Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Deadline Scheduling Class (SCHED_DEADLINE)
4 : *
5 : * Earliest Deadline First (EDF) + Constant Bandwidth Server (CBS).
6 : *
7 : * Tasks that periodically executes their instances for less than their
8 : * runtime won't miss any of their deadlines.
9 : * Tasks that are not periodic or sporadic or that tries to execute more
10 : * than their reserved bandwidth will be slowed down (and may potentially
11 : * miss some of their deadlines), and won't affect any other task.
12 : *
13 : * Copyright (C) 2012 Dario Faggioli <raistlin@linux.it>,
14 : * Juri Lelli <juri.lelli@gmail.com>,
15 : * Michael Trimarchi <michael@amarulasolutions.com>,
16 : * Fabio Checconi <fchecconi@gmail.com>
17 : */
18 :
19 : static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
20 : {
21 0 : return container_of(dl_se, struct task_struct, dl);
22 : }
23 :
24 : static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq)
25 : {
26 0 : return container_of(dl_rq, struct rq, dl);
27 : }
28 :
29 : static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
30 : {
31 0 : struct task_struct *p = dl_task_of(dl_se);
32 0 : struct rq *rq = task_rq(p);
33 :
34 : return &rq->dl;
35 : }
36 :
37 : static inline int on_dl_rq(struct sched_dl_entity *dl_se)
38 : {
39 0 : return !RB_EMPTY_NODE(&dl_se->rb_node);
40 : }
41 :
42 : #ifdef CONFIG_RT_MUTEXES
43 : static inline struct sched_dl_entity *pi_of(struct sched_dl_entity *dl_se)
44 : {
45 : return dl_se->pi_se;
46 : }
47 :
48 : static inline bool is_dl_boosted(struct sched_dl_entity *dl_se)
49 : {
50 0 : return pi_of(dl_se) != dl_se;
51 : }
52 : #else
53 : static inline struct sched_dl_entity *pi_of(struct sched_dl_entity *dl_se)
54 : {
55 : return dl_se;
56 : }
57 :
58 : static inline bool is_dl_boosted(struct sched_dl_entity *dl_se)
59 : {
60 : return false;
61 : }
62 : #endif
63 :
64 : #ifdef CONFIG_SMP
65 : static inline struct dl_bw *dl_bw_of(int i)
66 : {
67 : RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
68 : "sched RCU must be held");
69 : return &cpu_rq(i)->rd->dl_bw;
70 : }
71 :
72 : static inline int dl_bw_cpus(int i)
73 : {
74 : struct root_domain *rd = cpu_rq(i)->rd;
75 : int cpus;
76 :
77 : RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
78 : "sched RCU must be held");
79 :
80 : if (cpumask_subset(rd->span, cpu_active_mask))
81 : return cpumask_weight(rd->span);
82 :
83 : cpus = 0;
84 :
85 : for_each_cpu_and(i, rd->span, cpu_active_mask)
86 : cpus++;
87 :
88 : return cpus;
89 : }
90 :
91 : static inline unsigned long __dl_bw_capacity(int i)
92 : {
93 : struct root_domain *rd = cpu_rq(i)->rd;
94 : unsigned long cap = 0;
95 :
96 : RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
97 : "sched RCU must be held");
98 :
99 : for_each_cpu_and(i, rd->span, cpu_active_mask)
100 : cap += capacity_orig_of(i);
101 :
102 : return cap;
103 : }
104 :
105 : /*
106 : * XXX Fix: If 'rq->rd == def_root_domain' perform AC against capacity
107 : * of the CPU the task is running on rather rd's \Sum CPU capacity.
108 : */
109 : static inline unsigned long dl_bw_capacity(int i)
110 : {
111 : if (!static_branch_unlikely(&sched_asym_cpucapacity) &&
112 : capacity_orig_of(i) == SCHED_CAPACITY_SCALE) {
113 : return dl_bw_cpus(i) << SCHED_CAPACITY_SHIFT;
114 : } else {
115 : return __dl_bw_capacity(i);
116 : }
117 : }
118 :
119 : static inline bool dl_bw_visited(int cpu, u64 gen)
120 : {
121 : struct root_domain *rd = cpu_rq(cpu)->rd;
122 :
123 : if (rd->visit_gen == gen)
124 : return true;
125 :
126 : rd->visit_gen = gen;
127 : return false;
128 : }
129 :
130 : static inline
131 : void __dl_update(struct dl_bw *dl_b, s64 bw)
132 : {
133 : struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
134 : int i;
135 :
136 : RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
137 : "sched RCU must be held");
138 : for_each_cpu_and(i, rd->span, cpu_active_mask) {
139 : struct rq *rq = cpu_rq(i);
140 :
141 : rq->dl.extra_bw += bw;
142 : }
143 : }
144 : #else
145 : static inline struct dl_bw *dl_bw_of(int i)
146 : {
147 0 : return &cpu_rq(i)->dl.dl_bw;
148 : }
149 :
150 : static inline int dl_bw_cpus(int i)
151 : {
152 : return 1;
153 : }
154 :
155 : static inline unsigned long dl_bw_capacity(int i)
156 : {
157 : return SCHED_CAPACITY_SCALE;
158 : }
159 :
160 : static inline bool dl_bw_visited(int cpu, u64 gen)
161 : {
162 : return false;
163 : }
164 :
165 : static inline
166 : void __dl_update(struct dl_bw *dl_b, s64 bw)
167 : {
168 0 : struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
169 :
170 0 : dl->extra_bw += bw;
171 : }
172 : #endif
173 :
174 : static inline
175 : void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
176 : {
177 0 : dl_b->total_bw -= tsk_bw;
178 0 : __dl_update(dl_b, (s32)tsk_bw / cpus);
179 : }
180 :
181 : static inline
182 : void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
183 : {
184 0 : dl_b->total_bw += tsk_bw;
185 0 : __dl_update(dl_b, -((s32)tsk_bw / cpus));
186 : }
187 :
188 : static inline bool
189 : __dl_overflow(struct dl_bw *dl_b, unsigned long cap, u64 old_bw, u64 new_bw)
190 : {
191 0 : return dl_b->bw != -1 &&
192 0 : cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
193 : }
194 :
195 : static inline
196 0 : void __add_running_bw(u64 dl_bw, struct dl_rq *dl_rq)
197 : {
198 0 : u64 old = dl_rq->running_bw;
199 :
200 0 : lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
201 0 : dl_rq->running_bw += dl_bw;
202 0 : SCHED_WARN_ON(dl_rq->running_bw < old); /* overflow */
203 0 : SCHED_WARN_ON(dl_rq->running_bw > dl_rq->this_bw);
204 : /* kick cpufreq (see the comment in kernel/sched/sched.h). */
205 0 : cpufreq_update_util(rq_of_dl_rq(dl_rq), 0);
206 0 : }
207 :
208 : static inline
209 0 : void __sub_running_bw(u64 dl_bw, struct dl_rq *dl_rq)
210 : {
211 0 : u64 old = dl_rq->running_bw;
212 :
213 0 : lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
214 0 : dl_rq->running_bw -= dl_bw;
215 0 : SCHED_WARN_ON(dl_rq->running_bw > old); /* underflow */
216 0 : if (dl_rq->running_bw > old)
217 0 : dl_rq->running_bw = 0;
218 : /* kick cpufreq (see the comment in kernel/sched/sched.h). */
219 0 : cpufreq_update_util(rq_of_dl_rq(dl_rq), 0);
220 0 : }
221 :
222 : static inline
223 0 : void __add_rq_bw(u64 dl_bw, struct dl_rq *dl_rq)
224 : {
225 0 : u64 old = dl_rq->this_bw;
226 :
227 0 : lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
228 0 : dl_rq->this_bw += dl_bw;
229 0 : SCHED_WARN_ON(dl_rq->this_bw < old); /* overflow */
230 0 : }
231 :
232 : static inline
233 0 : void __sub_rq_bw(u64 dl_bw, struct dl_rq *dl_rq)
234 : {
235 0 : u64 old = dl_rq->this_bw;
236 :
237 0 : lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
238 0 : dl_rq->this_bw -= dl_bw;
239 0 : SCHED_WARN_ON(dl_rq->this_bw > old); /* underflow */
240 0 : if (dl_rq->this_bw > old)
241 0 : dl_rq->this_bw = 0;
242 0 : SCHED_WARN_ON(dl_rq->running_bw > dl_rq->this_bw);
243 0 : }
244 :
245 : static inline
246 : void add_rq_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
247 : {
248 0 : if (!dl_entity_is_special(dl_se))
249 0 : __add_rq_bw(dl_se->dl_bw, dl_rq);
250 : }
251 :
252 : static inline
253 : void sub_rq_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
254 : {
255 0 : if (!dl_entity_is_special(dl_se))
256 0 : __sub_rq_bw(dl_se->dl_bw, dl_rq);
257 : }
258 :
259 : static inline
260 : void add_running_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
261 : {
262 0 : if (!dl_entity_is_special(dl_se))
263 0 : __add_running_bw(dl_se->dl_bw, dl_rq);
264 : }
265 :
266 : static inline
267 : void sub_running_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
268 : {
269 0 : if (!dl_entity_is_special(dl_se))
270 0 : __sub_running_bw(dl_se->dl_bw, dl_rq);
271 : }
272 :
273 0 : static void dl_change_utilization(struct task_struct *p, u64 new_bw)
274 : {
275 : struct rq *rq;
276 :
277 0 : BUG_ON(p->dl.flags & SCHED_FLAG_SUGOV);
278 :
279 0 : if (task_on_rq_queued(p))
280 : return;
281 :
282 0 : rq = task_rq(p);
283 0 : if (p->dl.dl_non_contending) {
284 0 : sub_running_bw(&p->dl, &rq->dl);
285 0 : p->dl.dl_non_contending = 0;
286 : /*
287 : * If the timer handler is currently running and the
288 : * timer cannot be canceled, inactive_task_timer()
289 : * will see that dl_not_contending is not set, and
290 : * will not touch the rq's active utilization,
291 : * so we are still safe.
292 : */
293 0 : if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
294 0 : put_task_struct(p);
295 : }
296 0 : __sub_rq_bw(p->dl.dl_bw, &rq->dl);
297 0 : __add_rq_bw(new_bw, &rq->dl);
298 : }
299 :
300 : /*
301 : * The utilization of a task cannot be immediately removed from
302 : * the rq active utilization (running_bw) when the task blocks.
303 : * Instead, we have to wait for the so called "0-lag time".
304 : *
305 : * If a task blocks before the "0-lag time", a timer (the inactive
306 : * timer) is armed, and running_bw is decreased when the timer
307 : * fires.
308 : *
309 : * If the task wakes up again before the inactive timer fires,
310 : * the timer is canceled, whereas if the task wakes up after the
311 : * inactive timer fired (and running_bw has been decreased) the
312 : * task's utilization has to be added to running_bw again.
313 : * A flag in the deadline scheduling entity (dl_non_contending)
314 : * is used to avoid race conditions between the inactive timer handler
315 : * and task wakeups.
316 : *
317 : * The following diagram shows how running_bw is updated. A task is
318 : * "ACTIVE" when its utilization contributes to running_bw; an
319 : * "ACTIVE contending" task is in the TASK_RUNNING state, while an
320 : * "ACTIVE non contending" task is a blocked task for which the "0-lag time"
321 : * has not passed yet. An "INACTIVE" task is a task for which the "0-lag"
322 : * time already passed, which does not contribute to running_bw anymore.
323 : * +------------------+
324 : * wakeup | ACTIVE |
325 : * +------------------>+ contending |
326 : * | add_running_bw | |
327 : * | +----+------+------+
328 : * | | ^
329 : * | dequeue | |
330 : * +--------+-------+ | |
331 : * | | t >= 0-lag | | wakeup
332 : * | INACTIVE |<---------------+ |
333 : * | | sub_running_bw | |
334 : * +--------+-------+ | |
335 : * ^ | |
336 : * | t < 0-lag | |
337 : * | | |
338 : * | V |
339 : * | +----+------+------+
340 : * | sub_running_bw | ACTIVE |
341 : * +-------------------+ |
342 : * inactive timer | non contending |
343 : * fired +------------------+
344 : *
345 : * The task_non_contending() function is invoked when a task
346 : * blocks, and checks if the 0-lag time already passed or
347 : * not (in the first case, it directly updates running_bw;
348 : * in the second case, it arms the inactive timer).
349 : *
350 : * The task_contending() function is invoked when a task wakes
351 : * up, and checks if the task is still in the "ACTIVE non contending"
352 : * state or not (in the second case, it updates running_bw).
353 : */
354 0 : static void task_non_contending(struct task_struct *p)
355 : {
356 0 : struct sched_dl_entity *dl_se = &p->dl;
357 0 : struct hrtimer *timer = &dl_se->inactive_timer;
358 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
359 0 : struct rq *rq = rq_of_dl_rq(dl_rq);
360 : s64 zerolag_time;
361 :
362 : /*
363 : * If this is a non-deadline task that has been boosted,
364 : * do nothing
365 : */
366 0 : if (dl_se->dl_runtime == 0)
367 : return;
368 :
369 0 : if (dl_entity_is_special(dl_se))
370 : return;
371 :
372 0 : WARN_ON(dl_se->dl_non_contending);
373 :
374 0 : zerolag_time = dl_se->deadline -
375 0 : div64_long((dl_se->runtime * dl_se->dl_period),
376 : dl_se->dl_runtime);
377 :
378 : /*
379 : * Using relative times instead of the absolute "0-lag time"
380 : * allows to simplify the code
381 : */
382 0 : zerolag_time -= rq_clock(rq);
383 :
384 : /*
385 : * If the "0-lag time" already passed, decrease the active
386 : * utilization now, instead of starting a timer
387 : */
388 0 : if ((zerolag_time < 0) || hrtimer_active(&dl_se->inactive_timer)) {
389 0 : if (dl_task(p))
390 0 : sub_running_bw(dl_se, dl_rq);
391 0 : if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
392 0 : struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
393 :
394 0 : if (READ_ONCE(p->__state) == TASK_DEAD)
395 0 : sub_rq_bw(&p->dl, &rq->dl);
396 0 : raw_spin_lock(&dl_b->lock);
397 0 : __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
398 0 : __dl_clear_params(p);
399 0 : raw_spin_unlock(&dl_b->lock);
400 : }
401 :
402 : return;
403 : }
404 :
405 0 : dl_se->dl_non_contending = 1;
406 0 : get_task_struct(p);
407 0 : hrtimer_start(timer, ns_to_ktime(zerolag_time), HRTIMER_MODE_REL_HARD);
408 : }
409 :
410 0 : static void task_contending(struct sched_dl_entity *dl_se, int flags)
411 : {
412 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
413 :
414 : /*
415 : * If this is a non-deadline task that has been boosted,
416 : * do nothing
417 : */
418 0 : if (dl_se->dl_runtime == 0)
419 : return;
420 :
421 : if (flags & ENQUEUE_MIGRATED)
422 : add_rq_bw(dl_se, dl_rq);
423 :
424 0 : if (dl_se->dl_non_contending) {
425 0 : dl_se->dl_non_contending = 0;
426 : /*
427 : * If the timer handler is currently running and the
428 : * timer cannot be canceled, inactive_task_timer()
429 : * will see that dl_not_contending is not set, and
430 : * will not touch the rq's active utilization,
431 : * so we are still safe.
