sched: Implement on-demand (active) cfs_rq list
[linux-2.6.git] / kernel / sched_rt.c
1 /*
2  * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
3  * policies)
4  */
5
6 #ifdef CONFIG_RT_GROUP_SCHED
7
8 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
9
10 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
11 {
12 #ifdef CONFIG_SCHED_DEBUG
13         WARN_ON_ONCE(!rt_entity_is_task(rt_se));
14 #endif
15         return container_of(rt_se, struct task_struct, rt);
16 }
17
18 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
19 {
20         return rt_rq->rq;
21 }
22
23 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
24 {
25         return rt_se->rt_rq;
26 }
27
28 #else /* CONFIG_RT_GROUP_SCHED */
29
30 #define rt_entity_is_task(rt_se) (1)
31
32 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
33 {
34         return container_of(rt_se, struct task_struct, rt);
35 }
36
37 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
38 {
39         return container_of(rt_rq, struct rq, rt);
40 }
41
42 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
43 {
44         struct task_struct *p = rt_task_of(rt_se);
45         struct rq *rq = task_rq(p);
46
47         return &rq->rt;
48 }
49
50 #endif /* CONFIG_RT_GROUP_SCHED */
51
52 #ifdef CONFIG_SMP
53
54 static inline int rt_overloaded(struct rq *rq)
55 {
56         return atomic_read(&rq->rd->rto_count);
57 }
58
59 static inline void rt_set_overload(struct rq *rq)
60 {
61         if (!rq->online)
62                 return;
63
64         cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
65         /*
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
70          * updated yet.
71          */
72         wmb();
73         atomic_inc(&rq->rd->rto_count);
74 }
75
76 static inline void rt_clear_overload(struct rq *rq)
77 {
78         if (!rq->online)
79                 return;
80
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);
84 }
85
86 static void update_rt_migration(struct rt_rq *rt_rq)
87 {
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;
92                 }
93         } else if (rt_rq->overloaded) {
94                 rt_clear_overload(rq_of_rt_rq(rt_rq));
95                 rt_rq->overloaded = 0;
96         }
97 }
98
99 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
100 {
101         if (!rt_entity_is_task(rt_se))
102                 return;
103
104         rt_rq = &rq_of_rt_rq(rt_rq)->rt;
105
106         rt_rq->rt_nr_total++;
107         if (rt_se->nr_cpus_allowed > 1)
108                 rt_rq->rt_nr_migratory++;
109
110         update_rt_migration(rt_rq);
111 }
112
113 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
114 {
115         if (!rt_entity_is_task(rt_se))
116                 return;
117
118         rt_rq = &rq_of_rt_rq(rt_rq)->rt;
119
120         rt_rq->rt_nr_total--;
121         if (rt_se->nr_cpus_allowed > 1)
122                 rt_rq->rt_nr_migratory--;
123
124         update_rt_migration(rt_rq);
125 }
126
127 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
128 {
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);
132 }
133
134 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
135 {
136         plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
137 }
138
139 static inline int has_pushable_tasks(struct rq *rq)
140 {
141         return !plist_head_empty(&rq->rt.pushable_tasks);
142 }
143
144 #else
145
146 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
147 {
148 }
149
150 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
151 {
152 }
153
154 static inline
155 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
156 {
157 }
158
159 static inline
160 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
161 {
162 }
163
164 #endif /* CONFIG_SMP */
165
166 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
167 {
168         return !list_empty(&rt_se->run_list);
169 }
170
171 #ifdef CONFIG_RT_GROUP_SCHED
172
173 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
174 {
175         if (!rt_rq->tg)
176                 return RUNTIME_INF;
177
178         return rt_rq->rt_runtime;
179 }
180
181 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
182 {
183         return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
184 }
185
186 static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq)
187 {
188         list_add_rcu(&rt_rq->leaf_rt_rq_list,
189                         &rq_of_rt_rq(rt_rq)->leaf_rt_rq_list);
190 }
191
192 static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq)
193 {
194         list_del_rcu(&rt_rq->leaf_rt_rq_list);
195 }
196
197 #define for_each_leaf_rt_rq(rt_rq, rq) \
198         list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
199
200 #define for_each_sched_rt_entity(rt_se) \
201         for (; rt_se; rt_se = rt_se->parent)
202
203 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
204 {
205         return rt_se->my_q;
206 }
207
208 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head);
209 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
210
211 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
212 {
213         int this_cpu = smp_processor_id();
214         struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
215         struct sched_rt_entity *rt_se;
216
217         rt_se = rt_rq->tg->rt_se[this_cpu];
218
219         if (rt_rq->rt_nr_running) {
220                 if (rt_se && !on_rt_rq(rt_se))
221                         enqueue_rt_entity(rt_se, false);
222                 if (rt_rq->highest_prio.curr < curr->prio)
223                         resched_task(curr);
224         }
225 }
226
227 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
228 {
229         int this_cpu = smp_processor_id();
230         struct sched_rt_entity *rt_se;
231
232         rt_se = rt_rq->tg->rt_se[this_cpu];
233
234         if (rt_se && on_rt_rq(rt_se))
235                 dequeue_rt_entity(rt_se);
236 }
237
238 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
239 {
240         return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
241 }
242
243 static int rt_se_boosted(struct sched_rt_entity *rt_se)
244 {
245         struct rt_rq *rt_rq = group_rt_rq(rt_se);
246         struct task_struct *p;
247
248         if (rt_rq)
249                 return !!rt_rq->rt_nr_boosted;
250
251         p = rt_task_of(rt_se);
252         return p->prio != p->normal_prio;
253 }
254
255 #ifdef CONFIG_SMP
256 static inline const struct cpumask *sched_rt_period_mask(void)
257 {
258         return cpu_rq(smp_processor_id())->rd->span;
259 }
260 #else
261 static inline const struct cpumask *sched_rt_period_mask(void)
262 {
263         return cpu_online_mask;
264 }
265 #endif
266
267 static inline
268 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
269 {
270         return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
271 }
272
273 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
274 {
275         return &rt_rq->tg->rt_bandwidth;
276 }
277
278 #else /* !CONFIG_RT_GROUP_SCHED */
279
280 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
281 {
282         return rt_rq->rt_runtime;
283 }
284
285 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
286 {
287         return ktime_to_ns(def_rt_bandwidth.rt_period);
288 }
289
290 static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq)
291 {
292 }
293
294 static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq)
295 {
296 }
297
298 #define for_each_leaf_rt_rq(rt_rq, rq) \
299         for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
300
301 #define for_each_sched_rt_entity(rt_se) \
302         for (; rt_se; rt_se = NULL)
303
304 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
305 {
306         return NULL;
307 }
308
309 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
310 {
311         if (rt_rq->rt_nr_running)
312                 resched_task(rq_of_rt_rq(rt_rq)->curr);
313 }
314
315 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
316 {
317 }
318
319 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
320 {
321         return rt_rq->rt_throttled;
322 }
323
324 static inline const struct cpumask *sched_rt_period_mask(void)
325 {
326         return cpu_online_mask;
327 }
328
329 static inline
330 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
331 {
332         return &cpu_rq(cpu)->rt;
333 }
334
335 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
336 {
337         return &def_rt_bandwidth;
338 }
339
340 #endif /* CONFIG_RT_GROUP_SCHED */
341
342 #ifdef CONFIG_SMP
343 /*
344  * We ran out of runtime, see if we can borrow some from our neighbours.
