2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/block.h>
43 #include "blk-mq-sched.h"
46 #ifdef CONFIG_DEBUG_FS
47 struct dentry *blk_debugfs_root;
50 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
56 DEFINE_IDA(blk_queue_ida);
59 * For the allocated request tables
61 struct kmem_cache *request_cachep;
64 * For queue allocation
66 struct kmem_cache *blk_requestq_cachep;
69 * Controlling structure to kblockd
71 static struct workqueue_struct *kblockd_workqueue;
73 static void blk_clear_congested(struct request_list *rl, int sync)
75 #ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl->blkg->wb_congested, sync);
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
82 if (rl == &rl->q->root_rl)
83 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
87 static void blk_set_congested(struct request_list *rl, int sync)
89 #ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl->blkg->wb_congested, sync);
92 /* see blk_clear_congested() */
93 if (rl == &rl->q->root_rl)
94 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
98 void blk_queue_congestion_threshold(struct request_queue *q)
102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
103 if (nr > q->nr_requests)
105 q->nr_congestion_on = nr;
107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
110 q->nr_congestion_off = nr;
113 void blk_rq_init(struct request_queue *q, struct request *rq)
115 memset(rq, 0, sizeof(*rq));
117 INIT_LIST_HEAD(&rq->queuelist);
118 INIT_LIST_HEAD(&rq->timeout_list);
121 rq->__sector = (sector_t) -1;
122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
125 rq->internal_tag = -1;
126 rq->start_time = jiffies;
127 set_start_time_ns(rq);
130 EXPORT_SYMBOL(blk_rq_init);
132 static const struct {
136 [BLK_STS_OK] = { 0, "" },
137 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
138 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
139 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
140 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
141 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
142 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
143 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
144 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
145 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
146 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
151 /* everything else not covered above: */
152 [BLK_STS_IOERR] = { -EIO, "I/O" },
155 blk_status_t errno_to_blk_status(int errno)
159 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
160 if (blk_errors[i].errno == errno)
161 return (__force blk_status_t)i;
164 return BLK_STS_IOERR;
166 EXPORT_SYMBOL_GPL(errno_to_blk_status);
168 int blk_status_to_errno(blk_status_t status)
170 int idx = (__force int)status;
172 if (WARN_ON_ONCE(idx > ARRAY_SIZE(blk_errors)))
174 return blk_errors[idx].errno;
176 EXPORT_SYMBOL_GPL(blk_status_to_errno);
178 static void print_req_error(struct request *req, blk_status_t status)
180 int idx = (__force int)status;
182 if (WARN_ON_ONCE(idx > ARRAY_SIZE(blk_errors)))
185 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
186 __func__, blk_errors[idx].name, req->rq_disk ?
187 req->rq_disk->disk_name : "?",
188 (unsigned long long)blk_rq_pos(req));
191 static void req_bio_endio(struct request *rq, struct bio *bio,
192 unsigned int nbytes, blk_status_t error)
195 bio->bi_status = error;
197 if (unlikely(rq->rq_flags & RQF_QUIET))
198 bio_set_flag(bio, BIO_QUIET);
200 bio_advance(bio, nbytes);
202 /* don't actually finish bio if it's part of flush sequence */
203 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
207 void blk_dump_rq_flags(struct request *rq, char *msg)
209 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
210 rq->rq_disk ? rq->rq_disk->disk_name : "?",
211 (unsigned long long) rq->cmd_flags);
213 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
214 (unsigned long long)blk_rq_pos(rq),
215 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
216 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
217 rq->bio, rq->biotail, blk_rq_bytes(rq));
219 EXPORT_SYMBOL(blk_dump_rq_flags);
221 static void blk_delay_work(struct work_struct *work)
223 struct request_queue *q;
225 q = container_of(work, struct request_queue, delay_work.work);
226 spin_lock_irq(q->queue_lock);
228 spin_unlock_irq(q->queue_lock);
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
237 * Sometimes queueing needs to be postponed for a little while, to allow
238 * resources to come back. This function will make sure that queueing is
239 * restarted around the specified time. Queue lock must be held.
241 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
243 if (likely(!blk_queue_dead(q)))
244 queue_delayed_work(kblockd_workqueue, &q->delay_work,
245 msecs_to_jiffies(msecs));
247 EXPORT_SYMBOL(blk_delay_queue);
250 * blk_start_queue_async - asynchronously restart a previously stopped queue
251 * @q: The &struct request_queue in question
254 * blk_start_queue_async() will clear the stop flag on the queue, and
255 * ensure that the request_fn for the queue is run from an async
258 void blk_start_queue_async(struct request_queue *q)
260 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
261 blk_run_queue_async(q);
263 EXPORT_SYMBOL(blk_start_queue_async);
266 * blk_start_queue - restart a previously stopped queue
267 * @q: The &struct request_queue in question
270 * blk_start_queue() will clear the stop flag on the queue, and call
271 * the request_fn for the queue if it was in a stopped state when
272 * entered. Also see blk_stop_queue(). Queue lock must be held.
274 void blk_start_queue(struct request_queue *q)
276 WARN_ON(!irqs_disabled());
278 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
281 EXPORT_SYMBOL(blk_start_queue);
284 * blk_stop_queue - stop a queue
285 * @q: The &struct request_queue in question
288 * The Linux block layer assumes that a block driver will consume all
289 * entries on the request queue when the request_fn strategy is called.
290 * Often this will not happen, because of hardware limitations (queue
291 * depth settings). If a device driver gets a 'queue full' response,
292 * or if it simply chooses not to queue more I/O at one point, it can
293 * call this function to prevent the request_fn from being called until
294 * the driver has signalled it's ready to go again. This happens by calling
295 * blk_start_queue() to restart queue operations. Queue lock must be held.
297 void blk_stop_queue(struct request_queue *q)
299 cancel_delayed_work(&q->delay_work);
300 queue_flag_set(QUEUE_FLAG_STOPPED, q);
302 EXPORT_SYMBOL(blk_stop_queue);
305 * blk_sync_queue - cancel any pending callbacks on a queue
309 * The block layer may perform asynchronous callback activity
310 * on a queue, such as calling the unplug function after a timeout.
311 * A block device may call blk_sync_queue to ensure that any
312 * such activity is cancelled, thus allowing it to release resources
313 * that the callbacks might use. The caller must already have made sure
314 * that its ->make_request_fn will not re-add plugging prior to calling
317 * This function does not cancel any asynchronous activity arising
318 * out of elevator or throttling code. That would require elevator_exit()
319 * and blkcg_exit_queue() to be called with queue lock initialized.
322 void blk_sync_queue(struct request_queue *q)
324 del_timer_sync(&q->timeout);
327 struct blk_mq_hw_ctx *hctx;
330 queue_for_each_hw_ctx(q, hctx, i)
331 cancel_delayed_work_sync(&hctx->run_work);
333 cancel_delayed_work_sync(&q->delay_work);
336 EXPORT_SYMBOL(blk_sync_queue);
339 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
340 * @q: The queue to run
343 * Invoke request handling on a queue if there are any pending requests.
344 * May be used to restart request handling after a request has completed.
345 * This variant runs the queue whether or not the queue has been
346 * stopped. Must be called with the queue lock held and interrupts
347 * disabled. See also @blk_run_queue.
349 inline void __blk_run_queue_uncond(struct request_queue *q)
351 if (unlikely(blk_queue_dead(q)))
355 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
356 * the queue lock internally. As a result multiple threads may be
357 * running such a request function concurrently. Keep track of the
358 * number of active request_fn invocations such that blk_drain_queue()
359 * can wait until all these request_fn calls have finished.
361 q->request_fn_active++;
363 q->request_fn_active--;
365 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
368 * __blk_run_queue - run a single device queue
369 * @q: The queue to run
372 * See @blk_run_queue. This variant must be called with the queue lock
373 * held and interrupts disabled.
375 void __blk_run_queue(struct request_queue *q)
377 if (unlikely(blk_queue_stopped(q)))
380 __blk_run_queue_uncond(q);
382 EXPORT_SYMBOL(__blk_run_queue);
385 * blk_run_queue_async - run a single device queue in workqueue context
386 * @q: The queue to run
389 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
390 * of us. The caller must hold the queue lock.
392 void blk_run_queue_async(struct request_queue *q)
394 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
395 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
397 EXPORT_SYMBOL(blk_run_queue_async);
400 * blk_run_queue - run a single device queue
401 * @q: The queue to run
404 * Invoke request handling on this queue, if it has pending work to do.
405 * May be used to restart queueing when a request has completed.
407 void blk_run_queue(struct request_queue *q)
411 spin_lock_irqsave(q->queue_lock, flags);
413 spin_unlock_irqrestore(q->queue_lock, flags);
415 EXPORT_SYMBOL(blk_run_queue);
417 void blk_put_queue(struct request_queue *q)
419 kobject_put(&q->kobj);
421 EXPORT_SYMBOL(blk_put_queue);
424 * __blk_drain_queue - drain requests from request_queue
426 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
428 * Drain requests from @q. If @drain_all is set, all requests are drained.
429 * If not, only ELVPRIV requests are drained. The caller is responsible
430 * for ensuring that no new requests which need to be drained are queued.
432 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
433 __releases(q->queue_lock)
434 __acquires(q->queue_lock)
438 lockdep_assert_held(q->queue_lock);
444 * The caller might be trying to drain @q before its
445 * elevator is initialized.
448 elv_drain_elevator(q);
450 blkcg_drain_queue(q);
453 * This function might be called on a queue which failed
454 * driver init after queue creation or is not yet fully
455 * active yet. Some drivers (e.g. fd and loop) get unhappy
456 * in such cases. Kick queue iff dispatch queue has
457 * something on it and @q has request_fn set.
459 if (!list_empty(&q->queue_head) && q->request_fn)
462 drain |= q->nr_rqs_elvpriv;
463 drain |= q->request_fn_active;
466 * Unfortunately, requests are queued at and tracked from
467 * multiple places and there's no single counter which can
468 * be drained. Check all the queues and counters.
471 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
472 drain |= !list_empty(&q->queue_head);
473 for (i = 0; i < 2; i++) {
474 drain |= q->nr_rqs[i];
475 drain |= q->in_flight[i];
477 drain |= !list_empty(&fq->flush_queue[i]);
484 spin_unlock_irq(q->queue_lock);
488 spin_lock_irq(q->queue_lock);
492 * With queue marked dead, any woken up waiter will fail the
493 * allocation path, so the wakeup chaining is lost and we're
494 * left with hung waiters. We need to wake up those waiters.
