blk-mq.c 57.3 KB
Newer Older
1 2 3 4 5 6
/*
 * Block multiqueue core code
 *
 * Copyright (C) 2013-2014 Jens Axboe
 * Copyright (C) 2013-2014 Christoph Hellwig
 */
7 8 9 10 11
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/backing-dev.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
12
#include <linux/kmemleak.h>
13 14 15 16 17 18 19 20 21 22 23
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/workqueue.h>
#include <linux/smp.h>
#include <linux/llist.h>
#include <linux/list_sort.h>
#include <linux/cpu.h>
#include <linux/cache.h>
#include <linux/sched/sysctl.h>
#include <linux/delay.h>
24
#include <linux/crash_dump.h>
25
#include <linux/prefetch.h>
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43

#include <trace/events/block.h>

#include <linux/blk-mq.h>
#include "blk.h"
#include "blk-mq.h"
#include "blk-mq-tag.h"

static DEFINE_MUTEX(all_q_mutex);
static LIST_HEAD(all_q_list);

static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx);

/*
 * Check if any of the ctx's have pending work in this hardware queue
 */
static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx)
{
44
	return sbitmap_any_bit_set(&hctx->ctx_map);
45 46
}

47 48 49 50 51 52
/*
 * Mark this ctx as having pending work in this hardware queue
 */
static void blk_mq_hctx_mark_pending(struct blk_mq_hw_ctx *hctx,
				     struct blk_mq_ctx *ctx)
{
53 54
	if (!sbitmap_test_bit(&hctx->ctx_map, ctx->index_hw))
		sbitmap_set_bit(&hctx->ctx_map, ctx->index_hw);
55 56 57 58 59
}

static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx,
				      struct blk_mq_ctx *ctx)
{
60
	sbitmap_clear_bit(&hctx->ctx_map, ctx->index_hw);
61 62
}

63
void blk_mq_freeze_queue_start(struct request_queue *q)
64
{
65
	int freeze_depth;
66

67 68
	freeze_depth = atomic_inc_return(&q->mq_freeze_depth);
	if (freeze_depth == 1) {
69
		percpu_ref_kill(&q->q_usage_counter);
70
		blk_mq_run_hw_queues(q, false);
71
	}
72
}
73
EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_start);
74 75 76

static void blk_mq_freeze_queue_wait(struct request_queue *q)
{
77
	wait_event(q->mq_freeze_wq, percpu_ref_is_zero(&q->q_usage_counter));
78 79
}

80 81 82 83
/*
 * Guarantee no request is in use, so we can change any data structure of
 * the queue afterward.
 */
84
void blk_freeze_queue(struct request_queue *q)
85
{
86 87 88 89 90 91 92
	/*
	 * In the !blk_mq case we are only calling this to kill the
	 * q_usage_counter, otherwise this increases the freeze depth
	 * and waits for it to return to zero.  For this reason there is
	 * no blk_unfreeze_queue(), and blk_freeze_queue() is not
	 * exported to drivers as the only user for unfreeze is blk_mq.
	 */
93 94 95
	blk_mq_freeze_queue_start(q);
	blk_mq_freeze_queue_wait(q);
}
96 97 98 99 100 101 102 103 104

void blk_mq_freeze_queue(struct request_queue *q)
{
	/*
	 * ...just an alias to keep freeze and unfreeze actions balanced
	 * in the blk_mq_* namespace
	 */
	blk_freeze_queue(q);
}
105
EXPORT_SYMBOL_GPL(blk_mq_freeze_queue);
106

107
void blk_mq_unfreeze_queue(struct request_queue *q)
108
{
109
	int freeze_depth;
110

111 112 113
	freeze_depth = atomic_dec_return(&q->mq_freeze_depth);
	WARN_ON_ONCE(freeze_depth < 0);
	if (!freeze_depth) {
114
		percpu_ref_reinit(&q->q_usage_counter);
115
		wake_up_all(&q->mq_freeze_wq);
116
	}
117
}
118
EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue);
119

120 121 122 123 124 125 126 127
void blk_mq_wake_waiters(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	unsigned int i;

	queue_for_each_hw_ctx(q, hctx, i)
		if (blk_mq_hw_queue_mapped(hctx))
			blk_mq_tag_wakeup_all(hctx->tags, true);
128 129 130 131 132 133 134

	/*
	 * If we are called because the queue has now been marked as
	 * dying, we need to ensure that processes currently waiting on
	 * the queue are notified as well.
	 */
	wake_up_all(&q->mq_freeze_wq);
135 136
}

137 138 139 140 141 142
bool blk_mq_can_queue(struct blk_mq_hw_ctx *hctx)
{
	return blk_mq_has_free_tags(hctx->tags);
}
EXPORT_SYMBOL(blk_mq_can_queue);

143
static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
144 145
			       struct request *rq, int op,
			       unsigned int op_flags)
146
{
147
	if (blk_queue_io_stat(q))
148
		op_flags |= REQ_IO_STAT;
149

150 151 152
	INIT_LIST_HEAD(&rq->queuelist);
	/* csd/requeue_work/fifo_time is initialized before use */
	rq->q = q;
153
	rq->mq_ctx = ctx;
154
	req_set_op_attrs(rq, op, op_flags);
155 156 157 158 159 160
	/* do not touch atomic flags, it needs atomic ops against the timer */
	rq->cpu = -1;
	INIT_HLIST_NODE(&rq->hash);
	RB_CLEAR_NODE(&rq->rb_node);
	rq->rq_disk = NULL;
	rq->part = NULL;
161
	rq->start_time = jiffies;
162 163
#ifdef CONFIG_BLK_CGROUP
	rq->rl = NULL;
164
	set_start_time_ns(rq);
165 166 167 168 169 170 171 172 173 174
	rq->io_start_time_ns = 0;
#endif
	rq->nr_phys_segments = 0;
#if defined(CONFIG_BLK_DEV_INTEGRITY)
	rq->nr_integrity_segments = 0;
#endif
	rq->special = NULL;
	/* tag was already set */
	rq->errors = 0;

175 176
	rq->cmd = rq->__cmd;

177 178 179 180 181 182
	rq->extra_len = 0;
	rq->sense_len = 0;
	rq->resid_len = 0;
	rq->sense = NULL;

	INIT_LIST_HEAD(&rq->timeout_list);
183 184
	rq->timeout = 0;

185 186 187 188
	rq->end_io = NULL;
	rq->end_io_data = NULL;
	rq->next_rq = NULL;

189
	ctx->rq_dispatched[rw_is_sync(op, op_flags)]++;
190 191
}

192
static struct request *
193
__blk_mq_alloc_request(struct blk_mq_alloc_data *data, int op, int op_flags)
194 195 196 197
{
	struct request *rq;
	unsigned int tag;

198
	tag = blk_mq_get_tag(data);
199
	if (tag != BLK_MQ_TAG_FAIL) {
200
		rq = data->hctx->tags->rqs[tag];
201

202
		if (blk_mq_tag_busy(data->hctx)) {
203
			rq->cmd_flags = REQ_MQ_INFLIGHT;
204
			atomic_inc(&data->hctx->nr_active);
205 206 207
		}

		rq->tag = tag;
208
		blk_mq_rq_ctx_init(data->q, data->ctx, rq, op, op_flags);
209 210 211 212 213 214
		return rq;
	}

	return NULL;
}

215 216
struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
		unsigned int flags)
217
{
218 219
	struct blk_mq_ctx *ctx;
	struct blk_mq_hw_ctx *hctx;
220
	struct request *rq;
221
	struct blk_mq_alloc_data alloc_data;
222
	int ret;
223

224
	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
225 226
	if (ret)
		return ERR_PTR(ret);
227

228 229
	ctx = blk_mq_get_ctx(q);
	hctx = q->mq_ops->map_queue(q, ctx->cpu);
230
	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
231

232
	rq = __blk_mq_alloc_request(&alloc_data, rw, 0);
233
	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
234 235 236 237 238
		__blk_mq_run_hw_queue(hctx);
		blk_mq_put_ctx(ctx);

		ctx = blk_mq_get_ctx(q);
		hctx = q->mq_ops->map_queue(q, ctx->cpu);
239
		blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
240
		rq =  __blk_mq_alloc_request(&alloc_data, rw, 0);
241
		ctx = alloc_data.ctx;
242 243
	}
	blk_mq_put_ctx(ctx);
K
Keith Busch 已提交
244
	if (!rq) {
245
		blk_queue_exit(q);
246
		return ERR_PTR(-EWOULDBLOCK);
K
Keith Busch 已提交
247
	}
248 249 250 251

	rq->__data_len = 0;
	rq->__sector = (sector_t) -1;
	rq->bio = rq->biotail = NULL;
252 253
	return rq;
}
254
EXPORT_SYMBOL(blk_mq_alloc_request);
255

M
Ming Lin 已提交
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
		unsigned int flags, unsigned int hctx_idx)
{
	struct blk_mq_hw_ctx *hctx;
	struct blk_mq_ctx *ctx;
	struct request *rq;
	struct blk_mq_alloc_data alloc_data;
	int ret;

	/*
	 * If the tag allocator sleeps we could get an allocation for a
	 * different hardware context.  No need to complicate the low level
	 * allocator for this for the rare use case of a command tied to
	 * a specific queue.
	 */
	if (WARN_ON_ONCE(!(flags & BLK_MQ_REQ_NOWAIT)))
		return ERR_PTR(-EINVAL);

	if (hctx_idx >= q->nr_hw_queues)
		return ERR_PTR(-EIO);

	ret = blk_queue_enter(q, true);
	if (ret)
		return ERR_PTR(ret);

	hctx = q->queue_hw_ctx[hctx_idx];
	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));

	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
	rq = __blk_mq_alloc_request(&alloc_data, rw, 0);
	if (!rq) {
		blk_queue_exit(q);
		return ERR_PTR(-EWOULDBLOCK);
	}

	return rq;
}
EXPORT_SYMBOL_GPL(blk_mq_alloc_request_hctx);

295 296 297 298 299 300
static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
				  struct blk_mq_ctx *ctx, struct request *rq)
{
	const int tag = rq->tag;
	struct request_queue *q = rq->q;

301 302
	if (rq->cmd_flags & REQ_MQ_INFLIGHT)
		atomic_dec(&hctx->nr_active);
303
	rq->cmd_flags = 0;
304

305
	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
306
	blk_mq_put_tag(hctx, tag, &ctx->last_tag);
307
	blk_queue_exit(q);
308 309
}

310
void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *hctx, struct request *rq)
311 312 313 314 315
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;

	ctx->rq_completed[rq_is_sync(rq)]++;
	__blk_mq_free_request(hctx, ctx, rq);
316 317 318 319 320 321 322 323 324 325 326

