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>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/block.h>
40 #include "blk-cgroup.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_unplug
);
48 DEFINE_IDA(blk_queue_ida
);
51 * For the allocated request tables
53 struct kmem_cache
*request_cachep
= NULL
;
56 * For queue allocation
58 struct kmem_cache
*blk_requestq_cachep
;
61 * Controlling structure to kblockd
63 static struct workqueue_struct
*kblockd_workqueue
;
65 void blk_queue_congestion_threshold(struct request_queue
*q
)
69 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
70 if (nr
> q
->nr_requests
)
72 q
->nr_congestion_on
= nr
;
74 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
77 q
->nr_congestion_off
= nr
;
81 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
84 * Locates the passed device's request queue and returns the address of its
87 * Will return NULL if the request queue cannot be located.
89 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
91 struct backing_dev_info
*ret
= NULL
;
92 struct request_queue
*q
= bdev_get_queue(bdev
);
95 ret
= &q
->backing_dev_info
;
98 EXPORT_SYMBOL(blk_get_backing_dev_info
);
100 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
102 memset(rq
, 0, sizeof(*rq
));
104 INIT_LIST_HEAD(&rq
->queuelist
);
105 INIT_LIST_HEAD(&rq
->timeout_list
);
108 rq
->__sector
= (sector_t
) -1;
109 INIT_HLIST_NODE(&rq
->hash
);
110 RB_CLEAR_NODE(&rq
->rb_node
);
112 rq
->cmd_len
= BLK_MAX_CDB
;
114 rq
->start_time
= jiffies
;
115 set_start_time_ns(rq
);
118 EXPORT_SYMBOL(blk_rq_init
);
120 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
121 unsigned int nbytes
, int error
)
124 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
125 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
128 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
129 set_bit(BIO_QUIET
, &bio
->bi_flags
);
131 bio_advance(bio
, nbytes
);
133 /* don't actually finish bio if it's part of flush sequence */
134 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
135 bio_endio(bio
, error
);
138 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
142 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
143 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
144 (unsigned long long) rq
->cmd_flags
);
146 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
147 (unsigned long long)blk_rq_pos(rq
),
148 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
149 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
150 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
152 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
153 printk(KERN_INFO
" cdb: ");
154 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
155 printk("%02x ", rq
->cmd
[bit
]);
159 EXPORT_SYMBOL(blk_dump_rq_flags
);
161 static void blk_delay_work(struct work_struct
*work
)
163 struct request_queue
*q
;
165 q
= container_of(work
, struct request_queue
, delay_work
.work
);
166 spin_lock_irq(q
->queue_lock
);
168 spin_unlock_irq(q
->queue_lock
);
172 * blk_delay_queue - restart queueing after defined interval
173 * @q: The &struct request_queue in question
174 * @msecs: Delay in msecs
177 * Sometimes queueing needs to be postponed for a little while, to allow
178 * resources to come back. This function will make sure that queueing is
179 * restarted around the specified time. Queue lock must be held.
181 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
183 if (likely(!blk_queue_dead(q
)))
184 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
185 msecs_to_jiffies(msecs
));
187 EXPORT_SYMBOL(blk_delay_queue
);
190 * blk_start_queue - restart a previously stopped queue
191 * @q: The &struct request_queue in question
194 * blk_start_queue() will clear the stop flag on the queue, and call
195 * the request_fn for the queue if it was in a stopped state when
196 * entered. Also see blk_stop_queue(). Queue lock must be held.
198 void blk_start_queue(struct request_queue
*q
)
200 WARN_ON(!irqs_disabled());
202 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
205 EXPORT_SYMBOL(blk_start_queue
);
208 * blk_stop_queue - stop a queue
209 * @q: The &struct request_queue in question
212 * The Linux block layer assumes that a block driver will consume all
213 * entries on the request queue when the request_fn strategy is called.
214 * Often this will not happen, because of hardware limitations (queue
215 * depth settings). If a device driver gets a 'queue full' response,
216 * or if it simply chooses not to queue more I/O at one point, it can
217 * call this function to prevent the request_fn from being called until
218 * the driver has signalled it's ready to go again. This happens by calling
219 * blk_start_queue() to restart queue operations. Queue lock must be held.
221 void blk_stop_queue(struct request_queue
*q
)
223 cancel_delayed_work(&q
->delay_work
);
224 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
226 EXPORT_SYMBOL(blk_stop_queue
);
229 * blk_sync_queue - cancel any pending callbacks on a queue
233 * The block layer may perform asynchronous callback activity
234 * on a queue, such as calling the unplug function after a timeout.
235 * A block device may call blk_sync_queue to ensure that any
236 * such activity is cancelled, thus allowing it to release resources
237 * that the callbacks might use. The caller must already have made sure
238 * that its ->make_request_fn will not re-add plugging prior to calling
241 * This function does not cancel any asynchronous activity arising
242 * out of elevator or throttling code. That would require elevaotor_exit()
243 * and blkcg_exit_queue() to be called with queue lock initialized.
246 void blk_sync_queue(struct request_queue
*q
)
248 del_timer_sync(&q
->timeout
);
249 cancel_delayed_work_sync(&q
->delay_work
);
251 EXPORT_SYMBOL(blk_sync_queue
);
254 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
255 * @q: The queue to run
258 * Invoke request handling on a queue if there are any pending requests.
259 * May be used to restart request handling after a request has completed.
260 * This variant runs the queue whether or not the queue has been
261 * stopped. Must be called with the queue lock held and interrupts
262 * disabled. See also @blk_run_queue.
264 inline void __blk_run_queue_uncond(struct request_queue
*q
)
266 if (unlikely(blk_queue_dead(q
)))
270 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
271 * the queue lock internally. As a result multiple threads may be
272 * running such a request function concurrently. Keep track of the
273 * number of active request_fn invocations such that blk_drain_queue()
274 * can wait until all these request_fn calls have finished.
276 q
->request_fn_active
++;
278 q
->request_fn_active
--;
282 * __blk_run_queue - run a single device queue
283 * @q: The queue to run
286 * See @blk_run_queue. This variant must be called with the queue lock
287 * held and interrupts disabled.
289 void __blk_run_queue(struct request_queue
*q
)
291 if (unlikely(blk_queue_stopped(q
)))
294 __blk_run_queue_uncond(q
);
296 EXPORT_SYMBOL(__blk_run_queue
);
299 * blk_run_queue_async - run a single device queue in workqueue context
300 * @q: The queue to run
303 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
304 * of us. The caller must hold the queue lock.
306 void blk_run_queue_async(struct request_queue
*q
)
308 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
309 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
311 EXPORT_SYMBOL(blk_run_queue_async
);
314 * blk_run_queue - run a single device queue
315 * @q: The queue to run
318 * Invoke request handling on this queue, if it has pending work to do.
319 * May be used to restart queueing when a request has completed.
321 void blk_run_queue(struct request_queue
*q
)
325 spin_lock_irqsave(q
->queue_lock
, flags
);
327 spin_unlock_irqrestore(q
->queue_lock
, flags
);
329 EXPORT_SYMBOL(blk_run_queue
);
331 void blk_put_queue(struct request_queue
*q
)
333 kobject_put(&q
->kobj
);
335 EXPORT_SYMBOL(blk_put_queue
);
338 * __blk_drain_queue - drain requests from request_queue
340 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
342 * Drain requests from @q. If @drain_all is set, all requests are drained.
343 * If not, only ELVPRIV requests are drained. The caller is responsible
344 * for ensuring that no new requests which need to be drained are queued.
346 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
347 __releases(q
->queue_lock
)
348 __acquires(q
->queue_lock
)
352 lockdep_assert_held(q
->queue_lock
);
358 * The caller might be trying to drain @q before its
359 * elevator is initialized.
362 elv_drain_elevator(q
);
364 blkcg_drain_queue(q
);
367 * This function might be called on a queue which failed
368 * driver init after queue creation or is not yet fully
369 * active yet. Some drivers (e.g. fd and loop) get unhappy
370 * in such cases. Kick queue iff dispatch queue has
371 * something on it and @q has request_fn set.
373 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
376 drain
|= q
->nr_rqs_elvpriv
;
377 drain
|= q
->request_fn_active
;
380 * Unfortunately, requests are queued at and tracked from
381 * multiple places and there's no single counter which can
382 * be drained. Check all the queues and counters.
385 drain
|= !list_empty(&q
->queue_head
);
386 for (i
= 0; i
< 2; i
++) {
387 drain
|= q
->nr_rqs
[i
];
388 drain
|= q
->in_flight
[i
];
389 drain
|= !list_empty(&q
->flush_queue
[i
]);
396 spin_unlock_irq(q
->queue_lock
);
400 spin_lock_irq(q
->queue_lock
);
404 * With queue marked dead, any woken up waiter will fail the
405 * allocation path, so the wakeup chaining is lost and we're
406 * left with hung waiters. We need to wake up those waiters.
409 struct request_list
*rl
;
411 blk_queue_for_each_rl(rl
, q
)
412 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
413 wake_up_all(&rl
->wait
[i
]);
418 * blk_queue_bypass_start - enter queue bypass mode
419 * @q: queue of interest
421 * In bypass mode, only the dispatch FIFO queue of @q is used. This
422 * function makes @q enter bypass mode and drains all requests which were
423 * throttled or issued before. On return, it's guaranteed that no request
424 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
425 * inside queue or RCU read lock.