432 : */
433 0 : if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1)
434 0 : put_task_struct(dl_task_of(dl_se));
435 : } else {
436 : /*
437 : * Since "dl_non_contending" is not set, the
438 : * task's utilization has already been removed from
439 : * active utilization (either when the task blocked,
440 : * when the "inactive timer" fired).
441 : * So, add it back.
442 : */
443 0 : add_running_bw(dl_se, dl_rq);
444 : }
445 : }
446 :
447 : static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
448 : {
449 0 : struct sched_dl_entity *dl_se = &p->dl;
450 :
451 : return rb_first_cached(&dl_rq->root) == &dl_se->rb_node;
452 : }
453 :
454 : static void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
455 :
456 0 : void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime)
457 : {
458 : raw_spin_lock_init(&dl_b->dl_runtime_lock);
459 0 : dl_b->dl_period = period;
460 0 : dl_b->dl_runtime = runtime;
461 0 : }
462 :
463 1 : void init_dl_bw(struct dl_bw *dl_b)
464 : {
465 : raw_spin_lock_init(&dl_b->lock);
466 1 : if (global_rt_runtime() == RUNTIME_INF)
467 0 : dl_b->bw = -1;
468 : else
469 1 : dl_b->bw = to_ratio(global_rt_period(), global_rt_runtime());
470 1 : dl_b->total_bw = 0;
471 1 : }
472 :
473 1 : void init_dl_rq(struct dl_rq *dl_rq)
474 : {
475 1 : dl_rq->root = RB_ROOT_CACHED;
476 :
477 : #ifdef CONFIG_SMP
478 : /* zero means no -deadline tasks */
479 : dl_rq->earliest_dl.curr = dl_rq->earliest_dl.next = 0;
480 :
481 : dl_rq->dl_nr_migratory = 0;
482 : dl_rq->overloaded = 0;
483 : dl_rq->pushable_dl_tasks_root = RB_ROOT_CACHED;
484 : #else
485 1 : init_dl_bw(&dl_rq->dl_bw);
486 : #endif
487 :
488 1 : dl_rq->running_bw = 0;
489 1 : dl_rq->this_bw = 0;
490 1 : init_dl_rq_bw_ratio(dl_rq);
491 1 : }
492 :
493 : #ifdef CONFIG_SMP
494 :
495 : static inline int dl_overloaded(struct rq *rq)
496 : {
497 : return atomic_read(&rq->rd->dlo_count);
498 : }
499 :
500 : static inline void dl_set_overload(struct rq *rq)
501 : {
502 : if (!rq->online)
503 : return;
504 :
505 : cpumask_set_cpu(rq->cpu, rq->rd->dlo_mask);
506 : /*
507 : * Must be visible before the overload count is
508 : * set (as in sched_rt.c).
509 : *
510 : * Matched by the barrier in pull_dl_task().
511 : */
512 : smp_wmb();
513 : atomic_inc(&rq->rd->dlo_count);
514 : }
515 :
516 : static inline void dl_clear_overload(struct rq *rq)
517 : {
518 : if (!rq->online)
519 : return;
520 :
521 : atomic_dec(&rq->rd->dlo_count);
522 : cpumask_clear_cpu(rq->cpu, rq->rd->dlo_mask);
523 : }
524 :
525 : static void update_dl_migration(struct dl_rq *dl_rq)
526 : {
527 : if (dl_rq->dl_nr_migratory && dl_rq->dl_nr_running > 1) {
528 : if (!dl_rq->overloaded) {
529 : dl_set_overload(rq_of_dl_rq(dl_rq));
530 : dl_rq->overloaded = 1;
531 : }
532 : } else if (dl_rq->overloaded) {
533 : dl_clear_overload(rq_of_dl_rq(dl_rq));
534 : dl_rq->overloaded = 0;
535 : }
536 : }
537 :
538 : static void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
539 : {
540 : struct task_struct *p = dl_task_of(dl_se);
541 :
542 : if (p->nr_cpus_allowed > 1)
543 : dl_rq->dl_nr_migratory++;
544 :
545 : update_dl_migration(dl_rq);
546 : }
547 :
548 : static void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
549 : {
550 : struct task_struct *p = dl_task_of(dl_se);
551 :
552 : if (p->nr_cpus_allowed > 1)
553 : dl_rq->dl_nr_migratory--;
554 :
555 : update_dl_migration(dl_rq);
556 : }
557 :
558 : #define __node_2_pdl(node) \
559 : rb_entry((node), struct task_struct, pushable_dl_tasks)
560 :
561 : static inline bool __pushable_less(struct rb_node *a, const struct rb_node *b)
562 : {
563 : return dl_entity_preempt(&__node_2_pdl(a)->dl, &__node_2_pdl(b)->dl);
564 : }
565 :
566 : /*
567 : * The list of pushable -deadline task is not a plist, like in
568 : * sched_rt.c, it is an rb-tree with tasks ordered by deadline.
569 : */
570 : static void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
571 : {
572 : struct rb_node *leftmost;
573 :
574 : BUG_ON(!RB_EMPTY_NODE(&p->pushable_dl_tasks));
575 :
576 : leftmost = rb_add_cached(&p->pushable_dl_tasks,
577 : &rq->dl.pushable_dl_tasks_root,
578 : __pushable_less);
579 : if (leftmost)
580 : rq->dl.earliest_dl.next = p->dl.deadline;
581 : }
582 :
583 : static void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
584 : {
585 : struct dl_rq *dl_rq = &rq->dl;
586 : struct rb_root_cached *root = &dl_rq->pushable_dl_tasks_root;
587 : struct rb_node *leftmost;
588 :
589 : if (RB_EMPTY_NODE(&p->pushable_dl_tasks))
590 : return;
591 :
592 : leftmost = rb_erase_cached(&p->pushable_dl_tasks, root);
593 : if (leftmost)
594 : dl_rq->earliest_dl.next = __node_2_pdl(leftmost)->dl.deadline;
595 :
596 : RB_CLEAR_NODE(&p->pushable_dl_tasks);
597 : }
598 :
599 : static inline int has_pushable_dl_tasks(struct rq *rq)
600 : {
601 : return !RB_EMPTY_ROOT(&rq->dl.pushable_dl_tasks_root.rb_root);
602 : }
603 :
604 : static int push_dl_task(struct rq *rq);
605 :
606 : static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
607 : {
608 : return rq->online && dl_task(prev);
609 : }
610 :
611 : static DEFINE_PER_CPU(struct callback_head, dl_push_head);
612 : static DEFINE_PER_CPU(struct callback_head, dl_pull_head);
613 :
614 : static void push_dl_tasks(struct rq *);
615 : static void pull_dl_task(struct rq *);
616 :
617 : static inline void deadline_queue_push_tasks(struct rq *rq)
618 : {
619 : if (!has_pushable_dl_tasks(rq))
620 : return;
621 :
622 : queue_balance_callback(rq, &per_cpu(dl_push_head, rq->cpu), push_dl_tasks);
623 : }
624 :
625 : static inline void deadline_queue_pull_task(struct rq *rq)
626 : {
627 : queue_balance_callback(rq, &per_cpu(dl_pull_head, rq->cpu), pull_dl_task);
628 : }
629 :
630 : static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq);
631 :
632 : static struct rq *dl_task_offline_migration(struct rq *rq, struct task_struct *p)
633 : {
634 : struct rq *later_rq = NULL;
635 : struct dl_bw *dl_b;
636 :
637 : later_rq = find_lock_later_rq(p, rq);
638 : if (!later_rq) {
639 : int cpu;
640 :
641 : /*
642 : * If we cannot preempt any rq, fall back to pick any
643 : * online CPU:
644 : */
645 : cpu = cpumask_any_and(cpu_active_mask, p->cpus_ptr);
646 : if (cpu >= nr_cpu_ids) {
647 : /*
648 : * Failed to find any suitable CPU.
649 : * The task will never come back!
650 : */
651 : BUG_ON(dl_bandwidth_enabled());
652 :
653 : /*
654 : * If admission control is disabled we
655 : * try a little harder to let the task
656 : * run.
657 : */
658 : cpu = cpumask_any(cpu_active_mask);
659 : }
660 : later_rq = cpu_rq(cpu);
661 : double_lock_balance(rq, later_rq);
662 : }
663 :
664 : if (p->dl.dl_non_contending || p->dl.dl_throttled) {
665 : /*
666 : * Inactive timer is armed (or callback is running, but
667 : * waiting for us to release rq locks). In any case, when it
668 : * will fire (or continue), it will see running_bw of this
669 : * task migrated to later_rq (and correctly handle it).
670 : */
671 : sub_running_bw(&p->dl, &rq->dl);
672 : sub_rq_bw(&p->dl, &rq->dl);
673 :
674 : add_rq_bw(&p->dl, &later_rq->dl);
675 : add_running_bw(&p->dl, &later_rq->dl);
676 : } else {
677 : sub_rq_bw(&p->dl, &rq->dl);
678 : add_rq_bw(&p->dl, &later_rq->dl);
679 : }
680 :
681 : /*
682 : * And we finally need to fixup root_domain(s) bandwidth accounting,
683 : * since p is still hanging out in the old (now moved to default) root
684 : * domain.
685 : */
686 : dl_b = &rq->rd->dl_bw;
687 : raw_spin_lock(&dl_b->lock);
688 : __dl_sub(dl_b, p->dl.dl_bw, cpumask_weight(rq->rd->span));
689 : raw_spin_unlock(&dl_b->lock);
690 :
691 : dl_b = &later_rq->rd->dl_bw;
692 : raw_spin_lock(&dl_b->lock);
693 : __dl_add(dl_b, p->dl.dl_bw, cpumask_weight(later_rq->rd->span));
694 : raw_spin_unlock(&dl_b->lock);
695 :
696 : set_task_cpu(p, later_rq->cpu);
697 : double_unlock_balance(later_rq, rq);
698 :
699 : return later_rq;
700 : }
701 :
702 : #else
703 :
704 : static inline
705 : void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
706 : {
707 : }
708 :
709 : static inline
710 : void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
711 : {
712 : }
713 :
714 : static inline
715 : void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
716 : {
717 : }
718 :
719 : static inline
720 : void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
721 : {
722 : }
723 :
724 : static inline void deadline_queue_push_tasks(struct rq *rq)
725 : {
726 : }
727 :
728 : static inline void deadline_queue_pull_task(struct rq *rq)
729 : {
730 : }
731 : #endif /* CONFIG_SMP */
732 :
733 : static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags);
734 : static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags);
735 : static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p, int flags);
736 :
737 : /*
738 : * We are being explicitly informed that a new instance is starting,
739 : * and this means that:
740 : * - the absolute deadline of the entity has to be placed at
741 : * current time + relative deadline;
742 : * - the runtime of the entity has to be set to the maximum value.
743 : *
744 : * The capability of specifying such event is useful whenever a -deadline
745 : * entity wants to (try to!) synchronize its behaviour with the scheduler's
746 : * one, and to (try to!) reconcile itself with its own scheduling
747 : * parameters.
748 : */
749 0 : static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
750 : {
751 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
752 0 : struct rq *rq = rq_of_dl_rq(dl_rq);
753 :
754 0 : WARN_ON(is_dl_boosted(dl_se));
755 0 : WARN_ON(dl_time_before(rq_clock(rq), dl_se->deadline));
756 :
757 : /*
758 : * We are racing with the deadline timer. So, do nothing because
759 : * the deadline timer handler will take care of properly recharging
760 : * the runtime and postponing the deadline
761 : */
762 0 : if (dl_se->dl_throttled)
763 : return;
764 :
765 : /*
766 : * We use the regular wall clock time to set deadlines in the
767 : * future; in fact, we must consider execution overheads (time
768 : * spent on hardirq context, etc.).
769 : */
770 0 : dl_se->deadline = rq_clock(rq) + dl_se->dl_deadline;
771 0 : dl_se->runtime = dl_se->dl_runtime;
772 : }
773 :
774 : /*
775 : * Pure Earliest Deadline First (EDF) scheduling does not deal with the
776 : * possibility of a entity lasting more than what it declared, and thus
777 : * exhausting its runtime.
778 : *
779 : * Here we are interested in making runtime overrun possible, but we do
780 : * not want a entity which is misbehaving to affect the scheduling of all
781 : * other entities.
782 : * Therefore, a budgeting strategy called Constant Bandwidth Server (CBS)
783 : * is used, in order to confine each entity within its own bandwidth.
784 : *
785 : * This function deals exactly with that, and ensures that when the runtime
786 : * of a entity is replenished, its deadline is also postponed. That ensures
787 : * the overrunning entity can't interfere with other entity in the system and
788 : * can't make them miss their deadlines. Reasons why this kind of overruns
789 : * could happen are, typically, a entity voluntarily trying to overcome its
790 : * runtime, or it just underestimated it during sched_setattr().
791 : */
792 0 : static void replenish_dl_entity(struct sched_dl_entity *dl_se)
793 : {
794 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
795 0 : struct rq *rq = rq_of_dl_rq(dl_rq);
796 :
797 0 : BUG_ON(pi_of(dl_se)->dl_runtime <= 0);
798 :
799 : /*
800 : * This could be the case for a !-dl task that is boosted.
801 : * Just go with full inherited parameters.
802 : */
803 0 : if (dl_se->dl_deadline == 0) {
804 0 : dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
805 0 : dl_se->runtime = pi_of(dl_se)->dl_runtime;
806 : }
807 :
808 0 : if (dl_se->dl_yielded && dl_se->runtime > 0)
809 0 : dl_se->runtime = 0;
810 :
811 : /*
812 : * We keep moving the deadline away until we get some
813 : * available runtime for the entity. This ensures correct
814 : * handling of situations where the runtime overrun is
815 : * arbitrary large.
