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 void req_bio_endio(struct request *rq, struct bio *bio,
133 unsigned int nbytes, int error)
136 bio->bi_error = error;
138 if (unlikely(rq->rq_flags & RQF_QUIET))
139 bio_set_flag(bio, BIO_QUIET);
141 bio_advance(bio, nbytes);
143 /* don't actually finish bio if it's part of flush sequence */
144 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
148 void blk_dump_rq_flags(struct request *rq, char *msg)
150 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
151 rq->rq_disk ? rq->rq_disk->disk_name : "?",
152 (unsigned long long) rq->cmd_flags);
154 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
155 (unsigned long long)blk_rq_pos(rq),
156 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
157 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
158 rq->bio, rq->biotail, blk_rq_bytes(rq));
160 EXPORT_SYMBOL(blk_dump_rq_flags);
162 static void blk_delay_work(struct work_struct *work)
164 struct request_queue *q;
166 q = container_of(work, struct request_queue, delay_work.work);
167 spin_lock_irq(q->queue_lock);
169 spin_unlock_irq(q->queue_lock);
173 * blk_delay_queue - restart queueing after defined interval
174 * @q: The &struct request_queue in question
175 * @msecs: Delay in msecs
178 * Sometimes queueing needs to be postponed for a little while, to allow
179 * resources to come back. This function will make sure that queueing is
180 * restarted around the specified time. Queue lock must be held.
182 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
184 if (likely(!blk_queue_dead(q)))
185 queue_delayed_work(kblockd_workqueue, &q->delay_work,
186 msecs_to_jiffies(msecs));
188 EXPORT_SYMBOL(blk_delay_queue);
191 * blk_start_queue_async - asynchronously restart a previously stopped queue
192 * @q: The &struct request_queue in question
195 * blk_start_queue_async() will clear the stop flag on the queue, and
196 * ensure that the request_fn for the queue is run from an async
199 void blk_start_queue_async(struct request_queue *q)
201 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
202 blk_run_queue_async(q);
204 EXPORT_SYMBOL(blk_start_queue_async);
207 * blk_start_queue - restart a previously stopped queue
208 * @q: The &struct request_queue in question
211 * blk_start_queue() will clear the stop flag on the queue, and call
212 * the request_fn for the queue if it was in a stopped state when
213 * entered. Also see blk_stop_queue(). Queue lock must be held.
215 void blk_start_queue(struct request_queue *q)
217 WARN_ON(!irqs_disabled());
219 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
222 EXPORT_SYMBOL(blk_start_queue);
225 * blk_stop_queue - stop a queue
226 * @q: The &struct request_queue in question
229 * The Linux block layer assumes that a block driver will consume all
230 * entries on the request queue when the request_fn strategy is called.
231 * Often this will not happen, because of hardware limitations (queue
232 * depth settings). If a device driver gets a 'queue full' response,
233 * or if it simply chooses not to queue more I/O at one point, it can
234 * call this function to prevent the request_fn from being called until
235 * the driver has signalled it's ready to go again. This happens by calling
236 * blk_start_queue() to restart queue operations. Queue lock must be held.
238 void blk_stop_queue(struct request_queue *q)
240 cancel_delayed_work(&q->delay_work);
241 queue_flag_set(QUEUE_FLAG_STOPPED, q);
243 EXPORT_SYMBOL(blk_stop_queue);
246 * blk_sync_queue - cancel any pending callbacks on a queue
250 * The block layer may perform asynchronous callback activity
251 * on a queue, such as calling the unplug function after a timeout.
252 * A block device may call blk_sync_queue to ensure that any
253 * such activity is cancelled, thus allowing it to release resources
254 * that the callbacks might use. The caller must already have made sure
255 * that its ->make_request_fn will not re-add plugging prior to calling
258 * This function does not cancel any asynchronous activity arising
259 * out of elevator or throttling code. That would require elevator_exit()
260 * and blkcg_exit_queue() to be called with queue lock initialized.
263 void blk_sync_queue(struct request_queue *q)
265 del_timer_sync(&q->timeout);
268 struct blk_mq_hw_ctx *hctx;
271 queue_for_each_hw_ctx(q, hctx, i) {
272 cancel_work_sync(&hctx->run_work);
273 cancel_delayed_work_sync(&hctx->delay_work);
276 cancel_delayed_work_sync(&q->delay_work);
279 EXPORT_SYMBOL(blk_sync_queue);
282 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
283 * @q: The queue to run
286 * Invoke request handling on a queue if there are any pending requests.
287 * May be used to restart request handling after a request has completed.
288 * This variant runs the queue whether or not the queue has been
289 * stopped. Must be called with the queue lock held and interrupts
290 * disabled. See also @blk_run_queue.
292 inline void __blk_run_queue_uncond(struct request_queue *q)
294 if (unlikely(blk_queue_dead(q)))
298 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
299 * the queue lock internally. As a result multiple threads may be
300 * running such a request function concurrently. Keep track of the
301 * number of active request_fn invocations such that blk_drain_queue()
302 * can wait until all these request_fn calls have finished.
304 q->request_fn_active++;
306 q->request_fn_active--;
308 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
311 * __blk_run_queue - run a single device queue
312 * @q: The queue to run
315 * See @blk_run_queue. This variant must be called with the queue lock
316 * held and interrupts disabled.
318 void __blk_run_queue(struct request_queue *q)
320 if (unlikely(blk_queue_stopped(q)))
323 __blk_run_queue_uncond(q);
325 EXPORT_SYMBOL(__blk_run_queue);
328 * blk_run_queue_async - run a single device queue in workqueue context
329 * @q: The queue to run
332 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
333 * of us. The caller must hold the queue lock.
335 void blk_run_queue_async(struct request_queue *q)
337 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
338 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
340 EXPORT_SYMBOL(blk_run_queue_async);
343 * blk_run_queue - run a single device queue
344 * @q: The queue to run
347 * Invoke request handling on this queue, if it has pending work to do.
348 * May be used to restart queueing when a request has completed.
350 void blk_run_queue(struct request_queue *q)
354 spin_lock_irqsave(q->queue_lock, flags);
356 spin_unlock_irqrestore(q->queue_lock, flags);
358 EXPORT_SYMBOL(blk_run_queue);
360 void blk_put_queue(struct request_queue *q)
362 kobject_put(&q->kobj);
364 EXPORT_SYMBOL(blk_put_queue);
367 * __blk_drain_queue - drain requests from request_queue
369 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
371 * Drain requests from @q. If @drain_all is set, all requests are drained.
372 * If not, only ELVPRIV requests are drained. The caller is responsible
373 * for ensuring that no new requests which need to be drained are queued.
375 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
376 __releases(q->queue_lock)
377 __acquires(q->queue_lock)
381 lockdep_assert_held(q->queue_lock);
387 * The caller might be trying to drain @q before its
388 * elevator is initialized.
391 elv_drain_elevator(q);
393 blkcg_drain_queue(q);
396 * This function might be called on a queue which failed
397 * driver init after queue creation or is not yet fully
398 * active yet. Some drivers (e.g. fd and loop) get unhappy
399 * in such cases. Kick queue iff dispatch queue has
400 * something on it and @q has request_fn set.
402 if (!list_empty(&q->queue_head) && q->request_fn)
405 drain |= q->nr_rqs_elvpriv;
406 drain |= q->request_fn_active;
409 * Unfortunately, requests are queued at and tracked from
410 * multiple places and there's no single counter which can
411 * be drained. Check all the queues and counters.
414 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
415 drain |= !list_empty(&q->queue_head);
416 for (i = 0; i < 2; i++) {
417 drain |= q->nr_rqs[i];
418 drain |= q->in_flight[i];
420 drain |= !list_empty(&fq->flush_queue[i]);
427 spin_unlock_irq(q->queue_lock);
431 spin_lock_irq(q->queue_lock);
435 * With queue marked dead, any woken up waiter will fail the
436 * allocation path, so the wakeup chaining is lost and we're
437 * left with hung waiters. We need to wake up those waiters.
440 struct request_list *rl;
442 blk_queue_for_each_rl(rl, q)
443 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
444 wake_up_all(&rl->wait[i]);
449 * blk_queue_bypass_start - enter queue bypass mode
450 * @q: queue of interest
452 * In bypass mode, only the dispatch FIFO queue of @q is used. This
453 * function makes @q enter bypass mode and drains all requests which were
454 * throttled or issued before. On return, it's guaranteed that no request
455 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
456 * inside queue or RCU read lock.
458 void blk_queue_bypass_start(struct request_queue *q)
460 spin_lock_irq(q->queue_lock);
462 queue_flag_set(QUEUE_FLAG_BYPASS, q);
463 spin_unlock_irq(q->queue_lock);
466 * Queues start drained. Skip actual draining till init is
467 * complete. This avoids lenghty delays during queue init which
468 * can happen many times during boot.
470 if (blk_queue_init_done(q)) {
471 spin_lock_irq(q->queue_lock);
472 __blk_drain_queue(q, false);
473 spin_unlock_irq(q->queue_lock);
475 /* ensure blk_queue_bypass() is %true inside RCU read lock */
479 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
482 * blk_queue_bypass_end - leave queue bypass mode
483 * @q: queue of interest
485 * Leave bypass mode and restore the normal queueing behavior.
487 void blk_queue_bypass_end(struct request_queue *q)
489 spin_lock_irq(q->queue_lock);
490 if (!--q->bypass_depth)
491 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
492 WARN_ON_ONCE(q->bypass_depth < 0);
493 spin_unlock_irq(q->queue_lock);
495 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
497 void blk_set_queue_dying(struct request_queue *q)
499 spin_lock_irq(q->queue_lock);
500 queue_flag_set(QUEUE_FLAG_DYING, q);
501 spin_unlock_irq(q->queue_lock);
504 blk_mq_wake_waiters(q);
506 struct request_list *rl;
508 spin_lock_irq(q->queue_lock);
509 blk_queue_for_each_rl(rl, q) {
511 wake_up(&rl->wait[BLK_RW_SYNC]);
512 wake_up(&rl->wait[BLK_RW_ASYNC]);
515 spin_unlock_irq(q->queue_lock);
518 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
521 * blk_cleanup_queue - shutdown a request queue
522 * @q: request queue to shutdown
524 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
525 * put it. All future requests will be failed immediately with -ENODEV.
