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>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.h>
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
49 DEFINE_IDA(blk_queue_ida);
52 * For the allocated request tables
54 struct kmem_cache *request_cachep;
57 * For queue allocation
59 struct kmem_cache *blk_requestq_cachep;
62 * Controlling structure to kblockd
64 static struct workqueue_struct *kblockd_workqueue;
66 static void blk_clear_congested(struct request_list *rl, int sync)
68 #ifdef CONFIG_CGROUP_WRITEBACK
69 clear_wb_congested(rl->blkg->wb_congested, sync);
72 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
73 * flip its congestion state for events on other blkcgs.
75 if (rl == &rl->q->root_rl)
76 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
80 static void blk_set_congested(struct request_list *rl, int sync)
82 #ifdef CONFIG_CGROUP_WRITEBACK
83 set_wb_congested(rl->blkg->wb_congested, sync);
85 /* see blk_clear_congested() */
86 if (rl == &rl->q->root_rl)
87 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
91 void blk_queue_congestion_threshold(struct request_queue *q)
95 nr = q->nr_requests - (q->nr_requests / 8) + 1;
96 if (nr > q->nr_requests)
98 q->nr_congestion_on = nr;
100 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 q->nr_congestion_off = nr;
107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * Locates the passed device's request queue and returns the address of its
111 * backing_dev_info. This function can only be called if @bdev is opened
112 * and the return value is never NULL.
114 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
116 struct request_queue *q = bdev_get_queue(bdev);
118 return &q->backing_dev_info;
120 EXPORT_SYMBOL(blk_get_backing_dev_info);
122 void blk_rq_init(struct request_queue *q, struct request *rq)
124 memset(rq, 0, sizeof(*rq));
126 INIT_LIST_HEAD(&rq->queuelist);
127 INIT_LIST_HEAD(&rq->timeout_list);
130 rq->__sector = (sector_t) -1;
131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
134 rq->cmd_len = BLK_MAX_CDB;
136 rq->start_time = jiffies;
137 set_start_time_ns(rq);
140 EXPORT_SYMBOL(blk_rq_init);
142 static void req_bio_endio(struct request *rq, struct bio *bio,
143 unsigned int nbytes, int error)
146 bio->bi_error = error;
148 if (unlikely(rq->cmd_flags & REQ_QUIET))
149 bio_set_flag(bio, BIO_QUIET);
151 bio_advance(bio, nbytes);
153 /* don't actually finish bio if it's part of flush sequence */
154 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
158 void blk_dump_rq_flags(struct request *rq, char *msg)
162 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
163 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
164 (unsigned long long) rq->cmd_flags);
166 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
167 (unsigned long long)blk_rq_pos(rq),
168 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
169 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
170 rq->bio, rq->biotail, blk_rq_bytes(rq));
172 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
173 printk(KERN_INFO " cdb: ");
174 for (bit = 0; bit < BLK_MAX_CDB; bit++)
175 printk("%02x ", rq->cmd[bit]);
179 EXPORT_SYMBOL(blk_dump_rq_flags);
181 static void blk_delay_work(struct work_struct *work)
183 struct request_queue *q;
185 q = container_of(work, struct request_queue, delay_work.work);
186 spin_lock_irq(q->queue_lock);
188 spin_unlock_irq(q->queue_lock);
192 * blk_delay_queue - restart queueing after defined interval
193 * @q: The &struct request_queue in question
194 * @msecs: Delay in msecs
197 * Sometimes queueing needs to be postponed for a little while, to allow
198 * resources to come back. This function will make sure that queueing is
199 * restarted around the specified time. Queue lock must be held.
201 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
203 if (likely(!blk_queue_dead(q)))
204 queue_delayed_work(kblockd_workqueue, &q->delay_work,
205 msecs_to_jiffies(msecs));
207 EXPORT_SYMBOL(blk_delay_queue);
210 * blk_start_queue_async - asynchronously restart a previously stopped queue
211 * @q: The &struct request_queue in question
214 * blk_start_queue_async() will clear the stop flag on the queue, and
215 * ensure that the request_fn for the queue is run from an async
218 void blk_start_queue_async(struct request_queue *q)
220 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
221 blk_run_queue_async(q);
223 EXPORT_SYMBOL(blk_start_queue_async);
226 * blk_start_queue - restart a previously stopped queue
227 * @q: The &struct request_queue in question
230 * blk_start_queue() will clear the stop flag on the queue, and call
231 * the request_fn for the queue if it was in a stopped state when
232 * entered. Also see blk_stop_queue(). Queue lock must be held.
234 void blk_start_queue(struct request_queue *q)
236 WARN_ON(!irqs_disabled());
238 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
241 EXPORT_SYMBOL(blk_start_queue);
244 * blk_stop_queue - stop a queue
245 * @q: The &struct request_queue in question
248 * The Linux block layer assumes that a block driver will consume all
249 * entries on the request queue when the request_fn strategy is called.
250 * Often this will not happen, because of hardware limitations (queue
251 * depth settings). If a device driver gets a 'queue full' response,
252 * or if it simply chooses not to queue more I/O at one point, it can
253 * call this function to prevent the request_fn from being called until
254 * the driver has signalled it's ready to go again. This happens by calling
255 * blk_start_queue() to restart queue operations. Queue lock must be held.
257 void blk_stop_queue(struct request_queue *q)
259 cancel_delayed_work(&q->delay_work);
260 queue_flag_set(QUEUE_FLAG_STOPPED, q);
262 EXPORT_SYMBOL(blk_stop_queue);
265 * blk_sync_queue - cancel any pending callbacks on a queue
269 * The block layer may perform asynchronous callback activity
270 * on a queue, such as calling the unplug function after a timeout.
271 * A block device may call blk_sync_queue to ensure that any
272 * such activity is cancelled, thus allowing it to release resources
273 * that the callbacks might use. The caller must already have made sure
274 * that its ->make_request_fn will not re-add plugging prior to calling
277 * This function does not cancel any asynchronous activity arising
278 * out of elevator or throttling code. That would require elevator_exit()
279 * and blkcg_exit_queue() to be called with queue lock initialized.
282 void blk_sync_queue(struct request_queue *q)
284 del_timer_sync(&q->timeout);
287 struct blk_mq_hw_ctx *hctx;
290 queue_for_each_hw_ctx(q, hctx, i) {
291 cancel_delayed_work_sync(&hctx->run_work);
292 cancel_delayed_work_sync(&hctx->delay_work);
295 cancel_delayed_work_sync(&q->delay_work);
298 EXPORT_SYMBOL(blk_sync_queue);
301 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
302 * @q: The queue to run
305 * Invoke request handling on a queue if there are any pending requests.
306 * May be used to restart request handling after a request has completed.
307 * This variant runs the queue whether or not the queue has been
308 * stopped. Must be called with the queue lock held and interrupts
309 * disabled. See also @blk_run_queue.
311 inline void __blk_run_queue_uncond(struct request_queue *q)
313 if (unlikely(blk_queue_dead(q)))
317 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
318 * the queue lock internally. As a result multiple threads may be
319 * running such a request function concurrently. Keep track of the
320 * number of active request_fn invocations such that blk_drain_queue()
321 * can wait until all these request_fn calls have finished.
323 q->request_fn_active++;
325 q->request_fn_active--;
327 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
330 * __blk_run_queue - run a single device queue
331 * @q: The queue to run
334 * See @blk_run_queue. This variant must be called with the queue lock
335 * held and interrupts disabled.
337 void __blk_run_queue(struct request_queue *q)
339 if (unlikely(blk_queue_stopped(q)))
342 __blk_run_queue_uncond(q);
344 EXPORT_SYMBOL(__blk_run_queue);
347 * blk_run_queue_async - run a single device queue in workqueue context
348 * @q: The queue to run
351 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
352 * of us. The caller must hold the queue lock.
354 void blk_run_queue_async(struct request_queue *q)
356 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
357 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
359 EXPORT_SYMBOL(blk_run_queue_async);
362 * blk_run_queue - run a single device queue
363 * @q: The queue to run
366 * Invoke request handling on this queue, if it has pending work to do.
367 * May be used to restart queueing when a request has completed.
369 void blk_run_queue(struct request_queue *q)
373 spin_lock_irqsave(q->queue_lock, flags);
375 spin_unlock_irqrestore(q->queue_lock, flags);
377 EXPORT_SYMBOL(blk_run_queue);
379 void blk_put_queue(struct request_queue *q)
381 kobject_put(&q->kobj);
383 EXPORT_SYMBOL(blk_put_queue);
386 * __blk_drain_queue - drain requests from request_queue
388 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
390 * Drain requests from @q. If @drain_all is set, all requests are drained.
391 * If not, only ELVPRIV requests are drained. The caller is responsible
392 * for ensuring that no new requests which need to be drained are queued.
394 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
395 __releases(q->queue_lock)
396 __acquires(q->queue_lock)
400 lockdep_assert_held(q->queue_lock);
406 * The caller might be trying to drain @q before its
407 * elevator is initialized.
410 elv_drain_elevator(q);
412 blkcg_drain_queue(q);
415 * This function might be called on a queue which failed
416 * driver init after queue creation or is not yet fully
417 * active yet. Some drivers (e.g. fd and loop) get unhappy
418 * in such cases. Kick queue iff dispatch queue has
419 * something on it and @q has request_fn set.
421 if (!list_empty(&q->queue_head) && q->request_fn)
424 drain |= q->nr_rqs_elvpriv;
425 drain |= q->request_fn_active;
428 * Unfortunately, requests are queued at and tracked from
429 * multiple places and there's no single counter which can
430 * be drained. Check all the queues and counters.
433 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
434 drain |= !list_empty(&q->queue_head);
435 for (i = 0; i < 2; i++) {
436 drain |= q->nr_rqs[i];
437 drain |= q->in_flight[i];
439 drain |= !list_empty(&fq->flush_queue[i]);
446 spin_unlock_irq(q->queue_lock);
450 spin_lock_irq(q->queue_lock);
454 * With queue marked dead, any woken up waiter will fail the
455 * allocation path, so the wakeup chaining is lost and we're
456 * left with hung waiters. We need to wake up those waiters.
459 struct request_list *rl;
461 blk_queue_for_each_rl(rl, q)
462 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
463 wake_up_all(&rl->wait[i]);
468 * blk_queue_bypass_start - enter queue bypass mode
469 * @q: queue of interest
471 * In bypass mode, only the dispatch FIFO queue of @q is used. This
472 * function makes @q enter bypass mode and drains all requests which were
473 * throttled or issued before. On return, it's guaranteed that no request
474 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
475 * inside queue or RCU read lock.