}
EXPORT_SYMBOL_GPL(blk_mq_free_hctx_request);

void blk_mq_free_request(struct request *rq)
{
	struct blk_mq_hw_ctx *hctx;
	struct request_queue *q = rq->q;

	hctx = q->mq_ops->map_queue(q, rq->mq_ctx->cpu);
	blk_mq_free_hctx_request(hctx, rq);
327
}
J
Jens Axboe 已提交
328
EXPORT_SYMBOL_GPL(blk_mq_free_request);
329

330
inline void __blk_mq_end_request(struct request *rq, int error)
331
{
M
Ming Lei 已提交
332 333
	blk_account_io_done(rq);

C
Christoph Hellwig 已提交
334
	if (rq->end_io) {
335
		rq->end_io(rq, error);
C
Christoph Hellwig 已提交
336 337 338
	} else {
		if (unlikely(blk_bidi_rq(rq)))
			blk_mq_free_request(rq->next_rq);
339
		blk_mq_free_request(rq);
C
Christoph Hellwig 已提交
340
	}
341
}
342
EXPORT_SYMBOL(__blk_mq_end_request);
343

344
void blk_mq_end_request(struct request *rq, int error)
345 346 347
{
	if (blk_update_request(rq, error, blk_rq_bytes(rq)))
		BUG();
348
	__blk_mq_end_request(rq, error);
349
}
350
EXPORT_SYMBOL(blk_mq_end_request);
351

352
static void __blk_mq_complete_request_remote(void *data)
353
{
354
	struct request *rq = data;
355

356
	rq->q->softirq_done_fn(rq);
357 358
}

359
static void blk_mq_ipi_complete_request(struct request *rq)
360 361
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;
C
Christoph Hellwig 已提交
362
	bool shared = false;
363 364
	int cpu;

C
Christoph Hellwig 已提交
365
	if (!test_bit(QUEUE_FLAG_SAME_COMP, &rq->q->queue_flags)) {
366 367 368
		rq->q->softirq_done_fn(rq);
		return;
	}
369 370

	cpu = get_cpu();
C
Christoph Hellwig 已提交
371 372 373 374
	if (!test_bit(QUEUE_FLAG_SAME_FORCE, &rq->q->queue_flags))
		shared = cpus_share_cache(cpu, ctx->cpu);

	if (cpu != ctx->cpu && !shared && cpu_online(ctx->cpu)) {
375
		rq->csd.func = __blk_mq_complete_request_remote;
376 377
		rq->csd.info = rq;
		rq->csd.flags = 0;
378
		smp_call_function_single_async(ctx->cpu, &rq->csd);
379
	} else {
380
		rq->q->softirq_done_fn(rq);
381
	}
382 383
	put_cpu();
}
384

385
static void __blk_mq_complete_request(struct request *rq)
386 387 388 389
{
	struct request_queue *q = rq->q;

	if (!q->softirq_done_fn)
390
		blk_mq_end_request(rq, rq->errors);
391 392 393 394
	else
		blk_mq_ipi_complete_request(rq);
}

395 396 397 398 399 400 401 402
/**
 * blk_mq_complete_request - end I/O on a request
 * @rq:		the request being processed
 *
 * Description:
 *	Ends all I/O on a request. It does not handle partial completions.
 *	The actual completion happens out-of-order, through a IPI handler.
 **/
403
void blk_mq_complete_request(struct request *rq, int error)
404
{
405 406 407
	struct request_queue *q = rq->q;

	if (unlikely(blk_should_fake_timeout(q)))
408
		return;
409 410
	if (!blk_mark_rq_complete(rq)) {
		rq->errors = error;
411
		__blk_mq_complete_request(rq);
412
	}
413 414
}
EXPORT_SYMBOL(blk_mq_complete_request);
415

416 417 418 419 420 421
int blk_mq_request_started(struct request *rq)
{
	return test_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
}
EXPORT_SYMBOL_GPL(blk_mq_request_started);

422
void blk_mq_start_request(struct request *rq)
423 424 425 426 427
{
	struct request_queue *q = rq->q;

	trace_block_rq_issue(q, rq);

C
Christoph Hellwig 已提交
428
	rq->resid_len = blk_rq_bytes(rq);
C
Christoph Hellwig 已提交
429 430
	if (unlikely(blk_bidi_rq(rq)))
		rq->next_rq->resid_len = blk_rq_bytes(rq->next_rq);
C
Christoph Hellwig 已提交
431

432
	blk_add_timer(rq);
433

434 435 436 437 438 439
	/*
	 * Ensure that ->deadline is visible before set the started
	 * flag and clear the completed flag.
	 */
	smp_mb__before_atomic();

440 441 442 443 444 445
	/*
	 * Mark us as started and clear complete. Complete might have been
	 * set if requeue raced with timeout, which then marked it as
	 * complete. So be sure to clear complete again when we start
	 * the request, otherwise we'll ignore the completion event.
	 */
446 447 448 449
	if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
		set_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
	if (test_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags))
		clear_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
450 451 452 453 454 455 456 457 458

	if (q->dma_drain_size && blk_rq_bytes(rq)) {
		/*
		 * Make sure space for the drain appears.  We know we can do
		 * this because max_hw_segments has been adjusted to be one
		 * fewer than the device can handle.
		 */
		rq->nr_phys_segments++;
	}
459
}
460
EXPORT_SYMBOL(blk_mq_start_request);
461

462
static void __blk_mq_requeue_request(struct request *rq)
463 464 465 466
{
	struct request_queue *q = rq->q;

	trace_block_rq_requeue(q, rq);
467

468 469 470 471
	if (test_and_clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags)) {
		if (q->dma_drain_size && blk_rq_bytes(rq))
			rq->nr_phys_segments--;
	}
472 473
}

474 475 476 477 478
void blk_mq_requeue_request(struct request *rq)
{
	__blk_mq_requeue_request(rq);

	BUG_ON(blk_queued_rq(rq));
479
	blk_mq_add_to_requeue_list(rq, true);
480 481 482
}
EXPORT_SYMBOL(blk_mq_requeue_request);

483 484 485
static void blk_mq_requeue_work(struct work_struct *work)
{
	struct request_queue *q =
486
		container_of(work, struct request_queue, requeue_work.work);
487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509
	LIST_HEAD(rq_list);
	struct request *rq, *next;
	unsigned long flags;

	spin_lock_irqsave(&q->requeue_lock, flags);
	list_splice_init(&q->requeue_list, &rq_list);
	spin_unlock_irqrestore(&q->requeue_lock, flags);

	list_for_each_entry_safe(rq, next, &rq_list, queuelist) {
		if (!(rq->cmd_flags & REQ_SOFTBARRIER))
			continue;

		rq->cmd_flags &= ~REQ_SOFTBARRIER;
		list_del_init(&rq->queuelist);
		blk_mq_insert_request(rq, true, false, false);
	}

	while (!list_empty(&rq_list)) {
		rq = list_entry(rq_list.next, struct request, queuelist);
		list_del_init(&rq->queuelist);
		blk_mq_insert_request(rq, false, false, false);
	}

510 511 512 513 514
	/*
	 * Use the start variant of queue running here, so that running
	 * the requeue work will kick stopped queues.
	 */
	blk_mq_start_hw_queues(q);
515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
}

void blk_mq_add_to_requeue_list(struct request *rq, bool at_head)
{
	struct request_queue *q = rq->q;
	unsigned long flags;

	/*
	 * We abuse this flag that is otherwise used by the I/O scheduler to
	 * request head insertation from the workqueue.
	 */
	BUG_ON(rq->cmd_flags & REQ_SOFTBARRIER);

	spin_lock_irqsave(&q->requeue_lock, flags);
	if (at_head) {
		rq->cmd_flags |= REQ_SOFTBARRIER;
		list_add(&rq->queuelist, &q->requeue_list);
	} else {
		list_add_tail(&rq->queuelist, &q->requeue_list);
	}
	spin_unlock_irqrestore(&q->requeue_lock, flags);
}
EXPORT_SYMBOL(blk_mq_add_to_requeue_list);

539 540
void blk_mq_cancel_requeue_work(struct request_queue *q)
{
541
	cancel_delayed_work_sync(&q->requeue_work);
542 543 544
}
EXPORT_SYMBOL_GPL(blk_mq_cancel_requeue_work);

545 546
void blk_mq_kick_requeue_list(struct request_queue *q)
{
547
	kblockd_schedule_delayed_work(&q->requeue_work, 0);
548 549 550
}
EXPORT_SYMBOL(blk_mq_kick_requeue_list);

551 552 553 554 555 556 557 558
void blk_mq_delay_kick_requeue_list(struct request_queue *q,
				    unsigned long msecs)
{
	kblockd_schedule_delayed_work(&q->requeue_work,
				      msecs_to_jiffies(msecs));
}
EXPORT_SYMBOL(blk_mq_delay_kick_requeue_list);

559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578
void blk_mq_abort_requeue_list(struct request_queue *q)
{
	unsigned long flags;
	LIST_HEAD(rq_list);

	spin_lock_irqsave(&q->requeue_lock, flags);
	list_splice_init(&q->requeue_list, &rq_list);
	spin_unlock_irqrestore(&q->requeue_lock, flags);

	while (!list_empty(&rq_list)) {
		struct request *rq;

		rq = list_first_entry(&rq_list, struct request, queuelist);
		list_del_init(&rq->queuelist);
		rq->errors = -EIO;
		blk_mq_end_request(rq, rq->errors);
	}
}
EXPORT_SYMBOL(blk_mq_abort_requeue_list);

579 580
struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag)
{
581 582
	if (tag < tags->nr_tags) {
		prefetch(tags->rqs[tag]);
583
		return tags->rqs[tag];
584
	}
585 586

	return NULL;
587 588 589
}
EXPORT_SYMBOL(blk_mq_tag_to_rq);

590
struct blk_mq_timeout_data {
591 592
	unsigned long next;
	unsigned int next_set;
593 594
};

595
void blk_mq_rq_timed_out(struct request *req, bool reserved)
596
{
597 598
	struct blk_mq_ops *ops = req->q->mq_ops;
	enum blk_eh_timer_return ret = BLK_EH_RESET_TIMER;
599 600 601 602 603 604 605 606 607 608

	/*
	 * We know that complete is set at this point. If STARTED isn't set
	 * anymore, then the request isn't active and the "timeout" should
	 * just be ignored. This can happen due to the bitflag ordering.
	 * Timeout first checks if STARTED is set, and if it is, assumes
	 * the request is active. But if we race with completion, then
	 * we both flags will get cleared. So check here again, and ignore
	 * a timeout event with a request that isn't active.
	 */
609 610
	if (!test_bit(REQ_ATOM_STARTED, &req->atomic_flags))
		return;
611

612
	if (ops->timeout)
613
		ret = ops->timeout(req, reserved);
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628

	switch (ret) {
	case BLK_EH_HANDLED:
		__blk_mq_complete_request(req);
		break;
	case BLK_EH_RESET_TIMER:
		blk_add_timer(req);
		blk_clear_rq_complete(req);
		break;
	case BLK_EH_NOT_HANDLED:
		break;
	default:
		printk(KERN_ERR "block: bad eh return: %d\n", ret);
		break;
	}
629
}
630

