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, int op, unsigned int flags)
964 struct request_queue *q = rl->q;
965 int sync = rw_is_sync(op, 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_PREFLUSH | 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 * @op: REQ_OP_READ/REQ_OP_WRITE
1058 * @op_flags: rq_flag_bits
1059 * @bio: bio to allocate request for (can be %NULL)
1060 * @gfp_mask: allocation mask
1062 * Get a free request from @q. This function may fail under memory
1063 * pressure or if @q is dead.
1065 * Must be called with @q->queue_lock held and,
1066 * Returns ERR_PTR on failure, with @q->queue_lock held.
1067 * Returns request pointer on success, with @q->queue_lock *not held*.
1069 static struct request *__get_request(struct request_list *rl, int op,
1070 int op_flags, struct bio *bio,
1073 struct request_queue *q = rl->q;
1075 struct elevator_type *et = q->elevator->type;
1076 struct io_context *ioc = rq_ioc(bio);
1077 struct io_cq *icq = NULL;
1078 const bool is_sync = rw_is_sync(op, op_flags) != 0;
1081 if (unlikely(blk_queue_dying(q)))
1082 return ERR_PTR(-ENODEV);
1084 may_queue = elv_may_queue(q, op, op_flags);
1085 if (may_queue == ELV_MQUEUE_NO)
1088 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1089 if (rl->count[is_sync]+1 >= q->nr_requests) {
1091 * The queue will fill after this allocation, so set
1092 * it as full, and mark this process as "batching".
1093 * This process will be allowed to complete a batch of
1094 * requests, others will be blocked.
1096 if (!blk_rl_full(rl, is_sync)) {
1097 ioc_set_batching(q, ioc);
1098 blk_set_rl_full(rl, is_sync);
1100 if (may_queue != ELV_MQUEUE_MUST
1101 && !ioc_batching(q, ioc)) {
1103 * The queue is full and the allocating
1104 * process is not a "batcher", and not
1105 * exempted by the IO scheduler
1107 return ERR_PTR(-ENOMEM);
1111 blk_set_congested(rl, is_sync);
1115 * Only allow batching queuers to allocate up to 50% over the defined
1116 * limit of requests, otherwise we could have thousands of requests
1117 * allocated with any setting of ->nr_requests
1119 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1120 return ERR_PTR(-ENOMEM);
1122 q->nr_rqs[is_sync]++;
1123 rl->count[is_sync]++;
1124 rl->starved[is_sync] = 0;
1127 * Decide whether the new request will be managed by elevator. If
1128 * so, mark @op_flags and increment elvpriv. Non-zero elvpriv will
1129 * prevent the current elevator from being destroyed until the new
1130 * request is freed. This guarantees icq's won't be destroyed and
1131 * makes creating new ones safe.
1133 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1134 * it will be created after releasing queue_lock.
1136 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1137 op_flags |= REQ_ELVPRIV;
1138 q->nr_rqs_elvpriv++;
1139 if (et->icq_cache && ioc)
1140 icq = ioc_lookup_icq(ioc, q);
1143 if (blk_queue_io_stat(q))
1144 op_flags |= REQ_IO_STAT;
1145 spin_unlock_irq(q->queue_lock);
1147 /* allocate and init request */
1148 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1153 blk_rq_set_rl(rq, rl);
1154 req_set_op_attrs(rq, op, op_flags | REQ_ALLOCED);
1157 if (op_flags & REQ_ELVPRIV) {
1158 if (unlikely(et->icq_cache && !icq)) {
1160 icq = ioc_create_icq(ioc, q, gfp_mask);
1166 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1169 /* @rq->elv.icq holds io_context until @rq is freed */
1171 get_io_context(icq->ioc);
1175 * ioc may be NULL here, and ioc_batching will be false. That's
1176 * OK, if the queue is under the request limit then requests need
1177 * not count toward the nr_batch_requests limit. There will always
1178 * be some limit enforced by BLK_BATCH_TIME.
1180 if (ioc_batching(q, ioc))
1181 ioc->nr_batch_requests--;
1183 trace_block_getrq(q, bio, op);
1188 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1189 * and may fail indefinitely under memory pressure and thus
1190 * shouldn't stall IO. Treat this request as !elvpriv. This will
1191 * disturb iosched and blkcg but weird is bettern than dead.
1193 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1194 __func__, dev_name(q->backing_dev_info.dev));
1196 rq->cmd_flags &= ~REQ_ELVPRIV;
1199 spin_lock_irq(q->queue_lock);
1200 q->nr_rqs_elvpriv--;
1201 spin_unlock_irq(q->queue_lock);
1206 * Allocation failed presumably due to memory. Undo anything we
1207 * might have messed up.
1209 * Allocating task should really be put onto the front of the wait
1210 * queue, but this is pretty rare.
1212 spin_lock_irq(q->queue_lock);
1213 freed_request(rl, op, op_flags);
1216 * in the very unlikely event that allocation failed and no
1217 * requests for this direction was pending, mark us starved so that
1218 * freeing of a request in the other direction will notice
1219 * us. another possible fix would be to split the rq mempool into
1223 if (unlikely(rl->count[is_sync] == 0))
1224 rl->starved[is_sync] = 1;
1225 return ERR_PTR(-ENOMEM);
1229 * get_request - get a free request
1230 * @q: request_queue to allocate request from
1231 * @op: REQ_OP_READ/REQ_OP_WRITE
1232 * @op_flags: rq_flag_bits
1233 * @bio: bio to allocate request for (can be %NULL)
1234 * @gfp_mask: allocation mask
1236 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1237 * this function keeps retrying under memory pressure and fails iff @q is dead.
1239 * Must be called with @q->queue_lock held and,
1240 * Returns ERR_PTR on failure, with @q->queue_lock held.
1241 * Returns request pointer on success, with @q->queue_lock *not held*.
1243 static struct request *get_request(struct request_queue *q, int op,
1244 int op_flags, struct bio *bio,
1247 const bool is_sync = rw_is_sync(op, op_flags) != 0;
1249 struct request_list *rl;
1252 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1254 rq = __get_request(rl, op, op_flags, bio, gfp_mask);
1258 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1263 /* wait on @rl and retry */
1264 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1265 TASK_UNINTERRUPTIBLE);
1267 trace_block_sleeprq(q, bio, op);
1269 spin_unlock_irq(q->queue_lock);
1273 * After sleeping, we become a "batching" process and will be able
1274 * to allocate at least one request, and up to a big batch of them
1275 * for a small period time. See ioc_batching, ioc_set_batching
1277 ioc_set_batching(q, current->io_context);
1279 spin_lock_irq(q->queue_lock);
1280 finish_wait(&rl->wait[is_sync], &wait);
1285 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1290 BUG_ON(rw != READ && rw != WRITE);
1292 /* create ioc upfront */
1293 create_io_context(gfp_mask, q->node);
1295 spin_lock_irq(q->queue_lock);
1296 rq = get_request(q, rw, 0, NULL, gfp_mask);
1298 spin_unlock_irq(q->queue_lock);
1299 /* q->queue_lock is unlocked at this point */
1304 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1307 return blk_mq_alloc_request(q, rw,
1308 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1309 0 : BLK_MQ_REQ_NOWAIT);
1311 return blk_old_get_request(q, rw, gfp_mask);
1313 EXPORT_SYMBOL(blk_get_request);
1316 * blk_make_request - given a bio, allocate a corresponding struct request.
1317 * @q: target request queue
1318 * @bio: The bio describing the memory mappings that will be submitted for IO.
1319 * It may be a chained-bio properly constructed by block/bio layer.