497 struct request_list *rl;
499 blk_queue_for_each_rl(rl, q)
500 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
501 wake_up_all(&rl->wait[i]);
506 * blk_queue_bypass_start - enter queue bypass mode
507 * @q: queue of interest
509 * In bypass mode, only the dispatch FIFO queue of @q is used. This
510 * function makes @q enter bypass mode and drains all requests which were
511 * throttled or issued before. On return, it's guaranteed that no request
512 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
513 * inside queue or RCU read lock.
515 void blk_queue_bypass_start(struct request_queue *q)
517 spin_lock_irq(q->queue_lock);
519 queue_flag_set(QUEUE_FLAG_BYPASS, q);
520 spin_unlock_irq(q->queue_lock);
523 * Queues start drained. Skip actual draining till init is
524 * complete. This avoids lenghty delays during queue init which
525 * can happen many times during boot.
527 if (blk_queue_init_done(q)) {
528 spin_lock_irq(q->queue_lock);
529 __blk_drain_queue(q, false);
530 spin_unlock_irq(q->queue_lock);
532 /* ensure blk_queue_bypass() is %true inside RCU read lock */
536 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
539 * blk_queue_bypass_end - leave queue bypass mode
540 * @q: queue of interest
542 * Leave bypass mode and restore the normal queueing behavior.
544 void blk_queue_bypass_end(struct request_queue *q)
546 spin_lock_irq(q->queue_lock);
547 if (!--q->bypass_depth)
548 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
549 WARN_ON_ONCE(q->bypass_depth < 0);
550 spin_unlock_irq(q->queue_lock);
552 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
554 void blk_set_queue_dying(struct request_queue *q)
556 spin_lock_irq(q->queue_lock);
557 queue_flag_set(QUEUE_FLAG_DYING, q);
558 spin_unlock_irq(q->queue_lock);
561 * When queue DYING flag is set, we need to block new req
562 * entering queue, so we call blk_freeze_queue_start() to
563 * prevent I/O from crossing blk_queue_enter().
565 blk_freeze_queue_start(q);
568 blk_mq_wake_waiters(q);
570 struct request_list *rl;
572 spin_lock_irq(q->queue_lock);
573 blk_queue_for_each_rl(rl, q) {
575 wake_up(&rl->wait[BLK_RW_SYNC]);
576 wake_up(&rl->wait[BLK_RW_ASYNC]);
579 spin_unlock_irq(q->queue_lock);
582 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
585 * blk_cleanup_queue - shutdown a request queue
586 * @q: request queue to shutdown
588 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
589 * put it. All future requests will be failed immediately with -ENODEV.
591 void blk_cleanup_queue(struct request_queue *q)
593 spinlock_t *lock = q->queue_lock;
595 /* mark @q DYING, no new request or merges will be allowed afterwards */
596 mutex_lock(&q->sysfs_lock);
597 blk_set_queue_dying(q);
601 * A dying queue is permanently in bypass mode till released. Note
602 * that, unlike blk_queue_bypass_start(), we aren't performing
603 * synchronize_rcu() after entering bypass mode to avoid the delay
604 * as some drivers create and destroy a lot of queues while
605 * probing. This is still safe because blk_release_queue() will be
606 * called only after the queue refcnt drops to zero and nothing,
607 * RCU or not, would be traversing the queue by then.
610 queue_flag_set(QUEUE_FLAG_BYPASS, q);
612 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
613 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
614 queue_flag_set(QUEUE_FLAG_DYING, q);
615 spin_unlock_irq(lock);
616 mutex_unlock(&q->sysfs_lock);
619 * Drain all requests queued before DYING marking. Set DEAD flag to
620 * prevent that q->request_fn() gets invoked after draining finished.
625 __blk_drain_queue(q, true);
626 queue_flag_set(QUEUE_FLAG_DEAD, q);
627 spin_unlock_irq(lock);
629 /* for synchronous bio-based driver finish in-flight integrity i/o */
630 blk_flush_integrity();
632 /* @q won't process any more request, flush async actions */
633 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
637 blk_mq_free_queue(q);
638 percpu_ref_exit(&q->q_usage_counter);
641 if (q->queue_lock != &q->__queue_lock)
642 q->queue_lock = &q->__queue_lock;
643 spin_unlock_irq(lock);
645 /* @q is and will stay empty, shutdown and put */
648 EXPORT_SYMBOL(blk_cleanup_queue);
650 /* Allocate memory local to the request queue */
651 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
653 struct request_queue *q = data;
655 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
658 static void free_request_simple(void *element, void *data)
660 kmem_cache_free(request_cachep, element);
663 static void *alloc_request_size(gfp_t gfp_mask, void *data)
665 struct request_queue *q = data;
668 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
670 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
677 static void free_request_size(void *element, void *data)
679 struct request_queue *q = data;
682 q->exit_rq_fn(q, element);
686 int blk_init_rl(struct request_list *rl, struct request_queue *q,
689 if (unlikely(rl->rq_pool))
693 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
694 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
695 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
696 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
699 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
700 alloc_request_size, free_request_size,
701 q, gfp_mask, q->node);
703 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
704 alloc_request_simple, free_request_simple,
705 q, gfp_mask, q->node);
710 if (rl != &q->root_rl)
711 WARN_ON_ONCE(!blk_get_queue(q));
716 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
719 mempool_destroy(rl->rq_pool);
720 if (rl != &q->root_rl)
725 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
727 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
729 EXPORT_SYMBOL(blk_alloc_queue);
731 int blk_queue_enter(struct request_queue *q, bool nowait)
736 if (percpu_ref_tryget_live(&q->q_usage_counter))
743 * read pair of barrier in blk_freeze_queue_start(),
744 * we need to order reading __PERCPU_REF_DEAD flag of
745 * .q_usage_counter and reading .mq_freeze_depth or
746 * queue dying flag, otherwise the following wait may
747 * never return if the two reads are reordered.
751 ret = wait_event_interruptible(q->mq_freeze_wq,
752 !atomic_read(&q->mq_freeze_depth) ||
754 if (blk_queue_dying(q))
761 void blk_queue_exit(struct request_queue *q)
763 percpu_ref_put(&q->q_usage_counter);
766 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
768 struct request_queue *q =
769 container_of(ref, struct request_queue, q_usage_counter);
771 wake_up_all(&q->mq_freeze_wq);
774 static void blk_rq_timed_out_timer(unsigned long data)
776 struct request_queue *q = (struct request_queue *)data;
778 kblockd_schedule_work(&q->timeout_work);
781 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
783 struct request_queue *q;
785 q = kmem_cache_alloc_node(blk_requestq_cachep,
786 gfp_mask | __GFP_ZERO, node_id);
790 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
794 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
798 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
799 if (!q->backing_dev_info)
802 q->stats = blk_alloc_queue_stats();
806 q->backing_dev_info->ra_pages =
807 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
808 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
809 q->backing_dev_info->name = "block";
812 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
813 laptop_mode_timer_fn, (unsigned long) q);
814 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
815 INIT_LIST_HEAD(&q->queue_head);
816 INIT_LIST_HEAD(&q->timeout_list);
817 INIT_LIST_HEAD(&q->icq_list);
818 #ifdef CONFIG_BLK_CGROUP
819 INIT_LIST_HEAD(&q->blkg_list);
821 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
823 kobject_init(&q->kobj, &blk_queue_ktype);
825 mutex_init(&q->sysfs_lock);
826 spin_lock_init(&q->__queue_lock);
829 * By default initialize queue_lock to internal lock and driver can
830 * override it later if need be.
832 q->queue_lock = &q->__queue_lock;
835 * A queue starts its life with bypass turned on to avoid
836 * unnecessary bypass on/off overhead and nasty surprises during
837 * init. The initial bypass will be finished when the queue is
838 * registered by blk_register_queue().
841 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
843 init_waitqueue_head(&q->mq_freeze_wq);
846 * Init percpu_ref in atomic mode so that it's faster to shutdown.
847 * See blk_register_queue() for details.
849 if (percpu_ref_init(&q->q_usage_counter,
850 blk_queue_usage_counter_release,
851 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
854 if (blkcg_init_queue(q))
860 percpu_ref_exit(&q->q_usage_counter);
862 blk_free_queue_stats(q->stats);
864 bdi_put(q->backing_dev_info);
866 bioset_free(q->bio_split);
868 ida_simple_remove(&blk_queue_ida, q->id);
870 kmem_cache_free(blk_requestq_cachep, q);
873 EXPORT_SYMBOL(blk_alloc_queue_node);
876 * blk_init_queue - prepare a request queue for use with a block device
877 * @rfn: The function to be called to process requests that have been
878 * placed on the queue.
879 * @lock: Request queue spin lock
882 * If a block device wishes to use the standard request handling procedures,
883 * which sorts requests and coalesces adjacent requests, then it must
884 * call blk_init_queue(). The function @rfn will be called when there
885 * are requests on the queue that need to be processed. If the device
886 * supports plugging, then @rfn may not be called immediately when requests
887 * are available on the queue, but may be called at some time later instead.
888 * Plugged queues are generally unplugged when a buffer belonging to one
889 * of the requests on the queue is needed, or due to memory pressure.
891 * @rfn is not required, or even expected, to remove all requests off the
892 * queue, but only as many as it can handle at a time. If it does leave
893 * requests on the queue, it is responsible for arranging that the requests
894 * get dealt with eventually.
896 * The queue spin lock must be held while manipulating the requests on the
897 * request queue; this lock will be taken also from interrupt context, so irq
898 * disabling is needed for it.
900 * Function returns a pointer to the initialized request queue, or %NULL if
904 * blk_init_queue() must be paired with a blk_cleanup_queue() call
905 * when the block device is deactivated (such as at module unload).
