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 #include <linux/math64.h>
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
48 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
49 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
51 DEFINE_IDA(blk_queue_ida);
54 * For the allocated request tables
56 struct kmem_cache *request_cachep = NULL;
59 * For queue allocation
61 struct kmem_cache *blk_requestq_cachep;
64 * Controlling structure to kblockd
66 static struct workqueue_struct *kblockd_workqueue;
68 static void blk_clear_congested(struct request_list *rl, int sync)
70 #ifdef CONFIG_CGROUP_WRITEBACK
71 clear_wb_congested(rl->blkg->wb_congested, sync);
74 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
75 * flip its congestion state for events on other blkcgs.
77 if (rl == &rl->q->root_rl)
78 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
82 static void blk_set_congested(struct request_list *rl, int sync)
84 #ifdef CONFIG_CGROUP_WRITEBACK
85 set_wb_congested(rl->blkg->wb_congested, sync);
87 /* see blk_clear_congested() */
88 if (rl == &rl->q->root_rl)
89 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
93 void blk_queue_congestion_threshold(struct request_queue *q)
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
100 q->nr_congestion_on = nr;
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 q->nr_congestion_off = nr;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info. This function can only be called if @bdev is opened
114 * and the return value is never NULL.
116 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
118 struct request_queue *q = bdev_get_queue(bdev);
120 return &q->backing_dev_info;
122 EXPORT_SYMBOL(blk_get_backing_dev_info);
124 void blk_rq_init(struct request_queue *q, struct request *rq)
126 memset(rq, 0, sizeof(*rq));
128 INIT_LIST_HEAD(&rq->queuelist);
129 INIT_LIST_HEAD(&rq->timeout_list);
132 rq->__sector = (sector_t) -1;
133 INIT_HLIST_NODE(&rq->hash);
134 RB_CLEAR_NODE(&rq->rb_node);
136 rq->cmd_len = BLK_MAX_CDB;
138 rq->start_time = jiffies;
139 set_start_time_ns(rq);
142 EXPORT_SYMBOL(blk_rq_init);
144 static void req_bio_endio(struct request *rq, struct bio *bio,
145 unsigned int nbytes, int error)
148 bio->bi_error = error;
150 if (unlikely(rq->cmd_flags & REQ_QUIET))
151 bio_set_flag(bio, BIO_QUIET);
153 bio_advance(bio, nbytes);
155 /* don't actually finish bio if it's part of flush sequence */
156 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
160 void blk_dump_rq_flags(struct request *rq, char *msg)
164 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
165 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
166 (unsigned long long) rq->cmd_flags);
168 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
169 (unsigned long long)blk_rq_pos(rq),
170 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
171 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
172 rq->bio, rq->biotail, blk_rq_bytes(rq));
174 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
175 printk(KERN_INFO " cdb: ");
176 for (bit = 0; bit < BLK_MAX_CDB; bit++)
177 printk("%02x ", rq->cmd[bit]);
181 EXPORT_SYMBOL(blk_dump_rq_flags);
183 static void blk_delay_work(struct work_struct *work)
185 struct request_queue *q;
187 q = container_of(work, struct request_queue, delay_work.work);
188 spin_lock_irq(q->queue_lock);
190 spin_unlock_irq(q->queue_lock);
194 * blk_delay_queue - restart queueing after defined interval
195 * @q: The &struct request_queue in question
196 * @msecs: Delay in msecs
199 * Sometimes queueing needs to be postponed for a little while, to allow
200 * resources to come back. This function will make sure that queueing is
201 * restarted around the specified time. Queue lock must be held.
203 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
205 if (likely(!blk_queue_dead(q)))
206 queue_delayed_work(kblockd_workqueue, &q->delay_work,
207 msecs_to_jiffies(msecs));
209 EXPORT_SYMBOL(blk_delay_queue);
212 * blk_start_queue_async - asynchronously restart a previously stopped queue
213 * @q: The &struct request_queue in question
216 * blk_start_queue_async() will clear the stop flag on the queue, and
217 * ensure that the request_fn for the queue is run from an async
220 void blk_start_queue_async(struct request_queue *q)
222 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
223 blk_run_queue_async(q);
225 EXPORT_SYMBOL(blk_start_queue_async);
228 * blk_start_queue - restart a previously stopped queue
229 * @q: The &struct request_queue in question
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 void blk_start_queue(struct request_queue *q)
238 WARN_ON(!in_interrupt() && !irqs_disabled());
240 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
243 EXPORT_SYMBOL(blk_start_queue);
246 * blk_stop_queue - stop a queue
247 * @q: The &struct request_queue in question
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 void blk_stop_queue(struct request_queue *q)
261 cancel_delayed_work(&q->delay_work);
262 queue_flag_set(QUEUE_FLAG_STOPPED, q);
264 EXPORT_SYMBOL(blk_stop_queue);
267 * blk_sync_queue - cancel any pending callbacks on a queue
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
275 * that the callbacks might use. The caller must already have made sure
276 * that its ->make_request_fn will not re-add plugging prior to calling
279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevator_exit()
281 * and blkcg_exit_queue() to be called with queue lock initialized.
284 void blk_sync_queue(struct request_queue *q)
286 del_timer_sync(&q->timeout);
289 struct blk_mq_hw_ctx *hctx;
292 queue_for_each_hw_ctx(q, hctx, i) {
293 cancel_delayed_work_sync(&hctx->run_work);
294 cancel_delayed_work_sync(&hctx->delay_work);
297 cancel_delayed_work_sync(&q->delay_work);
300 EXPORT_SYMBOL(blk_sync_queue);
303 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
304 * @q: The queue to run
307 * Invoke request handling on a queue if there are any pending requests.
308 * May be used to restart request handling after a request has completed.
309 * This variant runs the queue whether or not the queue has been
310 * stopped. Must be called with the queue lock held and interrupts
311 * disabled. See also @blk_run_queue.
313 inline void __blk_run_queue_uncond(struct request_queue *q)
315 if (unlikely(blk_queue_dead(q)))
319 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
320 * the queue lock internally. As a result multiple threads may be
321 * running such a request function concurrently. Keep track of the
322 * number of active request_fn invocations such that blk_drain_queue()
323 * can wait until all these request_fn calls have finished.
325 q->request_fn_active++;
327 q->request_fn_active--;
329 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
332 * __blk_run_queue - run a single device queue
333 * @q: The queue to run
336 * See @blk_run_queue. This variant must be called with the queue lock
337 * held and interrupts disabled.
339 void __blk_run_queue(struct request_queue *q)
341 if (unlikely(blk_queue_stopped(q)))
344 __blk_run_queue_uncond(q);
346 EXPORT_SYMBOL(__blk_run_queue);
349 * blk_run_queue_async - run a single device queue in workqueue context
350 * @q: The queue to run
353 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
354 * of us. The caller must hold the queue lock.
356 void blk_run_queue_async(struct request_queue *q)
358 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
359 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
361 EXPORT_SYMBOL(blk_run_queue_async);
364 * blk_run_queue - run a single device queue
365 * @q: The queue to run
368 * Invoke request handling on this queue, if it has pending work to do.
369 * May be used to restart queueing when a request has completed.
371 void blk_run_queue(struct request_queue *q)
375 spin_lock_irqsave(q->queue_lock, flags);
377 spin_unlock_irqrestore(q->queue_lock, flags);
379 EXPORT_SYMBOL(blk_run_queue);
381 void blk_put_queue(struct request_queue *q)
383 kobject_put(&q->kobj);
385 EXPORT_SYMBOL(blk_put_queue);
388 * __blk_drain_queue - drain requests from request_queue
390 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
392 * Drain requests from @q. If @drain_all is set, all requests are drained.
393 * If not, only ELVPRIV requests are drained. The caller is responsible
394 * for ensuring that no new requests which need to be drained are queued.
396 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
397 __releases(q->queue_lock)
398 __acquires(q->queue_lock)
402 lockdep_assert_held(q->queue_lock);
408 * The caller might be trying to drain @q before its
409 * elevator is initialized.
412 elv_drain_elevator(q);
414 blkcg_drain_queue(q);
417 * This function might be called on a queue which failed
418 * driver init after queue creation or is not yet fully
419 * active yet. Some drivers (e.g. fd and loop) get unhappy
420 * in such cases. Kick queue iff dispatch queue has
421 * something on it and @q has request_fn set.
423 if (!list_empty(&q->queue_head) && q->request_fn)
426 drain |= q->nr_rqs_elvpriv;
427 drain |= q->request_fn_active;
430 * Unfortunately, requests are queued at and tracked from
431 * multiple places and there's no single counter which can
432 * be drained. Check all the queues and counters.
435 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
436 drain |= !list_empty(&q->queue_head);
437 for (i = 0; i < 2; i++) {
438 drain |= q->nr_rqs[i];
439 drain |= q->in_flight[i];
441 drain |= !list_empty(&fq->flush_queue[i]);
448 spin_unlock_irq(q->queue_lock);
452 spin_lock_irq(q->queue_lock);
456 * With queue marked dead, any woken up waiter will fail the
457 * allocation path, so the wakeup chaining is lost and we're
458 * left with hung waiters. We need to wake up those waiters.
461 struct request_list *rl;
463 blk_queue_for_each_rl(rl, q)
464 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
465 wake_up_all(&rl->wait[i]);
470 * blk_queue_bypass_start - enter queue bypass mode
471 * @q: queue of interest
473 * In bypass mode, only the dispatch FIFO queue of @q is used. This
474 * function makes @q enter bypass mode and drains all requests which were
475 * throttled or issued before. On return, it's guaranteed that no request
476 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
477 * inside queue or RCU read lock.
479 void blk_queue_bypass_start(struct request_queue *q)
481 spin_lock_irq(q->queue_lock);
483 queue_flag_set(QUEUE_FLAG_BYPASS, q);
484 spin_unlock_irq(q->queue_lock);
487 * Queues start drained. Skip actual draining till init is
488 * complete. This avoids lenghty delays during queue init which
489 * can happen many times during boot.
491 if (blk_queue_init_done(q)) {
492 spin_lock_irq(q->queue_lock);
493 __blk_drain_queue(q, false);
494 spin_unlock_irq(q->queue_lock);
496 /* ensure blk_queue_bypass() is %true inside RCU read lock */
500 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
503 * blk_queue_bypass_end - leave queue bypass mode
504 * @q: queue of interest
506 * Leave bypass mode and restore the normal queueing behavior.
