sched: rt: dont stop the period timer when there are tasks wanting to run
[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_SMP
7
8 static inline int rt_overloaded(struct rq *rq)
9 {
10         return atomic_read(&rq->rd->rto_count);
11 }
12
13 static inline void rt_set_overload(struct rq *rq)
14 {
15         cpu_set(rq->cpu, rq->rd->rto_mask);
16         /*
17          * Make sure the mask is visible before we set
18          * the overload count. That is checked to determine
19          * if we should look at the mask. It would be a shame
20          * if we looked at the mask, but the mask was not
21          * updated yet.
22          */
23         wmb();
24         atomic_inc(&rq->rd->rto_count);
25 }
26
27 static inline void rt_clear_overload(struct rq *rq)
28 {
29         /* the order here really doesn't matter */
30         atomic_dec(&rq->rd->rto_count);
31         cpu_clear(rq->cpu, rq->rd->rto_mask);
32 }
33
34 static void update_rt_migration(struct rq *rq)
35 {
36         if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1)) {
37                 if (!rq->rt.overloaded) {
38                         rt_set_overload(rq);
39                         rq->rt.overloaded = 1;
40                 }
41         } else if (rq->rt.overloaded) {
42                 rt_clear_overload(rq);
43                 rq->rt.overloaded = 0;
44         }
45 }
46 #endif /* CONFIG_SMP */
47
48 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
49 {
50         return container_of(rt_se, struct task_struct, rt);
51 }
52
53 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
54 {
55         return !list_empty(&rt_se->run_list);
56 }
57
58 #ifdef CONFIG_RT_GROUP_SCHED
59
60 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
61 {
62         if (!rt_rq->tg)
63                 return RUNTIME_INF;
64
65         return rt_rq->rt_runtime;
66 }
67
68 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
69 {
70         return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
71 }
72
73 #define for_each_leaf_rt_rq(rt_rq, rq) \
74         list_for_each_entry(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
75
76 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
77 {
78         return rt_rq->rq;
79 }
80
81 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
82 {
83         return rt_se->rt_rq;
84 }
85
86 #define for_each_sched_rt_entity(rt_se) \
87         for (; rt_se; rt_se = rt_se->parent)
88
89 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
90 {
91         return rt_se->my_q;
92 }
93
94 static void enqueue_rt_entity(struct sched_rt_entity *rt_se);
95 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
96
97 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
98 {
99         struct sched_rt_entity *rt_se = rt_rq->rt_se;
100
101         if (rt_se && !on_rt_rq(rt_se) && rt_rq->rt_nr_running) {
102                 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
103
104                 enqueue_rt_entity(rt_se);
105                 if (rt_rq->highest_prio < curr->prio)
106                         resched_task(curr);
107         }
108 }
109
110 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
111 {
112         struct sched_rt_entity *rt_se = rt_rq->rt_se;
113
114         if (rt_se && on_rt_rq(rt_se))
115                 dequeue_rt_entity(rt_se);
116 }
117
118 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
119 {
120         return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
121 }
122
123 static int rt_se_boosted(struct sched_rt_entity *rt_se)
124 {
125         struct rt_rq *rt_rq = group_rt_rq(rt_se);
126         struct task_struct *p;
127
128         if (rt_rq)
129                 return !!rt_rq->rt_nr_boosted;
130
131         p = rt_task_of(rt_se);
132         return p->prio != p->normal_prio;
133 }
134
135 #ifdef CONFIG_SMP
136 static inline cpumask_t sched_rt_period_mask(void)
137 {
138         return cpu_rq(smp_processor_id())->rd->span;
139 }
140 #else
141 static inline cpumask_t sched_rt_period_mask(void)
142 {
143         return cpu_online_map;
144 }
145 #endif
146
147 static inline
148 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
149 {
150         return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
151 }
152
153 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
154 {
155         return &rt_rq->tg->rt_bandwidth;
156 }
157
158 #else
159
160 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
161 {
162         return rt_rq->rt_runtime;
163 }
164
165 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
166 {
167         return ktime_to_ns(def_rt_bandwidth.rt_period);
168 }
169
170 #define for_each_leaf_rt_rq(rt_rq, rq) \
171         for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
172
173 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
174 {
175         return container_of(rt_rq, struct rq, rt);
176 }
177
178 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
179 {
180         struct task_struct *p = rt_task_of(rt_se);
181         struct rq *rq = task_rq(p);
182
183         return &rq->rt;
184 }
185
186 #define for_each_sched_rt_entity(rt_se) \
187         for (; rt_se; rt_se = NULL)
188
189 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
