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>
42 #include "blk-mq-sched.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
;
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
->internal_tag
= -1;
137 rq
->start_time
= jiffies
;
138 set_start_time_ns(rq
);
141 EXPORT_SYMBOL(blk_rq_init
);
143 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
144 unsigned int nbytes
, int error
)
147 bio
->bi_error
= error
;
149 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
150 bio_set_flag(bio
, BIO_QUIET
);
152 bio_advance(bio
, nbytes
);
154 /* don't actually finish bio if it's part of flush sequence */
155 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
159 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
161 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
162 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
163 (unsigned long long) rq
->cmd_flags
);
165 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
166 (unsigned long long)blk_rq_pos(rq
),
167 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
168 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
169 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
171 EXPORT_SYMBOL(blk_dump_rq_flags
);
173 static void blk_delay_work(struct work_struct
*work
)
175 struct request_queue
*q
;
177 q
= container_of(work
, struct request_queue
, delay_work
.work
);
178 spin_lock_irq(q
->queue_lock
);
180 spin_unlock_irq(q
->queue_lock
);
184 * blk_delay_queue - restart queueing after defined interval
185 * @q: The &struct request_queue in question
186 * @msecs: Delay in msecs
189 * Sometimes queueing needs to be postponed for a little while, to allow
190 * resources to come back. This function will make sure that queueing is
191 * restarted around the specified time. Queue lock must be held.
193 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
195 if (likely(!blk_queue_dead(q
)))
196 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
197 msecs_to_jiffies(msecs
));
199 EXPORT_SYMBOL(blk_delay_queue
);
202 * blk_start_queue_async - asynchronously restart a previously stopped queue
203 * @q: The &struct request_queue in question
206 * blk_start_queue_async() will clear the stop flag on the queue, and
207 * ensure that the request_fn for the queue is run from an async
210 void blk_start_queue_async(struct request_queue
*q
)
212 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
213 blk_run_queue_async(q
);
215 EXPORT_SYMBOL(blk_start_queue_async
);
218 * blk_start_queue - restart a previously stopped queue
219 * @q: The &struct request_queue in question
222 * blk_start_queue() will clear the stop flag on the queue, and call
223 * the request_fn for the queue if it was in a stopped state when
224 * entered. Also see blk_stop_queue(). Queue lock must be held.
226 void blk_start_queue(struct request_queue
*q
)
228 WARN_ON(!irqs_disabled());
230 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
233 EXPORT_SYMBOL(blk_start_queue
);
236 * blk_stop_queue - stop a queue
237 * @q: The &struct request_queue in question
240 * The Linux block layer assumes that a block driver will consume all
241 * entries on the request queue when the request_fn strategy is called.
242 * Often this will not happen, because of hardware limitations (queue
243 * depth settings). If a device driver gets a 'queue full' response,
244 * or if it simply chooses not to queue more I/O at one point, it can
245 * call this function to prevent the request_fn from being called until
246 * the driver has signalled it's ready to go again. This happens by calling
247 * blk_start_queue() to restart queue operations. Queue lock must be held.
249 void blk_stop_queue(struct request_queue
*q
)
251 cancel_delayed_work(&q
->delay_work
);
252 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
254 EXPORT_SYMBOL(blk_stop_queue
);
257 * blk_sync_queue - cancel any pending callbacks on a queue
261 * The block layer may perform asynchronous callback activity
262 * on a queue, such as calling the unplug function after a timeout.
263 * A block device may call blk_sync_queue to ensure that any
264 * such activity is cancelled, thus allowing it to release resources
265 * that the callbacks might use. The caller must already have made sure
266 * that its ->make_request_fn will not re-add plugging prior to calling
269 * This function does not cancel any asynchronous activity arising
270 * out of elevator or throttling code. That would require elevator_exit()
271 * and blkcg_exit_queue() to be called with queue lock initialized.
274 void blk_sync_queue(struct request_queue
*q
)
276 del_timer_sync(&q
->timeout
);
279 struct blk_mq_hw_ctx
*hctx
;
282 queue_for_each_hw_ctx(q
, hctx
, i
) {
283 cancel_work_sync(&hctx
->run_work
);
284 cancel_delayed_work_sync(&hctx
->delay_work
);
287 cancel_delayed_work_sync(&q
->delay_work
);
290 EXPORT_SYMBOL(blk_sync_queue
);
293 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
294 * @q: The queue to run
297 * Invoke request handling on a queue if there are any pending requests.
298 * May be used to restart request handling after a request has completed.
299 * This variant runs the queue whether or not the queue has been
300 * stopped. Must be called with the queue lock held and interrupts
301 * disabled. See also @blk_run_queue.
303 inline void __blk_run_queue_uncond(struct request_queue
*q
)
305 if (unlikely(blk_queue_dead(q
)))
309 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
310 * the queue lock internally. As a result multiple threads may be
311 * running such a request function concurrently. Keep track of the
312 * number of active request_fn invocations such that blk_drain_queue()
313 * can wait until all these request_fn calls have finished.
315 q
->request_fn_active
++;
317 q
->request_fn_active
--;
319 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
322 * __blk_run_queue - run a single device queue
323 * @q: The queue to run
326 * See @blk_run_queue. This variant must be called with the queue lock
327 * held and interrupts disabled.
329 void __blk_run_queue(struct request_queue
*q
)
331 if (unlikely(blk_queue_stopped(q
)))
334 __blk_run_queue_uncond(q
);
336 EXPORT_SYMBOL(__blk_run_queue
);
339 * blk_run_queue_async - run a single device queue in workqueue context
340 * @q: The queue to run
343 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
344 * of us. The caller must hold the queue lock.
346 void blk_run_queue_async(struct request_queue
*q
)
348 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
349 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
351 EXPORT_SYMBOL(blk_run_queue_async
);
354 * blk_run_queue - run a single device queue
355 * @q: The queue to run
358 * Invoke request handling on this queue, if it has pending work to do.
359 * May be used to restart queueing when a request has completed.
361 void blk_run_queue(struct request_queue
*q
)
365 spin_lock_irqsave(q
->queue_lock
, flags
);
367 spin_unlock_irqrestore(q
->queue_lock
, flags
);
369 EXPORT_SYMBOL(blk_run_queue
);
371 void blk_put_queue(struct request_queue
*q
)
373 kobject_put(&q
->kobj
);
375 EXPORT_SYMBOL(blk_put_queue
);
378 * __blk_drain_queue - drain requests from request_queue
380 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
382 * Drain requests from @q. If @drain_all is set, all requests are drained.
383 * If not, only ELVPRIV requests are drained. The caller is responsible
384 * for ensuring that no new requests which need to be drained are queued.
386 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
387 __releases(q
->queue_lock
)
388 __acquires(q
->queue_lock
)
392 lockdep_assert_held(q
->queue_lock
);
398 * The caller might be trying to drain @q before its
399 * elevator is initialized.
402 elv_drain_elevator(q
);
404 blkcg_drain_queue(q
);
407 * This function might be called on a queue which failed
408 * driver init after queue creation or is not yet fully
409 * active yet. Some drivers (e.g. fd and loop) get unhappy
410 * in such cases. Kick queue iff dispatch queue has
411 * something on it and @q has request_fn set.
413 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
416 drain
|= q
->nr_rqs_elvpriv
;
417 drain
|= q
->request_fn_active
;
420 * Unfortunately, requests are queued at and tracked from
421 * multiple places and there's no single counter which can
422 * be drained. Check all the queues and counters.
425 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
426 drain
|= !list_empty(&q
->queue_head
);
427 for (i
= 0; i
< 2; i
++) {
428 drain
|= q
->nr_rqs
[i
];
429 drain
|= q
->in_flight
[i
];
431 drain
|= !list_empty(&fq
->flush_queue
[i
]);
438 spin_unlock_irq(q
->queue_lock
);
442 spin_lock_irq(q
->queue_lock
);
446 * With queue marked dead, any woken up waiter will fail the
447 * allocation path, so the wakeup chaining is lost and we're
448 * left with hung waiters. We need to wake up those waiters.
451 struct request_list
*rl
;
453 blk_queue_for_each_rl(rl
, q
)
454 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
455 wake_up_all(&rl
->wait
[i
]);
460 * blk_queue_bypass_start - enter queue bypass mode
461 * @q: queue of interest
463 * In bypass mode, only the dispatch FIFO queue of @q is used. This
464 * function makes @q enter bypass mode and drains all requests which were
465 * throttled or issued before. On return, it's guaranteed that no request
466 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
467 * inside queue or RCU read lock.
469 void blk_queue_bypass_start(struct request_queue
*q
)
471 spin_lock_irq(q
->queue_lock
);
473 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
474 spin_unlock_irq(q
->queue_lock
);
477 * Queues start drained. Skip actual draining till init is
478 * complete. This avoids lenghty delays during queue init which
479 * can happen many times during boot.
481 if (blk_queue_init_done(q
)) {
482 spin_lock_irq(q
->queue_lock
);
483 __blk_drain_queue(q
, false);
484 spin_unlock_irq(q
->queue_lock
);
486 /* ensure blk_queue_bypass() is %true inside RCU read lock */
490 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
493 * blk_queue_bypass_end - leave queue bypass mode
494 * @q: queue of interest
496 * Leave bypass mode and restore the normal queueing behavior.
