blob: a2ce59015642c3ccc753006837a9485b2d9fbcd3 [file] [log] [blame]
Dario Faggioliaab03e02013-11-28 11:14:43 +01001/*
2 * Deadline Scheduling Class (SCHED_DEADLINE)
3 *
4 * Earliest Deadline First (EDF) + Constant Bandwidth Server (CBS).
5 *
6 * Tasks that periodically executes their instances for less than their
7 * runtime won't miss any of their deadlines.
8 * Tasks that are not periodic or sporadic or that tries to execute more
9 * than their reserved bandwidth will be slowed down (and may potentially
10 * miss some of their deadlines), and won't affect any other task.
11 *
12 * Copyright (C) 2012 Dario Faggioli <raistlin@linux.it>,
Juri Lelli1baca4c2013-11-07 14:43:38 +010013 * Juri Lelli <juri.lelli@gmail.com>,
Dario Faggioliaab03e02013-11-28 11:14:43 +010014 * Michael Trimarchi <michael@amarulasolutions.com>,
15 * Fabio Checconi <fchecconi@gmail.com>
16 */
17#include "sched.h"
18
Juri Lelli6bfd6d72013-11-07 14:43:47 +010019#include <linux/slab.h>
20
Dario Faggioli332ac172013-11-07 14:43:45 +010021struct dl_bandwidth def_dl_bandwidth;
22
Dario Faggioliaab03e02013-11-28 11:14:43 +010023static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
24{
25 return container_of(dl_se, struct task_struct, dl);
26}
27
28static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq)
29{
30 return container_of(dl_rq, struct rq, dl);
31}
32
33static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
34{
35 struct task_struct *p = dl_task_of(dl_se);
36 struct rq *rq = task_rq(p);
37
38 return &rq->dl;
39}
40
41static inline int on_dl_rq(struct sched_dl_entity *dl_se)
42{
43 return !RB_EMPTY_NODE(&dl_se->rb_node);
44}
45
46static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
47{
48 struct sched_dl_entity *dl_se = &p->dl;
49
50 return dl_rq->rb_leftmost == &dl_se->rb_node;
51}
52
Dario Faggioli332ac172013-11-07 14:43:45 +010053void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime)
54{
55 raw_spin_lock_init(&dl_b->dl_runtime_lock);
56 dl_b->dl_period = period;
57 dl_b->dl_runtime = runtime;
58}
59
Dario Faggioli332ac172013-11-07 14:43:45 +010060void init_dl_bw(struct dl_bw *dl_b)
61{
62 raw_spin_lock_init(&dl_b->lock);
63 raw_spin_lock(&def_dl_bandwidth.dl_runtime_lock);
Peter Zijlstra17248132013-12-17 12:44:49 +010064 if (global_rt_runtime() == RUNTIME_INF)
Dario Faggioli332ac172013-11-07 14:43:45 +010065 dl_b->bw = -1;
66 else
Peter Zijlstra17248132013-12-17 12:44:49 +010067 dl_b->bw = to_ratio(global_rt_period(), global_rt_runtime());
Dario Faggioli332ac172013-11-07 14:43:45 +010068 raw_spin_unlock(&def_dl_bandwidth.dl_runtime_lock);
69 dl_b->total_bw = 0;
70}
71
Abel Vesa07c54f72015-03-03 13:50:27 +020072void init_dl_rq(struct dl_rq *dl_rq)
Dario Faggioliaab03e02013-11-28 11:14:43 +010073{
74 dl_rq->rb_root = RB_ROOT;
Juri Lelli1baca4c2013-11-07 14:43:38 +010075
76#ifdef CONFIG_SMP
77 /* zero means no -deadline tasks */
78 dl_rq->earliest_dl.curr = dl_rq->earliest_dl.next = 0;
79
80 dl_rq->dl_nr_migratory = 0;
81 dl_rq->overloaded = 0;
82 dl_rq->pushable_dl_tasks_root = RB_ROOT;
Dario Faggioli332ac172013-11-07 14:43:45 +010083#else
84 init_dl_bw(&dl_rq->dl_bw);
Juri Lelli1baca4c2013-11-07 14:43:38 +010085#endif
Dario Faggioliaab03e02013-11-28 11:14:43 +010086}
87
Juri Lelli1baca4c2013-11-07 14:43:38 +010088#ifdef CONFIG_SMP
89
90static inline int dl_overloaded(struct rq *rq)
91{
92 return atomic_read(&rq->rd->dlo_count);
93}
94
95static inline void dl_set_overload(struct rq *rq)
96{
97 if (!rq->online)
98 return;
99
100 cpumask_set_cpu(rq->cpu, rq->rd->dlo_mask);
101 /*
102 * Must be visible before the overload count is
103 * set (as in sched_rt.c).
104 *
105 * Matched by the barrier in pull_dl_task().
106 */
107 smp_wmb();
108 atomic_inc(&rq->rd->dlo_count);
109}
110
111static inline void dl_clear_overload(struct rq *rq)
112{
113 if (!rq->online)
114 return;
115
116 atomic_dec(&rq->rd->dlo_count);
117 cpumask_clear_cpu(rq->cpu, rq->rd->dlo_mask);
118}
119
120static void update_dl_migration(struct dl_rq *dl_rq)
121{
Kirill Tkhai995b9ea2014-02-18 02:24:13 +0400122 if (dl_rq->dl_nr_migratory && dl_rq->dl_nr_running > 1) {
Juri Lelli1baca4c2013-11-07 14:43:38 +0100123 if (!dl_rq->overloaded) {
124 dl_set_overload(rq_of_dl_rq(dl_rq));
125 dl_rq->overloaded = 1;
126 }
127 } else if (dl_rq->overloaded) {
128 dl_clear_overload(rq_of_dl_rq(dl_rq));
129 dl_rq->overloaded = 0;
130 }
131}
132
133static void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
134{
135 struct task_struct *p = dl_task_of(dl_se);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100136
Ingo Molnar4b53a342017-02-05 15:41:03 +0100137 if (p->nr_cpus_allowed > 1)
Juri Lelli1baca4c2013-11-07 14:43:38 +0100138 dl_rq->dl_nr_migratory++;
139
140 update_dl_migration(dl_rq);
141}
142
143static void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
144{
145 struct task_struct *p = dl_task_of(dl_se);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100146
Ingo Molnar4b53a342017-02-05 15:41:03 +0100147 if (p->nr_cpus_allowed > 1)
Juri Lelli1baca4c2013-11-07 14:43:38 +0100148 dl_rq->dl_nr_migratory--;
149
150 update_dl_migration(dl_rq);
151}
152
153/*
154 * The list of pushable -deadline task is not a plist, like in
155 * sched_rt.c, it is an rb-tree with tasks ordered by deadline.
156 */
157static void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
158{
159 struct dl_rq *dl_rq = &rq->dl;
160 struct rb_node **link = &dl_rq->pushable_dl_tasks_root.rb_node;
161 struct rb_node *parent = NULL;
162 struct task_struct *entry;
163 int leftmost = 1;
164
165 BUG_ON(!RB_EMPTY_NODE(&p->pushable_dl_tasks));
166
167 while (*link) {
168 parent = *link;
169 entry = rb_entry(parent, struct task_struct,
170 pushable_dl_tasks);
171 if (dl_entity_preempt(&p->dl, &entry->dl))
172 link = &parent->rb_left;
173 else {
174 link = &parent->rb_right;
175 leftmost = 0;
176 }
177 }
178
Wanpeng Li7d92de32015-12-03 17:42:10 +0800179 if (leftmost) {
Juri Lelli1baca4c2013-11-07 14:43:38 +0100180 dl_rq->pushable_dl_tasks_leftmost = &p->pushable_dl_tasks;
Wanpeng Li7d92de32015-12-03 17:42:10 +0800181 dl_rq->earliest_dl.next = p->dl.deadline;
182 }
Juri Lelli1baca4c2013-11-07 14:43:38 +0100183
184 rb_link_node(&p->pushable_dl_tasks, parent, link);
185 rb_insert_color(&p->pushable_dl_tasks, &dl_rq->pushable_dl_tasks_root);
186}
187
188static void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
189{
190 struct dl_rq *dl_rq = &rq->dl;
191
192 if (RB_EMPTY_NODE(&p->pushable_dl_tasks))
193 return;
194
195 if (dl_rq->pushable_dl_tasks_leftmost == &p->pushable_dl_tasks) {
196 struct rb_node *next_node;
197
198 next_node = rb_next(&p->pushable_dl_tasks);
199 dl_rq->pushable_dl_tasks_leftmost = next_node;
Wanpeng Li7d92de32015-12-03 17:42:10 +0800200 if (next_node) {
201 dl_rq->earliest_dl.next = rb_entry(next_node,
202 struct task_struct, pushable_dl_tasks)->dl.deadline;
203 }
Juri Lelli1baca4c2013-11-07 14:43:38 +0100204 }
205
206 rb_erase(&p->pushable_dl_tasks, &dl_rq->pushable_dl_tasks_root);
207 RB_CLEAR_NODE(&p->pushable_dl_tasks);
208}
209
210static inline int has_pushable_dl_tasks(struct rq *rq)
211{
212 return !RB_EMPTY_ROOT(&rq->dl.pushable_dl_tasks_root);
213}
214
215static int push_dl_task(struct rq *rq);
216
Peter Zijlstradc877342014-02-12 15:47:29 +0100217static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
218{
219 return dl_task(prev);
220}
221
Peter Zijlstra9916e212015-06-11 14:46:43 +0200222static DEFINE_PER_CPU(struct callback_head, dl_push_head);
223static DEFINE_PER_CPU(struct callback_head, dl_pull_head);
Peter Zijlstrae3fca9e2015-06-11 14:46:37 +0200224
225static void push_dl_tasks(struct rq *);
Peter Zijlstra9916e212015-06-11 14:46:43 +0200226static void pull_dl_task(struct rq *);
Peter Zijlstrae3fca9e2015-06-11 14:46:37 +0200227
228static inline void queue_push_tasks(struct rq *rq)
Peter Zijlstradc877342014-02-12 15:47:29 +0100229{
Peter Zijlstrae3fca9e2015-06-11 14:46:37 +0200230 if (!has_pushable_dl_tasks(rq))
231 return;
232
Peter Zijlstra9916e212015-06-11 14:46:43 +0200233 queue_balance_callback(rq, &per_cpu(dl_push_head, rq->cpu), push_dl_tasks);
234}
235
236static inline void queue_pull_task(struct rq *rq)
237{
238 queue_balance_callback(rq, &per_cpu(dl_pull_head, rq->cpu), pull_dl_task);
Peter Zijlstradc877342014-02-12 15:47:29 +0100239}
240
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800241static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq);
242
Peter Zijlstraa649f232015-06-11 14:46:49 +0200243static struct rq *dl_task_offline_migration(struct rq *rq, struct task_struct *p)
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800244{
245 struct rq *later_rq = NULL;
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800246
247 later_rq = find_lock_later_rq(p, rq);
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800248 if (!later_rq) {
249 int cpu;
250
251 /*
252 * If we cannot preempt any rq, fall back to pick any
253 * online cpu.
