2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
6 #ifdef CONFIG_RT_GROUP_SCHED
8 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
10 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
12 #ifdef CONFIG_SCHED_DEBUG
13 WARN_ON_ONCE(!rt_entity_is_task(rt_se
));
15 return container_of(rt_se
, struct task_struct
, rt
);
18 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
23 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
28 #else /* CONFIG_RT_GROUP_SCHED */
30 #define rt_entity_is_task(rt_se) (1)
32 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
34 return container_of(rt_se
, struct task_struct
, rt
);
37 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
39 return container_of(rt_rq
, struct rq
, rt
);
42 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
44 struct task_struct
*p
= rt_task_of(rt_se
);
45 struct rq
*rq
= task_rq(p
);
50 #endif /* CONFIG_RT_GROUP_SCHED */
54 static inline int rt_overloaded(struct rq
*rq
)
56 return atomic_read(&rq
->rd
->rto_count
);
59 static inline void rt_set_overload(struct rq
*rq
)
64 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
66 * Make sure the mask is visible before we set
67 * the overload count. That is checked to determine
68 * if we should look at the mask. It would be a shame
69 * if we looked at the mask, but the mask was not
73 atomic_inc(&rq
->rd
->rto_count
);
76 static inline void rt_clear_overload(struct rq
*rq
)
81 /* the order here really doesn't matter */
82 atomic_dec(&rq
->rd
->rto_count
);
83 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
86 static void update_rt_migration(struct rt_rq
*rt_rq
)
88 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
89 if (!rt_rq
->overloaded
) {
90 rt_set_overload(rq_of_rt_rq(rt_rq
));
91 rt_rq
->overloaded
= 1;
93 } else if (rt_rq
->overloaded
) {
94 rt_clear_overload(rq_of_rt_rq(rt_rq
));
95 rt_rq
->overloaded
= 0;
99 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
101 if (!rt_entity_is_task(rt_se
))
104 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
106 rt_rq
->rt_nr_total
++;
107 if (rt_se
->nr_cpus_allowed
> 1)
108 rt_rq
->rt_nr_migratory
++;
110 update_rt_migration(rt_rq
);
113 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
115 if (!rt_entity_is_task(rt_se
))
118 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
120 rt_rq
->rt_nr_total
--;
121 if (rt_se
->nr_cpus_allowed
> 1)
122 rt_rq
->rt_nr_migratory
--;
124 update_rt_migration(rt_rq
);
127 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
129 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
130 plist_node_init(&p
->pushable_tasks
, p
->prio
);
131 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
134 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
136 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
139 static inline int has_pushable_tasks(struct rq
*rq
)
141 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
146 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
150 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
155 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
160 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
164 #endif /* CONFIG_SMP */
166 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
168 return !list_empty(&rt_se
->run_list
);
171 #ifdef CONFIG_RT_GROUP_SCHED
173 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
178 return rt_rq
->rt_runtime
;
181 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
183 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
186 #define for_each_leaf_rt_rq(rt_rq, rq) \
187 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
189 #define for_each_sched_rt_entity(rt_se) \
190 for (; rt_se; rt_se = rt_se->parent)
192 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
197 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
);
198 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
);
200 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
202 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
203 struct sched_rt_entity
*rt_se
;
205 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
207 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
209 if (rt_rq
->rt_nr_running
) {
210 if (rt_se
&& !on_rt_rq(rt_se
))
211 enqueue_rt_entity(rt_se
, false);
212 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
217 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
219 struct sched_rt_entity
*rt_se
;
220 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
222 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
224 if (rt_se
&& on_rt_rq(rt_se
))
225 dequeue_rt_entity(rt_se
);
228 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
230 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
233 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
235 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
236 struct task_struct
*p
;
239 return !!rt_rq
->rt_nr_boosted
;
241 p
= rt_task_of(rt_se
);
242 return p
->prio
!= p
->normal_prio
;
246 static inline const struct cpumask
*sched_rt_period_mask(void)
248 return cpu_rq(smp_processor_id())->rd
->span
;
251 static inline const struct cpumask
*sched_rt_period_mask(void)
253 return cpu_online_mask
;
258 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
260 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
263 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
265 return &rt_rq
->tg
->rt_bandwidth
;
268 #else /* !CONFIG_RT_GROUP_SCHED */
270 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
272 return rt_rq
->rt_runtime
;
275 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
277 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
280 #define for_each_leaf_rt_rq(rt_rq, rq) \
281 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
283 #define for_each_sched_rt_entity(rt_se) \
284 for (; rt_se; rt_se = NULL)
286 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
291 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
293 if (rt_rq
->rt_nr_running
)
294 resched_task(rq_of_rt_rq(rt_rq
)->curr
);
297 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
301 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
303 return rt_rq
->rt_throttled
;
306 static inline const struct cpumask
*sched_rt_period_mask(void)
308 return cpu_online_mask
;
312 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
314 return &cpu_rq(cpu
)->rt
;
317 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
319 return &def_rt_bandwidth
;
322 #endif /* CONFIG_RT_GROUP_SCHED */
326 * We ran out of runtime, see if we can borrow some from our neighbours.