498 void blk_queue_bypass_end(struct request_queue
*q
)
500 spin_lock_irq(q
->queue_lock
);
501 if (!--q
->bypass_depth
)
502 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
503 WARN_ON_ONCE(q
->bypass_depth
< 0);
504 spin_unlock_irq(q
->queue_lock
);
506 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
508 void blk_set_queue_dying(struct request_queue
*q
)
510 spin_lock_irq(q
->queue_lock
);
511 queue_flag_set(QUEUE_FLAG_DYING
, q
);
512 spin_unlock_irq(q
->queue_lock
);
515 blk_mq_wake_waiters(q
);
517 struct request_list
*rl
;
519 blk_queue_for_each_rl(rl
, q
) {
521 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
522 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
527 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
530 * blk_cleanup_queue - shutdown a request queue
531 * @q: request queue to shutdown
533 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
534 * put it. All future requests will be failed immediately with -ENODEV.
536 void blk_cleanup_queue(struct request_queue
*q
)
538 spinlock_t
*lock
= q
->queue_lock
;
540 /* mark @q DYING, no new request or merges will be allowed afterwards */
541 mutex_lock(&q
->sysfs_lock
);
542 blk_set_queue_dying(q
);
546 * A dying queue is permanently in bypass mode till released. Note
547 * that, unlike blk_queue_bypass_start(), we aren't performing
548 * synchronize_rcu() after entering bypass mode to avoid the delay
549 * as some drivers create and destroy a lot of queues while
550 * probing. This is still safe because blk_release_queue() will be
551 * called only after the queue refcnt drops to zero and nothing,
552 * RCU or not, would be traversing the queue by then.
555 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
557 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
558 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
559 queue_flag_set(QUEUE_FLAG_DYING
, q
);
560 spin_unlock_irq(lock
);
561 mutex_unlock(&q
->sysfs_lock
);
564 * Drain all requests queued before DYING marking. Set DEAD flag to
565 * prevent that q->request_fn() gets invoked after draining finished.
570 __blk_drain_queue(q
, true);
571 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
572 spin_unlock_irq(lock
);
574 /* for synchronous bio-based driver finish in-flight integrity i/o */
575 blk_flush_integrity();
577 /* @q won't process any more request, flush async actions */
578 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
582 blk_mq_free_queue(q
);
583 percpu_ref_exit(&q
->q_usage_counter
);
586 if (q
->queue_lock
!= &q
->__queue_lock
)
587 q
->queue_lock
= &q
->__queue_lock
;
588 spin_unlock_irq(lock
);
590 bdi_unregister(q
->backing_dev_info
);
592 /* @q is and will stay empty, shutdown and put */
595 EXPORT_SYMBOL(blk_cleanup_queue
);
597 /* Allocate memory local to the request queue */
598 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
600 struct request_queue
*q
= data
;
602 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
605 static void free_request_simple(void *element
, void *data
)
607 kmem_cache_free(request_cachep
, element
);
610 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
612 struct request_queue
*q
= data
;
615 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
617 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
624 static void free_request_size(void *element
, void *data
)
626 struct request_queue
*q
= data
;
629 q
->exit_rq_fn(q
, element
);
633 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
636 if (unlikely(rl
->rq_pool
))
640 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
641 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
642 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
643 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
646 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
647 alloc_request_size
, free_request_size
,
648 q
, gfp_mask
, q
->node
);
650 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
651 alloc_request_simple
, free_request_simple
,
652 q
, gfp_mask
, q
->node
);
660 void blk_exit_rl(struct request_list
*rl
)
663 mempool_destroy(rl
->rq_pool
);
666 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
668 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
670 EXPORT_SYMBOL(blk_alloc_queue
);
672 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
677 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
683 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
684 !atomic_read(&q
->mq_freeze_depth
) ||
686 if (blk_queue_dying(q
))
693 void blk_queue_exit(struct request_queue
*q
)
695 percpu_ref_put(&q
->q_usage_counter
);
698 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
700 struct request_queue
*q
=
701 container_of(ref
, struct request_queue
, q_usage_counter
);
703 wake_up_all(&q
->mq_freeze_wq
);
706 static void blk_rq_timed_out_timer(unsigned long data
)
708 struct request_queue
*q
= (struct request_queue
*)data
;
710 kblockd_schedule_work(&q
->timeout_work
);
713 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
715 struct request_queue
*q
;
717 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
718 gfp_mask
| __GFP_ZERO
, node_id
);
722 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
726 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
730 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
731 if (!q
->backing_dev_info
)
734 q
->backing_dev_info
->ra_pages
=
735 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
736 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
737 q
->backing_dev_info
->name
= "block";
740 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
741 laptop_mode_timer_fn
, (unsigned long) q
);
742 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
743 INIT_LIST_HEAD(&q
->queue_head
);
744 INIT_LIST_HEAD(&q
->timeout_list
);
745 INIT_LIST_HEAD(&q
->icq_list
);
746 #ifdef CONFIG_BLK_CGROUP
747 INIT_LIST_HEAD(&q
->blkg_list
);
749 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
751 kobject_init(&q
->kobj
, &blk_queue_ktype
);
753 mutex_init(&q
->sysfs_lock
);
754 spin_lock_init(&q
->__queue_lock
);
757 * By default initialize queue_lock to internal lock and driver can
758 * override it later if need be.
760 q
->queue_lock
= &q
->__queue_lock
;
763 * A queue starts its life with bypass turned on to avoid
764 * unnecessary bypass on/off overhead and nasty surprises during
765 * init. The initial bypass will be finished when the queue is
766 * registered by blk_register_queue().
769 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
771 init_waitqueue_head(&q
->mq_freeze_wq
);
774 * Init percpu_ref in atomic mode so that it's faster to shutdown.
775 * See blk_register_queue() for details.
777 if (percpu_ref_init(&q
->q_usage_counter
,
778 blk_queue_usage_counter_release
,
779 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
782 if (blkcg_init_queue(q
))
788 percpu_ref_exit(&q
->q_usage_counter
);
790 bdi_put(q
->backing_dev_info
);
792 bioset_free(q
->bio_split
);
794 ida_simple_remove(&blk_queue_ida
, q
->id
);
796 kmem_cache_free(blk_requestq_cachep
, q
);
799 EXPORT_SYMBOL(blk_alloc_queue_node
);
802 * blk_init_queue - prepare a request queue for use with a block device
803 * @rfn: The function to be called to process requests that have been
804 * placed on the queue.
805 * @lock: Request queue spin lock
808 * If a block device wishes to use the standard request handling procedures,
809 * which sorts requests and coalesces adjacent requests, then it must
810 * call blk_init_queue(). The function @rfn will be called when there
811 * are requests on the queue that need to be processed. If the device
812 * supports plugging, then @rfn may not be called immediately when requests
813 * are available on the queue, but may be called at some time later instead.
814 * Plugged queues are generally unplugged when a buffer belonging to one
815 * of the requests on the queue is needed, or due to memory pressure.
817 * @rfn is not required, or even expected, to remove all requests off the
818 * queue, but only as many as it can handle at a time. If it does leave
819 * requests on the queue, it is responsible for arranging that the requests
820 * get dealt with eventually.
822 * The queue spin lock must be held while manipulating the requests on the
823 * request queue; this lock will be taken also from interrupt context, so irq
824 * disabling is needed for it.
826 * Function returns a pointer to the initialized request queue, or %NULL if
830 * blk_init_queue() must be paired with a blk_cleanup_queue() call
831 * when the block device is deactivated (such as at module unload).
834 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
836 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
838 EXPORT_SYMBOL(blk_init_queue
);
840 struct request_queue
*
841 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
843 struct request_queue
*q
;
845 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
851 q
->queue_lock
= lock
;
852 if (blk_init_allocated_queue(q
) < 0) {
853 blk_cleanup_queue(q
);
859 EXPORT_SYMBOL(blk_init_queue_node
);
861 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
864 int blk_init_allocated_queue(struct request_queue
*q
)
866 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
870 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
871 goto out_free_flush_queue
;
873 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
874 goto out_exit_flush_rq
;
876 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
877 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
880 * This also sets hw/phys segments, boundary and size
882 blk_queue_make_request(q
, blk_queue_bio
);
884 q
->sg_reserved_size
= INT_MAX
;
886 /* Protect q->elevator from elevator_change */
887 mutex_lock(&q
->sysfs_lock
);
890 if (elevator_init(q
, NULL
)) {
891 mutex_unlock(&q
->sysfs_lock
);
892 goto out_exit_flush_rq
;
895 mutex_unlock(&q
->sysfs_lock
);
900 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
901 out_free_flush_queue
:
902 blk_free_flush_queue(q
->fq
);
906 EXPORT_SYMBOL(blk_init_allocated_queue
);
908 bool blk_get_queue(struct request_queue
*q
)
910 if (likely(!blk_queue_dying(q
))) {
917 EXPORT_SYMBOL(blk_get_queue
);
919 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
921 if (rq
->rq_flags
& RQF_ELVPRIV
) {
922 elv_put_request(rl
->q
, rq
);
924 put_io_context(rq
->elv
.icq
->ioc
);
927 mempool_free(rq
, rl
->rq_pool
);
931 * ioc_batching returns true if the ioc is a valid batching request and
932 * should be given priority access to a request.
