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
);
251 struct blk_mq_hw_ctx
*hctx
;
254 queue_for_each_hw_ctx(q
, hctx
, i
)
255 cancel_delayed_work_sync(&hctx
->delayed_work
);
257 cancel_delayed_work_sync(&q
->delay_work
);
260 EXPORT_SYMBOL(blk_sync_queue
);
263 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
264 * @q: The queue to run
267 * Invoke request handling on a queue if there are any pending requests.
268 * May be used to restart request handling after a request has completed.
269 * This variant runs the queue whether or not the queue has been
270 * stopped. Must be called with the queue lock held and interrupts
271 * disabled. See also @blk_run_queue.
273 inline void __blk_run_queue_uncond(struct request_queue
*q
)
275 if (unlikely(blk_queue_dead(q
)))
279 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
280 * the queue lock internally. As a result multiple threads may be
281 * running such a request function concurrently. Keep track of the
282 * number of active request_fn invocations such that blk_drain_queue()
283 * can wait until all these request_fn calls have finished.
285 q
->request_fn_active
++;
287 q
->request_fn_active
--;
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
298 void __blk_run_queue(struct request_queue
*q
)
300 if (unlikely(blk_queue_stopped(q
)))
303 __blk_run_queue_uncond(q
);
305 EXPORT_SYMBOL(__blk_run_queue
);
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
313 * of us. The caller must hold the queue lock.
315 void blk_run_queue_async(struct request_queue
*q
)
317 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
318 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
320 EXPORT_SYMBOL(blk_run_queue_async
);
323 * blk_run_queue - run a single device queue
324 * @q: The queue to run
327 * Invoke request handling on this queue, if it has pending work to do.
328 * May be used to restart queueing when a request has completed.
330 void blk_run_queue(struct request_queue
*q
)
334 spin_lock_irqsave(q
->queue_lock
, flags
);
336 spin_unlock_irqrestore(q
->queue_lock
, flags
);
338 EXPORT_SYMBOL(blk_run_queue
);
340 void blk_put_queue(struct request_queue
*q
)
342 kobject_put(&q
->kobj
);
344 EXPORT_SYMBOL(blk_put_queue
);
347 * __blk_drain_queue - drain requests from request_queue
349 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
351 * Drain requests from @q. If @drain_all is set, all requests are drained.
352 * If not, only ELVPRIV requests are drained. The caller is responsible
353 * for ensuring that no new requests which need to be drained are queued.
355 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
356 __releases(q
->queue_lock
)
357 __acquires(q
->queue_lock
)
361 lockdep_assert_held(q
->queue_lock
);
367 * The caller might be trying to drain @q before its
368 * elevator is initialized.
371 elv_drain_elevator(q
);
373 blkcg_drain_queue(q
);
376 * This function might be called on a queue which failed
377 * driver init after queue creation or is not yet fully
378 * active yet. Some drivers (e.g. fd and loop) get unhappy
379 * in such cases. Kick queue iff dispatch queue has
380 * something on it and @q has request_fn set.
382 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
385 drain
|= q
->nr_rqs_elvpriv
;
386 drain
|= q
->request_fn_active
;
389 * Unfortunately, requests are queued at and tracked from
390 * multiple places and there's no single counter which can
391 * be drained. Check all the queues and counters.
394 drain
|= !list_empty(&q
->queue_head
);
395 for (i
= 0; i
< 2; i
++) {
396 drain
|= q
->nr_rqs
[i
];
397 drain
|= q
->in_flight
[i
];
398 drain
|= !list_empty(&q
->flush_queue
[i
]);
405 spin_unlock_irq(q
->queue_lock
);
409 spin_lock_irq(q
->queue_lock
);
413 * With queue marked dead, any woken up waiter will fail the
414 * allocation path, so the wakeup chaining is lost and we're
415 * left with hung waiters. We need to wake up those waiters.
418 struct request_list
*rl
;
420 blk_queue_for_each_rl(rl
, q
)
421 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
422 wake_up_all(&rl
->wait
[i
]);
427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
432 * throttled or issued before. On return, it's guaranteed that no request
433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
436 void blk_queue_bypass_start(struct request_queue
*q
)
440 spin_lock_irq(q
->queue_lock
);
441 drain
= !q
->bypass_depth
++;
442 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
443 spin_unlock_irq(q
->queue_lock
);
446 spin_lock_irq(q
->queue_lock
);
447 __blk_drain_queue(q
, false);
448 spin_unlock_irq(q
->queue_lock
);
450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
454 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
460 * Leave bypass mode and restore the normal queueing behavior.
462 void blk_queue_bypass_end(struct request_queue
*q
)
464 spin_lock_irq(q
->queue_lock
);
465 if (!--q
->bypass_depth
)
466 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
467 WARN_ON_ONCE(q
->bypass_depth
< 0);
468 spin_unlock_irq(q
->queue_lock
);
470 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
476 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
477 * put it. All future requests will be failed immediately with -ENODEV.
479 void blk_cleanup_queue(struct request_queue
*q
)
481 spinlock_t
*lock
= q
->queue_lock
;
483 /* mark @q DYING, no new request or merges will be allowed afterwards */
484 mutex_lock(&q
->sysfs_lock
);
485 queue_flag_set_unlocked(QUEUE_FLAG_DYING
, q
);
489 * A dying queue is permanently in bypass mode till released. Note
490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
498 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
500 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
502 queue_flag_set(QUEUE_FLAG_DYING
, q
);
503 spin_unlock_irq(lock
);
504 mutex_unlock(&q
->sysfs_lock
);
507 * Drain all requests queued before DYING marking. Set DEAD flag to
508 * prevent that q->request_fn() gets invoked after draining finished.
511 blk_mq_drain_queue(q
);
515 __blk_drain_queue(q
, true);
517 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
518 spin_unlock_irq(lock
);
520 /* @q won't process any more request, flush async actions */
521 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
525 if (q
->queue_lock
!= &q
->__queue_lock
)
526 q
->queue_lock
= &q
->__queue_lock
;
527 spin_unlock_irq(lock
);
529 /* @q is and will stay empty, shutdown and put */
532 EXPORT_SYMBOL(blk_cleanup_queue
);
534 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
537 if (unlikely(rl
->rq_pool
))
541 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
542 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
543 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
544 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
546 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
547 mempool_free_slab
, request_cachep
,
555 void blk_exit_rl(struct request_list
*rl
)
558 mempool_destroy(rl
->rq_pool
);
561 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
563 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
565 EXPORT_SYMBOL(blk_alloc_queue
);
567 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
569 struct request_queue
*q
;
572 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
573 gfp_mask
| __GFP_ZERO
, node_id
);
577 if (percpu_counter_init(&q
->mq_usage_counter
, 0))
580 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
584 q
->backing_dev_info
.ra_pages
=
585 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
586 q
->backing_dev_info
.state
= 0;
587 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
588 q
->backing_dev_info
.name
= "block";
591 err
= bdi_init(&q
->backing_dev_info
);
595 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
596 laptop_mode_timer_fn
, (unsigned long) q
);
597 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
598 INIT_LIST_HEAD(&q
->queue_head
);
599 INIT_LIST_HEAD(&q
->timeout_list
);
600 INIT_LIST_HEAD(&q
->icq_list
);
601 #ifdef CONFIG_BLK_CGROUP
602 INIT_LIST_HEAD(&q
->blkg_list
);
604 INIT_LIST_HEAD(&q
->flush_queue
[0]);
605 INIT_LIST_HEAD(&q
->flush_queue
[1]);
606 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
607 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
609 kobject_init(&q
->kobj
, &blk_queue_ktype
);
611 mutex_init(&q
->sysfs_lock
);
612 spin_lock_init(&q
->__queue_lock
);
615 * By default initialize queue_lock to internal lock and driver can
616 * override it later if need be.
618 q
->queue_lock
= &q
->__queue_lock
;
621 * A queue starts its life with bypass turned on to avoid
622 * unnecessary bypass on/off overhead and nasty surprises during
623 * init. The initial bypass will be finished when the queue is
624 * registered by blk_register_queue().
627 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
629 init_waitqueue_head(&q
->mq_freeze_wq
);
631 if (blkcg_init_queue(q
))
637 bdi_destroy(&q
->backing_dev_info
);
639 ida_simple_remove(&blk_queue_ida
, q
->id
);
641 percpu_counter_destroy(&q
->mq_usage_counter
);
643 kmem_cache_free(blk_requestq_cachep
, q
);
646 EXPORT_SYMBOL(blk_alloc_queue_node
);
649 * blk_init_queue - prepare a request queue for use with a block device
650 * @rfn: The function to be called to process requests that have been
651 * placed on the queue.
