock(this_rq);
3037 double_rq_lock(this_rq, busiest);
3038
3039 return 1;
3040 }
3041
3042 #endif /* !CONFIG_PREEMPTION */
3043
3044 /*
3045 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3046 */
double_lock_balance(struct rq * this_rq,struct rq * busiest)3047 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3048 {
3049 lockdep_assert_irqs_disabled();
3050
3051 return _double_lock_balance(this_rq, busiest);
3052 }
3053
double_unlock_balance(struct rq * this_rq,struct rq * busiest)3054 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3055 __releases(busiest->lock)
3056 {
3057 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
3058 raw_spin_rq_unlock(busiest);
3059 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3060 }
3061
double_lock(spinlock_t * l1,spinlock_t * l2)3062 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3063 {
3064 if (l1 > l2)
3065 swap(l1, l2);
3066
3067 spin_lock(l1);
3068 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3069 }
3070
double_lock_irq(spinlock_t * l1,spinlock_t * l2)3071 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3072 {
3073 if (l1 > l2)
3074 swap(l1, l2);
3075
3076 spin_lock_irq(l1);
3077 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3078 }
3079
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)3080 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3081 {
3082 if (l1 > l2)
3083 swap(l1, l2);
3084
3085 raw_spin_lock(l1);
3086 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3087 }
3088
double_raw_unlock(raw_spinlock_t * l1,raw_spinlock_t * l2)3089 static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3090 {
3091 raw_spin_unlock(l1);
3092 raw_spin_unlock(l2);
3093 }
3094
3095 DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3096 double_raw_lock(_T->lock, _T->lock2),
3097 double_raw_unlock(_T->lock, _T->lock2))
3098
3099 /*
3100 * double_rq_unlock - safely unlock two runqueues
3101 *
3102 * Note this does not restore interrupts like task_rq_unlock,
3103 * you need to do so manually after calling.
3104 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)3105 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3106 __releases(rq1->lock)
3107 __releases(rq2->lock)
3108 {
3109 if (__rq_lockp(rq1) != __rq_lockp(rq2))
3110 raw_spin_rq_unlock(rq2);
3111 else
3112 __release(rq2->lock);
3113 raw_spin_rq_unlock(rq1);
3114 }
3115
3116 extern void set_rq_online (struct rq *rq);
3117 extern void set_rq_offline(struct rq *rq);
3118
3119 extern bool sched_smp_initialized;
3120
3121 #else /* !CONFIG_SMP: */
3122
3123 /*
3124 * double_rq_lock - safely lock two runqueues
3125 *
3126 * Note this does not disable interrupts like task_rq_lock,
3127 * you need to do so manually before calling.
3128 */
double_rq_lock(struct rq * rq1,struct rq * rq2)3129 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
3130 __acquires(rq1->lock)
3131 __acquires(rq2->lock)
3132 {
3133 WARN_ON_ONCE(!irqs_disabled());
3134 WARN_ON_ONCE(rq1 != rq2);
3135 raw_spin_rq_lock(rq1);
3136 __acquire(rq2->lock); /* Fake it out ;) */
3137 double_rq_clock_clear_update(rq1, rq2);
3138 }
3139
3140 /*
3141 * double_rq_unlock - safely unlock two runqueues
3142 *
3143 * Note this does not restore interrupts like task_rq_unlock,
3144 * you need to do so manually after calling.