254 */
Ingo Molnar0c98d342017-02-05 15:38:10 +0100255 cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed);
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800256 if (cpu >= nr_cpu_ids) {
257 /*
258 * Fail to find any suitable cpu.
259 * The task will never come back!
260 */
261 BUG_ON(dl_bandwidth_enabled());
262
263 /*
264 * If admission control is disabled we
265 * try a little harder to let the task
266 * run.
267 */
268 cpu = cpumask_any(cpu_active_mask);
269 }
270 later_rq = cpu_rq(cpu);
271 double_lock_balance(rq, later_rq);
272 }
273
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800274 set_task_cpu(p, later_rq->cpu);
Peter Zijlstraa649f232015-06-11 14:46:49 +0200275 double_unlock_balance(later_rq, rq);
276
277 return later_rq;
Wanpeng Lifa9c9d12015-03-27 07:08:35 +0800278}
279
Juri Lelli1baca4c2013-11-07 14:43:38 +0100280#else
281
282static inline
283void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
284{
285}
286
287static inline
288void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
289{
290}
291
292static inline
293void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
294{
295}
296
297static inline
298void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
299{
300}
301
Peter Zijlstradc877342014-02-12 15:47:29 +0100302static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
303{
304 return false;
305}
306
Peter Zijlstra0ea60c22015-06-11 14:46:42 +0200307static inline void pull_dl_task(struct rq *rq)
Peter Zijlstradc877342014-02-12 15:47:29 +0100308{
Peter Zijlstradc877342014-02-12 15:47:29 +0100309}
310
Peter Zijlstrae3fca9e2015-06-11 14:46:37 +0200311static inline void queue_push_tasks(struct rq *rq)
Peter Zijlstradc877342014-02-12 15:47:29 +0100312{
313}
Peter Zijlstra9916e212015-06-11 14:46:43 +0200314
315static inline void queue_pull_task(struct rq *rq)
Juri Lelli1baca4c2013-11-07 14:43:38 +0100316{
317}
318#endif /* CONFIG_SMP */
319
Dario Faggioliaab03e02013-11-28 11:14:43 +0100320static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags);
321static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags);
322static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
323 int flags);
324
325/*
326 * We are being explicitly informed that a new instance is starting,
327 * and this means that:
328 * - the absolute deadline of the entity has to be placed at
329 * current time + relative deadline;
330 * - the runtime of the entity has to be set to the maximum value.
331 *
332 * The capability of specifying such event is useful whenever a -deadline
333 * entity wants to (try to!) synchronize its behaviour with the scheduler's
334 * one, and to (try to!) reconcile itself with its own scheduling
335 * parameters.
336 */
Juri Lelli98b0a852016-08-05 16:07:55 +0100337static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100338{
339 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
340 struct rq *rq = rq_of_dl_rq(dl_rq);
341
Juri Lelli98b0a852016-08-05 16:07:55 +0100342 WARN_ON(dl_se->dl_boosted);
Luca Abeni72f9f3f2016-03-07 12:27:04 +0100343 WARN_ON(dl_time_before(rq_clock(rq), dl_se->deadline));
344
345 /*
346 * We are racing with the deadline timer. So, do nothing because
347 * the deadline timer handler will take care of properly recharging
348 * the runtime and postponing the deadline
349 */
350 if (dl_se->dl_throttled)
351 return;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100352
353 /*
354 * We use the regular wall clock time to set deadlines in the
355 * future; in fact, we must consider execution overheads (time
356 * spent on hardirq context, etc.).
357 */
Juri Lelli98b0a852016-08-05 16:07:55 +0100358 dl_se->deadline = rq_clock(rq) + dl_se->dl_deadline;
359 dl_se->runtime = dl_se->dl_runtime;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100360}
361
362/*
363 * Pure Earliest Deadline First (EDF) scheduling does not deal with the
364 * possibility of a entity lasting more than what it declared, and thus
365 * exhausting its runtime.
366 *
367 * Here we are interested in making runtime overrun possible, but we do
368 * not want a entity which is misbehaving to affect the scheduling of all
369 * other entities.
370 * Therefore, a budgeting strategy called Constant Bandwidth Server (CBS)
371 * is used, in order to confine each entity within its own bandwidth.
372 *
373 * This function deals exactly with that, and ensures that when the runtime
374 * of a entity is replenished, its deadline is also postponed. That ensures
375 * the overrunning entity can't interfere with other entity in the system and
376 * can't make them miss their deadlines. Reasons why this kind of overruns
377 * could happen are, typically, a entity voluntarily trying to overcome its
xiaofeng.yan1b09d292014-07-07 05:59:04 +0000378 * runtime, or it just underestimated it during sched_setattr().
Dario Faggioliaab03e02013-11-28 11:14:43 +0100379 */
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100380static void replenish_dl_entity(struct sched_dl_entity *dl_se,
381 struct sched_dl_entity *pi_se)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100382{
383 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
384 struct rq *rq = rq_of_dl_rq(dl_rq);
385
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100386 BUG_ON(pi_se->dl_runtime <= 0);
387
388 /*
389 * This could be the case for a !-dl task that is boosted.
390 * Just go with full inherited parameters.
391 */
392 if (dl_se->dl_deadline == 0) {
393 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
394 dl_se->runtime = pi_se->dl_runtime;
395 }
396
Peter Zijlstra48be3a62016-02-23 13:28:22 +0100397 if (dl_se->dl_yielded && dl_se->runtime > 0)
398 dl_se->runtime = 0;
399
Dario Faggioliaab03e02013-11-28 11:14:43 +0100400 /*
401 * We keep moving the deadline away until we get some
402 * available runtime for the entity. This ensures correct
403 * handling of situations where the runtime overrun is
404 * arbitrary large.
405 */
406 while (dl_se->runtime <= 0) {
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100407 dl_se->deadline += pi_se->dl_period;
408 dl_se->runtime += pi_se->dl_runtime;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100409 }
410
411 /*
412 * At this point, the deadline really should be "in
413 * the future" with respect to rq->clock. If it's
414 * not, we are, for some reason, lagging too much!
415 * Anyway, after having warn userspace abut that,
416 * we still try to keep the things running by
417 * resetting the deadline and the budget of the
418 * entity.
419 */
420 if (dl_time_before(dl_se->deadline, rq_clock(rq))) {
Steven Rostedtc219b7d2016-02-10 12:04:22 -0500421 printk_deferred_once("sched: DL replenish lagged too much\n");
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100422 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
423 dl_se->runtime = pi_se->dl_runtime;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100424 }
Peter Zijlstra1019a352014-11-26 08:44:03 +0800425
426 if (dl_se->dl_yielded)
427 dl_se->dl_yielded = 0;
428 if (dl_se->dl_throttled)
429 dl_se->dl_throttled = 0;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100430}
431
432/*
433 * Here we check if --at time t-- an entity (which is probably being
434 * [re]activated or, in general, enqueued) can use its remaining runtime
435 * and its current deadline _without_ exceeding the bandwidth it is
436 * assigned (function returns true if it can't). We are in fact applying
437 * one of the CBS rules: when a task wakes up, if the residual runtime
438 * over residual deadline fits within the allocated bandwidth, then we
439 * can keep the current (absolute) deadline and residual budget without
440 * disrupting the schedulability of the system. Otherwise, we should
441 * refill the runtime and set the deadline a period in the future,
442 * because keeping the current (absolute) deadline of the task would
Dario Faggioli712e5e32014-01-27 12:20:15 +0100443 * result in breaking guarantees promised to other tasks (refer to
444 * Documentation/scheduler/sched-deadline.txt for more informations).
Dario Faggioliaab03e02013-11-28 11:14:43 +0100445 *
446 * This function returns true if:
447 *
Steven Rostedt (VMware)2317d5f2017-03-02 15:10:59 +0100448 * runtime / (deadline - t) > dl_runtime / dl_deadline ,
Dario Faggioliaab03e02013-11-28 11:14:43 +0100449 *
450 * IOW we can't recycle current parameters.
Harald Gustafsson755378a2013-11-07 14:43:40 +0100451 *
Steven Rostedt (VMware)2317d5f2017-03-02 15:10:59 +0100452 * Notice that the bandwidth check is done against the deadline. For
Harald Gustafsson755378a2013-11-07 14:43:40 +0100453 * task with deadline equal to period this is the same of using
Steven Rostedt (VMware)2317d5f2017-03-02 15:10:59 +0100454 * dl_period instead of dl_deadline in the equation above.
Dario Faggioliaab03e02013-11-28 11:14:43 +0100455 */
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100456static bool dl_entity_overflow(struct sched_dl_entity *dl_se,
457 struct sched_dl_entity *pi_se, u64 t)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100458{
459 u64 left, right;
460
461 /*
462 * left and right are the two sides of the equation above,
463 * after a bit of shuffling to use multiplications instead
464 * of divisions.
465 *
466 * Note that none of the time values involved in the two
467 * multiplications are absolute: dl_deadline and dl_runtime
468 * are the relative deadline and the maximum runtime of each
469 * instance, runtime is the runtime left for the last instance
470 * and (deadline - t), since t is rq->clock, is the time left
471 * to the (absolute) deadline. Even if overflowing the u64 type
472 * is very unlikely to occur in both cases, here we scale down
473 * as we want to avoid that risk at all. Scaling down by 10
474 * means that we reduce granularity to 1us. We are fine with it,
475 * since this is only a true/false check and, anyway, thinking
476 * of anything below microseconds resolution is actually fiction
477 * (but still we want to give the user that illusion >;).