652 * @lock: Request queue spin lock
655 * If a block device wishes to use the standard request handling procedures,
656 * which sorts requests and coalesces adjacent requests, then it must
657 * call blk_init_queue(). The function @rfn will be called when there
658 * are requests on the queue that need to be processed. If the device
659 * supports plugging, then @rfn may not be called immediately when requests
660 * are available on the queue, but may be called at some time later instead.
661 * Plugged queues are generally unplugged when a buffer belonging to one
662 * of the requests on the queue is needed, or due to memory pressure.
664 * @rfn is not required, or even expected, to remove all requests off the
665 * queue, but only as many as it can handle at a time. If it does leave
666 * requests on the queue, it is responsible for arranging that the requests
667 * get dealt with eventually.
669 * The queue spin lock must be held while manipulating the requests on the
670 * request queue; this lock will be taken also from interrupt context, so irq
671 * disabling is needed for it.
673 * Function returns a pointer to the initialized request queue, or %NULL if
677 * blk_init_queue() must be paired with a blk_cleanup_queue() call
678 * when the block device is deactivated (such as at module unload).
681 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
683 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
685 EXPORT_SYMBOL(blk_init_queue
);
687 struct request_queue
*
688 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
690 struct request_queue
*uninit_q
, *q
;
692 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
696 uninit_q
->flush_rq
= kzalloc(sizeof(struct request
), GFP_KERNEL
);
697 if (!uninit_q
->flush_rq
)
698 goto out_cleanup_queue
;
700 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
702 goto out_free_flush_rq
;
706 kfree(uninit_q
->flush_rq
);
708 blk_cleanup_queue(uninit_q
);
711 EXPORT_SYMBOL(blk_init_queue_node
);
713 struct request_queue
*
714 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
720 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
724 q
->prep_rq_fn
= NULL
;
725 q
->unprep_rq_fn
= NULL
;
726 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
728 /* Override internal queue lock with supplied lock pointer */
730 q
->queue_lock
= lock
;
733 * This also sets hw/phys segments, boundary and size
735 blk_queue_make_request(q
, blk_queue_bio
);
737 q
->sg_reserved_size
= INT_MAX
;
739 /* Protect q->elevator from elevator_change */
740 mutex_lock(&q
->sysfs_lock
);
743 if (elevator_init(q
, NULL
)) {
744 mutex_unlock(&q
->sysfs_lock
);
748 mutex_unlock(&q
->sysfs_lock
);
752 EXPORT_SYMBOL(blk_init_allocated_queue
);
754 bool blk_get_queue(struct request_queue
*q
)
756 if (likely(!blk_queue_dying(q
))) {
763 EXPORT_SYMBOL(blk_get_queue
);
765 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
767 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
768 elv_put_request(rl
->q
, rq
);
770 put_io_context(rq
->elv
.icq
->ioc
);
773 mempool_free(rq
, rl
->rq_pool
);
777 * ioc_batching returns true if the ioc is a valid batching request and
778 * should be given priority access to a request.
780 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
786 * Make sure the process is able to allocate at least 1 request
787 * even if the batch times out, otherwise we could theoretically
790 return ioc
->nr_batch_requests
== q
->nr_batching
||
791 (ioc
->nr_batch_requests
> 0
792 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
796 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
797 * will cause the process to be a "batcher" on all queues in the system. This
798 * is the behaviour we want though - once it gets a wakeup it should be given
801 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
803 if (!ioc
|| ioc_batching(q
, ioc
))
806 ioc
->nr_batch_requests
= q
->nr_batching
;
807 ioc
->last_waited
= jiffies
;
810 static void __freed_request(struct request_list
*rl
, int sync
)
812 struct request_queue
*q
= rl
->q
;
815 * bdi isn't aware of blkcg yet. As all async IOs end up root
816 * blkcg anyway, just use root blkcg state.
818 if (rl
== &q
->root_rl
&&
819 rl
->count
[sync
] < queue_congestion_off_threshold(q
))
820 blk_clear_queue_congested(q
, sync
);
822 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
823 if (waitqueue_active(&rl
->wait
[sync
]))
824 wake_up(&rl
->wait
[sync
]);
826 blk_clear_rl_full(rl
, sync
);
831 * A request has just been released. Account for it, update the full and
832 * congestion status, wake up any waiters. Called under q->queue_lock.
834 static void freed_request(struct request_list
*rl
, unsigned int flags
)
836 struct request_queue
*q
= rl
->q
;
837 int sync
= rw_is_sync(flags
);
841 if (flags
& REQ_ELVPRIV
)
844 __freed_request(rl
, sync
);
846 if (unlikely(rl
->starved
[sync
^ 1]))
847 __freed_request(rl
, sync
^ 1);
851 * Determine if elevator data should be initialized when allocating the
852 * request associated with @bio.
854 static bool blk_rq_should_init_elevator(struct bio
*bio
)
860 * Flush requests do not use the elevator so skip initialization.
861 * This allows a request to share the flush and elevator data.
863 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
870 * rq_ioc - determine io_context for request allocation
871 * @bio: request being allocated is for this bio (can be %NULL)
873 * Determine io_context to use for request allocation for @bio. May return
874 * %NULL if %current->io_context doesn't exist.
876 static struct io_context
*rq_ioc(struct bio
*bio
)
878 #ifdef CONFIG_BLK_CGROUP
879 if (bio
&& bio
->bi_ioc
)
882 return current
->io_context
;
886 * __get_request - get a free request
887 * @rl: request list to allocate from
888 * @rw_flags: RW and SYNC flags
889 * @bio: bio to allocate request for (can be %NULL)
890 * @gfp_mask: allocation mask
892 * Get a free request from @q. This function may fail under memory
893 * pressure or if @q is dead.
895 * Must be callled with @q->queue_lock held and,
896 * Returns %NULL on failure, with @q->queue_lock held.
897 * Returns !%NULL on success, with @q->queue_lock *not held*.
899 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
900 struct bio
*bio
, gfp_t gfp_mask
)
902 struct request_queue
*q
= rl
->q
;
904 struct elevator_type
*et
= q
->elevator
->type
;
905 struct io_context
*ioc
= rq_ioc(bio
);
906 struct io_cq
*icq
= NULL
;
907 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
910 if (unlikely(blk_queue_dying(q
)))
913 may_queue
= elv_may_queue(q
, rw_flags
);
914 if (may_queue
== ELV_MQUEUE_NO
)
917 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
918 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
920 * The queue will fill after this allocation, so set
921 * it as full, and mark this process as "batching".
922 * This process will be allowed to complete a batch of
923 * requests, others will be blocked.
925 if (!blk_rl_full(rl
, is_sync
)) {
926 ioc_set_batching(q
, ioc
);
927 blk_set_rl_full(rl
, is_sync
);
929 if (may_queue
!= ELV_MQUEUE_MUST
930 && !ioc_batching(q
, ioc
)) {
932 * The queue is full and the allocating
933 * process is not a "batcher", and not
934 * exempted by the IO scheduler
941 * bdi isn't aware of blkcg yet. As all async IOs end up
942 * root blkcg anyway, just use root blkcg state.
944 if (rl
== &q
->root_rl
)
945 blk_set_queue_congested(q
, is_sync
);
949 * Only allow batching queuers to allocate up to 50% over the defined
950 * limit of requests, otherwise we could have thousands of requests
951 * allocated with any setting of ->nr_requests
953 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
956 q
->nr_rqs
[is_sync
]++;
957 rl
->count
[is_sync
]++;
958 rl
->starved
[is_sync
] = 0;
961 * Decide whether the new request will be managed by elevator. If
962 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
963 * prevent the current elevator from being destroyed until the new
964 * request is freed. This guarantees icq's won't be destroyed and
965 * makes creating new ones safe.
967 * Also, lookup icq while holding queue_lock. If it doesn't exist,
968 * it will be created after releasing queue_lock.
970 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
971 rw_flags
|= REQ_ELVPRIV
;
973 if (et
->icq_cache
&& ioc
)
974 icq
= ioc_lookup_icq(ioc
, q
);
977 if (blk_queue_io_stat(q
))
978 rw_flags
|= REQ_IO_STAT
;
979 spin_unlock_irq(q
->queue_lock
);
981 /* allocate and init request */
982 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
987 blk_rq_set_rl(rq
, rl
);
988 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
991 if (rw_flags
& REQ_ELVPRIV
) {
992 if (unlikely(et
->icq_cache
&& !icq
)) {
994 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1000 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1003 /* @rq->elv.icq holds io_context until @rq is freed */
1005 get_io_context(icq
->ioc
);
1009 * ioc may be NULL here, and ioc_batching will be false. That's
1010 * OK, if the queue is under the request limit then requests need
1011 * not count toward the nr_batch_requests limit. There will always
1012 * be some limit enforced by BLK_BATCH_TIME.
