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, len %u\n",
150 rq
->bio
, rq
->biotail
, 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 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
698 blk_cleanup_queue(uninit_q
);
702 EXPORT_SYMBOL(blk_init_queue_node
);
704 struct request_queue
*
705 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
711 q
->flush_rq
= kzalloc(sizeof(struct request
), GFP_KERNEL
);
715 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
719 q
->prep_rq_fn
= NULL
;
720 q
->unprep_rq_fn
= NULL
;
721 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
723 /* Override internal queue lock with supplied lock pointer */
725 q
->queue_lock
= lock
;
728 * This also sets hw/phys segments, boundary and size
730 blk_queue_make_request(q
, blk_queue_bio
);
732 q
->sg_reserved_size
= INT_MAX
;
734 /* Protect q->elevator from elevator_change */
735 mutex_lock(&q
->sysfs_lock
);
738 if (elevator_init(q
, NULL
)) {
739 mutex_unlock(&q
->sysfs_lock
);
743 mutex_unlock(&q
->sysfs_lock
);
751 EXPORT_SYMBOL(blk_init_allocated_queue
);
753 bool blk_get_queue(struct request_queue
*q
)
755 if (likely(!blk_queue_dying(q
))) {
762 EXPORT_SYMBOL(blk_get_queue
);
764 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
766 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
767 elv_put_request(rl
->q
, rq
);
769 put_io_context(rq
->elv
.icq
->ioc
);
772 mempool_free(rq
, rl
->rq_pool
);
776 * ioc_batching returns true if the ioc is a valid batching request and
777 * should be given priority access to a request.
779 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
785 * Make sure the process is able to allocate at least 1 request
786 * even if the batch times out, otherwise we could theoretically
789 return ioc
->nr_batch_requests
== q
->nr_batching
||
790 (ioc
->nr_batch_requests
> 0
791 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
795 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
796 * will cause the process to be a "batcher" on all queues in the system. This
797 * is the behaviour we want though - once it gets a wakeup it should be given
800 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
802 if (!ioc
|| ioc_batching(q
, ioc
))
805 ioc
->nr_batch_requests
= q
->nr_batching
;
806 ioc
->last_waited
= jiffies
;
809 static void __freed_request(struct request_list
*rl
, int sync
)
811 struct request_queue
*q
= rl
->q
;
814 * bdi isn't aware of blkcg yet. As all async IOs end up root
815 * blkcg anyway, just use root blkcg state.
817 if (rl
== &q
->root_rl
&&
818 rl
->count
[sync
] < queue_congestion_off_threshold(q
))
819 blk_clear_queue_congested(q
, sync
);
821 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
822 if (waitqueue_active(&rl
->wait
[sync
]))
823 wake_up(&rl
->wait
[sync
]);
825 blk_clear_rl_full(rl
, sync
);
830 * A request has just been released. Account for it, update the full and
831 * congestion status, wake up any waiters. Called under q->queue_lock.
833 static void freed_request(struct request_list
*rl
, unsigned int flags
)
835 struct request_queue
*q
= rl
->q
;
836 int sync
= rw_is_sync(flags
);
840 if (flags
& REQ_ELVPRIV
)
843 __freed_request(rl
, sync
);
845 if (unlikely(rl
->starved
[sync
^ 1]))
846 __freed_request(rl
, sync
^ 1);
850 * Determine if elevator data should be initialized when allocating the
851 * request associated with @bio.
853 static bool blk_rq_should_init_elevator(struct bio
*bio
)
859 * Flush requests do not use the elevator so skip initialization.
860 * This allows a request to share the flush and elevator data.
862 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
869 * rq_ioc - determine io_context for request allocation
870 * @bio: request being allocated is for this bio (can be %NULL)
872 * Determine io_context to use for request allocation for @bio. May return
873 * %NULL if %current->io_context doesn't exist.
875 static struct io_context
*rq_ioc(struct bio
*bio
)
877 #ifdef CONFIG_BLK_CGROUP
878 if (bio
&& bio
->bi_ioc
)
881 return current
->io_context
;
885 * __get_request - get a free request
886 * @rl: request list to allocate from
887 * @rw_flags: RW and SYNC flags
888 * @bio: bio to allocate request for (can be %NULL)
889 * @gfp_mask: allocation mask
891 * Get a free request from @q. This function may fail under memory
892 * pressure or if @q is dead.
894 * Must be callled with @q->queue_lock held and,
895 * Returns %NULL on failure, with @q->queue_lock held.
896 * Returns !%NULL on success, with @q->queue_lock *not held*.
898 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
899 struct bio
*bio
, gfp_t gfp_mask
)
901 struct request_queue
*q
= rl
->q
;
903 struct elevator_type
*et
= q
->elevator
->type
;
904 struct io_context
*ioc
= rq_ioc(bio
);
905 struct io_cq
*icq
= NULL
;
906 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
909 if (unlikely(blk_queue_dying(q
)))
912 may_queue
= elv_may_queue(q
, rw_flags
);
913 if (may_queue
== ELV_MQUEUE_NO
)
916 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
917 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
919 * The queue will fill after this allocation, so set
920 * it as full, and mark this process as "batching".
921 * This process will be allowed to complete a batch of
922 * requests, others will be blocked.
924 if (!blk_rl_full(rl
, is_sync
)) {
925 ioc_set_batching(q
, ioc
);
926 blk_set_rl_full(rl
, is_sync
);
928 if (may_queue
!= ELV_MQUEUE_MUST
929 && !ioc_batching(q
, ioc
)) {
931 * The queue is full and the allocating
932 * process is not a "batcher", and not
933 * exempted by the IO scheduler
940 * bdi isn't aware of blkcg yet. As all async IOs end up
941 * root blkcg anyway, just use root blkcg state.
943 if (rl
== &q
->root_rl
)
944 blk_set_queue_congested(q
, is_sync
);
948 * Only allow batching queuers to allocate up to 50% over the defined
949 * limit of requests, otherwise we could have thousands of requests
950 * allocated with any setting of ->nr_requests
952 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
955 q
->nr_rqs
[is_sync
]++;
956 rl
->count
[is_sync
]++;
957 rl
->starved
[is_sync
] = 0;
960 * Decide whether the new request will be managed by elevator. If
961 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
962 * prevent the current elevator from being destroyed until the new
963 * request is freed. This guarantees icq's won't be destroyed and
964 * makes creating new ones safe.
966 * Also, lookup icq while holding queue_lock. If it doesn't exist,
967 * it will be created after releasing queue_lock.
969 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
970 rw_flags
|= REQ_ELVPRIV
;
972 if (et
->icq_cache
&& ioc
)
973 icq
= ioc_lookup_icq(ioc
, q
);
976 if (blk_queue_io_stat(q
))
977 rw_flags
|= REQ_IO_STAT
;
978 spin_unlock_irq(q
->queue_lock
);
980 /* allocate and init request */
981 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
986 blk_rq_set_rl(rq
, rl
);
987 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
990 if (rw_flags
& REQ_ELVPRIV
) {
991 if (unlikely(et
->icq_cache
&& !icq
)) {
993 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
999 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1002 /* @rq->elv.icq holds io_context until @rq is freed */
1004 get_io_context(icq
->ioc
);
1008 * ioc may be NULL here, and ioc_batching will be false. That's
1009 * OK, if the queue is under the request limit then requests need
1010 * not count toward the nr_batch_requests limit. There will always
1011 * be some limit enforced by BLK_BATCH_TIME.