190 {
191         return NULL;
192 }
193
194 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
195 {
196 }
197
198 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
199 {
200 }
201
202 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
203 {
204         return rt_rq->rt_throttled;
205 }
206
207 static inline cpumask_t sched_rt_period_mask(void)
208 {
209         return cpu_online_map;
210 }
211
212 static inline
213 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
214 {
215         return &cpu_rq(cpu)->rt;
216 }
217
218 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
219 {
220         return &def_rt_bandwidth;
221 }
222
223 #endif
224
225 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
226 {
227         int i, idle = 1;
228         cpumask_t span;
229
230         if (rt_b->rt_runtime == RUNTIME_INF)
231                 return 1;
232
233         span = sched_rt_period_mask();
234         for_each_cpu_mask(i, span) {
235                 int enqueue = 0;
236                 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
237                 struct rq *rq = rq_of_rt_rq(rt_rq);
238
239                 spin_lock(&rq->lock);
240                 if (rt_rq->rt_time) {
241                         u64 runtime;
242
243                         spin_lock(&rt_rq->rt_runtime_lock);
244                         runtime = rt_rq->rt_runtime;
245                         rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
246                         if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
247                                 rt_rq->rt_throttled = 0;
248                                 enqueue = 1;
249                         }
250                         if (rt_rq->rt_time || rt_rq->rt_nr_running)
251                                 idle = 0;
252                         spin_unlock(&rt_rq->rt_runtime_lock);
253                 } else if (rt_rq->rt_nr_running)
254                         idle = 0;
255
256                 if (enqueue)
257                         sched_rt_rq_enqueue(rt_rq);
258                 spin_unlock(&rq->lock);
259         }
260
261         return idle;
262 }
263
264 #ifdef CONFIG_SMP
265 static int balance_runtime(struct rt_rq *rt_rq)
266 {
267         struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
268         struct root_domain *rd = cpu_rq(smp_processor_id())->rd;
269         int i, weight, more = 0;
270         u64 rt_period;
271
272         weight = cpus_weight(rd->span);
273
274         spin_lock(&rt_b->rt_runtime_lock);
275         rt_period = ktime_to_ns(rt_b->rt_period);
276         for_each_cpu_mask(i, rd->span) {
277                 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
278                 s64 diff;
279
280                 if (iter == rt_rq)
281                         continue;
282
283                 spin_lock(&iter->rt_runtime_lock);
284                 diff = iter->rt_runtime - iter->rt_time;
285                 if (diff > 0) {
286                         do_div(diff, weight);
287                         if (rt_rq->rt_runtime + diff > rt_period)
288                                 diff = rt_period - rt_rq->rt_runtime;
289                         iter->rt_runtime -= diff;
290                         rt_rq->rt_runtime += diff;
291                         more = 1;
292                         if (rt_rq->rt_runtime == rt_period) {
293                                 spin_unlock(&iter->rt_runtime_lock);
294                                 break;
295                         }
296                 }
297                 spin_unlock(&iter->rt_runtime_lock);
298         }
299         spin_unlock(&rt_b->rt_runtime_lock);
300
301         return more;
302 }
303 #endif
304
305 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
306 {
307 #ifdef CONFIG_RT_GROUP_SCHED
308         struct rt_rq *rt_rq = group_rt_rq(rt_se);
309
310         if (rt_rq)
311                 return rt_rq->highest_prio;
312 #endif
313
314         return rt_task_of(rt_se)->prio;
315 }
316
317 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
318 {
319         u64 runtime = sched_rt_runtime(rt_rq);
320
321         if (runtime == RUNTIME_INF)
322                 return 0;
323
324         if (rt_rq->rt_throttled)
325                 return rt_rq_throttled(rt_rq);
326
327         if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq))
328                 return 0;
329
330 #ifdef CONFIG_SMP
331         if (rt_rq->rt_time > runtime) {
332                 int more;
333
334                 spin_unlock(&rt_rq->rt_runtime_lock);
335                 more = balance_runtime(rt_rq);
336                 spin_lock(&rt_rq->rt_runtime_lock);
337
338                 if (more)
339                         runtime = sched_rt_runtime(rt_rq);
340         }
341 #endif
342
343         if (rt_rq->rt_time > runtime) {
344                 rt_rq->rt_throttled = 1;
345                 if (rt_rq_throttled(rt_rq)) {
346                         sched_rt_rq_dequeue(rt_rq);
347                         return 1;
348                 }
349         }
350
351         return 0;
352 }
353
354 /*
355  * Update the current task's runtime statistics. Skip current tasks that
356  * are not in our scheduling class.