1014 if (ioc_batching(q
, ioc
))
1015 ioc
->nr_batch_requests
--;
1017 trace_block_getrq(q
, bio
, rw_flags
& 1);
1022 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1023 * and may fail indefinitely under memory pressure and thus
1024 * shouldn't stall IO. Treat this request as !elvpriv. This will
1025 * disturb iosched and blkcg but weird is bettern than dead.
1027 printk_ratelimited(KERN_WARNING
"%s: request aux data allocation failed, iosched may be disturbed\n",
1028 dev_name(q
->backing_dev_info
.dev
));
1030 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1033 spin_lock_irq(q
->queue_lock
);
1034 q
->nr_rqs_elvpriv
--;
1035 spin_unlock_irq(q
->queue_lock
);
1040 * Allocation failed presumably due to memory. Undo anything we
1041 * might have messed up.
1043 * Allocating task should really be put onto the front of the wait
1044 * queue, but this is pretty rare.
1046 spin_lock_irq(q
->queue_lock
);
1047 freed_request(rl
, rw_flags
);
1050 * in the very unlikely event that allocation failed and no
1051 * requests for this direction was pending, mark us starved so that
1052 * freeing of a request in the other direction will notice
1053 * us. another possible fix would be to split the rq mempool into
1057 if (unlikely(rl
->count
[is_sync
] == 0))
1058 rl
->starved
[is_sync
] = 1;
1063 * get_request - get a free request
1064 * @q: request_queue to allocate request from
1065 * @rw_flags: RW and SYNC flags
1066 * @bio: bio to allocate request for (can be %NULL)
1067 * @gfp_mask: allocation mask
1069 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1070 * function keeps retrying under memory pressure and fails iff @q is dead.
1072 * Must be callled with @q->queue_lock held and,
1073 * Returns %NULL on failure, with @q->queue_lock held.
1074 * Returns !%NULL on success, with @q->queue_lock *not held*.
1076 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1077 struct bio
*bio
, gfp_t gfp_mask
)
1079 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1081 struct request_list
*rl
;
1084 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1086 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1090 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dying(q
))) {
1095 /* wait on @rl and retry */
1096 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1097 TASK_UNINTERRUPTIBLE
);
1099 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1101 spin_unlock_irq(q
->queue_lock
);
1105 * After sleeping, we become a "batching" process and will be able
1106 * to allocate at least one request, and up to a big batch of them
1107 * for a small period time. See ioc_batching, ioc_set_batching
1109 ioc_set_batching(q
, current
->io_context
);
1111 spin_lock_irq(q
->queue_lock
);
1112 finish_wait(&rl
->wait
[is_sync
], &wait
);
1117 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1122 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1124 /* create ioc upfront */
1125 create_io_context(gfp_mask
, q
->node
);
1127 spin_lock_irq(q
->queue_lock
);
1128 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1130 spin_unlock_irq(q
->queue_lock
);
1131 /* q->queue_lock is unlocked at this point */
1136 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1139 return blk_mq_alloc_request(q
, rw
, gfp_mask
);
1141 return blk_old_get_request(q
, rw
, gfp_mask
);
1143 EXPORT_SYMBOL(blk_get_request
);
1146 * blk_make_request - given a bio, allocate a corresponding struct request.
1147 * @q: target request queue
1148 * @bio: The bio describing the memory mappings that will be submitted for IO.
1149 * It may be a chained-bio properly constructed by block/bio layer.
1150 * @gfp_mask: gfp flags to be used for memory allocation
1152 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1153 * type commands. Where the struct request needs to be farther initialized by
1154 * the caller. It is passed a &struct bio, which describes the memory info of
1157 * The caller of blk_make_request must make sure that bi_io_vec
1158 * are set to describe the memory buffers. That bio_data_dir() will return
1159 * the needed direction of the request. (And all bio's in the passed bio-chain
1160 * are properly set accordingly)
1162 * If called under none-sleepable conditions, mapped bio buffers must not
1163 * need bouncing, by calling the appropriate masked or flagged allocator,
1164 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1167 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1168 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1169 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1170 * completion of a bio that hasn't been submitted yet, thus resulting in a
1171 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1172 * of bio_alloc(), as that avoids the mempool deadlock.
1173 * If possible a big IO should be split into smaller parts when allocation
1174 * fails. Partial allocation should not be an error, or you risk a live-lock.
1176 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1179 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1182 return ERR_PTR(-ENOMEM
);
1185 struct bio
*bounce_bio
= bio
;
1188 blk_queue_bounce(q
, &bounce_bio
);
1189 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1190 if (unlikely(ret
)) {
1191 blk_put_request(rq
);
1192 return ERR_PTR(ret
);
1198 EXPORT_SYMBOL(blk_make_request
);
1201 * blk_requeue_request - put a request back on queue
1202 * @q: request queue where request should be inserted
1203 * @rq: request to be inserted
1206 * Drivers often keep queueing requests until the hardware cannot accept
1207 * more, when that condition happens we need to put the request back
1208 * on the queue. Must be called with queue lock held.
1210 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1212 blk_delete_timer(rq
);
1213 blk_clear_rq_complete(rq
);
1214 trace_block_rq_requeue(q
, rq
);
1216 if (blk_rq_tagged(rq
))
1217 blk_queue_end_tag(q
, rq
);
1219 BUG_ON(blk_queued_rq(rq
));
1221 elv_requeue_request(q
, rq
);
1223 EXPORT_SYMBOL(blk_requeue_request
);
1225 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1228 blk_account_io_start(rq
, true);
1229 __elv_add_request(q
, rq
, where
);
1232 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1235 if (now
== part
->stamp
)
1238 if (part_in_flight(part
)) {
1239 __part_stat_add(cpu
, part
, time_in_queue
,
1240 part_in_flight(part
) * (now
- part
->stamp
));
1241 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1247 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1248 * @cpu: cpu number for stats access
1249 * @part: target partition
1251 * The average IO queue length and utilisation statistics are maintained
1252 * by observing the current state of the queue length and the amount of
1253 * time it has been in this state for.
1255 * Normally, that accounting is done on IO completion, but that can result
1256 * in more than a second's worth of IO being accounted for within any one
1257 * second, leading to >100% utilisation. To deal with that, we call this
1258 * function to do a round-off before returning the results when reading
1259 * /proc/diskstats. This accounts immediately for all queue usage up to
1260 * the current jiffies and restarts the counters again.
1262 void part_round_stats(int cpu
, struct hd_struct
*part
)
1264 unsigned long now
= jiffies
;
1267 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1268 part_round_stats_single(cpu
, part
, now
);
1270 EXPORT_SYMBOL_GPL(part_round_stats
);
1272 #ifdef CONFIG_PM_RUNTIME
1273 static void blk_pm_put_request(struct request
*rq
)
1275 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1276 pm_runtime_mark_last_busy(rq
->q
->dev
);
1279 static inline void blk_pm_put_request(struct request
*rq
) {}
1283 * queue lock must be held
1285 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1291 blk_mq_free_request(req
);
1295 blk_pm_put_request(req
);
1297 elv_completed_request(q
, req
);
1299 /* this is a bio leak */
1300 WARN_ON(req
->bio
!= NULL
);
1303 * Request may not have originated from ll_rw_blk. if not,
1304 * it didn't come out of our reserved rq pools
1306 if (req
->cmd_flags
& REQ_ALLOCED
) {
1307 unsigned int flags
= req
->cmd_flags
;
1308 struct request_list
*rl
= blk_rq_rl(req
);
1310 BUG_ON(!list_empty(&req
->queuelist
));
1311 BUG_ON(!hlist_unhashed(&req
->hash
));
1313 blk_free_request(rl
, req
);
1314 freed_request(rl
, flags
);
1318 EXPORT_SYMBOL_GPL(__blk_put_request
);
1320 void blk_put_request(struct request
*req
)
1322 struct request_queue
*q
= req
->q
;
1325 blk_mq_free_request(req
);
1327 unsigned long flags
;
1329 spin_lock_irqsave(q
->queue_lock
, flags
);
1330 __blk_put_request(q
, req
);
1331 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1334 EXPORT_SYMBOL(blk_put_request
);
1337 * blk_add_request_payload - add a payload to a request
1338 * @rq: request to update
1339 * @page: page backing the payload
1340 * @len: length of the payload.
1342 * This allows to later add a payload to an already submitted request by
1343 * a block driver. The driver needs to take care of freeing the payload
1346 * Note that this is a quite horrible hack and nothing but handling of
1347 * discard requests should ever use it.
