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