328 static int do_balance_runtime(struct rt_rq
*rt_rq
)
330 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
331 struct root_domain
*rd
= cpu_rq(smp_processor_id())->rd
;
332 int i
, weight
, more
= 0;
335 weight
= cpumask_weight(rd
->span
);
337 raw_spin_lock(&rt_b
->rt_runtime_lock
);
338 rt_period
= ktime_to_ns(rt_b
->rt_period
);
339 for_each_cpu(i
, rd
->span
) {
340 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
346 raw_spin_lock(&iter
->rt_runtime_lock
);
348 * Either all rqs have inf runtime and there's nothing to steal
349 * or __disable_runtime() below sets a specific rq to inf to
350 * indicate its been disabled and disalow stealing.
352 if (iter
->rt_runtime
== RUNTIME_INF
)
356 * From runqueues with spare time, take 1/n part of their
357 * spare time, but no more than our period.
359 diff
= iter
->rt_runtime
- iter
->rt_time
;
361 diff
= div_u64((u64
)diff
, weight
);
362 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
363 diff
= rt_period
- rt_rq
->rt_runtime
;
364 iter
->rt_runtime
-= diff
;
365 rt_rq
->rt_runtime
+= diff
;
367 if (rt_rq
->rt_runtime
== rt_period
) {
368 raw_spin_unlock(&iter
->rt_runtime_lock
);
373 raw_spin_unlock(&iter
->rt_runtime_lock
);
375 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
381 * Ensure this RQ takes back all the runtime it lend to its neighbours.
383 static void __disable_runtime(struct rq
*rq
)
385 struct root_domain
*rd
= rq
->rd
;
388 if (unlikely(!scheduler_running
))
391 for_each_leaf_rt_rq(rt_rq
, rq
) {
392 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
396 raw_spin_lock(&rt_b
->rt_runtime_lock
);
397 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
399 * Either we're all inf and nobody needs to borrow, or we're
400 * already disabled and thus have nothing to do, or we have
401 * exactly the right amount of runtime to take out.
403 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
404 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
406 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
409 * Calculate the difference between what we started out with
410 * and what we current have, that's the amount of runtime
411 * we lend and now have to reclaim.
413 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
416 * Greedy reclaim, take back as much as we can.
418 for_each_cpu(i
, rd
->span
) {
419 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
423 * Can't reclaim from ourselves or disabled runqueues.
425 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
428 raw_spin_lock(&iter
->rt_runtime_lock
);
430 diff
= min_t(s64
, iter
->rt_runtime
, want
);
431 iter
->rt_runtime
-= diff
;
434 iter
->rt_runtime
-= want
;
437 raw_spin_unlock(&iter
->rt_runtime_lock
);
443 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
445 * We cannot be left wanting - that would mean some runtime
446 * leaked out of the system.
451 * Disable all the borrow logic by pretending we have inf
452 * runtime - in which case borrowing doesn't make sense.
454 rt_rq
->rt_runtime
= RUNTIME_INF
;
455 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
456 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
460 static void disable_runtime(struct rq
*rq
)
464 raw_spin_lock_irqsave(&rq
->lock
, flags
);
465 __disable_runtime(rq
);
466 raw_spin_unlock_irqrestore(&rq
->lock
, flags
);
469 static void __enable_runtime(struct rq
*rq
)
473 if (unlikely(!scheduler_running
))
477 * Reset each runqueue's bandwidth settings
479 for_each_leaf_rt_rq(rt_rq
, rq
) {
480 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
482 raw_spin_lock(&rt_b
->rt_runtime_lock
);
483 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
484 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
486 rt_rq
->rt_throttled
= 0;
487 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
488 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
492 static void enable_runtime(struct rq
*rq
)
496 raw_spin_lock_irqsave(&rq
->lock
, flags
);
497 __enable_runtime(rq
);
498 raw_spin_unlock_irqrestore(&rq
->lock
, flags
);
501 static int balance_runtime(struct rt_rq
*rt_rq
)
505 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
506 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
507 more
= do_balance_runtime(rt_rq
);
508 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
513 #else /* !CONFIG_SMP */
514 static inline int balance_runtime(struct rt_rq
*rt_rq
)
518 #endif /* CONFIG_SMP */
520 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
523 const struct cpumask
*span
;
525 if (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
)
528 span
= sched_rt_period_mask();
529 for_each_cpu(i
, span
) {
531 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
532 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
534 raw_spin_lock(&rq
->lock
);
535 if (rt_rq
->rt_time
) {
538 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
539 if (rt_rq
->rt_throttled
)
540 balance_runtime(rt_rq
);
541 runtime
= rt_rq
->rt_runtime
;
542 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
543 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
544 rt_rq
->rt_throttled
= 0;
547 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
549 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
550 } else if (rt_rq
->rt_nr_running
) {
552 if (!rt_rq_throttled(rt_rq
))
557 sched_rt_rq_enqueue(rt_rq
);
558 raw_spin_unlock(&rq
->lock
);
564 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
566 #ifdef CONFIG_RT_GROUP_SCHED
567 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
570 return rt_rq
->highest_prio
.curr
;
573 return rt_task_of(rt_se
)->prio
;
576 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
578 u64 runtime
= sched_rt_runtime(rt_rq
);
580 if (rt_rq
->rt_throttled
)
581 return rt_rq_throttled(rt_rq
);
583 if (sched_rt_runtime(rt_rq
) >= sched_rt_period(rt_rq
))
586 balance_runtime(rt_rq
);
587 runtime
= sched_rt_runtime(rt_rq
);
588 if (runtime
== RUNTIME_INF
)
591 if (rt_rq
->rt_time
> runtime
) {
592 rt_rq
->rt_throttled
= 1;
593 if (rt_rq_throttled(rt_rq
)) {
594 sched_rt_rq_dequeue(rt_rq
);
603 * Update the current task's runtime statistics. Skip current tasks that
604 * are not in our scheduling class.
