2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
8 #include <linux/slab.h>
10 int sched_rr_timeslice
= RR_TIMESLICE
;
12 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
);
14 struct rt_bandwidth def_rt_bandwidth
;
16 static enum hrtimer_restart
sched_rt_period_timer(struct hrtimer
*timer
)
18 struct rt_bandwidth
*rt_b
=
19 container_of(timer
, struct rt_bandwidth
, rt_period_timer
);
25 now
= hrtimer_cb_get_time(timer
);
26 overrun
= hrtimer_forward(timer
, now
, rt_b
->rt_period
);
31 idle
= do_sched_rt_period_timer(rt_b
, overrun
);
34 return idle
? HRTIMER_NORESTART
: HRTIMER_RESTART
;
37 void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
)
39 rt_b
->rt_period
= ns_to_ktime(period
);
40 rt_b
->rt_runtime
= runtime
;
42 raw_spin_lock_init(&rt_b
->rt_runtime_lock
);
44 hrtimer_init(&rt_b
->rt_period_timer
,
45 CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
46 rt_b
->rt_period_timer
.function
= sched_rt_period_timer
;
49 static void start_rt_bandwidth(struct rt_bandwidth
*rt_b
)
51 if (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
)
54 if (hrtimer_active(&rt_b
->rt_period_timer
))
57 raw_spin_lock(&rt_b
->rt_runtime_lock
);
58 start_bandwidth_timer(&rt_b
->rt_period_timer
, rt_b
->rt_period
);
59 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
62 void init_rt_rq(struct rt_rq
*rt_rq
, struct rq
*rq
)
64 struct rt_prio_array
*array
;
67 array
= &rt_rq
->active
;
68 for (i
= 0; i
< MAX_RT_PRIO
; i
++) {
69 INIT_LIST_HEAD(array
->queue
+ i
);
70 __clear_bit(i
, array
->bitmap
);
72 /* delimiter for bitsearch: */
73 __set_bit(MAX_RT_PRIO
, array
->bitmap
);
75 #if defined CONFIG_SMP
76 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
77 rt_rq
->highest_prio
.next
= MAX_RT_PRIO
;
78 rt_rq
->rt_nr_migratory
= 0;
79 rt_rq
->overloaded
= 0;
80 plist_head_init(&rt_rq
->pushable_tasks
);
84 rt_rq
->rt_throttled
= 0;
85 rt_rq
->rt_runtime
= 0;
86 raw_spin_lock_init(&rt_rq
->rt_runtime_lock
);
89 #ifdef CONFIG_RT_GROUP_SCHED
90 static void destroy_rt_bandwidth(struct rt_bandwidth
*rt_b
)
92 hrtimer_cancel(&rt_b
->rt_period_timer
);
95 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
97 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
99 #ifdef CONFIG_SCHED_DEBUG
100 WARN_ON_ONCE(!rt_entity_is_task(rt_se
));
102 return container_of(rt_se
, struct task_struct
, rt
);
105 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
110 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
115 void free_rt_sched_group(struct task_group
*tg
)
120 destroy_rt_bandwidth(&tg
->rt_bandwidth
);
122 for_each_possible_cpu(i
) {
133 void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
134 struct sched_rt_entity
*rt_se
, int cpu
,
135 struct sched_rt_entity
*parent
)
137 struct rq
*rq
= cpu_rq(cpu
);
139 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
140 rt_rq
->rt_nr_boosted
= 0;
144 tg
->rt_rq
[cpu
] = rt_rq
;
145 tg
->rt_se
[cpu
] = rt_se
;
151 rt_se
->rt_rq
= &rq
->rt
;
153 rt_se
->rt_rq
= parent
->my_q
;
156 rt_se
->parent
= parent
;
157 INIT_LIST_HEAD(&rt_se
->run_list
);
160 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
163 struct sched_rt_entity
*rt_se
;
166 tg
->rt_rq
= kzalloc(sizeof(rt_rq
) * nr_cpu_ids
, GFP_KERNEL
);
169 tg
->rt_se
= kzalloc(sizeof(rt_se
) * nr_cpu_ids
, GFP_KERNEL
);
173 init_rt_bandwidth(&tg
->rt_bandwidth
,
174 ktime_to_ns(def_rt_bandwidth
.rt_period
), 0);
176 for_each_possible_cpu(i
) {
177 rt_rq
= kzalloc_node(sizeof(struct rt_rq
),
178 GFP_KERNEL
, cpu_to_node(i
));
182 rt_se
= kzalloc_node(sizeof(struct sched_rt_entity
),
183 GFP_KERNEL
, cpu_to_node(i
));
187 init_rt_rq(rt_rq
, cpu_rq(i
));
188 rt_rq
->rt_runtime
= tg
->rt_bandwidth
.rt_runtime
;
189 init_tg_rt_entry(tg
, rt_rq
, rt_se
, i
, parent
->rt_se
[i
]);
200 #else /* CONFIG_RT_GROUP_SCHED */
202 #define rt_entity_is_task(rt_se) (1)
204 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
206 return container_of(rt_se
, struct task_struct
, rt
);
209 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
211 return container_of(rt_rq
, struct rq
, rt
);
214 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
216 struct task_struct
*p
= rt_task_of(rt_se
);
217 struct rq
*rq
= task_rq(p
);
222 void free_rt_sched_group(struct task_group
*tg
) { }
224 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
228 #endif /* CONFIG_RT_GROUP_SCHED */
232 static inline int rt_overloaded(struct rq
*rq
)
234 return atomic_read(&rq
->rd
->rto_count
);
237 static inline void rt_set_overload(struct rq
*rq
)
242 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
244 * Make sure the mask is visible before we set
245 * the overload count. That is checked to determine
246 * if we should look at the mask. It would be a shame
247 * if we looked at the mask, but the mask was not
251 atomic_inc(&rq
->rd
->rto_count
);
254 static inline void rt_clear_overload(struct rq
*rq
)
259 /* the order here really doesn't matter */
260 atomic_dec(&rq
->rd
->rto_count
);
261 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
264 static void update_rt_migration(struct rt_rq
*rt_rq
)
266 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
267 if (!rt_rq
->overloaded
) {
268 rt_set_overload(rq_of_rt_rq(rt_rq
));
269 rt_rq
->overloaded
= 1;
271 } else if (rt_rq
->overloaded
) {
272 rt_clear_overload(rq_of_rt_rq(rt_rq
));
273 rt_rq
->overloaded
= 0;
277 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
279 struct task_struct
*p
;
281 if (!rt_entity_is_task(rt_se
))
284 p
= rt_task_of(rt_se