1349 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1352 struct bio
*bio
= rq
->bio
;
1354 bio
->bi_io_vec
->bv_page
= page
;
1355 bio
->bi_io_vec
->bv_offset
= 0;
1356 bio
->bi_io_vec
->bv_len
= len
;
1358 bio
->bi_iter
.bi_size
= len
;
1360 bio
->bi_phys_segments
= 1;
1362 rq
->__data_len
= rq
->resid_len
= len
;
1363 rq
->nr_phys_segments
= 1;
1364 rq
->buffer
= bio_data(bio
);
1366 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1368 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1371 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1373 if (!ll_back_merge_fn(q
, req
, bio
))
1376 trace_block_bio_backmerge(q
, req
, bio
);
1378 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1379 blk_rq_set_mixed_merge(req
);
1381 req
->biotail
->bi_next
= bio
;
1383 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1384 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1386 blk_account_io_start(req
, false);
1390 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1393 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1395 if (!ll_front_merge_fn(q
, req
, bio
))
1398 trace_block_bio_frontmerge(q
, req
, bio
);
1400 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1401 blk_rq_set_mixed_merge(req
);
1403 bio
->bi_next
= req
->bio
;
1407 * may not be valid. if the low level driver said
1408 * it didn't need a bounce buffer then it better
1409 * not touch req->buffer either...
1411 req
->buffer
= bio_data(bio
);
1412 req
->__sector
= bio
->bi_iter
.bi_sector
;
1413 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1414 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1416 blk_account_io_start(req
, false);
1421 * blk_attempt_plug_merge - try to merge with %current's plugged list
1422 * @q: request_queue new bio is being queued at
1423 * @bio: new bio being queued
1424 * @request_count: out parameter for number of traversed plugged requests
1426 * Determine whether @bio being queued on @q can be merged with a request
1427 * on %current's plugged list. Returns %true if merge was successful,
1430 * Plugging coalesces IOs from the same issuer for the same purpose without
1431 * going through @q->queue_lock. As such it's more of an issuing mechanism
1432 * than scheduling, and the request, while may have elvpriv data, is not
1433 * added on the elevator at this point. In addition, we don't have
1434 * reliable access to the elevator outside queue lock. Only check basic
1435 * merging parameters without querying the elevator.
1437 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1438 unsigned int *request_count
)
1440 struct blk_plug
*plug
;
1443 struct list_head
*plug_list
;
1445 if (blk_queue_nomerges(q
))
1448 plug
= current
->plug
;
1454 plug_list
= &plug
->mq_list
;
1456 plug_list
= &plug
->list
;
1458 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1464 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1467 el_ret
= blk_try_merge(rq
, bio
);
1468 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1469 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1472 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1473 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1482 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1484 req
->cmd_type
= REQ_TYPE_FS
;
1486 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1487 if (bio
->bi_rw
& REQ_RAHEAD
)
1488 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1491 req
->__sector
= bio
->bi_iter
.bi_sector
;
1492 req
->ioprio
= bio_prio(bio
);
1493 blk_rq_bio_prep(req
->q
, req
, bio
);
1496 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1498 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1499 struct blk_plug
*plug
;
1500 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1501 struct request
*req
;
1502 unsigned int request_count
= 0;
1505 * low level driver can indicate that it wants pages above a
1506 * certain limit bounced to low memory (ie for highmem, or even
1507 * ISA dma in theory)
1509 blk_queue_bounce(q
, &bio
);
1511 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1512 bio_endio(bio
, -EIO
);
1516 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1517 spin_lock_irq(q
->queue_lock
);
1518 where
= ELEVATOR_INSERT_FLUSH
;
1523 * Check if we can merge with the plugged list before grabbing
1526 if (blk_attempt_plug_merge(q
, bio
, &request_count
))
1529 spin_lock_irq(q
->queue_lock
);
1531 el_ret
= elv_merge(q
, &req
, bio
);
1532 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1533 if (bio_attempt_back_merge(q
, req
, bio
)) {
1534 elv_bio_merged(q
, req
, bio
);
1535 if (!attempt_back_merge(q
, req
))
1536 elv_merged_request(q
, req
, el_ret
);
1539 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1540 if (bio_attempt_front_merge(q
, req
, bio
)) {
1541 elv_bio_merged(q
, req
, bio
);
1542 if (!attempt_front_merge(q
, req
))
1543 elv_merged_request(q
, req
, el_ret
);
1550 * This sync check and mask will be re-done in init_request_from_bio(),
1551 * but we need to set it earlier to expose the sync flag to the
1552 * rq allocator and io schedulers.
1554 rw_flags
= bio_data_dir(bio
);
1556 rw_flags
|= REQ_SYNC
;
1559 * Grab a free request. This is might sleep but can not fail.
1560 * Returns with the queue unlocked.
1562 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1563 if (unlikely(!req
)) {
1564 bio_endio(bio
, -ENODEV
); /* @q is dead */
1569 * After dropping the lock and possibly sleeping here, our request
1570 * may now be mergeable after it had proven unmergeable (above).
1571 * We don't worry about that case for efficiency. It won't happen
1572 * often, and the elevators are able to handle it.
1574 init_request_from_bio(req
, bio
);
1576 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1577 req
->cpu
= raw_smp_processor_id();
1579 plug
= current
->plug
;
1582 * If this is the first request added after a plug, fire
1586 trace_block_plug(q
);
1588 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1589 blk_flush_plug_list(plug
, false);
1590 trace_block_plug(q
);
1593 list_add_tail(&req
->queuelist
, &plug
->list
);
1594 blk_account_io_start(req
, true);
1596 spin_lock_irq(q
->queue_lock
);
1597 add_acct_request(q
, req
, where
);
1600 spin_unlock_irq(q
->queue_lock
);
1603 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1606 * If bio->bi_dev is a partition, remap the location
1608 static inline void blk_partition_remap(struct bio
*bio
)
1610 struct block_device
*bdev
= bio
->bi_bdev
;
1612 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1613 struct hd_struct
*p
= bdev
->bd_part
;
1615 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1616 bio
->bi_bdev
= bdev
->bd_contains
;
1618 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1620 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1624 static void handle_bad_sector(struct bio
*bio
)
1626 char b
[BDEVNAME_SIZE
];
1628 printk(KERN_INFO
"attempt to access beyond end of device\n");
1629 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1630 bdevname(bio
->bi_bdev
, b
),
1632 (unsigned long long)bio_end_sector(bio
),
1633 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1635 set_bit(BIO_EOF
, &bio
->bi_flags
);
1638 #ifdef CONFIG_FAIL_MAKE_REQUEST
1640 static DECLARE_FAULT_ATTR(fail_make_request
);
1642 static int __init
setup_fail_make_request(char *str
)
1644 return setup_fault_attr(&fail_make_request
, str
);
1646 __setup("fail_make_request=", setup_fail_make_request
);
1648 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1650 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1653 static int __init
fail_make_request_debugfs(void)
1655 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1656 NULL
, &fail_make_request
);
1658 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1661 late_initcall(fail_make_request_debugfs
);
1663 #else /* CONFIG_FAIL_MAKE_REQUEST */
1665 static inline bool should_fail_request(struct hd_struct
*part
,
1671 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1674 * Check whether this bio extends beyond the end of the device.
1676 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1683 /* Test device or partition size, when known. */
1684 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1686 sector_t sector
= bio
->bi_iter
.bi_sector
;
1688 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1690 * This may well happen - the kernel calls bread()
1691 * without checking the size of the device, e.g., when
1692 * mounting a device.
1694 handle_bad_sector(bio
);
1702 static noinline_for_stack
bool
1703 generic_make_request_checks(struct bio
*bio
)
1705 struct request_queue
*q
;
1706 int nr_sectors
= bio_sectors(bio
);
1708 char b
[BDEVNAME_SIZE
];
1709 struct hd_struct
*part
;
1713 if (bio_check_eod(bio
, nr_sectors
))
1716 q
= bdev_get_queue(bio
->bi_bdev
);
1719 "generic_make_request: Trying to access "
1720 "nonexistent block-device %s (%Lu)\n",
1721 bdevname(bio
->bi_bdev
, b
),
1722 (long long) bio
->bi_iter
.bi_sector
);
1726 if (likely(bio_is_rw(bio
) &&
1727 nr_sectors
> queue_max_hw_sectors(q
))) {
1728 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1729 bdevname(bio
->bi_bdev
, b
),
1731 queue_max_hw_sectors(q
));
1735 part
= bio
->bi_bdev
->bd_part
;
1736 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1737 should_fail_request(&part_to_disk(part
)->part0
,
1738 bio
->bi_iter
.bi_size
))
1742 * If this device has partitions, remap block n
1743 * of partition p to block n+start(p) of the disk.
1745 blk_partition_remap(bio
);
1747 if (bio_check_eod(bio
, nr_sectors
))
1751 * Filter flush bio's early so that make_request based
1752 * drivers without flush support don't have to worry
1755 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1756 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1763 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1764 (!blk_queue_discard(q
) ||
1765 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1770 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1776 * Various block parts want %current->io_context and lazy ioc
1777 * allocation ends up trading a lot of pain for a small amount of
1778 * memory. Just allocate it upfront. This may fail and block
1779 * layer knows how to live with it.