934 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
940 * Make sure the process is able to allocate at least 1 request
941 * even if the batch times out, otherwise we could theoretically
944 return ioc
->nr_batch_requests
== q
->nr_batching
||
945 (ioc
->nr_batch_requests
> 0
946 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
950 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
951 * will cause the process to be a "batcher" on all queues in the system. This
952 * is the behaviour we want though - once it gets a wakeup it should be given
955 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
957 if (!ioc
|| ioc_batching(q
, ioc
))
960 ioc
->nr_batch_requests
= q
->nr_batching
;
961 ioc
->last_waited
= jiffies
;
964 static void __freed_request(struct request_list
*rl
, int sync
)
966 struct request_queue
*q
= rl
->q
;
968 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
969 blk_clear_congested(rl
, sync
);
971 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
972 if (waitqueue_active(&rl
->wait
[sync
]))
973 wake_up(&rl
->wait
[sync
]);
975 blk_clear_rl_full(rl
, sync
);
980 * A request has just been released. Account for it, update the full and
981 * congestion status, wake up any waiters. Called under q->queue_lock.
983 static void freed_request(struct request_list
*rl
, bool sync
,
984 req_flags_t rq_flags
)
986 struct request_queue
*q
= rl
->q
;
990 if (rq_flags
& RQF_ELVPRIV
)
993 __freed_request(rl
, sync
);
995 if (unlikely(rl
->starved
[sync
^ 1]))
996 __freed_request(rl
, sync
^ 1);
999 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1001 struct request_list
*rl
;
1002 int on_thresh
, off_thresh
;
1004 spin_lock_irq(q
->queue_lock
);
1005 q
->nr_requests
= nr
;
1006 blk_queue_congestion_threshold(q
);
1007 on_thresh
= queue_congestion_on_threshold(q
);
1008 off_thresh
= queue_congestion_off_threshold(q
);
1010 blk_queue_for_each_rl(rl
, q
) {
1011 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1012 blk_set_congested(rl
, BLK_RW_SYNC
);
1013 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1014 blk_clear_congested(rl
, BLK_RW_SYNC
);
1016 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1017 blk_set_congested(rl
, BLK_RW_ASYNC
);
1018 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1019 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1021 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1022 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1024 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1025 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1028 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1029 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1031 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1032 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1036 spin_unlock_irq(q
->queue_lock
);
1041 * __get_request - get a free request
1042 * @rl: request list to allocate from
1043 * @op: operation and flags
1044 * @bio: bio to allocate request for (can be %NULL)
1045 * @gfp_mask: allocation mask
1047 * Get a free request from @q. This function may fail under memory
1048 * pressure or if @q is dead.
1050 * Must be called with @q->queue_lock held and,
1051 * Returns ERR_PTR on failure, with @q->queue_lock held.
1052 * Returns request pointer on success, with @q->queue_lock *not held*.
1054 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1055 struct bio
*bio
, gfp_t gfp_mask
)
1057 struct request_queue
*q
= rl
->q
;
1059 struct elevator_type
*et
= q
->elevator
->type
;
1060 struct io_context
*ioc
= rq_ioc(bio
);
1061 struct io_cq
*icq
= NULL
;
1062 const bool is_sync
= op_is_sync(op
);
1064 req_flags_t rq_flags
= RQF_ALLOCED
;
1066 if (unlikely(blk_queue_dying(q
)))
1067 return ERR_PTR(-ENODEV
);
1069 may_queue
= elv_may_queue(q
, op
);
1070 if (may_queue
== ELV_MQUEUE_NO
)
1073 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1074 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1076 * The queue will fill after this allocation, so set
1077 * it as full, and mark this process as "batching".
1078 * This process will be allowed to complete a batch of
1079 * requests, others will be blocked.
1081 if (!blk_rl_full(rl
, is_sync
)) {
1082 ioc_set_batching(q
, ioc
);
1083 blk_set_rl_full(rl
, is_sync
);
1085 if (may_queue
!= ELV_MQUEUE_MUST
1086 && !ioc_batching(q
, ioc
)) {
1088 * The queue is full and the allocating
1089 * process is not a "batcher", and not
1090 * exempted by the IO scheduler
1092 return ERR_PTR(-ENOMEM
);
1096 blk_set_congested(rl
, is_sync
);
1100 * Only allow batching queuers to allocate up to 50% over the defined
1101 * limit of requests, otherwise we could have thousands of requests
1102 * allocated with any setting of ->nr_requests
1104 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1105 return ERR_PTR(-ENOMEM
);
1107 q
->nr_rqs
[is_sync
]++;
1108 rl
->count
[is_sync
]++;
1109 rl
->starved
[is_sync
] = 0;
1112 * Decide whether the new request will be managed by elevator. If
1113 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1114 * prevent the current elevator from being destroyed until the new
1115 * request is freed. This guarantees icq's won't be destroyed and
1116 * makes creating new ones safe.
1118 * Flush requests do not use the elevator so skip initialization.
1119 * This allows a request to share the flush and elevator data.
1121 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1122 * it will be created after releasing queue_lock.
1124 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1125 rq_flags
|= RQF_ELVPRIV
;
1126 q
->nr_rqs_elvpriv
++;
1127 if (et
->icq_cache
&& ioc
)
1128 icq
= ioc_lookup_icq(ioc
, q
);
1131 if (blk_queue_io_stat(q
))
1132 rq_flags
|= RQF_IO_STAT
;
1133 spin_unlock_irq(q
->queue_lock
);
1135 /* allocate and init request */
1136 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1141 blk_rq_set_rl(rq
, rl
);
1142 blk_rq_set_prio(rq
, ioc
);
1144 rq
->rq_flags
= rq_flags
;
1147 if (rq_flags
& RQF_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
, op
);
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
->rq_flags
&= ~RQF_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
, is_sync
, rq_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 * @op: operation and 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
, unsigned int op
,
1233 struct bio
*bio
, gfp_t gfp_mask
)
1235 const bool is_sync
= op_is_sync(op
);
1237 struct request_list
*rl
;
1240 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1242 rq
= __get_request(rl
, op
, 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
, op
);
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 /* create ioc upfront */
1279 create_io_context(gfp_mask
, q
->node
);
1281 spin_lock_irq(q
->queue_lock
);
1282 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1284 spin_unlock_irq(q
->queue_lock
);
1288 /* q->queue_lock is unlocked at this point */
1290 rq
->__sector
= (sector_t
) -1;
1291 rq
->bio
= rq
->biotail
= NULL
;
1295 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1298 return blk_mq_alloc_request(q
, rw
,
1299 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1300 0 : BLK_MQ_REQ_NOWAIT
);
1302 return blk_old_get_request(q
, rw
, gfp_mask
);
1304 EXPORT_SYMBOL(blk_get_request
);
1307 * blk_requeue_request - put a request back on queue
1308 * @q: request queue where request should be inserted
1309 * @rq: request to be inserted
1312 * Drivers often keep queueing requests until the hardware cannot accept
1313 * more, when that condition happens we need to put the request back
1314 * on the queue. Must be called with queue lock held.
1316 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1318 blk_delete_timer(rq
);
1319 blk_clear_rq_complete(rq
);
1320 trace_block_rq_requeue(q
, rq
);
1321 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1323 if (rq
->rq_flags
& RQF_QUEUED
)
1324 blk_queue_end_tag(q
, rq
);
1326 BUG_ON(blk_queued_rq(rq
));
1328 elv_requeue_request(q
, rq
);
1330 EXPORT_SYMBOL(blk_requeue_request
);
1332 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1335 blk_account_io_start(rq
, true);
1336 __elv_add_request(q
, rq
, where
);
1339 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1344 if (now
== part
->stamp
)
1347 inflight
= part_in_flight(part
);
1349 __part_stat_add(cpu
, part
, time_in_queue
,
1350 inflight
* (now
- part
->stamp
));
1351 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1357 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1358 * @cpu: cpu number for stats access
1359 * @part: target partition
1361 * The average IO queue length and utilisation statistics are maintained
1362 * by observing the current state of the queue length and the amount of
1363 * time it has been in this state for.
1365 * Normally, that accounting is done on IO completion, but that can result
1366 * in more than a second's worth of IO being accounted for within any one
1367 * second, leading to >100% utilisation. To deal with that, we call this
1368 * function to do a round-off before returning the results when reading
1369 * /proc/diskstats. This accounts immediately for all queue usage up to
1370 * the current jiffies and restarts the counters again.