908 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
910 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
912 EXPORT_SYMBOL(blk_init_queue);
914 struct request_queue *
915 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
917 struct request_queue *q;
919 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
925 q->queue_lock = lock;
926 if (blk_init_allocated_queue(q) < 0) {
927 blk_cleanup_queue(q);
933 EXPORT_SYMBOL(blk_init_queue_node);
935 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
938 int blk_init_allocated_queue(struct request_queue *q)
940 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
944 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
945 goto out_free_flush_queue;
947 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
948 goto out_exit_flush_rq;
950 INIT_WORK(&q->timeout_work, blk_timeout_work);
951 q->queue_flags |= QUEUE_FLAG_DEFAULT;
954 * This also sets hw/phys segments, boundary and size
956 blk_queue_make_request(q, blk_queue_bio);
958 q->sg_reserved_size = INT_MAX;
960 /* Protect q->elevator from elevator_change */
961 mutex_lock(&q->sysfs_lock);
964 if (elevator_init(q, NULL)) {
965 mutex_unlock(&q->sysfs_lock);
966 goto out_exit_flush_rq;
969 mutex_unlock(&q->sysfs_lock);
974 q->exit_rq_fn(q, q->fq->flush_rq);
975 out_free_flush_queue:
976 blk_free_flush_queue(q->fq);
979 EXPORT_SYMBOL(blk_init_allocated_queue);
981 bool blk_get_queue(struct request_queue *q)
983 if (likely(!blk_queue_dying(q))) {
990 EXPORT_SYMBOL(blk_get_queue);
992 static inline void blk_free_request(struct request_list *rl, struct request *rq)
994 if (rq->rq_flags & RQF_ELVPRIV) {
995 elv_put_request(rl->q, rq);
997 put_io_context(rq->elv.icq->ioc);
1000 mempool_free(rq, rl->rq_pool);
1004 * ioc_batching returns true if the ioc is a valid batching request and
1005 * should be given priority access to a request.
1007 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1013 * Make sure the process is able to allocate at least 1 request
1014 * even if the batch times out, otherwise we could theoretically
1017 return ioc->nr_batch_requests == q->nr_batching ||
1018 (ioc->nr_batch_requests > 0
1019 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1023 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1024 * will cause the process to be a "batcher" on all queues in the system. This
1025 * is the behaviour we want though - once it gets a wakeup it should be given
1028 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1030 if (!ioc || ioc_batching(q, ioc))
1033 ioc->nr_batch_requests = q->nr_batching;
1034 ioc->last_waited = jiffies;
1037 static void __freed_request(struct request_list *rl, int sync)
1039 struct request_queue *q = rl->q;
1041 if (rl->count[sync] < queue_congestion_off_threshold(q))
1042 blk_clear_congested(rl, sync);
1044 if (rl->count[sync] + 1 <= q->nr_requests) {
1045 if (waitqueue_active(&rl->wait[sync]))
1046 wake_up(&rl->wait[sync]);
1048 blk_clear_rl_full(rl, sync);
1053 * A request has just been released. Account for it, update the full and
1054 * congestion status, wake up any waiters. Called under q->queue_lock.
1056 static void freed_request(struct request_list *rl, bool sync,
1057 req_flags_t rq_flags)
1059 struct request_queue *q = rl->q;
1063 if (rq_flags & RQF_ELVPRIV)
1064 q->nr_rqs_elvpriv--;
1066 __freed_request(rl, sync);
1068 if (unlikely(rl->starved[sync ^ 1]))
1069 __freed_request(rl, sync ^ 1);
1072 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1074 struct request_list *rl;
1075 int on_thresh, off_thresh;
1077 spin_lock_irq(q->queue_lock);
1078 q->nr_requests = nr;
1079 blk_queue_congestion_threshold(q);
1080 on_thresh = queue_congestion_on_threshold(q);
1081 off_thresh = queue_congestion_off_threshold(q);
1083 blk_queue_for_each_rl(rl, q) {
1084 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1085 blk_set_congested(rl, BLK_RW_SYNC);
1086 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1087 blk_clear_congested(rl, BLK_RW_SYNC);
1089 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1090 blk_set_congested(rl, BLK_RW_ASYNC);
1091 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1092 blk_clear_congested(rl, BLK_RW_ASYNC);
1094 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1095 blk_set_rl_full(rl, BLK_RW_SYNC);
1097 blk_clear_rl_full(rl, BLK_RW_SYNC);
1098 wake_up(&rl->wait[BLK_RW_SYNC]);
1101 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1102 blk_set_rl_full(rl, BLK_RW_ASYNC);
1104 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1105 wake_up(&rl->wait[BLK_RW_ASYNC]);
1109 spin_unlock_irq(q->queue_lock);
1114 * __get_request - get a free request
1115 * @rl: request list to allocate from
1116 * @op: operation and flags
1117 * @bio: bio to allocate request for (can be %NULL)
1118 * @gfp_mask: allocation mask
1120 * Get a free request from @q. This function may fail under memory
1121 * pressure or if @q is dead.
1123 * Must be called with @q->queue_lock held and,
1124 * Returns ERR_PTR on failure, with @q->queue_lock held.
1125 * Returns request pointer on success, with @q->queue_lock *not held*.
1127 static struct request *__get_request(struct request_list *rl, unsigned int op,
1128 struct bio *bio, gfp_t gfp_mask)
1130 struct request_queue *q = rl->q;
1132 struct elevator_type *et = q->elevator->type;
1133 struct io_context *ioc = rq_ioc(bio);
1134 struct io_cq *icq = NULL;
1135 const bool is_sync = op_is_sync(op);
1137 req_flags_t rq_flags = RQF_ALLOCED;
1139 if (unlikely(blk_queue_dying(q)))
1140 return ERR_PTR(-ENODEV);
1142 may_queue = elv_may_queue(q, op);
1143 if (may_queue == ELV_MQUEUE_NO)
1146 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1147 if (rl->count[is_sync]+1 >= q->nr_requests) {
1149 * The queue will fill after this allocation, so set
1150 * it as full, and mark this process as "batching".
1151 * This process will be allowed to complete a batch of
1152 * requests, others will be blocked.
1154 if (!blk_rl_full(rl, is_sync)) {
1155 ioc_set_batching(q, ioc);
1156 blk_set_rl_full(rl, is_sync);
1158 if (may_queue != ELV_MQUEUE_MUST
1159 && !ioc_batching(q, ioc)) {
1161 * The queue is full and the allocating
1162 * process is not a "batcher", and not
1163 * exempted by the IO scheduler
1165 return ERR_PTR(-ENOMEM);
1169 blk_set_congested(rl, is_sync);
1173 * Only allow batching queuers to allocate up to 50% over the defined
1174 * limit of requests, otherwise we could have thousands of requests
1175 * allocated with any setting of ->nr_requests
1177 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1178 return ERR_PTR(-ENOMEM);
1180 q->nr_rqs[is_sync]++;
1181 rl->count[is_sync]++;
1182 rl->starved[is_sync] = 0;
1185 * Decide whether the new request will be managed by elevator. If
1186 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1187 * prevent the current elevator from being destroyed until the new
1188 * request is freed. This guarantees icq's won't be destroyed and
1189 * makes creating new ones safe.
1191 * Flush requests do not use the elevator so skip initialization.
1192 * This allows a request to share the flush and elevator data.
1194 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1195 * it will be created after releasing queue_lock.
1197 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1198 rq_flags |= RQF_ELVPRIV;
1199 q->nr_rqs_elvpriv++;
1200 if (et->icq_cache && ioc)
1201 icq = ioc_lookup_icq(ioc, q);
1204 if (blk_queue_io_stat(q))
1205 rq_flags |= RQF_IO_STAT;
1206 spin_unlock_irq(q->queue_lock);
1208 /* allocate and init request */
1209 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1214 blk_rq_set_rl(rq, rl);
1216 rq->rq_flags = rq_flags;
1219 if (rq_flags & RQF_ELVPRIV) {
1220 if (unlikely(et->icq_cache && !icq)) {
1222 icq = ioc_create_icq(ioc, q, gfp_mask);
1228 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1231 /* @rq->elv.icq holds io_context until @rq is freed */
1233 get_io_context(icq->ioc);
1237 * ioc may be NULL here, and ioc_batching will be false. That's
1238 * OK, if the queue is under the request limit then requests need
1239 * not count toward the nr_batch_requests limit. There will always
1240 * be some limit enforced by BLK_BATCH_TIME.
1242 if (ioc_batching(q, ioc))
1243 ioc->nr_batch_requests--;
1245 trace_block_getrq(q, bio, op);
1250 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1251 * and may fail indefinitely under memory pressure and thus
1252 * shouldn't stall IO. Treat this request as !elvpriv. This will
1253 * disturb iosched and blkcg but weird is bettern than dead.
1255 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1256 __func__, dev_name(q->backing_dev_info->dev));
1258 rq->rq_flags &= ~RQF_ELVPRIV;
1261 spin_lock_irq(q->queue_lock);
1262 q->nr_rqs_elvpriv--;
1263 spin_unlock_irq(q->queue_lock);
1268 * Allocation failed presumably due to memory. Undo anything we
1269 * might have messed up.
1271 * Allocating task should really be put onto the front of the wait
1272 * queue, but this is pretty rare.
1274 spin_lock_irq(q->queue_lock);
1275 freed_request(rl, is_sync, rq_flags);
1278 * in the very unlikely event that allocation failed and no
1279 * requests for this direction was pending, mark us starved so that
1280 * freeing of a request in the other direction will notice
1281 * us. another possible fix would be to split the rq mempool into
1285 if (unlikely(rl->count[is_sync] == 0))
1286 rl->starved[is_sync] = 1;
1287 return ERR_PTR(-ENOMEM);
1291 * get_request - get a free request
1292 * @q: request_queue to allocate request from
1293 * @op: operation and flags
1294 * @bio: bio to allocate request for (can be %NULL)
1295 * @gfp_mask: allocation mask
1297 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1298 * this function keeps retrying under memory pressure and fails iff @q is dead.
1300 * Must be called with @q->queue_lock held and,
1301 * Returns ERR_PTR on failure, with @q->queue_lock held.
1302 * Returns request pointer on success, with @q->queue_lock *not held*.
