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_PENDING = 1 << 2, /* balance hasn't run yet */
755 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */
756 SCX_RQ_BYPASSING = 1 << 4,
757
758 SCX_RQ_IN_WAKEUP = 1 << 16,
759 SCX_RQ_IN_BALANCE = 1 << 17,
760 };
761
762 struct scx_rq {
763 struct scx_dispatch_q local_dsq;
764 struct list_head runnable_list; /* runnable tasks on this rq */
765 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */
766 unsigned long ops_qseq;
767 u64 extra_enq_flags; /* see move_task_to_local_dsq() */
768 u32 nr_running;
769 u32 flags;
770 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */
771 bool cpu_released;
772 cpumask_var_t cpus_to_kick;
773 cpumask_var_t cpus_to_kick_if_idle;
774 cpumask_var_t cpus_to_preempt;
775 cpumask_var_t cpus_to_wait;
776 unsigned long pnt_seq;
777 struct balance_callback deferred_bal_cb;
778 struct irq_work deferred_irq_work;
779 struct irq_work kick_cpus_irq_work;
780 };
781 #endif /* CONFIG_SCHED_CLASS_EXT */
782
rt_bandwidth_enabled(void)783 static inline int rt_bandwidth_enabled(void)
784 {
785 return sysctl_sched_rt_runtime >= 0;
786 }
787
788 /* RT IPI pull logic requires IRQ_WORK */
789 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
790 # define HAVE_RT_PUSH_IPI
791 #endif
792
793 /* Real-Time classes' related field in a runqueue: */
794 struct rt_rq {
795 struct rt_prio_array active;
796 unsigned int rt_nr_running;
797 unsigned int rr_nr_running;
798 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
799 struct {
800 int curr; /* highest queued rt task prio */
801 #ifdef CONFIG_SMP
802 int next; /* next highest */
803 #endif
804 } highest_prio;
805 #endif
806 #ifdef CONFIG_SMP
807 bool overloaded;
808 struct plist_head pushable_tasks;
809
810 #endif /* CONFIG_SMP */
811 int rt_queued;
812
813 #ifdef CONFIG_RT_GROUP_SCHED
814 int rt_throttled;
815 u64 rt_time;
816 u64 rt_runtime;
817 /* Nests inside the rq lock: */
818 raw_spinlock_t rt_runtime_lock;
819
820 unsigned int rt_nr_boosted;
821
822 struct rq *rq;
823 struct task_group *tg;
824 #endif
825 };
826
rt_rq_is_runnable(struct rt_rq * rt_rq)827 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
828 {
829 return rt_rq->rt_queued && rt_rq->rt_nr_running;
830 }
831
832 /* Deadline class' related fields in a runqueue */
833 struct dl_rq {
834 /* runqueue is an rbtree, ordered by deadline */
835 struct rb_root_cached root;
836
837 unsigned int dl_nr_running;
838
839 #ifdef CONFIG_SMP
840 /*
841 * Deadline values of the currently executing and the
842 * earliest ready task on this rq. Caching these facilitates
843 * the decision whether or not a ready but not running task
844 * should migrate somewhere else.
845 */
846 struct {
847 u64 curr;
848 u64 next;
849 } earliest_dl;
850
851 bool overloaded;
852
853 /*
854 * Tasks on this rq that can be pushed away. They are kept in
855 * an rb-tree, ordered by tasks' deadlines, with caching
856 * of the leftmost (earliest deadline) element.
857 */
858 struct rb_root_cached pushable_dl_tasks_root;
859 #else
860 struct dl_bw dl_bw;
861 #endif
862 /*
863 * "Active utilization" for this runqueue: increased when a
864 * task wakes up (becomes TASK_RUNNING) and decreased when a
865 * task blocks
866 */
867 u64 running_bw;
868
869 /*
870 * Utilization of the tasks "assigned" to this runqueue (including
871 * the tasks that are in runqueue and the tasks that executed on this
872 * CPU and blocked). Increased when a task moves to this runqueue, and
873 * decreased when the task moves away (migrates, changes scheduling
874 * policy, or terminates).
875 * This is needed to compute the "inactive utilization" for the
876 * runqueue (inactive utilization = this_bw - running_bw).
877 */
878 u64 this_bw;
879 u64 extra_bw;
880
881 /*
882 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
883 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
884 */
885 u64 max_bw;
886
887 /*
888 * Inverse of the fraction of CPU utilization that can be reclaimed
889 * by the GRUB algorithm.
890 */
891 u64 bw_ratio;
892 };
893
894 #ifdef CONFIG_FAIR_GROUP_SCHED
895
896 /* An entity is a task if it doesn't "own" a runqueue */
897 #define entity_is_task(se) (!se->my_q)
898
se_update_runnable(struct sched_entity * se)899 static inline void se_update_runnable(struct sched_entity *se)
900 {
901 if (!entity_is_task(se))
902 se->runnable_weight = se->my_q->h_nr_running;
903 }
904
se_runnable(struct sched_entity * se)905 static inline long se_runnable(struct sched_entity *se)
906 {
907 if (se->sched_delayed)
908 return false;
909
910 if (entity_is_task(se))
911 return !!se->on_rq;
912 else
913 return se->runnable_weight;
914 }
915
916 #else /* !CONFIG_FAIR_GROUP_SCHED: */
917
918 #define entity_is_task(se) 1
919
se_update_runnable(struct sched_entity * se)920 static inline void se_update_runnable(struct sched_entity *se) { }
921
se_runnable(struct sched_entity * se)922 static inline long se_runnable(struct sched_entity *se)
923 {
924 if (se->sched_delayed)
925 return false;
926
927 return !!se->on_rq;
928 }
929
930 #endif /* !CONFIG_FAIR_GROUP_SCHED */
931
932 #ifdef CONFIG_SMP
933 /*
934 * XXX we want to get rid of these helpers and use the full load resolution.
935 */
se_weight(struct sched_entity * se)936 static inline long se_weight(struct sched_entity *se)
937 {
938 return scale_load_down(se->load.weight);
939 }
940
941
sched_asym_prefer(int a,int b)942 static inline bool sched_asym_prefer(int a, int b)
943 {
944 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
945 }
946
947 struct perf_domain {
948 struct em_perf_domain *em_pd;
949 struct perf_domain *next;
950 struct rcu_head rcu;
951 };
952
953 /*
954 * We add the notion of a root-domain which will be used to define per-domain
955 * variables. Each exclusive cpuset essentially defines an island domain by
956 * fully partitioning the member CPUs from any other cpuset. Whenever a new
957 * exclusive cpuset is created, we also create and attach a new root-domain
958 * object.
959 *
960 */
961 struct root_domain {
962 atomic_t refcount;
963 atomic_t rto_count;
964 struct rcu_head rcu;
965 cpumask_var_t span;
966 cpumask_var_t online;
967
968 /*
969 * Indicate pullable load on at least one CPU, e.g:
970 * - More than one runnable task
971 * - Running task is misfit
972 */
973 bool overloaded;
974
975 /* Indicate one or more CPUs over-utilized (tipping point) */
976 bool overutilized;
977
978 /*
979 * The bit corresponding to a CPU gets set here if such CPU has more
980 * than one runnable -deadline task (as it is below for RT tasks).
981 */
982 cpumask_var_t dlo_mask;
983 atomic_t dlo_count;
984 struct dl_bw dl_bw;
985 struct cpudl cpudl;
986
987 /*
988 * Indicate whether a root_domain's dl_bw has been checked or
989 * updated. It's monotonously increasing value.
990 *
991 * Also, some corner cases, like 'wrap around' is dangerous, but given
992 * that u64 is 'big enough'. So that shouldn't be a concern.
993 */
994 u64 visit_gen;
995
996 #ifdef HAVE_RT_PUSH_IPI
997 /*
998 * For IPI pull requests, loop across the rto_mask.
999 */
1000 struct irq_work rto_push_work;
1001 raw_spinlock_t rto_lock;
1002 /* These are only updated and read within rto_lock */
1003 int rto_loop;
1004 int rto_cpu;
1005 /* These atomics are updated outside of a lock */
1006 atomic_t rto_loop_next;
1007 atomic_t rto_loop_start;
1008 #endif
1009 /*
1010 * The "RT overload" flag: it gets set if a CPU has more than
1011 * one runnable RT task.
1012 */
1013 cpumask_var_t rto_mask;
1014 struct cpupri cpupri;
1015
1016 /*
1017 * NULL-terminated list of performance domains intersecting with the
1018 * CPUs of the rd. Protected by RCU.
1019 */
1020 struct perf_domain __rcu *pd;
1021 };
1022
1023 extern void init_defrootdomain(void);
1024 extern int sched_init_domains(const struct cpumask *cpu_map);
1025 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
1026 extern void sched_get_rd(struct root_domain *rd);
1027 extern void sched_put_rd(struct root_domain *rd);
1028
get_rd_overloaded(struct root_domain * rd)1029 static inline int get_rd_overloaded(struct root_domain *rd)
1030 {
1031 return READ_ONCE(rd->overloaded);
1032 }
1033
set_rd_overloaded(struct root_domain * rd,int status)1034 static inline void set_rd_overloaded(struct root_domain *rd, int status)
1035 {
1036 if (get_rd_overloaded(rd) != status)
1037 WRITE_ONCE(rd->overloaded, status);
1038 }
1039
1040 #ifdef HAVE_RT_PUSH_IPI
1041 extern void rto_push_irq_work_func(struct irq_work *work);
1042 #endif
1043 #endif /* CONFIG_SMP */
1044
1045 #ifdef CONFIG_UCLAMP_TASK
1046 /*
1047 * struct uclamp_bucket - Utilization clamp bucket
1048 * @value: utilization clamp value for tasks on this clamp bucket
1049 * @tasks: number of RUNNABLE tasks on this clamp bucket
1050 *
1051 * Keep track of how many tasks are RUNNABLE for a given utilization
1052 * clamp value.
1053 */
1054 struct uclamp_bucket {
1055 unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
1056 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
1057 };
1058
1059 /*
1060 * struct uclamp_rq - rq's utilization clamp
1061 * @value: currently active clamp values for a rq
1062 * @bucket: utilization clamp buckets affecting a rq
1063 *
1064 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
1065 * A clamp value is affecting a rq when there is at least one task RUNNABLE
1066 * (or actually running) with that value.
1067 *
1068 * There are up to UCLAMP_CNT possible different clamp values, currently there
1069 * are only two: minimum utilization and maximum utilization.
1070 *
1071 * All utilization clamping values are MAX aggregated, since:
1072 * - for util_min: we want to run the CPU at least at the max of the minimum
1073 * utilization required by its currently RUNNABLE tasks.
1074 * - for util_max: we want to allow the CPU to run up to the max of the
1075 * maximum utilization allowed by its currently RUNNABLE tasks.
1076 *
1077 * Since on each system we expect only a limited number of different
1078 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
1079 * the metrics required to compute all the per-rq utilization clamp values.
1080 */
1081 struct uclamp_rq {
1082 unsigned int value;
1083 struct uclamp_bucket bucket[UCLAMP_BUCKETS];
1084 };
1085
1086 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
1087 #endif /* CONFIG_UCLAMP_TASK */
1088
1089 /*
1090 * This is the main, per-CPU runqueue data structure.
1091 *
1092 * Locking rule: those places that want to lock multiple runqueues
1093 * (such as the load balancing or the thread migration code), lock
1094 * acquire operations must be ordered by ascending &runqueue.