477 void blk_queue_bypass_start(struct request_queue *q)
479 spin_lock_irq(q->queue_lock);
481 queue_flag_set(QUEUE_FLAG_BYPASS, q);
482 spin_unlock_irq(q->queue_lock);
485 * Queues start drained. Skip actual draining till init is
486 * complete. This avoids lenghty delays during queue init which
487 * can happen many times during boot.
489 if (blk_queue_init_done(q)) {
490 spin_lock_irq(q->queue_lock);
491 __blk_drain_queue(q, false);
492 spin_unlock_irq(q->queue_lock);
494 /* ensure blk_queue_bypass() is %true inside RCU read lock */
498 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
501 * blk_queue_bypass_end - leave queue bypass mode
502 * @q: queue of interest
504 * Leave bypass mode and restore the normal queueing behavior.
506 void blk_queue_bypass_end(struct request_queue *q)
508 spin_lock_irq(q->queue_lock);
509 if (!--q->bypass_depth)
510 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
511 WARN_ON_ONCE(q->bypass_depth < 0);
512 spin_unlock_irq(q->queue_lock);
514 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
516 void blk_set_queue_dying(struct request_queue *q)
518 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
521 blk_mq_wake_waiters(q);
523 struct request_list *rl;
525 blk_queue_for_each_rl(rl, q) {
527 wake_up(&rl->wait[BLK_RW_SYNC]);
528 wake_up(&rl->wait[BLK_RW_ASYNC]);
533 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
536 * blk_cleanup_queue - shutdown a request queue
537 * @q: request queue to shutdown
539 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
540 * put it. All future requests will be failed immediately with -ENODEV.
542 void blk_cleanup_queue(struct request_queue *q)
544 spinlock_t *lock = q->queue_lock;
546 /* mark @q DYING, no new request or merges will be allowed afterwards */
547 mutex_lock(&q->sysfs_lock);
548 blk_set_queue_dying(q);
552 * A dying queue is permanently in bypass mode till released. Note
553 * that, unlike blk_queue_bypass_start(), we aren't performing
554 * synchronize_rcu() after entering bypass mode to avoid the delay
555 * as some drivers create and destroy a lot of queues while
556 * probing. This is still safe because blk_release_queue() will be
557 * called only after the queue refcnt drops to zero and nothing,
558 * RCU or not, would be traversing the queue by then.
561 queue_flag_set(QUEUE_FLAG_BYPASS, q);
563 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
564 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
565 queue_flag_set(QUEUE_FLAG_DYING, q);
566 spin_unlock_irq(lock);
567 mutex_unlock(&q->sysfs_lock);
570 * Drain all requests queued before DYING marking. Set DEAD flag to
571 * prevent that q->request_fn() gets invoked after draining finished.
576 __blk_drain_queue(q, true);
577 queue_flag_set(QUEUE_FLAG_DEAD, q);
578 spin_unlock_irq(lock);
580 /* for synchronous bio-based driver finish in-flight integrity i/o */
581 blk_flush_integrity();
583 /* @q won't process any more request, flush async actions */
584 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
588 blk_mq_free_queue(q);
589 percpu_ref_exit(&q->q_usage_counter);
592 if (q->queue_lock != &q->__queue_lock)
593 q->queue_lock = &q->__queue_lock;
594 spin_unlock_irq(lock);
596 bdi_unregister(&q->backing_dev_info);
598 /* @q is and will stay empty, shutdown and put */
601 EXPORT_SYMBOL(blk_cleanup_queue);
603 /* Allocate memory local to the request queue */
604 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
606 int nid = (int)(long)data;
607 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
610 static void free_request_struct(void *element, void *unused)
612 kmem_cache_free(request_cachep, element);
615 int blk_init_rl(struct request_list *rl, struct request_queue *q,
618 if (unlikely(rl->rq_pool))
622 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
623 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
624 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
625 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
627 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
629 (void *)(long)q->node, gfp_mask,
637 void blk_exit_rl(struct request_list *rl)
640 mempool_destroy(rl->rq_pool);
643 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
645 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
647 EXPORT_SYMBOL(blk_alloc_queue);
649 int blk_queue_enter(struct request_queue *q, bool nowait)
654 if (percpu_ref_tryget_live(&q->q_usage_counter))
660 ret = wait_event_interruptible(q->mq_freeze_wq,
661 !atomic_read(&q->mq_freeze_depth) ||
663 if (blk_queue_dying(q))
670 void blk_queue_exit(struct request_queue *q)
672 percpu_ref_put(&q->q_usage_counter);
675 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
677 struct request_queue *q =
678 container_of(ref, struct request_queue, q_usage_counter);
680 wake_up_all(&q->mq_freeze_wq);
683 static void blk_rq_timed_out_timer(unsigned long data)
685 struct request_queue *q = (struct request_queue *)data;
687 kblockd_schedule_work(&q->timeout_work);
690 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
692 struct request_queue *q;
695 q = kmem_cache_alloc_node(blk_requestq_cachep,
696 gfp_mask | __GFP_ZERO, node_id);
700 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
704 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
708 q->backing_dev_info.ra_pages =
709 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
710 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
711 q->backing_dev_info.name = "block";
714 err = bdi_init(&q->backing_dev_info);
718 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
719 laptop_mode_timer_fn, (unsigned long) q);
720 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
721 INIT_LIST_HEAD(&q->queue_head);
722 INIT_LIST_HEAD(&q->timeout_list);
723 INIT_LIST_HEAD(&q->icq_list);
724 #ifdef CONFIG_BLK_CGROUP
725 INIT_LIST_HEAD(&q->blkg_list);
727 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
729 kobject_init(&q->kobj, &blk_queue_ktype);
731 mutex_init(&q->sysfs_lock);
732 spin_lock_init(&q->__queue_lock);
735 * By default initialize queue_lock to internal lock and driver can
736 * override it later if need be.
738 q->queue_lock = &q->__queue_lock;
741 * A queue starts its life with bypass turned on to avoid
742 * unnecessary bypass on/off overhead and nasty surprises during
743 * init. The initial bypass will be finished when the queue is
744 * registered by blk_register_queue().
747 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
749 init_waitqueue_head(&q->mq_freeze_wq);
752 * Init percpu_ref in atomic mode so that it's faster to shutdown.
753 * See blk_register_queue() for details.
755 if (percpu_ref_init(&q->q_usage_counter,
756 blk_queue_usage_counter_release,
757 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
760 if (blkcg_init_queue(q))
766 percpu_ref_exit(&q->q_usage_counter);
768 bdi_destroy(&q->backing_dev_info);
770 bioset_free(q->bio_split);
772 ida_simple_remove(&blk_queue_ida, q->id);
774 kmem_cache_free(blk_requestq_cachep, q);
777 EXPORT_SYMBOL(blk_alloc_queue_node);
780 * blk_init_queue - prepare a request queue for use with a block device
781 * @rfn: The function to be called to process requests that have been
782 * placed on the queue.
783 * @lock: Request queue spin lock
786 * If a block device wishes to use the standard request handling procedures,
787 * which sorts requests and coalesces adjacent requests, then it must
788 * call blk_init_queue(). The function @rfn will be called when there
789 * are requests on the queue that need to be processed. If the device
790 * supports plugging, then @rfn may not be called immediately when requests
791 * are available on the queue, but may be called at some time later instead.
792 * Plugged queues are generally unplugged when a buffer belonging to one
793 * of the requests on the queue is needed, or due to memory pressure.
795 * @rfn is not required, or even expected, to remove all requests off the
796 * queue, but only as many as it can handle at a time. If it does leave
797 * requests on the queue, it is responsible for arranging that the requests
798 * get dealt with eventually.
800 * The queue spin lock must be held while manipulating the requests on the
801 * request queue; this lock will be taken also from interrupt context, so irq
802 * disabling is needed for it.
804 * Function returns a pointer to the initialized request queue, or %NULL if
808 * blk_init_queue() must be paired with a blk_cleanup_queue() call
809 * when the block device is deactivated (such as at module unload).
812 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
814 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
816 EXPORT_SYMBOL(blk_init_queue);
818 struct request_queue *
819 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
821 struct request_queue *uninit_q, *q;
823 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
827 q = blk_init_allocated_queue(uninit_q, rfn, lock);
829 blk_cleanup_queue(uninit_q);
833 EXPORT_SYMBOL(blk_init_queue_node);
835 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
837 struct request_queue *
838 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
844 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
848 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
851 INIT_WORK(&q->timeout_work, blk_timeout_work);
853 q->prep_rq_fn = NULL;
854 q->unprep_rq_fn = NULL;
855 q->queue_flags |= QUEUE_FLAG_DEFAULT;
857 /* Override internal queue lock with supplied lock pointer */
859 q->queue_lock = lock;
862 * This also sets hw/phys segments, boundary and size
864 blk_queue_make_request(q, blk_queue_bio);
866 q->sg_reserved_size = INT_MAX;
868 /* Protect q->elevator from elevator_change */
869 mutex_lock(&q->sysfs_lock);
872 if (elevator_init(q, NULL)) {
873 mutex_unlock(&q->sysfs_lock);
877 mutex_unlock(&q->sysfs_lock);
882 blk_free_flush_queue(q->fq);
885 EXPORT_SYMBOL(blk_init_allocated_queue);
887 bool blk_get_queue(struct request_queue *q)
889 if (likely(!blk_queue_dying(q))) {
896 EXPORT_SYMBOL(blk_get_queue);
898 static inline void blk_free_request(struct request_list *rl, struct request *rq)
900 if (rq->cmd_flags & REQ_ELVPRIV) {
901 elv_put_request(rl->q, rq);
903 put_io_context(rq->elv.icq->ioc);
906 mempool_free(rq, rl->rq_pool);
910 * ioc_batching returns true if the ioc is a valid batching request and
911 * should be given priority access to a request.