1013 if (ioc_batching(q
, ioc
))
1014 ioc
->nr_batch_requests
--;
1016 trace_block_getrq(q
, bio
, rw_flags
& 1);
1021 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1022 * and may fail indefinitely under memory pressure and thus
1023 * shouldn't stall IO. Treat this request as !elvpriv. This will
1024 * disturb iosched and blkcg but weird is bettern than dead.
1026 printk_ratelimited(KERN_WARNING
"%s: request aux data allocation failed, iosched may be disturbed\n",
1027 dev_name(q
->backing_dev_info
.dev
));
1029 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1032 spin_lock_irq(q
->queue_lock
);
1033 q
->nr_rqs_elvpriv
--;
1034 spin_unlock_irq(q
->queue_lock
);
1039 * Allocation failed presumably due to memory. Undo anything we
1040 * might have messed up.
1042 * Allocating task should really be put onto the front of the wait
1043 * queue, but this is pretty rare.
1045 spin_lock_irq(q
->queue_lock
);
1046 freed_request(rl
, rw_flags
);
1049 * in the very unlikely event that allocation failed and no
1050 * requests for this direction was pending, mark us starved so that
1051 * freeing of a request in the other direction will notice
1052 * us. another possible fix would be to split the rq mempool into
1056 if (unlikely(rl
->count
[is_sync
] == 0))
1057 rl
->starved
[is_sync
] = 1;
1062 * get_request - get a free request
1063 * @q: request_queue to allocate request from
1064 * @rw_flags: RW and SYNC flags
1065 * @bio: bio to allocate request for (can be %NULL)
1066 * @gfp_mask: allocation mask
1068 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1069 * function keeps retrying under memory pressure and fails iff @q is dead.
1071 * Must be callled with @q->queue_lock held and,
1072 * Returns %NULL on failure, with @q->queue_lock held.
1073 * Returns !%NULL on success, with @q->queue_lock *not held*.
1075 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1076 struct bio
*bio
, gfp_t gfp_mask
)
1078 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1080 struct request_list
*rl
;
1083 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1085 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1089 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dying(q
))) {
1094 /* wait on @rl and retry */
1095 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1096 TASK_UNINTERRUPTIBLE
);
1098 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1100 spin_unlock_irq(q
->queue_lock
);
1104 * After sleeping, we become a "batching" process and will be able
1105 * to allocate at least one request, and up to a big batch of them
1106 * for a small period time. See ioc_batching, ioc_set_batching
1108 ioc_set_batching(q
, current
->io_context
);
1110 spin_lock_irq(q
->queue_lock
);
1111 finish_wait(&rl
->wait
[is_sync
], &wait
);
1116 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1121 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1123 /* create ioc upfront */
1124 create_io_context(gfp_mask
, q
->node
);
1126 spin_lock_irq(q
->queue_lock
);
1127 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1129 spin_unlock_irq(q
->queue_lock
);
1130 /* q->queue_lock is unlocked at this point */
1135 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1138 return blk_mq_alloc_request(q
, rw
, gfp_mask
);
1140 return blk_old_get_request(q
, rw
, gfp_mask
);
1142 EXPORT_SYMBOL(blk_get_request
);
1145 * blk_make_request - given a bio, allocate a corresponding struct request.
1146 * @q: target request queue
1147 * @bio: The bio describing the memory mappings that will be submitted for IO.
1148 * It may be a chained-bio properly constructed by block/bio layer.
1149 * @gfp_mask: gfp flags to be used for memory allocation
1151 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1152 * type commands. Where the struct request needs to be farther initialized by
1153 * the caller. It is passed a &struct bio, which describes the memory info of
1156 * The caller of blk_make_request must make sure that bi_io_vec
1157 * are set to describe the memory buffers. That bio_data_dir() will return
1158 * the needed direction of the request. (And all bio's in the passed bio-chain
1159 * are properly set accordingly)
1161 * If called under none-sleepable conditions, mapped bio buffers must not
1162 * need bouncing, by calling the appropriate masked or flagged allocator,
1163 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1166 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1167 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1168 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1169 * completion of a bio that hasn't been submitted yet, thus resulting in a
1170 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1171 * of bio_alloc(), as that avoids the mempool deadlock.
1172 * If possible a big IO should be split into smaller parts when allocation
1173 * fails. Partial allocation should not be an error, or you risk a live-lock.
1175 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1178 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1181 return ERR_PTR(-ENOMEM
);
1184 struct bio
*bounce_bio
= bio
;
1187 blk_queue_bounce(q
, &bounce_bio
);
1188 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1189 if (unlikely(ret
)) {
1190 blk_put_request(rq
);
1191 return ERR_PTR(ret
);
1197 EXPORT_SYMBOL(blk_make_request
);
1200 * blk_requeue_request - put a request back on queue
1201 * @q: request queue where request should be inserted
1202 * @rq: request to be inserted
1205 * Drivers often keep queueing requests until the hardware cannot accept
1206 * more, when that condition happens we need to put the request back
1207 * on the queue. Must be called with queue lock held.
1209 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1211 blk_delete_timer(rq
);
1212 blk_clear_rq_complete(rq
);
1213 trace_block_rq_requeue(q
, rq
);
1215 if (blk_rq_tagged(rq
))
1216 blk_queue_end_tag(q
, rq
);
1218 BUG_ON(blk_queued_rq(rq
));
1220 elv_requeue_request(q
, rq
);
1222 EXPORT_SYMBOL(blk_requeue_request
);
1224 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1227 blk_account_io_start(rq
, true);
1228 __elv_add_request(q
, rq
, where
);
1231 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1234 if (now
== part
->stamp
)
1237 if (part_in_flight(part
)) {
1238 __part_stat_add(cpu
, part
, time_in_queue
,
1239 part_in_flight(part
) * (now
- part
->stamp
));
1240 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1246 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1247 * @cpu: cpu number for stats access
1248 * @part: target partition
1250 * The average IO queue length and utilisation statistics are maintained
1251 * by observing the current state of the queue length and the amount of
1252 * time it has been in this state for.
1254 * Normally, that accounting is done on IO completion, but that can result
1255 * in more than a second's worth of IO being accounted for within any one
1256 * second, leading to >100% utilisation. To deal with that, we call this
1257 * function to do a round-off before returning the results when reading
1258 * /proc/diskstats. This accounts immediately for all queue usage up to
1259 * the current jiffies and restarts the counters again.
1261 void part_round_stats(int cpu
, struct hd_struct
*part
)
1263 unsigned long now
= jiffies
;
1266 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1267 part_round_stats_single(cpu
, part
, now
);
1269 EXPORT_SYMBOL_GPL(part_round_stats
);
1271 #ifdef CONFIG_PM_RUNTIME
1272 static void blk_pm_put_request(struct request
*rq
)
1274 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1275 pm_runtime_mark_last_busy(rq
->q
->dev
);
1278 static inline void blk_pm_put_request(struct request
*rq
) {}
1282 * queue lock must be held
1284 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1290 blk_mq_free_request(req
);
1294 blk_pm_put_request(req
);
1296 elv_completed_request(q
, req
);
1298 /* this is a bio leak */
1299 WARN_ON(req
->bio
!= NULL
);
1302 * Request may not have originated from ll_rw_blk. if not,
1303 * it didn't come out of our reserved rq pools
1305 if (req
->cmd_flags
& REQ_ALLOCED
) {
1306 unsigned int flags
= req
->cmd_flags
;
1307 struct request_list
*rl
= blk_rq_rl(req
);
1309 BUG_ON(!list_empty(&req
->queuelist
));
1310 BUG_ON(ELV_ON_HASH(req
));
1312 blk_free_request(rl
, req
);
1313 freed_request(rl
, flags
);
1317 EXPORT_SYMBOL_GPL(__blk_put_request
);
1319 void blk_put_request(struct request
*req
)
1321 struct request_queue
*q
= req
->q
;
1324 blk_mq_free_request(req
);
1326 unsigned long flags
;
1328 spin_lock_irqsave(q
->queue_lock
, flags
);
1329 __blk_put_request(q
, req
);
1330 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1333 EXPORT_SYMBOL(blk_put_request
);
1336 * blk_add_request_payload - add a payload to a request
1337 * @rq: request to update
1338 * @page: page backing the payload
1339 * @len: length of the payload.