1095 */
1096 struct rq {
1097 /* runqueue lock: */
1098 raw_spinlock_t __lock;
1099
1100 unsigned int nr_running;
1101 #ifdef CONFIG_NUMA_BALANCING
1102 unsigned int nr_numa_running;
1103 unsigned int nr_preferred_running;
1104 unsigned int numa_migrate_on;
1105 #endif
1106 #ifdef CONFIG_NO_HZ_COMMON
1107 #ifdef CONFIG_SMP
1108 unsigned long last_blocked_load_update_tick;
1109 unsigned int has_blocked_load;
1110 call_single_data_t nohz_csd;
1111 #endif /* CONFIG_SMP */
1112 unsigned int nohz_tick_stopped;
1113 atomic_t nohz_flags;
1114 #endif /* CONFIG_NO_HZ_COMMON */
1115
1116 #ifdef CONFIG_SMP
1117 unsigned int ttwu_pending;
1118 #endif
1119 u64 nr_switches;
1120
1121 #ifdef CONFIG_UCLAMP_TASK
1122 /* Utilization clamp values based on CPU's RUNNABLE tasks */
1123 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
1124 unsigned int uclamp_flags;
1125 #define UCLAMP_FLAG_IDLE 0x01
1126 #endif
1127
1128 struct cfs_rq cfs;
1129 struct rt_rq rt;
1130 struct dl_rq dl;
1131 #ifdef CONFIG_SCHED_CLASS_EXT
1132 struct scx_rq scx;
1133 #endif
1134
1135 struct sched_dl_entity fair_server;
1136
1137 #ifdef CONFIG_FAIR_GROUP_SCHED
1138 /* list of leaf cfs_rq on this CPU: */
1139 struct list_head leaf_cfs_rq_list;
1140 struct list_head *tmp_alone_branch;
1141 #endif /* CONFIG_FAIR_GROUP_SCHED */
1142
1143 /*
1144 * This is part of a global counter where only the total sum
1145 * over all CPUs matters. A task can increase this counter on
1146 * one CPU and if it got migrated afterwards it may decrease
1147 * it on another CPU. Always updated under the runqueue lock:
1148 */
1149 unsigned int nr_uninterruptible;
1150
1151 struct task_struct __rcu *curr;
1152 struct sched_dl_entity *dl_server;
1153 struct task_struct *idle;
1154 struct task_struct *stop;
1155 unsigned long next_balance;
1156 struct mm_struct *prev_mm;
1157
1158 unsigned int clock_update_flags;
1159 u64 clock;
1160 /* Ensure that all clocks are in the same cache line */
1161 u64 clock_task ____cacheline_aligned;
1162 u64 clock_pelt;
1163 unsigned long lost_idle_time;
1164 u64 clock_pelt_idle;
1165 u64 clock_idle;
1166 #ifndef CONFIG_64BIT
1167 u64 clock_pelt_idle_copy;
1168 u64 clock_idle_copy;
1169 #endif
1170
1171 atomic_t nr_iowait;
1172
1173 #ifdef CONFIG_SCHED_DEBUG
1174 u64 last_seen_need_resched_ns;
1175 int ticks_without_resched;
1176 #endif
1177
1178 #ifdef CONFIG_MEMBARRIER
1179 int membarrier_state;
1180 #endif
1181
1182 #ifdef CONFIG_SMP
1183 struct root_domain *rd;
1184 struct sched_domain __rcu *sd;
1185
1186 unsigned long cpu_capacity;
1187
1188 struct balance_callback *balance_callback;
1189
1190 unsigned char nohz_idle_balance;
1191 unsigned char idle_balance;
1192
1193 unsigned long misfit_task_load;
1194
1195 /* For active balancing */
1196 int active_balance;
1197 int push_cpu;
1198 struct cpu_stop_work active_balance_work;
1199
1200 /* CPU of this runqueue: */
1201 int cpu;
1202 int online;
1203
1204 struct list_head cfs_tasks;
1205
1206 struct sched_avg avg_rt;
1207 struct sched_avg avg_dl;
1208 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1209 struct sched_avg avg_irq;
1210 #endif
1211 #ifdef CONFIG_SCHED_HW_PRESSURE
1212 struct sched_avg avg_hw;
1213 #endif
1214 u64 idle_stamp;
1215 u64 avg_idle;
1216
1217 /* This is used to determine avg_idle's max value */
1218 u64 max_idle_balance_cost;
1219
1220 #ifdef CONFIG_HOTPLUG_CPU
1221 struct rcuwait hotplug_wait;
1222 #endif
1223 #endif /* CONFIG_SMP */
1224
1225 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1226 u64 prev_irq_time;
1227 u64 psi_irq_time;
1228 #endif
1229 #ifdef CONFIG_PARAVIRT
1230 u64 prev_steal_time;
1231 #endif
1232 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1233 u64 prev_steal_time_rq;
1234 #endif
1235
1236 /* calc_load related fields */
1237 unsigned long calc_load_update;
1238 long calc_load_active;
1239
1240 #ifdef CONFIG_SCHED_HRTICK
1241 #ifdef CONFIG_SMP
1242 call_single_data_t hrtick_csd;
1243 #endif
1244 struct hrtimer hrtick_timer;
1245 ktime_t hrtick_time;
1246 #endif
1247
1248 #ifdef CONFIG_SCHEDSTATS
1249 /* latency stats */
1250 struct sched_info rq_sched_info;
1251 unsigned long long rq_cpu_time;
1252
1253 /* sys_sched_yield() stats */
1254 unsigned int yld_count;
1255
1256 /* schedule() stats */
1257 unsigned int sched_count;
1258 unsigned int sched_goidle;
1259
1260 /* try_to_wake_up() stats */
1261 unsigned int ttwu_count;
1262 unsigned int ttwu_local;
1263 #endif
1264
1265 #ifdef CONFIG_CPU_IDLE
1266 /* Must be inspected within a RCU lock section */
1267 struct cpuidle_state *idle_state;
1268 #endif
1269
1270 #ifdef CONFIG_SMP
1271 unsigned int nr_pinned;
1272 #endif
1273 unsigned int push_busy;
1274 struct cpu_stop_work push_work;
1275
1276 #ifdef CONFIG_SCHED_CORE
1277 /* per rq */
1278 struct rq *core;
1279 struct task_struct *core_pick;
1280 struct sched_dl_entity *core_dl_server;
1281 unsigned int core_enabled;
1282 unsigned int core_sched_seq;
1283 struct rb_root core_tree;
1284
1285 /* shared state -- careful with sched_core_cpu_deactivate() */
1286 unsigned int core_task_seq;
1287 unsigned int core_pick_seq;
1288 unsigned long core_cookie;
1289 unsigned int core_forceidle_count;
1290 unsigned int core_forceidle_seq;
1291 unsigned int core_forceidle_occupation;
1292 u64 core_forceidle_start;
1293 #endif
1294
1295 /* Scratch cpumask to be temporarily used under rq_lock */
1296 cpumask_var_t scratch_mask;
1297
1298 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1299 call_single_data_t cfsb_csd;
1300 struct list_head cfsb_csd_list;
1301 #endif
1302 };
1303
1304 #ifdef CONFIG_FAIR_GROUP_SCHED
1305
1306 /* CPU runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)1307 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1308 {
1309 return cfs_rq->rq;
1310 }
1311
1312 #else
1313
rq_of(struct cfs_rq * cfs_rq)1314 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1315 {
1316 return container_of(cfs_rq, struct rq, cfs);
1317 }
1318 #endif
1319
cpu_of(struct rq * rq)1320 static inline int cpu_of(struct rq *rq)
1321 {
1322 #ifdef CONFIG_SMP
1323 return rq->cpu;
1324 #else
1325 return 0;
1326 #endif
1327 }
1328
1329 #define MDF_PUSH 0x01
1330
is_migration_disabled(struct task_struct * p)1331 static inline bool is_migration_disabled(struct task_struct *p)
1332 {
1333 #ifdef CONFIG_SMP
1334 return p->migration_disabled;
1335 #else
1336 return false;
1337 #endif
1338 }
1339
1340 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1341
1342 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1343 #define this_rq() this_cpu_ptr(&runqueues)
1344 #define task_rq(p) cpu_rq(task_cpu(p))
1345 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1346 #define raw_rq() raw_cpu_ptr(&runqueues)
1347
1348 #ifdef CONFIG_SCHED_CORE
1349 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1350
1351 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1352
sched_core_enabled(struct rq * rq)1353 static inline bool sched_core_enabled(struct rq *rq)
1354 {
1355 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1356 }
1357
sched_core_disabled(void)1358 static inline bool sched_core_disabled(void)
1359 {
1360 return !static_branch_unlikely(&__sched_core_enabled);
1361 }
1362
1363 /*
1364 * Be careful with this function; not for general use. The return value isn't
1365 * stable unless you actually hold a relevant rq->__lock.
1366 */
rq_lockp(struct rq * rq)1367 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1368 {
1369 if (sched_core_enabled(rq))
1370 return &rq->core->__lock;
1371
1372 return &rq->__lock;
1373 }
1374
__rq_lockp(struct rq * rq)1375 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1376 {
1377 if (rq->core_enabled)
1378 return &rq->core->__lock;
1379
1380 return &rq->__lock;
1381 }
1382
1383 extern bool
1384 cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi);
1385
1386 extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1387
1388 /*
1389 * Helpers to check if the CPU's core cookie matches with the task's cookie
1390 * when core scheduling is enabled.
1391 * A special case is that the task's cookie always matches with CPU's core
1392 * cookie if the CPU is in an idle core.
1393 */
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1394 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1395 {
1396 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1397 if (!sched_core_enabled(rq))
1398 return true;
1399
1400 return rq->core->core_cookie == p->core_cookie;
1401 }
1402
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1403 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1404 {
1405 bool idle_core = true;
1406 int cpu;
1407
1408 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1409 if (!sched_core_enabled(rq))
1410 return true;
1411
1412 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1413 if (!available_idle_cpu(cpu)) {
1414 idle_core = false;
1415 break;
1416 }
1417 }
1418
1419 /*
1420 * A CPU in an idle core is always the best choice for tasks with
1421 * cookies.
1422 */
1423 return idle_core || rq->core->core_cookie == p->core_cookie;
1424 }
1425
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1426 static inline bool sched_group_cookie_match(struct rq *rq,
1427 struct task_struct *p,
1428 struct sched_group *group)
1429 {
1430 int cpu;
1431
1432 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1433 if (!sched_core_enabled(rq))
1434 return true;
1435
1436 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1437 if (sched_core_cookie_match(cpu_rq(cpu), p))
1438 return true;
1439 }
1440 return false;
1441 }
1442
sched_core_enqueued(struct task_struct * p)1443 static inline bool sched_core_enqueued(struct task_struct *p)
1444 {
1445 return !RB_EMPTY_NODE(&p->core_node);
1446 }
1447
1448 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1449 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1450
1451 extern void sched_core_get(void);
1452 extern void sched_core_put(void);
1453
1454 #else /* !CONFIG_SCHED_CORE: */
1455
sched_core_enabled(struct rq * rq)1456 static inline bool sched_core_enabled(struct rq *rq)
1457 {
1458 return false;
1459 }
1460
sched_core_disabled(void)1461 static inline bool sched_core_disabled(void)
1462 {
1463 return true;
1464 }
1465
rq_lockp(struct rq * rq)1466 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1467 {
1468 return &rq->__lock;
1469 }
1470
__rq_lockp(struct rq * rq)1471 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1472 {
1473 return &rq->__lock;
1474 }
1475
sched_cpu_cookie_match(struct rq * rq,struct task_struct * p)1476 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1477 {
1478 return true;
1479 }
1480
sched_core_cookie_match(struct rq * rq,struct task_struct * p)1481 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1482 {
1483 return true;
1484 }
1485
sched_group_cookie_match(struct rq * rq,struct task_struct * p,struct sched_group * group)1486 static inline bool sched_group_cookie_match(struct rq *rq,
1487 struct task_struct *p,
1488 struct sched_group *group)
1489 {
1490 return true;
1491 }
1492
1493 #endif /* !CONFIG_SCHED_CORE */
1494
lockdep_assert_rq_held(struct rq * rq)1495 static inline void lockdep_assert_rq_held(struct rq *rq)
1496 {
1497 lockdep_assert_held(__rq_lockp(rq));
1498 }
1499
1500 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1501 extern bool raw_spin_rq_trylock(struct rq *rq);
1502 extern void raw_spin_rq_unlock(struct rq *rq);
1503
raw_spin_rq_lock(struct rq * rq)1504 static inline void raw_spin_rq_lock(struct rq *rq)
1505 {
1506 raw_spin_rq_lock_nested(rq, 0);
1507 }
1508
raw_spin_rq_lock_irq(struct rq * rq)1509 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1510 {
1511 local_irq_disable();
1512 raw_spin_rq_lock(rq);
1513 }
1514
raw_spin_rq_unlock_irq(struct rq * rq)1515 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1516 {
1517 raw_spin_rq_unlock(rq);
1518 local_irq_enable();
1519 }
1520
_raw_spin_rq_lock_irqsave(struct rq * rq)1521 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1522 {
1523 unsigned long flags;
1524
1525 local_irq_save(flags);
1526 raw_spin_rq_lock(rq);
1527
1528 return flags;
1529 }
1530
raw_spin_rq_unlock_irqrestore(struct rq * rq,unsigned long flags)1531 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1532 {
1533 raw_spin_rq_unlock(rq);
1534 local_irq_restore(flags);
1535 }
1536
1537 #define raw_spin_rq_lock_irqsave(rq, flags) \
1538 do { \
1539 flags = _raw_spin_rq_lock_irqsave(rq); \
1540 } while (0)
1541
1542 #ifdef CONFIG_SCHED_SMT
1543 extern void __update_idle_core(struct rq *rq);
1544
update_idle_core(struct rq * rq)1545 static inline void update_idle_core(struct rq *rq)
1546 {
1547 if (static_branch_unlikely(&sched_smt_present))
1548 __update_idle_core(rq);
1549 }
1550
1551 #else
update_idle_core(struct rq * rq)1552 static inline void update_idle_core(struct rq *rq) { }
1553 #endif
1554
1555 #ifdef CONFIG_FAIR_GROUP_SCHED
1556
task_of(struct sched_entity * se)1557 static inline struct task_struct *task_of(struct sched_entity *se)
1558 {
1559 SCHED_WARN_ON(!entity_is_task(se));
1560 return container_of(se, struct task_struct, se);
1561 }
1562
task_cfs_rq(struct task_struct * p)1563 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1564 {
1565 return p->se.cfs_rq;
1566 }
1567
1568 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(const struct sched_entity * se)1569 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1570 {
1571 return se->cfs_rq;
1572 }
1573
1574 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1575 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1576 {
1577 return grp->my_q;
1578 }
1579
1580 #else /* !CONFIG_FAIR_GROUP_SCHED: */
1581
1582 #define task_of(_se) container_of(_se, struct task_struct, se)
1583
task_cfs_rq(const struct task_struct * p)1584 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1585 {
1586 return &task_rq(p)->cfs;
1587 }
1588
cfs_rq_of(const struct sched_entity * se)1589 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1590 {
1591 const struct task_struct *p = task_of(se);
1592 struct rq *rq = task_rq(p);
1593
1594 return &rq->cfs;
1595 }
1596
1597 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)1598 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1599 {
1600 return NULL;
1601 }
1602
1603 #endif /* !CONFIG_FAIR_GROUP_SCHED */
1604
1605 extern void update_rq_clock(struct rq *rq);
1606
1607 /*
1608 * rq::clock_update_flags bits
1609 *
1610 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1611 * call to __schedule(). This is an optimisation to avoid
1612 * neighbouring rq clock updates.