1372 void part_round_stats(int cpu
, struct hd_struct
*part
)
1374 unsigned long now
= jiffies
;
1377 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1378 part_round_stats_single(cpu
, part
, now
);
1380 EXPORT_SYMBOL_GPL(part_round_stats
);
1383 static void blk_pm_put_request(struct request
*rq
)
1385 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1386 pm_runtime_mark_last_busy(rq
->q
->dev
);
1389 static inline void blk_pm_put_request(struct request
*rq
) {}
1393 * queue lock must be held
1395 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1397 req_flags_t rq_flags
= req
->rq_flags
;
1403 blk_mq_free_request(req
);
1407 blk_pm_put_request(req
);
1409 elv_completed_request(q
, req
);
1411 /* this is a bio leak */
1412 WARN_ON(req
->bio
!= NULL
);
1414 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1417 * Request may not have originated from ll_rw_blk. if not,
1418 * it didn't come out of our reserved rq pools
1420 if (rq_flags
& RQF_ALLOCED
) {
1421 struct request_list
*rl
= blk_rq_rl(req
);
1422 bool sync
= op_is_sync(req
->cmd_flags
);
1424 BUG_ON(!list_empty(&req
->queuelist
));
1425 BUG_ON(ELV_ON_HASH(req
));
1427 blk_free_request(rl
, req
);
1428 freed_request(rl
, sync
, rq_flags
);
1432 EXPORT_SYMBOL_GPL(__blk_put_request
);
1434 void blk_put_request(struct request
*req
)
1436 struct request_queue
*q
= req
->q
;
1439 blk_mq_free_request(req
);
1441 unsigned long flags
;
1443 spin_lock_irqsave(q
->queue_lock
, flags
);
1444 __blk_put_request(q
, req
);
1445 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1448 EXPORT_SYMBOL(blk_put_request
);
1450 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1453 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1455 if (!ll_back_merge_fn(q
, req
, bio
))
1458 trace_block_bio_backmerge(q
, req
, bio
);
1460 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1461 blk_rq_set_mixed_merge(req
);
1463 req
->biotail
->bi_next
= bio
;
1465 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1466 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1468 blk_account_io_start(req
, false);
1472 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1475 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1477 if (!ll_front_merge_fn(q
, req
, bio
))
1480 trace_block_bio_frontmerge(q
, req
, bio
);
1482 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1483 blk_rq_set_mixed_merge(req
);
1485 bio
->bi_next
= req
->bio
;
1488 req
->__sector
= bio
->bi_iter
.bi_sector
;
1489 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1490 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1492 blk_account_io_start(req
, false);
1497 * blk_attempt_plug_merge - try to merge with %current's plugged list
1498 * @q: request_queue new bio is being queued at
1499 * @bio: new bio being queued
1500 * @request_count: out parameter for number of traversed plugged requests
1501 * @same_queue_rq: pointer to &struct request that gets filled in when
1502 * another request associated with @q is found on the plug list
1503 * (optional, may be %NULL)
1505 * Determine whether @bio being queued on @q can be merged with a request
1506 * on %current's plugged list. Returns %true if merge was successful,
1509 * Plugging coalesces IOs from the same issuer for the same purpose without
1510 * going through @q->queue_lock. As such it's more of an issuing mechanism
1511 * than scheduling, and the request, while may have elvpriv data, is not
1512 * added on the elevator at this point. In addition, we don't have
1513 * reliable access to the elevator outside queue lock. Only check basic
1514 * merging parameters without querying the elevator.
1516 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1518 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1519 unsigned int *request_count
,
1520 struct request
**same_queue_rq
)
1522 struct blk_plug
*plug
;
1525 struct list_head
*plug_list
;
1527 plug
= current
->plug
;
1533 plug_list
= &plug
->mq_list
;
1535 plug_list
= &plug
->list
;
1537 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1543 * Only blk-mq multiple hardware queues case checks the
1544 * rq in the same queue, there should be only one such
1548 *same_queue_rq
= rq
;
1551 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1554 el_ret
= blk_try_merge(rq
, bio
);
1555 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1556 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1559 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1560 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1569 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1571 struct blk_plug
*plug
;
1573 struct list_head
*plug_list
;
1574 unsigned int ret
= 0;
1576 plug
= current
->plug
;
1581 plug_list
= &plug
->mq_list
;
1583 plug_list
= &plug
->list
;
1585 list_for_each_entry(rq
, plug_list
, queuelist
) {
1593 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1595 if (bio
->bi_opf
& REQ_RAHEAD
)
1596 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1599 req
->__sector
= bio
->bi_iter
.bi_sector
;
1600 if (ioprio_valid(bio_prio(bio
)))
1601 req
->ioprio
= bio_prio(bio
);
1602 blk_rq_bio_prep(req
->q
, req
, bio
);
1605 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1607 struct blk_plug
*plug
;
1608 int el_ret
, where
= ELEVATOR_INSERT_SORT
;
1609 struct request
*req
;
1610 unsigned int request_count
= 0;
1611 unsigned int wb_acct
;
1614 * low level driver can indicate that it wants pages above a
1615 * certain limit bounced to low memory (ie for highmem, or even
1616 * ISA dma in theory)
1618 blk_queue_bounce(q
, &bio
);
1620 blk_queue_split(q
, &bio
, q
->bio_split
);
1622 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1623 bio
->bi_error
= -EIO
;
1625 return BLK_QC_T_NONE
;
1628 if (op_is_flush(bio
->bi_opf
)) {
1629 spin_lock_irq(q
->queue_lock
);
1630 where
= ELEVATOR_INSERT_FLUSH
;
1635 * Check if we can merge with the plugged list before grabbing
1638 if (!blk_queue_nomerges(q
)) {
1639 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1640 return BLK_QC_T_NONE
;
1642 request_count
= blk_plug_queued_count(q
);
1644 spin_lock_irq(q
->queue_lock
);
1646 el_ret
= elv_merge(q
, &req
, bio
);
1647 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1648 if (bio_attempt_back_merge(q
, req
, bio
)) {
1649 elv_bio_merged(q
, req
, bio
);
1650 if (!attempt_back_merge(q
, req
))
1651 elv_merged_request(q
, req
, el_ret
);
1654 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1655 if (bio_attempt_front_merge(q
, req
, bio
)) {
1656 elv_bio_merged(q
, req
, bio
);
1657 if (!attempt_front_merge(q
, req
))
1658 elv_merged_request(q
, req
, el_ret
);
1664 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1667 * Grab a free request. This is might sleep but can not fail.
1668 * Returns with the queue unlocked.
1670 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1672 __wbt_done(q
->rq_wb
, wb_acct
);
1673 bio
->bi_error
= PTR_ERR(req
);
1678 wbt_track(&req
->issue_stat
, wb_acct
);
1681 * After dropping the lock and possibly sleeping here, our request
1682 * may now be mergeable after it had proven unmergeable (above).
1683 * We don't worry about that case for efficiency. It won't happen
1684 * often, and the elevators are able to handle it.
1686 init_request_from_bio(req
, bio
);
1688 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1689 req
->cpu
= raw_smp_processor_id();
1691 plug
= current
->plug
;
1694 * If this is the first request added after a plug, fire
1697 * @request_count may become stale because of schedule
1698 * out, so check plug list again.
1700 if (!request_count
|| list_empty(&plug
->list
))
1701 trace_block_plug(q
);
1703 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1704 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1705 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1706 blk_flush_plug_list(plug
, false);
1707 trace_block_plug(q
);
1710 list_add_tail(&req
->queuelist
, &plug
->list
);
1711 blk_account_io_start(req
, true);
1713 spin_lock_irq(q
->queue_lock
);
1714 add_acct_request(q
, req
, where
);
1717 spin_unlock_irq(q
->queue_lock
);
1720 return BLK_QC_T_NONE
;
1724 * If bio->bi_dev is a partition, remap the location
1726 static inline void blk_partition_remap(struct bio
*bio
)
1728 struct block_device
*bdev
= bio
->bi_bdev
;
1731 * Zone reset does not include bi_size so bio_sectors() is always 0.
1732 * Include a test for the reset op code and perform the remap if needed.
1734 if (bdev
!= bdev
->bd_contains
&&
1735 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1736 struct hd_struct
*p
= bdev
->bd_part
;
1738 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1739 bio
->bi_bdev
= bdev
->bd_contains
;
1741 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1743 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1747 static void handle_bad_sector(struct bio
*bio
)
1749 char b
[BDEVNAME_SIZE
];
1751 printk(KERN_INFO
"attempt to access beyond end of device\n");
1752 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1753 bdevname(bio
->bi_bdev
, b
),
1755 (unsigned long long)bio_end_sector(bio
),
1756 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1759 #ifdef CONFIG_FAIL_MAKE_REQUEST
1761 static DECLARE_FAULT_ATTR(fail_make_request
);
1763 static int __init
setup_fail_make_request(char *str
)
1765 return setup_fault_attr(&fail_make_request
, str
);
1767 __setup("fail_make_request=", setup_fail_make_request
);
1769 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1771 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1774 static int __init
fail_make_request_debugfs(void)
1776 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1777 NULL
, &fail_make_request
);
1779 return PTR_ERR_OR_ZERO(dir
);
1782 late_initcall(fail_make_request_debugfs
);
1784 #else /* CONFIG_FAIL_MAKE_REQUEST */
1786 static inline bool should_fail_request(struct hd_struct
*part
,
1792 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1795 * Check whether this bio extends beyond the end of the device.
1797 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1804 /* Test device or partition size, when known. */
1805 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1807 sector_t sector
= bio
->bi_iter
.bi_sector
;
1809 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1811 * This may well happen - the kernel calls bread()
1812 * without checking the size of the device, e.g., when
1813 * mounting a device.
1815 handle_bad_sector(bio
);
1823 static noinline_for_stack
bool
1824 generic_make_request_checks(struct bio
*bio
)
1826 struct request_queue
*q
;
1827 int nr_sectors
= bio_sectors(bio
);
1829 char b
[BDEVNAME_SIZE
];
1830 struct hd_struct
*part
;
1834 if (bio_check_eod(bio
, nr_sectors
))
1837 q
= bdev_get_queue(bio
->bi_bdev
);
1840 "generic_make_request: Trying to access "
1841 "nonexistent block-device %s (%Lu)\n",
1842 bdevname(bio
->bi_bdev
, b
),
1843 (long long) bio
->bi_iter
.bi_sector
);
1847 part
= bio
->bi_bdev
->bd_part
;
1848 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1849 should_fail_request(&part_to_disk(part
)->part0
,
1850 bio
->bi_iter
.bi_size
))
1854 * If this device has partitions, remap block n
1855 * of partition p to block n+start(p) of the disk.