1341 * This allows to later add a payload to an already submitted request by
1342 * a block driver. The driver needs to take care of freeing the payload
1345 * Note that this is a quite horrible hack and nothing but handling of
1346 * discard requests should ever use it.
1348 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1351 struct bio
*bio
= rq
->bio
;
1353 bio
->bi_io_vec
->bv_page
= page
;
1354 bio
->bi_io_vec
->bv_offset
= 0;
1355 bio
->bi_io_vec
->bv_len
= len
;
1357 bio
->bi_iter
.bi_size
= len
;
1359 bio
->bi_phys_segments
= 1;
1361 rq
->__data_len
= rq
->resid_len
= len
;
1362 rq
->nr_phys_segments
= 1;
1364 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1366 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1369 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1371 if (!ll_back_merge_fn(q
, req
, bio
))
1374 trace_block_bio_backmerge(q
, req
, bio
);
1376 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1377 blk_rq_set_mixed_merge(req
);
1379 req
->biotail
->bi_next
= bio
;
1381 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1382 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1384 blk_account_io_start(req
, false);
1388 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1391 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1393 if (!ll_front_merge_fn(q
, req
, bio
))
1396 trace_block_bio_frontmerge(q
, req
, bio
);
1398 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1399 blk_rq_set_mixed_merge(req
);
1401 bio
->bi_next
= req
->bio
;
1404 req
->__sector
= bio
->bi_iter
.bi_sector
;
1405 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1406 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1408 blk_account_io_start(req
, false);
1413 * blk_attempt_plug_merge - try to merge with %current's plugged list
1414 * @q: request_queue new bio is being queued at
1415 * @bio: new bio being queued
1416 * @request_count: out parameter for number of traversed plugged requests
1418 * Determine whether @bio being queued on @q can be merged with a request
1419 * on %current's plugged list. Returns %true if merge was successful,
1422 * Plugging coalesces IOs from the same issuer for the same purpose without
1423 * going through @q->queue_lock. As such it's more of an issuing mechanism
1424 * than scheduling, and the request, while may have elvpriv data, is not
1425 * added on the elevator at this point. In addition, we don't have
1426 * reliable access to the elevator outside queue lock. Only check basic
1427 * merging parameters without querying the elevator.
1429 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1430 unsigned int *request_count
)
1432 struct blk_plug
*plug
;
1435 struct list_head
*plug_list
;
1437 if (blk_queue_nomerges(q
))
1440 plug
= current
->plug
;
1446 plug_list
= &plug
->mq_list
;
1448 plug_list
= &plug
->list
;
1450 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1456 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1459 el_ret
= blk_try_merge(rq
, bio
);
1460 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1461 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1464 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1465 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1474 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1476 req
->cmd_type
= REQ_TYPE_FS
;
1478 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1479 if (bio
->bi_rw
& REQ_RAHEAD
)
1480 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1483 req
->__sector
= bio
->bi_iter
.bi_sector
;
1484 req
->ioprio
= bio_prio(bio
);
1485 blk_rq_bio_prep(req
->q
, req
, bio
);
1488 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1490 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1491 struct blk_plug
*plug
;
1492 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1493 struct request
*req
;
1494 unsigned int request_count
= 0;
1497 * low level driver can indicate that it wants pages above a
1498 * certain limit bounced to low memory (ie for highmem, or even
1499 * ISA dma in theory)
1501 blk_queue_bounce(q
, &bio
);
1503 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1504 bio_endio(bio
, -EIO
);
1508 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1509 spin_lock_irq(q
->queue_lock
);
1510 where
= ELEVATOR_INSERT_FLUSH
;
1515 * Check if we can merge with the plugged list before grabbing
1518 if (blk_attempt_plug_merge(q
, bio
, &request_count
))
1521 spin_lock_irq(q
->queue_lock
);
1523 el_ret
= elv_merge(q
, &req
, bio
);
1524 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1525 if (bio_attempt_back_merge(q
, req
, bio
)) {
1526 elv_bio_merged(q
, req
, bio
);
1527 if (!attempt_back_merge(q
, req
))
1528 elv_merged_request(q
, req
, el_ret
);
1531 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1532 if (bio_attempt_front_merge(q
, req
, bio
)) {
1533 elv_bio_merged(q
, req
, bio
);
1534 if (!attempt_front_merge(q
, req
))
1535 elv_merged_request(q
, req
, el_ret
);
1542 * This sync check and mask will be re-done in init_request_from_bio(),
1543 * but we need to set it earlier to expose the sync flag to the
1544 * rq allocator and io schedulers.
1546 rw_flags
= bio_data_dir(bio
);
1548 rw_flags
|= REQ_SYNC
;
1551 * Grab a free request. This is might sleep but can not fail.
1552 * Returns with the queue unlocked.
1554 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1555 if (unlikely(!req
)) {
1556 bio_endio(bio
, -ENODEV
); /* @q is dead */
1561 * After dropping the lock and possibly sleeping here, our request
1562 * may now be mergeable after it had proven unmergeable (above).
1563 * We don't worry about that case for efficiency. It won't happen
1564 * often, and the elevators are able to handle it.
1566 init_request_from_bio(req
, bio
);
1568 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1569 req
->cpu
= raw_smp_processor_id();
1571 plug
= current
->plug
;
1574 * If this is the first request added after a plug, fire
1578 trace_block_plug(q
);
1580 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1581 blk_flush_plug_list(plug
, false);
1582 trace_block_plug(q
);
1585 list_add_tail(&req
->queuelist
, &plug
->list
);
1586 blk_account_io_start(req
, true);
1588 spin_lock_irq(q
->queue_lock
);
1589 add_acct_request(q
, req
, where
);
1592 spin_unlock_irq(q
->queue_lock
);
1595 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1598 * If bio->bi_dev is a partition, remap the location
1600 static inline void blk_partition_remap(struct bio
*bio
)
1602 struct block_device
*bdev
= bio
->bi_bdev
;
1604 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1605 struct hd_struct
*p
= bdev
->bd_part
;
1607 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1608 bio
->bi_bdev
= bdev
->bd_contains
;
1610 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1612 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1616 static void handle_bad_sector(struct bio
*bio
)
1618 char b
[BDEVNAME_SIZE
];
1620 printk(KERN_INFO
"attempt to access beyond end of device\n");
1621 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1622 bdevname(bio
->bi_bdev
, b
),
1624 (unsigned long long)bio_end_sector(bio
),
1625 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1627 set_bit(BIO_EOF
, &bio
->bi_flags
);
1630 #ifdef CONFIG_FAIL_MAKE_REQUEST
1632 static DECLARE_FAULT_ATTR(fail_make_request
);
1634 static int __init
setup_fail_make_request(char *str
)
1636 return setup_fault_attr(&fail_make_request
, str
);
1638 __setup("fail_make_request=", setup_fail_make_request
);
1640 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1642 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1645 static int __init
fail_make_request_debugfs(void)
1647 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1648 NULL
, &fail_make_request
);
1650 return PTR_ERR_OR_ZERO(dir
);
1653 late_initcall(fail_make_request_debugfs
);
1655 #else /* CONFIG_FAIL_MAKE_REQUEST */
1657 static inline bool should_fail_request(struct hd_struct
*part
,
1663 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1666 * Check whether this bio extends beyond the end of the device.
