rt.c 47.0 KB
Newer Older
I
Ingo Molnar 已提交
1 2 3 4 5
/*
 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
 * policies)
 */

6 7 8 9
#include "sched.h"

#include <linux/slab.h>

10 11
int sched_rr_timeslice = RR_TIMESLICE;

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88
static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);

struct rt_bandwidth def_rt_bandwidth;

static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
{
	struct rt_bandwidth *rt_b =
		container_of(timer, struct rt_bandwidth, rt_period_timer);
	ktime_t now;
	int overrun;
	int idle = 0;

	for (;;) {
		now = hrtimer_cb_get_time(timer);
		overrun = hrtimer_forward(timer, now, rt_b->rt_period);

		if (!overrun)
			break;

		idle = do_sched_rt_period_timer(rt_b, overrun);
	}

	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
}

void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
{
	rt_b->rt_period = ns_to_ktime(period);
	rt_b->rt_runtime = runtime;

	raw_spin_lock_init(&rt_b->rt_runtime_lock);

	hrtimer_init(&rt_b->rt_period_timer,
			CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rt_b->rt_period_timer.function = sched_rt_period_timer;
}

static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
{
	if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
		return;

	if (hrtimer_active(&rt_b->rt_period_timer))
		return;

	raw_spin_lock(&rt_b->rt_runtime_lock);
	start_bandwidth_timer(&rt_b->rt_period_timer, rt_b->rt_period);
	raw_spin_unlock(&rt_b->rt_runtime_lock);
}

void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
{
	struct rt_prio_array *array;
	int i;

	array = &rt_rq->active;
	for (i = 0; i < MAX_RT_PRIO; i++) {
		INIT_LIST_HEAD(array->queue + i);
		__clear_bit(i, array->bitmap);
	}
	/* delimiter for bitsearch: */
	__set_bit(MAX_RT_PRIO, array->bitmap);

#if defined CONFIG_SMP
	rt_rq->highest_prio.curr = MAX_RT_PRIO;
	rt_rq->highest_prio.next = MAX_RT_PRIO;
	rt_rq->rt_nr_migratory = 0;
	rt_rq->overloaded = 0;
	plist_head_init(&rt_rq->pushable_tasks);
#endif

	rt_rq->rt_time = 0;
	rt_rq->rt_throttled = 0;
	rt_rq->rt_runtime = 0;
	raw_spin_lock_init(&rt_rq->rt_runtime_lock);
}

89
#ifdef CONFIG_RT_GROUP_SCHED
90 91 92 93
static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
{
	hrtimer_cancel(&rt_b->rt_period_timer);
}
94 95 96

#define rt_entity_is_task(rt_se) (!(rt_se)->my_q)

97 98
static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
{
99 100 101
#ifdef CONFIG_SCHED_DEBUG
	WARN_ON_ONCE(!rt_entity_is_task(rt_se));
#endif
102 103 104 105 106 107 108 109 110 111 112 113 114
	return container_of(rt_se, struct task_struct, rt);
}

static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
{
	return rt_rq->rq;
}

static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
{
	return rt_se->rt_rq;
}

115 116 117 118 119 120 121
static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
{
	struct rt_rq *rt_rq = rt_se->rt_rq;

	return rt_rq->rq;
}

122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
void free_rt_sched_group(struct task_group *tg)
{
	int i;

	if (tg->rt_se)
		destroy_rt_bandwidth(&tg->rt_bandwidth);

	for_each_possible_cpu(i) {
		if (tg->rt_rq)
			kfree(tg->rt_rq[i]);
		if (tg->rt_se)
			kfree(tg->rt_se[i]);
	}

	kfree(tg->rt_rq);
	kfree(tg->rt_se);
}

void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
		struct sched_rt_entity *rt_se, int cpu,
		struct sched_rt_entity *parent)
{
	struct rq *rq = cpu_rq(cpu);

	rt_rq->highest_prio.curr = MAX_RT_PRIO;
	rt_rq->rt_nr_boosted = 0;
	rt_rq->rq = rq;
	rt_rq->tg = tg;

	tg->rt_rq[cpu] = rt_rq;
	tg->rt_se[cpu] = rt_se;

	if (!rt_se)
		return;

	if (!parent)
		rt_se->rt_rq = &rq->rt;
	else
		rt_se->rt_rq = parent->my_q;

	rt_se->my_q = rt_rq;
	rt_se->parent = parent;
	INIT_LIST_HEAD(&rt_se->run_list);
}

int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
{
	struct rt_rq *rt_rq;
	struct sched_rt_entity *rt_se;
	int i;

	tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->rt_rq)
		goto err;
	tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->rt_se)
		goto err;

	init_rt_bandwidth(&tg->rt_bandwidth,
			ktime_to_ns(def_rt_bandwidth.rt_period), 0);

	for_each_possible_cpu(i) {
		rt_rq = kzalloc_node(sizeof(struct rt_rq),
				     GFP_KERNEL, cpu_to_node(i));
		if (!rt_rq)
			goto err;

		rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
				     GFP_KERNEL, cpu_to_node(i));
		if (!rt_se)
			goto err_free_rq;

		init_rt_rq(rt_rq, cpu_rq(i));
		rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
		init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
	}

	return 1;

err_free_rq:
	kfree(rt_rq);
err:
	return 0;
}

207 208
#else /* CONFIG_RT_GROUP_SCHED */

209 210
#define rt_entity_is_task(rt_se) (1)

211 212 213 214 215
static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
{
	return container_of(rt_se, struct task_struct, rt);
}

216 217 218 219 220
static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
{
	return container_of(rt_rq, struct rq, rt);
}

221
static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
222 223
{
	struct task_struct *p = rt_task_of(rt_se);
224 225 226 227 228 229 230

	return task_rq(p);
}

static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
{
	struct rq *rq = rq_of_rt_se(rt_se);
231 232 233 234

	return &rq->rt;
}

235 236 237 238 239 240
void free_rt_sched_group(struct task_group *tg) { }

int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
{
	return 1;
}
241 242
#endif /* CONFIG_RT_GROUP_SCHED */

S
Steven Rostedt 已提交
243
#ifdef CONFIG_SMP
I
Ingo Molnar 已提交
244

245 246
static int pull_rt_task(struct rq *this_rq);

P
Peter Zijlstra 已提交
247 248 249 250 251 252
static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
{
	/* Try to pull RT tasks here if we lower this rq's prio */
	return rq->rt.highest_prio.curr > prev->prio;
}

253
static inline int rt_overloaded(struct rq *rq)
S
Steven Rostedt 已提交
254
{
255
	return atomic_read(&rq->rd->rto_count);
S
Steven Rostedt 已提交
256
}
I
Ingo Molnar 已提交
257

S
Steven Rostedt 已提交
258 259
static inline void rt_set_overload(struct rq *rq)
{
260 261 262
	if (!rq->online)
		return;

263
	cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
S
Steven Rostedt 已提交
264 265 266 267 268 269
	/*
	 * Make sure the mask is visible before we set
	 * the overload count. That is checked to determine
	 * if we should look at the mask. It would be a shame
	 * if we looked at the mask, but the mask was not
	 * updated yet.
P
Peter Zijlstra 已提交
270 271
	 *
	 * Matched by the barrier in pull_rt_task().
S
Steven Rostedt 已提交
272
	 */
P
Peter Zijlstra 已提交
273
	smp_wmb();
274
	atomic_inc(&rq->rd->rto_count);
S
Steven Rostedt 已提交
275
}
I
Ingo Molnar 已提交
276

S
Steven Rostedt 已提交
277 278
static inline void rt_clear_overload(struct rq *rq)
{
279 280 281
	if (!rq->online)
		return;

S
Steven Rostedt 已提交
282
	/* the order here really doesn't matter */
283
	atomic_dec(&rq->rd->rto_count);
284
	cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
S
Steven Rostedt 已提交
285
}
286

287
static void update_rt_migration(struct rt_rq *rt_rq)
288
{
289
	if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
290 291 292
		if (!rt_rq->overloaded) {
			rt_set_overload(rq_of_rt_rq(rt_rq));
			rt_rq->overloaded = 1;
293
		}
294 295 296
	} else if (rt_rq->overloaded) {
		rt_clear_overload(rq_of_rt_rq(rt_rq));
		rt_rq->overloaded = 0;
297
	}
298
}
S
Steven Rostedt 已提交
299

300 301
static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
302 303
	struct task_struct *p;

304 305 306
	if (!rt_entity_is_task(rt_se))
		return;

307
	p = rt_task_of(rt_se);
308 309 310
	rt_rq = &rq_of_rt_rq(rt_rq)->rt;

	rt_rq->rt_nr_total++;
311
	if (p->nr_cpus_allowed > 1)
312 313 314 315 316 317 318
		rt_rq->rt_nr_migratory++;

	update_rt_migration(rt_rq);
}

static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
319 320
	struct task_struct *p;

321 322 323
	if (!rt_entity_is_task(rt_se))
		return;