357  */
358 static void update_curr_rt(struct rq *rq)
359 {
360         struct task_struct *curr = rq->curr;
361         struct sched_rt_entity *rt_se = &curr->rt;
362         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
363         u64 delta_exec;
364
365         if (!task_has_rt_policy(curr))
366                 return;
367
368         delta_exec = rq->clock - curr->se.exec_start;
369         if (unlikely((s64)delta_exec < 0))
370                 delta_exec = 0;
371
372         schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec));
373
374         curr->se.sum_exec_runtime += delta_exec;
375         curr->se.exec_start = rq->clock;
376         cpuacct_charge(curr, delta_exec);
377
378         for_each_sched_rt_entity(rt_se) {
379                 rt_rq = rt_rq_of_se(rt_se);
380
381                 spin_lock(&rt_rq->rt_runtime_lock);
382                 rt_rq->rt_time += delta_exec;
383                 if (sched_rt_runtime_exceeded(rt_rq))
384                         resched_task(curr);
385                 spin_unlock(&rt_rq->rt_runtime_lock);
386         }
387 }
388
389 static inline
390 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
391 {
392         WARN_ON(!rt_prio(rt_se_prio(rt_se)));
393         rt_rq->rt_nr_running++;
394 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
395         if (rt_se_prio(rt_se) < rt_rq->highest_prio)
396                 rt_rq->highest_prio = rt_se_prio(rt_se);
397 #endif
398 #ifdef CONFIG_SMP
399         if (rt_se->nr_cpus_allowed > 1) {
400                 struct rq *rq = rq_of_rt_rq(rt_rq);
401                 rq->rt.rt_nr_migratory++;
402         }
403
404         update_rt_migration(rq_of_rt_rq(rt_rq));
405 #endif
406 #ifdef CONFIG_RT_GROUP_SCHED
407         if (rt_se_boosted(rt_se))
408                 rt_rq->rt_nr_boosted++;
409
410         if (rt_rq->tg)
411                 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
412 #else
413         start_rt_bandwidth(&def_rt_bandwidth);
414 #endif
415 }
416
417 static inline
418 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
419 {
420         WARN_ON(!rt_prio(rt_se_prio(rt_se)));
421         WARN_ON(!rt_rq->rt_nr_running);
422         rt_rq->rt_nr_running--;
423 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
424         if (rt_rq->rt_nr_running) {
425                 struct rt_prio_array *array;
426
427                 WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio);
428                 if (rt_se_prio(rt_se) == rt_rq->highest_prio) {
429                         /* recalculate */
430                         array = &rt_rq->active;
431                         rt_rq->highest_prio =
432                                 sched_find_first_bit(array->bitmap);
433                 } /* otherwise leave rq->highest prio alone */
434         } else
435                 rt_rq->highest_prio = MAX_RT_PRIO;
436 #endif
437 #ifdef CONFIG_SMP
438         if (rt_se->nr_cpus_allowed > 1) {
439                 struct rq *rq = rq_of_rt_rq(rt_rq);
440                 rq->rt.rt_nr_migratory--;
441         }
442
443         update_rt_migration(rq_of_rt_rq(rt_rq));
444 #endif /* CONFIG_SMP */
445 #ifdef CONFIG_RT_GROUP_SCHED
446         if (rt_se_boosted(rt_se))
447                 rt_rq->rt_nr_boosted--;
448
449         WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
450 #endif
451 }
452
453 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
454 {
455         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
456         struct rt_prio_array *array = &rt_rq->active;
457         struct rt_rq *group_rq = group_rt_rq(rt_se);
458
459         /*
460          * Don't enqueue the group if its throttled, or when empty.
461          * The latter is a consequence of the former when a child group
462          * get throttled and the current group doesn't have any other
463          * active members.
464          */
465         if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
466                 return;
467
468         list_add_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se));
469         __set_bit(rt_se_prio(rt_se), array->bitmap);
470
471         inc_rt_tasks(rt_se, rt_rq);
472 }
473
474 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
475 {
476         struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
477         struct rt_prio_array *array = &rt_rq->active;
478
479         list_del_init(&rt_se->run_list);
480         if (list_empty(array->queue + rt_se_prio(rt_se)))
481                 __clear_bit(rt_se_prio(rt_se), array->bitmap);
482
483         dec_rt_tasks(rt_se, rt_rq);
484 }
485
486 /*
487  * Because the prio of an upper entry depends on the lower
488  * entries, we must remove entries top - down.
489  */
490 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
491 {
492         struct sched_rt_entity *back = NULL;
493
494         for_each_sched_rt_entity(rt_se) {
495                 rt_se->back = back;
496                 back = rt_se;
497         }
498
499         for (rt_se = back; rt_se; rt_se = rt_se->back) {
500                 if (on_rt_rq(rt_se))
501                         __dequeue_rt_entity(rt_se);
502         }
503 }
504
505 static void enqueue_rt_entity(struct sched_rt_entity *rt_se)
506 {
507         dequeue_rt_stack(rt_se);
508         for_each_sched_rt_entity(rt_se)
509                 __enqueue_rt_entity(rt_se);
510 }
511
512 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
513 {
514         dequeue_rt_stack(rt_se);
515
516         for_each_sched_rt_entity(rt_se) {
517                 struct rt_rq *rt_rq = group_rt_rq(rt_se);
518
519                 if (rt_rq && rt_rq->rt_nr_running)
520                         __enqueue_rt_entity(rt_se);
521         }
522 }
523
524 /*
525  * Adding/removing a task to/from a priority array:
526  */
527 static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
528 {
529         struct sched_rt_entity *rt_se = &p->rt;
530
531         if (wakeup)
532                 rt_se->timeout = 0;
533
534         enqueue_rt_entity(rt_se);
535 }
536
537 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
538 {
539         struct sched_rt_entity *rt_se = &p->rt;
540
541         update_curr_rt(rq);
542         dequeue_rt_entity(rt_se);
543 }
544
545 /*
546  * Put task to the end of the run list without the overhead of dequeue
547  * followed by enqueue.