427 void blk_queue_bypass_start(struct request_queue
*q
)
431 spin_lock_irq(q
->queue_lock
);
432 drain
= !q
->bypass_depth
++;
433 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
434 spin_unlock_irq(q
->queue_lock
);
437 spin_lock_irq(q
->queue_lock
);
438 __blk_drain_queue(q
, false);
439 spin_unlock_irq(q
->queue_lock
);
441 /* ensure blk_queue_bypass() is %true inside RCU read lock */
445 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
448 * blk_queue_bypass_end - leave queue bypass mode
449 * @q: queue of interest
451 * Leave bypass mode and restore the normal queueing behavior.
453 void blk_queue_bypass_end(struct request_queue
*q
)
455 spin_lock_irq(q
->queue_lock
);
456 if (!--q
->bypass_depth
)
457 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
458 WARN_ON_ONCE(q
->bypass_depth
< 0);
459 spin_unlock_irq(q
->queue_lock
);
461 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
464 * blk_cleanup_queue - shutdown a request queue
465 * @q: request queue to shutdown
467 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
468 * put it. All future requests will be failed immediately with -ENODEV.
470 void blk_cleanup_queue(struct request_queue
*q
)
472 spinlock_t
*lock
= q
->queue_lock
;
474 /* mark @q DYING, no new request or merges will be allowed afterwards */
475 mutex_lock(&q
->sysfs_lock
);
476 queue_flag_set_unlocked(QUEUE_FLAG_DYING
, q
);
480 * A dying queue is permanently in bypass mode till released. Note
481 * that, unlike blk_queue_bypass_start(), we aren't performing
482 * synchronize_rcu() after entering bypass mode to avoid the delay
483 * as some drivers create and destroy a lot of queues while
484 * probing. This is still safe because blk_release_queue() will be
485 * called only after the queue refcnt drops to zero and nothing,
486 * RCU or not, would be traversing the queue by then.
489 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
491 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
492 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
493 queue_flag_set(QUEUE_FLAG_DYING
, q
);
494 spin_unlock_irq(lock
);
495 mutex_unlock(&q
->sysfs_lock
);
498 * Drain all requests queued before DYING marking. Set DEAD flag to
499 * prevent that q->request_fn() gets invoked after draining finished.
502 blk_mq_drain_queue(q
);
506 __blk_drain_queue(q
, true);
508 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
509 spin_unlock_irq(lock
);
511 /* @q won't process any more request, flush async actions */
512 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
516 if (q
->queue_lock
!= &q
->__queue_lock
)
517 q
->queue_lock
= &q
->__queue_lock
;
518 spin_unlock_irq(lock
);
520 /* @q is and will stay empty, shutdown and put */
523 EXPORT_SYMBOL(blk_cleanup_queue
);
525 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
528 if (unlikely(rl
->rq_pool
))
532 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
533 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
534 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
535 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
537 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
538 mempool_free_slab
, request_cachep
,
546 void blk_exit_rl(struct request_list
*rl
)
549 mempool_destroy(rl
->rq_pool
);
552 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
554 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
556 EXPORT_SYMBOL(blk_alloc_queue
);
558 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
560 struct request_queue
*q
;
563 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
564 gfp_mask
| __GFP_ZERO
, node_id
);
568 if (percpu_counter_init(&q
->mq_usage_counter
, 0))
571 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
575 q
->backing_dev_info
.ra_pages
=
576 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
577 q
->backing_dev_info
.state
= 0;
578 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
579 q
->backing_dev_info
.name
= "block";
582 err
= bdi_init(&q
->backing_dev_info
);
586 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
587 laptop_mode_timer_fn
, (unsigned long) q
);
588 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
589 INIT_LIST_HEAD(&q
->queue_head
);
590 INIT_LIST_HEAD(&q
->timeout_list
);
591 INIT_LIST_HEAD(&q
->icq_list
);
592 #ifdef CONFIG_BLK_CGROUP
593 INIT_LIST_HEAD(&q
->blkg_list
);
595 INIT_LIST_HEAD(&q
->flush_queue
[0]);
596 INIT_LIST_HEAD(&q
->flush_queue
[1]);
597 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
598 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
600 kobject_init(&q
->kobj
, &blk_queue_ktype
);
602 mutex_init(&q
->sysfs_lock
);
603 spin_lock_init(&q
->__queue_lock
);
606 * By default initialize queue_lock to internal lock and driver can
607 * override it later if need be.
609 q
->queue_lock
= &q
->__queue_lock
;
612 * A queue starts its life with bypass turned on to avoid
613 * unnecessary bypass on/off overhead and nasty surprises during
614 * init. The initial bypass will be finished when the queue is
615 * registered by blk_register_queue().
618 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
620 init_waitqueue_head(&q
->mq_freeze_wq
);
622 if (blkcg_init_queue(q
))
628 bdi_destroy(&q
->backing_dev_info
);
630 ida_simple_remove(&blk_queue_ida
, q
->id
);
632 percpu_counter_destroy(&q
->mq_usage_counter
);
634 kmem_cache_free(blk_requestq_cachep
, q
);
637 EXPORT_SYMBOL(blk_alloc_queue_node
);
640 * blk_init_queue - prepare a request queue for use with a block device
641 * @rfn: The function to be called to process requests that have been
642 * placed on the queue.
643 * @lock: Request queue spin lock
646 * If a block device wishes to use the standard request handling procedures,
647 * which sorts requests and coalesces adjacent requests, then it must
648 * call blk_init_queue(). The function @rfn will be called when there
649 * are requests on the queue that need to be processed. If the device
650 * supports plugging, then @rfn may not be called immediately when requests
651 * are available on the queue, but may be called at some time later instead.
652 * Plugged queues are generally unplugged when a buffer belonging to one
653 * of the requests on the queue is needed, or due to memory pressure.
655 * @rfn is not required, or even expected, to remove all requests off the
656 * queue, but only as many as it can handle at a time. If it does leave
657 * requests on the queue, it is responsible for arranging that the requests
658 * get dealt with eventually.
660 * The queue spin lock must be held while manipulating the requests on the
661 * request queue; this lock will be taken also from interrupt context, so irq
662 * disabling is needed for it.
664 * Function returns a pointer to the initialized request queue, or %NULL if
668 * blk_init_queue() must be paired with a blk_cleanup_queue() call
669 * when the block device is deactivated (such as at module unload).
672 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
674 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
676 EXPORT_SYMBOL(blk_init_queue
);
678 struct request_queue
*
679 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
681 struct request_queue
*uninit_q
, *q
;
683 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
687 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
689 blk_cleanup_queue(uninit_q
);
693 EXPORT_SYMBOL(blk_init_queue_node
);
695 struct request_queue
*
696 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
702 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
706 q
->prep_rq_fn
= NULL
;
707 q
->unprep_rq_fn
= NULL
;
708 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
710 /* Override internal queue lock with supplied lock pointer */
712 q
->queue_lock
= lock
;
715 * This also sets hw/phys segments, boundary and size
717 blk_queue_make_request(q
, blk_queue_bio
);
719 q
->sg_reserved_size
= INT_MAX
;
721 /* Protect q->elevator from elevator_change */
722 mutex_lock(&q
->sysfs_lock
);
725 if (elevator_init(q
, NULL
)) {
726 mutex_unlock(&q
->sysfs_lock
);
730 mutex_unlock(&q
->sysfs_lock
);
734 EXPORT_SYMBOL(blk_init_allocated_queue
);
736 bool blk_get_queue(struct request_queue
*q
)
738 if (likely(!blk_queue_dying(q
))) {
745 EXPORT_SYMBOL(blk_get_queue
);
747 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
749 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
750 elv_put_request(rl
->q
, rq
);
752 put_io_context(rq
->elv
.icq
->ioc
);
755 mempool_free(rq
, rl
->rq_pool
);
759 * ioc_batching returns true if the ioc is a valid batching request and
760 * should be given priority access to a request.
762 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
768 * Make sure the process is able to allocate at least 1 request
769 * even if the batch times out, otherwise we could theoretically
772 return ioc
->nr_batch_requests
== q
->nr_batching
||
773 (ioc
->nr_batch_requests
> 0
774 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
778 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
779 * will cause the process to be a "batcher" on all queues in the system. This
780 * is the behaviour we want though - once it gets a wakeup it should be given
783 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
785 if (!ioc
|| ioc_batching(q
, ioc
))
788 ioc
->nr_batch_requests
= q
->nr_batching
;
789 ioc
->last_waited
= jiffies
;
792 static void __freed_request(struct request_list
*rl
, int sync
)
794 struct request_queue
*q
= rl
->q
;
797 * bdi isn't aware of blkcg yet. As all async IOs end up root
798 * blkcg anyway, just use root blkcg state.
800 if (rl
== &q
->root_rl
&&
801 rl
->count
[sync
] < queue_congestion_off_threshold(q
))
802 blk_clear_queue_congested(q
, sync
);
804 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
805 if (waitqueue_active(&rl
->wait
[sync
]))
806 wake_up(&rl
->wait
[sync
]);
808 blk_clear_rl_full(rl
, sync
);
813 * A request has just been released. Account for it, update the full and
814 * congestion status, wake up any waiters. Called under q->queue_lock.
816 static void freed_request(struct request_list
*rl
, unsigned int flags
)
818 struct request_queue
*q
= rl
->q
;
819 int sync
= rw_is_sync(flags
);
823 if (flags
& REQ_ELVPRIV
)
826 __freed_request(rl
, sync
);
828 if (unlikely(rl
->starved
[sync
^ 1]))
829 __freed_request(rl
, sync
^ 1);
833 * Determine if elevator data should be initialized when allocating the
834 * request associated with @bio.