1304 static struct request *get_request(struct request_queue *q, unsigned int op,
1305 struct bio *bio, gfp_t gfp_mask)
1307 const bool is_sync = op_is_sync(op);
1309 struct request_list *rl;
1312 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1314 rq = __get_request(rl, op, bio, gfp_mask);
1318 if (op & REQ_NOWAIT) {
1320 return ERR_PTR(-EAGAIN);
1323 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1328 /* wait on @rl and retry */
1329 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1330 TASK_UNINTERRUPTIBLE);
1332 trace_block_sleeprq(q, bio, op);
1334 spin_unlock_irq(q->queue_lock);
1338 * After sleeping, we become a "batching" process and will be able
1339 * to allocate at least one request, and up to a big batch of them
1340 * for a small period time. See ioc_batching, ioc_set_batching
1342 ioc_set_batching(q, current->io_context);
1344 spin_lock_irq(q->queue_lock);
1345 finish_wait(&rl->wait[is_sync], &wait);
1350 static struct request *blk_old_get_request(struct request_queue *q,
1351 unsigned int op, gfp_t gfp_mask)
1355 /* create ioc upfront */
1356 create_io_context(gfp_mask, q->node);
1358 spin_lock_irq(q->queue_lock);
1359 rq = get_request(q, op, NULL, gfp_mask);
1361 spin_unlock_irq(q->queue_lock);
1365 /* q->queue_lock is unlocked at this point */
1367 rq->__sector = (sector_t) -1;
1368 rq->bio = rq->biotail = NULL;
1372 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1375 struct request *req;
1378 req = blk_mq_alloc_request(q, op,
1379 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1380 0 : BLK_MQ_REQ_NOWAIT);
1381 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1382 q->mq_ops->initialize_rq_fn(req);
1384 req = blk_old_get_request(q, op, gfp_mask);
1385 if (!IS_ERR(req) && q->initialize_rq_fn)
1386 q->initialize_rq_fn(req);
1391 EXPORT_SYMBOL(blk_get_request);
1394 * blk_requeue_request - put a request back on queue
1395 * @q: request queue where request should be inserted
1396 * @rq: request to be inserted
1399 * Drivers often keep queueing requests until the hardware cannot accept
1400 * more, when that condition happens we need to put the request back
1401 * on the queue. Must be called with queue lock held.
1403 void blk_requeue_request(struct request_queue *q, struct request *rq)
1405 blk_delete_timer(rq);
1406 blk_clear_rq_complete(rq);
1407 trace_block_rq_requeue(q, rq);
1408 wbt_requeue(q->rq_wb, &rq->issue_stat);
1410 if (rq->rq_flags & RQF_QUEUED)
1411 blk_queue_end_tag(q, rq);
1413 BUG_ON(blk_queued_rq(rq));
1415 elv_requeue_request(q, rq);
1417 EXPORT_SYMBOL(blk_requeue_request);
1419 static void add_acct_request(struct request_queue *q, struct request *rq,
1422 blk_account_io_start(rq, true);
1423 __elv_add_request(q, rq, where);
1426 static void part_round_stats_single(int cpu, struct hd_struct *part,
1431 if (now == part->stamp)
1434 inflight = part_in_flight(part);
1436 __part_stat_add(cpu, part, time_in_queue,
1437 inflight * (now - part->stamp));
1438 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1444 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1445 * @cpu: cpu number for stats access
1446 * @part: target partition
1448 * The average IO queue length and utilisation statistics are maintained
1449 * by observing the current state of the queue length and the amount of
1450 * time it has been in this state for.
1452 * Normally, that accounting is done on IO completion, but that can result
1453 * in more than a second's worth of IO being accounted for within any one
1454 * second, leading to >100% utilisation. To deal with that, we call this
1455 * function to do a round-off before returning the results when reading
1456 * /proc/diskstats. This accounts immediately for all queue usage up to
1457 * the current jiffies and restarts the counters again.
1459 void part_round_stats(int cpu, struct hd_struct *part)
1461 unsigned long now = jiffies;
1464 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1465 part_round_stats_single(cpu, part, now);
1467 EXPORT_SYMBOL_GPL(part_round_stats);
1470 static void blk_pm_put_request(struct request *rq)
1472 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1473 pm_runtime_mark_last_busy(rq->q->dev);
1476 static inline void blk_pm_put_request(struct request *rq) {}
1480 * queue lock must be held
1482 void __blk_put_request(struct request_queue *q, struct request *req)
1484 req_flags_t rq_flags = req->rq_flags;
1490 blk_mq_free_request(req);
1494 blk_pm_put_request(req);
1496 elv_completed_request(q, req);
1498 /* this is a bio leak */
1499 WARN_ON(req->bio != NULL);
1501 wbt_done(q->rq_wb, &req->issue_stat);
1504 * Request may not have originated from ll_rw_blk. if not,
1505 * it didn't come out of our reserved rq pools
1507 if (rq_flags & RQF_ALLOCED) {
1508 struct request_list *rl = blk_rq_rl(req);
1509 bool sync = op_is_sync(req->cmd_flags);
1511 BUG_ON(!list_empty(&req->queuelist));
1512 BUG_ON(ELV_ON_HASH(req));
1514 blk_free_request(rl, req);
1515 freed_request(rl, sync, rq_flags);
1519 EXPORT_SYMBOL_GPL(__blk_put_request);
1521 void blk_put_request(struct request *req)
1523 struct request_queue *q = req->q;
1526 blk_mq_free_request(req);
1528 unsigned long flags;
1530 spin_lock_irqsave(q->queue_lock, flags);
1531 __blk_put_request(q, req);
1532 spin_unlock_irqrestore(q->queue_lock, flags);
1535 EXPORT_SYMBOL(blk_put_request);
1537 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1540 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1542 if (!ll_back_merge_fn(q, req, bio))
1545 trace_block_bio_backmerge(q, req, bio);
1547 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1548 blk_rq_set_mixed_merge(req);
1550 req->biotail->bi_next = bio;
1552 req->__data_len += bio->bi_iter.bi_size;
1553 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1555 blk_account_io_start(req, false);
1559 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1562 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1564 if (!ll_front_merge_fn(q, req, bio))
1567 trace_block_bio_frontmerge(q, req, bio);
1569 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1570 blk_rq_set_mixed_merge(req);
1572 bio->bi_next = req->bio;
1575 req->__sector = bio->bi_iter.bi_sector;
1576 req->__data_len += bio->bi_iter.bi_size;
1577 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1579 blk_account_io_start(req, false);
1583 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1586 unsigned short segments = blk_rq_nr_discard_segments(req);
1588 if (segments >= queue_max_discard_segments(q))
1590 if (blk_rq_sectors(req) + bio_sectors(bio) >
1591 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1594 req->biotail->bi_next = bio;
1596 req->__data_len += bio->bi_iter.bi_size;
1597 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1598 req->nr_phys_segments = segments + 1;
1600 blk_account_io_start(req, false);
1603 req_set_nomerge(q, req);
1608 * blk_attempt_plug_merge - try to merge with %current's plugged list
1609 * @q: request_queue new bio is being queued at
1610 * @bio: new bio being queued
1611 * @request_count: out parameter for number of traversed plugged requests
1612 * @same_queue_rq: pointer to &struct request that gets filled in when
1613 * another request associated with @q is found on the plug list
1614 * (optional, may be %NULL)
1616 * Determine whether @bio being queued on @q can be merged with a request
1617 * on %current's plugged list. Returns %true if merge was successful,
1620 * Plugging coalesces IOs from the same issuer for the same purpose without
1621 * going through @q->queue_lock. As such it's more of an issuing mechanism
1622 * than scheduling, and the request, while may have elvpriv data, is not
1623 * added on the elevator at this point. In addition, we don't have
1624 * reliable access to the elevator outside queue lock. Only check basic
1625 * merging parameters without querying the elevator.
1627 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1629 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1630 unsigned int *request_count,
1631 struct request **same_queue_rq)
1633 struct blk_plug *plug;
1635 struct list_head *plug_list;
1637 plug = current->plug;
1643 plug_list = &plug->mq_list;
1645 plug_list = &plug->list;
1647 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1648 bool merged = false;
1653 * Only blk-mq multiple hardware queues case checks the
1654 * rq in the same queue, there should be only one such
1658 *same_queue_rq = rq;
1661 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1664 switch (blk_try_merge(rq, bio)) {
1665 case ELEVATOR_BACK_MERGE:
1666 merged = bio_attempt_back_merge(q, rq, bio);
1668 case ELEVATOR_FRONT_MERGE:
1669 merged = bio_attempt_front_merge(q, rq, bio);
1671 case ELEVATOR_DISCARD_MERGE:
1672 merged = bio_attempt_discard_merge(q, rq, bio);
1685 unsigned int blk_plug_queued_count(struct request_queue *q)
1687 struct blk_plug *plug;
1689 struct list_head *plug_list;
1690 unsigned int ret = 0;
1692 plug = current->plug;
1697 plug_list = &plug->mq_list;
1699 plug_list = &plug->list;
1701 list_for_each_entry(rq, plug_list, queuelist) {
1709 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1711 struct io_context *ioc = rq_ioc(bio);
1713 if (bio->bi_opf & REQ_RAHEAD)
1714 req->cmd_flags |= REQ_FAILFAST_MASK;
1716 req->__sector = bio->bi_iter.bi_sector;
1717 if (ioprio_valid(bio_prio(bio)))
1718 req->ioprio = bio_prio(bio);
1720 req->ioprio = ioc->ioprio;
1722 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1723 blk_rq_bio_prep(req->q, req, bio);
1725 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1727 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1729 struct blk_plug *plug;
1730 int where = ELEVATOR_INSERT_SORT;
1731 struct request *req, *free;
1732 unsigned int request_count = 0;
1733 unsigned int wb_acct;
1736 * low level driver can indicate that it wants pages above a
1737 * certain limit bounced to low memory (ie for highmem, or even
1738 * ISA dma in theory)
1740 blk_queue_bounce(q, &bio);
1742 blk_queue_split(q, &bio);
1744 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1745 bio->bi_status = BLK_STS_IOERR;
1747 return BLK_QC_T_NONE;
1750 if (op_is_flush(bio->bi_opf)) {
1751 spin_lock_irq(q->queue_lock);
1752 where = ELEVATOR_INSERT_FLUSH;
1757 * Check if we can merge with the plugged list before grabbing
1760 if (!blk_queue_nomerges(q)) {
1761 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1762 return BLK_QC_T_NONE;
1764 request_count = blk_plug_queued_count(q);
1766 spin_lock_irq(q->queue_lock);
1768 switch (elv_merge(q, &req, bio)) {
1769 case ELEVATOR_BACK_MERGE:
1770 if (!bio_attempt_back_merge(q, req, bio))
1772 elv_bio_merged(q, req, bio);
1773 free = attempt_back_merge(q, req);
1775 __blk_put_request(q, free);
1777 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1779 case ELEVATOR_FRONT_MERGE:
1780 if (!bio_attempt_front_merge(q, req, bio))
1782 elv_bio_merged(q, req, bio);
1783 free = attempt_front_merge(q, req);
1785 __blk_put_request(q, free);
1787 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1794 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1797 * Grab a free request. This is might sleep but can not fail.
1798 * Returns with the queue unlocked.