1320 * @gfp_mask: gfp flags to be used for memory allocation
1322 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1323 * type commands. Where the struct request needs to be farther initialized by
1324 * the caller. It is passed a &struct bio, which describes the memory info of
1327 * The caller of blk_make_request must make sure that bi_io_vec
1328 * are set to describe the memory buffers. That bio_data_dir() will return
1329 * the needed direction of the request. (And all bio's in the passed bio-chain
1330 * are properly set accordingly)
1332 * If called under none-sleepable conditions, mapped bio buffers must not
1333 * need bouncing, by calling the appropriate masked or flagged allocator,
1334 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1337 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1338 * given to how you allocate bios. In particular, you cannot use
1339 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1340 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1341 * thus resulting in a deadlock. Alternatively bios should be allocated using
1342 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1343 * If possible a big IO should be split into smaller parts when allocation
1344 * fails. Partial allocation should not be an error, or you risk a live-lock.
1346 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1349 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1354 blk_rq_set_block_pc(rq);
1357 struct bio *bounce_bio = bio;
1360 blk_queue_bounce(q, &bounce_bio);
1361 ret = blk_rq_append_bio(q, rq, bounce_bio);
1362 if (unlikely(ret)) {
1363 blk_put_request(rq);
1364 return ERR_PTR(ret);
1370 EXPORT_SYMBOL(blk_make_request);
1373 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1374 * @rq: request to be initialized
1377 void blk_rq_set_block_pc(struct request *rq)
1379 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1381 rq->__sector = (sector_t) -1;
1382 rq->bio = rq->biotail = NULL;
1383 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1385 EXPORT_SYMBOL(blk_rq_set_block_pc);
1388 * blk_requeue_request - put a request back on queue
1389 * @q: request queue where request should be inserted
1390 * @rq: request to be inserted
1393 * Drivers often keep queueing requests until the hardware cannot accept
1394 * more, when that condition happens we need to put the request back
1395 * on the queue. Must be called with queue lock held.
1397 void blk_requeue_request(struct request_queue *q, struct request *rq)
1399 blk_delete_timer(rq);
1400 blk_clear_rq_complete(rq);
1401 trace_block_rq_requeue(q, rq);
1403 if (rq->cmd_flags & REQ_QUEUED)
1404 blk_queue_end_tag(q, rq);
1406 BUG_ON(blk_queued_rq(rq));
1408 elv_requeue_request(q, rq);
1410 EXPORT_SYMBOL(blk_requeue_request);
1412 static void add_acct_request(struct request_queue *q, struct request *rq,
1415 blk_account_io_start(rq, true);
1416 __elv_add_request(q, rq, where);
1419 static void part_round_stats_single(int cpu, struct hd_struct *part,
1424 if (now == part->stamp)
1427 inflight = part_in_flight(part);
1429 __part_stat_add(cpu, part, time_in_queue,
1430 inflight * (now - part->stamp));
1431 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1437 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1438 * @cpu: cpu number for stats access
1439 * @part: target partition
1441 * The average IO queue length and utilisation statistics are maintained
1442 * by observing the current state of the queue length and the amount of
1443 * time it has been in this state for.
1445 * Normally, that accounting is done on IO completion, but that can result
1446 * in more than a second's worth of IO being accounted for within any one
1447 * second, leading to >100% utilisation. To deal with that, we call this
1448 * function to do a round-off before returning the results when reading
1449 * /proc/diskstats. This accounts immediately for all queue usage up to
1450 * the current jiffies and restarts the counters again.
1452 void part_round_stats(int cpu, struct hd_struct *part)
1454 unsigned long now = jiffies;
1457 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1458 part_round_stats_single(cpu, part, now);
1460 EXPORT_SYMBOL_GPL(part_round_stats);
1463 static void blk_pm_put_request(struct request *rq)
1465 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1466 pm_runtime_mark_last_busy(rq->q->dev);
1469 static inline void blk_pm_put_request(struct request *rq) {}
1473 * queue lock must be held
1475 void __blk_put_request(struct request_queue *q, struct request *req)
1481 blk_mq_free_request(req);
1485 blk_pm_put_request(req);
1487 elv_completed_request(q, req);
1489 /* this is a bio leak */
1490 WARN_ON(req->bio != NULL);
1493 * Request may not have originated from ll_rw_blk. if not,
1494 * it didn't come out of our reserved rq pools
1496 if (req->cmd_flags & REQ_ALLOCED) {
1497 unsigned int flags = req->cmd_flags;
1498 int op = req_op(req);
1499 struct request_list *rl = blk_rq_rl(req);
1501 BUG_ON(!list_empty(&req->queuelist));
1502 BUG_ON(ELV_ON_HASH(req));
1504 blk_free_request(rl, req);
1505 freed_request(rl, op, flags);
1509 EXPORT_SYMBOL_GPL(__blk_put_request);
1511 void blk_put_request(struct request *req)
1513 struct request_queue *q = req->q;
1516 blk_mq_free_request(req);
1518 unsigned long flags;
1520 spin_lock_irqsave(q->queue_lock, flags);
1521 __blk_put_request(q, req);
1522 spin_unlock_irqrestore(q->queue_lock, flags);
1525 EXPORT_SYMBOL(blk_put_request);
1528 * blk_add_request_payload - add a payload to a request
1529 * @rq: request to update
1530 * @page: page backing the payload
1531 * @offset: offset in page
1532 * @len: length of the payload.
1534 * This allows to later add a payload to an already submitted request by
1535 * a block driver. The driver needs to take care of freeing the payload
1538 * Note that this is a quite horrible hack and nothing but handling of
1539 * discard requests should ever use it.
1541 void blk_add_request_payload(struct request *rq, struct page *page,
1542 int offset, unsigned int len)
1544 struct bio *bio = rq->bio;
1546 bio->bi_io_vec->bv_page = page;
1547 bio->bi_io_vec->bv_offset = offset;
1548 bio->bi_io_vec->bv_len = len;
1550 bio->bi_iter.bi_size = len;
1552 bio->bi_phys_segments = 1;
1554 rq->__data_len = rq->resid_len = len;
1555 rq->nr_phys_segments = 1;
1557 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1559 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1562 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1564 if (!ll_back_merge_fn(q, req, bio))
1567 trace_block_bio_backmerge(q, req, bio);
1569 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1570 blk_rq_set_mixed_merge(req);
1572 req->biotail->bi_next = bio;
1574 req->__data_len += bio->bi_iter.bi_size;
1575 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1577 blk_account_io_start(req, false);
1581 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1584 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1586 if (!ll_front_merge_fn(q, req, bio))
1589 trace_block_bio_frontmerge(q, req, bio);
1591 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1592 blk_rq_set_mixed_merge(req);
1594 bio->bi_next = req->bio;
1597 req->__sector = bio->bi_iter.bi_sector;
1598 req->__data_len += bio->bi_iter.bi_size;
1599 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1601 blk_account_io_start(req, false);
1606 * blk_attempt_plug_merge - try to merge with %current's plugged list
1607 * @q: request_queue new bio is being queued at
1608 * @bio: new bio being queued
1609 * @request_count: out parameter for number of traversed plugged requests
1610 * @same_queue_rq: pointer to &struct request that gets filled in when
1611 * another request associated with @q is found on the plug list
1612 * (optional, may be %NULL)
1614 * Determine whether @bio being queued on @q can be merged with a request
1615 * on %current's plugged list. Returns %true if merge was successful,
1618 * Plugging coalesces IOs from the same issuer for the same purpose without
1619 * going through @q->queue_lock. As such it's more of an issuing mechanism
1620 * than scheduling, and the request, while may have elvpriv data, is not
1621 * added on the elevator at this point. In addition, we don't have
1622 * reliable access to the elevator outside queue lock. Only check basic
1623 * merging parameters without querying the elevator.