1613 *
1614 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1615 * in effect and calls to update_rq_clock() are being ignored.
1616 *
1617 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1618 * made to update_rq_clock() since the last time rq::lock was pinned.
1619 *
1620 * If inside of __schedule(), clock_update_flags will have been
1621 * shifted left (a left shift is a cheap operation for the fast path
1622 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1623 *
1624 * if (rq-clock_update_flags >= RQCF_UPDATED)
1625 *
1626 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1627 * one position though, because the next rq_unpin_lock() will shift it
1628 * back.
1629 */
1630 #define RQCF_REQ_SKIP 0x01
1631 #define RQCF_ACT_SKIP 0x02
1632 #define RQCF_UPDATED 0x04
1633
assert_clock_updated(struct rq * rq)1634 static inline void assert_clock_updated(struct rq *rq)
1635 {
1636 /*
1637 * The only reason for not seeing a clock update since the
1638 * last rq_pin_lock() is if we're currently skipping updates.
1639 */
1640 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1641 }
1642
rq_clock(struct rq * rq)1643 static inline u64 rq_clock(struct rq *rq)
1644 {
1645 lockdep_assert_rq_held(rq);
1646 assert_clock_updated(rq);
1647
1648 return rq->clock;
1649 }
1650
rq_clock_task(struct rq * rq)1651 static inline u64 rq_clock_task(struct rq *rq)
1652 {
1653 lockdep_assert_rq_held(rq);
1654 assert_clock_updated(rq);
1655
1656 return rq->clock_task;
1657 }
1658
rq_clock_skip_update(struct rq * rq)1659 static inline void rq_clock_skip_update(struct rq *rq)
1660 {
1661 lockdep_assert_rq_held(rq);
1662 rq->clock_update_flags |= RQCF_REQ_SKIP;
1663 }
1664
1665 /*
1666 * See rt task throttling, which is the only time a skip
1667 * request is canceled.
1668 */
rq_clock_cancel_skipupdate(struct rq * rq)1669 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1670 {
1671 lockdep_assert_rq_held(rq);
1672 rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1673 }
1674
1675 /*
1676 * During cpu offlining and rq wide unthrottling, we can trigger
1677 * an update_rq_clock() for several cfs and rt runqueues (Typically
1678 * when using list_for_each_entry_*)
1679 * rq_clock_start_loop_update() can be called after updating the clock
1680 * once and before iterating over the list to prevent multiple update.
1681 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1682 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1683 */
rq_clock_start_loop_update(struct rq * rq)1684 static inline void rq_clock_start_loop_update(struct rq *rq)
1685 {
1686 lockdep_assert_rq_held(rq);
1687 SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1688 rq->clock_update_flags |= RQCF_ACT_SKIP;
1689 }
1690
rq_clock_stop_loop_update(struct rq * rq)1691 static inline void rq_clock_stop_loop_update(struct rq *rq)
1692 {
1693 lockdep_assert_rq_held(rq);
1694 rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1695 }
1696
1697 struct rq_flags {
1698 unsigned long flags;
1699 struct pin_cookie cookie;
1700 #ifdef CONFIG_SCHED_DEBUG
1701 /*
1702 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1703 * current pin context is stashed here in case it needs to be
1704 * restored in rq_repin_lock().
1705 */
1706 unsigned int clock_update_flags;
1707 #endif
1708 };
1709
1710 extern struct balance_callback balance_push_callback;
1711
1712 /*
1713 * Lockdep annotation that avoids accidental unlocks; it's like a
1714 * sticky/continuous lockdep_assert_held().
1715 *
1716 * This avoids code that has access to 'struct rq *rq' (basically everything in
1717 * the scheduler) from accidentally unlocking the rq if they do not also have a
1718 * copy of the (on-stack) 'struct rq_flags rf'.
1719 *
1720 * Also see Documentation/locking/lockdep-design.rst.
1721 */
rq_pin_lock(struct rq * rq,struct rq_flags * rf)1722 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1723 {
1724 rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1725
1726 #ifdef CONFIG_SCHED_DEBUG
1727 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1728 rf->clock_update_flags = 0;
1729 # ifdef CONFIG_SMP
1730 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1731 # endif
1732 #endif
1733 }
1734
rq_unpin_lock(struct rq * rq,struct rq_flags * rf)1735 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1736 {
1737 #ifdef CONFIG_SCHED_DEBUG
1738 if (rq->clock_update_flags > RQCF_ACT_SKIP)
1739 rf->clock_update_flags = RQCF_UPDATED;
1740 #endif
1741
1742 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1743 }
1744
rq_repin_lock(struct rq * rq,struct rq_flags * rf)1745 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1746 {
1747 lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1748
1749 #ifdef CONFIG_SCHED_DEBUG
1750 /*
1751 * Restore the value we stashed in @rf for this pin context.
1752 */
1753 rq->clock_update_flags |= rf->clock_update_flags;
1754 #endif
1755 }
1756
1757 extern
1758 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1759 __acquires(rq->lock);
1760
1761 extern
1762 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1763 __acquires(p->pi_lock)
1764 __acquires(rq->lock);
1765
__task_rq_unlock(struct rq * rq,struct rq_flags * rf)1766 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1767 __releases(rq->lock)
1768 {
1769 rq_unpin_lock(rq, rf);
1770 raw_spin_rq_unlock(rq);
1771 }
1772
1773 static inline void
task_rq_unlock(struct rq * rq,struct task_struct * p,struct rq_flags * rf)1774 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1775 __releases(rq->lock)
1776 __releases(p->pi_lock)
1777 {
1778 rq_unpin_lock(rq, rf);
1779 raw_spin_rq_unlock(rq);
1780 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1781 }
1782
1783 DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1784 _T->rq = task_rq_lock(_T->lock, &_T->rf),
1785 task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1786 struct rq *rq; struct rq_flags rf)
1787
rq_lock_irqsave(struct rq * rq,struct rq_flags * rf)1788 static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1789 __acquires(rq->lock)
1790 {
1791 raw_spin_rq_lock_irqsave(rq, rf->flags);
1792 rq_pin_lock(rq, rf);
1793 }
1794
rq_lock_irq(struct rq * rq,struct rq_flags * rf)1795 static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1796 __acquires(rq->lock)
1797 {
1798 raw_spin_rq_lock_irq(rq);
1799 rq_pin_lock(rq, rf);
1800 }
1801
rq_lock(struct rq * rq,struct rq_flags * rf)1802 static inline void rq_lock(struct rq *rq, struct rq_flags *rf)
1803 __acquires(rq->lock)
1804 {
1805 raw_spin_rq_lock(rq);
1806 rq_pin_lock(rq, rf);
1807 }
1808
rq_unlock_irqrestore(struct rq * rq,struct rq_flags * rf)1809 static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1810 __releases(rq->lock)
1811 {
1812 rq_unpin_lock(rq, rf);
1813 raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1814 }
1815
rq_unlock_irq(struct rq * rq,struct rq_flags * rf)1816 static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1817 __releases(rq->lock)
1818 {
1819 rq_unpin_lock(rq, rf);
1820 raw_spin_rq_unlock_irq(rq);
1821 }
1822
rq_unlock(struct rq * rq,struct rq_flags * rf)1823 static inline void rq_unlock(struct rq *rq, struct rq_flags *rf)
1824 __releases(rq->lock)
1825 {
1826 rq_unpin_lock(rq, rf);
1827 raw_spin_rq_unlock(rq);
1828 }
1829
1830 DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1831 rq_lock(_T->lock, &_T->rf),
1832 rq_unlock(_T->lock, &_T->rf),
1833 struct rq_flags rf)
1834
1835 DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1836 rq_lock_irq(_T->lock, &_T->rf),
1837 rq_unlock_irq(_T->lock, &_T->rf),
1838 struct rq_flags rf)
1839
1840 DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1841 rq_lock_irqsave(_T->lock, &_T->rf),
1842 rq_unlock_irqrestore(_T->lock, &_T->rf),
1843 struct rq_flags rf)
1844
this_rq_lock_irq(struct rq_flags * rf)1845 static inline struct rq *this_rq_lock_irq(struct rq_flags *rf)
1846 __acquires(rq->lock)
1847 {
1848 struct rq *rq;
1849
1850 local_irq_disable();
1851 rq = this_rq();
1852 rq_lock(rq, rf);
1853
1854 return rq;
1855 }
1856
1857 #ifdef CONFIG_NUMA
1858
1859 enum numa_topology_type {
1860 NUMA_DIRECT,
1861 NUMA_GLUELESS_MESH,
1862 NUMA_BACKPLANE,
1863 };
1864
1865 extern enum numa_topology_type sched_numa_topology_type;
1866 extern int sched_max_numa_distance;
1867 extern bool find_numa_distance(int distance);
1868 extern void sched_init_numa(int offline_node);
1869 extern void sched_update_numa(int cpu, bool online);
1870 extern void sched_domains_numa_masks_set(unsigned int cpu);
1871 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1872 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1873
1874 #else /* !CONFIG_NUMA: */
1875
sched_init_numa(int offline_node)1876 static inline void sched_init_numa(int offline_node) { }
sched_update_numa(int cpu,bool online)1877 static inline void sched_update_numa(int cpu, bool online) { }
sched_domains_numa_masks_set(unsigned int cpu)1878 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
sched_domains_numa_masks_clear(unsigned int cpu)1879 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1880
sched_numa_find_closest(const struct cpumask * cpus,int cpu)1881 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1882 {
1883 return nr_cpu_ids;
1884 }
1885
1886 #endif /* !CONFIG_NUMA */
1887
1888 #ifdef CONFIG_NUMA_BALANCING
1889
1890 /* The regions in numa_faults array from task_struct */
1891 enum numa_faults_stats {
1892 NUMA_MEM = 0,
1893 NUMA_CPU,
1894 NUMA_MEMBUF,
1895 NUMA_CPUBUF
1896 };
1897
1898 extern void sched_setnuma(struct task_struct *p, int node);
1899 extern int migrate_task_to(struct task_struct *p, int cpu);
1900 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1901 int cpu, int scpu);
1902 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1903
1904 #else /* !CONFIG_NUMA_BALANCING: */
1905
1906 static inline void
init_numa_balancing(unsigned long clone_flags,struct task_struct * p)1907 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1908 {
1909 }
1910
1911 #endif /* !CONFIG_NUMA_BALANCING */
1912
1913 #ifdef CONFIG_SMP
1914
1915 static inline void
queue_balance_callback(struct rq * rq,struct balance_callback * head,void (* func)(struct rq * rq))1916 queue_balance_callback(struct rq *rq,
1917 struct balance_callback *head,
1918 void (*func)(struct rq *rq))
1919 {
1920 lockdep_assert_rq_held(rq);
1921
1922 /*
1923 * Don't (re)queue an already queued item; nor queue anything when
1924 * balance_push() is active, see the comment with
1925 * balance_push_callback.