1800 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1802 __wbt_done(q->rq_wb, wb_acct);
1803 if (PTR_ERR(req) == -ENOMEM)
1804 bio->bi_status = BLK_STS_RESOURCE;
1806 bio->bi_status = BLK_STS_IOERR;
1811 wbt_track(&req->issue_stat, wb_acct);
1814 * After dropping the lock and possibly sleeping here, our request
1815 * may now be mergeable after it had proven unmergeable (above).
1816 * We don't worry about that case for efficiency. It won't happen
1817 * often, and the elevators are able to handle it.
1819 blk_init_request_from_bio(req, bio);
1821 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1822 req->cpu = raw_smp_processor_id();
1824 plug = current->plug;
1827 * If this is the first request added after a plug, fire
1830 * @request_count may become stale because of schedule
1831 * out, so check plug list again.
1833 if (!request_count || list_empty(&plug->list))
1834 trace_block_plug(q);
1836 struct request *last = list_entry_rq(plug->list.prev);
1837 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1838 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1839 blk_flush_plug_list(plug, false);
1840 trace_block_plug(q);
1843 list_add_tail(&req->queuelist, &plug->list);
1844 blk_account_io_start(req, true);
1846 spin_lock_irq(q->queue_lock);
1847 add_acct_request(q, req, where);
1850 spin_unlock_irq(q->queue_lock);
1853 return BLK_QC_T_NONE;
1857 * If bio->bi_dev is a partition, remap the location
1859 static inline void blk_partition_remap(struct bio *bio)
1861 struct block_device *bdev = bio->bi_bdev;
1864 * Zone reset does not include bi_size so bio_sectors() is always 0.
1865 * Include a test for the reset op code and perform the remap if needed.
1867 if (bdev != bdev->bd_contains &&
1868 (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1869 struct hd_struct *p = bdev->bd_part;
1871 bio->bi_iter.bi_sector += p->start_sect;
1872 bio->bi_bdev = bdev->bd_contains;
1874 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1876 bio->bi_iter.bi_sector - p->start_sect);
1880 static void handle_bad_sector(struct bio *bio)
1882 char b[BDEVNAME_SIZE];
1884 printk(KERN_INFO "attempt to access beyond end of device\n");
1885 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1886 bdevname(bio->bi_bdev, b),
1888 (unsigned long long)bio_end_sector(bio),
1889 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1892 #ifdef CONFIG_FAIL_MAKE_REQUEST
1894 static DECLARE_FAULT_ATTR(fail_make_request);
1896 static int __init setup_fail_make_request(char *str)
1898 return setup_fault_attr(&fail_make_request, str);
1900 __setup("fail_make_request=", setup_fail_make_request);
1902 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1904 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1907 static int __init fail_make_request_debugfs(void)
1909 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1910 NULL, &fail_make_request);
1912 return PTR_ERR_OR_ZERO(dir);
1915 late_initcall(fail_make_request_debugfs);
1917 #else /* CONFIG_FAIL_MAKE_REQUEST */
1919 static inline bool should_fail_request(struct hd_struct *part,
1925 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1928 * Check whether this bio extends beyond the end of the device.
1930 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1937 /* Test device or partition size, when known. */
1938 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1940 sector_t sector = bio->bi_iter.bi_sector;
1942 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1944 * This may well happen - the kernel calls bread()
1945 * without checking the size of the device, e.g., when
1946 * mounting a device.
1948 handle_bad_sector(bio);
1956 static noinline_for_stack bool
1957 generic_make_request_checks(struct bio *bio)
1959 struct request_queue *q;
1960 int nr_sectors = bio_sectors(bio);
1961 blk_status_t status = BLK_STS_IOERR;
1962 char b[BDEVNAME_SIZE];
1963 struct hd_struct *part;
1967 if (bio_check_eod(bio, nr_sectors))
1970 q = bdev_get_queue(bio->bi_bdev);
1973 "generic_make_request: Trying to access "
1974 "nonexistent block-device %s (%Lu)\n",
1975 bdevname(bio->bi_bdev, b),
1976 (long long) bio->bi_iter.bi_sector);
1981 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
1982 * if queue is not a request based queue.
1985 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
1988 part = bio->bi_bdev->bd_part;
1989 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1990 should_fail_request(&part_to_disk(part)->part0,
1991 bio->bi_iter.bi_size))
1995 * If this device has partitions, remap block n
1996 * of partition p to block n+start(p) of the disk.
1998 blk_partition_remap(bio);
2000 if (bio_check_eod(bio, nr_sectors))
2004 * Filter flush bio's early so that make_request based
2005 * drivers without flush support don't have to worry
2008 if (op_is_flush(bio->bi_opf) &&
2009 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2010 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2012 status = BLK_STS_OK;
2017 switch (bio_op(bio)) {
2018 case REQ_OP_DISCARD:
2019 if (!blk_queue_discard(q))
2022 case REQ_OP_SECURE_ERASE:
2023 if (!blk_queue_secure_erase(q))
2026 case REQ_OP_WRITE_SAME:
2027 if (!bdev_write_same(bio->bi_bdev))
2030 case REQ_OP_ZONE_REPORT:
2031 case REQ_OP_ZONE_RESET:
2032 if (!bdev_is_zoned(bio->bi_bdev))
2035 case REQ_OP_WRITE_ZEROES:
2036 if (!bdev_write_zeroes_sectors(bio->bi_bdev))
2044 * Various block parts want %current->io_context and lazy ioc
2045 * allocation ends up trading a lot of pain for a small amount of
2046 * memory. Just allocate it upfront. This may fail and block
2047 * layer knows how to live with it.
2049 create_io_context(GFP_ATOMIC, q->node);
2051 if (!blkcg_bio_issue_check(q, bio))
2054 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2055 trace_block_bio_queue(q, bio);
2056 /* Now that enqueuing has been traced, we need to trace
2057 * completion as well.
2059 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2064 status = BLK_STS_NOTSUPP;
2066 bio->bi_status = status;
2072 * generic_make_request - hand a buffer to its device driver for I/O
2073 * @bio: The bio describing the location in memory and on the device.
2075 * generic_make_request() is used to make I/O requests of block
2076 * devices. It is passed a &struct bio, which describes the I/O that needs
2079 * generic_make_request() does not return any status. The
2080 * success/failure status of the request, along with notification of
2081 * completion, is delivered asynchronously through the bio->bi_end_io
2082 * function described (one day) else where.
2084 * The caller of generic_make_request must make sure that bi_io_vec
2085 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2086 * set to describe the device address, and the
2087 * bi_end_io and optionally bi_private are set to describe how
2088 * completion notification should be signaled.
2090 * generic_make_request and the drivers it calls may use bi_next if this
2091 * bio happens to be merged with someone else, and may resubmit the bio to
2092 * a lower device by calling into generic_make_request recursively, which
2093 * means the bio should NOT be touched after the call to ->make_request_fn.
2095 blk_qc_t generic_make_request(struct bio *bio)
2098 * bio_list_on_stack[0] contains bios submitted by the current
2100 * bio_list_on_stack[1] contains bios that were submitted before
2101 * the current make_request_fn, but that haven't been processed
2104 struct bio_list bio_list_on_stack[2];
2105 blk_qc_t ret = BLK_QC_T_NONE;
2107 if (!generic_make_request_checks(bio))
2111 * We only want one ->make_request_fn to be active at a time, else
2112 * stack usage with stacked devices could be a problem. So use
2113 * current->bio_list to keep a list of requests submited by a
2114 * make_request_fn function. current->bio_list is also used as a
2115 * flag to say if generic_make_request is currently active in this
2116 * task or not. If it is NULL, then no make_request is active. If
2117 * it is non-NULL, then a make_request is active, and new requests
2118 * should be added at the tail
2120 if (current->bio_list) {
2121 bio_list_add(¤t->bio_list[0], bio);
2125 /* following loop may be a bit non-obvious, and so deserves some
2127 * Before entering the loop, bio->bi_next is NULL (as all callers
2128 * ensure that) so we have a list with a single bio.
2129 * We pretend that we have just taken it off a longer list, so
2130 * we assign bio_list to a pointer to the bio_list_on_stack,
2131 * thus initialising the bio_list of new bios to be
2132 * added. ->make_request() may indeed add some more bios
2133 * through a recursive call to generic_make_request. If it
2134 * did, we find a non-NULL value in bio_list and re-enter the loop
2135 * from the top. In this case we really did just take the bio
2136 * of the top of the list (no pretending) and so remove it from
2137 * bio_list, and call into ->make_request() again.
2139 BUG_ON(bio->bi_next);
2140 bio_list_init(&bio_list_on_stack[0]);
2141 current->bio_list = bio_list_on_stack;
2143 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2145 if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) {
2146 struct bio_list lower, same;
2148 /* Create a fresh bio_list for all subordinate requests */
2149 bio_list_on_stack[1] = bio_list_on_stack[0];
2150 bio_list_init(&bio_list_on_stack[0]);
2151 ret = q->make_request_fn(q, bio);
2155 /* sort new bios into those for a lower level
2156 * and those for the same level
2158 bio_list_init(&lower);
2159 bio_list_init(&same);
2160 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2161 if (q == bdev_get_queue(bio->bi_bdev))
2162 bio_list_add(&same, bio);
2164 bio_list_add(&lower, bio);
2165 /* now assemble so we handle the lowest level first */
2166 bio_list_merge(&bio_list_on_stack[0], &lower);
2167 bio_list_merge(&bio_list_on_stack[0], &same);
2168 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2170 if (unlikely(!blk_queue_dying(q) &&
2171 (bio->bi_opf & REQ_NOWAIT)))
2172 bio_wouldblock_error(bio);
2176 bio = bio_list_pop(&bio_list_on_stack[0]);
2178 current->bio_list = NULL; /* deactivate */
2183 EXPORT_SYMBOL(generic_make_request);
2186 * submit_bio - submit a bio to the block device layer for I/O
2187 * @bio: The &struct bio which describes the I/O
2189 * submit_bio() is very similar in purpose to generic_make_request(), and
2190 * uses that function to do most of the work. Both are fairly rough
2191 * interfaces; @bio must be presetup and ready for I/O.
2194 blk_qc_t submit_bio(struct bio *bio)
2197 * If it's a regular read/write or a barrier with data attached,
2198 * go through the normal accounting stuff before submission.