1625 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1627 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1628 unsigned int *request_count,
1629 struct request **same_queue_rq)
1631 struct blk_plug *plug;
1634 struct list_head *plug_list;
1636 plug = current->plug;
1642 plug_list = &plug->mq_list;
1644 plug_list = &plug->list;
1646 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1652 * Only blk-mq multiple hardware queues case checks the
1653 * rq in the same queue, there should be only one such
1657 *same_queue_rq = rq;
1660 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1663 el_ret = blk_try_merge(rq, bio);
1664 if (el_ret == ELEVATOR_BACK_MERGE) {
1665 ret = bio_attempt_back_merge(q, rq, bio);
1668 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1669 ret = bio_attempt_front_merge(q, rq, bio);
1678 unsigned int blk_plug_queued_count(struct request_queue *q)
1680 struct blk_plug *plug;
1682 struct list_head *plug_list;
1683 unsigned int ret = 0;
1685 plug = current->plug;
1690 plug_list = &plug->mq_list;
1692 plug_list = &plug->list;
1694 list_for_each_entry(rq, plug_list, queuelist) {
1702 void init_request_from_bio(struct request *req, struct bio *bio)
1704 req->cmd_type = REQ_TYPE_FS;
1706 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1707 if (bio->bi_rw & REQ_RAHEAD)
1708 req->cmd_flags |= REQ_FAILFAST_MASK;
1711 req->__sector = bio->bi_iter.bi_sector;
1712 req->ioprio = bio_prio(bio);
1713 blk_rq_bio_prep(req->q, req, bio);
1716 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1718 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1719 struct blk_plug *plug;
1720 int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
1721 struct request *req;
1722 unsigned int request_count = 0;
1725 * low level driver can indicate that it wants pages above a
1726 * certain limit bounced to low memory (ie for highmem, or even
1727 * ISA dma in theory)
1729 blk_queue_bounce(q, &bio);
1731 blk_queue_split(q, &bio, q->bio_split);
1733 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1734 bio->bi_error = -EIO;
1736 return BLK_QC_T_NONE;
1739 if (bio->bi_rw & (REQ_PREFLUSH | REQ_FUA)) {
1740 spin_lock_irq(q->queue_lock);
1741 where = ELEVATOR_INSERT_FLUSH;
1746 * Check if we can merge with the plugged list before grabbing
1749 if (!blk_queue_nomerges(q)) {
1750 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1751 return BLK_QC_T_NONE;
1753 request_count = blk_plug_queued_count(q);
1755 spin_lock_irq(q->queue_lock);
1757 el_ret = elv_merge(q, &req, bio);
1758 if (el_ret == ELEVATOR_BACK_MERGE) {
1759 if (bio_attempt_back_merge(q, req, bio)) {
1760 elv_bio_merged(q, req, bio);
1761 if (!attempt_back_merge(q, req))
1762 elv_merged_request(q, req, el_ret);
1765 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1766 if (bio_attempt_front_merge(q, req, bio)) {
1767 elv_bio_merged(q, req, bio);
1768 if (!attempt_front_merge(q, req))
1769 elv_merged_request(q, req, el_ret);
1776 * This sync check and mask will be re-done in init_request_from_bio(),
1777 * but we need to set it earlier to expose the sync flag to the
1778 * rq allocator and io schedulers.
1781 rw_flags |= REQ_SYNC;
1784 * Add in META/PRIO flags, if set, before we get to the IO scheduler
1786 rw_flags |= (bio->bi_rw & (REQ_META | REQ_PRIO));
1789 * Grab a free request. This is might sleep but can not fail.
1790 * Returns with the queue unlocked.
1792 req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
1794 bio->bi_error = PTR_ERR(req);
1800 * After dropping the lock and possibly sleeping here, our request
1801 * may now be mergeable after it had proven unmergeable (above).
1802 * We don't worry about that case for efficiency. It won't happen
1803 * often, and the elevators are able to handle it.
1805 init_request_from_bio(req, bio);
1807 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1808 req->cpu = raw_smp_processor_id();
1810 plug = current->plug;
1813 * If this is the first request added after a plug, fire
1817 trace_block_plug(q);
1819 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1820 blk_flush_plug_list(plug, false);
1821 trace_block_plug(q);
1824 list_add_tail(&req->queuelist, &plug->list);
1825 blk_account_io_start(req, true);
1827 spin_lock_irq(q->queue_lock);
1828 add_acct_request(q, req, where);
1831 spin_unlock_irq(q->queue_lock);
1834 return BLK_QC_T_NONE;
1838 * If bio->bi_dev is a partition, remap the location
1840 static inline void blk_partition_remap(struct bio *bio)
1842 struct block_device *bdev = bio->bi_bdev;
1844 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1845 struct hd_struct *p = bdev->bd_part;
1847 bio->bi_iter.bi_sector += p->start_sect;
1848 bio->bi_bdev = bdev->bd_contains;
1850 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1852 bio->bi_iter.bi_sector - p->start_sect);
1856 static void handle_bad_sector(struct bio *bio)
1858 char b[BDEVNAME_SIZE];
1860 printk(KERN_INFO "attempt to access beyond end of device\n");
1861 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1862 bdevname(bio->bi_bdev, b),
1864 (unsigned long long)bio_end_sector(bio),
1865 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1868 #ifdef CONFIG_FAIL_MAKE_REQUEST
1870 static DECLARE_FAULT_ATTR(fail_make_request);
1872 static int __init setup_fail_make_request(char *str)
1874 return setup_fault_attr(&fail_make_request, str);
1876 __setup("fail_make_request=", setup_fail_make_request);
1878 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1880 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1883 static int __init fail_make_request_debugfs(void)
1885 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1886 NULL, &fail_make_request);
1888 return PTR_ERR_OR_ZERO(dir);
1891 late_initcall(fail_make_request_debugfs);
1893 #else /* CONFIG_FAIL_MAKE_REQUEST */
1895 static inline bool should_fail_request(struct hd_struct *part,
1901 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1904 * Check whether this bio extends beyond the end of the device.
1906 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1913 /* Test device or partition size, when known. */
1914 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1916 sector_t sector = bio->bi_iter.bi_sector;
1918 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1920 * This may well happen - the kernel calls bread()
1921 * without checking the size of the device, e.g., when
1922 * mounting a device.
1924 handle_bad_sector(bio);
1932 static noinline_for_stack bool
1933 generic_make_request_checks(struct bio *bio)
1935 struct request_queue *q;
1936 int nr_sectors = bio_sectors(bio);
1938 char b[BDEVNAME_SIZE];
1939 struct hd_struct *part;
1943 if (bio_check_eod(bio, nr_sectors))
1946 q = bdev_get_queue(bio->bi_bdev);
1949 "generic_make_request: Trying to access "
1950 "nonexistent block-device %s (%Lu)\n",
1951 bdevname(bio->bi_bdev, b),
1952 (long long) bio->bi_iter.bi_sector);
1956 part = bio->bi_bdev->bd_part;
1957 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1958 should_fail_request(&part_to_disk(part)->part0,
1959 bio->bi_iter.bi_size))
1963 * If this device has partitions, remap block n
1964 * of partition p to block n+start(p) of the disk.