1781 create_io_context(GFP_ATOMIC
, q
->node
);
1783 if (blk_throtl_bio(q
, bio
))
1784 return false; /* throttled, will be resubmitted later */
1786 trace_block_bio_queue(q
, bio
);
1790 bio_endio(bio
, err
);
1795 * generic_make_request - hand a buffer to its device driver for I/O
1796 * @bio: The bio describing the location in memory and on the device.
1798 * generic_make_request() is used to make I/O requests of block
1799 * devices. It is passed a &struct bio, which describes the I/O that needs
1802 * generic_make_request() does not return any status. The
1803 * success/failure status of the request, along with notification of
1804 * completion, is delivered asynchronously through the bio->bi_end_io
1805 * function described (one day) else where.
1807 * The caller of generic_make_request must make sure that bi_io_vec
1808 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1809 * set to describe the device address, and the
1810 * bi_end_io and optionally bi_private are set to describe how
1811 * completion notification should be signaled.
1813 * generic_make_request and the drivers it calls may use bi_next if this
1814 * bio happens to be merged with someone else, and may resubmit the bio to
1815 * a lower device by calling into generic_make_request recursively, which
1816 * means the bio should NOT be touched after the call to ->make_request_fn.
1818 void generic_make_request(struct bio
*bio
)
1820 struct bio_list bio_list_on_stack
;
1822 if (!generic_make_request_checks(bio
))
1826 * We only want one ->make_request_fn to be active at a time, else
1827 * stack usage with stacked devices could be a problem. So use
1828 * current->bio_list to keep a list of requests submited by a
1829 * make_request_fn function. current->bio_list is also used as a
1830 * flag to say if generic_make_request is currently active in this
1831 * task or not. If it is NULL, then no make_request is active. If
1832 * it is non-NULL, then a make_request is active, and new requests
1833 * should be added at the tail
1835 if (current
->bio_list
) {
1836 bio_list_add(current
->bio_list
, bio
);
1840 /* following loop may be a bit non-obvious, and so deserves some
1842 * Before entering the loop, bio->bi_next is NULL (as all callers
1843 * ensure that) so we have a list with a single bio.
1844 * We pretend that we have just taken it off a longer list, so
1845 * we assign bio_list to a pointer to the bio_list_on_stack,
1846 * thus initialising the bio_list of new bios to be
1847 * added. ->make_request() may indeed add some more bios
1848 * through a recursive call to generic_make_request. If it
1849 * did, we find a non-NULL value in bio_list and re-enter the loop
1850 * from the top. In this case we really did just take the bio
1851 * of the top of the list (no pretending) and so remove it from
1852 * bio_list, and call into ->make_request() again.
1854 BUG_ON(bio
->bi_next
);
1855 bio_list_init(&bio_list_on_stack
);
1856 current
->bio_list
= &bio_list_on_stack
;
1858 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1860 q
->make_request_fn(q
, bio
);
1862 bio
= bio_list_pop(current
->bio_list
);
1864 current
->bio_list
= NULL
; /* deactivate */
1866 EXPORT_SYMBOL(generic_make_request
);
1869 * submit_bio - submit a bio to the block device layer for I/O
1870 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1871 * @bio: The &struct bio which describes the I/O
1873 * submit_bio() is very similar in purpose to generic_make_request(), and
1874 * uses that function to do most of the work. Both are fairly rough
1875 * interfaces; @bio must be presetup and ready for I/O.
1878 void submit_bio(int rw
, struct bio
*bio
)
1883 * If it's a regular read/write or a barrier with data attached,
1884 * go through the normal accounting stuff before submission.
1886 if (bio_has_data(bio
)) {
1889 if (unlikely(rw
& REQ_WRITE_SAME
))
1890 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
1892 count
= bio_sectors(bio
);
1895 count_vm_events(PGPGOUT
, count
);
1897 task_io_account_read(bio
->bi_iter
.bi_size
);
1898 count_vm_events(PGPGIN
, count
);
1901 if (unlikely(block_dump
)) {
1902 char b
[BDEVNAME_SIZE
];
1903 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1904 current
->comm
, task_pid_nr(current
),
1905 (rw
& WRITE
) ? "WRITE" : "READ",
1906 (unsigned long long)bio
->bi_iter
.bi_sector
,
1907 bdevname(bio
->bi_bdev
, b
),
1912 generic_make_request(bio
);
1914 EXPORT_SYMBOL(submit_bio
);
1917 * blk_rq_check_limits - Helper function to check a request for the queue limit
1919 * @rq: the request being checked
1922 * @rq may have been made based on weaker limitations of upper-level queues
1923 * in request stacking drivers, and it may violate the limitation of @q.
1924 * Since the block layer and the underlying device driver trust @rq
1925 * after it is inserted to @q, it should be checked against @q before
1926 * the insertion using this generic function.
1928 * This function should also be useful for request stacking drivers
1929 * in some cases below, so export this function.
1930 * Request stacking drivers like request-based dm may change the queue
1931 * limits while requests are in the queue (e.g. dm's table swapping).
1932 * Such request stacking drivers should check those requests agaist
1933 * the new queue limits again when they dispatch those requests,
1934 * although such checkings are also done against the old queue limits
1935 * when submitting requests.
1937 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1939 if (!rq_mergeable(rq
))
1942 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
1943 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1948 * queue's settings related to segment counting like q->bounce_pfn
1949 * may differ from that of other stacking queues.
1950 * Recalculate it to check the request correctly on this queue's
1953 blk_recalc_rq_segments(rq
);
1954 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1955 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1961 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1964 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1965 * @q: the queue to submit the request
1966 * @rq: the request being queued
1968 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1970 unsigned long flags
;
1971 int where
= ELEVATOR_INSERT_BACK
;
1973 if (blk_rq_check_limits(q
, rq
))
1977 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1980 spin_lock_irqsave(q
->queue_lock
, flags
);
1981 if (unlikely(blk_queue_dying(q
))) {
1982 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1987 * Submitting request must be dequeued before calling this function
1988 * because it will be linked to another request_queue
1990 BUG_ON(blk_queued_rq(rq
));
1992 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1993 where
= ELEVATOR_INSERT_FLUSH
;
1995 add_acct_request(q
, rq
, where
);
1996 if (where
== ELEVATOR_INSERT_FLUSH
)
1998 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2002 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2005 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2006 * @rq: request to examine
2009 * A request could be merge of IOs which require different failure
2010 * handling. This function determines the number of bytes which
2011 * can be failed from the beginning of the request without
2012 * crossing into area which need to be retried further.
2015 * The number of bytes to fail.
2018 * queue_lock must be held.
2020 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2022 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2023 unsigned int bytes
= 0;
2026 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2027 return blk_rq_bytes(rq
);
2030 * Currently the only 'mixing' which can happen is between
2031 * different fastfail types. We can safely fail portions
2032 * which have all the failfast bits that the first one has -
2033 * the ones which are at least as eager to fail as the first
2036 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2037 if ((bio
->bi_rw
& ff
) != ff
)
2039 bytes
+= bio
->bi_iter
.bi_size
;
2042 /* this could lead to infinite loop */
2043 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2046 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2048 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2050 if (blk_do_io_stat(req
)) {
2051 const int rw
= rq_data_dir(req
);
2052 struct hd_struct
*part
;
2055 cpu
= part_stat_lock();
2057 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2062 void blk_account_io_done(struct request
*req
)
2065 * Account IO completion. flush_rq isn't accounted as a
2066 * normal IO on queueing nor completion. Accounting the
2067 * containing request is enough.
2069 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2070 unsigned long duration
= jiffies
- req
->start_time
;
2071 const int rw
= rq_data_dir(req
);
2072 struct hd_struct
*part
;
2075 cpu
= part_stat_lock();
2078 part_stat_inc(cpu
, part
, ios
[rw
]);
2079 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2080 part_round_stats(cpu
, part
);
2081 part_dec_in_flight(part
, rw
);
2083 hd_struct_put(part
);
2088 #ifdef CONFIG_PM_RUNTIME
2090 * Don't process normal requests when queue is suspended
2091 * or in the process of suspending/resuming
2093 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2096 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2097 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2103 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2110 void blk_account_io_start(struct request
*rq
, bool new_io
)
2112 struct hd_struct
*part
;
2113 int rw
= rq_data_dir(rq
);
2116 if (!blk_do_io_stat(rq
))
2119 cpu
= part_stat_lock();
2123 part_stat_inc(cpu
, part
, merges
[rw
]);
2125 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2126 if (!hd_struct_try_get(part
)) {
2128 * The partition is already being removed,
2129 * the request will be accounted on the disk only
2131 * We take a reference on disk->part0 although that
2132 * partition will never be deleted, so we can treat
2133 * it as any other partition.