527 void blk_cleanup_queue(struct request_queue *q)
529 spinlock_t *lock = q->queue_lock;
531 /* mark @q DYING, no new request or merges will be allowed afterwards */
532 mutex_lock(&q->sysfs_lock);
533 blk_set_queue_dying(q);
537 * A dying queue is permanently in bypass mode till released. Note
538 * that, unlike blk_queue_bypass_start(), we aren't performing
539 * synchronize_rcu() after entering bypass mode to avoid the delay
540 * as some drivers create and destroy a lot of queues while
541 * probing. This is still safe because blk_release_queue() will be
542 * called only after the queue refcnt drops to zero and nothing,
543 * RCU or not, would be traversing the queue by then.
546 queue_flag_set(QUEUE_FLAG_BYPASS, q);
548 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
549 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
550 queue_flag_set(QUEUE_FLAG_DYING, q);
551 spin_unlock_irq(lock);
552 mutex_unlock(&q->sysfs_lock);
555 * Drain all requests queued before DYING marking. Set DEAD flag to
556 * prevent that q->request_fn() gets invoked after draining finished.
561 __blk_drain_queue(q, true);
562 queue_flag_set(QUEUE_FLAG_DEAD, q);
563 spin_unlock_irq(lock);
565 /* for synchronous bio-based driver finish in-flight integrity i/o */
566 blk_flush_integrity();
568 /* @q won't process any more request, flush async actions */
569 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
573 blk_mq_free_queue(q);
574 percpu_ref_exit(&q->q_usage_counter);
577 if (q->queue_lock != &q->__queue_lock)
578 q->queue_lock = &q->__queue_lock;
579 spin_unlock_irq(lock);
581 /* @q is and will stay empty, shutdown and put */
584 EXPORT_SYMBOL(blk_cleanup_queue);
586 /* Allocate memory local to the request queue */
587 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
589 struct request_queue *q = data;
591 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
594 static void free_request_simple(void *element, void *data)
596 kmem_cache_free(request_cachep, element);
599 static void *alloc_request_size(gfp_t gfp_mask, void *data)
601 struct request_queue *q = data;
604 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
606 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
613 static void free_request_size(void *element, void *data)
615 struct request_queue *q = data;
618 q->exit_rq_fn(q, element);
622 int blk_init_rl(struct request_list *rl, struct request_queue *q,
625 if (unlikely(rl->rq_pool))
629 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
630 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
631 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
632 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
635 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
636 alloc_request_size, free_request_size,
637 q, gfp_mask, q->node);
639 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
640 alloc_request_simple, free_request_simple,
641 q, gfp_mask, q->node);
649 void blk_exit_rl(struct request_list *rl)
652 mempool_destroy(rl->rq_pool);
655 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
657 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
659 EXPORT_SYMBOL(blk_alloc_queue);
661 int blk_queue_enter(struct request_queue *q, bool nowait)
666 if (percpu_ref_tryget_live(&q->q_usage_counter))
672 ret = wait_event_interruptible(q->mq_freeze_wq,
673 !atomic_read(&q->mq_freeze_depth) ||
675 if (blk_queue_dying(q))
682 void blk_queue_exit(struct request_queue *q)
684 percpu_ref_put(&q->q_usage_counter);
687 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
689 struct request_queue *q =
690 container_of(ref, struct request_queue, q_usage_counter);
692 wake_up_all(&q->mq_freeze_wq);
695 static void blk_rq_timed_out_timer(unsigned long data)
697 struct request_queue *q = (struct request_queue *)data;
699 kblockd_schedule_work(&q->timeout_work);
702 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
704 struct request_queue *q;
706 q = kmem_cache_alloc_node(blk_requestq_cachep,
707 gfp_mask | __GFP_ZERO, node_id);
711 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
715 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
719 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
720 if (!q->backing_dev_info)
723 q->stats = blk_alloc_queue_stats();
727 q->backing_dev_info->ra_pages =
728 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
729 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
730 q->backing_dev_info->name = "block";
733 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
734 laptop_mode_timer_fn, (unsigned long) q);
735 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
736 INIT_LIST_HEAD(&q->queue_head);
737 INIT_LIST_HEAD(&q->timeout_list);
738 INIT_LIST_HEAD(&q->icq_list);
739 #ifdef CONFIG_BLK_CGROUP
740 INIT_LIST_HEAD(&q->blkg_list);
742 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
744 kobject_init(&q->kobj, &blk_queue_ktype);
746 mutex_init(&q->sysfs_lock);
747 spin_lock_init(&q->__queue_lock);
750 * By default initialize queue_lock to internal lock and driver can
751 * override it later if need be.
753 q->queue_lock = &q->__queue_lock;
756 * A queue starts its life with bypass turned on to avoid
757 * unnecessary bypass on/off overhead and nasty surprises during
758 * init. The initial bypass will be finished when the queue is
759 * registered by blk_register_queue().
762 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
764 init_waitqueue_head(&q->mq_freeze_wq);
767 * Init percpu_ref in atomic mode so that it's faster to shutdown.
768 * See blk_register_queue() for details.
770 if (percpu_ref_init(&q->q_usage_counter,
771 blk_queue_usage_counter_release,
772 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
775 if (blkcg_init_queue(q))
781 percpu_ref_exit(&q->q_usage_counter);
783 blk_free_queue_stats(q->stats);
785 bdi_put(q->backing_dev_info);
787 bioset_free(q->bio_split);
789 ida_simple_remove(&blk_queue_ida, q->id);
791 kmem_cache_free(blk_requestq_cachep, q);
794 EXPORT_SYMBOL(blk_alloc_queue_node);
797 * blk_init_queue - prepare a request queue for use with a block device
798 * @rfn: The function to be called to process requests that have been
799 * placed on the queue.
800 * @lock: Request queue spin lock
803 * If a block device wishes to use the standard request handling procedures,
804 * which sorts requests and coalesces adjacent requests, then it must
805 * call blk_init_queue(). The function @rfn will be called when there
806 * are requests on the queue that need to be processed. If the device
807 * supports plugging, then @rfn may not be called immediately when requests
808 * are available on the queue, but may be called at some time later instead.
809 * Plugged queues are generally unplugged when a buffer belonging to one
810 * of the requests on the queue is needed, or due to memory pressure.
812 * @rfn is not required, or even expected, to remove all requests off the
813 * queue, but only as many as it can handle at a time. If it does leave
814 * requests on the queue, it is responsible for arranging that the requests
815 * get dealt with eventually.
817 * The queue spin lock must be held while manipulating the requests on the
818 * request queue; this lock will be taken also from interrupt context, so irq
819 * disabling is needed for it.
821 * Function returns a pointer to the initialized request queue, or %NULL if
825 * blk_init_queue() must be paired with a blk_cleanup_queue() call
826 * when the block device is deactivated (such as at module unload).
829 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
831 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
833 EXPORT_SYMBOL(blk_init_queue);
835 struct request_queue *
836 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
838 struct request_queue *q;
840 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
846 q->queue_lock = lock;
847 if (blk_init_allocated_queue(q) < 0) {
848 blk_cleanup_queue(q);
854 EXPORT_SYMBOL(blk_init_queue_node);
856 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
859 int blk_init_allocated_queue(struct request_queue *q)
861 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
865 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
866 goto out_free_flush_queue;
868 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
869 goto out_exit_flush_rq;
871 INIT_WORK(&q->timeout_work, blk_timeout_work);
872 q->queue_flags |= QUEUE_FLAG_DEFAULT;
875 * This also sets hw/phys segments, boundary and size
877 blk_queue_make_request(q, blk_queue_bio);
879 q->sg_reserved_size = INT_MAX;
881 /* Protect q->elevator from elevator_change */
882 mutex_lock(&q->sysfs_lock);
885 if (elevator_init(q, NULL)) {
886 mutex_unlock(&q->sysfs_lock);
887 goto out_exit_flush_rq;
890 mutex_unlock(&q->sysfs_lock);
895 q->exit_rq_fn(q, q->fq->flush_rq);
896 out_free_flush_queue:
897 blk_free_flush_queue(q->fq);
900 EXPORT_SYMBOL(blk_init_allocated_queue);
902 bool blk_get_queue(struct request_queue *q)
904 if (likely(!blk_queue_dying(q))) {
911 EXPORT_SYMBOL(blk_get_queue);
913 static inline void blk_free_request(struct request_list *rl, struct request *rq)
915 if (rq->rq_flags & RQF_ELVPRIV) {
916 elv_put_request(rl->q, rq);
918 put_io_context(rq->elv.icq->ioc);
921 mempool_free(rq, rl->rq_pool);
925 * ioc_batching returns true if the ioc is a valid batching request and
926 * should be given priority access to a request.
928 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
934 * Make sure the process is able to allocate at least 1 request
935 * even if the batch times out, otherwise we could theoretically
938 return ioc->nr_batch_requests == q->nr_batching ||
939 (ioc->nr_batch_requests > 0
940 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
944 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
945 * will cause the process to be a "batcher" on all queues in the system. This
946 * is the behaviour we want though - once it gets a wakeup it should be given
949 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
951 if (!ioc || ioc_batching(q, ioc))
954 ioc->nr_batch_requests = q->nr_batching;
955 ioc->last_waited = jiffies;
958 static void __freed_request(struct request_list *rl, int sync)
960 struct request_queue *q = rl->q;
962 if (rl->count[sync] < queue_congestion_off_threshold(q))
963 blk_clear_congested(rl, sync);
965 if (rl->count[sync] + 1 <= q->nr_requests) {
966 if (waitqueue_active(&rl->wait[sync]))
967 wake_up(&rl->wait[sync]);
969 blk_clear_rl_full(rl, sync);
974 * A request has just been released. Account for it, update the full and
975 * congestion status, wake up any waiters. Called under q->queue_lock.
