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" },
147 /* device mapper special case, should not leak out: */
148 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
150 /* everything else not covered above: */
151 [BLK_STS_IOERR] = { -EIO, "I/O" },
154 blk_status_t errno_to_blk_status(int errno)
158 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
159 if (blk_errors[i].errno == errno)
160 return (__force blk_status_t)i;
163 return BLK_STS_IOERR;
165 EXPORT_SYMBOL_GPL(errno_to_blk_status);
167 int blk_status_to_errno(blk_status_t status)
169 int idx = (__force int)status;
171 if (WARN_ON_ONCE(idx > ARRAY_SIZE(blk_errors)))
173 return blk_errors[idx].errno;
175 EXPORT_SYMBOL_GPL(blk_status_to_errno);
177 static void print_req_error(struct request *req, blk_status_t status)
179 int idx = (__force int)status;
181 if (WARN_ON_ONCE(idx > ARRAY_SIZE(blk_errors)))
184 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
185 __func__, blk_errors[idx].name, req->rq_disk ?
186 req->rq_disk->disk_name : "?",
187 (unsigned long long)blk_rq_pos(req));
190 static void req_bio_endio(struct request *rq, struct bio *bio,
191 unsigned int nbytes, blk_status_t error)
194 bio->bi_status = error;
196 if (unlikely(rq->rq_flags & RQF_QUIET))
197 bio_set_flag(bio, BIO_QUIET);
199 bio_advance(bio, nbytes);
201 /* don't actually finish bio if it's part of flush sequence */
202 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
206 void blk_dump_rq_flags(struct request *rq, char *msg)
208 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
209 rq->rq_disk ? rq->rq_disk->disk_name : "?",
210 (unsigned long long) rq->cmd_flags);
212 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
213 (unsigned long long)blk_rq_pos(rq),
214 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
215 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
216 rq->bio, rq->biotail, blk_rq_bytes(rq));
218 EXPORT_SYMBOL(blk_dump_rq_flags);
220 static void blk_delay_work(struct work_struct *work)
222 struct request_queue *q;
224 q = container_of(work, struct request_queue, delay_work.work);
225 spin_lock_irq(q->queue_lock);
227 spin_unlock_irq(q->queue_lock);
231 * blk_delay_queue - restart queueing after defined interval
232 * @q: The &struct request_queue in question
233 * @msecs: Delay in msecs
236 * Sometimes queueing needs to be postponed for a little while, to allow
237 * resources to come back. This function will make sure that queueing is
238 * restarted around the specified time. Queue lock must be held.
240 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
242 if (likely(!blk_queue_dead(q)))
243 queue_delayed_work(kblockd_workqueue, &q->delay_work,
244 msecs_to_jiffies(msecs));
246 EXPORT_SYMBOL(blk_delay_queue);
249 * blk_start_queue_async - asynchronously restart a previously stopped queue
250 * @q: The &struct request_queue in question
253 * blk_start_queue_async() will clear the stop flag on the queue, and
254 * ensure that the request_fn for the queue is run from an async
257 void blk_start_queue_async(struct request_queue *q)
259 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
260 blk_run_queue_async(q);
262 EXPORT_SYMBOL(blk_start_queue_async);
265 * blk_start_queue - restart a previously stopped queue
266 * @q: The &struct request_queue in question
269 * blk_start_queue() will clear the stop flag on the queue, and call
270 * the request_fn for the queue if it was in a stopped state when
271 * entered. Also see blk_stop_queue(). Queue lock must be held.
273 void blk_start_queue(struct request_queue *q)
275 WARN_ON(!irqs_disabled());
277 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
280 EXPORT_SYMBOL(blk_start_queue);
283 * blk_stop_queue - stop a queue
284 * @q: The &struct request_queue in question
287 * The Linux block layer assumes that a block driver will consume all
288 * entries on the request queue when the request_fn strategy is called.
289 * Often this will not happen, because of hardware limitations (queue
290 * depth settings). If a device driver gets a 'queue full' response,
291 * or if it simply chooses not to queue more I/O at one point, it can
292 * call this function to prevent the request_fn from being called until
293 * the driver has signalled it's ready to go again. This happens by calling
294 * blk_start_queue() to restart queue operations. Queue lock must be held.
296 void blk_stop_queue(struct request_queue *q)
298 cancel_delayed_work(&q->delay_work);
299 queue_flag_set(QUEUE_FLAG_STOPPED, q);
301 EXPORT_SYMBOL(blk_stop_queue);
304 * blk_sync_queue - cancel any pending callbacks on a queue
308 * The block layer may perform asynchronous callback activity
309 * on a queue, such as calling the unplug function after a timeout.
310 * A block device may call blk_sync_queue to ensure that any
311 * such activity is cancelled, thus allowing it to release resources
312 * that the callbacks might use. The caller must already have made sure
313 * that its ->make_request_fn will not re-add plugging prior to calling
316 * This function does not cancel any asynchronous activity arising
317 * out of elevator or throttling code. That would require elevator_exit()
318 * and blkcg_exit_queue() to be called with queue lock initialized.
321 void blk_sync_queue(struct request_queue *q)
323 del_timer_sync(&q->timeout);
326 struct blk_mq_hw_ctx *hctx;
329 queue_for_each_hw_ctx(q, hctx, i)
330 cancel_delayed_work_sync(&hctx->run_work);
332 cancel_delayed_work_sync(&q->delay_work);
335 EXPORT_SYMBOL(blk_sync_queue);
338 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
339 * @q: The queue to run
342 * Invoke request handling on a queue if there are any pending requests.
343 * May be used to restart request handling after a request has completed.
344 * This variant runs the queue whether or not the queue has been
345 * stopped. Must be called with the queue lock held and interrupts
346 * disabled. See also @blk_run_queue.
348 inline void __blk_run_queue_uncond(struct request_queue *q)
350 if (unlikely(blk_queue_dead(q)))
354 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
355 * the queue lock internally. As a result multiple threads may be
356 * running such a request function concurrently. Keep track of the
357 * number of active request_fn invocations such that blk_drain_queue()
358 * can wait until all these request_fn calls have finished.
360 q->request_fn_active++;
362 q->request_fn_active--;
364 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
367 * __blk_run_queue - run a single device queue
368 * @q: The queue to run
371 * See @blk_run_queue. This variant must be called with the queue lock
372 * held and interrupts disabled.
374 void __blk_run_queue(struct request_queue *q)
376 if (unlikely(blk_queue_stopped(q)))
379 __blk_run_queue_uncond(q);
381 EXPORT_SYMBOL(__blk_run_queue);
384 * blk_run_queue_async - run a single device queue in workqueue context
385 * @q: The queue to run
388 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
389 * of us. The caller must hold the queue lock.
391 void blk_run_queue_async(struct request_queue *q)
393 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
394 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
396 EXPORT_SYMBOL(blk_run_queue_async);
399 * blk_run_queue - run a single device queue
400 * @q: The queue to run
403 * Invoke request handling on this queue, if it has pending work to do.
404 * May be used to restart queueing when a request has completed.
406 void blk_run_queue(struct request_queue *q)
410 spin_lock_irqsave(q->queue_lock, flags);
412 spin_unlock_irqrestore(q->queue_lock, flags);
414 EXPORT_SYMBOL(blk_run_queue);
416 void blk_put_queue(struct request_queue *q)
418 kobject_put(&q->kobj);
420 EXPORT_SYMBOL(blk_put_queue);
423 * __blk_drain_queue - drain requests from request_queue
425 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
427 * Drain requests from @q. If @drain_all is set, all requests are drained.
428 * If not, only ELVPRIV requests are drained. The caller is responsible
429 * for ensuring that no new requests which need to be drained are queued.
431 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
432 __releases(q->queue_lock)
433 __acquires(q->queue_lock)
437 lockdep_assert_held(q->queue_lock);
443 * The caller might be trying to drain @q before its
444 * elevator is initialized.
447 elv_drain_elevator(q);
449 blkcg_drain_queue(q);
452 * This function might be called on a queue which failed
453 * driver init after queue creation or is not yet fully
454 * active yet. Some drivers (e.g. fd and loop) get unhappy
455 * in such cases. Kick queue iff dispatch queue has
456 * something on it and @q has request_fn set.
458 if (!list_empty(&q->queue_head) && q->request_fn)
461 drain |= q->nr_rqs_elvpriv;
462 drain |= q->request_fn_active;
465 * Unfortunately, requests are queued at and tracked from
466 * multiple places and there's no single counter which can
467 * be drained. Check all the queues and counters.
470 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
471 drain |= !list_empty(&q->queue_head);
472 for (i = 0; i < 2; i++) {
473 drain |= q->nr_rqs[i];
474 drain |= q->in_flight[i];
476 drain |= !list_empty(&fq->flush_queue[i]);
483 spin_unlock_irq(q->queue_lock);
487 spin_lock_irq(q->queue_lock);
491 * With queue marked dead, any woken up waiter will fail the
492 * allocation path, so the wakeup chaining is lost and we're
493 * left with hung waiters. We need to wake up those waiters.
496 struct request_list *rl;
498 blk_queue_for_each_rl(rl, q)
499 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
500 wake_up_all(&rl->wait[i]);
505 * blk_queue_bypass_start - enter queue bypass mode
506 * @q: queue of interest
508 * In bypass mode, only the dispatch FIFO queue of @q is used. This
509 * function makes @q enter bypass mode and drains all requests which were
510 * throttled or issued before. On return, it's guaranteed that no request
511 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
512 * inside queue or RCU read lock.
514 void blk_queue_bypass_start(struct request_queue *q)
516 spin_lock_irq(q->queue_lock);
518 queue_flag_set(QUEUE_FLAG_BYPASS, q);
519 spin_unlock_irq(q->queue_lock);
522 * Queues start drained. Skip actual draining till init is
523 * complete. This avoids lenghty delays during queue init which
524 * can happen many times during boot.
526 if (blk_queue_init_done(q)) {
527 spin_lock_irq(q->queue_lock);
528 __blk_drain_queue(q, false);
529 spin_unlock_irq(q->queue_lock);
531 /* ensure blk_queue_bypass() is %true inside RCU read lock */
535 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
538 * blk_queue_bypass_end - leave queue bypass mode
539 * @q: queue of interest
541 * Leave bypass mode and restore the normal queueing behavior.