478 */
Steven Rostedt (VMware)2317d5f2017-03-02 15:10:59 +0100479 left = (pi_se->dl_deadline >> DL_SCALE) * (dl_se->runtime >> DL_SCALE);
Dario Faggioli332ac172013-11-07 14:43:45 +0100480 right = ((dl_se->deadline - t) >> DL_SCALE) *
481 (pi_se->dl_runtime >> DL_SCALE);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100482
483 return dl_time_before(right, left);
484}
485
486/*
487 * When a -deadline entity is queued back on the runqueue, its runtime and
488 * deadline might need updating.
489 *
490 * The policy here is that we update the deadline of the entity only if:
491 * - the current deadline is in the past,
492 * - using the remaining runtime with the current deadline would make
493 * the entity exceed its bandwidth.
494 */
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100495static void update_dl_entity(struct sched_dl_entity *dl_se,
496 struct sched_dl_entity *pi_se)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100497{
498 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
499 struct rq *rq = rq_of_dl_rq(dl_rq);
500
Dario Faggioliaab03e02013-11-28 11:14:43 +0100501 if (dl_time_before(dl_se->deadline, rq_clock(rq)) ||
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100502 dl_entity_overflow(dl_se, pi_se, rq_clock(rq))) {
503 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
504 dl_se->runtime = pi_se->dl_runtime;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100505 }
506}
507
Daniel Bristot de Oliveira5ac69d32017-03-02 15:10:57 +0100508static inline u64 dl_next_period(struct sched_dl_entity *dl_se)
509{
510 return dl_se->deadline - dl_se->dl_deadline + dl_se->dl_period;
511}
512
Dario Faggioliaab03e02013-11-28 11:14:43 +0100513/*
514 * If the entity depleted all its runtime, and if we want it to sleep
515 * while waiting for some new execution time to become available, we
Daniel Bristot de Oliveira5ac69d32017-03-02 15:10:57 +0100516 * set the bandwidth replenishment timer to the replenishment instant
Dario Faggioliaab03e02013-11-28 11:14:43 +0100517 * and try to activate it.
518 *
519 * Notice that it is important for the caller to know if the timer
520 * actually started or not (i.e., the replenishment instant is in
521 * the future or in the past).
522 */
Peter Zijlstraa649f232015-06-11 14:46:49 +0200523static int start_dl_timer(struct task_struct *p)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100524{
Peter Zijlstraa649f232015-06-11 14:46:49 +0200525 struct sched_dl_entity *dl_se = &p->dl;
526 struct hrtimer *timer = &dl_se->dl_timer;
527 struct rq *rq = task_rq(p);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100528 ktime_t now, act;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100529 s64 delta;
530
Peter Zijlstraa649f232015-06-11 14:46:49 +0200531 lockdep_assert_held(&rq->lock);
532
Dario Faggioliaab03e02013-11-28 11:14:43 +0100533 /*
534 * We want the timer to fire at the deadline, but considering
535 * that it is actually coming from rq->clock and not from
536 * hrtimer's time base reading.
537 */
Daniel Bristot de Oliveira5ac69d32017-03-02 15:10:57 +0100538 act = ns_to_ktime(dl_next_period(dl_se));
Peter Zijlstraa649f232015-06-11 14:46:49 +0200539 now = hrtimer_cb_get_time(timer);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100540 delta = ktime_to_ns(now) - rq_clock(rq);
541 act = ktime_add_ns(act, delta);
542
543 /*
544 * If the expiry time already passed, e.g., because the value
545 * chosen as the deadline is too small, don't even try to
546 * start the timer in the past!
547 */
548 if (ktime_us_delta(act, now) < 0)
549 return 0;
550
Peter Zijlstraa649f232015-06-11 14:46:49 +0200551 /*
552 * !enqueued will guarantee another callback; even if one is already in
553 * progress. This ensures a balanced {get,put}_task_struct().
554 *
555 * The race against __run_timer() clearing the enqueued state is
556 * harmless because we're holding task_rq()->lock, therefore the timer
557 * expiring after we've done the check will wait on its task_rq_lock()
558 * and observe our state.
559 */
560 if (!hrtimer_is_queued(timer)) {
561 get_task_struct(p);
562 hrtimer_start(timer, act, HRTIMER_MODE_ABS);
563 }
Dario Faggioliaab03e02013-11-28 11:14:43 +0100564
Thomas Gleixnercc9684d2015-04-14 21:09:06 +0000565 return 1;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100566}
567
568/*
569 * This is the bandwidth enforcement timer callback. If here, we know
570 * a task is not on its dl_rq, since the fact that the timer was running
571 * means the task is throttled and needs a runtime replenishment.
572 *
573 * However, what we actually do depends on the fact the task is active,
574 * (it is on its rq) or has been removed from there by a call to
575 * dequeue_task_dl(). In the former case we must issue the runtime
576 * replenishment and add the task back to the dl_rq; in the latter, we just
577 * do nothing but clearing dl_throttled, so that runtime and deadline
578 * updating (and the queueing back to dl_rq) will be done by the
579 * next call to enqueue_task_dl().
580 */
581static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
582{
583 struct sched_dl_entity *dl_se = container_of(timer,
584 struct sched_dl_entity,
585 dl_timer);
586 struct task_struct *p = dl_task_of(dl_se);
Peter Zijlstraeb580752015-07-31 21:28:18 +0200587 struct rq_flags rf;
Kirill Tkhai0f397f22014-05-20 13:33:42 +0400588 struct rq *rq;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100589
Peter Zijlstraeb580752015-07-31 21:28:18 +0200590 rq = task_rq_lock(p, &rf);
Kirill Tkhai0f397f22014-05-20 13:33:42 +0400591
Dario Faggioliaab03e02013-11-28 11:14:43 +0100592 /*
Peter Zijlstraa649f232015-06-11 14:46:49 +0200593 * The task might have changed its scheduling policy to something
Daniel Bristot de Oliveira9846d502016-11-08 11:15:23 +0100594 * different than SCHED_DEADLINE (through switched_from_dl()).
Dario Faggioliaab03e02013-11-28 11:14:43 +0100595 */
Peter Zijlstraa649f232015-06-11 14:46:49 +0200596 if (!dl_task(p)) {
597 __dl_clear_params(p);
598 goto unlock;
599 }
600
601 /*
Peter Zijlstraa649f232015-06-11 14:46:49 +0200602 * The task might have been boosted by someone else and might be in the
603 * boosting/deboosting path, its not throttled.
604 */
605 if (dl_se->dl_boosted)
606 goto unlock;
607
608 /*
609 * Spurious timer due to start_dl_timer() race; or we already received
610 * a replenishment from rt_mutex_setprio().
611 */
612 if (!dl_se->dl_throttled)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100613 goto unlock;
614
615 sched_clock_tick();
616 update_rq_clock(rq);
Kirill Tkhaia79ec892015-02-16 15:38:34 +0300617
618 /*
619 * If the throttle happened during sched-out; like:
620 *
621 * schedule()
622 * deactivate_task()
623 * dequeue_task_dl()
624 * update_curr_dl()
625 * start_dl_timer()
626 * __dequeue_task_dl()
627 * prev->on_rq = 0;
628 *
629 * We can be both throttled and !queued. Replenish the counter
630 * but do not enqueue -- wait for our wakeup to do that.
631 */
632 if (!task_on_rq_queued(p)) {
633 replenish_dl_entity(dl_se, dl_se);
634 goto unlock;
635 }
636
Wanpeng Li61c7aca2016-08-31 18:27:44 +0800637#ifdef CONFIG_SMP
638 if (unlikely(!rq->online)) {
639 /*
640 * If the runqueue is no longer available, migrate the
641 * task elsewhere. This necessarily changes rq.
642 */
643 lockdep_unpin_lock(&rq->lock, rf.cookie);
644 rq = dl_task_offline_migration(rq, p);
645 rf.cookie = lockdep_pin_lock(&rq->lock);
Wanpeng Lidcc3b5f2017-03-06 21:51:28 -0800646 update_rq_clock(rq);
Wanpeng Li61c7aca2016-08-31 18:27:44 +0800647
648 /*
649 * Now that the task has been migrated to the new RQ and we
650 * have that locked, proceed as normal and enqueue the task
651 * there.
652 */
653 }
654#endif
655
Peter Zijlstra1019a352014-11-26 08:44:03 +0800656 enqueue_task_dl(rq, p, ENQUEUE_REPLENISH);
657 if (dl_task(rq->curr))
658 check_preempt_curr_dl(rq, p, 0);
659 else
660 resched_curr(rq);
Peter Zijlstraa649f232015-06-11 14:46:49 +0200661
Juri Lelli1baca4c2013-11-07 14:43:38 +0100662#ifdef CONFIG_SMP
Peter Zijlstra1019a352014-11-26 08:44:03 +0800663 /*
Peter Zijlstraa649f232015-06-11 14:46:49 +0200664 * Queueing this task back might have overloaded rq, check if we need
665 * to kick someone away.
Peter Zijlstra1019a352014-11-26 08:44:03 +0800666 */
Peter Zijlstra0aaafaa2015-10-23 11:50:08 +0200667 if (has_pushable_dl_tasks(rq)) {
668 /*
669 * Nothing relies on rq->lock after this, so its safe to drop
670 * rq->lock.
671 */
Matt Flemingd8ac8972016-09-21 14:38:10 +0100672 rq_unpin_lock(rq, &rf);
Peter Zijlstra1019a352014-11-26 08:44:03 +0800673 push_dl_task(rq);
Matt Flemingd8ac8972016-09-21 14:38:10 +0100674 rq_repin_lock(rq, &rf);
Peter Zijlstra0aaafaa2015-10-23 11:50:08 +0200675 }
Juri Lelli1baca4c2013-11-07 14:43:38 +0100676#endif
Peter Zijlstraa649f232015-06-11 14:46:49 +0200677
Dario Faggioliaab03e02013-11-28 11:14:43 +0100678unlock:
Peter Zijlstraeb580752015-07-31 21:28:18 +0200679 task_rq_unlock(rq, p, &rf);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100680
Peter Zijlstraa649f232015-06-11 14:46:49 +0200681 /*
682 * This can free the task_struct, including this hrtimer, do not touch
683 * anything related to that after this.