977 static void freed_request(struct request_list *rl, bool sync,
978 req_flags_t rq_flags)
980 struct request_queue *q = rl->q;
984 if (rq_flags & RQF_ELVPRIV)
987 __freed_request(rl, sync);
989 if (unlikely(rl->starved[sync ^ 1]))
990 __freed_request(rl, sync ^ 1);
993 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
995 struct request_list *rl;
996 int on_thresh, off_thresh;
998 spin_lock_irq(q->queue_lock);
1000 blk_queue_congestion_threshold(q);
1001 on_thresh = queue_congestion_on_threshold(q);
1002 off_thresh = queue_congestion_off_threshold(q);
1004 blk_queue_for_each_rl(rl, q) {
1005 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1006 blk_set_congested(rl, BLK_RW_SYNC);
1007 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1008 blk_clear_congested(rl, BLK_RW_SYNC);
1010 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1011 blk_set_congested(rl, BLK_RW_ASYNC);
1012 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1013 blk_clear_congested(rl, BLK_RW_ASYNC);
1015 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1016 blk_set_rl_full(rl, BLK_RW_SYNC);
1018 blk_clear_rl_full(rl, BLK_RW_SYNC);
1019 wake_up(&rl->wait[BLK_RW_SYNC]);
1022 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1023 blk_set_rl_full(rl, BLK_RW_ASYNC);
1025 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1026 wake_up(&rl->wait[BLK_RW_ASYNC]);
1030 spin_unlock_irq(q->queue_lock);
1035 * __get_request - get a free request
1036 * @rl: request list to allocate from
1037 * @op: operation and flags
1038 * @bio: bio to allocate request for (can be %NULL)
1039 * @gfp_mask: allocation mask
1041 * Get a free request from @q. This function may fail under memory
1042 * pressure or if @q is dead.
1044 * Must be called with @q->queue_lock held and,
1045 * Returns ERR_PTR on failure, with @q->queue_lock held.
1046 * Returns request pointer on success, with @q->queue_lock *not held*.
1048 static struct request *__get_request(struct request_list *rl, unsigned int op,
1049 struct bio *bio, gfp_t gfp_mask)
1051 struct request_queue *q = rl->q;
1053 struct elevator_type *et = q->elevator->type;
1054 struct io_context *ioc = rq_ioc(bio);
1055 struct io_cq *icq = NULL;
1056 const bool is_sync = op_is_sync(op);
1058 req_flags_t rq_flags = RQF_ALLOCED;
1060 if (unlikely(blk_queue_dying(q)))
1061 return ERR_PTR(-ENODEV);
1063 may_queue = elv_may_queue(q, op);
1064 if (may_queue == ELV_MQUEUE_NO)
1067 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1068 if (rl->count[is_sync]+1 >= q->nr_requests) {
1070 * The queue will fill after this allocation, so set
1071 * it as full, and mark this process as "batching".
1072 * This process will be allowed to complete a batch of
1073 * requests, others will be blocked.
1075 if (!blk_rl_full(rl, is_sync)) {
1076 ioc_set_batching(q, ioc);
1077 blk_set_rl_full(rl, is_sync);
1079 if (may_queue != ELV_MQUEUE_MUST
1080 && !ioc_batching(q, ioc)) {
1082 * The queue is full and the allocating
1083 * process is not a "batcher", and not
1084 * exempted by the IO scheduler
1086 return ERR_PTR(-ENOMEM);
1090 blk_set_congested(rl, is_sync);
1094 * Only allow batching queuers to allocate up to 50% over the defined
1095 * limit of requests, otherwise we could have thousands of requests
1096 * allocated with any setting of ->nr_requests
1098 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1099 return ERR_PTR(-ENOMEM);
1101 q->nr_rqs[is_sync]++;
1102 rl->count[is_sync]++;
1103 rl->starved[is_sync] = 0;
1106 * Decide whether the new request will be managed by elevator. If
1107 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1108 * prevent the current elevator from being destroyed until the new
1109 * request is freed. This guarantees icq's won't be destroyed and
1110 * makes creating new ones safe.
1112 * Flush requests do not use the elevator so skip initialization.
1113 * This allows a request to share the flush and elevator data.
1115 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1116 * it will be created after releasing queue_lock.
1118 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1119 rq_flags |= RQF_ELVPRIV;
1120 q->nr_rqs_elvpriv++;
1121 if (et->icq_cache && ioc)
1122 icq = ioc_lookup_icq(ioc, q);
1125 if (blk_queue_io_stat(q))
1126 rq_flags |= RQF_IO_STAT;
1127 spin_unlock_irq(q->queue_lock);
1129 /* allocate and init request */
1130 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1135 blk_rq_set_rl(rq, rl);
1136 blk_rq_set_prio(rq, ioc);
1138 rq->rq_flags = rq_flags;
1141 if (rq_flags & RQF_ELVPRIV) {
1142 if (unlikely(et->icq_cache && !icq)) {
1144 icq = ioc_create_icq(ioc, q, gfp_mask);
1150 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1153 /* @rq->elv.icq holds io_context until @rq is freed */
1155 get_io_context(icq->ioc);
1159 * ioc may be NULL here, and ioc_batching will be false. That's
1160 * OK, if the queue is under the request limit then requests need
1161 * not count toward the nr_batch_requests limit. There will always
1162 * be some limit enforced by BLK_BATCH_TIME.
1164 if (ioc_batching(q, ioc))
1165 ioc->nr_batch_requests--;
1167 trace_block_getrq(q, bio, op);
1172 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1173 * and may fail indefinitely under memory pressure and thus
1174 * shouldn't stall IO. Treat this request as !elvpriv. This will
1175 * disturb iosched and blkcg but weird is bettern than dead.
1177 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1178 __func__, dev_name(q->backing_dev_info->dev));
1180 rq->rq_flags &= ~RQF_ELVPRIV;
1183 spin_lock_irq(q->queue_lock);
1184 q->nr_rqs_elvpriv--;
1185 spin_unlock_irq(q->queue_lock);
1190 * Allocation failed presumably due to memory. Undo anything we
1191 * might have messed up.
1193 * Allocating task should really be put onto the front of the wait
1194 * queue, but this is pretty rare.
1196 spin_lock_irq(q->queue_lock);
1197 freed_request(rl, is_sync, rq_flags);
1200 * in the very unlikely event that allocation failed and no
1201 * requests for this direction was pending, mark us starved so that
1202 * freeing of a request in the other direction will notice
1203 * us. another possible fix would be to split the rq mempool into
1207 if (unlikely(rl->count[is_sync] == 0))
1208 rl->starved[is_sync] = 1;
1209 return ERR_PTR(-ENOMEM);
1213 * get_request - get a free request
1214 * @q: request_queue to allocate request from
1215 * @op: operation and flags
1216 * @bio: bio to allocate request for (can be %NULL)
1217 * @gfp_mask: allocation mask
1219 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1220 * this function keeps retrying under memory pressure and fails iff @q is dead.
1222 * Must be called with @q->queue_lock held and,
1223 * Returns ERR_PTR on failure, with @q->queue_lock held.
1224 * Returns request pointer on success, with @q->queue_lock *not held*.
1226 static struct request *get_request(struct request_queue *q, unsigned int op,
1227 struct bio *bio, gfp_t gfp_mask)
1229 const bool is_sync = op_is_sync(op);
1231 struct request_list *rl;
1234 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1236 rq = __get_request(rl, op, bio, gfp_mask);
1240 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1245 /* wait on @rl and retry */
1246 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1247 TASK_UNINTERRUPTIBLE);
1249 trace_block_sleeprq(q, bio, op);
1251 spin_unlock_irq(q->queue_lock);
1255 * After sleeping, we become a "batching" process and will be able
1256 * to allocate at least one request, and up to a big batch of them
1257 * for a small period time. See ioc_batching, ioc_set_batching
1259 ioc_set_batching(q, current->io_context);
1261 spin_lock_irq(q->queue_lock);
1262 finish_wait(&rl->wait[is_sync], &wait);
1267 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1272 /* create ioc upfront */
1273 create_io_context(gfp_mask, q->node);
1275 spin_lock_irq(q->queue_lock);
1276 rq = get_request(q, rw, NULL, gfp_mask);
1278 spin_unlock_irq(q->queue_lock);
1282 /* q->queue_lock is unlocked at this point */
1284 rq->__sector = (sector_t) -1;
1285 rq->bio = rq->biotail = NULL;
1289 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1292 return blk_mq_alloc_request(q, rw,
1293 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1294 0 : BLK_MQ_REQ_NOWAIT);
1296 return blk_old_get_request(q, rw, gfp_mask);
1298 EXPORT_SYMBOL(blk_get_request);
1301 * blk_requeue_request - put a request back on queue
1302 * @q: request queue where request should be inserted
1303 * @rq: request to be inserted
1306 * Drivers often keep queueing requests until the hardware cannot accept
1307 * more, when that condition happens we need to put the request back
1308 * on the queue. Must be called with queue lock held.
1310 void blk_requeue_request(struct request_queue *q, struct request *rq)
1312 blk_delete_timer(rq);
1313 blk_clear_rq_complete(rq);
1314 trace_block_rq_requeue(q, rq);
1315 wbt_requeue(q->rq_wb, &rq->issue_stat);
1317 if (rq->rq_flags & RQF_QUEUED)
1318 blk_queue_end_tag(q, rq);
1320 BUG_ON(blk_queued_rq(rq));
1322 elv_requeue_request(q, rq);
1324 EXPORT_SYMBOL(blk_requeue_request);
1326 static void add_acct_request(struct request_queue *q, struct request *rq,
1329 blk_account_io_start(rq, true);
1330 __elv_add_request(q, rq, where);
1333 static void part_round_stats_single(int cpu, struct hd_struct *part,
1338 if (now == part->stamp)
1341 inflight = part_in_flight(part);
1343 __part_stat_add(cpu, part, time_in_queue,
1344 inflight * (now - part->stamp));
1345 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1351 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1352 * @cpu: cpu number for stats access
1353 * @part: target partition
1355 * The average IO queue length and utilisation statistics are maintained
1356 * by observing the current state of the queue length and the amount of
1357 * time it has been in this state for.
1359 * Normally, that accounting is done on IO completion, but that can result
1360 * in more than a second's worth of IO being accounted for within any one
1361 * second, leading to >100% utilisation. To deal with that, we call this
1362 * function to do a round-off before returning the results when reading
1363 * /proc/diskstats. This accounts immediately for all queue usage up to
1364 * the current jiffies and restarts the counters again.