543 void blk_queue_bypass_end(struct request_queue *q)
545 spin_lock_irq(q->queue_lock);
546 if (!--q->bypass_depth)
547 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
548 WARN_ON_ONCE(q->bypass_depth < 0);
549 spin_unlock_irq(q->queue_lock);
551 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
553 void blk_set_queue_dying(struct request_queue *q)
555 spin_lock_irq(q->queue_lock);
556 queue_flag_set(QUEUE_FLAG_DYING, q);
557 spin_unlock_irq(q->queue_lock);
560 * When queue DYING flag is set, we need to block new req
561 * entering queue, so we call blk_freeze_queue_start() to
562 * prevent I/O from crossing blk_queue_enter().
564 blk_freeze_queue_start(q);
567 blk_mq_wake_waiters(q);
569 struct request_list *rl;
571 spin_lock_irq(q->queue_lock);
572 blk_queue_for_each_rl(rl, q) {
574 wake_up(&rl->wait[BLK_RW_SYNC]);
575 wake_up(&rl->wait[BLK_RW_ASYNC]);
578 spin_unlock_irq(q->queue_lock);
581 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
584 * blk_cleanup_queue - shutdown a request queue
585 * @q: request queue to shutdown
587 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
588 * put it. All future requests will be failed immediately with -ENODEV.
590 void blk_cleanup_queue(struct request_queue *q)
592 spinlock_t *lock = q->queue_lock;
594 /* mark @q DYING, no new request or merges will be allowed afterwards */
595 mutex_lock(&q->sysfs_lock);
596 blk_set_queue_dying(q);
600 * A dying queue is permanently in bypass mode till released. Note
601 * that, unlike blk_queue_bypass_start(), we aren't performing
602 * synchronize_rcu() after entering bypass mode to avoid the delay
603 * as some drivers create and destroy a lot of queues while
604 * probing. This is still safe because blk_release_queue() will be
605 * called only after the queue refcnt drops to zero and nothing,
606 * RCU or not, would be traversing the queue by then.
609 queue_flag_set(QUEUE_FLAG_BYPASS, q);
611 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
612 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
613 queue_flag_set(QUEUE_FLAG_DYING, q);
614 spin_unlock_irq(lock);
615 mutex_unlock(&q->sysfs_lock);
618 * Drain all requests queued before DYING marking. Set DEAD flag to
619 * prevent that q->request_fn() gets invoked after draining finished.
624 __blk_drain_queue(q, true);
625 queue_flag_set(QUEUE_FLAG_DEAD, q);
626 spin_unlock_irq(lock);
628 /* for synchronous bio-based driver finish in-flight integrity i/o */
629 blk_flush_integrity();
631 /* @q won't process any more request, flush async actions */
632 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
636 blk_mq_free_queue(q);
637 percpu_ref_exit(&q->q_usage_counter);
640 if (q->queue_lock != &q->__queue_lock)
641 q->queue_lock = &q->__queue_lock;
642 spin_unlock_irq(lock);
644 /* @q is and will stay empty, shutdown and put */
647 EXPORT_SYMBOL(blk_cleanup_queue);
649 /* Allocate memory local to the request queue */
650 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
652 struct request_queue *q = data;
654 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
657 static void free_request_simple(void *element, void *data)
659 kmem_cache_free(request_cachep, element);
662 static void *alloc_request_size(gfp_t gfp_mask, void *data)
664 struct request_queue *q = data;
667 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
669 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
676 static void free_request_size(void *element, void *data)
678 struct request_queue *q = data;
681 q->exit_rq_fn(q, element);
685 int blk_init_rl(struct request_list *rl, struct request_queue *q,
688 if (unlikely(rl->rq_pool))
692 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
693 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
694 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
695 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
698 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
699 alloc_request_size, free_request_size,
700 q, gfp_mask, q->node);
702 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
703 alloc_request_simple, free_request_simple,
704 q, gfp_mask, q->node);
712 void blk_exit_rl(struct request_list *rl)
715 mempool_destroy(rl->rq_pool);
718 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
720 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
722 EXPORT_SYMBOL(blk_alloc_queue);
724 int blk_queue_enter(struct request_queue *q, bool nowait)
729 if (percpu_ref_tryget_live(&q->q_usage_counter))
736 * read pair of barrier in blk_freeze_queue_start(),
737 * we need to order reading __PERCPU_REF_DEAD flag of
738 * .q_usage_counter and reading .mq_freeze_depth or
739 * queue dying flag, otherwise the following wait may
740 * never return if the two reads are reordered.
744 ret = wait_event_interruptible(q->mq_freeze_wq,
745 !atomic_read(&q->mq_freeze_depth) ||
747 if (blk_queue_dying(q))
754 void blk_queue_exit(struct request_queue *q)
756 percpu_ref_put(&q->q_usage_counter);
759 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
761 struct request_queue *q =
762 container_of(ref, struct request_queue, q_usage_counter);
764 wake_up_all(&q->mq_freeze_wq);
767 static void blk_rq_timed_out_timer(unsigned long data)
769 struct request_queue *q = (struct request_queue *)data;
771 kblockd_schedule_work(&q->timeout_work);
774 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
776 struct request_queue *q;
778 q = kmem_cache_alloc_node(blk_requestq_cachep,
779 gfp_mask | __GFP_ZERO, node_id);
783 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
787 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
791 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
792 if (!q->backing_dev_info)
795 q->stats = blk_alloc_queue_stats();
799 q->backing_dev_info->ra_pages =
800 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
801 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
802 q->backing_dev_info->name = "block";
805 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
806 laptop_mode_timer_fn, (unsigned long) q);
807 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
808 INIT_LIST_HEAD(&q->queue_head);
809 INIT_LIST_HEAD(&q->timeout_list);
810 INIT_LIST_HEAD(&q->icq_list);
811 #ifdef CONFIG_BLK_CGROUP
812 INIT_LIST_HEAD(&q->blkg_list);
814 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
816 kobject_init(&q->kobj, &blk_queue_ktype);
818 mutex_init(&q->sysfs_lock);
819 spin_lock_init(&q->__queue_lock);
822 * By default initialize queue_lock to internal lock and driver can
823 * override it later if need be.
825 q->queue_lock = &q->__queue_lock;
828 * A queue starts its life with bypass turned on to avoid
829 * unnecessary bypass on/off overhead and nasty surprises during
830 * init. The initial bypass will be finished when the queue is
831 * registered by blk_register_queue().
834 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
836 init_waitqueue_head(&q->mq_freeze_wq);
839 * Init percpu_ref in atomic mode so that it's faster to shutdown.
840 * See blk_register_queue() for details.
842 if (percpu_ref_init(&q->q_usage_counter,
843 blk_queue_usage_counter_release,
844 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
847 if (blkcg_init_queue(q))
853 percpu_ref_exit(&q->q_usage_counter);
855 blk_free_queue_stats(q->stats);
857 bdi_put(q->backing_dev_info);
859 bioset_free(q->bio_split);
861 ida_simple_remove(&blk_queue_ida, q->id);
863 kmem_cache_free(blk_requestq_cachep, q);
866 EXPORT_SYMBOL(blk_alloc_queue_node);
869 * blk_init_queue - prepare a request queue for use with a block device
870 * @rfn: The function to be called to process requests that have been
871 * placed on the queue.
872 * @lock: Request queue spin lock
875 * If a block device wishes to use the standard request handling procedures,
876 * which sorts requests and coalesces adjacent requests, then it must
877 * call blk_init_queue(). The function @rfn will be called when there
878 * are requests on the queue that need to be processed. If the device
879 * supports plugging, then @rfn may not be called immediately when requests
880 * are available on the queue, but may be called at some time later instead.
881 * Plugged queues are generally unplugged when a buffer belonging to one
882 * of the requests on the queue is needed, or due to memory pressure.
884 * @rfn is not required, or even expected, to remove all requests off the
885 * queue, but only as many as it can handle at a time. If it does leave
886 * requests on the queue, it is responsible for arranging that the requests
887 * get dealt with eventually.
889 * The queue spin lock must be held while manipulating the requests on the
890 * request queue; this lock will be taken also from interrupt context, so irq
891 * disabling is needed for it.
893 * Function returns a pointer to the initialized request queue, or %NULL if
897 * blk_init_queue() must be paired with a blk_cleanup_queue() call
898 * when the block device is deactivated (such as at module unload).
901 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
903 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
905 EXPORT_SYMBOL(blk_init_queue);
907 struct request_queue *
908 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
910 struct request_queue *q;
912 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
918 q->queue_lock = lock;
919 if (blk_init_allocated_queue(q) < 0) {
920 blk_cleanup_queue(q);
926 EXPORT_SYMBOL(blk_init_queue_node);
928 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
931 int blk_init_allocated_queue(struct request_queue *q)
933 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
937 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
938 goto out_free_flush_queue;
940 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
941 goto out_exit_flush_rq;
943 INIT_WORK(&q->timeout_work, blk_timeout_work);
944 q->queue_flags |= QUEUE_FLAG_DEFAULT;
947 * This also sets hw/phys segments, boundary and size
949 blk_queue_make_request(q, blk_queue_bio);
951 q->sg_reserved_size = INT_MAX;
953 /* Protect q->elevator from elevator_change */
954 mutex_lock(&q->sysfs_lock);
957 if (elevator_init(q, NULL)) {
958 mutex_unlock(&q->sysfs_lock);
959 goto out_exit_flush_rq;
962 mutex_unlock(&q->sysfs_lock);
967 q->exit_rq_fn(q, q->fq->flush_rq);
968 out_free_flush_queue:
969 blk_free_flush_queue(q->fq);
972 EXPORT_SYMBOL(blk_init_allocated_queue);
974 bool blk_get_queue(struct request_queue *q)
976 if (likely(!blk_queue_dying(q))) {
983 EXPORT_SYMBOL(blk_get_queue);
985 static inline void blk_free_request(struct request_list *rl, struct request *rq)
987 if (rq->rq_flags & RQF_ELVPRIV) {
988 elv_put_request(rl->q, rq);
990 put_io_context(rq->elv.icq->ioc);
993 mempool_free(rq, rl->rq_pool);
997 * ioc_batching returns true if the ioc is a valid batching request and
998 * should be given priority access to a request.