684 */
685 put_task_struct(p);
686
Dario Faggioliaab03e02013-11-28 11:14:43 +0100687 return HRTIMER_NORESTART;
688}
689
690void init_dl_task_timer(struct sched_dl_entity *dl_se)
691{
692 struct hrtimer *timer = &dl_se->dl_timer;
693
Dario Faggioliaab03e02013-11-28 11:14:43 +0100694 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
695 timer->function = dl_task_timer;
696}
697
Daniel Bristot de Oliveiradf8eac82017-03-02 15:10:58 +0100698/*
699 * During the activation, CBS checks if it can reuse the current task's
700 * runtime and period. If the deadline of the task is in the past, CBS
701 * cannot use the runtime, and so it replenishes the task. This rule
702 * works fine for implicit deadline tasks (deadline == period), and the
703 * CBS was designed for implicit deadline tasks. However, a task with
704 * constrained deadline (deadine < period) might be awakened after the
705 * deadline, but before the next period. In this case, replenishing the
706 * task would allow it to run for runtime / deadline. As in this case
707 * deadline < period, CBS enables a task to run for more than the
708 * runtime / period. In a very loaded system, this can cause a domino
709 * effect, making other tasks miss their deadlines.
710 *
711 * To avoid this problem, in the activation of a constrained deadline
712 * task after the deadline but before the next period, throttle the
713 * task and set the replenishing timer to the begin of the next period,
714 * unless it is boosted.
715 */
716static inline void dl_check_constrained_dl(struct sched_dl_entity *dl_se)
717{
718 struct task_struct *p = dl_task_of(dl_se);
719 struct rq *rq = rq_of_dl_rq(dl_rq_of_se(dl_se));
720
721 if (dl_time_before(dl_se->deadline, rq_clock(rq)) &&
722 dl_time_before(rq_clock(rq), dl_next_period(dl_se))) {
723 if (unlikely(dl_se->dl_boosted || !start_dl_timer(p)))
724 return;
725 dl_se->dl_throttled = 1;
726 }
727}
728
Dario Faggioliaab03e02013-11-28 11:14:43 +0100729static
Zhiqiang Zhang6fab5412015-06-15 11:15:20 +0800730int dl_runtime_exceeded(struct sched_dl_entity *dl_se)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100731{
Luca Abeni269ad802014-12-17 11:50:32 +0100732 return (dl_se->runtime <= 0);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100733}
734
Juri Lellifaa59932014-02-21 11:37:15 +0100735extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
736
Dario Faggioliaab03e02013-11-28 11:14:43 +0100737/*
738 * Update the current task's runtime statistics (provided it is still
739 * a -deadline task and has not been removed from the dl_rq).
740 */
741static void update_curr_dl(struct rq *rq)
742{
743 struct task_struct *curr = rq->curr;
744 struct sched_dl_entity *dl_se = &curr->dl;
745 u64 delta_exec;
746
747 if (!dl_task(curr) || !on_dl_rq(dl_se))
748 return;
749
750 /*
751 * Consumed budget is computed considering the time as
752 * observed by schedulable tasks (excluding time spent
753 * in hardirq context, etc.). Deadlines are instead
754 * computed using hard walltime. This seems to be the more
755 * natural solution, but the full ramifications of this
756 * approach need further study.
757 */
758 delta_exec = rq_clock_task(rq) - curr->se.exec_start;
Peter Zijlstra48be3a62016-02-23 13:28:22 +0100759 if (unlikely((s64)delta_exec <= 0)) {
760 if (unlikely(dl_se->dl_yielded))
761 goto throttle;
Kirill Tkhai734ff2a2014-03-04 19:25:46 +0400762 return;
Peter Zijlstra48be3a62016-02-23 13:28:22 +0100763 }
Dario Faggioliaab03e02013-11-28 11:14:43 +0100764
Rafael J. Wysocki58919e82016-08-16 22:14:55 +0200765 /* kick cpufreq (see the comment in kernel/sched/sched.h). */
Rafael J. Wysocki12bde332016-08-10 03:11:17 +0200766 cpufreq_update_this_cpu(rq, SCHED_CPUFREQ_DL);
Wanpeng Li594dd292016-04-22 17:07:24 +0800767
Dario Faggioliaab03e02013-11-28 11:14:43 +0100768 schedstat_set(curr->se.statistics.exec_max,
769 max(curr->se.statistics.exec_max, delta_exec));
770
771 curr->se.sum_exec_runtime += delta_exec;
772 account_group_exec_runtime(curr, delta_exec);
773
774 curr->se.exec_start = rq_clock_task(rq);
775 cpuacct_charge(curr, delta_exec);
776
Dario Faggioli239be4a2013-11-07 14:43:39 +0100777 sched_rt_avg_update(rq, delta_exec);
778
Peter Zijlstra48be3a62016-02-23 13:28:22 +0100779 dl_se->runtime -= delta_exec;
780
781throttle:
782 if (dl_runtime_exceeded(dl_se) || dl_se->dl_yielded) {
Peter Zijlstra1019a352014-11-26 08:44:03 +0800783 dl_se->dl_throttled = 1;
Dario Faggioliaab03e02013-11-28 11:14:43 +0100784 __dequeue_task_dl(rq, curr, 0);
Peter Zijlstraa649f232015-06-11 14:46:49 +0200785 if (unlikely(dl_se->dl_boosted || !start_dl_timer(curr)))
Dario Faggioliaab03e02013-11-28 11:14:43 +0100786 enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH);
787
788 if (!is_leftmost(curr, &rq->dl))
Kirill Tkhai88751252014-06-29 00:03:57 +0400789 resched_curr(rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100790 }
Peter Zijlstra17248132013-12-17 12:44:49 +0100791
792 /*
793 * Because -- for now -- we share the rt bandwidth, we need to
794 * account our runtime there too, otherwise actual rt tasks
795 * would be able to exceed the shared quota.
796 *
797 * Account to the root rt group for now.
798 *
799 * The solution we're working towards is having the RT groups scheduled
800 * using deadline servers -- however there's a few nasties to figure
801 * out before that can happen.
802 */
803 if (rt_bandwidth_enabled()) {
804 struct rt_rq *rt_rq = &rq->rt;
805
806 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstra17248132013-12-17 12:44:49 +0100807 /*
808 * We'll let actual RT tasks worry about the overflow here, we
Juri Lellifaa59932014-02-21 11:37:15 +0100809 * have our own CBS to keep us inline; only account when RT
810 * bandwidth is relevant.
Peter Zijlstra17248132013-12-17 12:44:49 +0100811 */
Juri Lellifaa59932014-02-21 11:37:15 +0100812 if (sched_rt_bandwidth_account(rt_rq))
813 rt_rq->rt_time += delta_exec;
Peter Zijlstra17248132013-12-17 12:44:49 +0100814 raw_spin_unlock(&rt_rq->rt_runtime_lock);
815 }
Dario Faggioliaab03e02013-11-28 11:14:43 +0100816}
817
Juri Lelli1baca4c2013-11-07 14:43:38 +0100818#ifdef CONFIG_SMP
819
Juri Lelli1baca4c2013-11-07 14:43:38 +0100820static void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
821{
822 struct rq *rq = rq_of_dl_rq(dl_rq);
823
824 if (dl_rq->earliest_dl.curr == 0 ||
825 dl_time_before(deadline, dl_rq->earliest_dl.curr)) {
Juri Lelli1baca4c2013-11-07 14:43:38 +0100826 dl_rq->earliest_dl.curr = deadline;
Tommaso Cucinottad8206bb2016-08-14 16:27:08 +0200827 cpudl_set(&rq->rd->cpudl, rq->cpu, deadline);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100828 }
829}
830
831static void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
832{
833 struct rq *rq = rq_of_dl_rq(dl_rq);
834
835 /*
836 * Since we may have removed our earliest (and/or next earliest)
837 * task we must recompute them.
838 */
839 if (!dl_rq->dl_nr_running) {
840 dl_rq->earliest_dl.curr = 0;
841 dl_rq->earliest_dl.next = 0;
Tommaso Cucinottad8206bb2016-08-14 16:27:08 +0200842 cpudl_clear(&rq->rd->cpudl, rq->cpu);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100843 } else {
844 struct rb_node *leftmost = dl_rq->rb_leftmost;
845 struct sched_dl_entity *entry;
846
847 entry = rb_entry(leftmost, struct sched_dl_entity, rb_node);
848 dl_rq->earliest_dl.curr = entry->deadline;
Tommaso Cucinottad8206bb2016-08-14 16:27:08 +0200849 cpudl_set(&rq->rd->cpudl, rq->cpu, entry->deadline);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100850 }
851}
852
853#else
854
855static inline void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
856static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
857
858#endif /* CONFIG_SMP */
859
860static inline
861void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
862{
863 int prio = dl_task_of(dl_se)->prio;
864 u64 deadline = dl_se->deadline;
865
866 WARN_ON(!dl_prio(prio));
867 dl_rq->dl_nr_running++;
Kirill Tkhai72465442014-05-09 03:00:14 +0400868 add_nr_running(rq_of_dl_rq(dl_rq), 1);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100869
870 inc_dl_deadline(dl_rq, deadline);
871 inc_dl_migration(dl_se, dl_rq);
872}
873
874static inline
875void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
876{
877 int prio = dl_task_of(dl_se)->prio;
878
879 WARN_ON(!dl_prio(prio));
880 WARN_ON(!dl_rq->dl_nr_running);
881 dl_rq->dl_nr_running--;
Kirill Tkhai72465442014-05-09 03:00:14 +0400882 sub_nr_running(rq_of_dl_rq(dl_rq), 1);
Juri Lelli1baca4c2013-11-07 14:43:38 +0100883
884 dec_dl_deadline(dl_rq, dl_se->deadline);
885 dec_dl_migration(dl_se, dl_rq);
886}
887
Dario Faggioliaab03e02013-11-28 11:14:43 +0100888static void __enqueue_dl_entity(struct sched_dl_entity *dl_se)
889{
890 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
891 struct rb_node **link = &dl_rq->rb_root.rb_node;
892 struct rb_node *parent = NULL;
893 struct sched_dl_entity *entry;
894 int leftmost = 1;
895
896 BUG_ON(!RB_EMPTY_NODE(&dl_se->rb_node));
897
898 while (*link) {
899 parent = *link;
900 entry = rb_entry(parent, struct sched_dl_entity, rb_node);
901 if (dl_time_before(dl_se->deadline, entry->deadline))
902 link = &parent->rb_left;
903 else {
904 link = &parent->rb_right;
905 leftmost = 0;
906 }
907 }
908
909 if (leftmost)
910 dl_rq->rb_leftmost = &dl_se->rb_node;
911
912 rb_link_node(&dl_se->rb_node, parent, link);
913 rb_insert_color(&dl_se->rb_node, &dl_rq->rb_root);
914
Juri Lelli1baca4c2013-11-07 14:43:38 +0100915 inc_dl_tasks(dl_se, dl_rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100916}
917
918static void __dequeue_dl_entity(struct sched_dl_entity *dl_se)
919{
920 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
921
922 if (RB_EMPTY_NODE(&dl_se->rb_node))
923 return;
924
925 if (dl_rq->rb_leftmost == &dl_se->rb_node) {
926 struct rb_node *next_node;
927
928 next_node = rb_next(&dl_se->rb_node);
929 dl_rq->rb_leftmost = next_node;
930 }
931
932 rb_erase(&dl_se->rb_node, &dl_rq->rb_root);
933 RB_CLEAR_NODE(&dl_se->rb_node);
934
Juri Lelli1baca4c2013-11-07 14:43:38 +0100935 dec_dl_tasks(dl_se, dl_rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100936}
937
938static void
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100939enqueue_dl_entity(struct sched_dl_entity *dl_se,
940 struct sched_dl_entity *pi_se, int flags)
Dario Faggioliaab03e02013-11-28 11:14:43 +0100941{
942 BUG_ON(on_dl_rq(dl_se));
943
944 /*
945 * If this is a wakeup or a new instance, the scheduling
946 * parameters of the task might need updating. Otherwise,
947 * we want a replenishment of its runtime.