1366 void part_round_stats(int cpu, struct hd_struct *part)
1368 unsigned long now = jiffies;
1371 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1372 part_round_stats_single(cpu, part, now);
1374 EXPORT_SYMBOL_GPL(part_round_stats);
1377 static void blk_pm_put_request(struct request *rq)
1379 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1380 pm_runtime_mark_last_busy(rq->q->dev);
1383 static inline void blk_pm_put_request(struct request *rq) {}
1387 * queue lock must be held
1389 void __blk_put_request(struct request_queue *q, struct request *req)
1391 req_flags_t rq_flags = req->rq_flags;
1397 blk_mq_free_request(req);
1401 blk_pm_put_request(req);
1403 elv_completed_request(q, req);
1405 /* this is a bio leak */
1406 WARN_ON(req->bio != NULL);
1408 wbt_done(q->rq_wb, &req->issue_stat);
1411 * Request may not have originated from ll_rw_blk. if not,
1412 * it didn't come out of our reserved rq pools
1414 if (rq_flags & RQF_ALLOCED) {
1415 struct request_list *rl = blk_rq_rl(req);
1416 bool sync = op_is_sync(req->cmd_flags);
1418 BUG_ON(!list_empty(&req->queuelist));
1419 BUG_ON(ELV_ON_HASH(req));
1421 blk_free_request(rl, req);
1422 freed_request(rl, sync, rq_flags);
1426 EXPORT_SYMBOL_GPL(__blk_put_request);
1428 void blk_put_request(struct request *req)
1430 struct request_queue *q = req->q;
1433 blk_mq_free_request(req);
1435 unsigned long flags;
1437 spin_lock_irqsave(q->queue_lock, flags);
1438 __blk_put_request(q, req);
1439 spin_unlock_irqrestore(q->queue_lock, flags);
1442 EXPORT_SYMBOL(blk_put_request);
1444 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1447 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1449 if (!ll_back_merge_fn(q, req, bio))
1452 trace_block_bio_backmerge(q, req, bio);
1454 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1455 blk_rq_set_mixed_merge(req);
1457 req->biotail->bi_next = bio;
1459 req->__data_len += bio->bi_iter.bi_size;
1460 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1462 blk_account_io_start(req, false);
1466 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1469 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1471 if (!ll_front_merge_fn(q, req, bio))
1474 trace_block_bio_frontmerge(q, req, bio);
1476 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1477 blk_rq_set_mixed_merge(req);
1479 bio->bi_next = req->bio;
1482 req->__sector = bio->bi_iter.bi_sector;
1483 req->__data_len += bio->bi_iter.bi_size;
1484 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1486 blk_account_io_start(req, false);
1490 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1493 unsigned short segments = blk_rq_nr_discard_segments(req);
1495 if (segments >= queue_max_discard_segments(q))
1497 if (blk_rq_sectors(req) + bio_sectors(bio) >
1498 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1501 req->biotail->bi_next = bio;
1503 req->__data_len += bio->bi_iter.bi_size;
1504 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1505 req->nr_phys_segments = segments + 1;
1507 blk_account_io_start(req, false);
1510 req_set_nomerge(q, req);
1515 * blk_attempt_plug_merge - try to merge with %current's plugged list
1516 * @q: request_queue new bio is being queued at
1517 * @bio: new bio being queued
1518 * @request_count: out parameter for number of traversed plugged requests
1519 * @same_queue_rq: pointer to &struct request that gets filled in when
1520 * another request associated with @q is found on the plug list
1521 * (optional, may be %NULL)
1523 * Determine whether @bio being queued on @q can be merged with a request
1524 * on %current's plugged list. Returns %true if merge was successful,
1527 * Plugging coalesces IOs from the same issuer for the same purpose without
1528 * going through @q->queue_lock. As such it's more of an issuing mechanism
1529 * than scheduling, and the request, while may have elvpriv data, is not
1530 * added on the elevator at this point. In addition, we don't have
1531 * reliable access to the elevator outside queue lock. Only check basic
1532 * merging parameters without querying the elevator.
1534 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1536 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1537 unsigned int *request_count,
1538 struct request **same_queue_rq)
1540 struct blk_plug *plug;
1542 struct list_head *plug_list;
1544 plug = current->plug;
1550 plug_list = &plug->mq_list;
1552 plug_list = &plug->list;
1554 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1555 bool merged = false;
1560 * Only blk-mq multiple hardware queues case checks the
1561 * rq in the same queue, there should be only one such
1565 *same_queue_rq = rq;
1568 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1571 switch (blk_try_merge(rq, bio)) {
1572 case ELEVATOR_BACK_MERGE:
1573 merged = bio_attempt_back_merge(q, rq, bio);
1575 case ELEVATOR_FRONT_MERGE:
1576 merged = bio_attempt_front_merge(q, rq, bio);
1578 case ELEVATOR_DISCARD_MERGE:
1579 merged = bio_attempt_discard_merge(q, rq, bio);
1592 unsigned int blk_plug_queued_count(struct request_queue *q)
1594 struct blk_plug *plug;
1596 struct list_head *plug_list;
1597 unsigned int ret = 0;
1599 plug = current->plug;
1604 plug_list = &plug->mq_list;
1606 plug_list = &plug->list;
1608 list_for_each_entry(rq, plug_list, queuelist) {
1616 void init_request_from_bio(struct request *req, struct bio *bio)
1618 if (bio->bi_opf & REQ_RAHEAD)
1619 req->cmd_flags |= REQ_FAILFAST_MASK;
1622 req->__sector = bio->bi_iter.bi_sector;
1623 if (ioprio_valid(bio_prio(bio)))
1624 req->ioprio = bio_prio(bio);
1625 blk_rq_bio_prep(req->q, req, bio);
1628 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1630 struct blk_plug *plug;
1631 int where = ELEVATOR_INSERT_SORT;
1632 struct request *req, *free;
1633 unsigned int request_count = 0;
1634 unsigned int wb_acct;
1637 * low level driver can indicate that it wants pages above a
1638 * certain limit bounced to low memory (ie for highmem, or even
1639 * ISA dma in theory)
1641 blk_queue_bounce(q, &bio);
1643 blk_queue_split(q, &bio, q->bio_split);
1645 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1646 bio->bi_error = -EIO;
1648 return BLK_QC_T_NONE;
1651 if (op_is_flush(bio->bi_opf)) {
1652 spin_lock_irq(q->queue_lock);
1653 where = ELEVATOR_INSERT_FLUSH;
1658 * Check if we can merge with the plugged list before grabbing
1661 if (!blk_queue_nomerges(q)) {
1662 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1663 return BLK_QC_T_NONE;
1665 request_count = blk_plug_queued_count(q);
1667 spin_lock_irq(q->queue_lock);
1669 switch (elv_merge(q, &req, bio)) {
1670 case ELEVATOR_BACK_MERGE:
1671 if (!bio_attempt_back_merge(q, req, bio))
1673 elv_bio_merged(q, req, bio);
1674 free = attempt_back_merge(q, req);
1676 __blk_put_request(q, free);
1678 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1680 case ELEVATOR_FRONT_MERGE:
1681 if (!bio_attempt_front_merge(q, req, bio))
1683 elv_bio_merged(q, req, bio);
1684 free = attempt_front_merge(q, req);
1686 __blk_put_request(q, free);
1688 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1695 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1698 * Grab a free request. This is might sleep but can not fail.
1699 * Returns with the queue unlocked.
1701 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1703 __wbt_done(q->rq_wb, wb_acct);
1704 bio->bi_error = PTR_ERR(req);
1709 wbt_track(&req->issue_stat, wb_acct);
1712 * After dropping the lock and possibly sleeping here, our request
1713 * may now be mergeable after it had proven unmergeable (above).
1714 * We don't worry about that case for efficiency. It won't happen
1715 * often, and the elevators are able to handle it.
1717 init_request_from_bio(req, bio);
1719 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1720 req->cpu = raw_smp_processor_id();
1722 plug = current->plug;
1725 * If this is the first request added after a plug, fire
1728 * @request_count may become stale because of schedule
1729 * out, so check plug list again.
1731 if (!request_count || list_empty(&plug->list))
1732 trace_block_plug(q);
1734 struct request *last = list_entry_rq(plug->list.prev);
1735 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1736 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1737 blk_flush_plug_list(plug, false);
1738 trace_block_plug(q);
1741 list_add_tail(&req->queuelist, &plug->list);
1742 blk_account_io_start(req, true);
1744 spin_lock_irq(q->queue_lock);
1745 add_acct_request(q, req, where);
1748 spin_unlock_irq(q->queue_lock);
1751 return BLK_QC_T_NONE;
1755 * If bio->bi_dev is a partition, remap the location
1757 static inline void blk_partition_remap(struct bio *bio)
1759 struct block_device *bdev = bio->bi_bdev;
1762 * Zone reset does not include bi_size so bio_sectors() is always 0.
1763 * Include a test for the reset op code and perform the remap if needed.
1765 if (bdev != bdev->bd_contains &&
1766 (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1767 struct hd_struct *p = bdev->bd_part;
1769 bio->bi_iter.bi_sector += p->start_sect;
1770 bio->bi_bdev = bdev->bd_contains;
1772 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1774 bio->bi_iter.bi_sector - p->start_sect);
1778 static void handle_bad_sector(struct bio *bio)
1780 char b[BDEVNAME_SIZE];
1782 printk(KERN_INFO "attempt to access beyond end of device\n");
1783 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1784 bdevname(bio->bi_bdev, b),
1786 (unsigned long long)bio_end_sector(bio),
1787 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1790 #ifdef CONFIG_FAIL_MAKE_REQUEST
1792 static DECLARE_FAULT_ATTR(fail_make_request);
1794 static int __init setup_fail_make_request(char *str)
1796 return setup_fault_attr(&fail_make_request, str);
1798 __setup("fail_make_request=", setup_fail_make_request);
1800 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1802 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1805 static int __init fail_make_request_debugfs(void)
1807 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1808 NULL, &fail_make_request);
1810 return PTR_ERR_OR_ZERO(dir);
1813 late_initcall(fail_make_request_debugfs);
1815 #else /* CONFIG_FAIL_MAKE_REQUEST */
1817 static inline bool should_fail_request(struct hd_struct *part,
1823 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1826 * Check whether this bio extends beyond the end of the device.