1926 */
1927 if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1928 return;
1929
1930 head->func = func;
1931 head->next = rq->balance_callback;
1932 rq->balance_callback = head;
1933 }
1934
1935 #define rcu_dereference_check_sched_domain(p) \
1936 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex))
1937
1938 /*
1939 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1940 * See destroy_sched_domains: call_rcu for details.
1941 *
1942 * The domain tree of any CPU may only be accessed from within
1943 * preempt-disabled sections.
1944 */
1945 #define for_each_domain(cpu, __sd) \
1946 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1947 __sd; __sd = __sd->parent)
1948
1949 /* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1950 #define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1951 static const unsigned int SD_SHARED_CHILD_MASK =
1952 #include <linux/sched/sd_flags.h>
1953 0;
1954 #undef SD_FLAG
1955
1956 /**
1957 * highest_flag_domain - Return highest sched_domain containing flag.
1958 * @cpu: The CPU whose highest level of sched domain is to
1959 * be returned.
1960 * @flag: The flag to check for the highest sched_domain
1961 * for the given CPU.
1962 *
1963 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1964 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1965 */
highest_flag_domain(int cpu,int flag)1966 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1967 {
1968 struct sched_domain *sd, *hsd = NULL;
1969
1970 for_each_domain(cpu, sd) {
1971 if (sd->flags & flag) {
1972 hsd = sd;
1973 continue;
1974 }
1975
1976 /*
1977 * Stop the search if @flag is known to be shared at lower
1978 * levels. It will not be found further up.
1979 */
1980 if (flag & SD_SHARED_CHILD_MASK)
1981 break;
1982 }
1983
1984 return hsd;
1985 }
1986
lowest_flag_domain(int cpu,int flag)1987 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1988 {
1989 struct sched_domain *sd;
1990
1991 for_each_domain(cpu, sd) {
1992 if (sd->flags & flag)
1993 break;
1994 }
1995
1996 return sd;
1997 }
1998
1999 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
2000 DECLARE_PER_CPU(int, sd_llc_size);
2001 DECLARE_PER_CPU(int, sd_llc_id);
2002 DECLARE_PER_CPU(int, sd_share_id);
2003 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
2004 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
2005 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
2006 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
2007
2008 extern struct static_key_false sched_asym_cpucapacity;
2009 extern struct static_key_false sched_cluster_active;
2010
sched_asym_cpucap_active(void)2011 static __always_inline bool sched_asym_cpucap_active(void)
2012 {
2013 return static_branch_unlikely(&sched_asym_cpucapacity);
2014 }
2015
2016 struct sched_group_capacity {
2017 atomic_t ref;
2018 /*
2019 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
2020 * for a single CPU.
2021 */
2022 unsigned long capacity;
2023 unsigned long min_capacity; /* Min per-CPU capacity in group */
2024 unsigned long max_capacity; /* Max per-CPU capacity in group */
2025 unsigned long next_update;
2026 int imbalance; /* XXX unrelated to capacity but shared group state */
2027
2028 #ifdef CONFIG_SCHED_DEBUG
2029 int id;
2030 #endif
2031
2032 unsigned long cpumask[]; /* Balance mask */
2033 };
2034
2035 struct sched_group {
2036 struct sched_group *next; /* Must be a circular list */
2037 atomic_t ref;
2038
2039 unsigned int group_weight;
2040 unsigned int cores;
2041 struct sched_group_capacity *sgc;
2042 int asym_prefer_cpu; /* CPU of highest priority in group */
2043 int flags;
2044
2045 /*
2046 * The CPUs this group covers.
2047 *
2048 * NOTE: this field is variable length. (Allocated dynamically
2049 * by attaching extra space to the end of the structure,
2050 * depending on how many CPUs the kernel has booted up with)
2051 */
2052 unsigned long cpumask[];
2053 };
2054
sched_group_span(struct sched_group * sg)2055 static inline struct cpumask *sched_group_span(struct sched_group *sg)
2056 {
2057 return to_cpumask(sg->cpumask);
2058 }
2059
2060 /*
2061 * See build_balance_mask().
2062 */
group_balance_mask(struct sched_group * sg)2063 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
2064 {
2065 return to_cpumask(sg->sgc->cpumask);
2066 }
2067
2068 extern int group_balance_cpu(struct sched_group *sg);
2069
2070 #ifdef CONFIG_SCHED_DEBUG
2071 extern void update_sched_domain_debugfs(void);
2072 extern void dirty_sched_domain_sysctl(int cpu);
2073 #else
update_sched_domain_debugfs(void)2074 static inline void update_sched_domain_debugfs(void) { }
dirty_sched_domain_sysctl(int cpu)2075 static inline void dirty_sched_domain_sysctl(int cpu) { }
2076 #endif
2077
2078 extern int sched_update_scaling(void);
2079
task_user_cpus(struct task_struct * p)2080 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
2081 {
2082 if (!p->user_cpus_ptr)
2083 return cpu_possible_mask; /* &init_task.cpus_mask */
2084 return p->user_cpus_ptr;
2085 }
2086
2087 #endif /* CONFIG_SMP */
2088
2089 #include "stats.h"
2090
2091 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
2092
2093 extern void __sched_core_account_forceidle(struct rq *rq);
2094
sched_core_account_forceidle(struct rq * rq)2095 static inline void sched_core_account_forceidle(struct rq *rq)
2096 {
2097 if (schedstat_enabled())
2098 __sched_core_account_forceidle(rq);
2099 }
2100
2101 extern void __sched_core_tick(struct rq *rq);
2102
sched_core_tick(struct rq * rq)2103 static inline void sched_core_tick(struct rq *rq)
2104 {
2105 if (sched_core_enabled(rq) && schedstat_enabled())
2106 __sched_core_tick(rq);
2107 }
2108
2109 #else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */
2110
sched_core_account_forceidle(struct rq * rq)2111 static inline void sched_core_account_forceidle(struct rq *rq) { }
2112
sched_core_tick(struct rq * rq)2113 static inline void sched_core_tick(struct rq *rq) { }
2114
2115 #endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */
2116
2117 #ifdef CONFIG_CGROUP_SCHED
2118
2119 /*
2120 * Return the group to which this tasks belongs.
2121 *
2122 * We cannot use task_css() and friends because the cgroup subsystem
2123 * changes that value before the cgroup_subsys::attach() method is called,
2124 * therefore we cannot pin it and might observe the wrong value.
2125 *
2126 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2127 * core changes this before calling sched_move_task().
2128 *
2129 * Instead we use a 'copy' which is updated from sched_move_task() while
2130 * holding both task_struct::pi_lock and rq::lock.
2131 */
task_group(struct task_struct * p)2132 static inline struct task_group *task_group(struct task_struct *p)
2133 {
2134 return p->sched_task_group;
2135 }
2136
2137 /* 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)2138 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2139 {
2140 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2141 struct task_group *tg = task_group(p);
2142 #endif
2143
2144 #ifdef CONFIG_FAIR_GROUP_SCHED
2145 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2146 p->se.cfs_rq = tg->cfs_rq[cpu];
2147 p->se.parent = tg->se[cpu];
2148 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2149 #endif
2150
2151 #ifdef CONFIG_RT_GROUP_SCHED
2152 p->rt.rt_rq = tg->rt_rq[cpu];
2153 p->rt.parent = tg->rt_se[cpu];
2154 #endif
2155 }
2156
2157 #else /* !CONFIG_CGROUP_SCHED: */
2158
set_task_rq(struct task_struct * p,unsigned int cpu)2159 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2160
task_group(struct task_struct * p)2161 static inline struct task_group *task_group(struct task_struct *p)
2162 {
2163 return NULL;
2164 }
2165
2166 #endif /* !CONFIG_CGROUP_SCHED */
2167
__set_task_cpu(struct task_struct * p,unsigned int cpu)2168 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2169 {
2170 set_task_rq(p, cpu);
2171 #ifdef CONFIG_SMP
2172 /*
2173 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2174 * successfully executed on another CPU. We must ensure that updates of
2175 * per-task data have been completed by this moment.
2176 */
2177 smp_wmb();
2178 WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2179 p->wake_cpu = cpu;
2180 #endif
2181 }
2182
2183 /*
2184 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2185 */
2186 #ifdef CONFIG_SCHED_DEBUG
2187 # define const_debug __read_mostly
2188 #else
2189 # define const_debug const
2190 #endif
2191
2192 #define SCHED_FEAT(name, enabled) \
2193 __SCHED_FEAT_##name ,
2194
2195 enum {
2196 #include "features.h"
2197 __SCHED_FEAT_NR,
2198 };
2199
2200 #undef SCHED_FEAT
2201
2202 #ifdef CONFIG_SCHED_DEBUG
2203
2204 /*
2205 * To support run-time toggling of sched features, all the translation units
2206 * (but core.c) reference the sysctl_sched_features defined in core.c.
2207 */
2208 extern const_debug unsigned int sysctl_sched_features;
2209
2210 #ifdef CONFIG_JUMP_LABEL
2211
2212 #define SCHED_FEAT(name, enabled) \
2213 static __always_inline bool static_branch_##name(struct static_key *key) \
2214 { \
2215 return static_key_##enabled(key); \
2216 }
2217
2218 #include "features.h"
2219 #undef SCHED_FEAT
2220
2221 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2222 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2223
2224 #else /* !CONFIG_JUMP_LABEL: */
2225
2226 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2227
2228 #endif /* !CONFIG_JUMP_LABEL */
2229
2230 #else /* !SCHED_DEBUG: */
2231
2232 /*
2233 * Each translation unit has its own copy of sysctl_sched_features to allow
2234 * constants propagation at compile time and compiler optimization based on
2235 * features default.
2236 */
2237 #define SCHED_FEAT(name, enabled) \
2238 (1UL << __SCHED_FEAT_##name) * enabled |
2239 static const_debug __maybe_unused unsigned int sysctl_sched_features =
2240 #include "features.h"
2241 0;
2242 #undef SCHED_FEAT
2243
2244 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2245
2246 #endif /* !SCHED_DEBUG */
2247
2248 extern struct static_key_false sched_numa_balancing;
2249 extern struct static_key_false sched_schedstats;
2250
global_rt_period(void)2251 static inline u64 global_rt_period(void)
2252 {
2253 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2254 }
2255
global_rt_runtime(void)2256 static inline u64 global_rt_runtime(void)
2257 {
2258 if (sysctl_sched_rt_runtime < 0)
2259 return RUNTIME_INF;
2260
2261 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2262 }
2263
task_current(struct rq * rq,struct task_struct * p)2264 static inline int task_current(struct rq *rq, struct task_struct *p)
2265 {
2266 return rq->curr == p;
2267 }
2268
task_on_cpu(struct rq * rq,struct task_struct * p)2269 static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2270 {
2271 #ifdef CONFIG_SMP
2272 return p->on_cpu;
2273 #else
2274 return task_current(rq, p);
2275 #endif
2276 }
2277
task_on_rq_queued(struct task_struct * p)2278 static inline int task_on_rq_queued(struct task_struct *p)
2279 {
2280 return p->on_rq == TASK_ON_RQ_QUEUED;
2281 }
2282
task_on_rq_migrating(struct task_struct * p)2283 static inline int task_on_rq_migrating(struct task_struct *p)
2284 {
2285 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2286 }
2287
2288 /* Wake flags. The first three directly map to some SD flag value */
2289 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2290 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2291 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2292
2293 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2294 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2295 #define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */
2296 #define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */
2297
2298 #ifdef CONFIG_SMP
2299 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2300 static_assert(WF_FORK == SD_BALANCE_FORK);
2301 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2302 #endif
2303
2304 /*
2305 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2306 * of tasks with abnormal "nice" values across CPUs the contribution that
2307 * each task makes to its run queue's load is weighted according to its
2308 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2309 * scaled version of the new time slice allocation that they receive on time
2310 * slice expiry etc.
2311 */
2312
2313 #define WEIGHT_IDLEPRIO 3
2314 #define WMULT_IDLEPRIO 1431655765
2315
2316 extern const int sched_prio_to_weight[40];
2317 extern const u32 sched_prio_to_wmult[40];
2318
2319 /*
2320 * {de,en}queue flags:
2321 *
2322 * DEQUEUE_SLEEP - task is no longer runnable
2323 * ENQUEUE_WAKEUP - task just became runnable
2324 *
2325 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2326 * are in a known state which allows modification. Such pairs
2327 * should preserve as much state as possible.
2328 *
2329 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2330 * in the runqueue.