324
	p = rt_task_of(rt_se);
325 326 327
	rt_rq = &rq_of_rt_rq(rt_rq)->rt;

	rt_rq->rt_nr_total--;
328
	if (p->nr_cpus_allowed > 1)
329 330 331 332 333
		rt_rq->rt_nr_migratory--;

	update_rt_migration(rt_rq);
}

334 335 336 337 338
static inline int has_pushable_tasks(struct rq *rq)
{
	return !plist_head_empty(&rq->rt.pushable_tasks);
}

P
Peter Zijlstra 已提交
339 340 341 342 343 344 345 346 347
static inline void set_post_schedule(struct rq *rq)
{
	/*
	 * We detect this state here so that we can avoid taking the RQ
	 * lock again later if there is no need to push
	 */
	rq->post_schedule = has_pushable_tasks(rq);
}

348 349 350 351 352
static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
{
	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
	plist_node_init(&p->pushable_tasks, p->prio);
	plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
353 354 355 356

	/* Update the highest prio pushable task */
	if (p->prio < rq->rt.highest_prio.next)
		rq->rt.highest_prio.next = p->prio;
357 358 359 360 361 362
}

static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
{
	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);

363 364 365 366 367 368 369
	/* Update the new highest prio pushable task */
	if (has_pushable_tasks(rq)) {
		p = plist_first_entry(&rq->rt.pushable_tasks,
				      struct task_struct, pushable_tasks);
		rq->rt.highest_prio.next = p->prio;
	} else
		rq->rt.highest_prio.next = MAX_RT_PRIO;
370 371
}

372 373
#else

374
static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
P
Peter Zijlstra 已提交
375
{
P
Peter Zijlstra 已提交
376 377
}

378 379 380 381
static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
{
}

382
static inline
383 384 385 386
void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
}

387
static inline
388 389 390
void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
}
391

P
Peter Zijlstra 已提交
392 393 394 395 396 397 398 399 400 401 402 403 404
static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
{
	return false;
}

static inline int pull_rt_task(struct rq *this_rq)
{
	return 0;
}

static inline void set_post_schedule(struct rq *rq)
{
}
S
Steven Rostedt 已提交
405 406
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
407 408 409 410 411
static inline int on_rt_rq(struct sched_rt_entity *rt_se)
{
	return !list_empty(&rt_se->run_list);
}

412
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
413

P
Peter Zijlstra 已提交
414
static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
415 416
{
	if (!rt_rq->tg)
P
Peter Zijlstra 已提交
417
		return RUNTIME_INF;
P
Peter Zijlstra 已提交
418

P
Peter Zijlstra 已提交
419 420 421 422 423 424
	return rt_rq->rt_runtime;
}

static inline u64 sched_rt_period(struct rt_rq *rt_rq)
{
	return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
P
Peter Zijlstra 已提交
425 426
}

C
Cheng Xu 已提交
427 428
typedef struct task_group *rt_rq_iter_t;

429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445
static inline struct task_group *next_task_group(struct task_group *tg)
{
	do {
		tg = list_entry_rcu(tg->list.next,
			typeof(struct task_group), list);
	} while (&tg->list != &task_groups && task_group_is_autogroup(tg));

	if (&tg->list == &task_groups)
		tg = NULL;

	return tg;
}

#define for_each_rt_rq(rt_rq, iter, rq)					\
	for (iter = container_of(&task_groups, typeof(*iter), list);	\
		(iter = next_task_group(iter)) &&			\
		(rt_rq = iter->rt_rq[cpu_of(rq)]);)
C
Cheng Xu 已提交
446

P
Peter Zijlstra 已提交
447 448 449 450 451 452 453 454
#define for_each_sched_rt_entity(rt_se) \
	for (; rt_se; rt_se = rt_se->parent)

static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
{
	return rt_se->my_q;
}

455
static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head);
P
Peter Zijlstra 已提交
456 457
static void dequeue_rt_entity(struct sched_rt_entity *rt_se);

P
Peter Zijlstra 已提交
458
static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
459
{
460
	struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
461 462
	struct sched_rt_entity *rt_se;

463 464 465
	int cpu = cpu_of(rq_of_rt_rq(rt_rq));

	rt_se = rt_rq->tg->rt_se[cpu];
P
Peter Zijlstra 已提交
466

467 468
	if (rt_rq->rt_nr_running) {
		if (rt_se && !on_rt_rq(rt_se))
469
			enqueue_rt_entity(rt_se, false);
470
		if (rt_rq->highest_prio.curr < curr->prio)
471
			resched_task(curr);
P
Peter Zijlstra 已提交
472 473 474
	}
}

P
Peter Zijlstra 已提交
475
static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
476
{
477
	struct sched_rt_entity *rt_se;
478
	int cpu = cpu_of(rq_of_rt_rq(rt_rq));
479

480
	rt_se = rt_rq->tg->rt_se[cpu];
P
Peter Zijlstra 已提交
481 482 483 484 485

	if (rt_se && on_rt_rq(rt_se))
		dequeue_rt_entity(rt_se);
}

P
Peter Zijlstra 已提交
486 487 488 489 490 491 492 493 494 495 496 497
static int rt_se_boosted(struct sched_rt_entity *rt_se)
{
	struct rt_rq *rt_rq = group_rt_rq(rt_se);
	struct task_struct *p;

	if (rt_rq)
		return !!rt_rq->rt_nr_boosted;

	p = rt_task_of(rt_se);
	return p->prio != p->normal_prio;
}

498
#ifdef CONFIG_SMP
499
static inline const struct cpumask *sched_rt_period_mask(void)
500
{
N
Nathan Zimmer 已提交
501
	return this_rq()->rd->span;
502
}
P
Peter Zijlstra 已提交
503
#else
504
static inline const struct cpumask *sched_rt_period_mask(void)
505
{
506
	return cpu_online_mask;
507 508
}
#endif
P
Peter Zijlstra 已提交
509

510 511
static inline
struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
P
Peter Zijlstra 已提交
512
{
513 514
	return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
}
P
Peter Zijlstra 已提交
515

P
Peter Zijlstra 已提交
516 517 518 519 520
static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
{
	return &rt_rq->tg->rt_bandwidth;
}

521
#else /* !CONFIG_RT_GROUP_SCHED */
522 523 524

static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
{
P
Peter Zijlstra 已提交
525 526 527 528 529 530
	return rt_rq->rt_runtime;
}

static inline u64 sched_rt_period(struct rt_rq *rt_rq)
{
	return ktime_to_ns(def_rt_bandwidth.rt_period);
P
Peter Zijlstra 已提交
531 532
}

C
Cheng Xu 已提交
533 534 535 536 537
typedef struct rt_rq *rt_rq_iter_t;

#define for_each_rt_rq(rt_rq, iter, rq) \
	for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)

P
Peter Zijlstra 已提交
538 539 540 541 542 543 544 545
#define for_each_sched_rt_entity(rt_se) \
	for (; rt_se; rt_se = NULL)

static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
{
	return NULL;
}

P
Peter Zijlstra 已提交
546
static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
547
{
548 549
	if (rt_rq->rt_nr_running)
		resched_task(rq_of_rt_rq(rt_rq)->curr);
P
Peter Zijlstra 已提交
550 551
}

P
Peter Zijlstra 已提交
552
static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
553 554 555
{
}

556
static inline const struct cpumask *sched_rt_period_mask(void)
557
{
558
	return cpu_online_mask;
559 560 561 562 563 564 565 566
}

static inline
struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
{
	return &cpu_rq(cpu)->rt;
}

P
Peter Zijlstra 已提交
567 568 569 570 571
static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
{
	return &def_rt_bandwidth;
}

572
#endif /* CONFIG_RT_GROUP_SCHED */
573

574 575 576 577 578 579 580 581
bool sched_rt_bandwidth_account(struct rt_rq *rt_rq)
{
	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);

	return (hrtimer_active(&rt_b->rt_period_timer) ||
		rt_rq->rt_time < rt_b->rt_runtime);
}

P
Peter Zijlstra 已提交
582
#ifdef CONFIG_SMP
583 584 585
/*
 * We ran out of runtime, see if we can borrow some from our neighbours.
 */
586
static int do_balance_runtime(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
587 588
{
	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
589
	struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
P
Peter Zijlstra 已提交
590 591 592
	int i, weight, more = 0;
	u64 rt_period;

593
	weight = cpumask_weight(rd->span);
P
Peter Zijlstra 已提交
594

595
	raw_spin_lock(&rt_b->rt_runtime_lock);
P
Peter Zijlstra 已提交
596
	rt_period = ktime_to_ns(rt_b->rt_period);
597
	for_each_cpu(i, rd->span) {
P
Peter Zijlstra 已提交
598 599 600 601 602 603
		struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
		s64 diff;

		if (iter == rt_rq)
			continue;