508 void blk_queue_bypass_end(struct request_queue *q)
510 spin_lock_irq(q->queue_lock);
511 if (!--q->bypass_depth)
512 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
513 WARN_ON_ONCE(q->bypass_depth < 0);
514 spin_unlock_irq(q->queue_lock);
516 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
518 void blk_set_queue_dying(struct request_queue *q)
520 spin_lock_irq(q->queue_lock);
521 queue_flag_set(QUEUE_FLAG_DYING, q);
522 spin_unlock_irq(q->queue_lock);
525 blk_mq_wake_waiters(q);
527 struct request_list *rl;
529 blk_queue_for_each_rl(rl, q) {
531 wake_up_all(&rl->wait[BLK_RW_SYNC]);
532 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
537 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
540 * blk_cleanup_queue - shutdown a request queue
541 * @q: request queue to shutdown
543 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
544 * put it. All future requests will be failed immediately with -ENODEV.
546 void blk_cleanup_queue(struct request_queue *q)
548 spinlock_t *lock = q->queue_lock;
550 /* mark @q DYING, no new request or merges will be allowed afterwards */
551 mutex_lock(&q->sysfs_lock);
552 blk_set_queue_dying(q);
556 * A dying queue is permanently in bypass mode till released. Note
557 * that, unlike blk_queue_bypass_start(), we aren't performing
558 * synchronize_rcu() after entering bypass mode to avoid the delay
559 * as some drivers create and destroy a lot of queues while
560 * probing. This is still safe because blk_release_queue() will be
561 * called only after the queue refcnt drops to zero and nothing,
562 * RCU or not, would be traversing the queue by then.
565 queue_flag_set(QUEUE_FLAG_BYPASS, q);
567 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
568 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
569 queue_flag_set(QUEUE_FLAG_DYING, q);
570 spin_unlock_irq(lock);
571 mutex_unlock(&q->sysfs_lock);
574 * Drain all requests queued before DYING marking. Set DEAD flag to
575 * prevent that q->request_fn() gets invoked after draining finished.
580 __blk_drain_queue(q, true);
581 queue_flag_set(QUEUE_FLAG_DEAD, q);
582 spin_unlock_irq(lock);
584 /* for synchronous bio-based driver finish in-flight integrity i/o */
585 blk_flush_integrity();
587 /* @q won't process any more request, flush async actions */
588 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
592 blk_mq_free_queue(q);
593 percpu_ref_exit(&q->q_usage_counter);
596 if (q->queue_lock != &q->__queue_lock)
597 q->queue_lock = &q->__queue_lock;
598 spin_unlock_irq(lock);
600 bdi_unregister(&q->backing_dev_info);
602 /* @q is and will stay empty, shutdown and put */
605 EXPORT_SYMBOL(blk_cleanup_queue);
607 /* Allocate memory local to the request queue */
608 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
610 int nid = (int)(long)data;
611 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
614 static void free_request_struct(void *element, void *unused)
616 kmem_cache_free(request_cachep, element);
619 int blk_init_rl(struct request_list *rl, struct request_queue *q,
622 if (unlikely(rl->rq_pool))
626 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
627 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
628 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
629 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
631 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
633 (void *)(long)q->node, gfp_mask,
641 void blk_exit_rl(struct request_list *rl)
644 mempool_destroy(rl->rq_pool);
647 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
649 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
651 EXPORT_SYMBOL(blk_alloc_queue);
653 int blk_queue_enter(struct request_queue *q, gfp_t gfp)
656 if (percpu_ref_tryget_live(&q->q_usage_counter))
659 if (!gfpflags_allow_blocking(gfp))
662 wait_event(q->mq_freeze_wq,
663 !atomic_read(&q->mq_freeze_depth) ||
665 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 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
685 struct request_queue *q;
688 q = kmem_cache_alloc_node(blk_requestq_cachep,
689 gfp_mask | __GFP_ZERO, node_id);
693 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
697 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
701 q->backing_dev_info.ra_pages =
702 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
703 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
704 q->backing_dev_info.name = "block";
707 err = bdi_init(&q->backing_dev_info);
711 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
712 laptop_mode_timer_fn, (unsigned long) q);
713 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
714 INIT_LIST_HEAD(&q->queue_head);
715 INIT_LIST_HEAD(&q->timeout_list);
716 INIT_LIST_HEAD(&q->icq_list);
717 #ifdef CONFIG_BLK_CGROUP
718 INIT_LIST_HEAD(&q->blkg_list);
720 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
722 kobject_init(&q->kobj, &blk_queue_ktype);
724 mutex_init(&q->sysfs_lock);
725 spin_lock_init(&q->__queue_lock);
728 * By default initialize queue_lock to internal lock and driver can
729 * override it later if need be.
731 q->queue_lock = &q->__queue_lock;
734 * A queue starts its life with bypass turned on to avoid
735 * unnecessary bypass on/off overhead and nasty surprises during
736 * init. The initial bypass will be finished when the queue is
737 * registered by blk_register_queue().
740 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
742 init_waitqueue_head(&q->mq_freeze_wq);
745 * Init percpu_ref in atomic mode so that it's faster to shutdown.
746 * See blk_register_queue() for details.
748 if (percpu_ref_init(&q->q_usage_counter,
749 blk_queue_usage_counter_release,
750 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
753 if (blkcg_init_queue(q))
759 percpu_ref_exit(&q->q_usage_counter);
761 bdi_destroy(&q->backing_dev_info);
763 bioset_free(q->bio_split);
765 ida_simple_remove(&blk_queue_ida, q->id);
767 kmem_cache_free(blk_requestq_cachep, q);
770 EXPORT_SYMBOL(blk_alloc_queue_node);
773 * blk_init_queue - prepare a request queue for use with a block device
774 * @rfn: The function to be called to process requests that have been
775 * placed on the queue.
776 * @lock: Request queue spin lock
779 * If a block device wishes to use the standard request handling procedures,
780 * which sorts requests and coalesces adjacent requests, then it must
781 * call blk_init_queue(). The function @rfn will be called when there
782 * are requests on the queue that need to be processed. If the device
783 * supports plugging, then @rfn may not be called immediately when requests
784 * are available on the queue, but may be called at some time later instead.
785 * Plugged queues are generally unplugged when a buffer belonging to one
786 * of the requests on the queue is needed, or due to memory pressure.
788 * @rfn is not required, or even expected, to remove all requests off the
789 * queue, but only as many as it can handle at a time. If it does leave
790 * requests on the queue, it is responsible for arranging that the requests
791 * get dealt with eventually.
793 * The queue spin lock must be held while manipulating the requests on the
794 * request queue; this lock will be taken also from interrupt context, so irq
795 * disabling is needed for it.
797 * Function returns a pointer to the initialized request queue, or %NULL if
801 * blk_init_queue() must be paired with a blk_cleanup_queue() call
802 * when the block device is deactivated (such as at module unload).
805 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
807 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
809 EXPORT_SYMBOL(blk_init_queue);
811 struct request_queue *
812 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
814 struct request_queue *uninit_q, *q;
816 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
820 q = blk_init_allocated_queue(uninit_q, rfn, lock);
822 blk_cleanup_queue(uninit_q);
826 EXPORT_SYMBOL(blk_init_queue_node);
828 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
830 struct request_queue *
831 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
837 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
841 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
845 q->prep_rq_fn = NULL;
846 q->unprep_rq_fn = NULL;
847 q->queue_flags |= QUEUE_FLAG_DEFAULT;
849 /* Override internal queue lock with supplied lock pointer */
851 q->queue_lock = lock;
854 * This also sets hw/phys segments, boundary and size
856 blk_queue_make_request(q, blk_queue_bio);
858 q->sg_reserved_size = INT_MAX;
860 /* Protect q->elevator from elevator_change */
861 mutex_lock(&q->sysfs_lock);
864 if (elevator_init(q, NULL)) {
865 mutex_unlock(&q->sysfs_lock);
869 mutex_unlock(&q->sysfs_lock);
874 blk_free_flush_queue(q->fq);
877 EXPORT_SYMBOL(blk_init_allocated_queue);
879 bool blk_get_queue(struct request_queue *q)
881 if (likely(!blk_queue_dying(q))) {
888 EXPORT_SYMBOL(blk_get_queue);
890 static inline void blk_free_request(struct request_list *rl, struct request *rq)
892 if (rq->cmd_flags & REQ_ELVPRIV) {
893 elv_put_request(rl->q, rq);
895 put_io_context(rq->elv.icq->ioc);
898 mempool_free(rq, rl->rq_pool);
902 * ioc_batching returns true if the ioc is a valid batching request and
903 * should be given priority access to a request.
905 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
911 * Make sure the process is able to allocate at least 1 request
912 * even if the batch times out, otherwise we could theoretically
915 return ioc->nr_batch_requests == q->nr_batching ||
916 (ioc->nr_batch_requests > 0
917 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
921 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
922 * will cause the process to be a "batcher" on all queues in the system. This
923 * is the behaviour we want though - once it gets a wakeup it should be given
926 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
928 if (!ioc || ioc_batching(q, ioc))
931 ioc->nr_batch_requests = q->nr_batching;
932 ioc->last_waited = jiffies;
935 static void __freed_request(struct request_list *rl, int sync)
937 struct request_queue *q = rl->q;
939 if (rl->count[sync] < queue_congestion_off_threshold(q))
940 blk_clear_congested(rl, sync);
942 if (rl->count[sync] + 1 <= q->nr_requests) {
943 if (waitqueue_active(&rl->wait[sync]))
944 wake_up(&rl->wait[sync]);
946 blk_clear_rl_full(rl, sync);
951 * A request has just been released. Account for it, update the full and
952 * congestion status, wake up any waiters. Called under q->queue_lock.