2200 if (bio_has_data(bio)) {
2203 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2204 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2206 count = bio_sectors(bio);
2208 if (op_is_write(bio_op(bio))) {
2209 count_vm_events(PGPGOUT, count);
2211 task_io_account_read(bio->bi_iter.bi_size);
2212 count_vm_events(PGPGIN, count);
2215 if (unlikely(block_dump)) {
2216 char b[BDEVNAME_SIZE];
2217 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2218 current->comm, task_pid_nr(current),
2219 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2220 (unsigned long long)bio->bi_iter.bi_sector,
2221 bdevname(bio->bi_bdev, b),
2226 return generic_make_request(bio);
2228 EXPORT_SYMBOL(submit_bio);
2231 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2232 * for new the queue limits
2234 * @rq: the request being checked
2237 * @rq may have been made based on weaker limitations of upper-level queues
2238 * in request stacking drivers, and it may violate the limitation of @q.
2239 * Since the block layer and the underlying device driver trust @rq
2240 * after it is inserted to @q, it should be checked against @q before
2241 * the insertion using this generic function.
2243 * Request stacking drivers like request-based dm may change the queue
2244 * limits when retrying requests on other queues. Those requests need
2245 * to be checked against the new queue limits again during dispatch.
2247 static int blk_cloned_rq_check_limits(struct request_queue *q,
2250 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2251 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2256 * queue's settings related to segment counting like q->bounce_pfn
2257 * may differ from that of other stacking queues.
2258 * Recalculate it to check the request correctly on this queue's
2261 blk_recalc_rq_segments(rq);
2262 if (rq->nr_phys_segments > queue_max_segments(q)) {
2263 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2271 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2272 * @q: the queue to submit the request
2273 * @rq: the request being queued
2275 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2277 unsigned long flags;
2278 int where = ELEVATOR_INSERT_BACK;
2280 if (blk_cloned_rq_check_limits(q, rq))
2281 return BLK_STS_IOERR;
2284 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2285 return BLK_STS_IOERR;
2288 if (blk_queue_io_stat(q))
2289 blk_account_io_start(rq, true);
2290 blk_mq_sched_insert_request(rq, false, true, false, false);
2294 spin_lock_irqsave(q->queue_lock, flags);
2295 if (unlikely(blk_queue_dying(q))) {
2296 spin_unlock_irqrestore(q->queue_lock, flags);
2297 return BLK_STS_IOERR;
2301 * Submitting request must be dequeued before calling this function
2302 * because it will be linked to another request_queue
2304 BUG_ON(blk_queued_rq(rq));
2306 if (op_is_flush(rq->cmd_flags))
2307 where = ELEVATOR_INSERT_FLUSH;
2309 add_acct_request(q, rq, where);
2310 if (where == ELEVATOR_INSERT_FLUSH)
2312 spin_unlock_irqrestore(q->queue_lock, flags);
2316 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2319 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2320 * @rq: request to examine
2323 * A request could be merge of IOs which require different failure
2324 * handling. This function determines the number of bytes which
2325 * can be failed from the beginning of the request without
2326 * crossing into area which need to be retried further.
2329 * The number of bytes to fail.
2332 * queue_lock must be held.
2334 unsigned int blk_rq_err_bytes(const struct request *rq)
2336 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2337 unsigned int bytes = 0;
2340 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2341 return blk_rq_bytes(rq);
2344 * Currently the only 'mixing' which can happen is between
2345 * different fastfail types. We can safely fail portions
2346 * which have all the failfast bits that the first one has -
2347 * the ones which are at least as eager to fail as the first
2350 for (bio = rq->bio; bio; bio = bio->bi_next) {
2351 if ((bio->bi_opf & ff) != ff)
2353 bytes += bio->bi_iter.bi_size;
2356 /* this could lead to infinite loop */
2357 BUG_ON(blk_rq_bytes(rq) && !bytes);
2360 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2362 void blk_account_io_completion(struct request *req, unsigned int bytes)
2364 if (blk_do_io_stat(req)) {
2365 const int rw = rq_data_dir(req);
2366 struct hd_struct *part;
2369 cpu = part_stat_lock();
2371 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2376 void blk_account_io_done(struct request *req)
2379 * Account IO completion. flush_rq isn't accounted as a
2380 * normal IO on queueing nor completion. Accounting the
2381 * containing request is enough.
2383 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2384 unsigned long duration = jiffies - req->start_time;
2385 const int rw = rq_data_dir(req);
2386 struct hd_struct *part;
2389 cpu = part_stat_lock();
2392 part_stat_inc(cpu, part, ios[rw]);
2393 part_stat_add(cpu, part, ticks[rw], duration);
2394 part_round_stats(cpu, part);
2395 part_dec_in_flight(part, rw);
2397 hd_struct_put(part);
2404 * Don't process normal requests when queue is suspended
2405 * or in the process of suspending/resuming
2407 static struct request *blk_pm_peek_request(struct request_queue *q,
2410 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2411 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2417 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2424 void blk_account_io_start(struct request *rq, bool new_io)
2426 struct hd_struct *part;
2427 int rw = rq_data_dir(rq);
2430 if (!blk_do_io_stat(rq))
2433 cpu = part_stat_lock();
2437 part_stat_inc(cpu, part, merges[rw]);
2439 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2440 if (!hd_struct_try_get(part)) {
2442 * The partition is already being removed,
2443 * the request will be accounted on the disk only
2445 * We take a reference on disk->part0 although that
2446 * partition will never be deleted, so we can treat
2447 * it as any other partition.
2449 part = &rq->rq_disk->part0;
2450 hd_struct_get(part);
2452 part_round_stats(cpu, part);
2453 part_inc_in_flight(part, rw);
2461 * blk_peek_request - peek at the top of a request queue
2462 * @q: request queue to peek at
2465 * Return the request at the top of @q. The returned request
2466 * should be started using blk_start_request() before LLD starts
2470 * Pointer to the request at the top of @q if available. Null
2474 * queue_lock must be held.
2476 struct request *blk_peek_request(struct request_queue *q)
2481 while ((rq = __elv_next_request(q)) != NULL) {
2483 rq = blk_pm_peek_request(q, rq);
2487 if (!(rq->rq_flags & RQF_STARTED)) {
2489 * This is the first time the device driver
2490 * sees this request (possibly after
2491 * requeueing). Notify IO scheduler.
2493 if (rq->rq_flags & RQF_SORTED)
2494 elv_activate_rq(q, rq);
2497 * just mark as started even if we don't start
2498 * it, a request that has been delayed should
2499 * not be passed by new incoming requests
2501 rq->rq_flags |= RQF_STARTED;
2502 trace_block_rq_issue(q, rq);
2505 if (!q->boundary_rq || q->boundary_rq == rq) {
2506 q->end_sector = rq_end_sector(rq);
2507 q->boundary_rq = NULL;
2510 if (rq->rq_flags & RQF_DONTPREP)
2513 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2515 * make sure space for the drain appears we
2516 * know we can do this because max_hw_segments
2517 * has been adjusted to be one fewer than the
2520 rq->nr_phys_segments++;
2526 ret = q->prep_rq_fn(q, rq);
2527 if (ret == BLKPREP_OK) {
2529 } else if (ret == BLKPREP_DEFER) {
2531 * the request may have been (partially) prepped.
2532 * we need to keep this request in the front to
2533 * avoid resource deadlock. RQF_STARTED will
2534 * prevent other fs requests from passing this one.
2536 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2537 !(rq->rq_flags & RQF_DONTPREP)) {
2539 * remove the space for the drain we added
2540 * so that we don't add it again
2542 --rq->nr_phys_segments;
2547 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2548 rq->rq_flags |= RQF_QUIET;
2550 * Mark this request as started so we don't trigger
2551 * any debug logic in the end I/O path.
2553 blk_start_request(rq);
2554 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2555 BLK_STS_TARGET : BLK_STS_IOERR);
2557 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2564 EXPORT_SYMBOL(blk_peek_request);
2566 void blk_dequeue_request(struct request *rq)
2568 struct request_queue *q = rq->q;
2570 BUG_ON(list_empty(&rq->queuelist));
2571 BUG_ON(ELV_ON_HASH(rq));
2573 list_del_init(&rq->queuelist);
2576 * the time frame between a request being removed from the lists
2577 * and to it is freed is accounted as io that is in progress at
2580 if (blk_account_rq(rq)) {
2581 q->in_flight[rq_is_sync(rq)]++;
2582 set_io_start_time_ns(rq);
2587 * blk_start_request - start request processing on the driver
2588 * @req: request to dequeue
2591 * Dequeue @req and start timeout timer on it. This hands off the
2592 * request to the driver.
2594 * Block internal functions which don't want to start timer should
2595 * call blk_dequeue_request().
2598 * queue_lock must be held.
2600 void blk_start_request(struct request *req)
2602 blk_dequeue_request(req);
2604 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2605 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2606 req->rq_flags |= RQF_STATS;
2607 wbt_issue(req->q->rq_wb, &req->issue_stat);
2610 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2613 EXPORT_SYMBOL(blk_start_request);
2616 * blk_fetch_request - fetch a request from a request queue
2617 * @q: request queue to fetch a request from
2620 * Return the request at the top of @q. The request is started on
2621 * return and LLD can start processing it immediately.
2624 * Pointer to the request at the top of @q if available. Null
2628 * queue_lock must be held.
2630 struct request *blk_fetch_request(struct request_queue *q)
2634 rq = blk_peek_request(q);
2636 blk_start_request(rq);
2639 EXPORT_SYMBOL(blk_fetch_request);
2642 * blk_update_request - Special helper function for request stacking drivers
2643 * @req: the request being processed
2644 * @error: block status code
2645 * @nr_bytes: number of bytes to complete @req
2648 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2649 * the request structure even if @req doesn't have leftover.
2650 * If @req has leftover, sets it up for the next range of segments.
2652 * This special helper function is only for request stacking drivers
2653 * (e.g. request-based dm) so that they can handle partial completion.
2654 * Actual device drivers should use blk_end_request instead.
2656 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2657 * %false return from this function.