631 632 633 634
static void blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
		struct request *rq, void *priv, bool reserved)
{
	struct blk_mq_timeout_data *data = priv;
635

636 637 638 639 640
	if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags)) {
		/*
		 * If a request wasn't started before the queue was
		 * marked dying, kill it here or it'll go unnoticed.
		 */
641 642 643 644
		if (unlikely(blk_queue_dying(rq->q))) {
			rq->errors = -EIO;
			blk_mq_end_request(rq, rq->errors);
		}
645
		return;
646
	}
647

648 649
	if (time_after_eq(jiffies, rq->deadline)) {
		if (!blk_mark_rq_complete(rq))
650
			blk_mq_rq_timed_out(rq, reserved);
651 652 653 654
	} else if (!data->next_set || time_after(data->next, rq->deadline)) {
		data->next = rq->deadline;
		data->next_set = 1;
	}
655 656
}

657
static void blk_mq_timeout_work(struct work_struct *work)
658
{
659 660
	struct request_queue *q =
		container_of(work, struct request_queue, timeout_work);
661 662 663 664 665
	struct blk_mq_timeout_data data = {
		.next		= 0,
		.next_set	= 0,
	};
	int i;
666

667 668 669 670 671 672 673 674 675 676 677 678 679 680
	/* A deadlock might occur if a request is stuck requiring a
	 * timeout at the same time a queue freeze is waiting
	 * completion, since the timeout code would not be able to
	 * acquire the queue reference here.
	 *
	 * That's why we don't use blk_queue_enter here; instead, we use
	 * percpu_ref_tryget directly, because we need to be able to
	 * obtain a reference even in the short window between the queue
	 * starting to freeze, by dropping the first reference in
	 * blk_mq_freeze_queue_start, and the moment the last request is
	 * consumed, marked by the instant q_usage_counter reaches
	 * zero.
	 */
	if (!percpu_ref_tryget(&q->q_usage_counter))
681 682
		return;

683
	blk_mq_queue_tag_busy_iter(q, blk_mq_check_expired, &data);
684

685 686 687
	if (data.next_set) {
		data.next = blk_rq_timeout(round_jiffies_up(data.next));
		mod_timer(&q->timeout, data.next);
688
	} else {
689 690
		struct blk_mq_hw_ctx *hctx;

691 692 693 694 695
		queue_for_each_hw_ctx(q, hctx, i) {
			/* the hctx may be unmapped, so check it here */
			if (blk_mq_hw_queue_mapped(hctx))
				blk_mq_tag_idle(hctx);
		}
696
	}
697
	blk_queue_exit(q);
698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738
}

/*
 * Reverse check our software queue for entries that we could potentially
 * merge with. Currently includes a hand-wavy stop count of 8, to not spend
 * too much time checking for merges.
 */
static bool blk_mq_attempt_merge(struct request_queue *q,
				 struct blk_mq_ctx *ctx, struct bio *bio)
{
	struct request *rq;
	int checked = 8;

	list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) {
		int el_ret;

		if (!checked--)
			break;

		if (!blk_rq_merge_ok(rq, bio))
			continue;

		el_ret = blk_try_merge(rq, bio);
		if (el_ret == ELEVATOR_BACK_MERGE) {
			if (bio_attempt_back_merge(q, rq, bio)) {
				ctx->rq_merged++;
				return true;
			}
			break;
		} else if (el_ret == ELEVATOR_FRONT_MERGE) {
			if (bio_attempt_front_merge(q, rq, bio)) {
				ctx->rq_merged++;
				return true;
			}
			break;
		}
	}

	return false;
}

739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
struct flush_busy_ctx_data {
	struct blk_mq_hw_ctx *hctx;
	struct list_head *list;
};

static bool flush_busy_ctx(struct sbitmap *sb, unsigned int bitnr, void *data)
{
	struct flush_busy_ctx_data *flush_data = data;
	struct blk_mq_hw_ctx *hctx = flush_data->hctx;
	struct blk_mq_ctx *ctx = hctx->ctxs[bitnr];

	sbitmap_clear_bit(sb, bitnr);
	spin_lock(&ctx->lock);
	list_splice_tail_init(&ctx->rq_list, flush_data->list);
	spin_unlock(&ctx->lock);
	return true;
}

757 758 759 760 761 762
/*
 * Process software queues that have been marked busy, splicing them
 * to the for-dispatch
 */
static void flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list)
{
763 764 765 766
	struct flush_busy_ctx_data data = {
		.hctx = hctx,
		.list = list,
	};
767

768
	sbitmap_for_each_set(&hctx->ctx_map, flush_busy_ctx, &data);
769 770
}

771 772 773 774 775 776 777 778
static inline unsigned int queued_to_index(unsigned int queued)
{
	if (!queued)
		return 0;

	return min(BLK_MQ_MAX_DISPATCH_ORDER - 1, ilog2(queued) + 1);
}

779 780 781 782 783 784 785 786 787 788 789
/*
 * Run this hardware queue, pulling any software queues mapped to it in.
 * Note that this function currently has various problems around ordering
 * of IO. In particular, we'd like FIFO behaviour on handling existing
 * items on the hctx->dispatch list. Ignore that for now.
 */
static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
{
	struct request_queue *q = hctx->queue;
	struct request *rq;
	LIST_HEAD(rq_list);
790 791
	LIST_HEAD(driver_list);
	struct list_head *dptr;
792
	int queued;
793

794
	if (unlikely(test_bit(BLK_MQ_S_STOPPED, &hctx->state)))
795 796
		return;

797 798 799
	WARN_ON(!cpumask_test_cpu(raw_smp_processor_id(), hctx->cpumask) &&
		cpu_online(hctx->next_cpu));

800 801 802 803 804
	hctx->run++;

	/*
	 * Touch any software queue that has pending entries.
	 */
805
	flush_busy_ctxs(hctx, &rq_list);
806 807 808 809 810 811 812 813 814 815 816 817

	/*
	 * If we have previous entries on our dispatch list, grab them
	 * and stuff them at the front for more fair dispatch.
	 */
	if (!list_empty_careful(&hctx->dispatch)) {
		spin_lock(&hctx->lock);
		if (!list_empty(&hctx->dispatch))
			list_splice_init(&hctx->dispatch, &rq_list);
		spin_unlock(&hctx->lock);
	}

818 819 820 821 822 823
	/*
	 * Start off with dptr being NULL, so we start the first request
	 * immediately, even if we have more pending.
	 */
	dptr = NULL;

824 825 826
	/*
	 * Now process all the entries, sending them to the driver.
	 */
827
	queued = 0;
828
	while (!list_empty(&rq_list)) {
829
		struct blk_mq_queue_data bd;
830 831 832 833 834
		int ret;

		rq = list_first_entry(&rq_list, struct request, queuelist);
		list_del_init(&rq->queuelist);

835 836 837 838 839
		bd.rq = rq;
		bd.list = dptr;
		bd.last = list_empty(&rq_list);

		ret = q->mq_ops->queue_rq(hctx, &bd);
840 841 842
		switch (ret) {
		case BLK_MQ_RQ_QUEUE_OK:
			queued++;
843
			break;
844 845
		case BLK_MQ_RQ_QUEUE_BUSY:
			list_add(&rq->queuelist, &rq_list);
846
			__blk_mq_requeue_request(rq);
847 848 849 850
			break;
		default:
			pr_err("blk-mq: bad return on queue: %d\n", ret);
		case BLK_MQ_RQ_QUEUE_ERROR:
851
			rq->errors = -EIO;
852
			blk_mq_end_request(rq, rq->errors);
853 854 855 856 857
			break;
		}

		if (ret == BLK_MQ_RQ_QUEUE_BUSY)
			break;
858 859 860 861 862 863 864

		/*
		 * We've done the first request. If we have more than 1
		 * left in the list, set dptr to defer issue.
		 */
		if (!dptr && rq_list.next != rq_list.prev)
			dptr = &driver_list;
865 866
	}

867
	hctx->dispatched[queued_to_index(queued)]++;
868 869 870 871 872 873 874 875 876

	/*
	 * Any items that need requeuing? Stuff them into hctx->dispatch,
	 * that is where we will continue on next queue run.
	 */
	if (!list_empty(&rq_list)) {
		spin_lock(&hctx->lock);
		list_splice(&rq_list, &hctx->dispatch);
		spin_unlock(&hctx->lock);
877 878 879 880 881 882 883 884 885 886
		/*
		 * the queue is expected stopped with BLK_MQ_RQ_QUEUE_BUSY, but
		 * it's possible the queue is stopped and restarted again
		 * before this. Queue restart will dispatch requests. And since
		 * requests in rq_list aren't added into hctx->dispatch yet,
		 * the requests in rq_list might get lost.
		 *
		 * blk_mq_run_hw_queue() already checks the STOPPED bit
		 **/
		blk_mq_run_hw_queue(hctx, true);
887 888 889
	}
}

890 891 892 893 894 895 896 897
/*
 * It'd be great if the workqueue API had a way to pass
 * in a mask and had some smarts for more clever placement.
 * For now we just round-robin here, switching for every
 * BLK_MQ_CPU_WORK_BATCH queued items.
 */
static int blk_mq_hctx_next_cpu(struct blk_mq_hw_ctx *hctx)
{
898 899
	if (hctx->queue->nr_hw_queues == 1)
		return WORK_CPU_UNBOUND;
900 901

	if (--hctx->next_cpu_batch <= 0) {
902
		int cpu = hctx->next_cpu, next_cpu;
903 904 905 906 907 908 909

		next_cpu = cpumask_next(hctx->next_cpu, hctx->cpumask);
		if (next_cpu >= nr_cpu_ids)
			next_cpu = cpumask_first(hctx->cpumask);

		hctx->next_cpu = next_cpu;
		hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH;
910 911

		return cpu;
912 913
	}

914
	return hctx->next_cpu;
915 916
}

917 918
void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async)
{
919 920
	if (unlikely(test_bit(BLK_MQ_S_STOPPED, &hctx->state) ||
	    !blk_mq_hw_queue_mapped(hctx)))
921 922
		return;

923
	if (!async) {
924 925
		int cpu = get_cpu();
		if (cpumask_test_cpu(cpu, hctx->cpumask)) {
926
			__blk_mq_run_hw_queue(hctx);
927
			put_cpu();
928 929
			return;
		}
930

931
		put_cpu();
932
	}
933

934
	kblockd_schedule_work_on(blk_mq_hctx_next_cpu(hctx), &hctx->run_work);
935 936
}