3145 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)3146 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3147 __releases(rq1->lock)
3148 __releases(rq2->lock)
3149 {
3150 WARN_ON_ONCE(rq1 != rq2);
3151 raw_spin_rq_unlock(rq1);
3152 __release(rq2->lock);
3153 }
3154
3155 #endif /* !CONFIG_SMP */
3156
3157 DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3158 double_rq_lock(_T->lock, _T->lock2),
3159 double_rq_unlock(_T->lock, _T->lock2))
3160
3161 extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3162 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3163 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3164
3165 #ifdef CONFIG_SCHED_DEBUG
3166 extern bool sched_debug_verbose;
3167
3168 extern void print_cfs_stats(struct seq_file *m, int cpu);
3169 extern void print_rt_stats(struct seq_file *m, int cpu);
3170 extern void print_dl_stats(struct seq_file *m, int cpu);
3171 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3172 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3173 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3174
3175 extern void resched_latency_warn(int cpu, u64 latency);
3176 # ifdef CONFIG_NUMA_BALANCING
3177 extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3178 extern void
3179 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3180 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3181 # endif /* CONFIG_NUMA_BALANCING */
3182 #else /* !CONFIG_SCHED_DEBUG: */
resched_latency_warn(int cpu,u64 latency)3183 static inline void resched_latency_warn(int cpu, u64 latency) { }
3184 #endif /* !CONFIG_SCHED_DEBUG */
3185
3186 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3187 extern void init_rt_rq(struct rt_rq *rt_rq);
3188 extern void init_dl_rq(struct dl_rq *dl_rq);
3189
3190 extern void cfs_bandwidth_usage_inc(void);
3191 extern void cfs_bandwidth_usage_dec(void);
3192
3193 #ifdef CONFIG_NO_HZ_COMMON
3194
3195 #define NOHZ_BALANCE_KICK_BIT 0
3196 #define NOHZ_STATS_KICK_BIT 1
3197 #define NOHZ_NEWILB_KICK_BIT 2
3198 #define NOHZ_NEXT_KICK_BIT 3
3199
3200 /* Run sched_balance_domains() */
3201 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3202 /* Update blocked load */
3203 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3204 /* Update blocked load when entering idle */
3205 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3206 /* Update nohz.next_balance */
3207 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3208
3209 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3210
3211 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3212
3213 extern void nohz_balance_exit_idle(struct rq *rq);
3214 #else /* !CONFIG_NO_HZ_COMMON: */
nohz_balance_exit_idle(struct rq * rq)3215 static inline void nohz_balance_exit_idle(struct rq *rq) { }
3216 #endif /* !CONFIG_NO_HZ_COMMON */
3217
3218 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
3219 extern void nohz_run_idle_balance(int cpu);
3220 #else
nohz_run_idle_balance(int cpu)3221 static inline void nohz_run_idle_balance(int cpu) { }
3222 #endif
3223
3224 #include "stats.h"
3225
3226 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
3227
3228 extern void __sched_core_account_forceidle(struct rq *rq);
3229
sched_core_account_forceidle(struct rq * rq)3230 static inline void sched_core_account_forceidle(struct rq *rq)
3231 {
3232 if (schedstat_enabled())
3233 __sched_core_account_forceidle(rq);
3234 }
3235
3236 extern void __sched_core_tick(struct rq *rq);
3237
sched_core_tick(struct rq * rq)3238 static inline void sched_core_tick(struct rq *rq)
3239 {
3240 if (sched_core_enabled(rq) && schedstat_enabled())
3241 __sched_core_tick(rq);
3242 }
3243
3244 #else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
3245
sched_core_account_forceidle(struct rq * rq)3246 static inline void sched_core_account_forceidle(struct rq *rq) { }
3247
sched_core_tick(struct rq * rq)3248 static inline void sched_core_tick(struct rq *rq) { }
3249
3250 #endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
3251
3252 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
3253
3254 struct irqtime {
3255 u64 total;
3256 u64 tick_delta;
3257 u64 irq_start_time;
3258 struct u64_stats_sync sync;
3259 };
3260
3261 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3262 DECLARE_STATIC_KEY_FALSE(sched_clock_irqtime);
3263
irqtime_enabled(void)3264 static inline int irqtime_enabled(void)
3265 {
3266 return static_branch_likely(&sched_clock_irqtime);
3267 }
3268
3269 /*
3270 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3271 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3272 * and never move forward.