954 static void freed_request(struct request_list *rl, unsigned int flags)
956 struct request_queue *q = rl->q;
957 int sync = rw_is_sync(flags);
961 if (flags & REQ_ELVPRIV)
964 __freed_request(rl, sync);
966 if (unlikely(rl->starved[sync ^ 1]))
967 __freed_request(rl, sync ^ 1);
970 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
972 struct request_list *rl;
973 int on_thresh, off_thresh;
975 spin_lock_irq(q->queue_lock);
977 blk_queue_congestion_threshold(q);
978 on_thresh = queue_congestion_on_threshold(q);
979 off_thresh = queue_congestion_off_threshold(q);
981 blk_queue_for_each_rl(rl, q) {
982 if (rl->count[BLK_RW_SYNC] >= on_thresh)
983 blk_set_congested(rl, BLK_RW_SYNC);
984 else if (rl->count[BLK_RW_SYNC] < off_thresh)
985 blk_clear_congested(rl, BLK_RW_SYNC);
987 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
988 blk_set_congested(rl, BLK_RW_ASYNC);
989 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
990 blk_clear_congested(rl, BLK_RW_ASYNC);
992 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
993 blk_set_rl_full(rl, BLK_RW_SYNC);
995 blk_clear_rl_full(rl, BLK_RW_SYNC);
996 wake_up(&rl->wait[BLK_RW_SYNC]);
999 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1000 blk_set_rl_full(rl, BLK_RW_ASYNC);
1002 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1003 wake_up(&rl->wait[BLK_RW_ASYNC]);
1007 spin_unlock_irq(q->queue_lock);
1012 * Determine if elevator data should be initialized when allocating the
1013 * request associated with @bio.
1015 static bool blk_rq_should_init_elevator(struct bio *bio)
1021 * Flush requests do not use the elevator so skip initialization.
1022 * This allows a request to share the flush and elevator data.
1024 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1031 * rq_ioc - determine io_context for request allocation
1032 * @bio: request being allocated is for this bio (can be %NULL)
1034 * Determine io_context to use for request allocation for @bio. May return
1035 * %NULL if %current->io_context doesn't exist.
1037 static struct io_context *rq_ioc(struct bio *bio)
1039 #ifdef CONFIG_BLK_CGROUP
1040 if (bio && bio->bi_ioc)
1043 return current->io_context;
1047 * __get_request - get a free request
1048 * @rl: request list to allocate from
1049 * @rw_flags: RW and SYNC flags
1050 * @bio: bio to allocate request for (can be %NULL)
1051 * @gfp_mask: allocation mask
1053 * Get a free request from @q. This function may fail under memory
1054 * pressure or if @q is dead.
1056 * Must be called with @q->queue_lock held and,
1057 * Returns ERR_PTR on failure, with @q->queue_lock held.
1058 * Returns request pointer on success, with @q->queue_lock *not held*.
1060 static struct request *__get_request(struct request_list *rl, int rw_flags,
1061 struct bio *bio, gfp_t gfp_mask)
1063 struct request_queue *q = rl->q;
1065 struct elevator_type *et = q->elevator->type;
1066 struct io_context *ioc = rq_ioc(bio);
1067 struct io_cq *icq = NULL;
1068 const bool is_sync = rw_is_sync(rw_flags) != 0;
1071 if (unlikely(blk_queue_dying(q)))
1072 return ERR_PTR(-ENODEV);
1074 may_queue = elv_may_queue(q, rw_flags);
1075 if (may_queue == ELV_MQUEUE_NO)
1078 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1079 if (rl->count[is_sync]+1 >= q->nr_requests) {
1081 * The queue will fill after this allocation, so set
1082 * it as full, and mark this process as "batching".
1083 * This process will be allowed to complete a batch of
1084 * requests, others will be blocked.
1086 if (!blk_rl_full(rl, is_sync)) {
1087 ioc_set_batching(q, ioc);
1088 blk_set_rl_full(rl, is_sync);
1090 if (may_queue != ELV_MQUEUE_MUST
1091 && !ioc_batching(q, ioc)) {
1093 * The queue is full and the allocating
1094 * process is not a "batcher", and not
1095 * exempted by the IO scheduler
1097 return ERR_PTR(-ENOMEM);
1101 blk_set_congested(rl, is_sync);
1105 * Only allow batching queuers to allocate up to 50% over the defined
1106 * limit of requests, otherwise we could have thousands of requests
1107 * allocated with any setting of ->nr_requests
1109 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1110 return ERR_PTR(-ENOMEM);
1112 q->nr_rqs[is_sync]++;
1113 rl->count[is_sync]++;
1114 rl->starved[is_sync] = 0;
1117 * Decide whether the new request will be managed by elevator. If
1118 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1119 * prevent the current elevator from being destroyed until the new
1120 * request is freed. This guarantees icq's won't be destroyed and
1121 * makes creating new ones safe.
1123 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1124 * it will be created after releasing queue_lock.
1126 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1127 rw_flags |= REQ_ELVPRIV;
1128 q->nr_rqs_elvpriv++;
1129 if (et->icq_cache && ioc)
1130 icq = ioc_lookup_icq(ioc, q);
1133 if (blk_queue_io_stat(q))
1134 rw_flags |= REQ_IO_STAT;
1135 spin_unlock_irq(q->queue_lock);
1137 /* allocate and init request */
1138 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1143 blk_rq_set_rl(rq, rl);
1144 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1147 if (rw_flags & REQ_ELVPRIV) {
1148 if (unlikely(et->icq_cache && !icq)) {
1150 icq = ioc_create_icq(ioc, q, gfp_mask);
1156 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1159 /* @rq->elv.icq holds io_context until @rq is freed */
1161 get_io_context(icq->ioc);
1165 * ioc may be NULL here, and ioc_batching will be false. That's
1166 * OK, if the queue is under the request limit then requests need
1167 * not count toward the nr_batch_requests limit. There will always
1168 * be some limit enforced by BLK_BATCH_TIME.
1170 if (ioc_batching(q, ioc))
1171 ioc->nr_batch_requests--;
1173 trace_block_getrq(q, bio, rw_flags & 1);
1178 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1179 * and may fail indefinitely under memory pressure and thus
1180 * shouldn't stall IO. Treat this request as !elvpriv. This will
1181 * disturb iosched and blkcg but weird is bettern than dead.
1183 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1184 __func__, dev_name(q->backing_dev_info.dev));
1186 rq->cmd_flags &= ~REQ_ELVPRIV;
1189 spin_lock_irq(q->queue_lock);
1190 q->nr_rqs_elvpriv--;
1191 spin_unlock_irq(q->queue_lock);
1196 * Allocation failed presumably due to memory. Undo anything we
1197 * might have messed up.
1199 * Allocating task should really be put onto the front of the wait
1200 * queue, but this is pretty rare.
1202 spin_lock_irq(q->queue_lock);
1203 freed_request(rl, rw_flags);
1206 * in the very unlikely event that allocation failed and no
1207 * requests for this direction was pending, mark us starved so that
1208 * freeing of a request in the other direction will notice
1209 * us. another possible fix would be to split the rq mempool into
1213 if (unlikely(rl->count[is_sync] == 0))
1214 rl->starved[is_sync] = 1;
1215 return ERR_PTR(-ENOMEM);
1219 * get_request - get a free request
1220 * @q: request_queue to allocate request from
1221 * @rw_flags: RW and SYNC flags
1222 * @bio: bio to allocate request for (can be %NULL)
1223 * @gfp_mask: allocation mask
1225 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1226 * this function keeps retrying under memory pressure and fails iff @q is dead.
1228 * Must be called with @q->queue_lock held and,
1229 * Returns ERR_PTR on failure, with @q->queue_lock held.
1230 * Returns request pointer on success, with @q->queue_lock *not held*.
1232 static struct request *get_request(struct request_queue *q, int rw_flags,
1233 struct bio *bio, gfp_t gfp_mask)
1235 const bool is_sync = rw_is_sync(rw_flags) != 0;
1237 struct request_list *rl;
1240 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1242 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1246 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1251 /* wait on @rl and retry */
1252 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1253 TASK_UNINTERRUPTIBLE);
1255 trace_block_sleeprq(q, bio, rw_flags & 1);
1257 spin_unlock_irq(q->queue_lock);
1261 * After sleeping, we become a "batching" process and will be able
1262 * to allocate at least one request, and up to a big batch of them
1263 * for a small period time. See ioc_batching, ioc_set_batching
1265 ioc_set_batching(q, current->io_context);
1267 spin_lock_irq(q->queue_lock);
1268 finish_wait(&rl->wait[is_sync], &wait);
1273 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1278 BUG_ON(rw != READ && rw != WRITE);
1280 /* create ioc upfront */
1281 create_io_context(gfp_mask, q->node);
1283 spin_lock_irq(q->queue_lock);
1284 rq = get_request(q, rw, NULL, gfp_mask);
1286 spin_unlock_irq(q->queue_lock);
1287 /* q->queue_lock is unlocked at this point */
1292 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1295 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1297 return blk_old_get_request(q, rw, gfp_mask);
1299 EXPORT_SYMBOL(blk_get_request);
1302 * blk_make_request - given a bio, allocate a corresponding struct request.
1303 * @q: target request queue
1304 * @bio: The bio describing the memory mappings that will be submitted for IO.
1305 * It may be a chained-bio properly constructed by block/bio layer.
1306 * @gfp_mask: gfp flags to be used for memory allocation
1308 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1309 * type commands. Where the struct request needs to be farther initialized by
1310 * the caller. It is passed a &struct bio, which describes the memory info of
1313 * The caller of blk_make_request must make sure that bi_io_vec
1314 * are set to describe the memory buffers. That bio_data_dir() will return
1315 * the needed direction of the request. (And all bio's in the passed bio-chain
1316 * are properly set accordingly)
1318 * If called under none-sleepable conditions, mapped bio buffers must not
1319 * need bouncing, by calling the appropriate masked or flagged allocator,
1320 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1323 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1324 * given to how you allocate bios. In particular, you cannot use
1325 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1326 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1327 * thus resulting in a deadlock. Alternatively bios should be allocated using
1328 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1329 * If possible a big IO should be split into smaller parts when allocation
1330 * fails. Partial allocation should not be an error, or you risk a live-lock.
1332 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1335 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1340 blk_rq_set_block_pc(rq);
1343 struct bio *bounce_bio = bio;
1346 blk_queue_bounce(q, &bounce_bio);
1347 ret = blk_rq_append_bio(q, rq, bounce_bio);
1348 if (unlikely(ret)) {
1349 blk_put_request(rq);
1350 return ERR_PTR(ret);
1356 EXPORT_SYMBOL(blk_make_request);
1359 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1360 * @rq: request to be initialized
1363 void blk_rq_set_block_pc(struct request *rq)
1365 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1367 rq->__sector = (sector_t) -1;
1368 rq->bio = rq->biotail = NULL;
1369 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1371 EXPORT_SYMBOL(blk_rq_set_block_pc);
1374 * blk_requeue_request - put a request back on queue
1375 * @q: request queue where request should be inserted
1376 * @rq: request to be inserted
1379 * Drivers often keep queueing requests until the hardware cannot accept
1380 * more, when that condition happens we need to put the request back
1381 * on the queue. Must be called with queue lock held.