1668 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1675 /* Test device or partition size, when known. */
1676 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1678 sector_t sector
= bio
->bi_iter
.bi_sector
;
1680 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1682 * This may well happen - the kernel calls bread()
1683 * without checking the size of the device, e.g., when
1684 * mounting a device.
1686 handle_bad_sector(bio
);
1694 static noinline_for_stack
bool
1695 generic_make_request_checks(struct bio
*bio
)
1697 struct request_queue
*q
;
1698 int nr_sectors
= bio_sectors(bio
);
1700 char b
[BDEVNAME_SIZE
];
1701 struct hd_struct
*part
;
1705 if (bio_check_eod(bio
, nr_sectors
))
1708 q
= bdev_get_queue(bio
->bi_bdev
);
1711 "generic_make_request: Trying to access "
1712 "nonexistent block-device %s (%Lu)\n",
1713 bdevname(bio
->bi_bdev
, b
),
1714 (long long) bio
->bi_iter
.bi_sector
);
1718 if (likely(bio_is_rw(bio
) &&
1719 nr_sectors
> queue_max_hw_sectors(q
))) {
1720 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1721 bdevname(bio
->bi_bdev
, b
),
1723 queue_max_hw_sectors(q
));
1727 part
= bio
->bi_bdev
->bd_part
;
1728 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1729 should_fail_request(&part_to_disk(part
)->part0
,
1730 bio
->bi_iter
.bi_size
))
1734 * If this device has partitions, remap block n
1735 * of partition p to block n+start(p) of the disk.
1737 blk_partition_remap(bio
);
1739 if (bio_check_eod(bio
, nr_sectors
))
1743 * Filter flush bio's early so that make_request based
1744 * drivers without flush support don't have to worry
1747 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1748 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1755 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1756 (!blk_queue_discard(q
) ||
1757 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1762 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1768 * Various block parts want %current->io_context and lazy ioc
1769 * allocation ends up trading a lot of pain for a small amount of
1770 * memory. Just allocate it upfront. This may fail and block
1771 * layer knows how to live with it.
1773 create_io_context(GFP_ATOMIC
, q
->node
);
1775 if (blk_throtl_bio(q
, bio
))
1776 return false; /* throttled, will be resubmitted later */
1778 trace_block_bio_queue(q
, bio
);
1782 bio_endio(bio
, err
);
1787 * generic_make_request - hand a buffer to its device driver for I/O
1788 * @bio: The bio describing the location in memory and on the device.
1790 * generic_make_request() is used to make I/O requests of block
1791 * devices. It is passed a &struct bio, which describes the I/O that needs
1794 * generic_make_request() does not return any status. The
1795 * success/failure status of the request, along with notification of
1796 * completion, is delivered asynchronously through the bio->bi_end_io
1797 * function described (one day) else where.
1799 * The caller of generic_make_request must make sure that bi_io_vec
1800 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1801 * set to describe the device address, and the
1802 * bi_end_io and optionally bi_private are set to describe how
1803 * completion notification should be signaled.
1805 * generic_make_request and the drivers it calls may use bi_next if this
1806 * bio happens to be merged with someone else, and may resubmit the bio to
1807 * a lower device by calling into generic_make_request recursively, which
1808 * means the bio should NOT be touched after the call to ->make_request_fn.
1810 void generic_make_request(struct bio
*bio
)
1812 struct bio_list bio_list_on_stack
;
1814 if (!generic_make_request_checks(bio
))
1818 * We only want one ->make_request_fn to be active at a time, else
1819 * stack usage with stacked devices could be a problem. So use
1820 * current->bio_list to keep a list of requests submited by a
1821 * make_request_fn function. current->bio_list is also used as a
1822 * flag to say if generic_make_request is currently active in this
1823 * task or not. If it is NULL, then no make_request is active. If
1824 * it is non-NULL, then a make_request is active, and new requests
1825 * should be added at the tail
1827 if (current
->bio_list
) {
1828 bio_list_add(current
->bio_list
, bio
);
1832 /* following loop may be a bit non-obvious, and so deserves some
1834 * Before entering the loop, bio->bi_next is NULL (as all callers
1835 * ensure that) so we have a list with a single bio.
1836 * We pretend that we have just taken it off a longer list, so
1837 * we assign bio_list to a pointer to the bio_list_on_stack,
1838 * thus initialising the bio_list of new bios to be
1839 * added. ->make_request() may indeed add some more bios
1840 * through a recursive call to generic_make_request. If it
1841 * did, we find a non-NULL value in bio_list and re-enter the loop
1842 * from the top. In this case we really did just take the bio
1843 * of the top of the list (no pretending) and so remove it from
1844 * bio_list, and call into ->make_request() again.
1846 BUG_ON(bio
->bi_next
);
1847 bio_list_init(&bio_list_on_stack
);
1848 current
->bio_list
= &bio_list_on_stack
;
1850 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1852 q
->make_request_fn(q
, bio
);
1854 bio
= bio_list_pop(current
->bio_list
);
1856 current
->bio_list
= NULL
; /* deactivate */
1858 EXPORT_SYMBOL(generic_make_request
);
1861 * submit_bio - submit a bio to the block device layer for I/O
1862 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1863 * @bio: The &struct bio which describes the I/O
1865 * submit_bio() is very similar in purpose to generic_make_request(), and
1866 * uses that function to do most of the work. Both are fairly rough
1867 * interfaces; @bio must be presetup and ready for I/O.
1870 void submit_bio(int rw
, struct bio
*bio
)
1875 * If it's a regular read/write or a barrier with data attached,
1876 * go through the normal accounting stuff before submission.
1878 if (bio_has_data(bio
)) {
1881 if (unlikely(rw
& REQ_WRITE_SAME
))
1882 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
1884 count
= bio_sectors(bio
);
1887 count_vm_events(PGPGOUT
, count
);
1889 task_io_account_read(bio
->bi_iter
.bi_size
);
1890 count_vm_events(PGPGIN
, count
);
1893 if (unlikely(block_dump
)) {
1894 char b
[BDEVNAME_SIZE
];
1895 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1896 current
->comm
, task_pid_nr(current
),
1897 (rw
& WRITE
) ? "WRITE" : "READ",
1898 (unsigned long long)bio
->bi_iter
.bi_sector
,
1899 bdevname(bio
->bi_bdev
, b
),
1904 generic_make_request(bio
);
1906 EXPORT_SYMBOL(submit_bio
);
1909 * blk_rq_check_limits - Helper function to check a request for the queue limit
1911 * @rq: the request being checked
1914 * @rq may have been made based on weaker limitations of upper-level queues
1915 * in request stacking drivers, and it may violate the limitation of @q.
1916 * Since the block layer and the underlying device driver trust @rq
1917 * after it is inserted to @q, it should be checked against @q before
1918 * the insertion using this generic function.
1920 * This function should also be useful for request stacking drivers
1921 * in some cases below, so export this function.
1922 * Request stacking drivers like request-based dm may change the queue
1923 * limits while requests are in the queue (e.g. dm's table swapping).