345  */
346 static int do_balance_runtime(struct rt_rq *rt_rq)
347 {
348         struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
349         struct root_domain *rd = cpu_rq(smp_processor_id())->rd;
350         int i, weight, more = 0;
351         u64 rt_period;
352
353         weight = cpumask_weight(rd->span);
354
355         raw_spin_lock(&rt_b->rt_runtime_lock);
356         rt_period = ktime_to_ns(rt_b->rt_period);
357         for_each_cpu(i, rd->span) {
358                 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
359                 s64 diff;
360
361                 if (iter == rt_rq)
362                         continue;
363
364                 raw_spin_lock(&iter->rt_runtime_lock);
365                 /*
366                  * Either all rqs have inf runtime and there's nothing to steal
367                  * or __disable_runtime() below sets a specific rq to inf to
368                  * indicate its been disabled and disalow stealing.
369                  */
370                 if (iter->rt_runtime == RUNTIME_INF)
371                         goto next;
372
373                 /*
374                  * From runqueues with spare time, take 1/n part of their
375                  * spare time, but no more than our period.
376                  */
377                 diff = iter->rt_runtime - iter->rt_time;
378                 if (diff > 0) {
379                         diff = div_u64((u64)diff, weight);
380                         if (rt_rq->rt_runtime + diff > rt_period)
381                                 diff = rt_period - rt_rq->rt_runtime;
382                         iter->rt_runtime -= diff;
383                         rt_rq->rt_runtime += diff;
384                         more = 1;
385                         if (rt_rq->rt_runtime == rt_period) {
386                                 raw_spin_unlock(&iter->rt_runtime_lock);
387                                 break;
388                         }
389                 }
390 next:
391                 raw_spin_unlock(&iter->rt_runtime_lock);
392         }
393         raw_spin_unlock(&rt_b->rt_runtime_lock);
394
395         return more;
396 }
397
398 /*
399  * Ensure this RQ takes back all the runtime it lend to its neighbours.
400  */
401 static void __disable_runtime(struct rq *rq)
402 {
403         struct root_domain *rd = rq->rd;
404         struct rt_rq *rt_rq;
405
406         if (unlikely(!scheduler_running))
407                 return;
408
409         for_each_leaf_rt_rq(rt_rq, rq) {
410                 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
411                 s64 want;
412                 int i;
413
414                 raw_spin_lock(&rt_b->rt_runtime_lock);
415                 raw_spin_lock(&rt_rq->rt_runtime_lock);
416                 /*
417                  * Either we're all inf and nobody needs to borrow, or we're
418                  * already disabled and thus have nothing to do, or we have
419                  * exactly the right amount of runtime to take out.
420                  */
421                 if (rt_rq->rt_runtime == RUNTIME_INF ||
422                                 rt_rq->rt_runtime == rt_b->rt_runtime)
423                         goto balanced;
424                 raw_spin_unlock(&rt_rq->rt_runtime_lock);
425
426                 /*
427                  * Calculate the difference between what we started out with
428                  * and what we current have, that's the amount of runtime
429                  * we lend and now have to reclaim.
430                  */
431                 want = rt_b->rt_runtime - rt_rq->rt_runtime;
432
433                 /*
434                  * Greedy reclaim, take back as much as we can.
435                  */
436                 for_each_cpu(i, rd->span) {
437                         struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
438                         s64 diff;
439
440                         /*
441                          * Can't reclaim from ourselves or disabled runqueues.
442                          */
443                         if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
444                                 continue;
445
446                         raw_spin_lock(&iter->rt_runtime_lock);
447                         if (want > 0) {
448                                 diff = min_t(s64, iter->rt_runtime, want);
449                                 iter->rt_runtime -= diff;
450                                 want -= diff;
451                         } else {
452                                 iter->rt_runtime -= want;
453                                 want -= want;
454                         }
455                         raw_spin_unlock(&iter->rt_runtime_lock);
456
457                         if (!want)
458                                 break;
459                 }
460
461                 raw_spin_lock(&rt_rq->rt_runtime_lock);
462                 /*
463                  * We cannot be left wanting - that would mean some runtime
464                  * leaked out of the system.
465                  */
466                 BUG_ON(want);
467 balanced:
468                 /*
469                  * Disable all the borrow logic by pretending we have inf
470                  * runtime - in which case borrowing doesn't make sense.