1000 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1006 * Make sure the process is able to allocate at least 1 request
1007 * even if the batch times out, otherwise we could theoretically
1010 return ioc->nr_batch_requests == q->nr_batching ||
1011 (ioc->nr_batch_requests > 0
1012 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1016 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1017 * will cause the process to be a "batcher" on all queues in the system. This
1018 * is the behaviour we want though - once it gets a wakeup it should be given
1021 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1023 if (!ioc || ioc_batching(q, ioc))
1026 ioc->nr_batch_requests = q->nr_batching;
1027 ioc->last_waited = jiffies;
1030 static void __freed_request(struct request_list *rl, int sync)
1032 struct request_queue *q = rl->q;
1034 if (rl->count[sync] < queue_congestion_off_threshold(q))
1035 blk_clear_congested(rl, sync);
1037 if (rl->count[sync] + 1 <= q->nr_requests) {
1038 if (waitqueue_active(&rl->wait[sync]))
1039 wake_up(&rl->wait[sync]);
1041 blk_clear_rl_full(rl, sync);
1046 * A request has just been released. Account for it, update the full and
1047 * congestion status, wake up any waiters. Called under q->queue_lock.
1049 static void freed_request(struct request_list *rl, bool sync,
1050 req_flags_t rq_flags)
1052 struct request_queue *q = rl->q;
1056 if (rq_flags & RQF_ELVPRIV)
1057 q->nr_rqs_elvpriv--;
1059 __freed_request(rl, sync);
1061 if (unlikely(rl->starved[sync ^ 1]))
1062 __freed_request(rl, sync ^ 1);
1065 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1067 struct request_list *rl;
1068 int on_thresh, off_thresh;
1070 spin_lock_irq(q->queue_lock);
1071 q->nr_requests = nr;
1072 blk_queue_congestion_threshold(q);
1073 on_thresh = queue_congestion_on_threshold(q);
1074 off_thresh = queue_congestion_off_threshold(q);
1076 blk_queue_for_each_rl(rl, q) {
1077 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1078 blk_set_congested(rl, BLK_RW_SYNC);
1079 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1080 blk_clear_congested(rl, BLK_RW_SYNC);
1082 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1083 blk_set_congested(rl, BLK_RW_ASYNC);
1084 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1085 blk_clear_congested(rl, BLK_RW_ASYNC);
1087 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1088 blk_set_rl_full(rl, BLK_RW_SYNC);
1090 blk_clear_rl_full(rl, BLK_RW_SYNC);
1091 wake_up(&rl->wait[BLK_RW_SYNC]);
1094 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1095 blk_set_rl_full(rl, BLK_RW_ASYNC);
1097 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1098 wake_up(&rl->wait[BLK_RW_ASYNC]);
1102 spin_unlock_irq(q->queue_lock);
1107 * __get_request - get a free request
1108 * @rl: request list to allocate from
1109 * @op: operation and flags
1110 * @bio: bio to allocate request for (can be %NULL)
1111 * @gfp_mask: allocation mask
1113 * Get a free request from @q. This function may fail under memory
1114 * pressure or if @q is dead.
1116 * Must be called with @q->queue_lock held and,
1117 * Returns ERR_PTR on failure, with @q->queue_lock held.
1118 * Returns request pointer on success, with @q->queue_lock *not held*.
1120 static struct request *__get_request(struct request_list *rl, unsigned int op,
1121 struct bio *bio, gfp_t gfp_mask)
1123 struct request_queue *q = rl->q;
1125 struct elevator_type *et = q->elevator->type;
1126 struct io_context *ioc = rq_ioc(bio);
1127 struct io_cq *icq = NULL;
1128 const bool is_sync = op_is_sync(op);
1130 req_flags_t rq_flags = RQF_ALLOCED;
1132 if (unlikely(blk_queue_dying(q)))
1133 return ERR_PTR(-ENODEV);
1135 may_queue = elv_may_queue(q, op);
1136 if (may_queue == ELV_MQUEUE_NO)
1139 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1140 if (rl->count[is_sync]+1 >= q->nr_requests) {
1142 * The queue will fill after this allocation, so set
1143 * it as full, and mark this process as "batching".
1144 * This process will be allowed to complete a batch of
1145 * requests, others will be blocked.
1147 if (!blk_rl_full(rl, is_sync)) {
1148 ioc_set_batching(q, ioc);
1149 blk_set_rl_full(rl, is_sync);
1151 if (may_queue != ELV_MQUEUE_MUST
1152 && !ioc_batching(q, ioc)) {
1154 * The queue is full and the allocating
1155 * process is not a "batcher", and not
1156 * exempted by the IO scheduler
1158 return ERR_PTR(-ENOMEM);
1162 blk_set_congested(rl, is_sync);
1166 * Only allow batching queuers to allocate up to 50% over the defined
1167 * limit of requests, otherwise we could have thousands of requests
1168 * allocated with any setting of ->nr_requests
1170 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1171 return ERR_PTR(-ENOMEM);
1173 q->nr_rqs[is_sync]++;
1174 rl->count[is_sync]++;
1175 rl->starved[is_sync] = 0;
1178 * Decide whether the new request will be managed by elevator. If
1179 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1180 * prevent the current elevator from being destroyed until the new
1181 * request is freed. This guarantees icq's won't be destroyed and
1182 * makes creating new ones safe.
1184 * Flush requests do not use the elevator so skip initialization.
1185 * This allows a request to share the flush and elevator data.
1187 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1188 * it will be created after releasing queue_lock.
1190 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1191 rq_flags |= RQF_ELVPRIV;
1192 q->nr_rqs_elvpriv++;
1193 if (et->icq_cache && ioc)
1194 icq = ioc_lookup_icq(ioc, q);
1197 if (blk_queue_io_stat(q))
1198 rq_flags |= RQF_IO_STAT;
1199 spin_unlock_irq(q->queue_lock);
1201 /* allocate and init request */
1202 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1207 blk_rq_set_rl(rq, rl);
1209 rq->rq_flags = rq_flags;
1212 if (rq_flags & RQF_ELVPRIV) {
1213 if (unlikely(et->icq_cache && !icq)) {
1215 icq = ioc_create_icq(ioc, q, gfp_mask);
1221 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1224 /* @rq->elv.icq holds io_context until @rq is freed */
1226 get_io_context(icq->ioc);
1230 * ioc may be NULL here, and ioc_batching will be false. That's
1231 * OK, if the queue is under the request limit then requests need
1232 * not count toward the nr_batch_requests limit. There will always
1233 * be some limit enforced by BLK_BATCH_TIME.
1235 if (ioc_batching(q, ioc))
1236 ioc->nr_batch_requests--;
1238 trace_block_getrq(q, bio, op);
1243 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1244 * and may fail indefinitely under memory pressure and thus
1245 * shouldn't stall IO. Treat this request as !elvpriv. This will
1246 * disturb iosched and blkcg but weird is bettern than dead.
1248 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1249 __func__, dev_name(q->backing_dev_info->dev));
1251 rq->rq_flags &= ~RQF_ELVPRIV;
1254 spin_lock_irq(q->queue_lock);
1255 q->nr_rqs_elvpriv--;
1256 spin_unlock_irq(q->queue_lock);
1261 * Allocation failed presumably due to memory. Undo anything we
1262 * might have messed up.
1264 * Allocating task should really be put onto the front of the wait
1265 * queue, but this is pretty rare.
1267 spin_lock_irq(q->queue_lock);
1268 freed_request(rl, is_sync, rq_flags);
1271 * in the very unlikely event that allocation failed and no
1272 * requests for this direction was pending, mark us starved so that
1273 * freeing of a request in the other direction will notice
1274 * us. another possible fix would be to split the rq mempool into
1278 if (unlikely(rl->count[is_sync] == 0))
1279 rl->starved[is_sync] = 1;
1280 return ERR_PTR(-ENOMEM);
1284 * get_request - get a free request
1285 * @q: request_queue to allocate request from
1286 * @op: operation and flags
1287 * @bio: bio to allocate request for (can be %NULL)
1288 * @gfp_mask: allocation mask
1290 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1291 * this function keeps retrying under memory pressure and fails iff @q is dead.
1293 * Must be called with @q->queue_lock held and,
1294 * Returns ERR_PTR on failure, with @q->queue_lock held.
1295 * Returns request pointer on success, with @q->queue_lock *not held*.
1297 static struct request *get_request(struct request_queue *q, unsigned int op,
1298 struct bio *bio, gfp_t gfp_mask)
1300 const bool is_sync = op_is_sync(op);
1302 struct request_list *rl;
1305 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1307 rq = __get_request(rl, op, bio, gfp_mask);
1311 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1316 /* wait on @rl and retry */
1317 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1318 TASK_UNINTERRUPTIBLE);
1320 trace_block_sleeprq(q, bio, op);
1322 spin_unlock_irq(q->queue_lock);
1326 * After sleeping, we become a "batching" process and will be able
1327 * to allocate at least one request, and up to a big batch of them
1328 * for a small period time. See ioc_batching, ioc_set_batching
1330 ioc_set_batching(q, current->io_context);
1332 spin_lock_irq(q->queue_lock);
1333 finish_wait(&rl->wait[is_sync], &wait);
1338 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1343 /* create ioc upfront */
1344 create_io_context(gfp_mask, q->node);
1346 spin_lock_irq(q->queue_lock);
1347 rq = get_request(q, rw, NULL, gfp_mask);
1349 spin_unlock_irq(q->queue_lock);
1353 /* q->queue_lock is unlocked at this point */
1355 rq->__sector = (sector_t) -1;
1356 rq->bio = rq->biotail = NULL;
1360 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1363 return blk_mq_alloc_request(q, rw,
1364 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1365 0 : BLK_MQ_REQ_NOWAIT);
1367 return blk_old_get_request(q, rw, gfp_mask);
1369 EXPORT_SYMBOL(blk_get_request);
1372 * blk_requeue_request - put a request back on queue
1373 * @q: request queue where request should be inserted
1374 * @rq: request to be inserted
1377 * Drivers often keep queueing requests until the hardware cannot accept
1378 * more, when that condition happens we need to put the request back
1379 * on the queue. Must be called with queue lock held.