606 static void update_curr_rt(struct rq
*rq
)
608 struct task_struct
*curr
= rq
->curr
;
609 struct sched_rt_entity
*rt_se
= &curr
->rt
;
610 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
613 if (curr
->sched_class
!= &rt_sched_class
)
616 delta_exec
= rq
->clock_task
- curr
->se
.exec_start
;
617 if (unlikely((s64
)delta_exec
< 0))
620 schedstat_set(curr
->se
.statistics
.exec_max
, max(curr
->se
.statistics
.exec_max
, delta_exec
));
622 curr
->se
.sum_exec_runtime
+= delta_exec
;
623 account_group_exec_runtime(curr
, delta_exec
);
625 curr
->se
.exec_start
= rq
->clock_task
;
626 cpuacct_charge(curr
, delta_exec
);
628 sched_rt_avg_update(rq
, delta_exec
);
630 if (!rt_bandwidth_enabled())
633 for_each_sched_rt_entity(rt_se
) {
634 rt_rq
= rt_rq_of_se(rt_se
);
636 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
637 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
638 rt_rq
->rt_time
+= delta_exec
;
639 if (sched_rt_runtime_exceeded(rt_rq
))
641 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
646 #if defined CONFIG_SMP
648 static struct task_struct
*pick_next_highest_task_rt(struct rq
*rq
, int cpu
);
650 static inline int next_prio(struct rq
*rq
)
652 struct task_struct
*next
= pick_next_highest_task_rt(rq
, rq
->cpu
);
654 if (next
&& rt_prio(next
->prio
))
661 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
663 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
665 if (prio
< prev_prio
) {
668 * If the new task is higher in priority than anything on the
669 * run-queue, we know that the previous high becomes our
672 rt_rq
->highest_prio
.next
= prev_prio
;
675 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
677 } else if (prio
== rt_rq
->highest_prio
.curr
)
679 * If the next task is equal in priority to the highest on
680 * the run-queue, then we implicitly know that the next highest
681 * task cannot be any lower than current
683 rt_rq
->highest_prio
.next
= prio
;
684 else if (prio
< rt_rq
->highest_prio
.next
)
686 * Otherwise, we need to recompute next-highest
688 rt_rq
->highest_prio
.next
= next_prio(rq
);
692 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
694 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
696 if (rt_rq
->rt_nr_running
&& (prio
<= rt_rq
->highest_prio
.next
))
697 rt_rq
->highest_prio
.next
= next_prio(rq
);
699 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
700 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
703 #else /* CONFIG_SMP */
706 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
708 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
710 #endif /* CONFIG_SMP */
712 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
714 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
716 int prev_prio
= rt_rq
->highest_prio
.curr
;
718 if (prio
< prev_prio
)
719 rt_rq
->highest_prio
.curr
= prio
;
721 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
725 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
727 int prev_prio
= rt_rq
->highest_prio
.curr
;
729 if (rt_rq
->rt_nr_running
) {
731 WARN_ON(prio
< prev_prio
);
734 * This may have been our highest task, and therefore
735 * we may have some recomputation to do
737 if (prio
== prev_prio
) {
738 struct rt_prio_array
*array
= &rt_rq
->active
;
740 rt_rq
->highest_prio
.curr
=
741 sched_find_first_bit(array
->bitmap
);
745 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
747 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
752 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
753 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
755 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
757 #ifdef CONFIG_RT_GROUP_SCHED
760 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
762 if (rt_se_boosted(rt_se
))
763 rt_rq
->rt_nr_boosted
++;
766 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
770 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
772 if (rt_se_boosted(rt_se
))
773 rt_rq
->rt_nr_boosted
--;
775 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
778 #else /* CONFIG_RT_GROUP_SCHED */
781 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
783 start_rt_bandwidth(&def_rt_bandwidth
);
787 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
789 #endif /* CONFIG_RT_GROUP_SCHED */
792 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
794 int prio
= rt_se_prio(rt_se
);
796 WARN_ON(!rt_prio(prio
));
797 rt_rq
->rt_nr_running
++;
799 inc_rt_prio(rt_rq
, prio
);
800 inc_rt_migration(rt_se
, rt_rq
);
801 inc_rt_group(rt_se
, rt_rq
);
805 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
807 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
808 WARN_ON(!rt_rq
->rt_nr_running
);
809 rt_rq
->rt_nr_running
--;
811 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
812 dec_rt_migration(rt_se
, rt_rq
);
813 dec_rt_group(rt_se
, rt_rq
);
816 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
818 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
819 struct rt_prio_array
*array
= &rt_rq
->active
;
820 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
821 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
824 * Don't enqueue the group if its throttled, or when empty.