1383 void blk_requeue_request(struct request_queue *q, struct request *rq)
1385 blk_delete_timer(rq);
1386 blk_clear_rq_complete(rq);
1387 trace_block_rq_requeue(q, rq);
1389 if (rq->cmd_flags & REQ_QUEUED)
1390 blk_queue_end_tag(q, rq);
1392 BUG_ON(blk_queued_rq(rq));
1394 elv_requeue_request(q, rq);
1396 EXPORT_SYMBOL(blk_requeue_request);
1398 static void add_acct_request(struct request_queue *q, struct request *rq,
1401 blk_account_io_start(rq, true);
1402 __elv_add_request(q, rq, where);
1405 static void part_round_stats_single(int cpu, struct hd_struct *part,
1410 if (now == part->stamp)
1413 inflight = part_in_flight(part);
1415 __part_stat_add(cpu, part, time_in_queue,
1416 inflight * (now - part->stamp));
1417 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1423 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1424 * @cpu: cpu number for stats access
1425 * @part: target partition
1427 * The average IO queue length and utilisation statistics are maintained
1428 * by observing the current state of the queue length and the amount of
1429 * time it has been in this state for.
1431 * Normally, that accounting is done on IO completion, but that can result
1432 * in more than a second's worth of IO being accounted for within any one
1433 * second, leading to >100% utilisation. To deal with that, we call this
1434 * function to do a round-off before returning the results when reading
1435 * /proc/diskstats. This accounts immediately for all queue usage up to
1436 * the current jiffies and restarts the counters again.
1438 void part_round_stats(int cpu, struct hd_struct *part)
1440 unsigned long now = jiffies;
1443 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1444 part_round_stats_single(cpu, part, now);
1446 EXPORT_SYMBOL_GPL(part_round_stats);
1449 static void blk_pm_put_request(struct request *rq)
1451 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1452 pm_runtime_mark_last_busy(rq->q->dev);
1455 static inline void blk_pm_put_request(struct request *rq) {}
1459 * queue lock must be held
1461 void __blk_put_request(struct request_queue *q, struct request *req)
1467 blk_mq_free_request(req);
1471 blk_pm_put_request(req);
1473 elv_completed_request(q, req);
1475 /* this is a bio leak */
1476 WARN_ON(req->bio != NULL);
1479 * Request may not have originated from ll_rw_blk. if not,
1480 * it didn't come out of our reserved rq pools
1482 if (req->cmd_flags & REQ_ALLOCED) {
1483 unsigned int flags = req->cmd_flags;
1484 struct request_list *rl = blk_rq_rl(req);
1486 BUG_ON(!list_empty(&req->queuelist));
1487 BUG_ON(ELV_ON_HASH(req));
1489 blk_free_request(rl, req);
1490 freed_request(rl, flags);
1494 EXPORT_SYMBOL_GPL(__blk_put_request);
1496 void blk_put_request(struct request *req)
1498 struct request_queue *q = req->q;
1501 blk_mq_free_request(req);
1503 unsigned long flags;
1505 spin_lock_irqsave(q->queue_lock, flags);
1506 __blk_put_request(q, req);
1507 spin_unlock_irqrestore(q->queue_lock, flags);
1510 EXPORT_SYMBOL(blk_put_request);
1513 * blk_add_request_payload - add a payload to a request
1514 * @rq: request to update
1515 * @page: page backing the payload
1516 * @len: length of the payload.
1518 * This allows to later add a payload to an already submitted request by
1519 * a block driver. The driver needs to take care of freeing the payload
1522 * Note that this is a quite horrible hack and nothing but handling of
1523 * discard requests should ever use it.
1525 void blk_add_request_payload(struct request *rq, struct page *page,
1528 struct bio *bio = rq->bio;
1530 bio->bi_io_vec->bv_page = page;
1531 bio->bi_io_vec->bv_offset = 0;
1532 bio->bi_io_vec->bv_len = len;
1534 bio->bi_iter.bi_size = len;
1536 bio->bi_phys_segments = 1;
1538 rq->__data_len = rq->resid_len = len;
1539 rq->nr_phys_segments = 1;
1541 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1543 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1546 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1548 if (!ll_back_merge_fn(q, req, bio))
1551 trace_block_bio_backmerge(q, req, bio);
1553 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1554 blk_rq_set_mixed_merge(req);
1556 req->biotail->bi_next = bio;
1558 req->__data_len += bio->bi_iter.bi_size;
1559 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1561 blk_account_io_start(req, false);
1565 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1568 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1570 if (!ll_front_merge_fn(q, req, bio))
1573 trace_block_bio_frontmerge(q, req, bio);
1575 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1576 blk_rq_set_mixed_merge(req);
1578 bio->bi_next = req->bio;
1581 req->__sector = bio->bi_iter.bi_sector;
1582 req->__data_len += bio->bi_iter.bi_size;
1583 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1585 blk_account_io_start(req, false);
1590 * blk_attempt_plug_merge - try to merge with %current's plugged list
1591 * @q: request_queue new bio is being queued at
1592 * @bio: new bio being queued
1593 * @request_count: out parameter for number of traversed plugged requests
1594 * @same_queue_rq: pointer to &struct request that gets filled in when
1595 * another request associated with @q is found on the plug list
1596 * (optional, may be %NULL)
1598 * Determine whether @bio being queued on @q can be merged with a request
1599 * on %current's plugged list. Returns %true if merge was successful,
1602 * Plugging coalesces IOs from the same issuer for the same purpose without
1603 * going through @q->queue_lock. As such it's more of an issuing mechanism
1604 * than scheduling, and the request, while may have elvpriv data, is not
1605 * added on the elevator at this point. In addition, we don't have
1606 * reliable access to the elevator outside queue lock. Only check basic
1607 * merging parameters without querying the elevator.
1609 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1611 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1612 unsigned int *request_count,
1613 struct request **same_queue_rq)
1615 struct blk_plug *plug;
1618 struct list_head *plug_list;
1620 plug = current->plug;
1626 plug_list = &plug->mq_list;
1628 plug_list = &plug->list;
1630 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1636 * Only blk-mq multiple hardware queues case checks the
1637 * rq in the same queue, there should be only one such
1641 *same_queue_rq = rq;
1644 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1647 el_ret = blk_try_merge(rq, bio);
1648 if (el_ret == ELEVATOR_BACK_MERGE) {
1649 ret = bio_attempt_back_merge(q, rq, bio);
1652 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1653 ret = bio_attempt_front_merge(q, rq, bio);
1662 unsigned int blk_plug_queued_count(struct request_queue *q)
1664 struct blk_plug *plug;
1666 struct list_head *plug_list;
1667 unsigned int ret = 0;
1669 plug = current->plug;
1674 plug_list = &plug->mq_list;
1676 plug_list = &plug->list;
1678 list_for_each_entry(rq, plug_list, queuelist) {
1686 void init_request_from_bio(struct request *req, struct bio *bio)
1688 req->cmd_type = REQ_TYPE_FS;
1690 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1691 if (bio->bi_rw & REQ_RAHEAD)
1692 req->cmd_flags |= REQ_FAILFAST_MASK;
1695 req->__sector = bio->bi_iter.bi_sector;
1696 req->ioprio = bio_prio(bio);
1697 blk_rq_bio_prep(req->q, req, bio);
1700 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1702 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1703 struct blk_plug *plug;
1704 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1705 struct request *req;
1706 unsigned int request_count = 0;
1709 * low level driver can indicate that it wants pages above a
1710 * certain limit bounced to low memory (ie for highmem, or even
1711 * ISA dma in theory)
1713 blk_queue_bounce(q, &bio);
1715 blk_queue_split(q, &bio, q->bio_split);
1717 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1718 bio->bi_error = -EIO;
1720 return BLK_QC_T_NONE;
1723 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1724 spin_lock_irq(q->queue_lock);
1725 where = ELEVATOR_INSERT_FLUSH;
1730 * Check if we can merge with the plugged list before grabbing
1733 if (!blk_queue_nomerges(q)) {
1734 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1735 return BLK_QC_T_NONE;
1737 request_count = blk_plug_queued_count(q);
1739 spin_lock_irq(q->queue_lock);
1741 el_ret = elv_merge(q, &req, bio);
1742 if (el_ret == ELEVATOR_BACK_MERGE) {
1743 if (bio_attempt_back_merge(q, req, bio)) {
1744 elv_bio_merged(q, req, bio);
1745 if (!attempt_back_merge(q, req))
1746 elv_merged_request(q, req, el_ret);
1749 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1750 if (bio_attempt_front_merge(q, req, bio)) {
1751 elv_bio_merged(q, req, bio);
1752 if (!attempt_front_merge(q, req))
1753 elv_merged_request(q, req, el_ret);
1760 * This sync check and mask will be re-done in init_request_from_bio(),
1761 * but we need to set it earlier to expose the sync flag to the
1762 * rq allocator and io schedulers.
1764 rw_flags = bio_data_dir(bio);
1766 rw_flags |= REQ_SYNC;
1769 * Grab a free request. This is might sleep but can not fail.
1770 * Returns with the queue unlocked.
1772 req = get_request(q, rw_flags, bio, GFP_NOIO);
1774 bio->bi_error = PTR_ERR(req);
1780 * After dropping the lock and possibly sleeping here, our request
1781 * may now be mergeable after it had proven unmergeable (above).
1782 * We don't worry about that case for efficiency. It won't happen
1783 * often, and the elevators are able to handle it.