);
285 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
287 rt_rq
->rt_nr_total
++;
288 if (p
->nr_cpus_allowed
> 1)
289 rt_rq
->rt_nr_migratory
++;
291 update_rt_migration(rt_rq
);
294 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
296 struct task_struct
*p
;
298 if (!rt_entity_is_task(rt_se
))
301 p
= rt_task_of(rt_se
);
302 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
304 rt_rq
->rt_nr_total
--;
305 if (p
->nr_cpus_allowed
> 1)
306 rt_rq
->rt_nr_migratory
--;
308 update_rt_migration(rt_rq
);
311 static inline int has_pushable_tasks(struct rq
*rq
)
313 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
316 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
318 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
319 plist_node_init(&p
->pushable_tasks
, p
->prio
);
320 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
322 /* Update the highest prio pushable task */
323 if (p
->prio
< rq
->rt
.highest_prio
.next
)
324 rq
->rt
.highest_prio
.next
= p
->prio
;
327 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
329 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
331 /* Update the new highest prio pushable task */
332 if (has_pushable_tasks(rq
)) {
333 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
334 struct task_struct
, pushable_tasks
);
335 rq
->rt
.highest_prio
.next
= p
->prio
;
337 rq
->rt
.highest_prio
.next
= MAX_RT_PRIO
;
342 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
346 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
351 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
356 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
360 #endif /* CONFIG_SMP */
362 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
364 return !list_empty(&rt_se
->run_list
);
367 #ifdef CONFIG_RT_GROUP_SCHED
369 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
374 return rt_rq
->rt_runtime
;
377 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
379 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
382 typedef struct task_group
*rt_rq_iter_t
;
384 static inline struct task_group
*next_task_group(struct task_group
*tg
)
387 tg
= list_entry_rcu(tg
->list
.next
,
388 typeof(struct task_group
), list
);
389 } while (&tg
->list
!= &task_groups
&& task_group_is_autogroup(tg
));
391 if (&tg
->list
== &task_groups
)
397 #define for_each_rt_rq(rt_rq, iter, rq) \
398 for (iter = container_of(&task_groups, typeof(*iter), list); \
399 (iter = next_task_group(iter)) && \
400 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
402 #define for_each_sched_rt_entity(rt_se) \
403 for (; rt_se; rt_se = rt_se->parent)
405 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
410 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
);
411 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
);
413 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
415 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
416 struct sched_rt_entity
*rt_se
;
418 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
420 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
422 if (rt_rq
->rt_nr_running
) {
423 if (rt_se
&& !on_rt_rq(rt_se
))
424 enqueue_rt_entity(rt_se
, false);
425 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
430 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
432 struct sched_rt_entity
*rt_se
;
433 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
435 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
437 if (rt_se
&& on_rt_rq(rt_se
))
438 dequeue_rt_entity(rt_se
);
441 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
443 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
446 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
448 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
449 struct task_struct
*p
;
452 return !!rt_rq
->rt_nr_boosted
;
454 p
= rt_task_of(rt_se
);
455 return p
->prio
!= p
->normal_prio
;
459 static inline const struct cpumask
*sched_rt_period_mask(void)
461 return this_rq()->rd
->span
;
464 static inline const struct cpumask
*sched_rt_period_mask(void)
466 return cpu_online_mask
;
471 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
473 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
476 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
478 return &rt_rq
->tg
->rt_bandwidth
;
481 #else /* !CONFIG_RT_GROUP_SCHED */
483 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
485 return rt_rq
->rt_runtime
;
488 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
490 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
493 typedef struct rt_rq
*rt_rq_iter_t
;
495 #define for_each_rt_rq(rt_rq, iter, rq) \
496 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
498 #define for_each_sched_rt_entity(rt_se) \
499 for (; rt_se; rt_se = NULL)
501 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
506 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
508 if (rt_rq
->rt_nr_running
)
509 resched_task(rq_of_rt_rq(rt_rq
)->curr
);
512 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
516 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
518 return rt_rq
->rt_throttled
;
521 static inline const struct cpumask
*sched_rt_period_mask(void)
523 return cpu_online_mask
;
527 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
529 return &cpu_rq(cpu
)->rt
;
532 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
534 return &def_rt_bandwidth
;
537 #endif /* CONFIG_RT_GROUP_SCHED */
541 * We ran out of runtime, see if we can borrow some from our neighbours.