3273 */
irq_time_read(int cpu)3274 static inline u64 irq_time_read(int cpu)
3275 {
3276 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3277 unsigned int seq;
3278 u64 total;
3279
3280 do {
3281 seq = __u64_stats_fetch_begin(&irqtime->sync);
3282 total = irqtime->total;
3283 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3284
3285 return total;
3286 }
3287
3288 #else
3289
irqtime_enabled(void)3290 static inline int irqtime_enabled(void)
3291 {
3292 return 0;
3293 }
3294
3295 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3296
3297 #ifdef CONFIG_CPU_FREQ
3298
3299 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3300
3301 /**
3302 * cpufreq_update_util - Take a note about CPU utilization changes.
3303 * @rq: Runqueue to carry out the update for.
3304 * @flags: Update reason flags.
3305 *
3306 * This function is called by the scheduler on the CPU whose utilization is
3307 * being updated.
3308 *
3309 * It can only be called from RCU-sched read-side critical sections.
3310 *
3311 * The way cpufreq is currently arranged requires it to evaluate the CPU
3312 * performance state (frequency/voltage) on a regular basis to prevent it from
3313 * being stuck in a completely inadequate performance level for too long.
3314 * That is not guaranteed to happen if the updates are only triggered from CFS
3315 * and DL, though, because they may not be coming in if only RT tasks are
3316 * active all the time (or there are RT tasks only).
3317 *
3318 * As a workaround for that issue, this function is called periodically by the
3319 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3320 * but that really is a band-aid. Going forward it should be replaced with
3321 * solutions targeted more specifically at RT tasks.
3322 */
cpufreq_update_util(struct rq * rq,unsigned int flags)3323 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3324 {
3325 struct update_util_data *data;
3326
3327 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3328 cpu_of(rq)));
3329 if (data)
3330 data->func(data, rq_clock(rq), flags);
3331 }
3332 #else /* !CONFIG_CPU_FREQ: */
cpufreq_update_util(struct rq * rq,unsigned int flags)3333 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3334 #endif /* !CONFIG_CPU_FREQ */
3335
3336 #ifdef arch_scale_freq_capacity
3337 # ifndef arch_scale_freq_invariant
3338 # define arch_scale_freq_invariant() true
3339 # endif
3340 #else
3341 # define arch_scale_freq_invariant() false
3342 #endif
3343
3344 #ifdef CONFIG_SMP
3345
3346 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3347 unsigned long *min,
3348 unsigned long *max);
3349
3350 unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3351 unsigned long min,
3352 unsigned long max);
3353
3354
3355 /*
3356 * Verify the fitness of task @p to run on @cpu taking into account the
3357 * CPU original capacity and the runtime/deadline ratio of the task.
3358 *
3359 * The function will return true if the original capacity of @cpu is
3360 * greater than or equal to task's deadline density right shifted by
3361 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3362 */
dl_task_fits_capacity(struct task_struct * p,int cpu)3363 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3364 {
3365 unsigned long cap = arch_scale_cpu_capacity(cpu);
3366
3367 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3368 }
3369
cpu_bw_dl(struct rq * rq)3370 static inline unsigned long cpu_bw_dl(struct rq *rq)
3371 {
3372 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3373 }
3374
cpu_util_dl(struct rq * rq)3375 static inline unsigned long cpu_util_dl(struct rq *rq)
3376 {
3377 return READ_ONCE(rq->avg_dl.util_avg);
3378 }
3379
3380
3381 extern unsigned long cpu_util_cfs(int cpu);
3382 extern unsigned long cpu_util_cfs_boost(int cpu);
3383
cpu_util_rt(struct rq * rq)3384 static inline unsigned long cpu_util_rt(struct rq *rq)
3385 {
3386 return READ_ONCE(rq->avg_rt.util_avg);
3387 }
3388
3389 #else /* !CONFIG_SMP */
update_other_load_avgs(struct rq * rq)3390 static inline bool update_other_load_avgs(struct rq *rq) { return false; }
3391 #endif /* CONFIG_SMP */
3392
3393 #ifdef CONFIG_UCLAMP_TASK
3394
3395 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3396
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3397 static inline unsigned long uclamp_rq_get(struct rq *rq,
3398 enum uclamp_id clamp_id)
3399 {
3400 return READ_ONCE(rq->uclamp[clamp_id].value);
3401 }
3402
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3403 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3404 unsigned int value)
3405 {
3406 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3407 }
3408
uclamp_rq_is_idle(struct rq * rq)3409 static inline bool uclamp_rq_is_idle(struct rq *rq)
3410 {
3411 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3412 }
3413
3414 /* Is the rq being capped/throttled by uclamp_max? */
uclamp_rq_is_capped(struct rq * rq)3415 static inline bool uclamp_rq_is_capped(struct rq *rq)
3416 {
3417 unsigned long rq_util;
3418 unsigned long max_util;
3419
3420 if (!static_branch_likely(&sched_uclamp_used))
3421 return false;
3422
3423 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3424 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3425
3426 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3427 }
3428
3429 /*
3430 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3431 * by default in the fast path and only gets turned on once userspace performs
3432 * an operation that requires it.
