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->rq_flags & RQF_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->rq_flags & RQF_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_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 spin_lock_irq(q->queue_lock);
519 queue_flag_set(QUEUE_FLAG_DYING, q);
520 spin_unlock_irq(q->queue_lock);
523 blk_mq_wake_waiters(q);
525 struct request_list *rl;
527 blk_queue_for_each_rl(rl, q) {
529 wake_up(&rl->wait[BLK_RW_SYNC]);
530 wake_up(&rl->wait[BLK_RW_ASYNC]);
535 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
538 * blk_cleanup_queue - shutdown a request queue
539 * @q: request queue to shutdown
541 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
542 * put it. All future requests will be failed immediately with -ENODEV.
544 void blk_cleanup_queue(struct request_queue *q)
546 spinlock_t *lock = q->queue_lock;
548 /* mark @q DYING, no new request or merges will be allowed afterwards */
549 mutex_lock(&q->sysfs_lock);
550 blk_set_queue_dying(q);
554 * A dying queue is permanently in bypass mode till released. Note
555 * that, unlike blk_queue_bypass_start(), we aren't performing
556 * synchronize_rcu() after entering bypass mode to avoid the delay
557 * as some drivers create and destroy a lot of queues while
558 * probing. This is still safe because blk_release_queue() will be
559 * called only after the queue refcnt drops to zero and nothing,
560 * RCU or not, would be traversing the queue by then.
563 queue_flag_set(QUEUE_FLAG_BYPASS, q);
565 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
566 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
567 queue_flag_set(QUEUE_FLAG_DYING, q);
568 spin_unlock_irq(lock);
569 mutex_unlock(&q->sysfs_lock);
572 * Drain all requests queued before DYING marking. Set DEAD flag to
573 * prevent that q->request_fn() gets invoked after draining finished.
578 __blk_drain_queue(q, true);
579 queue_flag_set(QUEUE_FLAG_DEAD, q);
580 spin_unlock_irq(lock);
582 /* for synchronous bio-based driver finish in-flight integrity i/o */
583 blk_flush_integrity();
585 /* @q won't process any more request, flush async actions */
586 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
590 blk_mq_free_queue(q);
591 percpu_ref_exit(&q->q_usage_counter);
594 if (q->queue_lock != &q->__queue_lock)
595 q->queue_lock = &q->__queue_lock;
596 spin_unlock_irq(lock);
598 bdi_unregister(&q->backing_dev_info);
600 /* @q is and will stay empty, shutdown and put */
603 EXPORT_SYMBOL(blk_cleanup_queue);
605 /* Allocate memory local to the request queue */
606 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
608 int nid = (int)(long)data;
609 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
612 static void free_request_struct(void *element, void *unused)
614 kmem_cache_free(request_cachep, element);
617 int blk_init_rl(struct request_list *rl, struct request_queue *q,
620 if (unlikely(rl->rq_pool))
624 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
625 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
626 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
627 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
629 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
631 (void *)(long)q->node, gfp_mask,
639 void blk_exit_rl(struct request_list *rl)
642 mempool_destroy(rl->rq_pool);
645 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
647 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
649 EXPORT_SYMBOL(blk_alloc_queue);
651 int blk_queue_enter(struct request_queue *q, bool nowait)
656 if (percpu_ref_tryget_live(&q->q_usage_counter))
662 ret = wait_event_interruptible(q->mq_freeze_wq,
663 !atomic_read(&q->mq_freeze_depth) ||
665 if (blk_queue_dying(q))
672 void blk_queue_exit(struct request_queue *q)
674 percpu_ref_put(&q->q_usage_counter);
677 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
679 struct request_queue *q =
680 container_of(ref, struct request_queue, q_usage_counter);
682 wake_up_all(&q->mq_freeze_wq);
685 static void blk_rq_timed_out_timer(unsigned long data)
687 struct request_queue *q = (struct request_queue *)data;
689 kblockd_schedule_work(&q->timeout_work);
692 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
694 struct request_queue *q;
697 q = kmem_cache_alloc_node(blk_requestq_cachep,
698 gfp_mask | __GFP_ZERO, node_id);
702 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
706 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
710 q->backing_dev_info.ra_pages =
711 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
712 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
713 q->backing_dev_info.name = "block";
716 err = bdi_init(&q->backing_dev_info);
720 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
721 laptop_mode_timer_fn, (unsigned long) q);
722 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
723 INIT_LIST_HEAD(&q->queue_head);
724 INIT_LIST_HEAD(&q->timeout_list);
725 INIT_LIST_HEAD(&q->icq_list);
726 #ifdef CONFIG_BLK_CGROUP
727 INIT_LIST_HEAD(&q->blkg_list);
729 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
731 kobject_init(&q->kobj, &blk_queue_ktype);
733 mutex_init(&q->sysfs_lock);
734 spin_lock_init(&q->__queue_lock);
737 * By default initialize queue_lock to internal lock and driver can
738 * override it later if need be.
740 q->queue_lock = &q->__queue_lock;
743 * A queue starts its life with bypass turned on to avoid
744 * unnecessary bypass on/off overhead and nasty surprises during
745 * init. The initial bypass will be finished when the queue is
746 * registered by blk_register_queue().
749 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
751 init_waitqueue_head(&q->mq_freeze_wq);
754 * Init percpu_ref in atomic mode so that it's faster to shutdown.
755 * See blk_register_queue() for details.
757 if (percpu_ref_init(&q->q_usage_counter,
758 blk_queue_usage_counter_release,
759 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
762 if (blkcg_init_queue(q))
768 percpu_ref_exit(&q->q_usage_counter);
770 bdi_destroy(&q->backing_dev_info);
772 bioset_free(q->bio_split);
774 ida_simple_remove(&blk_queue_ida, q->id);
776 kmem_cache_free(blk_requestq_cachep, q);
779 EXPORT_SYMBOL(blk_alloc_queue_node);
782 * blk_init_queue - prepare a request queue for use with a block device
783 * @rfn: The function to be called to process requests that have been
784 * placed on the queue.
785 * @lock: Request queue spin lock
788 * If a block device wishes to use the standard request handling procedures,
789 * which sorts requests and coalesces adjacent requests, then it must
790 * call blk_init_queue(). The function @rfn will be called when there
791 * are requests on the queue that need to be processed. If the device
792 * supports plugging, then @rfn may not be called immediately when requests
793 * are available on the queue, but may be called at some time later instead.
794 * Plugged queues are generally unplugged when a buffer belonging to one
795 * of the requests on the queue is needed, or due to memory pressure.
797 * @rfn is not required, or even expected, to remove all requests off the
798 * queue, but only as many as it can handle at a time. If it does leave
799 * requests on the queue, it is responsible for arranging that the requests
800 * get dealt with eventually.
802 * The queue spin lock must be held while manipulating the requests on the
803 * request queue; this lock will be taken also from interrupt context, so irq
804 * disabling is needed for it.
806 * Function returns a pointer to the initialized request queue, or %NULL if
810 * blk_init_queue() must be paired with a blk_cleanup_queue() call
811 * when the block device is deactivated (such as at module unload).
814 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
816 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
818 EXPORT_SYMBOL(blk_init_queue);
820 struct request_queue *
821 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
823 struct request_queue *uninit_q, *q;
825 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
829 q = blk_init_allocated_queue(uninit_q, rfn, lock);
831 blk_cleanup_queue(uninit_q);
835 EXPORT_SYMBOL(blk_init_queue_node);
837 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
839 struct request_queue *
840 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
846 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
850 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
853 INIT_WORK(&q->timeout_work, blk_timeout_work);
855 q->prep_rq_fn = NULL;
856 q->unprep_rq_fn = NULL;
857 q->queue_flags |= QUEUE_FLAG_DEFAULT;
859 /* Override internal queue lock with supplied lock pointer */
861 q->queue_lock = lock;
864 * This also sets hw/phys segments, boundary and size
866 blk_queue_make_request(q, blk_queue_bio);
868 q->sg_reserved_size = INT_MAX;
870 /* Protect q->elevator from elevator_change */
871 mutex_lock(&q->sysfs_lock);
874 if (elevator_init(q, NULL)) {
875 mutex_unlock(&q->sysfs_lock);
879 mutex_unlock(&q->sysfs_lock);
884 blk_free_flush_queue(q->fq);
887 EXPORT_SYMBOL(blk_init_allocated_queue);
889 bool blk_get_queue(struct request_queue *q)
891 if (likely(!blk_queue_dying(q))) {
898 EXPORT_SYMBOL(blk_get_queue);
900 static inline void blk_free_request(struct request_list *rl, struct request *rq)
902 if (rq->rq_flags & RQF_ELVPRIV) {
903 elv_put_request(rl->q, rq);
905 put_io_context(rq->elv.icq->ioc);
908 mempool_free(rq, rl->rq_pool);
912 * ioc_batching returns true if the ioc is a valid batching request and
913 * should be given priority access to a request.