836 static bool blk_rq_should_init_elevator(struct bio
*bio
)
842 * Flush requests do not use the elevator so skip initialization.
843 * This allows a request to share the flush and elevator data.
845 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
852 * rq_ioc - determine io_context for request allocation
853 * @bio: request being allocated is for this bio (can be %NULL)
855 * Determine io_context to use for request allocation for @bio. May return
856 * %NULL if %current->io_context doesn't exist.
858 static struct io_context
*rq_ioc(struct bio
*bio
)
860 #ifdef CONFIG_BLK_CGROUP
861 if (bio
&& bio
->bi_ioc
)
864 return current
->io_context
;
868 * __get_request - get a free request
869 * @rl: request list to allocate from
870 * @rw_flags: RW and SYNC flags
871 * @bio: bio to allocate request for (can be %NULL)
872 * @gfp_mask: allocation mask
874 * Get a free request from @q. This function may fail under memory
875 * pressure or if @q is dead.
877 * Must be callled with @q->queue_lock held and,
878 * Returns %NULL on failure, with @q->queue_lock held.
879 * Returns !%NULL on success, with @q->queue_lock *not held*.
881 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
882 struct bio
*bio
, gfp_t gfp_mask
)
884 struct request_queue
*q
= rl
->q
;
886 struct elevator_type
*et
= q
->elevator
->type
;
887 struct io_context
*ioc
= rq_ioc(bio
);
888 struct io_cq
*icq
= NULL
;
889 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
892 if (unlikely(blk_queue_dying(q
)))
895 may_queue
= elv_may_queue(q
, rw_flags
);
896 if (may_queue
== ELV_MQUEUE_NO
)
899 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
900 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
902 * The queue will fill after this allocation, so set
903 * it as full, and mark this process as "batching".
904 * This process will be allowed to complete a batch of
905 * requests, others will be blocked.
907 if (!blk_rl_full(rl
, is_sync
)) {
908 ioc_set_batching(q
, ioc
);
909 blk_set_rl_full(rl
, is_sync
);
911 if (may_queue
!= ELV_MQUEUE_MUST
912 && !ioc_batching(q
, ioc
)) {
914 * The queue is full and the allocating
915 * process is not a "batcher", and not
916 * exempted by the IO scheduler
923 * bdi isn't aware of blkcg yet. As all async IOs end up
924 * root blkcg anyway, just use root blkcg state.
926 if (rl
== &q
->root_rl
)
927 blk_set_queue_congested(q
, is_sync
);
931 * Only allow batching queuers to allocate up to 50% over the defined
932 * limit of requests, otherwise we could have thousands of requests
933 * allocated with any setting of ->nr_requests
935 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
938 q
->nr_rqs
[is_sync
]++;
939 rl
->count
[is_sync
]++;
940 rl
->starved
[is_sync
] = 0;
943 * Decide whether the new request will be managed by elevator. If
944 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
945 * prevent the current elevator from being destroyed until the new
946 * request is freed. This guarantees icq's won't be destroyed and
947 * makes creating new ones safe.
949 * Also, lookup icq while holding queue_lock. If it doesn't exist,
950 * it will be created after releasing queue_lock.
952 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
953 rw_flags
|= REQ_ELVPRIV
;
955 if (et
->icq_cache
&& ioc
)
956 icq
= ioc_lookup_icq(ioc
, q
);
959 if (blk_queue_io_stat(q
))
960 rw_flags
|= REQ_IO_STAT
;
961 spin_unlock_irq(q
->queue_lock
);
963 /* allocate and init request */
964 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
969 blk_rq_set_rl(rq
, rl
);
970 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
973 if (rw_flags
& REQ_ELVPRIV
) {
974 if (unlikely(et
->icq_cache
&& !icq
)) {
976 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
982 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
985 /* @rq->elv.icq holds io_context until @rq is freed */
987 get_io_context(icq
->ioc
);
991 * ioc may be NULL here, and ioc_batching will be false. That's
992 * OK, if the queue is under the request limit then requests need
993 * not count toward the nr_batch_requests limit. There will always
994 * be some limit enforced by BLK_BATCH_TIME.
996 if (ioc_batching(q
, ioc
))
997 ioc
->nr_batch_requests
--;
999 trace_block_getrq(q
, bio
, rw_flags
& 1);
1004 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1005 * and may fail indefinitely under memory pressure and thus
1006 * shouldn't stall IO. Treat this request as !elvpriv. This will
1007 * disturb iosched and blkcg but weird is bettern than dead.
1009 printk_ratelimited(KERN_WARNING
"%s: request aux data allocation failed, iosched may be disturbed\n",
1010 dev_name(q
->backing_dev_info
.dev
));
1012 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1015 spin_lock_irq(q
->queue_lock
);
1016 q
->nr_rqs_elvpriv
--;
1017 spin_unlock_irq(q
->queue_lock
);
1022 * Allocation failed presumably due to memory. Undo anything we
1023 * might have messed up.
1025 * Allocating task should really be put onto the front of the wait
1026 * queue, but this is pretty rare.
1028 spin_lock_irq(q
->queue_lock
);
1029 freed_request(rl
, rw_flags
);
1032 * in the very unlikely event that allocation failed and no
1033 * requests for this direction was pending, mark us starved so that
1034 * freeing of a request in the other direction will notice
1035 * us. another possible fix would be to split the rq mempool into
1039 if (unlikely(rl
->count
[is_sync
] == 0))
1040 rl
->starved
[is_sync
] = 1;
1045 * get_request - get a free request
1046 * @q: request_queue to allocate request from
1047 * @rw_flags: RW and SYNC flags
1048 * @bio: bio to allocate request for (can be %NULL)
1049 * @gfp_mask: allocation mask
1051 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1052 * function keeps retrying under memory pressure and fails iff @q is dead.
1054 * Must be callled with @q->queue_lock held and,
1055 * Returns %NULL on failure, with @q->queue_lock held.
1056 * Returns !%NULL on success, with @q->queue_lock *not held*.
1058 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1059 struct bio
*bio
, gfp_t gfp_mask
)
1061 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1063 struct request_list
*rl
;
1066 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1068 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1072 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dying(q
))) {
1077 /* wait on @rl and retry */
1078 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1079 TASK_UNINTERRUPTIBLE
);
1081 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1083 spin_unlock_irq(q
->queue_lock
);
1087 * After sleeping, we become a "batching" process and will be able
1088 * to allocate at least one request, and up to a big batch of them
1089 * for a small period time. See ioc_batching, ioc_set_batching
1091 ioc_set_batching(q
, current
->io_context
);
1093 spin_lock_irq(q
->queue_lock
);
1094 finish_wait(&rl
->wait
[is_sync
], &wait
);
1099 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1104 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1106 /* create ioc upfront */
1107 create_io_context(gfp_mask
, q
->node
);
1109 spin_lock_irq(q
->queue_lock
);
1110 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1112 spin_unlock_irq(q
->queue_lock
);
1113 /* q->queue_lock is unlocked at this point */
1118 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1121 return blk_mq_alloc_request(q
, rw
, gfp_mask
, false);
1123 return blk_old_get_request(q
, rw
, gfp_mask
);
1125 EXPORT_SYMBOL(blk_get_request
);
1128 * blk_make_request - given a bio, allocate a corresponding struct request.
1129 * @q: target request queue
1130 * @bio: The bio describing the memory mappings that will be submitted for IO.
1131 * It may be a chained-bio properly constructed by block/bio layer.
1132 * @gfp_mask: gfp flags to be used for memory allocation
1134 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1135 * type commands. Where the struct request needs to be farther initialized by
1136 * the caller. It is passed a &struct bio, which describes the memory info of
1139 * The caller of blk_make_request must make sure that bi_io_vec
1140 * are set to describe the memory buffers. That bio_data_dir() will return
1141 * the needed direction of the request. (And all bio's in the passed bio-chain
1142 * are properly set accordingly)
1144 * If called under none-sleepable conditions, mapped bio buffers must not
1145 * need bouncing, by calling the appropriate masked or flagged allocator,
1146 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1149 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1150 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1151 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1152 * completion of a bio that hasn't been submitted yet, thus resulting in a
1153 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1154 * of bio_alloc(), as that avoids the mempool deadlock.
1155 * If possible a big IO should be split into smaller parts when allocation
1156 * fails. Partial allocation should not be an error, or you risk a live-lock.
1158 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1161 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1164 return ERR_PTR(-ENOMEM
);
1167 struct bio
*bounce_bio
= bio
;
1170 blk_queue_bounce(q
, &bounce_bio
);
1171 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1172 if (unlikely(ret
)) {
1173 blk_put_request(rq
);
1174 return ERR_PTR(ret
);
1180 EXPORT_SYMBOL(blk_make_request
);
1183 * blk_requeue_request - put a request back on queue
1184 * @q: request queue where request should be inserted
1185 * @rq: request to be inserted
1188 * Drivers often keep queueing requests until the hardware cannot accept
1189 * more, when that condition happens we need to put the request back
1190 * on the queue. Must be called with queue lock held.