1966 blk_partition_remap(bio);
1968 if (bio_check_eod(bio, nr_sectors))
1972 * Filter flush bio's early so that make_request based
1973 * drivers without flush support don't have to worry
1976 if ((bio->bi_rw & (REQ_PREFLUSH | REQ_FUA)) &&
1977 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1978 bio->bi_rw &= ~(REQ_PREFLUSH | REQ_FUA);
1985 if ((bio_op(bio) == REQ_OP_DISCARD) &&
1986 (!blk_queue_discard(q) ||
1987 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1992 if (bio_op(bio) == REQ_OP_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1998 * Various block parts want %current->io_context and lazy ioc
1999 * allocation ends up trading a lot of pain for a small amount of
2000 * memory. Just allocate it upfront. This may fail and block
2001 * layer knows how to live with it.
2003 create_io_context(GFP_ATOMIC, q->node);
2005 if (!blkcg_bio_issue_check(q, bio))
2008 trace_block_bio_queue(q, bio);
2012 bio->bi_error = err;
2018 * generic_make_request - hand a buffer to its device driver for I/O
2019 * @bio: The bio describing the location in memory and on the device.
2021 * generic_make_request() is used to make I/O requests of block
2022 * devices. It is passed a &struct bio, which describes the I/O that needs
2025 * generic_make_request() does not return any status. The
2026 * success/failure status of the request, along with notification of
2027 * completion, is delivered asynchronously through the bio->bi_end_io
2028 * function described (one day) else where.
2030 * The caller of generic_make_request must make sure that bi_io_vec
2031 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2032 * set to describe the device address, and the
2033 * bi_end_io and optionally bi_private are set to describe how
2034 * completion notification should be signaled.
2036 * generic_make_request and the drivers it calls may use bi_next if this
2037 * bio happens to be merged with someone else, and may resubmit the bio to
2038 * a lower device by calling into generic_make_request recursively, which
2039 * means the bio should NOT be touched after the call to ->make_request_fn.
2041 blk_qc_t generic_make_request(struct bio *bio)
2043 struct bio_list bio_list_on_stack;
2044 blk_qc_t ret = BLK_QC_T_NONE;
2046 if (!generic_make_request_checks(bio))
2050 * We only want one ->make_request_fn to be active at a time, else
2051 * stack usage with stacked devices could be a problem. So use
2052 * current->bio_list to keep a list of requests submited by a
2053 * make_request_fn function. current->bio_list is also used as a
2054 * flag to say if generic_make_request is currently active in this
2055 * task or not. If it is NULL, then no make_request is active. If
2056 * it is non-NULL, then a make_request is active, and new requests
2057 * should be added at the tail
2059 if (current->bio_list) {
2060 bio_list_add(current->bio_list, bio);
2064 /* following loop may be a bit non-obvious, and so deserves some
2066 * Before entering the loop, bio->bi_next is NULL (as all callers
2067 * ensure that) so we have a list with a single bio.
2068 * We pretend that we have just taken it off a longer list, so
2069 * we assign bio_list to a pointer to the bio_list_on_stack,
2070 * thus initialising the bio_list of new bios to be
2071 * added. ->make_request() may indeed add some more bios
2072 * through a recursive call to generic_make_request. If it
2073 * did, we find a non-NULL value in bio_list and re-enter the loop
2074 * from the top. In this case we really did just take the bio
2075 * of the top of the list (no pretending) and so remove it from
2076 * bio_list, and call into ->make_request() again.
2078 BUG_ON(bio->bi_next);
2079 bio_list_init(&bio_list_on_stack);
2080 current->bio_list = &bio_list_on_stack;
2082 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2084 if (likely(blk_queue_enter(q, false) == 0)) {
2085 ret = q->make_request_fn(q, bio);
2089 bio = bio_list_pop(current->bio_list);
2091 struct bio *bio_next = bio_list_pop(current->bio_list);
2097 current->bio_list = NULL; /* deactivate */
2102 EXPORT_SYMBOL(generic_make_request);
2105 * submit_bio - submit a bio to the block device layer for I/O
2106 * @bio: The &struct bio which describes the I/O
2108 * submit_bio() is very similar in purpose to generic_make_request(), and
2109 * uses that function to do most of the work. Both are fairly rough
2110 * interfaces; @bio must be presetup and ready for I/O.
2113 blk_qc_t submit_bio(struct bio *bio)
2116 * If it's a regular read/write or a barrier with data attached,
2117 * go through the normal accounting stuff before submission.
2119 if (bio_has_data(bio)) {
2122 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2123 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2125 count = bio_sectors(bio);
2127 if (op_is_write(bio_op(bio))) {
2128 count_vm_events(PGPGOUT, count);
2130 task_io_account_read(bio->bi_iter.bi_size);
2131 count_vm_events(PGPGIN, count);
2134 if (unlikely(block_dump)) {
2135 char b[BDEVNAME_SIZE];
2136 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2137 current->comm, task_pid_nr(current),
2138 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2139 (unsigned long long)bio->bi_iter.bi_sector,
2140 bdevname(bio->bi_bdev, b),
2145 return generic_make_request(bio);
2147 EXPORT_SYMBOL(submit_bio);
2150 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2151 * for new the queue limits
2153 * @rq: the request being checked
2156 * @rq may have been made based on weaker limitations of upper-level queues
2157 * in request stacking drivers, and it may violate the limitation of @q.
2158 * Since the block layer and the underlying device driver trust @rq
2159 * after it is inserted to @q, it should be checked against @q before
2160 * the insertion using this generic function.
2162 * Request stacking drivers like request-based dm may change the queue
2163 * limits when retrying requests on other queues. Those requests need
2164 * to be checked against the new queue limits again during dispatch.
2166 static int blk_cloned_rq_check_limits(struct request_queue *q,
2169 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2170 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2175 * queue's settings related to segment counting like q->bounce_pfn
2176 * may differ from that of other stacking queues.
2177 * Recalculate it to check the request correctly on this queue's
2180 blk_recalc_rq_segments(rq);
2181 if (rq->nr_phys_segments > queue_max_segments(q)) {
2182 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2190 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2191 * @q: the queue to submit the request
2192 * @rq: the request being queued
2194 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2196 unsigned long flags;
2197 int where = ELEVATOR_INSERT_BACK;
2199 if (blk_cloned_rq_check_limits(q, rq))
2203 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2207 if (blk_queue_io_stat(q))
2208 blk_account_io_start(rq, true);
2209 blk_mq_insert_request(rq, false, true, false);
2213 spin_lock_irqsave(q->queue_lock, flags);
2214 if (unlikely(blk_queue_dying(q))) {
2215 spin_unlock_irqrestore(q->queue_lock, flags);
2220 * Submitting request must be dequeued before calling this function
2221 * because it will be linked to another request_queue
2223 BUG_ON(blk_queued_rq(rq));
2225 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2226 where = ELEVATOR_INSERT_FLUSH;
2228 add_acct_request(q, rq, where);
2229 if (where == ELEVATOR_INSERT_FLUSH)
2231 spin_unlock_irqrestore(q->queue_lock, flags);
2235 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2238 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2239 * @rq: request to examine
2242 * A request could be merge of IOs which require different failure
2243 * handling. This function determines the number of bytes which
2244 * can be failed from the beginning of the request without
2245 * crossing into area which need to be retried further.
2248 * The number of bytes to fail.
2251 * queue_lock must be held.