1857 blk_partition_remap(bio
);
1859 if (bio_check_eod(bio
, nr_sectors
))
1863 * Filter flush bio's early so that make_request based
1864 * drivers without flush support don't have to worry
1867 if (op_is_flush(bio
->bi_opf
) &&
1868 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1869 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1876 switch (bio_op(bio
)) {
1877 case REQ_OP_DISCARD
:
1878 if (!blk_queue_discard(q
))
1881 case REQ_OP_SECURE_ERASE
:
1882 if (!blk_queue_secure_erase(q
))
1885 case REQ_OP_WRITE_SAME
:
1886 if (!bdev_write_same(bio
->bi_bdev
))
1889 case REQ_OP_ZONE_REPORT
:
1890 case REQ_OP_ZONE_RESET
:
1891 if (!bdev_is_zoned(bio
->bi_bdev
))
1894 case REQ_OP_WRITE_ZEROES
:
1895 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1903 * Various block parts want %current->io_context and lazy ioc
1904 * allocation ends up trading a lot of pain for a small amount of
1905 * memory. Just allocate it upfront. This may fail and block
1906 * layer knows how to live with it.
1908 create_io_context(GFP_ATOMIC
, q
->node
);
1910 if (!blkcg_bio_issue_check(q
, bio
))
1913 trace_block_bio_queue(q
, bio
);
1919 bio
->bi_error
= err
;
1925 * generic_make_request - hand a buffer to its device driver for I/O
1926 * @bio: The bio describing the location in memory and on the device.
1928 * generic_make_request() is used to make I/O requests of block
1929 * devices. It is passed a &struct bio, which describes the I/O that needs
1932 * generic_make_request() does not return any status. The
1933 * success/failure status of the request, along with notification of
1934 * completion, is delivered asynchronously through the bio->bi_end_io
1935 * function described (one day) else where.
1937 * The caller of generic_make_request must make sure that bi_io_vec
1938 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1939 * set to describe the device address, and the
1940 * bi_end_io and optionally bi_private are set to describe how
1941 * completion notification should be signaled.
1943 * generic_make_request and the drivers it calls may use bi_next if this
1944 * bio happens to be merged with someone else, and may resubmit the bio to
1945 * a lower device by calling into generic_make_request recursively, which
1946 * means the bio should NOT be touched after the call to ->make_request_fn.
1948 blk_qc_t
generic_make_request(struct bio
*bio
)
1950 struct bio_list bio_list_on_stack
;
1951 blk_qc_t ret
= BLK_QC_T_NONE
;
1953 if (!generic_make_request_checks(bio
))
1957 * We only want one ->make_request_fn to be active at a time, else
1958 * stack usage with stacked devices could be a problem. So use
1959 * current->bio_list to keep a list of requests submited by a
1960 * make_request_fn function. current->bio_list is also used as a
1961 * flag to say if generic_make_request is currently active in this
1962 * task or not. If it is NULL, then no make_request is active. If
1963 * it is non-NULL, then a make_request is active, and new requests
1964 * should be added at the tail
1966 if (current
->bio_list
) {
1967 bio_list_add(current
->bio_list
, bio
);
1971 /* following loop may be a bit non-obvious, and so deserves some
1973 * Before entering the loop, bio->bi_next is NULL (as all callers
1974 * ensure that) so we have a list with a single bio.
1975 * We pretend that we have just taken it off a longer list, so
1976 * we assign bio_list to a pointer to the bio_list_on_stack,
1977 * thus initialising the bio_list of new bios to be
1978 * added. ->make_request() may indeed add some more bios
1979 * through a recursive call to generic_make_request. If it
1980 * did, we find a non-NULL value in bio_list and re-enter the loop
1981 * from the top. In this case we really did just take the bio
1982 * of the top of the list (no pretending) and so remove it from
1983 * bio_list, and call into ->make_request() again.
1985 BUG_ON(bio
->bi_next
);
1986 bio_list_init(&bio_list_on_stack
);
1987 current
->bio_list
= &bio_list_on_stack
;
1989 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1991 if (likely(blk_queue_enter(q
, false) == 0)) {
1992 ret
= q
->make_request_fn(q
, bio
);
1996 bio
= bio_list_pop(current
->bio_list
);
1998 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2004 current
->bio_list
= NULL
; /* deactivate */
2009 EXPORT_SYMBOL(generic_make_request
);
2012 * submit_bio - submit a bio to the block device layer for I/O
2013 * @bio: The &struct bio which describes the I/O
2015 * submit_bio() is very similar in purpose to generic_make_request(), and
2016 * uses that function to do most of the work. Both are fairly rough
2017 * interfaces; @bio must be presetup and ready for I/O.
2020 blk_qc_t
submit_bio(struct bio
*bio
)
2023 * If it's a regular read/write or a barrier with data attached,
2024 * go through the normal accounting stuff before submission.
2026 if (bio_has_data(bio
)) {
2029 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2030 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2032 count
= bio_sectors(bio
);
2034 if (op_is_write(bio_op(bio
))) {
2035 count_vm_events(PGPGOUT
, count
);
2037 task_io_account_read(bio
->bi_iter
.bi_size
);
2038 count_vm_events(PGPGIN
, count
);
2041 if (unlikely(block_dump
)) {
2042 char b
[BDEVNAME_SIZE
];
2043 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2044 current
->comm
, task_pid_nr(current
),
2045 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2046 (unsigned long long)bio
->bi_iter
.bi_sector
,
2047 bdevname(bio
->bi_bdev
, b
),
2052 return generic_make_request(bio
);
2054 EXPORT_SYMBOL(submit_bio
);
2057 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2058 * for new the queue limits
2060 * @rq: the request being checked
2063 * @rq may have been made based on weaker limitations of upper-level queues
2064 * in request stacking drivers, and it may violate the limitation of @q.
2065 * Since the block layer and the underlying device driver trust @rq
2066 * after it is inserted to @q, it should be checked against @q before
2067 * the insertion using this generic function.
2069 * Request stacking drivers like request-based dm may change the queue
2070 * limits when retrying requests on other queues. Those requests need
2071 * to be checked against the new queue limits again during dispatch.
2073 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2076 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2077 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2082 * queue's settings related to segment counting like q->bounce_pfn
2083 * may differ from that of other stacking queues.
2084 * Recalculate it to check the request correctly on this queue's
2087 blk_recalc_rq_segments(rq
);
2088 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2089 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2097 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2098 * @q: the queue to submit the request
2099 * @rq: the request being queued
2101 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2103 unsigned long flags
;
2104 int where
= ELEVATOR_INSERT_BACK
;
2106 if (blk_cloned_rq_check_limits(q
, rq
))
2110 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2114 if (blk_queue_io_stat(q
))
2115 blk_account_io_start(rq
, true);
2116 blk_mq_sched_insert_request(rq
, false, true, false, false);
2120 spin_lock_irqsave(q
->queue_lock
, flags
);
2121 if (unlikely(blk_queue_dying(q
))) {
2122 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2127 * Submitting request must be dequeued before calling this function
2128 * because it will be linked to another request_queue
2130 BUG_ON(blk_queued_rq(rq
));
2132 if (op_is_flush(rq
->cmd_flags
))
2133 where
= ELEVATOR_INSERT_FLUSH
;
2135 add_acct_request(q
, rq
, where
);
2136 if (where
== ELEVATOR_INSERT_FLUSH
)
2138 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2142 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2145 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2146 * @rq: request to examine
2149 * A request could be merge of IOs which require different failure
2150 * handling. This function determines the number of bytes which
2151 * can be failed from the beginning of the request without
2152 * crossing into area which need to be retried further.
2155 * The number of bytes to fail.
2158 * queue_lock must be held.
2160 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2162 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2163 unsigned int bytes
= 0;
2166 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2167 return blk_rq_bytes(rq
);
2170 * Currently the only 'mixing' which can happen is between
2171 * different fastfail types. We can safely fail portions
2172 * which have all the failfast bits that the first one has -
2173 * the ones which are at least as eager to fail as the first
2176 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2177 if ((bio
->bi_opf
& ff
) != ff
)
2179 bytes
+= bio
->bi_iter
.bi_size
;
2182 /* this could lead to infinite loop */
2183 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2186 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2188 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2190 if (blk_do_io_stat(req
)) {
2191 const int rw
= rq_data_dir(req
);
2192 struct hd_struct
*part
;
2195 cpu
= part_stat_lock();
2197 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2202 void blk_account_io_done(struct request
*req
)
2205 * Account IO completion. flush_rq isn't accounted as a
2206 * normal IO on queueing nor completion. Accounting the
2207 * containing request is enough.