1785 init_request_from_bio(req, bio);
1787 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1788 req->cpu = raw_smp_processor_id();
1790 plug = current->plug;
1793 * If this is the first request added after a plug, fire
1797 trace_block_plug(q);
1799 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1800 blk_flush_plug_list(plug, false);
1801 trace_block_plug(q);
1804 list_add_tail(&req->queuelist, &plug->list);
1805 blk_account_io_start(req, true);
1807 spin_lock_irq(q->queue_lock);
1808 add_acct_request(q, req, where);
1811 spin_unlock_irq(q->queue_lock);
1814 return BLK_QC_T_NONE;
1818 * If bio->bi_dev is a partition, remap the location
1820 static inline void blk_partition_remap(struct bio *bio)
1822 struct block_device *bdev = bio->bi_bdev;
1824 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1825 struct hd_struct *p = bdev->bd_part;
1827 bio->bi_iter.bi_sector += p->start_sect;
1828 bio->bi_bdev = bdev->bd_contains;
1830 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1832 bio->bi_iter.bi_sector - p->start_sect);
1836 static void handle_bad_sector(struct bio *bio)
1838 char b[BDEVNAME_SIZE];
1840 printk(KERN_INFO "attempt to access beyond end of device\n");
1841 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1842 bdevname(bio->bi_bdev, b),
1844 (unsigned long long)bio_end_sector(bio),
1845 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1848 #ifdef CONFIG_FAIL_MAKE_REQUEST
1850 static DECLARE_FAULT_ATTR(fail_make_request);
1852 static int __init setup_fail_make_request(char *str)
1854 return setup_fault_attr(&fail_make_request, str);
1856 __setup("fail_make_request=", setup_fail_make_request);
1858 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1860 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1863 static int __init fail_make_request_debugfs(void)
1865 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1866 NULL, &fail_make_request);
1868 return PTR_ERR_OR_ZERO(dir);
1871 late_initcall(fail_make_request_debugfs);
1873 #else /* CONFIG_FAIL_MAKE_REQUEST */
1875 static inline bool should_fail_request(struct hd_struct *part,
1881 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1884 * Check whether this bio extends beyond the end of the device.
1886 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1893 /* Test device or partition size, when known. */
1894 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1896 sector_t sector = bio->bi_iter.bi_sector;
1898 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1900 * This may well happen - the kernel calls bread()
1901 * without checking the size of the device, e.g., when
1902 * mounting a device.
1904 handle_bad_sector(bio);
1912 static noinline_for_stack bool
1913 generic_make_request_checks(struct bio *bio)
1915 struct request_queue *q;
1916 int nr_sectors = bio_sectors(bio);
1918 char b[BDEVNAME_SIZE];
1919 struct hd_struct *part;
1923 if (bio_check_eod(bio, nr_sectors))
1926 q = bdev_get_queue(bio->bi_bdev);
1929 "generic_make_request: Trying to access "
1930 "nonexistent block-device %s (%Lu)\n",
1931 bdevname(bio->bi_bdev, b),
1932 (long long) bio->bi_iter.bi_sector);
1936 part = bio->bi_bdev->bd_part;
1937 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1938 should_fail_request(&part_to_disk(part)->part0,
1939 bio->bi_iter.bi_size))
1943 * If this device has partitions, remap block n
1944 * of partition p to block n+start(p) of the disk.
1946 blk_partition_remap(bio);
1948 if (bio_check_eod(bio, nr_sectors))
1952 * Filter flush bio's early so that make_request based
1953 * drivers without flush support don't have to worry
1956 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1957 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1964 if ((bio->bi_rw & REQ_DISCARD) &&
1965 (!blk_queue_discard(q) ||
1966 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1971 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1977 * Various block parts want %current->io_context and lazy ioc
1978 * allocation ends up trading a lot of pain for a small amount of
1979 * memory. Just allocate it upfront. This may fail and block
1980 * layer knows how to live with it.
1982 create_io_context(GFP_ATOMIC, q->node);
1984 if (!blkcg_bio_issue_check(q, bio))
1987 trace_block_bio_queue(q, bio);
1991 bio->bi_error = err;
1997 * generic_make_request - hand a buffer to its device driver for I/O
1998 * @bio: The bio describing the location in memory and on the device.
2000 * generic_make_request() is used to make I/O requests of block
2001 * devices. It is passed a &struct bio, which describes the I/O that needs
2004 * generic_make_request() does not return any status. The
2005 * success/failure status of the request, along with notification of
2006 * completion, is delivered asynchronously through the bio->bi_end_io
2007 * function described (one day) else where.
2009 * The caller of generic_make_request must make sure that bi_io_vec
2010 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2011 * set to describe the device address, and the
2012 * bi_end_io and optionally bi_private are set to describe how
2013 * completion notification should be signaled.
2015 * generic_make_request and the drivers it calls may use bi_next if this
2016 * bio happens to be merged with someone else, and may resubmit the bio to
2017 * a lower device by calling into generic_make_request recursively, which
2018 * means the bio should NOT be touched after the call to ->make_request_fn.
2020 blk_qc_t generic_make_request(struct bio *bio)
2023 * bio_list_on_stack[0] contains bios submitted by the current
2025 * bio_list_on_stack[1] contains bios that were submitted before
2026 * the current make_request_fn, but that haven't been processed
2029 struct bio_list bio_list_on_stack[2];
2030 blk_qc_t ret = BLK_QC_T_NONE;
2032 if (!generic_make_request_checks(bio))
2036 * We only want one ->make_request_fn to be active at a time, else
2037 * stack usage with stacked devices could be a problem. So use
2038 * current->bio_list to keep a list of requests submited by a
2039 * make_request_fn function. current->bio_list is also used as a
2040 * flag to say if generic_make_request is currently active in this
2041 * task or not. If it is NULL, then no make_request is active. If
2042 * it is non-NULL, then a make_request is active, and new requests
2043 * should be added at the tail
2045 if (current->bio_list) {
2046 bio_list_add(¤t->bio_list[0], bio);
2050 /* following loop may be a bit non-obvious, and so deserves some
2052 * Before entering the loop, bio->bi_next is NULL (as all callers
2053 * ensure that) so we have a list with a single bio.
2054 * We pretend that we have just taken it off a longer list, so
2055 * we assign bio_list to a pointer to the bio_list_on_stack,
2056 * thus initialising the bio_list of new bios to be
2057 * added. ->make_request() may indeed add some more bios
2058 * through a recursive call to generic_make_request. If it
2059 * did, we find a non-NULL value in bio_list and re-enter the loop
2060 * from the top. In this case we really did just take the bio
2061 * of the top of the list (no pretending) and so remove it from
2062 * bio_list, and call into ->make_request() again.
2064 BUG_ON(bio->bi_next);
2065 bio_list_init(&bio_list_on_stack[0]);
2066 current->bio_list = bio_list_on_stack;
2068 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2070 if (likely(blk_queue_enter(q, __GFP_DIRECT_RECLAIM) == 0)) {
2071 struct bio_list lower, same;
2073 /* Create a fresh bio_list for all subordinate requests */
2074 bio_list_on_stack[1] = bio_list_on_stack[0];
2075 bio_list_init(&bio_list_on_stack[0]);
2077 ret = q->make_request_fn(q, bio);
2080 /* sort new bios into those for a lower level
2081 * and those for the same level
2083 bio_list_init(&lower);
2084 bio_list_init(&same);
2085 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2086 if (q == bdev_get_queue(bio->bi_bdev))
2087 bio_list_add(&same, bio);
2089 bio_list_add(&lower, bio);
2090 /* now assemble so we handle the lowest level first */
2091 bio_list_merge(&bio_list_on_stack[0], &lower);
2092 bio_list_merge(&bio_list_on_stack[0], &same);
2093 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2097 bio = bio_list_pop(&bio_list_on_stack[0]);
2099 current->bio_list = NULL; /* deactivate */
2104 EXPORT_SYMBOL(generic_make_request);
2107 * submit_bio - submit a bio to the block device layer for I/O
2108 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2109 * @bio: The &struct bio which describes the I/O
2111 * submit_bio() is very similar in purpose to generic_make_request(), and
2112 * uses that function to do most of the work. Both are fairly rough
2113 * interfaces; @bio must be presetup and ready for I/O.
2116 blk_qc_t submit_bio(int rw, struct bio *bio)
2121 * If it's a regular read/write or a barrier with data attached,
2122 * go through the normal accounting stuff before submission.
2124 if (bio_has_data(bio)) {
2127 if (unlikely(rw & REQ_WRITE_SAME))
2128 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2130 count = bio_sectors(bio);
2133 count_vm_events(PGPGOUT, count);
2135 task_io_account_read(bio->bi_iter.bi_size);
2136 count_vm_events(PGPGIN, count);
2139 if (unlikely(block_dump)) {
2140 char b[BDEVNAME_SIZE];
2141 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2142 current->comm, task_pid_nr(current),
2143 (rw & WRITE) ? "WRITE" : "READ",
2144 (unsigned long long)bio->bi_iter.bi_sector,
2145 bdevname(bio->bi_bdev, b),
2150 return generic_make_request(bio);
2152 EXPORT_SYMBOL(submit_bio);
2155 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2156 * for new the queue limits
2158 * @rq: the request being checked
2161 * @rq may have been made based on weaker limitations of upper-level queues
2162 * in request stacking drivers, and it may violate the limitation of @q.
2163 * Since the block layer and the underlying device driver trust @rq
2164 * after it is inserted to @q, it should be checked against @q before
2165 * the insertion using this generic function.
2167 * Request stacking drivers like request-based dm may change the queue
2168 * limits when retrying requests on other queues. Those requests need
2169 * to be checked against the new queue limits again during dispatch.
2171 static int blk_cloned_rq_check_limits(struct request_queue *q,
2174 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2175 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2180 * queue's settings related to segment counting like q->bounce_pfn
2181 * may differ from that of other stacking queues.
2182 * Recalculate it to check the request correctly on this queue's
2185 blk_recalc_rq_segments(rq);
2186 if (rq->nr_phys_segments > queue_max_segments(q)) {
2187 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2195 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2196 * @q: the queue to submit the request
2197 * @rq: the request being queued
2199 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2201 unsigned long flags;
2202 int where = ELEVATOR_INSERT_BACK;
2204 if (blk_cloned_rq_check_limits(q, rq))
2208 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2212 if (blk_queue_io_stat(q))
2213 blk_account_io_start(rq, true);
2214 blk_mq_insert_request(rq, false, true, false);
2218 spin_lock_irqsave(q->queue_lock, flags);
2219 if (unlikely(blk_queue_dying(q))) {
2220 spin_unlock_irqrestore(q->queue_lock, flags);
2225 * Submitting request must be dequeued before calling this function
2226 * because it will be linked to another request_queue
2228 BUG_ON(blk_queued_rq(rq));
2230 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2231 where = ELEVATOR_INSERT_FLUSH;
2233 add_acct_request(q, rq, where);
2234 if (where == ELEVATOR_INSERT_FLUSH)
2236 spin_unlock_irqrestore(q->queue_lock, flags);
2240 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2243 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2244 * @rq: request to examine
2247 * A request could be merge of IOs which require different failure
2248 * handling. This function determines the number of bytes which
2249 * can be failed from the beginning of the request without
2250 * crossing into area which need to be retried further.
2253 * The number of bytes to fail.