2253 unsigned int blk_rq_err_bytes(const struct request *rq)
2255 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2256 unsigned int bytes = 0;
2259 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2260 return blk_rq_bytes(rq);
2263 * Currently the only 'mixing' which can happen is between
2264 * different fastfail types. We can safely fail portions
2265 * which have all the failfast bits that the first one has -
2266 * the ones which are at least as eager to fail as the first
2269 for (bio = rq->bio; bio; bio = bio->bi_next) {
2270 if ((bio->bi_rw & ff) != ff)
2272 bytes += bio->bi_iter.bi_size;
2275 /* this could lead to infinite loop */
2276 BUG_ON(blk_rq_bytes(rq) && !bytes);
2279 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2281 void blk_account_io_completion(struct request *req, unsigned int bytes)
2283 if (blk_do_io_stat(req)) {
2284 const int rw = rq_data_dir(req);
2285 struct hd_struct *part;
2288 cpu = part_stat_lock();
2290 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2295 void blk_account_io_done(struct request *req)
2298 * Account IO completion. flush_rq isn't accounted as a
2299 * normal IO on queueing nor completion. Accounting the
2300 * containing request is enough.
2302 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2303 unsigned long duration = jiffies - req->start_time;
2304 const int rw = rq_data_dir(req);
2305 struct hd_struct *part;
2308 cpu = part_stat_lock();
2311 part_stat_inc(cpu, part, ios[rw]);
2312 part_stat_add(cpu, part, ticks[rw], duration);
2313 part_round_stats(cpu, part);
2314 part_dec_in_flight(part, rw);
2316 hd_struct_put(part);
2323 * Don't process normal requests when queue is suspended
2324 * or in the process of suspending/resuming
2326 static struct request *blk_pm_peek_request(struct request_queue *q,
2329 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2330 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2336 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2343 void blk_account_io_start(struct request *rq, bool new_io)
2345 struct hd_struct *part;
2346 int rw = rq_data_dir(rq);
2349 if (!blk_do_io_stat(rq))
2352 cpu = part_stat_lock();
2356 part_stat_inc(cpu, part, merges[rw]);
2358 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2359 if (!hd_struct_try_get(part)) {
2361 * The partition is already being removed,
2362 * the request will be accounted on the disk only
2364 * We take a reference on disk->part0 although that
2365 * partition will never be deleted, so we can treat
2366 * it as any other partition.
2368 part = &rq->rq_disk->part0;
2369 hd_struct_get(part);
2371 part_round_stats(cpu, part);
2372 part_inc_in_flight(part, rw);
2380 * blk_peek_request - peek at the top of a request queue
2381 * @q: request queue to peek at
2384 * Return the request at the top of @q. The returned request
2385 * should be started using blk_start_request() before LLD starts
2389 * Pointer to the request at the top of @q if available. Null
2393 * queue_lock must be held.
2395 struct request *blk_peek_request(struct request_queue *q)
2400 while ((rq = __elv_next_request(q)) != NULL) {
2402 rq = blk_pm_peek_request(q, rq);
2406 if (!(rq->cmd_flags & REQ_STARTED)) {
2408 * This is the first time the device driver
2409 * sees this request (possibly after
2410 * requeueing). Notify IO scheduler.
2412 if (rq->cmd_flags & REQ_SORTED)
2413 elv_activate_rq(q, rq);
2416 * just mark as started even if we don't start
2417 * it, a request that has been delayed should
2418 * not be passed by new incoming requests
2420 rq->cmd_flags |= REQ_STARTED;
2421 trace_block_rq_issue(q, rq);
2424 if (!q->boundary_rq || q->boundary_rq == rq) {
2425 q->end_sector = rq_end_sector(rq);
2426 q->boundary_rq = NULL;
2429 if (rq->cmd_flags & REQ_DONTPREP)
2432 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2434 * make sure space for the drain appears we
2435 * know we can do this because max_hw_segments
2436 * has been adjusted to be one fewer than the
2439 rq->nr_phys_segments++;
2445 ret = q->prep_rq_fn(q, rq);
2446 if (ret == BLKPREP_OK) {
2448 } else if (ret == BLKPREP_DEFER) {
2450 * the request may have been (partially) prepped.
2451 * we need to keep this request in the front to
2452 * avoid resource deadlock. REQ_STARTED will
2453 * prevent other fs requests from passing this one.
2455 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2456 !(rq->cmd_flags & REQ_DONTPREP)) {
2458 * remove the space for the drain we added
2459 * so that we don't add it again
2461 --rq->nr_phys_segments;
2466 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2467 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2469 rq->cmd_flags |= REQ_QUIET;
2471 * Mark this request as started so we don't trigger
2472 * any debug logic in the end I/O path.
2474 blk_start_request(rq);
2475 __blk_end_request_all(rq, err);
2477 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2484 EXPORT_SYMBOL(blk_peek_request);
2486 void blk_dequeue_request(struct request *rq)
2488 struct request_queue *q = rq->q;
2490 BUG_ON(list_empty(&rq->queuelist));
2491 BUG_ON(ELV_ON_HASH(rq));
2493 list_del_init(&rq->queuelist);
2496 * the time frame between a request being removed from the lists
2497 * and to it is freed is accounted as io that is in progress at
2500 if (blk_account_rq(rq)) {
2501 q->in_flight[rq_is_sync(rq)]++;
2502 set_io_start_time_ns(rq);
2507 * blk_start_request - start request processing on the driver
2508 * @req: request to dequeue
2511 * Dequeue @req and start timeout timer on it. This hands off the
2512 * request to the driver.
2514 * Block internal functions which don't want to start timer should
2515 * call blk_dequeue_request().
2518 * queue_lock must be held.
2520 void blk_start_request(struct request *req)
2522 blk_dequeue_request(req);
2525 * We are now handing the request to the hardware, initialize
2526 * resid_len to full count and add the timeout handler.
2528 req->resid_len = blk_rq_bytes(req);
2529 if (unlikely(blk_bidi_rq(req)))
2530 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2532 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2535 EXPORT_SYMBOL(blk_start_request);
2538 * blk_fetch_request - fetch a request from a request queue
2539 * @q: request queue to fetch a request from
2542 * Return the request at the top of @q. The request is started on
2543 * return and LLD can start processing it immediately.
2546 * Pointer to the request at the top of @q if available. Null
2550 * queue_lock must be held.
2552 struct request *blk_fetch_request(struct request_queue *q)
2556 rq = blk_peek_request(q);
2558 blk_start_request(rq);
2561 EXPORT_SYMBOL(blk_fetch_request);
2564 * blk_update_request - Special helper function for request stacking drivers
2565 * @req: the request being processed
2566 * @error: %0 for success, < %0 for error
2567 * @nr_bytes: number of bytes to complete @req
2570 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2571 * the request structure even if @req doesn't have leftover.
2572 * If @req has leftover, sets it up for the next range of segments.
2574 * This special helper function is only for request stacking drivers
2575 * (e.g. request-based dm) so that they can handle partial completion.
2576 * Actual device drivers should use blk_end_request instead.
2578 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2579 * %false return from this function.
2582 * %false - this request doesn't have any more data
2583 * %true - this request has more data
2585 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2589 trace_block_rq_complete(req->q, req, nr_bytes);
2595 * For fs requests, rq is just carrier of independent bio's
2596 * and each partial completion should be handled separately.
2597 * Reset per-request error on each partial completion.
2599 * TODO: tj: This is too subtle. It would be better to let
2600 * low level drivers do what they see fit.
2602 if (req->cmd_type == REQ_TYPE_FS)
2605 if (error && req->cmd_type == REQ_TYPE_FS &&
2606 !(req->cmd_flags & REQ_QUIET)) {
2611 error_type = "recoverable transport";
2614 error_type = "critical target";
2617 error_type = "critical nexus";
2620 error_type = "timeout";
2623 error_type = "critical space allocation";
2626 error_type = "critical medium";
2633 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2634 __func__, error_type, req->rq_disk ?