1828 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1835 /* Test device or partition size, when known. */
1836 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1838 sector_t sector = bio->bi_iter.bi_sector;
1840 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1842 * This may well happen - the kernel calls bread()
1843 * without checking the size of the device, e.g., when
1844 * mounting a device.
1846 handle_bad_sector(bio);
1854 static noinline_for_stack bool
1855 generic_make_request_checks(struct bio *bio)
1857 struct request_queue *q;
1858 int nr_sectors = bio_sectors(bio);
1860 char b[BDEVNAME_SIZE];
1861 struct hd_struct *part;
1865 if (bio_check_eod(bio, nr_sectors))
1868 q = bdev_get_queue(bio->bi_bdev);
1871 "generic_make_request: Trying to access "
1872 "nonexistent block-device %s (%Lu)\n",
1873 bdevname(bio->bi_bdev, b),
1874 (long long) bio->bi_iter.bi_sector);
1878 part = bio->bi_bdev->bd_part;
1879 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1880 should_fail_request(&part_to_disk(part)->part0,
1881 bio->bi_iter.bi_size))
1885 * If this device has partitions, remap block n
1886 * of partition p to block n+start(p) of the disk.
1888 blk_partition_remap(bio);
1890 if (bio_check_eod(bio, nr_sectors))
1894 * Filter flush bio's early so that make_request based
1895 * drivers without flush support don't have to worry
1898 if (op_is_flush(bio->bi_opf) &&
1899 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1900 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1907 switch (bio_op(bio)) {
1908 case REQ_OP_DISCARD:
1909 if (!blk_queue_discard(q))
1912 case REQ_OP_SECURE_ERASE:
1913 if (!blk_queue_secure_erase(q))
1916 case REQ_OP_WRITE_SAME:
1917 if (!bdev_write_same(bio->bi_bdev))
1920 case REQ_OP_ZONE_REPORT:
1921 case REQ_OP_ZONE_RESET:
1922 if (!bdev_is_zoned(bio->bi_bdev))
1925 case REQ_OP_WRITE_ZEROES:
1926 if (!bdev_write_zeroes_sectors(bio->bi_bdev))
1934 * Various block parts want %current->io_context and lazy ioc
1935 * allocation ends up trading a lot of pain for a small amount of
1936 * memory. Just allocate it upfront. This may fail and block
1937 * layer knows how to live with it.
1939 create_io_context(GFP_ATOMIC, q->node);
1941 if (!blkcg_bio_issue_check(q, bio))
1944 trace_block_bio_queue(q, bio);
1950 bio->bi_error = err;
1956 * generic_make_request - hand a buffer to its device driver for I/O
1957 * @bio: The bio describing the location in memory and on the device.
1959 * generic_make_request() is used to make I/O requests of block
1960 * devices. It is passed a &struct bio, which describes the I/O that needs
1963 * generic_make_request() does not return any status. The
1964 * success/failure status of the request, along with notification of
1965 * completion, is delivered asynchronously through the bio->bi_end_io
1966 * function described (one day) else where.
1968 * The caller of generic_make_request must make sure that bi_io_vec
1969 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1970 * set to describe the device address, and the
1971 * bi_end_io and optionally bi_private are set to describe how
1972 * completion notification should be signaled.
1974 * generic_make_request and the drivers it calls may use bi_next if this
1975 * bio happens to be merged with someone else, and may resubmit the bio to
1976 * a lower device by calling into generic_make_request recursively, which
1977 * means the bio should NOT be touched after the call to ->make_request_fn.
1979 blk_qc_t generic_make_request(struct bio *bio)
1982 * bio_list_on_stack[0] contains bios submitted by the current
1984 * bio_list_on_stack[1] contains bios that were submitted before
1985 * the current make_request_fn, but that haven't been processed
1988 struct bio_list bio_list_on_stack[2];
1989 blk_qc_t ret = BLK_QC_T_NONE;
1991 if (!generic_make_request_checks(bio))
1995 * We only want one ->make_request_fn to be active at a time, else
1996 * stack usage with stacked devices could be a problem. So use
1997 * current->bio_list to keep a list of requests submited by a
1998 * make_request_fn function. current->bio_list is also used as a
1999 * flag to say if generic_make_request is currently active in this
2000 * task or not. If it is NULL, then no make_request is active. If
2001 * it is non-NULL, then a make_request is active, and new requests
2002 * should be added at the tail
2004 if (current->bio_list) {
2005 bio_list_add(¤t->bio_list[0], bio);
2009 /* following loop may be a bit non-obvious, and so deserves some
2011 * Before entering the loop, bio->bi_next is NULL (as all callers
2012 * ensure that) so we have a list with a single bio.
2013 * We pretend that we have just taken it off a longer list, so
2014 * we assign bio_list to a pointer to the bio_list_on_stack,
2015 * thus initialising the bio_list of new bios to be
2016 * added. ->make_request() may indeed add some more bios
2017 * through a recursive call to generic_make_request. If it
2018 * did, we find a non-NULL value in bio_list and re-enter the loop
2019 * from the top. In this case we really did just take the bio
2020 * of the top of the list (no pretending) and so remove it from
2021 * bio_list, and call into ->make_request() again.
2023 BUG_ON(bio->bi_next);
2024 bio_list_init(&bio_list_on_stack[0]);
2025 current->bio_list = bio_list_on_stack;
2027 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2029 if (likely(blk_queue_enter(q, false) == 0)) {
2030 struct bio_list lower, same;
2032 /* Create a fresh bio_list for all subordinate requests */
2033 bio_list_on_stack[1] = bio_list_on_stack[0];
2034 bio_list_init(&bio_list_on_stack[0]);
2035 ret = q->make_request_fn(q, bio);
2039 /* sort new bios into those for a lower level
2040 * and those for the same level
2042 bio_list_init(&lower);
2043 bio_list_init(&same);
2044 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2045 if (q == bdev_get_queue(bio->bi_bdev))
2046 bio_list_add(&same, bio);
2048 bio_list_add(&lower, bio);
2049 /* now assemble so we handle the lowest level first */
2050 bio_list_merge(&bio_list_on_stack[0], &lower);
2051 bio_list_merge(&bio_list_on_stack[0], &same);
2052 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2056 bio = bio_list_pop(&bio_list_on_stack[0]);
2058 current->bio_list = NULL; /* deactivate */
2063 EXPORT_SYMBOL(generic_make_request);
2066 * submit_bio - submit a bio to the block device layer for I/O
2067 * @bio: The &struct bio which describes the I/O
2069 * submit_bio() is very similar in purpose to generic_make_request(), and
2070 * uses that function to do most of the work. Both are fairly rough
2071 * interfaces; @bio must be presetup and ready for I/O.
2074 blk_qc_t submit_bio(struct bio *bio)
2077 * If it's a regular read/write or a barrier with data attached,
2078 * go through the normal accounting stuff before submission.
2080 if (bio_has_data(bio)) {
2083 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2084 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2086 count = bio_sectors(bio);
2088 if (op_is_write(bio_op(bio))) {
2089 count_vm_events(PGPGOUT, count);
2091 task_io_account_read(bio->bi_iter.bi_size);
2092 count_vm_events(PGPGIN, count);
2095 if (unlikely(block_dump)) {
2096 char b[BDEVNAME_SIZE];
2097 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2098 current->comm, task_pid_nr(current),
2099 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2100 (unsigned long long)bio->bi_iter.bi_sector,
2101 bdevname(bio->bi_bdev, b),
2106 return generic_make_request(bio);
2108 EXPORT_SYMBOL(submit_bio);
2111 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2112 * for new the queue limits
2114 * @rq: the request being checked
2117 * @rq may have been made based on weaker limitations of upper-level queues
2118 * in request stacking drivers, and it may violate the limitation of @q.
2119 * Since the block layer and the underlying device driver trust @rq
2120 * after it is inserted to @q, it should be checked against @q before
2121 * the insertion using this generic function.
2123 * Request stacking drivers like request-based dm may change the queue
2124 * limits when retrying requests on other queues. Those requests need
2125 * to be checked against the new queue limits again during dispatch.
2127 static int blk_cloned_rq_check_limits(struct request_queue *q,
2130 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2131 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2136 * queue's settings related to segment counting like q->bounce_pfn
2137 * may differ from that of other stacking queues.
2138 * Recalculate it to check the request correctly on this queue's
2141 blk_recalc_rq_segments(rq);
2142 if (rq->nr_phys_segments > queue_max_segments(q)) {
2143 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2151 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2152 * @q: the queue to submit the request
2153 * @rq: the request being queued
2155 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2157 unsigned long flags;
2158 int where = ELEVATOR_INSERT_BACK;
2160 if (blk_cloned_rq_check_limits(q, rq))
2164 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2168 if (blk_queue_io_stat(q))
2169 blk_account_io_start(rq, true);
2170 blk_mq_sched_insert_request(rq, false, true, false, false);
2174 spin_lock_irqsave(q->queue_lock, flags);
2175 if (unlikely(blk_queue_dying(q))) {
2176 spin_unlock_irqrestore(q->queue_lock, flags);
2181 * Submitting request must be dequeued before calling this function
2182 * because it will be linked to another request_queue
2184 BUG_ON(blk_queued_rq(rq));
2186 if (op_is_flush(rq->cmd_flags))
2187 where = ELEVATOR_INSERT_FLUSH;
2189 add_acct_request(q, rq, where);
2190 if (where == ELEVATOR_INSERT_FLUSH)
2192 spin_unlock_irqrestore(q->queue_lock, flags);
2196 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2199 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2200 * @rq: request to examine
2203 * A request could be merge of IOs which require different failure
2204 * handling. This function determines the number of bytes which
2205 * can be failed from the beginning of the request without
2206 * crossing into area which need to be retried further.
2209 * The number of bytes to fail.
2212 * queue_lock must be held.