816 : */
817 0 : while (dl_se->runtime <= 0) {
818 0 : dl_se->deadline += pi_of(dl_se)->dl_period;
819 0 : dl_se->runtime += pi_of(dl_se)->dl_runtime;
820 : }
821 :
822 : /*
823 : * At this point, the deadline really should be "in
824 : * the future" with respect to rq->clock. If it's
825 : * not, we are, for some reason, lagging too much!
826 : * Anyway, after having warn userspace abut that,
827 : * we still try to keep the things running by
828 : * resetting the deadline and the budget of the
829 : * entity.
830 : */
831 0 : if (dl_time_before(dl_se->deadline, rq_clock(rq))) {
832 0 : printk_deferred_once("sched: DL replenish lagged too much\n");
833 0 : dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
834 0 : dl_se->runtime = pi_of(dl_se)->dl_runtime;
835 : }
836 :
837 0 : if (dl_se->dl_yielded)
838 0 : dl_se->dl_yielded = 0;
839 0 : if (dl_se->dl_throttled)
840 0 : dl_se->dl_throttled = 0;
841 0 : }
842 :
843 : /*
844 : * Here we check if --at time t-- an entity (which is probably being
845 : * [re]activated or, in general, enqueued) can use its remaining runtime
846 : * and its current deadline _without_ exceeding the bandwidth it is
847 : * assigned (function returns true if it can't). We are in fact applying
848 : * one of the CBS rules: when a task wakes up, if the residual runtime
849 : * over residual deadline fits within the allocated bandwidth, then we
850 : * can keep the current (absolute) deadline and residual budget without
851 : * disrupting the schedulability of the system. Otherwise, we should
852 : * refill the runtime and set the deadline a period in the future,
853 : * because keeping the current (absolute) deadline of the task would
854 : * result in breaking guarantees promised to other tasks (refer to
855 : * Documentation/scheduler/sched-deadline.rst for more information).
856 : *
857 : * This function returns true if:
858 : *
859 : * runtime / (deadline - t) > dl_runtime / dl_deadline ,
860 : *
861 : * IOW we can't recycle current parameters.
862 : *
863 : * Notice that the bandwidth check is done against the deadline. For
864 : * task with deadline equal to period this is the same of using
865 : * dl_period instead of dl_deadline in the equation above.
866 : */
867 : static bool dl_entity_overflow(struct sched_dl_entity *dl_se, u64 t)
868 : {
869 : u64 left, right;
870 :
871 : /*
872 : * left and right are the two sides of the equation above,
873 : * after a bit of shuffling to use multiplications instead
874 : * of divisions.
875 : *
876 : * Note that none of the time values involved in the two
877 : * multiplications are absolute: dl_deadline and dl_runtime
878 : * are the relative deadline and the maximum runtime of each
879 : * instance, runtime is the runtime left for the last instance
880 : * and (deadline - t), since t is rq->clock, is the time left
881 : * to the (absolute) deadline. Even if overflowing the u64 type
882 : * is very unlikely to occur in both cases, here we scale down
883 : * as we want to avoid that risk at all. Scaling down by 10
884 : * means that we reduce granularity to 1us. We are fine with it,
885 : * since this is only a true/false check and, anyway, thinking
886 : * of anything below microseconds resolution is actually fiction
887 : * (but still we want to give the user that illusion >;).
888 : */
889 0 : left = (pi_of(dl_se)->dl_deadline >> DL_SCALE) * (dl_se->runtime >> DL_SCALE);
890 0 : right = ((dl_se->deadline - t) >> DL_SCALE) *
891 0 : (pi_of(dl_se)->dl_runtime >> DL_SCALE);
892 :
893 0 : return dl_time_before(right, left);
894 : }
895 :
896 : /*
897 : * Revised wakeup rule [1]: For self-suspending tasks, rather then
898 : * re-initializing task's runtime and deadline, the revised wakeup
899 : * rule adjusts the task's runtime to avoid the task to overrun its
900 : * density.
901 : *
902 : * Reasoning: a task may overrun the density if:
903 : * runtime / (deadline - t) > dl_runtime / dl_deadline
904 : *
905 : * Therefore, runtime can be adjusted to:
906 : * runtime = (dl_runtime / dl_deadline) * (deadline - t)
907 : *
908 : * In such way that runtime will be equal to the maximum density
909 : * the task can use without breaking any rule.
910 : *
911 : * [1] Luca Abeni, Giuseppe Lipari, and Juri Lelli. 2015. Constant
912 : * bandwidth server revisited. SIGBED Rev. 11, 4 (January 2015), 19-24.
913 : */
914 : static void
915 0 : update_dl_revised_wakeup(struct sched_dl_entity *dl_se, struct rq *rq)
916 : {
917 0 : u64 laxity = dl_se->deadline - rq_clock(rq);
918 :
919 : /*
920 : * If the task has deadline < period, and the deadline is in the past,
921 : * it should already be throttled before this check.
922 : *
923 : * See update_dl_entity() comments for further details.
924 : */
925 0 : WARN_ON(dl_time_before(dl_se->deadline, rq_clock(rq)));
926 :
927 0 : dl_se->runtime = (dl_se->dl_density * laxity) >> BW_SHIFT;
928 0 : }
929 :
930 : /*
931 : * Regarding the deadline, a task with implicit deadline has a relative
932 : * deadline == relative period. A task with constrained deadline has a
933 : * relative deadline <= relative period.
934 : *
935 : * We support constrained deadline tasks. However, there are some restrictions
936 : * applied only for tasks which do not have an implicit deadline. See
937 : * update_dl_entity() to know more about such restrictions.
938 : *
939 : * The dl_is_implicit() returns true if the task has an implicit deadline.
940 : */
941 : static inline bool dl_is_implicit(struct sched_dl_entity *dl_se)
942 : {
943 : return dl_se->dl_deadline == dl_se->dl_period;
944 : }
945 :
946 : /*
947 : * When a deadline entity is placed in the runqueue, its runtime and deadline
948 : * might need to be updated. This is done by a CBS wake up rule. There are two
949 : * different rules: 1) the original CBS; and 2) the Revisited CBS.
950 : *
951 : * When the task is starting a new period, the Original CBS is used. In this
952 : * case, the runtime is replenished and a new absolute deadline is set.
953 : *
954 : * When a task is queued before the begin of the next period, using the
955 : * remaining runtime and deadline could make the entity to overflow, see
956 : * dl_entity_overflow() to find more about runtime overflow. When such case
957 : * is detected, the runtime and deadline need to be updated.
958 : *
959 : * If the task has an implicit deadline, i.e., deadline == period, the Original
960 : * CBS is applied. the runtime is replenished and a new absolute deadline is
961 : * set, as in the previous cases.
962 : *
963 : * However, the Original CBS does not work properly for tasks with
964 : * deadline < period, which are said to have a constrained deadline. By
965 : * applying the Original CBS, a constrained deadline task would be able to run
966 : * runtime/deadline in a period. With deadline < period, the task would
967 : * overrun the runtime/period allowed bandwidth, breaking the admission test.
968 : *
969 : * In order to prevent this misbehave, the Revisited CBS is used for
970 : * constrained deadline tasks when a runtime overflow is detected. In the
971 : * Revisited CBS, rather than replenishing & setting a new absolute deadline,
972 : * the remaining runtime of the task is reduced to avoid runtime overflow.
973 : * Please refer to the comments update_dl_revised_wakeup() function to find
974 : * more about the Revised CBS rule.
975 : */
976 0 : static void update_dl_entity(struct sched_dl_entity *dl_se)
977 : {
978 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
979 0 : struct rq *rq = rq_of_dl_rq(dl_rq);
980 :
981 0 : if (dl_time_before(dl_se->deadline, rq_clock(rq)) ||
982 0 : dl_entity_overflow(dl_se, rq_clock(rq))) {
983 :
984 0 : if (unlikely(!dl_is_implicit(dl_se) &&
985 : !dl_time_before(dl_se->deadline, rq_clock(rq)) &&
986 : !is_dl_boosted(dl_se))) {
987 0 : update_dl_revised_wakeup(dl_se, rq);
988 0 : return;
989 : }
990 :
991 0 : dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
992 0 : dl_se->runtime = pi_of(dl_se)->dl_runtime;
993 : }
994 : }
995 :
996 : static inline u64 dl_next_period(struct sched_dl_entity *dl_se)
997 : {
998 0 : return dl_se->deadline - dl_se->dl_deadline + dl_se->dl_period;
999 : }
1000 :
1001 : /*
1002 : * If the entity depleted all its runtime, and if we want it to sleep
1003 : * while waiting for some new execution time to become available, we
1004 : * set the bandwidth replenishment timer to the replenishment instant
1005 : * and try to activate it.
1006 : *
1007 : * Notice that it is important for the caller to know if the timer
1008 : * actually started or not (i.e., the replenishment instant is in
1009 : * the future or in the past).
1010 : */
1011 0 : static int start_dl_timer(struct task_struct *p)
1012 : {
1013 0 : struct sched_dl_entity *dl_se = &p->dl;
1014 0 : struct hrtimer *timer = &dl_se->dl_timer;
1015 0 : struct rq *rq = task_rq(p);
1016 : ktime_t now, act;
1017 : s64 delta;
1018 :
1019 0 : lockdep_assert_rq_held(rq);
1020 :
1021 : /*
1022 : * We want the timer to fire at the deadline, but considering
1023 : * that it is actually coming from rq->clock and not from
1024 : * hrtimer's time base reading.
1025 : */
1026 0 : act = ns_to_ktime(dl_next_period(dl_se));
1027 0 : now = hrtimer_cb_get_time(timer);
1028 0 : delta = ktime_to_ns(now) - rq_clock(rq);
1029 0 : act = ktime_add_ns(act, delta);
1030 :
1031 : /*
1032 : * If the expiry time already passed, e.g., because the value
1033 : * chosen as the deadline is too small, don't even try to
1034 : * start the timer in the past!
1035 : */
1036 0 : if (ktime_us_delta(act, now) < 0)
1037 : return 0;
1038 :
1039 : /*
1040 : * !enqueued will guarantee another callback; even if one is already in
1041 : * progress. This ensures a balanced {get,put}_task_struct().
1042 : *
1043 : * The race against __run_timer() clearing the enqueued state is
1044 : * harmless because we're holding task_rq()->lock, therefore the timer
1045 : * expiring after we've done the check will wait on its task_rq_lock()
1046 : * and observe our state.
1047 : */
1048 0 : if (!hrtimer_is_queued(timer)) {
1049 0 : get_task_struct(p);
1050 : hrtimer_start(timer, act, HRTIMER_MODE_ABS_HARD);
1051 : }
1052 :
1053 : return 1;
1054 : }
1055 :
1056 : /*
1057 : * This is the bandwidth enforcement timer callback. If here, we know
1058 : * a task is not on its dl_rq, since the fact that the timer was running
1059 : * means the task is throttled and needs a runtime replenishment.
1060 : *
1061 : * However, what we actually do depends on the fact the task is active,
1062 : * (it is on its rq) or has been removed from there by a call to
1063 : * dequeue_task_dl(). In the former case we must issue the runtime
1064 : * replenishment and add the task back to the dl_rq; in the latter, we just
1065 : * do nothing but clearing dl_throttled, so that runtime and deadline
1066 : * updating (and the queueing back to dl_rq) will be done by the
1067 : * next call to enqueue_task_dl().
1068 : */
1069 0 : static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
1070 : {
1071 0 : struct sched_dl_entity *dl_se = container_of(timer,
1072 : struct sched_dl_entity,
1073 : dl_timer);
1074 0 : struct task_struct *p = dl_task_of(dl_se);
1075 : struct rq_flags rf;
1076 : struct rq *rq;
1077 :
1078 0 : rq = task_rq_lock(p, &rf);
1079 :
1080 : /*
1081 : * The task might have changed its scheduling policy to something
1082 : * different than SCHED_DEADLINE (through switched_from_dl()).
1083 : */
1084 0 : if (!dl_task(p))
1085 : goto unlock;
1086 :
1087 : /*
1088 : * The task might have been boosted by someone else and might be in the
1089 : * boosting/deboosting path, its not throttled.
1090 : */
1091 0 : if (is_dl_boosted(dl_se))
1092 : goto unlock;
1093 :
1094 : /*
1095 : * Spurious timer due to start_dl_timer() race; or we already received
1096 : * a replenishment from rt_mutex_setprio().
1097 : */
1098 0 : if (!dl_se->dl_throttled)
1099 : goto unlock;
1100 :
1101 : sched_clock_tick();
1102 0 : update_rq_clock(rq);
1103 :
1104 : /*
1105 : * If the throttle happened during sched-out; like:
1106 : *
1107 : * schedule()
1108 : * deactivate_task()
1109 : * dequeue_task_dl()
1110 : * update_curr_dl()
1111 : * start_dl_timer()
1112 : * __dequeue_task_dl()
1113 : * prev->on_rq = 0;
1114 : *
1115 : * We can be both throttled and !queued. Replenish the counter
1116 : * but do not enqueue -- wait for our wakeup to do that.
1117 : */
1118 0 : if (!task_on_rq_queued(p)) {
1119 0 : replenish_dl_entity(dl_se);
1120 0 : goto unlock;
1121 : }
1122 :
1123 : #ifdef CONFIG_SMP
1124 : if (unlikely(!rq->online)) {
1125 : /*
1126 : * If the runqueue is no longer available, migrate the
1127 : * task elsewhere. This necessarily changes rq.
1128 : */
1129 : lockdep_unpin_lock(__rq_lockp(rq), rf.cookie);
1130 : rq = dl_task_offline_migration(rq, p);
1131 : rf.cookie = lockdep_pin_lock(__rq_lockp(rq));
1132 : update_rq_clock(rq);
1133 :
1134 : /*
1135 : * Now that the task has been migrated to the new RQ and we
1136 : * have that locked, proceed as normal and enqueue the task
1137 : * there.
1138 : */
1139 : }
1140 : #endif
1141 :
1142 0 : enqueue_task_dl(rq, p, ENQUEUE_REPLENISH);
1143 0 : if (dl_task(rq->curr))
1144 : check_preempt_curr_dl(rq, p, 0);
1145 : else
1146 0 : resched_curr(rq);
1147 :
1148 : #ifdef CONFIG_SMP
1149 : /*
1150 : * Queueing this task back might have overloaded rq, check if we need
1151 : * to kick someone away.
1152 : */
1153 : if (has_pushable_dl_tasks(rq)) {
1154 : /*
1155 : * Nothing relies on rq->lock after this, so its safe to drop
1156 : * rq->lock.
1157 : */
1158 : rq_unpin_lock(rq, &rf);
1159 : push_dl_task(rq);
1160 : rq_repin_lock(rq, &rf);
1161 : }
1162 : #endif
1163 :
1164 : unlock:
1165 0 : task_rq_unlock(rq, p, &rf);
1166 :
1167 : /*
1168 : * This can free the task_struct, including this hrtimer, do not touch
1169 : * anything related to that after this.