825 * The latter is a consequence of the former when a child group
826 * get throttled and the current group doesn't have any other
829 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
))
833 list_add(&rt_se
->run_list
, queue
);
835 list_add_tail(&rt_se
->run_list
, queue
);
836 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
838 inc_rt_tasks(rt_se
, rt_rq
);
841 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
843 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
844 struct rt_prio_array
*array
= &rt_rq
->active
;
846 list_del_init(&rt_se
->run_list
);
847 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
848 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
850 dec_rt_tasks(rt_se
, rt_rq
);
854 * Because the prio of an upper entry depends on the lower
855 * entries, we must remove entries top - down.
857 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
)
859 struct sched_rt_entity
*back
= NULL
;
861 for_each_sched_rt_entity(rt_se
) {
866 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
868 __dequeue_rt_entity(rt_se
);
872 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
874 dequeue_rt_stack(rt_se
);
875 for_each_sched_rt_entity(rt_se
)
876 __enqueue_rt_entity(rt_se
, head
);
879 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
881 dequeue_rt_stack(rt_se
);
883 for_each_sched_rt_entity(rt_se
) {
884 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
886 if (rt_rq
&& rt_rq
->rt_nr_running
)
887 __enqueue_rt_entity(rt_se
, false);
892 * Adding/removing a task to/from a priority array:
895 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
897 struct sched_rt_entity
*rt_se
= &p
->rt
;
899 if (flags
& ENQUEUE_WAKEUP
)
902 enqueue_rt_entity(rt_se
, flags
& ENQUEUE_HEAD
);
904 if (!task_current(rq
, p
) && p
->rt
.nr_cpus_allowed
> 1)
905 enqueue_pushable_task(rq
, p
);
908 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
910 struct sched_rt_entity
*rt_se
= &p
->rt
;
913 dequeue_rt_entity(rt_se
);
915 dequeue_pushable_task(rq
, p
);
919 * Put task to the end of the run list without the overhead of dequeue
920 * followed by enqueue.
923 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
925 if (on_rt_rq(rt_se
)) {
926 struct rt_prio_array
*array
= &rt_rq
->active
;
927 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
930 list_move(&rt_se
->run_list
, queue
);
932 list_move_tail(&rt_se
->run_list
, queue
);
936 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
938 struct sched_rt_entity
*rt_se
= &p
->rt
;
941 for_each_sched_rt_entity(rt_se
) {
942 rt_rq
= rt_rq_of_se(rt_se
);
943 requeue_rt_entity(rt_rq
, rt_se
, head
);
947 static void yield_task_rt(struct rq
*rq
)
949 requeue_task_rt(rq
, rq
->curr
, 0);
953 static int find_lowest_rq(struct task_struct
*task
);
956 select_task_rq_rt(struct rq
*rq
, struct task_struct
*p
, int sd_flag
, int flags
)
958 if (sd_flag
!= SD_BALANCE_WAKE
)
959 return smp_processor_id();
962 * If the current task is an RT task, then
963 * try to see if we can wake this RT task up on another
964 * runqueue. Otherwise simply start this RT task
965 * on its current runqueue.
967 * We want to avoid overloading runqueues. If the woken
968 * task is a higher priority, then it will stay on this CPU
969 * and the lower prio task should be moved to another CPU.
970 * Even though this will probably make the lower prio task
971 * lose its cache, we do not want to bounce a higher task
972 * around just because it gave up its CPU, perhaps for a
975 * For equal prio tasks, we just let the scheduler sort it out.