948 */
Luca Abeni72f9f3f2016-03-07 12:27:04 +0100949 if (flags & ENQUEUE_WAKEUP)
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100950 update_dl_entity(dl_se, pi_se);
Luca Abeni6a503c32014-12-17 11:50:31 +0100951 else if (flags & ENQUEUE_REPLENISH)
952 replenish_dl_entity(dl_se, pi_se);
Dario Faggioliaab03e02013-11-28 11:14:43 +0100953
954 __enqueue_dl_entity(dl_se);
955}
956
957static void dequeue_dl_entity(struct sched_dl_entity *dl_se)
958{
959 __dequeue_dl_entity(dl_se);
960}
961
Daniel Bristot de Oliveiradf8eac82017-03-02 15:10:58 +0100962static inline bool dl_is_constrained(struct sched_dl_entity *dl_se)
963{
964 return dl_se->dl_deadline < dl_se->dl_period;
965}
966
Dario Faggioliaab03e02013-11-28 11:14:43 +0100967static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
968{
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100969 struct task_struct *pi_task = rt_mutex_get_top_task(p);
970 struct sched_dl_entity *pi_se = &p->dl;
971
972 /*
973 * Use the scheduling parameters of the top pi-waiter
Andrea Parriff277d42015-08-05 15:56:19 +0200974 * task if we have one and its (absolute) deadline is
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100975 * smaller than our one... OTW we keep our runtime and
976 * deadline.
977 */
Juri Lelli64be6f12014-10-24 10:16:37 +0100978 if (pi_task && p->dl.dl_boosted && dl_prio(pi_task->normal_prio)) {
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100979 pi_se = &pi_task->dl;
Juri Lelli64be6f12014-10-24 10:16:37 +0100980 } else if (!dl_prio(p->normal_prio)) {
981 /*
982 * Special case in which we have a !SCHED_DEADLINE task
983 * that is going to be deboosted, but exceedes its
984 * runtime while doing so. No point in replenishing
985 * it, as it's going to return back to its original
986 * scheduling class after this.
987 */
988 BUG_ON(!p->dl.dl_boosted || flags != ENQUEUE_REPLENISH);
989 return;
990 }
Dario Faggioli2d3d8912013-11-07 14:43:44 +0100991
Dario Faggioliaab03e02013-11-28 11:14:43 +0100992 /*
Daniel Bristot de Oliveiradf8eac82017-03-02 15:10:58 +0100993 * Check if a constrained deadline task was activated
994 * after the deadline but before the next period.
995 * If that is the case, the task will be throttled and
996 * the replenishment timer will be set to the next period.
997 */
998 if (!p->dl.dl_throttled && dl_is_constrained(&p->dl))
999 dl_check_constrained_dl(&p->dl);
1000
1001 /*
Dario Faggioliaab03e02013-11-28 11:14:43 +01001002 * If p is throttled, we do nothing. In fact, if it exhausted
1003 * its budget it needs a replenishment and, since it now is on
1004 * its rq, the bandwidth timer callback (which clearly has not
1005 * run yet) will take care of this.
1006 */
Peter Zijlstra1019a352014-11-26 08:44:03 +08001007 if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH))
Dario Faggioliaab03e02013-11-28 11:14:43 +01001008 return;
1009
Dario Faggioli2d3d8912013-11-07 14:43:44 +01001010 enqueue_dl_entity(&p->dl, pi_se, flags);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001011
Ingo Molnar4b53a342017-02-05 15:41:03 +01001012 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001013 enqueue_pushable_dl_task(rq, p);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001014}
1015
1016static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1017{
1018 dequeue_dl_entity(&p->dl);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001019 dequeue_pushable_dl_task(rq, p);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001020}
1021
1022static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1023{
1024 update_curr_dl(rq);
1025 __dequeue_task_dl(rq, p, flags);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001026}
1027
1028/*
1029 * Yield task semantic for -deadline tasks is:
1030 *
1031 * get off from the CPU until our next instance, with
1032 * a new runtime. This is of little use now, since we
1033 * don't have a bandwidth reclaiming mechanism. Anyway,
1034 * bandwidth reclaiming is planned for the future, and
1035 * yield_task_dl will indicate that some spare budget
1036 * is available for other task instances to use it.
1037 */
1038static void yield_task_dl(struct rq *rq)
1039{
Dario Faggioliaab03e02013-11-28 11:14:43 +01001040 /*
1041 * We make the task go to sleep until its current deadline by
1042 * forcing its runtime to zero. This way, update_curr_dl() stops
1043 * it and the bandwidth timer will wake it up and will give it
Juri Lelli5bfd1262014-04-15 13:49:04 +02001044 * new scheduling parameters (thanks to dl_yielded=1).
Dario Faggioliaab03e02013-11-28 11:14:43 +01001045 */
Peter Zijlstra48be3a62016-02-23 13:28:22 +01001046 rq->curr->dl.dl_yielded = 1;
1047
Kirill Tkhai6f1607f2015-02-04 12:09:32 +03001048 update_rq_clock(rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001049 update_curr_dl(rq);
Wanpeng Li44fb0852015-03-10 12:20:00 +08001050 /*
1051 * Tell update_rq_clock() that we've just updated,
1052 * so we don't do microscopic update in schedule()
1053 * and double the fastpath cost.
1054 */
1055 rq_clock_skip_update(rq, true);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001056}
1057
Juri Lelli1baca4c2013-11-07 14:43:38 +01001058#ifdef CONFIG_SMP
1059
1060static int find_later_rq(struct task_struct *task);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001061
1062static int
1063select_task_rq_dl(struct task_struct *p, int cpu, int sd_flag, int flags)
1064{
1065 struct task_struct *curr;
1066 struct rq *rq;
1067
Wanpeng Li1d7e9742014-10-14 10:22:39 +08001068 if (sd_flag != SD_BALANCE_WAKE)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001069 goto out;
1070
1071 rq = cpu_rq(cpu);
1072
1073 rcu_read_lock();
Jason Low316c1608d2015-04-28 13:00:20 -07001074 curr = READ_ONCE(rq->curr); /* unlocked access */
Juri Lelli1baca4c2013-11-07 14:43:38 +01001075
1076 /*
1077 * If we are dealing with a -deadline task, we must
1078 * decide where to wake it up.
1079 * If it has a later deadline and the current task
1080 * on this rq can't move (provided the waking task
1081 * can!) we prefer to send it somewhere else. On the
1082 * other hand, if it has a shorter deadline, we
1083 * try to make it stay here, it might be important.
1084 */
1085 if (unlikely(dl_task(curr)) &&
Ingo Molnar4b53a342017-02-05 15:41:03 +01001086 (curr->nr_cpus_allowed < 2 ||
Juri Lelli1baca4c2013-11-07 14:43:38 +01001087 !dl_entity_preempt(&p->dl, &curr->dl)) &&
Ingo Molnar4b53a342017-02-05 15:41:03 +01001088 (p->nr_cpus_allowed > 1)) {
Juri Lelli1baca4c2013-11-07 14:43:38 +01001089 int target = find_later_rq(p);
1090
Wanpeng Li9d514262015-05-13 14:01:03 +08001091 if (target != -1 &&
Luca Abeni5aa50502015-10-16 10:06:21 +02001092 (dl_time_before(p->dl.deadline,
1093 cpu_rq(target)->dl.earliest_dl.curr) ||
1094 (cpu_rq(target)->dl.dl_nr_running == 0)))
Juri Lelli1baca4c2013-11-07 14:43:38 +01001095 cpu = target;
1096 }
1097 rcu_read_unlock();
1098
1099out:
1100 return cpu;
1101}
1102
1103static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
1104{
1105 /*
1106 * Current can't be migrated, useless to reschedule,
1107 * let's hope p can move out.
1108 */
Ingo Molnar4b53a342017-02-05 15:41:03 +01001109 if (rq->curr->nr_cpus_allowed == 1 ||
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001110 cpudl_find(&rq->rd->cpudl, rq->curr, NULL) == -1)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001111 return;
1112
1113 /*
1114 * p is migratable, so let's not schedule it and
1115 * see if it is pushed or pulled somewhere else.
1116 */
Ingo Molnar4b53a342017-02-05 15:41:03 +01001117 if (p->nr_cpus_allowed != 1 &&
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001118 cpudl_find(&rq->rd->cpudl, p, NULL) != -1)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001119 return;
1120
Kirill Tkhai88751252014-06-29 00:03:57 +04001121 resched_curr(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001122}
1123
1124#endif /* CONFIG_SMP */
1125
Dario Faggioliaab03e02013-11-28 11:14:43 +01001126/*
1127 * Only called when both the current and waking task are -deadline
1128 * tasks.
1129 */
1130static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
1131 int flags)
1132{
Juri Lelli1baca4c2013-11-07 14:43:38 +01001133 if (dl_entity_preempt(&p->dl, &rq->curr->dl)) {
Kirill Tkhai88751252014-06-29 00:03:57 +04001134 resched_curr(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001135 return;
1136 }
1137
1138#ifdef CONFIG_SMP
1139 /*
1140 * In the unlikely case current and p have the same deadline
1141 * let us try to decide what's the best thing to do...