2660 * %false - this request doesn't have any more data
2661 * %true - this request has more data
2663 bool blk_update_request(struct request *req, blk_status_t error,
2664 unsigned int nr_bytes)
2668 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2673 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2674 !(req->rq_flags & RQF_QUIET)))
2675 print_req_error(req, error);
2677 blk_account_io_completion(req, nr_bytes);
2681 struct bio *bio = req->bio;
2682 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2684 if (bio_bytes == bio->bi_iter.bi_size)
2685 req->bio = bio->bi_next;
2687 /* Completion has already been traced */
2688 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2689 req_bio_endio(req, bio, bio_bytes, error);
2691 total_bytes += bio_bytes;
2692 nr_bytes -= bio_bytes;
2703 * Reset counters so that the request stacking driver
2704 * can find how many bytes remain in the request
2707 req->__data_len = 0;
2711 req->__data_len -= total_bytes;
2713 /* update sector only for requests with clear definition of sector */
2714 if (!blk_rq_is_passthrough(req))
2715 req->__sector += total_bytes >> 9;
2717 /* mixed attributes always follow the first bio */
2718 if (req->rq_flags & RQF_MIXED_MERGE) {
2719 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2720 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2723 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2725 * If total number of sectors is less than the first segment
2726 * size, something has gone terribly wrong.
2728 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2729 blk_dump_rq_flags(req, "request botched");
2730 req->__data_len = blk_rq_cur_bytes(req);
2733 /* recalculate the number of segments */
2734 blk_recalc_rq_segments(req);
2739 EXPORT_SYMBOL_GPL(blk_update_request);
2741 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2742 unsigned int nr_bytes,
2743 unsigned int bidi_bytes)
2745 if (blk_update_request(rq, error, nr_bytes))
2748 /* Bidi request must be completed as a whole */
2749 if (unlikely(blk_bidi_rq(rq)) &&
2750 blk_update_request(rq->next_rq, error, bidi_bytes))
2753 if (blk_queue_add_random(rq->q))
2754 add_disk_randomness(rq->rq_disk);
2760 * blk_unprep_request - unprepare a request
2763 * This function makes a request ready for complete resubmission (or
2764 * completion). It happens only after all error handling is complete,
2765 * so represents the appropriate moment to deallocate any resources
2766 * that were allocated to the request in the prep_rq_fn. The queue
2767 * lock is held when calling this.
2769 void blk_unprep_request(struct request *req)
2771 struct request_queue *q = req->q;
2773 req->rq_flags &= ~RQF_DONTPREP;
2774 if (q->unprep_rq_fn)
2775 q->unprep_rq_fn(q, req);
2777 EXPORT_SYMBOL_GPL(blk_unprep_request);
2780 * queue lock must be held
2782 void blk_finish_request(struct request *req, blk_status_t error)
2784 struct request_queue *q = req->q;
2786 if (req->rq_flags & RQF_STATS)
2789 if (req->rq_flags & RQF_QUEUED)
2790 blk_queue_end_tag(q, req);
2792 BUG_ON(blk_queued_rq(req));
2794 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
2795 laptop_io_completion(req->q->backing_dev_info);
2797 blk_delete_timer(req);
2799 if (req->rq_flags & RQF_DONTPREP)
2800 blk_unprep_request(req);
2802 blk_account_io_done(req);
2805 wbt_done(req->q->rq_wb, &req->issue_stat);
2806 req->end_io(req, error);
2808 if (blk_bidi_rq(req))
2809 __blk_put_request(req->next_rq->q, req->next_rq);
2811 __blk_put_request(q, req);
2814 EXPORT_SYMBOL(blk_finish_request);
2817 * blk_end_bidi_request - Complete a bidi request
2818 * @rq: the request to complete
2819 * @error: block status code
2820 * @nr_bytes: number of bytes to complete @rq
2821 * @bidi_bytes: number of bytes to complete @rq->next_rq
2824 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2825 * Drivers that supports bidi can safely call this member for any
2826 * type of request, bidi or uni. In the later case @bidi_bytes is
2830 * %false - we are done with this request
2831 * %true - still buffers pending for this request
2833 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
2834 unsigned int nr_bytes, unsigned int bidi_bytes)
2836 struct request_queue *q = rq->q;
2837 unsigned long flags;
2839 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2842 spin_lock_irqsave(q->queue_lock, flags);
2843 blk_finish_request(rq, error);
2844 spin_unlock_irqrestore(q->queue_lock, flags);
2850 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2851 * @rq: the request to complete
2852 * @error: block status code
2853 * @nr_bytes: number of bytes to complete @rq
2854 * @bidi_bytes: number of bytes to complete @rq->next_rq
2857 * Identical to blk_end_bidi_request() except that queue lock is
2858 * assumed to be locked on entry and remains so on return.
2861 * %false - we are done with this request
2862 * %true - still buffers pending for this request
2864 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
2865 unsigned int nr_bytes, unsigned int bidi_bytes)
2867 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2870 blk_finish_request(rq, error);
2876 * blk_end_request - Helper function for drivers to complete the request.
2877 * @rq: the request being processed
2878 * @error: block status code
2879 * @nr_bytes: number of bytes to complete
2882 * Ends I/O on a number of bytes attached to @rq.
2883 * If @rq has leftover, sets it up for the next range of segments.
2886 * %false - we are done with this request
2887 * %true - still buffers pending for this request
2889 bool blk_end_request(struct request *rq, blk_status_t error,
2890 unsigned int nr_bytes)
2892 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2894 EXPORT_SYMBOL(blk_end_request);
2897 * blk_end_request_all - Helper function for drives to finish the request.
2898 * @rq: the request to finish
2899 * @error: block status code
2902 * Completely finish @rq.
2904 void blk_end_request_all(struct request *rq, blk_status_t error)
2907 unsigned int bidi_bytes = 0;
2909 if (unlikely(blk_bidi_rq(rq)))
2910 bidi_bytes = blk_rq_bytes(rq->next_rq);
2912 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2915 EXPORT_SYMBOL(blk_end_request_all);
2918 * __blk_end_request - Helper function for drivers to complete the request.
2919 * @rq: the request being processed
2920 * @error: block status code
2921 * @nr_bytes: number of bytes to complete
2924 * Must be called with queue lock held unlike blk_end_request().
2927 * %false - we are done with this request
2928 * %true - still buffers pending for this request
2930 bool __blk_end_request(struct request *rq, blk_status_t error,
2931 unsigned int nr_bytes)
2933 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2935 EXPORT_SYMBOL(__blk_end_request);
2938 * __blk_end_request_all - Helper function for drives to finish the request.
2939 * @rq: the request to finish
2940 * @error: block status code
2943 * Completely finish @rq. Must be called with queue lock held.
2945 void __blk_end_request_all(struct request *rq, blk_status_t error)
2948 unsigned int bidi_bytes = 0;
2950 if (unlikely(blk_bidi_rq(rq)))
2951 bidi_bytes = blk_rq_bytes(rq->next_rq);
2953 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2956 EXPORT_SYMBOL(__blk_end_request_all);
2959 * __blk_end_request_cur - Helper function to finish the current request chunk.
2960 * @rq: the request to finish the current chunk for
2961 * @error: block status code
2964 * Complete the current consecutively mapped chunk from @rq. Must
2965 * be called with queue lock held.
2968 * %false - we are done with this request
2969 * %true - still buffers pending for this request
2971 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
2973 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2975 EXPORT_SYMBOL(__blk_end_request_cur);
2977 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2980 if (bio_has_data(bio))
2981 rq->nr_phys_segments = bio_phys_segments(q, bio);
2983 rq->__data_len = bio->bi_iter.bi_size;
2984 rq->bio = rq->biotail = bio;
2987 rq->rq_disk = bio->bi_bdev->bd_disk;
2990 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2992 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2993 * @rq: the request to be flushed
2996 * Flush all pages in @rq.
2998 void rq_flush_dcache_pages(struct request *rq)
3000 struct req_iterator iter;
3001 struct bio_vec bvec;
3003 rq_for_each_segment(bvec, rq, iter)
3004 flush_dcache_page(bvec.bv_page);
3006 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3010 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3011 * @q : the queue of the device being checked
3014 * Check if underlying low-level drivers of a device are busy.
3015 * If the drivers want to export their busy state, they must set own
3016 * exporting function using blk_queue_lld_busy() first.
3018 * Basically, this function is used only by request stacking drivers
3019 * to stop dispatching requests to underlying devices when underlying
3020 * devices are busy. This behavior helps more I/O merging on the queue
3021 * of the request stacking driver and prevents I/O throughput regression
3022 * on burst I/O load.
3025 * 0 - Not busy (The request stacking driver should dispatch request)
3026 * 1 - Busy (The request stacking driver should stop dispatching request)
3028 int blk_lld_busy(struct request_queue *q)
3031 return q->lld_busy_fn(q);
3035 EXPORT_SYMBOL_GPL(blk_lld_busy);
3038 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3039 * @rq: the clone request to be cleaned up
3042 * Free all bios in @rq for a cloned request.
3044 void blk_rq_unprep_clone(struct request *rq)
3048 while ((bio = rq->bio) != NULL) {
3049 rq->bio = bio->bi_next;
3054 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3057 * Copy attributes of the original request to the clone request.
3058 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3060 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3062 dst->cpu = src->cpu;
3063 dst->__sector = blk_rq_pos(src);
3064 dst->__data_len = blk_rq_bytes(src);
3065 dst->nr_phys_segments = src->nr_phys_segments;
3066 dst->ioprio = src->ioprio;
3067 dst->extra_len = src->extra_len;
3071 * blk_rq_prep_clone - Helper function to setup clone request
3072 * @rq: the request to be setup
3073 * @rq_src: original request to be cloned
3074 * @bs: bio_set that bios for clone are allocated from
3075 * @gfp_mask: memory allocation mask for bio
3076 * @bio_ctr: setup function to be called for each clone bio.
3077 * Returns %0 for success, non %0 for failure.
3078 * @data: private data to be passed to @bio_ctr
3081 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3082 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3083 * are not copied, and copying such parts is the caller's responsibility.
3084 * Also, pages which the original bios are pointing to are not copied
3085 * and the cloned bios just point same pages.
3086 * So cloned bios must be completed before original bios, which means
3087 * the caller must complete @rq before @rq_src.