915 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
921 * Make sure the process is able to allocate at least 1 request
922 * even if the batch times out, otherwise we could theoretically
925 return ioc->nr_batch_requests == q->nr_batching ||
926 (ioc->nr_batch_requests > 0
927 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
931 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
932 * will cause the process to be a "batcher" on all queues in the system. This
933 * is the behaviour we want though - once it gets a wakeup it should be given
936 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
938 if (!ioc || ioc_batching(q, ioc))
941 ioc->nr_batch_requests = q->nr_batching;
942 ioc->last_waited = jiffies;
945 static void __freed_request(struct request_list *rl, int sync)
947 struct request_queue *q = rl->q;
949 if (rl->count[sync] < queue_congestion_off_threshold(q))
950 blk_clear_congested(rl, sync);
952 if (rl->count[sync] + 1 <= q->nr_requests) {
953 if (waitqueue_active(&rl->wait[sync]))
954 wake_up(&rl->wait[sync]);
956 blk_clear_rl_full(rl, sync);
961 * A request has just been released. Account for it, update the full and
962 * congestion status, wake up any waiters. Called under q->queue_lock.
964 static void freed_request(struct request_list *rl, bool sync,
965 req_flags_t rq_flags)
967 struct request_queue *q = rl->q;
971 if (rq_flags & RQF_ELVPRIV)
974 __freed_request(rl, sync);
976 if (unlikely(rl->starved[sync ^ 1]))
977 __freed_request(rl, sync ^ 1);
980 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
982 struct request_list *rl;
983 int on_thresh, off_thresh;
985 spin_lock_irq(q->queue_lock);
987 blk_queue_congestion_threshold(q);
988 on_thresh = queue_congestion_on_threshold(q);
989 off_thresh = queue_congestion_off_threshold(q);
991 blk_queue_for_each_rl(rl, q) {
992 if (rl->count[BLK_RW_SYNC] >= on_thresh)
993 blk_set_congested(rl, BLK_RW_SYNC);
994 else if (rl->count[BLK_RW_SYNC] < off_thresh)
995 blk_clear_congested(rl, BLK_RW_SYNC);
997 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
998 blk_set_congested(rl, BLK_RW_ASYNC);
999 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1000 blk_clear_congested(rl, BLK_RW_ASYNC);
1002 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1003 blk_set_rl_full(rl, BLK_RW_SYNC);
1005 blk_clear_rl_full(rl, BLK_RW_SYNC);
1006 wake_up(&rl->wait[BLK_RW_SYNC]);
1009 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1010 blk_set_rl_full(rl, BLK_RW_ASYNC);
1012 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1013 wake_up(&rl->wait[BLK_RW_ASYNC]);
1017 spin_unlock_irq(q->queue_lock);
1022 * Determine if elevator data should be initialized when allocating the
1023 * request associated with @bio.
1025 static bool blk_rq_should_init_elevator(struct bio *bio)
1031 * Flush requests do not use the elevator so skip initialization.
1032 * This allows a request to share the flush and elevator data.
1034 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
1041 * rq_ioc - determine io_context for request allocation
1042 * @bio: request being allocated is for this bio (can be %NULL)
1044 * Determine io_context to use for request allocation for @bio. May return
1045 * %NULL if %current->io_context doesn't exist.
1047 static struct io_context *rq_ioc(struct bio *bio)
1049 #ifdef CONFIG_BLK_CGROUP
1050 if (bio && bio->bi_ioc)
1053 return current->io_context;
1057 * __get_request - get a free request
1058 * @rl: request list to allocate from
1059 * @op: REQ_OP_READ/REQ_OP_WRITE
1060 * @op_flags: rq_flag_bits
1061 * @bio: bio to allocate request for (can be %NULL)
1062 * @gfp_mask: allocation mask
1064 * Get a free request from @q. This function may fail under memory
1065 * pressure or if @q is dead.
1067 * Must be called with @q->queue_lock held and,
1068 * Returns ERR_PTR on failure, with @q->queue_lock held.
1069 * Returns request pointer on success, with @q->queue_lock *not held*.
1071 static struct request *__get_request(struct request_list *rl, int op,
1072 int op_flags, struct bio *bio,
1075 struct request_queue *q = rl->q;
1077 struct elevator_type *et = q->elevator->type;
1078 struct io_context *ioc = rq_ioc(bio);
1079 struct io_cq *icq = NULL;
1080 const bool is_sync = rw_is_sync(op, op_flags) != 0;
1082 req_flags_t rq_flags = RQF_ALLOCED;
1084 if (unlikely(blk_queue_dying(q)))
1085 return ERR_PTR(-ENODEV);
1087 may_queue = elv_may_queue(q, op, op_flags);
1088 if (may_queue == ELV_MQUEUE_NO)
1091 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1092 if (rl->count[is_sync]+1 >= q->nr_requests) {
1094 * The queue will fill after this allocation, so set
1095 * it as full, and mark this process as "batching".
1096 * This process will be allowed to complete a batch of
1097 * requests, others will be blocked.
1099 if (!blk_rl_full(rl, is_sync)) {
1100 ioc_set_batching(q, ioc);
1101 blk_set_rl_full(rl, is_sync);
1103 if (may_queue != ELV_MQUEUE_MUST
1104 && !ioc_batching(q, ioc)) {
1106 * The queue is full and the allocating
1107 * process is not a "batcher", and not
1108 * exempted by the IO scheduler
1110 return ERR_PTR(-ENOMEM);
1114 blk_set_congested(rl, is_sync);
1118 * Only allow batching queuers to allocate up to 50% over the defined
1119 * limit of requests, otherwise we could have thousands of requests
1120 * allocated with any setting of ->nr_requests
1122 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1123 return ERR_PTR(-ENOMEM);
1125 q->nr_rqs[is_sync]++;
1126 rl->count[is_sync]++;
1127 rl->starved[is_sync] = 0;
1130 * Decide whether the new request will be managed by elevator. If
1131 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1132 * prevent the current elevator from being destroyed until the new
1133 * request is freed. This guarantees icq's won't be destroyed and
1134 * makes creating new ones safe.
1136 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1137 * it will be created after releasing queue_lock.
1139 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1140 rq_flags |= RQF_ELVPRIV;
1141 q->nr_rqs_elvpriv++;
1142 if (et->icq_cache && ioc)
1143 icq = ioc_lookup_icq(ioc, q);
1146 if (blk_queue_io_stat(q))
1147 rq_flags |= RQF_IO_STAT;
1148 spin_unlock_irq(q->queue_lock);
1150 /* allocate and init request */
1151 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1156 blk_rq_set_rl(rq, rl);
1157 req_set_op_attrs(rq, op, op_flags);
1158 rq->rq_flags = rq_flags;
1161 if (rq_flags & RQF_ELVPRIV) {
1162 if (unlikely(et->icq_cache && !icq)) {
1164 icq = ioc_create_icq(ioc, q, gfp_mask);
1170 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1173 /* @rq->elv.icq holds io_context until @rq is freed */
1175 get_io_context(icq->ioc);
1179 * ioc may be NULL here, and ioc_batching will be false. That's
1180 * OK, if the queue is under the request limit then requests need
1181 * not count toward the nr_batch_requests limit. There will always
1182 * be some limit enforced by BLK_BATCH_TIME.
1184 if (ioc_batching(q, ioc))
1185 ioc->nr_batch_requests--;
1187 trace_block_getrq(q, bio, op);
1192 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1193 * and may fail indefinitely under memory pressure and thus
1194 * shouldn't stall IO. Treat this request as !elvpriv. This will
1195 * disturb iosched and blkcg but weird is bettern than dead.
1197 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1198 __func__, dev_name(q->backing_dev_info.dev));
1200 rq->rq_flags &= ~RQF_ELVPRIV;
1203 spin_lock_irq(q->queue_lock);
1204 q->nr_rqs_elvpriv--;
1205 spin_unlock_irq(q->queue_lock);
1210 * Allocation failed presumably due to memory. Undo anything we
1211 * might have messed up.
1213 * Allocating task should really be put onto the front of the wait
1214 * queue, but this is pretty rare.
1216 spin_lock_irq(q->queue_lock);
1217 freed_request(rl, is_sync, rq_flags);
1220 * in the very unlikely event that allocation failed and no
1221 * requests for this direction was pending, mark us starved so that
1222 * freeing of a request in the other direction will notice
1223 * us. another possible fix would be to split the rq mempool into
1227 if (unlikely(rl->count[is_sync] == 0))
1228 rl->starved[is_sync] = 1;
1229 return ERR_PTR(-ENOMEM);
1233 * get_request - get a free request
1234 * @q: request_queue to allocate request from
1235 * @op: REQ_OP_READ/REQ_OP_WRITE
1236 * @op_flags: rq_flag_bits
1237 * @bio: bio to allocate request for (can be %NULL)
1238 * @gfp_mask: allocation mask
1240 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1241 * this function keeps retrying under memory pressure and fails iff @q is dead.
1243 * Must be called with @q->queue_lock held and,
1244 * Returns ERR_PTR on failure, with @q->queue_lock held.