2256 * queue_lock must be held.
2258 unsigned int blk_rq_err_bytes(const struct request *rq)
2260 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2261 unsigned int bytes = 0;
2264 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2265 return blk_rq_bytes(rq);
2268 * Currently the only 'mixing' which can happen is between
2269 * different fastfail types. We can safely fail portions
2270 * which have all the failfast bits that the first one has -
2271 * the ones which are at least as eager to fail as the first
2274 for (bio = rq->bio; bio; bio = bio->bi_next) {
2275 if ((bio->bi_rw & ff) != ff)
2277 bytes += bio->bi_iter.bi_size;
2280 /* this could lead to infinite loop */
2281 BUG_ON(blk_rq_bytes(rq) && !bytes);
2284 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2286 void blk_account_io_completion(struct request *req, unsigned int bytes)
2288 if (blk_do_io_stat(req)) {
2289 const int rw = rq_data_dir(req);
2290 struct hd_struct *part;
2293 cpu = part_stat_lock();
2295 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2300 void blk_account_io_done(struct request *req)
2303 * Account IO completion. flush_rq isn't accounted as a
2304 * normal IO on queueing nor completion. Accounting the
2305 * containing request is enough.
2307 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2308 unsigned long duration = jiffies - req->start_time;
2309 const int rw = rq_data_dir(req);
2310 struct hd_struct *part;
2313 cpu = part_stat_lock();
2316 part_stat_inc(cpu, part, ios[rw]);
2317 part_stat_add(cpu, part, ticks[rw], duration);
2318 part_round_stats(cpu, part);
2319 part_dec_in_flight(part, rw);
2321 hd_struct_put(part);
2328 * Don't process normal requests when queue is suspended
2329 * or in the process of suspending/resuming
2331 static struct request *blk_pm_peek_request(struct request_queue *q,
2334 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2335 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2341 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2348 void blk_account_io_start(struct request *rq, bool new_io)
2350 struct hd_struct *part;
2351 int rw = rq_data_dir(rq);
2354 if (!blk_do_io_stat(rq))
2357 cpu = part_stat_lock();
2361 part_stat_inc(cpu, part, merges[rw]);
2363 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2364 if (!hd_struct_try_get(part)) {
2366 * The partition is already being removed,
2367 * the request will be accounted on the disk only
2369 * We take a reference on disk->part0 although that
2370 * partition will never be deleted, so we can treat
2371 * it as any other partition.
2373 part = &rq->rq_disk->part0;
2374 hd_struct_get(part);
2376 part_round_stats(cpu, part);
2377 part_inc_in_flight(part, rw);
2385 * blk_peek_request - peek at the top of a request queue
2386 * @q: request queue to peek at
2389 * Return the request at the top of @q. The returned request
2390 * should be started using blk_start_request() before LLD starts
2394 * Pointer to the request at the top of @q if available. Null
2398 * queue_lock must be held.
2400 struct request *blk_peek_request(struct request_queue *q)
2405 while ((rq = __elv_next_request(q)) != NULL) {
2407 rq = blk_pm_peek_request(q, rq);
2411 if (!(rq->cmd_flags & REQ_STARTED)) {
2413 * This is the first time the device driver
2414 * sees this request (possibly after
2415 * requeueing). Notify IO scheduler.
2417 if (rq->cmd_flags & REQ_SORTED)
2418 elv_activate_rq(q, rq);
2421 * just mark as started even if we don't start
2422 * it, a request that has been delayed should
2423 * not be passed by new incoming requests
2425 rq->cmd_flags |= REQ_STARTED;
2426 trace_block_rq_issue(q, rq);
2429 if (!q->boundary_rq || q->boundary_rq == rq) {
2430 q->end_sector = rq_end_sector(rq);
2431 q->boundary_rq = NULL;
2434 if (rq->cmd_flags & REQ_DONTPREP)
2437 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2439 * make sure space for the drain appears we
2440 * know we can do this because max_hw_segments
2441 * has been adjusted to be one fewer than the
2444 rq->nr_phys_segments++;
2450 ret = q->prep_rq_fn(q, rq);
2451 if (ret == BLKPREP_OK) {
2453 } else if (ret == BLKPREP_DEFER) {
2455 * the request may have been (partially) prepped.
2456 * we need to keep this request in the front to
2457 * avoid resource deadlock. REQ_STARTED will
2458 * prevent other fs requests from passing this one.
2460 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2461 !(rq->cmd_flags & REQ_DONTPREP)) {
2463 * remove the space for the drain we added
2464 * so that we don't add it again
2466 --rq->nr_phys_segments;
2471 } else if (ret == BLKPREP_KILL) {
2472 rq->cmd_flags |= REQ_QUIET;
2474 * Mark this request as started so we don't trigger
2475 * any debug logic in the end I/O path.
2477 blk_start_request(rq);
2478 __blk_end_request_all(rq, -EIO);
2480 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2487 EXPORT_SYMBOL(blk_peek_request);
2489 void blk_dequeue_request(struct request *rq)
2491 struct request_queue *q = rq->q;
2493 BUG_ON(list_empty(&rq->queuelist));
2494 BUG_ON(ELV_ON_HASH(rq));
2496 list_del_init(&rq->queuelist);
2499 * the time frame between a request being removed from the lists
2500 * and to it is freed is accounted as io that is in progress at
2503 if (blk_account_rq(rq)) {
2504 q->in_flight[rq_is_sync(rq)]++;
2505 set_io_start_time_ns(rq);
2510 * blk_start_request - start request processing on the driver
2511 * @req: request to dequeue
2514 * Dequeue @req and start timeout timer on it. This hands off the
2515 * request to the driver.
2517 * Block internal functions which don't want to start timer should
2518 * call blk_dequeue_request().
2521 * queue_lock must be held.
2523 void blk_start_request(struct request *req)
2525 blk_dequeue_request(req);
2528 * We are now handing the request to the hardware, initialize
2529 * resid_len to full count and add the timeout handler.
2531 req->resid_len = blk_rq_bytes(req);
2532 if (unlikely(blk_bidi_rq(req)))
2533 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2535 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2538 EXPORT_SYMBOL(blk_start_request);
2541 * blk_fetch_request - fetch a request from a request queue
2542 * @q: request queue to fetch a request from
2545 * Return the request at the top of @q. The request is started on
2546 * return and LLD can start processing it immediately.
2549 * Pointer to the request at the top of @q if available. Null
2553 * queue_lock must be held.
2555 struct request *blk_fetch_request(struct request_queue *q)
2559 rq = blk_peek_request(q);
2561 blk_start_request(rq);
2564 EXPORT_SYMBOL(blk_fetch_request);
2567 * blk_update_request - Special helper function for request stacking drivers
2568 * @req: the request being processed
2569 * @error: %0 for success, < %0 for error
2570 * @nr_bytes: number of bytes to complete @req
2573 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2574 * the request structure even if @req doesn't have leftover.
2575 * If @req has leftover, sets it up for the next range of segments.
2577 * This special helper function is only for request stacking drivers
2578 * (e.g. request-based dm) so that they can handle partial completion.
2579 * Actual device drivers should use blk_end_request instead.
2581 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2582 * %false return from this function.
2585 * %false - this request doesn't have any more data
2586 * %true - this request has more data
2588 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2592 trace_block_rq_complete(req->q, req, nr_bytes);
2598 * For fs requests, rq is just carrier of independent bio's
2599 * and each partial completion should be handled separately.
2600 * Reset per-request error on each partial completion.
2602 * TODO: tj: This is too subtle. It would be better to let
2603 * low level drivers do what they see fit.
2605 if (req->cmd_type == REQ_TYPE_FS)
2608 if (error && req->cmd_type == REQ_TYPE_FS &&
2609 !(req->cmd_flags & REQ_QUIET)) {
2614 error_type = "recoverable transport";
2617 error_type = "critical target";
2620 error_type = "critical nexus";
2623 error_type = "timeout";
2626 error_type = "critical space allocation";
2629 error_type = "critical medium";
2636 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2637 __func__, error_type, req->rq_disk ?
2638 req->rq_disk->disk_name : "?",
2639 (unsigned long long)blk_rq_pos(req));
2643 blk_account_io_completion(req, nr_bytes);
2647 struct bio *bio = req->bio;
2648 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2650 if (bio_bytes == bio->bi_iter.bi_size)
2651 req->bio = bio->bi_next;
2653 req_bio_endio(req, bio, bio_bytes, error);
2655 total_bytes += bio_bytes;
2656 nr_bytes -= bio_bytes;
2667 * Reset counters so that the request stacking driver
2668 * can find how many bytes remain in the request
2671 req->__data_len = 0;
2675 req->__data_len -= total_bytes;
2677 /* update sector only for requests with clear definition of sector */
2678 if (req->cmd_type == REQ_TYPE_FS)
2679 req->__sector += total_bytes >> 9;
2681 /* mixed attributes always follow the first bio */
2682 if (req->cmd_flags & REQ_MIXED_MERGE) {
2683 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2684 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2688 * If total number of sectors is less than the first segment
2689 * size, something has gone terribly wrong.
2691 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2692 blk_dump_rq_flags(req, "request botched");
2693 req->__data_len = blk_rq_cur_bytes(req);
2696 /* recalculate the number of segments */
2697 blk_recalc_rq_segments(req);
2701 EXPORT_SYMBOL_GPL(blk_update_request);
2703 static bool blk_update_bidi_request(struct request *rq, int error,
2704 unsigned int nr_bytes,
2705 unsigned int bidi_bytes)
2707 if (blk_update_request(rq, error, nr_bytes))
2710 /* Bidi request must be completed as a whole */
2711 if (unlikely(blk_bidi_rq(rq)) &&
2712 blk_update_request(rq->next_rq, error, bidi_bytes))
2715 if (blk_queue_add_random(rq->q))
2716 add_disk_randomness(rq->rq_disk);
2722 * blk_unprep_request - unprepare a request
2725 * This function makes a request ready for complete resubmission (or
2726 * completion). It happens only after all error handling is complete,
2727 * so represents the appropriate moment to deallocate any resources
2728 * that were allocated to the request in the prep_rq_fn. The queue
2729 * lock is held when calling this.