2135 part
= &rq
->rq_disk
->part0
;
2136 hd_struct_get(part
);
2138 part_round_stats(cpu
, part
);
2139 part_inc_in_flight(part
, rw
);
2147 * blk_peek_request - peek at the top of a request queue
2148 * @q: request queue to peek at
2151 * Return the request at the top of @q. The returned request
2152 * should be started using blk_start_request() before LLD starts
2156 * Pointer to the request at the top of @q if available. Null
2160 * queue_lock must be held.
2162 struct request
*blk_peek_request(struct request_queue
*q
)
2167 while ((rq
= __elv_next_request(q
)) != NULL
) {
2169 rq
= blk_pm_peek_request(q
, rq
);
2173 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2175 * This is the first time the device driver
2176 * sees this request (possibly after
2177 * requeueing). Notify IO scheduler.
2179 if (rq
->cmd_flags
& REQ_SORTED
)
2180 elv_activate_rq(q
, rq
);
2183 * just mark as started even if we don't start
2184 * it, a request that has been delayed should
2185 * not be passed by new incoming requests
2187 rq
->cmd_flags
|= REQ_STARTED
;
2188 trace_block_rq_issue(q
, rq
);
2191 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2192 q
->end_sector
= rq_end_sector(rq
);
2193 q
->boundary_rq
= NULL
;
2196 if (rq
->cmd_flags
& REQ_DONTPREP
)
2199 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2201 * make sure space for the drain appears we
2202 * know we can do this because max_hw_segments
2203 * has been adjusted to be one fewer than the
2206 rq
->nr_phys_segments
++;
2212 ret
= q
->prep_rq_fn(q
, rq
);
2213 if (ret
== BLKPREP_OK
) {
2215 } else if (ret
== BLKPREP_DEFER
) {
2217 * the request may have been (partially) prepped.
2218 * we need to keep this request in the front to
2219 * avoid resource deadlock. REQ_STARTED will
2220 * prevent other fs requests from passing this one.
2222 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2223 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2225 * remove the space for the drain we added
2226 * so that we don't add it again
2228 --rq
->nr_phys_segments
;
2233 } else if (ret
== BLKPREP_KILL
) {
2234 rq
->cmd_flags
|= REQ_QUIET
;
2236 * Mark this request as started so we don't trigger
2237 * any debug logic in the end I/O path.
2239 blk_start_request(rq
);
2240 __blk_end_request_all(rq
, -EIO
);
2242 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2249 EXPORT_SYMBOL(blk_peek_request
);
2251 void blk_dequeue_request(struct request
*rq
)
2253 struct request_queue
*q
= rq
->q
;
2255 BUG_ON(list_empty(&rq
->queuelist
));
2256 BUG_ON(ELV_ON_HASH(rq
));
2258 list_del_init(&rq
->queuelist
);
2261 * the time frame between a request being removed from the lists
2262 * and to it is freed is accounted as io that is in progress at
2265 if (blk_account_rq(rq
)) {
2266 q
->in_flight
[rq_is_sync(rq
)]++;
2267 set_io_start_time_ns(rq
);
2272 * blk_start_request - start request processing on the driver
2273 * @req: request to dequeue
2276 * Dequeue @req and start timeout timer on it. This hands off the
2277 * request to the driver.
2279 * Block internal functions which don't want to start timer should
2280 * call blk_dequeue_request().
2283 * queue_lock must be held.
2285 void blk_start_request(struct request
*req
)
2287 blk_dequeue_request(req
);
2290 * We are now handing the request to the hardware, initialize
2291 * resid_len to full count and add the timeout handler.
2293 req
->resid_len
= blk_rq_bytes(req
);
2294 if (unlikely(blk_bidi_rq(req
)))
2295 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2297 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2300 EXPORT_SYMBOL(blk_start_request
);
2303 * blk_fetch_request - fetch a request from a request queue
2304 * @q: request queue to fetch a request from
2307 * Return the request at the top of @q. The request is started on
2308 * return and LLD can start processing it immediately.
2311 * Pointer to the request at the top of @q if available. Null
2315 * queue_lock must be held.
2317 struct request
*blk_fetch_request(struct request_queue
*q
)
2321 rq
= blk_peek_request(q
);
2323 blk_start_request(rq
);
2326 EXPORT_SYMBOL(blk_fetch_request
);
2329 * blk_update_request - Special helper function for request stacking drivers
2330 * @req: the request being processed
2331 * @error: %0 for success, < %0 for error
2332 * @nr_bytes: number of bytes to complete @req
2335 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2336 * the request structure even if @req doesn't have leftover.
2337 * If @req has leftover, sets it up for the next range of segments.
2339 * This special helper function is only for request stacking drivers
2340 * (e.g. request-based dm) so that they can handle partial completion.
2341 * Actual device drivers should use blk_end_request instead.
2343 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2344 * %false return from this function.
2347 * %false - this request doesn't have any more data
2348 * %true - this request has more data
2350 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2357 trace_block_rq_complete(req
->q
, req
);
2360 * For fs requests, rq is just carrier of independent bio's
2361 * and each partial completion should be handled separately.
2362 * Reset per-request error on each partial completion.
2364 * TODO: tj: This is too subtle. It would be better to let
2365 * low level drivers do what they see fit.
2367 if (req
->cmd_type
== REQ_TYPE_FS
)
2370 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2371 !(req
->cmd_flags
& REQ_QUIET
)) {
2376 error_type
= "recoverable transport";
2379 error_type
= "critical target";
2382 error_type
= "critical nexus";
2385 error_type
= "timeout";
2388 error_type
= "critical space allocation";
2391 error_type
= "critical medium";
2398 printk_ratelimited(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2399 error_type
, req
->rq_disk
?
2400 req
->rq_disk
->disk_name
: "?",
2401 (unsigned long long)blk_rq_pos(req
));
2405 blk_account_io_completion(req
, nr_bytes
);
2409 struct bio
*bio
= req
->bio
;
2410 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2412 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2413 req
->bio
= bio
->bi_next
;
2415 req_bio_endio(req
, bio
, bio_bytes
, error
);
2417 total_bytes
+= bio_bytes
;
2418 nr_bytes
-= bio_bytes
;
2429 * Reset counters so that the request stacking driver
2430 * can find how many bytes remain in the request
2433 req
->__data_len
= 0;
2437 req
->__data_len
-= total_bytes
;
2438 req
->buffer
= bio_data(req
->bio
);
2440 /* update sector only for requests with clear definition of sector */
2441 if (req
->cmd_type
== REQ_TYPE_FS
)
2442 req
->__sector
+= total_bytes
>> 9;
2444 /* mixed attributes always follow the first bio */
2445 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2446 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2447 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2451 * If total number of sectors is less than the first segment
2452 * size, something has gone terribly wrong.
2454 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2455 blk_dump_rq_flags(req
, "request botched");
2456 req
->__data_len
= blk_rq_cur_bytes(req
);
2459 /* recalculate the number of segments */
2460 blk_recalc_rq_segments(req
);
2464 EXPORT_SYMBOL_GPL(blk_update_request
);
2466 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2467 unsigned int nr_bytes
,
2468 unsigned int bidi_bytes
)
2470 if (blk_update_request(rq
, error
, nr_bytes
))
2473 /* Bidi request must be completed as a whole */
2474 if (unlikely(blk_bidi_rq(rq
)) &&
2475 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2478 if (blk_queue_add_random(rq
->q
))
2479 add_disk_randomness(rq
->rq_disk
);
2485 * blk_unprep_request - unprepare a request
2488 * This function makes a request ready for complete resubmission (or
2489 * completion). It happens only after all error handling is complete,
2490 * so represents the appropriate moment to deallocate any resources
2491 * that were allocated to the request in the prep_rq_fn. The queue
2492 * lock is held when calling this.
2494 void blk_unprep_request(struct request
*req
)
2496 struct request_queue
*q
= req
->q
;
2498 req
->cmd_flags
&= ~REQ_DONTPREP
;
2499 if (q
->unprep_rq_fn
)
2500 q
->unprep_rq_fn(q
, req
);
2502 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2505 * queue lock must be held
2507 static void blk_finish_request(struct request
*req
, int error
)
2509 if (blk_rq_tagged(req
))
2510 blk_queue_end_tag(req
->q
, req
);
2512 BUG_ON(blk_queued_rq(req
));
2514 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2515 laptop_io_completion(&req
->q
->backing_dev_info
);
2517 blk_delete_timer(req
);
2519 if (req
->cmd_flags
& REQ_DONTPREP
)
2520 blk_unprep_request(req
);
2522 blk_account_io_done(req
);
2525 req
->end_io(req
, error
);
2527 if (blk_bidi_rq(req
))
2528 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2530 __blk_put_request(req
->q
, req
);
2535 * blk_end_bidi_request - Complete a bidi request
2536 * @rq: the request to complete
2537 * @error: %0 for success, < %0 for error
2538 * @nr_bytes: number of bytes to complete @rq
2539 * @bidi_bytes: number of bytes to complete @rq->next_rq
2542 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2543 * Drivers that supports bidi can safely call this member for any
2544 * type of request, bidi or uni. In the later case @bidi_bytes is
2548 * %false - we are done with this request
2549 * %true - still buffers pending for this request
2551 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2552 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2554 struct request_queue
*q
= rq
->q
;
2555 unsigned long flags
;
2557 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2560 spin_lock_irqsave(q
->queue_lock
, flags
);
2561 blk_finish_request(rq
, error
);
2562 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2568 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2569 * @rq: the request to complete
2570 * @error: %0 for success, < %0 for error
2571 * @nr_bytes: number of bytes to complete @rq
2572 * @bidi_bytes: number of bytes to complete @rq->next_rq
2575 * Identical to blk_end_bidi_request() except that queue lock is
2576 * assumed to be locked on entry and remains so on return.