913 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
919 * Make sure the process is able to allocate at least 1 request
920 * even if the batch times out, otherwise we could theoretically
923 return ioc->nr_batch_requests == q->nr_batching ||
924 (ioc->nr_batch_requests > 0
925 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
929 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
930 * will cause the process to be a "batcher" on all queues in the system. This
931 * is the behaviour we want though - once it gets a wakeup it should be given
934 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
936 if (!ioc || ioc_batching(q, ioc))
939 ioc->nr_batch_requests = q->nr_batching;
940 ioc->last_waited = jiffies;
943 static void __freed_request(struct request_list *rl, int sync)
945 struct request_queue *q = rl->q;
947 if (rl->count[sync] < queue_congestion_off_threshold(q))
948 blk_clear_congested(rl, sync);
950 if (rl->count[sync] + 1 <= q->nr_requests) {
951 if (waitqueue_active(&rl->wait[sync]))
952 wake_up(&rl->wait[sync]);
954 blk_clear_rl_full(rl, sync);
959 * A request has just been released. Account for it, update the full and
960 * congestion status, wake up any waiters. Called under q->queue_lock.
962 static void freed_request(struct request_list *rl, unsigned int flags)
964 struct request_queue *q = rl->q;
965 int sync = rw_is_sync(flags);
969 if (flags & REQ_ELVPRIV)
972 __freed_request(rl, sync);
974 if (unlikely(rl->starved[sync ^ 1]))
975 __freed_request(rl, sync ^ 1);
978 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
980 struct request_list *rl;
981 int on_thresh, off_thresh;
983 spin_lock_irq(q->queue_lock);
985 blk_queue_congestion_threshold(q);
986 on_thresh = queue_congestion_on_threshold(q);
987 off_thresh = queue_congestion_off_threshold(q);
989 blk_queue_for_each_rl(rl, q) {
990 if (rl->count[BLK_RW_SYNC] >= on_thresh)
991 blk_set_congested(rl, BLK_RW_SYNC);
992 else if (rl->count[BLK_RW_SYNC] < off_thresh)
993 blk_clear_congested(rl, BLK_RW_SYNC);
995 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
996 blk_set_congested(rl, BLK_RW_ASYNC);
997 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
998 blk_clear_congested(rl, BLK_RW_ASYNC);
1000 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1001 blk_set_rl_full(rl, BLK_RW_SYNC);
1003 blk_clear_rl_full(rl, BLK_RW_SYNC);
1004 wake_up(&rl->wait[BLK_RW_SYNC]);
1007 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1008 blk_set_rl_full(rl, BLK_RW_ASYNC);
1010 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1011 wake_up(&rl->wait[BLK_RW_ASYNC]);
1015 spin_unlock_irq(q->queue_lock);
1020 * Determine if elevator data should be initialized when allocating the
1021 * request associated with @bio.
1023 static bool blk_rq_should_init_elevator(struct bio *bio)
1029 * Flush requests do not use the elevator so skip initialization.
1030 * This allows a request to share the flush and elevator data.
1032 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1039 * rq_ioc - determine io_context for request allocation
1040 * @bio: request being allocated is for this bio (can be %NULL)
1042 * Determine io_context to use for request allocation for @bio. May return
1043 * %NULL if %current->io_context doesn't exist.
1045 static struct io_context *rq_ioc(struct bio *bio)
1047 #ifdef CONFIG_BLK_CGROUP
1048 if (bio && bio->bi_ioc)
1051 return current->io_context;
1055 * __get_request - get a free request
1056 * @rl: request list to allocate from
1057 * @rw_flags: RW and SYNC flags
1058 * @bio: bio to allocate request for (can be %NULL)
1059 * @gfp_mask: allocation mask
1061 * Get a free request from @q. This function may fail under memory
1062 * pressure or if @q is dead.
1064 * Must be called with @q->queue_lock held and,
1065 * Returns ERR_PTR on failure, with @q->queue_lock held.
1066 * Returns request pointer on success, with @q->queue_lock *not held*.
1068 static struct request *__get_request(struct request_list *rl, int rw_flags,
1069 struct bio *bio, gfp_t gfp_mask)
1071 struct request_queue *q = rl->q;
1073 struct elevator_type *et = q->elevator->type;
1074 struct io_context *ioc = rq_ioc(bio);
1075 struct io_cq *icq = NULL;
1076 const bool is_sync = rw_is_sync(rw_flags) != 0;
1079 if (unlikely(blk_queue_dying(q)))
1080 return ERR_PTR(-ENODEV);
1082 may_queue = elv_may_queue(q, rw_flags);
1083 if (may_queue == ELV_MQUEUE_NO)
1086 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1087 if (rl->count[is_sync]+1 >= q->nr_requests) {
1089 * The queue will fill after this allocation, so set
1090 * it as full, and mark this process as "batching".
1091 * This process will be allowed to complete a batch of
1092 * requests, others will be blocked.
1094 if (!blk_rl_full(rl, is_sync)) {
1095 ioc_set_batching(q, ioc);
1096 blk_set_rl_full(rl, is_sync);
1098 if (may_queue != ELV_MQUEUE_MUST
1099 && !ioc_batching(q, ioc)) {
1101 * The queue is full and the allocating
1102 * process is not a "batcher", and not
1103 * exempted by the IO scheduler
1105 return ERR_PTR(-ENOMEM);
1109 blk_set_congested(rl, is_sync);
1113 * Only allow batching queuers to allocate up to 50% over the defined
1114 * limit of requests, otherwise we could have thousands of requests
1115 * allocated with any setting of ->nr_requests
1117 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1118 return ERR_PTR(-ENOMEM);
1120 q->nr_rqs[is_sync]++;
1121 rl->count[is_sync]++;
1122 rl->starved[is_sync] = 0;
1125 * Decide whether the new request will be managed by elevator. If
1126 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1127 * prevent the current elevator from being destroyed until the new
1128 * request is freed. This guarantees icq's won't be destroyed and
1129 * makes creating new ones safe.
1131 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1132 * it will be created after releasing queue_lock.
1134 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1135 rw_flags |= REQ_ELVPRIV;
1136 q->nr_rqs_elvpriv++;
1137 if (et->icq_cache && ioc)
1138 icq = ioc_lookup_icq(ioc, q);
1141 if (blk_queue_io_stat(q))
1142 rw_flags |= REQ_IO_STAT;
1143 spin_unlock_irq(q->queue_lock);
1145 /* allocate and init request */
1146 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1151 blk_rq_set_rl(rq, rl);
1152 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1155 if (rw_flags & REQ_ELVPRIV) {
1156 if (unlikely(et->icq_cache && !icq)) {
1158 icq = ioc_create_icq(ioc, q, gfp_mask);
1164 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1167 /* @rq->elv.icq holds io_context until @rq is freed */
1169 get_io_context(icq->ioc);
1173 * ioc may be NULL here, and ioc_batching will be false. That's
1174 * OK, if the queue is under the request limit then requests need
1175 * not count toward the nr_batch_requests limit. There will always
1176 * be some limit enforced by BLK_BATCH_TIME.
1178 if (ioc_batching(q, ioc))
1179 ioc->nr_batch_requests--;
1181 trace_block_getrq(q, bio, rw_flags & 1);
1186 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1187 * and may fail indefinitely under memory pressure and thus
1188 * shouldn't stall IO. Treat this request as !elvpriv. This will
1189 * disturb iosched and blkcg but weird is bettern than dead.
1191 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1192 __func__, dev_name(q->backing_dev_info.dev));
1194 rq->cmd_flags &= ~REQ_ELVPRIV;
1197 spin_lock_irq(q->queue_lock);
1198 q->nr_rqs_elvpriv--;
1199 spin_unlock_irq(q->queue_lock);
1204 * Allocation failed presumably due to memory. Undo anything we
1205 * might have messed up.
1207 * Allocating task should really be put onto the front of the wait
1208 * queue, but this is pretty rare.
1210 spin_lock_irq(q->queue_lock);
1211 freed_request(rl, rw_flags);
1214 * in the very unlikely event that allocation failed and no
1215 * requests for this direction was pending, mark us starved so that
1216 * freeing of a request in the other direction will notice
1217 * us. another possible fix would be to split the rq mempool into
1221 if (unlikely(rl->count[is_sync] == 0))
1222 rl->starved[is_sync] = 1;
1223 return ERR_PTR(-ENOMEM);
1227 * get_request - get a free request
1228 * @q: request_queue to allocate request from
1229 * @rw_flags: RW and SYNC flags
1230 * @bio: bio to allocate request for (can be %NULL)
1231 * @gfp_mask: allocation mask
1233 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1234 * this function keeps retrying under memory pressure and fails iff @q is dead.
1236 * Must be called with @q->queue_lock held and,
1237 * Returns ERR_PTR on failure, with @q->queue_lock held.
1238 * Returns request pointer on success, with @q->queue_lock *not held*.
1240 static struct request *get_request(struct request_queue *q, int rw_flags,
1241 struct bio *bio, gfp_t gfp_mask)
1243 const bool is_sync = rw_is_sync(rw_flags) != 0;
1245 struct request_list *rl;
1248 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1250 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1254 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1259 /* wait on @rl and retry */
1260 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1261 TASK_UNINTERRUPTIBLE);
1263 trace_block_sleeprq(q, bio, rw_flags & 1);
1265 spin_unlock_irq(q->queue_lock);
1269 * After sleeping, we become a "batching" process and will be able
1270 * to allocate at least one request, and up to a big batch of them
1271 * for a small period time. See ioc_batching, ioc_set_batching
1273 ioc_set_batching(q, current->io_context);
1275 spin_lock_irq(q->queue_lock);
1276 finish_wait(&rl->wait[is_sync], &wait);
1281 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1286 BUG_ON(rw != READ && rw != WRITE);
1288 /* create ioc upfront */
1289 create_io_context(gfp_mask, q->node);
1291 spin_lock_irq(q->queue_lock);
1292 rq = get_request(q, rw, NULL, gfp_mask);
1294 spin_unlock_irq(q->queue_lock);
1295 /* q->queue_lock is unlocked at this point */
1300 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1303 return blk_mq_alloc_request(q, rw,
1304 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1305 0 : BLK_MQ_REQ_NOWAIT);
1307 return blk_old_get_request(q, rw, gfp_mask);
1309 EXPORT_SYMBOL(blk_get_request);
1312 * blk_make_request - given a bio, allocate a corresponding struct request.
1313 * @q: target request queue
1314 * @bio: The bio describing the memory mappings that will be submitted for IO.
1315 * It may be a chained-bio properly constructed by block/bio layer.