1192 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1194 blk_delete_timer(rq
);
1195 blk_clear_rq_complete(rq
);
1196 trace_block_rq_requeue(q
, rq
);
1198 if (blk_rq_tagged(rq
))
1199 blk_queue_end_tag(q
, rq
);
1201 BUG_ON(blk_queued_rq(rq
));
1203 elv_requeue_request(q
, rq
);
1205 EXPORT_SYMBOL(blk_requeue_request
);
1207 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1210 blk_account_io_start(rq
, true);
1211 __elv_add_request(q
, rq
, where
);
1214 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1217 if (now
== part
->stamp
)
1220 if (part_in_flight(part
)) {
1221 __part_stat_add(cpu
, part
, time_in_queue
,
1222 part_in_flight(part
) * (now
- part
->stamp
));
1223 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1229 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1230 * @cpu: cpu number for stats access
1231 * @part: target partition
1233 * The average IO queue length and utilisation statistics are maintained
1234 * by observing the current state of the queue length and the amount of
1235 * time it has been in this state for.
1237 * Normally, that accounting is done on IO completion, but that can result
1238 * in more than a second's worth of IO being accounted for within any one
1239 * second, leading to >100% utilisation. To deal with that, we call this
1240 * function to do a round-off before returning the results when reading
1241 * /proc/diskstats. This accounts immediately for all queue usage up to
1242 * the current jiffies and restarts the counters again.
1244 void part_round_stats(int cpu
, struct hd_struct
*part
)
1246 unsigned long now
= jiffies
;
1249 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1250 part_round_stats_single(cpu
, part
, now
);
1252 EXPORT_SYMBOL_GPL(part_round_stats
);
1254 #ifdef CONFIG_PM_RUNTIME
1255 static void blk_pm_put_request(struct request
*rq
)
1257 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1258 pm_runtime_mark_last_busy(rq
->q
->dev
);
1261 static inline void blk_pm_put_request(struct request
*rq
) {}
1265 * queue lock must be held
1267 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1272 blk_pm_put_request(req
);
1274 elv_completed_request(q
, req
);
1276 /* this is a bio leak */
1277 WARN_ON(req
->bio
!= NULL
);
1280 * Request may not have originated from ll_rw_blk. if not,
1281 * it didn't come out of our reserved rq pools
1283 if (req
->cmd_flags
& REQ_ALLOCED
) {
1284 unsigned int flags
= req
->cmd_flags
;
1285 struct request_list
*rl
= blk_rq_rl(req
);
1287 BUG_ON(!list_empty(&req
->queuelist
));
1288 BUG_ON(!hlist_unhashed(&req
->hash
));
1290 blk_free_request(rl
, req
);
1291 freed_request(rl
, flags
);
1295 EXPORT_SYMBOL_GPL(__blk_put_request
);
1297 void blk_put_request(struct request
*req
)
1299 struct request_queue
*q
= req
->q
;
1302 blk_mq_free_request(req
);
1304 unsigned long flags
;
1306 spin_lock_irqsave(q
->queue_lock
, flags
);
1307 __blk_put_request(q
, req
);
1308 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1311 EXPORT_SYMBOL(blk_put_request
);
1314 * blk_add_request_payload - add a payload to a request
1315 * @rq: request to update
1316 * @page: page backing the payload
1317 * @len: length of the payload.
1319 * This allows to later add a payload to an already submitted request by
1320 * a block driver. The driver needs to take care of freeing the payload
1323 * Note that this is a quite horrible hack and nothing but handling of
1324 * discard requests should ever use it.
1326 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1329 struct bio
*bio
= rq
->bio
;
1331 bio
->bi_io_vec
->bv_page
= page
;
1332 bio
->bi_io_vec
->bv_offset
= 0;
1333 bio
->bi_io_vec
->bv_len
= len
;
1335 bio
->bi_iter
.bi_size
= len
;
1337 bio
->bi_phys_segments
= 1;
1339 rq
->__data_len
= rq
->resid_len
= len
;
1340 rq
->nr_phys_segments
= 1;
1341 rq
->buffer
= bio_data(bio
);
1343 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1345 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1348 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1350 if (!ll_back_merge_fn(q
, req
, bio
))
1353 trace_block_bio_backmerge(q
, req
, bio
);
1355 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1356 blk_rq_set_mixed_merge(req
);
1358 req
->biotail
->bi_next
= bio
;
1360 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1361 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1363 blk_account_io_start(req
, false);
1367 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1370 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1372 if (!ll_front_merge_fn(q
, req
, bio
))
1375 trace_block_bio_frontmerge(q
, req
, bio
);
1377 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1378 blk_rq_set_mixed_merge(req
);
1380 bio
->bi_next
= req
->bio
;
1384 * may not be valid. if the low level driver said
1385 * it didn't need a bounce buffer then it better
1386 * not touch req->buffer either...
1388 req
->buffer
= bio_data(bio
);
1389 req
->__sector
= bio
->bi_iter
.bi_sector
;
1390 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1391 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1393 blk_account_io_start(req
, false);
1398 * blk_attempt_plug_merge - try to merge with %current's plugged list
1399 * @q: request_queue new bio is being queued at
1400 * @bio: new bio being queued
1401 * @request_count: out parameter for number of traversed plugged requests
1403 * Determine whether @bio being queued on @q can be merged with a request
1404 * on %current's plugged list. Returns %true if merge was successful,
1407 * Plugging coalesces IOs from the same issuer for the same purpose without
1408 * going through @q->queue_lock. As such it's more of an issuing mechanism
1409 * than scheduling, and the request, while may have elvpriv data, is not
1410 * added on the elevator at this point. In addition, we don't have
1411 * reliable access to the elevator outside queue lock. Only check basic
1412 * merging parameters without querying the elevator.
1414 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1415 unsigned int *request_count
)
1417 struct blk_plug
*plug
;
1420 struct list_head
*plug_list
;
1422 if (blk_queue_nomerges(q
))
1425 plug
= current
->plug
;
1431 plug_list
= &plug
->mq_list
;
1433 plug_list
= &plug
->list
;
1435 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1441 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1444 el_ret
= blk_try_merge(rq
, bio
);
1445 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1446 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1449 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1450 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1459 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1461 req
->cmd_type
= REQ_TYPE_FS
;
1463 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1464 if (bio
->bi_rw
& REQ_RAHEAD
)
1465 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1468 req
->__sector
= bio
->bi_iter
.bi_sector
;
1469 req
->ioprio
= bio_prio(bio
);
1470 blk_rq_bio_prep(req
->q
, req
, bio
);
1473 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1475 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1476 struct blk_plug
*plug
;
1477 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1478 struct request
*req
;
1479 unsigned int request_count
= 0;
1482 * low level driver can indicate that it wants pages above a
1483 * certain limit bounced to low memory (ie for highmem, or even
1484 * ISA dma in theory)
1486 blk_queue_bounce(q
, &bio
);
1488 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1489 bio_endio(bio
, -EIO
);
1493 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1494 spin_lock_irq(q
->queue_lock
);
1495 where
= ELEVATOR_INSERT_FLUSH
;
1500 * Check if we can merge with the plugged list before grabbing
1503 if (blk_attempt_plug_merge(q
, bio
, &request_count
))
1506 spin_lock_irq(q
->queue_lock
);
1508 el_ret
= elv_merge(q
, &req
, bio
);
1509 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1510 if (bio_attempt_back_merge(q
, req
, bio
)) {
1511 elv_bio_merged(q
, req
, bio
);
1512 if (!attempt_back_merge(q
, req
))
1513 elv_merged_request(q
, req
, el_ret
);
1516 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1517 if (bio_attempt_front_merge(q
, req
, bio
)) {
1518 elv_bio_merged(q
, req
, bio
);
1519 if (!attempt_front_merge(q
, req
))
1520 elv_merged_request(q
, req
, el_ret
);
1527 * This sync check and mask will be re-done in init_request_from_bio(),
1528 * but we need to set it earlier to expose the sync flag to the
1529 * rq allocator and io schedulers.
1531 rw_flags
= bio_data_dir(bio
);
1533 rw_flags
|= REQ_SYNC
;
1536 * Grab a free request. This is might sleep but can not fail.
1537 * Returns with the queue unlocked.
1539 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1540 if (unlikely(!req
)) {
1541 bio_endio(bio
, -ENODEV
); /* @q is dead */
1546 * After dropping the lock and possibly sleeping here, our request
1547 * may now be mergeable after it had proven unmergeable (above).
1548 * We don't worry about that case for efficiency. It won't happen
1549 * often, and the elevators are able to handle it.
1551 init_request_from_bio(req
, bio
);
1553 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1554 req
->cpu
= raw_smp_processor_id();
1556 plug
= current
->plug
;
1559 * If this is the first request added after a plug, fire
1563 trace_block_plug(q
);
1565 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1566 blk_flush_plug_list(plug
, false);
1567 trace_block_plug(q
);
1570 list_add_tail(&req
->queuelist
, &plug
->list
);
1571 blk_account_io_start(req
, true);
1573 spin_lock_irq(q
->queue_lock
);
1574 add_acct_request(q
, req
, where
);
1577 spin_unlock_irq(q
->queue_lock
);
1580 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1583 * If bio->bi_dev is a partition, remap the location
1585 static inline void blk_partition_remap(struct bio
*bio
)
1587 struct block_device
*bdev
= bio
->bi_bdev
;
1589 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1590 struct hd_struct
*p
= bdev
->bd_part
;
1592 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1593 bio
->bi_bdev
= bdev
->bd_contains
;
1595 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1597 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1601 static void handle_bad_sector(struct bio
*bio
)
1603 char b
[BDEVNAME_SIZE
];
1605 printk(KERN_INFO
"attempt to access beyond end of device\n");
1606 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1607 bdevname(bio
->bi_bdev
, b
),
1609 (unsigned long long)bio_end_sector(bio
),
1610 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1612 set_bit(BIO_EOF
, &bio
->bi_flags
);
1615 #ifdef CONFIG_FAIL_MAKE_REQUEST
1617 static DECLARE_FAULT_ATTR(fail_make_request
);
1619 static int __init
setup_fail_make_request(char *str
)
1621 return setup_fault_attr(&fail_make_request
, str
);
1623 __setup("fail_make_request=", setup_fail_make_request
);
1625 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1627 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1630 static int __init
fail_make_request_debugfs(void)
1632 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1633 NULL
, &fail_make_request
);
1635 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1638 late_initcall(fail_make_request_debugfs
);
1640 #else /* CONFIG_FAIL_MAKE_REQUEST */
1642 static inline bool should_fail_request(struct hd_struct
*part
,
1648 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1651 * Check whether this bio extends beyond the end of the device.