548  */
549 static
550 void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
551 {
552         struct rt_prio_array *array = &rt_rq->active;
553         struct list_head *queue = array->queue + rt_se_prio(rt_se);
554
555         if (on_rt_rq(rt_se))
556                 list_move_tail(&rt_se->run_list, queue);
557 }
558
559 static void requeue_task_rt(struct rq *rq, struct task_struct *p)
560 {
561         struct sched_rt_entity *rt_se = &p->rt;
562         struct rt_rq *rt_rq;
563
564         for_each_sched_rt_entity(rt_se) {
565                 rt_rq = rt_rq_of_se(rt_se);
566                 requeue_rt_entity(rt_rq, rt_se);
567         }
568 }
569
570 static void yield_task_rt(struct rq *rq)
571 {
572         requeue_task_rt(rq, rq->curr);
573 }
574
575 #ifdef CONFIG_SMP
576 static int find_lowest_rq(struct task_struct *task);
577
578 static int select_task_rq_rt(struct task_struct *p, int sync)
579 {
580         struct rq *rq = task_rq(p);
581
582         /*
583          * If the current task is an RT task, then
584          * try to see if we can wake this RT task up on another
585          * runqueue. Otherwise simply start this RT task
586          * on its current runqueue.
587          *
588          * We want to avoid overloading runqueues. Even if
589          * the RT task is of higher priority than the current RT task.
590          * RT tasks behave differently than other tasks. If
591          * one gets preempted, we try to push it off to another queue.
592          * So trying to keep a preempting RT task on the same
593          * cache hot CPU will force the running RT task to
594          * a cold CPU. So we waste all the cache for the lower
595          * RT task in hopes of saving some of a RT task
596          * that is just being woken and probably will have
597          * cold cache anyway.
598          */
599         if (unlikely(rt_task(rq->curr)) &&
600             (p->rt.nr_cpus_allowed > 1)) {
601                 int cpu = find_lowest_rq(p);
602
603                 return (cpu == -1) ? task_cpu(p) : cpu;
604         }
605
606         /*
607          * Otherwise, just let it ride on the affined RQ and the
608          * post-schedule router will push the preempted task away
609          */
610         return task_cpu(p);
611 }
612 #endif /* CONFIG_SMP */
613
614 /*
615  * Preempt the current task with a newly woken task if needed:
616  */
617 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)
618 {
619         if (p->prio < rq->curr->prio)
620                 resched_task(rq->curr);
621 }
622
623 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
624                                                    struct rt_rq *rt_rq)
625 {
626         struct rt_prio_array *array = &rt_rq->active;
627         struct sched_rt_entity *next = NULL;
628         struct list_head *queue;
629         int idx;
630
631         idx = sched_find_first_bit(array->bitmap);
632         BUG_ON(idx >= MAX_RT_PRIO);
633
634         queue = array->queue + idx;
635         next = list_entry(queue->next, struct sched_rt_entity, run_list);
636
637         return next;
638 }
639
640 static struct task_struct *pick_next_task_rt(struct rq *rq)
641 {
642         struct sched_rt_entity *rt_se;
643         struct task_struct *p;
644         struct rt_rq *rt_rq;
645
646         rt_rq = &rq->rt;
647
648         if (unlikely(!rt_rq->rt_nr_running))
649                 return NULL;
650
651         if (rt_rq_throttled(rt_rq))
652                 return NULL;
653
654         do {
655                 rt_se = pick_next_rt_entity(rq, rt_rq);
656                 BUG_ON(!rt_se);
657                 rt_rq = group_rt_rq(rt_se);
658         } while (rt_rq);
659
660         p = rt_task_of(rt_se);
661         p->se.exec_start = rq->clock;
662         return p;
663 }
664
665 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
666 {
667         update_curr_rt(rq);
668         p->se.exec_start = 0;
669 }
670
671 #ifdef CONFIG_SMP
672
673 /* Only try algorithms three times */
674 #define RT_MAX_TRIES 3
675
676 static int double_lock_balance(struct rq *this_rq, struct rq *busiest);
677 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
678
679 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
680 {
681         if (!task_running(rq, p) &&
682             (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) &&
683             (p->rt.nr_cpus_allowed > 1))
684                 return 1;
685         return 0;
686 }
687
688 /* Return the second highest RT task, NULL otherwise */
689 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
690 {
691         struct task_struct *next = NULL;
692         struct sched_rt_entity *rt_se;
693         struct rt_prio_array *array;
694         struct rt_rq *rt_rq;
695         int idx;
696
697         for_each_leaf_rt_rq(rt_rq, rq) {
698                 array = &rt_rq->active;
699                 idx = sched_find_first_bit(array->bitmap);
700  next_idx:
701                 if (idx >= MAX_RT_PRIO)
702                         continue;
703                 if (next && next->prio < idx)
704                         continue;
705                 list_for_each_entry(rt_se, array->queue + idx, run_list) {
706                         struct task_struct *p = rt_task_of(rt_se);
707                         if (pick_rt_task(rq, p, cpu)) {
708                                 next = p;
709                                 break;
710                         }
711                 }
712                 if (!next) {
713                         idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
714                         goto next_idx;
715                 }
716         }
717
718         return next;
719 }
720
721 static DEFINE_PER_CPU(cpumask_t, local_cpu_mask);
722
723 static int find_lowest_cpus(struct task_struct *task, cpumask_t *lowest_mask)
724 {
725         int       lowest_prio = -1;
726         int       lowest_cpu  = -1;
727         int       count       = 0;
728         int       cpu;
729
730         cpus_and(*lowest_mask, task_rq(task)->rd->online, task->cpus_allowed);
731
732         /*
733          * Scan each rq for the lowest prio.