1316 * @gfp_mask: gfp flags to be used for memory allocation
1318 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1319 * type commands. Where the struct request needs to be farther initialized by
1320 * the caller. It is passed a &struct bio, which describes the memory info of
1323 * The caller of blk_make_request must make sure that bi_io_vec
1324 * are set to describe the memory buffers. That bio_data_dir() will return
1325 * the needed direction of the request. (And all bio's in the passed bio-chain
1326 * are properly set accordingly)
1328 * If called under none-sleepable conditions, mapped bio buffers must not
1329 * need bouncing, by calling the appropriate masked or flagged allocator,
1330 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1333 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1334 * given to how you allocate bios. In particular, you cannot use
1335 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1336 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1337 * thus resulting in a deadlock. Alternatively bios should be allocated using
1338 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1339 * If possible a big IO should be split into smaller parts when allocation
1340 * fails. Partial allocation should not be an error, or you risk a live-lock.
1342 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1345 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1350 blk_rq_set_block_pc(rq);
1353 struct bio *bounce_bio = bio;
1356 blk_queue_bounce(q, &bounce_bio);
1357 ret = blk_rq_append_bio(q, rq, bounce_bio);
1358 if (unlikely(ret)) {
1359 blk_put_request(rq);
1360 return ERR_PTR(ret);
1366 EXPORT_SYMBOL(blk_make_request);
1369 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1370 * @rq: request to be initialized
1373 void blk_rq_set_block_pc(struct request *rq)
1375 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1377 rq->__sector = (sector_t) -1;
1378 rq->bio = rq->biotail = NULL;
1379 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1381 EXPORT_SYMBOL(blk_rq_set_block_pc);
1384 * blk_requeue_request - put a request back on queue
1385 * @q: request queue where request should be inserted
1386 * @rq: request to be inserted
1389 * Drivers often keep queueing requests until the hardware cannot accept
1390 * more, when that condition happens we need to put the request back
1391 * on the queue. Must be called with queue lock held.
1393 void blk_requeue_request(struct request_queue *q, struct request *rq)
1395 blk_delete_timer(rq);
1396 blk_clear_rq_complete(rq);
1397 trace_block_rq_requeue(q, rq);
1399 if (rq->cmd_flags & REQ_QUEUED)
1400 blk_queue_end_tag(q, rq);
1402 BUG_ON(blk_queued_rq(rq));
1404 elv_requeue_request(q, rq);
1406 EXPORT_SYMBOL(blk_requeue_request);
1408 static void add_acct_request(struct request_queue *q, struct request *rq,
1411 blk_account_io_start(rq, true);
1412 __elv_add_request(q, rq, where);
1415 static void part_round_stats_single(int cpu, struct hd_struct *part,
1420 if (now == part->stamp)
1423 inflight = part_in_flight(part);
1425 __part_stat_add(cpu, part, time_in_queue,
1426 inflight * (now - part->stamp));
1427 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1433 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1434 * @cpu: cpu number for stats access
1435 * @part: target partition
1437 * The average IO queue length and utilisation statistics are maintained
1438 * by observing the current state of the queue length and the amount of
1439 * time it has been in this state for.
1441 * Normally, that accounting is done on IO completion, but that can result
1442 * in more than a second's worth of IO being accounted for within any one
1443 * second, leading to >100% utilisation. To deal with that, we call this
1444 * function to do a round-off before returning the results when reading
1445 * /proc/diskstats. This accounts immediately for all queue usage up to
1446 * the current jiffies and restarts the counters again.
1448 void part_round_stats(int cpu, struct hd_struct *part)
1450 unsigned long now = jiffies;
1453 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1454 part_round_stats_single(cpu, part, now);
1456 EXPORT_SYMBOL_GPL(part_round_stats);
1459 static void blk_pm_put_request(struct request *rq)
1461 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1462 pm_runtime_mark_last_busy(rq->q->dev);
1465 static inline void blk_pm_put_request(struct request *rq) {}
1469 * queue lock must be held
1471 void __blk_put_request(struct request_queue *q, struct request *req)
1477 blk_mq_free_request(req);
1481 blk_pm_put_request(req);
1483 elv_completed_request(q, req);
1485 /* this is a bio leak */
1486 WARN_ON(req->bio != NULL);
1489 * Request may not have originated from ll_rw_blk. if not,
1490 * it didn't come out of our reserved rq pools
1492 if (req->cmd_flags & REQ_ALLOCED) {
1493 unsigned int flags = req->cmd_flags;
1494 struct request_list *rl = blk_rq_rl(req);
1496 BUG_ON(!list_empty(&req->queuelist));
1497 BUG_ON(ELV_ON_HASH(req));
1499 blk_free_request(rl, req);
1500 freed_request(rl, flags);
1504 EXPORT_SYMBOL_GPL(__blk_put_request);
1506 void blk_put_request(struct request *req)
1508 struct request_queue *q = req->q;
1511 blk_mq_free_request(req);
1513 unsigned long flags;
1515 spin_lock_irqsave(q->queue_lock, flags);
1516 __blk_put_request(q, req);
1517 spin_unlock_irqrestore(q->queue_lock, flags);
1520 EXPORT_SYMBOL(blk_put_request);
1523 * blk_add_request_payload - add a payload to a request
1524 * @rq: request to update
1525 * @page: page backing the payload
1526 * @offset: offset in page
1527 * @len: length of the payload.
1529 * This allows to later add a payload to an already submitted request by
1530 * a block driver. The driver needs to take care of freeing the payload
1533 * Note that this is a quite horrible hack and nothing but handling of
1534 * discard requests should ever use it.
1536 void blk_add_request_payload(struct request *rq, struct page *page,
1537 int offset, unsigned int len)
1539 struct bio *bio = rq->bio;
1541 bio->bi_io_vec->bv_page = page;
1542 bio->bi_io_vec->bv_offset = offset;
1543 bio->bi_io_vec->bv_len = len;
1545 bio->bi_iter.bi_size = len;
1547 bio->bi_phys_segments = 1;
1549 rq->__data_len = rq->resid_len = len;
1550 rq->nr_phys_segments = 1;
1552 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1554 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1557 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1559 if (!ll_back_merge_fn(q, req, bio))
1562 trace_block_bio_backmerge(q, req, bio);
1564 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1565 blk_rq_set_mixed_merge(req);
1567 req->biotail->bi_next = bio;
1569 req->__data_len += bio->bi_iter.bi_size;
1570 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1572 blk_account_io_start(req, false);
1576 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1579 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1581 if (!ll_front_merge_fn(q, req, bio))
1584 trace_block_bio_frontmerge(q, req, bio);
1586 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1587 blk_rq_set_mixed_merge(req);
1589 bio->bi_next = req->bio;
1592 req->__sector = bio->bi_iter.bi_sector;
1593 req->__data_len += bio->bi_iter.bi_size;
1594 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1596 blk_account_io_start(req, false);
1601 * blk_attempt_plug_merge - try to merge with %current's plugged list
1602 * @q: request_queue new bio is being queued at
1603 * @bio: new bio being queued
1604 * @request_count: out parameter for number of traversed plugged requests
1605 * @same_queue_rq: pointer to &struct request that gets filled in when
1606 * another request associated with @q is found on the plug list
1607 * (optional, may be %NULL)
1609 * Determine whether @bio being queued on @q can be merged with a request
1610 * on %current's plugged list. Returns %true if merge was successful,
1613 * Plugging coalesces IOs from the same issuer for the same purpose without
1614 * going through @q->queue_lock. As such it's more of an issuing mechanism
1615 * than scheduling, and the request, while may have elvpriv data, is not
1616 * added on the elevator at this point. In addition, we don't have
1617 * reliable access to the elevator outside queue lock. Only check basic
1618 * merging parameters without querying the elevator.
1620 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1622 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1623 unsigned int *request_count,
1624 struct request **same_queue_rq)
1626 struct blk_plug *plug;
1629 struct list_head *plug_list;
1631 plug = current->plug;
1637 plug_list = &plug->mq_list;
1639 plug_list = &plug->list;
1641 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1647 * Only blk-mq multiple hardware queues case checks the
1648 * rq in the same queue, there should be only one such
1652 *same_queue_rq = rq;
1655 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1658 el_ret = blk_try_merge(rq, bio);
1659 if (el_ret == ELEVATOR_BACK_MERGE) {
1660 ret = bio_attempt_back_merge(q, rq, bio);
1663 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1664 ret = bio_attempt_front_merge(q, rq, bio);
1673 unsigned int blk_plug_queued_count(struct request_queue *q)
1675 struct blk_plug *plug;
1677 struct list_head *plug_list;
1678 unsigned int ret = 0;
1680 plug = current->plug;
1685 plug_list = &plug->mq_list;
1687 plug_list = &plug->list;
1689 list_for_each_entry(rq, plug_list, queuelist) {
1697 void init_request_from_bio(struct request *req, struct bio *bio)
1699 req->cmd_type = REQ_TYPE_FS;
1701 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1702 if (bio->bi_rw & REQ_RAHEAD)
1703 req->cmd_flags |= REQ_FAILFAST_MASK;
1706 req->__sector = bio->bi_iter.bi_sector;
1707 req->ioprio = bio_prio(bio);
1708 blk_rq_bio_prep(req->q, req, bio);
1711 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1713 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1714 struct blk_plug *plug;
1715 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1716 struct request *req;
1717 unsigned int request_count = 0;
1720 * low level driver can indicate that it wants pages above a
1721 * certain limit bounced to low memory (ie for highmem, or even
1722 * ISA dma in theory)
1724 blk_queue_bounce(q, &bio);
1726 blk_queue_split(q, &bio, q->bio_split);
1728 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1729 bio->bi_error = -EIO;
1731 return BLK_QC_T_NONE;
1734 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1735 spin_lock_irq(q->queue_lock);
1736 where = ELEVATOR_INSERT_FLUSH;
1741 * Check if we can merge with the plugged list before grabbing
1744 if (!blk_queue_nomerges(q)) {
1745 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1746 return BLK_QC_T_NONE;
1748 request_count = blk_plug_queued_count(q);
1750 spin_lock_irq(q->queue_lock);
1752 el_ret = elv_merge(q, &req, bio);
1753 if (el_ret == ELEVATOR_BACK_MERGE) {
1754 if (bio_attempt_back_merge(q, req, bio)) {
1755 elv_bio_merged(q, req, bio);
1756 if (!attempt_back_merge(q, req))
1757 elv_merged_request(q, req, el_ret);
1760 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1761 if (bio_attempt_front_merge(q, req, bio)) {
1762 elv_bio_merged(q, req, bio);
1763 if (!attempt_front_merge(q, req))
1764 elv_merged_request(q, req, el_ret);
1771 * This sync check and mask will be re-done in init_request_from_bio(),
1772 * but we need to set it earlier to expose the sync flag to the
1773 * rq allocator and io schedulers.