977 if (unlikely(rt_task(rq
->curr
)) &&
978 (rq
->curr
->rt
.nr_cpus_allowed
< 2 ||
979 rq
->curr
->prio
< p
->prio
) &&
980 (p
->rt
.nr_cpus_allowed
> 1)) {
981 int cpu
= find_lowest_rq(p
);
983 return (cpu
== -1) ? task_cpu(p
) : cpu
;
987 * Otherwise, just let it ride on the affined RQ and the
988 * post-schedule router will push the preempted task away
993 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
995 if (rq
->curr
->rt
.nr_cpus_allowed
== 1)
998 if (p
->rt
.nr_cpus_allowed
!= 1
999 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
1002 if (!cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
1006 * There appears to be other cpus that can accept
1007 * current and none to run 'p', so lets reschedule
1008 * to try and push current away:
1010 requeue_task_rt(rq
, p
, 1);
1011 resched_task(rq
->curr
);
1014 #endif /* CONFIG_SMP */
1017 * Preempt the current task with a newly woken task if needed:
1019 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1021 if (p
->prio
< rq
->curr
->prio
) {
1022 resched_task(rq
->curr
);
1030 * - the newly woken task is of equal priority to the current task
1031 * - the newly woken task is non-migratable while current is migratable
1032 * - current will be preempted on the next reschedule
1034 * we should check to see if current can readily move to a different
1035 * cpu. If so, we will reschedule to allow the push logic to try
1036 * to move current somewhere else, making room for our non-migratable
1039 if (p
->prio
== rq
->curr
->prio
&& !need_resched())
1040 check_preempt_equal_prio(rq
, p
);
1044 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1045 struct rt_rq
*rt_rq
)
1047 struct rt_prio_array
*array
= &rt_rq
->active
;
1048 struct sched_rt_entity
*next
= NULL
;
1049 struct list_head
*queue
;
1052 idx
= sched_find_first_bit(array
->bitmap
);
1053 BUG_ON(idx
>= MAX_RT_PRIO
);
1055 queue
= array
->queue
+ idx
;
1056 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1061 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1063 struct sched_rt_entity
*rt_se
;
1064 struct task_struct
*p
;
1065 struct rt_rq
*rt_rq
;
1069 if (unlikely(!rt_rq
->rt_nr_running
))
1072 if (rt_rq_throttled(rt_rq
))
1076 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1078 rt_rq
= group_rt_rq(rt_se
);
1081 p
= rt_task_of(rt_se
);
1082 p
->se
.exec_start
= rq
->clock_task
;
1087 static struct task_struct
*pick_next_task_rt(struct rq
*rq
)
1089 struct task_struct
*p
= _pick_next_task_rt(rq
);
1091 /* The running task is never eligible for pushing */
1093 dequeue_pushable_task(rq
, p
);
1097 * We detect this state here so that we can avoid taking the RQ
1098 * lock again later if there is no need to push
1100 rq
->post_schedule
= has_pushable_tasks(rq
);
1106 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1109 p
->se
.exec_start
= 0;
1112 * The previous task needs to be made eligible for pushing
1113 * if it is still active
1115 if (p
->se
.on_rq
&& p
->rt
.nr_cpus_allowed
> 1)
1116 enqueue_pushable_task(rq
, p
);
1121 /* Only try algorithms three times */
1122 #define RT_MAX_TRIES 3
1124 static void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int sleep
);
1126 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1128 if (!task_running(rq
, p
) &&
1129 (cpu
< 0 || cpumask_test_cpu(cpu
, &p
->cpus_allowed
)) &&
1130 (p
->rt
.nr_cpus_allowed
> 1))
1135 /* Return the second highest RT task, NULL otherwise */
1136 static struct task_struct
*pick_next_highest_task_rt(struct rq
*rq
, int cpu
)
1138 struct task_struct
*next
= NULL
;
1139 struct sched_rt_entity
*rt_se
;
1140 struct rt_prio_array
*array
;
1141 struct rt_rq
*rt_rq
;
1144 for_each_leaf_rt_rq(rt_rq
, rq
) {
1145 array
= &rt_rq
->active
;
1146 idx
= sched_find_first_bit(array
->bitmap
);
1148 if (idx
>= MAX_RT_PRIO
)
1150 if (next
&& next
->prio
< idx
)
1152 list_for_each_entry(rt_se
, array
->queue
+ idx
, run_list
) {
1153 struct task_struct
*p
;
1155 if (!rt_entity_is_task(rt_se
))
1158 p
= rt_task_of(rt_se
);
1159 if (pick_rt_task(rq
, p
, cpu
)) {
1165 idx
= find_next_bit(array
->bitmap
, MAX_RT_PRIO
, idx
+1);
1173 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1175 static int find_lowest_rq(struct task_struct
*task
)
1177 struct sched_domain
*sd
;
1178 struct cpumask
*lowest_mask
= __get_cpu_var(local_cpu_mask
);
1179 int this_cpu
= smp_processor_id();
1180 int cpu
= task_cpu(task
);
1182 if (task
->rt
.nr_cpus_allowed
== 1)
1183 return -1; /* No other targets possible */
1185 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1186 return -1; /* No targets found */
1189 * At this point we have built a mask of cpus representing the
1190 * lowest priority tasks in the system. Now we want to elect
1191 * the best one based on our affinity and topology.
1193 * We prioritize the last cpu that the task executed on since
1194 * it is most likely cache-hot in that location.