3433 *
3434 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3435 * hence is active.
3436 */
uclamp_is_used(void)3437 static inline bool uclamp_is_used(void)
3438 {
3439 return static_branch_likely(&sched_uclamp_used);
3440 }
3441
3442 #define for_each_clamp_id(clamp_id) \
3443 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3444
3445 extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3446
3447
uclamp_none(enum uclamp_id clamp_id)3448 static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3449 {
3450 if (clamp_id == UCLAMP_MIN)
3451 return 0;
3452 return SCHED_CAPACITY_SCALE;
3453 }
3454
3455 /* Integer rounded range for each bucket */
3456 #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3457
uclamp_bucket_id(unsigned int clamp_value)3458 static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3459 {
3460 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3461 }
3462
3463 static inline void
uclamp_se_set(struct uclamp_se * uc_se,unsigned int value,bool user_defined)3464 uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3465 {
3466 uc_se->value = value;
3467 uc_se->bucket_id = uclamp_bucket_id(value);
3468 uc_se->user_defined = user_defined;
3469 }
3470
3471 #else /* !CONFIG_UCLAMP_TASK: */
3472
3473 static inline unsigned long
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)3474 uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3475 {
3476 if (clamp_id == UCLAMP_MIN)
3477 return 0;
3478
3479 return SCHED_CAPACITY_SCALE;
3480 }
3481
uclamp_rq_is_capped(struct rq * rq)3482 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3483
uclamp_is_used(void)3484 static inline bool uclamp_is_used(void)
3485 {
3486 return false;
3487 }
3488
3489 static inline unsigned long
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3490 uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3491 {
3492 if (clamp_id == UCLAMP_MIN)
3493 return 0;
3494
3495 return SCHED_CAPACITY_SCALE;
3496 }
3497
3498 static inline void
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3499 uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3500 {
3501 }
3502
uclamp_rq_is_idle(struct rq * rq)3503 static inline bool uclamp_rq_is_idle(struct rq *rq)
3504 {
3505 return false;
3506 }
3507
3508 #endif /* !CONFIG_UCLAMP_TASK */
3509
3510 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3511
cpu_util_irq(struct rq * rq)3512 static inline unsigned long cpu_util_irq(struct rq *rq)
3513 {
3514 return READ_ONCE(rq->avg_irq.util_avg);
3515 }
3516
3517 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3518 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3519 {
3520 util *= (max - irq);
3521 util /= max;
3522
3523 return util;
3524
3525 }
3526
3527 #else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3528
cpu_util_irq(struct rq * rq)3529 static inline unsigned long cpu_util_irq(struct rq *rq)
3530 {
3531 return 0;
3532 }
3533
3534 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3535 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3536 {
3537 return util;
3538 }
3539
3540 #endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3541
3542 extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr);
3543
3544 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3545
3546 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3547
3548 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3549
sched_energy_enabled(void)3550 static inline bool sched_energy_enabled(void)
3551 {
3552 return static_branch_unlikely(&sched_energy_present);
3553 }
3554
3555 extern struct cpufreq_governor schedutil_gov;
3556
3557 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3558
3559 #define perf_domain_span(pd) NULL
3560
sched_energy_enabled(void)3561 static inline bool sched_energy_enabled(void) { return false; }
3562
3563 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3564
3565 #ifdef CONFIG_MEMBARRIER
3566
3567 /*
3568 * The scheduler provides memory barriers required by membarrier between:
3569 * - prior user-space memory accesses and store to rq->membarrier_state,
3570 * - store to rq->membarrier_state and following user-space memory accesses.