734          */
735         for_each_cpu_mask(cpu, *lowest_mask) {
736                 struct rq *rq = cpu_rq(cpu);
737
738                 /* We look for lowest RT prio or non-rt CPU */
739                 if (rq->rt.highest_prio >= MAX_RT_PRIO) {
740                         /*
741                          * if we already found a low RT queue
742                          * and now we found this non-rt queue
743                          * clear the mask and set our bit.
744                          * Otherwise just return the queue as is
745                          * and the count==1 will cause the algorithm
746                          * to use the first bit found.
747                          */
748                         if (lowest_cpu != -1) {
749                                 cpus_clear(*lowest_mask);
750                                 cpu_set(rq->cpu, *lowest_mask);
751                         }
752                         return 1;
753                 }
754
755                 /* no locking for now */
756                 if ((rq->rt.highest_prio > task->prio)
757                     && (rq->rt.highest_prio >= lowest_prio)) {
758                         if (rq->rt.highest_prio > lowest_prio) {
759                                 /* new low - clear old data */
760                                 lowest_prio = rq->rt.highest_prio;
761                                 lowest_cpu = cpu;
762                                 count = 0;
763                         }
764                         count++;
765                 } else
766                         cpu_clear(cpu, *lowest_mask);
767         }
768
769         /*
770          * Clear out all the set bits that represent
771          * runqueues that were of higher prio than
772          * the lowest_prio.
773          */
774         if (lowest_cpu > 0) {
775                 /*
776                  * Perhaps we could add another cpumask op to
777                  * zero out bits. Like cpu_zero_bits(cpumask, nrbits);
778                  * Then that could be optimized to use memset and such.
779                  */
780                 for_each_cpu_mask(cpu, *lowest_mask) {
781                         if (cpu >= lowest_cpu)
782                                 break;
783                         cpu_clear(cpu, *lowest_mask);
784                 }
785         }
786
787         return count;
788 }
789
790 static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask)
791 {
792         int first;
793
794         /* "this_cpu" is cheaper to preempt than a remote processor */
795         if ((this_cpu != -1) && cpu_isset(this_cpu, *mask))
796                 return this_cpu;
797
798         first = first_cpu(*mask);
799         if (first != NR_CPUS)
800                 return first;
801
802         return -1;
803 }
804
805 static int find_lowest_rq(struct task_struct *task)
806 {
807         struct sched_domain *sd;
808         cpumask_t *lowest_mask = &__get_cpu_var(local_cpu_mask);
809         int this_cpu = smp_processor_id();
810         int cpu      = task_cpu(task);
811         int count    = find_lowest_cpus(task, lowest_mask);
812
813         if (!count)
814                 return -1; /* No targets found */
815
816         /*
817          * There is no sense in performing an optimal search if only one
818          * target is found.
819          */
820         if (count == 1)
821                 return first_cpu(*lowest_mask);
822
823         /*
824          * At this point we have built a mask of cpus representing the
825          * lowest priority tasks in the system.  Now we want to elect
826          * the best one based on our affinity and topology.
827          *
828          * We prioritize the last cpu that the task executed on since
829          * it is most likely cache-hot in that location.
830          */
831         if (cpu_isset(cpu, *lowest_mask))
832                 return cpu;
833
834         /*
835          * Otherwise, we consult the sched_domains span maps to figure
836          * out which cpu is logically closest to our hot cache data.
837          */
838         if (this_cpu == cpu)
839                 this_cpu = -1; /* Skip this_cpu opt if the same */
840
841         for_each_domain(cpu, sd) {
842                 if (sd->flags & SD_WAKE_AFFINE) {
843                         cpumask_t domain_mask;
844                         int       best_cpu;
845
846                         cpus_and(domain_mask, sd->span, *lowest_mask);
847
848                         best_cpu = pick_optimal_cpu(this_cpu,
849                                                     &domain_mask);
850                         if (best_cpu != -1)
851                                 return best_cpu;
852                 }
853         }
854
855         /*
856          * And finally, if there were no matches within the domains
857          * just give the caller *something* to work with from the compatible
858          * locations.
859          */
860         return pick_optimal_cpu(this_cpu, lowest_mask);
861 }
862
863 /* Will lock the rq it finds */
864 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
865 {
866         struct rq *lowest_rq = NULL;
867         int tries;
868         int cpu;
869
870         for (tries = 0; tries < RT_MAX_TRIES; tries++) {
871                 cpu = find_lowest_rq(task);
872
873                 if ((cpu == -1) || (cpu == rq->cpu))
874                         break;
875
876                 lowest_rq = cpu_rq(cpu);
877
878                 /* if the prio of this runqueue changed, try again */
879                 if (double_lock_balance(rq, lowest_rq)) {
880                         /*
881                          * We had to unlock the run queue. In
882                          * the mean time, task could have
883                          * migrated already or had its affinity changed.