1775 rw_flags = bio_data_dir(bio);
1777 rw_flags |= REQ_SYNC;
1780 * Grab a free request. This is might sleep but can not fail.
1781 * Returns with the queue unlocked.
1783 req = get_request(q, rw_flags, bio, GFP_NOIO);
1785 bio->bi_error = PTR_ERR(req);
1791 * After dropping the lock and possibly sleeping here, our request
1792 * may now be mergeable after it had proven unmergeable (above).
1793 * We don't worry about that case for efficiency. It won't happen
1794 * often, and the elevators are able to handle it.
1796 init_request_from_bio(req, bio);
1798 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1799 req->cpu = raw_smp_processor_id();
1801 plug = current->plug;
1804 * If this is the first request added after a plug, fire
1808 trace_block_plug(q);
1810 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1811 blk_flush_plug_list(plug, false);
1812 trace_block_plug(q);
1815 list_add_tail(&req->queuelist, &plug->list);
1816 blk_account_io_start(req, true);
1818 spin_lock_irq(q->queue_lock);
1819 add_acct_request(q, req, where);
1822 spin_unlock_irq(q->queue_lock);
1825 return BLK_QC_T_NONE;
1829 * If bio->bi_dev is a partition, remap the location
1831 static inline void blk_partition_remap(struct bio *bio)
1833 struct block_device *bdev = bio->bi_bdev;
1835 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1836 struct hd_struct *p = bdev->bd_part;
1838 bio->bi_iter.bi_sector += p->start_sect;
1839 bio->bi_bdev = bdev->bd_contains;
1841 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1843 bio->bi_iter.bi_sector - p->start_sect);
1847 static void handle_bad_sector(struct bio *bio)
1849 char b[BDEVNAME_SIZE];
1851 printk(KERN_INFO "attempt to access beyond end of device\n");
1852 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1853 bdevname(bio->bi_bdev, b),
1855 (unsigned long long)bio_end_sector(bio),
1856 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1859 #ifdef CONFIG_FAIL_MAKE_REQUEST
1861 static DECLARE_FAULT_ATTR(fail_make_request);
1863 static int __init setup_fail_make_request(char *str)
1865 return setup_fault_attr(&fail_make_request, str);
1867 __setup("fail_make_request=", setup_fail_make_request);
1869 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1871 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1874 static int __init fail_make_request_debugfs(void)
1876 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1877 NULL, &fail_make_request);
1879 return PTR_ERR_OR_ZERO(dir);
1882 late_initcall(fail_make_request_debugfs);
1884 #else /* CONFIG_FAIL_MAKE_REQUEST */
1886 static inline bool should_fail_request(struct hd_struct *part,
1892 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1895 * Check whether this bio extends beyond the end of the device.
1897 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1904 /* Test device or partition size, when known. */
1905 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1907 sector_t sector = bio->bi_iter.bi_sector;
1909 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1911 * This may well happen - the kernel calls bread()
1912 * without checking the size of the device, e.g., when
1913 * mounting a device.
1915 handle_bad_sector(bio);
1923 static noinline_for_stack bool
1924 generic_make_request_checks(struct bio *bio)
1926 struct request_queue *q;
1927 int nr_sectors = bio_sectors(bio);
1929 char b[BDEVNAME_SIZE];
1930 struct hd_struct *part;
1934 if (bio_check_eod(bio, nr_sectors))
1937 q = bdev_get_queue(bio->bi_bdev);
1940 "generic_make_request: Trying to access "
1941 "nonexistent block-device %s (%Lu)\n",
1942 bdevname(bio->bi_bdev, b),
1943 (long long) bio->bi_iter.bi_sector);
1947 part = bio->bi_bdev->bd_part;
1948 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1949 should_fail_request(&part_to_disk(part)->part0,
1950 bio->bi_iter.bi_size))
1954 * If this device has partitions, remap block n
1955 * of partition p to block n+start(p) of the disk.
1957 blk_partition_remap(bio);
1959 if (bio_check_eod(bio, nr_sectors))
1963 * Filter flush bio's early so that make_request based
1964 * drivers without flush support don't have to worry
1967 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1968 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1975 if ((bio->bi_rw & REQ_DISCARD) &&
1976 (!blk_queue_discard(q) ||
1977 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1982 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1988 * Various block parts want %current->io_context and lazy ioc
1989 * allocation ends up trading a lot of pain for a small amount of
1990 * memory. Just allocate it upfront. This may fail and block
1991 * layer knows how to live with it.
1993 create_io_context(GFP_ATOMIC, q->node);
1995 if (!blkcg_bio_issue_check(q, bio))
1998 trace_block_bio_queue(q, bio);
2002 bio->bi_error = err;
2008 * generic_make_request - hand a buffer to its device driver for I/O
2009 * @bio: The bio describing the location in memory and on the device.
2011 * generic_make_request() is used to make I/O requests of block
2012 * devices. It is passed a &struct bio, which describes the I/O that needs
2015 * generic_make_request() does not return any status. The
2016 * success/failure status of the request, along with notification of
2017 * completion, is delivered asynchronously through the bio->bi_end_io
2018 * function described (one day) else where.
2020 * The caller of generic_make_request must make sure that bi_io_vec
2021 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2022 * set to describe the device address, and the
2023 * bi_end_io and optionally bi_private are set to describe how
2024 * completion notification should be signaled.
2026 * generic_make_request and the drivers it calls may use bi_next if this
2027 * bio happens to be merged with someone else, and may resubmit the bio to
2028 * a lower device by calling into generic_make_request recursively, which
2029 * means the bio should NOT be touched after the call to ->make_request_fn.
2031 blk_qc_t generic_make_request(struct bio *bio)
2033 struct bio_list bio_list_on_stack;
2034 blk_qc_t ret = BLK_QC_T_NONE;
2036 if (!generic_make_request_checks(bio))
2040 * We only want one ->make_request_fn to be active at a time, else
2041 * stack usage with stacked devices could be a problem. So use
2042 * current->bio_list to keep a list of requests submited by a
2043 * make_request_fn function. current->bio_list is also used as a
2044 * flag to say if generic_make_request is currently active in this
2045 * task or not. If it is NULL, then no make_request is active. If
2046 * it is non-NULL, then a make_request is active, and new requests
2047 * should be added at the tail
2049 if (current->bio_list) {
2050 bio_list_add(current->bio_list, bio);
2054 /* following loop may be a bit non-obvious, and so deserves some
2056 * Before entering the loop, bio->bi_next is NULL (as all callers
2057 * ensure that) so we have a list with a single bio.
2058 * We pretend that we have just taken it off a longer list, so
2059 * we assign bio_list to a pointer to the bio_list_on_stack,
2060 * thus initialising the bio_list of new bios to be
2061 * added. ->make_request() may indeed add some more bios
2062 * through a recursive call to generic_make_request. If it
2063 * did, we find a non-NULL value in bio_list and re-enter the loop
2064 * from the top. In this case we really did just take the bio
2065 * of the top of the list (no pretending) and so remove it from
2066 * bio_list, and call into ->make_request() again.
2068 BUG_ON(bio->bi_next);
2069 bio_list_init(&bio_list_on_stack);
2070 current->bio_list = &bio_list_on_stack;
2072 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2074 if (likely(blk_queue_enter(q, false) == 0)) {
2075 ret = q->make_request_fn(q, bio);
2079 bio = bio_list_pop(current->bio_list);
2081 struct bio *bio_next = bio_list_pop(current->bio_list);
2087 current->bio_list = NULL; /* deactivate */
2092 EXPORT_SYMBOL(generic_make_request);
2095 * submit_bio - submit a bio to the block device layer for I/O
2096 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2097 * @bio: The &struct bio which describes the I/O
2099 * submit_bio() is very similar in purpose to generic_make_request(), and
2100 * uses that function to do most of the work. Both are fairly rough
2101 * interfaces; @bio must be presetup and ready for I/O.
2104 blk_qc_t submit_bio(int rw, struct bio *bio)
2109 * If it's a regular read/write or a barrier with data attached,
2110 * go through the normal accounting stuff before submission.
2112 if (bio_has_data(bio)) {
2115 if (unlikely(rw & REQ_WRITE_SAME))
2116 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2118 count = bio_sectors(bio);
2121 count_vm_events(PGPGOUT, count);
2123 task_io_account_read(bio->bi_iter.bi_size);
2124 count_vm_events(PGPGIN, count);
2127 if (unlikely(block_dump)) {
2128 char b[BDEVNAME_SIZE];
2129 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2130 current->comm, task_pid_nr(current),
2131 (rw & WRITE) ? "WRITE" : "READ",
2132 (unsigned long long)bio->bi_iter.bi_sector,
2133 bdevname(bio->bi_bdev, b),
2138 return generic_make_request(bio);
2140 EXPORT_SYMBOL(submit_bio);
2143 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2144 * for new the queue limits
2146 * @rq: the request being checked
2149 * @rq may have been made based on weaker limitations of upper-level queues
2150 * in request stacking drivers, and it may violate the limitation of @q.
2151 * Since the block layer and the underlying device driver trust @rq
2152 * after it is inserted to @q, it should be checked against @q before
2153 * the insertion using this generic function.
2155 * Request stacking drivers like request-based dm may change the queue
2156 * limits when retrying requests on other queues. Those requests need
2157 * to be checked against the new queue limits again during dispatch.
2159 static int blk_cloned_rq_check_limits(struct request_queue *q,
2162 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2163 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2168 * queue's settings related to segment counting like q->bounce_pfn
2169 * may differ from that of other stacking queues.
2170 * Recalculate it to check the request correctly on this queue's
2173 blk_recalc_rq_segments(rq);
2174 if (rq->nr_phys_segments > queue_max_segments(q)) {
2175 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2183 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2184 * @q: the queue to submit the request
2185 * @rq: the request being queued
2187 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2189 unsigned long flags;
2190 int where = ELEVATOR_INSERT_BACK;
2192 if (blk_cloned_rq_check_limits(q, rq))
2196 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2200 if (blk_queue_io_stat(q))
2201 blk_account_io_start(rq, true);
2202 blk_mq_insert_request(rq, false, true, false);
2206 spin_lock_irqsave(q->queue_lock, flags);
2207 if (unlikely(blk_queue_dying(q))) {
2208 spin_unlock_irqrestore(q->queue_lock, flags);
2213 * Submitting request must be dequeued before calling this function
2214 * because it will be linked to another request_queue
2216 BUG_ON(blk_queued_rq(rq));
2218 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2219 where = ELEVATOR_INSERT_FLUSH;
2221 add_acct_request(q, rq, where);
2222 if (where == ELEVATOR_INSERT_FLUSH)
2224 spin_unlock_irqrestore(q->queue_lock, flags);
2228 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2231 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2232 * @rq: request to examine
2235 * A request could be merge of IOs which require different failure
2236 * handling. This function determines the number of bytes which
2237 * can be failed from the beginning of the request without
2238 * crossing into area which need to be retried further.