937
void blk_mq_run_hw_queues(struct request_queue *q, bool async)
938 939 940 941 942 943 944
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if ((!blk_mq_hctx_has_pending(hctx) &&
		    list_empty_careful(&hctx->dispatch)) ||
945
		    test_bit(BLK_MQ_S_STOPPED, &hctx->state))
946 947
			continue;

948
		blk_mq_run_hw_queue(hctx, async);
949 950
	}
}
951
EXPORT_SYMBOL(blk_mq_run_hw_queues);
952 953 954

void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx)
{
955
	cancel_work(&hctx->run_work);
956
	cancel_delayed_work(&hctx->delay_work);
957 958 959 960
	set_bit(BLK_MQ_S_STOPPED, &hctx->state);
}
EXPORT_SYMBOL(blk_mq_stop_hw_queue);

961 962 963 964 965 966 967 968 969 970
void blk_mq_stop_hw_queues(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i)
		blk_mq_stop_hw_queue(hctx);
}
EXPORT_SYMBOL(blk_mq_stop_hw_queues);

971 972 973
void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx)
{
	clear_bit(BLK_MQ_S_STOPPED, &hctx->state);
974

975
	blk_mq_run_hw_queue(hctx, false);
976 977 978
}
EXPORT_SYMBOL(blk_mq_start_hw_queue);

979 980 981 982 983 984 985 986 987 988
void blk_mq_start_hw_queues(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i)
		blk_mq_start_hw_queue(hctx);
}
EXPORT_SYMBOL(blk_mq_start_hw_queues);

989
void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async)
990 991 992 993 994 995 996 997 998
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if (!test_bit(BLK_MQ_S_STOPPED, &hctx->state))
			continue;

		clear_bit(BLK_MQ_S_STOPPED, &hctx->state);
999
		blk_mq_run_hw_queue(hctx, async);
1000 1001 1002 1003
	}
}
EXPORT_SYMBOL(blk_mq_start_stopped_hw_queues);

1004
static void blk_mq_run_work_fn(struct work_struct *work)
1005 1006 1007
{
	struct blk_mq_hw_ctx *hctx;

1008
	hctx = container_of(work, struct blk_mq_hw_ctx, run_work);
1009

1010 1011 1012
	__blk_mq_run_hw_queue(hctx);
}

1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
static void blk_mq_delay_work_fn(struct work_struct *work)
{
	struct blk_mq_hw_ctx *hctx;

	hctx = container_of(work, struct blk_mq_hw_ctx, delay_work.work);

	if (test_and_clear_bit(BLK_MQ_S_STOPPED, &hctx->state))
		__blk_mq_run_hw_queue(hctx);
}

void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs)
{
1025 1026
	if (unlikely(!blk_mq_hw_queue_mapped(hctx)))
		return;
1027

1028 1029
	kblockd_schedule_delayed_work_on(blk_mq_hctx_next_cpu(hctx),
			&hctx->delay_work, msecs_to_jiffies(msecs));
1030 1031 1032
}
EXPORT_SYMBOL(blk_mq_delay_queue);

1033 1034 1035
static inline void __blk_mq_insert_req_list(struct blk_mq_hw_ctx *hctx,
					    struct request *rq,
					    bool at_head)
1036
{
J
Jens Axboe 已提交
1037 1038
	struct blk_mq_ctx *ctx = rq->mq_ctx;

1039 1040
	trace_block_rq_insert(hctx->queue, rq);

1041 1042 1043 1044
	if (at_head)
		list_add(&rq->queuelist, &ctx->rq_list);
	else
		list_add_tail(&rq->queuelist, &ctx->rq_list);
1045
}
1046

1047 1048 1049 1050 1051
static void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx,
				    struct request *rq, bool at_head)
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;

J
Jens Axboe 已提交
1052
	__blk_mq_insert_req_list(hctx, rq, at_head);
1053 1054 1055
	blk_mq_hctx_mark_pending(hctx, ctx);
}

1056
void blk_mq_insert_request(struct request *rq, bool at_head, bool run_queue,
J
Jens Axboe 已提交
1057
			   bool async)
1058
{
J
Jens Axboe 已提交
1059
	struct blk_mq_ctx *ctx = rq->mq_ctx;
1060
	struct request_queue *q = rq->q;
1061 1062 1063 1064
	struct blk_mq_hw_ctx *hctx;

	hctx = q->mq_ops->map_queue(q, ctx->cpu);

1065 1066 1067
	spin_lock(&ctx->lock);
	__blk_mq_insert_request(hctx, rq, at_head);
	spin_unlock(&ctx->lock);
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094

	if (run_queue)
		blk_mq_run_hw_queue(hctx, async);
}

static void blk_mq_insert_requests(struct request_queue *q,
				     struct blk_mq_ctx *ctx,
				     struct list_head *list,
				     int depth,
				     bool from_schedule)

{
	struct blk_mq_hw_ctx *hctx;

	trace_block_unplug(q, depth, !from_schedule);

	hctx = q->mq_ops->map_queue(q, ctx->cpu);

	/*
	 * preemption doesn't flush plug list, so it's possible ctx->cpu is
	 * offline now
	 */
	spin_lock(&ctx->lock);
	while (!list_empty(list)) {
		struct request *rq;

		rq = list_first_entry(list, struct request, queuelist);
J
Jens Axboe 已提交
1095
		BUG_ON(rq->mq_ctx != ctx);
1096
		list_del_init(&rq->queuelist);
J
Jens Axboe 已提交
1097
		__blk_mq_insert_req_list(hctx, rq, false);
1098
	}
1099
	blk_mq_hctx_mark_pending(hctx, ctx);
1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
	spin_unlock(&ctx->lock);

	blk_mq_run_hw_queue(hctx, from_schedule);
}

static int plug_ctx_cmp(void *priv, struct list_head *a, struct list_head *b)
{
	struct request *rqa = container_of(a, struct request, queuelist);
	struct request *rqb = container_of(b, struct request, queuelist);

	return !(rqa->mq_ctx < rqb->mq_ctx ||
		 (rqa->mq_ctx == rqb->mq_ctx &&
		  blk_rq_pos(rqa) < blk_rq_pos(rqb)));
}

void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
{
	struct blk_mq_ctx *this_ctx;
	struct request_queue *this_q;
	struct request *rq;
	LIST_HEAD(list);
	LIST_HEAD(ctx_list);
	unsigned int depth;

	list_splice_init(&plug->mq_list, &list);

	list_sort(NULL, &list, plug_ctx_cmp);

	this_q = NULL;
	this_ctx = NULL;
	depth = 0;

	while (!list_empty(&list)) {
		rq = list_entry_rq(list.next);
		list_del_init(&rq->queuelist);
		BUG_ON(!rq->q);
		if (rq->mq_ctx != this_ctx) {
			if (this_ctx) {
				blk_mq_insert_requests(this_q, this_ctx,
							&ctx_list, depth,
							from_schedule);
			}

			this_ctx = rq->mq_ctx;
			this_q = rq->q;
			depth = 0;
		}

		depth++;
		list_add_tail(&rq->queuelist, &ctx_list);
	}

	/*
	 * If 'this_ctx' is set, we know we have entries to complete
	 * on 'ctx_list'. Do those.
	 */
	if (this_ctx) {
		blk_mq_insert_requests(this_q, this_ctx, &ctx_list, depth,
				       from_schedule);
	}
}

static void blk_mq_bio_to_request(struct request *rq, struct bio *bio)
{
	init_request_from_bio(rq, bio);
1165

1166
	blk_account_io_start(rq, 1);
1167 1168
}

1169 1170 1171 1172 1173 1174
static inline bool hctx_allow_merges(struct blk_mq_hw_ctx *hctx)
{
	return (hctx->flags & BLK_MQ_F_SHOULD_MERGE) &&
		!blk_queue_nomerges(hctx->queue);
}

1175 1176 1177
static inline bool blk_mq_merge_queue_io(struct blk_mq_hw_ctx *hctx,
					 struct blk_mq_ctx *ctx,
					 struct request *rq, struct bio *bio)
1178
{
1179
	if (!hctx_allow_merges(hctx) || !bio_mergeable(bio)) {
1180 1181 1182 1183 1184 1185 1186
		blk_mq_bio_to_request(rq, bio);
		spin_lock(&ctx->lock);
insert_rq:
		__blk_mq_insert_request(hctx, rq, false);
		spin_unlock(&ctx->lock);
		return false;
	} else {
1187 1188
		struct request_queue *q = hctx->queue;

1189 1190 1191 1192 1193
		spin_lock(&ctx->lock);
		if (!blk_mq_attempt_merge(q, ctx, bio)) {
			blk_mq_bio_to_request(rq, bio);
			goto insert_rq;
		}
1194

1195 1196 1197
		spin_unlock(&ctx->lock);
		__blk_mq_free_request(hctx, ctx, rq);
		return true;
1198
	}
1199
}
1200

1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212
struct blk_map_ctx {
	struct blk_mq_hw_ctx *hctx;
	struct blk_mq_ctx *ctx;
};

static struct request *blk_mq_map_request(struct request_queue *q,
					  struct bio *bio,
					  struct blk_map_ctx *data)
{
	struct blk_mq_hw_ctx *hctx;
	struct blk_mq_ctx *ctx;
	struct request *rq;
1213 1214
	int op = bio_data_dir(bio);
	int op_flags = 0;
1215
	struct blk_mq_alloc_data alloc_data;
1216

1217
	blk_queue_enter_live(q);
1218 1219 1220
	ctx = blk_mq_get_ctx(q);
	hctx = q->mq_ops->map_queue(q, ctx->cpu);

J
Jens Axboe 已提交
1221
	if (rw_is_sync(bio_op(bio), bio->bi_opf))
1222
		op_flags |= REQ_SYNC;
1223

1224
	trace_block_getrq(q, bio, op);
1225
	blk_mq_set_alloc_data(&alloc_data, q, BLK_MQ_REQ_NOWAIT, ctx, hctx);
1226
	rq = __blk_mq_alloc_request(&alloc_data, op, op_flags);
1227
	if (unlikely(!rq)) {
1228
		__blk_mq_run_hw_queue(hctx);
1229
		blk_mq_put_ctx(ctx);
1230
		trace_block_sleeprq(q, bio, op);
1231 1232

		ctx = blk_mq_get_ctx(q);
1233
		hctx = q->mq_ops->map_queue(q, ctx->cpu);
1234
		blk_mq_set_alloc_data(&alloc_data, q, 0, ctx, hctx);
1235
		rq = __blk_mq_alloc_request(&alloc_data, op, op_flags);
1236 1237
		ctx = alloc_data.ctx;
		hctx = alloc_data.hctx;
1238 1239 1240
	}

	hctx->queued++;
1241 1242 1243 1244 1245
	data->hctx = hctx;
	data->ctx = ctx;
	return rq;
}