884                          * Also make sure that it wasn't scheduled on its rq.
885                          */
886                         if (unlikely(task_rq(task) != rq ||
887                                      !cpu_isset(lowest_rq->cpu,
888                                                 task->cpus_allowed) ||
889                                      task_running(rq, task) ||
890                                      !task->se.on_rq)) {
891
892                                 spin_unlock(&lowest_rq->lock);
893                                 lowest_rq = NULL;
894                                 break;
895                         }
896                 }
897
898                 /* If this rq is still suitable use it. */
899                 if (lowest_rq->rt.highest_prio > task->prio)
900                         break;
901
902                 /* try again */
903                 spin_unlock(&lowest_rq->lock);
904                 lowest_rq = NULL;
905         }
906
907         return lowest_rq;
908 }
909
910 /*
911  * If the current CPU has more than one RT task, see if the non
912  * running task can migrate over to a CPU that is running a task
913  * of lesser priority.
914  */
915 static int push_rt_task(struct rq *rq)
916 {
917         struct task_struct *next_task;
918         struct rq *lowest_rq;
919         int ret = 0;
920         int paranoid = RT_MAX_TRIES;
921
922         if (!rq->rt.overloaded)
923                 return 0;
924
925         next_task = pick_next_highest_task_rt(rq, -1);
926         if (!next_task)
927                 return 0;
928
929  retry:
930         if (unlikely(next_task == rq->curr)) {
931                 WARN_ON(1);
932                 return 0;
933         }
934
935         /*
936          * It's possible that the next_task slipped in of
937          * higher priority than current. If that's the case
938          * just reschedule current.
939          */
940         if (unlikely(next_task->prio < rq->curr->prio)) {
941                 resched_task(rq->curr);
942                 return 0;
943         }
944
945         /* We might release rq lock */
946         get_task_struct(next_task);
947
948         /* find_lock_lowest_rq locks the rq if found */
949         lowest_rq = find_lock_lowest_rq(next_task, rq);
950         if (!lowest_rq) {
951                 struct task_struct *task;
952                 /*
953                  * find lock_lowest_rq releases rq->lock
954                  * so it is possible that next_task has changed.
955                  * If it has, then try again.
956                  */
957                 task = pick_next_highest_task_rt(rq, -1);
958                 if (unlikely(task != next_task) && task && paranoid--) {
959                         put_task_struct(next_task);
960                         next_task = task;
961                         goto retry;
962                 }
963                 goto out;
964         }
965
966         deactivate_task(rq, next_task, 0);
967         set_task_cpu(next_task, lowest_rq->cpu);
968         activate_task(lowest_rq, next_task, 0);
969
970         resched_task(lowest_rq->curr);
971
972         spin_unlock(&lowest_rq->lock);
973
974         ret = 1;
975 out:
976         put_task_struct(next_task);
977
978         return ret;
979 }
980
981 /*
982  * TODO: Currently we just use the second highest prio task on
983  *       the queue, and stop when it can't migrate (or there's
984  *       no more RT tasks).  There may be a case where a lower
985  *       priority RT task has a different affinity than the
986  *       higher RT task. In this case the lower RT task could
987  *       possibly be able to migrate where as the higher priority
988  *       RT task could not.  We currently ignore this issue.
989  *       Enhancements are welcome!
990  */
991 static void push_rt_tasks(struct rq *rq)
992 {
993         /* push_rt_task will return true if it moved an RT */
994         while (push_rt_task(rq))
995                 ;
996 }
997
998 static int pull_rt_task(struct rq *this_rq)
999 {
1000         int this_cpu = this_rq->cpu, ret = 0, cpu;
1001         struct task_struct *p, *next;
1002         struct rq *src_rq;
1003
1004         if (likely(!rt_overloaded(this_rq)))
1005                 return 0;
1006
1007         next = pick_next_task_rt(this_rq);
1008
1009         for_each_cpu_mask(cpu, this_rq->rd->rto_mask) {
1010                 if (this_cpu == cpu)
1011                         continue;
1012
1013                 src_rq = cpu_rq(cpu);
1014                 /*
1015                  * We can potentially drop this_rq's lock in
1016                  * double_lock_balance, and another CPU could
1017                  * steal our next task - hence we must cause
1018                  * the caller to recalculate the next task
1019                  * in that case:
1020                  */
1021                 if (double_lock_balance(this_rq, src_rq)) {
1022                         struct task_struct *old_next = next;
1023
1024                         next = pick_next_task_rt(this_rq);
1025                         if (next != old_next)
1026                                 ret = 1;
1027                 }
1028
1029                 /*
1030                  * Are there still pullable RT tasks?
1031                  */
1032                 if (src_rq->rt.rt_nr_running <= 1)
1033                         goto skip;
1034
1035                 p = pick_next_highest_task_rt(src_rq, this_cpu);
1036
1037                 /*
1038                  * Do we have an RT task that preempts
1039                  * the to-be-scheduled task?