1170 : */
1171 0 : put_task_struct(p);
1172 :
1173 0 : return HRTIMER_NORESTART;
1174 : }
1175 :
1176 108 : void init_dl_task_timer(struct sched_dl_entity *dl_se)
1177 : {
1178 108 : struct hrtimer *timer = &dl_se->dl_timer;
1179 :
1180 108 : hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1181 108 : timer->function = dl_task_timer;
1182 108 : }
1183 :
1184 : /*
1185 : * During the activation, CBS checks if it can reuse the current task's
1186 : * runtime and period. If the deadline of the task is in the past, CBS
1187 : * cannot use the runtime, and so it replenishes the task. This rule
1188 : * works fine for implicit deadline tasks (deadline == period), and the
1189 : * CBS was designed for implicit deadline tasks. However, a task with
1190 : * constrained deadline (deadline < period) might be awakened after the
1191 : * deadline, but before the next period. In this case, replenishing the
1192 : * task would allow it to run for runtime / deadline. As in this case
1193 : * deadline < period, CBS enables a task to run for more than the
1194 : * runtime / period. In a very loaded system, this can cause a domino
1195 : * effect, making other tasks miss their deadlines.
1196 : *
1197 : * To avoid this problem, in the activation of a constrained deadline
1198 : * task after the deadline but before the next period, throttle the
1199 : * task and set the replenishing timer to the begin of the next period,
1200 : * unless it is boosted.
1201 : */
1202 0 : static inline void dl_check_constrained_dl(struct sched_dl_entity *dl_se)
1203 : {
1204 0 : struct task_struct *p = dl_task_of(dl_se);
1205 0 : struct rq *rq = rq_of_dl_rq(dl_rq_of_se(dl_se));
1206 :
1207 0 : if (dl_time_before(dl_se->deadline, rq_clock(rq)) &&
1208 0 : dl_time_before(rq_clock(rq), dl_next_period(dl_se))) {
1209 0 : if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(p)))
1210 : return;
1211 0 : dl_se->dl_throttled = 1;
1212 0 : if (dl_se->runtime > 0)
1213 0 : dl_se->runtime = 0;
1214 : }
1215 : }
1216 :
1217 : static
1218 : int dl_runtime_exceeded(struct sched_dl_entity *dl_se)
1219 : {
1220 : return (dl_se->runtime <= 0);
1221 : }
1222 :
1223 : extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
1224 :
1225 : /*
1226 : * This function implements the GRUB accounting rule:
1227 : * according to the GRUB reclaiming algorithm, the runtime is
1228 : * not decreased as "dq = -dt", but as
1229 : * "dq = -max{u / Umax, (1 - Uinact - Uextra)} dt",
1230 : * where u is the utilization of the task, Umax is the maximum reclaimable
1231 : * utilization, Uinact is the (per-runqueue) inactive utilization, computed
1232 : * as the difference between the "total runqueue utilization" and the
1233 : * runqueue active utilization, and Uextra is the (per runqueue) extra
1234 : * reclaimable utilization.
1235 : * Since rq->dl.running_bw and rq->dl.this_bw contain utilizations
1236 : * multiplied by 2^BW_SHIFT, the result has to be shifted right by
1237 : * BW_SHIFT.
1238 : * Since rq->dl.bw_ratio contains 1 / Umax multiplied by 2^RATIO_SHIFT,
1239 : * dl_bw is multiped by rq->dl.bw_ratio and shifted right by RATIO_SHIFT.
1240 : * Since delta is a 64 bit variable, to have an overflow its value
1241 : * should be larger than 2^(64 - 20 - 8), which is more than 64 seconds.
1242 : * So, overflow is not an issue here.
1243 : */
1244 : static u64 grub_reclaim(u64 delta, struct rq *rq, struct sched_dl_entity *dl_se)
1245 : {
1246 0 : u64 u_inact = rq->dl.this_bw - rq->dl.running_bw; /* Utot - Uact */
1247 : u64 u_act;
1248 0 : u64 u_act_min = (dl_se->dl_bw * rq->dl.bw_ratio) >> RATIO_SHIFT;
1249 :
1250 : /*
1251 : * Instead of computing max{u * bw_ratio, (1 - u_inact - u_extra)},
1252 : * we compare u_inact + rq->dl.extra_bw with
1253 : * 1 - (u * rq->dl.bw_ratio >> RATIO_SHIFT), because
1254 : * u_inact + rq->dl.extra_bw can be larger than
1255 : * 1 * (so, 1 - u_inact - rq->dl.extra_bw would be negative
1256 : * leading to wrong results)
1257 : */
1258 0 : if (u_inact + rq->dl.extra_bw > BW_UNIT - u_act_min)
1259 : u_act = u_act_min;
1260 : else
1261 0 : u_act = BW_UNIT - u_inact - rq->dl.extra_bw;
1262 :
1263 0 : return (delta * u_act) >> BW_SHIFT;
1264 : }
1265 :
1266 : /*
1267 : * Update the current task's runtime statistics (provided it is still
1268 : * a -deadline task and has not been removed from the dl_rq).
1269 : */
1270 0 : static void update_curr_dl(struct rq *rq)
1271 : {
1272 0 : struct task_struct *curr = rq->curr;
1273 0 : struct sched_dl_entity *dl_se = &curr->dl;
1274 : u64 delta_exec, scaled_delta_exec;
1275 0 : int cpu = cpu_of(rq);
1276 : u64 now;
1277 :
1278 0 : if (!dl_task(curr) || !on_dl_rq(dl_se))
1279 : return;
1280 :
1281 : /*
1282 : * Consumed budget is computed considering the time as
1283 : * observed by schedulable tasks (excluding time spent
1284 : * in hardirq context, etc.). Deadlines are instead
1285 : * computed using hard walltime. This seems to be the more
1286 : * natural solution, but the full ramifications of this
1287 : * approach need further study.
1288 : */
1289 0 : now = rq_clock_task(rq);
1290 0 : delta_exec = now - curr->se.exec_start;
1291 0 : if (unlikely((s64)delta_exec <= 0)) {
1292 0 : if (unlikely(dl_se->dl_yielded))
1293 : goto throttle;
1294 : return;
1295 : }
1296 :
1297 : schedstat_set(curr->stats.exec_max,
1298 : max(curr->stats.exec_max, delta_exec));
1299 :
1300 0 : trace_sched_stat_runtime(curr, delta_exec, 0);
1301 :
1302 0 : curr->se.sum_exec_runtime += delta_exec;
1303 0 : account_group_exec_runtime(curr, delta_exec);
1304 :
1305 0 : curr->se.exec_start = now;
1306 0 : cgroup_account_cputime(curr, delta_exec);
1307 :
1308 0 : if (dl_entity_is_special(dl_se))
1309 : return;
1310 :
1311 : /*
1312 : * For tasks that participate in GRUB, we implement GRUB-PA: the
1313 : * spare reclaimed bandwidth is used to clock down frequency.
1314 : *
1315 : * For the others, we still need to scale reservation parameters
1316 : * according to current frequency and CPU maximum capacity.
1317 : */
1318 0 : if (unlikely(dl_se->flags & SCHED_FLAG_RECLAIM)) {
1319 0 : scaled_delta_exec = grub_reclaim(delta_exec,
1320 : rq,
1321 : &curr->dl);
1322 : } else {
1323 0 : unsigned long scale_freq = arch_scale_freq_capacity(cpu);
1324 0 : unsigned long scale_cpu = arch_scale_cpu_capacity(cpu);
1325 :
1326 0 : scaled_delta_exec = cap_scale(delta_exec, scale_freq);
1327 0 : scaled_delta_exec = cap_scale(scaled_delta_exec, scale_cpu);
1328 : }
1329 :
1330 0 : dl_se->runtime -= scaled_delta_exec;
1331 :
1332 : throttle:
1333 0 : if (dl_runtime_exceeded(dl_se) || dl_se->dl_yielded) {
1334 0 : dl_se->dl_throttled = 1;
1335 :
1336 : /* If requested, inform the user about runtime overruns. */
1337 0 : if (dl_runtime_exceeded(dl_se) &&
1338 0 : (dl_se->flags & SCHED_FLAG_DL_OVERRUN))
1339 0 : dl_se->dl_overrun = 1;
1340 :
1341 0 : __dequeue_task_dl(rq, curr, 0);
1342 0 : if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(curr)))
1343 0 : enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH);
1344 :
1345 0 : if (!is_leftmost(curr, &rq->dl))
1346 0 : resched_curr(rq);
1347 : }
1348 :
1349 : /*
1350 : * Because -- for now -- we share the rt bandwidth, we need to
1351 : * account our runtime there too, otherwise actual rt tasks
1352 : * would be able to exceed the shared quota.
1353 : *
1354 : * Account to the root rt group for now.
1355 : *
1356 : * The solution we're working towards is having the RT groups scheduled
1357 : * using deadline servers -- however there's a few nasties to figure
1358 : * out before that can happen.
1359 : */
1360 0 : if (rt_bandwidth_enabled()) {
1361 0 : struct rt_rq *rt_rq = &rq->rt;
1362 :
1363 0 : raw_spin_lock(&rt_rq->rt_runtime_lock);
1364 : /*
1365 : * We'll let actual RT tasks worry about the overflow here, we
1366 : * have our own CBS to keep us inline; only account when RT
1367 : * bandwidth is relevant.
1368 : */
1369 0 : if (sched_rt_bandwidth_account(rt_rq))
1370 0 : rt_rq->rt_time += delta_exec;
1371 0 : raw_spin_unlock(&rt_rq->rt_runtime_lock);
1372 : }
1373 : }
1374 :
1375 0 : static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer)
1376 : {
1377 0 : struct sched_dl_entity *dl_se = container_of(timer,
1378 : struct sched_dl_entity,
1379 : inactive_timer);
1380 0 : struct task_struct *p = dl_task_of(dl_se);
1381 : struct rq_flags rf;
1382 : struct rq *rq;
1383 :
1384 0 : rq = task_rq_lock(p, &rf);
1385 :
1386 : sched_clock_tick();
1387 0 : update_rq_clock(rq);
1388 :
1389 0 : if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
1390 0 : struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
1391 :
1392 0 : if (READ_ONCE(p->__state) == TASK_DEAD && dl_se->dl_non_contending) {
1393 0 : sub_running_bw(&p->dl, dl_rq_of_se(&p->dl));
1394 0 : sub_rq_bw(&p->dl, dl_rq_of_se(&p->dl));
1395 0 : dl_se->dl_non_contending = 0;
1396 : }
1397 :
1398 0 : raw_spin_lock(&dl_b->lock);
1399 0 : __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
1400 0 : raw_spin_unlock(&dl_b->lock);
1401 : __dl_clear_params(p);
1402 :
1403 : goto unlock;
1404 : }
1405 0 : if (dl_se->dl_non_contending == 0)
1406 : goto unlock;
1407 :
1408 0 : sub_running_bw(dl_se, &rq->dl);
1409 0 : dl_se->dl_non_contending = 0;
1410 : unlock:
1411 0 : task_rq_unlock(rq, p, &rf);
1412 0 : put_task_struct(p);
1413 :
1414 0 : return HRTIMER_NORESTART;
1415 : }
1416 :
1417 108 : void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se)
1418 : {
1419 108 : struct hrtimer *timer = &dl_se->inactive_timer;
1420 :
1421 108 : hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1422 108 : timer->function = inactive_task_timer;
1423 108 : }
1424 :
1425 : #define __node_2_dle(node) \
1426 : rb_entry((node), struct sched_dl_entity, rb_node)
1427 :
1428 : #ifdef CONFIG_SMP
1429 :
1430 : static void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
1431 : {
1432 : struct rq *rq = rq_of_dl_rq(dl_rq);
1433 :
1434 : if (dl_rq->earliest_dl.curr == 0 ||
1435 : dl_time_before(deadline, dl_rq->earliest_dl.curr)) {
1436 : if (dl_rq->earliest_dl.curr == 0)
1437 : cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_HIGHER);
1438 : dl_rq->earliest_dl.curr = deadline;
1439 : cpudl_set(&rq->rd->cpudl, rq->cpu, deadline);
1440 : }
1441 : }
1442 :
1443 : static void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
1444 : {
1445 : struct rq *rq = rq_of_dl_rq(dl_rq);
1446 :
1447 : /*
1448 : * Since we may have removed our earliest (and/or next earliest)
1449 : * task we must recompute them.