2214 unsigned int blk_rq_err_bytes(const struct request *rq)
2216 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2217 unsigned int bytes = 0;
2220 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2221 return blk_rq_bytes(rq);
2224 * Currently the only 'mixing' which can happen is between
2225 * different fastfail types. We can safely fail portions
2226 * which have all the failfast bits that the first one has -
2227 * the ones which are at least as eager to fail as the first
2230 for (bio = rq->bio; bio; bio = bio->bi_next) {
2231 if ((bio->bi_opf & ff) != ff)
2233 bytes += bio->bi_iter.bi_size;
2236 /* this could lead to infinite loop */
2237 BUG_ON(blk_rq_bytes(rq) && !bytes);
2240 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2242 void blk_account_io_completion(struct request *req, unsigned int bytes)
2244 if (blk_do_io_stat(req)) {
2245 const int rw = rq_data_dir(req);
2246 struct hd_struct *part;
2249 cpu = part_stat_lock();
2251 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2256 void blk_account_io_done(struct request *req)
2259 * Account IO completion. flush_rq isn't accounted as a
2260 * normal IO on queueing nor completion. Accounting the
2261 * containing request is enough.
2263 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2264 unsigned long duration = jiffies - req->start_time;
2265 const int rw = rq_data_dir(req);
2266 struct hd_struct *part;
2269 cpu = part_stat_lock();
2272 part_stat_inc(cpu, part, ios[rw]);
2273 part_stat_add(cpu, part, ticks[rw], duration);
2274 part_round_stats(cpu, part);
2275 part_dec_in_flight(part, rw);
2277 hd_struct_put(part);
2284 * Don't process normal requests when queue is suspended
2285 * or in the process of suspending/resuming
2287 static struct request *blk_pm_peek_request(struct request_queue *q,
2290 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2291 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2297 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2304 void blk_account_io_start(struct request *rq, bool new_io)
2306 struct hd_struct *part;
2307 int rw = rq_data_dir(rq);
2310 if (!blk_do_io_stat(rq))
2313 cpu = part_stat_lock();
2317 part_stat_inc(cpu, part, merges[rw]);
2319 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2320 if (!hd_struct_try_get(part)) {
2322 * The partition is already being removed,
2323 * the request will be accounted on the disk only
2325 * We take a reference on disk->part0 although that
2326 * partition will never be deleted, so we can treat
2327 * it as any other partition.
2329 part = &rq->rq_disk->part0;
2330 hd_struct_get(part);
2332 part_round_stats(cpu, part);
2333 part_inc_in_flight(part, rw);
2341 * blk_peek_request - peek at the top of a request queue
2342 * @q: request queue to peek at
2345 * Return the request at the top of @q. The returned request
2346 * should be started using blk_start_request() before LLD starts
2350 * Pointer to the request at the top of @q if available. Null
2354 * queue_lock must be held.
2356 struct request *blk_peek_request(struct request_queue *q)
2361 while ((rq = __elv_next_request(q)) != NULL) {
2363 rq = blk_pm_peek_request(q, rq);
2367 if (!(rq->rq_flags & RQF_STARTED)) {
2369 * This is the first time the device driver
2370 * sees this request (possibly after
2371 * requeueing). Notify IO scheduler.
2373 if (rq->rq_flags & RQF_SORTED)
2374 elv_activate_rq(q, rq);
2377 * just mark as started even if we don't start
2378 * it, a request that has been delayed should
2379 * not be passed by new incoming requests
2381 rq->rq_flags |= RQF_STARTED;
2382 trace_block_rq_issue(q, rq);
2385 if (!q->boundary_rq || q->boundary_rq == rq) {
2386 q->end_sector = rq_end_sector(rq);
2387 q->boundary_rq = NULL;
2390 if (rq->rq_flags & RQF_DONTPREP)
2393 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2395 * make sure space for the drain appears we
2396 * know we can do this because max_hw_segments
2397 * has been adjusted to be one fewer than the
2400 rq->nr_phys_segments++;
2406 ret = q->prep_rq_fn(q, rq);
2407 if (ret == BLKPREP_OK) {
2409 } else if (ret == BLKPREP_DEFER) {
2411 * the request may have been (partially) prepped.
2412 * we need to keep this request in the front to
2413 * avoid resource deadlock. RQF_STARTED will
2414 * prevent other fs requests from passing this one.
2416 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2417 !(rq->rq_flags & RQF_DONTPREP)) {
2419 * remove the space for the drain we added
2420 * so that we don't add it again
2422 --rq->nr_phys_segments;
2427 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2428 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2430 rq->rq_flags |= RQF_QUIET;
2432 * Mark this request as started so we don't trigger
2433 * any debug logic in the end I/O path.
2435 blk_start_request(rq);
2436 __blk_end_request_all(rq, err);
2438 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2445 EXPORT_SYMBOL(blk_peek_request);
2447 void blk_dequeue_request(struct request *rq)
2449 struct request_queue *q = rq->q;
2451 BUG_ON(list_empty(&rq->queuelist));
2452 BUG_ON(ELV_ON_HASH(rq));
2454 list_del_init(&rq->queuelist);
2457 * the time frame between a request being removed from the lists
2458 * and to it is freed is accounted as io that is in progress at
2461 if (blk_account_rq(rq)) {
2462 q->in_flight[rq_is_sync(rq)]++;
2463 set_io_start_time_ns(rq);
2468 * blk_start_request - start request processing on the driver
2469 * @req: request to dequeue
2472 * Dequeue @req and start timeout timer on it. This hands off the
2473 * request to the driver.
2475 * Block internal functions which don't want to start timer should
2476 * call blk_dequeue_request().
2479 * queue_lock must be held.
2481 void blk_start_request(struct request *req)
2483 blk_dequeue_request(req);
2485 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2486 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2487 req->rq_flags |= RQF_STATS;
2488 wbt_issue(req->q->rq_wb, &req->issue_stat);
2491 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2494 EXPORT_SYMBOL(blk_start_request);
2497 * blk_fetch_request - fetch a request from a request queue
2498 * @q: request queue to fetch a request from
2501 * Return the request at the top of @q. The request is started on
2502 * return and LLD can start processing it immediately.
2505 * Pointer to the request at the top of @q if available. Null
2509 * queue_lock must be held.
2511 struct request *blk_fetch_request(struct request_queue *q)
2515 rq = blk_peek_request(q);
2517 blk_start_request(rq);
2520 EXPORT_SYMBOL(blk_fetch_request);
2523 * blk_update_request - Special helper function for request stacking drivers
2524 * @req: the request being processed
2525 * @error: %0 for success, < %0 for error
2526 * @nr_bytes: number of bytes to complete @req
2529 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2530 * the request structure even if @req doesn't have leftover.
2531 * If @req has leftover, sets it up for the next range of segments.
2533 * This special helper function is only for request stacking drivers
2534 * (e.g. request-based dm) so that they can handle partial completion.
2535 * Actual device drivers should use blk_end_request instead.
2537 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2538 * %false return from this function.
2541 * %false - this request doesn't have any more data
2542 * %true - this request has more data
2544 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2548 trace_block_rq_complete(req->q, req, nr_bytes);
2554 * For fs requests, rq is just carrier of independent bio's
2555 * and each partial completion should be handled separately.
2556 * Reset per-request error on each partial completion.
2558 * TODO: tj: This is too subtle. It would be better to let
2559 * low level drivers do what they see fit.
2561 if (!blk_rq_is_passthrough(req))
2564 if (error && !blk_rq_is_passthrough(req) &&
2565 !(req->rq_flags & RQF_QUIET)) {
2570 error_type = "recoverable transport";
2573 error_type = "critical target";
2576 error_type = "critical nexus";
2579 error_type = "timeout";
2582 error_type = "critical space allocation";
2585 error_type = "critical medium";
2592 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2593 __func__, error_type, req->rq_disk ?
2594 req->rq_disk->disk_name : "?",
2595 (unsigned long long)blk_rq_pos(req));
2599 blk_account_io_completion(req, nr_bytes);
2603 struct bio *bio = req->bio;
2604 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2606 if (bio_bytes == bio->bi_iter.bi_size)
2607 req->bio = bio->bi_next;
2609 req_bio_endio(req, bio, bio_bytes, error);
2611 total_bytes += bio_bytes;
2612 nr_bytes -= bio_bytes;
2623 * Reset counters so that the request stacking driver
2624 * can find how many bytes remain in the request
2627 req->__data_len = 0;
2631 WARN_ON_ONCE(req->rq_flags & RQF_SPECIAL_PAYLOAD);
2633 req->__data_len -= total_bytes;
2635 /* update sector only for requests with clear definition of sector */
2636 if (!blk_rq_is_passthrough(req))
2637 req->__sector += total_bytes >> 9;
2639 /* mixed attributes always follow the first bio */
2640 if (req->rq_flags & RQF_MIXED_MERGE) {
2641 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2642 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2646 * If total number of sectors is less than the first segment
2647 * size, something has gone terribly wrong.
2649 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2650 blk_dump_rq_flags(req, "request botched");
2651 req->__data_len = blk_rq_cur_bytes(req);
2654 /* recalculate the number of segments */
2655 blk_recalc_rq_segments(req);
2659 EXPORT_SYMBOL_GPL(blk_update_request);
2661 static bool blk_update_bidi_request(struct request *rq, int error,
2662 unsigned int nr_bytes,
2663 unsigned int bidi_bytes)
2665 if (blk_update_request(rq, error, nr_bytes))
2668 /* Bidi request must be completed as a whole */
2669 if (unlikely(blk_bidi_rq(rq)) &&
2670 blk_update_request(rq->next_rq, error, bidi_bytes))
2673 if (blk_queue_add_random(rq->q))
2674 add_disk_randomness(rq->rq_disk);
2680 * blk_unprep_request - unprepare a request
2683 * This function makes a request ready for complete resubmission (or
2684 * completion). It happens only after all error handling is complete,
2685 * so represents the appropriate moment to deallocate any resources
2686 * that were allocated to the request in the prep_rq_fn. The queue
2687 * lock is held when calling this.