543 static int do_balance_runtime(struct rt_rq
*rt_rq
)
545 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
546 struct root_domain
*rd
= rq_of_rt_rq(rt_rq
)->rd
;
547 int i
, weight
, more
= 0;
550 weight
= cpumask_weight(rd
->span
);
552 raw_spin_lock(&rt_b
->rt_runtime_lock
);
553 rt_period
= ktime_to_ns(rt_b
->rt_period
);
554 for_each_cpu(i
, rd
->span
) {
555 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
561 raw_spin_lock(&iter
->rt_runtime_lock
);
563 * Either all rqs have inf runtime and there's nothing to steal
564 * or __disable_runtime() below sets a specific rq to inf to
565 * indicate its been disabled and disalow stealing.
567 if (iter
->rt_runtime
== RUNTIME_INF
)
571 * From runqueues with spare time, take 1/n part of their
572 * spare time, but no more than our period.
574 diff
= iter
->rt_runtime
- iter
->rt_time
;
576 diff
= div_u64((u64
)diff
, weight
);
577 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
578 diff
= rt_period
- rt_rq
->rt_runtime
;
579 iter
->rt_runtime
-= diff
;
580 rt_rq
->rt_runtime
+= diff
;
582 if (rt_rq
->rt_runtime
== rt_period
) {
583 raw_spin_unlock(&iter
->rt_runtime_lock
);
588 raw_spin_unlock(&iter
->rt_runtime_lock
);
590 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
596 * Ensure this RQ takes back all the runtime it lend to its neighbours.
598 static void __disable_runtime(struct rq
*rq
)
600 struct root_domain
*rd
= rq
->rd
;
604 if (unlikely(!scheduler_running
))
607 for_each_rt_rq(rt_rq
, iter
, rq
) {
608 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
612 raw_spin_lock(&rt_b
->rt_runtime_lock
);
613 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
615 * Either we're all inf and nobody needs to borrow, or we're
616 * already disabled and thus have nothing to do, or we have
617 * exactly the right amount of runtime to take out.
619 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
620 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
622 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
625 * Calculate the difference between what we started out with
626 * and what we current have, that's the amount of runtime
627 * we lend and now have to reclaim.
629 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
632 * Greedy reclaim, take back as much as we can.
634 for_each_cpu(i
, rd
->span
) {
635 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
639 * Can't reclaim from ourselves or disabled runqueues.
641 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
644 raw_spin_lock(&iter
->rt_runtime_lock
);
646 diff
= min_t(s64
, iter
->rt_runtime
, want
);
647 iter
->rt_runtime
-= diff
;
650 iter
->rt_runtime
-= want
;
653 raw_spin_unlock(&iter
->rt_runtime_lock
);
659 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
661 * We cannot be left wanting - that would mean some runtime
662 * leaked out of the system.
667 * Disable all the borrow logic by pretending we have inf
668 * runtime - in which case borrowing doesn't make sense.
670 rt_rq
->rt_runtime
= RUNTIME_INF
;
671 rt_rq
->rt_throttled
= 0;
672 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
673 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
677 static void __enable_runtime(struct rq
*rq
)
682 if (unlikely(!scheduler_running
))
686 * Reset each runqueue's bandwidth settings
688 for_each_rt_rq(rt_rq
, iter
, rq
) {
689 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
691 raw_spin_lock(&rt_b
->rt_runtime_lock
);
692 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
693 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
695 rt_rq
->rt_throttled
= 0;
696 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
697 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
701 static int balance_runtime(struct rt_rq
*rt_rq
)
705 if (!sched_feat(RT_RUNTIME_SHARE
))
708 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
709 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
710 more
= do_balance_runtime(rt_rq
);
711 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
716 #else /* !CONFIG_SMP */
717 static inline int balance_runtime(struct rt_rq
*rt_rq
)
721 #endif /* CONFIG_SMP */
723 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
725 int i
, idle
= 1, throttled
= 0;
726 const struct cpumask
*span
;
728 span
= sched_rt_period_mask();
729 #ifdef CONFIG_RT_GROUP_SCHED
731 * FIXME: isolated CPUs should really leave the root task group,
732 * whether they are isolcpus or were isolated via cpusets, lest
733 * the timer run on a CPU which does not service all runqueues,
734 * potentially leaving other CPUs indefinitely throttled. If
735 * isolation is really required, the user will turn the throttle
736 * off to kill the perturbations it causes anyway. Meanwhile,
737 * this maintains functionality for boot and/or troubleshooting.
739 if (rt_b
== &root_task_group
.rt_bandwidth
)
740 span
= cpu_online_mask
;
742 for_each_cpu(i
, span
) {
744 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
745 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
747 raw_spin_lock(&rq
->lock
);
748 if (rt_rq
->rt_time
) {
751 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
752 if (rt_rq
->rt_throttled
)
753 balance_runtime(rt_rq
);
754 runtime
= rt_rq
->rt_runtime
;
755 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
756 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
757 rt_rq
->rt_throttled
= 0;
761 * Force a clock update if the CPU was idle,
762 * lest wakeup -> unthrottle time accumulate.
764 if (rt_rq
->rt_nr_running
&& rq
->curr
== rq
->idle
)
765 rq
->skip_clock_update
= -1;
767 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
769 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
770 } else if (rt_rq
->rt_nr_running
) {
772 if (!rt_rq_throttled(rt_rq
))
775 if (rt_rq
->rt_throttled
)
779 sched_rt_rq_enqueue(rt_rq
);
780 raw_spin_unlock(&rq
->lock
);
783 if (!throttled
&& (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
))
789 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
791 #ifdef CONFIG_RT_GROUP_SCHED
792 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
795 return rt_rq
->highest_prio
.curr
;
798 return rt_task_of(rt_se
)->prio
;
801 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
803 u64 runtime
= sched_rt_runtime(rt_rq
);
805 if (rt_rq
->rt_throttled
)
806 return rt_rq_throttled(rt_rq
);
808 if (runtime
>= sched_rt_period(rt_rq
))
811 balance_runtime(rt_rq
);
812 runtime
= sched_rt_runtime(rt_rq
);
813 if (runtime
== RUNTIME_INF
)
816 if (rt_rq
->rt_time
> runtime
) {
817 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
820 * Don't actually throttle groups that have no runtime assigned
821 * but accrue some time due to boosting.