2331 *
2332 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2333 *
2334 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2335 *
2336 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2337 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2338 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2339 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called
2340 *
2341 */
2342
2343 #define DEQUEUE_SLEEP 0x01 /* Matches ENQUEUE_WAKEUP */
2344 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2345 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2346 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2347 #define DEQUEUE_SPECIAL 0x10
2348 #define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */
2349 #define DEQUEUE_DELAYED 0x200 /* Matches ENQUEUE_DELAYED */
2350
2351 #define ENQUEUE_WAKEUP 0x01
2352 #define ENQUEUE_RESTORE 0x02
2353 #define ENQUEUE_MOVE 0x04
2354 #define ENQUEUE_NOCLOCK 0x08
2355
2356 #define ENQUEUE_HEAD 0x10
2357 #define ENQUEUE_REPLENISH 0x20
2358 #ifdef CONFIG_SMP
2359 #define ENQUEUE_MIGRATED 0x40
2360 #else
2361 #define ENQUEUE_MIGRATED 0x00
2362 #endif
2363 #define ENQUEUE_INITIAL 0x80
2364 #define ENQUEUE_MIGRATING 0x100
2365 #define ENQUEUE_DELAYED 0x200
2366 #define ENQUEUE_RQ_SELECTED 0x400
2367
2368 #define RETRY_TASK ((void *)-1UL)
2369
2370 struct affinity_context {
2371 const struct cpumask *new_mask;
2372 struct cpumask *user_mask;
2373 unsigned int flags;
2374 };
2375
2376 extern s64 update_curr_common(struct rq *rq);
2377
2378 struct sched_class {
2379
2380 #ifdef CONFIG_UCLAMP_TASK
2381 int uclamp_enabled;
2382 #endif
2383
2384 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2385 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2386 void (*yield_task) (struct rq *rq);
2387 bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2388
2389 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2390
2391 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2392 struct task_struct *(*pick_task)(struct rq *rq);
2393 /*
2394 * Optional! When implemented pick_next_task() should be equivalent to:
2395 *
2396 * next = pick_task();
2397 * if (next) {
2398 * put_prev_task(prev);
2399 * set_next_task_first(next);
2400 * }
2401 */
2402 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev);
2403
2404 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next);
2405 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2406
2407 #ifdef CONFIG_SMP
2408 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2409
2410 void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2411
2412 void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2413
2414 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2415
2416 void (*rq_online)(struct rq *rq);
2417 void (*rq_offline)(struct rq *rq);
2418
2419 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2420 #endif
2421
2422 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2423 void (*task_fork)(struct task_struct *p);
2424 void (*task_dead)(struct task_struct *p);
2425
2426 /*
2427 * The switched_from() call is allowed to drop rq->lock, therefore we
2428 * cannot assume the switched_from/switched_to pair is serialized by
2429 * rq->lock. They are however serialized by p->pi_lock.
2430 */
2431 void (*switching_to) (struct rq *this_rq, struct task_struct *task);
2432 void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2433 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
2434 void (*reweight_task)(struct rq *this_rq, struct task_struct *task,
2435 const struct load_weight *lw);
2436 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2437 int oldprio);
2438
2439 unsigned int (*get_rr_interval)(struct rq *rq,
2440 struct task_struct *task);
2441
2442 void (*update_curr)(struct rq *rq);
2443
2444 #ifdef CONFIG_FAIR_GROUP_SCHED
2445 void (*task_change_group)(struct task_struct *p);
2446 #endif
2447
2448 #ifdef CONFIG_SCHED_CORE
2449 int (*task_is_throttled)(struct task_struct *p, int cpu);
2450 #endif
2451 };
2452
put_prev_task(struct rq * rq,struct task_struct * prev)2453 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2454 {
2455 WARN_ON_ONCE(rq->curr != prev);
2456 prev->sched_class->put_prev_task(rq, prev, NULL);
2457 }
2458
set_next_task(struct rq * rq,struct task_struct * next)2459 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2460 {
2461 next->sched_class->set_next_task(rq, next, false);
2462 }
2463
2464 static inline void
__put_prev_set_next_dl_server(struct rq * rq,struct task_struct * prev,struct task_struct * next)2465 __put_prev_set_next_dl_server(struct rq *rq,
2466 struct task_struct *prev,
2467 struct task_struct *next)
2468 {
2469 prev->dl_server = NULL;
2470 next->dl_server = rq->dl_server;
2471 rq->dl_server = NULL;
2472 }
2473
put_prev_set_next_task(struct rq * rq,struct task_struct * prev,struct task_struct * next)2474 static inline void put_prev_set_next_task(struct rq *rq,
2475 struct task_struct *prev,
2476 struct task_struct *next)
2477 {
2478 WARN_ON_ONCE(rq->curr != prev);
2479
2480 __put_prev_set_next_dl_server(rq, prev, next);
2481
2482 if (next == prev)
2483 return;
2484
2485 prev->sched_class->put_prev_task(rq, prev, next);
2486 next->sched_class->set_next_task(rq, next, true);
2487 }
2488
2489 /*
2490 * Helper to define a sched_class instance; each one is placed in a separate
2491 * section which is ordered by the linker script:
2492 *
2493 * include/asm-generic/vmlinux.lds.h
2494 *
2495 * *CAREFUL* they are laid out in *REVERSE* order!!!
2496 *
2497 * Also enforce alignment on the instance, not the type, to guarantee layout.
2498 */
2499 #define DEFINE_SCHED_CLASS(name) \
2500 const struct sched_class name##_sched_class \
2501 __aligned(__alignof__(struct sched_class)) \
2502 __section("__" #name "_sched_class")
2503
2504 /* Defined in include/asm-generic/vmlinux.lds.h */
2505 extern struct sched_class __sched_class_highest[];
2506 extern struct sched_class __sched_class_lowest[];
2507
2508 extern const struct sched_class stop_sched_class;
2509 extern const struct sched_class dl_sched_class;
2510 extern const struct sched_class rt_sched_class;
2511 extern const struct sched_class fair_sched_class;
2512 extern const struct sched_class idle_sched_class;
2513
2514 #ifdef CONFIG_SCHED_CLASS_EXT
2515 extern const struct sched_class ext_sched_class;
2516
2517 DECLARE_STATIC_KEY_FALSE(__scx_ops_enabled); /* SCX BPF scheduler loaded */
2518 DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */
2519
2520 #define scx_enabled() static_branch_unlikely(&__scx_ops_enabled)
2521 #define scx_switched_all() static_branch_unlikely(&__scx_switched_all)
2522 #else /* !CONFIG_SCHED_CLASS_EXT */
2523 #define scx_enabled() false
2524 #define scx_switched_all() false
2525 #endif /* !CONFIG_SCHED_CLASS_EXT */
2526
2527 /*
2528 * Iterate only active classes. SCX can take over all fair tasks or be
2529 * completely disabled. If the former, skip fair. If the latter, skip SCX.
2530 */
next_active_class(const struct sched_class * class)2531 static inline const struct sched_class *next_active_class(const struct sched_class *class)
2532 {
2533 class++;
2534 #ifdef CONFIG_SCHED_CLASS_EXT
2535 if (scx_switched_all() && class == &fair_sched_class)
2536 class++;
2537 if (!scx_enabled() && class == &ext_sched_class)
2538 class++;
2539 #endif
2540 return class;
2541 }
2542
2543 #define for_class_range(class, _from, _to) \
2544 for (class = (_from); class < (_to); class++)
2545
2546 #define for_each_class(class) \
2547 for_class_range(class, __sched_class_highest, __sched_class_lowest)
2548
2549 #define for_active_class_range(class, _from, _to) \
2550 for (class = (_from); class != (_to); class = next_active_class(class))
2551
2552 #define for_each_active_class(class) \
2553 for_active_class_range(class, __sched_class_highest, __sched_class_lowest)
2554
2555 #define sched_class_above(_a, _b) ((_a) < (_b))
2556
sched_stop_runnable(struct rq * rq)2557 static inline bool sched_stop_runnable(struct rq *rq)
2558 {
2559 return rq->stop && task_on_rq_queued(rq->stop);
2560 }
2561
sched_dl_runnable(struct rq * rq)2562 static inline bool sched_dl_runnable(struct rq *rq)
2563 {
2564 return rq->dl.dl_nr_running > 0;
2565 }
2566
sched_rt_runnable(struct rq * rq)2567 static inline bool sched_rt_runnable(struct rq *rq)
2568 {
2569 return rq->rt.rt_queued > 0;
2570 }
2571
sched_fair_runnable(struct rq * rq)2572 static inline bool sched_fair_runnable(struct rq *rq)
2573 {
2574 return rq->cfs.nr_running > 0;
2575 }
2576
2577 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2578 extern struct task_struct *pick_task_idle(struct rq *rq);
2579
2580 #define SCA_CHECK 0x01
2581 #define SCA_MIGRATE_DISABLE 0x02
2582 #define SCA_MIGRATE_ENABLE 0x04
2583 #define SCA_USER 0x08
2584
2585 #ifdef CONFIG_SMP
2586
2587 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2588
2589 extern void sched_balance_trigger(struct rq *rq);
2590
2591 extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx);
2592 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2593
task_allowed_on_cpu(struct task_struct * p,int cpu)2594 static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2595 {
2596 /* When not in the task's cpumask, no point in looking further. */
2597 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2598 return false;
2599
2600 /* Can @cpu run a user thread? */
2601 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p))
2602 return false;
2603
2604 return true;
2605 }
2606
alloc_user_cpus_ptr(int node)2607 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2608 {
2609 /*
2610 * See do_set_cpus_allowed() above for the rcu_head usage.
2611 */
2612 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head));
2613
2614 return kmalloc_node(size, GFP_KERNEL, node);
2615 }
2616
get_push_task(struct rq * rq)2617 static inline struct task_struct *get_push_task(struct rq *rq)
2618 {
2619 struct task_struct *p = rq->curr;
2620
2621 lockdep_assert_rq_held(rq);
2622
2623 if (rq->push_busy)
2624 return NULL;
2625
2626 if (p->nr_cpus_allowed == 1)
2627 return NULL;
2628
2629 if (p->migration_disabled)
2630 return NULL;
2631
2632 rq->push_busy = true;
2633 return get_task_struct(p);
2634 }
2635
2636 extern int push_cpu_stop(void *arg);
2637
2638 #else /* !CONFIG_SMP: */
2639
task_allowed_on_cpu(struct task_struct * p,int cpu)2640 static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu)
2641 {
2642 return true;
2643 }
2644
__set_cpus_allowed_ptr(struct task_struct * p,struct affinity_context * ctx)2645 static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2646 struct affinity_context *ctx)
2647 {
2648 return set_cpus_allowed_ptr(p, ctx->new_mask);
2649 }
2650
alloc_user_cpus_ptr(int node)2651 static inline cpumask_t *alloc_user_cpus_ptr(int node)
2652 {
2653 return NULL;
2654 }
2655
2656 #endif /* !CONFIG_SMP */
2657
2658 #ifdef CONFIG_CPU_IDLE
2659
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2660 static inline void idle_set_state(struct rq *rq,
2661 struct cpuidle_state *idle_state)
2662 {
2663 rq->idle_state = idle_state;
2664 }
2665
idle_get_state(struct rq * rq)2666 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2667 {
2668 SCHED_WARN_ON(!rcu_read_lock_held());
2669
2670 return rq->idle_state;
2671 }
2672
2673 #else /* !CONFIG_CPU_IDLE: */
2674
idle_set_state(struct rq * rq,struct cpuidle_state * idle_state)2675 static inline void idle_set_state(struct rq *rq,
2676 struct cpuidle_state *idle_state)
2677 {
2678 }
2679
idle_get_state(struct rq * rq)2680 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2681 {
2682 return NULL;
2683 }
2684
2685 #endif /* !CONFIG_CPU_IDLE */
2686
2687 extern void schedule_idle(void);
2688 asmlinkage void schedule_user(void);
2689
2690 extern void sysrq_sched_debug_show(void);
2691 extern void sched_init_granularity(void);
2692 extern void update_max_interval(void);
2693
2694 extern void init_sched_dl_class(void);
2695 extern void init_sched_rt_class(void);
2696 extern void init_sched_fair_class(void);
2697
2698 extern void resched_curr(struct rq *rq);
2699 extern void resched_cpu(int cpu);
2700
2701 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2702 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2703
2704 extern void init_dl_entity(struct sched_dl_entity *dl_se);
2705
2706 #define BW_SHIFT 20
2707 #define BW_UNIT (1 << BW_SHIFT)
2708 #define RATIO_SHIFT 8
2709 #define MAX_BW_BITS (64 - BW_SHIFT)
2710 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2711
2712 extern unsigned long to_ratio(u64 period, u64 runtime);
2713
2714 extern void init_entity_runnable_average(struct sched_entity *se);
2715 extern void post_init_entity_util_avg(struct task_struct *p);
2716
2717 #ifdef CONFIG_NO_HZ_FULL
2718 extern bool sched_can_stop_tick(struct rq *rq);
2719 extern int __init sched_tick_offload_init(void);
2720
2721 /*
2722 * Tick may be needed by tasks in the runqueue depending on their policy and
2723 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2724 * nohz mode if necessary.