1653 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1660 /* Test device or partition size, when known. */
1661 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1663 sector_t sector
= bio
->bi_iter
.bi_sector
;
1665 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1667 * This may well happen - the kernel calls bread()
1668 * without checking the size of the device, e.g., when
1669 * mounting a device.
1671 handle_bad_sector(bio
);
1679 static noinline_for_stack
bool
1680 generic_make_request_checks(struct bio
*bio
)
1682 struct request_queue
*q
;
1683 int nr_sectors
= bio_sectors(bio
);
1685 char b
[BDEVNAME_SIZE
];
1686 struct hd_struct
*part
;
1690 if (bio_check_eod(bio
, nr_sectors
))
1693 q
= bdev_get_queue(bio
->bi_bdev
);
1696 "generic_make_request: Trying to access "
1697 "nonexistent block-device %s (%Lu)\n",
1698 bdevname(bio
->bi_bdev
, b
),
1699 (long long) bio
->bi_iter
.bi_sector
);
1703 if (likely(bio_is_rw(bio
) &&
1704 nr_sectors
> queue_max_hw_sectors(q
))) {
1705 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1706 bdevname(bio
->bi_bdev
, b
),
1708 queue_max_hw_sectors(q
));
1712 part
= bio
->bi_bdev
->bd_part
;
1713 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1714 should_fail_request(&part_to_disk(part
)->part0
,
1715 bio
->bi_iter
.bi_size
))
1719 * If this device has partitions, remap block n
1720 * of partition p to block n+start(p) of the disk.
1722 blk_partition_remap(bio
);
1724 if (bio_check_eod(bio
, nr_sectors
))
1728 * Filter flush bio's early so that make_request based
1729 * drivers without flush support don't have to worry
1732 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1733 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1740 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1741 (!blk_queue_discard(q
) ||
1742 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1747 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1753 * Various block parts want %current->io_context and lazy ioc
1754 * allocation ends up trading a lot of pain for a small amount of
1755 * memory. Just allocate it upfront. This may fail and block
1756 * layer knows how to live with it.
1758 create_io_context(GFP_ATOMIC
, q
->node
);
1760 if (blk_throtl_bio(q
, bio
))
1761 return false; /* throttled, will be resubmitted later */
1763 trace_block_bio_queue(q
, bio
);
1767 bio_endio(bio
, err
);
1772 * generic_make_request - hand a buffer to its device driver for I/O
1773 * @bio: The bio describing the location in memory and on the device.
1775 * generic_make_request() is used to make I/O requests of block
1776 * devices. It is passed a &struct bio, which describes the I/O that needs
1779 * generic_make_request() does not return any status. The
1780 * success/failure status of the request, along with notification of
1781 * completion, is delivered asynchronously through the bio->bi_end_io
1782 * function described (one day) else where.
1784 * The caller of generic_make_request must make sure that bi_io_vec
1785 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1786 * set to describe the device address, and the
1787 * bi_end_io and optionally bi_private are set to describe how
1788 * completion notification should be signaled.
1790 * generic_make_request and the drivers it calls may use bi_next if this
1791 * bio happens to be merged with someone else, and may resubmit the bio to
1792 * a lower device by calling into generic_make_request recursively, which
1793 * means the bio should NOT be touched after the call to ->make_request_fn.
1795 void generic_make_request(struct bio
*bio
)
1797 struct bio_list bio_list_on_stack
;
1799 if (!generic_make_request_checks(bio
))
1803 * We only want one ->make_request_fn to be active at a time, else
1804 * stack usage with stacked devices could be a problem. So use
1805 * current->bio_list to keep a list of requests submited by a
1806 * make_request_fn function. current->bio_list is also used as a
1807 * flag to say if generic_make_request is currently active in this
1808 * task or not. If it is NULL, then no make_request is active. If
1809 * it is non-NULL, then a make_request is active, and new requests
1810 * should be added at the tail
1812 if (current
->bio_list
) {
1813 bio_list_add(current
->bio_list
, bio
);
1817 /* following loop may be a bit non-obvious, and so deserves some
1819 * Before entering the loop, bio->bi_next is NULL (as all callers
1820 * ensure that) so we have a list with a single bio.
1821 * We pretend that we have just taken it off a longer list, so
1822 * we assign bio_list to a pointer to the bio_list_on_stack,
1823 * thus initialising the bio_list of new bios to be
1824 * added. ->make_request() may indeed add some more bios
1825 * through a recursive call to generic_make_request. If it
1826 * did, we find a non-NULL value in bio_list and re-enter the loop
1827 * from the top. In this case we really did just take the bio
1828 * of the top of the list (no pretending) and so remove it from
1829 * bio_list, and call into ->make_request() again.
1831 BUG_ON(bio
->bi_next
);
1832 bio_list_init(&bio_list_on_stack
);
1833 current
->bio_list
= &bio_list_on_stack
;
1835 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1837 q
->make_request_fn(q
, bio
);
1839 bio
= bio_list_pop(current
->bio_list
);
1841 current
->bio_list
= NULL
; /* deactivate */
1843 EXPORT_SYMBOL(generic_make_request
);
1846 * submit_bio - submit a bio to the block device layer for I/O
1847 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1848 * @bio: The &struct bio which describes the I/O
1850 * submit_bio() is very similar in purpose to generic_make_request(), and
1851 * uses that function to do most of the work. Both are fairly rough
1852 * interfaces; @bio must be presetup and ready for I/O.
1855 void submit_bio(int rw
, struct bio
*bio
)
1860 * If it's a regular read/write or a barrier with data attached,
1861 * go through the normal accounting stuff before submission.
1863 if (bio_has_data(bio
)) {
1866 if (unlikely(rw
& REQ_WRITE_SAME
))
1867 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
1869 count
= bio_sectors(bio
);
1872 count_vm_events(PGPGOUT
, count
);
1874 task_io_account_read(bio
->bi_iter
.bi_size
);
1875 count_vm_events(PGPGIN
, count
);
1878 if (unlikely(block_dump
)) {
1879 char b
[BDEVNAME_SIZE
];
1880 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1881 current
->comm
, task_pid_nr(current
),
1882 (rw
& WRITE
) ? "WRITE" : "READ",
1883 (unsigned long long)bio
->bi_iter
.bi_sector
,
1884 bdevname(bio
->bi_bdev
, b
),
1889 generic_make_request(bio
);
1891 EXPORT_SYMBOL(submit_bio
);
1894 * blk_rq_check_limits - Helper function to check a request for the queue limit
1896 * @rq: the request being checked
1899 * @rq may have been made based on weaker limitations of upper-level queues
1900 * in request stacking drivers, and it may violate the limitation of @q.
1901 * Since the block layer and the underlying device driver trust @rq
1902 * after it is inserted to @q, it should be checked against @q before
1903 * the insertion using this generic function.
1905 * This function should also be useful for request stacking drivers
1906 * in some cases below, so export this function.
1907 * Request stacking drivers like request-based dm may change the queue
1908 * limits while requests are in the queue (e.g. dm's table swapping).
1909 * Such request stacking drivers should check those requests agaist
1910 * the new queue limits again when they dispatch those requests,
1911 * although such checkings are also done against the old queue limits
1912 * when submitting requests.
1914 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1916 if (!rq_mergeable(rq
))
1919 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
1920 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1925 * queue's settings related to segment counting like q->bounce_pfn
1926 * may differ from that of other stacking queues.
1927 * Recalculate it to check the request correctly on this queue's
1930 blk_recalc_rq_segments(rq
);
1931 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1932 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1938 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1941 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1942 * @q: the queue to submit the request
1943 * @rq: the request being queued
1945 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1947 unsigned long flags
;
1948 int where
= ELEVATOR_INSERT_BACK
;
1950 if (blk_rq_check_limits(q
, rq
))
1954 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1957 spin_lock_irqsave(q
->queue_lock
, flags
);
1958 if (unlikely(blk_queue_dying(q
))) {
1959 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1964 * Submitting request must be dequeued before calling this function
1965 * because it will be linked to another request_queue
1967 BUG_ON(blk_queued_rq(rq
));
1969 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1970 where
= ELEVATOR_INSERT_FLUSH
;
1972 add_acct_request(q
, rq
, where
);
1973 if (where
== ELEVATOR_INSERT_FLUSH
)
1975 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1979 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1982 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1983 * @rq: request to examine
1986 * A request could be merge of IOs which require different failure
1987 * handling. This function determines the number of bytes which
1988 * can be failed from the beginning of the request without
1989 * crossing into area which need to be retried further.
1992 * The number of bytes to fail.
1995 * queue_lock must be held.