471                  */
472                 rt_rq->rt_runtime = RUNTIME_INF;
473                 raw_spin_unlock(&rt_rq->rt_runtime_lock);
474                 raw_spin_unlock(&rt_b->rt_runtime_lock);
475         }
476 }
477
478 static void disable_runtime(struct rq *rq)
479 {
480         unsigned long flags;
481
482         raw_spin_lock_irqsave(&rq->lock, flags);
483         __disable_runtime(rq);
484         raw_spin_unlock_irqrestore(&rq->lock, flags);
485 }
486
487 static void __enable_runtime(struct rq *rq)
488 {
489         struct rt_rq *rt_rq;
490
491         if (unlikely(!scheduler_running))
492                 return;
493
494         /*
495          * Reset each runqueue's bandwidth settings
496          */
497         for_each_leaf_rt_rq(rt_rq, rq) {
498                 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
499
500                 raw_spin_lock(&rt_b->rt_runtime_lock);
501                 raw_spin_lock(&rt_rq->rt_runtime_lock);
502                 rt_rq->rt_runtime = rt_b->rt_runtime;
503                 rt_rq->rt_time = 0;
504                 rt_rq->rt_throttled = 0;
505                 raw_spin_unlock(&rt_rq->rt_runtime_lock);
506                 raw_spin_unlock(&rt_b->rt_runtime_lock);
507         }
508 }
509
510 static void enable_runtime(struct rq *rq)
511 {
512         unsigned long flags;
513
514         raw_spin_lock_irqsave(&rq->lock, flags);
515         __enable_runtime(rq);
516         raw_spin_unlock_irqrestore(&rq->lock, flags);
517 }
518
519 static int balance_runtime(struct rt_rq *rt_rq)
520 {
521         int more = 0;
522
523         if (rt_rq->rt_time > rt_rq->rt_runtime) {
524                 raw_spin_unlock(&rt_rq->rt_runtime_lock);
525                 more = do_balance_runtime(rt_rq);
526                 raw_spin_lock(&rt_rq->rt_runtime_lock);
527         }
528
529         return more;
530 }
531 #else /* !CONFIG_SMP */
532 static inline int balance_runtime(struct rt_rq *rt_rq)
533 {
534         return 0;
535 }
536 #endif /* CONFIG_SMP */
537
538 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
539 {
540         int i, idle = 1;
541         const struct cpumask *span;
542
543         if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
544                 return 1;
545
546         span = sched_rt_period_mask();
547         for_each_cpu(i, span) {
548                 int enqueue = 0;
549                 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
550                 struct rq *rq = rq_of_rt_rq(rt_rq);
551
552                 raw_spin_lock(&rq->lock);
553                 if (rt_rq->rt_time) {
554                         u64 runtime;
555
556                         raw_spin_lock(&rt_rq->rt_runtime_lock);
557                         if (rt_rq->rt_throttled)
558                                 balance_runtime(rt_rq);
559                         runtime = rt_rq->rt_runtime;
560                         rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
561                         if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
562                                 rt_rq->rt_throttled = 0;
563                                 enqueue = 1;
564                         }
565                         if (rt_rq->rt_time || rt_rq->rt_nr_running)
566                                 idle = 0;
567                         raw_spin_unlock(&rt_rq->rt_runtime_lock);
568                 } else if (rt_rq->rt_nr_running)
569                         idle = 0;
570
571                 if (enqueue)
572                         sched_rt_rq_enqueue(rt_rq);
573                 raw_spin_unlock(&rq->lock);
574         }
575
576         return idle;
577 }
578
579 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
580 {
581 #ifdef CONFIG_RT_GROUP_SCHED
582         struct rt_rq *rt_rq = group_rt_rq(rt_se);
583
584         if (rt_rq)
585                 return rt_rq->highest_prio.curr;
586 #endif
587
588         return rt_task_of(rt_se)->prio;
589 }
590
591 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
592 {
593         u64 runtime = sched_rt_runtime(rt_rq);
594
595         if (rt_rq->rt_throttled)
596                 return rt_rq_throttled(rt_rq);
597
598         if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq))
599                 return 0;
600
601         balance_runtime(rt_rq);
602         runtime = sched_rt_runtime(rt_rq);
603         if (runtime == RUNTIME_INF)
604                 return 0;
605
606         if (rt_rq->rt_time > runtime) {
607                 rt_rq->rt_throttled = 1;
608                 if (rt_rq_throttled(rt_rq)) {
609                         sched_rt_rq_dequeue(rt_rq);
610                         return 1;
611                 }
612         }
613
614         return 0;
615 }
616
617 /*
618  * Update the current task's runtime statistics. Skip current tasks that
619  * are not in our scheduling class.
620  */
621 static void update_curr_rt(struct rq *rq)
622 {
623         struct task_struct *curr = rq->curr;
624         struct sched_rt_entity *rt_se = &curr->rt;
625         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
626         u64 delta_exec;
627
628         if (!task_has_rt_policy(curr))
629                 return;
630
631         delta_exec = rq->clock_task - curr->se.exec_start;
632         if (unlikely((s64)delta_exec < 0))
633                 delta_exec = 0;
634
635         schedstat_set(curr->se.statistics.exec_max, max(curr->se.statistics.exec_max, delta_exec));
636
637         curr->se.sum_exec_runtime += delta_exec;
638         account_group_exec_runtime(curr, delta_exec);
639
640         curr->se.exec_start = rq->clock_task;
641         cpuacct_charge(curr, delta_exec);
642
643         sched_rt_avg_update(rq, delta_exec);
644
645         if (!rt_bandwidth_enabled())
646                 return;
647
648         for_each_sched_rt_entity(rt_se) {
649                 rt_rq = rt_rq_of_se(rt_se);
650
651                 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
652                         raw_spin_lock(&rt_rq->rt_runtime_lock);
653                         rt_rq->rt_time += delta_exec;
654                         if (sched_rt_runtime_exceeded(rt_rq))
655                                 resched_task(curr);
656                         raw_spin_unlock(&rt_rq->rt_runtime_lock);
657                 }
658         }
659 }
660
661 #if defined CONFIG_SMP
662
663 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu);
664
665 static inline int next_prio(struct rq *rq)
666 {
667         struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu);
668
669         if (next && rt_prio(next->prio))
670                 return next->prio;
671         else
672                 return MAX_RT_PRIO;
673 }
674
675 static void
676 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
677 {
678         struct rq *rq = rq_of_rt_rq(rt_rq);
679
680         if (prio < prev_prio) {
681
682                 /*
683                  * If the new task is higher in priority than anything on the
684                  * run-queue, we know that the previous high becomes our
685                  * next-highest.