1924 * Such request stacking drivers should check those requests against
1925 * the new queue limits again when they dispatch those requests,
1926 * although such checkings are also done against the old queue limits
1927 * when submitting requests.
1929 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1931 if (!rq_mergeable(rq
))
1934 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
1935 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1940 * queue's settings related to segment counting like q->bounce_pfn
1941 * may differ from that of other stacking queues.
1942 * Recalculate it to check the request correctly on this queue's
1945 blk_recalc_rq_segments(rq
);
1946 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1947 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1953 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1956 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1957 * @q: the queue to submit the request
1958 * @rq: the request being queued
1960 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1962 unsigned long flags
;
1963 int where
= ELEVATOR_INSERT_BACK
;
1965 if (blk_rq_check_limits(q
, rq
))
1969 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1972 spin_lock_irqsave(q
->queue_lock
, flags
);
1973 if (unlikely(blk_queue_dying(q
))) {
1974 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1979 * Submitting request must be dequeued before calling this function
1980 * because it will be linked to another request_queue
1982 BUG_ON(blk_queued_rq(rq
));
1984 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1985 where
= ELEVATOR_INSERT_FLUSH
;
1987 add_acct_request(q
, rq
, where
);
1988 if (where
== ELEVATOR_INSERT_FLUSH
)
1990 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1994 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1997 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1998 * @rq: request to examine
2001 * A request could be merge of IOs which require different failure
2002 * handling. This function determines the number of bytes which
2003 * can be failed from the beginning of the request without
2004 * crossing into area which need to be retried further.
2007 * The number of bytes to fail.
2010 * queue_lock must be held.
2012 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2014 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2015 unsigned int bytes
= 0;
2018 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2019 return blk_rq_bytes(rq
);
2022 * Currently the only 'mixing' which can happen is between
2023 * different fastfail types. We can safely fail portions
2024 * which have all the failfast bits that the first one has -
2025 * the ones which are at least as eager to fail as the first
2028 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2029 if ((bio
->bi_rw
& ff
) != ff
)
2031 bytes
+= bio
->bi_iter
.bi_size
;
2034 /* this could lead to infinite loop */
2035 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2038 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2040 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2042 if (blk_do_io_stat(req
)) {
2043 const int rw
= rq_data_dir(req
);
2044 struct hd_struct
*part
;
2047 cpu
= part_stat_lock();
2049 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2054 void blk_account_io_done(struct request
*req
)
2057 * Account IO completion. flush_rq isn't accounted as a
2058 * normal IO on queueing nor completion. Accounting the
2059 * containing request is enough.
2061 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2062 unsigned long duration
= jiffies
- req
->start_time
;
2063 const int rw
= rq_data_dir(req
);
2064 struct hd_struct
*part
;
2067 cpu
= part_stat_lock();
2070 part_stat_inc(cpu
, part
, ios
[rw
]);
2071 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2072 part_round_stats(cpu
, part
);
2073 part_dec_in_flight(part
, rw
);
2075 hd_struct_put(part
);
2080 #ifdef CONFIG_PM_RUNTIME
2082 * Don't process normal requests when queue is suspended
2083 * or in the process of suspending/resuming
2085 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2088 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2089 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2095 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2102 void blk_account_io_start(struct request
*rq
, bool new_io
)
2104 struct hd_struct
*part
;
2105 int rw
= rq_data_dir(rq
);
2108 if (!blk_do_io_stat(rq
))
2111 cpu
= part_stat_lock();
2115 part_stat_inc(cpu
, part
, merges
[rw
]);
2117 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2118 if (!hd_struct_try_get(part
)) {
2120 * The partition is already being removed,
2121 * the request will be accounted on the disk only
2123 * We take a reference on disk->part0 although that
2124 * partition will never be deleted, so we can treat
2125 * it as any other partition.
2127 part
= &rq
->rq_disk
->part0
;
2128 hd_struct_get(part
);
2130 part_round_stats(cpu
, part
);
2131 part_inc_in_flight(part
, rw
);
2139 * blk_peek_request - peek at the top of a request queue
2140 * @q: request queue to peek at
2143 * Return the request at the top of @q. The returned request
2144 * should be started using blk_start_request() before LLD starts
2148 * Pointer to the request at the top of @q if available. Null
2152 * queue_lock must be held.
2154 struct request
*blk_peek_request(struct request_queue
*q
)
2159 while ((rq
= __elv_next_request(q
)) != NULL
) {
2161 rq
= blk_pm_peek_request(q
, rq
);
2165 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2167 * This is the first time the device driver
2168 * sees this request (possibly after
2169 * requeueing). Notify IO scheduler.
2171 if (rq
->cmd_flags
& REQ_SORTED
)
2172 elv_activate_rq(q
, rq
);
2175 * just mark as started even if we don't start
2176 * it, a request that has been delayed should
2177 * not be passed by new incoming requests
2179 rq
->cmd_flags
|= REQ_STARTED
;
2180 trace_block_rq_issue(q
, rq
);
2183 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2184 q
->end_sector
= rq_end_sector(rq
);
2185 q
->boundary_rq
= NULL
;
2188 if (rq
->cmd_flags
& REQ_DONTPREP
)
2191 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2193 * make sure space for the drain appears we
2194 * know we can do this because max_hw_segments
2195 * has been adjusted to be one fewer than the
2198 rq
->nr_phys_segments
++;
2204 ret
= q
->prep_rq_fn(q
, rq
);
2205 if (ret
== BLKPREP_OK
) {
2207 } else if (ret
== BLKPREP_DEFER
) {
2209 * the request may have been (partially) prepped.
2210 * we need to keep this request in the front to
2211 * avoid resource deadlock. REQ_STARTED will
2212 * prevent other fs requests from passing this one.
2214 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2215 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2217 * remove the space for the drain we added
2218 * so that we don't add it again
2220 --rq
->nr_phys_segments
;
2225 } else if (ret
== BLKPREP_KILL
) {
2226 rq
->cmd_flags
|= REQ_QUIET
;
2228 * Mark this request as started so we don't trigger
2229 * any debug logic in the end I/O path.
2231 blk_start_request(rq
);
2232 __blk_end_request_all(rq
, -EIO
);
2234 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2241 EXPORT_SYMBOL(blk_peek_request
);
2243 void blk_dequeue_request(struct request
*rq
)
2245 struct request_queue
*q
= rq
->q
;
2247 BUG_ON(list_empty(&rq
->queuelist
));
2248 BUG_ON(ELV_ON_HASH(rq
));
2250 list_del_init(&rq
->queuelist
);
2253 * the time frame between a request being removed from the lists
2254 * and to it is freed is accounted as io that is in progress at
2257 if (blk_account_rq(rq
)) {
2258 q
->in_flight
[rq_is_sync(rq
)]++;
2259 set_io_start_time_ns(rq
);
2264 * blk_start_request - start request processing on the driver
2265 * @req: request to dequeue
2268 * Dequeue @req and start timeout timer on it. This hands off the
2269 * request to the driver.
2271 * Block internal functions which don't want to start timer should
2272 * call blk_dequeue_request().
2275 * queue_lock must be held.
2277 void blk_start_request(struct request
*req
)
2279 blk_dequeue_request(req
);
2282 * We are now handing the request to the hardware, initialize
2283 * resid_len to full count and add the timeout handler.
2285 req
->resid_len
= blk_rq_bytes(req
);
2286 if (unlikely(blk_bidi_rq(req
)))
2287 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2289 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2292 EXPORT_SYMBOL(blk_start_request
);
2295 * blk_fetch_request - fetch a request from a request queue
2296 * @q: request queue to fetch a request from
2299 * Return the request at the top of @q. The request is started on
2300 * return and LLD can start processing it immediately.