1450 : */
1451 : if (!dl_rq->dl_nr_running) {
1452 : dl_rq->earliest_dl.curr = 0;
1453 : dl_rq->earliest_dl.next = 0;
1454 : cpudl_clear(&rq->rd->cpudl, rq->cpu);
1455 : cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1456 : } else {
1457 : struct rb_node *leftmost = rb_first_cached(&dl_rq->root);
1458 : struct sched_dl_entity *entry = __node_2_dle(leftmost);
1459 :
1460 : dl_rq->earliest_dl.curr = entry->deadline;
1461 : cpudl_set(&rq->rd->cpudl, rq->cpu, entry->deadline);
1462 : }
1463 : }
1464 :
1465 : #else
1466 :
1467 : static inline void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
1468 : static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
1469 :
1470 : #endif /* CONFIG_SMP */
1471 :
1472 : static inline
1473 0 : void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
1474 : {
1475 0 : int prio = dl_task_of(dl_se)->prio;
1476 0 : u64 deadline = dl_se->deadline;
1477 :
1478 0 : WARN_ON(!dl_prio(prio));
1479 0 : dl_rq->dl_nr_running++;
1480 0 : add_nr_running(rq_of_dl_rq(dl_rq), 1);
1481 :
1482 0 : inc_dl_deadline(dl_rq, deadline);
1483 0 : inc_dl_migration(dl_se, dl_rq);
1484 0 : }
1485 :
1486 : static inline
1487 0 : void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
1488 : {
1489 0 : int prio = dl_task_of(dl_se)->prio;
1490 :
1491 0 : WARN_ON(!dl_prio(prio));
1492 0 : WARN_ON(!dl_rq->dl_nr_running);
1493 0 : dl_rq->dl_nr_running--;
1494 0 : sub_nr_running(rq_of_dl_rq(dl_rq), 1);
1495 :
1496 0 : dec_dl_deadline(dl_rq, dl_se->deadline);
1497 0 : dec_dl_migration(dl_se, dl_rq);
1498 0 : }
1499 :
1500 : static inline bool __dl_less(struct rb_node *a, const struct rb_node *b)
1501 : {
1502 0 : return dl_time_before(__node_2_dle(a)->deadline, __node_2_dle(b)->deadline);
1503 : }
1504 :
1505 : static inline struct sched_statistics *
1506 : __schedstats_from_dl_se(struct sched_dl_entity *dl_se)
1507 : {
1508 : return &dl_task_of(dl_se)->stats;
1509 : }
1510 :
1511 : static inline void
1512 : update_stats_wait_start_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1513 : {
1514 : struct sched_statistics *stats;
1515 :
1516 : if (!schedstat_enabled())
1517 : return;
1518 :
1519 : stats = __schedstats_from_dl_se(dl_se);
1520 : __update_stats_wait_start(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1521 : }
1522 :
1523 : static inline void
1524 : update_stats_wait_end_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1525 : {
1526 : struct sched_statistics *stats;
1527 :
1528 : if (!schedstat_enabled())
1529 : return;
1530 :
1531 : stats = __schedstats_from_dl_se(dl_se);
1532 : __update_stats_wait_end(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1533 : }
1534 :
1535 : static inline void
1536 : update_stats_enqueue_sleeper_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1537 : {
1538 : struct sched_statistics *stats;
1539 :
1540 : if (!schedstat_enabled())
1541 : return;
1542 :
1543 : stats = __schedstats_from_dl_se(dl_se);
1544 : __update_stats_enqueue_sleeper(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1545 : }
1546 :
1547 : static inline void
1548 : update_stats_enqueue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
1549 : int flags)
1550 : {
1551 : if (!schedstat_enabled())
1552 : return;
1553 :
1554 : if (flags & ENQUEUE_WAKEUP)
1555 : update_stats_enqueue_sleeper_dl(dl_rq, dl_se);
1556 : }
1557 :
1558 : static inline void
1559 : update_stats_dequeue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
1560 : int flags)
1561 : {
1562 0 : struct task_struct *p = dl_task_of(dl_se);
1563 :
1564 : if (!schedstat_enabled())
1565 : return;
1566 :
1567 : if ((flags & DEQUEUE_SLEEP)) {
1568 : unsigned int state;
1569 :
1570 : state = READ_ONCE(p->__state);
1571 : if (state & TASK_INTERRUPTIBLE)
1572 : __schedstat_set(p->stats.sleep_start,
1573 : rq_clock(rq_of_dl_rq(dl_rq)));
1574 :
1575 : if (state & TASK_UNINTERRUPTIBLE)
1576 : __schedstat_set(p->stats.block_start,
1577 : rq_clock(rq_of_dl_rq(dl_rq)));
1578 : }
1579 : }
1580 :
1581 0 : static void __enqueue_dl_entity(struct sched_dl_entity *dl_se)
1582 : {
1583 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
1584 :
1585 0 : BUG_ON(!RB_EMPTY_NODE(&dl_se->rb_node));
1586 :
1587 0 : rb_add_cached(&dl_se->rb_node, &dl_rq->root, __dl_less);
1588 :
1589 0 : inc_dl_tasks(dl_se, dl_rq);
1590 0 : }
1591 :
1592 0 : static void __dequeue_dl_entity(struct sched_dl_entity *dl_se)
1593 : {
1594 0 : struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
1595 :
1596 0 : if (RB_EMPTY_NODE(&dl_se->rb_node))
1597 : return;
1598 :
1599 0 : rb_erase_cached(&dl_se->rb_node, &dl_rq->root);
1600 :
1601 0 : RB_CLEAR_NODE(&dl_se->rb_node);
1602 :
1603 0 : dec_dl_tasks(dl_se, dl_rq);
1604 : }
1605 :
1606 : static void
1607 0 : enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags)
1608 : {
1609 0 : BUG_ON(on_dl_rq(dl_se));
1610 :
1611 0 : update_stats_enqueue_dl(dl_rq_of_se(dl_se), dl_se, flags);
1612 :
1613 : /*
1614 : * If this is a wakeup or a new instance, the scheduling
1615 : * parameters of the task might need updating. Otherwise,
1616 : * we want a replenishment of its runtime.
1617 : */
1618 0 : if (flags & ENQUEUE_WAKEUP) {
1619 0 : task_contending(dl_se, flags);
1620 0 : update_dl_entity(dl_se);
1621 0 : } else if (flags & ENQUEUE_REPLENISH) {
1622 0 : replenish_dl_entity(dl_se);
1623 0 : } else if ((flags & ENQUEUE_RESTORE) &&
1624 0 : dl_time_before(dl_se->deadline,
1625 : rq_clock(rq_of_dl_rq(dl_rq_of_se(dl_se))))) {
1626 0 : setup_new_dl_entity(dl_se);
1627 : }
1628 :
1629 0 : __enqueue_dl_entity(dl_se);
1630 0 : }
1631 :
1632 : static void dequeue_dl_entity(struct sched_dl_entity *dl_se)
1633 : {
1634 0 : __dequeue_dl_entity(dl_se);
1635 : }
1636 :
1637 0 : static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1638 : {
1639 0 : if (is_dl_boosted(&p->dl)) {
1640 : /*
1641 : * Because of delays in the detection of the overrun of a
1642 : * thread's runtime, it might be the case that a thread
1643 : * goes to sleep in a rt mutex with negative runtime. As
1644 : * a consequence, the thread will be throttled.
1645 : *
1646 : * While waiting for the mutex, this thread can also be
1647 : * boosted via PI, resulting in a thread that is throttled
1648 : * and boosted at the same time.
1649 : *
1650 : * In this case, the boost overrides the throttle.
1651 : */
1652 0 : if (p->dl.dl_throttled) {
1653 : /*
1654 : * The replenish timer needs to be canceled. No
1655 : * problem if it fires concurrently: boosted threads
1656 : * are ignored in dl_task_timer().
1657 : */
1658 0 : hrtimer_try_to_cancel(&p->dl.dl_timer);
1659 0 : p->dl.dl_throttled = 0;
1660 : }
1661 0 : } else if (!dl_prio(p->normal_prio)) {
1662 : /*
1663 : * Special case in which we have a !SCHED_DEADLINE task that is going
1664 : * to be deboosted, but exceeds its runtime while doing so. No point in
1665 : * replenishing it, as it's going to return back to its original
1666 : * scheduling class after this. If it has been throttled, we need to
1667 : * clear the flag, otherwise the task may wake up as throttled after
1668 : * being boosted again with no means to replenish the runtime and clear
1669 : * the throttle.
1670 : */
1671 0 : p->dl.dl_throttled = 0;
1672 0 : BUG_ON(!is_dl_boosted(&p->dl) || flags != ENQUEUE_REPLENISH);
1673 : return;
1674 : }
1675 :
1676 : /*
1677 : * Check if a constrained deadline task was activated
1678 : * after the deadline but before the next period.
1679 : * If that is the case, the task will be throttled and
1680 : * the replenishment timer will be set to the next period.
1681 : */
1682 0 : if (!p->dl.dl_throttled && !dl_is_implicit(&p->dl))
1683 0 : dl_check_constrained_dl(&p->dl);
1684 :
1685 0 : if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & ENQUEUE_RESTORE) {
1686 0 : add_rq_bw(&p->dl, &rq->dl);
1687 0 : add_running_bw(&p->dl, &rq->dl);
1688 : }
1689 :
1690 : /*
1691 : * If p is throttled, we do not enqueue it. In fact, if it exhausted
1692 : * its budget it needs a replenishment and, since it now is on
1693 : * its rq, the bandwidth timer callback (which clearly has not
1694 : * run yet) will take care of this.
1695 : * However, the active utilization does not depend on the fact
1696 : * that the task is on the runqueue or not (but depends on the
1697 : * task's state - in GRUB parlance, "inactive" vs "active contending").
1698 : * In other words, even if a task is throttled its utilization must
1699 : * be counted in the active utilization; hence, we need to call
1700 : * add_running_bw().
1701 : */
1702 0 : if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH)) {
1703 0 : if (flags & ENQUEUE_WAKEUP)
1704 0 : task_contending(&p->dl, flags);
1705 :
1706 : return;
1707 : }
1708 :
1709 : check_schedstat_required();
1710 0 : update_stats_wait_start_dl(dl_rq_of_se(&p->dl), &p->dl);
1711 :
1712 0 : enqueue_dl_entity(&p->dl, flags);
1713 :
1714 0 : if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1715 : enqueue_pushable_dl_task(rq, p);
1716 : }
1717 :
1718 : static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1719 : {
1720 0 : update_stats_dequeue_dl(&rq->dl, &p->dl, flags);
1721 0 : dequeue_dl_entity(&p->dl);
1722 0 : dequeue_pushable_dl_task(rq, p);
1723 : }
1724 :
1725 0 : static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1726 : {
1727 0 : update_curr_dl(rq);
1728 0 : __dequeue_task_dl(rq, p, flags);
1729 :
1730 0 : if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & DEQUEUE_SAVE) {
1731 0 : sub_running_bw(&p->dl, &rq->dl);
1732 0 : sub_rq_bw(&p->dl, &rq->dl);
1733 : }
1734 :
1735 : /*
1736 : * This check allows to start the inactive timer (or to immediately
1737 : * decrease the active utilization, if needed) in two cases:
1738 : * when the task blocks and when it is terminating
1739 : * (p->state == TASK_DEAD). We can handle the two cases in the same
1740 : * way, because from GRUB's point of view the same thing is happening
1741 : * (the task moves from "active contending" to "active non contending"
1742 : * or "inactive")
1743 : */
1744 0 : if (flags & DEQUEUE_SLEEP)
1745 0 : task_non_contending(p);
1746 0 : }
1747 :
1748 : /*
1749 : * Yield task semantic for -deadline tasks is:
1750 : *
1751 : * get off from the CPU until our next instance, with
1752 : * a new runtime. This is of little use now, since we
1753 : * don't have a bandwidth reclaiming mechanism. Anyway,
1754 : * bandwidth reclaiming is planned for the future, and
1755 : * yield_task_dl will indicate that some spare budget
1756 : * is available for other task instances to use it.
1757 : */
1758 0 : static void yield_task_dl(struct rq *rq)
1759 : {
1760 : /*
1761 : * We make the task go to sleep until its current deadline by
1762 : * forcing its runtime to zero. This way, update_curr_dl() stops
1763 : * it and the bandwidth timer will wake it up and will give it
1764 : * new scheduling parameters (thanks to dl_yielded=1).
1765 : */
1766 0 : rq->curr->dl.dl_yielded = 1;
1767 :
1768 0 : update_rq_clock(rq);
1769 0 : update_curr_dl(rq);
1770 : /*
1771 : * Tell update_rq_clock() that we've just updated,
1772 : * so we don't do microscopic update in schedule()
1773 : * and double the fastpath cost.
1774 : */
1775 0 : rq_clock_skip_update(rq);
1776 0 : }
1777 :
1778 : #ifdef CONFIG_SMP
1779 :
1780 : static int find_later_rq(struct task_struct *task);
1781 :
1782 : static int
1783 : select_task_rq_dl(struct task_struct *p, int cpu, int flags)
1784 : {
1785 : struct task_struct *curr;
1786 : bool select_rq;
1787 : struct rq *rq;
1788 :
1789 : if (!(flags & WF_TTWU))
1790 : goto out;
1791 :
1792 : rq = cpu_rq(cpu);
1793 :
1794 : rcu_read_lock();
1795 : curr = READ_ONCE(rq->curr); /* unlocked access */
1796 :
1797 : /*
1798 : * If we are dealing with a -deadline task, we must
1799 : * decide where to wake it up.
1800 : * If it has a later deadline and the current task
1801 : * on this rq can't move (provided the waking task
1802 : * can!) we prefer to send it somewhere else. On the
1803 : * other hand, if it has a shorter deadline, we
1804 : * try to make it stay here, it might be important.
1805 : */
1806 : select_rq = unlikely(dl_task(curr)) &&
1807 : (curr->nr_cpus_allowed < 2 ||
1808 : !dl_entity_preempt(&p->dl, &curr->dl)) &&
1809 : p->nr_cpus_allowed > 1;
1810 :
1811 : /*
1812 : * Take the capacity of the CPU into account to
1813 : * ensure it fits the requirement of the task.
1814 : */
1815 : if (static_branch_unlikely(&sched_asym_cpucapacity))
1816 : select_rq |= !dl_task_fits_capacity(p, cpu);
1817 :
1818 : if (select_rq) {
1819 : int target = find_later_rq(p);
1820 :
1821 : if (target != -1 &&
1822 : (dl_time_before(p->dl.deadline,
1823 : cpu_rq(target)->dl.earliest_dl.curr) ||
1824 : (cpu_rq(target)->dl.dl_nr_running == 0)))
1825 : cpu = target;
1826 : }
1827 : rcu_read_unlock();
1828 :
1829 : out:
1830 : return cpu;
1831 : }
1832 :
1833 : static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused)
1834 : {
1835 : struct rq *rq;
1836 :
1837 : if (READ_ONCE(p->__state) != TASK_WAKING)
1838 : return;
1839 :
1840 : rq = task_rq(p);
1841 : /*
1842 : * Since p->state == TASK_WAKING, set_task_cpu() has been called
1843 : * from try_to_wake_up(). Hence, p->pi_lock is locked, but
1844 : * rq->lock is not... So, lock it
1845 : */
1846 : raw_spin_rq_lock(rq);
1847 : if (p->dl.dl_non_contending) {
1848 : update_rq_clock(rq);
1849 : sub_running_bw(&p->dl, &rq->dl);
1850 : p->dl.dl_non_contending = 0;
1851 : /*
1852 : * If the timer handler is currently running and the
1853 : * timer cannot be canceled, inactive_task_timer()
1854 : * will see that dl_not_contending is not set, and
1855 : * will not touch the rq's active utilization,
1856 : * so we are still safe.
1857 : */
1858 : if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
1859 : put_task_struct(p);
1860 : }
1861 : sub_rq_bw(&p->dl, &rq->dl);
1862 : raw_spin_rq_unlock(rq);
1863 : }
1864 :
1865 : static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
1866 : {
1867 : /*
1868 : * Current can't be migrated, useless to reschedule,
1869 : * let's hope p can move out.
1870 : */
1871 : if (rq->curr->nr_cpus_allowed == 1 ||
1872 : !cpudl_find(&rq->rd->cpudl, rq->curr, NULL))
1873 : return;
1874 :
1875 : /*
1876 : * p is migratable, so let's not schedule it and
1877 : * see if it is pushed or pulled somewhere else.