1196 if (cpumask_test_cpu(cpu
, lowest_mask
))
1200 * Otherwise, we consult the sched_domains span maps to figure
1201 * out which cpu is logically closest to our hot cache data.
1203 if (!cpumask_test_cpu(this_cpu
, lowest_mask
))
1204 this_cpu
= -1; /* Skip this_cpu opt if not among lowest */
1206 for_each_domain(cpu
, sd
) {
1207 if (sd
->flags
& SD_WAKE_AFFINE
) {
1211 * "this_cpu" is cheaper to preempt than a
1214 if (this_cpu
!= -1 &&
1215 cpumask_test_cpu(this_cpu
, sched_domain_span(sd
)))
1218 best_cpu
= cpumask_first_and(lowest_mask
,
1219 sched_domain_span(sd
));
1220 if (best_cpu
< nr_cpu_ids
)
1226 * And finally, if there were no matches within the domains
1227 * just give the caller *something* to work with from the compatible
1233 cpu
= cpumask_any(lowest_mask
);
1234 if (cpu
< nr_cpu_ids
)
1239 /* Will lock the rq it finds */
1240 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1242 struct rq
*lowest_rq
= NULL
;
1246 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1247 cpu
= find_lowest_rq(task
);
1249 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1252 lowest_rq
= cpu_rq(cpu
);
1254 /* if the prio of this runqueue changed, try again */
1255 if (double_lock_balance(rq
, lowest_rq
)) {
1257 * We had to unlock the run queue. In
1258 * the mean time, task could have
1259 * migrated already or had its affinity changed.
1260 * Also make sure that it wasn't scheduled on its rq.
1262 if (unlikely(task_rq(task
) != rq
||
1263 !cpumask_test_cpu(lowest_rq
->cpu
,
1264 &task
->cpus_allowed
) ||
1265 task_running(rq
, task
) ||
1268 raw_spin_unlock(&lowest_rq
->lock
);
1274 /* If this rq is still suitable use it. */
1275 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1279 double_unlock_balance(rq
, lowest_rq
);
1286 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1288 struct task_struct
*p
;
1290 if (!has_pushable_tasks(rq
))
1293 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1294 struct task_struct
, pushable_tasks
);
1296 BUG_ON(rq
->cpu
!= task_cpu(p
));
1297 BUG_ON(task_current(rq
, p
));
1298 BUG_ON(p
->rt
.nr_cpus_allowed
<= 1);
1300 BUG_ON(!p
->se
.on_rq
);
1301 BUG_ON(!rt_task(p
));
1307 * If the current CPU has more than one RT task, see if the non
1308 * running task can migrate over to a CPU that is running a task
1309 * of lesser priority.
1311 static int push_rt_task(struct rq
*rq
)
1313 struct task_struct
*next_task
;
1314 struct rq
*lowest_rq
;
1316 if (!rq
->rt
.overloaded
)
1319 next_task
= pick_next_pushable_task(rq
);
1324 if (unlikely(next_task
== rq
->curr
)) {
1330 * It's possible that the next_task slipped in of
1331 * higher priority than current. If that's the case
1332 * just reschedule current.
1334 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1335 resched_task(rq
->curr
);
1339 /* We might release rq lock */
1340 get_task_struct(next_task
);
1342 /* find_lock_lowest_rq locks the rq if found */
1343 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1345 struct task_struct
*task
;
1347 * find lock_lowest_rq releases rq->lock
1348 * so it is possible that next_task has migrated.
1350 * We need to make sure that the task is still on the same
1351 * run-queue and is also still the next task eligible for
1354 task
= pick_next_pushable_task(rq
);
1355 if (task_cpu(next_task
) == rq
->cpu
&& task
== next_task
) {
1357 * If we get here, the task hasnt moved at all, but
1358 * it has failed to push. We will not try again,
1359 * since the other cpus will pull from us when they
1362 dequeue_pushable_task(rq
, next_task
);
1367 /* No more tasks, just exit */
1371 * Something has shifted, try again.
1373 put_task_struct(next_task
);
1378 deactivate_task(rq
, next_task
, 0);
1379 set_task_cpu(next_task
, lowest_rq
->cpu
);
1380 activate_task(lowest_rq
, next_task
, 0);
1382 resched_task(lowest_rq
->curr
);
1384 double_unlock_balance(rq
, lowest_rq
);
1387 put_task_struct(next_task
);
1392 static void push_rt_tasks(struct rq
*rq
)
1394 /* push_rt_task will return true if it moved an RT */
1395 while (push_rt_task(rq
))
1399 static int pull_rt_task(struct rq
*this_rq
)
1401 int this_cpu
= this_rq
->cpu
, ret
= 0, cpu
;
1402 struct task_struct
*p
;
1405 if (likely(!rt_overloaded(this_rq
)))
1408 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
1409 if (this_cpu
== cpu
)
1412 src_rq
= cpu_rq(cpu
);
1415 * Don't bother taking the src_rq->lock if the next highest
1416 * task is known to be lower-priority than our current task.