1997 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1999 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2000 unsigned int bytes
= 0;
2003 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2004 return blk_rq_bytes(rq
);
2007 * Currently the only 'mixing' which can happen is between
2008 * different fastfail types. We can safely fail portions
2009 * which have all the failfast bits that the first one has -
2010 * the ones which are at least as eager to fail as the first
2013 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2014 if ((bio
->bi_rw
& ff
) != ff
)
2016 bytes
+= bio
->bi_iter
.bi_size
;
2019 /* this could lead to infinite loop */
2020 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2023 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2025 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2027 if (blk_do_io_stat(req
)) {
2028 const int rw
= rq_data_dir(req
);
2029 struct hd_struct
*part
;
2032 cpu
= part_stat_lock();
2034 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2039 void blk_account_io_done(struct request
*req
)
2042 * Account IO completion. flush_rq isn't accounted as a
2043 * normal IO on queueing nor completion. Accounting the
2044 * containing request is enough.
2046 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2047 unsigned long duration
= jiffies
- req
->start_time
;
2048 const int rw
= rq_data_dir(req
);
2049 struct hd_struct
*part
;
2052 cpu
= part_stat_lock();
2055 part_stat_inc(cpu
, part
, ios
[rw
]);
2056 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2057 part_round_stats(cpu
, part
);
2058 part_dec_in_flight(part
, rw
);
2060 hd_struct_put(part
);
2065 #ifdef CONFIG_PM_RUNTIME
2067 * Don't process normal requests when queue is suspended
2068 * or in the process of suspending/resuming
2070 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2073 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2074 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2080 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2087 void blk_account_io_start(struct request
*rq
, bool new_io
)
2089 struct hd_struct
*part
;
2090 int rw
= rq_data_dir(rq
);
2093 if (!blk_do_io_stat(rq
))
2096 cpu
= part_stat_lock();
2100 part_stat_inc(cpu
, part
, merges
[rw
]);
2102 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2103 if (!hd_struct_try_get(part
)) {
2105 * The partition is already being removed,
2106 * the request will be accounted on the disk only
2108 * We take a reference on disk->part0 although that
2109 * partition will never be deleted, so we can treat
2110 * it as any other partition.
2112 part
= &rq
->rq_disk
->part0
;
2113 hd_struct_get(part
);
2115 part_round_stats(cpu
, part
);
2116 part_inc_in_flight(part
, rw
);
2124 * blk_peek_request - peek at the top of a request queue
2125 * @q: request queue to peek at
2128 * Return the request at the top of @q. The returned request
2129 * should be started using blk_start_request() before LLD starts
2133 * Pointer to the request at the top of @q if available. Null
2137 * queue_lock must be held.
2139 struct request
*blk_peek_request(struct request_queue
*q
)
2144 while ((rq
= __elv_next_request(q
)) != NULL
) {
2146 rq
= blk_pm_peek_request(q
, rq
);
2150 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2152 * This is the first time the device driver
2153 * sees this request (possibly after
2154 * requeueing). Notify IO scheduler.
2156 if (rq
->cmd_flags
& REQ_SORTED
)
2157 elv_activate_rq(q
, rq
);
2160 * just mark as started even if we don't start
2161 * it, a request that has been delayed should
2162 * not be passed by new incoming requests
2164 rq
->cmd_flags
|= REQ_STARTED
;
2165 trace_block_rq_issue(q
, rq
);
2168 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2169 q
->end_sector
= rq_end_sector(rq
);
2170 q
->boundary_rq
= NULL
;
2173 if (rq
->cmd_flags
& REQ_DONTPREP
)
2176 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2178 * make sure space for the drain appears we
2179 * know we can do this because max_hw_segments
2180 * has been adjusted to be one fewer than the
2183 rq
->nr_phys_segments
++;
2189 ret
= q
->prep_rq_fn(q
, rq
);
2190 if (ret
== BLKPREP_OK
) {
2192 } else if (ret
== BLKPREP_DEFER
) {
2194 * the request may have been (partially) prepped.
2195 * we need to keep this request in the front to
2196 * avoid resource deadlock. REQ_STARTED will
2197 * prevent other fs requests from passing this one.
2199 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2200 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2202 * remove the space for the drain we added
2203 * so that we don't add it again
2205 --rq
->nr_phys_segments
;
2210 } else if (ret
== BLKPREP_KILL
) {
2211 rq
->cmd_flags
|= REQ_QUIET
;
2213 * Mark this request as started so we don't trigger
2214 * any debug logic in the end I/O path.
2216 blk_start_request(rq
);
2217 __blk_end_request_all(rq
, -EIO
);
2219 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2226 EXPORT_SYMBOL(blk_peek_request
);
2228 void blk_dequeue_request(struct request
*rq
)
2230 struct request_queue
*q
= rq
->q
;
2232 BUG_ON(list_empty(&rq
->queuelist
));
2233 BUG_ON(ELV_ON_HASH(rq
));
2235 list_del_init(&rq
->queuelist
);
2238 * the time frame between a request being removed from the lists
2239 * and to it is freed is accounted as io that is in progress at
2242 if (blk_account_rq(rq
)) {
2243 q
->in_flight
[rq_is_sync(rq
)]++;
2244 set_io_start_time_ns(rq
);
2249 * blk_start_request - start request processing on the driver
2250 * @req: request to dequeue
2253 * Dequeue @req and start timeout timer on it. This hands off the
2254 * request to the driver.
2256 * Block internal functions which don't want to start timer should
2257 * call blk_dequeue_request().
2260 * queue_lock must be held.
2262 void blk_start_request(struct request
*req
)
2264 blk_dequeue_request(req
);
2267 * We are now handing the request to the hardware, initialize
2268 * resid_len to full count and add the timeout handler.
2270 req
->resid_len
= blk_rq_bytes(req
);
2271 if (unlikely(blk_bidi_rq(req
)))
2272 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2274 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2277 EXPORT_SYMBOL(blk_start_request
);
2280 * blk_fetch_request - fetch a request from a request queue
2281 * @q: request queue to fetch a request from
2284 * Return the request at the top of @q. The request is started on
2285 * return and LLD can start processing it immediately.
2288 * Pointer to the request at the top of @q if available. Null
2292 * queue_lock must be held.
2294 struct request
*blk_fetch_request(struct request_queue
*q
)
2298 rq
= blk_peek_request(q
);
2300 blk_start_request(rq
);
2303 EXPORT_SYMBOL(blk_fetch_request
);
2306 * blk_update_request - Special helper function for request stacking drivers
2307 * @req: the request being processed
2308 * @error: %0 for success, < %0 for error
2309 * @nr_bytes: number of bytes to complete @req
2312 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2313 * the request structure even if @req doesn't have leftover.
2314 * If @req has leftover, sets it up for the next range of segments.
2316 * This special helper function is only for request stacking drivers
2317 * (e.g. request-based dm) so that they can handle partial completion.
2318 * Actual device drivers should use blk_end_request instead.
2320 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2321 * %false return from this function.
2324 * %false - this request doesn't have any more data
2325 * %true - this request has more data
2327 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2334 trace_block_rq_complete(req
->q
, req
);
2337 * For fs requests, rq is just carrier of independent bio's
2338 * and each partial completion should be handled separately.
2339 * Reset per-request error on each partial completion.
2341 * TODO: tj: This is too subtle. It would be better to let
2342 * low level drivers do what they see fit.
2344 if (req
->cmd_type
== REQ_TYPE_FS
)
2347 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2348 !(req
->cmd_flags
& REQ_QUIET
)) {
2353 error_type
= "recoverable transport";
2356 error_type
= "critical target";
2359 error_type
= "critical nexus";
2362 error_type
= "timeout";
2365 error_type
= "critical space allocation";
2368 error_type
= "critical medium";
2375 printk_ratelimited(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2376 error_type
, req
->rq_disk
?
2377 req
->rq_disk
->disk_name
: "?",
2378 (unsigned long long)blk_rq_pos(req
));
2382 blk_account_io_completion(req
, nr_bytes
);
2386 struct bio
*bio
= req
->bio
;
2387 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2389 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2390 req
->bio
= bio
->bi_next
;
2392 req_bio_endio(req
, bio
, bio_bytes
, error
);
2394 total_bytes
+= bio_bytes
;
2395 nr_bytes
-= bio_bytes
;
2406 * Reset counters so that the request stacking driver
2407 * can find how many bytes remain in the request
2410 req
->__data_len
= 0;
2414 req
->__data_len
-= total_bytes
;
2415 req
->buffer
= bio_data(req
->bio
);
2417 /* update sector only for requests with clear definition of sector */
2418 if (req
->cmd_type
== REQ_TYPE_FS
)
2419 req
->__sector
+= total_bytes
>> 9;
2421 /* mixed attributes always follow the first bio */
2422 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2423 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2424 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2428 * If total number of sectors is less than the first segment
2429 * size, something has gone terribly wrong.
2431 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2432 blk_dump_rq_flags(req
, "request botched");
2433 req
->__data_len
= blk_rq_cur_bytes(req
);
2436 /* recalculate the number of segments */
2437 blk_recalc_rq_segments(req
);
2441 EXPORT_SYMBOL_GPL(blk_update_request
);
2443 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2444 unsigned int nr_bytes
,
2445 unsigned int bidi_bytes
)
2447 if (blk_update_request(rq
, error
, nr_bytes
))
2450 /* Bidi request must be completed as a whole */
2451 if (unlikely(blk_bidi_rq(rq
)) &&
2452 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2455 if (blk_queue_add_random(rq
->q
))
2456 add_disk_randomness(rq
->rq_disk
);
2462 * blk_unprep_request - unprepare a request
2465 * This function makes a request ready for complete resubmission (or
2466 * completion). It happens only after all error handling is complete,
2467 * so represents the appropriate moment to deallocate any resources
2468 * that were allocated to the request in the prep_rq_fn. The queue
2469 * lock is held when calling this.