604
		raw_spin_lock(&iter->rt_runtime_lock);
605 606 607 608 609
		/*
		 * Either all rqs have inf runtime and there's nothing to steal
		 * or __disable_runtime() below sets a specific rq to inf to
		 * indicate its been disabled and disalow stealing.
		 */
P
Peter Zijlstra 已提交
610 611 612
		if (iter->rt_runtime == RUNTIME_INF)
			goto next;

613 614 615 616
		/*
		 * From runqueues with spare time, take 1/n part of their
		 * spare time, but no more than our period.
		 */
P
Peter Zijlstra 已提交
617 618
		diff = iter->rt_runtime - iter->rt_time;
		if (diff > 0) {
619
			diff = div_u64((u64)diff, weight);
P
Peter Zijlstra 已提交
620 621 622 623 624 625
			if (rt_rq->rt_runtime + diff > rt_period)
				diff = rt_period - rt_rq->rt_runtime;
			iter->rt_runtime -= diff;
			rt_rq->rt_runtime += diff;
			more = 1;
			if (rt_rq->rt_runtime == rt_period) {
626
				raw_spin_unlock(&iter->rt_runtime_lock);
P
Peter Zijlstra 已提交
627 628 629
				break;
			}
		}
P
Peter Zijlstra 已提交
630
next:
631
		raw_spin_unlock(&iter->rt_runtime_lock);
P
Peter Zijlstra 已提交
632
	}
633
	raw_spin_unlock(&rt_b->rt_runtime_lock);
P
Peter Zijlstra 已提交
634 635 636

	return more;
}
P
Peter Zijlstra 已提交
637

638 639 640
/*
 * Ensure this RQ takes back all the runtime it lend to its neighbours.
 */
P
Peter Zijlstra 已提交
641 642 643
static void __disable_runtime(struct rq *rq)
{
	struct root_domain *rd = rq->rd;
C
Cheng Xu 已提交
644
	rt_rq_iter_t iter;
P
Peter Zijlstra 已提交
645 646 647 648 649
	struct rt_rq *rt_rq;

	if (unlikely(!scheduler_running))
		return;

C
Cheng Xu 已提交
650
	for_each_rt_rq(rt_rq, iter, rq) {
P
Peter Zijlstra 已提交
651 652 653 654
		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
		s64 want;
		int i;

655 656
		raw_spin_lock(&rt_b->rt_runtime_lock);
		raw_spin_lock(&rt_rq->rt_runtime_lock);
657 658 659 660 661
		/*
		 * Either we're all inf and nobody needs to borrow, or we're
		 * already disabled and thus have nothing to do, or we have
		 * exactly the right amount of runtime to take out.
		 */
P
Peter Zijlstra 已提交
662 663 664
		if (rt_rq->rt_runtime == RUNTIME_INF ||
				rt_rq->rt_runtime == rt_b->rt_runtime)
			goto balanced;
665
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
666

667 668 669 670 671
		/*
		 * Calculate the difference between what we started out with
		 * and what we current have, that's the amount of runtime
		 * we lend and now have to reclaim.
		 */
P
Peter Zijlstra 已提交
672 673
		want = rt_b->rt_runtime - rt_rq->rt_runtime;

674 675 676
		/*
		 * Greedy reclaim, take back as much as we can.
		 */
677
		for_each_cpu(i, rd->span) {
P
Peter Zijlstra 已提交
678 679 680
			struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
			s64 diff;

681 682 683
			/*
			 * Can't reclaim from ourselves or disabled runqueues.
			 */
684
			if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
685 686
				continue;

687
			raw_spin_lock(&iter->rt_runtime_lock);
P
Peter Zijlstra 已提交
688 689 690 691 692 693 694 695
			if (want > 0) {
				diff = min_t(s64, iter->rt_runtime, want);
				iter->rt_runtime -= diff;
				want -= diff;
			} else {
				iter->rt_runtime -= want;
				want -= want;
			}
696
			raw_spin_unlock(&iter->rt_runtime_lock);
P
Peter Zijlstra 已提交
697 698 699 700 701

			if (!want)
				break;
		}

702
		raw_spin_lock(&rt_rq->rt_runtime_lock);
703 704 705 706
		/*
		 * We cannot be left wanting - that would mean some runtime
		 * leaked out of the system.
		 */
P
Peter Zijlstra 已提交
707 708
		BUG_ON(want);
balanced:
709 710 711 712
		/*
		 * Disable all the borrow logic by pretending we have inf
		 * runtime - in which case borrowing doesn't make sense.
		 */
P
Peter Zijlstra 已提交
713
		rt_rq->rt_runtime = RUNTIME_INF;
714
		rt_rq->rt_throttled = 0;
715 716
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
		raw_spin_unlock(&rt_b->rt_runtime_lock);
P
Peter Zijlstra 已提交
717 718 719 720 721
	}
}

static void __enable_runtime(struct rq *rq)
{
C
Cheng Xu 已提交
722
	rt_rq_iter_t iter;
P
Peter Zijlstra 已提交
723 724 725 726 727
	struct rt_rq *rt_rq;

	if (unlikely(!scheduler_running))
		return;

728 729 730
	/*
	 * Reset each runqueue's bandwidth settings
	 */
C
Cheng Xu 已提交
731
	for_each_rt_rq(rt_rq, iter, rq) {
P
Peter Zijlstra 已提交
732 733
		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);

734 735
		raw_spin_lock(&rt_b->rt_runtime_lock);
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
736 737
		rt_rq->rt_runtime = rt_b->rt_runtime;
		rt_rq->rt_time = 0;
738
		rt_rq->rt_throttled = 0;
739 740
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
		raw_spin_unlock(&rt_b->rt_runtime_lock);
P
Peter Zijlstra 已提交
741 742 743
	}
}

744 745 746 747
static int balance_runtime(struct rt_rq *rt_rq)
{
	int more = 0;

748 749 750
	if (!sched_feat(RT_RUNTIME_SHARE))
		return more;

751
	if (rt_rq->rt_time > rt_rq->rt_runtime) {
752
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
753
		more = do_balance_runtime(rt_rq);
754
		raw_spin_lock(&rt_rq->rt_runtime_lock);
755 756 757 758
	}

	return more;
}
759
#else /* !CONFIG_SMP */
760 761 762 763
static inline int balance_runtime(struct rt_rq *rt_rq)
{
	return 0;
}
764
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
765

766 767
static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
{
768
	int i, idle = 1, throttled = 0;
769
	const struct cpumask *span;
770 771

	span = sched_rt_period_mask();
772 773 774 775 776 777 778 779 780 781 782 783 784
#ifdef CONFIG_RT_GROUP_SCHED
	/*
	 * FIXME: isolated CPUs should really leave the root task group,
	 * whether they are isolcpus or were isolated via cpusets, lest
	 * the timer run on a CPU which does not service all runqueues,
	 * potentially leaving other CPUs indefinitely throttled.  If
	 * isolation is really required, the user will turn the throttle
	 * off to kill the perturbations it causes anyway.  Meanwhile,
	 * this maintains functionality for boot and/or troubleshooting.
	 */
	if (rt_b == &root_task_group.rt_bandwidth)
		span = cpu_online_mask;
#endif
785
	for_each_cpu(i, span) {
786 787 788 789
		int enqueue = 0;
		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
		struct rq *rq = rq_of_rt_rq(rt_rq);

790
		raw_spin_lock(&rq->lock);
791 792 793
		if (rt_rq->rt_time) {
			u64 runtime;

794
			raw_spin_lock(&rt_rq->rt_runtime_lock);
795 796 797 798 799 800 801
			if (rt_rq->rt_throttled)
				balance_runtime(rt_rq);
			runtime = rt_rq->rt_runtime;
			rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
				rt_rq->rt_throttled = 0;
				enqueue = 1;
802 803 804 805 806 807 808

				/*
				 * Force a clock update if the CPU was idle,
				 * lest wakeup -> unthrottle time accumulate.
				 */
				if (rt_rq->rt_nr_running && rq->curr == rq->idle)
					rq->skip_clock_update = -1;
809 810 811
			}
			if (rt_rq->rt_time || rt_rq->rt_nr_running)
				idle = 0;
812
			raw_spin_unlock(&rt_rq->rt_runtime_lock);
813
		} else if (rt_rq->rt_nr_running) {
814
			idle = 0;
815 816 817
			if (!rt_rq_throttled(rt_rq))
				enqueue = 1;
		}
818 819
		if (rt_rq->rt_throttled)
			throttled = 1;
820 821 822

		if (enqueue)
			sched_rt_rq_enqueue(rt_rq);
823
		raw_spin_unlock(&rq->lock);
824 825
	}

826 827 828
	if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
		return 1;