686                  */
687                 rt_rq->highest_prio.next = prev_prio;
688
689                 if (rq->online)
690                         cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
691
692         } else if (prio == rt_rq->highest_prio.curr)
693                 /*
694                  * If the next task is equal in priority to the highest on
695                  * the run-queue, then we implicitly know that the next highest
696                  * task cannot be any lower than current
697                  */
698                 rt_rq->highest_prio.next = prio;
699         else if (prio < rt_rq->highest_prio.next)
700                 /*
701                  * Otherwise, we need to recompute next-highest
702                  */
703                 rt_rq->highest_prio.next = next_prio(rq);
704 }
705
706 static void
707 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
708 {
709         struct rq *rq = rq_of_rt_rq(rt_rq);
710
711         if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next))
712                 rt_rq->highest_prio.next = next_prio(rq);
713
714         if (rq->online && rt_rq->highest_prio.curr != prev_prio)
715                 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
716 }
717
718 #else /* CONFIG_SMP */
719
720 static inline
721 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
722 static inline
723 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
724
725 #endif /* CONFIG_SMP */
726
727 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
728 static void
729 inc_rt_prio(struct rt_rq *rt_rq, int prio)
730 {
731         int prev_prio = rt_rq->highest_prio.curr;
732
733         if (prio < prev_prio)
734                 rt_rq->highest_prio.curr = prio;
735
736         inc_rt_prio_smp(rt_rq, prio, prev_prio);
737 }
738
739 static void
740 dec_rt_prio(struct rt_rq *rt_rq, int prio)
741 {
742         int prev_prio = rt_rq->highest_prio.curr;
743
744         if (rt_rq->rt_nr_running) {
745
746                 WARN_ON(prio < prev_prio);
747
748                 /*
749                  * This may have been our highest task, and therefore
750                  * we may have some recomputation to do
751                  */
752                 if (prio == prev_prio) {
753                         struct rt_prio_array *array = &rt_rq->active;
754
755                         rt_rq->highest_prio.curr =
756                                 sched_find_first_bit(array->bitmap);
757                 }
758
759         } else
760                 rt_rq->highest_prio.curr = MAX_RT_PRIO;
761
762         dec_rt_prio_smp(rt_rq, prio, prev_prio);
763 }
764
765 #else
766
767 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
768 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
769
770 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
771
772 #ifdef CONFIG_RT_GROUP_SCHED
773
774 static void
775 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
776 {
777         if (rt_se_boosted(rt_se))
778                 rt_rq->rt_nr_boosted++;
779
780         if (rt_rq->tg)
781                 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
782 }
783
784 static void
785 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
786 {
787         if (rt_se_boosted(rt_se))
788                 rt_rq->rt_nr_boosted--;
789
790         WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
791 }
792
793 #else /* CONFIG_RT_GROUP_SCHED */
794
795 static void
796 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
797 {
798         start_rt_bandwidth(&def_rt_bandwidth);
799 }
800
801 static inline
802 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
803
804 #endif /* CONFIG_RT_GROUP_SCHED */
805
806 static inline
807 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
808 {
809         int prio = rt_se_prio(rt_se);
810
811         WARN_ON(!rt_prio(prio));
812         rt_rq->rt_nr_running++;
813
814         inc_rt_prio(rt_rq, prio);
815         inc_rt_migration(rt_se, rt_rq);
816         inc_rt_group(rt_se, rt_rq);
817 }
818
819 static inline
820 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
821 {
822         WARN_ON(!rt_prio(rt_se_prio(rt_se)));
823         WARN_ON(!rt_rq->rt_nr_running);
824         rt_rq->rt_nr_running--;
825
826         dec_rt_prio(rt_rq, rt_se_prio(rt_se));
827         dec_rt_migration(rt_se, rt_rq);
828         dec_rt_group(rt_se, rt_rq);
829 }
830
831 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
832 {
833         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
834         struct rt_prio_array *array = &rt_rq->active;
835         struct rt_rq *group_rq = group_rt_rq(rt_se);
836         struct list_head *queue = array->queue + rt_se_prio(rt_se);
837
838         /*
839          * Don't enqueue the group if its throttled, or when empty.
840          * The latter is a consequence of the former when a child group
841          * get throttled and the current group doesn't have any other
842          * active members.
843          */
844         if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
845                 return;
846
847         if (!rt_rq->rt_nr_running)
848                 list_add_leaf_rt_rq(rt_rq);
849
850         if (head)
851                 list_add(&rt_se->run_list, queue);
852         else
853                 list_add_tail(&rt_se->run_list, queue);
854         __set_bit(rt_se_prio(rt_se), array->bitmap);
855
856         inc_rt_tasks(rt_se, rt_rq);
857 }
858
859 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
860 {
861         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
862         struct rt_prio_array *array = &rt_rq->active;
863
864         list_del_init(&rt_se->run_list);
865         if (list_empty(array->queue + rt_se_prio(rt_se)))
866                 __clear_bit(rt_se_prio(rt_se), array->bitmap);
867
868         dec_rt_tasks(rt_se, rt_rq);
869         if (!rt_rq->rt_nr_running)
870                 list_del_leaf_rt_rq(rt_rq);
871 }
872
873 /*
874  * Because the prio of an upper entry depends on the lower
875  * entries, we must remove entries top - down.
876  */
877 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
878 {
879         struct sched_rt_entity *back = NULL;
880
881         for_each_sched_rt_entity(rt_se) {
882                 rt_se->back = back;
883                 back = rt_se;
884         }
885
886         for (rt_se = back; rt_se; rt_se = rt_se->back) {
887                 if (on_rt_rq(rt_se))
888                         __dequeue_rt_entity(rt_se);
889         }
890 }
891
892 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
893 {
894         dequeue_rt_stack(rt_se);
895         for_each_sched_rt_entity(rt_se)
896                 __enqueue_rt_entity(rt_se, head);
897 }
898
899 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
900 {
901         dequeue_rt_stack(rt_se);
902
903         for_each_sched_rt_entity(rt_se) {
904                 struct rt_rq *rt_rq = group_rt_rq(rt_se);
905
906                 if (rt_rq && rt_rq->rt_nr_running)
907                         __enqueue_rt_entity(rt_se, false);
908         }
909 }
910
911 /*
912  * Adding/removing a task to/from a priority array:
913  */
914 static void
915 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
916 {
917         struct sched_rt_entity *rt_se = &p->rt;
918
919         if (flags & ENQUEUE_WAKEUP)
920                 rt_se->timeout = 0;
921
922         enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
923
924         if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1)
925                 enqueue_pushable_task(rq, p);
926 }
927
928 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
929 {
930         struct sched_rt_entity *rt_se = &p->rt;
931
932         update_curr_rt(rq);
933         dequeue_rt_entity(rt_se);
934
935         dequeue_pushable_task(rq, p);
936 }
937
938 /*
939  * Put task to the end of the run list without the overhead of dequeue
940  * followed by enqueue.
941  */
942 static void
943 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
944 {
945         if (on_rt_rq(rt_se)) {
946                 struct rt_prio_array *array = &rt_rq->active;
947                 struct list_head *queue = array->queue + rt_se_prio(rt_se);
948
949                 if (head)
950                         list_move(&rt_se->run_list, queue);
951                 else
952                         list_move_tail(&rt_se->run_list, queue);
953         }
954 }
955
956 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
957 {
958         struct sched_rt_entity *rt_se = &p->rt;
959         struct rt_rq *rt_rq;
960
961         for_each_sched_rt_entity(rt_se) {
962                 rt_rq = rt_rq_of_se(rt_se);
963                 requeue_rt_entity(rt_rq, rt_se, head);
964         }
965 }
966
967 static void yield_task_rt(struct rq *rq)
968 {
969         requeue_task_rt(rq, rq->curr, 0);
970 }
971
972 #ifdef CONFIG_SMP
973 static int find_lowest_rq(struct task_struct *task);
974
975 static int
976 select_task_rq_rt(struct rq *rq, struct task_struct *p, int sd_flag, int flags)
977 {
978         if (sd_flag != SD_BALANCE_WAKE)
979                 return smp_processor_id();
980
981         /*
982          * If the current task is an RT task, then
983          * try to see if we can wake this RT task up on another
984          * runqueue. Otherwise simply start this RT task
985          * on its current runqueue.