1246
static int blk_mq_direct_issue_request(struct request *rq, blk_qc_t *cookie)
1247 1248 1249 1250 1251 1252 1253 1254 1255 1256
{
	int ret;
	struct request_queue *q = rq->q;
	struct blk_mq_hw_ctx *hctx = q->mq_ops->map_queue(q,
			rq->mq_ctx->cpu);
	struct blk_mq_queue_data bd = {
		.rq = rq,
		.list = NULL,
		.last = 1
	};
1257
	blk_qc_t new_cookie = blk_tag_to_qc_t(rq->tag, hctx->queue_num);
1258 1259 1260 1261 1262 1263 1264

	/*
	 * For OK queue, we are done. For error, kill it. Any other
	 * error (busy), just add it to our list as we previously
	 * would have done
	 */
	ret = q->mq_ops->queue_rq(hctx, &bd);
1265 1266
	if (ret == BLK_MQ_RQ_QUEUE_OK) {
		*cookie = new_cookie;
1267
		return 0;
1268
	}
1269

1270 1271 1272 1273 1274 1275 1276
	__blk_mq_requeue_request(rq);

	if (ret == BLK_MQ_RQ_QUEUE_ERROR) {
		*cookie = BLK_QC_T_NONE;
		rq->errors = -EIO;
		blk_mq_end_request(rq, rq->errors);
		return 0;
1277
	}
1278 1279

	return -1;
1280 1281
}

1282 1283 1284 1285 1286
/*
 * Multiple hardware queue variant. This will not use per-process plugs,
 * but will attempt to bypass the hctx queueing if we can go straight to
 * hardware for SYNC IO.
 */
1287
static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
1288
{
J
Jens Axboe 已提交
1289 1290
	const int is_sync = rw_is_sync(bio_op(bio), bio->bi_opf);
	const int is_flush_fua = bio->bi_opf & (REQ_PREFLUSH | REQ_FUA);
1291 1292
	struct blk_map_ctx data;
	struct request *rq;
1293 1294
	unsigned int request_count = 0;
	struct blk_plug *plug;
1295
	struct request *same_queue_rq = NULL;
1296
	blk_qc_t cookie;
1297 1298 1299 1300

	blk_queue_bounce(q, &bio);

	if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1301
		bio_io_error(bio);
1302
		return BLK_QC_T_NONE;
1303 1304
	}

1305 1306
	blk_queue_split(q, &bio, q->bio_split);

1307 1308 1309
	if (!is_flush_fua && !blk_queue_nomerges(q) &&
	    blk_attempt_plug_merge(q, bio, &request_count, &same_queue_rq))
		return BLK_QC_T_NONE;
1310

1311 1312
	rq = blk_mq_map_request(q, bio, &data);
	if (unlikely(!rq))
1313
		return BLK_QC_T_NONE;
1314

1315
	cookie = blk_tag_to_qc_t(rq->tag, data.hctx->queue_num);
1316 1317 1318 1319 1320 1321 1322

	if (unlikely(is_flush_fua)) {
		blk_mq_bio_to_request(rq, bio);
		blk_insert_flush(rq);
		goto run_queue;
	}

1323
	plug = current->plug;
1324 1325 1326 1327 1328
	/*
	 * If the driver supports defer issued based on 'last', then
	 * queue it up like normal since we can potentially save some
	 * CPU this way.
	 */
1329 1330 1331
	if (((plug && !blk_queue_nomerges(q)) || is_sync) &&
	    !(data.hctx->flags & BLK_MQ_F_DEFER_ISSUE)) {
		struct request *old_rq = NULL;
1332 1333 1334 1335

		blk_mq_bio_to_request(rq, bio);

		/*
1336
		 * We do limited pluging. If the bio can be merged, do that.
1337 1338
		 * Otherwise the existing request in the plug list will be
		 * issued. So the plug list will have one request at most
1339
		 */
1340
		if (plug) {
1341 1342
			/*
			 * The plug list might get flushed before this. If that
1343 1344 1345
			 * happens, same_queue_rq is invalid and plug list is
			 * empty
			 */
1346 1347
			if (same_queue_rq && !list_empty(&plug->mq_list)) {
				old_rq = same_queue_rq;
1348
				list_del_init(&old_rq->queuelist);
1349
			}
1350 1351 1352 1353 1354
			list_add_tail(&rq->queuelist, &plug->mq_list);
		} else /* is_sync */
			old_rq = rq;
		blk_mq_put_ctx(data.ctx);
		if (!old_rq)
1355 1356 1357
			goto done;
		if (!blk_mq_direct_issue_request(old_rq, &cookie))
			goto done;
1358
		blk_mq_insert_request(old_rq, false, true, true);
1359
		goto done;
1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372
	}

	if (!blk_mq_merge_queue_io(data.hctx, data.ctx, rq, bio)) {
		/*
		 * For a SYNC request, send it to the hardware immediately. For
		 * an ASYNC request, just ensure that we run it later on. The
		 * latter allows for merging opportunities and more efficient
		 * dispatching.
		 */
run_queue:
		blk_mq_run_hw_queue(data.hctx, !is_sync || is_flush_fua);
	}
	blk_mq_put_ctx(data.ctx);
1373 1374
done:
	return cookie;
1375 1376 1377 1378 1379 1380
}

/*
 * Single hardware queue variant. This will attempt to use any per-process
 * plug for merging and IO deferral.
 */
1381
static blk_qc_t blk_sq_make_request(struct request_queue *q, struct bio *bio)
1382
{
J
Jens Axboe 已提交
1383 1384
	const int is_sync = rw_is_sync(bio_op(bio), bio->bi_opf);
	const int is_flush_fua = bio->bi_opf & (REQ_PREFLUSH | REQ_FUA);
1385 1386
	struct blk_plug *plug;
	unsigned int request_count = 0;
1387 1388
	struct blk_map_ctx data;
	struct request *rq;
1389
	blk_qc_t cookie;
1390 1391 1392 1393

	blk_queue_bounce(q, &bio);

	if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1394
		bio_io_error(bio);
1395
		return BLK_QC_T_NONE;
1396 1397
	}

1398 1399
	blk_queue_split(q, &bio, q->bio_split);

1400 1401 1402 1403 1404
	if (!is_flush_fua && !blk_queue_nomerges(q)) {
		if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
			return BLK_QC_T_NONE;
	} else
		request_count = blk_plug_queued_count(q);
1405 1406

	rq = blk_mq_map_request(q, bio, &data);
1407
	if (unlikely(!rq))
1408
		return BLK_QC_T_NONE;
1409

1410
	cookie = blk_tag_to_qc_t(rq->tag, data.hctx->queue_num);
1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422

	if (unlikely(is_flush_fua)) {
		blk_mq_bio_to_request(rq, bio);
		blk_insert_flush(rq);
		goto run_queue;
	}

	/*
	 * A task plug currently exists. Since this is completely lockless,
	 * utilize that to temporarily store requests until the task is
	 * either done or scheduled away.
	 */
1423 1424 1425
	plug = current->plug;
	if (plug) {
		blk_mq_bio_to_request(rq, bio);
M
Ming Lei 已提交
1426
		if (!request_count)
1427
			trace_block_plug(q);
1428 1429 1430 1431

		blk_mq_put_ctx(data.ctx);

		if (request_count >= BLK_MAX_REQUEST_COUNT) {
1432 1433
			blk_flush_plug_list(plug, false);
			trace_block_plug(q);
1434
		}
1435

1436
		list_add_tail(&rq->queuelist, &plug->mq_list);
1437
		return cookie;
1438 1439
	}

1440 1441 1442 1443 1444 1445 1446 1447 1448
	if (!blk_mq_merge_queue_io(data.hctx, data.ctx, rq, bio)) {
		/*
		 * For a SYNC request, send it to the hardware immediately. For
		 * an ASYNC request, just ensure that we run it later on. The
		 * latter allows for merging opportunities and more efficient
		 * dispatching.
		 */
run_queue:
		blk_mq_run_hw_queue(data.hctx, !is_sync || is_flush_fua);
1449 1450
	}

1451
	blk_mq_put_ctx(data.ctx);
1452
	return cookie;
1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
}

/*
 * Default mapping to a software queue, since we use one per CPU.
 */
struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q, const int cpu)
{
	return q->queue_hw_ctx[q->mq_map[cpu]];
}
EXPORT_SYMBOL(blk_mq_map_queue);

1464 1465
static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
		struct blk_mq_tags *tags, unsigned int hctx_idx)
1466
{
1467
	struct page *page;
1468

1469
	if (tags->rqs && set->ops->exit_request) {
1470
		int i;
1471

1472 1473
		for (i = 0; i < tags->nr_tags; i++) {
			if (!tags->rqs[i])
1474
				continue;
1475 1476
			set->ops->exit_request(set->driver_data, tags->rqs[i],
						hctx_idx, i);
1477
			tags->rqs[i] = NULL;
1478
		}
1479 1480
	}

1481 1482
	while (!list_empty(&tags->page_list)) {
		page = list_first_entry(&tags->page_list, struct page, lru);
1483
		list_del_init(&page->lru);
1484 1485 1486 1487 1488
		/*
		 * Remove kmemleak object previously allocated in
		 * blk_mq_init_rq_map().
		 */
		kmemleak_free(page_address(page));
1489 1490 1491
		__free_pages(page, page->private);
	}

1492
	kfree(tags->rqs);
1493

1494
	blk_mq_free_tags(tags);
1495 1496 1497 1498
}

static size_t order_to_size(unsigned int order)
{
1499
	return (size_t)PAGE_SIZE << order;
1500 1501
}

1502 1503
static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
		unsigned int hctx_idx)
1504
{
1505
	struct blk_mq_tags *tags;
1506 1507 1508
	unsigned int i, j, entries_per_page, max_order = 4;
	size_t rq_size, left;

1509
	tags = blk_mq_init_tags(set->queue_depth, set->reserved_tags,
S
Shaohua Li 已提交
1510 1511
				set->numa_node,
				BLK_MQ_FLAG_TO_ALLOC_POLICY(set->flags));
1512 1513
	if (!tags)
		return NULL;
1514

1515 1516
	INIT_LIST_HEAD(&tags->page_list);

1517 1518 1519
	tags->rqs = kzalloc_node(set->queue_depth * sizeof(struct request *),
				 GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY,
				 set->numa_node);
1520 1521 1522 1523
	if (!tags->rqs) {
		blk_mq_free_tags(tags);
		return NULL;
	}
1524 1525 1526 1527 1528

	/*
	 * rq_size is the size of the request plus driver payload, rounded
	 * to the cacheline size
	 */
1529
	rq_size = round_up(sizeof(struct request) + set->cmd_size,
1530
				cache_line_size());
1531
	left = rq_size * set->queue_depth;
1532

1533
	for (i = 0; i < set->queue_depth; ) {
1534 1535 1536 1537 1538
		int this_order = max_order;
		struct page *page;
		int to_do;
		void *p;