2725 */
sched_update_tick_dependency(struct rq * rq)2726 static inline void sched_update_tick_dependency(struct rq *rq)
2727 {
2728 int cpu = cpu_of(rq);
2729
2730 if (!tick_nohz_full_cpu(cpu))
2731 return;
2732
2733 if (sched_can_stop_tick(rq))
2734 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2735 else
2736 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2737 }
2738 #else /* !CONFIG_NO_HZ_FULL: */
sched_tick_offload_init(void)2739 static inline int sched_tick_offload_init(void) { return 0; }
sched_update_tick_dependency(struct rq * rq)2740 static inline void sched_update_tick_dependency(struct rq *rq) { }
2741 #endif /* !CONFIG_NO_HZ_FULL */
2742
add_nr_running(struct rq * rq,unsigned count)2743 static inline void add_nr_running(struct rq *rq, unsigned count)
2744 {
2745 unsigned prev_nr = rq->nr_running;
2746
2747 rq->nr_running = prev_nr + count;
2748 if (trace_sched_update_nr_running_tp_enabled()) {
2749 call_trace_sched_update_nr_running(rq, count);
2750 }
2751
2752 #ifdef CONFIG_SMP
2753 if (prev_nr < 2 && rq->nr_running >= 2)
2754 set_rd_overloaded(rq->rd, 1);
2755 #endif
2756
2757 sched_update_tick_dependency(rq);
2758 }
2759
sub_nr_running(struct rq * rq,unsigned count)2760 static inline void sub_nr_running(struct rq *rq, unsigned count)
2761 {
2762 rq->nr_running -= count;
2763 if (trace_sched_update_nr_running_tp_enabled()) {
2764 call_trace_sched_update_nr_running(rq, -count);
2765 }
2766
2767 /* Check if we still need preemption */
2768 sched_update_tick_dependency(rq);
2769 }
2770
__block_task(struct rq * rq,struct task_struct * p)2771 static inline void __block_task(struct rq *rq, struct task_struct *p)
2772 {
2773 if (p->sched_contributes_to_load)
2774 rq->nr_uninterruptible++;
2775
2776 if (p->in_iowait) {
2777 atomic_inc(&rq->nr_iowait);
2778 delayacct_blkio_start();
2779 }
2780
2781 ASSERT_EXCLUSIVE_WRITER(p->on_rq);
2782
2783 /*
2784 * The moment this write goes through, ttwu() can swoop in and migrate
2785 * this task, rendering our rq->__lock ineffective.
2786 *
2787 * __schedule() try_to_wake_up()
2788 * LOCK rq->__lock LOCK p->pi_lock
2789 * pick_next_task()
2790 * pick_next_task_fair()
2791 * pick_next_entity()
2792 * dequeue_entities()
2793 * __block_task()
2794 * RELEASE p->on_rq = 0 if (p->on_rq && ...)
2795 * break;
2796 *
2797 * ACQUIRE (after ctrl-dep)
2798 *
2799 * cpu = select_task_rq();
2800 * set_task_cpu(p, cpu);
2801 * ttwu_queue()
2802 * ttwu_do_activate()
2803 * LOCK rq->__lock
2804 * activate_task()
2805 * STORE p->on_rq = 1
2806 * UNLOCK rq->__lock
2807 *
2808 * Callers must ensure to not reference @p after this -- we no longer
2809 * own it.
2810 */
2811 smp_store_release(&p->on_rq, 0);
2812 }
2813
2814 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2815 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2816
2817 extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2818
2819 #ifdef CONFIG_PREEMPT_RT
2820 # define SCHED_NR_MIGRATE_BREAK 8
2821 #else
2822 # define SCHED_NR_MIGRATE_BREAK 32
2823 #endif
2824
2825 extern const_debug unsigned int sysctl_sched_nr_migrate;
2826 extern const_debug unsigned int sysctl_sched_migration_cost;
2827
2828 extern unsigned int sysctl_sched_base_slice;
2829
2830 #ifdef CONFIG_SCHED_DEBUG
2831 extern int sysctl_resched_latency_warn_ms;
2832 extern int sysctl_resched_latency_warn_once;
2833
2834 extern unsigned int sysctl_sched_tunable_scaling;
2835
2836 extern unsigned int sysctl_numa_balancing_scan_delay;
2837 extern unsigned int sysctl_numa_balancing_scan_period_min;
2838 extern unsigned int sysctl_numa_balancing_scan_period_max;
2839 extern unsigned int sysctl_numa_balancing_scan_size;
2840 extern unsigned int sysctl_numa_balancing_hot_threshold;
2841 #endif
2842
2843 #ifdef CONFIG_SCHED_HRTICK
2844
2845 /*
2846 * Use hrtick when:
2847 * - enabled by features
2848 * - hrtimer is actually high res
2849 */
hrtick_enabled(struct rq * rq)2850 static inline int hrtick_enabled(struct rq *rq)
2851 {
2852 if (!cpu_active(cpu_of(rq)))
2853 return 0;
2854 return hrtimer_is_hres_active(&rq->hrtick_timer);
2855 }
2856
hrtick_enabled_fair(struct rq * rq)2857 static inline int hrtick_enabled_fair(struct rq *rq)
2858 {
2859 if (!sched_feat(HRTICK))
2860 return 0;
2861 return hrtick_enabled(rq);
2862 }
2863
hrtick_enabled_dl(struct rq * rq)2864 static inline int hrtick_enabled_dl(struct rq *rq)
2865 {
2866 if (!sched_feat(HRTICK_DL))
2867 return 0;
2868 return hrtick_enabled(rq);
2869 }
2870
2871 extern void hrtick_start(struct rq *rq, u64 delay);
2872
2873 #else /* !CONFIG_SCHED_HRTICK: */
2874
hrtick_enabled_fair(struct rq * rq)2875 static inline int hrtick_enabled_fair(struct rq *rq)
2876 {
2877 return 0;
2878 }
2879
hrtick_enabled_dl(struct rq * rq)2880 static inline int hrtick_enabled_dl(struct rq *rq)
2881 {
2882 return 0;
2883 }
2884
hrtick_enabled(struct rq * rq)2885 static inline int hrtick_enabled(struct rq *rq)
2886 {
2887 return 0;
2888 }
2889
2890 #endif /* !CONFIG_SCHED_HRTICK */
2891
2892 #ifndef arch_scale_freq_tick
arch_scale_freq_tick(void)2893 static __always_inline void arch_scale_freq_tick(void) { }
2894 #endif
2895
2896 #ifndef arch_scale_freq_capacity
2897 /**
2898 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2899 * @cpu: the CPU in question.
2900 *
2901 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2902 *
2903 * f_curr
2904 * ------ * SCHED_CAPACITY_SCALE
2905 * f_max
2906 */
2907 static __always_inline
arch_scale_freq_capacity(int cpu)2908 unsigned long arch_scale_freq_capacity(int cpu)
2909 {
2910 return SCHED_CAPACITY_SCALE;
2911 }
2912 #endif
2913
2914 #ifdef CONFIG_SCHED_DEBUG
2915 /*
2916 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2917 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2918 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2919 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2920 */
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2921 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2922 {
2923 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2924 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2925 #ifdef CONFIG_SMP
2926 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2927 #endif
2928 }
2929 #else
double_rq_clock_clear_update(struct rq * rq1,struct rq * rq2)2930 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) { }
2931 #endif
2932
2933 #define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \
2934 __DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2935 static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2936 { class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \
2937 _lock; return _t; }
2938
2939 #ifdef CONFIG_SMP
2940
rq_order_less(struct rq * rq1,struct rq * rq2)2941 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2942 {
2943 #ifdef CONFIG_SCHED_CORE
2944 /*
2945 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2946 * order by core-id first and cpu-id second.
2947 *
2948 * Notably:
2949 *
2950 * double_rq_lock(0,3); will take core-0, core-1 lock
2951 * double_rq_lock(1,2); will take core-1, core-0 lock
2952 *
2953 * when only cpu-id is considered.
2954 */
2955 if (rq1->core->cpu < rq2->core->cpu)
2956 return true;
2957 if (rq1->core->cpu > rq2->core->cpu)
2958 return false;
2959
2960 /*
2961 * __sched_core_flip() relies on SMT having cpu-id lock order.
2962 */
2963 #endif
2964 return rq1->cpu < rq2->cpu;
2965 }
2966
2967 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2968
2969 #ifdef CONFIG_PREEMPTION
2970
2971 /*
2972 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2973 * way at the expense of forcing extra atomic operations in all
2974 * invocations. This assures that the double_lock is acquired using the
2975 * same underlying policy as the spinlock_t on this architecture, which
2976 * reduces latency compared to the unfair variant below. However, it
2977 * also adds more overhead and therefore may reduce throughput.
2978 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2979 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2980 __releases(this_rq->lock)
2981 __acquires(busiest->lock)
2982 __acquires(this_rq->lock)
2983 {
2984 raw_spin_rq_unlock(this_rq);
2985 double_rq_lock(this_rq, busiest);
2986
2987 return 1;
2988 }
2989
2990 #else /* !CONFIG_PREEMPTION: */
2991 /*
2992 * Unfair double_lock_balance: Optimizes throughput at the expense of
2993 * latency by eliminating extra atomic operations when the locks are
2994 * already in proper order on entry. This favors lower CPU-ids and will
2995 * grant the double lock to lower CPUs over higher ids under contention,
2996 * regardless of entry order into the function.
2997 */
_double_lock_balance(struct rq * this_rq,struct rq * busiest)2998 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2999 __releases(this_rq->lock)
3000 __acquires(busiest->lock)
3001 __acquires(this_rq->lock)
3002 {
3003 if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
3004 likely(raw_spin_rq_trylock(busiest))) {
3005 double_rq_clock_clear_update(this_rq, busiest);
3006 return 0;
3007 }
3008
3009 if (rq_order_less(this_rq, busiest)) {
3010 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
3011 double_rq_clock_clear_update(this_rq, busiest);
3012 return 0;
3013 }
3014
3015 raw_spin_rq_unlock(this_rq);
3016 double_rq_lock(this_rq, busiest);
3017
3018 return 1;
3019 }
3020
3021 #endif /* !CONFIG_PREEMPTION */
3022
3023 /*
3024 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
3025 */
double_lock_balance(struct rq * this_rq,struct rq * busiest)3026 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
3027 {
3028 lockdep_assert_irqs_disabled();
3029
3030 return _double_lock_balance(this_rq, busiest);
3031 }
3032
double_unlock_balance(struct rq * this_rq,struct rq * busiest)3033 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
3034 __releases(busiest->lock)
3035 {
3036 if (__rq_lockp(this_rq) != __rq_lockp(busiest))
3037 raw_spin_rq_unlock(busiest);
3038 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
3039 }
3040
double_lock(spinlock_t * l1,spinlock_t * l2)3041 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
3042 {
3043 if (l1 > l2)
3044 swap(l1, l2);
3045
3046 spin_lock(l1);
3047 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3048 }
3049
double_lock_irq(spinlock_t * l1,spinlock_t * l2)3050 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
3051 {
3052 if (l1 > l2)
3053 swap(l1, l2);
3054
3055 spin_lock_irq(l1);
3056 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3057 }
3058
double_raw_lock(raw_spinlock_t * l1,raw_spinlock_t * l2)3059 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3060 {
3061 if (l1 > l2)
3062 swap(l1, l2);
3063
3064 raw_spin_lock(l1);
3065 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
3066 }
3067
double_raw_unlock(raw_spinlock_t * l1,raw_spinlock_t * l2)3068 static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
3069 {
3070 raw_spin_unlock(l1);
3071 raw_spin_unlock(l2);
3072 }
3073
3074 DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
3075 double_raw_lock(_T->lock, _T->lock2),
3076 double_raw_unlock(_T->lock, _T->lock2))
3077
3078 /*
3079 * double_rq_unlock - safely unlock two runqueues
3080 *
3081 * Note this does not restore interrupts like task_rq_unlock,
3082 * you need to do so manually after calling.
3083 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)3084 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3085 __releases(rq1->lock)
3086 __releases(rq2->lock)
3087 {
3088 if (__rq_lockp(rq1) != __rq_lockp(rq2))
3089 raw_spin_rq_unlock(rq2);
3090 else
3091 __release(rq2->lock);
3092 raw_spin_rq_unlock(rq1);
3093 }
3094
3095 extern void set_rq_online (struct rq *rq);
3096 extern void set_rq_offline(struct rq *rq);
3097
3098 extern bool sched_smp_initialized;
3099
3100 #else /* !CONFIG_SMP: */
3101
3102 /*
3103 * double_rq_lock - safely lock two runqueues
3104 *
3105 * Note this does not disable interrupts like task_rq_lock,
3106 * you need to do so manually before calling.