1245 * Returns request pointer on success, with @q->queue_lock *not held*.
1247 static struct request *get_request(struct request_queue *q, int op,
1248 int op_flags, struct bio *bio,
1251 const bool is_sync = rw_is_sync(op, op_flags) != 0;
1253 struct request_list *rl;
1256 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1258 rq = __get_request(rl, op, op_flags, bio, gfp_mask);
1262 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1267 /* wait on @rl and retry */
1268 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1269 TASK_UNINTERRUPTIBLE);
1271 trace_block_sleeprq(q, bio, op);
1273 spin_unlock_irq(q->queue_lock);
1277 * After sleeping, we become a "batching" process and will be able
1278 * to allocate at least one request, and up to a big batch of them
1279 * for a small period time. See ioc_batching, ioc_set_batching
1281 ioc_set_batching(q, current->io_context);
1283 spin_lock_irq(q->queue_lock);
1284 finish_wait(&rl->wait[is_sync], &wait);
1289 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1294 BUG_ON(rw != READ && rw != WRITE);
1296 /* create ioc upfront */
1297 create_io_context(gfp_mask, q->node);
1299 spin_lock_irq(q->queue_lock);
1300 rq = get_request(q, rw, 0, NULL, gfp_mask);
1302 spin_unlock_irq(q->queue_lock);
1306 /* q->queue_lock is unlocked at this point */
1308 rq->__sector = (sector_t) -1;
1309 rq->bio = rq->biotail = NULL;
1313 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1316 return blk_mq_alloc_request(q, rw,
1317 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1318 0 : BLK_MQ_REQ_NOWAIT);
1320 return blk_old_get_request(q, rw, gfp_mask);
1322 EXPORT_SYMBOL(blk_get_request);
1325 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1326 * @rq: request to be initialized
1329 void blk_rq_set_block_pc(struct request *rq)
1331 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1332 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1334 EXPORT_SYMBOL(blk_rq_set_block_pc);
1337 * blk_requeue_request - put a request back on queue
1338 * @q: request queue where request should be inserted
1339 * @rq: request to be inserted
1342 * Drivers often keep queueing requests until the hardware cannot accept
1343 * more, when that condition happens we need to put the request back
1344 * on the queue. Must be called with queue lock held.
1346 void blk_requeue_request(struct request_queue *q, struct request *rq)
1348 blk_delete_timer(rq);
1349 blk_clear_rq_complete(rq);
1350 trace_block_rq_requeue(q, rq);
1352 if (rq->rq_flags & RQF_QUEUED)
1353 blk_queue_end_tag(q, rq);
1355 BUG_ON(blk_queued_rq(rq));
1357 elv_requeue_request(q, rq);
1359 EXPORT_SYMBOL(blk_requeue_request);
1361 static void add_acct_request(struct request_queue *q, struct request *rq,
1364 blk_account_io_start(rq, true);
1365 __elv_add_request(q, rq, where);
1368 static void part_round_stats_single(int cpu, struct hd_struct *part,
1373 if (now == part->stamp)
1376 inflight = part_in_flight(part);
1378 __part_stat_add(cpu, part, time_in_queue,
1379 inflight * (now - part->stamp));
1380 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1386 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1387 * @cpu: cpu number for stats access
1388 * @part: target partition
1390 * The average IO queue length and utilisation statistics are maintained
1391 * by observing the current state of the queue length and the amount of
1392 * time it has been in this state for.
1394 * Normally, that accounting is done on IO completion, but that can result
1395 * in more than a second's worth of IO being accounted for within any one
1396 * second, leading to >100% utilisation. To deal with that, we call this
1397 * function to do a round-off before returning the results when reading
1398 * /proc/diskstats. This accounts immediately for all queue usage up to
1399 * the current jiffies and restarts the counters again.
1401 void part_round_stats(int cpu, struct hd_struct *part)
1403 unsigned long now = jiffies;
1406 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1407 part_round_stats_single(cpu, part, now);
1409 EXPORT_SYMBOL_GPL(part_round_stats);
1412 static void blk_pm_put_request(struct request *rq)
1414 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1415 pm_runtime_mark_last_busy(rq->q->dev);
1418 static inline void blk_pm_put_request(struct request *rq) {}
1422 * queue lock must be held
1424 void __blk_put_request(struct request_queue *q, struct request *req)
1426 req_flags_t rq_flags = req->rq_flags;
1432 blk_mq_free_request(req);
1436 blk_pm_put_request(req);
1438 elv_completed_request(q, req);
1440 /* this is a bio leak */
1441 WARN_ON(req->bio != NULL);
1444 * Request may not have originated from ll_rw_blk. if not,
1445 * it didn't come out of our reserved rq pools
1447 if (rq_flags & RQF_ALLOCED) {
1448 struct request_list *rl = blk_rq_rl(req);
1449 bool sync = rw_is_sync(req_op(req), req->cmd_flags);
1451 BUG_ON(!list_empty(&req->queuelist));
1452 BUG_ON(ELV_ON_HASH(req));
1454 blk_free_request(rl, req);
1455 freed_request(rl, sync, rq_flags);
1459 EXPORT_SYMBOL_GPL(__blk_put_request);
1461 void blk_put_request(struct request *req)
1463 struct request_queue *q = req->q;
1466 blk_mq_free_request(req);
1468 unsigned long flags;
1470 spin_lock_irqsave(q->queue_lock, flags);
1471 __blk_put_request(q, req);
1472 spin_unlock_irqrestore(q->queue_lock, flags);
1475 EXPORT_SYMBOL(blk_put_request);
1478 * blk_add_request_payload - add a payload to a request
1479 * @rq: request to update
1480 * @page: page backing the payload
1481 * @offset: offset in page
1482 * @len: length of the payload.
1484 * This allows to later add a payload to an already submitted request by
1485 * a block driver. The driver needs to take care of freeing the payload
1488 * Note that this is a quite horrible hack and nothing but handling of
1489 * discard requests should ever use it.
1491 void blk_add_request_payload(struct request *rq, struct page *page,
1492 int offset, unsigned int len)
1494 struct bio *bio = rq->bio;
1496 bio->bi_io_vec->bv_page = page;
1497 bio->bi_io_vec->bv_offset = offset;
1498 bio->bi_io_vec->bv_len = len;
1500 bio->bi_iter.bi_size = len;
1502 bio->bi_phys_segments = 1;
1504 rq->__data_len = rq->resid_len = len;
1505 rq->nr_phys_segments = 1;
1507 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1509 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1512 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1514 if (!ll_back_merge_fn(q, req, bio))
1517 trace_block_bio_backmerge(q, req, bio);
1519 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1520 blk_rq_set_mixed_merge(req);
1522 req->biotail->bi_next = bio;
1524 req->__data_len += bio->bi_iter.bi_size;
1525 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1527 blk_account_io_start(req, false);
1531 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1534 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1536 if (!ll_front_merge_fn(q, req, bio))
1539 trace_block_bio_frontmerge(q, req, bio);
1541 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1542 blk_rq_set_mixed_merge(req);
1544 bio->bi_next = req->bio;
1547 req->__sector = bio->bi_iter.bi_sector;
1548 req->__data_len += bio->bi_iter.bi_size;
1549 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1551 blk_account_io_start(req, false);
1556 * blk_attempt_plug_merge - try to merge with %current's plugged list
1557 * @q: request_queue new bio is being queued at
1558 * @bio: new bio being queued
1559 * @request_count: out parameter for number of traversed plugged requests
1560 * @same_queue_rq: pointer to &struct request that gets filled in when
1561 * another request associated with @q is found on the plug list
1562 * (optional, may be %NULL)
1564 * Determine whether @bio being queued on @q can be merged with a request
1565 * on %current's plugged list. Returns %true if merge was successful,
1568 * Plugging coalesces IOs from the same issuer for the same purpose without
1569 * going through @q->queue_lock. As such it's more of an issuing mechanism
1570 * than scheduling, and the request, while may have elvpriv data, is not
1571 * added on the elevator at this point. In addition, we don't have
1572 * reliable access to the elevator outside queue lock. Only check basic
1573 * merging parameters without querying the elevator.