1381 void blk_requeue_request(struct request_queue *q, struct request *rq)
1383 blk_delete_timer(rq);
1384 blk_clear_rq_complete(rq);
1385 trace_block_rq_requeue(q, rq);
1386 wbt_requeue(q->rq_wb, &rq->issue_stat);
1388 if (rq->rq_flags & RQF_QUEUED)
1389 blk_queue_end_tag(q, rq);
1391 BUG_ON(blk_queued_rq(rq));
1393 elv_requeue_request(q, rq);
1395 EXPORT_SYMBOL(blk_requeue_request);
1397 static void add_acct_request(struct request_queue *q, struct request *rq,
1400 blk_account_io_start(rq, true);
1401 __elv_add_request(q, rq, where);
1404 static void part_round_stats_single(int cpu, struct hd_struct *part,
1409 if (now == part->stamp)
1412 inflight = part_in_flight(part);
1414 __part_stat_add(cpu, part, time_in_queue,
1415 inflight * (now - part->stamp));
1416 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1422 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1423 * @cpu: cpu number for stats access
1424 * @part: target partition
1426 * The average IO queue length and utilisation statistics are maintained
1427 * by observing the current state of the queue length and the amount of
1428 * time it has been in this state for.
1430 * Normally, that accounting is done on IO completion, but that can result
1431 * in more than a second's worth of IO being accounted for within any one
1432 * second, leading to >100% utilisation. To deal with that, we call this
1433 * function to do a round-off before returning the results when reading
1434 * /proc/diskstats. This accounts immediately for all queue usage up to
1435 * the current jiffies and restarts the counters again.
1437 void part_round_stats(int cpu, struct hd_struct *part)
1439 unsigned long now = jiffies;
1442 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1443 part_round_stats_single(cpu, part, now);
1445 EXPORT_SYMBOL_GPL(part_round_stats);
1448 static void blk_pm_put_request(struct request *rq)
1450 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1451 pm_runtime_mark_last_busy(rq->q->dev);
1454 static inline void blk_pm_put_request(struct request *rq) {}
1458 * queue lock must be held
1460 void __blk_put_request(struct request_queue *q, struct request *req)
1462 req_flags_t rq_flags = req->rq_flags;
1468 blk_mq_free_request(req);
1472 blk_pm_put_request(req);
1474 elv_completed_request(q, req);
1476 /* this is a bio leak */
1477 WARN_ON(req->bio != NULL);
1479 wbt_done(q->rq_wb, &req->issue_stat);
1482 * Request may not have originated from ll_rw_blk. if not,
1483 * it didn't come out of our reserved rq pools
1485 if (rq_flags & RQF_ALLOCED) {
1486 struct request_list *rl = blk_rq_rl(req);
1487 bool sync = op_is_sync(req->cmd_flags);
1489 BUG_ON(!list_empty(&req->queuelist));
1490 BUG_ON(ELV_ON_HASH(req));
1492 blk_free_request(rl, req);
1493 freed_request(rl, sync, rq_flags);
1497 EXPORT_SYMBOL_GPL(__blk_put_request);
1499 void blk_put_request(struct request *req)
1501 struct request_queue *q = req->q;
1504 blk_mq_free_request(req);
1506 unsigned long flags;
1508 spin_lock_irqsave(q->queue_lock, flags);
1509 __blk_put_request(q, req);
1510 spin_unlock_irqrestore(q->queue_lock, flags);
1513 EXPORT_SYMBOL(blk_put_request);
1515 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1518 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1520 if (!ll_back_merge_fn(q, req, bio))
1523 trace_block_bio_backmerge(q, req, bio);
1525 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1526 blk_rq_set_mixed_merge(req);
1528 req->biotail->bi_next = bio;
1530 req->__data_len += bio->bi_iter.bi_size;
1531 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1533 blk_account_io_start(req, false);
1537 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1540 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1542 if (!ll_front_merge_fn(q, req, bio))
1545 trace_block_bio_frontmerge(q, req, bio);
1547 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1548 blk_rq_set_mixed_merge(req);
1550 bio->bi_next = req->bio;
1553 req->__sector = bio->bi_iter.bi_sector;
1554 req->__data_len += bio->bi_iter.bi_size;
1555 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1557 blk_account_io_start(req, false);
1561 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1564 unsigned short segments = blk_rq_nr_discard_segments(req);
1566 if (segments >= queue_max_discard_segments(q))
1568 if (blk_rq_sectors(req) + bio_sectors(bio) >
1569 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1572 req->biotail->bi_next = bio;
1574 req->__data_len += bio->bi_iter.bi_size;
1575 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1576 req->nr_phys_segments = segments + 1;
1578 blk_account_io_start(req, false);
1581 req_set_nomerge(q, req);
1586 * blk_attempt_plug_merge - try to merge with %current's plugged list
1587 * @q: request_queue new bio is being queued at
1588 * @bio: new bio being queued
1589 * @request_count: out parameter for number of traversed plugged requests
1590 * @same_queue_rq: pointer to &struct request that gets filled in when
1591 * another request associated with @q is found on the plug list
1592 * (optional, may be %NULL)
1594 * Determine whether @bio being queued on @q can be merged with a request
1595 * on %current's plugged list. Returns %true if merge was successful,
1598 * Plugging coalesces IOs from the same issuer for the same purpose without
1599 * going through @q->queue_lock. As such it's more of an issuing mechanism
1600 * than scheduling, and the request, while may have elvpriv data, is not
1601 * added on the elevator at this point. In addition, we don't have
1602 * reliable access to the elevator outside queue lock. Only check basic
1603 * merging parameters without querying the elevator.
1605 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1607 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1608 unsigned int *request_count,
1609 struct request **same_queue_rq)
1611 struct blk_plug *plug;
1613 struct list_head *plug_list;
1615 plug = current->plug;
1621 plug_list = &plug->mq_list;
1623 plug_list = &plug->list;
1625 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1626 bool merged = false;
1631 * Only blk-mq multiple hardware queues case checks the
1632 * rq in the same queue, there should be only one such
1636 *same_queue_rq = rq;
1639 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1642 switch (blk_try_merge(rq, bio)) {
1643 case ELEVATOR_BACK_MERGE:
1644 merged = bio_attempt_back_merge(q, rq, bio);
1646 case ELEVATOR_FRONT_MERGE:
1647 merged = bio_attempt_front_merge(q, rq, bio);
1649 case ELEVATOR_DISCARD_MERGE:
1650 merged = bio_attempt_discard_merge(q, rq, bio);
1663 unsigned int blk_plug_queued_count(struct request_queue *q)
1665 struct blk_plug *plug;
1667 struct list_head *plug_list;
1668 unsigned int ret = 0;
1670 plug = current->plug;
1675 plug_list = &plug->mq_list;
1677 plug_list = &plug->list;
1679 list_for_each_entry(rq, plug_list, queuelist) {
1687 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1689 struct io_context *ioc = rq_ioc(bio);
1691 if (bio->bi_opf & REQ_RAHEAD)
1692 req->cmd_flags |= REQ_FAILFAST_MASK;
1694 req->__sector = bio->bi_iter.bi_sector;
1695 if (ioprio_valid(bio_prio(bio)))
1696 req->ioprio = bio_prio(bio);
1698 req->ioprio = ioc->ioprio;
1700 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1701 blk_rq_bio_prep(req->q, req, bio);
1703 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1705 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1707 struct blk_plug *plug;
1708 int where = ELEVATOR_INSERT_SORT;
1709 struct request *req, *free;
1710 unsigned int request_count = 0;
1711 unsigned int wb_acct;
1714 * low level driver can indicate that it wants pages above a
1715 * certain limit bounced to low memory (ie for highmem, or even
1716 * ISA dma in theory)
1718 blk_queue_bounce(q, &bio);
1720 blk_queue_split(q, &bio, q->bio_split);
1722 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1723 bio->bi_status = BLK_STS_IOERR;
1725 return BLK_QC_T_NONE;
1728 if (op_is_flush(bio->bi_opf)) {
1729 spin_lock_irq(q->queue_lock);
1730 where = ELEVATOR_INSERT_FLUSH;
1735 * Check if we can merge with the plugged list before grabbing
1738 if (!blk_queue_nomerges(q)) {
1739 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1740 return BLK_QC_T_NONE;
1742 request_count = blk_plug_queued_count(q);
1744 spin_lock_irq(q->queue_lock);
1746 switch (elv_merge(q, &req, bio)) {
1747 case ELEVATOR_BACK_MERGE:
1748 if (!bio_attempt_back_merge(q, req, bio))
1750 elv_bio_merged(q, req, bio);
1751 free = attempt_back_merge(q, req);
1753 __blk_put_request(q, free);
1755 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1757 case ELEVATOR_FRONT_MERGE:
1758 if (!bio_attempt_front_merge(q, req, bio))
1760 elv_bio_merged(q, req, bio);
1761 free = attempt_front_merge(q, req);
1763 __blk_put_request(q, free);
1765 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1772 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1775 * Grab a free request. This is might sleep but can not fail.
1776 * Returns with the queue unlocked.
1778 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1780 __wbt_done(q->rq_wb, wb_acct);
1781 if (PTR_ERR(req) == -ENOMEM)
1782 bio->bi_status = BLK_STS_RESOURCE;
1784 bio->bi_status = BLK_STS_IOERR;
1789 wbt_track(&req->issue_stat, wb_acct);
1792 * After dropping the lock and possibly sleeping here, our request
1793 * may now be mergeable after it had proven unmergeable (above).
1794 * We don't worry about that case for efficiency. It won't happen
1795 * often, and the elevators are able to handle it.
1797 blk_init_request_from_bio(req, bio);
1799 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1800 req->cpu = raw_smp_processor_id();
1802 plug = current->plug;
1805 * If this is the first request added after a plug, fire
1808 * @request_count may become stale because of schedule
1809 * out, so check plug list again.