829 830
	return idle;
}
P
Peter Zijlstra 已提交
831

P
Peter Zijlstra 已提交
832 833
static inline int rt_se_prio(struct sched_rt_entity *rt_se)
{
834
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
835 836 837
	struct rt_rq *rt_rq = group_rt_rq(rt_se);

	if (rt_rq)
838
		return rt_rq->highest_prio.curr;
P
Peter Zijlstra 已提交
839 840 841 842 843
#endif

	return rt_task_of(rt_se)->prio;
}

P
Peter Zijlstra 已提交
844
static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
P
Peter Zijlstra 已提交
845
{
P
Peter Zijlstra 已提交
846
	u64 runtime = sched_rt_runtime(rt_rq);
P
Peter Zijlstra 已提交
847 848

	if (rt_rq->rt_throttled)
P
Peter Zijlstra 已提交
849
		return rt_rq_throttled(rt_rq);
P
Peter Zijlstra 已提交
850

851
	if (runtime >= sched_rt_period(rt_rq))
P
Peter Zijlstra 已提交
852 853
		return 0;

854 855 856 857
	balance_runtime(rt_rq);
	runtime = sched_rt_runtime(rt_rq);
	if (runtime == RUNTIME_INF)
		return 0;
P
Peter Zijlstra 已提交
858

P
Peter Zijlstra 已提交
859
	if (rt_rq->rt_time > runtime) {
860 861 862 863 864 865 866
		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);

		/*
		 * Don't actually throttle groups that have no runtime assigned
		 * but accrue some time due to boosting.
		 */
		if (likely(rt_b->rt_runtime)) {
867 868
			static bool once = false;

869
			rt_rq->rt_throttled = 1;
870 871 872 873 874

			if (!once) {
				once = true;
				printk_sched("sched: RT throttling activated\n");
			}
875 876 877 878 879 880 881 882 883
		} else {
			/*
			 * In case we did anyway, make it go away,
			 * replenishment is a joke, since it will replenish us
			 * with exactly 0 ns.
			 */
			rt_rq->rt_time = 0;
		}

P
Peter Zijlstra 已提交
884
		if (rt_rq_throttled(rt_rq)) {
P
Peter Zijlstra 已提交
885
			sched_rt_rq_dequeue(rt_rq);
P
Peter Zijlstra 已提交
886 887
			return 1;
		}
P
Peter Zijlstra 已提交
888 889 890 891 892
	}

	return 0;
}

I
Ingo Molnar 已提交
893 894 895 896
/*
 * Update the current task's runtime statistics. Skip current tasks that
 * are not in our scheduling class.
 */
A
Alexey Dobriyan 已提交
897
static void update_curr_rt(struct rq *rq)
I
Ingo Molnar 已提交
898 899
{
	struct task_struct *curr = rq->curr;
P
Peter Zijlstra 已提交
900 901
	struct sched_rt_entity *rt_se = &curr->rt;
	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
I
Ingo Molnar 已提交
902 903
	u64 delta_exec;

P
Peter Zijlstra 已提交
904
	if (curr->sched_class != &rt_sched_class)
I
Ingo Molnar 已提交
905 906
		return;

907
	delta_exec = rq_clock_task(rq) - curr->se.exec_start;
908 909
	if (unlikely((s64)delta_exec <= 0))
		return;
I
Ingo Molnar 已提交
910

911 912
	schedstat_set(curr->se.statistics.exec_max,
		      max(curr->se.statistics.exec_max, delta_exec));
I
Ingo Molnar 已提交
913 914

	curr->se.sum_exec_runtime += delta_exec;
915 916
	account_group_exec_runtime(curr, delta_exec);

917
	curr->se.exec_start = rq_clock_task(rq);
918
	cpuacct_charge(curr, delta_exec);
P
Peter Zijlstra 已提交
919

920 921
	sched_rt_avg_update(rq, delta_exec);

922 923 924
	if (!rt_bandwidth_enabled())
		return;

D
Dhaval Giani 已提交
925 926 927
	for_each_sched_rt_entity(rt_se) {
		rt_rq = rt_rq_of_se(rt_se);

928
		if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
929
			raw_spin_lock(&rt_rq->rt_runtime_lock);
930 931 932
			rt_rq->rt_time += delta_exec;
			if (sched_rt_runtime_exceeded(rt_rq))
				resched_task(curr);
933
			raw_spin_unlock(&rt_rq->rt_runtime_lock);
934
		}
D
Dhaval Giani 已提交
935
	}
I
Ingo Molnar 已提交
936 937
}

938
#if defined CONFIG_SMP
939

940 941
static void
inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
942
{
G
Gregory Haskins 已提交
943
	struct rq *rq = rq_of_rt_rq(rt_rq);
944

945 946 947 948 949 950 951
#ifdef CONFIG_RT_GROUP_SCHED
	/*
	 * Change rq's cpupri only if rt_rq is the top queue.
	 */
	if (&rq->rt != rt_rq)
		return;
#endif
952 953
	if (rq->online && prio < prev_prio)
		cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
954
}
955

956 957 958 959
static void
dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
{
	struct rq *rq = rq_of_rt_rq(rt_rq);
960

961 962 963 964 965 966 967
#ifdef CONFIG_RT_GROUP_SCHED
	/*
	 * Change rq's cpupri only if rt_rq is the top queue.
	 */
	if (&rq->rt != rt_rq)
		return;
#endif
968 969
	if (rq->online && rt_rq->highest_prio.curr != prev_prio)
		cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
970 971
}

972 973
#else /* CONFIG_SMP */

P
Peter Zijlstra 已提交
974
static inline
975 976 977 978 979
void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
static inline
void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}

#endif /* CONFIG_SMP */
980

981
#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
static void
inc_rt_prio(struct rt_rq *rt_rq, int prio)
{
	int prev_prio = rt_rq->highest_prio.curr;

	if (prio < prev_prio)
		rt_rq->highest_prio.curr = prio;

	inc_rt_prio_smp(rt_rq, prio, prev_prio);
}

static void
dec_rt_prio(struct rt_rq *rt_rq, int prio)
{
	int prev_prio = rt_rq->highest_prio.curr;

P
Peter Zijlstra 已提交
998
	if (rt_rq->rt_nr_running) {
999

1000
		WARN_ON(prio < prev_prio);
1001

1002
		/*
1003 1004
		 * This may have been our highest task, and therefore
		 * we may have some recomputation to do
1005
		 */
1006
		if (prio == prev_prio) {
1007 1008 1009
			struct rt_prio_array *array = &rt_rq->active;

			rt_rq->highest_prio.curr =
1010
				sched_find_first_bit(array->bitmap);
1011 1012
		}

1013
	} else
1014
		rt_rq->highest_prio.curr = MAX_RT_PRIO;
1015

1016 1017
	dec_rt_prio_smp(rt_rq, prio, prev_prio);
}
1018

1019 1020 1021 1022 1023 1024
#else

static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}

#endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1025

1026
#ifdef CONFIG_RT_GROUP_SCHED
1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040

static void
inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
	if (rt_se_boosted(rt_se))
		rt_rq->rt_nr_boosted++;

	if (rt_rq->tg)
		start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
}

static void
dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
P
Peter Zijlstra 已提交
1041 1042 1043 1044
	if (rt_se_boosted(rt_se))
		rt_rq->rt_nr_boosted--;

	WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
}

#else /* CONFIG_RT_GROUP_SCHED */

static void
inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
	start_rt_bandwidth(&def_rt_bandwidth);
}

static inline
void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}

#endif /* CONFIG_RT_GROUP_SCHED */

1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070
static inline
unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
{
	struct rt_rq *group_rq = group_rt_rq(rt_se);

	if (group_rq)
		return group_rq->rt_nr_running;
	else
		return 1;
}

1071 1072 1073 1074 1075 1076
static inline
void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
	int prio = rt_se_prio(rt_se);

	WARN_ON(!rt_prio(prio));
1077
	rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088

	inc_rt_prio(rt_rq, prio);
	inc_rt_migration(rt_se, rt_rq);
	inc_rt_group(rt_se, rt_rq);
}

static inline
void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
{
	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
	WARN_ON(!rt_rq->rt_nr_running);
1089
	rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1090 1091 1092 1093

	dec_rt_prio(rt_rq, rt_se_prio(rt_se));
	dec_rt_migration(rt_se, rt_rq);
	dec_rt_group(rt_se, rt_rq);
1094 1095
}

1096
static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
I
Ingo Molnar 已提交
1097
{
P
Peter Zijlstra 已提交
1098 1099 1100
	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
	struct rt_prio_array *array = &rt_rq->active;
	struct rt_rq *group_rq = group_rt_rq(rt_se);
1101
	struct list_head *queue = array->queue + rt_se_prio(rt_se);
I
Ingo Molnar 已提交
1102

1103 1104 1105 1106 1107 1108 1109
	/*
	 * Don't enqueue the group if its throttled, or when empty.
	 * The latter is a consequence of the former when a child group
	 * get throttled and the current group doesn't have any other
	 * active members.
	 */
	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
P
Peter Zijlstra 已提交
1110
		return;
1111

1112 1113 1114 1115
	if (head)
		list_add(&rt_se->run_list, queue);
	else
		list_add_tail(&rt_se->run_list, queue);
P
Peter Zijlstra 已提交
1116
	__set_bit(rt_se_prio(rt_se), array->bitmap);
1117