1040                  */
1041                 if (p && (!next || (p->prio < next->prio))) {
1042                         WARN_ON(p == src_rq->curr);
1043                         WARN_ON(!p->se.on_rq);
1044
1045                         /*
1046                          * There's a chance that p is higher in priority
1047                          * than what's currently running on its cpu.
1048                          * This is just that p is wakeing up and hasn't
1049                          * had a chance to schedule. We only pull
1050                          * p if it is lower in priority than the
1051                          * current task on the run queue or
1052                          * this_rq next task is lower in prio than
1053                          * the current task on that rq.
1054                          */
1055                         if (p->prio < src_rq->curr->prio ||
1056                             (next && next->prio < src_rq->curr->prio))
1057                                 goto skip;
1058
1059                         ret = 1;
1060
1061                         deactivate_task(src_rq, p, 0);
1062                         set_task_cpu(p, this_cpu);
1063                         activate_task(this_rq, p, 0);
1064                         /*
1065                          * We continue with the search, just in
1066                          * case there's an even higher prio task
1067                          * in another runqueue. (low likelyhood
1068                          * but possible)
1069                          *
1070                          * Update next so that we won't pick a task
1071                          * on another cpu with a priority lower (or equal)
1072                          * than the one we just picked.
1073                          */
1074                         next = p;
1075
1076                 }
1077  skip:
1078                 spin_unlock(&src_rq->lock);
1079         }
1080
1081         return ret;
1082 }
1083
1084 static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1085 {
1086         /* Try to pull RT tasks here if we lower this rq's prio */
1087         if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio)
1088                 pull_rt_task(rq);
1089 }
1090
1091 static void post_schedule_rt(struct rq *rq)
1092 {
1093         /*
1094          * If we have more than one rt_task queued, then
1095          * see if we can push the other rt_tasks off to other CPUS.
1096          * Note we may release the rq lock, and since
1097          * the lock was owned by prev, we need to release it
1098          * first via finish_lock_switch and then reaquire it here.
1099          */
1100         if (unlikely(rq->rt.overloaded)) {
1101                 spin_lock_irq(&rq->lock);
1102                 push_rt_tasks(rq);
1103                 spin_unlock_irq(&rq->lock);
1104         }
1105 }
1106
1107 /*
1108  * If we are not running and we are not going to reschedule soon, we should
1109  * try to push tasks away now
1110  */
1111 static void task_wake_up_rt(struct rq *rq, struct task_struct *p)
1112 {
1113         if (!task_running(rq, p) &&
1114             !test_tsk_need_resched(rq->curr) &&
1115             rq->rt.overloaded)
1116                 push_rt_tasks(rq);
1117 }
1118
1119 static unsigned long
1120 load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1121                 unsigned long max_load_move,
1122                 struct sched_domain *sd, enum cpu_idle_type idle,
1123                 int *all_pinned, int *this_best_prio)
1124 {
1125         /* don't touch RT tasks */
1126         return 0;
1127 }
1128
1129 static int
1130 move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1131                  struct sched_domain *sd, enum cpu_idle_type idle)
1132 {
1133         /* don't touch RT tasks */
1134         return 0;
1135 }
1136
1137 static void set_cpus_allowed_rt(struct task_struct *p,
1138                                 const cpumask_t *new_mask)
1139 {
1140         int weight = cpus_weight(*new_mask);
1141
1142         BUG_ON(!rt_task(p));
1143
1144         /*
1145          * Update the migration status of the RQ if we have an RT task
1146          * which is running AND changing its weight value.
1147          */
1148         if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
1149                 struct rq *rq = task_rq(p);
1150
1151                 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
1152                         rq->rt.rt_nr_migratory++;
1153                 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
1154                         BUG_ON(!rq->rt.rt_nr_migratory);
1155                         rq->rt.rt_nr_migratory--;
1156                 }
1157
1158                 update_rt_migration(rq);
1159         }
1160
1161         p->cpus_allowed    = *new_mask;
1162         p->rt.nr_cpus_allowed = weight;
1163 }
1164
1165 /* Assumes rq->lock is held */
1166 static void join_domain_rt(struct rq *rq)
1167 {
1168         if (rq->rt.overloaded)
1169                 rt_set_overload(rq);
1170 }
1171
1172 /* Assumes rq->lock is held */
1173 static void leave_domain_rt(struct rq *rq)
1174 {
1175         if (rq->rt.overloaded)
1176                 rt_clear_overload(rq);
1177 }
1178
1179 /*
1180  * When switch from the rt queue, we bring ourselves to a position
1181  * that we might want to pull RT tasks from other runqueues.
1182  */
1183 static void switched_from_rt(struct rq *rq, struct task_struct *p,
1184                            int running)
1185 {
1186         /*
1187          * If there are other RT tasks then we will reschedule
1188          * and the scheduling of the other RT tasks will handle
1189          * the balancing. But if we are the last RT task
1190          * we may need to handle the pulling of RT tasks
1191          * now.
1192          */
1193         if (!rq->rt.rt_nr_running)
1194                 pull_rt_task(rq);
1195 }
1196 #endif /* CONFIG_SMP */
1197
1198 /*
1199  * When switching a task to RT, we may overload the runqueue
1200  * with RT tasks. In this case we try to push them off to
1201  * other runqueues.