2241 * The number of bytes to fail.
2244 * queue_lock must be held.
2246 unsigned int blk_rq_err_bytes(const struct request *rq)
2248 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2249 unsigned int bytes = 0;
2252 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2253 return blk_rq_bytes(rq);
2256 * Currently the only 'mixing' which can happen is between
2257 * different fastfail types. We can safely fail portions
2258 * which have all the failfast bits that the first one has -
2259 * the ones which are at least as eager to fail as the first
2262 for (bio = rq->bio; bio; bio = bio->bi_next) {
2263 if ((bio->bi_rw & ff) != ff)
2265 bytes += bio->bi_iter.bi_size;
2268 /* this could lead to infinite loop */
2269 BUG_ON(blk_rq_bytes(rq) && !bytes);
2272 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2274 void blk_account_io_completion(struct request *req, unsigned int bytes)
2276 if (blk_do_io_stat(req)) {
2277 const int rw = rq_data_dir(req);
2278 struct hd_struct *part;
2281 cpu = part_stat_lock();
2283 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2288 void blk_account_io_done(struct request *req)
2291 * Account IO completion. flush_rq isn't accounted as a
2292 * normal IO on queueing nor completion. Accounting the
2293 * containing request is enough.
2295 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2296 unsigned long duration = jiffies - req->start_time;
2297 const int rw = rq_data_dir(req);
2298 struct hd_struct *part;
2301 cpu = part_stat_lock();
2304 part_stat_inc(cpu, part, ios[rw]);
2305 part_stat_add(cpu, part, ticks[rw], duration);
2306 part_round_stats(cpu, part);
2307 part_dec_in_flight(part, rw);
2309 hd_struct_put(part);
2316 * Don't process normal requests when queue is suspended
2317 * or in the process of suspending/resuming
2319 static struct request *blk_pm_peek_request(struct request_queue *q,
2322 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2323 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2329 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2336 void blk_account_io_start(struct request *rq, bool new_io)
2338 struct hd_struct *part;
2339 int rw = rq_data_dir(rq);
2342 if (!blk_do_io_stat(rq))
2345 cpu = part_stat_lock();
2349 part_stat_inc(cpu, part, merges[rw]);
2351 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2352 if (!hd_struct_try_get(part)) {
2354 * The partition is already being removed,
2355 * the request will be accounted on the disk only
2357 * We take a reference on disk->part0 although that
2358 * partition will never be deleted, so we can treat
2359 * it as any other partition.
2361 part = &rq->rq_disk->part0;
2362 hd_struct_get(part);
2364 part_round_stats(cpu, part);
2365 part_inc_in_flight(part, rw);
2373 * blk_peek_request - peek at the top of a request queue
2374 * @q: request queue to peek at
2377 * Return the request at the top of @q. The returned request
2378 * should be started using blk_start_request() before LLD starts
2382 * Pointer to the request at the top of @q if available. Null
2386 * queue_lock must be held.
2388 struct request *blk_peek_request(struct request_queue *q)
2393 while ((rq = __elv_next_request(q)) != NULL) {
2395 rq = blk_pm_peek_request(q, rq);
2399 if (!(rq->cmd_flags & REQ_STARTED)) {
2401 * This is the first time the device driver
2402 * sees this request (possibly after
2403 * requeueing). Notify IO scheduler.
2405 if (rq->cmd_flags & REQ_SORTED)
2406 elv_activate_rq(q, rq);
2409 * just mark as started even if we don't start
2410 * it, a request that has been delayed should
2411 * not be passed by new incoming requests
2413 rq->cmd_flags |= REQ_STARTED;
2414 trace_block_rq_issue(q, rq);
2417 if (!q->boundary_rq || q->boundary_rq == rq) {
2418 q->end_sector = rq_end_sector(rq);
2419 q->boundary_rq = NULL;
2422 if (rq->cmd_flags & REQ_DONTPREP)
2425 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2427 * make sure space for the drain appears we
2428 * know we can do this because max_hw_segments
2429 * has been adjusted to be one fewer than the
2432 rq->nr_phys_segments++;
2438 ret = q->prep_rq_fn(q, rq);
2439 if (ret == BLKPREP_OK) {
2441 } else if (ret == BLKPREP_DEFER) {
2443 * the request may have been (partially) prepped.
2444 * we need to keep this request in the front to
2445 * avoid resource deadlock. REQ_STARTED will
2446 * prevent other fs requests from passing this one.
2448 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2449 !(rq->cmd_flags & REQ_DONTPREP)) {
2451 * remove the space for the drain we added
2452 * so that we don't add it again
2454 --rq->nr_phys_segments;
2459 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2460 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2462 rq->cmd_flags |= REQ_QUIET;
2464 * Mark this request as started so we don't trigger
2465 * any debug logic in the end I/O path.
2467 blk_start_request(rq);
2468 __blk_end_request_all(rq, err);
2470 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2477 EXPORT_SYMBOL(blk_peek_request);
2479 void blk_dequeue_request(struct request *rq)
2481 struct request_queue *q = rq->q;
2483 BUG_ON(list_empty(&rq->queuelist));
2484 BUG_ON(ELV_ON_HASH(rq));
2486 list_del_init(&rq->queuelist);
2489 * the time frame between a request being removed from the lists
2490 * and to it is freed is accounted as io that is in progress at
2493 if (blk_account_rq(rq)) {
2494 q->in_flight[rq_is_sync(rq)]++;
2495 set_io_start_time_ns(rq);
2500 * blk_start_request - start request processing on the driver
2501 * @req: request to dequeue
2504 * Dequeue @req and start timeout timer on it. This hands off the
2505 * request to the driver.
2507 * Block internal functions which don't want to start timer should
2508 * call blk_dequeue_request().
2511 * queue_lock must be held.
2513 void blk_start_request(struct request *req)
2515 blk_dequeue_request(req);
2518 * We are now handing the request to the hardware, initialize
2519 * resid_len to full count and add the timeout handler.
2521 req->resid_len = blk_rq_bytes(req);
2522 if (unlikely(blk_bidi_rq(req)))
2523 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2525 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2528 EXPORT_SYMBOL(blk_start_request);
2531 * blk_fetch_request - fetch a request from a request queue
2532 * @q: request queue to fetch a request from
2535 * Return the request at the top of @q. The request is started on
2536 * return and LLD can start processing it immediately.
2539 * Pointer to the request at the top of @q if available. Null
2543 * queue_lock must be held.
2545 struct request *blk_fetch_request(struct request_queue *q)
2549 rq = blk_peek_request(q);
2551 blk_start_request(rq);
2554 EXPORT_SYMBOL(blk_fetch_request);
2557 * blk_update_request - Special helper function for request stacking drivers
2558 * @req: the request being processed
2559 * @error: %0 for success, < %0 for error
2560 * @nr_bytes: number of bytes to complete @req
2563 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2564 * the request structure even if @req doesn't have leftover.
2565 * If @req has leftover, sets it up for the next range of segments.
2567 * This special helper function is only for request stacking drivers
2568 * (e.g. request-based dm) so that they can handle partial completion.
2569 * Actual device drivers should use blk_end_request instead.
2571 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2572 * %false return from this function.
2575 * %false - this request doesn't have any more data
2576 * %true - this request has more data
2578 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2582 trace_block_rq_complete(req->q, req, nr_bytes);
2588 * For fs requests, rq is just carrier of independent bio's
2589 * and each partial completion should be handled separately.
2590 * Reset per-request error on each partial completion.
2592 * TODO: tj: This is too subtle. It would be better to let
2593 * low level drivers do what they see fit.
2595 if (req->cmd_type == REQ_TYPE_FS)
2598 if (error && req->cmd_type == REQ_TYPE_FS &&
2599 !(req->cmd_flags & REQ_QUIET)) {
2604 error_type = "recoverable transport";
2607 error_type = "critical target";
2610 error_type = "critical nexus";
2613 error_type = "timeout";
2616 error_type = "critical space allocation";
2619 error_type = "critical medium";
2626 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2627 __func__, error_type, req->rq_disk ?
2628 req->rq_disk->disk_name : "?",
2629 (unsigned long long)blk_rq_pos(req));
2633 blk_account_io_completion(req, nr_bytes);
2637 struct bio *bio = req->bio;
2638 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2640 if (bio_bytes == bio->bi_iter.bi_size)
2641 req->bio = bio->bi_next;
2643 req_bio_endio(req, bio, bio_bytes, error);
2645 total_bytes += bio_bytes;
2646 nr_bytes -= bio_bytes;
2657 * Reset counters so that the request stacking driver
2658 * can find how many bytes remain in the request
2661 req->__data_len = 0;
2665 req->__data_len -= total_bytes;
2667 /* update sector only for requests with clear definition of sector */
2668 if (req->cmd_type == REQ_TYPE_FS)
2669 req->__sector += total_bytes >> 9;
2671 /* mixed attributes always follow the first bio */
2672 if (req->cmd_flags & REQ_MIXED_MERGE) {
2673 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2674 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2678 * If total number of sectors is less than the first segment
2679 * size, something has gone terribly wrong.
2681 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2682 blk_dump_rq_flags(req, "request botched");
2683 req->__data_len = blk_rq_cur_bytes(req);
2686 /* recalculate the number of segments */
2687 blk_recalc_rq_segments(req);
2691 EXPORT_SYMBOL_GPL(blk_update_request);
2693 static bool blk_update_bidi_request(struct request *rq, int error,
2694 unsigned int nr_bytes,
2695 unsigned int bidi_bytes)
2697 if (blk_update_request(rq, error, nr_bytes))
2700 /* Bidi request must be completed as a whole */
2701 if (unlikely(blk_bidi_rq(rq)) &&
2702 blk_update_request(rq->next_rq, error, bidi_bytes))
2705 if (blk_queue_add_random(rq->q))
2706 add_disk_randomness(rq->rq_disk);
2712 * blk_unprep_request - unprepare a request
2715 * This function makes a request ready for complete resubmission (or
2716 * completion). It happens only after all error handling is complete,
2717 * so represents the appropriate moment to deallocate any resources
2718 * that were allocated to the request in the prep_rq_fn. The queue
2719 * lock is held when calling this.