1142 */
Dario Faggioli332ac172013-11-07 14:43:45 +01001143 if ((p->dl.deadline == rq->curr->dl.deadline) &&
1144 !test_tsk_need_resched(rq->curr))
Juri Lelli1baca4c2013-11-07 14:43:38 +01001145 check_preempt_equal_dl(rq, p);
1146#endif /* CONFIG_SMP */
Dario Faggioliaab03e02013-11-28 11:14:43 +01001147}
1148
1149#ifdef CONFIG_SCHED_HRTICK
1150static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1151{
xiaofeng.yan177ef2a2014-08-26 03:15:41 +00001152 hrtick_start(rq, p->dl.runtime);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001153}
Wanpeng Li36ce9882014-11-11 09:52:26 +08001154#else /* !CONFIG_SCHED_HRTICK */
1155static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1156{
1157}
Dario Faggioliaab03e02013-11-28 11:14:43 +01001158#endif
1159
1160static struct sched_dl_entity *pick_next_dl_entity(struct rq *rq,
1161 struct dl_rq *dl_rq)
1162{
1163 struct rb_node *left = dl_rq->rb_leftmost;
1164
1165 if (!left)
1166 return NULL;
1167
1168 return rb_entry(left, struct sched_dl_entity, rb_node);
1169}
1170
Peter Zijlstrae7904a22015-08-01 19:25:08 +02001171struct task_struct *
Matt Flemingd8ac8972016-09-21 14:38:10 +01001172pick_next_task_dl(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
Dario Faggioliaab03e02013-11-28 11:14:43 +01001173{
1174 struct sched_dl_entity *dl_se;
1175 struct task_struct *p;
1176 struct dl_rq *dl_rq;
1177
1178 dl_rq = &rq->dl;
1179
Kirill Tkhaia1d9a322014-04-10 17:38:36 +04001180 if (need_pull_dl_task(rq, prev)) {
Peter Zijlstracbce1a62015-06-11 14:46:54 +02001181 /*
1182 * This is OK, because current is on_cpu, which avoids it being
1183 * picked for load-balance and preemption/IRQs are still
1184 * disabled avoiding further scheduler activity on it and we're
1185 * being very careful to re-start the picking loop.
1186 */
Matt Flemingd8ac8972016-09-21 14:38:10 +01001187 rq_unpin_lock(rq, rf);
Peter Zijlstra38033c32014-01-23 20:32:21 +01001188 pull_dl_task(rq);
Matt Flemingd8ac8972016-09-21 14:38:10 +01001189 rq_repin_lock(rq, rf);
Kirill Tkhaia1d9a322014-04-10 17:38:36 +04001190 /*
T.Zhou176cedc2016-11-23 08:48:32 +08001191 * pull_dl_task() can drop (and re-acquire) rq->lock; this
Kirill Tkhaia1d9a322014-04-10 17:38:36 +04001192 * means a stop task can slip in, in which case we need to
1193 * re-start task selection.
1194 */
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001195 if (rq->stop && task_on_rq_queued(rq->stop))
Kirill Tkhaia1d9a322014-04-10 17:38:36 +04001196 return RETRY_TASK;
1197 }
1198
Kirill Tkhai734ff2a2014-03-04 19:25:46 +04001199 /*
1200 * When prev is DL, we may throttle it in put_prev_task().
1201 * So, we update time before we check for dl_nr_running.
1202 */
1203 if (prev->sched_class == &dl_sched_class)
1204 update_curr_dl(rq);
Peter Zijlstra38033c32014-01-23 20:32:21 +01001205
Dario Faggioliaab03e02013-11-28 11:14:43 +01001206 if (unlikely(!dl_rq->dl_nr_running))
1207 return NULL;
1208
Peter Zijlstra3f1d2a32014-02-12 10:49:30 +01001209 put_prev_task(rq, prev);
Peter Zijlstra606dba22012-02-11 06:05:00 +01001210
Dario Faggioliaab03e02013-11-28 11:14:43 +01001211 dl_se = pick_next_dl_entity(rq, dl_rq);
1212 BUG_ON(!dl_se);
1213
1214 p = dl_task_of(dl_se);
1215 p->se.exec_start = rq_clock_task(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001216
1217 /* Running task will never be pushed. */
Juri Lelli71362652014-01-14 12:03:51 +01001218 dequeue_pushable_dl_task(rq, p);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001219
Dario Faggioliaab03e02013-11-28 11:14:43 +01001220 if (hrtick_enabled(rq))
1221 start_hrtick_dl(rq, p);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001222
Peter Zijlstrae3fca9e2015-06-11 14:46:37 +02001223 queue_push_tasks(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001224
Dario Faggioliaab03e02013-11-28 11:14:43 +01001225 return p;
1226}
1227
1228static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
1229{
1230 update_curr_dl(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001231
Ingo Molnar4b53a342017-02-05 15:41:03 +01001232 if (on_dl_rq(&p->dl) && p->nr_cpus_allowed > 1)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001233 enqueue_pushable_dl_task(rq, p);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001234}
1235
1236static void task_tick_dl(struct rq *rq, struct task_struct *p, int queued)
1237{
1238 update_curr_dl(rq);
1239
Wanpeng Lia7bebf42014-11-26 08:44:01 +08001240 /*
1241 * Even when we have runtime, update_curr_dl() might have resulted in us
1242 * not being the leftmost task anymore. In that case NEED_RESCHED will
1243 * be set and schedule() will start a new hrtick for the next task.
1244 */
1245 if (hrtick_enabled(rq) && queued && p->dl.runtime > 0 &&
1246 is_leftmost(p, &rq->dl))
Dario Faggioliaab03e02013-11-28 11:14:43 +01001247 start_hrtick_dl(rq, p);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001248}
1249
1250static void task_fork_dl(struct task_struct *p)
1251{
1252 /*
1253 * SCHED_DEADLINE tasks cannot fork and this is achieved through
1254 * sched_fork()
1255 */
1256}
1257
1258static void task_dead_dl(struct task_struct *p)
1259{
Dario Faggioli332ac172013-11-07 14:43:45 +01001260 struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
1261
1262 /*
1263 * Since we are TASK_DEAD we won't slip out of the domain!
1264 */
1265 raw_spin_lock_irq(&dl_b->lock);
Peter Zijlstra40767b02015-01-28 15:08:03 +01001266 /* XXX we should retain the bw until 0-lag */
Dario Faggioli332ac172013-11-07 14:43:45 +01001267 dl_b->total_bw -= p->dl.dl_bw;
1268 raw_spin_unlock_irq(&dl_b->lock);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001269}
1270
1271static void set_curr_task_dl(struct rq *rq)
1272{
1273 struct task_struct *p = rq->curr;
1274
1275 p->se.exec_start = rq_clock_task(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001276
1277 /* You can't push away the running task */
1278 dequeue_pushable_dl_task(rq, p);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001279}
1280
Juri Lelli1baca4c2013-11-07 14:43:38 +01001281#ifdef CONFIG_SMP
1282
1283/* Only try algorithms three times */
1284#define DL_MAX_TRIES 3
1285
1286static int pick_dl_task(struct rq *rq, struct task_struct *p, int cpu)
1287{
1288 if (!task_running(rq, p) &&
Ingo Molnar0c98d342017-02-05 15:38:10 +01001289 cpumask_test_cpu(cpu, &p->cpus_allowed))
Juri Lelli1baca4c2013-11-07 14:43:38 +01001290 return 1;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001291 return 0;
1292}
1293
Wanpeng Li8b5e7702015-05-13 14:01:01 +08001294/*
1295 * Return the earliest pushable rq's task, which is suitable to be executed
1296 * on the CPU, NULL otherwise:
1297 */
1298static struct task_struct *pick_earliest_pushable_dl_task(struct rq *rq, int cpu)
1299{
1300 struct rb_node *next_node = rq->dl.pushable_dl_tasks_leftmost;
1301 struct task_struct *p = NULL;
1302
1303 if (!has_pushable_dl_tasks(rq))
1304 return NULL;
1305
1306next_node:
1307 if (next_node) {
1308 p = rb_entry(next_node, struct task_struct, pushable_dl_tasks);
1309
1310 if (pick_dl_task(rq, p, cpu))
1311 return p;
1312
1313 next_node = rb_next(next_node);
1314 goto next_node;
1315 }
1316
1317 return NULL;
1318}
1319
Juri Lelli1baca4c2013-11-07 14:43:38 +01001320static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
1321
1322static int find_later_rq(struct task_struct *task)
1323{
1324 struct sched_domain *sd;
Christoph Lameter4ba29682014-08-26 19:12:21 -05001325 struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001326 int this_cpu = smp_processor_id();
1327 int best_cpu, cpu = task_cpu(task);
1328
1329 /* Make sure the mask is initialized first */
1330 if (unlikely(!later_mask))
1331 return -1;
1332
Ingo Molnar4b53a342017-02-05 15:41:03 +01001333 if (task->nr_cpus_allowed == 1)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001334 return -1;
1335
Juri Lelli91ec6772014-09-19 10:22:41 +01001336 /*
1337 * We have to consider system topology and task affinity
1338 * first, then we can look for a suitable cpu.
1339 */
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001340 best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
1341 task, later_mask);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001342 if (best_cpu == -1)
1343 return -1;
1344
1345 /*
1346 * If we are here, some target has been found,
1347 * the most suitable of which is cached in best_cpu.
1348 * This is, among the runqueues where the current tasks
1349 * have later deadlines than the task's one, the rq
1350 * with the latest possible one.
1351 *
1352 * Now we check how well this matches with task's
1353 * affinity and system topology.
1354 *
1355 * The last cpu where the task run is our first
1356 * guess, since it is most likely cache-hot there.
1357 */
1358 if (cpumask_test_cpu(cpu, later_mask))
1359 return cpu;
1360 /*
1361 * Check if this_cpu is to be skipped (i.e., it is
1362 * not in the mask) or not.
1363 */
1364 if (!cpumask_test_cpu(this_cpu, later_mask))
1365 this_cpu = -1;
1366
1367 rcu_read_lock();
1368 for_each_domain(cpu, sd) {
1369 if (sd->flags & SD_WAKE_AFFINE) {
1370
1371 /*
1372 * If possible, preempting this_cpu is
1373 * cheaper than migrating.