1202  */
1203 static void switched_to_rt(struct rq *rq, struct task_struct *p,
1204                            int running)
1205 {
1206         int check_resched = 1;
1207
1208         /*
1209          * If we are already running, then there's nothing
1210          * that needs to be done. But if we are not running
1211          * we may need to preempt the current running task.
1212          * If that current running task is also an RT task
1213          * then see if we can move to another run queue.
1214          */
1215         if (!running) {
1216 #ifdef CONFIG_SMP
1217                 if (rq->rt.overloaded && push_rt_task(rq) &&
1218                     /* Don't resched if we changed runqueues */
1219                     rq != task_rq(p))
1220                         check_resched = 0;
1221 #endif /* CONFIG_SMP */
1222                 if (check_resched && p->prio < rq->curr->prio)
1223                         resched_task(rq->curr);
1224         }
1225 }
1226
1227 /*
1228  * Priority of the task has changed. This may cause
1229  * us to initiate a push or pull.
1230  */
1231 static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1232                             int oldprio, int running)
1233 {
1234         if (running) {
1235 #ifdef CONFIG_SMP
1236                 /*
1237                  * If our priority decreases while running, we
1238                  * may need to pull tasks to this runqueue.
1239                  */
1240                 if (oldprio < p->prio)
1241                         pull_rt_task(rq);
1242                 /*
1243                  * If there's a higher priority task waiting to run
1244                  * then reschedule. Note, the above pull_rt_task
1245                  * can release the rq lock and p could migrate.
1246                  * Only reschedule if p is still on the same runqueue.
1247                  */
1248                 if (p->prio > rq->rt.highest_prio && rq->curr == p)
1249                         resched_task(p);
1250 #else
1251                 /* For UP simply resched on drop of prio */
1252                 if (oldprio < p->prio)
1253                         resched_task(p);
1254 #endif /* CONFIG_SMP */
1255         } else {
1256                 /*
1257                  * This task is not running, but if it is
1258                  * greater than the current running task
1259                  * then reschedule.
1260                  */
1261                 if (p->prio < rq->curr->prio)
1262                         resched_task(rq->curr);
1263         }
1264 }
1265
1266 static void watchdog(struct rq *rq, struct task_struct *p)
1267 {
1268         unsigned long soft, hard;
1269
1270         if (!p->signal)
1271                 return;
1272
1273         soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur;
1274         hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max;
1275
1276         if (soft != RLIM_INFINITY) {
1277                 unsigned long next;
1278
1279                 p->rt.timeout++;
1280                 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1281                 if (p->rt.timeout > next)
1282                         p->it_sched_expires = p->se.sum_exec_runtime;
1283         }
1284 }
1285
1286 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
1287 {
1288         update_curr_rt(rq);
1289
1290         watchdog(rq, p);
1291
1292         /*
1293          * RR tasks need a special form of timeslice management.
1294          * FIFO tasks have no timeslices.
1295          */
1296         if (p->policy != SCHED_RR)
1297                 return;
1298
1299         if (--p->rt.time_slice)
1300                 return;
1301
1302         p->rt.time_slice = DEF_TIMESLICE;
1303
1304         /*
1305          * Requeue to the end of queue if we are not the only element
1306          * on the queue:
1307          */
1308         if (p->rt.run_list.prev != p->rt.run_list.next) {
1309                 requeue_task_rt(rq, p);
1310                 set_tsk_need_resched(p);
1311         }
1312 }
1313
1314 static void set_curr_task_rt(struct rq *rq)
1315 {
1316         struct task_struct *p = rq->curr;
1317
1318         p->se.exec_start = rq->clock;
1319 }
1320
1321 static const struct sched_class rt_sched_class = {
1322         .next                   = &fair_sched_class,
1323         .enqueue_task           = enqueue_task_rt,
1324         .dequeue_task           = dequeue_task_rt,
1325         .yield_task             = yield_task_rt,
1326 #ifdef CONFIG_SMP
1327         .select_task_rq         = select_task_rq_rt,
1328 #endif /* CONFIG_SMP */
1329
1330         .check_preempt_curr     = check_preempt_curr_rt,
1331
1332         .pick_next_task         = pick_next_task_rt,
1333         .put_prev_task          = put_prev_task_rt,
1334
1335 #ifdef CONFIG_SMP
1336         .load_balance           = load_balance_rt,
1337         .move_one_task          = move_one_task_rt,
1338         .set_cpus_allowed       = set_cpus_allowed_rt,
1339         .join_domain            = join_domain_rt,
1340         .leave_domain           = leave_domain_rt,
1341         .pre_schedule           = pre_schedule_rt,
1342         .post_schedule          = post_schedule_rt,
1343         .task_wake_up           = task_wake_up_rt,
1344         .switched_from          = switched_from_rt,
1345 #endif
1346
1347         .set_curr_task          = set_curr_task_rt,
1348         .task_tick              = task_tick_rt,
1349
1350         .prio_changed           = prio_changed_rt,
1351         .switched_to            = switched_to_rt,
1352 };