823 if (likely(rt_b
->rt_runtime
)) {
824 static bool once
= false;
826 rt_rq
->rt_throttled
= 1;
830 printk_sched("sched: RT throttling activated\n");
834 * In case we did anyway, make it go away,
835 * replenishment is a joke, since it will replenish us
841 if (rt_rq_throttled(rt_rq
)) {
842 sched_rt_rq_dequeue(rt_rq
);
851 * Update the current task's runtime statistics. Skip current tasks that
852 * are not in our scheduling class.
854 static void update_curr_rt(struct rq
*rq
)
856 struct task_struct
*curr
= rq
->curr
;
857 struct sched_rt_entity
*rt_se
= &curr
->rt
;
858 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
861 if (curr
->sched_class
!= &rt_sched_class
)
864 delta_exec
= rq_clock_task(rq
) - curr
->se
.exec_start
;
865 if (unlikely((s64
)delta_exec
<= 0))
868 schedstat_set(curr
->se
.statistics
.exec_max
,
869 max(curr
->se
.statistics
.exec_max
, delta_exec
));
871 curr
->se
.sum_exec_runtime
+= delta_exec
;
872 account_group_exec_runtime(curr
, delta_exec
);
874 curr
->se
.exec_start
= rq_clock_task(rq
);
875 cpuacct_charge(curr
, delta_exec
);
877 sched_rt_avg_update(rq
, delta_exec
);
879 if (!rt_bandwidth_enabled())
882 for_each_sched_rt_entity(rt_se
) {
883 rt_rq
= rt_rq_of_se(rt_se
);
885 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
886 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
887 rt_rq
->rt_time
+= delta_exec
;
888 if (sched_rt_runtime_exceeded(rt_rq
))
890 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
895 #if defined CONFIG_SMP
898 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
900 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
902 if (rq
->online
&& prio
< prev_prio
)
903 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
907 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
909 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
911 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
912 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
915 #else /* CONFIG_SMP */
918 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
920 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
922 #endif /* CONFIG_SMP */
924 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
926 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
928 int prev_prio
= rt_rq
->highest_prio
.curr
;
930 if (prio
< prev_prio
)
931 rt_rq
->highest_prio
.curr
= prio
;
933 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
937 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
939 int prev_prio
= rt_rq
->highest_prio
.curr
;
941 if (rt_rq
->rt_nr_running
) {
943 WARN_ON(prio
< prev_prio
);
946 * This may have been our highest task, and therefore
947 * we may have some recomputation to do
949 if (prio
== prev_prio
) {
950 struct rt_prio_array
*array
= &rt_rq
->active
;
952 rt_rq
->highest_prio
.curr
=
953 sched_find_first_bit(array
->bitmap
);
957 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
959 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
964 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
965 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
967 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
969 #ifdef CONFIG_RT_GROUP_SCHED
972 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
974 if (rt_se_boosted(rt_se
))
975 rt_rq
->rt_nr_boosted
++;
978 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
982 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
984 if (rt_se_boosted(rt_se
))
985 rt_rq
->rt_nr_boosted
--;
987 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
990 #else /* CONFIG_RT_GROUP_SCHED */
993 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
995 start_rt_bandwidth(&def_rt_bandwidth
);
999 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
1001 #endif /* CONFIG_RT_GROUP_SCHED */
1004 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1006 int prio
= rt_se_prio(rt_se
);
1008 WARN_ON(!rt_prio(prio
));
1009 rt_rq
->rt_nr_running
++;
1011 inc_rt_prio(rt_rq
, prio
);
1012 inc_rt_migration(rt_se
, rt_rq
);
1013 inc_rt_group(rt_se
, rt_rq
);
1017 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1019 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
1020 WARN_ON(!rt_rq
->rt_nr_running
);
1021 rt_rq
->rt_nr_running
--;
1023 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
1024 dec_rt_migration(rt_se
, rt_rq
);
1025 dec_rt_group(rt_se
, rt_rq
);
1028 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1030 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1031 struct rt_prio_array
*array
= &rt_rq
->active
;
1032 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1033 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1036 * Don't enqueue the group if its throttled, or when empty.
1037 * The latter is a consequence of the former when a child group
1038 * get throttled and the current group doesn't have any other
1041 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
))
1045 list_add(&rt_se
->run_list
, queue
);
1047 list_add_tail(&rt_se
->run_list
, queue
);
1048 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
1050 inc_rt_tasks(rt_se
, rt_rq
);
1053 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1055 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1056 struct rt_prio_array
*array
= &rt_rq
->active
;
1058 list_del_init(&rt_se
->run_list
);
1059 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
1060 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
1062 dec_rt_tasks(rt_se
, rt_rq
);
1066 * Because the prio of an upper entry depends on the lower
1067 * entries, we must remove entries top - down.