2721 void blk_unprep_request(struct request *req)
2723 struct request_queue *q = req->q;
2725 req->cmd_flags &= ~REQ_DONTPREP;
2726 if (q->unprep_rq_fn)
2727 q->unprep_rq_fn(q, req);
2729 EXPORT_SYMBOL_GPL(blk_unprep_request);
2732 * queue lock must be held
2734 void blk_finish_request(struct request *req, int error)
2736 if (req->cmd_flags & REQ_QUEUED)
2737 blk_queue_end_tag(req->q, req);
2739 BUG_ON(blk_queued_rq(req));
2741 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2742 laptop_io_completion(&req->q->backing_dev_info);
2744 blk_delete_timer(req);
2746 if (req->cmd_flags & REQ_DONTPREP)
2747 blk_unprep_request(req);
2749 blk_account_io_done(req);
2752 req->end_io(req, error);
2754 if (blk_bidi_rq(req))
2755 __blk_put_request(req->next_rq->q, req->next_rq);
2757 __blk_put_request(req->q, req);
2760 EXPORT_SYMBOL(blk_finish_request);
2763 * blk_end_bidi_request - Complete a bidi request
2764 * @rq: the request to complete
2765 * @error: %0 for success, < %0 for error
2766 * @nr_bytes: number of bytes to complete @rq
2767 * @bidi_bytes: number of bytes to complete @rq->next_rq
2770 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2771 * Drivers that supports bidi can safely call this member for any
2772 * type of request, bidi or uni. In the later case @bidi_bytes is
2776 * %false - we are done with this request
2777 * %true - still buffers pending for this request
2779 static bool blk_end_bidi_request(struct request *rq, int error,
2780 unsigned int nr_bytes, unsigned int bidi_bytes)
2782 struct request_queue *q = rq->q;
2783 unsigned long flags;
2785 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2788 spin_lock_irqsave(q->queue_lock, flags);
2789 blk_finish_request(rq, error);
2790 spin_unlock_irqrestore(q->queue_lock, flags);
2796 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2797 * @rq: the request to complete
2798 * @error: %0 for success, < %0 for error
2799 * @nr_bytes: number of bytes to complete @rq
2800 * @bidi_bytes: number of bytes to complete @rq->next_rq
2803 * Identical to blk_end_bidi_request() except that queue lock is
2804 * assumed to be locked on entry and remains so on return.
2807 * %false - we are done with this request
2808 * %true - still buffers pending for this request
2810 bool __blk_end_bidi_request(struct request *rq, int error,
2811 unsigned int nr_bytes, unsigned int bidi_bytes)
2813 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2816 blk_finish_request(rq, error);
2822 * blk_end_request - Helper function for drivers to complete the request.
2823 * @rq: the request being processed
2824 * @error: %0 for success, < %0 for error
2825 * @nr_bytes: number of bytes to complete
2828 * Ends I/O on a number of bytes attached to @rq.
2829 * If @rq has leftover, sets it up for the next range of segments.
2832 * %false - we are done with this request
2833 * %true - still buffers pending for this request
2835 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2837 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2839 EXPORT_SYMBOL(blk_end_request);
2842 * blk_end_request_all - Helper function for drives to finish the request.
2843 * @rq: the request to finish
2844 * @error: %0 for success, < %0 for error
2847 * Completely finish @rq.
2849 void blk_end_request_all(struct request *rq, int error)
2852 unsigned int bidi_bytes = 0;
2854 if (unlikely(blk_bidi_rq(rq)))
2855 bidi_bytes = blk_rq_bytes(rq->next_rq);
2857 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2860 EXPORT_SYMBOL(blk_end_request_all);
2863 * blk_end_request_cur - Helper function to finish the current request chunk.
2864 * @rq: the request to finish the current chunk for
2865 * @error: %0 for success, < %0 for error
2868 * Complete the current consecutively mapped chunk from @rq.
2871 * %false - we are done with this request
2872 * %true - still buffers pending for this request
2874 bool blk_end_request_cur(struct request *rq, int error)
2876 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2878 EXPORT_SYMBOL(blk_end_request_cur);
2881 * blk_end_request_err - Finish a request till the next failure boundary.
2882 * @rq: the request to finish till the next failure boundary for
2883 * @error: must be negative errno
2886 * Complete @rq till the next failure boundary.
2889 * %false - we are done with this request
2890 * %true - still buffers pending for this request
2892 bool blk_end_request_err(struct request *rq, int error)
2894 WARN_ON(error >= 0);
2895 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2897 EXPORT_SYMBOL_GPL(blk_end_request_err);
2900 * __blk_end_request - Helper function for drivers to complete the request.
2901 * @rq: the request being processed
2902 * @error: %0 for success, < %0 for error
2903 * @nr_bytes: number of bytes to complete
2906 * Must be called with queue lock held unlike blk_end_request().
2909 * %false - we are done with this request
2910 * %true - still buffers pending for this request
2912 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2914 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2916 EXPORT_SYMBOL(__blk_end_request);
2919 * __blk_end_request_all - Helper function for drives to finish the request.
2920 * @rq: the request to finish
2921 * @error: %0 for success, < %0 for error
2924 * Completely finish @rq. Must be called with queue lock held.
2926 void __blk_end_request_all(struct request *rq, int error)
2929 unsigned int bidi_bytes = 0;
2931 if (unlikely(blk_bidi_rq(rq)))
2932 bidi_bytes = blk_rq_bytes(rq->next_rq);
2934 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2937 EXPORT_SYMBOL(__blk_end_request_all);
2940 * __blk_end_request_cur - Helper function to finish the current request chunk.
2941 * @rq: the request to finish the current chunk for
2942 * @error: %0 for success, < %0 for error
2945 * Complete the current consecutively mapped chunk from @rq. Must
2946 * be called with queue lock held.
2949 * %false - we are done with this request
2950 * %true - still buffers pending for this request
2952 bool __blk_end_request_cur(struct request *rq, int error)
2954 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2956 EXPORT_SYMBOL(__blk_end_request_cur);
2959 * __blk_end_request_err - Finish a request till the next failure boundary.
2960 * @rq: the request to finish till the next failure boundary for
2961 * @error: must be negative errno
2964 * Complete @rq till the next failure boundary. Must be called
2965 * with queue lock held.
2968 * %false - we are done with this request
2969 * %true - still buffers pending for this request
2971 bool __blk_end_request_err(struct request *rq, int error)
2973 WARN_ON(error >= 0);
2974 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2976 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2978 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2981 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2982 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2984 if (bio_has_data(bio))
2985 rq->nr_phys_segments = bio_phys_segments(q, bio);
2987 rq->__data_len = bio->bi_iter.bi_size;
2988 rq->bio = rq->biotail = bio;
2991 rq->rq_disk = bio->bi_bdev->bd_disk;
2994 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2996 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2997 * @rq: the request to be flushed
3000 * Flush all pages in @rq.
3002 void rq_flush_dcache_pages(struct request *rq)
3004 struct req_iterator iter;
3005 struct bio_vec bvec;
3007 rq_for_each_segment(bvec, rq, iter)
3008 flush_dcache_page(bvec.bv_page);
3010 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3014 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3015 * @q : the queue of the device being checked
3018 * Check if underlying low-level drivers of a device are busy.
3019 * If the drivers want to export their busy state, they must set own
3020 * exporting function using blk_queue_lld_busy() first.
3022 * Basically, this function is used only by request stacking drivers
3023 * to stop dispatching requests to underlying devices when underlying
3024 * devices are busy. This behavior helps more I/O merging on the queue
3025 * of the request stacking driver and prevents I/O throughput regression
3026 * on burst I/O load.
3029 * 0 - Not busy (The request stacking driver should dispatch request)
3030 * 1 - Busy (The request stacking driver should stop dispatching request)
3032 int blk_lld_busy(struct request_queue *q)
3035 return q->lld_busy_fn(q);
3039 EXPORT_SYMBOL_GPL(blk_lld_busy);
3042 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3043 * @rq: the clone request to be cleaned up
3046 * Free all bios in @rq for a cloned request.
3048 void blk_rq_unprep_clone(struct request *rq)
3052 while ((bio = rq->bio) != NULL) {
3053 rq->bio = bio->bi_next;
3058 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3061 * Copy attributes of the original request to the clone request.
3062 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3064 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3066 dst->cpu = src->cpu;
3067 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3068 dst->cmd_type = src->cmd_type;
3069 dst->__sector = blk_rq_pos(src);
3070 dst->__data_len = blk_rq_bytes(src);
3071 dst->nr_phys_segments = src->nr_phys_segments;
3072 dst->ioprio = src->ioprio;
3073 dst->extra_len = src->extra_len;
3077 * blk_rq_prep_clone - Helper function to setup clone request
3078 * @rq: the request to be setup
3079 * @rq_src: original request to be cloned
3080 * @bs: bio_set that bios for clone are allocated from
3081 * @gfp_mask: memory allocation mask for bio
3082 * @bio_ctr: setup function to be called for each clone bio.
3083 * Returns %0 for success, non %0 for failure.
3084 * @data: private data to be passed to @bio_ctr
3087 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3088 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3089 * are not copied, and copying such parts is the caller's responsibility.
3090 * Also, pages which the original bios are pointing to are not copied
3091 * and the cloned bios just point same pages.
3092 * So cloned bios must be completed before original bios, which means
3093 * the caller must complete @rq before @rq_src.
3095 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3096 struct bio_set *bs, gfp_t gfp_mask,
3097 int (*bio_ctr)(struct bio *, struct bio *, void *),
3100 struct bio *bio, *bio_src;
3105 __rq_for_each_bio(bio_src, rq_src) {
3106 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3110 if (bio_ctr && bio_ctr(bio, bio_src, data))
3114 rq->biotail->bi_next = bio;
3117 rq->bio = rq->biotail = bio;
3120 __blk_rq_prep_clone(rq, rq_src);
3127 blk_rq_unprep_clone(rq);
3131 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3133 int kblockd_schedule_work(struct work_struct *work)
3135 return queue_work(kblockd_workqueue, work);
3137 EXPORT_SYMBOL(kblockd_schedule_work);
3139 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3140 unsigned long delay)
3142 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3144 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3146 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3147 unsigned long delay)
3149 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3151 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3154 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3155 * @plug: The &struct blk_plug that needs to be initialized
3158 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3159 * pending I/O should the task end up blocking between blk_start_plug() and
3160 * blk_finish_plug(). This is important from a performance perspective, but
3161 * also ensures that we don't deadlock. For instance, if the task is blocking
3162 * for a memory allocation, memory reclaim could end up wanting to free a
3163 * page belonging to that request that is currently residing in our private
3164 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3165 * this kind of deadlock.