2579 * %false - we are done with this request
2580 * %true - still buffers pending for this request
2582 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2583 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2585 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2588 blk_finish_request(rq
, error
);
2594 * blk_end_request - Helper function for drivers to complete the request.
2595 * @rq: the request being processed
2596 * @error: %0 for success, < %0 for error
2597 * @nr_bytes: number of bytes to complete
2600 * Ends I/O on a number of bytes attached to @rq.
2601 * If @rq has leftover, sets it up for the next range of segments.
2604 * %false - we are done with this request
2605 * %true - still buffers pending for this request
2607 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2609 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2611 EXPORT_SYMBOL(blk_end_request
);
2614 * blk_end_request_all - Helper function for drives to finish the request.
2615 * @rq: the request to finish
2616 * @error: %0 for success, < %0 for error
2619 * Completely finish @rq.
2621 void blk_end_request_all(struct request
*rq
, int error
)
2624 unsigned int bidi_bytes
= 0;
2626 if (unlikely(blk_bidi_rq(rq
)))
2627 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2629 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2632 EXPORT_SYMBOL(blk_end_request_all
);
2635 * blk_end_request_cur - Helper function to finish the current request chunk.
2636 * @rq: the request to finish the current chunk for
2637 * @error: %0 for success, < %0 for error
2640 * Complete the current consecutively mapped chunk from @rq.
2643 * %false - we are done with this request
2644 * %true - still buffers pending for this request
2646 bool blk_end_request_cur(struct request
*rq
, int error
)
2648 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2650 EXPORT_SYMBOL(blk_end_request_cur
);
2653 * blk_end_request_err - Finish a request till the next failure boundary.
2654 * @rq: the request to finish till the next failure boundary for
2655 * @error: must be negative errno
2658 * Complete @rq till the next failure boundary.
2661 * %false - we are done with this request
2662 * %true - still buffers pending for this request
2664 bool blk_end_request_err(struct request
*rq
, int error
)
2666 WARN_ON(error
>= 0);
2667 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2669 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2672 * __blk_end_request - Helper function for drivers to complete the request.
2673 * @rq: the request being processed
2674 * @error: %0 for success, < %0 for error
2675 * @nr_bytes: number of bytes to complete
2678 * Must be called with queue lock held unlike blk_end_request().
2681 * %false - we are done with this request
2682 * %true - still buffers pending for this request
2684 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2686 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2688 EXPORT_SYMBOL(__blk_end_request
);
2691 * __blk_end_request_all - Helper function for drives to finish the request.
2692 * @rq: the request to finish
2693 * @error: %0 for success, < %0 for error
2696 * Completely finish @rq. Must be called with queue lock held.
2698 void __blk_end_request_all(struct request
*rq
, int error
)
2701 unsigned int bidi_bytes
= 0;
2703 if (unlikely(blk_bidi_rq(rq
)))
2704 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2706 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2709 EXPORT_SYMBOL(__blk_end_request_all
);
2712 * __blk_end_request_cur - Helper function to finish the current request chunk.
2713 * @rq: the request to finish the current chunk for
2714 * @error: %0 for success, < %0 for error
2717 * Complete the current consecutively mapped chunk from @rq. Must
2718 * be called with queue lock held.
2721 * %false - we are done with this request
2722 * %true - still buffers pending for this request
2724 bool __blk_end_request_cur(struct request
*rq
, int error
)
2726 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2728 EXPORT_SYMBOL(__blk_end_request_cur
);
2731 * __blk_end_request_err - Finish a request till the next failure boundary.
2732 * @rq: the request to finish till the next failure boundary for
2733 * @error: must be negative errno
2736 * Complete @rq till the next failure boundary. Must be called
2737 * with queue lock held.
2740 * %false - we are done with this request
2741 * %true - still buffers pending for this request
2743 bool __blk_end_request_err(struct request
*rq
, int error
)
2745 WARN_ON(error
>= 0);
2746 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2748 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2750 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2753 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2754 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2756 if (bio_has_data(bio
)) {
2757 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2758 rq
->buffer
= bio_data(bio
);
2760 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2761 rq
->bio
= rq
->biotail
= bio
;
2764 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2767 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2769 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2770 * @rq: the request to be flushed
2773 * Flush all pages in @rq.
2775 void rq_flush_dcache_pages(struct request
*rq
)
2777 struct req_iterator iter
;
2778 struct bio_vec bvec
;
2780 rq_for_each_segment(bvec
, rq
, iter
)
2781 flush_dcache_page(bvec
.bv_page
);
2783 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2787 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2788 * @q : the queue of the device being checked
2791 * Check if underlying low-level drivers of a device are busy.
2792 * If the drivers want to export their busy state, they must set own
2793 * exporting function using blk_queue_lld_busy() first.
2795 * Basically, this function is used only by request stacking drivers
2796 * to stop dispatching requests to underlying devices when underlying
2797 * devices are busy. This behavior helps more I/O merging on the queue
2798 * of the request stacking driver and prevents I/O throughput regression
2799 * on burst I/O load.
2802 * 0 - Not busy (The request stacking driver should dispatch request)
2803 * 1 - Busy (The request stacking driver should stop dispatching request)
2805 int blk_lld_busy(struct request_queue
*q
)
2808 return q
->lld_busy_fn(q
);
2812 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2815 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2816 * @rq: the clone request to be cleaned up
2819 * Free all bios in @rq for a cloned request.
2821 void blk_rq_unprep_clone(struct request
*rq
)
2825 while ((bio
= rq
->bio
) != NULL
) {
2826 rq
->bio
= bio
->bi_next
;
2831 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2834 * Copy attributes of the original request to the clone request.
2835 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2837 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2839 dst
->cpu
= src
->cpu
;
2840 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2841 dst
->cmd_type
= src
->cmd_type
;
2842 dst
->__sector
= blk_rq_pos(src
);
2843 dst
->__data_len
= blk_rq_bytes(src
);
2844 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2845 dst
->ioprio
= src
->ioprio
;
2846 dst
->extra_len
= src
->extra_len
;
2850 * blk_rq_prep_clone - Helper function to setup clone request
2851 * @rq: the request to be setup
2852 * @rq_src: original request to be cloned
2853 * @bs: bio_set that bios for clone are allocated from
2854 * @gfp_mask: memory allocation mask for bio
2855 * @bio_ctr: setup function to be called for each clone bio.
2856 * Returns %0 for success, non %0 for failure.
2857 * @data: private data to be passed to @bio_ctr
2860 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2861 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2862 * are not copied, and copying such parts is the caller's responsibility.
2863 * Also, pages which the original bios are pointing to are not copied
2864 * and the cloned bios just point same pages.
2865 * So cloned bios must be completed before original bios, which means
2866 * the caller must complete @rq before @rq_src.
2868 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2869 struct bio_set
*bs
, gfp_t gfp_mask
,
2870 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2873 struct bio
*bio
, *bio_src
;
2878 blk_rq_init(NULL
, rq
);
2880 __rq_for_each_bio(bio_src
, rq_src
) {
2881 bio
= bio_clone_bioset(bio_src
, gfp_mask
, bs
);
2885 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2889 rq
->biotail
->bi_next
= bio
;
2892 rq
->bio
= rq
->biotail
= bio
;
2895 __blk_rq_prep_clone(rq
, rq_src
);
2902 blk_rq_unprep_clone(rq
);
2906 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2908 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2910 return queue_work(kblockd_workqueue
, work
);
2912 EXPORT_SYMBOL(kblockd_schedule_work
);
2914 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2915 struct delayed_work
*dwork
, unsigned long delay
)
2917 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2919 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2921 #define PLUG_MAGIC 0x91827364
2924 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2925 * @plug: The &struct blk_plug that needs to be initialized
2928 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2929 * pending I/O should the task end up blocking between blk_start_plug() and
2930 * blk_finish_plug(). This is important from a performance perspective, but
2931 * also ensures that we don't deadlock. For instance, if the task is blocking
2932 * for a memory allocation, memory reclaim could end up wanting to free a
2933 * page belonging to that request that is currently residing in our private
2934 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2935 * this kind of deadlock.