2689 void blk_unprep_request(struct request *req)
2691 struct request_queue *q = req->q;
2693 req->rq_flags &= ~RQF_DONTPREP;
2694 if (q->unprep_rq_fn)
2695 q->unprep_rq_fn(q, req);
2697 EXPORT_SYMBOL_GPL(blk_unprep_request);
2700 * queue lock must be held
2702 void blk_finish_request(struct request *req, int error)
2704 struct request_queue *q = req->q;
2706 if (req->rq_flags & RQF_STATS)
2709 if (req->rq_flags & RQF_QUEUED)
2710 blk_queue_end_tag(q, req);
2712 BUG_ON(blk_queued_rq(req));
2714 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
2715 laptop_io_completion(req->q->backing_dev_info);
2717 blk_delete_timer(req);
2719 if (req->rq_flags & RQF_DONTPREP)
2720 blk_unprep_request(req);
2722 blk_account_io_done(req);
2725 wbt_done(req->q->rq_wb, &req->issue_stat);
2726 req->end_io(req, error);
2728 if (blk_bidi_rq(req))
2729 __blk_put_request(req->next_rq->q, req->next_rq);
2731 __blk_put_request(q, req);
2734 EXPORT_SYMBOL(blk_finish_request);
2737 * blk_end_bidi_request - Complete a bidi request
2738 * @rq: the request to complete
2739 * @error: %0 for success, < %0 for error
2740 * @nr_bytes: number of bytes to complete @rq
2741 * @bidi_bytes: number of bytes to complete @rq->next_rq
2744 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2745 * Drivers that supports bidi can safely call this member for any
2746 * type of request, bidi or uni. In the later case @bidi_bytes is
2750 * %false - we are done with this request
2751 * %true - still buffers pending for this request
2753 static bool blk_end_bidi_request(struct request *rq, int error,
2754 unsigned int nr_bytes, unsigned int bidi_bytes)
2756 struct request_queue *q = rq->q;
2757 unsigned long flags;
2759 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2762 spin_lock_irqsave(q->queue_lock, flags);
2763 blk_finish_request(rq, error);
2764 spin_unlock_irqrestore(q->queue_lock, flags);
2770 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2771 * @rq: the request to complete
2772 * @error: %0 for success, < %0 for error
2773 * @nr_bytes: number of bytes to complete @rq
2774 * @bidi_bytes: number of bytes to complete @rq->next_rq
2777 * Identical to blk_end_bidi_request() except that queue lock is
2778 * assumed to be locked on entry and remains so on return.
2781 * %false - we are done with this request
2782 * %true - still buffers pending for this request
2784 bool __blk_end_bidi_request(struct request *rq, int error,
2785 unsigned int nr_bytes, unsigned int bidi_bytes)
2787 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2790 blk_finish_request(rq, error);
2796 * blk_end_request - Helper function for drivers to complete the request.
2797 * @rq: the request being processed
2798 * @error: %0 for success, < %0 for error
2799 * @nr_bytes: number of bytes to complete
2802 * Ends I/O on a number of bytes attached to @rq.
2803 * If @rq has leftover, sets it up for the next range of segments.
2806 * %false - we are done with this request
2807 * %true - still buffers pending for this request
2809 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2811 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2813 EXPORT_SYMBOL(blk_end_request);
2816 * blk_end_request_all - Helper function for drives to finish the request.
2817 * @rq: the request to finish
2818 * @error: %0 for success, < %0 for error
2821 * Completely finish @rq.
2823 void blk_end_request_all(struct request *rq, int error)
2826 unsigned int bidi_bytes = 0;
2828 if (unlikely(blk_bidi_rq(rq)))
2829 bidi_bytes = blk_rq_bytes(rq->next_rq);
2831 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2834 EXPORT_SYMBOL(blk_end_request_all);
2837 * blk_end_request_cur - Helper function to finish the current request chunk.
2838 * @rq: the request to finish the current chunk for
2839 * @error: %0 for success, < %0 for error
2842 * Complete the current consecutively mapped chunk from @rq.
2845 * %false - we are done with this request
2846 * %true - still buffers pending for this request
2848 bool blk_end_request_cur(struct request *rq, int error)
2850 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2852 EXPORT_SYMBOL(blk_end_request_cur);
2855 * blk_end_request_err - Finish a request till the next failure boundary.
2856 * @rq: the request to finish till the next failure boundary for
2857 * @error: must be negative errno
2860 * Complete @rq till the next failure boundary.
2863 * %false - we are done with this request
2864 * %true - still buffers pending for this request
2866 bool blk_end_request_err(struct request *rq, int error)
2868 WARN_ON(error >= 0);
2869 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2871 EXPORT_SYMBOL_GPL(blk_end_request_err);
2874 * __blk_end_request - Helper function for drivers to complete the request.
2875 * @rq: the request being processed
2876 * @error: %0 for success, < %0 for error
2877 * @nr_bytes: number of bytes to complete
2880 * Must be called with queue lock held unlike blk_end_request().
2883 * %false - we are done with this request
2884 * %true - still buffers pending for this request
2886 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2888 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2890 EXPORT_SYMBOL(__blk_end_request);
2893 * __blk_end_request_all - Helper function for drives to finish the request.
2894 * @rq: the request to finish
2895 * @error: %0 for success, < %0 for error
2898 * Completely finish @rq. Must be called with queue lock held.
2900 void __blk_end_request_all(struct request *rq, int error)
2903 unsigned int bidi_bytes = 0;
2905 if (unlikely(blk_bidi_rq(rq)))
2906 bidi_bytes = blk_rq_bytes(rq->next_rq);
2908 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2911 EXPORT_SYMBOL(__blk_end_request_all);
2914 * __blk_end_request_cur - Helper function to finish the current request chunk.
2915 * @rq: the request to finish the current chunk for
2916 * @error: %0 for success, < %0 for error
2919 * Complete the current consecutively mapped chunk from @rq. Must
2920 * be called with queue lock held.
2923 * %false - we are done with this request
2924 * %true - still buffers pending for this request
2926 bool __blk_end_request_cur(struct request *rq, int error)
2928 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2930 EXPORT_SYMBOL(__blk_end_request_cur);
2933 * __blk_end_request_err - Finish a request till the next failure boundary.
2934 * @rq: the request to finish till the next failure boundary for
2935 * @error: must be negative errno
2938 * Complete @rq till the next failure boundary. Must be called
2939 * with queue lock held.
2942 * %false - we are done with this request
2943 * %true - still buffers pending for this request
2945 bool __blk_end_request_err(struct request *rq, int error)
2947 WARN_ON(error >= 0);
2948 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2950 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2952 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2955 if (bio_has_data(bio))
2956 rq->nr_phys_segments = bio_phys_segments(q, bio);
2958 rq->__data_len = bio->bi_iter.bi_size;
2959 rq->bio = rq->biotail = bio;
2962 rq->rq_disk = bio->bi_bdev->bd_disk;
2965 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2967 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2968 * @rq: the request to be flushed
2971 * Flush all pages in @rq.
2973 void rq_flush_dcache_pages(struct request *rq)
2975 struct req_iterator iter;
2976 struct bio_vec bvec;
2978 rq_for_each_segment(bvec, rq, iter)
2979 flush_dcache_page(bvec.bv_page);
2981 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2985 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2986 * @q : the queue of the device being checked
2989 * Check if underlying low-level drivers of a device are busy.
2990 * If the drivers want to export their busy state, they must set own
2991 * exporting function using blk_queue_lld_busy() first.
2993 * Basically, this function is used only by request stacking drivers
2994 * to stop dispatching requests to underlying devices when underlying
2995 * devices are busy. This behavior helps more I/O merging on the queue
2996 * of the request stacking driver and prevents I/O throughput regression
2997 * on burst I/O load.
3000 * 0 - Not busy (The request stacking driver should dispatch request)
3001 * 1 - Busy (The request stacking driver should stop dispatching request)
3003 int blk_lld_busy(struct request_queue *q)
3006 return q->lld_busy_fn(q);
3010 EXPORT_SYMBOL_GPL(blk_lld_busy);
3013 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3014 * @rq: the clone request to be cleaned up
3017 * Free all bios in @rq for a cloned request.
3019 void blk_rq_unprep_clone(struct request *rq)
3023 while ((bio = rq->bio) != NULL) {
3024 rq->bio = bio->bi_next;
3029 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3032 * Copy attributes of the original request to the clone request.
3033 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3035 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3037 dst->cpu = src->cpu;
3038 dst->__sector = blk_rq_pos(src);
3039 dst->__data_len = blk_rq_bytes(src);
3040 dst->nr_phys_segments = src->nr_phys_segments;
3041 dst->ioprio = src->ioprio;
3042 dst->extra_len = src->extra_len;
3046 * blk_rq_prep_clone - Helper function to setup clone request
3047 * @rq: the request to be setup
3048 * @rq_src: original request to be cloned
3049 * @bs: bio_set that bios for clone are allocated from
3050 * @gfp_mask: memory allocation mask for bio
3051 * @bio_ctr: setup function to be called for each clone bio.
3052 * Returns %0 for success, non %0 for failure.
3053 * @data: private data to be passed to @bio_ctr
3056 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3057 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3058 * are not copied, and copying such parts is the caller's responsibility.
3059 * Also, pages which the original bios are pointing to are not copied
3060 * and the cloned bios just point same pages.
3061 * So cloned bios must be completed before original bios, which means
3062 * the caller must complete @rq before @rq_src.
3064 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3065 struct bio_set *bs, gfp_t gfp_mask,
3066 int (*bio_ctr)(struct bio *, struct bio *, void *),
3069 struct bio *bio, *bio_src;
3074 __rq_for_each_bio(bio_src, rq_src) {
3075 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3079 if (bio_ctr && bio_ctr(bio, bio_src, data))
3083 rq->biotail->bi_next = bio;
3086 rq->bio = rq->biotail = bio;
3089 __blk_rq_prep_clone(rq, rq_src);
3096 blk_rq_unprep_clone(rq);
3100 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3102 int kblockd_schedule_work(struct work_struct *work)
3104 return queue_work(kblockd_workqueue, work);
3106 EXPORT_SYMBOL(kblockd_schedule_work);
3108 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3110 return queue_work_on(cpu, kblockd_workqueue, work);
3112 EXPORT_SYMBOL(kblockd_schedule_work_on);
3114 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3115 unsigned long delay)
3117 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3119 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3121 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3122 unsigned long delay)
3124 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3126 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3129 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3130 * @plug: The &struct blk_plug that needs to be initialized
3133 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3134 * pending I/O should the task end up blocking between blk_start_plug() and
3135 * blk_finish_plug(). This is important from a performance perspective, but
3136 * also ensures that we don't deadlock. For instance, if the task is blocking
3137 * for a memory allocation, memory reclaim could end up wanting to free a
3138 * page belonging to that request that is currently residing in our private
3139 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3140 * this kind of deadlock.