1575 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1577 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1578 unsigned int *request_count,
1579 struct request **same_queue_rq)
1581 struct blk_plug *plug;
1584 struct list_head *plug_list;
1586 plug = current->plug;
1592 plug_list = &plug->mq_list;
1594 plug_list = &plug->list;
1596 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1602 * Only blk-mq multiple hardware queues case checks the
1603 * rq in the same queue, there should be only one such
1607 *same_queue_rq = rq;
1610 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1613 el_ret = blk_try_merge(rq, bio);
1614 if (el_ret == ELEVATOR_BACK_MERGE) {
1615 ret = bio_attempt_back_merge(q, rq, bio);
1618 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1619 ret = bio_attempt_front_merge(q, rq, bio);
1628 unsigned int blk_plug_queued_count(struct request_queue *q)
1630 struct blk_plug *plug;
1632 struct list_head *plug_list;
1633 unsigned int ret = 0;
1635 plug = current->plug;
1640 plug_list = &plug->mq_list;
1642 plug_list = &plug->list;
1644 list_for_each_entry(rq, plug_list, queuelist) {
1652 void init_request_from_bio(struct request *req, struct bio *bio)
1654 req->cmd_type = REQ_TYPE_FS;
1656 req->cmd_flags |= bio->bi_opf & REQ_COMMON_MASK;
1657 if (bio->bi_opf & REQ_RAHEAD)
1658 req->cmd_flags |= REQ_FAILFAST_MASK;
1661 req->__sector = bio->bi_iter.bi_sector;
1662 req->ioprio = bio_prio(bio);
1663 blk_rq_bio_prep(req->q, req, bio);
1666 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1668 const bool sync = !!(bio->bi_opf & REQ_SYNC);
1669 struct blk_plug *plug;
1670 int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
1671 struct request *req;
1672 unsigned int request_count = 0;
1675 * low level driver can indicate that it wants pages above a
1676 * certain limit bounced to low memory (ie for highmem, or even
1677 * ISA dma in theory)
1679 blk_queue_bounce(q, &bio);
1681 blk_queue_split(q, &bio, q->bio_split);
1683 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1684 bio->bi_error = -EIO;
1686 return BLK_QC_T_NONE;
1689 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
1690 spin_lock_irq(q->queue_lock);
1691 where = ELEVATOR_INSERT_FLUSH;
1696 * Check if we can merge with the plugged list before grabbing
1699 if (!blk_queue_nomerges(q)) {
1700 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1701 return BLK_QC_T_NONE;
1703 request_count = blk_plug_queued_count(q);
1705 spin_lock_irq(q->queue_lock);
1707 el_ret = elv_merge(q, &req, bio);
1708 if (el_ret == ELEVATOR_BACK_MERGE) {
1709 if (bio_attempt_back_merge(q, req, bio)) {
1710 elv_bio_merged(q, req, bio);
1711 if (!attempt_back_merge(q, req))
1712 elv_merged_request(q, req, el_ret);
1715 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1716 if (bio_attempt_front_merge(q, req, bio)) {
1717 elv_bio_merged(q, req, bio);
1718 if (!attempt_front_merge(q, req))
1719 elv_merged_request(q, req, el_ret);
1726 * This sync check and mask will be re-done in init_request_from_bio(),
1727 * but we need to set it earlier to expose the sync flag to the
1728 * rq allocator and io schedulers.
1731 rw_flags |= REQ_SYNC;
1734 * Add in META/PRIO flags, if set, before we get to the IO scheduler
1736 rw_flags |= (bio->bi_opf & (REQ_META | REQ_PRIO));
1739 * Grab a free request. This is might sleep but can not fail.
1740 * Returns with the queue unlocked.
1742 req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
1744 bio->bi_error = PTR_ERR(req);
1750 * After dropping the lock and possibly sleeping here, our request
1751 * may now be mergeable after it had proven unmergeable (above).
1752 * We don't worry about that case for efficiency. It won't happen
1753 * often, and the elevators are able to handle it.
1755 init_request_from_bio(req, bio);
1757 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1758 req->cpu = raw_smp_processor_id();
1760 plug = current->plug;
1763 * If this is the first request added after a plug, fire
1767 trace_block_plug(q);
1769 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1770 blk_flush_plug_list(plug, false);
1771 trace_block_plug(q);
1774 list_add_tail(&req->queuelist, &plug->list);
1775 blk_account_io_start(req, true);
1777 spin_lock_irq(q->queue_lock);
1778 add_acct_request(q, req, where);
1781 spin_unlock_irq(q->queue_lock);
1784 return BLK_QC_T_NONE;
1788 * If bio->bi_dev is a partition, remap the location
1790 static inline void blk_partition_remap(struct bio *bio)
1792 struct block_device *bdev = bio->bi_bdev;
1794 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1795 struct hd_struct *p = bdev->bd_part;
1797 bio->bi_iter.bi_sector += p->start_sect;
1798 bio->bi_bdev = bdev->bd_contains;
1800 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1802 bio->bi_iter.bi_sector - p->start_sect);
1806 static void handle_bad_sector(struct bio *bio)
1808 char b[BDEVNAME_SIZE];
1810 printk(KERN_INFO "attempt to access beyond end of device\n");
1811 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1812 bdevname(bio->bi_bdev, b),
1814 (unsigned long long)bio_end_sector(bio),
1815 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1818 #ifdef CONFIG_FAIL_MAKE_REQUEST
1820 static DECLARE_FAULT_ATTR(fail_make_request);
1822 static int __init setup_fail_make_request(char *str)
1824 return setup_fault_attr(&fail_make_request, str);
1826 __setup("fail_make_request=", setup_fail_make_request);
1828 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1830 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1833 static int __init fail_make_request_debugfs(void)
1835 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1836 NULL, &fail_make_request);
1838 return PTR_ERR_OR_ZERO(dir);
1841 late_initcall(fail_make_request_debugfs);
1843 #else /* CONFIG_FAIL_MAKE_REQUEST */
1845 static inline bool should_fail_request(struct hd_struct *part,
1851 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1854 * Check whether this bio extends beyond the end of the device.
1856 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1863 /* Test device or partition size, when known. */
1864 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1866 sector_t sector = bio->bi_iter.bi_sector;
1868 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1870 * This may well happen - the kernel calls bread()
1871 * without checking the size of the device, e.g., when
1872 * mounting a device.
1874 handle_bad_sector(bio);
1882 static noinline_for_stack bool
1883 generic_make_request_checks(struct bio *bio)
1885 struct request_queue *q;
1886 int nr_sectors = bio_sectors(bio);
1888 char b[BDEVNAME_SIZE];
1889 struct hd_struct *part;
1893 if (bio_check_eod(bio, nr_sectors))
1896 q = bdev_get_queue(bio->bi_bdev);
1899 "generic_make_request: Trying to access "
1900 "nonexistent block-device %s (%Lu)\n",
1901 bdevname(bio->bi_bdev, b),
1902 (long long) bio->bi_iter.bi_sector);
1906 part = bio->bi_bdev->bd_part;
1907 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1908 should_fail_request(&part_to_disk(part)->part0,
1909 bio->bi_iter.bi_size))
1913 * If this device has partitions, remap block n
1914 * of partition p to block n+start(p) of the disk.
1916 blk_partition_remap(bio);
1918 if (bio_check_eod(bio, nr_sectors))
1922 * Filter flush bio's early so that make_request based
1923 * drivers without flush support don't have to worry
1926 if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
1927 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1928 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1935 switch (bio_op(bio)) {
1936 case REQ_OP_DISCARD:
1937 if (!blk_queue_discard(q))
1940 case REQ_OP_SECURE_ERASE:
1941 if (!blk_queue_secure_erase(q))
1944 case REQ_OP_WRITE_SAME:
1945 if (!bdev_write_same(bio->bi_bdev))
1947 case REQ_OP_ZONE_REPORT:
1948 case REQ_OP_ZONE_RESET:
1949 if (!bdev_is_zoned(bio->bi_bdev))
1957 * Various block parts want %current->io_context and lazy ioc
1958 * allocation ends up trading a lot of pain for a small amount of
1959 * memory. Just allocate it upfront. This may fail and block
1960 * layer knows how to live with it.
1962 create_io_context(GFP_ATOMIC, q->node);
1964 if (!blkcg_bio_issue_check(q, bio))
1967 trace_block_bio_queue(q, bio);
1973 bio->bi_error = err;
1979 * generic_make_request - hand a buffer to its device driver for I/O
1980 * @bio: The bio describing the location in memory and on the device.
1982 * generic_make_request() is used to make I/O requests of block
1983 * devices. It is passed a &struct bio, which describes the I/O that needs
1986 * generic_make_request() does not return any status. The
1987 * success/failure status of the request, along with notification of
1988 * completion, is delivered asynchronously through the bio->bi_end_io
1989 * function described (one day) else where.
1991 * The caller of generic_make_request must make sure that bi_io_vec
1992 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1993 * set to describe the device address, and the
1994 * bi_end_io and optionally bi_private are set to describe how
1995 * completion notification should be signaled.
1997 * generic_make_request and the drivers it calls may use bi_next if this
1998 * bio happens to be merged with someone else, and may resubmit the bio to
1999 * a lower device by calling into generic_make_request recursively, which
2000 * means the bio should NOT be touched after the call to ->make_request_fn.
2002 blk_qc_t generic_make_request(struct bio *bio)
2004 struct bio_list bio_list_on_stack;
2005 blk_qc_t ret = BLK_QC_T_NONE;
2007 if (!generic_make_request_checks(bio))
2011 * We only want one ->make_request_fn to be active at a time, else
2012 * stack usage with stacked devices could be a problem. So use
2013 * current->bio_list to keep a list of requests submited by a
2014 * make_request_fn function. current->bio_list is also used as a
2015 * flag to say if generic_make_request is currently active in this
2016 * task or not. If it is NULL, then no make_request is active. If
2017 * it is non-NULL, then a make_request is active, and new requests
2018 * should be added at the tail
2020 if (current->bio_list) {
2021 bio_list_add(current->bio_list, bio);
2025 /* following loop may be a bit non-obvious, and so deserves some
2027 * Before entering the loop, bio->bi_next is NULL (as all callers
2028 * ensure that) so we have a list with a single bio.