2635 req->rq_disk->disk_name : "?",
2636 (unsigned long long)blk_rq_pos(req));
2640 blk_account_io_completion(req, nr_bytes);
2644 struct bio *bio = req->bio;
2645 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2647 if (bio_bytes == bio->bi_iter.bi_size)
2648 req->bio = bio->bi_next;
2650 req_bio_endio(req, bio, bio_bytes, error);
2652 total_bytes += bio_bytes;
2653 nr_bytes -= bio_bytes;
2664 * Reset counters so that the request stacking driver
2665 * can find how many bytes remain in the request
2668 req->__data_len = 0;
2672 req->__data_len -= total_bytes;
2674 /* update sector only for requests with clear definition of sector */
2675 if (req->cmd_type == REQ_TYPE_FS)
2676 req->__sector += total_bytes >> 9;
2678 /* mixed attributes always follow the first bio */
2679 if (req->cmd_flags & REQ_MIXED_MERGE) {
2680 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2681 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2685 * If total number of sectors is less than the first segment
2686 * size, something has gone terribly wrong.
2688 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2689 blk_dump_rq_flags(req, "request botched");
2690 req->__data_len = blk_rq_cur_bytes(req);
2693 /* recalculate the number of segments */
2694 blk_recalc_rq_segments(req);
2698 EXPORT_SYMBOL_GPL(blk_update_request);
2700 static bool blk_update_bidi_request(struct request *rq, int error,
2701 unsigned int nr_bytes,
2702 unsigned int bidi_bytes)
2704 if (blk_update_request(rq, error, nr_bytes))
2707 /* Bidi request must be completed as a whole */
2708 if (unlikely(blk_bidi_rq(rq)) &&
2709 blk_update_request(rq->next_rq, error, bidi_bytes))
2712 if (blk_queue_add_random(rq->q))
2713 add_disk_randomness(rq->rq_disk);
2719 * blk_unprep_request - unprepare a request
2722 * This function makes a request ready for complete resubmission (or
2723 * completion). It happens only after all error handling is complete,
2724 * so represents the appropriate moment to deallocate any resources
2725 * that were allocated to the request in the prep_rq_fn. The queue
2726 * lock is held when calling this.
2728 void blk_unprep_request(struct request *req)
2730 struct request_queue *q = req->q;
2732 req->cmd_flags &= ~REQ_DONTPREP;
2733 if (q->unprep_rq_fn)
2734 q->unprep_rq_fn(q, req);
2736 EXPORT_SYMBOL_GPL(blk_unprep_request);
2739 * queue lock must be held
2741 void blk_finish_request(struct request *req, int error)
2743 if (req->cmd_flags & REQ_QUEUED)
2744 blk_queue_end_tag(req->q, req);
2746 BUG_ON(blk_queued_rq(req));
2748 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2749 laptop_io_completion(&req->q->backing_dev_info);
2751 blk_delete_timer(req);
2753 if (req->cmd_flags & REQ_DONTPREP)
2754 blk_unprep_request(req);
2756 blk_account_io_done(req);
2759 req->end_io(req, error);
2761 if (blk_bidi_rq(req))
2762 __blk_put_request(req->next_rq->q, req->next_rq);
2764 __blk_put_request(req->q, req);
2767 EXPORT_SYMBOL(blk_finish_request);
2770 * blk_end_bidi_request - Complete a bidi request
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 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2778 * Drivers that supports bidi can safely call this member for any
2779 * type of request, bidi or uni. In the later case @bidi_bytes is
2783 * %false - we are done with this request
2784 * %true - still buffers pending for this request
2786 static bool blk_end_bidi_request(struct request *rq, int error,
2787 unsigned int nr_bytes, unsigned int bidi_bytes)
2789 struct request_queue *q = rq->q;
2790 unsigned long flags;
2792 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2795 spin_lock_irqsave(q->queue_lock, flags);
2796 blk_finish_request(rq, error);
2797 spin_unlock_irqrestore(q->queue_lock, flags);
2803 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2804 * @rq: the request to complete
2805 * @error: %0 for success, < %0 for error
2806 * @nr_bytes: number of bytes to complete @rq
2807 * @bidi_bytes: number of bytes to complete @rq->next_rq
2810 * Identical to blk_end_bidi_request() except that queue lock is
2811 * assumed to be locked on entry and remains so on return.
2814 * %false - we are done with this request
2815 * %true - still buffers pending for this request
2817 bool __blk_end_bidi_request(struct request *rq, int error,
2818 unsigned int nr_bytes, unsigned int bidi_bytes)
2820 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2823 blk_finish_request(rq, error);
2829 * blk_end_request - Helper function for drivers to complete the request.
2830 * @rq: the request being processed
2831 * @error: %0 for success, < %0 for error
2832 * @nr_bytes: number of bytes to complete
2835 * Ends I/O on a number of bytes attached to @rq.
2836 * If @rq has leftover, sets it up for the next range of segments.
2839 * %false - we are done with this request
2840 * %true - still buffers pending for this request
2842 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2844 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2846 EXPORT_SYMBOL(blk_end_request);
2849 * blk_end_request_all - Helper function for drives to finish the request.
2850 * @rq: the request to finish
2851 * @error: %0 for success, < %0 for error
2854 * Completely finish @rq.
2856 void blk_end_request_all(struct request *rq, int error)
2859 unsigned int bidi_bytes = 0;
2861 if (unlikely(blk_bidi_rq(rq)))
2862 bidi_bytes = blk_rq_bytes(rq->next_rq);
2864 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2867 EXPORT_SYMBOL(blk_end_request_all);
2870 * blk_end_request_cur - Helper function to finish the current request chunk.
2871 * @rq: the request to finish the current chunk for
2872 * @error: %0 for success, < %0 for error
2875 * Complete the current consecutively mapped chunk from @rq.
2878 * %false - we are done with this request
2879 * %true - still buffers pending for this request
2881 bool blk_end_request_cur(struct request *rq, int error)
2883 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2885 EXPORT_SYMBOL(blk_end_request_cur);
2888 * blk_end_request_err - Finish a request till the next failure boundary.
2889 * @rq: the request to finish till the next failure boundary for
2890 * @error: must be negative errno
2893 * Complete @rq till the next failure boundary.
2896 * %false - we are done with this request
2897 * %true - still buffers pending for this request
2899 bool blk_end_request_err(struct request *rq, int error)
2901 WARN_ON(error >= 0);
2902 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2904 EXPORT_SYMBOL_GPL(blk_end_request_err);
2907 * __blk_end_request - Helper function for drivers to complete the request.
2908 * @rq: the request being processed
2909 * @error: %0 for success, < %0 for error
2910 * @nr_bytes: number of bytes to complete
2913 * Must be called with queue lock held unlike blk_end_request().
2916 * %false - we are done with this request
2917 * %true - still buffers pending for this request
2919 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2921 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2923 EXPORT_SYMBOL(__blk_end_request);
2926 * __blk_end_request_all - Helper function for drives to finish the request.
2927 * @rq: the request to finish
2928 * @error: %0 for success, < %0 for error
2931 * Completely finish @rq. Must be called with queue lock held.
2933 void __blk_end_request_all(struct request *rq, int error)
2936 unsigned int bidi_bytes = 0;
2938 if (unlikely(blk_bidi_rq(rq)))
2939 bidi_bytes = blk_rq_bytes(rq->next_rq);
2941 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2944 EXPORT_SYMBOL(__blk_end_request_all);
2947 * __blk_end_request_cur - Helper function to finish the current request chunk.