3089 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3090 struct bio_set *bs, gfp_t gfp_mask,
3091 int (*bio_ctr)(struct bio *, struct bio *, void *),
3094 struct bio *bio, *bio_src;
3099 __rq_for_each_bio(bio_src, rq_src) {
3100 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3104 if (bio_ctr && bio_ctr(bio, bio_src, data))
3108 rq->biotail->bi_next = bio;
3111 rq->bio = rq->biotail = bio;
3114 __blk_rq_prep_clone(rq, rq_src);
3121 blk_rq_unprep_clone(rq);
3125 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3127 int kblockd_schedule_work(struct work_struct *work)
3129 return queue_work(kblockd_workqueue, work);
3131 EXPORT_SYMBOL(kblockd_schedule_work);
3133 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3135 return queue_work_on(cpu, kblockd_workqueue, work);
3137 EXPORT_SYMBOL(kblockd_schedule_work_on);
3139 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3140 unsigned long delay)
3142 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3144 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3146 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3147 unsigned long delay)
3149 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3151 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3153 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3154 unsigned long delay)
3156 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3158 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3161 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3162 * @plug: The &struct blk_plug that needs to be initialized
3165 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3166 * pending I/O should the task end up blocking between blk_start_plug() and
3167 * blk_finish_plug(). This is important from a performance perspective, but
3168 * also ensures that we don't deadlock. For instance, if the task is blocking
3169 * for a memory allocation, memory reclaim could end up wanting to free a
3170 * page belonging to that request that is currently residing in our private
3171 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3172 * this kind of deadlock.
3174 void blk_start_plug(struct blk_plug *plug)
3176 struct task_struct *tsk = current;
3179 * If this is a nested plug, don't actually assign it.
3184 INIT_LIST_HEAD(&plug->list);
3185 INIT_LIST_HEAD(&plug->mq_list);
3186 INIT_LIST_HEAD(&plug->cb_list);
3188 * Store ordering should not be needed here, since a potential
3189 * preempt will imply a full memory barrier
3193 EXPORT_SYMBOL(blk_start_plug);
3195 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3197 struct request *rqa = container_of(a, struct request, queuelist);
3198 struct request *rqb = container_of(b, struct request, queuelist);
3200 return !(rqa->q < rqb->q ||
3201 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3205 * If 'from_schedule' is true, then postpone the dispatch of requests
3206 * until a safe kblockd context. We due this to avoid accidental big
3207 * additional stack usage in driver dispatch, in places where the originally
3208 * plugger did not intend it.
3210 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3212 __releases(q->queue_lock)
3214 trace_block_unplug(q, depth, !from_schedule);
3217 blk_run_queue_async(q);
3220 spin_unlock(q->queue_lock);
3223 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3225 LIST_HEAD(callbacks);
3227 while (!list_empty(&plug->cb_list)) {
3228 list_splice_init(&plug->cb_list, &callbacks);
3230 while (!list_empty(&callbacks)) {
3231 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3234 list_del(&cb->list);
3235 cb->callback(cb, from_schedule);
3240 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3243 struct blk_plug *plug = current->plug;
3244 struct blk_plug_cb *cb;
3249 list_for_each_entry(cb, &plug->cb_list, list)
3250 if (cb->callback == unplug && cb->data == data)
3253 /* Not currently on the callback list */
3254 BUG_ON(size < sizeof(*cb));
3255 cb = kzalloc(size, GFP_ATOMIC);
3258 cb->callback = unplug;
3259 list_add(&cb->list, &plug->cb_list);
3263 EXPORT_SYMBOL(blk_check_plugged);
3265 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3267 struct request_queue *q;
3268 unsigned long flags;
3273 flush_plug_callbacks(plug, from_schedule);
3275 if (!list_empty(&plug->mq_list))
3276 blk_mq_flush_plug_list(plug, from_schedule);
3278 if (list_empty(&plug->list))
3281 list_splice_init(&plug->list, &list);
3283 list_sort(NULL, &list, plug_rq_cmp);
3289 * Save and disable interrupts here, to avoid doing it for every
3290 * queue lock we have to take.
3292 local_irq_save(flags);
3293 while (!list_empty(&list)) {
3294 rq = list_entry_rq(list.next);
3295 list_del_init(&rq->queuelist);
3299 * This drops the queue lock
3302 queue_unplugged(q, depth, from_schedule);
3305 spin_lock(q->queue_lock);
3309 * Short-circuit if @q is dead
3311 if (unlikely(blk_queue_dying(q))) {
3312 __blk_end_request_all(rq, BLK_STS_IOERR);
3317 * rq is already accounted, so use raw insert
3319 if (op_is_flush(rq->cmd_flags))
3320 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3322 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3328 * This drops the queue lock
3331 queue_unplugged(q, depth, from_schedule);
3333 local_irq_restore(flags);
3336 void blk_finish_plug(struct blk_plug *plug)
3338 if (plug != current->plug)
3340 blk_flush_plug_list(plug, false);
3342 current->plug = NULL;
3344 EXPORT_SYMBOL(blk_finish_plug);
3348 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3349 * @q: the queue of the device
3350 * @dev: the device the queue belongs to
3353 * Initialize runtime-PM-related fields for @q and start auto suspend for
3354 * @dev. Drivers that want to take advantage of request-based runtime PM
3355 * should call this function after @dev has been initialized, and its
3356 * request queue @q has been allocated, and runtime PM for it can not happen
3357 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3358 * cases, driver should call this function before any I/O has taken place.
3360 * This function takes care of setting up using auto suspend for the device,
3361 * the autosuspend delay is set to -1 to make runtime suspend impossible
3362 * until an updated value is either set by user or by driver. Drivers do
3363 * not need to touch other autosuspend settings.
3365 * The block layer runtime PM is request based, so only works for drivers
3366 * that use request as their IO unit instead of those directly use bio's.
3368 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3371 q->rpm_status = RPM_ACTIVE;
3372 pm_runtime_set_autosuspend_delay(q->dev, -1);
3373 pm_runtime_use_autosuspend(q->dev);
3375 EXPORT_SYMBOL(blk_pm_runtime_init);
3378 * blk_pre_runtime_suspend - Pre runtime suspend check
3379 * @q: the queue of the device
3382 * This function will check if runtime suspend is allowed for the device
3383 * by examining if there are any requests pending in the queue. If there
3384 * are requests pending, the device can not be runtime suspended; otherwise,
3385 * the queue's status will be updated to SUSPENDING and the driver can
3386 * proceed to suspend the device.
3388 * For the not allowed case, we mark last busy for the device so that
3389 * runtime PM core will try to autosuspend it some time later.
3391 * This function should be called near the start of the device's
3392 * runtime_suspend callback.
3395 * 0 - OK to runtime suspend the device
3396 * -EBUSY - Device should not be runtime suspended
3398 int blk_pre_runtime_suspend(struct request_queue *q)
3405 spin_lock_irq(q->queue_lock);
3406 if (q->nr_pending) {
3408 pm_runtime_mark_last_busy(q->dev);
3410 q->rpm_status = RPM_SUSPENDING;
3412 spin_unlock_irq(q->queue_lock);
3415 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3418 * blk_post_runtime_suspend - Post runtime suspend processing
3419 * @q: the queue of the device
3420 * @err: return value of the device's runtime_suspend function
3423 * Update the queue's runtime status according to the return value of the
3424 * device's runtime suspend function and mark last busy for the device so
3425 * that PM core will try to auto suspend the device at a later time.
3427 * This function should be called near the end of the device's
3428 * runtime_suspend callback.
3430 void blk_post_runtime_suspend(struct request_queue *q, int err)
3435 spin_lock_irq(q->queue_lock);
3437 q->rpm_status = RPM_SUSPENDED;
3439 q->rpm_status = RPM_ACTIVE;
3440 pm_runtime_mark_last_busy(q->dev);
3442 spin_unlock_irq(q->queue_lock);
3444 EXPORT_SYMBOL(blk_post_runtime_suspend);
3447 * blk_pre_runtime_resume - Pre runtime resume processing
3448 * @q: the queue of the device
3451 * Update the queue's runtime status to RESUMING in preparation for the
3452 * runtime resume of the device.
3454 * This function should be called near the start of the device's
3455 * runtime_resume callback.
3457 void blk_pre_runtime_resume(struct request_queue *q)
3462 spin_lock_irq(q->queue_lock);
3463 q->rpm_status = RPM_RESUMING;
3464 spin_unlock_irq(q->queue_lock);
3466 EXPORT_SYMBOL(blk_pre_runtime_resume);
3469 * blk_post_runtime_resume - Post runtime resume processing
3470 * @q: the queue of the device
3471 * @err: return value of the device's runtime_resume function
3474 * Update the queue's runtime status according to the return value of the
3475 * device's runtime_resume function. If it is successfully resumed, process
3476 * the requests that are queued into the device's queue when it is resuming
3477 * and then mark last busy and initiate autosuspend for it.
3479 * This function should be called near the end of the device's
3480 * runtime_resume callback.
3482 void blk_post_runtime_resume(struct request_queue *q, int err)
3487 spin_lock_irq(q->queue_lock);
3489 q->rpm_status = RPM_ACTIVE;
3491 pm_runtime_mark_last_busy(q->dev);
3492 pm_request_autosuspend(q->dev);
3494 q->rpm_status = RPM_SUSPENDED;
3496 spin_unlock_irq(q->queue_lock);
3498 EXPORT_SYMBOL(blk_post_runtime_resume);
3501 * blk_set_runtime_active - Force runtime status of the queue to be active
3502 * @q: the queue of the device
3504 * If the device is left runtime suspended during system suspend the resume
3505 * hook typically resumes the device and corrects runtime status
3506 * accordingly. However, that does not affect the queue runtime PM status
3507 * which is still "suspended". This prevents processing requests from the
3510 * This function can be used in driver's resume hook to correct queue
3511 * runtime PM status and re-enable peeking requests from the queue. It
3512 * should be called before first request is added to the queue.
3514 void blk_set_runtime_active(struct request_queue *q)
3516 spin_lock_irq(q->queue_lock);
3517 q->rpm_status = RPM_ACTIVE;
3518 pm_runtime_mark_last_busy(q->dev);
3519 pm_request_autosuspend(q->dev);
3520 spin_unlock_irq(q->queue_lock);
3522 EXPORT_SYMBOL(blk_set_runtime_active);
3525 int __init blk_dev_init(void)
3527 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3528 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3529 FIELD_SIZEOF(struct request, cmd_flags));
3530 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3531 FIELD_SIZEOF(struct bio, bi_opf));
3533 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3534 kblockd_workqueue = alloc_workqueue("kblockd",
3535 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3536 if (!kblockd_workqueue)
3537 panic("Failed to create kblockd\n");
3539 request_cachep = kmem_cache_create("blkdev_requests",
3540 sizeof(struct request), 0, SLAB_PANIC, NULL);
3542 blk_requestq_cachep = kmem_cache_create("request_queue",
3543 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3545 #ifdef CONFIG_DEBUG_FS
3546 blk_debugfs_root = debugfs_create_dir("block", NULL);