1539
		while (this_order && left < order_to_size(this_order - 1))
1540 1541 1542
			this_order--;

		do {
1543
			page = alloc_pages_node(set->numa_node,
1544
				GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO,
1545
				this_order);
1546 1547 1548 1549 1550 1551 1552 1553 1554
			if (page)
				break;
			if (!this_order--)
				break;
			if (order_to_size(this_order) < rq_size)
				break;
		} while (1);

		if (!page)
1555
			goto fail;
1556 1557

		page->private = this_order;
1558
		list_add_tail(&page->lru, &tags->page_list);
1559 1560

		p = page_address(page);
1561 1562 1563 1564 1565
		/*
		 * Allow kmemleak to scan these pages as they contain pointers
		 * to additional allocations like via ops->init_request().
		 */
		kmemleak_alloc(p, order_to_size(this_order), 1, GFP_KERNEL);
1566
		entries_per_page = order_to_size(this_order) / rq_size;
1567
		to_do = min(entries_per_page, set->queue_depth - i);
1568 1569
		left -= to_do * rq_size;
		for (j = 0; j < to_do; j++) {
1570 1571 1572 1573
			tags->rqs[i] = p;
			if (set->ops->init_request) {
				if (set->ops->init_request(set->driver_data,
						tags->rqs[i], hctx_idx, i,
1574 1575
						set->numa_node)) {
					tags->rqs[i] = NULL;
1576
					goto fail;
1577
				}
1578 1579
			}

1580 1581 1582 1583
			p += rq_size;
			i++;
		}
	}
1584
	return tags;
1585

1586 1587 1588
fail:
	blk_mq_free_rq_map(set, tags, hctx_idx);
	return NULL;
1589 1590
}

J
Jens Axboe 已提交
1591 1592 1593 1594 1595
/*
 * 'cpu' is going away. splice any existing rq_list entries from this
 * software queue to the hw queue dispatch list, and ensure that it
 * gets run.
 */
1596 1597 1598 1599 1600
static int blk_mq_hctx_cpu_offline(struct blk_mq_hw_ctx *hctx, int cpu)
{
	struct blk_mq_ctx *ctx;
	LIST_HEAD(tmp);

J
Jens Axboe 已提交
1601
	ctx = __blk_mq_get_ctx(hctx->queue, cpu);
1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612

	spin_lock(&ctx->lock);
	if (!list_empty(&ctx->rq_list)) {
		list_splice_init(&ctx->rq_list, &tmp);
		blk_mq_hctx_clear_pending(hctx, ctx);
	}
	spin_unlock(&ctx->lock);

	if (list_empty(&tmp))
		return NOTIFY_OK;

J
Jens Axboe 已提交
1613 1614 1615
	spin_lock(&hctx->lock);
	list_splice_tail_init(&tmp, &hctx->dispatch);
	spin_unlock(&hctx->lock);
1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627

	blk_mq_run_hw_queue(hctx, true);
	return NOTIFY_OK;
}

static int blk_mq_hctx_notify(void *data, unsigned long action,
			      unsigned int cpu)
{
	struct blk_mq_hw_ctx *hctx = data;

	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
		return blk_mq_hctx_cpu_offline(hctx, cpu);
M
Ming Lei 已提交
1628 1629 1630 1631 1632

	/*
	 * In case of CPU online, tags may be reallocated
	 * in blk_mq_map_swqueue() after mapping is updated.
	 */
1633 1634 1635 1636

	return NOTIFY_OK;
}

1637
/* hctx->ctxs will be freed in queue's release handler */
1638 1639 1640 1641
static void blk_mq_exit_hctx(struct request_queue *q,
		struct blk_mq_tag_set *set,
		struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
{
1642 1643
	unsigned flush_start_tag = set->queue_depth;

1644 1645
	blk_mq_tag_idle(hctx);

1646 1647 1648 1649 1650
	if (set->ops->exit_request)
		set->ops->exit_request(set->driver_data,
				       hctx->fq->flush_rq, hctx_idx,
				       flush_start_tag + hctx_idx);

1651 1652 1653 1654
	if (set->ops->exit_hctx)
		set->ops->exit_hctx(hctx, hctx_idx);

	blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
1655
	blk_free_flush_queue(hctx->fq);
1656
	sbitmap_free(&hctx->ctx_map);
1657 1658
}

M
Ming Lei 已提交
1659 1660 1661 1662 1663 1664 1665 1666 1667
static void blk_mq_exit_hw_queues(struct request_queue *q,
		struct blk_mq_tag_set *set, int nr_queue)
{
	struct blk_mq_hw_ctx *hctx;
	unsigned int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if (i == nr_queue)
			break;
1668
		blk_mq_exit_hctx(q, set, hctx, i);
M
Ming Lei 已提交
1669 1670 1671 1672 1673 1674 1675 1676 1677
	}
}

static void blk_mq_free_hw_queues(struct request_queue *q,
		struct blk_mq_tag_set *set)
{
	struct blk_mq_hw_ctx *hctx;
	unsigned int i;

1678
	queue_for_each_hw_ctx(q, hctx, i)
M
Ming Lei 已提交
1679 1680 1681
		free_cpumask_var(hctx->cpumask);
}

1682 1683 1684
static int blk_mq_init_hctx(struct request_queue *q,
		struct blk_mq_tag_set *set,
		struct blk_mq_hw_ctx *hctx, unsigned hctx_idx)
1685
{
1686
	int node;
1687
	unsigned flush_start_tag = set->queue_depth;
1688 1689 1690 1691 1692

	node = hctx->numa_node;
	if (node == NUMA_NO_NODE)
		node = hctx->numa_node = set->numa_node;

1693
	INIT_WORK(&hctx->run_work, blk_mq_run_work_fn);
1694 1695 1696 1697 1698
	INIT_DELAYED_WORK(&hctx->delay_work, blk_mq_delay_work_fn);
	spin_lock_init(&hctx->lock);
	INIT_LIST_HEAD(&hctx->dispatch);
	hctx->queue = q;
	hctx->queue_num = hctx_idx;
1699
	hctx->flags = set->flags & ~BLK_MQ_F_TAG_SHARED;
1700 1701 1702 1703 1704 1705

	blk_mq_init_cpu_notifier(&hctx->cpu_notifier,
					blk_mq_hctx_notify, hctx);
	blk_mq_register_cpu_notifier(&hctx->cpu_notifier);

	hctx->tags = set->tags[hctx_idx];
1706 1707

	/*
1708 1709
	 * Allocate space for all possible cpus to avoid allocation at
	 * runtime
1710
	 */
1711 1712 1713 1714
	hctx->ctxs = kmalloc_node(nr_cpu_ids * sizeof(void *),
					GFP_KERNEL, node);
	if (!hctx->ctxs)
		goto unregister_cpu_notifier;
1715

1716 1717
	if (sbitmap_init_node(&hctx->ctx_map, nr_cpu_ids, ilog2(8), GFP_KERNEL,
			      node))
1718
		goto free_ctxs;
1719

1720
	hctx->nr_ctx = 0;
1721

1722 1723 1724
	if (set->ops->init_hctx &&
	    set->ops->init_hctx(hctx, set->driver_data, hctx_idx))
		goto free_bitmap;
1725

1726 1727 1728
	hctx->fq = blk_alloc_flush_queue(q, hctx->numa_node, set->cmd_size);
	if (!hctx->fq)
		goto exit_hctx;
1729

1730 1731 1732 1733 1734
	if (set->ops->init_request &&
	    set->ops->init_request(set->driver_data,
				   hctx->fq->flush_rq, hctx_idx,
				   flush_start_tag + hctx_idx, node))
		goto free_fq;
1735

1736
	return 0;
1737

1738 1739 1740 1741 1742
 free_fq:
	kfree(hctx->fq);
 exit_hctx:
	if (set->ops->exit_hctx)
		set->ops->exit_hctx(hctx, hctx_idx);
1743
 free_bitmap:
1744
	sbitmap_free(&hctx->ctx_map);
1745 1746 1747 1748
 free_ctxs:
	kfree(hctx->ctxs);
 unregister_cpu_notifier:
	blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
1749

1750 1751
	return -1;
}
1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771

static void blk_mq_init_cpu_queues(struct request_queue *q,
				   unsigned int nr_hw_queues)
{
	unsigned int i;

	for_each_possible_cpu(i) {
		struct blk_mq_ctx *__ctx = per_cpu_ptr(q->queue_ctx, i);
		struct blk_mq_hw_ctx *hctx;

		memset(__ctx, 0, sizeof(*__ctx));
		__ctx->cpu = i;
		spin_lock_init(&__ctx->lock);
		INIT_LIST_HEAD(&__ctx->rq_list);
		__ctx->queue = q;

		/* If the cpu isn't online, the cpu is mapped to first hctx */
		if (!cpu_online(i))
			continue;

1772 1773
		hctx = q->mq_ops->map_queue(q, i);

1774 1775 1776 1777 1778
		/*
		 * Set local node, IFF we have more than one hw queue. If
		 * not, we remain on the home node of the device
		 */
		if (nr_hw_queues > 1 && hctx->numa_node == NUMA_NO_NODE)
1779
			hctx->numa_node = local_memory_node(cpu_to_node(i));
1780 1781 1782
	}
}

1783 1784
static void blk_mq_map_swqueue(struct request_queue *q,
			       const struct cpumask *online_mask)
1785 1786 1787 1788
{
	unsigned int i;
	struct blk_mq_hw_ctx *hctx;
	struct blk_mq_ctx *ctx;
M
Ming Lei 已提交
1789
	struct blk_mq_tag_set *set = q->tag_set;
1790

1791 1792 1793 1794 1795
	/*
	 * Avoid others reading imcomplete hctx->cpumask through sysfs
	 */
	mutex_lock(&q->sysfs_lock);

1796
	queue_for_each_hw_ctx(q, hctx, i) {
1797
		cpumask_clear(hctx->cpumask);
1798 1799 1800 1801 1802 1803
		hctx->nr_ctx = 0;
	}

	/*
	 * Map software to hardware queues
	 */
1804
	for_each_possible_cpu(i) {
1805
		/* If the cpu isn't online, the cpu is mapped to first hctx */
1806
		if (!cpumask_test_cpu(i, online_mask))
1807 1808
			continue;

1809
		ctx = per_cpu_ptr(q->queue_ctx, i);
1810
		hctx = q->mq_ops->map_queue(q, i);
K
Keith Busch 已提交
1811

1812
		cpumask_set_cpu(i, hctx->cpumask);
1813 1814 1815
		ctx->index_hw = hctx->nr_ctx;
		hctx->ctxs[hctx->nr_ctx++] = ctx;
	}
1816

1817 1818
	mutex_unlock(&q->sysfs_lock);