2029 * We pretend that we have just taken it off a longer list, so
2030 * we assign bio_list to a pointer to the bio_list_on_stack,
2031 * thus initialising the bio_list of new bios to be
2032 * added. ->make_request() may indeed add some more bios
2033 * through a recursive call to generic_make_request. If it
2034 * did, we find a non-NULL value in bio_list and re-enter the loop
2035 * from the top. In this case we really did just take the bio
2036 * of the top of the list (no pretending) and so remove it from
2037 * bio_list, and call into ->make_request() again.
2039 BUG_ON(bio->bi_next);
2040 bio_list_init(&bio_list_on_stack);
2041 current->bio_list = &bio_list_on_stack;
2043 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2045 if (likely(blk_queue_enter(q, false) == 0)) {
2046 ret = q->make_request_fn(q, bio);
2050 bio = bio_list_pop(current->bio_list);
2052 struct bio *bio_next = bio_list_pop(current->bio_list);
2058 current->bio_list = NULL; /* deactivate */
2063 EXPORT_SYMBOL(generic_make_request);
2066 * submit_bio - submit a bio to the block device layer for I/O
2067 * @bio: The &struct bio which describes the I/O
2069 * submit_bio() is very similar in purpose to generic_make_request(), and
2070 * uses that function to do most of the work. Both are fairly rough
2071 * interfaces; @bio must be presetup and ready for I/O.
2074 blk_qc_t submit_bio(struct bio *bio)
2077 * If it's a regular read/write or a barrier with data attached,
2078 * go through the normal accounting stuff before submission.
2080 if (bio_has_data(bio)) {
2083 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2084 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2086 count = bio_sectors(bio);
2088 if (op_is_write(bio_op(bio))) {
2089 count_vm_events(PGPGOUT, count);
2091 task_io_account_read(bio->bi_iter.bi_size);
2092 count_vm_events(PGPGIN, count);
2095 if (unlikely(block_dump)) {
2096 char b[BDEVNAME_SIZE];
2097 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2098 current->comm, task_pid_nr(current),
2099 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2100 (unsigned long long)bio->bi_iter.bi_sector,
2101 bdevname(bio->bi_bdev, b),
2106 return generic_make_request(bio);
2108 EXPORT_SYMBOL(submit_bio);
2111 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2112 * for new the queue limits
2114 * @rq: the request being checked
2117 * @rq may have been made based on weaker limitations of upper-level queues
2118 * in request stacking drivers, and it may violate the limitation of @q.
2119 * Since the block layer and the underlying device driver trust @rq
2120 * after it is inserted to @q, it should be checked against @q before
2121 * the insertion using this generic function.
2123 * Request stacking drivers like request-based dm may change the queue
2124 * limits when retrying requests on other queues. Those requests need
2125 * to be checked against the new queue limits again during dispatch.
2127 static int blk_cloned_rq_check_limits(struct request_queue *q,
2130 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2131 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2136 * queue's settings related to segment counting like q->bounce_pfn
2137 * may differ from that of other stacking queues.
2138 * Recalculate it to check the request correctly on this queue's
2141 blk_recalc_rq_segments(rq);
2142 if (rq->nr_phys_segments > queue_max_segments(q)) {
2143 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2151 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2152 * @q: the queue to submit the request
2153 * @rq: the request being queued
2155 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2157 unsigned long flags;
2158 int where = ELEVATOR_INSERT_BACK;
2160 if (blk_cloned_rq_check_limits(q, rq))
2164 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2168 if (blk_queue_io_stat(q))
2169 blk_account_io_start(rq, true);
2170 blk_mq_insert_request(rq, false, true, false);
2174 spin_lock_irqsave(q->queue_lock, flags);
2175 if (unlikely(blk_queue_dying(q))) {
2176 spin_unlock_irqrestore(q->queue_lock, flags);
2181 * Submitting request must be dequeued before calling this function
2182 * because it will be linked to another request_queue
2184 BUG_ON(blk_queued_rq(rq));
2186 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2187 where = ELEVATOR_INSERT_FLUSH;
2189 add_acct_request(q, rq, where);
2190 if (where == ELEVATOR_INSERT_FLUSH)
2192 spin_unlock_irqrestore(q->queue_lock, flags);
2196 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2199 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2200 * @rq: request to examine
2203 * A request could be merge of IOs which require different failure
2204 * handling. This function determines the number of bytes which
2205 * can be failed from the beginning of the request without
2206 * crossing into area which need to be retried further.
2209 * The number of bytes to fail.
2212 * queue_lock must be held.
2214 unsigned int blk_rq_err_bytes(const struct request *rq)
2216 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2217 unsigned int bytes = 0;
2220 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2221 return blk_rq_bytes(rq);
2224 * Currently the only 'mixing' which can happen is between
2225 * different fastfail types. We can safely fail portions
2226 * which have all the failfast bits that the first one has -
2227 * the ones which are at least as eager to fail as the first
2230 for (bio = rq->bio; bio; bio = bio->bi_next) {
2231 if ((bio->bi_opf & ff) != ff)
2233 bytes += bio->bi_iter.bi_size;
2236 /* this could lead to infinite loop */
2237 BUG_ON(blk_rq_bytes(rq) && !bytes);
2240 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2242 void blk_account_io_completion(struct request *req, unsigned int bytes)
2244 if (blk_do_io_stat(req)) {
2245 const int rw = rq_data_dir(req);
2246 struct hd_struct *part;
2249 cpu = part_stat_lock();
2251 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2256 void blk_account_io_done(struct request *req)
2259 * Account IO completion. flush_rq isn't accounted as a
2260 * normal IO on queueing nor completion. Accounting the
2261 * containing request is enough.
2263 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2264 unsigned long duration = jiffies - req->start_time;
2265 const int rw = rq_data_dir(req);
2266 struct hd_struct *part;
2269 cpu = part_stat_lock();
2272 part_stat_inc(cpu, part, ios[rw]);
2273 part_stat_add(cpu, part, ticks[rw], duration);
2274 part_round_stats(cpu, part);
2275 part_dec_in_flight(part, rw);
2277 hd_struct_put(part);
2284 * Don't process normal requests when queue is suspended
2285 * or in the process of suspending/resuming
2287 static struct request *blk_pm_peek_request(struct request_queue *q,
2290 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2291 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2297 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2304 void blk_account_io_start(struct request *rq, bool new_io)
2306 struct hd_struct *part;
2307 int rw = rq_data_dir(rq);
2310 if (!blk_do_io_stat(rq))
2313 cpu = part_stat_lock();
2317 part_stat_inc(cpu, part, merges[rw]);
2319 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2320 if (!hd_struct_try_get(part)) {
2322 * The partition is already being removed,
2323 * the request will be accounted on the disk only
2325 * We take a reference on disk->part0 although that
2326 * partition will never be deleted, so we can treat
2327 * it as any other partition.
2329 part = &rq->rq_disk->part0;
2330 hd_struct_get(part);
2332 part_round_stats(cpu, part);
2333 part_inc_in_flight(part, rw);
2341 * blk_peek_request - peek at the top of a request queue
2342 * @q: request queue to peek at
2345 * Return the request at the top of @q. The returned request
2346 * should be started using blk_start_request() before LLD starts
2350 * Pointer to the request at the top of @q if available. Null
2354 * queue_lock must be held.
2356 struct request *blk_peek_request(struct request_queue *q)
2361 while ((rq = __elv_next_request(q)) != NULL) {
2363 rq = blk_pm_peek_request(q, rq);
2367 if (!(rq->rq_flags & RQF_STARTED)) {
2369 * This is the first time the device driver
2370 * sees this request (possibly after
2371 * requeueing). Notify IO scheduler.
2373 if (rq->rq_flags & RQF_SORTED)
2374 elv_activate_rq(q, rq);
2377 * just mark as started even if we don't start
2378 * it, a request that has been delayed should
2379 * not be passed by new incoming requests
2381 rq->rq_flags |= RQF_STARTED;
2382 trace_block_rq_issue(q, rq);
2385 if (!q->boundary_rq || q->boundary_rq == rq) {
2386 q->end_sector = rq_end_sector(rq);
2387 q->boundary_rq = NULL;
2390 if (rq->rq_flags & RQF_DONTPREP)
2393 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2395 * make sure space for the drain appears we
2396 * know we can do this because max_hw_segments
2397 * has been adjusted to be one fewer than the
2400 rq->nr_phys_segments++;
2406 ret = q->prep_rq_fn(q, rq);
2407 if (ret == BLKPREP_OK) {
2409 } else if (ret == BLKPREP_DEFER) {
2411 * the request may have been (partially) prepped.