1811 if (!request_count || list_empty(&plug->list))
1812 trace_block_plug(q);
1814 struct request *last = list_entry_rq(plug->list.prev);
1815 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1816 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1817 blk_flush_plug_list(plug, false);
1818 trace_block_plug(q);
1821 list_add_tail(&req->queuelist, &plug->list);
1822 blk_account_io_start(req, true);
1824 spin_lock_irq(q->queue_lock);
1825 add_acct_request(q, req, where);
1828 spin_unlock_irq(q->queue_lock);
1831 return BLK_QC_T_NONE;
1835 * If bio->bi_dev is a partition, remap the location
1837 static inline void blk_partition_remap(struct bio *bio)
1839 struct block_device *bdev = bio->bi_bdev;
1842 * Zone reset does not include bi_size so bio_sectors() is always 0.
1843 * Include a test for the reset op code and perform the remap if needed.
1845 if (bdev != bdev->bd_contains &&
1846 (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1847 struct hd_struct *p = bdev->bd_part;
1849 bio->bi_iter.bi_sector += p->start_sect;
1850 bio->bi_bdev = bdev->bd_contains;
1852 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1854 bio->bi_iter.bi_sector - p->start_sect);
1858 static void handle_bad_sector(struct bio *bio)
1860 char b[BDEVNAME_SIZE];
1862 printk(KERN_INFO "attempt to access beyond end of device\n");
1863 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1864 bdevname(bio->bi_bdev, b),
1866 (unsigned long long)bio_end_sector(bio),
1867 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1870 #ifdef CONFIG_FAIL_MAKE_REQUEST
1872 static DECLARE_FAULT_ATTR(fail_make_request);
1874 static int __init setup_fail_make_request(char *str)
1876 return setup_fault_attr(&fail_make_request, str);
1878 __setup("fail_make_request=", setup_fail_make_request);
1880 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1882 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1885 static int __init fail_make_request_debugfs(void)
1887 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1888 NULL, &fail_make_request);
1890 return PTR_ERR_OR_ZERO(dir);
1893 late_initcall(fail_make_request_debugfs);
1895 #else /* CONFIG_FAIL_MAKE_REQUEST */
1897 static inline bool should_fail_request(struct hd_struct *part,
1903 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1906 * Check whether this bio extends beyond the end of the device.
1908 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1915 /* Test device or partition size, when known. */
1916 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1918 sector_t sector = bio->bi_iter.bi_sector;
1920 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1922 * This may well happen - the kernel calls bread()
1923 * without checking the size of the device, e.g., when
1924 * mounting a device.
1926 handle_bad_sector(bio);
1934 static noinline_for_stack bool
1935 generic_make_request_checks(struct bio *bio)
1937 struct request_queue *q;
1938 int nr_sectors = bio_sectors(bio);
1939 blk_status_t status = BLK_STS_IOERR;
1940 char b[BDEVNAME_SIZE];
1941 struct hd_struct *part;
1945 if (bio_check_eod(bio, nr_sectors))
1948 q = bdev_get_queue(bio->bi_bdev);
1951 "generic_make_request: Trying to access "
1952 "nonexistent block-device %s (%Lu)\n",
1953 bdevname(bio->bi_bdev, b),
1954 (long long) bio->bi_iter.bi_sector);
1958 part = bio->bi_bdev->bd_part;
1959 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1960 should_fail_request(&part_to_disk(part)->part0,
1961 bio->bi_iter.bi_size))
1965 * If this device has partitions, remap block n
1966 * of partition p to block n+start(p) of the disk.
1968 blk_partition_remap(bio);
1970 if (bio_check_eod(bio, nr_sectors))
1974 * Filter flush bio's early so that make_request based
1975 * drivers without flush support don't have to worry
1978 if (op_is_flush(bio->bi_opf) &&
1979 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1980 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1982 status = BLK_STS_OK;
1987 switch (bio_op(bio)) {
1988 case REQ_OP_DISCARD:
1989 if (!blk_queue_discard(q))
1992 case REQ_OP_SECURE_ERASE:
1993 if (!blk_queue_secure_erase(q))
1996 case REQ_OP_WRITE_SAME:
1997 if (!bdev_write_same(bio->bi_bdev))
2000 case REQ_OP_ZONE_REPORT:
2001 case REQ_OP_ZONE_RESET:
2002 if (!bdev_is_zoned(bio->bi_bdev))
2005 case REQ_OP_WRITE_ZEROES:
2006 if (!bdev_write_zeroes_sectors(bio->bi_bdev))
2014 * Various block parts want %current->io_context and lazy ioc
2015 * allocation ends up trading a lot of pain for a small amount of
2016 * memory. Just allocate it upfront. This may fail and block
2017 * layer knows how to live with it.
2019 create_io_context(GFP_ATOMIC, q->node);
2021 if (!blkcg_bio_issue_check(q, bio))
2024 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2025 trace_block_bio_queue(q, bio);
2026 /* Now that enqueuing has been traced, we need to trace
2027 * completion as well.
2029 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2034 status = BLK_STS_NOTSUPP;
2036 bio->bi_status = status;
2042 * generic_make_request - hand a buffer to its device driver for I/O
2043 * @bio: The bio describing the location in memory and on the device.
2045 * generic_make_request() is used to make I/O requests of block
2046 * devices. It is passed a &struct bio, which describes the I/O that needs
2049 * generic_make_request() does not return any status. The
2050 * success/failure status of the request, along with notification of
2051 * completion, is delivered asynchronously through the bio->bi_end_io
2052 * function described (one day) else where.
2054 * The caller of generic_make_request must make sure that bi_io_vec
2055 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2056 * set to describe the device address, and the
2057 * bi_end_io and optionally bi_private are set to describe how
2058 * completion notification should be signaled.
2060 * generic_make_request and the drivers it calls may use bi_next if this
2061 * bio happens to be merged with someone else, and may resubmit the bio to
2062 * a lower device by calling into generic_make_request recursively, which
2063 * means the bio should NOT be touched after the call to ->make_request_fn.
2065 blk_qc_t generic_make_request(struct bio *bio)
2068 * bio_list_on_stack[0] contains bios submitted by the current
2070 * bio_list_on_stack[1] contains bios that were submitted before
2071 * the current make_request_fn, but that haven't been processed
2074 struct bio_list bio_list_on_stack[2];
2075 blk_qc_t ret = BLK_QC_T_NONE;
2077 if (!generic_make_request_checks(bio))
2081 * We only want one ->make_request_fn to be active at a time, else
2082 * stack usage with stacked devices could be a problem. So use
2083 * current->bio_list to keep a list of requests submited by a
2084 * make_request_fn function. current->bio_list is also used as a
2085 * flag to say if generic_make_request is currently active in this
2086 * task or not. If it is NULL, then no make_request is active. If
2087 * it is non-NULL, then a make_request is active, and new requests
2088 * should be added at the tail
2090 if (current->bio_list) {
2091 bio_list_add(¤t->bio_list[0], bio);
2095 /* following loop may be a bit non-obvious, and so deserves some
2097 * Before entering the loop, bio->bi_next is NULL (as all callers
2098 * ensure that) so we have a list with a single bio.
2099 * We pretend that we have just taken it off a longer list, so
2100 * we assign bio_list to a pointer to the bio_list_on_stack,
2101 * thus initialising the bio_list of new bios to be
2102 * added. ->make_request() may indeed add some more bios
2103 * through a recursive call to generic_make_request. If it
2104 * did, we find a non-NULL value in bio_list and re-enter the loop
2105 * from the top. In this case we really did just take the bio
2106 * of the top of the list (no pretending) and so remove it from
2107 * bio_list, and call into ->make_request() again.
2109 BUG_ON(bio->bi_next);
2110 bio_list_init(&bio_list_on_stack[0]);
2111 current->bio_list = bio_list_on_stack;
2113 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2115 if (likely(blk_queue_enter(q, false) == 0)) {
2116 struct bio_list lower, same;
2118 /* Create a fresh bio_list for all subordinate requests */
2119 bio_list_on_stack[1] = bio_list_on_stack[0];
2120 bio_list_init(&bio_list_on_stack[0]);
2121 ret = q->make_request_fn(q, bio);
2125 /* sort new bios into those for a lower level
2126 * and those for the same level
2128 bio_list_init(&lower);
2129 bio_list_init(&same);
2130 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2131 if (q == bdev_get_queue(bio->bi_bdev))
2132 bio_list_add(&same, bio);
2134 bio_list_add(&lower, bio);
2135 /* now assemble so we handle the lowest level first */
2136 bio_list_merge(&bio_list_on_stack[0], &lower);
2137 bio_list_merge(&bio_list_on_stack[0], &same);
2138 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2142 bio = bio_list_pop(&bio_list_on_stack[0]);
2144 current->bio_list = NULL; /* deactivate */
2149 EXPORT_SYMBOL(generic_make_request);
2152 * submit_bio - submit a bio to the block device layer for I/O
2153 * @bio: The &struct bio which describes the I/O
2155 * submit_bio() is very similar in purpose to generic_make_request(), and
2156 * uses that function to do most of the work. Both are fairly rough
2157 * interfaces; @bio must be presetup and ready for I/O.
2160 blk_qc_t submit_bio(struct bio *bio)
2163 * If it's a regular read/write or a barrier with data attached,
2164 * go through the normal accounting stuff before submission.
2166 if (bio_has_data(bio)) {
2169 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2170 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2172 count = bio_sectors(bio);
2174 if (op_is_write(bio_op(bio))) {
2175 count_vm_events(PGPGOUT, count);
2177 task_io_account_read(bio->bi_iter.bi_size);
2178 count_vm_events(PGPGIN, count);
2181 if (unlikely(block_dump)) {
2182 char b[BDEVNAME_SIZE];
2183 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2184 current->comm, task_pid_nr(current),
2185 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2186 (unsigned long long)bio->bi_iter.bi_sector,
2187 bdevname(bio->bi_bdev, b),
2192 return generic_make_request(bio);
2194 EXPORT_SYMBOL(submit_bio);
2197 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2198 * for new the queue limits
2200 * @rq: the request being checked
2203 * @rq may have been made based on weaker limitations of upper-level queues
2204 * in request stacking drivers, and it may violate the limitation of @q.