1819
	queue_for_each_hw_ctx(q, hctx, i) {
1820
		/*
1821 1822
		 * If no software queues are mapped to this hardware queue,
		 * disable it and free the request entries.
1823 1824 1825 1826 1827 1828
		 */
		if (!hctx->nr_ctx) {
			if (set->tags[i]) {
				blk_mq_free_rq_map(set, set->tags[i], i);
				set->tags[i] = NULL;
			}
M
Ming Lei 已提交
1829
			hctx->tags = NULL;
1830 1831 1832
			continue;
		}

M
Ming Lei 已提交
1833 1834 1835 1836 1837 1838
		/* unmapped hw queue can be remapped after CPU topo changed */
		if (!set->tags[i])
			set->tags[i] = blk_mq_init_rq_map(set, i);
		hctx->tags = set->tags[i];
		WARN_ON(!hctx->tags);

1839
		cpumask_copy(hctx->tags->cpumask, hctx->cpumask);
1840 1841 1842 1843 1844
		/*
		 * Set the map size to the number of mapped software queues.
		 * This is more accurate and more efficient than looping
		 * over all possibly mapped software queues.
		 */
1845
		sbitmap_resize(&hctx->ctx_map, hctx->nr_ctx);
1846

1847 1848 1849
		/*
		 * Initialize batch roundrobin counts
		 */
1850 1851 1852
		hctx->next_cpu = cpumask_first(hctx->cpumask);
		hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH;
	}
1853 1854
}

1855
static void queue_set_hctx_shared(struct request_queue *q, bool shared)
1856 1857 1858 1859
{
	struct blk_mq_hw_ctx *hctx;
	int i;

1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
	queue_for_each_hw_ctx(q, hctx, i) {
		if (shared)
			hctx->flags |= BLK_MQ_F_TAG_SHARED;
		else
			hctx->flags &= ~BLK_MQ_F_TAG_SHARED;
	}
}

static void blk_mq_update_tag_set_depth(struct blk_mq_tag_set *set, bool shared)
{
	struct request_queue *q;
1871 1872 1873

	list_for_each_entry(q, &set->tag_list, tag_set_list) {
		blk_mq_freeze_queue(q);
1874
		queue_set_hctx_shared(q, shared);
1875 1876 1877 1878 1879 1880 1881 1882 1883 1884
		blk_mq_unfreeze_queue(q);
	}
}

static void blk_mq_del_queue_tag_set(struct request_queue *q)
{
	struct blk_mq_tag_set *set = q->tag_set;

	mutex_lock(&set->tag_list_lock);
	list_del_init(&q->tag_set_list);
1885 1886 1887 1888 1889 1890
	if (list_is_singular(&set->tag_list)) {
		/* just transitioned to unshared */
		set->flags &= ~BLK_MQ_F_TAG_SHARED;
		/* update existing queue */
		blk_mq_update_tag_set_depth(set, false);
	}
1891 1892 1893 1894 1895 1896 1897 1898 1899
	mutex_unlock(&set->tag_list_lock);
}

static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set,
				     struct request_queue *q)
{
	q->tag_set = set;

	mutex_lock(&set->tag_list_lock);
1900 1901 1902 1903 1904 1905 1906 1907 1908

	/* Check to see if we're transitioning to shared (from 1 to 2 queues). */
	if (!list_empty(&set->tag_list) && !(set->flags & BLK_MQ_F_TAG_SHARED)) {
		set->flags |= BLK_MQ_F_TAG_SHARED;
		/* update existing queue */
		blk_mq_update_tag_set_depth(set, true);
	}
	if (set->flags & BLK_MQ_F_TAG_SHARED)
		queue_set_hctx_shared(q, true);
1909
	list_add_tail(&q->tag_set_list, &set->tag_list);
1910

1911 1912 1913
	mutex_unlock(&set->tag_list_lock);
}

1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925
/*
 * It is the actual release handler for mq, but we do it from
 * request queue's release handler for avoiding use-after-free
 * and headache because q->mq_kobj shouldn't have been introduced,
 * but we can't group ctx/kctx kobj without it.
 */
void blk_mq_release(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	unsigned int i;

	/* hctx kobj stays in hctx */
1926 1927 1928 1929
	queue_for_each_hw_ctx(q, hctx, i) {
		if (!hctx)
			continue;
		kfree(hctx->ctxs);
1930
		kfree(hctx);
1931
	}
1932

1933 1934 1935
	kfree(q->mq_map);
	q->mq_map = NULL;

1936 1937 1938 1939 1940 1941
	kfree(q->queue_hw_ctx);

	/* ctx kobj stays in queue_ctx */
	free_percpu(q->queue_ctx);
}

1942
struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957
{
	struct request_queue *uninit_q, *q;

	uninit_q = blk_alloc_queue_node(GFP_KERNEL, set->numa_node);
	if (!uninit_q)
		return ERR_PTR(-ENOMEM);

	q = blk_mq_init_allocated_queue(set, uninit_q);
	if (IS_ERR(q))
		blk_cleanup_queue(uninit_q);

	return q;
}
EXPORT_SYMBOL(blk_mq_init_queue);

K
Keith Busch 已提交
1958 1959
static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,
						struct request_queue *q)
1960
{
K
Keith Busch 已提交
1961 1962
	int i, j;
	struct blk_mq_hw_ctx **hctxs = q->queue_hw_ctx;
1963

K
Keith Busch 已提交
1964
	blk_mq_sysfs_unregister(q);
1965
	for (i = 0; i < set->nr_hw_queues; i++) {
K
Keith Busch 已提交
1966
		int node;
1967

K
Keith Busch 已提交
1968 1969 1970 1971
		if (hctxs[i])
			continue;

		node = blk_mq_hw_queue_to_node(q->mq_map, i);
1972 1973
		hctxs[i] = kzalloc_node(sizeof(struct blk_mq_hw_ctx),
					GFP_KERNEL, node);
1974
		if (!hctxs[i])
K
Keith Busch 已提交
1975
			break;
1976

1977
		if (!zalloc_cpumask_var_node(&hctxs[i]->cpumask, GFP_KERNEL,
K
Keith Busch 已提交
1978 1979 1980 1981 1982
						node)) {
			kfree(hctxs[i]);
			hctxs[i] = NULL;
			break;
		}
1983

1984
		atomic_set(&hctxs[i]->nr_active, 0);
1985
		hctxs[i]->numa_node = node;
1986
		hctxs[i]->queue_num = i;
K
Keith Busch 已提交
1987 1988 1989 1990 1991 1992 1993 1994

		if (blk_mq_init_hctx(q, set, hctxs[i], i)) {
			free_cpumask_var(hctxs[i]->cpumask);
			kfree(hctxs[i]);
			hctxs[i] = NULL;
			break;
		}
		blk_mq_hctx_kobj_init(hctxs[i]);
1995
	}
K
Keith Busch 已提交
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019
	for (j = i; j < q->nr_hw_queues; j++) {
		struct blk_mq_hw_ctx *hctx = hctxs[j];

		if (hctx) {
			if (hctx->tags) {
				blk_mq_free_rq_map(set, hctx->tags, j);
				set->tags[j] = NULL;
			}
			blk_mq_exit_hctx(q, set, hctx, j);
			free_cpumask_var(hctx->cpumask);
			kobject_put(&hctx->kobj);
			kfree(hctx->ctxs);
			kfree(hctx);
			hctxs[j] = NULL;

		}
	}
	q->nr_hw_queues = i;
	blk_mq_sysfs_register(q);
}

struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
						  struct request_queue *q)
{
M
Ming Lei 已提交
2020 2021 2022
	/* mark the queue as mq asap */
	q->mq_ops = set->ops;

K
Keith Busch 已提交
2023 2024
	q->queue_ctx = alloc_percpu(struct blk_mq_ctx);
	if (!q->queue_ctx)
M
Ming Lin 已提交
2025
		goto err_exit;
K
Keith Busch 已提交
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038

	q->queue_hw_ctx = kzalloc_node(nr_cpu_ids * sizeof(*(q->queue_hw_ctx)),
						GFP_KERNEL, set->numa_node);
	if (!q->queue_hw_ctx)
		goto err_percpu;

	q->mq_map = blk_mq_make_queue_map(set);
	if (!q->mq_map)
		goto err_map;

	blk_mq_realloc_hw_ctxs(set, q);
	if (!q->nr_hw_queues)
		goto err_hctxs;
2039

2040
	INIT_WORK(&q->timeout_work, blk_mq_timeout_work);
2041
	blk_queue_rq_timeout(q, set->timeout ? set->timeout : 30 * HZ);
2042 2043 2044

	q->nr_queues = nr_cpu_ids;

2045
	q->queue_flags |= QUEUE_FLAG_MQ_DEFAULT;
2046

2047 2048 2049
	if (!(set->flags & BLK_MQ_F_SG_MERGE))
		q->queue_flags |= 1 << QUEUE_FLAG_NO_SG_MERGE;

2050 2051
	q->sg_reserved_size = INT_MAX;

2052
	INIT_DELAYED_WORK(&q->requeue_work, blk_mq_requeue_work);
2053 2054 2055
	INIT_LIST_HEAD(&q->requeue_list);
	spin_lock_init(&q->requeue_lock);

2056 2057 2058 2059 2060
	if (q->nr_hw_queues > 1)
		blk_queue_make_request(q, blk_mq_make_request);
	else
		blk_queue_make_request(q, blk_sq_make_request);

2061 2062 2063 2064 2065
	/*
	 * Do this after blk_queue_make_request() overrides it...
	 */
	q->nr_requests = set->queue_depth;

2066 2067
	if (set->ops->complete)
		blk_queue_softirq_done(q, set->ops->complete);
2068

2069
	blk_mq_init_cpu_queues(q, set->nr_hw_queues);
2070

2071
	get_online_cpus();
2072 2073
	mutex_lock(&all_q_mutex);

2074
	list_add_tail(&q->all_q_node, &all_q_list);
2075
	blk_mq_add_queue_tag_set(set, q);
2076
	blk_mq_map_swqueue(q, cpu_online_mask);
2077

2078
	mutex_unlock(&all_q_mutex);
2079
	put_online_cpus();
2080

2081
	return q;
2082

2083
err_hctxs:
K
Keith Busch 已提交
2084
	kfree(q->mq_map);
2085
err_map:
K
Keith Busch 已提交
2086
	kfree(q->queue_hw_ctx);
2087
err_percpu:
K
Keith Busch 已提交
2088
	free_percpu(q->queue_ctx);
M
Ming Lin 已提交
2089 2090
err_exit:
	q->mq_ops = NULL;
2091 2092
	return ERR_PTR(-ENOMEM);
}
2093
EXPORT_SYMBOL(blk_mq_init_allocated_queue);
2094 2095 2096

void blk_mq_free_queue(struct request_queue *q)
{
M
Ming Lei 已提交
2097
	struct blk_mq_tag_set	*set = q->tag_set;
2098

2099 2100 2101 2102
	mutex_lock(&all_q_mutex);
	list_del_init(&q->all_q_node);
	mutex_unlock(&all_q_mutex);