1878 : */
1879 : if (p->nr_cpus_allowed != 1 &&
1880 : cpudl_find(&rq->rd->cpudl, p, NULL))
1881 : return;
1882 :
1883 : resched_curr(rq);
1884 : }
1885 :
1886 : static int balance_dl(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1887 : {
1888 : if (!on_dl_rq(&p->dl) && need_pull_dl_task(rq, p)) {
1889 : /*
1890 : * This is OK, because current is on_cpu, which avoids it being
1891 : * picked for load-balance and preemption/IRQs are still
1892 : * disabled avoiding further scheduler activity on it and we've
1893 : * not yet started the picking loop.
1894 : */
1895 : rq_unpin_lock(rq, rf);
1896 : pull_dl_task(rq);
1897 : rq_repin_lock(rq, rf);
1898 : }
1899 :
1900 : return sched_stop_runnable(rq) || sched_dl_runnable(rq);
1901 : }
1902 : #endif /* CONFIG_SMP */
1903 :
1904 : /*
1905 : * Only called when both the current and waking task are -deadline
1906 : * tasks.
1907 : */
1908 0 : static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
1909 : int flags)
1910 : {
1911 0 : if (dl_entity_preempt(&p->dl, &rq->curr->dl)) {
1912 0 : resched_curr(rq);
1913 0 : return;
1914 : }
1915 :
1916 : #ifdef CONFIG_SMP
1917 : /*
1918 : * In the unlikely case current and p have the same deadline
1919 : * let us try to decide what's the best thing to do...
1920 : */
1921 : if ((p->dl.deadline == rq->curr->dl.deadline) &&
1922 : !test_tsk_need_resched(rq->curr))
1923 : check_preempt_equal_dl(rq, p);
1924 : #endif /* CONFIG_SMP */
1925 : }
1926 :
1927 : #ifdef CONFIG_SCHED_HRTICK
1928 : static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1929 : {
1930 : hrtick_start(rq, p->dl.runtime);
1931 : }
1932 : #else /* !CONFIG_SCHED_HRTICK */
1933 : static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1934 : {
1935 : }
1936 : #endif
1937 :
1938 0 : static void set_next_task_dl(struct rq *rq, struct task_struct *p, bool first)
1939 : {
1940 0 : struct sched_dl_entity *dl_se = &p->dl;
1941 0 : struct dl_rq *dl_rq = &rq->dl;
1942 :
1943 0 : p->se.exec_start = rq_clock_task(rq);
1944 0 : if (on_dl_rq(&p->dl))
1945 : update_stats_wait_end_dl(dl_rq, dl_se);
1946 :
1947 : /* You can't push away the running task */
1948 0 : dequeue_pushable_dl_task(rq, p);
1949 :
1950 0 : if (!first)
1951 : return;
1952 :
1953 0 : if (hrtick_enabled_dl(rq))
1954 : start_hrtick_dl(rq, p);
1955 :
1956 0 : if (rq->curr->sched_class != &dl_sched_class)
1957 0 : update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0);
1958 :
1959 : deadline_queue_push_tasks(rq);
1960 : }
1961 :
1962 : static struct sched_dl_entity *pick_next_dl_entity(struct dl_rq *dl_rq)
1963 : {
1964 0 : struct rb_node *left = rb_first_cached(&dl_rq->root);
1965 :
1966 0 : if (!left)
1967 : return NULL;
1968 :
1969 0 : return __node_2_dle(left);
1970 : }
1971 :
1972 0 : static struct task_struct *pick_task_dl(struct rq *rq)
1973 : {
1974 : struct sched_dl_entity *dl_se;
1975 0 : struct dl_rq *dl_rq = &rq->dl;
1976 : struct task_struct *p;
1977 :
1978 0 : if (!sched_dl_runnable(rq))
1979 : return NULL;
1980 :
1981 0 : dl_se = pick_next_dl_entity(dl_rq);
1982 0 : BUG_ON(!dl_se);
1983 0 : p = dl_task_of(dl_se);
1984 :
1985 0 : return p;
1986 : }
1987 :
1988 0 : static struct task_struct *pick_next_task_dl(struct rq *rq)
1989 : {
1990 : struct task_struct *p;
1991 :
1992 0 : p = pick_task_dl(rq);
1993 0 : if (p)
1994 0 : set_next_task_dl(rq, p, true);
1995 :
1996 0 : return p;
1997 : }
1998 :
1999 0 : static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
2000 : {
2001 0 : struct sched_dl_entity *dl_se = &p->dl;
2002 0 : struct dl_rq *dl_rq = &rq->dl;
2003 :
2004 0 : if (on_dl_rq(&p->dl))
2005 : update_stats_wait_start_dl(dl_rq, dl_se);
2006 :
2007 0 : update_curr_dl(rq);
2008 :
2009 0 : update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 1);
2010 0 : if (on_dl_rq(&p->dl) && p->nr_cpus_allowed > 1)
2011 : enqueue_pushable_dl_task(rq, p);
2012 0 : }
2013 :
2014 : /*
2015 : * scheduler tick hitting a task of our scheduling class.
2016 : *
2017 : * NOTE: This function can be called remotely by the tick offload that
2018 : * goes along full dynticks. Therefore no local assumption can be made
2019 : * and everything must be accessed through the @rq and @curr passed in
2020 : * parameters.
2021 : */
2022 0 : static void task_tick_dl(struct rq *rq, struct task_struct *p, int queued)
2023 : {
2024 0 : update_curr_dl(rq);
2025 :
2026 0 : update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 1);
2027 : /*
2028 : * Even when we have runtime, update_curr_dl() might have resulted in us
2029 : * not being the leftmost task anymore. In that case NEED_RESCHED will
2030 : * be set and schedule() will start a new hrtick for the next task.
2031 : */
2032 0 : if (hrtick_enabled_dl(rq) && queued && p->dl.runtime > 0 &&
2033 : is_leftmost(p, &rq->dl))
2034 : start_hrtick_dl(rq, p);
2035 0 : }
2036 :
2037 0 : static void task_fork_dl(struct task_struct *p)
2038 : {
2039 : /*
2040 : * SCHED_DEADLINE tasks cannot fork and this is achieved through
2041 : * sched_fork()
2042 : */
2043 0 : }
2044 :
2045 : #ifdef CONFIG_SMP
2046 :
2047 : /* Only try algorithms three times */
2048 : #define DL_MAX_TRIES 3
2049 :
2050 : static int pick_dl_task(struct rq *rq, struct task_struct *p, int cpu)
2051 : {
2052 : if (!task_running(rq, p) &&
2053 : cpumask_test_cpu(cpu, &p->cpus_mask))
2054 : return 1;
2055 : return 0;
2056 : }
2057 :
2058 : /*
2059 : * Return the earliest pushable rq's task, which is suitable to be executed
2060 : * on the CPU, NULL otherwise:
2061 : */
2062 : static struct task_struct *pick_earliest_pushable_dl_task(struct rq *rq, int cpu)
2063 : {
2064 : struct task_struct *p = NULL;
2065 : struct rb_node *next_node;
2066 :
2067 : if (!has_pushable_dl_tasks(rq))
2068 : return NULL;
2069 :
2070 : next_node = rb_first_cached(&rq->dl.pushable_dl_tasks_root);
2071 :
2072 : next_node:
2073 : if (next_node) {
2074 : p = __node_2_pdl(next_node);
2075 :
2076 : if (pick_dl_task(rq, p, cpu))
2077 : return p;
2078 :
2079 : next_node = rb_next(next_node);
2080 : goto next_node;
2081 : }
2082 :
2083 : return NULL;
2084 : }
2085 :
2086 : static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
2087 :
2088 : static int find_later_rq(struct task_struct *task)
2089 : {
2090 : struct sched_domain *sd;
2091 : struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
2092 : int this_cpu = smp_processor_id();
2093 : int cpu = task_cpu(task);
2094 :
2095 : /* Make sure the mask is initialized first */
2096 : if (unlikely(!later_mask))
2097 : return -1;
2098 :
2099 : if (task->nr_cpus_allowed == 1)
2100 : return -1;
2101 :
2102 : /*
2103 : * We have to consider system topology and task affinity
2104 : * first, then we can look for a suitable CPU.
2105 : */
2106 : if (!cpudl_find(&task_rq(task)->rd->cpudl, task, later_mask))
2107 : return -1;
2108 :
2109 : /*
2110 : * If we are here, some targets have been found, including
2111 : * the most suitable which is, among the runqueues where the
2112 : * current tasks have later deadlines than the task's one, the
2113 : * rq with the latest possible one.
2114 : *
2115 : * Now we check how well this matches with task's
2116 : * affinity and system topology.
2117 : *
2118 : * The last CPU where the task run is our first
2119 : * guess, since it is most likely cache-hot there.
2120 : */
2121 : if (cpumask_test_cpu(cpu, later_mask))
2122 : return cpu;
2123 : /*
2124 : * Check if this_cpu is to be skipped (i.e., it is
2125 : * not in the mask) or not.
2126 : */
2127 : if (!cpumask_test_cpu(this_cpu, later_mask))
2128 : this_cpu = -1;
2129 :
2130 : rcu_read_lock();
2131 : for_each_domain(cpu, sd) {
2132 : if (sd->flags & SD_WAKE_AFFINE) {
2133 : int best_cpu;
2134 :
2135 : /*
2136 : * If possible, preempting this_cpu is
2137 : * cheaper than migrating.
2138 : */
2139 : if (this_cpu != -1 &&
2140 : cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
2141 : rcu_read_unlock();
2142 : return this_cpu;
2143 : }
2144 :
2145 : best_cpu = cpumask_any_and_distribute(later_mask,
2146 : sched_domain_span(sd));
2147 : /*
2148 : * Last chance: if a CPU being in both later_mask
2149 : * and current sd span is valid, that becomes our
2150 : * choice. Of course, the latest possible CPU is
2151 : * already under consideration through later_mask.
2152 : */
2153 : if (best_cpu < nr_cpu_ids) {
2154 : rcu_read_unlock();
2155 : return best_cpu;
2156 : }
2157 : }
2158 : }
2159 : rcu_read_unlock();
2160 :
2161 : /*
2162 : * At this point, all our guesses failed, we just return
2163 : * 'something', and let the caller sort the things out.
2164 : */
2165 : if (this_cpu != -1)
2166 : return this_cpu;
2167 :
2168 : cpu = cpumask_any_distribute(later_mask);
2169 : if (cpu < nr_cpu_ids)
2170 : return cpu;
2171 :
2172 : return -1;
2173 : }
2174 :
2175 : /* Locks the rq it finds */
2176 : static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq)
2177 : {
2178 : struct rq *later_rq = NULL;
2179 : int tries;
2180 : int cpu;
2181 :
2182 : for (tries = 0; tries < DL_MAX_TRIES; tries++) {
2183 : cpu = find_later_rq(task);
2184 :
2185 : if ((cpu == -1) || (cpu == rq->cpu))
2186 : break;
2187 :
2188 : later_rq = cpu_rq(cpu);
2189 :
2190 : if (later_rq->dl.dl_nr_running &&
2191 : !dl_time_before(task->dl.deadline,
2192 : later_rq->dl.earliest_dl.curr)) {
2193 : /*
2194 : * Target rq has tasks of equal or earlier deadline,
2195 : * retrying does not release any lock and is unlikely
2196 : * to yield a different result.
2197 : */
2198 : later_rq = NULL;
2199 : break;
2200 : }
2201 :
2202 : /* Retry if something changed. */
2203 : if (double_lock_balance(rq, later_rq)) {
2204 : if (unlikely(task_rq(task) != rq ||
2205 : !cpumask_test_cpu(later_rq->cpu, &task->cpus_mask) ||
2206 : task_running(rq, task) ||
2207 : !dl_task(task) ||
2208 : !task_on_rq_queued(task))) {
2209 : double_unlock_balance(rq, later_rq);
2210 : later_rq = NULL;
2211 : break;
2212 : }
2213 : }
2214 :
2215 : /*
2216 : * If the rq we found has no -deadline task, or
2217 : * its earliest one has a later deadline than our
2218 : * task, the rq is a good one.
2219 : */
2220 : if (!later_rq->dl.dl_nr_running ||
2221 : dl_time_before(task->dl.deadline,
2222 : later_rq->dl.earliest_dl.curr))
2223 : break;
2224 :
2225 : /* Otherwise we try again. */
2226 : double_unlock_balance(rq, later_rq);
2227 : later_rq = NULL;
2228 : }
2229 :
2230 : return later_rq;
2231 : }
2232 :
2233 : static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
2234 : {
2235 : struct task_struct *p;
2236 :
2237 : if (!has_pushable_dl_tasks(rq))
2238 : return NULL;
2239 :
2240 : p = __node_2_pdl(rb_first_cached(&rq->dl.pushable_dl_tasks_root));
2241 :
2242 : BUG_ON(rq->cpu != task_cpu(p));
2243 : BUG_ON(task_current(rq, p));
2244 : BUG_ON(p->nr_cpus_allowed <= 1);
2245 :
2246 : BUG_ON(!task_on_rq_queued(p));
2247 : BUG_ON(!dl_task(p));
2248 :
2249 : return p;
2250 : }
2251 :
2252 : /*
2253 : * See if the non running -deadline tasks on this rq
2254 : * can be sent to some other CPU where they can preempt
2255 : * and start executing.
2256 : */
2257 : static int push_dl_task(struct rq *rq)
2258 : {
2259 : struct task_struct *next_task;
2260 : struct rq *later_rq;
2261 : int ret = 0;
2262 :
2263 : if (!rq->dl.overloaded)
2264 : return 0;
2265 :
2266 : next_task = pick_next_pushable_dl_task(rq);
2267 : if (!next_task)
2268 : return 0;
2269 :
2270 : retry:
2271 : /*
2272 : * If next_task preempts rq->curr, and rq->curr
2273 : * can move away, it makes sense to just reschedule
2274 : * without going further in pushing next_task.
2275 : */
2276 : if (dl_task(rq->curr) &&
2277 : dl_time_before(next_task->dl.deadline, rq->curr->dl.deadline) &&
2278 : rq->curr->nr_cpus_allowed > 1) {
2279 : resched_curr(rq);
2280 : return 0;
2281 : }
2282 :
2283 : if (is_migration_disabled(next_task))
2284 : return 0;
2285 :
2286 : if (WARN_ON(next_task == rq->curr))
2287 : return 0;
2288 :
2289 : /* We might release rq lock */
2290 : get_task_struct(next_task);
2291 :
2292 : /* Will lock the rq it'll find */
2293 : later_rq = find_lock_later_rq(next_task, rq);
2294 : if (!later_rq) {
2295 : struct task_struct *task;
2296 :
2297 : /*
2298 : * We must check all this again, since
2299 : * find_lock_later_rq releases rq->lock and it is
2300 : * then possible that next_task has migrated.