3142 void blk_start_plug(struct blk_plug *plug)
3144 struct task_struct *tsk = current;
3147 * If this is a nested plug, don't actually assign it.
3152 INIT_LIST_HEAD(&plug->list);
3153 INIT_LIST_HEAD(&plug->mq_list);
3154 INIT_LIST_HEAD(&plug->cb_list);
3156 * Store ordering should not be needed here, since a potential
3157 * preempt will imply a full memory barrier
3161 EXPORT_SYMBOL(blk_start_plug);
3163 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3165 struct request *rqa = container_of(a, struct request, queuelist);
3166 struct request *rqb = container_of(b, struct request, queuelist);
3168 return !(rqa->q < rqb->q ||
3169 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3173 * If 'from_schedule' is true, then postpone the dispatch of requests
3174 * until a safe kblockd context. We due this to avoid accidental big
3175 * additional stack usage in driver dispatch, in places where the originally
3176 * plugger did not intend it.
3178 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3180 __releases(q->queue_lock)
3182 trace_block_unplug(q, depth, !from_schedule);
3185 blk_run_queue_async(q);
3188 spin_unlock(q->queue_lock);
3191 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3193 LIST_HEAD(callbacks);
3195 while (!list_empty(&plug->cb_list)) {
3196 list_splice_init(&plug->cb_list, &callbacks);
3198 while (!list_empty(&callbacks)) {
3199 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3202 list_del(&cb->list);
3203 cb->callback(cb, from_schedule);
3208 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3211 struct blk_plug *plug = current->plug;
3212 struct blk_plug_cb *cb;
3217 list_for_each_entry(cb, &plug->cb_list, list)
3218 if (cb->callback == unplug && cb->data == data)
3221 /* Not currently on the callback list */
3222 BUG_ON(size < sizeof(*cb));
3223 cb = kzalloc(size, GFP_ATOMIC);
3226 cb->callback = unplug;
3227 list_add(&cb->list, &plug->cb_list);
3231 EXPORT_SYMBOL(blk_check_plugged);
3233 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3235 struct request_queue *q;
3236 unsigned long flags;
3241 flush_plug_callbacks(plug, from_schedule);
3243 if (!list_empty(&plug->mq_list))
3244 blk_mq_flush_plug_list(plug, from_schedule);
3246 if (list_empty(&plug->list))
3249 list_splice_init(&plug->list, &list);
3251 list_sort(NULL, &list, plug_rq_cmp);
3257 * Save and disable interrupts here, to avoid doing it for every
3258 * queue lock we have to take.
3260 local_irq_save(flags);
3261 while (!list_empty(&list)) {
3262 rq = list_entry_rq(list.next);
3263 list_del_init(&rq->queuelist);
3267 * This drops the queue lock
3270 queue_unplugged(q, depth, from_schedule);
3273 spin_lock(q->queue_lock);
3277 * Short-circuit if @q is dead
3279 if (unlikely(blk_queue_dying(q))) {
3280 __blk_end_request_all(rq, -ENODEV);
3285 * rq is already accounted, so use raw insert
3287 if (op_is_flush(rq->cmd_flags))
3288 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3290 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3296 * This drops the queue lock
3299 queue_unplugged(q, depth, from_schedule);
3301 local_irq_restore(flags);
3304 void blk_finish_plug(struct blk_plug *plug)
3306 if (plug != current->plug)
3308 blk_flush_plug_list(plug, false);
3310 current->plug = NULL;
3312 EXPORT_SYMBOL(blk_finish_plug);
3316 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3317 * @q: the queue of the device
3318 * @dev: the device the queue belongs to
3321 * Initialize runtime-PM-related fields for @q and start auto suspend for
3322 * @dev. Drivers that want to take advantage of request-based runtime PM
3323 * should call this function after @dev has been initialized, and its
3324 * request queue @q has been allocated, and runtime PM for it can not happen
3325 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3326 * cases, driver should call this function before any I/O has taken place.
3328 * This function takes care of setting up using auto suspend for the device,
3329 * the autosuspend delay is set to -1 to make runtime suspend impossible
3330 * until an updated value is either set by user or by driver. Drivers do
3331 * not need to touch other autosuspend settings.
3333 * The block layer runtime PM is request based, so only works for drivers
3334 * that use request as their IO unit instead of those directly use bio's.
3336 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3339 q->rpm_status = RPM_ACTIVE;
3340 pm_runtime_set_autosuspend_delay(q->dev, -1);
3341 pm_runtime_use_autosuspend(q->dev);
3343 EXPORT_SYMBOL(blk_pm_runtime_init);
3346 * blk_pre_runtime_suspend - Pre runtime suspend check
3347 * @q: the queue of the device
3350 * This function will check if runtime suspend is allowed for the device
3351 * by examining if there are any requests pending in the queue. If there
3352 * are requests pending, the device can not be runtime suspended; otherwise,
3353 * the queue's status will be updated to SUSPENDING and the driver can
3354 * proceed to suspend the device.
3356 * For the not allowed case, we mark last busy for the device so that
3357 * runtime PM core will try to autosuspend it some time later.
3359 * This function should be called near the start of the device's
3360 * runtime_suspend callback.
3363 * 0 - OK to runtime suspend the device
3364 * -EBUSY - Device should not be runtime suspended
3366 int blk_pre_runtime_suspend(struct request_queue *q)
3373 spin_lock_irq(q->queue_lock);
3374 if (q->nr_pending) {
3376 pm_runtime_mark_last_busy(q->dev);
3378 q->rpm_status = RPM_SUSPENDING;
3380 spin_unlock_irq(q->queue_lock);
3383 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3386 * blk_post_runtime_suspend - Post runtime suspend processing
3387 * @q: the queue of the device
3388 * @err: return value of the device's runtime_suspend function
3391 * Update the queue's runtime status according to the return value of the
3392 * device's runtime suspend function and mark last busy for the device so
3393 * that PM core will try to auto suspend the device at a later time.
3395 * This function should be called near the end of the device's
3396 * runtime_suspend callback.
3398 void blk_post_runtime_suspend(struct request_queue *q, int err)
3403 spin_lock_irq(q->queue_lock);
3405 q->rpm_status = RPM_SUSPENDED;
3407 q->rpm_status = RPM_ACTIVE;
3408 pm_runtime_mark_last_busy(q->dev);
3410 spin_unlock_irq(q->queue_lock);
3412 EXPORT_SYMBOL(blk_post_runtime_suspend);
3415 * blk_pre_runtime_resume - Pre runtime resume processing
3416 * @q: the queue of the device
3419 * Update the queue's runtime status to RESUMING in preparation for the
3420 * runtime resume of the device.
3422 * This function should be called near the start of the device's
3423 * runtime_resume callback.
3425 void blk_pre_runtime_resume(struct request_queue *q)
3430 spin_lock_irq(q->queue_lock);
3431 q->rpm_status = RPM_RESUMING;
3432 spin_unlock_irq(q->queue_lock);
3434 EXPORT_SYMBOL(blk_pre_runtime_resume);
3437 * blk_post_runtime_resume - Post runtime resume processing
3438 * @q: the queue of the device
3439 * @err: return value of the device's runtime_resume function
3442 * Update the queue's runtime status according to the return value of the
3443 * device's runtime_resume function. If it is successfully resumed, process
3444 * the requests that are queued into the device's queue when it is resuming
3445 * and then mark last busy and initiate autosuspend for it.
3447 * This function should be called near the end of the device's
3448 * runtime_resume callback.
3450 void blk_post_runtime_resume(struct request_queue *q, int err)
3455 spin_lock_irq(q->queue_lock);
3457 q->rpm_status = RPM_ACTIVE;
3459 pm_runtime_mark_last_busy(q->dev);
3460 pm_request_autosuspend(q->dev);
3462 q->rpm_status = RPM_SUSPENDED;
3464 spin_unlock_irq(q->queue_lock);
3466 EXPORT_SYMBOL(blk_post_runtime_resume);
3469 * blk_set_runtime_active - Force runtime status of the queue to be active
3470 * @q: the queue of the device
3472 * If the device is left runtime suspended during system suspend the resume
3473 * hook typically resumes the device and corrects runtime status
3474 * accordingly. However, that does not affect the queue runtime PM status
3475 * which is still "suspended". This prevents processing requests from the
3478 * This function can be used in driver's resume hook to correct queue
3479 * runtime PM status and re-enable peeking requests from the queue. It
3480 * should be called before first request is added to the queue.
3482 void blk_set_runtime_active(struct request_queue *q)
3484 spin_lock_irq(q->queue_lock);
3485 q->rpm_status = RPM_ACTIVE;
3486 pm_runtime_mark_last_busy(q->dev);
3487 pm_request_autosuspend(q->dev);
3488 spin_unlock_irq(q->queue_lock);
3490 EXPORT_SYMBOL(blk_set_runtime_active);
3493 int __init blk_dev_init(void)
3495 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3496 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3497 FIELD_SIZEOF(struct request, cmd_flags));
3498 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3499 FIELD_SIZEOF(struct bio, bi_opf));
3501 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3502 kblockd_workqueue = alloc_workqueue("kblockd",
3503 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3504 if (!kblockd_workqueue)
3505 panic("Failed to create kblockd\n");
3507 request_cachep = kmem_cache_create("blkdev_requests",
3508 sizeof(struct request), 0, SLAB_PANIC, NULL);
3510 blk_requestq_cachep = kmem_cache_create("request_queue",
3511 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3513 #ifdef CONFIG_DEBUG_FS
3514 blk_debugfs_root = debugfs_create_dir("block", NULL);