1069 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
)
1071 struct sched_rt_entity
*back
= NULL
;
1073 for_each_sched_rt_entity(rt_se
) {
1078 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
1079 if (on_rt_rq(rt_se
))
1080 __dequeue_rt_entity(rt_se
);
1084 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1086 dequeue_rt_stack(rt_se
);
1087 for_each_sched_rt_entity(rt_se
)
1088 __enqueue_rt_entity(rt_se
, head
);
1091 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1093 dequeue_rt_stack(rt_se
);
1095 for_each_sched_rt_entity(rt_se
) {
1096 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
1098 if (rt_rq
&& rt_rq
->rt_nr_running
)
1099 __enqueue_rt_entity(rt_se
, false);
1104 * Adding/removing a task to/from a priority array:
1107 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1109 struct sched_rt_entity
*rt_se
= &p
->rt
;
1111 if (flags
& ENQUEUE_WAKEUP
)
1114 enqueue_rt_entity(rt_se
, flags
& ENQUEUE_HEAD
);
1116 if (!task_current(rq
, p
) && p
->nr_cpus_allowed
> 1)
1117 enqueue_pushable_task(rq
, p
);
1122 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1124 struct sched_rt_entity
*rt_se
= &p
->rt
;
1127 dequeue_rt_entity(rt_se
);
1129 dequeue_pushable_task(rq
, p
);
1135 * Put task to the head or the end of the run list without the overhead of
1136 * dequeue followed by enqueue.
1139 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
1141 if (on_rt_rq(rt_se
)) {
1142 struct rt_prio_array
*array
= &rt_rq
->active
;
1143 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1146 list_move(&rt_se
->run_list
, queue
);
1148 list_move_tail(&rt_se
->run_list
, queue
);
1152 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
1154 struct sched_rt_entity
*rt_se
= &p
->rt
;
1155 struct rt_rq
*rt_rq
;
1157 for_each_sched_rt_entity(rt_se
) {
1158 rt_rq
= rt_rq_of_se(rt_se
);
1159 requeue_rt_entity(rt_rq
, rt_se
, head
);
1163 static void yield_task_rt(struct rq
*rq
)
1165 requeue_task_rt(rq
, rq
->curr
, 0);
1169 static int find_lowest_rq(struct task_struct
*task
);
1172 select_task_rq_rt(struct task_struct
*p
, int sd_flag
, int flags
)
1174 struct task_struct
*curr
;
1180 if (p
->nr_cpus_allowed
== 1)
1183 /* For anything but wake ups, just return the task_cpu */
1184 if (sd_flag
!= SD_BALANCE_WAKE
&& sd_flag
!= SD_BALANCE_FORK
)
1190 curr
= ACCESS_ONCE(rq
->curr
); /* unlocked access */
1193 * If the current task on @p's runqueue is an RT task, then
1194 * try to see if we can wake this RT task up on another
1195 * runqueue. Otherwise simply start this RT task
1196 * on its current runqueue.
1198 * We want to avoid overloading runqueues. If the woken
1199 * task is a higher priority, then it will stay on this CPU
1200 * and the lower prio task should be moved to another CPU.
1201 * Even though this will probably make the lower prio task
1202 * lose its cache, we do not want to bounce a higher task
1203 * around just because it gave up its CPU, perhaps for a
1206 * For equal prio tasks, we just let the scheduler sort it out.
1208 * Otherwise, just let it ride on the affined RQ and the
1209 * post-schedule router will push the preempted task away
1211 * This test is optimistic, if we get it wrong the load-balancer
1212 * will have to sort it out.
1214 if (curr
&& unlikely(rt_task(curr
)) &&
1215 (curr
->nr_cpus_allowed
< 2 ||
1216 curr
->prio
<= p
->prio
)) {
1217 int target
= find_lowest_rq(p
);
1228 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
1230 if (rq
->curr
->nr_cpus_allowed
== 1)
1233 if (p
->nr_cpus_allowed
!= 1
1234 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
1237 if (!cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
1241 * There appears to be other cpus that can accept
1242 * current and none to run 'p', so lets reschedule
1243 * to try and push current away:
1245 requeue_task_rt(rq
, p
, 1);
1246 resched_task(rq
->curr
);
1249 #endif /* CONFIG_SMP */
1252 * Preempt the current task with a newly woken task if needed:
1254 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1256 if (p
->prio
< rq
->curr
->prio
) {
1257 resched_task(rq
->curr
);
1265 * - the newly woken task is of equal priority to the current task
1266 * - the newly woken task is non-migratable while current is migratable
1267 * - current will be preempted on the next reschedule
1269 * we should check to see if current can readily move to a different
1270 * cpu. If so, we will reschedule to allow the push logic to try
1271 * to move current somewhere else, making room for our non-migratable
1274 if (p
->prio
== rq
->curr
->prio
&& !test_tsk_need_resched(rq
->curr
))
1275 check_preempt_equal_prio(rq
, p
);
1279 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1280 struct rt_rq
*rt_rq
)
1282 struct rt_prio_array
*array
= &rt_rq
->active
;
1283 struct sched_rt_entity
*next
= NULL
;
1284 struct list_head
*queue
;
1287 idx
= sched_find_first_bit(array
->bitmap
);
1288 BUG_ON(idx
>= MAX_RT_PRIO
);
1290 queue
= array
->queue
+ idx
;
1291 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1296 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1298 struct sched_rt_entity
*rt_se
;
1299 struct task_struct
*p
;
1300 struct rt_rq
*rt_rq
;
1304 if (!rt_rq
->rt_nr_running
)
1307 if (rt_rq_throttled(rt_rq
))
1311 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1313 rt_rq