2948 * @rq: the request to finish the current chunk for
2949 * @error: %0 for success, < %0 for error
2952 * Complete the current consecutively mapped chunk from @rq. Must
2953 * be called with queue lock held.
2956 * %false - we are done with this request
2957 * %true - still buffers pending for this request
2959 bool __blk_end_request_cur(struct request *rq, int error)
2961 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2963 EXPORT_SYMBOL(__blk_end_request_cur);
2966 * __blk_end_request_err - Finish a request till the next failure boundary.
2967 * @rq: the request to finish till the next failure boundary for
2968 * @error: must be negative errno
2971 * Complete @rq till the next failure boundary. Must be called
2972 * with queue lock held.
2975 * %false - we are done with this request
2976 * %true - still buffers pending for this request
2978 bool __blk_end_request_err(struct request *rq, int error)
2980 WARN_ON(error >= 0);
2981 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2983 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2985 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2988 req_set_op(rq, bio_op(bio));
2990 if (bio_has_data(bio))
2991 rq->nr_phys_segments = bio_phys_segments(q, bio);
2993 rq->__data_len = bio->bi_iter.bi_size;
2994 rq->bio = rq->biotail = bio;
2997 rq->rq_disk = bio->bi_bdev->bd_disk;
3000 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3002 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3003 * @rq: the request to be flushed
3006 * Flush all pages in @rq.
3008 void rq_flush_dcache_pages(struct request *rq)
3010 struct req_iterator iter;
3011 struct bio_vec bvec;
3013 rq_for_each_segment(bvec, rq, iter)
3014 flush_dcache_page(bvec.bv_page);
3016 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3020 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3021 * @q : the queue of the device being checked
3024 * Check if underlying low-level drivers of a device are busy.
3025 * If the drivers want to export their busy state, they must set own
3026 * exporting function using blk_queue_lld_busy() first.
3028 * Basically, this function is used only by request stacking drivers
3029 * to stop dispatching requests to underlying devices when underlying
3030 * devices are busy. This behavior helps more I/O merging on the queue
3031 * of the request stacking driver and prevents I/O throughput regression
3032 * on burst I/O load.
3035 * 0 - Not busy (The request stacking driver should dispatch request)
3036 * 1 - Busy (The request stacking driver should stop dispatching request)
3038 int blk_lld_busy(struct request_queue *q)
3041 return q->lld_busy_fn(q);
3045 EXPORT_SYMBOL_GPL(blk_lld_busy);
3048 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3049 * @rq: the clone request to be cleaned up
3052 * Free all bios in @rq for a cloned request.
3054 void blk_rq_unprep_clone(struct request *rq)
3058 while ((bio = rq->bio) != NULL) {
3059 rq->bio = bio->bi_next;
3064 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3067 * Copy attributes of the original request to the clone request.
3068 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3070 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3072 dst->cpu = src->cpu;
3073 req_set_op_attrs(dst, req_op(src),
3074 (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
3075 dst->cmd_type = src->cmd_type;
3076 dst->__sector = blk_rq_pos(src);
3077 dst->__data_len = blk_rq_bytes(src);
3078 dst->nr_phys_segments = src->nr_phys_segments;
3079 dst->ioprio = src->ioprio;
3080 dst->extra_len = src->extra_len;
3084 * blk_rq_prep_clone - Helper function to setup clone request
3085 * @rq: the request to be setup
3086 * @rq_src: original request to be cloned
3087 * @bs: bio_set that bios for clone are allocated from
3088 * @gfp_mask: memory allocation mask for bio
3089 * @bio_ctr: setup function to be called for each clone bio.
3090 * Returns %0 for success, non %0 for failure.
3091 * @data: private data to be passed to @bio_ctr
3094 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3095 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3096 * are not copied, and copying such parts is the caller's responsibility.
3097 * Also, pages which the original bios are pointing to are not copied
3098 * and the cloned bios just point same pages.
3099 * So cloned bios must be completed before original bios, which means
3100 * the caller must complete @rq before @rq_src.
3102 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3103 struct bio_set *bs, gfp_t gfp_mask,
3104 int (*bio_ctr)(struct bio *, struct bio *, void *),
3107 struct bio *bio, *bio_src;
3112 __rq_for_each_bio(bio_src, rq_src) {
3113 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3117 if (bio_ctr && bio_ctr(bio, bio_src, data))
3121 rq->biotail->bi_next = bio;
3124 rq->bio = rq->biotail = bio;
3127 __blk_rq_prep_clone(rq, rq_src);
3134 blk_rq_unprep_clone(rq);
3138 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3140 int kblockd_schedule_work(struct work_struct *work)
3142 return queue_work(kblockd_workqueue, work);
3144 EXPORT_SYMBOL(kblockd_schedule_work);
3146 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3147 unsigned long delay)
3149 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3151 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3153 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3154 unsigned long delay)
3156 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3158 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3161 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3162 * @plug: The &struct blk_plug that needs to be initialized
3165 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3166 * pending I/O should the task end up blocking between blk_start_plug() and
3167 * blk_finish_plug(). This is important from a performance perspective, but
3168 * also ensures that we don't deadlock. For instance, if the task is blocking
3169 * for a memory allocation, memory reclaim could end up wanting to free a
3170 * page belonging to that request that is currently residing in our private
3171 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3172 * this kind of deadlock.
3174 void blk_start_plug(struct blk_plug *plug)
3176 struct task_struct *tsk = current;
3179 * If this is a nested plug, don't actually assign it.
3184 INIT_LIST_HEAD(&plug->list);
3185 INIT_LIST_HEAD(&plug->mq_list);
3186 INIT_LIST_HEAD(&plug->cb_list);
3188 * Store ordering should not be needed here, since a potential
3189 * preempt will imply a full memory barrier
3193 EXPORT_SYMBOL(blk_start_plug);
3195 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3197 struct request *rqa = container_of(a, struct request, queuelist);
3198 struct request *rqb = container_of(b, struct request, queuelist);
3200 return !(rqa->q < rqb->q ||
3201 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3205 * If 'from_schedule' is true, then postpone the dispatch of requests
3206 * until a safe kblockd context. We due this to avoid accidental big
3207 * additional stack usage in driver dispatch, in places where the originally
3208 * plugger did not intend it.
3210 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3212 __releases(q->queue_lock)
3214 trace_block_unplug(q, depth, !from_schedule);
3217 blk_run_queue_async(q);
3220 spin_unlock(q->queue_lock);
3223 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3225 LIST_HEAD(callbacks);
3227 while (!list_empty(&plug->cb_list)) {
3228 list_splice_init(&plug->cb_list, &callbacks);
3230 while (!list_empty(&callbacks)) {
3231 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3234 list_del(&cb->list);
3235 cb->callback(cb, from_schedule);
3240 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3243 struct blk_plug *plug = current->plug;
3244 struct blk_plug_cb *cb;
3249 list_for_each_entry(cb, &plug->cb_list, list)
3250 if (cb->callback == unplug && cb->data == data)
3253 /* Not currently on the callback list */
3254 BUG_ON(size < sizeof(*cb));
3255 cb = kzalloc(size, GFP_ATOMIC);
3258 cb->callback = unplug;
3259 list_add(&cb->list, &plug->cb_list);
3263 EXPORT_SYMBOL(blk_check_plugged);
3265 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3267 struct request_queue *q;
3268 unsigned long flags;
3273 flush_plug_callbacks(plug, from_schedule);
3275 if (!list_empty(&plug->mq_list))
3276 blk_mq_flush_plug_list(plug, from_schedule);
3278 if (list_empty(&plug->list))
3281 list_splice_init(&plug->list, &list);
3283 list_sort(NULL, &list, plug_rq_cmp);
3289 * Save and disable interrupts here, to avoid doing it for every
3290 * queue lock we have to take.