2301 : */
2302 : task = pick_next_pushable_dl_task(rq);
2303 : if (task == next_task) {
2304 : /*
2305 : * The task is still there. We don't try
2306 : * again, some other CPU will pull it when ready.
2307 : */
2308 : goto out;
2309 : }
2310 :
2311 : if (!task)
2312 : /* No more tasks */
2313 : goto out;
2314 :
2315 : put_task_struct(next_task);
2316 : next_task = task;
2317 : goto retry;
2318 : }
2319 :
2320 : deactivate_task(rq, next_task, 0);
2321 : set_task_cpu(next_task, later_rq->cpu);
2322 :
2323 : /*
2324 : * Update the later_rq clock here, because the clock is used
2325 : * by the cpufreq_update_util() inside __add_running_bw().
2326 : */
2327 : update_rq_clock(later_rq);
2328 : activate_task(later_rq, next_task, ENQUEUE_NOCLOCK);
2329 : ret = 1;
2330 :
2331 : resched_curr(later_rq);
2332 :
2333 : double_unlock_balance(rq, later_rq);
2334 :
2335 : out:
2336 : put_task_struct(next_task);
2337 :
2338 : return ret;
2339 : }
2340 :
2341 : static void push_dl_tasks(struct rq *rq)
2342 : {
2343 : /* push_dl_task() will return true if it moved a -deadline task */
2344 : while (push_dl_task(rq))
2345 : ;
2346 : }
2347 :
2348 : static void pull_dl_task(struct rq *this_rq)
2349 : {
2350 : int this_cpu = this_rq->cpu, cpu;
2351 : struct task_struct *p, *push_task;
2352 : bool resched = false;
2353 : struct rq *src_rq;
2354 : u64 dmin = LONG_MAX;
2355 :
2356 : if (likely(!dl_overloaded(this_rq)))
2357 : return;
2358 :
2359 : /*
2360 : * Match the barrier from dl_set_overloaded; this guarantees that if we
2361 : * see overloaded we must also see the dlo_mask bit.
2362 : */
2363 : smp_rmb();
2364 :
2365 : for_each_cpu(cpu, this_rq->rd->dlo_mask) {
2366 : if (this_cpu == cpu)
2367 : continue;
2368 :
2369 : src_rq = cpu_rq(cpu);
2370 :
2371 : /*
2372 : * It looks racy, abd it is! However, as in sched_rt.c,
2373 : * we are fine with this.
2374 : */
2375 : if (this_rq->dl.dl_nr_running &&
2376 : dl_time_before(this_rq->dl.earliest_dl.curr,
2377 : src_rq->dl.earliest_dl.next))
2378 : continue;
2379 :
2380 : /* Might drop this_rq->lock */
2381 : push_task = NULL;
2382 : double_lock_balance(this_rq, src_rq);
2383 :
2384 : /*
2385 : * If there are no more pullable tasks on the
2386 : * rq, we're done with it.
2387 : */
2388 : if (src_rq->dl.dl_nr_running <= 1)
2389 : goto skip;
2390 :
2391 : p = pick_earliest_pushable_dl_task(src_rq, this_cpu);
2392 :
2393 : /*
2394 : * We found a task to be pulled if:
2395 : * - it preempts our current (if there's one),
2396 : * - it will preempt the last one we pulled (if any).
2397 : */
2398 : if (p && dl_time_before(p->dl.deadline, dmin) &&
2399 : (!this_rq->dl.dl_nr_running ||
2400 : dl_time_before(p->dl.deadline,
2401 : this_rq->dl.earliest_dl.curr))) {
2402 : WARN_ON(p == src_rq->curr);
2403 : WARN_ON(!task_on_rq_queued(p));
2404 :
2405 : /*
2406 : * Then we pull iff p has actually an earlier
2407 : * deadline than the current task of its runqueue.
2408 : */
2409 : if (dl_time_before(p->dl.deadline,
2410 : src_rq->curr->dl.deadline))
2411 : goto skip;
2412 :
2413 : if (is_migration_disabled(p)) {
2414 : push_task = get_push_task(src_rq);
2415 : } else {
2416 : deactivate_task(src_rq, p, 0);
2417 : set_task_cpu(p, this_cpu);
2418 : activate_task(this_rq, p, 0);
2419 : dmin = p->dl.deadline;
2420 : resched = true;
2421 : }
2422 :
2423 : /* Is there any other task even earlier? */
2424 : }
2425 : skip:
2426 : double_unlock_balance(this_rq, src_rq);
2427 :
2428 : if (push_task) {
2429 : raw_spin_rq_unlock(this_rq);
2430 : stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
2431 : push_task, &src_rq->push_work);
2432 : raw_spin_rq_lock(this_rq);
2433 : }
2434 : }
2435 :
2436 : if (resched)
2437 : resched_curr(this_rq);
2438 : }
2439 :
2440 : /*
2441 : * Since the task is not running and a reschedule is not going to happen
2442 : * anytime soon on its runqueue, we try pushing it away now.
2443 : */
2444 : static void task_woken_dl(struct rq *rq, struct task_struct *p)
2445 : {
2446 : if (!task_running(rq, p) &&
2447 : !test_tsk_need_resched(rq->curr) &&
2448 : p->nr_cpus_allowed > 1 &&
2449 : dl_task(rq->curr) &&
2450 : (rq->curr->nr_cpus_allowed < 2 ||
2451 : !dl_entity_preempt(&p->dl, &rq->curr->dl))) {
2452 : push_dl_tasks(rq);
2453 : }
2454 : }
2455 :
2456 : static void set_cpus_allowed_dl(struct task_struct *p,
2457 : const struct cpumask *new_mask,
2458 : u32 flags)
2459 : {
2460 : struct root_domain *src_rd;
2461 : struct rq *rq;
2462 :
2463 : BUG_ON(!dl_task(p));
2464 :
2465 : rq = task_rq(p);
2466 : src_rd = rq->rd;
2467 : /*
2468 : * Migrating a SCHED_DEADLINE task between exclusive
2469 : * cpusets (different root_domains) entails a bandwidth
2470 : * update. We already made space for us in the destination
2471 : * domain (see cpuset_can_attach()).
2472 : */
2473 : if (!cpumask_intersects(src_rd->span, new_mask)) {
2474 : struct dl_bw *src_dl_b;
2475 :
2476 : src_dl_b = dl_bw_of(cpu_of(rq));
2477 : /*
2478 : * We now free resources of the root_domain we are migrating
2479 : * off. In the worst case, sched_setattr() may temporary fail
2480 : * until we complete the update.
2481 : */
2482 : raw_spin_lock(&src_dl_b->lock);
2483 : __dl_sub(src_dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
2484 : raw_spin_unlock(&src_dl_b->lock);
2485 : }
2486 :
2487 : set_cpus_allowed_common(p, new_mask, flags);
2488 : }
2489 :
2490 : /* Assumes rq->lock is held */
2491 : static void rq_online_dl(struct rq *rq)
2492 : {
2493 : if (rq->dl.overloaded)
2494 : dl_set_overload(rq);
2495 :
2496 : cpudl_set_freecpu(&rq->rd->cpudl, rq->cpu);
2497 : if (rq->dl.dl_nr_running > 0)
2498 : cpudl_set(&rq->rd->cpudl, rq->cpu, rq->dl.earliest_dl.curr);
2499 : }
2500 :
2501 : /* Assumes rq->lock is held */
2502 : static void rq_offline_dl(struct rq *rq)
2503 : {
2504 : if (rq->dl.overloaded)
2505 : dl_clear_overload(rq);
2506 :
2507 : cpudl_clear(&rq->rd->cpudl, rq->cpu);
2508 : cpudl_clear_freecpu(&rq->rd->cpudl, rq->cpu);
2509 : }
2510 :
2511 : void __init init_sched_dl_class(void)
2512 : {
2513 : unsigned int i;
2514 :
2515 : for_each_possible_cpu(i)
2516 : zalloc_cpumask_var_node(&per_cpu(local_cpu_mask_dl, i),
2517 : GFP_KERNEL, cpu_to_node(i));
2518 : }
2519 :
2520 : void dl_add_task_root_domain(struct task_struct *p)
2521 : {
2522 : struct rq_flags rf;
2523 : struct rq *rq;
2524 : struct dl_bw *dl_b;
2525 :
2526 : raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2527 : if (!dl_task(p)) {
2528 : raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags);
2529 : return;
2530 : }
2531 :
2532 : rq = __task_rq_lock(p, &rf);
2533 :
2534 : dl_b = &rq->rd->dl_bw;
2535 : raw_spin_lock(&dl_b->lock);
2536 :
2537 : __dl_add(dl_b, p->dl.dl_bw, cpumask_weight(rq->rd->span));
2538 :
2539 : raw_spin_unlock(&dl_b->lock);
2540 :
2541 : task_rq_unlock(rq, p, &rf);
2542 : }
2543 :
2544 : void dl_clear_root_domain(struct root_domain *rd)
2545 : {
2546 : unsigned long flags;
2547 :
2548 : raw_spin_lock_irqsave(&rd->dl_bw.lock, flags);
2549 : rd->dl_bw.total_bw = 0;
2550 : raw_spin_unlock_irqrestore(&rd->dl_bw.lock, flags);
2551 : }
2552 :
2553 : #endif /* CONFIG_SMP */
2554 :
2555 0 : static void switched_from_dl(struct rq *rq, struct task_struct *p)
2556 : {
2557 : /*
2558 : * task_non_contending() can start the "inactive timer" (if the 0-lag
2559 : * time is in the future). If the task switches back to dl before
2560 : * the "inactive timer" fires, it can continue to consume its current
2561 : * runtime using its current deadline. If it stays outside of
2562 : * SCHED_DEADLINE until the 0-lag time passes, inactive_task_timer()
2563 : * will reset the task parameters.
2564 : */
2565 0 : if (task_on_rq_queued(p) && p->dl.dl_runtime)
2566 0 : task_non_contending(p);
2567 :
2568 0 : if (!task_on_rq_queued(p)) {
2569 : /*
2570 : * Inactive timer is armed. However, p is leaving DEADLINE and
2571 : * might migrate away from this rq while continuing to run on
2572 : * some other class. We need to remove its contribution from
2573 : * this rq running_bw now, or sub_rq_bw (below) will complain.
2574 : */
2575 0 : if (p->dl.dl_non_contending)
2576 0 : sub_running_bw(&p->dl, &rq->dl);
2577 0 : sub_rq_bw(&p->dl, &rq->dl);
2578 : }
2579 :
2580 : /*
2581 : * We cannot use inactive_task_timer() to invoke sub_running_bw()
2582 : * at the 0-lag time, because the task could have been migrated
2583 : * while SCHED_OTHER in the meanwhile.
2584 : */
2585 0 : if (p->dl.dl_non_contending)
2586 0 : p->dl.dl_non_contending = 0;
2587 :
2588 : /*
2589 : * Since this might be the only -deadline task on the rq,
2590 : * this is the right place to try to pull some other one
2591 : * from an overloaded CPU, if any.
2592 : */
2593 0 : if (!task_on_rq_queued(p) || rq->dl.dl_nr_running)
2594 : return;
2595 :
2596 : deadline_queue_pull_task(rq);
2597 : }
2598 :
2599 : /*
2600 : * When switching to -deadline, we may overload the rq, then
2601 : * we try to push someone off, if possible.
2602 : */
2603 0 : static void switched_to_dl(struct rq *rq, struct task_struct *p)
2604 : {
2605 0 : if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
2606 0 : put_task_struct(p);
2607 :
2608 : /* If p is not queued we will update its parameters at next wakeup. */
2609 0 : if (!task_on_rq_queued(p)) {
2610 0 : add_rq_bw(&p->dl, &rq->dl);
2611 :
2612 : return;
2613 : }
2614 :
2615 0 : if (rq->curr != p) {
2616 : #ifdef CONFIG_SMP
2617 : if (p->nr_cpus_allowed > 1 && rq->dl.overloaded)
2618 : deadline_queue_push_tasks(rq);
2619 : #endif
2620 0 : if (dl_task(rq->curr))
2621 : check_preempt_curr_dl(rq, p, 0);
2622 : else
2623 0 : resched_curr(rq);
2624 : } else {
2625 0 : update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0);
2626 : }
2627 : }
2628 :
2629 : /*
2630 : * If the scheduling parameters of a -deadline task changed,
2631 : * a push or pull operation might be needed.
2632 : */
2633 0 : static void prio_changed_dl(struct rq *rq, struct task_struct *p,
2634 : int oldprio)
2635 : {
2636 0 : if (task_on_rq_queued(p) || task_current(rq, p)) {
2637 : #ifdef CONFIG_SMP
2638 : /*
2639 : * This might be too much, but unfortunately
2640 : * we don't have the old deadline value, and
2641 : * we can't argue if the task is increasing
2642 : * or lowering its prio, so...
2643 : */
2644 : if (!rq->dl.overloaded)
2645 : deadline_queue_pull_task(rq);
2646 :
2647 : /*
2648 : * If we now have a earlier deadline task than p,
2649 : * then reschedule, provided p is still on this
2650 : * runqueue.
2651 : */
2652 : if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline))
2653 : resched_curr(rq);
2654 : #else
2655 : /*
2656 : * Again, we don't know if p has a earlier
2657 : * or later deadline, so let's blindly set a
2658 : * (maybe not needed) rescheduling point.
2659 : */
2660 0 : resched_curr(rq);
2661 : #endif /* CONFIG_SMP */
2662 : }
2663 0 : }
2664 :
2665 : DEFINE_SCHED_CLASS(dl) = {
2666 :
2667 : .enqueue_task = enqueue_task_dl,
2668 : .dequeue_task = dequeue_task_dl,
2669 : .yield_task = yield_task_dl,
2670 :
2671 : .check_preempt_curr = check_preempt_curr_dl,
2672 :
2673 : .pick_next_task = pick_next_task_dl,
2674 : .put_prev_task = put_prev_task_dl,
2675 : .set_next_task = set_next_task_dl,
2676 :
2677 : #ifdef CONFIG_SMP
2678 : .balance = balance_dl,
2679 : .pick_task = pick_task_dl,
2680 : .select_task_rq = select_task_rq_dl,
2681 : .migrate_task_rq = migrate_task_rq_dl,
2682 : .set_cpus_allowed = set_cpus_allowed_dl,
2683 : .rq_online = rq_online_dl,
2684 : .rq_offline = rq_offline_dl,
2685 : .task_woken = task_woken_dl,
2686 : .find_lock_rq = find_lock_later_rq,
2687 : #endif
2688 :
2689 : .task_tick = task_tick_dl,
2690 : .task_fork = task_fork_dl,
2691 :
2692 : .prio_changed = prio_changed_dl,
2693 : .switched_from = switched_from_dl,
2694 : .switched_to = switched_to_dl,
2695 :
2696 : .update_curr = update_curr_dl,
2697 : };
2698 :
2699 : /* Used for dl_bw check and update, used under sched_rt_handler()::mutex */
2700 : static u64 dl_generation;
2701 :
2702 0 : int sched_dl_global_validate(void)
2703 : {
2704 0 : u64 runtime = global_rt_runtime();
2705 0 : u64 period = global_rt_period();
2706 0 : u64 new_bw = to_ratio(period, runtime);
2707 0 : u64 gen = ++dl_generation;
2708 : struct dl_bw *dl_b;
2709 0 : int cpu, cpus, ret = 0;
2710 : unsigned long flags;
2711 :
2712 : /*
2713 : * Here we want to check the bandwidth not being set to some
2714 : * value smaller than the currently allocated bandwidth in
2715 : * any of the root_domains.