1374 */
1375 if (this_cpu != -1 &&
1376 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1377 rcu_read_unlock();
1378 return this_cpu;
1379 }
1380
1381 /*
1382 * Last chance: if best_cpu is valid and is
1383 * in the mask, that becomes our choice.
1384 */
1385 if (best_cpu < nr_cpu_ids &&
1386 cpumask_test_cpu(best_cpu, sched_domain_span(sd))) {
1387 rcu_read_unlock();
1388 return best_cpu;
1389 }
1390 }
1391 }
1392 rcu_read_unlock();
1393
1394 /*
1395 * At this point, all our guesses failed, we just return
1396 * 'something', and let the caller sort the things out.
1397 */
1398 if (this_cpu != -1)
1399 return this_cpu;
1400
1401 cpu = cpumask_any(later_mask);
1402 if (cpu < nr_cpu_ids)
1403 return cpu;
1404
1405 return -1;
1406}
1407
1408/* Locks the rq it finds */
1409static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq)
1410{
1411 struct rq *later_rq = NULL;
1412 int tries;
1413 int cpu;
1414
1415 for (tries = 0; tries < DL_MAX_TRIES; tries++) {
1416 cpu = find_later_rq(task);
1417
1418 if ((cpu == -1) || (cpu == rq->cpu))
1419 break;
1420
1421 later_rq = cpu_rq(cpu);
1422
Luca Abeni5aa50502015-10-16 10:06:21 +02001423 if (later_rq->dl.dl_nr_running &&
1424 !dl_time_before(task->dl.deadline,
Wanpeng Li9d514262015-05-13 14:01:03 +08001425 later_rq->dl.earliest_dl.curr)) {
1426 /*
1427 * Target rq has tasks of equal or earlier deadline,
1428 * retrying does not release any lock and is unlikely
1429 * to yield a different result.
1430 */
1431 later_rq = NULL;
1432 break;
1433 }
1434
Juri Lelli1baca4c2013-11-07 14:43:38 +01001435 /* Retry if something changed. */
1436 if (double_lock_balance(rq, later_rq)) {
1437 if (unlikely(task_rq(task) != rq ||
Ingo Molnar0c98d342017-02-05 15:38:10 +01001438 !cpumask_test_cpu(later_rq->cpu, &task->cpus_allowed) ||
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001439 task_running(rq, task) ||
Xunlei Pang13b5ab02016-05-09 12:11:31 +08001440 !dl_task(task) ||
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001441 !task_on_rq_queued(task))) {
Juri Lelli1baca4c2013-11-07 14:43:38 +01001442 double_unlock_balance(rq, later_rq);
1443 later_rq = NULL;
1444 break;
1445 }
1446 }
1447
1448 /*
1449 * If the rq we found has no -deadline task, or
1450 * its earliest one has a later deadline than our
1451 * task, the rq is a good one.
1452 */
1453 if (!later_rq->dl.dl_nr_running ||
1454 dl_time_before(task->dl.deadline,
1455 later_rq->dl.earliest_dl.curr))
1456 break;
1457
1458 /* Otherwise we try again. */
1459 double_unlock_balance(rq, later_rq);
1460 later_rq = NULL;
1461 }
1462
1463 return later_rq;
1464}
1465
1466static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
1467{
1468 struct task_struct *p;
1469
1470 if (!has_pushable_dl_tasks(rq))
1471 return NULL;
1472
1473 p = rb_entry(rq->dl.pushable_dl_tasks_leftmost,
1474 struct task_struct, pushable_dl_tasks);
1475
1476 BUG_ON(rq->cpu != task_cpu(p));
1477 BUG_ON(task_current(rq, p));
Ingo Molnar4b53a342017-02-05 15:41:03 +01001478 BUG_ON(p->nr_cpus_allowed <= 1);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001479
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001480 BUG_ON(!task_on_rq_queued(p));
Juri Lelli1baca4c2013-11-07 14:43:38 +01001481 BUG_ON(!dl_task(p));
1482
1483 return p;
1484}
1485
1486/*
1487 * See if the non running -deadline tasks on this rq
1488 * can be sent to some other CPU where they can preempt
1489 * and start executing.
1490 */
1491static int push_dl_task(struct rq *rq)
1492{
1493 struct task_struct *next_task;
1494 struct rq *later_rq;
Wanpeng Lic51b8ab2014-11-06 15:22:44 +08001495 int ret = 0;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001496
1497 if (!rq->dl.overloaded)
1498 return 0;
1499
1500 next_task = pick_next_pushable_dl_task(rq);
1501 if (!next_task)
1502 return 0;
1503
1504retry:
1505 if (unlikely(next_task == rq->curr)) {
1506 WARN_ON(1);
1507 return 0;
1508 }
1509
1510 /*
1511 * If next_task preempts rq->curr, and rq->curr
1512 * can move away, it makes sense to just reschedule
1513 * without going further in pushing next_task.
1514 */
1515 if (dl_task(rq->curr) &&
1516 dl_time_before(next_task->dl.deadline, rq->curr->dl.deadline) &&
Ingo Molnar4b53a342017-02-05 15:41:03 +01001517 rq->curr->nr_cpus_allowed > 1) {
Kirill Tkhai88751252014-06-29 00:03:57 +04001518 resched_curr(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001519 return 0;
1520 }
1521
1522 /* We might release rq lock */
1523 get_task_struct(next_task);
1524
1525 /* Will lock the rq it'll find */
1526 later_rq = find_lock_later_rq(next_task, rq);
1527 if (!later_rq) {
1528 struct task_struct *task;
1529
1530 /*
1531 * We must check all this again, since
1532 * find_lock_later_rq releases rq->lock and it is
1533 * then possible that next_task has migrated.
1534 */
1535 task = pick_next_pushable_dl_task(rq);
1536 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1537 /*
1538 * The task is still there. We don't try
1539 * again, some other cpu will pull it when ready.
1540 */
Juri Lelli1baca4c2013-11-07 14:43:38 +01001541 goto out;
1542 }
1543
1544 if (!task)
1545 /* No more tasks */
1546 goto out;
1547
1548 put_task_struct(next_task);
1549 next_task = task;
1550 goto retry;
1551 }
1552
1553 deactivate_task(rq, next_task, 0);
1554 set_task_cpu(next_task, later_rq->cpu);
1555 activate_task(later_rq, next_task, 0);
Wanpeng Lic51b8ab2014-11-06 15:22:44 +08001556 ret = 1;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001557
Kirill Tkhai88751252014-06-29 00:03:57 +04001558 resched_curr(later_rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001559
1560 double_unlock_balance(rq, later_rq);
1561
1562out:
1563 put_task_struct(next_task);
1564
Wanpeng Lic51b8ab2014-11-06 15:22:44 +08001565 return ret;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001566}
1567
1568static void push_dl_tasks(struct rq *rq)
1569{
Andrea Parri4ffa08e2015-08-05 15:56:18 +02001570 /* push_dl_task() will return true if it moved a -deadline task */
Juri Lelli1baca4c2013-11-07 14:43:38 +01001571 while (push_dl_task(rq))
1572 ;
1573}
1574
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001575static void pull_dl_task(struct rq *this_rq)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001576{
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001577 int this_cpu = this_rq->cpu, cpu;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001578 struct task_struct *p;
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001579 bool resched = false;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001580 struct rq *src_rq;
1581 u64 dmin = LONG_MAX;
1582
1583 if (likely(!dl_overloaded(this_rq)))
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001584 return;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001585
1586 /*
1587 * Match the barrier from dl_set_overloaded; this guarantees that if we
1588 * see overloaded we must also see the dlo_mask bit.
1589 */
1590 smp_rmb();
1591
1592 for_each_cpu(cpu, this_rq->rd->dlo_mask) {
1593 if (this_cpu == cpu)
1594 continue;
1595
1596 src_rq = cpu_rq(cpu);
1597
1598 /*
1599 * It looks racy, abd it is! However, as in sched_rt.c,
1600 * we are fine with this.
1601 */
1602 if (this_rq->dl.dl_nr_running &&
1603 dl_time_before(this_rq->dl.earliest_dl.curr,
1604 src_rq->dl.earliest_dl.next))
1605 continue;
1606
1607 /* Might drop this_rq->lock */
1608 double_lock_balance(this_rq, src_rq);
1609
1610 /*
1611 * If there are no more pullable tasks on the
1612 * rq, we're done with it.
1613 */
1614 if (src_rq->dl.dl_nr_running <= 1)
1615 goto skip;
1616
Wanpeng Li8b5e7702015-05-13 14:01:01 +08001617 p = pick_earliest_pushable_dl_task(src_rq, this_cpu);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001618
1619 /*
1620 * We found a task to be pulled if:
1621 * - it preempts our current (if there's one),
1622 * - it will preempt the last one we pulled (if any).
1623 */
1624 if (p && dl_time_before(p->dl.deadline, dmin) &&
1625 (!this_rq->dl.dl_nr_running ||
1626 dl_time_before(p->dl.deadline,
1627 this_rq->dl.earliest_dl.curr))) {
1628 WARN_ON(p == src_rq->curr);
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001629 WARN_ON(!task_on_rq_queued(p));
Juri Lelli1baca4c2013-11-07 14:43:38 +01001630
1631 /*
1632 * Then we pull iff p has actually an earlier
1633 * deadline than the current task of its runqueue.
1634 */
1635 if (dl_time_before(p->dl.deadline,
1636 src_rq->curr->dl.deadline))
1637 goto skip;
1638
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001639 resched = true;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001640
1641 deactivate_task(src_rq, p, 0);
1642 set_task_cpu(p, this_cpu);
1643 activate_task(this_rq, p, 0);
1644 dmin = p->dl.deadline;
1645
1646 /* Is there any other task even earlier? */
1647 }
1648skip:
1649 double_unlock_balance(this_rq, src_rq);
1650 }
1651
Peter Zijlstra0ea60c22015-06-11 14:46:42 +02001652 if (resched)
1653 resched_curr(this_rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001654}
1655
1656/*
1657 * Since the task is not running and a reschedule is not going to happen
1658 * anytime soon on its runqueue, we try pushing it away now.