986          *
987          * We want to avoid overloading runqueues. If the woken
988          * task is a higher priority, then it will stay on this CPU
989          * and the lower prio task should be moved to another CPU.
990          * Even though this will probably make the lower prio task
991          * lose its cache, we do not want to bounce a higher task
992          * around just because it gave up its CPU, perhaps for a
993          * lock?
994          *
995          * For equal prio tasks, we just let the scheduler sort it out.
996          */
997         if (unlikely(rt_task(rq->curr)) &&
998             (rq->curr->rt.nr_cpus_allowed < 2 ||
999              rq->curr->prio < p->prio) &&
1000             (p->rt.nr_cpus_allowed > 1)) {
1001                 int cpu = find_lowest_rq(p);
1002
1003                 return (cpu == -1) ? task_cpu(p) : cpu;
1004         }
1005
1006         /*
1007          * Otherwise, just let it ride on the affined RQ and the
1008          * post-schedule router will push the preempted task away
1009          */
1010         return task_cpu(p);
1011 }
1012
1013 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1014 {
1015         if (rq->curr->rt.nr_cpus_allowed == 1)
1016                 return;
1017
1018         if (p->rt.nr_cpus_allowed != 1
1019             && cpupri_find(&rq->rd->cpupri, p, NULL))
1020                 return;
1021
1022         if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1023                 return;
1024
1025         /*
1026          * There appears to be other cpus that can accept
1027          * current and none to run 'p', so lets reschedule
1028          * to try and push current away:
1029          */
1030         requeue_task_rt(rq, p, 1);
1031         resched_task(rq->curr);
1032 }
1033
1034 #endif /* CONFIG_SMP */
1035
1036 /*
1037  * Preempt the current task with a newly woken task if needed:
1038  */
1039 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1040 {
1041         if (p->prio < rq->curr->prio) {
1042                 resched_task(rq->curr);
1043                 return;
1044         }
1045
1046 #ifdef CONFIG_SMP
1047         /*
1048          * If:
1049          *
1050          * - the newly woken task is of equal priority to the current task
1051          * - the newly woken task is non-migratable while current is migratable
1052          * - current will be preempted on the next reschedule
1053          *
1054          * we should check to see if current can readily move to a different
1055          * cpu.  If so, we will reschedule to allow the push logic to try
1056          * to move current somewhere else, making room for our non-migratable
1057          * task.
1058          */
1059         if (p->prio == rq->curr->prio && !need_resched())
1060                 check_preempt_equal_prio(rq, p);
1061 #endif
1062 }
1063
1064 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1065                                                    struct rt_rq *rt_rq)
1066 {
1067         struct rt_prio_array *array = &rt_rq->active;
1068         struct sched_rt_entity *next = NULL;
1069         struct list_head *queue;
1070         int idx;
1071
1072         idx = sched_find_first_bit(array->bitmap);
1073         BUG_ON(idx >= MAX_RT_PRIO);
1074
1075         queue = array->queue + idx;
1076         next = list_entry(queue->next, struct sched_rt_entity, run_list);
1077
1078         return next;
1079 }
1080
1081 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1082 {
1083         struct sched_rt_entity *rt_se;
1084         struct task_struct *p;
1085         struct rt_rq *rt_rq;
1086
1087         rt_rq = &rq->rt;
1088
1089         if (unlikely(!rt_rq->rt_nr_running))
1090                 return NULL;
1091
1092         if (rt_rq_throttled(rt_rq))
1093                 return NULL;
1094
1095         do {
1096                 rt_se = pick_next_rt_entity(rq, rt_rq);
1097                 BUG_ON(!rt_se);
1098                 rt_rq = group_rt_rq(rt_se);
1099         } while (rt_rq);
1100
1101         p = rt_task_of(rt_se);
1102         p->se.exec_start = rq->clock_task;
1103
1104         return p;
1105 }
1106
1107 static struct task_struct *pick_next_task_rt(struct rq *rq)
1108 {
1109         struct task_struct *p = _pick_next_task_rt(rq);
1110
1111         /* The running task is never eligible for pushing */
1112         if (p)
1113                 dequeue_pushable_task(rq, p);
1114
1115 #ifdef CONFIG_SMP
1116         /*
1117          * We detect this state here so that we can avoid taking the RQ
1118          * lock again later if there is no need to push
1119          */
1120         rq->post_schedule = has_pushable_tasks(rq);
1121 #endif
1122
1123         return p;
1124 }
1125
1126 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1127 {
1128         update_curr_rt(rq);
1129         p->se.exec_start = 0;
1130
1131         /*
1132          * The previous task needs to be made eligible for pushing
1133          * if it is still active
1134          */
1135         if (p->se.on_rq && p->rt.nr_cpus_allowed > 1)
1136                 enqueue_pushable_task(rq, p);
1137 }
1138
1139 #ifdef CONFIG_SMP
1140
1141 /* Only try algorithms three times */
1142 #define RT_MAX_TRIES 3
1143
1144 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
1145
1146 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1147 {
1148         if (!task_running(rq, p) &&
1149             (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) &&
1150             (p->rt.nr_cpus_allowed > 1))
1151                 return 1;
1152         return 0;
1153 }
1154
1155 /* Return the second highest RT task, NULL otherwise */
1156 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
1157 {
1158         struct task_struct *next = NULL;
1159         struct sched_rt_entity *rt_se;
1160         struct rt_prio_array *array;
1161         struct rt_rq *rt_rq;
1162         int idx;
1163
1164         for_each_leaf_rt_rq(rt_rq, rq) {
1165                 array = &rt_rq->active;
1166                 idx = sched_find_first_bit(array->bitmap);
1167 next_idx:
1168                 if (idx >= MAX_RT_PRIO)
1169                         continue;
1170                 if (next && next->prio < idx)
1171                         continue;
1172                 list_for_each_entry(rt_se, array->queue + idx, run_list) {
1173                         struct task_struct *p;
1174
1175                         if (!rt_entity_is_task(rt_se))
1176                                 continue;
1177
1178                         p = rt_task_of(rt_se);
1179                         if (pick_rt_task(rq, p, cpu)) {
1180                                 next = p;
1181                                 break;
1182                         }
1183                 }
1184                 if (!next) {
1185                         idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
1186                         goto next_idx;
1187                 }
1188         }
1189
1190         return next;
1191 }
1192
1193 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1194
1195 static int find_lowest_rq(struct task_struct *task)
1196 {
1197         struct sched_domain *sd;
1198         struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
1199         int this_cpu = smp_processor_id();
1200         int cpu      = task_cpu(task);
1201
1202         if (task->rt.nr_cpus_allowed == 1)
1203                 return -1; /* No other targets possible */
1204
1205         if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1206                 return -1; /* No targets found */
1207
1208         /*
1209          * At this point we have built a mask of cpus representing the
1210          * lowest priority tasks in the system.  Now we want to elect
1211          * the best one based on our affinity and topology.