2731 void blk_unprep_request(struct request *req)
2733 struct request_queue *q = req->q;
2735 req->cmd_flags &= ~REQ_DONTPREP;
2736 if (q->unprep_rq_fn)
2737 q->unprep_rq_fn(q, req);
2739 EXPORT_SYMBOL_GPL(blk_unprep_request);
2742 * queue lock must be held
2744 void blk_finish_request(struct request *req, int error)
2746 if (req->cmd_flags & REQ_QUEUED)
2747 blk_queue_end_tag(req->q, req);
2749 BUG_ON(blk_queued_rq(req));
2751 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2752 laptop_io_completion(&req->q->backing_dev_info);
2754 blk_delete_timer(req);
2756 if (req->cmd_flags & REQ_DONTPREP)
2757 blk_unprep_request(req);
2759 blk_account_io_done(req);
2762 req->end_io(req, error);
2764 if (blk_bidi_rq(req))
2765 __blk_put_request(req->next_rq->q, req->next_rq);
2767 __blk_put_request(req->q, req);
2770 EXPORT_SYMBOL(blk_finish_request);
2773 * blk_end_bidi_request - Complete a bidi request
2774 * @rq: the request to complete
2775 * @error: %0 for success, < %0 for error
2776 * @nr_bytes: number of bytes to complete @rq
2777 * @bidi_bytes: number of bytes to complete @rq->next_rq
2780 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2781 * Drivers that supports bidi can safely call this member for any
2782 * type of request, bidi or uni. In the later case @bidi_bytes is
2786 * %false - we are done with this request
2787 * %true - still buffers pending for this request
2789 static bool blk_end_bidi_request(struct request *rq, int error,
2790 unsigned int nr_bytes, unsigned int bidi_bytes)
2792 struct request_queue *q = rq->q;
2793 unsigned long flags;
2795 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2798 spin_lock_irqsave(q->queue_lock, flags);
2799 blk_finish_request(rq, error);
2800 spin_unlock_irqrestore(q->queue_lock, flags);
2806 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2807 * @rq: the request to complete
2808 * @error: %0 for success, < %0 for error
2809 * @nr_bytes: number of bytes to complete @rq
2810 * @bidi_bytes: number of bytes to complete @rq->next_rq
2813 * Identical to blk_end_bidi_request() except that queue lock is
2814 * assumed to be locked on entry and remains so on return.
2817 * %false - we are done with this request
2818 * %true - still buffers pending for this request
2820 bool __blk_end_bidi_request(struct request *rq, int error,
2821 unsigned int nr_bytes, unsigned int bidi_bytes)
2823 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2826 blk_finish_request(rq, error);
2832 * blk_end_request - Helper function for drivers to complete the request.
2833 * @rq: the request being processed
2834 * @error: %0 for success, < %0 for error
2835 * @nr_bytes: number of bytes to complete
2838 * Ends I/O on a number of bytes attached to @rq.
2839 * If @rq has leftover, sets it up for the next range of segments.
2842 * %false - we are done with this request
2843 * %true - still buffers pending for this request
2845 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2847 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2849 EXPORT_SYMBOL(blk_end_request);
2852 * blk_end_request_all - Helper function for drives to finish the request.
2853 * @rq: the request to finish
2854 * @error: %0 for success, < %0 for error
2857 * Completely finish @rq.
2859 void blk_end_request_all(struct request *rq, int error)
2862 unsigned int bidi_bytes = 0;
2864 if (unlikely(blk_bidi_rq(rq)))
2865 bidi_bytes = blk_rq_bytes(rq->next_rq);
2867 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2870 EXPORT_SYMBOL(blk_end_request_all);
2873 * blk_end_request_cur - Helper function to finish the current request chunk.
2874 * @rq: the request to finish the current chunk for
2875 * @error: %0 for success, < %0 for error
2878 * Complete the current consecutively mapped chunk from @rq.
2881 * %false - we are done with this request
2882 * %true - still buffers pending for this request
2884 bool blk_end_request_cur(struct request *rq, int error)
2886 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2888 EXPORT_SYMBOL(blk_end_request_cur);
2891 * blk_end_request_err - Finish a request till the next failure boundary.
2892 * @rq: the request to finish till the next failure boundary for
2893 * @error: must be negative errno
2896 * Complete @rq till the next failure boundary.
2899 * %false - we are done with this request
2900 * %true - still buffers pending for this request
2902 bool blk_end_request_err(struct request *rq, int error)
2904 WARN_ON(error >= 0);
2905 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2907 EXPORT_SYMBOL_GPL(blk_end_request_err);
2910 * __blk_end_request - Helper function for drivers to complete the request.
2911 * @rq: the request being processed
2912 * @error: %0 for success, < %0 for error
2913 * @nr_bytes: number of bytes to complete
2916 * Must be called with queue lock held unlike blk_end_request().
2919 * %false - we are done with this request
2920 * %true - still buffers pending for this request
2922 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2924 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2926 EXPORT_SYMBOL(__blk_end_request);
2929 * __blk_end_request_all - Helper function for drives to finish the request.
2930 * @rq: the request to finish
2931 * @error: %0 for success, < %0 for error
2934 * Completely finish @rq. Must be called with queue lock held.
2936 void __blk_end_request_all(struct request *rq, int error)
2939 unsigned int bidi_bytes = 0;
2941 if (unlikely(blk_bidi_rq(rq)))
2942 bidi_bytes = blk_rq_bytes(rq->next_rq);
2944 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2947 EXPORT_SYMBOL(__blk_end_request_all);
2950 * __blk_end_request_cur - Helper function to finish the current request chunk.
2951 * @rq: the request to finish the current chunk for
2952 * @error: %0 for success, < %0 for error
2955 * Complete the current consecutively mapped chunk from @rq. Must
2956 * be called with queue lock held.
2959 * %false - we are done with this request
2960 * %true - still buffers pending for this request
2962 bool __blk_end_request_cur(struct request *rq, int error)
2964 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2966 EXPORT_SYMBOL(__blk_end_request_cur);
2969 * __blk_end_request_err - Finish a request till the next failure boundary.
2970 * @rq: the request to finish till the next failure boundary for
2971 * @error: must be negative errno
2974 * Complete @rq till the next failure boundary. Must be called
2975 * with queue lock held.
2978 * %false - we are done with this request
2979 * %true - still buffers pending for this request
2981 bool __blk_end_request_err(struct request *rq, int error)
2983 WARN_ON(error >= 0);
2984 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2986 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2988 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2991 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2992 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2994 if (bio_has_data(bio))
2995 rq->nr_phys_segments = bio_phys_segments(q, bio);
2997 rq->__data_len = bio->bi_iter.bi_size;
2998 rq->bio = rq->biotail = bio;
3001 rq->rq_disk = bio->bi_bdev->bd_disk;
3004 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3006 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3007 * @rq: the request to be flushed
3010 * Flush all pages in @rq.
3012 void rq_flush_dcache_pages(struct request *rq)
3014 struct req_iterator iter;
3015 struct bio_vec bvec;
3017 rq_for_each_segment(bvec, rq, iter)
3018 flush_dcache_page(bvec.bv_page);
3020 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3024 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3025 * @q : the queue of the device being checked
3028 * Check if underlying low-level drivers of a device are busy.
3029 * If the drivers want to export their busy state, they must set own
3030 * exporting function using blk_queue_lld_busy() first.
3032 * Basically, this function is used only by request stacking drivers
3033 * to stop dispatching requests to underlying devices when underlying
3034 * devices are busy. This behavior helps more I/O merging on the queue
3035 * of the request stacking driver and prevents I/O throughput regression
3036 * on burst I/O load.
3039 * 0 - Not busy (The request stacking driver should dispatch request)
3040 * 1 - Busy (The request stacking driver should stop dispatching request)
3042 int blk_lld_busy(struct request_queue *q)
3045 return q->lld_busy_fn(q);
3049 EXPORT_SYMBOL_GPL(blk_lld_busy);
3052 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3053 * @rq: the clone request to be cleaned up
3056 * Free all bios in @rq for a cloned request.
3058 void blk_rq_unprep_clone(struct request *rq)
3062 while ((bio = rq->bio) != NULL) {
3063 rq->bio = bio->bi_next;
3068 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3071 * Copy attributes of the original request to the clone request.
3072 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3074 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3076 dst->cpu = src->cpu;
3077 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3078 dst->cmd_type = src->cmd_type;
3079 dst->__sector = blk_rq_pos(src);
3080 dst->__data_len = blk_rq_bytes(src);
3081 dst->nr_phys_segments = src->nr_phys_segments;
3082 dst->ioprio = src->ioprio;
3083 dst->extra_len = src->extra_len;
3087 * blk_rq_prep_clone - Helper function to setup clone request
3088 * @rq: the request to be setup
3089 * @rq_src: original request to be cloned
3090 * @bs: bio_set that bios for clone are allocated from
3091 * @gfp_mask: memory allocation mask for bio
3092 * @bio_ctr: setup function to be called for each clone bio.
3093 * Returns %0 for success, non %0 for failure.
3094 * @data: private data to be passed to @bio_ctr
3097 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3098 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3099 * are not copied, and copying such parts is the caller's responsibility.
3100 * Also, pages which the original bios are pointing to are not copied
3101 * and the cloned bios just point same pages.
3102 * So cloned bios must be completed before original bios, which means
3103 * the caller must complete @rq before @rq_src.
3105 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3106 struct bio_set *bs, gfp_t gfp_mask,
3107 int (*bio_ctr)(struct bio *, struct bio *, void *),
3110 struct bio *bio, *bio_src;
3115 __rq_for_each_bio(bio_src, rq_src) {
3116 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3120 if (bio_ctr && bio_ctr(bio, bio_src, data))
3124 rq->biotail->bi_next = bio;
3127 rq->bio = rq->biotail = bio;
3130 __blk_rq_prep_clone(rq, rq_src);
3137 blk_rq_unprep_clone(rq);
3141 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3143 int kblockd_schedule_work(struct work_struct *work)
3145 return queue_work(kblockd_workqueue, work);
3147 EXPORT_SYMBOL(kblockd_schedule_work);
3149 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3150 unsigned long delay)
3152 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3154 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3156 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3157 unsigned long delay)
3159 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3161 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3164 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3165 * @plug: The &struct blk_plug that needs to be initialized
3168 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3169 * pending I/O should the task end up blocking between blk_start_plug() and
3170 * blk_finish_plug(). This is important from a performance perspective, but
3171 * also ensures that we don't deadlock. For instance, if the task is blocking
3172 * for a memory allocation, memory reclaim could end up wanting to free a
3173 * page belonging to that request that is currently residing in our private
3174 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3175 * this kind of deadlock.