3292 local_irq_save(flags);
3293 while (!list_empty(&list)) {
3294 rq = list_entry_rq(list.next);
3295 list_del_init(&rq->queuelist);
3299 * This drops the queue lock
3302 queue_unplugged(q, depth, from_schedule);
3305 spin_lock(q->queue_lock);
3309 * Short-circuit if @q is dead
3311 if (unlikely(blk_queue_dying(q))) {
3312 __blk_end_request_all(rq, -ENODEV);
3317 * rq is already accounted, so use raw insert
3319 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3320 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3322 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3328 * This drops the queue lock
3331 queue_unplugged(q, depth, from_schedule);
3333 local_irq_restore(flags);
3336 void blk_finish_plug(struct blk_plug *plug)
3338 if (plug != current->plug)
3340 blk_flush_plug_list(plug, false);
3342 current->plug = NULL;
3344 EXPORT_SYMBOL(blk_finish_plug);
3346 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3348 struct blk_plug *plug;
3351 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3352 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3355 plug = current->plug;
3357 blk_flush_plug_list(plug, false);
3359 state = current->state;
3360 while (!need_resched()) {
3361 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3362 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3365 hctx->poll_invoked++;
3367 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3369 hctx->poll_success++;
3370 set_current_state(TASK_RUNNING);
3374 if (signal_pending_state(state, current))
3375 set_current_state(TASK_RUNNING);
3377 if (current->state == TASK_RUNNING)
3389 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3390 * @q: the queue of the device
3391 * @dev: the device the queue belongs to
3394 * Initialize runtime-PM-related fields for @q and start auto suspend for
3395 * @dev. Drivers that want to take advantage of request-based runtime PM
3396 * should call this function after @dev has been initialized, and its
3397 * request queue @q has been allocated, and runtime PM for it can not happen
3398 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3399 * cases, driver should call this function before any I/O has taken place.
3401 * This function takes care of setting up using auto suspend for the device,
3402 * the autosuspend delay is set to -1 to make runtime suspend impossible
3403 * until an updated value is either set by user or by driver. Drivers do
3404 * not need to touch other autosuspend settings.
3406 * The block layer runtime PM is request based, so only works for drivers
3407 * that use request as their IO unit instead of those directly use bio's.
3409 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3412 q->rpm_status = RPM_ACTIVE;
3413 pm_runtime_set_autosuspend_delay(q->dev, -1);
3414 pm_runtime_use_autosuspend(q->dev);
3416 EXPORT_SYMBOL(blk_pm_runtime_init);
3419 * blk_pre_runtime_suspend - Pre runtime suspend check
3420 * @q: the queue of the device
3423 * This function will check if runtime suspend is allowed for the device
3424 * by examining if there are any requests pending in the queue. If there
3425 * are requests pending, the device can not be runtime suspended; otherwise,
3426 * the queue's status will be updated to SUSPENDING and the driver can
3427 * proceed to suspend the device.
3429 * For the not allowed case, we mark last busy for the device so that
3430 * runtime PM core will try to autosuspend it some time later.
3432 * This function should be called near the start of the device's
3433 * runtime_suspend callback.
3436 * 0 - OK to runtime suspend the device
3437 * -EBUSY - Device should not be runtime suspended
3439 int blk_pre_runtime_suspend(struct request_queue *q)
3446 spin_lock_irq(q->queue_lock);
3447 if (q->nr_pending) {
3449 pm_runtime_mark_last_busy(q->dev);
3451 q->rpm_status = RPM_SUSPENDING;
3453 spin_unlock_irq(q->queue_lock);
3456 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3459 * blk_post_runtime_suspend - Post runtime suspend processing
3460 * @q: the queue of the device
3461 * @err: return value of the device's runtime_suspend function
3464 * Update the queue's runtime status according to the return value of the
3465 * device's runtime suspend function and mark last busy for the device so
3466 * that PM core will try to auto suspend the device at a later time.
3468 * This function should be called near the end of the device's
3469 * runtime_suspend callback.
3471 void blk_post_runtime_suspend(struct request_queue *q, int err)
3476 spin_lock_irq(q->queue_lock);
3478 q->rpm_status = RPM_SUSPENDED;
3480 q->rpm_status = RPM_ACTIVE;
3481 pm_runtime_mark_last_busy(q->dev);
3483 spin_unlock_irq(q->queue_lock);
3485 EXPORT_SYMBOL(blk_post_runtime_suspend);
3488 * blk_pre_runtime_resume - Pre runtime resume processing
3489 * @q: the queue of the device
3492 * Update the queue's runtime status to RESUMING in preparation for the
3493 * runtime resume of the device.
3495 * This function should be called near the start of the device's
3496 * runtime_resume callback.
3498 void blk_pre_runtime_resume(struct request_queue *q)
3503 spin_lock_irq(q->queue_lock);
3504 q->rpm_status = RPM_RESUMING;
3505 spin_unlock_irq(q->queue_lock);
3507 EXPORT_SYMBOL(blk_pre_runtime_resume);
3510 * blk_post_runtime_resume - Post runtime resume processing
3511 * @q: the queue of the device
3512 * @err: return value of the device's runtime_resume function
3515 * Update the queue's runtime status according to the return value of the
3516 * device's runtime_resume function. If it is successfully resumed, process
3517 * the requests that are queued into the device's queue when it is resuming
3518 * and then mark last busy and initiate autosuspend for it.
3520 * This function should be called near the end of the device's
3521 * runtime_resume callback.
3523 void blk_post_runtime_resume(struct request_queue *q, int err)
3528 spin_lock_irq(q->queue_lock);
3530 q->rpm_status = RPM_ACTIVE;
3532 pm_runtime_mark_last_busy(q->dev);
3533 pm_request_autosuspend(q->dev);
3535 q->rpm_status = RPM_SUSPENDED;
3537 spin_unlock_irq(q->queue_lock);
3539 EXPORT_SYMBOL(blk_post_runtime_resume);
3542 * blk_set_runtime_active - Force runtime status of the queue to be active
3543 * @q: the queue of the device
3545 * If the device is left runtime suspended during system suspend the resume
3546 * hook typically resumes the device and corrects runtime status
3547 * accordingly. However, that does not affect the queue runtime PM status
3548 * which is still "suspended". This prevents processing requests from the
3551 * This function can be used in driver's resume hook to correct queue
3552 * runtime PM status and re-enable peeking requests from the queue. It
3553 * should be called before first request is added to the queue.
3555 void blk_set_runtime_active(struct request_queue *q)
3557 spin_lock_irq(q->queue_lock);
3558 q->rpm_status = RPM_ACTIVE;
3559 pm_runtime_mark_last_busy(q->dev);
3560 pm_request_autosuspend(q->dev);
3561 spin_unlock_irq(q->queue_lock);
3563 EXPORT_SYMBOL(blk_set_runtime_active);
3566 int __init blk_dev_init(void)
3568 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3569 FIELD_SIZEOF(struct request, cmd_flags));
3571 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3572 kblockd_workqueue = alloc_workqueue("kblockd",
3573 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3574 if (!kblockd_workqueue)
3575 panic("Failed to create kblockd\n");
3577 request_cachep = kmem_cache_create("blkdev_requests",
3578 sizeof(struct request), 0, SLAB_PANIC, NULL);
3580 blk_requestq_cachep = kmem_cache_create("request_queue",
3581 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);