2209 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2210 unsigned long duration
= jiffies
- req
->start_time
;
2211 const int rw
= rq_data_dir(req
);
2212 struct hd_struct
*part
;
2215 cpu
= part_stat_lock();
2218 part_stat_inc(cpu
, part
, ios
[rw
]);
2219 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2220 part_round_stats(cpu
, part
);
2221 part_dec_in_flight(part
, rw
);
2223 hd_struct_put(part
);
2230 * Don't process normal requests when queue is suspended
2231 * or in the process of suspending/resuming
2233 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2236 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2237 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2243 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2250 void blk_account_io_start(struct request
*rq
, bool new_io
)
2252 struct hd_struct
*part
;
2253 int rw
= rq_data_dir(rq
);
2256 if (!blk_do_io_stat(rq
))
2259 cpu
= part_stat_lock();
2263 part_stat_inc(cpu
, part
, merges
[rw
]);
2265 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2266 if (!hd_struct_try_get(part
)) {
2268 * The partition is already being removed,
2269 * the request will be accounted on the disk only
2271 * We take a reference on disk->part0 although that
2272 * partition will never be deleted, so we can treat
2273 * it as any other partition.
2275 part
= &rq
->rq_disk
->part0
;
2276 hd_struct_get(part
);
2278 part_round_stats(cpu
, part
);
2279 part_inc_in_flight(part
, rw
);
2287 * blk_peek_request - peek at the top of a request queue
2288 * @q: request queue to peek at
2291 * Return the request at the top of @q. The returned request
2292 * should be started using blk_start_request() before LLD starts
2296 * Pointer to the request at the top of @q if available. Null
2300 * queue_lock must be held.
2302 struct request
*blk_peek_request(struct request_queue
*q
)
2307 while ((rq
= __elv_next_request(q
)) != NULL
) {
2309 rq
= blk_pm_peek_request(q
, rq
);
2313 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2315 * This is the first time the device driver
2316 * sees this request (possibly after
2317 * requeueing). Notify IO scheduler.
2319 if (rq
->rq_flags
& RQF_SORTED
)
2320 elv_activate_rq(q
, rq
);
2323 * just mark as started even if we don't start
2324 * it, a request that has been delayed should
2325 * not be passed by new incoming requests
2327 rq
->rq_flags
|= RQF_STARTED
;
2328 trace_block_rq_issue(q
, rq
);
2331 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2332 q
->end_sector
= rq_end_sector(rq
);
2333 q
->boundary_rq
= NULL
;
2336 if (rq
->rq_flags
& RQF_DONTPREP
)
2339 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2341 * make sure space for the drain appears we
2342 * know we can do this because max_hw_segments
2343 * has been adjusted to be one fewer than the
2346 rq
->nr_phys_segments
++;
2352 ret
= q
->prep_rq_fn(q
, rq
);
2353 if (ret
== BLKPREP_OK
) {
2355 } else if (ret
== BLKPREP_DEFER
) {
2357 * the request may have been (partially) prepped.
2358 * we need to keep this request in the front to
2359 * avoid resource deadlock. RQF_STARTED will
2360 * prevent other fs requests from passing this one.
2362 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2363 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2365 * remove the space for the drain we added
2366 * so that we don't add it again
2368 --rq
->nr_phys_segments
;
2373 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2374 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2376 rq
->rq_flags
|= RQF_QUIET
;
2378 * Mark this request as started so we don't trigger
2379 * any debug logic in the end I/O path.
2381 blk_start_request(rq
);
2382 __blk_end_request_all(rq
, err
);
2384 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2391 EXPORT_SYMBOL(blk_peek_request
);
2393 void blk_dequeue_request(struct request
*rq
)
2395 struct request_queue
*q
= rq
->q
;
2397 BUG_ON(list_empty(&rq
->queuelist
));
2398 BUG_ON(ELV_ON_HASH(rq
));
2400 list_del_init(&rq
->queuelist
);
2403 * the time frame between a request being removed from the lists
2404 * and to it is freed is accounted as io that is in progress at
2407 if (blk_account_rq(rq
)) {
2408 q
->in_flight
[rq_is_sync(rq
)]++;
2409 set_io_start_time_ns(rq
);
2414 * blk_start_request - start request processing on the driver
2415 * @req: request to dequeue
2418 * Dequeue @req and start timeout timer on it. This hands off the
2419 * request to the driver.
2421 * Block internal functions which don't want to start timer should
2422 * call blk_dequeue_request().
2425 * queue_lock must be held.
2427 void blk_start_request(struct request
*req
)
2429 blk_dequeue_request(req
);
2431 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2432 blk_stat_set_issue_time(&req
->issue_stat
);
2433 req
->rq_flags
|= RQF_STATS
;
2434 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2437 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2440 EXPORT_SYMBOL(blk_start_request
);
2443 * blk_fetch_request - fetch a request from a request queue
2444 * @q: request queue to fetch a request from
2447 * Return the request at the top of @q. The request is started on
2448 * return and LLD can start processing it immediately.
2451 * Pointer to the request at the top of @q if available. Null
2455 * queue_lock must be held.
2457 struct request
*blk_fetch_request(struct request_queue
*q
)
2461 rq
= blk_peek_request(q
);
2463 blk_start_request(rq
);
2466 EXPORT_SYMBOL(blk_fetch_request
);
2469 * blk_update_request - Special helper function for request stacking drivers
2470 * @req: the request being processed
2471 * @error: %0 for success, < %0 for error
2472 * @nr_bytes: number of bytes to complete @req
2475 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2476 * the request structure even if @req doesn't have leftover.
2477 * If @req has leftover, sets it up for the next range of segments.
2479 * This special helper function is only for request stacking drivers
2480 * (e.g. request-based dm) so that they can handle partial completion.
2481 * Actual device drivers should use blk_end_request instead.
2483 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2484 * %false return from this function.
2487 * %false - this request doesn't have any more data
2488 * %true - this request has more data
2490 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2494 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2500 * For fs requests, rq is just carrier of independent bio's
2501 * and each partial completion should be handled separately.
2502 * Reset per-request error on each partial completion.
2504 * TODO: tj: This is too subtle. It would be better to let
2505 * low level drivers do what they see fit.
2507 if (!blk_rq_is_passthrough(req
))
2510 if (error
&& !blk_rq_is_passthrough(req
) &&
2511 !(req
->rq_flags
& RQF_QUIET
)) {
2516 error_type
= "recoverable transport";
2519 error_type
= "critical target";
2522 error_type
= "critical nexus";
2525 error_type
= "timeout";
2528 error_type
= "critical space allocation";
2531 error_type
= "critical medium";
2538 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2539 __func__
, error_type
, req
->rq_disk
?
2540 req
->rq_disk
->disk_name
: "?",
2541 (unsigned long long)blk_rq_pos(req
));
2545 blk_account_io_completion(req
, nr_bytes
);
2549 struct bio
*bio
= req
->bio
;
2550 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2552 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2553 req
->bio
= bio
->bi_next
;
2555 req_bio_endio(req
, bio
, bio_bytes
, error
);
2557 total_bytes
+= bio_bytes
;
2558 nr_bytes
-= bio_bytes
;
2569 * Reset counters so that the request stacking driver
2570 * can find how many bytes remain in the request
2573 req
->__data_len
= 0;
2577 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2579 req
->__data_len
-= total_bytes
;
2581 /* update sector only for requests with clear definition of sector */
2582 if (!blk_rq_is_passthrough(req
))
2583 req
->__sector
+= total_bytes
>> 9;
2585 /* mixed attributes always follow the first bio */
2586 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2587 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2588 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2592 * If total number of sectors is less than the first segment
2593 * size, something has gone terribly wrong.
2595 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2596 blk_dump_rq_flags(req
, "request botched");
2597 req
->__data_len
= blk_rq_cur_bytes(req
);
2600 /* recalculate the number of segments */
2601 blk_recalc_rq_segments(req
);
2605 EXPORT_SYMBOL_GPL(blk_update_request
);
2607 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2608 unsigned int nr_bytes
,
2609 unsigned int bidi_bytes
)
2611 if (blk_update_request(rq
, error
, nr_bytes
))
2614 /* Bidi request must be completed as a whole */
2615 if (unlikely(blk_bidi_rq(rq
)) &&
2616 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2619 if (blk_queue_add_random(rq
->q
))
2620 add_disk_randomness(rq
->rq_disk
);
2626 * blk_unprep_request - unprepare a request
2629 * This function makes a request ready for complete resubmission (or
2630 * completion). It happens only after all error handling is complete,
2631 * so represents the appropriate moment to deallocate any resources
2632 * that were allocated to the request in the prep_rq_fn. The queue
2633 * lock is held when calling this.