2937 void blk_start_plug(struct blk_plug
*plug
)
2939 struct task_struct
*tsk
= current
;
2941 plug
->magic
= PLUG_MAGIC
;
2942 INIT_LIST_HEAD(&plug
->list
);
2943 INIT_LIST_HEAD(&plug
->mq_list
);
2944 INIT_LIST_HEAD(&plug
->cb_list
);
2947 * If this is a nested plug, don't actually assign it. It will be
2948 * flushed on its own.
2952 * Store ordering should not be needed here, since a potential
2953 * preempt will imply a full memory barrier
2958 EXPORT_SYMBOL(blk_start_plug
);
2960 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2962 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2963 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2965 return !(rqa
->q
< rqb
->q
||
2966 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
2970 * If 'from_schedule' is true, then postpone the dispatch of requests
2971 * until a safe kblockd context. We due this to avoid accidental big
2972 * additional stack usage in driver dispatch, in places where the originally
2973 * plugger did not intend it.
2975 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2977 __releases(q
->queue_lock
)
2979 trace_block_unplug(q
, depth
, !from_schedule
);
2982 blk_run_queue_async(q
);
2985 spin_unlock(q
->queue_lock
);
2988 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
2990 LIST_HEAD(callbacks
);
2992 while (!list_empty(&plug
->cb_list
)) {
2993 list_splice_init(&plug
->cb_list
, &callbacks
);
2995 while (!list_empty(&callbacks
)) {
2996 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2999 list_del(&cb
->list
);
3000 cb
->callback(cb
, from_schedule
);
3005 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3008 struct blk_plug
*plug
= current
->plug
;
3009 struct blk_plug_cb
*cb
;
3014 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3015 if (cb
->callback
== unplug
&& cb
->data
== data
)
3018 /* Not currently on the callback list */
3019 BUG_ON(size
< sizeof(*cb
));
3020 cb
= kzalloc(size
, GFP_ATOMIC
);
3023 cb
->callback
= unplug
;
3024 list_add(&cb
->list
, &plug
->cb_list
);
3028 EXPORT_SYMBOL(blk_check_plugged
);
3030 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3032 struct request_queue
*q
;
3033 unsigned long flags
;
3038 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
3040 flush_plug_callbacks(plug
, from_schedule
);
3042 if (!list_empty(&plug
->mq_list
))
3043 blk_mq_flush_plug_list(plug
, from_schedule
);
3045 if (list_empty(&plug
->list
))
3048 list_splice_init(&plug
->list
, &list
);
3050 list_sort(NULL
, &list
, plug_rq_cmp
);
3056 * Save and disable interrupts here, to avoid doing it for every
3057 * queue lock we have to take.
3059 local_irq_save(flags
);
3060 while (!list_empty(&list
)) {
3061 rq
= list_entry_rq(list
.next
);
3062 list_del_init(&rq
->queuelist
);
3066 * This drops the queue lock
3069 queue_unplugged(q
, depth
, from_schedule
);
3072 spin_lock(q
->queue_lock
);
3076 * Short-circuit if @q is dead
3078 if (unlikely(blk_queue_dying(q
))) {
3079 __blk_end_request_all(rq
, -ENODEV
);
3084 * rq is already accounted, so use raw insert
3086 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3087 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3089 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3095 * This drops the queue lock
3098 queue_unplugged(q
, depth
, from_schedule
);
3100 local_irq_restore(flags
);
3103 void blk_finish_plug(struct blk_plug
*plug
)
3105 blk_flush_plug_list(plug
, false);
3107 if (plug
== current
->plug
)
3108 current
->plug
= NULL
;
3110 EXPORT_SYMBOL(blk_finish_plug
);
3112 #ifdef CONFIG_PM_RUNTIME
3114 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3115 * @q: the queue of the device
3116 * @dev: the device the queue belongs to
3119 * Initialize runtime-PM-related fields for @q and start auto suspend for
3120 * @dev. Drivers that want to take advantage of request-based runtime PM
3121 * should call this function after @dev has been initialized, and its
3122 * request queue @q has been allocated, and runtime PM for it can not happen
3123 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3124 * cases, driver should call this function before any I/O has taken place.
3126 * This function takes care of setting up using auto suspend for the device,
3127 * the autosuspend delay is set to -1 to make runtime suspend impossible
3128 * until an updated value is either set by user or by driver. Drivers do
3129 * not need to touch other autosuspend settings.
3131 * The block layer runtime PM is request based, so only works for drivers
3132 * that use request as their IO unit instead of those directly use bio's.
3134 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3137 q
->rpm_status
= RPM_ACTIVE
;
3138 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3139 pm_runtime_use_autosuspend(q
->dev
);
3141 EXPORT_SYMBOL(blk_pm_runtime_init
);
3144 * blk_pre_runtime_suspend - Pre runtime suspend check
3145 * @q: the queue of the device
3148 * This function will check if runtime suspend is allowed for the device
3149 * by examining if there are any requests pending in the queue. If there
3150 * are requests pending, the device can not be runtime suspended; otherwise,
3151 * the queue's status will be updated to SUSPENDING and the driver can
3152 * proceed to suspend the device.
3154 * For the not allowed case, we mark last busy for the device so that
3155 * runtime PM core will try to autosuspend it some time later.
3157 * This function should be called near the start of the device's
3158 * runtime_suspend callback.
3161 * 0 - OK to runtime suspend the device
3162 * -EBUSY - Device should not be runtime suspended
3164 int blk_pre_runtime_suspend(struct request_queue
*q
)
3168 spin_lock_irq(q
->queue_lock
);
3169 if (q
->nr_pending
) {
3171 pm_runtime_mark_last_busy(q
->dev
);
3173 q
->rpm_status
= RPM_SUSPENDING
;
3175 spin_unlock_irq(q
->queue_lock
);
3178 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3181 * blk_post_runtime_suspend - Post runtime suspend processing
3182 * @q: the queue of the device
3183 * @err: return value of the device's runtime_suspend function
3186 * Update the queue's runtime status according to the return value of the
3187 * device's runtime suspend function and mark last busy for the device so
3188 * that PM core will try to auto suspend the device at a later time.
3190 * This function should be called near the end of the device's
3191 * runtime_suspend callback.
3193 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3195 spin_lock_irq(q
->queue_lock
);
3197 q
->rpm_status
= RPM_SUSPENDED
;
3199 q
->rpm_status
= RPM_ACTIVE
;
3200 pm_runtime_mark_last_busy(q
->dev
);
3202 spin_unlock_irq(q
->queue_lock
);
3204 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3207 * blk_pre_runtime_resume - Pre runtime resume processing
3208 * @q: the queue of the device
3211 * Update the queue's runtime status to RESUMING in preparation for the
3212 * runtime resume of the device.
3214 * This function should be called near the start of the device's
3215 * runtime_resume callback.
3217 void blk_pre_runtime_resume(struct request_queue
*q
)
3219 spin_lock_irq(q
->queue_lock
);
3220 q
->rpm_status
= RPM_RESUMING
;
3221 spin_unlock_irq(q
->queue_lock
);
3223 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3226 * blk_post_runtime_resume - Post runtime resume processing
3227 * @q: the queue of the device
3228 * @err: return value of the device's runtime_resume function
3231 * Update the queue's runtime status according to the return value of the
3232 * device's runtime_resume function. If it is successfully resumed, process
3233 * the requests that are queued into the device's queue when it is resuming
3234 * and then mark last busy and initiate autosuspend for it.
3236 * This function should be called near the end of the device's
3237 * runtime_resume callback.
3239 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3241 spin_lock_irq(q
->queue_lock
);
3243 q
->rpm_status
= RPM_ACTIVE
;
3245 pm_runtime_mark_last_busy(q
->dev
);
3246 pm_request_autosuspend(q
->dev
);
3248 q
->rpm_status
= RPM_SUSPENDED
;
3250 spin_unlock_irq(q
->queue_lock
);
3252 EXPORT_SYMBOL(blk_post_runtime_resume
);
3255 int __init
blk_dev_init(void)
3257 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3258 sizeof(((struct request
*)0)->cmd_flags
));
3260 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3261 kblockd_workqueue
= alloc_workqueue("kblockd",
3262 WQ_MEM_RECLAIM
| WQ_HIGHPRI
|
3263 WQ_POWER_EFFICIENT
, 0);
3264 if (!kblockd_workqueue
)
3265 panic("Failed to create kblockd\n");
3267 request_cachep
= kmem_cache_create("blkdev_requests",
3268 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3270 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3271 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);