2205 * Since the block layer and the underlying device driver trust @rq
2206 * after it is inserted to @q, it should be checked against @q before
2207 * the insertion using this generic function.
2209 * Request stacking drivers like request-based dm may change the queue
2210 * limits when retrying requests on other queues. Those requests need
2211 * to be checked against the new queue limits again during dispatch.
2213 static int blk_cloned_rq_check_limits(struct request_queue *q,
2216 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2217 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2222 * queue's settings related to segment counting like q->bounce_pfn
2223 * may differ from that of other stacking queues.
2224 * Recalculate it to check the request correctly on this queue's
2227 blk_recalc_rq_segments(rq);
2228 if (rq->nr_phys_segments > queue_max_segments(q)) {
2229 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2237 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2238 * @q: the queue to submit the request
2239 * @rq: the request being queued
2241 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2243 unsigned long flags;
2244 int where = ELEVATOR_INSERT_BACK;
2246 if (blk_cloned_rq_check_limits(q, rq))
2247 return BLK_STS_IOERR;
2250 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2251 return BLK_STS_IOERR;
2254 if (blk_queue_io_stat(q))
2255 blk_account_io_start(rq, true);
2256 blk_mq_sched_insert_request(rq, false, true, false, false);
2260 spin_lock_irqsave(q->queue_lock, flags);
2261 if (unlikely(blk_queue_dying(q))) {
2262 spin_unlock_irqrestore(q->queue_lock, flags);
2263 return BLK_STS_IOERR;
2267 * Submitting request must be dequeued before calling this function
2268 * because it will be linked to another request_queue
2270 BUG_ON(blk_queued_rq(rq));
2272 if (op_is_flush(rq->cmd_flags))
2273 where = ELEVATOR_INSERT_FLUSH;
2275 add_acct_request(q, rq, where);
2276 if (where == ELEVATOR_INSERT_FLUSH)
2278 spin_unlock_irqrestore(q->queue_lock, flags);
2282 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2285 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2286 * @rq: request to examine
2289 * A request could be merge of IOs which require different failure
2290 * handling. This function determines the number of bytes which
2291 * can be failed from the beginning of the request without
2292 * crossing into area which need to be retried further.
2295 * The number of bytes to fail.
2298 * queue_lock must be held.
2300 unsigned int blk_rq_err_bytes(const struct request *rq)
2302 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2303 unsigned int bytes = 0;
2306 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2307 return blk_rq_bytes(rq);
2310 * Currently the only 'mixing' which can happen is between
2311 * different fastfail types. We can safely fail portions
2312 * which have all the failfast bits that the first one has -
2313 * the ones which are at least as eager to fail as the first
2316 for (bio = rq->bio; bio; bio = bio->bi_next) {
2317 if ((bio->bi_opf & ff) != ff)
2319 bytes += bio->bi_iter.bi_size;
2322 /* this could lead to infinite loop */
2323 BUG_ON(blk_rq_bytes(rq) && !bytes);
2326 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2328 void blk_account_io_completion(struct request *req, unsigned int bytes)
2330 if (blk_do_io_stat(req)) {
2331 const int rw = rq_data_dir(req);
2332 struct hd_struct *part;
2335 cpu = part_stat_lock();
2337 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2342 void blk_account_io_done(struct request *req)
2345 * Account IO completion. flush_rq isn't accounted as a
2346 * normal IO on queueing nor completion. Accounting the
2347 * containing request is enough.
2349 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2350 unsigned long duration = jiffies - req->start_time;
2351 const int rw = rq_data_dir(req);
2352 struct hd_struct *part;
2355 cpu = part_stat_lock();
2358 part_stat_inc(cpu, part, ios[rw]);
2359 part_stat_add(cpu, part, ticks[rw], duration);
2360 part_round_stats(cpu, part);
2361 part_dec_in_flight(part, rw);
2363 hd_struct_put(part);
2370 * Don't process normal requests when queue is suspended
2371 * or in the process of suspending/resuming
2373 static struct request *blk_pm_peek_request(struct request_queue *q,
2376 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2377 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2383 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2390 void blk_account_io_start(struct request *rq, bool new_io)
2392 struct hd_struct *part;
2393 int rw = rq_data_dir(rq);
2396 if (!blk_do_io_stat(rq))
2399 cpu = part_stat_lock();
2403 part_stat_inc(cpu, part, merges[rw]);
2405 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2406 if (!hd_struct_try_get(part)) {
2408 * The partition is already being removed,
2409 * the request will be accounted on the disk only
2411 * We take a reference on disk->part0 although that
2412 * partition will never be deleted, so we can treat
2413 * it as any other partition.
2415 part = &rq->rq_disk->part0;
2416 hd_struct_get(part);
2418 part_round_stats(cpu, part);
2419 part_inc_in_flight(part, rw);
2427 * blk_peek_request - peek at the top of a request queue
2428 * @q: request queue to peek at
2431 * Return the request at the top of @q. The returned request
2432 * should be started using blk_start_request() before LLD starts
2436 * Pointer to the request at the top of @q if available. Null
2440 * queue_lock must be held.
2442 struct request *blk_peek_request(struct request_queue *q)
2447 while ((rq = __elv_next_request(q)) != NULL) {
2449 rq = blk_pm_peek_request(q, rq);
2453 if (!(rq->rq_flags & RQF_STARTED)) {
2455 * This is the first time the device driver
2456 * sees this request (possibly after
2457 * requeueing). Notify IO scheduler.
2459 if (rq->rq_flags & RQF_SORTED)
2460 elv_activate_rq(q, rq);
2463 * just mark as started even if we don't start
2464 * it, a request that has been delayed should
2465 * not be passed by new incoming requests
2467 rq->rq_flags |= RQF_STARTED;
2468 trace_block_rq_issue(q, rq);
2471 if (!q->boundary_rq || q->boundary_rq == rq) {
2472 q->end_sector = rq_end_sector(rq);
2473 q->boundary_rq = NULL;
2476 if (rq->rq_flags & RQF_DONTPREP)
2479 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2481 * make sure space for the drain appears we
2482 * know we can do this because max_hw_segments
2483 * has been adjusted to be one fewer than the
2486 rq->nr_phys_segments++;
2492 ret = q->prep_rq_fn(q, rq);
2493 if (ret == BLKPREP_OK) {
2495 } else if (ret == BLKPREP_DEFER) {
2497 * the request may have been (partially) prepped.
2498 * we need to keep this request in the front to
2499 * avoid resource deadlock. RQF_STARTED will
2500 * prevent other fs requests from passing this one.
2502 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2503 !(rq->rq_flags & RQF_DONTPREP)) {
2505 * remove the space for the drain we added
2506 * so that we don't add it again
2508 --rq->nr_phys_segments;
2513 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2514 rq->rq_flags |= RQF_QUIET;
2516 * Mark this request as started so we don't trigger
2517 * any debug logic in the end I/O path.
2519 blk_start_request(rq);
2520 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2521 BLK_STS_TARGET : BLK_STS_IOERR);
2523 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2530 EXPORT_SYMBOL(blk_peek_request);
2532 void blk_dequeue_request(struct request *rq)
2534 struct request_queue *q = rq->q;
2536 BUG_ON(list_empty(&rq->queuelist));
2537 BUG_ON(ELV_ON_HASH(rq));
2539 list_del_init(&rq->queuelist);
2542 * the time frame between a request being removed from the lists
2543 * and to it is freed is accounted as io that is in progress at
2546 if (blk_account_rq(rq)) {
2547 q->in_flight[rq_is_sync(rq)]++;
2548 set_io_start_time_ns(rq);
2553 * blk_start_request - start request processing on the driver
2554 * @req: request to dequeue
2557 * Dequeue @req and start timeout timer on it. This hands off the
2558 * request to the driver.
2560 * Block internal functions which don't want to start timer should
2561 * call blk_dequeue_request().
2564 * queue_lock must be held.
2566 void blk_start_request(struct request *req)
2568 blk_dequeue_request(req);
2570 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2571 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2572 req->rq_flags |= RQF_STATS;
2573 wbt_issue(req->q->rq_wb, &req->issue_stat);
2576 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2579 EXPORT_SYMBOL(blk_start_request);
2582 * blk_fetch_request - fetch a request from a request queue
2583 * @q: request queue to fetch a request from
2586 * Return the request at the top of @q. The request is started on
2587 * return and LLD can start processing it immediately.
2590 * Pointer to the request at the top of @q if available. Null
2594 * queue_lock must be held.
2596 struct request *blk_fetch_request(struct request_queue *q)
2600 rq = blk_peek_request(q);
2602 blk_start_request(rq);
2605 EXPORT_SYMBOL(blk_fetch_request);
2608 * blk_update_request - Special helper function for request stacking drivers
2609 * @req: the request being processed
2610 * @error: block status code
2611 * @nr_bytes: number of bytes to complete @req
2614 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2615 * the request structure even if @req doesn't have leftover.
2616 * If @req has leftover, sets it up for the next range of segments.
2618 * This special helper function is only for request stacking drivers
2619 * (e.g. request-based dm) so that they can handle partial completion.
2620 * Actual device drivers should use blk_end_request instead.
2622 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2623 * %false return from this function.
2626 * %false - this request doesn't have any more data
2627 * %true - this request has more data
2629 bool blk_update_request(struct request *req, blk_status_t error,
2630 unsigned int nr_bytes)
2634 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2639 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2640 !(req->rq_flags & RQF_QUIET)))
2641 print_req_error(req, error);
2643 blk_account_io_completion(req, nr_bytes);
2647 struct bio *bio = req->bio;
2648 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2650 if (bio_bytes == bio->bi_iter.bi_size)
2651 req->bio = bio->bi_next;
2653 /* Completion has already been traced */
2654 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2655 req_bio_endio(req, bio, bio_bytes, error);
2657 total_bytes += bio_bytes;
2658 nr_bytes -= bio_bytes;
2669 * Reset counters so that the request stacking driver
2670 * can find how many bytes remain in the request
2673 req->__data_len = 0;
2677 req->__data_len -= total_bytes;
2679 /* update sector only for requests with clear definition of sector */
2680 if (!blk_rq_is_passthrough(req))
2681 req->__sector += total_bytes >> 9;
2683 /* mixed attributes always follow the first bio */
2684 if (req->rq_flags & RQF_MIXED_MERGE) {
2685 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2686 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2689 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2691 * If total number of sectors is less than the first segment
2692 * size, something has gone terribly wrong.