2716 : */
2717 0 : for_each_possible_cpu(cpu) {
2718 : rcu_read_lock_sched();
2719 :
2720 0 : if (dl_bw_visited(cpu, gen))
2721 : goto next;
2722 :
2723 0 : dl_b = dl_bw_of(cpu);
2724 0 : cpus = dl_bw_cpus(cpu);
2725 :
2726 0 : raw_spin_lock_irqsave(&dl_b->lock, flags);
2727 0 : if (new_bw * cpus < dl_b->total_bw)
2728 0 : ret = -EBUSY;
2729 0 : raw_spin_unlock_irqrestore(&dl_b->lock, flags);
2730 :
2731 : next:
2732 : rcu_read_unlock_sched();
2733 :
2734 0 : if (ret)
2735 : break;
2736 : }
2737 :
2738 0 : return ret;
2739 : }
2740 :
2741 1 : static void init_dl_rq_bw_ratio(struct dl_rq *dl_rq)
2742 : {
2743 1 : if (global_rt_runtime() == RUNTIME_INF) {
2744 0 : dl_rq->bw_ratio = 1 << RATIO_SHIFT;
2745 0 : dl_rq->extra_bw = 1 << BW_SHIFT;
2746 : } else {
2747 2 : dl_rq->bw_ratio = to_ratio(global_rt_runtime(),
2748 1 : global_rt_period()) >> (BW_SHIFT - RATIO_SHIFT);
2749 1 : dl_rq->extra_bw = to_ratio(global_rt_period(),
2750 : global_rt_runtime());
2751 : }
2752 1 : }
2753 :
2754 0 : void sched_dl_do_global(void)
2755 : {
2756 0 : u64 new_bw = -1;
2757 0 : u64 gen = ++dl_generation;
2758 : struct dl_bw *dl_b;
2759 : int cpu;
2760 : unsigned long flags;
2761 :
2762 0 : if (global_rt_runtime() != RUNTIME_INF)
2763 0 : new_bw = to_ratio(global_rt_period(), global_rt_runtime());
2764 :
2765 0 : for_each_possible_cpu(cpu) {
2766 : rcu_read_lock_sched();
2767 :
2768 0 : if (dl_bw_visited(cpu, gen)) {
2769 : rcu_read_unlock_sched();
2770 : continue;
2771 : }
2772 :
2773 0 : dl_b = dl_bw_of(cpu);
2774 :
2775 0 : raw_spin_lock_irqsave(&dl_b->lock, flags);
2776 0 : dl_b->bw = new_bw;
2777 0 : raw_spin_unlock_irqrestore(&dl_b->lock, flags);
2778 :
2779 : rcu_read_unlock_sched();
2780 0 : init_dl_rq_bw_ratio(&cpu_rq(cpu)->dl);
2781 : }
2782 0 : }
2783 :
2784 : /*
2785 : * We must be sure that accepting a new task (or allowing changing the
2786 : * parameters of an existing one) is consistent with the bandwidth
2787 : * constraints. If yes, this function also accordingly updates the currently
2788 : * allocated bandwidth to reflect the new situation.
2789 : *
2790 : * This function is called while holding p's rq->lock.
2791 : */
2792 0 : int sched_dl_overflow(struct task_struct *p, int policy,
2793 : const struct sched_attr *attr)
2794 : {
2795 0 : u64 period = attr->sched_period ?: attr->sched_deadline;
2796 0 : u64 runtime = attr->sched_runtime;
2797 0 : u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2798 0 : int cpus, err = -1, cpu = task_cpu(p);
2799 0 : struct dl_bw *dl_b = dl_bw_of(cpu);
2800 : unsigned long cap;
2801 :
2802 0 : if (attr->sched_flags & SCHED_FLAG_SUGOV)
2803 : return 0;
2804 :
2805 : /* !deadline task may carry old deadline bandwidth */
2806 0 : if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2807 : return 0;
2808 :
2809 : /*
2810 : * Either if a task, enters, leave, or stays -deadline but changes
2811 : * its parameters, we may need to update accordingly the total
2812 : * allocated bandwidth of the container.
2813 : */
2814 0 : raw_spin_lock(&dl_b->lock);
2815 0 : cpus = dl_bw_cpus(cpu);
2816 0 : cap = dl_bw_capacity(cpu);
2817 :
2818 0 : if (dl_policy(policy) && !task_has_dl_policy(p) &&
2819 0 : !__dl_overflow(dl_b, cap, 0, new_bw)) {
2820 0 : if (hrtimer_active(&p->dl.inactive_timer))
2821 0 : __dl_sub(dl_b, p->dl.dl_bw, cpus);
2822 0 : __dl_add(dl_b, new_bw, cpus);
2823 0 : err = 0;
2824 0 : } else if (dl_policy(policy) && task_has_dl_policy(p) &&
2825 0 : !__dl_overflow(dl_b, cap, p->dl.dl_bw, new_bw)) {
2826 : /*
2827 : * XXX this is slightly incorrect: when the task
2828 : * utilization decreases, we should delay the total
2829 : * utilization change until the task's 0-lag point.
2830 : * But this would require to set the task's "inactive
2831 : * timer" when the task is not inactive.
2832 : */
2833 0 : __dl_sub(dl_b, p->dl.dl_bw, cpus);
2834 0 : __dl_add(dl_b, new_bw, cpus);
2835 0 : dl_change_utilization(p, new_bw);
2836 0 : err = 0;
2837 0 : } else if (!dl_policy(policy) && task_has_dl_policy(p)) {
2838 : /*
2839 : * Do not decrease the total deadline utilization here,
2840 : * switched_from_dl() will take care to do it at the correct
2841 : * (0-lag) time.
2842 : */
2843 0 : err = 0;
2844 : }
2845 0 : raw_spin_unlock(&dl_b->lock);
2846 :
2847 0 : return err;
2848 : }
2849 :
2850 : /*
2851 : * This function initializes the sched_dl_entity of a newly becoming
2852 : * SCHED_DEADLINE task.
2853 : *
2854 : * Only the static values are considered here, the actual runtime and the
2855 : * absolute deadline will be properly calculated when the task is enqueued
2856 : * for the first time with its new policy.
2857 : */
2858 0 : void __setparam_dl(struct task_struct *p, const struct sched_attr *attr)
2859 : {
2860 0 : struct sched_dl_entity *dl_se = &p->dl;
2861 :
2862 0 : dl_se->dl_runtime = attr->sched_runtime;
2863 0 : dl_se->dl_deadline = attr->sched_deadline;
2864 0 : dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
2865 0 : dl_se->flags = attr->sched_flags & SCHED_DL_FLAGS;
2866 0 : dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
2867 0 : dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime);
2868 0 : }
2869 :
2870 0 : void __getparam_dl(struct task_struct *p, struct sched_attr *attr)
2871 : {
2872 0 : struct sched_dl_entity *dl_se = &p->dl;
2873 :
2874 0 : attr->sched_priority = p->rt_priority;
2875 0 : attr->sched_runtime = dl_se->dl_runtime;
2876 0 : attr->sched_deadline = dl_se->dl_deadline;
2877 0 : attr->sched_period = dl_se->dl_period;
2878 0 : attr->sched_flags &= ~SCHED_DL_FLAGS;
2879 0 : attr->sched_flags |= dl_se->flags;
2880 0 : }
2881 :
2882 : /*
2883 : * Default limits for DL period; on the top end we guard against small util
2884 : * tasks still getting ridiculously long effective runtimes, on the bottom end we
2885 : * guard against timer DoS.
2886 : */
2887 : unsigned int sysctl_sched_dl_period_max = 1 << 22; /* ~4 seconds */
2888 : unsigned int sysctl_sched_dl_period_min = 100; /* 100 us */
2889 :
2890 : /*
2891 : * This function validates the new parameters of a -deadline task.
2892 : * We ask for the deadline not being zero, and greater or equal
2893 : * than the runtime, as well as the period of being zero or
2894 : * greater than deadline. Furthermore, we have to be sure that
2895 : * user parameters are above the internal resolution of 1us (we
2896 : * check sched_runtime only since it is always the smaller one) and
2897 : * below 2^63 ns (we have to check both sched_deadline and
2898 : * sched_period, as the latter can be zero).
2899 : */
2900 0 : bool __checkparam_dl(const struct sched_attr *attr)
2901 : {
2902 : u64 period, max, min;
2903 :
2904 : /* special dl tasks don't actually use any parameter */
2905 0 : if (attr->sched_flags & SCHED_FLAG_SUGOV)
2906 : return true;
2907 :
2908 : /* deadline != 0 */
2909 0 : if (attr->sched_deadline == 0)
2910 : return false;
2911 :
2912 : /*
2913 : * Since we truncate DL_SCALE bits, make sure we're at least
2914 : * that big.
2915 : */
2916 0 : if (attr->sched_runtime < (1ULL << DL_SCALE))
2917 : return false;
2918 :
2919 : /*
2920 : * Since we use the MSB for wrap-around and sign issues, make
2921 : * sure it's not set (mind that period can be equal to zero).
2922 : */
2923 0 : if (attr->sched_deadline & (1ULL << 63) ||
2924 0 : attr->sched_period & (1ULL << 63))
2925 : return false;
2926 :
2927 0 : period = attr->sched_period;
2928 0 : if (!period)
2929 0 : period = attr->sched_deadline;
2930 :
2931 : /* runtime <= deadline <= period (if period != 0) */
2932 0 : if (period < attr->sched_deadline ||
2933 : attr->sched_deadline < attr->sched_runtime)
2934 : return false;
2935 :
2936 0 : max = (u64)READ_ONCE(sysctl_sched_dl_period_max) * NSEC_PER_USEC;
2937 0 : min = (u64)READ_ONCE(sysctl_sched_dl_period_min) * NSEC_PER_USEC;
2938 :
2939 0 : if (period < min || period > max)
2940 : return false;
2941 :
2942 0 : return true;
2943 : }
2944 :
2945 : /*
2946 : * This function clears the sched_dl_entity static params.
2947 : */
2948 108 : void __dl_clear_params(struct task_struct *p)
2949 : {
2950 108 : struct sched_dl_entity *dl_se = &p->dl;
2951 :
2952 108 : dl_se->dl_runtime = 0;
2953 108 : dl_se->dl_deadline = 0;
2954 108 : dl_se->dl_period = 0;
2955 108 : dl_se->flags = 0;
2956 108 : dl_se->dl_bw = 0;
2957 108 : dl_se->dl_density = 0;
2958 :
2959 108 : dl_se->dl_throttled = 0;
2960 108 : dl_se->dl_yielded = 0;
2961 108 : dl_se->dl_non_contending = 0;
2962 108 : dl_se->dl_overrun = 0;
2963 :
2964 : #ifdef CONFIG_RT_MUTEXES
2965 108 : dl_se->pi_se = dl_se;
2966 : #endif
2967 108 : }
2968 :
2969 0 : bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr)
2970 : {
2971 0 : struct sched_dl_entity *dl_se = &p->dl;
2972 :
2973 0 : if (dl_se->dl_runtime != attr->sched_runtime ||
2974 0 : dl_se->dl_deadline != attr->sched_deadline ||
2975 0 : dl_se->dl_period != attr->sched_period ||
2976 0 : dl_se->flags != (attr->sched_flags & SCHED_DL_FLAGS))
2977 : return true;
2978 :
2979 0 : return false;
2980 : }
2981 :
2982 : #ifdef CONFIG_SMP
2983 : int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur,
2984 : const struct cpumask *trial)
2985 : {
2986 : int ret = 1, trial_cpus;
2987 : struct dl_bw *cur_dl_b;
2988 : unsigned long flags;
2989 :
2990 : rcu_read_lock_sched();
2991 : cur_dl_b = dl_bw_of(cpumask_any(cur));
2992 : trial_cpus = cpumask_weight(trial);
2993 :
2994 : raw_spin_lock_irqsave(&cur_dl_b->lock, flags);
2995 : if (cur_dl_b->bw != -1 &&
2996 : cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw)
2997 : ret = 0;
2998 : raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags);
2999 : rcu_read_unlock_sched();
3000 :
3001 : return ret;
3002 : }
3003 :
3004 : int dl_cpu_busy(int cpu, struct task_struct *p)
3005 : {
3006 : unsigned long flags, cap;
3007 : struct dl_bw *dl_b;
3008 : bool overflow;
3009 :
3010 : rcu_read_lock_sched();
3011 : dl_b = dl_bw_of(cpu);
3012 : raw_spin_lock_irqsave(&dl_b->lock, flags);
3013 : cap = dl_bw_capacity(cpu);
3014 : overflow = __dl_overflow(dl_b, cap, 0, p ? p->dl.dl_bw : 0);
3015 :
3016 : if (!overflow && p) {
3017 : /*
3018 : * We reserve space for this task in the destination
3019 : * root_domain, as we can't fail after this point.
3020 : * We will free resources in the source root_domain
3021 : * later on (see set_cpus_allowed_dl()).
3022 : */
3023 : __dl_add(dl_b, p->dl.dl_bw, dl_bw_cpus(cpu));
3024 : }
3025 :
3026 : raw_spin_unlock_irqrestore(&dl_b->lock, flags);
3027 : rcu_read_unlock_sched();
3028 :
3029 : return overflow ? -EBUSY : 0;
3030 : }
3031 : #endif
3032 :
3033 : #ifdef CONFIG_SCHED_DEBUG
3034 0 : void print_dl_stats(struct seq_file *m, int cpu)
3035 : {
3036 0 : print_dl_rq(m, cpu, &cpu_rq(cpu)->dl);
3037 0 : }
3038 : #endif /* CONFIG_SCHED_DEBUG */
|