2412 * we need to keep this request in the front to
2413 * avoid resource deadlock. RQF_STARTED will
2414 * prevent other fs requests from passing this one.
2416 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2417 !(rq->rq_flags & RQF_DONTPREP)) {
2419 * remove the space for the drain we added
2420 * so that we don't add it again
2422 --rq->nr_phys_segments;
2427 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2428 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2430 rq->rq_flags |= RQF_QUIET;
2432 * Mark this request as started so we don't trigger
2433 * any debug logic in the end I/O path.
2435 blk_start_request(rq);
2436 __blk_end_request_all(rq, err);
2438 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2445 EXPORT_SYMBOL(blk_peek_request);
2447 void blk_dequeue_request(struct request *rq)
2449 struct request_queue *q = rq->q;
2451 BUG_ON(list_empty(&rq->queuelist));
2452 BUG_ON(ELV_ON_HASH(rq));
2454 list_del_init(&rq->queuelist);
2457 * the time frame between a request being removed from the lists
2458 * and to it is freed is accounted as io that is in progress at
2461 if (blk_account_rq(rq)) {
2462 q->in_flight[rq_is_sync(rq)]++;
2463 set_io_start_time_ns(rq);
2468 * blk_start_request - start request processing on the driver
2469 * @req: request to dequeue
2472 * Dequeue @req and start timeout timer on it. This hands off the
2473 * request to the driver.
2475 * Block internal functions which don't want to start timer should
2476 * call blk_dequeue_request().
2479 * queue_lock must be held.
2481 void blk_start_request(struct request *req)
2483 blk_dequeue_request(req);
2486 * We are now handing the request to the hardware, initialize
2487 * resid_len to full count and add the timeout handler.
2489 req->resid_len = blk_rq_bytes(req);
2490 if (unlikely(blk_bidi_rq(req)))
2491 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2493 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2496 EXPORT_SYMBOL(blk_start_request);
2499 * blk_fetch_request - fetch a request from a request queue
2500 * @q: request queue to fetch a request from
2503 * Return the request at the top of @q. The request is started on
2504 * return and LLD can start processing it immediately.
2507 * Pointer to the request at the top of @q if available. Null
2511 * queue_lock must be held.
2513 struct request *blk_fetch_request(struct request_queue *q)
2517 rq = blk_peek_request(q);
2519 blk_start_request(rq);
2522 EXPORT_SYMBOL(blk_fetch_request);
2525 * blk_update_request - Special helper function for request stacking drivers
2526 * @req: the request being processed
2527 * @error: %0 for success, < %0 for error
2528 * @nr_bytes: number of bytes to complete @req
2531 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2532 * the request structure even if @req doesn't have leftover.
2533 * If @req has leftover, sets it up for the next range of segments.
2535 * This special helper function is only for request stacking drivers
2536 * (e.g. request-based dm) so that they can handle partial completion.
2537 * Actual device drivers should use blk_end_request instead.
2539 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2540 * %false return from this function.
2543 * %false - this request doesn't have any more data
2544 * %true - this request has more data
2546 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2550 trace_block_rq_complete(req->q, req, nr_bytes);
2556 * For fs requests, rq is just carrier of independent bio's
2557 * and each partial completion should be handled separately.
2558 * Reset per-request error on each partial completion.
2560 * TODO: tj: This is too subtle. It would be better to let
2561 * low level drivers do what they see fit.
2563 if (req->cmd_type == REQ_TYPE_FS)
2566 if (error && req->cmd_type == REQ_TYPE_FS &&
2567 !(req->rq_flags & RQF_QUIET)) {
2572 error_type = "recoverable transport";
2575 error_type = "critical target";
2578 error_type = "critical nexus";
2581 error_type = "timeout";
2584 error_type = "critical space allocation";
2587 error_type = "critical medium";
2594 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2595 __func__, error_type, req->rq_disk ?
2596 req->rq_disk->disk_name : "?",
2597 (unsigned long long)blk_rq_pos(req));
2601 blk_account_io_completion(req, nr_bytes);
2605 struct bio *bio = req->bio;
2606 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2608 if (bio_bytes == bio->bi_iter.bi_size)
2609 req->bio = bio->bi_next;
2611 req_bio_endio(req, bio, bio_bytes, error);
2613 total_bytes += bio_bytes;
2614 nr_bytes -= bio_bytes;
2625 * Reset counters so that the request stacking driver
2626 * can find how many bytes remain in the request
2629 req->__data_len = 0;
2633 req->__data_len -= total_bytes;
2635 /* update sector only for requests with clear definition of sector */
2636 if (req->cmd_type == REQ_TYPE_FS)
2637 req->__sector += total_bytes >> 9;
2639 /* mixed attributes always follow the first bio */
2640 if (req->rq_flags & RQF_MIXED_MERGE) {
2641 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2642 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2646 * If total number of sectors is less than the first segment
2647 * size, something has gone terribly wrong.
2649 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2650 blk_dump_rq_flags(req, "request botched");
2651 req->__data_len = blk_rq_cur_bytes(req);
2654 /* recalculate the number of segments */
2655 blk_recalc_rq_segments(req);
2659 EXPORT_SYMBOL_GPL(blk_update_request);
2661 static bool blk_update_bidi_request(struct request *rq, int error,
2662 unsigned int nr_bytes,
2663 unsigned int bidi_bytes)
2665 if (blk_update_request(rq, error, nr_bytes))
2668 /* Bidi request must be completed as a whole */
2669 if (unlikely(blk_bidi_rq(rq)) &&
2670 blk_update_request(rq->next_rq, error, bidi_bytes))
2673 if (blk_queue_add_random(rq->q))
2674 add_disk_randomness(rq->rq_disk);
2680 * blk_unprep_request - unprepare a request
2683 * This function makes a request ready for complete resubmission (or
2684 * completion). It happens only after all error handling is complete,
2685 * so represents the appropriate moment to deallocate any resources
2686 * that were allocated to the request in the prep_rq_fn. The queue
2687 * lock is held when calling this.
2689 void blk_unprep_request(struct request *req)
2691 struct request_queue *q = req->q;
2693 req->rq_flags &= ~RQF_DONTPREP;
2694 if (q->unprep_rq_fn)
2695 q->unprep_rq_fn(q, req);
2697 EXPORT_SYMBOL_GPL(blk_unprep_request);
2700 * queue lock must be held
2702 void blk_finish_request(struct request *req, int error)
2704 if (req->rq_flags & RQF_QUEUED)
2705 blk_queue_end_tag(req->q, req);
2707 BUG_ON(blk_queued_rq(req));
2709 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2710 laptop_io_completion(&req->q->backing_dev_info);
2712 blk_delete_timer(req);
2714 if (req->rq_flags & RQF_DONTPREP)
2715 blk_unprep_request(req);
2717 blk_account_io_done(req);
2720 req->end_io(req, error);
2722 if (blk_bidi_rq(req))
2723 __blk_put_request(req->next_rq->q, req->next_rq);
2725 __blk_put_request(req->q, req);
2728 EXPORT_SYMBOL(blk_finish_request);
2731 * blk_end_bidi_request - Complete a bidi request
2732 * @rq: the request to complete
2733 * @error: %0 for success, < %0 for error
2734 * @nr_bytes: number of bytes to complete @rq
2735 * @bidi_bytes: number of bytes to complete @rq->next_rq
2738 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2739 * Drivers that supports bidi can safely call this member for any
2740 * type of request, bidi or uni. In the later case @bidi_bytes is
2744 * %false - we are done with this request
2745 * %true - still buffers pending for this request
2747 static bool blk_end_bidi_request(struct request *rq, int error,
2748 unsigned int nr_bytes, unsigned int bidi_bytes)
2750 struct request_queue *q = rq->q;
2751 unsigned long flags;
2753 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2756 spin_lock_irqsave(q->queue_lock, flags);
2757 blk_finish_request(rq, error);
2758 spin_unlock_irqrestore(q->queue_lock, flags);
2764 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2765 * @rq: the request to complete
2766 * @error: %0 for success, < %0 for error
2767 * @nr_bytes: number of bytes to complete @rq
2768 * @bidi_bytes: number of bytes to complete @rq->next_rq
2771 * Identical to blk_end_bidi_request() except that queue lock is
2772 * assumed to be locked on entry and remains so on return.
2775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
2778 bool __blk_end_bidi_request(struct request *rq, int error,
2779 unsigned int nr_bytes, unsigned int bidi_bytes)
2781 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2784 blk_finish_request(rq, error);
2790 * blk_end_request - Helper function for drivers to complete the request.
2791 * @rq: the request being processed
2792 * @error: %0 for success, < %0 for error
2793 * @nr_bytes: number of bytes to complete
2796 * Ends I/O on a number of bytes attached to @rq.
2797 * If @rq has leftover, sets it up for the next range of segments.
2800 * %false - we are done with this request
2801 * %true - still buffers pending for this request
2803 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2805 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2807 EXPORT_SYMBOL(blk_end_request);
2810 * blk_end_request_all - Helper function for drives to finish the request.