P
Peter Zijlstra 已提交
1118 1119 1120
	inc_rt_tasks(rt_se, rt_rq);
}

1121
static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
P
Peter Zijlstra 已提交
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
{
	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
	struct rt_prio_array *array = &rt_rq->active;

	list_del_init(&rt_se->run_list);
	if (list_empty(array->queue + rt_se_prio(rt_se)))
		__clear_bit(rt_se_prio(rt_se), array->bitmap);

	dec_rt_tasks(rt_se, rt_rq);
}

/*
 * Because the prio of an upper entry depends on the lower
 * entries, we must remove entries top - down.
 */
1137
static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
P
Peter Zijlstra 已提交
1138
{
1139
	struct sched_rt_entity *back = NULL;
P
Peter Zijlstra 已提交
1140

1141 1142 1143 1144 1145 1146 1147
	for_each_sched_rt_entity(rt_se) {
		rt_se->back = back;
		back = rt_se;
	}

	for (rt_se = back; rt_se; rt_se = rt_se->back) {
		if (on_rt_rq(rt_se))
1148 1149 1150 1151
			__dequeue_rt_entity(rt_se);
	}
}

1152
static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
1153 1154 1155
{
	dequeue_rt_stack(rt_se);
	for_each_sched_rt_entity(rt_se)
1156
		__enqueue_rt_entity(rt_se, head);
1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
}

static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
{
	dequeue_rt_stack(rt_se);

	for_each_sched_rt_entity(rt_se) {
		struct rt_rq *rt_rq = group_rt_rq(rt_se);

		if (rt_rq && rt_rq->rt_nr_running)
1167
			__enqueue_rt_entity(rt_se, false);
1168
	}
I
Ingo Molnar 已提交
1169 1170 1171 1172 1173
}

/*
 * Adding/removing a task to/from a priority array:
 */
1174
static void
1175
enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
P
Peter Zijlstra 已提交
1176 1177 1178
{
	struct sched_rt_entity *rt_se = &p->rt;

1179
	if (flags & ENQUEUE_WAKEUP)
P
Peter Zijlstra 已提交
1180 1181
		rt_se->timeout = 0;

1182
	enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
1183

1184
	if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1185
		enqueue_pushable_task(rq, p);
1186 1187

	inc_nr_running(rq);
P
Peter Zijlstra 已提交
1188 1189
}

1190
static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
I
Ingo Molnar 已提交
1191
{
P
Peter Zijlstra 已提交
1192
	struct sched_rt_entity *rt_se = &p->rt;
I
Ingo Molnar 已提交
1193

1194
	update_curr_rt(rq);
1195
	dequeue_rt_entity(rt_se);
1196

1197
	dequeue_pushable_task(rq, p);
1198 1199

	dec_nr_running(rq);
I
Ingo Molnar 已提交
1200 1201 1202
}

/*
1203 1204
 * Put task to the head or the end of the run list without the overhead of
 * dequeue followed by enqueue.
I
Ingo Molnar 已提交
1205
 */
1206 1207
static void
requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
P
Peter Zijlstra 已提交
1208
{
1209
	if (on_rt_rq(rt_se)) {
1210 1211 1212 1213 1214 1215 1216
		struct rt_prio_array *array = &rt_rq->active;
		struct list_head *queue = array->queue + rt_se_prio(rt_se);

		if (head)
			list_move(&rt_se->run_list, queue);
		else
			list_move_tail(&rt_se->run_list, queue);
1217
	}
P
Peter Zijlstra 已提交
1218 1219
}

1220
static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
I
Ingo Molnar 已提交
1221
{
P
Peter Zijlstra 已提交
1222 1223
	struct sched_rt_entity *rt_se = &p->rt;
	struct rt_rq *rt_rq;
I
Ingo Molnar 已提交
1224

P
Peter Zijlstra 已提交
1225 1226
	for_each_sched_rt_entity(rt_se) {
		rt_rq = rt_rq_of_se(rt_se);
1227
		requeue_rt_entity(rt_rq, rt_se, head);
P
Peter Zijlstra 已提交
1228
	}
I
Ingo Molnar 已提交
1229 1230
}

P
Peter Zijlstra 已提交
1231
static void yield_task_rt(struct rq *rq)
I
Ingo Molnar 已提交
1232
{
1233
	requeue_task_rt(rq, rq->curr, 0);
I
Ingo Molnar 已提交
1234 1235
}

1236
#ifdef CONFIG_SMP
1237 1238
static int find_lowest_rq(struct task_struct *task);

1239
static int
1240
select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1241
{
1242 1243
	struct task_struct *curr;
	struct rq *rq;
1244

1245
	if (p->nr_cpus_allowed == 1)
1246 1247
		goto out;

1248 1249 1250 1251
	/* For anything but wake ups, just return the task_cpu */
	if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
		goto out;

1252 1253 1254 1255 1256
	rq = cpu_rq(cpu);

	rcu_read_lock();
	curr = ACCESS_ONCE(rq->curr); /* unlocked access */

1257
	/*
1258
	 * If the current task on @p's runqueue is an RT task, then
1259 1260 1261 1262
	 * try to see if we can wake this RT task up on another
	 * runqueue. Otherwise simply start this RT task
	 * on its current runqueue.
	 *
1263 1264 1265 1266 1267 1268 1269 1270 1271
	 * We want to avoid overloading runqueues. If the woken
	 * task is a higher priority, then it will stay on this CPU
	 * and the lower prio task should be moved to another CPU.
	 * Even though this will probably make the lower prio task
	 * lose its cache, we do not want to bounce a higher task
	 * around just because it gave up its CPU, perhaps for a
	 * lock?
	 *
	 * For equal prio tasks, we just let the scheduler sort it out.
1272 1273 1274 1275 1276 1277
	 *
	 * Otherwise, just let it ride on the affined RQ and the
	 * post-schedule router will push the preempted task away
	 *
	 * This test is optimistic, if we get it wrong the load-balancer
	 * will have to sort it out.
1278
	 */
1279
	if (curr && unlikely(rt_task(curr)) &&
1280
	    (curr->nr_cpus_allowed < 2 ||
1281
	     curr->prio <= p->prio)) {
1282
		int target = find_lowest_rq(p);
1283

1284 1285
		if (target != -1)
			cpu = target;
1286
	}
1287
	rcu_read_unlock();
1288

1289
out:
1290
	return cpu;
1291
}
1292 1293 1294

static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
{
1295
	if (rq->curr->nr_cpus_allowed == 1)
1296 1297
		return;

1298
	if (p->nr_cpus_allowed != 1
1299 1300
	    && cpupri_find(&rq->rd->cpupri, p, NULL))
		return;
1301

1302 1303
	if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
		return;
1304 1305 1306 1307 1308 1309 1310 1311 1312 1313

	/*
	 * There appears to be other cpus that can accept
	 * current and none to run 'p', so lets reschedule
	 * to try and push current away:
	 */
	requeue_task_rt(rq, p, 1);
	resched_task(rq->curr);
}

1314 1315
#endif /* CONFIG_SMP */

I
Ingo Molnar 已提交
1316 1317 1318
/*
 * Preempt the current task with a newly woken task if needed:
 */
P
Peter Zijlstra 已提交
1319
static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
I
Ingo Molnar 已提交
1320
{
1321
	if (p->prio < rq->curr->prio) {
I
Ingo Molnar 已提交
1322
		resched_task(rq->curr);
1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338
		return;
	}

#ifdef CONFIG_SMP
	/*
	 * If:
	 *
	 * - the newly woken task is of equal priority to the current task
	 * - the newly woken task is non-migratable while current is migratable
	 * - current will be preempted on the next reschedule
	 *
	 * we should check to see if current can readily move to a different
	 * cpu.  If so, we will reschedule to allow the push logic to try
	 * to move current somewhere else, making room for our non-migratable
	 * task.
	 */
1339
	if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1340
		check_preempt_equal_prio(rq, p);
1341
#endif
I
Ingo Molnar 已提交
1342 1343
}

P
Peter Zijlstra 已提交
1344 1345
static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
						   struct rt_rq *rt_rq)
I
Ingo Molnar 已提交
1346
{
P
Peter Zijlstra 已提交
1347 1348
	struct rt_prio_array *array = &rt_rq->active;
	struct sched_rt_entity *next = NULL;
I
Ingo Molnar 已提交
1349 1350 1351 1352
	struct list_head *queue;
	int idx;

	idx = sched_find_first_bit(array->bitmap);
P
Peter Zijlstra 已提交
1353
	BUG_ON(idx >= MAX_RT_PRIO);
I
Ingo Molnar 已提交
1354 1355

	queue = array->queue + idx;
P
Peter Zijlstra 已提交
1356
	next = list_entry(queue->next, struct sched_rt_entity, run_list);
1357