2635 void blk_unprep_request(struct request
*req
)
2637 struct request_queue
*q
= req
->q
;
2639 req
->rq_flags
&= ~RQF_DONTPREP
;
2640 if (q
->unprep_rq_fn
)
2641 q
->unprep_rq_fn(q
, req
);
2643 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2646 * queue lock must be held
2648 void blk_finish_request(struct request
*req
, int error
)
2650 struct request_queue
*q
= req
->q
;
2652 if (req
->rq_flags
& RQF_STATS
)
2653 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2655 if (req
->rq_flags
& RQF_QUEUED
)
2656 blk_queue_end_tag(q
, req
);
2658 BUG_ON(blk_queued_rq(req
));
2660 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2661 laptop_io_completion(req
->q
->backing_dev_info
);
2663 blk_delete_timer(req
);
2665 if (req
->rq_flags
& RQF_DONTPREP
)
2666 blk_unprep_request(req
);
2668 blk_account_io_done(req
);
2671 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2672 req
->end_io(req
, error
);
2674 if (blk_bidi_rq(req
))
2675 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2677 __blk_put_request(q
, req
);
2680 EXPORT_SYMBOL(blk_finish_request
);
2683 * blk_end_bidi_request - Complete a bidi request
2684 * @rq: the request to complete
2685 * @error: %0 for success, < %0 for error
2686 * @nr_bytes: number of bytes to complete @rq
2687 * @bidi_bytes: number of bytes to complete @rq->next_rq
2690 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2691 * Drivers that supports bidi can safely call this member for any
2692 * type of request, bidi or uni. In the later case @bidi_bytes is
2696 * %false - we are done with this request
2697 * %true - still buffers pending for this request
2699 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2700 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2702 struct request_queue
*q
= rq
->q
;
2703 unsigned long flags
;
2705 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2708 spin_lock_irqsave(q
->queue_lock
, flags
);
2709 blk_finish_request(rq
, error
);
2710 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2716 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2717 * @rq: the request to complete
2718 * @error: %0 for success, < %0 for error
2719 * @nr_bytes: number of bytes to complete @rq
2720 * @bidi_bytes: number of bytes to complete @rq->next_rq
2723 * Identical to blk_end_bidi_request() except that queue lock is
2724 * assumed to be locked on entry and remains so on return.
2727 * %false - we are done with this request
2728 * %true - still buffers pending for this request
2730 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2731 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2733 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2736 blk_finish_request(rq
, error
);
2742 * blk_end_request - Helper function for drivers to complete the request.
2743 * @rq: the request being processed
2744 * @error: %0 for success, < %0 for error
2745 * @nr_bytes: number of bytes to complete
2748 * Ends I/O on a number of bytes attached to @rq.
2749 * If @rq has leftover, sets it up for the next range of segments.
2752 * %false - we are done with this request
2753 * %true - still buffers pending for this request
2755 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2757 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2759 EXPORT_SYMBOL(blk_end_request
);
2762 * blk_end_request_all - Helper function for drives to finish the request.
2763 * @rq: the request to finish
2764 * @error: %0 for success, < %0 for error
2767 * Completely finish @rq.
2769 void blk_end_request_all(struct request
*rq
, int error
)
2772 unsigned int bidi_bytes
= 0;
2774 if (unlikely(blk_bidi_rq(rq
)))
2775 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2777 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2780 EXPORT_SYMBOL(blk_end_request_all
);
2783 * blk_end_request_cur - Helper function to finish the current request chunk.
2784 * @rq: the request to finish the current chunk for
2785 * @error: %0 for success, < %0 for error
2788 * Complete the current consecutively mapped chunk from @rq.
2791 * %false - we are done with this request
2792 * %true - still buffers pending for this request
2794 bool blk_end_request_cur(struct request
*rq
, int error
)
2796 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2798 EXPORT_SYMBOL(blk_end_request_cur
);
2801 * blk_end_request_err - Finish a request till the next failure boundary.
2802 * @rq: the request to finish till the next failure boundary for
2803 * @error: must be negative errno
2806 * Complete @rq till the next failure boundary.
2809 * %false - we are done with this request
2810 * %true - still buffers pending for this request
2812 bool blk_end_request_err(struct request
*rq
, int error
)
2814 WARN_ON(error
>= 0);
2815 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2817 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2820 * __blk_end_request - Helper function for drivers to complete the request.
2821 * @rq: the request being processed
2822 * @error: %0 for success, < %0 for error
2823 * @nr_bytes: number of bytes to complete
2826 * Must be called with queue lock held unlike blk_end_request().
2829 * %false - we are done with this request
2830 * %true - still buffers pending for this request
2832 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2834 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2836 EXPORT_SYMBOL(__blk_end_request
);
2839 * __blk_end_request_all - Helper function for drives to finish the request.
2840 * @rq: the request to finish
2841 * @error: %0 for success, < %0 for error
2844 * Completely finish @rq. Must be called with queue lock held.
2846 void __blk_end_request_all(struct request
*rq
, int error
)
2849 unsigned int bidi_bytes
= 0;
2851 if (unlikely(blk_bidi_rq(rq
)))
2852 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2854 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2857 EXPORT_SYMBOL(__blk_end_request_all
);
2860 * __blk_end_request_cur - Helper function to finish the current request chunk.
2861 * @rq: the request to finish the current chunk for
2862 * @error: %0 for success, < %0 for error
2865 * Complete the current consecutively mapped chunk from @rq. Must
2866 * be called with queue lock held.
2869 * %false - we are done with this request
2870 * %true - still buffers pending for this request
2872 bool __blk_end_request_cur(struct request
*rq
, int error
)
2874 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2876 EXPORT_SYMBOL(__blk_end_request_cur
);
2879 * __blk_end_request_err - Finish a request till the next failure boundary.
2880 * @rq: the request to finish till the next failure boundary for
2881 * @error: must be negative errno
2884 * Complete @rq till the next failure boundary. Must be called
2885 * with queue lock held.
2888 * %false - we are done with this request
2889 * %true - still buffers pending for this request
2891 bool __blk_end_request_err(struct request
*rq
, int error
)
2893 WARN_ON(error
>= 0);
2894 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2896 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2898 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2901 if (bio_has_data(bio
))
2902 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2904 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2905 rq
->bio
= rq
->biotail
= bio
;
2908 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2911 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2913 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2914 * @rq: the request to be flushed
2917 * Flush all pages in @rq.
2919 void rq_flush_dcache_pages(struct request
*rq
)
2921 struct req_iterator iter
;
2922 struct bio_vec bvec
;
2924 rq_for_each_segment(bvec
, rq
, iter
)
2925 flush_dcache_page(bvec
.bv_page
);
2927 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2931 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2932 * @q : the queue of the device being checked
2935 * Check if underlying low-level drivers of a device are busy.
2936 * If the drivers want to export their busy state, they must set own
2937 * exporting function using blk_queue_lld_busy() first.
2939 * Basically, this function is used only by request stacking drivers
2940 * to stop dispatching requests to underlying devices when underlying
2941 * devices are busy. This behavior helps more I/O merging on the queue
2942 * of the request stacking driver and prevents I/O throughput regression
2943 * on burst I/O load.
2946 * 0 - Not busy (The request stacking driver should dispatch request)
2947 * 1 - Busy (The request stacking driver should stop dispatching request)
2949 int blk_lld_busy(struct request_queue
*q
)
2952 return q
->lld_busy_fn(q
);
2956 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2959 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2960 * @rq: the clone request to be cleaned up
2963 * Free all bios in @rq for a cloned request.
2965 void blk_rq_unprep_clone(struct request
*rq
)
2969 while ((bio
= rq
->bio
) != NULL
) {
2970 rq
->bio
= bio
->bi_next
;
2975 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2978 * Copy attributes of the original request to the clone request.
2979 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2981 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2983 dst
->cpu
= src
->cpu
;
2984 dst
->__sector
= blk_rq_pos(src
);
2985 dst
->__data_len
= blk_rq_bytes(src
);
2986 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2987 dst
->ioprio
= src
->ioprio
;
2988 dst
->extra_len
= src
->extra_len
;
2992 * blk_rq_prep_clone - Helper function to setup clone request
2993 * @rq: the request to be setup
2994 * @rq_src: original request to be cloned
2995 * @bs: bio_set that bios for clone are allocated from
2996 * @gfp_mask: memory allocation mask for bio
2997 * @bio_ctr: setup function to be called for each clone bio.
2998 * Returns %0 for success, non %0 for failure.
2999 * @data: private data to be passed to @bio_ctr
3002 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3003 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3004 * are not copied, and copying such parts is the caller's responsibility.
3005 * Also, pages which the original bios are pointing to are not copied
3006 * and the cloned bios just point same pages.
3007 * So cloned bios must be completed before original bios, which means
3008 * the caller must complete @rq before @rq_src.
3010 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3011 struct bio_set
*bs
, gfp_t gfp_mask
,
3012 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3015 struct bio
*bio
, *bio_src
;
3020 __rq_for_each_bio(bio_src
, rq_src
) {
3021 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3025 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3029 rq
->biotail
->bi_next
= bio
;
3032 rq
->bio
= rq
->biotail
= bio
;
3035 __blk_rq_prep_clone(rq
, rq_src
);
3042 blk_rq_unprep_clone(rq
);
3046 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3048 int kblockd_schedule_work(struct work_struct
*work
)
3050 return queue_work(kblockd_workqueue
, work
);
3052 EXPORT_SYMBOL(kblockd_schedule_work
);
3054 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3056 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3058 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3060 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3061 unsigned long delay
)
3063 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3065 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3067 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3068 unsigned long delay
)
3070 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3072 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3075 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3076 * @plug: The &struct blk_plug that needs to be initialized
3079 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3080 * pending I/O should the task end up blocking between blk_start_plug() and
3081 * blk_finish_plug(). This is important from a performance perspective, but
3082 * also ensures that we don't deadlock. For instance, if the task is blocking
3083 * for a memory allocation, memory reclaim could end up wanting to free a
3084 * page belonging to that request that is currently residing in our private
3085 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3086 * this kind of deadlock.
