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

Archive Download this file



interactive