1212          *
1213          * We prioritize the last cpu that the task executed on since
1214          * it is most likely cache-hot in that location.
1215          */
1216         if (cpumask_test_cpu(cpu, lowest_mask))
1217                 return cpu;
1218
1219         /*
1220          * Otherwise, we consult the sched_domains span maps to figure
1221          * out which cpu is logically closest to our hot cache data.
1222          */
1223         if (!cpumask_test_cpu(this_cpu, lowest_mask))
1224                 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1225
1226         for_each_domain(cpu, sd) {
1227                 if (sd->flags & SD_WAKE_AFFINE) {
1228                         int best_cpu;
1229
1230                         /*
1231                          * "this_cpu" is cheaper to preempt than a
1232                          * remote processor.
1233                          */
1234                         if (this_cpu != -1 &&
1235                             cpumask_test_cpu(this_cpu, sched_domain_span(sd)))
1236                                 return this_cpu;
1237
1238                         best_cpu = cpumask_first_and(lowest_mask,
1239                                                      sched_domain_span(sd));
1240                         if (best_cpu < nr_cpu_ids)
1241                                 return best_cpu;
1242                 }
1243         }
1244
1245         /*
1246          * And finally, if there were no matches within the domains
1247          * just give the caller *something* to work with from the compatible
1248          * locations.
1249          */
1250         if (this_cpu != -1)
1251                 return this_cpu;
1252
1253         cpu = cpumask_any(lowest_mask);
1254         if (cpu < nr_cpu_ids)
1255                 return cpu;
1256         return -1;
1257 }
1258
1259 /* Will lock the rq it finds */
1260 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1261 {
1262         struct rq *lowest_rq = NULL;
1263         int tries;
1264         int cpu;
1265
1266         for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1267                 cpu = find_lowest_rq(task);
1268
1269                 if ((cpu == -1) || (cpu == rq->cpu))
1270                         break;
1271
1272                 lowest_rq = cpu_rq(cpu);
1273
1274                 /* if the prio of this runqueue changed, try again */
1275                 if (double_lock_balance(rq, lowest_rq)) {
1276                         /*
1277                          * We had to unlock the run queue. In
1278                          * the mean time, task could have
1279                          * migrated already or had its affinity changed.
1280                          * Also make sure that it wasn't scheduled on its rq.
1281                          */
1282                         if (unlikely(task_rq(task) != rq ||
1283                                      !cpumask_test_cpu(lowest_rq->cpu,
1284                                                        &task->cpus_allowed) ||
1285                                      task_running(rq, task) ||
1286                                      !task->se.on_rq)) {
1287
1288                                 raw_spin_unlock(&lowest_rq->lock);
1289                                 lowest_rq = NULL;
1290                                 break;
1291                         }
1292                 }
1293
1294                 /* If this rq is still suitable use it. */
1295                 if (lowest_rq->rt.highest_prio.curr > task->prio)
1296                         break;
1297
1298                 /* try again */
1299                 double_unlock_balance(rq, lowest_rq);
1300                 lowest_rq = NULL;
1301         }
1302
1303         return lowest_rq;
1304 }
1305
1306 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1307 {
1308         struct task_struct *p;
1309
1310         if (!has_pushable_tasks(rq))
1311                 return NULL;
1312
1313         p = plist_first_entry(&rq->rt.pushable_tasks,
1314                               struct task_struct, pushable_tasks);
1315
1316         BUG_ON(rq->cpu != task_cpu(p));
1317         BUG_ON(task_current(rq, p));
1318         BUG_ON(p->rt.nr_cpus_allowed <= 1);
1319
1320         BUG_ON(!p->se.on_rq);
1321         BUG_ON(!rt_task(p));
1322
1323         return p;
1324 }
1325
1326 /*
1327  * If the current CPU has more than one RT task, see if the non
1328  * running task can migrate over to a CPU that is running a task
1329  * of lesser priority.
1330  */
1331 static int push_rt_task(struct rq *rq)
1332 {
1333         struct task_struct *next_task;
1334         struct rq *lowest_rq;
1335
1336         if (!rq->rt.overloaded)
1337                 return 0;
1338
1339         next_task = pick_next_pushable_task(rq);
1340         if (!next_task)
1341                 return 0;
1342
1343 retry:
1344         if (unlikely(next_task == rq->curr)) {
1345                 WARN_ON(1);
1346                 return 0;
1347         }
1348
1349         /*
1350          * It's possible that the next_task slipped in of
1351          * higher priority than current. If that's the case
1352          * just reschedule current.
1353          */
1354         if (unlikely(next_task->prio < rq->curr->prio)) {
1355                 resched_task(rq->curr);
1356                 return 0;
1357         }
1358
1359         /* We might release rq lock */
1360         get_task_struct(next_task);
1361
1362         /* find_lock_lowest_rq locks the rq if found */
1363         lowest_rq = find_lock_lowest_rq(next_task, rq);
1364         if (!lowest_rq) {
1365                 struct task_struct *task;
1366                 /*
1367                  * find lock_lowest_rq releases rq->lock
1368                  * so it is possible that next_task has migrated.
1369                  *
1370                  * We need to make sure that the task is still on the same
1371                  * run-queue and is also still the next task eligible for
1372                  * pushing.
1373                  */
1374                 task = pick_next_pushable_task(rq);
1375                 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1376                         /*
1377                          * If we get here, the task hasnt moved at all, but
1378                          * it has failed to push.  We will not try again,
1379                          * since the other cpus will pull from us when they
1380                          * are ready.
1381                          */
1382                         dequeue_pushable_task(rq, next_task);
1383                         goto out;
1384                 }
1385
1386                 if (!task)
1387                         /* No more tasks, just exit */
1388                         goto out;
1389
1390                 /*
1391                  * Something has shifted, try again.