P
Peter Zijlstra 已提交
1358 1359
	return next;
}
I
Ingo Molnar 已提交
1360

1361
static struct task_struct *_pick_next_task_rt(struct rq *rq)
P
Peter Zijlstra 已提交
1362 1363 1364
{
	struct sched_rt_entity *rt_se;
	struct task_struct *p;
1365
	struct rt_rq *rt_rq  = &rq->rt;
P
Peter Zijlstra 已提交
1366 1367 1368

	do {
		rt_se = pick_next_rt_entity(rq, rt_rq);
1369
		BUG_ON(!rt_se);
P
Peter Zijlstra 已提交
1370 1371 1372 1373
		rt_rq = group_rt_rq(rt_se);
	} while (rt_rq);

	p = rt_task_of(rt_se);
1374
	p->se.exec_start = rq_clock_task(rq);
1375 1376 1377 1378

	return p;
}

1379 1380
static struct task_struct *
pick_next_task_rt(struct rq *rq, struct task_struct *prev)
1381
{
1382 1383 1384
	struct task_struct *p;
	struct rt_rq *rt_rq = &rq->rt;

1385
	if (need_pull_rt_task(rq, prev)) {
1386
		pull_rt_task(rq);
1387 1388
		/*
		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1389 1390
		 * means a dl or stop task can slip in, in which case we need
		 * to re-start task selection.
1391
		 */
1392 1393
		if (unlikely((rq->stop && rq->stop->on_rq) ||
			     rq->dl.dl_nr_running))
1394 1395
			return RETRY_TASK;
	}
1396

1397 1398 1399 1400 1401 1402 1403
	/*
	 * We may dequeue prev's rt_rq in put_prev_task().
	 * So, we update time before rt_nr_running check.
	 */
	if (prev->sched_class == &rt_sched_class)
		update_curr_rt(rq);

1404 1405 1406 1407 1408 1409
	if (!rt_rq->rt_nr_running)
		return NULL;

	if (rt_rq_throttled(rt_rq))
		return NULL;

1410
	put_prev_task(rq, prev);
1411 1412

	p = _pick_next_task_rt(rq);
1413 1414 1415 1416 1417

	/* The running task is never eligible for pushing */
	if (p)
		dequeue_pushable_task(rq, p);

P
Peter Zijlstra 已提交
1418
	set_post_schedule(rq);
1419

P
Peter Zijlstra 已提交
1420
	return p;
I
Ingo Molnar 已提交
1421 1422
}

1423
static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
I
Ingo Molnar 已提交
1424
{
1425
	update_curr_rt(rq);
1426 1427 1428 1429 1430

	/*
	 * The previous task needs to be made eligible for pushing
	 * if it is still active
	 */
1431
	if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1432
		enqueue_pushable_task(rq, p);
I
Ingo Molnar 已提交
1433 1434
}

1435
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
1436

S
Steven Rostedt 已提交
1437 1438 1439
/* Only try algorithms three times */
#define RT_MAX_TRIES 3

1440 1441 1442
static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
{
	if (!task_running(rq, p) &&
1443
	    cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
1444 1445 1446 1447
		return 1;
	return 0;
}

1448 1449 1450 1451 1452
/*
 * Return the highest pushable rq's task, which is suitable to be executed
 * on the cpu, NULL otherwise
 */
static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
S
Steven Rostedt 已提交
1453
{
1454 1455
	struct plist_head *head = &rq->rt.pushable_tasks;
	struct task_struct *p;
1456

1457 1458
	if (!has_pushable_tasks(rq))
		return NULL;
1459

1460 1461 1462
	plist_for_each_entry(p, head, pushable_tasks) {
		if (pick_rt_task(rq, p, cpu))
			return p;
1463 1464
	}

1465
	return NULL;
S
Steven Rostedt 已提交
1466 1467
}

1468
static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
S
Steven Rostedt 已提交
1469

G
Gregory Haskins 已提交
1470 1471 1472
static int find_lowest_rq(struct task_struct *task)
{
	struct sched_domain *sd;
1473
	struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
G
Gregory Haskins 已提交
1474 1475
	int this_cpu = smp_processor_id();
	int cpu      = task_cpu(task);
G
Gregory Haskins 已提交
1476

1477 1478 1479 1480
	/* Make sure the mask is initialized first */
	if (unlikely(!lowest_mask))
		return -1;

1481
	if (task->nr_cpus_allowed == 1)
1482
		return -1; /* No other targets possible */
G
Gregory Haskins 已提交
1483

1484 1485
	if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
		return -1; /* No targets found */
G
Gregory Haskins 已提交
1486 1487 1488 1489 1490 1491 1492 1493 1494

	/*
	 * At this point we have built a mask of cpus representing the
	 * lowest priority tasks in the system.  Now we want to elect
	 * the best one based on our affinity and topology.
	 *
	 * We prioritize the last cpu that the task executed on since
	 * it is most likely cache-hot in that location.
	 */
1495
	if (cpumask_test_cpu(cpu, lowest_mask))
G
Gregory Haskins 已提交
1496 1497 1498 1499 1500 1501
		return cpu;

	/*
	 * Otherwise, we consult the sched_domains span maps to figure
	 * out which cpu is logically closest to our hot cache data.
	 */
R
Rusty Russell 已提交
1502 1503
	if (!cpumask_test_cpu(this_cpu, lowest_mask))
		this_cpu = -1; /* Skip this_cpu opt if not among lowest */
G
Gregory Haskins 已提交
1504

1505
	rcu_read_lock();
R
Rusty Russell 已提交
1506 1507 1508
	for_each_domain(cpu, sd) {
		if (sd->flags & SD_WAKE_AFFINE) {
			int best_cpu;
G
Gregory Haskins 已提交
1509

R
Rusty Russell 已提交
1510 1511 1512 1513 1514
			/*
			 * "this_cpu" is cheaper to preempt than a
			 * remote processor.
			 */
			if (this_cpu != -1 &&
1515 1516
			    cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
				rcu_read_unlock();
R
Rusty Russell 已提交
1517
				return this_cpu;
1518
			}
R
Rusty Russell 已提交
1519 1520 1521

			best_cpu = cpumask_first_and(lowest_mask,
						     sched_domain_span(sd));
1522 1523
			if (best_cpu < nr_cpu_ids) {
				rcu_read_unlock();
R
Rusty Russell 已提交
1524
				return best_cpu;
1525
			}
G
Gregory Haskins 已提交
1526 1527
		}
	}
1528
	rcu_read_unlock();
G
Gregory Haskins 已提交
1529 1530 1531 1532 1533 1534

	/*
	 * And finally, if there were no matches within the domains
	 * just give the caller *something* to work with from the compatible
	 * locations.
	 */
R
Rusty Russell 已提交
1535 1536 1537 1538 1539 1540 1541
	if (this_cpu != -1)
		return this_cpu;

	cpu = cpumask_any(lowest_mask);
	if (cpu < nr_cpu_ids)
		return cpu;
	return -1;
1542 1543 1544
}

/* Will lock the rq it finds */
1545
static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1546 1547 1548
{
	struct rq *lowest_rq = NULL;
	int tries;
1549
	int cpu;
S
Steven Rostedt 已提交
1550

1551 1552 1553
	for (tries = 0; tries < RT_MAX_TRIES; tries++) {
		cpu = find_lowest_rq(task);

1554
		if ((cpu == -1) || (cpu == rq->cpu))
S
Steven Rostedt 已提交
1555 1556
			break;

1557 1558
		lowest_rq = cpu_rq(cpu);

S
Steven Rostedt 已提交
1559
		/* if the prio of this runqueue changed, try again */
1560
		if (double_lock_balance(rq, lowest_rq)) {
S
Steven Rostedt 已提交
1561 1562 1563 1564 1565 1566
			/*
			 * We had to unlock the run queue. In
			 * the mean time, task could have
			 * migrated already or had its affinity changed.
			 * Also make sure that it wasn't scheduled on its rq.
			 */
1567
			if (unlikely(task_rq(task) != rq ||
1568
				     !cpumask_test_cpu(lowest_rq->cpu,
1569
						       tsk_cpus_allowed(task)) ||
1570
				     task_running(rq, task) ||
P
Peter Zijlstra 已提交
1571
				     !task->on_rq)) {
1572

1573
				double_unlock_balance(rq, lowest_rq);
S
Steven Rostedt 已提交
1574 1575 1576 1577 1578 1579
				lowest_rq = NULL;
				break;
			}
		}

		/* If this rq is still suitable use it. */
1580
		if (lowest_rq->rt.highest_prio.curr > task->prio)
S
Steven Rostedt 已提交
1581 1582 1583
			break;

		/* try again */
1584
		double_unlock_balance(rq, lowest_rq);
S
Steven Rostedt 已提交
1585 1586 1587 1588 1589 1590
		lowest_rq = NULL;
	}

	return lowest_rq;
}

1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602
static struct task_struct *pick_next_pushable_task(struct rq *rq)
{
	struct task_struct *p;

	if (!has_pushable_tasks(rq))
		return NULL;

	p = plist_first_entry(&rq->rt.pushable_tasks,
			      struct task_struct, pushable_tasks);