2811 * @rq: the request to finish
2812 * @error: %0 for success, < %0 for error
2815 * Completely finish @rq.
2817 void blk_end_request_all(struct request *rq, int error)
2820 unsigned int bidi_bytes = 0;
2822 if (unlikely(blk_bidi_rq(rq)))
2823 bidi_bytes = blk_rq_bytes(rq->next_rq);
2825 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2828 EXPORT_SYMBOL(blk_end_request_all);
2831 * blk_end_request_cur - Helper function to finish the current request chunk.
2832 * @rq: the request to finish the current chunk for
2833 * @error: %0 for success, < %0 for error
2836 * Complete the current consecutively mapped chunk from @rq.
2839 * %false - we are done with this request
2840 * %true - still buffers pending for this request
2842 bool blk_end_request_cur(struct request *rq, int error)
2844 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2846 EXPORT_SYMBOL(blk_end_request_cur);
2849 * blk_end_request_err - Finish a request till the next failure boundary.
2850 * @rq: the request to finish till the next failure boundary for
2851 * @error: must be negative errno
2854 * Complete @rq till the next failure boundary.
2857 * %false - we are done with this request
2858 * %true - still buffers pending for this request
2860 bool blk_end_request_err(struct request *rq, int error)
2862 WARN_ON(error >= 0);
2863 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2865 EXPORT_SYMBOL_GPL(blk_end_request_err);
2868 * __blk_end_request - Helper function for drivers to complete the request.
2869 * @rq: the request being processed
2870 * @error: %0 for success, < %0 for error
2871 * @nr_bytes: number of bytes to complete
2874 * Must be called with queue lock held unlike blk_end_request().
2877 * %false - we are done with this request
2878 * %true - still buffers pending for this request
2880 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2882 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2884 EXPORT_SYMBOL(__blk_end_request);
2887 * __blk_end_request_all - Helper function for drives to finish the request.
2888 * @rq: the request to finish
2889 * @error: %0 for success, < %0 for error
2892 * Completely finish @rq. Must be called with queue lock held.
2894 void __blk_end_request_all(struct request *rq, int error)
2897 unsigned int bidi_bytes = 0;
2899 if (unlikely(blk_bidi_rq(rq)))
2900 bidi_bytes = blk_rq_bytes(rq->next_rq);
2902 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2905 EXPORT_SYMBOL(__blk_end_request_all);
2908 * __blk_end_request_cur - Helper function to finish the current request chunk.
2909 * @rq: the request to finish the current chunk for
2910 * @error: %0 for success, < %0 for error
2913 * Complete the current consecutively mapped chunk from @rq. Must
2914 * be called with queue lock held.
2917 * %false - we are done with this request
2918 * %true - still buffers pending for this request
2920 bool __blk_end_request_cur(struct request *rq, int error)
2922 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2924 EXPORT_SYMBOL(__blk_end_request_cur);
2927 * __blk_end_request_err - Finish a request till the next failure boundary.
2928 * @rq: the request to finish till the next failure boundary for
2929 * @error: must be negative errno
2932 * Complete @rq till the next failure boundary. Must be called
2933 * with queue lock held.
2936 * %false - we are done with this request
2937 * %true - still buffers pending for this request
2939 bool __blk_end_request_err(struct request *rq, int error)
2941 WARN_ON(error >= 0);
2942 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2944 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2946 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2949 req_set_op(rq, bio_op(bio));
2951 if (bio_has_data(bio))
2952 rq->nr_phys_segments = bio_phys_segments(q, bio);
2954 rq->__data_len = bio->bi_iter.bi_size;
2955 rq->bio = rq->biotail = bio;
2958 rq->rq_disk = bio->bi_bdev->bd_disk;
2961 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2963 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2964 * @rq: the request to be flushed
2967 * Flush all pages in @rq.
2969 void rq_flush_dcache_pages(struct request *rq)
2971 struct req_iterator iter;
2972 struct bio_vec bvec;
2974 rq_for_each_segment(bvec, rq, iter)
2975 flush_dcache_page(bvec.bv_page);
2977 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2981 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2982 * @q : the queue of the device being checked
2985 * Check if underlying low-level drivers of a device are busy.
2986 * If the drivers want to export their busy state, they must set own
2987 * exporting function using blk_queue_lld_busy() first.
2989 * Basically, this function is used only by request stacking drivers
2990 * to stop dispatching requests to underlying devices when underlying
2991 * devices are busy. This behavior helps more I/O merging on the queue
2992 * of the request stacking driver and prevents I/O throughput regression
2993 * on burst I/O load.
2996 * 0 - Not busy (The request stacking driver should dispatch request)
2997 * 1 - Busy (The request stacking driver should stop dispatching request)
2999 int blk_lld_busy(struct request_queue *q)
3002 return q->lld_busy_fn(q);
3006 EXPORT_SYMBOL_GPL(blk_lld_busy);
3009 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3010 * @rq: the clone request to be cleaned up
3013 * Free all bios in @rq for a cloned request.
3015 void blk_rq_unprep_clone(struct request *rq)
3019 while ((bio = rq->bio) != NULL) {
3020 rq->bio = bio->bi_next;
3025 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3028 * Copy attributes of the original request to the clone request.
3029 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3031 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3033 dst->cpu = src->cpu;
3034 req_set_op_attrs(dst, req_op(src),
3035 (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
3036 dst->cmd_type = src->cmd_type;
3037 dst->__sector = blk_rq_pos(src);
3038 dst->__data_len = blk_rq_bytes(src);
3039 dst->nr_phys_segments = src->nr_phys_segments;
3040 dst->ioprio = src->ioprio;
3041 dst->extra_len = src->extra_len;
3045 * blk_rq_prep_clone - Helper function to setup clone request
3046 * @rq: the request to be setup
3047 * @rq_src: original request to be cloned
3048 * @bs: bio_set that bios for clone are allocated from
3049 * @gfp_mask: memory allocation mask for bio
3050 * @bio_ctr: setup function to be called for each clone bio.
3051 * Returns %0 for success, non %0 for failure.
3052 * @data: private data to be passed to @bio_ctr
3055 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3056 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3057 * are not copied, and copying such parts is the caller's responsibility.
3058 * Also, pages which the original bios are pointing to are not copied
3059 * and the cloned bios just point same pages.
3060 * So cloned bios must be completed before original bios, which means
3061 * the caller must complete @rq before @rq_src.
3063 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3064 struct bio_set *bs, gfp_t gfp_mask,
3065 int (*bio_ctr)(struct bio *, struct bio *, void *),
3068 struct bio *bio, *bio_src;
3073 __rq_for_each_bio(bio_src, rq_src) {
3074 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3078 if (bio_ctr && bio_ctr(bio, bio_src, data))
3082 rq->biotail->bi_next = bio;
3085 rq->bio = rq->biotail = bio;
3088 __blk_rq_prep_clone(rq, rq_src);
3095 blk_rq_unprep_clone(rq);
3099 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3101 int kblockd_schedule_work(struct work_struct *work)
3103 return queue_work(kblockd_workqueue, work);
3105 EXPORT_SYMBOL(kblockd_schedule_work);
3107 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3109 return queue_work_on(cpu, kblockd_workqueue, work);
3111 EXPORT_SYMBOL(kblockd_schedule_work_on);
3113 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3114 unsigned long delay)
3116 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3118 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3120 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3121 unsigned long delay)
3123 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3125 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3128 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3129 * @plug: The &struct blk_plug that needs to be initialized
3132 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3133 * pending I/O should the task end up blocking between blk_start_plug() and
3134 * blk_finish_plug(). This is important from a performance perspective, but
3135 * also ensures that we don't deadlock. For instance, if the task is blocking
3136 * for a memory allocation, memory reclaim could end up wanting to free a
3137 * page belonging to that request that is currently residing in our private
3138 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3139 * this kind of deadlock.
3141 void blk_start_plug(struct blk_plug *plug)
3143 struct task_struct *tsk = current;
3146 * If this is a nested plug, don't actually assign it.
3151 INIT_LIST_HEAD(&plug->list);
3152 INIT_LIST_HEAD(&plug->mq_list);
3153 INIT_LIST_HEAD(&plug->cb_list);
3155 * Store ordering should not be needed here, since a potential
3156 * preempt will imply a full memory barrier
3160 EXPORT_SYMBOL(blk_start_plug);
3162 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3164 struct request *rqa = container_of(a, struct request, queuelist);
3165 struct request *rqb = container_of(b, struct request, queuelist);
3167 return !(rqa->q < rqb->q ||
3168 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3172 * If 'from_schedule' is true, then postpone the dispatch of requests
3173 * until a safe kblockd context. We due this to avoid accidental big
3174 * additional stack usage in driver dispatch, in places where the originally
3175 * plugger did not intend it.