= group_rt_rq(rt_se
);
1316 p
= rt_task_of(rt_se
);
1317 p
->se
.exec_start
= rq_clock_task(rq
);
1322 static struct task_struct
*pick_next_task_rt(struct rq
*rq
)
1324 struct task_struct
*p
= _pick_next_task_rt(rq
);
1326 /* The running task is never eligible for pushing */
1328 dequeue_pushable_task(rq
, p
);
1332 * We detect this state here so that we can avoid taking the RQ
1333 * lock again later if there is no need to push
1335 rq
->post_schedule
= has_pushable_tasks(rq
);
1341 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1346 * The previous task needs to be made eligible for pushing
1347 * if it is still active
1349 if (on_rt_rq(&p
->rt
) && p
->nr_cpus_allowed
> 1)
1350 enqueue_pushable_task(rq
, p
);
1355 /* Only try algorithms three times */
1356 #define RT_MAX_TRIES 3
1358 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1360 if (!task_running(rq
, p
) &&
1361 cpumask_test_cpu(cpu
, tsk_cpus_allowed(p
)))
1367 * Return the highest pushable rq's task, which is suitable to be executed
1368 * on the cpu, NULL otherwise
1370 static struct task_struct
*pick_highest_pushable_task(struct rq
*rq
, int cpu
)
1372 struct plist_head
*head
= &rq
->rt
.pushable_tasks
;
1373 struct task_struct
*p
;
1375 if (!has_pushable_tasks(rq
))
1378 plist_for_each_entry(p
, head
, pushable_tasks
) {
1379 if (pick_rt_task(rq
, p
, cpu
))
1386 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1388 static int find_lowest_rq(struct task_struct
*task
)
1390 struct sched_domain
*sd
;
1391 struct cpumask
*lowest_mask
= __get_cpu_var(local_cpu_mask
);
1392 int this_cpu
= smp_processor_id();
1393 int cpu
= task_cpu(task
);
1395 /* Make sure the mask is initialized first */
1396 if (unlikely(!lowest_mask
))
1399 if (task
->nr_cpus_allowed
== 1)
1400 return -1; /* No other targets possible */
1402 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1403 return -1; /* No targets found */
1406 * At this point we have built a mask of cpus representing the
1407 * lowest priority tasks in the system. Now we want to elect
1408 * the best one based on our affinity and topology.
1410 * We prioritize the last cpu that the task executed on since
1411 * it is most likely cache-hot in that location.
1413 if (cpumask_test_cpu(cpu
, lowest_mask
))
1417 * Otherwise, we consult the sched_domains span maps to figure
1418 * out which cpu is logically closest to our hot cache data.
1420 if (!cpumask_test_cpu(this_cpu
, lowest_mask
))
1421 this_cpu
= -1; /* Skip this_cpu opt if not among lowest */
1424 for_each_domain(cpu
, sd
) {
1425 if (sd
->flags
& SD_WAKE_AFFINE
) {
1429 * "this_cpu" is cheaper to preempt than a
1432 if (this_cpu
!= -1 &&
1433 cpumask_test_cpu(this_cpu
, sched_domain_span(sd
))) {
1438 best_cpu
= cpumask_first_and(lowest_mask
,
1439 sched_domain_span(sd
));
1440 if (best_cpu
< nr_cpu_ids
) {
1449 * And finally, if there were no matches within the domains
1450 * just give the caller *something* to work with from the compatible
1456 cpu
= cpumask_any(lowest_mask
);
1457 if (cpu
< nr_cpu_ids
)
1462 /* Will lock the rq it finds */
1463 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1465 struct rq
*lowest_rq
= NULL
;
1469 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1470 cpu
= find_lowest_rq(task
);
1472 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1475 lowest_rq
= cpu_rq(cpu
);
1477 /* if the prio of this runqueue changed, try again */
1478 if (double_lock_balance(rq
, lowest_rq
)) {
1480 * We had to unlock the run queue. In
1481 * the mean time, task could have
1482 * migrated already or had its affinity changed.
1483 * Also make sure that it wasn't scheduled on its rq.
1485 if (unlikely(task_rq(task
) != rq
||
1486 !cpumask_test_cpu(lowest_rq
->cpu
,
1487 tsk_cpus_allowed(task
)) ||
1488 task_running(rq
, task
) ||
1491 double_unlock_balance(rq
, lowest_rq
);
1497 /* If this rq is still suitable use it. */
1498 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1502 double_unlock_balance(rq
, lowest_rq
);
1509 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1511 struct task_struct
*p
;
1513 if (!has_pushable_tasks(rq
))
1516 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1517 struct task_struct
, pushable_tasks
);
1519 BUG_ON(rq
->cpu
!= task_cpu(p
));
1520 BUG_ON(task_current(rq
, p
));
1521 BUG_ON(p
->nr_cpus_allowed
<= 1);
1524 BUG_ON(!rt_task(p
));
1530 * If the current CPU has more than one RT task, see if the non
1531 * running task can migrate over to a CPU that is running a task
1532 * of lesser priority.
1534 static int push_rt_task(struct rq
*rq
)
1536 struct task_struct
*next_task
;
1537 struct rq
*lowest_rq
;
1540 if (!rq
->rt
.overloaded
)
1543 next_task
= pick_next_pushable_task(rq
);
1548 if (unlikely(next_task
== rq
->curr
)) {
1554 * It's possible that the next_task slipped in of
1555 * higher priority than current. If that's the case
1556 * just reschedule current.
1558 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1559 resched_task(rq
->curr
);
1563 /* We might release rq lock */
1564 get_task_struct(next_task
);
1566 /* find_lock_lowest_rq locks the rq if found */
1567 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1569 struct task_struct
*task
;
1571 * find_lock_lowest_rq releases rq->lock
1572 * so it is possible that next_task has migrated.