3167 void blk_start_plug(struct blk_plug *plug)
3169 struct task_struct *tsk = current;
3172 * If this is a nested plug, don't actually assign it.
3177 INIT_LIST_HEAD(&plug->list);
3178 INIT_LIST_HEAD(&plug->mq_list);
3179 INIT_LIST_HEAD(&plug->cb_list);
3181 * Store ordering should not be needed here, since a potential
3182 * preempt will imply a full memory barrier
3186 EXPORT_SYMBOL(blk_start_plug);
3188 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3190 struct request *rqa = container_of(a, struct request, queuelist);
3191 struct request *rqb = container_of(b, struct request, queuelist);
3193 return !(rqa->q < rqb->q ||
3194 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3198 * If 'from_schedule' is true, then postpone the dispatch of requests
3199 * until a safe kblockd context. We due this to avoid accidental big
3200 * additional stack usage in driver dispatch, in places where the originally
3201 * plugger did not intend it.
3203 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3205 __releases(q->queue_lock)
3207 trace_block_unplug(q, depth, !from_schedule);
3210 blk_run_queue_async(q);
3213 spin_unlock(q->queue_lock);
3216 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3218 LIST_HEAD(callbacks);
3220 while (!list_empty(&plug->cb_list)) {
3221 list_splice_init(&plug->cb_list, &callbacks);
3223 while (!list_empty(&callbacks)) {
3224 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3227 list_del(&cb->list);
3228 cb->callback(cb, from_schedule);
3233 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3236 struct blk_plug *plug = current->plug;
3237 struct blk_plug_cb *cb;
3242 list_for_each_entry(cb, &plug->cb_list, list)
3243 if (cb->callback == unplug && cb->data == data)
3246 /* Not currently on the callback list */
3247 BUG_ON(size < sizeof(*cb));
3248 cb = kzalloc(size, GFP_ATOMIC);
3251 cb->callback = unplug;
3252 list_add(&cb->list, &plug->cb_list);
3256 EXPORT_SYMBOL(blk_check_plugged);
3258 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3260 struct request_queue *q;
3261 unsigned long flags;
3266 flush_plug_callbacks(plug, from_schedule);
3268 if (!list_empty(&plug->mq_list))
3269 blk_mq_flush_plug_list(plug, from_schedule);
3271 if (list_empty(&plug->list))
3274 list_splice_init(&plug->list, &list);
3276 list_sort(NULL, &list, plug_rq_cmp);
3282 * Save and disable interrupts here, to avoid doing it for every
3283 * queue lock we have to take.
3285 local_irq_save(flags);
3286 while (!list_empty(&list)) {
3287 rq = list_entry_rq(list.next);
3288 list_del_init(&rq->queuelist);
3292 * This drops the queue lock
3295 queue_unplugged(q, depth, from_schedule);
3298 spin_lock(q->queue_lock);
3302 * Short-circuit if @q is dead
3304 if (unlikely(blk_queue_dying(q))) {
3305 __blk_end_request_all(rq, -ENODEV);
3310 * rq is already accounted, so use raw insert
3312 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3313 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3315 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3321 * This drops the queue lock
3324 queue_unplugged(q, depth, from_schedule);
3326 local_irq_restore(flags);
3329 void blk_finish_plug(struct blk_plug *plug)
3331 if (plug != current->plug)
3333 blk_flush_plug_list(plug, false);
3335 current->plug = NULL;
3337 EXPORT_SYMBOL(blk_finish_plug);
3339 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3341 struct blk_plug *plug;
3344 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3345 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3348 plug = current->plug;
3350 blk_flush_plug_list(plug, false);
3352 state = current->state;
3353 while (!need_resched()) {
3354 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3355 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3358 hctx->poll_invoked++;
3360 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3362 hctx->poll_success++;
3363 set_current_state(TASK_RUNNING);
3367 if (signal_pending_state(state, current))
3368 set_current_state(TASK_RUNNING);
3370 if (current->state == TASK_RUNNING)
3382 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3383 * @q: the queue of the device
3384 * @dev: the device the queue belongs to
3387 * Initialize runtime-PM-related fields for @q and start auto suspend for
3388 * @dev. Drivers that want to take advantage of request-based runtime PM
3389 * should call this function after @dev has been initialized, and its
3390 * request queue @q has been allocated, and runtime PM for it can not happen
3391 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3392 * cases, driver should call this function before any I/O has taken place.
3394 * This function takes care of setting up using auto suspend for the device,
3395 * the autosuspend delay is set to -1 to make runtime suspend impossible
3396 * until an updated value is either set by user or by driver. Drivers do
3397 * not need to touch other autosuspend settings.
3399 * The block layer runtime PM is request based, so only works for drivers
3400 * that use request as their IO unit instead of those directly use bio's.
3402 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3405 q->rpm_status = RPM_ACTIVE;
3406 pm_runtime_set_autosuspend_delay(q->dev, -1);
3407 pm_runtime_use_autosuspend(q->dev);
3409 EXPORT_SYMBOL(blk_pm_runtime_init);
3412 * blk_pre_runtime_suspend - Pre runtime suspend check
3413 * @q: the queue of the device
3416 * This function will check if runtime suspend is allowed for the device
3417 * by examining if there are any requests pending in the queue. If there
3418 * are requests pending, the device can not be runtime suspended; otherwise,
3419 * the queue's status will be updated to SUSPENDING and the driver can
3420 * proceed to suspend the device.
3422 * For the not allowed case, we mark last busy for the device so that
3423 * runtime PM core will try to autosuspend it some time later.
3425 * This function should be called near the start of the device's
3426 * runtime_suspend callback.
3429 * 0 - OK to runtime suspend the device
3430 * -EBUSY - Device should not be runtime suspended
3432 int blk_pre_runtime_suspend(struct request_queue *q)
3439 spin_lock_irq(q->queue_lock);
3440 if (q->nr_pending) {
3442 pm_runtime_mark_last_busy(q->dev);
3444 q->rpm_status = RPM_SUSPENDING;
3446 spin_unlock_irq(q->queue_lock);
3449 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3452 * blk_post_runtime_suspend - Post runtime suspend processing
3453 * @q: the queue of the device
3454 * @err: return value of the device's runtime_suspend function
3457 * Update the queue's runtime status according to the return value of the
3458 * device's runtime suspend function and mark last busy for the device so
3459 * that PM core will try to auto suspend the device at a later time.
3461 * This function should be called near the end of the device's
3462 * runtime_suspend callback.
3464 void blk_post_runtime_suspend(struct request_queue *q, int err)
3469 spin_lock_irq(q->queue_lock);
3471 q->rpm_status = RPM_SUSPENDED;
3473 q->rpm_status = RPM_ACTIVE;
3474 pm_runtime_mark_last_busy(q->dev);
3476 spin_unlock_irq(q->queue_lock);
3478 EXPORT_SYMBOL(blk_post_runtime_suspend);
3481 * blk_pre_runtime_resume - Pre runtime resume processing
3482 * @q: the queue of the device
3485 * Update the queue's runtime status to RESUMING in preparation for the
3486 * runtime resume of the device.
3488 * This function should be called near the start of the device's
3489 * runtime_resume callback.
3491 void blk_pre_runtime_resume(struct request_queue *q)
3496 spin_lock_irq(q->queue_lock);
3497 q->rpm_status = RPM_RESUMING;
3498 spin_unlock_irq(q->queue_lock);
3500 EXPORT_SYMBOL(blk_pre_runtime_resume);
3503 * blk_post_runtime_resume - Post runtime resume processing
3504 * @q: the queue of the device
3505 * @err: return value of the device's runtime_resume function
3508 * Update the queue's runtime status according to the return value of the
3509 * device's runtime_resume function. If it is successfully resumed, process
3510 * the requests that are queued into the device's queue when it is resuming
3511 * and then mark last busy and initiate autosuspend for it.
3513 * This function should be called near the end of the device's
3514 * runtime_resume callback.
3516 void blk_post_runtime_resume(struct request_queue *q, int err)
3521 spin_lock_irq(q->queue_lock);
3523 q->rpm_status = RPM_ACTIVE;
3525 pm_runtime_mark_last_busy(q->dev);
3526 pm_request_autosuspend(q->dev);
3528 q->rpm_status = RPM_SUSPENDED;
3530 spin_unlock_irq(q->queue_lock);
3532 EXPORT_SYMBOL(blk_post_runtime_resume);
3535 * blk_set_runtime_active - Force runtime status of the queue to be active
3536 * @q: the queue of the device
3538 * If the device is left runtime suspended during system suspend the resume
3539 * hook typically resumes the device and corrects runtime status
3540 * accordingly. However, that does not affect the queue runtime PM status
3541 * which is still "suspended". This prevents processing requests from the
3544 * This function can be used in driver's resume hook to correct queue
3545 * runtime PM status and re-enable peeking requests from the queue. It
3546 * should be called before first request is added to the queue.
3548 void blk_set_runtime_active(struct request_queue *q)
3550 spin_lock_irq(q->queue_lock);
3551 q->rpm_status = RPM_ACTIVE;
3552 pm_runtime_mark_last_busy(q->dev);
3553 pm_request_autosuspend(q->dev);
3554 spin_unlock_irq(q->queue_lock);
3556 EXPORT_SYMBOL(blk_set_runtime_active);
3559 int __init blk_dev_init(void)
3561 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3562 FIELD_SIZEOF(struct request, cmd_flags));
3564 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3565 kblockd_workqueue = alloc_workqueue("kblockd",
3566 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3567 if (!kblockd_workqueue)
3568 panic("Failed to create kblockd\n");
3570 request_cachep = kmem_cache_create("blkdev_requests",
3571 sizeof(struct request), 0, SLAB_PANIC, NULL);
3573 blk_requestq_cachep = kmem_cache_create("request_queue",
3574 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);