2303 * Pointer to the request at the top of @q if available. Null
2307 * queue_lock must be held.
2309 struct request
*blk_fetch_request(struct request_queue
*q
)
2313 rq
= blk_peek_request(q
);
2315 blk_start_request(rq
);
2318 EXPORT_SYMBOL(blk_fetch_request
);
2321 * blk_update_request - Special helper function for request stacking drivers
2322 * @req: the request being processed
2323 * @error: %0 for success, < %0 for error
2324 * @nr_bytes: number of bytes to complete @req
2327 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2328 * the request structure even if @req doesn't have leftover.
2329 * If @req has leftover, sets it up for the next range of segments.
2331 * This special helper function is only for request stacking drivers
2332 * (e.g. request-based dm) so that they can handle partial completion.
2333 * Actual device drivers should use blk_end_request instead.
2335 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2336 * %false return from this function.
2339 * %false - this request doesn't have any more data
2340 * %true - this request has more data
2342 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2349 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2352 * For fs requests, rq is just carrier of independent bio's
2353 * and each partial completion should be handled separately.
2354 * Reset per-request error on each partial completion.
2356 * TODO: tj: This is too subtle. It would be better to let
2357 * low level drivers do what they see fit.
2359 if (req
->cmd_type
== REQ_TYPE_FS
)
2362 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2363 !(req
->cmd_flags
& REQ_QUIET
)) {
2368 error_type
= "recoverable transport";
2371 error_type
= "critical target";
2374 error_type
= "critical nexus";
2377 error_type
= "timeout";
2380 error_type
= "critical space allocation";
2383 error_type
= "critical medium";
2390 printk_ratelimited(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2391 error_type
, req
->rq_disk
?
2392 req
->rq_disk
->disk_name
: "?",
2393 (unsigned long long)blk_rq_pos(req
));
2397 blk_account_io_completion(req
, nr_bytes
);
2401 struct bio
*bio
= req
->bio
;
2402 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2404 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2405 req
->bio
= bio
->bi_next
;
2407 req_bio_endio(req
, bio
, bio_bytes
, error
);
2409 total_bytes
+= bio_bytes
;
2410 nr_bytes
-= bio_bytes
;
2421 * Reset counters so that the request stacking driver
2422 * can find how many bytes remain in the request
2425 req
->__data_len
= 0;
2429 req
->__data_len
-= total_bytes
;
2431 /* update sector only for requests with clear definition of sector */
2432 if (req
->cmd_type
== REQ_TYPE_FS
)
2433 req
->__sector
+= total_bytes
>> 9;
2435 /* mixed attributes always follow the first bio */
2436 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2437 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2438 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2442 * If total number of sectors is less than the first segment
2443 * size, something has gone terribly wrong.
2445 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2446 blk_dump_rq_flags(req
, "request botched");
2447 req
->__data_len
= blk_rq_cur_bytes(req
);
2450 /* recalculate the number of segments */
2451 blk_recalc_rq_segments(req
);
2455 EXPORT_SYMBOL_GPL(blk_update_request
);
2457 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2458 unsigned int nr_bytes
,
2459 unsigned int bidi_bytes
)
2461 if (blk_update_request(rq
, error
, nr_bytes
))
2464 /* Bidi request must be completed as a whole */
2465 if (unlikely(blk_bidi_rq(rq
)) &&
2466 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2469 if (blk_queue_add_random(rq
->q
))
2470 add_disk_randomness(rq
->rq_disk
);
2476 * blk_unprep_request - unprepare a request
2479 * This function makes a request ready for complete resubmission (or
2480 * completion). It happens only after all error handling is complete,
2481 * so represents the appropriate moment to deallocate any resources
2482 * that were allocated to the request in the prep_rq_fn. The queue
2483 * lock is held when calling this.
2485 void blk_unprep_request(struct request
*req
)
2487 struct request_queue
*q
= req
->q
;
2489 req
->cmd_flags
&= ~REQ_DONTPREP
;
2490 if (q
->unprep_rq_fn
)
2491 q
->unprep_rq_fn(q
, req
);
2493 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2496 * queue lock must be held
2498 static void blk_finish_request(struct request
*req
, int error
)
2500 if (blk_rq_tagged(req
))
2501 blk_queue_end_tag(req
->q
, req
);
2503 BUG_ON(blk_queued_rq(req
));
2505 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2506 laptop_io_completion(&req
->q
->backing_dev_info
);
2508 blk_delete_timer(req
);
2510 if (req
->cmd_flags
& REQ_DONTPREP
)
2511 blk_unprep_request(req
);
2513 blk_account_io_done(req
);
2516 req
->end_io(req
, error
);
2518 if (blk_bidi_rq(req
))
2519 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2521 __blk_put_request(req
->q
, req
);
2526 * blk_end_bidi_request - Complete a bidi request
2527 * @rq: the request to complete
2528 * @error: %0 for success, < %0 for error
2529 * @nr_bytes: number of bytes to complete @rq
2530 * @bidi_bytes: number of bytes to complete @rq->next_rq
2533 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2534 * Drivers that supports bidi can safely call this member for any
2535 * type of request, bidi or uni. In the later case @bidi_bytes is
2539 * %false - we are done with this request
2540 * %true - still buffers pending for this request
2542 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2543 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2545 struct request_queue
*q
= rq
->q
;
2546 unsigned long flags
;
2548 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2551 spin_lock_irqsave(q
->queue_lock
, flags
);
2552 blk_finish_request(rq
, error
);
2553 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2559 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2560 * @rq: the request to complete
2561 * @error: %0 for success, < %0 for error
2562 * @nr_bytes: number of bytes to complete @rq
2563 * @bidi_bytes: number of bytes to complete @rq->next_rq
2566 * Identical to blk_end_bidi_request() except that queue lock is
2567 * assumed to be locked on entry and remains so on return.
2570 * %false - we are done with this request
2571 * %true - still buffers pending for this request
2573 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2574 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2576 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2579 blk_finish_request(rq
, error
);
2585 * blk_end_request - Helper function for drivers to complete the request.
2586 * @rq: the request being processed
2587 * @error: %0 for success, < %0 for error
2588 * @nr_bytes: number of bytes to complete
2591 * Ends I/O on a number of bytes attached to @rq.
2592 * If @rq has leftover, sets it up for the next range of segments.
2595 * %false - we are done with this request
2596 * %true - still buffers pending for this request
2598 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2600 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2602 EXPORT_SYMBOL(blk_end_request
);
2605 * blk_end_request_all - Helper function for drives to finish the request.
2606 * @rq: the request to finish
2607 * @error: %0 for success, < %0 for error
2610 * Completely finish @rq.
2612 void blk_end_request_all(struct request
*rq
, int error
)
2615 unsigned int bidi_bytes
= 0;
2617 if (unlikely(blk_bidi_rq(rq
)))
2618 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2620 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2623 EXPORT_SYMBOL(blk_end_request_all
);
2626 * blk_end_request_cur - Helper function to finish the current request chunk.
2627 * @rq: the request to finish the current chunk for
2628 * @error: %0 for success, < %0 for error
2631 * Complete the current consecutively mapped chunk from @rq.
2634 * %false - we are done with this request
2635 * %true - still buffers pending for this request
2637 bool blk_end_request_cur(struct request
*rq
, int error
)
2639 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2641 EXPORT_SYMBOL(blk_end_request_cur
);
2644 * blk_end_request_err - Finish a request till the next failure boundary.