1417 * This may look racy, but if this value is about to go
1418 * logically higher, the src_rq will push this task away.
1419 * And if its going logically lower, we do not care
1421 if (src_rq
->rt
.highest_prio
.next
>=
1422 this_rq
->rt
.highest_prio
.curr
)
1426 * We can potentially drop this_rq's lock in
1427 * double_lock_balance, and another CPU could
1430 double_lock_balance(this_rq
, src_rq
);
1433 * Are there still pullable RT tasks?
1435 if (src_rq
->rt
.rt_nr_running
<= 1)
1438 p
= pick_next_highest_task_rt(src_rq
, this_cpu
);
1441 * Do we have an RT task that preempts
1442 * the to-be-scheduled task?
1444 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
1445 WARN_ON(p
== src_rq
->curr
);
1446 WARN_ON(!p
->se
.on_rq
);
1449 * There's a chance that p is higher in priority
1450 * than what's currently running on its cpu.
1451 * This is just that p is wakeing up and hasn't
1452 * had a chance to schedule. We only pull
1453 * p if it is lower in priority than the
1454 * current task on the run queue
1456 if (p
->prio
< src_rq
->curr
->prio
)
1461 deactivate_task(src_rq
, p
, 0);
1462 set_task_cpu(p
, this_cpu
);
1463 activate_task(this_rq
, p
, 0);
1465 * We continue with the search, just in
1466 * case there's an even higher prio task
1467 * in another runqueue. (low likelyhood
1472 double_unlock_balance(this_rq
, src_rq
);
1478 static void pre_schedule_rt(struct rq
*rq
, struct task_struct
*prev
)
1480 /* Try to pull RT tasks here if we lower this rq's prio */
1481 if (unlikely(rt_task(prev
)) && rq
->rt
.highest_prio
.curr
> prev
->prio
)
1485 static void post_schedule_rt(struct rq
*rq
)
1491 * If we are not running and we are not going to reschedule soon, we should
1492 * try to push tasks away now
1494 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
1496 if (!task_running(rq
, p
) &&
1497 !test_tsk_need_resched(rq
->curr
) &&
1498 has_pushable_tasks(rq
) &&
1499 p
->rt
.nr_cpus_allowed
> 1 &&
1500 rt_task(rq
->curr
) &&
1501 (rq
->curr
->rt
.nr_cpus_allowed
< 2 ||
1502 rq
->curr
->prio
< p
->prio
))
1506 static void set_cpus_allowed_rt(struct task_struct
*p
,
1507 const struct cpumask
*new_mask
)
1509 int weight
= cpumask_weight(new_mask
);
1511 BUG_ON(!rt_task(p
));
1514 * Update the migration status of the RQ if we have an RT task
1515 * which is running AND changing its weight value.
1517 if (p
->se
.on_rq
&& (weight
!= p
->rt
.nr_cpus_allowed
)) {
1518 struct rq
*rq
= task_rq(p
);
1520 if (!task_current(rq
, p
)) {
1522 * Make sure we dequeue this task from the pushable list
1523 * before going further. It will either remain off of
1524 * the list because we are no longer pushable, or it
1527 if (p
->rt
.nr_cpus_allowed
> 1)
1528 dequeue_pushable_task(rq
, p
);
1531 * Requeue if our weight is changing and still > 1
1534 enqueue_pushable_task(rq
, p
);
1538 if ((p
->rt
.nr_cpus_allowed
<= 1) && (weight
> 1)) {
1539 rq
->rt
.rt_nr_migratory
++;
1540 } else if ((p
->rt
.nr_cpus_allowed
> 1) && (weight
<= 1)) {
1541 BUG_ON(!rq
->rt
.rt_nr_migratory
);
1542 rq
->rt
.rt_nr_migratory
--;
1545 update_rt_migration(&rq
->rt
);
1548 cpumask_copy(&p
->cpus_allowed
, new_mask
);
1549 p
->rt
.nr_cpus_allowed
= weight
;
1552 /* Assumes rq->lock is held */
1553 static void rq_online_rt(struct rq
*rq
)
1555 if (rq
->rt
.overloaded
)
1556 rt_set_overload(rq
);
1558 __enable_runtime(rq
);
1560 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
1563 /* Assumes rq->lock is held */
1564 static void rq_offline_rt(struct rq
*rq
)
1566 if (rq
->rt
.overloaded
)
1567 rt_clear_overload(rq
);
1569 __disable_runtime(rq
);
1571 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
1575 * When switch from the rt queue, we bring ourselves to a position
1576 * that we might want to pull RT tasks from other runqueues.