2694 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2695 blk_dump_rq_flags(req, "request botched");
2696 req->__data_len = blk_rq_cur_bytes(req);
2699 /* recalculate the number of segments */
2700 blk_recalc_rq_segments(req);
2705 EXPORT_SYMBOL_GPL(blk_update_request);
2707 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2708 unsigned int nr_bytes,
2709 unsigned int bidi_bytes)
2711 if (blk_update_request(rq, error, nr_bytes))
2714 /* Bidi request must be completed as a whole */
2715 if (unlikely(blk_bidi_rq(rq)) &&
2716 blk_update_request(rq->next_rq, error, bidi_bytes))
2719 if (blk_queue_add_random(rq->q))
2720 add_disk_randomness(rq->rq_disk);
2726 * blk_unprep_request - unprepare a request
2729 * This function makes a request ready for complete resubmission (or
2730 * completion). It happens only after all error handling is complete,
2731 * so represents the appropriate moment to deallocate any resources
2732 * that were allocated to the request in the prep_rq_fn. The queue
2733 * lock is held when calling this.
2735 void blk_unprep_request(struct request *req)
2737 struct request_queue *q = req->q;
2739 req->rq_flags &= ~RQF_DONTPREP;
2740 if (q->unprep_rq_fn)
2741 q->unprep_rq_fn(q, req);
2743 EXPORT_SYMBOL_GPL(blk_unprep_request);
2746 * queue lock must be held
2748 void blk_finish_request(struct request *req, blk_status_t error)
2750 struct request_queue *q = req->q;
2752 if (req->rq_flags & RQF_STATS)
2755 if (req->rq_flags & RQF_QUEUED)
2756 blk_queue_end_tag(q, req);
2758 BUG_ON(blk_queued_rq(req));
2760 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
2761 laptop_io_completion(req->q->backing_dev_info);
2763 blk_delete_timer(req);
2765 if (req->rq_flags & RQF_DONTPREP)
2766 blk_unprep_request(req);
2768 blk_account_io_done(req);
2771 wbt_done(req->q->rq_wb, &req->issue_stat);
2772 req->end_io(req, error);
2774 if (blk_bidi_rq(req))
2775 __blk_put_request(req->next_rq->q, req->next_rq);
2777 __blk_put_request(q, req);
2780 EXPORT_SYMBOL(blk_finish_request);
2783 * blk_end_bidi_request - Complete a bidi request
2784 * @rq: the request to complete
2785 * @error: block status code
2786 * @nr_bytes: number of bytes to complete @rq
2787 * @bidi_bytes: number of bytes to complete @rq->next_rq
2790 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2791 * Drivers that supports bidi can safely call this member for any
2792 * type of request, bidi or uni. In the later case @bidi_bytes is
2796 * %false - we are done with this request
2797 * %true - still buffers pending for this request
2799 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
2800 unsigned int nr_bytes, unsigned int bidi_bytes)
2802 struct request_queue *q = rq->q;
2803 unsigned long flags;
2805 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2808 spin_lock_irqsave(q->queue_lock, flags);
2809 blk_finish_request(rq, error);
2810 spin_unlock_irqrestore(q->queue_lock, flags);
2816 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2817 * @rq: the request to complete
2818 * @error: block status code
2819 * @nr_bytes: number of bytes to complete @rq
2820 * @bidi_bytes: number of bytes to complete @rq->next_rq
2823 * Identical to blk_end_bidi_request() except that queue lock is
2824 * assumed to be locked on entry and remains so on return.
2827 * %false - we are done with this request
2828 * %true - still buffers pending for this request
2830 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
2831 unsigned int nr_bytes, unsigned int bidi_bytes)
2833 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2836 blk_finish_request(rq, error);
2842 * blk_end_request - Helper function for drivers to complete the request.
2843 * @rq: the request being processed
2844 * @error: block status code
2845 * @nr_bytes: number of bytes to complete
2848 * Ends I/O on a number of bytes attached to @rq.
2849 * If @rq has leftover, sets it up for the next range of segments.
2852 * %false - we are done with this request
2853 * %true - still buffers pending for this request
2855 bool blk_end_request(struct request *rq, blk_status_t error,
2856 unsigned int nr_bytes)
2858 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2860 EXPORT_SYMBOL(blk_end_request);
2863 * blk_end_request_all - Helper function for drives to finish the request.
2864 * @rq: the request to finish
2865 * @error: block status code
2868 * Completely finish @rq.
2870 void blk_end_request_all(struct request *rq, blk_status_t error)
2873 unsigned int bidi_bytes = 0;
2875 if (unlikely(blk_bidi_rq(rq)))
2876 bidi_bytes = blk_rq_bytes(rq->next_rq);
2878 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2881 EXPORT_SYMBOL(blk_end_request_all);
2884 * __blk_end_request - Helper function for drivers to complete the request.
2885 * @rq: the request being processed
2886 * @error: block status code
2887 * @nr_bytes: number of bytes to complete
2890 * Must be called with queue lock held unlike blk_end_request().
2893 * %false - we are done with this request
2894 * %true - still buffers pending for this request
2896 bool __blk_end_request(struct request *rq, blk_status_t error,
2897 unsigned int nr_bytes)
2899 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2901 EXPORT_SYMBOL(__blk_end_request);
2904 * __blk_end_request_all - Helper function for drives to finish the request.
2905 * @rq: the request to finish
2906 * @error: block status code
2909 * Completely finish @rq. Must be called with queue lock held.
2911 void __blk_end_request_all(struct request *rq, blk_status_t error)
2914 unsigned int bidi_bytes = 0;
2916 if (unlikely(blk_bidi_rq(rq)))
2917 bidi_bytes = blk_rq_bytes(rq->next_rq);
2919 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2922 EXPORT_SYMBOL(__blk_end_request_all);
2925 * __blk_end_request_cur - Helper function to finish the current request chunk.
2926 * @rq: the request to finish the current chunk for
2927 * @error: block status code
2930 * Complete the current consecutively mapped chunk from @rq. Must
2931 * be called with queue lock held.
2934 * %false - we are done with this request
2935 * %true - still buffers pending for this request
2937 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
2939 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2941 EXPORT_SYMBOL(__blk_end_request_cur);
2943 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2946 if (bio_has_data(bio))
2947 rq->nr_phys_segments = bio_phys_segments(q, bio);
2949 rq->__data_len = bio->bi_iter.bi_size;
2950 rq->bio = rq->biotail = bio;
2953 rq->rq_disk = bio->bi_bdev->bd_disk;
2956 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2958 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2959 * @rq: the request to be flushed
2962 * Flush all pages in @rq.
2964 void rq_flush_dcache_pages(struct request *rq)
2966 struct req_iterator iter;
2967 struct bio_vec bvec;
2969 rq_for_each_segment(bvec, rq, iter)
2970 flush_dcache_page(bvec.bv_page);
2972 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2976 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2977 * @q : the queue of the device being checked
2980 * Check if underlying low-level drivers of a device are busy.
2981 * If the drivers want to export their busy state, they must set own
2982 * exporting function using blk_queue_lld_busy() first.
2984 * Basically, this function is used only by request stacking drivers
2985 * to stop dispatching requests to underlying devices when underlying
2986 * devices are busy. This behavior helps more I/O merging on the queue
2987 * of the request stacking driver and prevents I/O throughput regression
2988 * on burst I/O load.
2991 * 0 - Not busy (The request stacking driver should dispatch request)
2992 * 1 - Busy (The request stacking driver should stop dispatching request)
2994 int blk_lld_busy(struct request_queue *q)
2997 return q->lld_busy_fn(q);
3001 EXPORT_SYMBOL_GPL(blk_lld_busy);
3004 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3005 * @rq: the clone request to be cleaned up
3008 * Free all bios in @rq for a cloned request.
3010 void blk_rq_unprep_clone(struct request *rq)
3014 while ((bio = rq->bio) != NULL) {
3015 rq->bio = bio->bi_next;
3020 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3023 * Copy attributes of the original request to the clone request.
3024 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3026 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3028 dst->cpu = src->cpu;
3029 dst->__sector = blk_rq_pos(src);
3030 dst->__data_len = blk_rq_bytes(src);
3031 dst->nr_phys_segments = src->nr_phys_segments;
3032 dst->ioprio = src->ioprio;
3033 dst->extra_len = src->extra_len;
3037 * blk_rq_prep_clone - Helper function to setup clone request
3038 * @rq: the request to be setup
3039 * @rq_src: original request to be cloned
3040 * @bs: bio_set that bios for clone are allocated from
3041 * @gfp_mask: memory allocation mask for bio
3042 * @bio_ctr: setup function to be called for each clone bio.
3043 * Returns %0 for success, non %0 for failure.
3044 * @data: private data to be passed to @bio_ctr
3047 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3048 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3049 * are not copied, and copying such parts is the caller's responsibility.
3050 * Also, pages which the original bios are pointing to are not copied
3051 * and the cloned bios just point same pages.
3052 * So cloned bios must be completed before original bios, which means
3053 * the caller must complete @rq before @rq_src.
3055 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3056 struct bio_set *bs, gfp_t gfp_mask,
3057 int (*bio_ctr)(struct bio *, struct bio *, void *),
3060 struct bio *bio, *bio_src;
3065 __rq_for_each_bio(bio_src, rq_src) {
3066 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3070 if (bio_ctr && bio_ctr(bio, bio_src, data))
3074 rq->biotail->bi_next = bio;
3077 rq->bio = rq->biotail = bio;
3080 __blk_rq_prep_clone(rq, rq_src);
3087 blk_rq_unprep_clone(rq);
3091 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3093 int kblockd_schedule_work(struct work_struct *work)
3095 return queue_work(kblockd_workqueue, work);
3097 EXPORT_SYMBOL(kblockd_schedule_work);
3099 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3101 return queue_work_on(cpu, kblockd_workqueue, work);
3103 EXPORT_SYMBOL(kblockd_schedule_work_on);
3105 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3106 unsigned long delay)
3108 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3110 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3112 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3113 unsigned long delay)
3115 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3117 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3119 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3120 unsigned long delay)
3122 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3124 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3127 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3128 * @plug: The &struct blk_plug that needs to be initialized
3131 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3132 * pending I/O should the task end up blocking between blk_start_plug() and
3133 * blk_finish_plug(). This is important from a performance perspective, but
3134 * also ensures that we don't deadlock. For instance, if the task is blocking
3135 * for a memory allocation, memory reclaim could end up wanting to free a
3136 * page belonging to that request that is currently residing in our private
3137 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3138 * this kind of deadlock.