3107 */
double_rq_lock(struct rq * rq1,struct rq * rq2)3108 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
3109 __acquires(rq1->lock)
3110 __acquires(rq2->lock)
3111 {
3112 WARN_ON_ONCE(!irqs_disabled());
3113 WARN_ON_ONCE(rq1 != rq2);
3114 raw_spin_rq_lock(rq1);
3115 __acquire(rq2->lock); /* Fake it out ;) */
3116 double_rq_clock_clear_update(rq1, rq2);
3117 }
3118
3119 /*
3120 * double_rq_unlock - safely unlock two runqueues
3121 *
3122 * Note this does not restore interrupts like task_rq_unlock,
3123 * you need to do so manually after calling.
3124 */
double_rq_unlock(struct rq * rq1,struct rq * rq2)3125 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
3126 __releases(rq1->lock)
3127 __releases(rq2->lock)
3128 {
3129 WARN_ON_ONCE(rq1 != rq2);
3130 raw_spin_rq_unlock(rq1);
3131 __release(rq2->lock);
3132 }
3133
3134 #endif /* !CONFIG_SMP */
3135
3136 DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
3137 double_rq_lock(_T->lock, _T->lock2),
3138 double_rq_unlock(_T->lock, _T->lock2))
3139
3140 extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
3141 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
3142 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
3143
3144 #ifdef CONFIG_SCHED_DEBUG
3145 extern bool sched_debug_verbose;
3146
3147 extern void print_cfs_stats(struct seq_file *m, int cpu);
3148 extern void print_rt_stats(struct seq_file *m, int cpu);
3149 extern void print_dl_stats(struct seq_file *m, int cpu);
3150 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
3151 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
3152 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
3153
3154 extern void resched_latency_warn(int cpu, u64 latency);
3155 # ifdef CONFIG_NUMA_BALANCING
3156 extern void show_numa_stats(struct task_struct *p, struct seq_file *m);
3157 extern void
3158 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
3159 unsigned long tpf, unsigned long gsf, unsigned long gpf);
3160 # endif /* CONFIG_NUMA_BALANCING */
3161 #else /* !CONFIG_SCHED_DEBUG: */
resched_latency_warn(int cpu,u64 latency)3162 static inline void resched_latency_warn(int cpu, u64 latency) { }
3163 #endif /* !CONFIG_SCHED_DEBUG */
3164
3165 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
3166 extern void init_rt_rq(struct rt_rq *rt_rq);
3167 extern void init_dl_rq(struct dl_rq *dl_rq);
3168
3169 extern void cfs_bandwidth_usage_inc(void);
3170 extern void cfs_bandwidth_usage_dec(void);
3171
3172 #ifdef CONFIG_NO_HZ_COMMON
3173
3174 #define NOHZ_BALANCE_KICK_BIT 0
3175 #define NOHZ_STATS_KICK_BIT 1
3176 #define NOHZ_NEWILB_KICK_BIT 2
3177 #define NOHZ_NEXT_KICK_BIT 3
3178
3179 /* Run sched_balance_domains() */
3180 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
3181 /* Update blocked load */
3182 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
3183 /* Update blocked load when entering idle */
3184 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
3185 /* Update nohz.next_balance */
3186 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
3187
3188 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
3189
3190 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
3191
3192 extern void nohz_balance_exit_idle(struct rq *rq);
3193 #else /* !CONFIG_NO_HZ_COMMON: */
nohz_balance_exit_idle(struct rq * rq)3194 static inline void nohz_balance_exit_idle(struct rq *rq) { }
3195 #endif /* !CONFIG_NO_HZ_COMMON */
3196
3197 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
3198 extern void nohz_run_idle_balance(int cpu);
3199 #else
nohz_run_idle_balance(int cpu)3200 static inline void nohz_run_idle_balance(int cpu) { }
3201 #endif
3202
3203 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
3204
3205 struct irqtime {
3206 u64 total;
3207 u64 tick_delta;
3208 u64 irq_start_time;
3209 struct u64_stats_sync sync;
3210 };
3211
3212 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
3213
3214 /*
3215 * Returns the irqtime minus the softirq time computed by ksoftirqd.
3216 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
3217 * and never move forward.
3218 */
irq_time_read(int cpu)3219 static inline u64 irq_time_read(int cpu)
3220 {
3221 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
3222 unsigned int seq;
3223 u64 total;
3224
3225 do {
3226 seq = __u64_stats_fetch_begin(&irqtime->sync);
3227 total = irqtime->total;
3228 } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
3229
3230 return total;
3231 }
3232
3233 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3234
3235 #ifdef CONFIG_CPU_FREQ
3236
3237 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
3238
3239 /**
3240 * cpufreq_update_util - Take a note about CPU utilization changes.
3241 * @rq: Runqueue to carry out the update for.
3242 * @flags: Update reason flags.
3243 *
3244 * This function is called by the scheduler on the CPU whose utilization is
3245 * being updated.
3246 *
3247 * It can only be called from RCU-sched read-side critical sections.
3248 *
3249 * The way cpufreq is currently arranged requires it to evaluate the CPU
3250 * performance state (frequency/voltage) on a regular basis to prevent it from
3251 * being stuck in a completely inadequate performance level for too long.
3252 * That is not guaranteed to happen if the updates are only triggered from CFS
3253 * and DL, though, because they may not be coming in if only RT tasks are
3254 * active all the time (or there are RT tasks only).
3255 *
3256 * As a workaround for that issue, this function is called periodically by the
3257 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
3258 * but that really is a band-aid. Going forward it should be replaced with
3259 * solutions targeted more specifically at RT tasks.
3260 */
cpufreq_update_util(struct rq * rq,unsigned int flags)3261 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
3262 {
3263 struct update_util_data *data;
3264
3265 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
3266 cpu_of(rq)));
3267 if (data)
3268 data->func(data, rq_clock(rq), flags);
3269 }
3270 #else /* !CONFIG_CPU_FREQ: */
cpufreq_update_util(struct rq * rq,unsigned int flags)3271 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { }
3272 #endif /* !CONFIG_CPU_FREQ */
3273
3274 #ifdef arch_scale_freq_capacity
3275 # ifndef arch_scale_freq_invariant
3276 # define arch_scale_freq_invariant() true
3277 # endif
3278 #else
3279 # define arch_scale_freq_invariant() false
3280 #endif
3281
3282 #ifdef CONFIG_SMP
3283
3284 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3285 unsigned long *min,
3286 unsigned long *max);
3287
3288 unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3289 unsigned long min,
3290 unsigned long max);
3291
3292
3293 /*
3294 * Verify the fitness of task @p to run on @cpu taking into account the
3295 * CPU original capacity and the runtime/deadline ratio of the task.
3296 *
3297 * The function will return true if the original capacity of @cpu is
3298 * greater than or equal to task's deadline density right shifted by
3299 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3300 */
dl_task_fits_capacity(struct task_struct * p,int cpu)3301 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3302 {
3303 unsigned long cap = arch_scale_cpu_capacity(cpu);
3304
3305 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3306 }
3307
cpu_bw_dl(struct rq * rq)3308 static inline unsigned long cpu_bw_dl(struct rq *rq)
3309 {
3310 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3311 }
3312
cpu_util_dl(struct rq * rq)3313 static inline unsigned long cpu_util_dl(struct rq *rq)
3314 {
3315 return READ_ONCE(rq->avg_dl.util_avg);
3316 }
3317
3318
3319 extern unsigned long cpu_util_cfs(int cpu);
3320 extern unsigned long cpu_util_cfs_boost(int cpu);
3321
cpu_util_rt(struct rq * rq)3322 static inline unsigned long cpu_util_rt(struct rq *rq)
3323 {
3324 return READ_ONCE(rq->avg_rt.util_avg);
3325 }
3326
3327 #else /* !CONFIG_SMP */
update_other_load_avgs(struct rq * rq)3328 static inline bool update_other_load_avgs(struct rq *rq) { return false; }
3329 #endif /* CONFIG_SMP */
3330
3331 #ifdef CONFIG_UCLAMP_TASK
3332
3333 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3334
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3335 static inline unsigned long uclamp_rq_get(struct rq *rq,
3336 enum uclamp_id clamp_id)
3337 {
3338 return READ_ONCE(rq->uclamp[clamp_id].value);
3339 }
3340
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3341 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3342 unsigned int value)
3343 {
3344 WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3345 }
3346
uclamp_rq_is_idle(struct rq * rq)3347 static inline bool uclamp_rq_is_idle(struct rq *rq)
3348 {
3349 return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3350 }
3351
3352 /* Is the rq being capped/throttled by uclamp_max? */
uclamp_rq_is_capped(struct rq * rq)3353 static inline bool uclamp_rq_is_capped(struct rq *rq)
3354 {
3355 unsigned long rq_util;
3356 unsigned long max_util;
3357
3358 if (!static_branch_likely(&sched_uclamp_used))
3359 return false;
3360
3361 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3362 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3363
3364 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3365 }
3366
3367 /*
3368 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3369 * by default in the fast path and only gets turned on once userspace performs
3370 * an operation that requires it.
3371 *
3372 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3373 * hence is active.
3374 */
uclamp_is_used(void)3375 static inline bool uclamp_is_used(void)
3376 {
3377 return static_branch_likely(&sched_uclamp_used);
3378 }
3379
3380 #define for_each_clamp_id(clamp_id) \
3381 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
3382
3383 extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
3384
3385
uclamp_none(enum uclamp_id clamp_id)3386 static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
3387 {
3388 if (clamp_id == UCLAMP_MIN)
3389 return 0;
3390 return SCHED_CAPACITY_SCALE;
3391 }
3392
3393 /* Integer rounded range for each bucket */
3394 #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
3395
uclamp_bucket_id(unsigned int clamp_value)3396 static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
3397 {
3398 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1);
3399 }
3400
3401 static inline void
uclamp_se_set(struct uclamp_se * uc_se,unsigned int value,bool user_defined)3402 uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined)
3403 {
3404 uc_se->value = value;
3405 uc_se->bucket_id = uclamp_bucket_id(value);
3406 uc_se->user_defined = user_defined;
3407 }
3408
3409 #else /* !CONFIG_UCLAMP_TASK: */
3410
3411 static inline unsigned long
uclamp_eff_value(struct task_struct * p,enum uclamp_id clamp_id)3412 uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
3413 {
3414 if (clamp_id == UCLAMP_MIN)
3415 return 0;
3416
3417 return SCHED_CAPACITY_SCALE;
3418 }
3419
uclamp_rq_is_capped(struct rq * rq)3420 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3421
uclamp_is_used(void)3422 static inline bool uclamp_is_used(void)
3423 {
3424 return false;
3425 }
3426
3427 static inline unsigned long
uclamp_rq_get(struct rq * rq,enum uclamp_id clamp_id)3428 uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id)
3429 {
3430 if (clamp_id == UCLAMP_MIN)
3431 return 0;
3432
3433 return SCHED_CAPACITY_SCALE;
3434 }
3435
3436 static inline void
uclamp_rq_set(struct rq * rq,enum uclamp_id clamp_id,unsigned int value)3437 uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value)
3438 {
3439 }
3440
uclamp_rq_is_idle(struct rq * rq)3441 static inline bool uclamp_rq_is_idle(struct rq *rq)
3442 {
3443 return false;
3444 }
3445
3446 #endif /* !CONFIG_UCLAMP_TASK */
3447
3448 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3449
cpu_util_irq(struct rq * rq)3450 static inline unsigned long cpu_util_irq(struct rq *rq)
3451 {
3452 return READ_ONCE(rq->avg_irq.util_avg);
3453 }
3454
3455 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3456 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3457 {
3458 util *= (max - irq);
3459 util /= max;
3460
3461 return util;
3462
3463 }
3464
3465 #else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */
3466
cpu_util_irq(struct rq * rq)3467 static inline unsigned long cpu_util_irq(struct rq *rq)
3468 {
3469 return 0;
3470 }
3471
3472 static inline
scale_irq_capacity(unsigned long util,unsigned long irq,unsigned long max)3473 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3474 {
3475 return util;
3476 }
3477
3478 #endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */
3479
3480 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3481
3482 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3483
3484 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3485
sched_energy_enabled(void)3486 static inline bool sched_energy_enabled(void)
3487 {
3488 return static_branch_unlikely(&sched_energy_present);
3489 }
3490
3491 extern struct cpufreq_governor schedutil_gov;
3492
3493 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3494
3495 #define perf_domain_span(pd) NULL
3496
sched_energy_enabled(void)3497 static inline bool sched_energy_enabled(void) { return false; }
3498
3499 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3500
3501 #ifdef CONFIG_MEMBARRIER
3502
3503 /*
3504 * The scheduler provides memory barriers required by membarrier between:
3505 * - prior user-space memory accesses and store to rq->membarrier_state,
3506 * - store to rq->membarrier_state and following user-space memory accesses.