1392                  */
1393                 put_task_struct(next_task);
1394                 next_task = task;
1395                 goto retry;
1396         }
1397
1398         deactivate_task(rq, next_task, 0);
1399         set_task_cpu(next_task, lowest_rq->cpu);
1400         activate_task(lowest_rq, next_task, 0);
1401
1402         resched_task(lowest_rq->curr);
1403
1404         double_unlock_balance(rq, lowest_rq);
1405
1406 out:
1407         put_task_struct(next_task);
1408
1409         return 1;
1410 }
1411
1412 static void push_rt_tasks(struct rq *rq)
1413 {
1414         /* push_rt_task will return true if it moved an RT */
1415         while (push_rt_task(rq))
1416                 ;
1417 }
1418
1419 static int pull_rt_task(struct rq *this_rq)
1420 {
1421         int this_cpu = this_rq->cpu, ret = 0, cpu;
1422         struct task_struct *p;
1423         struct rq *src_rq;
1424
1425         if (likely(!rt_overloaded(this_rq)))
1426                 return 0;
1427
1428         for_each_cpu(cpu, this_rq->rd->rto_mask) {
1429                 if (this_cpu == cpu)
1430                         continue;
1431
1432                 src_rq = cpu_rq(cpu);
1433
1434                 /*
1435                  * Don't bother taking the src_rq->lock if the next highest
1436                  * task is known to be lower-priority than our current task.
1437                  * This may look racy, but if this value is about to go
1438                  * logically higher, the src_rq will push this task away.
1439                  * And if its going logically lower, we do not care
1440                  */
1441                 if (src_rq->rt.highest_prio.next >=
1442                     this_rq->rt.highest_prio.curr)
1443                         continue;
1444
1445                 /*
1446                  * We can potentially drop this_rq's lock in
1447                  * double_lock_balance, and another CPU could
1448                  * alter this_rq
1449                  */
1450                 double_lock_balance(this_rq, src_rq);
1451
1452                 /*
1453                  * Are there still pullable RT tasks?
1454                  */
1455                 if (src_rq->rt.rt_nr_running <= 1)
1456                         goto skip;
1457
1458                 p = pick_next_highest_task_rt(src_rq, this_cpu);
1459
1460                 /*
1461                  * Do we have an RT task that preempts
1462                  * the to-be-scheduled task?
1463                  */
1464                 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1465                         WARN_ON(p == src_rq->curr);
1466                         WARN_ON(!p->se.on_rq);
1467
1468                         /*
1469                          * There's a chance that p is higher in priority
1470                          * than what's currently running on its cpu.
1471                          * This is just that p is wakeing up and hasn't
1472                          * had a chance to schedule. We only pull
1473                          * p if it is lower in priority than the
1474                          * current task on the run queue
1475                          */
1476                         if (p->prio < src_rq->curr->prio)
1477                                 goto skip;
1478
1479                         ret = 1;
1480
1481                         deactivate_task(src_rq, p, 0);
1482                         set_task_cpu(p, this_cpu);
1483                         activate_task(this_rq, p, 0);
1484                         /*
1485                          * We continue with the search, just in
1486                          * case there's an even higher prio task
1487                          * in another runqueue. (low likelyhood
1488                          * but possible)
1489                          */
1490                 }
1491 skip:
1492                 double_unlock_balance(this_rq, src_rq);
1493         }
1494
1495         return ret;
1496 }
1497
1498 static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1499 {
1500         /* Try to pull RT tasks here if we lower this rq's prio */
1501         if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio)
1502                 pull_rt_task(rq);
1503 }
1504
1505 static void post_schedule_rt(struct rq *rq)
1506 {
1507         push_rt_tasks(rq);
1508 }
1509
1510 /*
1511  * If we are not running and we are not going to reschedule soon, we should
1512  * try to push tasks away now
1513  */
1514 static void task_woken_rt(struct rq *rq, struct task_struct *p)
1515 {
1516         if (!task_running(rq, p) &&
1517             !test_tsk_need_resched(rq->curr) &&
1518             has_pushable_tasks(rq) &&
1519             p->rt.nr_cpus_allowed > 1 &&
1520             rt_task(rq->curr) &&
1521             (rq->curr->rt.nr_cpus_allowed < 2 ||
1522              rq->curr->prio < p->prio))
1523                 push_rt_tasks(rq);
1524 }
1525
1526 static void set_cpus_allowed_rt(struct task_struct *p,
1527                                 const struct cpumask *new_mask)
1528 {
1529         int weight = cpumask_weight(new_mask);
1530
1531         BUG_ON(!rt_task(p));
1532
1533         /*
1534          * Update the migration status of the RQ if we have an RT task
1535          * which is running AND changing its weight value.
1536          */
1537         if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
1538                 struct rq *rq = task_rq(p);
1539
1540                 if (!task_current(rq, p)) {
1541                         /*
1542                          * Make sure we dequeue this task from the pushable list
1543                          * before going further.  It will either remain off of
1544                          * the list because we are no longer pushable, or it
1545                          * will be requeued.
1546                          */
1547                         if (p->rt.nr_cpus_allowed > 1)
1548                                 dequeue_pushable_task(rq, p);
1549
1550                         /*
1551                          * Requeue if our weight is changing and still > 1
1552                          */
1553                         if (weight > 1)
1554                                 enqueue_pushable_task(rq, p);
1555
1556                 }
1557
1558                 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
1559                         rq->rt.rt_nr_migratory++;
1560                 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
1561                         BUG_ON(!rq->rt.rt_nr_migratory);
1562                         rq->rt.rt_nr_migratory--;
1563                 }
1564
1565                 update_rt_migration(&rq->rt);
1566         }
1567
1568         cpumask_copy(&p->cpus_allowed, new_mask);
1569         p->rt.nr_cpus_allowed = weight;
1570 }
1571
1572 /* Assumes rq->lock is held */
1573 static void rq_online_rt(struct rq *rq)
1574 {
1575         if (rq->rt.overloaded)
1576                 rt_set_overload(rq);
1577
1578         __enable_runtime(rq);
1579
1580         cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1581 }
1582
1583 /* Assumes rq->lock is held */
1584 static void rq_offline_rt(struct rq *rq)
1585 {
1586         if (rq->rt.overloaded)
1587                 rt_clear_overload(rq);
1588
1589         __disable_runtime(rq);
1590
1591         cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
1592 }
1593
1594 /*
1595  * When switch from the rt queue, we bring ourselves to a position
1596  * that we might want to pull RT tasks from other runqueues.