2471 void blk_unprep_request(struct request
*req
)
2473 struct request_queue
*q
= req
->q
;
2475 req
->cmd_flags
&= ~REQ_DONTPREP
;
2476 if (q
->unprep_rq_fn
)
2477 q
->unprep_rq_fn(q
, req
);
2479 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2482 * queue lock must be held
2484 static void blk_finish_request(struct request
*req
, int error
)
2486 if (blk_rq_tagged(req
))
2487 blk_queue_end_tag(req
->q
, req
);
2489 BUG_ON(blk_queued_rq(req
));
2491 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2492 laptop_io_completion(&req
->q
->backing_dev_info
);
2494 blk_delete_timer(req
);
2496 if (req
->cmd_flags
& REQ_DONTPREP
)
2497 blk_unprep_request(req
);
2499 blk_account_io_done(req
);
2502 req
->end_io(req
, error
);
2504 if (blk_bidi_rq(req
))
2505 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2507 __blk_put_request(req
->q
, req
);
2512 * blk_end_bidi_request - Complete a bidi request
2513 * @rq: the request to complete
2514 * @error: %0 for success, < %0 for error
2515 * @nr_bytes: number of bytes to complete @rq
2516 * @bidi_bytes: number of bytes to complete @rq->next_rq
2519 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2520 * Drivers that supports bidi can safely call this member for any
2521 * type of request, bidi or uni. In the later case @bidi_bytes is
2525 * %false - we are done with this request
2526 * %true - still buffers pending for this request
2528 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2529 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2531 struct request_queue
*q
= rq
->q
;
2532 unsigned long flags
;
2534 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2537 spin_lock_irqsave(q
->queue_lock
, flags
);
2538 blk_finish_request(rq
, error
);
2539 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2545 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2546 * @rq: the request to complete
2547 * @error: %0 for success, < %0 for error
2548 * @nr_bytes: number of bytes to complete @rq
2549 * @bidi_bytes: number of bytes to complete @rq->next_rq
2552 * Identical to blk_end_bidi_request() except that queue lock is
2553 * assumed to be locked on entry and remains so on return.
2556 * %false - we are done with this request
2557 * %true - still buffers pending for this request
2559 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2560 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2562 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2565 blk_finish_request(rq
, error
);
2571 * blk_end_request - Helper function for drivers to complete the request.
2572 * @rq: the request being processed
2573 * @error: %0 for success, < %0 for error
2574 * @nr_bytes: number of bytes to complete
2577 * Ends I/O on a number of bytes attached to @rq.
2578 * If @rq has leftover, sets it up for the next range of segments.
2581 * %false - we are done with this request
2582 * %true - still buffers pending for this request
2584 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2586 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2588 EXPORT_SYMBOL(blk_end_request
);
2591 * blk_end_request_all - Helper function for drives to finish the request.
2592 * @rq: the request to finish
2593 * @error: %0 for success, < %0 for error
2596 * Completely finish @rq.
2598 void blk_end_request_all(struct request
*rq
, int error
)
2601 unsigned int bidi_bytes
= 0;
2603 if (unlikely(blk_bidi_rq(rq
)))
2604 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2606 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2609 EXPORT_SYMBOL(blk_end_request_all
);
2612 * blk_end_request_cur - Helper function to finish the current request chunk.
2613 * @rq: the request to finish the current chunk for
2614 * @error: %0 for success, < %0 for error
2617 * Complete the current consecutively mapped chunk from @rq.
2620 * %false - we are done with this request
2621 * %true - still buffers pending for this request
2623 bool blk_end_request_cur(struct request
*rq
, int error
)
2625 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2627 EXPORT_SYMBOL(blk_end_request_cur
);
2630 * blk_end_request_err - Finish a request till the next failure boundary.
2631 * @rq: the request to finish till the next failure boundary for
2632 * @error: must be negative errno
2635 * Complete @rq till the next failure boundary.
2638 * %false - we are done with this request
2639 * %true - still buffers pending for this request
2641 bool blk_end_request_err(struct request
*rq
, int error
)
2643 WARN_ON(error
>= 0);
2644 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2646 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2649 * __blk_end_request - Helper function for drivers to complete the request.
2650 * @rq: the request being processed
2651 * @error: %0 for success, < %0 for error
2652 * @nr_bytes: number of bytes to complete
2655 * Must be called with queue lock held unlike blk_end_request().
2658 * %false - we are done with this request
2659 * %true - still buffers pending for this request
2661 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2663 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2665 EXPORT_SYMBOL(__blk_end_request
);
2668 * __blk_end_request_all - Helper function for drives to finish the request.
2669 * @rq: the request to finish
2670 * @error: %0 for success, < %0 for error
2673 * Completely finish @rq. Must be called with queue lock held.
2675 void __blk_end_request_all(struct request
*rq
, int error
)
2678 unsigned int bidi_bytes
= 0;
2680 if (unlikely(blk_bidi_rq(rq
)))
2681 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2683 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2686 EXPORT_SYMBOL(__blk_end_request_all
);
2689 * __blk_end_request_cur - Helper function to finish the current request chunk.
2690 * @rq: the request to finish the current chunk for
2691 * @error: %0 for success, < %0 for error
2694 * Complete the current consecutively mapped chunk from @rq. Must
2695 * be called with queue lock held.
2698 * %false - we are done with this request
2699 * %true - still buffers pending for this request
2701 bool __blk_end_request_cur(struct request
*rq
, int error
)
2703 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2705 EXPORT_SYMBOL(__blk_end_request_cur
);
2708 * __blk_end_request_err - Finish a request till the next failure boundary.
2709 * @rq: the request to finish till the next failure boundary for
2710 * @error: must be negative errno
2713 * Complete @rq till the next failure boundary. Must be called
2714 * with queue lock held.
2717 * %false - we are done with this request
2718 * %true - still buffers pending for this request
2720 bool __blk_end_request_err(struct request
*rq
, int error
)
2722 WARN_ON(error
>= 0);
2723 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2725 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2727 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2730 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2731 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2733 if (bio_has_data(bio
)) {
2734 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2735 rq
->buffer
= bio_data(bio
);
2737 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2738 rq
->bio
= rq
->biotail
= bio
;
2741 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2746 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2747 * @rq: the request to be flushed
2750 * Flush all pages in @rq.
2752 void rq_flush_dcache_pages(struct request
*rq
)
2754 struct req_iterator iter
;
2755 struct bio_vec bvec
;
2757 rq_for_each_segment(bvec
, rq
, iter
)
2758 flush_dcache_page(bvec
.bv_page
);
2760 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2764 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2765 * @q : the queue of the device being checked
2768 * Check if underlying low-level drivers of a device are busy.
2769 * If the drivers want to export their busy state, they must set own
2770 * exporting function using blk_queue_lld_busy() first.
2772 * Basically, this function is used only by request stacking drivers
2773 * to stop dispatching requests to underlying devices when underlying
2774 * devices are busy. This behavior helps more I/O merging on the queue
2775 * of the request stacking driver and prevents I/O throughput regression
2776 * on burst I/O load.
2779 * 0 - Not busy (The request stacking driver should dispatch request)
2780 * 1 - Busy (The request stacking driver should stop dispatching request)
2782 int blk_lld_busy(struct request_queue
*q
)
2785 return q
->lld_busy_fn(q
);
2789 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2792 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2793 * @rq: the clone request to be cleaned up
2796 * Free all bios in @rq for a cloned request.
2798 void blk_rq_unprep_clone(struct request
*rq
)
2802 while ((bio
= rq
->bio
) != NULL
) {
2803 rq
->bio
= bio
->bi_next
;
2808 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2811 * Copy attributes of the original request to the clone request.
2812 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2814 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2816 dst
->cpu
= src
->cpu
;
2817 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2818 dst
->cmd_type
= src
->cmd_type
;
2819 dst
->__sector
= blk_rq_pos(src
);
2820 dst
->__data_len
= blk_rq_bytes(src
);
2821 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2822 dst
->ioprio
= src
->ioprio
;
2823 dst
->extra_len
= src
->extra_len
;
2827 * blk_rq_prep_clone - Helper function to setup clone request
2828 * @rq: the request to be setup
2829 * @rq_src: original request to be cloned
2830 * @bs: bio_set that bios for clone are allocated from
2831 * @gfp_mask: memory allocation mask for bio
2832 * @bio_ctr: setup function to be called for each clone bio.
2833 * Returns %0 for success, non %0 for failure.
2834 * @data: private data to be passed to @bio_ctr
2837 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2838 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2839 * are not copied, and copying such parts is the caller's responsibility.
2840 * Also, pages which the original bios are pointing to are not copied
2841 * and the cloned bios just point same pages.
2842 * So cloned bios must be completed before original bios, which means
2843 * the caller must complete @rq before @rq_src.