3571 * In the same way it provides those guarantees around store to rq->curr.
3572 */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3573 static inline void membarrier_switch_mm(struct rq *rq,
3574 struct mm_struct *prev_mm,
3575 struct mm_struct *next_mm)
3576 {
3577 int membarrier_state;
3578
3579 if (prev_mm == next_mm)
3580 return;
3581
3582 membarrier_state = atomic_read(&next_mm->membarrier_state);
3583 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3584 return;
3585
3586 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3587 }
3588
3589 #else /* !CONFIG_MEMBARRIER :*/
3590
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3591 static inline void membarrier_switch_mm(struct rq *rq,
3592 struct mm_struct *prev_mm,
3593 struct mm_struct *next_mm)
3594 {
3595 }
3596
3597 #endif /* !CONFIG_MEMBARRIER */
3598
3599 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3600 static inline bool is_per_cpu_kthread(struct task_struct *p)
3601 {
3602 if (!(p->flags & PF_KTHREAD))
3603 return false;
3604
3605 if (p->nr_cpus_allowed != 1)
3606 return false;
3607
3608 return true;
3609 }
3610 #endif
3611
3612 extern void swake_up_all_locked(struct swait_queue_head *q);
3613 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3614
3615 extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3616
3617 #ifdef CONFIG_PREEMPT_DYNAMIC
3618 extern int preempt_dynamic_mode;
3619 extern int sched_dynamic_mode(const char *str);
3620 extern void sched_dynamic_update(int mode);
3621 #endif
3622
3623 #ifdef CONFIG_SCHED_MM_CID
3624
3625 #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3626 #define MM_CID_SCAN_DELAY 100 /* 100ms */
3627
3628 extern raw_spinlock_t cid_lock;
3629 extern int use_cid_lock;
3630
3631 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3632 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3633 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3634 extern void init_sched_mm_cid(struct task_struct *t);
3635
__mm_cid_put(struct mm_struct * mm,int cid)3636 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3637 {
3638 if (cid < 0)
3639 return;
3640 cpumask_clear_cpu(cid, mm_cidmask(mm));
3641 }
3642
3643 /*
3644 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3645 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3646 * be held to transition to other states.
3647 *
3648 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3649 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3650 */
mm_cid_put_lazy(struct task_struct * t)3651 static inline void mm_cid_put_lazy(struct task_struct *t)
3652 {
3653 struct mm_struct *mm = t->mm;
3654 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3655 int cid;
3656
3657 lockdep_assert_irqs_disabled();
3658 cid = __this_cpu_read(pcpu_cid->cid);
3659 if (!mm_cid_is_lazy_put(cid) ||
3660 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3661 return;
3662 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3663 }
3664
mm_cid_pcpu_unset(struct mm_struct * mm)3665 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3666 {
3667 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3668 int cid, res;
3669
3670 lockdep_assert_irqs_disabled();
3671 cid = __this_cpu_read(pcpu_cid->cid);
3672 for (;;) {
3673 if (mm_cid_is_unset(cid))
3674 return MM_CID_UNSET;
3675 /*
3676 * Attempt transition from valid or lazy-put to unset.