3140 void blk_start_plug(struct blk_plug *plug)
3142 struct task_struct *tsk = current;
3145 * If this is a nested plug, don't actually assign it.
3150 INIT_LIST_HEAD(&plug->list);
3151 INIT_LIST_HEAD(&plug->mq_list);
3152 INIT_LIST_HEAD(&plug->cb_list);
3154 * Store ordering should not be needed here, since a potential
3155 * preempt will imply a full memory barrier
3159 EXPORT_SYMBOL(blk_start_plug);
3161 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3163 struct request *rqa = container_of(a, struct request, queuelist);
3164 struct request *rqb = container_of(b, struct request, queuelist);
3166 return !(rqa->q < rqb->q ||
3167 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3171 * If 'from_schedule' is true, then postpone the dispatch of requests
3172 * until a safe kblockd context. We due this to avoid accidental big
3173 * additional stack usage in driver dispatch, in places where the originally
3174 * plugger did not intend it.
3176 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3178 __releases(q->queue_lock)
3180 trace_block_unplug(q, depth, !from_schedule);
3183 blk_run_queue_async(q);
3186 spin_unlock(q->queue_lock);
3189 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3191 LIST_HEAD(callbacks);
3193 while (!list_empty(&plug->cb_list)) {
3194 list_splice_init(&plug->cb_list, &callbacks);
3196 while (!list_empty(&callbacks)) {
3197 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3200 list_del(&cb->list);
3201 cb->callback(cb, from_schedule);
3206 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3209 struct blk_plug *plug = current->plug;
3210 struct blk_plug_cb *cb;
3215 list_for_each_entry(cb, &plug->cb_list, list)
3216 if (cb->callback == unplug && cb->data == data)
3219 /* Not currently on the callback list */
3220 BUG_ON(size < sizeof(*cb));
3221 cb = kzalloc(size, GFP_ATOMIC);
3224 cb->callback = unplug;
3225 list_add(&cb->list, &plug->cb_list);
3229 EXPORT_SYMBOL(blk_check_plugged);
3231 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3233 struct request_queue *q;
3234 unsigned long flags;
3239 flush_plug_callbacks(plug, from_schedule);
3241 if (!list_empty(&plug->mq_list))
3242 blk_mq_flush_plug_list(plug, from_schedule);
3244 if (list_empty(&plug->list))
3247 list_splice_init(&plug->list, &list);
3249 list_sort(NULL, &list, plug_rq_cmp);
3255 * Save and disable interrupts here, to avoid doing it for every
3256 * queue lock we have to take.
3258 local_irq_save(flags);
3259 while (!list_empty(&list)) {
3260 rq = list_entry_rq(list.next);
3261 list_del_init(&rq->queuelist);
3265 * This drops the queue lock
3268 queue_unplugged(q, depth, from_schedule);
3271 spin_lock(q->queue_lock);
3275 * Short-circuit if @q is dead
3277 if (unlikely(blk_queue_dying(q))) {
3278 __blk_end_request_all(rq, BLK_STS_IOERR);
3283 * rq is already accounted, so use raw insert
3285 if (op_is_flush(rq->cmd_flags))
3286 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3288 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3294 * This drops the queue lock
3297 queue_unplugged(q, depth, from_schedule);
3299 local_irq_restore(flags);
3302 void blk_finish_plug(struct blk_plug *plug)
3304 if (plug != current->plug)
3306 blk_flush_plug_list(plug, false);
3308 current->plug = NULL;
3310 EXPORT_SYMBOL(blk_finish_plug);
3314 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3315 * @q: the queue of the device
3316 * @dev: the device the queue belongs to
3319 * Initialize runtime-PM-related fields for @q and start auto suspend for
3320 * @dev. Drivers that want to take advantage of request-based runtime PM
3321 * should call this function after @dev has been initialized, and its
3322 * request queue @q has been allocated, and runtime PM for it can not happen
3323 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3324 * cases, driver should call this function before any I/O has taken place.
3326 * This function takes care of setting up using auto suspend for the device,
3327 * the autosuspend delay is set to -1 to make runtime suspend impossible
3328 * until an updated value is either set by user or by driver. Drivers do
3329 * not need to touch other autosuspend settings.
3331 * The block layer runtime PM is request based, so only works for drivers
3332 * that use request as their IO unit instead of those directly use bio's.
3334 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3337 q->rpm_status = RPM_ACTIVE;
3338 pm_runtime_set_autosuspend_delay(q->dev, -1);
3339 pm_runtime_use_autosuspend(q->dev);
3341 EXPORT_SYMBOL(blk_pm_runtime_init);
3344 * blk_pre_runtime_suspend - Pre runtime suspend check
3345 * @q: the queue of the device
3348 * This function will check if runtime suspend is allowed for the device
3349 * by examining if there are any requests pending in the queue. If there
3350 * are requests pending, the device can not be runtime suspended; otherwise,
3351 * the queue's status will be updated to SUSPENDING and the driver can
3352 * proceed to suspend the device.
3354 * For the not allowed case, we mark last busy for the device so that
3355 * runtime PM core will try to autosuspend it some time later.
3357 * This function should be called near the start of the device's
3358 * runtime_suspend callback.
3361 * 0 - OK to runtime suspend the device
3362 * -EBUSY - Device should not be runtime suspended
3364 int blk_pre_runtime_suspend(struct request_queue *q)
3371 spin_lock_irq(q->queue_lock);
3372 if (q->nr_pending) {
3374 pm_runtime_mark_last_busy(q->dev);
3376 q->rpm_status = RPM_SUSPENDING;
3378 spin_unlock_irq(q->queue_lock);
3381 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3384 * blk_post_runtime_suspend - Post runtime suspend processing
3385 * @q: the queue of the device
3386 * @err: return value of the device's runtime_suspend function
3389 * Update the queue's runtime status according to the return value of the
3390 * device's runtime suspend function and mark last busy for the device so
3391 * that PM core will try to auto suspend the device at a later time.
3393 * This function should be called near the end of the device's
3394 * runtime_suspend callback.
3396 void blk_post_runtime_suspend(struct request_queue *q, int err)
3401 spin_lock_irq(q->queue_lock);
3403 q->rpm_status = RPM_SUSPENDED;
3405 q->rpm_status = RPM_ACTIVE;
3406 pm_runtime_mark_last_busy(q->dev);
3408 spin_unlock_irq(q->queue_lock);
3410 EXPORT_SYMBOL(blk_post_runtime_suspend);
3413 * blk_pre_runtime_resume - Pre runtime resume processing
3414 * @q: the queue of the device
3417 * Update the queue's runtime status to RESUMING in preparation for the
3418 * runtime resume of the device.
3420 * This function should be called near the start of the device's
3421 * runtime_resume callback.
3423 void blk_pre_runtime_resume(struct request_queue *q)
3428 spin_lock_irq(q->queue_lock);
3429 q->rpm_status = RPM_RESUMING;
3430 spin_unlock_irq(q->queue_lock);
3432 EXPORT_SYMBOL(blk_pre_runtime_resume);
3435 * blk_post_runtime_resume - Post runtime resume processing
3436 * @q: the queue of the device
3437 * @err: return value of the device's runtime_resume function
3440 * Update the queue's runtime status according to the return value of the
3441 * device's runtime_resume function. If it is successfully resumed, process
3442 * the requests that are queued into the device's queue when it is resuming
3443 * and then mark last busy and initiate autosuspend for it.
3445 * This function should be called near the end of the device's
3446 * runtime_resume callback.
3448 void blk_post_runtime_resume(struct request_queue *q, int err)
3453 spin_lock_irq(q->queue_lock);
3455 q->rpm_status = RPM_ACTIVE;
3457 pm_runtime_mark_last_busy(q->dev);
3458 pm_request_autosuspend(q->dev);
3460 q->rpm_status = RPM_SUSPENDED;
3462 spin_unlock_irq(q->queue_lock);
3464 EXPORT_SYMBOL(blk_post_runtime_resume);
3467 * blk_set_runtime_active - Force runtime status of the queue to be active
3468 * @q: the queue of the device
3470 * If the device is left runtime suspended during system suspend the resume
3471 * hook typically resumes the device and corrects runtime status
3472 * accordingly. However, that does not affect the queue runtime PM status
3473 * which is still "suspended". This prevents processing requests from the
3476 * This function can be used in driver's resume hook to correct queue
3477 * runtime PM status and re-enable peeking requests from the queue. It
3478 * should be called before first request is added to the queue.
3480 void blk_set_runtime_active(struct request_queue *q)
3482 spin_lock_irq(q->queue_lock);
3483 q->rpm_status = RPM_ACTIVE;
3484 pm_runtime_mark_last_busy(q->dev);
3485 pm_request_autosuspend(q->dev);
3486 spin_unlock_irq(q->queue_lock);
3488 EXPORT_SYMBOL(blk_set_runtime_active);
3491 int __init blk_dev_init(void)
3493 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3494 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3495 FIELD_SIZEOF(struct request, cmd_flags));
3496 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3497 FIELD_SIZEOF(struct bio, bi_opf));
3499 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3500 kblockd_workqueue = alloc_workqueue("kblockd",
3501 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3502 if (!kblockd_workqueue)
3503 panic("Failed to create kblockd\n");
3505 request_cachep = kmem_cache_create("blkdev_requests",
3506 sizeof(struct request), 0, SLAB_PANIC, NULL);
3508 blk_requestq_cachep = kmem_cache_create("request_queue",
3509 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3511 #ifdef CONFIG_DEBUG_FS
3512 blk_debugfs_root = debugfs_create_dir("block", NULL);