1574 * We need to make sure that the task is still on the same
1575 * run-queue and is also still the next task eligible for
1578 task
= pick_next_pushable_task(rq
);
1579 if (task_cpu(next_task
) == rq
->cpu
&& task
== next_task
) {
1581 * The task hasn't migrated, and is still the next
1582 * eligible task, but we failed to find a run-queue
1583 * to push it to. Do not retry in this case, since
1584 * other cpus will pull from us when ready.
1590 /* No more tasks, just exit */
1594 * Something has shifted, try again.
1596 put_task_struct(next_task
);
1601 deactivate_task(rq
, next_task
, 0);
1602 set_task_cpu(next_task
, lowest_rq
->cpu
);
1603 activate_task(lowest_rq
, next_task
, 0);
1606 resched_task(lowest_rq
->curr
);
1608 double_unlock_balance(rq
, lowest_rq
);
1611 put_task_struct(next_task
);
1616 static void push_rt_tasks(struct rq
*rq
)
1618 /* push_rt_task will return true if it moved an RT */
1619 while (push_rt_task(rq
))
1623 static int pull_rt_task(struct rq
*this_rq
)
1625 int this_cpu
= this_rq
->cpu
, ret
= 0, cpu
;
1626 struct task_struct
*p
;
1629 if (likely(!rt_overloaded(this_rq
)))
1632 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
1633 if (this_cpu
== cpu
)
1636 src_rq
= cpu_rq(cpu
);
1639 * Don't bother taking the src_rq->lock if the next highest
1640 * task is known to be lower-priority than our current task.
1641 * This may look racy, but if this value is about to go
1642 * logically higher, the src_rq will push this task away.
1643 * And if its going logically lower, we do not care
1645 if (src_rq
->rt
.highest_prio
.next
>=
1646 this_rq
->rt
.highest_prio
.curr
)
1650 * We can potentially drop this_rq's lock in
1651 * double_lock_balance, and another CPU could
1654 double_lock_balance(this_rq
, src_rq
);
1657 * We can pull only a task, which is pushable
1658 * on its rq, and no others.
1660 p
= pick_highest_pushable_task(src_rq
, this_cpu
);
1663 * Do we have an RT task that preempts
1664 * the to-be-scheduled task?
1666 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
1667 WARN_ON(p
== src_rq
->curr
);
1671 * There's a chance that p is higher in priority
1672 * than what's currently running on its cpu.
1673 * This is just that p is wakeing up and hasn't
1674 * had a chance to schedule. We only pull
1675 * p if it is lower in priority than the
1676 * current task on the run queue
1678 if (p
->prio
< src_rq
->curr
->prio
)
1683 deactivate_task(src_rq
, p
, 0);
1684 set_task_cpu(p
, this_cpu
);
1685 activate_task(this_rq
, p
, 0);
1687 * We continue with the search, just in
1688 * case there's an even higher prio task
1689 * in another runqueue. (low likelihood
1694 double_unlock_balance(this_rq
, src_rq
);
1700 static void pre_schedule_rt(struct rq
*rq
, struct task_struct
*prev
)
1702 /* Try to pull RT tasks here if we lower this rq's prio */
1703 if (rq
->rt
.highest_prio
.curr
> prev
->prio
)
1707 static void post_schedule_rt(struct rq
*rq
)
1713 * If we are not running and we are not going to reschedule soon, we should
1714 * try to push tasks away now
1716 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
1718 if (!task_running(rq
, p
) &&
1719 !test_tsk_need_resched(rq
->curr
) &&
1720 has_pushable_tasks(rq
) &&
1721 p
->nr_cpus_allowed
> 1 &&
1722 rt_task(rq
->curr
) &&
1723 (rq
->curr
->nr_cpus_allowed
< 2 ||
1724 rq
->curr
->prio
<= p
->prio
))
1728 static void set_cpus_allowed_rt(struct task_struct
*p
,
1729 const struct cpumask
*new_mask
)
1734 BUG_ON(!rt_task(p
));
1739 weight
= cpumask_weight(new_mask
);
1742 * Only update if the process changes its state from whether it
1743 * can migrate or not.
1745 if ((p
->nr_cpus_allowed
> 1) == (weight
> 1))
1751 * The process used to be able to migrate OR it can now migrate
1754 if (!task_current(rq
, p
))
1755 dequeue_pushable_task(rq
, p
);
1756 BUG_ON(!rq
->rt
.rt_nr_migratory
);
1757 rq
->rt
.rt_nr_migratory
--;
1759 if (!task_current(rq
, p
))
1760 enqueue_pushable_task(rq
, p
);
1761 rq
->rt
.rt_nr_migratory
++;
1764 update_rt_migration(&rq
->rt
);
1767 /* Assumes rq->lock is held */
1768 static void rq_online_rt(struct rq
*rq
)
1770 if (rq
->rt
.overloaded
)
1771 rt_set_overload(rq
);
1773 __enable_runtime(rq
);
1775 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
1778 /* Assumes rq->lock is held */
1779 static void rq_offline_rt(struct rq
*rq
)
1781 if (rq
->rt
.overloaded
)
1782 rt_clear_overload(rq
);
1784 __disable_runtime(rq
);
1786 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
1790 * When switch from the rt queue, we bring ourselves to a position
1791 * that we might want to pull RT tasks from other runqueues.