3677 */
3678 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3679 if (res == cid)
3680 break;
3681 cid = res;
3682 }
3683 return cid;
3684 }
3685
mm_cid_put(struct mm_struct * mm)3686 static inline void mm_cid_put(struct mm_struct *mm)
3687 {
3688 int cid;
3689
3690 lockdep_assert_irqs_disabled();
3691 cid = mm_cid_pcpu_unset(mm);
3692 if (cid == MM_CID_UNSET)
3693 return;
3694 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3695 }
3696
__mm_cid_try_get(struct task_struct * t,struct mm_struct * mm)3697 static inline int __mm_cid_try_get(struct task_struct *t, struct mm_struct *mm)
3698 {
3699 struct cpumask *cidmask = mm_cidmask(mm);
3700 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3701 int cid, max_nr_cid, allowed_max_nr_cid;
3702
3703 /*
3704 * After shrinking the number of threads or reducing the number
3705 * of allowed cpus, reduce the value of max_nr_cid so expansion
3706 * of cid allocation will preserve cache locality if the number
3707 * of threads or allowed cpus increase again.
3708 */
3709 max_nr_cid = atomic_read(&mm->max_nr_cid);
3710 while ((allowed_max_nr_cid = min_t(int, READ_ONCE(mm->nr_cpus_allowed),
3711 atomic_read(&mm->mm_users))),
3712 max_nr_cid > allowed_max_nr_cid) {
3713 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into max_nr_cid. */
3714 if (atomic_try_cmpxchg(&mm->max_nr_cid, &max_nr_cid, allowed_max_nr_cid)) {
3715 max_nr_cid = allowed_max_nr_cid;
3716 break;
3717 }
3718 }
3719 /* Try to re-use recent cid. This improves cache locality. */
3720 cid = __this_cpu_read(pcpu_cid->recent_cid);
3721 if (!mm_cid_is_unset(cid) && cid < max_nr_cid &&
3722 !cpumask_test_and_set_cpu(cid, cidmask))
3723 return cid;
3724 /*
3725 * Expand cid allocation if the maximum number of concurrency
3726 * IDs allocated (max_nr_cid) is below the number cpus allowed
3727 * and number of threads. Expanding cid allocation as much as
3728 * possible improves cache locality.
3729 */
3730 cid = max_nr_cid;
3731 while (cid < READ_ONCE(mm->nr_cpus_allowed) && cid < atomic_read(&mm->mm_users)) {
3732 /* atomic_try_cmpxchg loads previous mm->max_nr_cid into cid. */
3733 if (!atomic_try_cmpxchg(&mm->max_nr_cid, &cid, cid + 1))
3734 continue;
3735 if (!cpumask_test_and_set_cpu(cid, cidmask))
3736 return cid;
3737 }
3738 /*
3739 * Find the first available concurrency id.
3740 * Retry finding first zero bit if the mask is temporarily
3741 * filled. This only happens during concurrent remote-clear
3742 * which owns a cid without holding a rq lock.
3743 */
3744 for (;;) {
3745 cid = cpumask_first_zero(cidmask);
3746 if (cid < READ_ONCE(mm->nr_cpus_allowed))
3747 break;
3748 cpu_relax();
3749 }
3750 if (cpumask_test_and_set_cpu(cid, cidmask))
3751 return -1;
3752
3753 return cid;
3754 }
3755
3756 /*
3757 * Save a snapshot of the current runqueue time of this cpu
3758 * with the per-cpu cid value, allowing to estimate how recently it was used.
3759 */
mm_cid_snapshot_time(struct rq * rq,struct mm_struct * mm)3760 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3761 {
3762 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3763
3764 lockdep_assert_rq_held(rq);
3765 WRITE_ONCE(pcpu_cid->time, rq->clock);
3766 }
3767
__mm_cid_get(struct rq * rq,struct task_struct * t,struct mm_struct * mm)3768 static inline int __mm_cid_get(struct rq *rq, struct task_struct *t,
3769 struct mm_struct *mm)
3770 {
3771 int cid;
3772
3773 /*
3774 * All allocations (even those using the cid_lock) are lock-free. If
3775 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3776 * guarantee forward progress.