2103 2104
	blk_mq_del_queue_tag_set(q);

M
Ming Lei 已提交
2105 2106
	blk_mq_exit_hw_queues(q, set, set->nr_hw_queues);
	blk_mq_free_hw_queues(q, set);
2107 2108 2109
}

/* Basically redo blk_mq_init_queue with queue frozen */
2110 2111
static void blk_mq_queue_reinit(struct request_queue *q,
				const struct cpumask *online_mask)
2112
{
2113
	WARN_ON_ONCE(!atomic_read(&q->mq_freeze_depth));
2114

2115 2116
	blk_mq_sysfs_unregister(q);

2117
	blk_mq_update_queue_map(q->mq_map, q->nr_hw_queues, online_mask);
2118 2119 2120 2121 2122 2123 2124

	/*
	 * redo blk_mq_init_cpu_queues and blk_mq_init_hw_queues. FIXME: maybe
	 * we should change hctx numa_node according to new topology (this
	 * involves free and re-allocate memory, worthy doing?)
	 */

2125
	blk_mq_map_swqueue(q, online_mask);
2126

2127
	blk_mq_sysfs_register(q);
2128 2129
}

2130 2131
static int blk_mq_queue_reinit_notify(struct notifier_block *nb,
				      unsigned long action, void *hcpu)
2132 2133
{
	struct request_queue *q;
2134 2135 2136 2137 2138 2139 2140
	int cpu = (unsigned long)hcpu;
	/*
	 * New online cpumask which is going to be set in this hotplug event.
	 * Declare this cpumasks as global as cpu-hotplug operation is invoked
	 * one-by-one and dynamically allocating this could result in a failure.
	 */
	static struct cpumask online_new;
2141 2142

	/*
2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157
	 * Before hotadded cpu starts handling requests, new mappings must
	 * be established.  Otherwise, these requests in hw queue might
	 * never be dispatched.
	 *
	 * For example, there is a single hw queue (hctx) and two CPU queues
	 * (ctx0 for CPU0, and ctx1 for CPU1).
	 *
	 * Now CPU1 is just onlined and a request is inserted into
	 * ctx1->rq_list and set bit0 in pending bitmap as ctx1->index_hw is
	 * still zero.
	 *
	 * And then while running hw queue, flush_busy_ctxs() finds bit0 is
	 * set in pending bitmap and tries to retrieve requests in
	 * hctx->ctxs[0]->rq_list.  But htx->ctxs[0] is a pointer to ctx0,
	 * so the request in ctx1->rq_list is ignored.
2158
	 */
2159 2160 2161 2162 2163 2164 2165 2166 2167 2168
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DEAD:
	case CPU_UP_CANCELED:
		cpumask_copy(&online_new, cpu_online_mask);
		break;
	case CPU_UP_PREPARE:
		cpumask_copy(&online_new, cpu_online_mask);
		cpumask_set_cpu(cpu, &online_new);
		break;
	default:
2169
		return NOTIFY_OK;
2170
	}
2171 2172

	mutex_lock(&all_q_mutex);
2173 2174 2175 2176 2177 2178 2179 2180 2181 2182

	/*
	 * We need to freeze and reinit all existing queues.  Freezing
	 * involves synchronous wait for an RCU grace period and doing it
	 * one by one may take a long time.  Start freezing all queues in
	 * one swoop and then wait for the completions so that freezing can
	 * take place in parallel.
	 */
	list_for_each_entry(q, &all_q_list, all_q_node)
		blk_mq_freeze_queue_start(q);
2183
	list_for_each_entry(q, &all_q_list, all_q_node) {
2184 2185
		blk_mq_freeze_queue_wait(q);

2186 2187 2188 2189 2190 2191 2192
		/*
		 * timeout handler can't touch hw queue during the
		 * reinitialization
		 */
		del_timer_sync(&q->timeout);
	}

2193
	list_for_each_entry(q, &all_q_list, all_q_node)
2194
		blk_mq_queue_reinit(q, &online_new);
2195 2196 2197 2198

	list_for_each_entry(q, &all_q_list, all_q_node)
		blk_mq_unfreeze_queue(q);

2199 2200 2201 2202
	mutex_unlock(&all_q_mutex);
	return NOTIFY_OK;
}

2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256
static int __blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set)
{
	int i;

	for (i = 0; i < set->nr_hw_queues; i++) {
		set->tags[i] = blk_mq_init_rq_map(set, i);
		if (!set->tags[i])
			goto out_unwind;
	}

	return 0;

out_unwind:
	while (--i >= 0)
		blk_mq_free_rq_map(set, set->tags[i], i);

	return -ENOMEM;
}

/*
 * Allocate the request maps associated with this tag_set. Note that this
 * may reduce the depth asked for, if memory is tight. set->queue_depth
 * will be updated to reflect the allocated depth.
 */
static int blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set)
{
	unsigned int depth;
	int err;

	depth = set->queue_depth;
	do {
		err = __blk_mq_alloc_rq_maps(set);
		if (!err)
			break;

		set->queue_depth >>= 1;
		if (set->queue_depth < set->reserved_tags + BLK_MQ_TAG_MIN) {
			err = -ENOMEM;
			break;
		}
	} while (set->queue_depth);

	if (!set->queue_depth || err) {
		pr_err("blk-mq: failed to allocate request map\n");
		return -ENOMEM;
	}

	if (depth != set->queue_depth)
		pr_info("blk-mq: reduced tag depth (%u -> %u)\n",
						depth, set->queue_depth);

	return 0;
}

K
Keith Busch 已提交
2257 2258 2259 2260 2261 2262
struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags)
{
	return tags->cpumask;
}
EXPORT_SYMBOL_GPL(blk_mq_tags_cpumask);

2263 2264 2265 2266 2267 2268
/*
 * Alloc a tag set to be associated with one or more request queues.
 * May fail with EINVAL for various error conditions. May adjust the
 * requested depth down, if if it too large. In that case, the set
 * value will be stored in set->queue_depth.
 */
2269 2270
int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
{
B
Bart Van Assche 已提交
2271 2272
	BUILD_BUG_ON(BLK_MQ_MAX_DEPTH > 1 << BLK_MQ_UNIQUE_TAG_BITS);

2273 2274
	if (!set->nr_hw_queues)
		return -EINVAL;
2275
	if (!set->queue_depth)
2276 2277 2278 2279
		return -EINVAL;
	if (set->queue_depth < set->reserved_tags + BLK_MQ_TAG_MIN)
		return -EINVAL;

2280
	if (!set->ops->queue_rq || !set->ops->map_queue)
2281 2282
		return -EINVAL;

2283 2284 2285 2286 2287
	if (set->queue_depth > BLK_MQ_MAX_DEPTH) {
		pr_info("blk-mq: reduced tag depth to %u\n",
			BLK_MQ_MAX_DEPTH);
		set->queue_depth = BLK_MQ_MAX_DEPTH;
	}
2288

2289 2290 2291 2292 2293 2294 2295 2296 2297
	/*
	 * If a crashdump is active, then we are potentially in a very
	 * memory constrained environment. Limit us to 1 queue and
	 * 64 tags to prevent using too much memory.
	 */
	if (is_kdump_kernel()) {
		set->nr_hw_queues = 1;
		set->queue_depth = min(64U, set->queue_depth);
	}
K
Keith Busch 已提交
2298 2299 2300 2301 2302
	/*
	 * There is no use for more h/w queues than cpus.
	 */
	if (set->nr_hw_queues > nr_cpu_ids)
		set->nr_hw_queues = nr_cpu_ids;
2303

K
Keith Busch 已提交
2304
	set->tags = kzalloc_node(nr_cpu_ids * sizeof(struct blk_mq_tags *),
2305 2306
				 GFP_KERNEL, set->numa_node);
	if (!set->tags)
2307
		return -ENOMEM;
2308

2309 2310
	if (blk_mq_alloc_rq_maps(set))
		goto enomem;
2311

2312 2313 2314
	mutex_init(&set->tag_list_lock);
	INIT_LIST_HEAD(&set->tag_list);

2315
	return 0;
2316
enomem:
2317 2318
	kfree(set->tags);
	set->tags = NULL;
2319 2320 2321 2322 2323 2324 2325 2326
	return -ENOMEM;
}
EXPORT_SYMBOL(blk_mq_alloc_tag_set);

void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
{
	int i;

K
Keith Busch 已提交
2327
	for (i = 0; i < nr_cpu_ids; i++) {
2328
		if (set->tags[i])
2329 2330 2331
			blk_mq_free_rq_map(set, set->tags[i], i);
	}

M
Ming Lei 已提交
2332
	kfree(set->tags);
2333
	set->tags = NULL;
2334 2335 2336
}
EXPORT_SYMBOL(blk_mq_free_tag_set);

2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347
int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr)
{
	struct blk_mq_tag_set *set = q->tag_set;
	struct blk_mq_hw_ctx *hctx;
	int i, ret;

	if (!set || nr > set->queue_depth)
		return -EINVAL;

	ret = 0;
	queue_for_each_hw_ctx(q, hctx, i) {
2348 2349
		if (!hctx->tags)
			continue;
2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360
		ret = blk_mq_tag_update_depth(hctx->tags, nr);
		if (ret)
			break;
	}

	if (!ret)
		q->nr_requests = nr;

	return ret;
}

K
Keith Busch 已提交
2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues)
{
	struct request_queue *q;

	if (nr_hw_queues > nr_cpu_ids)
		nr_hw_queues = nr_cpu_ids;
	if (nr_hw_queues < 1 || nr_hw_queues == set->nr_hw_queues)
		return;

	list_for_each_entry(q, &set->tag_list, tag_set_list)
		blk_mq_freeze_queue(q);

	set->nr_hw_queues = nr_hw_queues;
	list_for_each_entry(q, &set->tag_list, tag_set_list) {
		blk_mq_realloc_hw_ctxs(set, q);

		if (q->nr_hw_queues > 1)
			blk_queue_make_request(q, blk_mq_make_request);
		else
			blk_queue_make_request(q, blk_sq_make_request);

		blk_mq_queue_reinit(q, cpu_online_mask);
	}

	list_for_each_entry(q, &set->tag_list, tag_set_list)
		blk_mq_unfreeze_queue(q);
}
EXPORT_SYMBOL_GPL(blk_mq_update_nr_hw_queues);

2390 2391 2392 2393 2394 2395 2396 2397 2398 2399
void blk_mq_disable_hotplug(void)
{
	mutex_lock(&all_q_mutex);
}

void blk_mq_enable_hotplug(void)
{
	mutex_unlock(&all_q_mutex);
}

2400 2401 2402 2403
static int __init blk_mq_init(void)
{
	blk_mq_cpu_init();

2404
	hotcpu_notifier(blk_mq_queue_reinit_notify, 0);
2405 2406 2407 2408

	return 0;
}
subsys_initcall(blk_mq_init);