3088 void blk_start_plug(struct blk_plug
*plug
)
3090 struct task_struct
*tsk
= current
;
3093 * If this is a nested plug, don't actually assign it.
3098 INIT_LIST_HEAD(&plug
->list
);
3099 INIT_LIST_HEAD(&plug
->mq_list
);
3100 INIT_LIST_HEAD(&plug
->cb_list
);
3102 * Store ordering should not be needed here, since a potential
3103 * preempt will imply a full memory barrier
3107 EXPORT_SYMBOL(blk_start_plug
);
3109 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3111 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3112 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3114 return !(rqa
->q
< rqb
->q
||
3115 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3119 * If 'from_schedule' is true, then postpone the dispatch of requests
3120 * until a safe kblockd context. We due this to avoid accidental big
3121 * additional stack usage in driver dispatch, in places where the originally
3122 * plugger did not intend it.
3124 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3126 __releases(q
->queue_lock
)
3128 trace_block_unplug(q
, depth
, !from_schedule
);
3131 blk_run_queue_async(q
);
3134 spin_unlock(q
->queue_lock
);
3137 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3139 LIST_HEAD(callbacks
);
3141 while (!list_empty(&plug
->cb_list
)) {
3142 list_splice_init(&plug
->cb_list
, &callbacks
);
3144 while (!list_empty(&callbacks
)) {
3145 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3148 list_del(&cb
->list
);
3149 cb
->callback(cb
, from_schedule
);
3154 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3157 struct blk_plug
*plug
= current
->plug
;
3158 struct blk_plug_cb
*cb
;
3163 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3164 if (cb
->callback
== unplug
&& cb
->data
== data
)
3167 /* Not currently on the callback list */
3168 BUG_ON(size
< sizeof(*cb
));
3169 cb
= kzalloc(size
, GFP_ATOMIC
);
3172 cb
->callback
= unplug
;
3173 list_add(&cb
->list
, &plug
->cb_list
);
3177 EXPORT_SYMBOL(blk_check_plugged
);
3179 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3181 struct request_queue
*q
;
3182 unsigned long flags
;
3187 flush_plug_callbacks(plug
, from_schedule
);
3189 if (!list_empty(&plug
->mq_list
))
3190 blk_mq_flush_plug_list(plug
, from_schedule
);
3192 if (list_empty(&plug
->list
))
3195 list_splice_init(&plug
->list
, &list
);
3197 list_sort(NULL
, &list
, plug_rq_cmp
);
3203 * Save and disable interrupts here, to avoid doing it for every
3204 * queue lock we have to take.
3206 local_irq_save(flags
);
3207 while (!list_empty(&list
)) {
3208 rq
= list_entry_rq(list
.next
);
3209 list_del_init(&rq
->queuelist
);
3213 * This drops the queue lock
3216 queue_unplugged(q
, depth
, from_schedule
);
3219 spin_lock(q
->queue_lock
);
3223 * Short-circuit if @q is dead
3225 if (unlikely(blk_queue_dying(q
))) {
3226 __blk_end_request_all(rq
, -ENODEV
);
3231 * rq is already accounted, so use raw insert
3233 if (op_is_flush(rq
->cmd_flags
))
3234 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3236 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3242 * This drops the queue lock
3245 queue_unplugged(q
, depth
, from_schedule
);
3247 local_irq_restore(flags
);
3250 void blk_finish_plug(struct blk_plug
*plug
)
3252 if (plug
!= current
->plug
)
3254 blk_flush_plug_list(plug
, false);
3256 current
->plug
= NULL
;
3258 EXPORT_SYMBOL(blk_finish_plug
);
3262 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3263 * @q: the queue of the device
3264 * @dev: the device the queue belongs to
3267 * Initialize runtime-PM-related fields for @q and start auto suspend for
3268 * @dev. Drivers that want to take advantage of request-based runtime PM
3269 * should call this function after @dev has been initialized, and its
3270 * request queue @q has been allocated, and runtime PM for it can not happen
3271 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3272 * cases, driver should call this function before any I/O has taken place.
3274 * This function takes care of setting up using auto suspend for the device,
3275 * the autosuspend delay is set to -1 to make runtime suspend impossible
3276 * until an updated value is either set by user or by driver. Drivers do
3277 * not need to touch other autosuspend settings.
3279 * The block layer runtime PM is request based, so only works for drivers
3280 * that use request as their IO unit instead of those directly use bio's.
3282 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3285 q
->rpm_status
= RPM_ACTIVE
;
3286 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3287 pm_runtime_use_autosuspend(q
->dev
);
3289 EXPORT_SYMBOL(blk_pm_runtime_init
);
3292 * blk_pre_runtime_suspend - Pre runtime suspend check
3293 * @q: the queue of the device
3296 * This function will check if runtime suspend is allowed for the device
3297 * by examining if there are any requests pending in the queue. If there
3298 * are requests pending, the device can not be runtime suspended; otherwise,
3299 * the queue's status will be updated to SUSPENDING and the driver can
3300 * proceed to suspend the device.
3302 * For the not allowed case, we mark last busy for the device so that
3303 * runtime PM core will try to autosuspend it some time later.
3305 * This function should be called near the start of the device's
3306 * runtime_suspend callback.
3309 * 0 - OK to runtime suspend the device
3310 * -EBUSY - Device should not be runtime suspended
3312 int blk_pre_runtime_suspend(struct request_queue
*q
)
3319 spin_lock_irq(q
->queue_lock
);
3320 if (q
->nr_pending
) {
3322 pm_runtime_mark_last_busy(q
->dev
);
3324 q
->rpm_status
= RPM_SUSPENDING
;
3326 spin_unlock_irq(q
->queue_lock
);
3329 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3332 * blk_post_runtime_suspend - Post runtime suspend processing
3333 * @q: the queue of the device
3334 * @err: return value of the device's runtime_suspend function
3337 * Update the queue's runtime status according to the return value of the
3338 * device's runtime suspend function and mark last busy for the device so
3339 * that PM core will try to auto suspend the device at a later time.
3341 * This function should be called near the end of the device's
3342 * runtime_suspend callback.
3344 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3349 spin_lock_irq(q
->queue_lock
);
3351 q
->rpm_status
= RPM_SUSPENDED
;
3353 q
->rpm_status
= RPM_ACTIVE
;
3354 pm_runtime_mark_last_busy(q
->dev
);
3356 spin_unlock_irq(q
->queue_lock
);
3358 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3361 * blk_pre_runtime_resume - Pre runtime resume processing
3362 * @q: the queue of the device
3365 * Update the queue's runtime status to RESUMING in preparation for the
3366 * runtime resume of the device.
3368 * This function should be called near the start of the device's
3369 * runtime_resume callback.
3371 void blk_pre_runtime_resume(struct request_queue
*q
)
3376 spin_lock_irq(q
->queue_lock
);
3377 q
->rpm_status
= RPM_RESUMING
;
3378 spin_unlock_irq(q
->queue_lock
);
3380 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3383 * blk_post_runtime_resume - Post runtime resume processing
3384 * @q: the queue of the device
3385 * @err: return value of the device's runtime_resume function
3388 * Update the queue's runtime status according to the return value of the
3389 * device's runtime_resume function. If it is successfully resumed, process
3390 * the requests that are queued into the device's queue when it is resuming
3391 * and then mark last busy and initiate autosuspend for it.
3393 * This function should be called near the end of the device's
3394 * runtime_resume callback.
3396 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3401 spin_lock_irq(q
->queue_lock
);
3403 q
->rpm_status
= RPM_ACTIVE
;
3405 pm_runtime_mark_last_busy(q
->dev
);
3406 pm_request_autosuspend(q
->dev
);
3408 q
->rpm_status
= RPM_SUSPENDED
;
3410 spin_unlock_irq(q
->queue_lock
);
3412 EXPORT_SYMBOL(blk_post_runtime_resume
);
3415 * blk_set_runtime_active - Force runtime status of the queue to be active
3416 * @q: the queue of the device
3418 * If the device is left runtime suspended during system suspend the resume
3419 * hook typically resumes the device and corrects runtime status
3420 * accordingly. However, that does not affect the queue runtime PM status
3421 * which is still "suspended". This prevents processing requests from the
3424 * This function can be used in driver's resume hook to correct queue
3425 * runtime PM status and re-enable peeking requests from the queue. It
3426 * should be called before first request is added to the queue.
3428 void blk_set_runtime_active(struct request_queue
*q
)
3430 spin_lock_irq(q
->queue_lock
);
3431 q
->rpm_status
= RPM_ACTIVE
;
3432 pm_runtime_mark_last_busy(q
->dev
);
3433 pm_request_autosuspend(q
->dev
);
3434 spin_unlock_irq(q
->queue_lock
);
3436 EXPORT_SYMBOL(blk_set_runtime_active
);
3439 int __init
blk_dev_init(void)
3441 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3442 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3443 FIELD_SIZEOF(struct request
, cmd_flags
));
3444 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3445 FIELD_SIZEOF(struct bio
, bi_opf
));
3447 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3448 kblockd_workqueue
= alloc_workqueue("kblockd",
3449 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3450 if (!kblockd_workqueue
)
3451 panic("Failed to create kblockd\n");
3453 request_cachep
= kmem_cache_create("blkdev_requests",
3454 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3456 blk_requestq_cachep
= kmem_cache_create("request_queue",
3457 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);