1793 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
)
1796 * If there are other RT tasks then we will reschedule
1797 * and the scheduling of the other RT tasks will handle
1798 * the balancing. But if we are the last RT task
1799 * we may need to handle the pulling of RT tasks
1802 if (!p
->on_rq
|| rq
->rt
.rt_nr_running
)
1805 if (pull_rt_task(rq
))
1806 resched_task(rq
->curr
);
1809 void init_sched_rt_class(void)
1813 for_each_possible_cpu(i
) {
1814 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
1815 GFP_KERNEL
, cpu_to_node(i
));
1818 #endif /* CONFIG_SMP */
1821 * When switching a task to RT, we may overload the runqueue
1822 * with RT tasks. In this case we try to push them off to
1825 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
)
1827 int check_resched
= 1;
1830 * If we are already running, then there's nothing
1831 * that needs to be done. But if we are not running
1832 * we may need to preempt the current running task.
1833 * If that current running task is also an RT task
1834 * then see if we can move to another run queue.
1836 if (p
->on_rq
&& rq
->curr
!= p
) {
1838 if (rq
->rt
.overloaded
&& push_rt_task(rq
) &&
1839 /* Don't resched if we changed runqueues */
1842 #endif /* CONFIG_SMP */
1843 if (check_resched
&& p
->prio
< rq
->curr
->prio
)
1844 resched_task(rq
->curr
);
1849 * Priority of the task has changed. This may cause
1850 * us to initiate a push or pull.
1853 prio_changed_rt(struct rq
*rq
, struct task_struct
*p
, int oldprio
)
1858 if (rq
->curr
== p
) {
1861 * If our priority decreases while running, we
1862 * may need to pull tasks to this runqueue.
1864 if (oldprio
< p
->prio
)
1867 * If there's a higher priority task waiting to run
1868 * then reschedule. Note, the above pull_rt_task
1869 * can release the rq lock and p could migrate.
1870 * Only reschedule if p is still on the same runqueue.
1872 if (p
->prio
> rq
->rt
.highest_prio
.curr
&& rq
->curr
== p
)
1875 /* For UP simply resched on drop of prio */
1876 if (oldprio
< p
->prio
)
1878 #endif /* CONFIG_SMP */
1881 * This task is not running, but if it is
1882 * greater than the current running task
1885 if (p
->prio
< rq
->curr
->prio
)
1886 resched_task(rq
->curr
);
1890 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
1892 unsigned long soft
, hard
;
1894 /* max may change after cur was read, this will be fixed next tick */
1895 soft
= task_rlimit(p
, RLIMIT_RTTIME
);
1896 hard
= task_rlimit_max(p
, RLIMIT_RTTIME
);
1898 if (soft
!= RLIM_INFINITY
) {
1901 if (p
->rt
.watchdog_stamp
!= jiffies
) {
1903 p
->rt
.watchdog_stamp
= jiffies
;
1906 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
1907 if (p
->rt
.timeout
> next
)
1908 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
1912 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
1914 struct sched_rt_entity
*rt_se
= &p
->rt
;
1921 * RR tasks need a special form of timeslice management.
1922 * FIFO tasks have no timeslices.
1924 if (p
->policy
!= SCHED_RR
)
1927 if (--p
->rt
.time_slice
)
1930 p
->rt
.time_slice
= sched_rr_timeslice
;
1933 * Requeue to the end of queue if we (and all of our ancestors) are the
1934 * only element on the queue
1936 for_each_sched_rt_entity(rt_se
) {
1937 if (rt_se
->run_list
.prev
!= rt_se
->run_list
.next
) {
1938 requeue_task_rt(rq
, p
, 0);
1939 set_tsk_need_resched(p
);
1945 static void set_curr_task_rt(struct rq
*rq
)
1947 struct task_struct
*p
= rq
->curr
;
1949 p
->se
.exec_start
= rq_clock_task(rq
);
1951 /* The running task is never eligible for pushing */
1952 dequeue_pushable_task(rq
, p
);
1955 static unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
1958 * Time slice is 0 for SCHED_FIFO tasks
1960 if (task
->policy
== SCHED_RR
)
1961 return sched_rr_timeslice
;
1966 const struct sched_class rt_sched_class
= {
1967 .next
= &fair_sched_class
,
1968 .enqueue_task
= enqueue_task_rt
,
1969 .dequeue_task
= dequeue_task_rt
,
1970 .yield_task
= yield_task_rt
,
1972 .check_preempt_curr
= check_preempt_curr_rt
,
1974 .pick_next_task
= pick_next_task_rt
,
1975 .put_prev_task
= put_prev_task_rt
,
1978 .select_task_rq
= select_task_rq_rt
,
1980 .set_cpus_allowed
= set_cpus_allowed_rt
,
1981 .rq_online
= rq_online_rt
,
1982 .rq_offline
= rq_offline_rt
,
1983 .pre_schedule
= pre_schedule_rt
,
1984 .post_schedule
= post_schedule_rt
,
1985 .task_woken
= task_woken_rt
,
1986 .switched_from
= switched_from_rt
,
1989 .set_curr_task
= set_curr_task_rt
,
1990 .task_tick
= task_tick_rt
,
1992 .get_rr_interval
= get_rr_interval_rt
,
1994 .prio_changed
= prio_changed_rt
,
1995 .switched_to
= switched_to_rt
,
1998 #ifdef CONFIG_SCHED_DEBUG
1999 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
2001 void print_rt_stats(struct seq_file
*m
, int cpu
)
2004 struct rt_rq
*rt_rq
;
2007 for_each_rt_rq(rt_rq
, iter
, cpu_rq(cpu
))
2008 print_rt_rq(m
, cpu
, rt_rq
);
2011 #endif /* CONFIG_SCHED_DEBUG */