1578 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
,
1582 * If there are other RT tasks then we will reschedule
1583 * and the scheduling of the other RT tasks will handle
1584 * the balancing. But if we are the last RT task
1585 * we may need to handle the pulling of RT tasks
1588 if (!rq
->rt
.rt_nr_running
)
1592 static inline void init_sched_rt_class(void)
1596 for_each_possible_cpu(i
)
1597 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
1598 GFP_KERNEL
, cpu_to_node(i
));
1600 #endif /* CONFIG_SMP */
1603 * When switching a task to RT, we may overload the runqueue
1604 * with RT tasks. In this case we try to push them off to
1607 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
,
1610 int check_resched
= 1;
1613 * If we are already running, then there's nothing
1614 * that needs to be done. But if we are not running
1615 * we may need to preempt the current running task.
1616 * If that current running task is also an RT task
1617 * then see if we can move to another run queue.
1621 if (rq
->rt
.overloaded
&& push_rt_task(rq
) &&
1622 /* Don't resched if we changed runqueues */
1625 #endif /* CONFIG_SMP */
1626 if (check_resched
&& p
->prio
< rq
->curr
->prio
)
1627 resched_task(rq
->curr
);
1632 * Priority of the task has changed. This may cause
1633 * us to initiate a push or pull.
1635 static void prio_changed_rt(struct rq
*rq
, struct task_struct
*p
,
1636 int oldprio
, int running
)
1641 * If our priority decreases while running, we
1642 * may need to pull tasks to this runqueue.
1644 if (oldprio
< p
->prio
)
1647 * If there's a higher priority task waiting to run
1648 * then reschedule. Note, the above pull_rt_task
1649 * can release the rq lock and p could migrate.
1650 * Only reschedule if p is still on the same runqueue.
1652 if (p
->prio
> rq
->rt
.highest_prio
.curr
&& rq
->curr
== p
)
1655 /* For UP simply resched on drop of prio */
1656 if (oldprio
< p
->prio
)
1658 #endif /* CONFIG_SMP */
1661 * This task is not running, but if it is
1662 * greater than the current running task
1665 if (p
->prio
< rq
->curr
->prio
)
1666 resched_task(rq
->curr
);
1670 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
1672 unsigned long soft
, hard
;
1674 /* max may change after cur was read, this will be fixed next tick */
1675 soft
= task_rlimit(p
, RLIMIT_RTTIME
);
1676 hard
= task_rlimit_max(p
, RLIMIT_RTTIME
);
1678 if (soft
!= RLIM_INFINITY
) {
1682 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
1683 if (p
->rt
.timeout
> next
)
1684 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
1688 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
1695 * RR tasks need a special form of timeslice management.
1696 * FIFO tasks have no timeslices.
1698 if (p
->policy
!= SCHED_RR
)
1701 if (--p
->rt
.time_slice
)
1704 p
->rt
.time_slice
= DEF_TIMESLICE
;
1707 * Requeue to the end of queue if we are not the only element
1710 if (p
->rt
.run_list
.prev
!= p
->rt
.run_list
.next
) {
1711 requeue_task_rt(rq
, p
, 0);
1712 set_tsk_need_resched(p
);
1716 static void set_curr_task_rt(struct rq
*rq
)
1718 struct task_struct
*p
= rq
->curr
;
1720 p
->se
.exec_start
= rq
->clock_task
;
1722 /* The running task is never eligible for pushing */
1723 dequeue_pushable_task(rq
, p
);
1726 static unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
1729 * Time slice is 0 for SCHED_FIFO tasks
1731 if (task
->policy
== SCHED_RR
)
1732 return DEF_TIMESLICE
;
1737 static const struct sched_class rt_sched_class
= {
1738 .next
= &fair_sched_class
,
1739 .enqueue_task
= enqueue_task_rt
,
1740 .dequeue_task
= dequeue_task_rt
,
1741 .yield_task
= yield_task_rt
,
1743 .check_preempt_curr
= check_preempt_curr_rt
,
1745 .pick_next_task
= pick_next_task_rt
,
1746 .put_prev_task
= put_prev_task_rt
,
1749 .select_task_rq
= select_task_rq_rt
,
1751 .set_cpus_allowed
= set_cpus_allowed_rt
,
1752 .rq_online
= rq_online_rt
,
1753 .rq_offline
= rq_offline_rt
,
1754 .pre_schedule
= pre_schedule_rt
,
1755 .post_schedule
= post_schedule_rt
,
1756 .task_woken
= task_woken_rt
,
1757 .switched_from
= switched_from_rt
,
1760 .set_curr_task
= set_curr_task_rt
,
1761 .task_tick
= task_tick_rt
,
1763 .get_rr_interval
= get_rr_interval_rt
,
1765 .prio_changed
= prio_changed_rt
,
1766 .switched_to
= switched_to_rt
,
1769 #ifdef CONFIG_SCHED_DEBUG
1770 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
1772 static void print_rt_stats(struct seq_file
*m
, int cpu
)
1774 struct rt_rq
*rt_rq
;
1777 for_each_leaf_rt_rq(rt_rq
, cpu_rq(cpu
))
1778 print_rt_rq(m
, cpu
, rt_rq
);
1781 #endif /* CONFIG_SCHED_DEBUG */