3177 void blk_start_plug(struct blk_plug *plug)
3179 struct task_struct *tsk = current;
3182 * If this is a nested plug, don't actually assign it.
3187 INIT_LIST_HEAD(&plug->list);
3188 INIT_LIST_HEAD(&plug->mq_list);
3189 INIT_LIST_HEAD(&plug->cb_list);
3191 * Store ordering should not be needed here, since a potential
3192 * preempt will imply a full memory barrier
3196 EXPORT_SYMBOL(blk_start_plug);
3198 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3200 struct request *rqa = container_of(a, struct request, queuelist);
3201 struct request *rqb = container_of(b, struct request, queuelist);
3203 return !(rqa->q < rqb->q ||
3204 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3208 * If 'from_schedule' is true, then postpone the dispatch of requests
3209 * until a safe kblockd context. We due this to avoid accidental big
3210 * additional stack usage in driver dispatch, in places where the originally
3211 * plugger did not intend it.
3213 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3215 __releases(q->queue_lock)
3217 trace_block_unplug(q, depth, !from_schedule);
3220 blk_run_queue_async(q);
3223 spin_unlock(q->queue_lock);
3226 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3228 LIST_HEAD(callbacks);
3230 while (!list_empty(&plug->cb_list)) {
3231 list_splice_init(&plug->cb_list, &callbacks);
3233 while (!list_empty(&callbacks)) {
3234 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3237 list_del(&cb->list);
3238 cb->callback(cb, from_schedule);
3243 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3246 struct blk_plug *plug = current->plug;
3247 struct blk_plug_cb *cb;
3252 list_for_each_entry(cb, &plug->cb_list, list)
3253 if (cb->callback == unplug && cb->data == data)
3256 /* Not currently on the callback list */
3257 BUG_ON(size < sizeof(*cb));
3258 cb = kzalloc(size, GFP_ATOMIC);
3261 cb->callback = unplug;
3262 list_add(&cb->list, &plug->cb_list);
3266 EXPORT_SYMBOL(blk_check_plugged);
3268 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3270 struct request_queue *q;
3271 unsigned long flags;
3276 flush_plug_callbacks(plug, from_schedule);
3278 if (!list_empty(&plug->mq_list))
3279 blk_mq_flush_plug_list(plug, from_schedule);
3281 if (list_empty(&plug->list))
3284 list_splice_init(&plug->list, &list);
3286 list_sort(NULL, &list, plug_rq_cmp);
3292 * Save and disable interrupts here, to avoid doing it for every
3293 * queue lock we have to take.
3295 local_irq_save(flags);
3296 while (!list_empty(&list)) {
3297 rq = list_entry_rq(list.next);
3298 list_del_init(&rq->queuelist);
3302 * This drops the queue lock
3305 queue_unplugged(q, depth, from_schedule);
3308 spin_lock(q->queue_lock);
3312 * Short-circuit if @q is dead
3314 if (unlikely(blk_queue_dying(q))) {
3315 __blk_end_request_all(rq, -ENODEV);
3320 * rq is already accounted, so use raw insert
3322 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3323 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3325 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3331 * This drops the queue lock
3334 queue_unplugged(q, depth, from_schedule);
3336 local_irq_restore(flags);
3339 void blk_finish_plug(struct blk_plug *plug)
3341 if (plug != current->plug)
3343 blk_flush_plug_list(plug, false);
3345 current->plug = NULL;
3347 EXPORT_SYMBOL(blk_finish_plug);
3349 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3351 struct blk_plug *plug;
3354 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3355 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3358 plug = current->plug;
3360 blk_flush_plug_list(plug, false);
3362 state = current->state;
3363 while (!need_resched()) {
3364 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3365 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3368 hctx->poll_invoked++;
3370 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3372 hctx->poll_success++;
3373 set_current_state(TASK_RUNNING);
3377 if (signal_pending_state(state, current))
3378 set_current_state(TASK_RUNNING);
3380 if (current->state == TASK_RUNNING)
3392 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3393 * @q: the queue of the device
3394 * @dev: the device the queue belongs to
3397 * Initialize runtime-PM-related fields for @q and start auto suspend for
3398 * @dev. Drivers that want to take advantage of request-based runtime PM
3399 * should call this function after @dev has been initialized, and its
3400 * request queue @q has been allocated, and runtime PM for it can not happen
3401 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3402 * cases, driver should call this function before any I/O has taken place.
3404 * This function takes care of setting up using auto suspend for the device,
3405 * the autosuspend delay is set to -1 to make runtime suspend impossible
3406 * until an updated value is either set by user or by driver. Drivers do
3407 * not need to touch other autosuspend settings.
3409 * The block layer runtime PM is request based, so only works for drivers
3410 * that use request as their IO unit instead of those directly use bio's.
3412 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3415 q->rpm_status = RPM_ACTIVE;
3416 pm_runtime_set_autosuspend_delay(q->dev, -1);
3417 pm_runtime_use_autosuspend(q->dev);
3419 EXPORT_SYMBOL(blk_pm_runtime_init);
3422 * blk_pre_runtime_suspend - Pre runtime suspend check
3423 * @q: the queue of the device
3426 * This function will check if runtime suspend is allowed for the device
3427 * by examining if there are any requests pending in the queue. If there
3428 * are requests pending, the device can not be runtime suspended; otherwise,
3429 * the queue's status will be updated to SUSPENDING and the driver can
3430 * proceed to suspend the device.
3432 * For the not allowed case, we mark last busy for the device so that
3433 * runtime PM core will try to autosuspend it some time later.
3435 * This function should be called near the start of the device's
3436 * runtime_suspend callback.
3439 * 0 - OK to runtime suspend the device
3440 * -EBUSY - Device should not be runtime suspended
3442 int blk_pre_runtime_suspend(struct request_queue *q)
3449 spin_lock_irq(q->queue_lock);
3450 if (q->nr_pending) {
3452 pm_runtime_mark_last_busy(q->dev);
3454 q->rpm_status = RPM_SUSPENDING;
3456 spin_unlock_irq(q->queue_lock);
3459 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3462 * blk_post_runtime_suspend - Post runtime suspend processing
3463 * @q: the queue of the device
3464 * @err: return value of the device's runtime_suspend function
3467 * Update the queue's runtime status according to the return value of the
3468 * device's runtime suspend function and mark last busy for the device so
3469 * that PM core will try to auto suspend the device at a later time.
3471 * This function should be called near the end of the device's
3472 * runtime_suspend callback.
3474 void blk_post_runtime_suspend(struct request_queue *q, int err)
3479 spin_lock_irq(q->queue_lock);
3481 q->rpm_status = RPM_SUSPENDED;
3483 q->rpm_status = RPM_ACTIVE;
3484 pm_runtime_mark_last_busy(q->dev);
3486 spin_unlock_irq(q->queue_lock);
3488 EXPORT_SYMBOL(blk_post_runtime_suspend);
3491 * blk_pre_runtime_resume - Pre runtime resume processing
3492 * @q: the queue of the device
3495 * Update the queue's runtime status to RESUMING in preparation for the
3496 * runtime resume of the device.
3498 * This function should be called near the start of the device's
3499 * runtime_resume callback.
3501 void blk_pre_runtime_resume(struct request_queue *q)
3506 spin_lock_irq(q->queue_lock);
3507 q->rpm_status = RPM_RESUMING;
3508 spin_unlock_irq(q->queue_lock);
3510 EXPORT_SYMBOL(blk_pre_runtime_resume);
3513 * blk_post_runtime_resume - Post runtime resume processing
3514 * @q: the queue of the device
3515 * @err: return value of the device's runtime_resume function
3518 * Update the queue's runtime status according to the return value of the
3519 * device's runtime_resume function. If it is successfully resumed, process
3520 * the requests that are queued into the device's queue when it is resuming
3521 * and then mark last busy and initiate autosuspend for it.
3523 * This function should be called near the end of the device's
3524 * runtime_resume callback.
3526 void blk_post_runtime_resume(struct request_queue *q, int err)
3531 spin_lock_irq(q->queue_lock);
3533 q->rpm_status = RPM_ACTIVE;
3535 pm_runtime_mark_last_busy(q->dev);
3536 pm_request_autosuspend(q->dev);
3538 q->rpm_status = RPM_SUSPENDED;
3540 spin_unlock_irq(q->queue_lock);
3542 EXPORT_SYMBOL(blk_post_runtime_resume);
3545 int __init blk_dev_init(void)
3547 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3548 FIELD_SIZEOF(struct request, cmd_flags));
3550 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3551 kblockd_workqueue = alloc_workqueue("kblockd",
3552 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3553 if (!kblockd_workqueue)
3554 panic("Failed to create kblockd\n");
3556 request_cachep = kmem_cache_create("blkdev_requests",
3557 sizeof(struct request), 0, SLAB_PANIC, NULL);
3559 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3560 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3566 * Blk IO latency support. We want this to be as cheap as possible, so doing
3567 * this lockless (and avoiding atomics), a few off by a few errors in this
3568 * code is not harmful, and we don't want to do anything that is
3570 * TODO : If necessary, we can make the histograms per-cpu and aggregate
3571 * them when printing them out.
3574 blk_latency_hist_show(char* name, struct io_latency_state *s, char *buf,
3578 int bytes_written = 0;
3579 u_int64_t num_elem, elem;
3583 num_elem = s->latency_elems;
3585 average = div64_u64(s->latency_sum, s->latency_elems);
3586 bytes_written += scnprintf(buf + bytes_written,
3587 buf_size - bytes_written,
3588 "IO svc_time %s Latency Histogram (n = %llu,"
3589 " average = %llu):\n", name, num_elem, average);
3591 i < ARRAY_SIZE(latency_x_axis_us);
3593 elem = s->latency_y_axis[i];
3594 pct = div64_u64(elem * 100, num_elem);
3595 bytes_written += scnprintf(buf + bytes_written,
3596 PAGE_SIZE - bytes_written,
3597 "\t< %6lluus%15llu%15d%%\n",
3598 latency_x_axis_us[i],
3601 /* Last element in y-axis table is overflow */
3602 elem = s->latency_y_axis[i];
3603 pct = div64_u64(elem * 100, num_elem);
3604 bytes_written += scnprintf(buf + bytes_written,
3605 PAGE_SIZE - bytes_written,
3606 "\t>=%6lluus%15llu%15d%%\n",
3607 latency_x_axis_us[i - 1], elem, pct);
3610 return bytes_written;
3612 EXPORT_SYMBOL(blk_latency_hist_show);