2845 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2846 struct bio_set
*bs
, gfp_t gfp_mask
,
2847 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2850 struct bio
*bio
, *bio_src
;
2855 blk_rq_init(NULL
, rq
);
2857 __rq_for_each_bio(bio_src
, rq_src
) {
2858 bio
= bio_clone_bioset(bio_src
, gfp_mask
, bs
);
2862 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2866 rq
->biotail
->bi_next
= bio
;
2869 rq
->bio
= rq
->biotail
= bio
;
2872 __blk_rq_prep_clone(rq
, rq_src
);
2879 blk_rq_unprep_clone(rq
);
2883 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2885 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2887 return queue_work(kblockd_workqueue
, work
);
2889 EXPORT_SYMBOL(kblockd_schedule_work
);
2891 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2892 struct delayed_work
*dwork
, unsigned long delay
)
2894 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2896 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2898 #define PLUG_MAGIC 0x91827364
2901 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2902 * @plug: The &struct blk_plug that needs to be initialized
2905 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2906 * pending I/O should the task end up blocking between blk_start_plug() and
2907 * blk_finish_plug(). This is important from a performance perspective, but
2908 * also ensures that we don't deadlock. For instance, if the task is blocking
2909 * for a memory allocation, memory reclaim could end up wanting to free a
2910 * page belonging to that request that is currently residing in our private
2911 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2912 * this kind of deadlock.
2914 void blk_start_plug(struct blk_plug
*plug
)
2916 struct task_struct
*tsk
= current
;
2918 plug
->magic
= PLUG_MAGIC
;
2919 INIT_LIST_HEAD(&plug
->list
);
2920 INIT_LIST_HEAD(&plug
->mq_list
);
2921 INIT_LIST_HEAD(&plug
->cb_list
);
2924 * If this is a nested plug, don't actually assign it. It will be
2925 * flushed on its own.
2929 * Store ordering should not be needed here, since a potential
2930 * preempt will imply a full memory barrier
2935 EXPORT_SYMBOL(blk_start_plug
);
2937 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2939 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2940 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2942 return !(rqa
->q
< rqb
->q
||
2943 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
2947 * If 'from_schedule' is true, then postpone the dispatch of requests
2948 * until a safe kblockd context. We due this to avoid accidental big
2949 * additional stack usage in driver dispatch, in places where the originally
2950 * plugger did not intend it.
2952 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2954 __releases(q
->queue_lock
)
2956 trace_block_unplug(q
, depth
, !from_schedule
);
2959 blk_run_queue_async(q
);
2962 spin_unlock(q
->queue_lock
);
2965 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
2967 LIST_HEAD(callbacks
);
2969 while (!list_empty(&plug
->cb_list
)) {
2970 list_splice_init(&plug
->cb_list
, &callbacks
);
2972 while (!list_empty(&callbacks
)) {
2973 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2976 list_del(&cb
->list
);
2977 cb
->callback(cb
, from_schedule
);
2982 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
2985 struct blk_plug
*plug
= current
->plug
;
2986 struct blk_plug_cb
*cb
;
2991 list_for_each_entry(cb
, &plug
->cb_list
, list
)
2992 if (cb
->callback
== unplug
&& cb
->data
== data
)
2995 /* Not currently on the callback list */
2996 BUG_ON(size
< sizeof(*cb
));
2997 cb
= kzalloc(size
, GFP_ATOMIC
);
3000 cb
->callback
= unplug
;
3001 list_add(&cb
->list
, &plug
->cb_list
);
3005 EXPORT_SYMBOL(blk_check_plugged
);
3007 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3009 struct request_queue
*q
;
3010 unsigned long flags
;
3015 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
3017 flush_plug_callbacks(plug
, from_schedule
);
3019 if (!list_empty(&plug
->mq_list
))
3020 blk_mq_flush_plug_list(plug
, from_schedule
);
3022 if (list_empty(&plug
->list
))
3025 list_splice_init(&plug
->list
, &list
);
3027 list_sort(NULL
, &list
, plug_rq_cmp
);
3033 * Save and disable interrupts here, to avoid doing it for every
3034 * queue lock we have to take.
3036 local_irq_save(flags
);
3037 while (!list_empty(&list
)) {
3038 rq
= list_entry_rq(list
.next
);
3039 list_del_init(&rq
->queuelist
);
3043 * This drops the queue lock
3046 queue_unplugged(q
, depth
, from_schedule
);
3049 spin_lock(q
->queue_lock
);
3053 * Short-circuit if @q is dead
3055 if (unlikely(blk_queue_dying(q
))) {
3056 __blk_end_request_all(rq
, -ENODEV
);
3061 * rq is already accounted, so use raw insert
3063 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3064 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3066 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3072 * This drops the queue lock
3075 queue_unplugged(q
, depth
, from_schedule
);
3077 local_irq_restore(flags
);
3080 void blk_finish_plug(struct blk_plug
*plug
)
3082 blk_flush_plug_list(plug
, false);
3084 if (plug
== current
->plug
)
3085 current
->plug
= NULL
;
3087 EXPORT_SYMBOL(blk_finish_plug
);
3089 #ifdef CONFIG_PM_RUNTIME
3091 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3092 * @q: the queue of the device
3093 * @dev: the device the queue belongs to
3096 * Initialize runtime-PM-related fields for @q and start auto suspend for
3097 * @dev. Drivers that want to take advantage of request-based runtime PM
3098 * should call this function after @dev has been initialized, and its
3099 * request queue @q has been allocated, and runtime PM for it can not happen
3100 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3101 * cases, driver should call this function before any I/O has taken place.
3103 * This function takes care of setting up using auto suspend for the device,
3104 * the autosuspend delay is set to -1 to make runtime suspend impossible
3105 * until an updated value is either set by user or by driver. Drivers do
3106 * not need to touch other autosuspend settings.
3108 * The block layer runtime PM is request based, so only works for drivers
3109 * that use request as their IO unit instead of those directly use bio's.
3111 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3114 q
->rpm_status
= RPM_ACTIVE
;
3115 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3116 pm_runtime_use_autosuspend(q
->dev
);
3118 EXPORT_SYMBOL(blk_pm_runtime_init
);
3121 * blk_pre_runtime_suspend - Pre runtime suspend check
3122 * @q: the queue of the device
3125 * This function will check if runtime suspend is allowed for the device
3126 * by examining if there are any requests pending in the queue. If there
3127 * are requests pending, the device can not be runtime suspended; otherwise,
3128 * the queue's status will be updated to SUSPENDING and the driver can
3129 * proceed to suspend the device.
3131 * For the not allowed case, we mark last busy for the device so that
3132 * runtime PM core will try to autosuspend it some time later.
3134 * This function should be called near the start of the device's
3135 * runtime_suspend callback.
3138 * 0 - OK to runtime suspend the device
3139 * -EBUSY - Device should not be runtime suspended
3141 int blk_pre_runtime_suspend(struct request_queue
*q
)
3145 spin_lock_irq(q
->queue_lock
);
3146 if (q
->nr_pending
) {
3148 pm_runtime_mark_last_busy(q
->dev
);
3150 q
->rpm_status
= RPM_SUSPENDING
;
3152 spin_unlock_irq(q
->queue_lock
);
3155 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3158 * blk_post_runtime_suspend - Post runtime suspend processing
3159 * @q: the queue of the device
3160 * @err: return value of the device's runtime_suspend function
3163 * Update the queue's runtime status according to the return value of the
3164 * device's runtime suspend function and mark last busy for the device so
3165 * that PM core will try to auto suspend the device at a later time.
3167 * This function should be called near the end of the device's
3168 * runtime_suspend callback.
3170 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3172 spin_lock_irq(q
->queue_lock
);
3174 q
->rpm_status
= RPM_SUSPENDED
;
3176 q
->rpm_status
= RPM_ACTIVE
;
3177 pm_runtime_mark_last_busy(q
->dev
);
3179 spin_unlock_irq(q
->queue_lock
);
3181 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3184 * blk_pre_runtime_resume - Pre runtime resume processing
3185 * @q: the queue of the device
3188 * Update the queue's runtime status to RESUMING in preparation for the
3189 * runtime resume of the device.
3191 * This function should be called near the start of the device's
3192 * runtime_resume callback.
3194 void blk_pre_runtime_resume(struct request_queue
*q
)
3196 spin_lock_irq(q
->queue_lock
);
3197 q
->rpm_status
= RPM_RESUMING
;
3198 spin_unlock_irq(q
->queue_lock
);
3200 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3203 * blk_post_runtime_resume - Post runtime resume processing
3204 * @q: the queue of the device
3205 * @err: return value of the device's runtime_resume function
3208 * Update the queue's runtime status according to the return value of the
3209 * device's runtime_resume function. If it is successfully resumed, process
3210 * the requests that are queued into the device's queue when it is resuming
3211 * and then mark last busy and initiate autosuspend for it.
3213 * This function should be called near the end of the device's
3214 * runtime_resume callback.
3216 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3218 spin_lock_irq(q
->queue_lock
);
3220 q
->rpm_status
= RPM_ACTIVE
;
3222 pm_runtime_mark_last_busy(q
->dev
);
3223 pm_request_autosuspend(q
->dev
);
3225 q
->rpm_status
= RPM_SUSPENDED
;
3227 spin_unlock_irq(q
->queue_lock
);
3229 EXPORT_SYMBOL(blk_post_runtime_resume
);
3232 int __init
blk_dev_init(void)
3234 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3235 sizeof(((struct request
*)0)->cmd_flags
));
3237 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3238 kblockd_workqueue
= alloc_workqueue("kblockd",
3239 WQ_MEM_RECLAIM
| WQ_HIGHPRI
|
3240 WQ_POWER_EFFICIENT
, 0);
3241 if (!kblockd_workqueue
)
3242 panic("Failed to create kblockd\n");
3244 request_cachep
= kmem_cache_create("blkdev_requests",
3245 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3247 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3248 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);