2645 * @rq: the request to finish till the next failure boundary for
2646 * @error: must be negative errno
2649 * Complete @rq till the next failure boundary.
2652 * %false - we are done with this request
2653 * %true - still buffers pending for this request
2655 bool blk_end_request_err(struct request
*rq
, int error
)
2657 WARN_ON(error
>= 0);
2658 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2660 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2663 * __blk_end_request - Helper function for drivers to complete the request.
2664 * @rq: the request being processed
2665 * @error: %0 for success, < %0 for error
2666 * @nr_bytes: number of bytes to complete
2669 * Must be called with queue lock held unlike blk_end_request().
2672 * %false - we are done with this request
2673 * %true - still buffers pending for this request
2675 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2677 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2679 EXPORT_SYMBOL(__blk_end_request
);
2682 * __blk_end_request_all - Helper function for drives to finish the request.
2683 * @rq: the request to finish
2684 * @error: %0 for success, < %0 for error
2687 * Completely finish @rq. Must be called with queue lock held.
2689 void __blk_end_request_all(struct request
*rq
, int error
)
2692 unsigned int bidi_bytes
= 0;
2694 if (unlikely(blk_bidi_rq(rq
)))
2695 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2697 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2700 EXPORT_SYMBOL(__blk_end_request_all
);
2703 * __blk_end_request_cur - Helper function to finish the current request chunk.
2704 * @rq: the request to finish the current chunk for
2705 * @error: %0 for success, < %0 for error
2708 * Complete the current consecutively mapped chunk from @rq. Must
2709 * be called with queue lock held.
2712 * %false - we are done with this request
2713 * %true - still buffers pending for this request
2715 bool __blk_end_request_cur(struct request
*rq
, int error
)
2717 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2719 EXPORT_SYMBOL(__blk_end_request_cur
);
2722 * __blk_end_request_err - Finish a request till the next failure boundary.
2723 * @rq: the request to finish till the next failure boundary for
2724 * @error: must be negative errno
2727 * Complete @rq till the next failure boundary. Must be called
2728 * with queue lock held.
2731 * %false - we are done with this request
2732 * %true - still buffers pending for this request
2734 bool __blk_end_request_err(struct request
*rq
, int error
)
2736 WARN_ON(error
>= 0);
2737 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2739 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2741 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2744 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2745 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2747 if (bio_has_data(bio
))
2748 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2750 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2751 rq
->bio
= rq
->biotail
= bio
;
2754 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2757 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2759 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2760 * @rq: the request to be flushed
2763 * Flush all pages in @rq.
2765 void rq_flush_dcache_pages(struct request
*rq
)
2767 struct req_iterator iter
;
2768 struct bio_vec bvec
;
2770 rq_for_each_segment(bvec
, rq
, iter
)
2771 flush_dcache_page(bvec
.bv_page
);
2773 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2777 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2778 * @q : the queue of the device being checked
2781 * Check if underlying low-level drivers of a device are busy.
2782 * If the drivers want to export their busy state, they must set own
2783 * exporting function using blk_queue_lld_busy() first.
2785 * Basically, this function is used only by request stacking drivers
2786 * to stop dispatching requests to underlying devices when underlying
2787 * devices are busy. This behavior helps more I/O merging on the queue
2788 * of the request stacking driver and prevents I/O throughput regression
2789 * on burst I/O load.
2792 * 0 - Not busy (The request stacking driver should dispatch request)
2793 * 1 - Busy (The request stacking driver should stop dispatching request)
2795 int blk_lld_busy(struct request_queue
*q
)
2798 return q
->lld_busy_fn(q
);
2802 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2805 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2806 * @rq: the clone request to be cleaned up
2809 * Free all bios in @rq for a cloned request.
2811 void blk_rq_unprep_clone(struct request
*rq
)
2815 while ((bio
= rq
->bio
) != NULL
) {
2816 rq
->bio
= bio
->bi_next
;
2821 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2824 * Copy attributes of the original request to the clone request.
2825 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2827 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2829 dst
->cpu
= src
->cpu
;
2830 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2831 dst
->cmd_type
= src
->cmd_type
;
2832 dst
->__sector
= blk_rq_pos(src
);
2833 dst
->__data_len
= blk_rq_bytes(src
);
2834 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2835 dst
->ioprio
= src
->ioprio
;
2836 dst
->extra_len
= src
->extra_len
;
2840 * blk_rq_prep_clone - Helper function to setup clone request
2841 * @rq: the request to be setup
2842 * @rq_src: original request to be cloned
2843 * @bs: bio_set that bios for clone are allocated from
2844 * @gfp_mask: memory allocation mask for bio
2845 * @bio_ctr: setup function to be called for each clone bio.
2846 * Returns %0 for success, non %0 for failure.
2847 * @data: private data to be passed to @bio_ctr
2850 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2851 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2852 * are not copied, and copying such parts is the caller's responsibility.
2853 * Also, pages which the original bios are pointing to are not copied
2854 * and the cloned bios just point same pages.
2855 * So cloned bios must be completed before original bios, which means
2856 * the caller must complete @rq before @rq_src.
2858 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2859 struct bio_set
*bs
, gfp_t gfp_mask
,
2860 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2863 struct bio
*bio
, *bio_src
;
2868 blk_rq_init(NULL
, rq
);
2870 __rq_for_each_bio(bio_src
, rq_src
) {
2871 bio
= bio_clone_bioset(bio_src
, gfp_mask
, bs
);
2875 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2879 rq
->biotail
->bi_next
= bio
;
2882 rq
->bio
= rq
->biotail
= bio
;
2885 __blk_rq_prep_clone(rq
, rq_src
);
2892 blk_rq_unprep_clone(rq
);
2896 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2898 int kblockd_schedule_work(struct work_struct
*work
)
2900 return queue_work(kblockd_workqueue
, work
);
2902 EXPORT_SYMBOL(kblockd_schedule_work
);
2904 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
2905 unsigned long delay
)
2907 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2909 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2911 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
2912 unsigned long delay
)
2914 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
2916 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
2918 #define PLUG_MAGIC 0x91827364
2921 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2922 * @plug: The &struct blk_plug that needs to be initialized
2925 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2926 * pending I/O should the task end up blocking between blk_start_plug() and
2927 * blk_finish_plug(). This is important from a performance perspective, but
2928 * also ensures that we don't deadlock. For instance, if the task is blocking
2929 * for a memory allocation, memory reclaim could end up wanting to free a
2930 * page belonging to that request that is currently residing in our private
2931 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2932 * this kind of deadlock.
2934 void blk_start_plug(struct blk_plug
*plug
)
2936 struct task_struct
*tsk
= current
;
2938 plug
->magic
= PLUG_MAGIC
;
2939 INIT_LIST_HEAD(&plug
->list
);
2940 INIT_LIST_HEAD(&plug
->mq_list
);
2941 INIT_LIST_HEAD(&plug
->cb_list
);
2944 * If this is a nested plug, don't actually assign it. It will be
2945 * flushed on its own.
2949 * Store ordering should not be needed here, since a potential
2950 * preempt will imply a full memory barrier
2955 EXPORT_SYMBOL(blk_start_plug
);
2957 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2959 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2960 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2962 return !(rqa
->q
< rqb
->q
||
2963 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
2967 * If 'from_schedule' is true, then postpone the dispatch of requests
2968 * until a safe kblockd context. We due this to avoid accidental big
2969 * additional stack usage in driver dispatch, in places where the originally
2970 * plugger did not intend it.