	BUG_ON(rq->cpu != task_cpu(p));
	BUG_ON(task_current(rq, p));
1603
	BUG_ON(p->nr_cpus_allowed <= 1);
1604

P
Peter Zijlstra 已提交
1605
	BUG_ON(!p->on_rq);
1606 1607 1608 1609 1610
	BUG_ON(!rt_task(p));

	return p;
}

S
Steven Rostedt 已提交
1611 1612 1613 1614 1615
/*
 * If the current CPU has more than one RT task, see if the non
 * running task can migrate over to a CPU that is running a task
 * of lesser priority.
 */
1616
static int push_rt_task(struct rq *rq)
S
Steven Rostedt 已提交
1617 1618 1619
{
	struct task_struct *next_task;
	struct rq *lowest_rq;
1620
	int ret = 0;
S
Steven Rostedt 已提交
1621

G
Gregory Haskins 已提交
1622 1623 1624
	if (!rq->rt.overloaded)
		return 0;

1625
	next_task = pick_next_pushable_task(rq);
S
Steven Rostedt 已提交
1626 1627 1628
	if (!next_task)
		return 0;

P
Peter Zijlstra 已提交
1629
retry:
1630
	if (unlikely(next_task == rq->curr)) {
1631
		WARN_ON(1);
S
Steven Rostedt 已提交
1632
		return 0;
1633
	}
S
Steven Rostedt 已提交
1634 1635 1636 1637 1638 1639

	/*
	 * It's possible that the next_task slipped in of
	 * higher priority than current. If that's the case
	 * just reschedule current.
	 */
1640 1641
	if (unlikely(next_task->prio < rq->curr->prio)) {
		resched_task(rq->curr);
S
Steven Rostedt 已提交
1642 1643 1644
		return 0;
	}

1645
	/* We might release rq lock */
S
Steven Rostedt 已提交
1646 1647 1648
	get_task_struct(next_task);

	/* find_lock_lowest_rq locks the rq if found */
1649
	lowest_rq = find_lock_lowest_rq(next_task, rq);
S
Steven Rostedt 已提交
1650 1651 1652
	if (!lowest_rq) {
		struct task_struct *task;
		/*
1653
		 * find_lock_lowest_rq releases rq->lock
1654 1655 1656 1657 1658
		 * so it is possible that next_task has migrated.
		 *
		 * We need to make sure that the task is still on the same
		 * run-queue and is also still the next task eligible for
		 * pushing.
S
Steven Rostedt 已提交
1659
		 */
1660
		task = pick_next_pushable_task(rq);
1661 1662
		if (task_cpu(next_task) == rq->cpu && task == next_task) {
			/*
1663 1664 1665 1666
			 * The task hasn't migrated, and is still the next
			 * eligible task, but we failed to find a run-queue
			 * to push it to.  Do not retry in this case, since
			 * other cpus will pull from us when ready.
1667 1668
			 */
			goto out;
S
Steven Rostedt 已提交
1669
		}
1670

1671 1672 1673 1674
		if (!task)
			/* No more tasks, just exit */
			goto out;

1675
		/*
1676
		 * Something has shifted, try again.
1677
		 */
1678 1679 1680
		put_task_struct(next_task);
		next_task = task;
		goto retry;
S
Steven Rostedt 已提交
1681 1682
	}

1683
	deactivate_task(rq, next_task, 0);
S
Steven Rostedt 已提交
1684 1685
	set_task_cpu(next_task, lowest_rq->cpu);
	activate_task(lowest_rq, next_task, 0);
1686
	ret = 1;
S
Steven Rostedt 已提交
1687 1688 1689

	resched_task(lowest_rq->curr);

1690
	double_unlock_balance(rq, lowest_rq);
S
Steven Rostedt 已提交
1691 1692 1693 1694

out:
	put_task_struct(next_task);

1695
	return ret;
S
Steven Rostedt 已提交
1696 1697 1698 1699 1700 1701 1702 1703 1704
}

static void push_rt_tasks(struct rq *rq)
{
	/* push_rt_task will return true if it moved an RT */
	while (push_rt_task(rq))
		;
}

1705 1706
static int pull_rt_task(struct rq *this_rq)
{
I
Ingo Molnar 已提交
1707
	int this_cpu = this_rq->cpu, ret = 0, cpu;
1708
	struct task_struct *p;
1709 1710
	struct rq *src_rq;

1711
	if (likely(!rt_overloaded(this_rq)))
1712 1713
		return 0;

P
Peter Zijlstra 已提交
1714 1715 1716 1717 1718 1719
	/*
	 * Match the barrier from rt_set_overloaded; this guarantees that if we
	 * see overloaded we must also see the rto_mask bit.
	 */
	smp_rmb();

1720
	for_each_cpu(cpu, this_rq->rd->rto_mask) {
1721 1722 1723 1724
		if (this_cpu == cpu)
			continue;

		src_rq = cpu_rq(cpu);
1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736

		/*
		 * Don't bother taking the src_rq->lock if the next highest
		 * task is known to be lower-priority than our current task.
		 * This may look racy, but if this value is about to go
		 * logically higher, the src_rq will push this task away.
		 * And if its going logically lower, we do not care
		 */
		if (src_rq->rt.highest_prio.next >=
		    this_rq->rt.highest_prio.curr)
			continue;

1737 1738 1739
		/*
		 * We can potentially drop this_rq's lock in
		 * double_lock_balance, and another CPU could
1740
		 * alter this_rq
1741
		 */
1742
		double_lock_balance(this_rq, src_rq);
1743 1744

		/*
1745 1746
		 * We can pull only a task, which is pushable
		 * on its rq, and no others.
1747
		 */
1748
		p = pick_highest_pushable_task(src_rq, this_cpu);
1749 1750 1751 1752 1753

		/*
		 * Do we have an RT task that preempts
		 * the to-be-scheduled task?
		 */
1754
		if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1755
			WARN_ON(p == src_rq->curr);
P
Peter Zijlstra 已提交
1756
			WARN_ON(!p->on_rq);
1757 1758 1759 1760 1761 1762 1763

			/*
			 * There's a chance that p is higher in priority
			 * than what's currently running on its cpu.
			 * This is just that p is wakeing up and hasn't
			 * had a chance to schedule. We only pull
			 * p if it is lower in priority than the
1764
			 * current task on the run queue
1765
			 */
1766
			if (p->prio < src_rq->curr->prio)
M
Mike Galbraith 已提交
1767
				goto skip;
1768 1769 1770 1771 1772 1773 1774 1775 1776

			ret = 1;

			deactivate_task(src_rq, p, 0);
			set_task_cpu(p, this_cpu);
			activate_task(this_rq, p, 0);
			/*
			 * We continue with the search, just in
			 * case there's an even higher prio task
L
Lucas De Marchi 已提交
1777
			 * in another runqueue. (low likelihood
1778 1779 1780
			 * but possible)
			 */
		}
P
Peter Zijlstra 已提交
1781
skip:
1782
		double_unlock_balance(this_rq, src_rq);
1783 1784 1785 1786 1787
	}

	return ret;
}

1788
static void post_schedule_rt(struct rq *rq)
S
Steven Rostedt 已提交
1789
{
1790
	push_rt_tasks(rq);
S
Steven Rostedt 已提交
1791 1792
}

1793 1794 1795 1796
/*
 * If we are not running and we are not going to reschedule soon, we should
 * try to push tasks away now
 */
1797
static void task_woken_rt(struct rq *rq, struct task_struct *p)
1798
{
1799
	if (!task_running(rq, p) &&
1800
	    !test_tsk_need_resched(rq->curr) &&
1801
	    has_pushable_tasks(rq) &&
1802
	    p->nr_cpus_allowed > 1 &&
1803
	    (dl_task(rq->curr) || rt_task(rq->curr)) &&
1804
	    (rq->curr->nr_cpus_allowed < 2 ||
1805
	     rq->curr->prio <= p->prio))
1806 1807 1808
		push_rt_tasks(rq);
}

1809
static void set_cpus_allowed_rt(struct task_struct *p,
1810
				const struct cpumask *new_mask)
1811
{
1812 1813
	struct rq *rq;
	int weight;
1814 1815 1816

	BUG_ON(!rt_task(p));

1817 1818
	if (!p->on_rq)
		return;
1819

1820
	weight = cpumask_weight(new_mask);
1821

1822 1823 1824 1825
	/*
	 * Only update if the process changes its state from whether it
	 * can migrate or not.
	 */
1826
	if ((p->nr_cpus_allowed > 1) == (weight > 1))
1827
		return;
1828

1829
	rq = task_rq(p);
1830

1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842
	/*
	 * The process used to be able to migrate OR it can now migrate
	 */
	if (weight <= 1) {
		if (!task_current(rq, p))
			dequeue_pushable_task(rq, p);
		BUG_ON(!rq->rt.rt_nr_migratory);
		rq->rt.rt_nr_migratory--;
	} else {
		if (!task_current(rq, p))
			enqueue_pushable_task(rq, p);
		rq->rt.rt_nr_migratory++;
1843
	}
1844 1845

	update_rt_migration(&rq->rt);
1846
}
1847

1848
/* Assumes rq->lock is held */
1849
static void rq_online_rt(struct rq *rq)
1850 1851 1852
{
	if (rq->rt.overloaded)
		rt_set_overload(rq);
1853