3177 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3179 __releases(q->queue_lock)
3181 trace_block_unplug(q, depth, !from_schedule);
3184 blk_run_queue_async(q);
3187 spin_unlock(q->queue_lock);
3190 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3192 LIST_HEAD(callbacks);
3194 while (!list_empty(&plug->cb_list)) {
3195 list_splice_init(&plug->cb_list, &callbacks);
3197 while (!list_empty(&callbacks)) {
3198 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3201 list_del(&cb->list);
3202 cb->callback(cb, from_schedule);
3207 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3210 struct blk_plug *plug = current->plug;
3211 struct blk_plug_cb *cb;
3216 list_for_each_entry(cb, &plug->cb_list, list)
3217 if (cb->callback == unplug && cb->data == data)
3220 /* Not currently on the callback list */
3221 BUG_ON(size < sizeof(*cb));
3222 cb = kzalloc(size, GFP_ATOMIC);
3225 cb->callback = unplug;
3226 list_add(&cb->list, &plug->cb_list);
3230 EXPORT_SYMBOL(blk_check_plugged);
3232 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3234 struct request_queue *q;
3235 unsigned long flags;
3240 flush_plug_callbacks(plug, from_schedule);
3242 if (!list_empty(&plug->mq_list))
3243 blk_mq_flush_plug_list(plug, from_schedule);
3245 if (list_empty(&plug->list))
3248 list_splice_init(&plug->list, &list);
3250 list_sort(NULL, &list, plug_rq_cmp);
3256 * Save and disable interrupts here, to avoid doing it for every
3257 * queue lock we have to take.
3259 local_irq_save(flags);
3260 while (!list_empty(&list)) {
3261 rq = list_entry_rq(list.next);
3262 list_del_init(&rq->queuelist);
3266 * This drops the queue lock
3269 queue_unplugged(q, depth, from_schedule);
3272 spin_lock(q->queue_lock);
3276 * Short-circuit if @q is dead
3278 if (unlikely(blk_queue_dying(q))) {
3279 __blk_end_request_all(rq, -ENODEV);
3284 * rq is already accounted, so use raw insert
3286 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3287 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3289 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3295 * This drops the queue lock
3298 queue_unplugged(q, depth, from_schedule);
3300 local_irq_restore(flags);
3303 void blk_finish_plug(struct blk_plug *plug)
3305 if (plug != current->plug)
3307 blk_flush_plug_list(plug, false);
3309 current->plug = NULL;
3311 EXPORT_SYMBOL(blk_finish_plug);
3313 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3315 struct blk_plug *plug;
3317 unsigned int queue_num;
3318 struct blk_mq_hw_ctx *hctx;
3320 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3321 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3324 queue_num = blk_qc_t_to_queue_num(cookie);
3325 hctx = q->queue_hw_ctx[queue_num];
3326 hctx->poll_considered++;
3328 plug = current->plug;
3330 blk_flush_plug_list(plug, false);
3332 state = current->state;
3333 while (!need_resched()) {
3336 hctx->poll_invoked++;
3338 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3340 hctx->poll_success++;
3341 set_current_state(TASK_RUNNING);
3345 if (signal_pending_state(state, current))
3346 set_current_state(TASK_RUNNING);
3348 if (current->state == TASK_RUNNING)
3357 EXPORT_SYMBOL_GPL(blk_poll);
3361 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3362 * @q: the queue of the device
3363 * @dev: the device the queue belongs to
3366 * Initialize runtime-PM-related fields for @q and start auto suspend for
3367 * @dev. Drivers that want to take advantage of request-based runtime PM
3368 * should call this function after @dev has been initialized, and its
3369 * request queue @q has been allocated, and runtime PM for it can not happen
3370 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3371 * cases, driver should call this function before any I/O has taken place.
3373 * This function takes care of setting up using auto suspend for the device,
3374 * the autosuspend delay is set to -1 to make runtime suspend impossible
3375 * until an updated value is either set by user or by driver. Drivers do
3376 * not need to touch other autosuspend settings.
3378 * The block layer runtime PM is request based, so only works for drivers
3379 * that use request as their IO unit instead of those directly use bio's.
3381 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3384 q->rpm_status = RPM_ACTIVE;
3385 pm_runtime_set_autosuspend_delay(q->dev, -1);
3386 pm_runtime_use_autosuspend(q->dev);
3388 EXPORT_SYMBOL(blk_pm_runtime_init);
3391 * blk_pre_runtime_suspend - Pre runtime suspend check
3392 * @q: the queue of the device
3395 * This function will check if runtime suspend is allowed for the device
3396 * by examining if there are any requests pending in the queue. If there
3397 * are requests pending, the device can not be runtime suspended; otherwise,
3398 * the queue's status will be updated to SUSPENDING and the driver can
3399 * proceed to suspend the device.
3401 * For the not allowed case, we mark last busy for the device so that
3402 * runtime PM core will try to autosuspend it some time later.
3404 * This function should be called near the start of the device's
3405 * runtime_suspend callback.
3408 * 0 - OK to runtime suspend the device
3409 * -EBUSY - Device should not be runtime suspended
3411 int blk_pre_runtime_suspend(struct request_queue *q)
3418 spin_lock_irq(q->queue_lock);
3419 if (q->nr_pending) {
3421 pm_runtime_mark_last_busy(q->dev);
3423 q->rpm_status = RPM_SUSPENDING;
3425 spin_unlock_irq(q->queue_lock);
3428 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3431 * blk_post_runtime_suspend - Post runtime suspend processing
3432 * @q: the queue of the device
3433 * @err: return value of the device's runtime_suspend function
3436 * Update the queue's runtime status according to the return value of the
3437 * device's runtime suspend function and mark last busy for the device so
3438 * that PM core will try to auto suspend the device at a later time.
3440 * This function should be called near the end of the device's
3441 * runtime_suspend callback.
3443 void blk_post_runtime_suspend(struct request_queue *q, int err)
3448 spin_lock_irq(q->queue_lock);
3450 q->rpm_status = RPM_SUSPENDED;
3452 q->rpm_status = RPM_ACTIVE;
3453 pm_runtime_mark_last_busy(q->dev);
3455 spin_unlock_irq(q->queue_lock);
3457 EXPORT_SYMBOL(blk_post_runtime_suspend);
3460 * blk_pre_runtime_resume - Pre runtime resume processing
3461 * @q: the queue of the device
3464 * Update the queue's runtime status to RESUMING in preparation for the
3465 * runtime resume of the device.
3467 * This function should be called near the start of the device's
3468 * runtime_resume callback.
3470 void blk_pre_runtime_resume(struct request_queue *q)
3475 spin_lock_irq(q->queue_lock);
3476 q->rpm_status = RPM_RESUMING;
3477 spin_unlock_irq(q->queue_lock);
3479 EXPORT_SYMBOL(blk_pre_runtime_resume);
3482 * blk_post_runtime_resume - Post runtime resume processing
3483 * @q: the queue of the device
3484 * @err: return value of the device's runtime_resume function
3487 * Update the queue's runtime status according to the return value of the
3488 * device's runtime_resume function. If it is successfully resumed, process
3489 * the requests that are queued into the device's queue when it is resuming
3490 * and then mark last busy and initiate autosuspend for it.
3492 * This function should be called near the end of the device's
3493 * runtime_resume callback.
3495 void blk_post_runtime_resume(struct request_queue *q, int err)
3500 spin_lock_irq(q->queue_lock);
3502 q->rpm_status = RPM_ACTIVE;
3504 pm_runtime_mark_last_busy(q->dev);
3505 pm_request_autosuspend(q->dev);
3507 q->rpm_status = RPM_SUSPENDED;
3509 spin_unlock_irq(q->queue_lock);
3511 EXPORT_SYMBOL(blk_post_runtime_resume);
3514 * blk_set_runtime_active - Force runtime status of the queue to be active
3515 * @q: the queue of the device
3517 * If the device is left runtime suspended during system suspend the resume
3518 * hook typically resumes the device and corrects runtime status
3519 * accordingly. However, that does not affect the queue runtime PM status
3520 * which is still "suspended". This prevents processing requests from the
3523 * This function can be used in driver's resume hook to correct queue
3524 * runtime PM status and re-enable peeking requests from the queue. It
3525 * should be called before first request is added to the queue.
3527 void blk_set_runtime_active(struct request_queue *q)
3529 spin_lock_irq(q->queue_lock);
3530 q->rpm_status = RPM_ACTIVE;
3531 pm_runtime_mark_last_busy(q->dev);
3532 pm_request_autosuspend(q->dev);
3533 spin_unlock_irq(q->queue_lock);
3535 EXPORT_SYMBOL(blk_set_runtime_active);
3538 int __init blk_dev_init(void)
3540 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3541 FIELD_SIZEOF(struct request, cmd_flags));
3543 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3544 kblockd_workqueue = alloc_workqueue("kblockd",
3545 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3546 if (!kblockd_workqueue)
3547 panic("Failed to create kblockd\n");
3549 request_cachep = kmem_cache_create("blkdev_requests",
3550 sizeof(struct request), 0, SLAB_PANIC, NULL);
3552 blk_requestq_cachep = kmem_cache_create("request_queue",
3553 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);