1659 */
1660static void task_woken_dl(struct rq *rq, struct task_struct *p)
1661{
1662 if (!task_running(rq, p) &&
1663 !test_tsk_need_resched(rq->curr) &&
Ingo Molnar4b53a342017-02-05 15:41:03 +01001664 p->nr_cpus_allowed > 1 &&
Juri Lelli1baca4c2013-11-07 14:43:38 +01001665 dl_task(rq->curr) &&
Ingo Molnar4b53a342017-02-05 15:41:03 +01001666 (rq->curr->nr_cpus_allowed < 2 ||
Wanpeng Li6b0a5632014-10-31 06:39:34 +08001667 !dl_entity_preempt(&p->dl, &rq->curr->dl))) {
Juri Lelli1baca4c2013-11-07 14:43:38 +01001668 push_dl_tasks(rq);
1669 }
1670}
1671
1672static void set_cpus_allowed_dl(struct task_struct *p,
1673 const struct cpumask *new_mask)
1674{
Juri Lelli7f514122014-09-19 10:22:40 +01001675 struct root_domain *src_rd;
Peter Zijlstra6c370672015-05-15 17:43:36 +02001676 struct rq *rq;
Juri Lelli1baca4c2013-11-07 14:43:38 +01001677
1678 BUG_ON(!dl_task(p));
1679
Juri Lelli7f514122014-09-19 10:22:40 +01001680 rq = task_rq(p);
1681 src_rd = rq->rd;
1682 /*
1683 * Migrating a SCHED_DEADLINE task between exclusive
1684 * cpusets (different root_domains) entails a bandwidth
1685 * update. We already made space for us in the destination
1686 * domain (see cpuset_can_attach()).
1687 */
1688 if (!cpumask_intersects(src_rd->span, new_mask)) {
1689 struct dl_bw *src_dl_b;
1690
1691 src_dl_b = dl_bw_of(cpu_of(rq));
1692 /*
1693 * We now free resources of the root_domain we are migrating
1694 * off. In the worst case, sched_setattr() may temporary fail
1695 * until we complete the update.
1696 */
1697 raw_spin_lock(&src_dl_b->lock);
1698 __dl_clear(src_dl_b, p->dl.dl_bw);
1699 raw_spin_unlock(&src_dl_b->lock);
1700 }
1701
Peter Zijlstra6c370672015-05-15 17:43:36 +02001702 set_cpus_allowed_common(p, new_mask);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001703}
1704
1705/* Assumes rq->lock is held */
1706static void rq_online_dl(struct rq *rq)
1707{
1708 if (rq->dl.overloaded)
1709 dl_set_overload(rq);
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001710
Xunlei Pang16b26942015-01-19 04:49:36 +00001711 cpudl_set_freecpu(&rq->rd->cpudl, rq->cpu);
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001712 if (rq->dl.dl_nr_running > 0)
Tommaso Cucinottad8206bb2016-08-14 16:27:08 +02001713 cpudl_set(&rq->rd->cpudl, rq->cpu, rq->dl.earliest_dl.curr);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001714}
1715
1716/* Assumes rq->lock is held */
1717static void rq_offline_dl(struct rq *rq)
1718{
1719 if (rq->dl.overloaded)
1720 dl_clear_overload(rq);
Juri Lelli6bfd6d72013-11-07 14:43:47 +01001721
Tommaso Cucinottad8206bb2016-08-14 16:27:08 +02001722 cpudl_clear(&rq->rd->cpudl, rq->cpu);
Xunlei Pang16b26942015-01-19 04:49:36 +00001723 cpudl_clear_freecpu(&rq->rd->cpudl, rq->cpu);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001724}
1725
Wanpeng Lia6c0e742015-05-13 14:01:02 +08001726void __init init_sched_dl_class(void)
Juri Lelli1baca4c2013-11-07 14:43:38 +01001727{
1728 unsigned int i;
1729
1730 for_each_possible_cpu(i)
1731 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask_dl, i),
1732 GFP_KERNEL, cpu_to_node(i));
1733}
1734
1735#endif /* CONFIG_SMP */
1736
Dario Faggioliaab03e02013-11-28 11:14:43 +01001737static void switched_from_dl(struct rq *rq, struct task_struct *p)
1738{
Peter Zijlstraa649f232015-06-11 14:46:49 +02001739 /*
1740 * Start the deadline timer; if we switch back to dl before this we'll
1741 * continue consuming our current CBS slice. If we stay outside of
1742 * SCHED_DEADLINE until the deadline passes, the timer will reset the
1743 * task.
1744 */
1745 if (!start_dl_timer(p))
1746 __dl_clear_params(p);
Juri Lellia5e7be32014-09-19 10:22:39 +01001747
Juri Lelli1baca4c2013-11-07 14:43:38 +01001748 /*
1749 * Since this might be the only -deadline task on the rq,
1750 * this is the right place to try to pull some other one
1751 * from an overloaded cpu, if any.
1752 */
Wanpeng Licd660912014-10-31 06:39:35 +08001753 if (!task_on_rq_queued(p) || rq->dl.dl_nr_running)
1754 return;
1755
Peter Zijlstra9916e212015-06-11 14:46:43 +02001756 queue_pull_task(rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001757}
1758
Juri Lelli1baca4c2013-11-07 14:43:38 +01001759/*
1760 * When switching to -deadline, we may overload the rq, then
1761 * we try to push someone off, if possible.
1762 */
Dario Faggioliaab03e02013-11-28 11:14:43 +01001763static void switched_to_dl(struct rq *rq, struct task_struct *p)
1764{
Luca Abeni72f9f3f2016-03-07 12:27:04 +01001765
Juri Lelli98b0a852016-08-05 16:07:55 +01001766 /* If p is not queued we will update its parameters at next wakeup. */
1767 if (!task_on_rq_queued(p))
1768 return;
1769
1770 /*
1771 * If p is boosted we already updated its params in
1772 * rt_mutex_setprio()->enqueue_task(..., ENQUEUE_REPLENISH),
1773 * p's deadline being now already after rq_clock(rq).
1774 */
1775 if (dl_time_before(p->dl.deadline, rq_clock(rq)))
1776 setup_new_dl_entity(&p->dl);
1777
1778 if (rq->curr != p) {
Juri Lelli1baca4c2013-11-07 14:43:38 +01001779#ifdef CONFIG_SMP
Ingo Molnar4b53a342017-02-05 15:41:03 +01001780 if (p->nr_cpus_allowed > 1 && rq->dl.overloaded)
Peter Zijlstra9916e212015-06-11 14:46:43 +02001781 queue_push_tasks(rq);
Sebastian Andrzej Siewior619bd4a2017-01-24 15:40:06 +01001782#endif
Peter Zijlstra9916e212015-06-11 14:46:43 +02001783 if (dl_task(rq->curr))
1784 check_preempt_curr_dl(rq, p, 0);
1785 else
1786 resched_curr(rq);
Dario Faggioliaab03e02013-11-28 11:14:43 +01001787 }
1788}
1789
Juri Lelli1baca4c2013-11-07 14:43:38 +01001790/*
1791 * If the scheduling parameters of a -deadline task changed,
1792 * a push or pull operation might be needed.
1793 */
Dario Faggioliaab03e02013-11-28 11:14:43 +01001794static void prio_changed_dl(struct rq *rq, struct task_struct *p,
1795 int oldprio)
1796{
Kirill Tkhaida0c1e62014-08-20 13:47:32 +04001797 if (task_on_rq_queued(p) || rq->curr == p) {
Dario Faggioliaab03e02013-11-28 11:14:43 +01001798#ifdef CONFIG_SMP
Juri Lelli1baca4c2013-11-07 14:43:38 +01001799 /*
1800 * This might be too much, but unfortunately
1801 * we don't have the old deadline value, and
1802 * we can't argue if the task is increasing
1803 * or lowering its prio, so...
1804 */
1805 if (!rq->dl.overloaded)
Peter Zijlstra9916e212015-06-11 14:46:43 +02001806 queue_pull_task(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001807
1808 /*
1809 * If we now have a earlier deadline task than p,
1810 * then reschedule, provided p is still on this
1811 * runqueue.
1812 */
Peter Zijlstra9916e212015-06-11 14:46:43 +02001813 if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline))
Kirill Tkhai88751252014-06-29 00:03:57 +04001814 resched_curr(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001815#else
1816 /*
1817 * Again, we don't know if p has a earlier
1818 * or later deadline, so let's blindly set a
1819 * (maybe not needed) rescheduling point.
1820 */
Kirill Tkhai88751252014-06-29 00:03:57 +04001821 resched_curr(rq);
Juri Lelli1baca4c2013-11-07 14:43:38 +01001822#endif /* CONFIG_SMP */
Peter Zijlstra801ccdb2016-02-25 15:01:49 +01001823 }
Dario Faggioliaab03e02013-11-28 11:14:43 +01001824}
Dario Faggioliaab03e02013-11-28 11:14:43 +01001825
1826const struct sched_class dl_sched_class = {
1827 .next = &rt_sched_class,
1828 .enqueue_task = enqueue_task_dl,
1829 .dequeue_task = dequeue_task_dl,
1830 .yield_task = yield_task_dl,
1831
1832 .check_preempt_curr = check_preempt_curr_dl,
1833
1834 .pick_next_task = pick_next_task_dl,
1835 .put_prev_task = put_prev_task_dl,
1836
1837#ifdef CONFIG_SMP
1838 .select_task_rq = select_task_rq_dl,
Juri Lelli1baca4c2013-11-07 14:43:38 +01001839 .set_cpus_allowed = set_cpus_allowed_dl,
1840 .rq_online = rq_online_dl,
1841 .rq_offline = rq_offline_dl,
Juri Lelli1baca4c2013-11-07 14:43:38 +01001842 .task_woken = task_woken_dl,
Dario Faggioliaab03e02013-11-28 11:14:43 +01001843#endif
1844
1845 .set_curr_task = set_curr_task_dl,
1846 .task_tick = task_tick_dl,
1847 .task_fork = task_fork_dl,
1848 .task_dead = task_dead_dl,
1849
1850 .prio_changed = prio_changed_dl,
1851 .switched_from = switched_from_dl,
1852 .switched_to = switched_to_dl,
Stanislaw Gruszka6e998912014-11-12 16:58:44 +01001853
1854 .update_curr = update_curr_dl,
Dario Faggioliaab03e02013-11-28 11:14:43 +01001855};
Wanpeng Liacb32132014-10-31 06:39:33 +08001856
1857#ifdef CONFIG_SCHED_DEBUG
1858extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
1859
1860void print_dl_stats(struct seq_file *m, int cpu)
1861{
1862 print_dl_rq(m, cpu, &cpu_rq(cpu)->dl);
1863}
1864#endif /* CONFIG_SCHED_DEBUG */