2972 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2974 __releases(q
->queue_lock
)
2976 trace_block_unplug(q
, depth
, !from_schedule
);
2979 blk_run_queue_async(q
);
2982 spin_unlock(q
->queue_lock
);
2985 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
2987 LIST_HEAD(callbacks
);
2989 while (!list_empty(&plug
->cb_list
)) {
2990 list_splice_init(&plug
->cb_list
, &callbacks
);
2992 while (!list_empty(&callbacks
)) {
2993 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2996 list_del(&cb
->list
);
2997 cb
->callback(cb
, from_schedule
);
3002 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3005 struct blk_plug
*plug
= current
->plug
;
3006 struct blk_plug_cb
*cb
;
3011 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3012 if (cb
->callback
== unplug
&& cb
->data
== data
)
3015 /* Not currently on the callback list */
3016 BUG_ON(size
< sizeof(*cb
));
3017 cb
= kzalloc(size
, GFP_ATOMIC
);
3020 cb
->callback
= unplug
;
3021 list_add(&cb
->list
, &plug
->cb_list
);
3025 EXPORT_SYMBOL(blk_check_plugged
);
3027 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3029 struct request_queue
*q
;
3030 unsigned long flags
;
3035 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
3037 flush_plug_callbacks(plug
, from_schedule
);
3039 if (!list_empty(&plug
->mq_list
))
3040 blk_mq_flush_plug_list(plug
, from_schedule
);
3042 if (list_empty(&plug
->list
))
3045 list_splice_init(&plug
->list
, &list
);
3047 list_sort(NULL
, &list
, plug_rq_cmp
);
3053 * Save and disable interrupts here, to avoid doing it for every
3054 * queue lock we have to take.
3056 local_irq_save(flags
);
3057 while (!list_empty(&list
)) {
3058 rq
= list_entry_rq(list
.next
);
3059 list_del_init(&rq
->queuelist
);
3063 * This drops the queue lock
3066 queue_unplugged(q
, depth
, from_schedule
);
3069 spin_lock(q
->queue_lock
);
3073 * Short-circuit if @q is dead
3075 if (unlikely(blk_queue_dying(q
))) {
3076 __blk_end_request_all(rq
, -ENODEV
);
3081 * rq is already accounted, so use raw insert
3083 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3084 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3086 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3092 * This drops the queue lock
3095 queue_unplugged(q
, depth
, from_schedule
);
3097 local_irq_restore(flags
);
3100 void blk_finish_plug(struct blk_plug
*plug
)
3102 blk_flush_plug_list(plug
, false);
3104 if (plug
== current
->plug
)
3105 current
->plug
= NULL
;
3107 EXPORT_SYMBOL(blk_finish_plug
);
3109 #ifdef CONFIG_PM_RUNTIME
3111 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3112 * @q: the queue of the device
3113 * @dev: the device the queue belongs to
3116 * Initialize runtime-PM-related fields for @q and start auto suspend for
3117 * @dev. Drivers that want to take advantage of request-based runtime PM
3118 * should call this function after @dev has been initialized, and its
3119 * request queue @q has been allocated, and runtime PM for it can not happen
3120 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3121 * cases, driver should call this function before any I/O has taken place.
3123 * This function takes care of setting up using auto suspend for the device,
3124 * the autosuspend delay is set to -1 to make runtime suspend impossible
3125 * until an updated value is either set by user or by driver. Drivers do
3126 * not need to touch other autosuspend settings.
3128 * The block layer runtime PM is request based, so only works for drivers
3129 * that use request as their IO unit instead of those directly use bio's.
3131 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3134 q
->rpm_status
= RPM_ACTIVE
;
3135 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3136 pm_runtime_use_autosuspend(q
->dev
);
3138 EXPORT_SYMBOL(blk_pm_runtime_init
);
3141 * blk_pre_runtime_suspend - Pre runtime suspend check
3142 * @q: the queue of the device
3145 * This function will check if runtime suspend is allowed for the device
3146 * by examining if there are any requests pending in the queue. If there
3147 * are requests pending, the device can not be runtime suspended; otherwise,
3148 * the queue's status will be updated to SUSPENDING and the driver can
3149 * proceed to suspend the device.
3151 * For the not allowed case, we mark last busy for the device so that
3152 * runtime PM core will try to autosuspend it some time later.
3154 * This function should be called near the start of the device's
3155 * runtime_suspend callback.
3158 * 0 - OK to runtime suspend the device
3159 * -EBUSY - Device should not be runtime suspended
3161 int blk_pre_runtime_suspend(struct request_queue
*q
)
3165 spin_lock_irq(q
->queue_lock
);
3166 if (q
->nr_pending
) {
3168 pm_runtime_mark_last_busy(q
->dev
);
3170 q
->rpm_status
= RPM_SUSPENDING
;
3172 spin_unlock_irq(q
->queue_lock
);
3175 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3178 * blk_post_runtime_suspend - Post runtime suspend processing
3179 * @q: the queue of the device
3180 * @err: return value of the device's runtime_suspend function
3183 * Update the queue's runtime status according to the return value of the
3184 * device's runtime suspend function and mark last busy for the device so
3185 * that PM core will try to auto suspend the device at a later time.
3187 * This function should be called near the end of the device's
3188 * runtime_suspend callback.
3190 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3192 spin_lock_irq(q
->queue_lock
);
3194 q
->rpm_status
= RPM_SUSPENDED
;
3196 q
->rpm_status
= RPM_ACTIVE
;
3197 pm_runtime_mark_last_busy(q
->dev
);
3199 spin_unlock_irq(q
->queue_lock
);
3201 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3204 * blk_pre_runtime_resume - Pre runtime resume processing
3205 * @q: the queue of the device
3208 * Update the queue's runtime status to RESUMING in preparation for the
3209 * runtime resume of the device.
3211 * This function should be called near the start of the device's
3212 * runtime_resume callback.
3214 void blk_pre_runtime_resume(struct request_queue
*q
)
3216 spin_lock_irq(q
->queue_lock
);
3217 q
->rpm_status
= RPM_RESUMING
;
3218 spin_unlock_irq(q
->queue_lock
);
3220 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3223 * blk_post_runtime_resume - Post runtime resume processing
3224 * @q: the queue of the device
3225 * @err: return value of the device's runtime_resume function
3228 * Update the queue's runtime status according to the return value of the
3229 * device's runtime_resume function. If it is successfully resumed, process
3230 * the requests that are queued into the device's queue when it is resuming
3231 * and then mark last busy and initiate autosuspend for it.
3233 * This function should be called near the end of the device's
3234 * runtime_resume callback.
3236 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3238 spin_lock_irq(q
->queue_lock
);
3240 q
->rpm_status
= RPM_ACTIVE
;
3242 pm_runtime_mark_last_busy(q
->dev
);
3243 pm_request_autosuspend(q
->dev
);
3245 q
->rpm_status
= RPM_SUSPENDED
;
3247 spin_unlock_irq(q
->queue_lock
);
3249 EXPORT_SYMBOL(blk_post_runtime_resume
);
3252 int __init
blk_dev_init(void)
3254 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3255 sizeof(((struct request
*)0)->cmd_flags
));
3257 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3258 kblockd_workqueue
= alloc_workqueue("kblockd",
3259 WQ_MEM_RECLAIM
| WQ_HIGHPRI
|
3260 WQ_POWER_EFFICIENT
, 0);
3261 if (!kblockd_workqueue
)
3262 panic("Failed to create kblockd\n");
3264 request_cachep
= kmem_cache_create("blkdev_requests",
3265 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3267 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3268 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);