1597  */
1598 static void switched_from_rt(struct rq *rq, struct task_struct *p,
1599                            int running)
1600 {
1601         /*
1602          * If there are other RT tasks then we will reschedule
1603          * and the scheduling of the other RT tasks will handle
1604          * the balancing. But if we are the last RT task
1605          * we may need to handle the pulling of RT tasks
1606          * now.
1607          */
1608         if (!rq->rt.rt_nr_running)
1609                 pull_rt_task(rq);
1610 }
1611
1612 static inline void init_sched_rt_class(void)
1613 {
1614         unsigned int i;
1615
1616         for_each_possible_cpu(i)
1617                 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
1618                                         GFP_KERNEL, cpu_to_node(i));
1619 }
1620 #endif /* CONFIG_SMP */
1621
1622 /*
1623  * When switching a task to RT, we may overload the runqueue
1624  * with RT tasks. In this case we try to push them off to
1625  * other runqueues.
1626  */
1627 static void switched_to_rt(struct rq *rq, struct task_struct *p,
1628                            int running)
1629 {
1630         int check_resched = 1;
1631
1632         /*
1633          * If we are already running, then there's nothing
1634          * that needs to be done. But if we are not running
1635          * we may need to preempt the current running task.
1636          * If that current running task is also an RT task
1637          * then see if we can move to another run queue.
1638          */
1639         if (!running) {
1640 #ifdef CONFIG_SMP
1641                 if (rq->rt.overloaded && push_rt_task(rq) &&
1642                     /* Don't resched if we changed runqueues */
1643                     rq != task_rq(p))
1644                         check_resched = 0;
1645 #endif /* CONFIG_SMP */
1646                 if (check_resched && p->prio < rq->curr->prio)
1647                         resched_task(rq->curr);
1648         }
1649 }
1650
1651 /*
1652  * Priority of the task has changed. This may cause
1653  * us to initiate a push or pull.
1654  */
1655 static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1656                             int oldprio, int running)
1657 {
1658         if (running) {
1659 #ifdef CONFIG_SMP
1660                 /*
1661                  * If our priority decreases while running, we
1662                  * may need to pull tasks to this runqueue.
1663                  */
1664                 if (oldprio < p->prio)
1665                         pull_rt_task(rq);
1666                 /*
1667                  * If there's a higher priority task waiting to run
1668                  * then reschedule. Note, the above pull_rt_task
1669                  * can release the rq lock and p could migrate.
1670                  * Only reschedule if p is still on the same runqueue.
1671                  */
1672                 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1673                         resched_task(p);
1674 #else
1675                 /* For UP simply resched on drop of prio */
1676                 if (oldprio < p->prio)
1677                         resched_task(p);
1678 #endif /* CONFIG_SMP */
1679         } else {
1680                 /*
1681                  * This task is not running, but if it is
1682                  * greater than the current running task
1683                  * then reschedule.
1684                  */
1685                 if (p->prio < rq->curr->prio)
1686                         resched_task(rq->curr);
1687         }
1688 }
1689
1690 static void watchdog(struct rq *rq, struct task_struct *p)
1691 {
1692         unsigned long soft, hard;
1693
1694         /* max may change after cur was read, this will be fixed next tick */
1695         soft = task_rlimit(p, RLIMIT_RTTIME);
1696         hard = task_rlimit_max(p, RLIMIT_RTTIME);
1697
1698         if (soft != RLIM_INFINITY) {
1699                 unsigned long next;
1700
1701                 p->rt.timeout++;
1702                 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1703                 if (p->rt.timeout > next)
1704                         p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
1705         }
1706 }
1707
1708 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
1709 {
1710         update_curr_rt(rq);
1711
1712         watchdog(rq, p);
1713
1714         /*
1715          * RR tasks need a special form of timeslice management.
1716          * FIFO tasks have no timeslices.
1717          */
1718         if (p->policy != SCHED_RR)
1719                 return;
1720
1721         if (--p->rt.time_slice)
1722                 return;
1723
1724         p->rt.time_slice = DEF_TIMESLICE;
1725
1726         /*
1727          * Requeue to the end of queue if we are not the only element
1728          * on the queue:
1729          */
1730         if (p->rt.run_list.prev != p->rt.run_list.next) {
1731                 requeue_task_rt(rq, p, 0);
1732                 set_tsk_need_resched(p);
1733         }
1734 }
1735
1736 static void set_curr_task_rt(struct rq *rq)
1737 {
1738         struct task_struct *p = rq->curr;
1739
1740         p->se.exec_start = rq->clock_task;
1741
1742         /* The running task is never eligible for pushing */
1743         dequeue_pushable_task(rq, p);
1744 }
1745
1746 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
1747 {
1748         /*
1749          * Time slice is 0 for SCHED_FIFO tasks
1750          */
1751         if (task->policy == SCHED_RR)
1752                 return DEF_TIMESLICE;
1753         else
1754                 return 0;
1755 }
1756
1757 static const struct sched_class rt_sched_class = {
1758         .next                   = &fair_sched_class,
1759         .enqueue_task           = enqueue_task_rt,
1760         .dequeue_task           = dequeue_task_rt,
1761         .yield_task             = yield_task_rt,
1762
1763         .check_preempt_curr     = check_preempt_curr_rt,
1764
1765         .pick_next_task         = pick_next_task_rt,
1766         .put_prev_task          = put_prev_task_rt,
1767
1768 #ifdef CONFIG_SMP
1769         .select_task_rq         = select_task_rq_rt,
1770
1771         .set_cpus_allowed       = set_cpus_allowed_rt,
1772         .rq_online              = rq_online_rt,
1773         .rq_offline             = rq_offline_rt,
1774         .pre_schedule           = pre_schedule_rt,
1775         .post_schedule          = post_schedule_rt,
1776         .task_woken             = task_woken_rt,
1777         .switched_from          = switched_from_rt,
1778 #endif
1779
1780         .set_curr_task          = set_curr_task_rt,
1781         .task_tick              = task_tick_rt,
1782
1783         .get_rr_interval        = get_rr_interval_rt,
1784
1785         .prio_changed           = prio_changed_rt,
1786         .switched_to            = switched_to_rt,
1787 };
1788
1789 #ifdef CONFIG_SCHED_DEBUG
1790 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
1791
1792 static void print_rt_stats(struct seq_file *m, int cpu)
1793 {
1794         struct rt_rq *rt_rq;
1795
1796         rcu_read_lock();
1797         for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu))
1798                 print_rt_rq(m, cpu, rt_rq);
1799         rcu_read_unlock();
1800 }
1801 #endif /* CONFIG_SCHED_DEBUG */
1802