3777 */
3778 if (!READ_ONCE(use_cid_lock)) {
3779 cid = __mm_cid_try_get(t, mm);
3780 if (cid >= 0)
3781 goto end;
3782 raw_spin_lock(&cid_lock);
3783 } else {
3784 raw_spin_lock(&cid_lock);
3785 cid = __mm_cid_try_get(t, mm);
3786 if (cid >= 0)
3787 goto unlock;
3788 }
3789
3790 /*
3791 * cid concurrently allocated. Retry while forcing following
3792 * allocations to use the cid_lock to ensure forward progress.
3793 */
3794 WRITE_ONCE(use_cid_lock, 1);
3795 /*
3796 * Set use_cid_lock before allocation. Only care about program order
3797 * because this is only required for forward progress.
3798 */
3799 barrier();
3800 /*
3801 * Retry until it succeeds. It is guaranteed to eventually succeed once
3802 * all newcoming allocations observe the use_cid_lock flag set.
3803 */
3804 do {
3805 cid = __mm_cid_try_get(t, mm);
3806 cpu_relax();
3807 } while (cid < 0);
3808 /*
3809 * Allocate before clearing use_cid_lock. Only care about
3810 * program order because this is for forward progress.
3811 */
3812 barrier();
3813 WRITE_ONCE(use_cid_lock, 0);
3814 unlock:
3815 raw_spin_unlock(&cid_lock);
3816 end:
3817 mm_cid_snapshot_time(rq, mm);
3818
3819 return cid;
3820 }
3821
mm_cid_get(struct rq * rq,struct task_struct * t,struct mm_struct * mm)3822 static inline int mm_cid_get(struct rq *rq, struct task_struct *t,
3823 struct mm_struct *mm)
3824 {
3825 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3826 struct cpumask *cpumask;
3827 int cid;
3828
3829 lockdep_assert_rq_held(rq);
3830 cpumask = mm_cidmask(mm);
3831 cid = __this_cpu_read(pcpu_cid->cid);
3832 if (mm_cid_is_valid(cid)) {
3833 mm_cid_snapshot_time(rq, mm);
3834 return cid;
3835 }
3836 if (mm_cid_is_lazy_put(cid)) {
3837 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3838 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3839 }
3840 cid = __mm_cid_get(rq, t, mm);
3841 __this_cpu_write(pcpu_cid->cid, cid);
3842 __this_cpu_write(pcpu_cid->recent_cid, cid);
3843
3844 return cid;
3845 }
3846
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3847 static inline void switch_mm_cid(struct rq *rq,
3848 struct task_struct *prev,
3849 struct task_struct *next)
3850 {
3851 /*
3852 * Provide a memory barrier between rq->curr store and load of
3853 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3854 *
3855 * Should be adapted if context_switch() is modified.
3856 */
3857 if (!next->mm) { // to kernel
3858 /*
3859 * user -> kernel transition does not guarantee a barrier, but
3860 * we can use the fact that it performs an atomic operation in
3861 * mmgrab().
3862 */
3863 if (prev->mm) // from user
3864 smp_mb__after_mmgrab();
3865 /*
3866 * kernel -> kernel transition does not change rq->curr->mm
3867 * state. It stays NULL.
3868 */
3869 } else { // to user
3870 /*
3871 * kernel -> user transition does not provide a barrier
3872 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3873 * Provide it here.
3874 */
3875 if (!prev->mm) { // from kernel
3876 smp_mb();
3877 } else { // from user
3878 /*
3879 * user->user transition relies on an implicit
3880 * memory barrier in switch_mm() when
3881 * current->mm changes. If the architecture
3882 * switch_mm() does not have an implicit memory
3883 * barrier, it is emitted here. If current->mm
3884 * is unchanged, no barrier is needed.