P
Peter Zijlstra 已提交
1854 1855
	__enable_runtime(rq);

1856
	cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1857 1858 1859
}

/* Assumes rq->lock is held */
1860
static void rq_offline_rt(struct rq *rq)
1861 1862 1863
{
	if (rq->rt.overloaded)
		rt_clear_overload(rq);
1864

P
Peter Zijlstra 已提交
1865 1866
	__disable_runtime(rq);

1867
	cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
1868
}
1869 1870 1871 1872 1873

/*
 * When switch from the rt queue, we bring ourselves to a position
 * that we might want to pull RT tasks from other runqueues.
 */
P
Peter Zijlstra 已提交
1874
static void switched_from_rt(struct rq *rq, struct task_struct *p)
1875 1876 1877 1878 1879 1880 1881 1882
{
	/*
	 * If there are other RT tasks then we will reschedule
	 * and the scheduling of the other RT tasks will handle
	 * the balancing. But if we are the last RT task
	 * we may need to handle the pulling of RT tasks
	 * now.
	 */
1883 1884 1885 1886 1887
	if (!p->on_rq || rq->rt.rt_nr_running)
		return;

	if (pull_rt_task(rq))
		resched_task(rq->curr);
1888
}
1889

1890
void __init init_sched_rt_class(void)
1891 1892 1893
{
	unsigned int i;

1894
	for_each_possible_cpu(i) {
1895
		zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
1896
					GFP_KERNEL, cpu_to_node(i));
1897
	}
1898
}
1899 1900 1901 1902 1903 1904 1905
#endif /* CONFIG_SMP */

/*
 * When switching a task to RT, we may overload the runqueue
 * with RT tasks. In this case we try to push them off to
 * other runqueues.
 */
P
Peter Zijlstra 已提交
1906
static void switched_to_rt(struct rq *rq, struct task_struct *p)
1907 1908 1909 1910 1911 1912 1913 1914 1915 1916
{
	int check_resched = 1;

	/*
	 * If we are already running, then there's nothing
	 * that needs to be done. But if we are not running
	 * we may need to preempt the current running task.
	 * If that current running task is also an RT task
	 * then see if we can move to another run queue.
	 */
P
Peter Zijlstra 已提交
1917
	if (p->on_rq && rq->curr != p) {
1918
#ifdef CONFIG_SMP
1919
		if (p->nr_cpus_allowed > 1 && rq->rt.overloaded &&
1920
		    /* Don't resched if we changed runqueues */
1921
		    push_rt_task(rq) && rq != task_rq(p))
1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932
			check_resched = 0;
#endif /* CONFIG_SMP */
		if (check_resched && p->prio < rq->curr->prio)
			resched_task(rq->curr);
	}
}

/*
 * Priority of the task has changed. This may cause
 * us to initiate a push or pull.
 */
P
Peter Zijlstra 已提交
1933 1934
static void
prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
1935
{
P
Peter Zijlstra 已提交
1936
	if (!p->on_rq)
P
Peter Zijlstra 已提交
1937 1938 1939
		return;

	if (rq->curr == p) {
1940 1941 1942 1943 1944 1945 1946 1947 1948
#ifdef CONFIG_SMP
		/*
		 * If our priority decreases while running, we
		 * may need to pull tasks to this runqueue.
		 */
		if (oldprio < p->prio)
			pull_rt_task(rq);
		/*
		 * If there's a higher priority task waiting to run
1949 1950 1951
		 * then reschedule. Note, the above pull_rt_task
		 * can release the rq lock and p could migrate.
		 * Only reschedule if p is still on the same runqueue.
1952
		 */
1953
		if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1954 1955 1956 1957 1958
			resched_task(p);
#else
		/* For UP simply resched on drop of prio */
		if (oldprio < p->prio)
			resched_task(p);
S
Steven Rostedt 已提交
1959
#endif /* CONFIG_SMP */
1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970
	} else {
		/*
		 * This task is not running, but if it is
		 * greater than the current running task
		 * then reschedule.
		 */
		if (p->prio < rq->curr->prio)
			resched_task(rq->curr);
	}
}

1971 1972 1973 1974
static void watchdog(struct rq *rq, struct task_struct *p)
{
	unsigned long soft, hard;

1975 1976 1977
	/* max may change after cur was read, this will be fixed next tick */
	soft = task_rlimit(p, RLIMIT_RTTIME);
	hard = task_rlimit_max(p, RLIMIT_RTTIME);
1978 1979 1980 1981

	if (soft != RLIM_INFINITY) {
		unsigned long next;

1982 1983 1984 1985 1986
		if (p->rt.watchdog_stamp != jiffies) {
			p->rt.timeout++;
			p->rt.watchdog_stamp = jiffies;
		}

1987
		next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1988
		if (p->rt.timeout > next)
1989
			p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
1990 1991
	}
}
I
Ingo Molnar 已提交
1992

P
Peter Zijlstra 已提交
1993
static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
I
Ingo Molnar 已提交
1994
{
1995 1996
	struct sched_rt_entity *rt_se = &p->rt;

1997 1998
	update_curr_rt(rq);

1999 2000
	watchdog(rq, p);

I
Ingo Molnar 已提交
2001 2002 2003 2004 2005 2006 2007
	/*
	 * RR tasks need a special form of timeslice management.
	 * FIFO tasks have no timeslices.
	 */
	if (p->policy != SCHED_RR)
		return;

P
Peter Zijlstra 已提交
2008
	if (--p->rt.time_slice)
I
Ingo Molnar 已提交
2009 2010
		return;

2011
	p->rt.time_slice = sched_rr_timeslice;
I
Ingo Molnar 已提交
2012

2013
	/*
L
Li Bin 已提交
2014 2015
	 * Requeue to the end of queue if we (and all of our ancestors) are not
	 * the only element on the queue
2016
	 */
2017 2018 2019 2020 2021 2022
	for_each_sched_rt_entity(rt_se) {
		if (rt_se->run_list.prev != rt_se->run_list.next) {
			requeue_task_rt(rq, p, 0);
			set_tsk_need_resched(p);
			return;
		}
2023
	}
I
Ingo Molnar 已提交
2024 2025
}

2026 2027 2028 2029
static void set_curr_task_rt(struct rq *rq)
{
	struct task_struct *p = rq->curr;

2030
	p->se.exec_start = rq_clock_task(rq);
2031 2032 2033

	/* The running task is never eligible for pushing */
	dequeue_pushable_task(rq, p);
2034 2035
}

2036
static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2037 2038 2039 2040 2041
{
	/*
	 * Time slice is 0 for SCHED_FIFO tasks
	 */
	if (task->policy == SCHED_RR)
2042
		return sched_rr_timeslice;
2043 2044 2045 2046
	else
		return 0;
}

2047
const struct sched_class rt_sched_class = {
2048
	.next			= &fair_sched_class,
I
Ingo Molnar 已提交
2049 2050 2051 2052 2053 2054 2055 2056 2057
	.enqueue_task		= enqueue_task_rt,
	.dequeue_task		= dequeue_task_rt,
	.yield_task		= yield_task_rt,

	.check_preempt_curr	= check_preempt_curr_rt,

	.pick_next_task		= pick_next_task_rt,
	.put_prev_task		= put_prev_task_rt,

2058
#ifdef CONFIG_SMP
L
Li Zefan 已提交
2059 2060
	.select_task_rq		= select_task_rq_rt,

2061
	.set_cpus_allowed       = set_cpus_allowed_rt,
2062 2063
	.rq_online              = rq_online_rt,
	.rq_offline             = rq_offline_rt,
2064
	.post_schedule		= post_schedule_rt,
2065
	.task_woken		= task_woken_rt,
2066
	.switched_from		= switched_from_rt,
2067
#endif
I
Ingo Molnar 已提交
2068

2069
	.set_curr_task          = set_curr_task_rt,
I
Ingo Molnar 已提交
2070
	.task_tick		= task_tick_rt,
2071

2072 2073
	.get_rr_interval	= get_rr_interval_rt,

2074 2075
	.prio_changed		= prio_changed_rt,
	.switched_to		= switched_to_rt,
I
Ingo Molnar 已提交
2076
};
2077 2078 2079 2080

#ifdef CONFIG_SCHED_DEBUG
extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);

2081
void print_rt_stats(struct seq_file *m, int cpu)
2082
{
C
Cheng Xu 已提交
2083
	rt_rq_iter_t iter;
2084 2085 2086
	struct rt_rq *rt_rq;

	rcu_read_lock();
C
Cheng Xu 已提交
2087
	for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2088 2089 2090
		print_rt_rq(m, cpu, rt_rq);
	rcu_read_unlock();
}
2091
#endif /* CONFIG_SCHED_DEBUG */