3507 * In the same way it provides those guarantees around store to rq->curr.
3508 */
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3509 static inline void membarrier_switch_mm(struct rq *rq,
3510 struct mm_struct *prev_mm,
3511 struct mm_struct *next_mm)
3512 {
3513 int membarrier_state;
3514
3515 if (prev_mm == next_mm)
3516 return;
3517
3518 membarrier_state = atomic_read(&next_mm->membarrier_state);
3519 if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3520 return;
3521
3522 WRITE_ONCE(rq->membarrier_state, membarrier_state);
3523 }
3524
3525 #else /* !CONFIG_MEMBARRIER :*/
3526
membarrier_switch_mm(struct rq * rq,struct mm_struct * prev_mm,struct mm_struct * next_mm)3527 static inline void membarrier_switch_mm(struct rq *rq,
3528 struct mm_struct *prev_mm,
3529 struct mm_struct *next_mm)
3530 {
3531 }
3532
3533 #endif /* !CONFIG_MEMBARRIER */
3534
3535 #ifdef CONFIG_SMP
is_per_cpu_kthread(struct task_struct * p)3536 static inline bool is_per_cpu_kthread(struct task_struct *p)
3537 {
3538 if (!(p->flags & PF_KTHREAD))
3539 return false;
3540
3541 if (p->nr_cpus_allowed != 1)
3542 return false;
3543
3544 return true;
3545 }
3546 #endif
3547
3548 extern void swake_up_all_locked(struct swait_queue_head *q);
3549 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3550
3551 extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3552
3553 #ifdef CONFIG_PREEMPT_DYNAMIC
3554 extern int preempt_dynamic_mode;
3555 extern int sched_dynamic_mode(const char *str);
3556 extern void sched_dynamic_update(int mode);
3557 #endif
3558
3559 #ifdef CONFIG_SCHED_MM_CID
3560
3561 #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */
3562 #define MM_CID_SCAN_DELAY 100 /* 100ms */
3563
3564 extern raw_spinlock_t cid_lock;
3565 extern int use_cid_lock;
3566
3567 extern void sched_mm_cid_migrate_from(struct task_struct *t);
3568 extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3569 extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3570 extern void init_sched_mm_cid(struct task_struct *t);
3571
__mm_cid_put(struct mm_struct * mm,int cid)3572 static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3573 {
3574 if (cid < 0)
3575 return;
3576 cpumask_clear_cpu(cid, mm_cidmask(mm));
3577 }
3578
3579 /*
3580 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3581 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3582 * be held to transition to other states.
3583 *
3584 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3585 * consistent across CPUs, which prevents use of this_cpu_cmpxchg.
3586 */
mm_cid_put_lazy(struct task_struct * t)3587 static inline void mm_cid_put_lazy(struct task_struct *t)
3588 {
3589 struct mm_struct *mm = t->mm;
3590 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3591 int cid;
3592
3593 lockdep_assert_irqs_disabled();
3594 cid = __this_cpu_read(pcpu_cid->cid);
3595 if (!mm_cid_is_lazy_put(cid) ||
3596 !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3597 return;
3598 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3599 }
3600
mm_cid_pcpu_unset(struct mm_struct * mm)3601 static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3602 {
3603 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3604 int cid, res;
3605
3606 lockdep_assert_irqs_disabled();
3607 cid = __this_cpu_read(pcpu_cid->cid);
3608 for (;;) {
3609 if (mm_cid_is_unset(cid))
3610 return MM_CID_UNSET;
3611 /*
3612 * Attempt transition from valid or lazy-put to unset.
3613 */
3614 res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3615 if (res == cid)
3616 break;
3617 cid = res;
3618 }
3619 return cid;
3620 }
3621
mm_cid_put(struct mm_struct * mm)3622 static inline void mm_cid_put(struct mm_struct *mm)
3623 {
3624 int cid;
3625
3626 lockdep_assert_irqs_disabled();
3627 cid = mm_cid_pcpu_unset(mm);
3628 if (cid == MM_CID_UNSET)
3629 return;
3630 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3631 }
3632
__mm_cid_try_get(struct mm_struct * mm)3633 static inline int __mm_cid_try_get(struct mm_struct *mm)
3634 {
3635 struct cpumask *cpumask;
3636 int cid;
3637
3638 cpumask = mm_cidmask(mm);
3639 /*
3640 * Retry finding first zero bit if the mask is temporarily
3641 * filled. This only happens during concurrent remote-clear
3642 * which owns a cid without holding a rq lock.
3643 */
3644 for (;;) {
3645 cid = cpumask_first_zero(cpumask);
3646 if (cid < nr_cpu_ids)
3647 break;
3648 cpu_relax();
3649 }
3650 if (cpumask_test_and_set_cpu(cid, cpumask))
3651 return -1;
3652
3653 return cid;
3654 }
3655
3656 /*
3657 * Save a snapshot of the current runqueue time of this cpu
3658 * with the per-cpu cid value, allowing to estimate how recently it was used.
3659 */
mm_cid_snapshot_time(struct rq * rq,struct mm_struct * mm)3660 static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3661 {
3662 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3663
3664 lockdep_assert_rq_held(rq);
3665 WRITE_ONCE(pcpu_cid->time, rq->clock);
3666 }
3667
__mm_cid_get(struct rq * rq,struct mm_struct * mm)3668 static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3669 {
3670 int cid;
3671
3672 /*
3673 * All allocations (even those using the cid_lock) are lock-free. If
3674 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3675 * guarantee forward progress.
3676 */
3677 if (!READ_ONCE(use_cid_lock)) {
3678 cid = __mm_cid_try_get(mm);
3679 if (cid >= 0)
3680 goto end;
3681 raw_spin_lock(&cid_lock);
3682 } else {
3683 raw_spin_lock(&cid_lock);
3684 cid = __mm_cid_try_get(mm);
3685 if (cid >= 0)
3686 goto unlock;
3687 }
3688
3689 /*
3690 * cid concurrently allocated. Retry while forcing following
3691 * allocations to use the cid_lock to ensure forward progress.
3692 */
3693 WRITE_ONCE(use_cid_lock, 1);
3694 /*
3695 * Set use_cid_lock before allocation. Only care about program order
3696 * because this is only required for forward progress.
3697 */
3698 barrier();
3699 /*
3700 * Retry until it succeeds. It is guaranteed to eventually succeed once
3701 * all newcoming allocations observe the use_cid_lock flag set.
3702 */
3703 do {
3704 cid = __mm_cid_try_get(mm);
3705 cpu_relax();
3706 } while (cid < 0);
3707 /*
3708 * Allocate before clearing use_cid_lock. Only care about
3709 * program order because this is for forward progress.
3710 */
3711 barrier();
3712 WRITE_ONCE(use_cid_lock, 0);
3713 unlock:
3714 raw_spin_unlock(&cid_lock);
3715 end:
3716 mm_cid_snapshot_time(rq, mm);
3717
3718 return cid;
3719 }
3720
mm_cid_get(struct rq * rq,struct mm_struct * mm)3721 static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3722 {
3723 struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3724 struct cpumask *cpumask;
3725 int cid;
3726
3727 lockdep_assert_rq_held(rq);
3728 cpumask = mm_cidmask(mm);
3729 cid = __this_cpu_read(pcpu_cid->cid);
3730 if (mm_cid_is_valid(cid)) {
3731 mm_cid_snapshot_time(rq, mm);
3732 return cid;
3733 }
3734 if (mm_cid_is_lazy_put(cid)) {
3735 if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3736 __mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3737 }
3738 cid = __mm_cid_get(rq, mm);
3739 __this_cpu_write(pcpu_cid->cid, cid);
3740
3741 return cid;
3742 }
3743
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3744 static inline void switch_mm_cid(struct rq *rq,
3745 struct task_struct *prev,
3746 struct task_struct *next)
3747 {
3748 /*
3749 * Provide a memory barrier between rq->curr store and load of
3750 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3751 *
3752 * Should be adapted if context_switch() is modified.
3753 */
3754 if (!next->mm) { // to kernel
3755 /*
3756 * user -> kernel transition does not guarantee a barrier, but
3757 * we can use the fact that it performs an atomic operation in
3758 * mmgrab().
3759 */
3760 if (prev->mm) // from user
3761 smp_mb__after_mmgrab();
3762 /*
3763 * kernel -> kernel transition does not change rq->curr->mm
3764 * state. It stays NULL.
3765 */
3766 } else { // to user
3767 /*
3768 * kernel -> user transition does not provide a barrier
3769 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3770 * Provide it here.
3771 */
3772 if (!prev->mm) { // from kernel
3773 smp_mb();
3774 } else { // from user
3775 /*
3776 * user->user transition relies on an implicit
3777 * memory barrier in switch_mm() when
3778 * current->mm changes. If the architecture
3779 * switch_mm() does not have an implicit memory
3780 * barrier, it is emitted here. If current->mm
3781 * is unchanged, no barrier is needed.
3782 */
3783 smp_mb__after_switch_mm();
3784 }
3785 }
3786 if (prev->mm_cid_active) {
3787 mm_cid_snapshot_time(rq, prev->mm);
3788 mm_cid_put_lazy(prev);
3789 prev->mm_cid = -1;
3790 }
3791 if (next->mm_cid_active)
3792 next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3793 }
3794
3795 #else /* !CONFIG_SCHED_MM_CID: */
switch_mm_cid(struct rq * rq,struct task_struct * prev,struct task_struct * next)3796 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)3797 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)3798 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)3799 static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
init_sched_mm_cid(struct task_struct * t)3800 static inline void init_sched_mm_cid(struct task_struct *t) { }
3801 #endif /* !CONFIG_SCHED_MM_CID */
3802
3803 extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3804 extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3805
3806 #ifdef CONFIG_RT_MUTEXES
3807
__rt_effective_prio(struct task_struct * pi_task,int prio)3808 static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3809 {
3810 if (pi_task)
3811 prio = min(prio, pi_task->prio);
3812
3813 return prio;
3814 }
3815
rt_effective_prio(struct task_struct * p,int prio)3816 static inline int rt_effective_prio(struct task_struct *p, int prio)
3817 {
3818 struct task_struct *pi_task = rt_mutex_get_top_task(p);
3819
3820 return __rt_effective_prio(pi_task, prio);
3821 }
3822
3823 #else /* !CONFIG_RT_MUTEXES: */
3824
rt_effective_prio(struct task_struct * p,int prio)3825 static inline int rt_effective_prio(struct task_struct *p, int prio)
3826 {
3827 return prio;
3828 }
3829
3830 #endif /* !CONFIG_RT_MUTEXES */
3831
3832 extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi);
3833 extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx);
3834 extern const struct sched_class *__setscheduler_class(int policy, int prio);
3835 extern void set_load_weight(struct task_struct *p, bool update_load);
3836 extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
3837 extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags);
3838
3839 extern void check_class_changing(struct rq *rq, struct task_struct *p,
3840 const struct sched_class *prev_class);
3841 extern void check_class_changed(struct rq *rq, struct task_struct *p,
3842 const struct sched_class *prev_class,
3843 int oldprio);
3844
3845 #ifdef CONFIG_SMP
3846 extern struct balance_callback *splice_balance_callbacks(struct rq *rq);
3847 extern void balance_callbacks(struct rq *rq, struct balance_callback *head);
3848 #else
3849
splice_balance_callbacks(struct rq * rq)3850 static inline struct balance_callback *splice_balance_callbacks(struct rq *rq)
3851 {
3852 return NULL;
3853 }
3854
balance_callbacks(struct rq * rq,struct balance_callback * head)3855 static inline void balance_callbacks(struct rq *rq, struct balance_callback *head)
3856 {
3857 }
3858
3859 #endif
3860
3861 #ifdef CONFIG_SCHED_CLASS_EXT
3862 /*
3863 * Used by SCX in the enable/disable paths to move tasks between sched_classes
3864 * and establish invariants.
3865 */
3866 struct sched_enq_and_set_ctx {
3867 struct task_struct *p;
3868 int queue_flags;
3869 bool queued;
3870 bool running;
3871 };
3872
3873 void sched_deq_and_put_task(struct task_struct *p, int queue_flags,
3874 struct sched_enq_and_set_ctx *ctx);
3875 void sched_enq_and_set_task(struct sched_enq_and_set_ctx *ctx);
3876
3877 #endif /* CONFIG_SCHED_CLASS_EXT */
3878
3879 #include "ext.h"
3880
3881 #endif /* _KERNEL_SCHED_SCHED_H */
3882