3885 */
3886 smp_mb__after_switch_mm();
3887 }
3888 }
3889 if (prev->mm_cid_active) {
3890 mm_cid_snapshot_time(rq, prev->mm);
3891 mm_cid_put_lazy(prev);
3892 prev->mm_cid = -1;
3893 }
3894 if (next->mm_cid_active)
3895 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next, next->mm);
3896 }
3897
3898 #else /* !CONFIG_SCHED_MM_CID: */
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3899 static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
sched_mm_cid_migrate_from(struct task_struct * t)3900 static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
sched_mm_cid_migrate_to(struct rq * dst_rq,struct task_struct * t)3901 static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
task_tick_mm_cid(struct rq * rq,struct task_struct * curr)3902 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
init_sched_mm_cid(struct task_struct * t)3903 static inline void init_sched_mm_cid(struct task_struct *t) { }
3904 #endif /* !CONFIG_SCHED_MM_CID */
3905
3906 extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3907 extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3908 #ifdef CONFIG_SMP
3909 static inline
move_queued_task_locked(struct rq * src_rq,struct rq * dst_rq,struct task_struct * task)3910 void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task)
3911 {
3912 lockdep_assert_rq_held(src_rq);
3913 lockdep_assert_rq_held(dst_rq);
3914
3915 deactivate_task(src_rq, task, 0);
3916 set_task_cpu(task, dst_rq->cpu);
3917 activate_task(dst_rq, task, 0);
3918 }
3919
3920 static inline
task_is_pushable(struct rq * rq,struct task_struct * p,int cpu)3921 bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu)
3922 {
3923 if (!task_on_cpu(rq, p) &&
3924 cpumask_test_cpu(cpu, &p->cpus_mask))
3925 return true;
3926
3927 return false;
3928 }
3929 #endif
3930
3931 #ifdef CONFIG_RT_MUTEXES
3932
__rt_effective_prio(struct task_struct * pi_task,int prio)3933 static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3934 {
3935 if (pi_task)
3936 prio = min(prio, pi_task->prio);
3937
3938 return prio;
3939 }
3940
rt_effective_prio(struct task_struct * p,int prio)3941 static inline int rt_effective_prio(struct task_struct *p, int prio)
3942 {
3943 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3944
3945 return __rt_effective_prio(pi_task, prio);
3946 }
3947
3948 #else /* !CONFIG_RT_MUTEXES: */
3949
rt_effective_prio(struct task_struct * p,int prio)3950 static inline int rt_effective_prio(struct task_struct *p, int prio)
3951 {
3952 return prio;
3953 }
3954
3955 #endif /* !CONFIG_RT_MUTEXES */
3956
3957 extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3958 extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3959 extern const struct sched_class *__setscheduler_class(int policy, int prio);
3960 extern void set_load_weight(struct task_struct *p, bool update_load);
3961 extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3962 extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3963
3964 extern void check_class_changing(struct rq *rq, struct task_struct *p,
3965 const struct sched_class *prev_class);
3966 extern void check_class_changed(struct rq *rq, struct task_struct *p,
3967 const struct sched_class *prev_class,
3968 int oldprio);
3969
3970 #ifdef CONFIG_SMP
3971 extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3972 extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3973 #else
3974
splice_balance_callbacks(struct rq * rq)3975 static inline struct balance_callback *splice_balance_callbacks(struct rq *rq)
3976 {
3977 return NULL;
3978 }
3979
balance_callbacks(struct rq * rq,struct balance_callback * head)3980 static inline void balance_callbacks(struct rq *rq, struct balance_callback *head)
3981 {
3982 }
3983
3984 #endif
3985
3986 #ifdef CONFIG_SCHED_CLASS_EXT
3987 /*
3988 * Used by SCX in the enable/disable paths to move tasks between sched_classes
3989 * and establish invariants.
3990 */
3991 struct sched_enq_and_set_ctx {
3992 struct task_struct *p;
3993 int queue_flags;
3994 bool queued;
3995 bool running;
3996 };
3997
3998 void sched_deq_and_put_task(struct task_struct *p, int queue_flags,
3999 struct sched_enq_and_set_ctx *ctx);
4000 void sched_enq_and_set_task(struct sched_enq_and_set_ctx *ctx);
4001
4002 #endif /* CONFIG_SCHED_CLASS_EXT */
4003
4004 #include "ext.h"
4005
4006 #endif /* _KERNEL_SCHED_SCHED_H */
4007