2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.h>
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
49 DEFINE_IDA(blk_queue_ida
);
52 * For the allocated request tables
54 struct kmem_cache
*request_cachep
= NULL
;
57 * For queue allocation
59 struct kmem_cache
*blk_requestq_cachep
;
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 void blk_queue_congestion_threshold(struct request_queue
*q
)
70 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
71 if (nr
> q
->nr_requests
)
73 q
->nr_congestion_on
= nr
;
75 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
78 q
->nr_congestion_off
= nr
;
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
85 * Locates the passed device's request queue and returns the address of its
86 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
89 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
91 struct request_queue
*q
= bdev_get_queue(bdev
);
93 return &q
->backing_dev_info
;
95 EXPORT_SYMBOL(blk_get_backing_dev_info
);
97 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
99 memset(rq
, 0, sizeof(*rq
));
101 INIT_LIST_HEAD(&rq
->queuelist
);
102 INIT_LIST_HEAD(&rq
->timeout_list
);
105 rq
->__sector
= (sector_t
) -1;
106 INIT_HLIST_NODE(&rq
->hash
);
107 RB_CLEAR_NODE(&rq
->rb_node
);
109 rq
->cmd_len
= BLK_MAX_CDB
;
111 rq
->start_time
= jiffies
;
112 set_start_time_ns(rq
);
115 EXPORT_SYMBOL(blk_rq_init
);
117 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
118 unsigned int nbytes
, int error
)
120 if (error
&& !(rq
->cmd_flags
& REQ_CLONE
))
121 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
122 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
125 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
126 set_bit(BIO_QUIET
, &bio
->bi_flags
);
128 bio_advance(bio
, nbytes
);
130 /* don't actually finish bio if it's part of flush sequence */
131 if (bio
->bi_iter
.bi_size
== 0 &&
132 !(rq
->cmd_flags
& (REQ_FLUSH_SEQ
|REQ_CLONE
)))
133 bio_endio(bio
, error
);
136 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
140 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
141 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
142 (unsigned long long) rq
->cmd_flags
);
144 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
145 (unsigned long long)blk_rq_pos(rq
),
146 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
147 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
148 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
150 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
151 printk(KERN_INFO
" cdb: ");
152 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
153 printk("%02x ", rq
->cmd
[bit
]);
157 EXPORT_SYMBOL(blk_dump_rq_flags
);
159 static void blk_delay_work(struct work_struct
*work
)
161 struct request_queue
*q
;
163 q
= container_of(work
, struct request_queue
, delay_work
.work
);
164 spin_lock_irq(q
->queue_lock
);
166 spin_unlock_irq(q
->queue_lock
);
170 * blk_delay_queue - restart queueing after defined interval
171 * @q: The &struct request_queue in question
172 * @msecs: Delay in msecs
175 * Sometimes queueing needs to be postponed for a little while, to allow
176 * resources to come back. This function will make sure that queueing is
177 * restarted around the specified time. Queue lock must be held.
179 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
181 if (likely(!blk_queue_dead(q
)))
182 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
183 msecs_to_jiffies(msecs
));
185 EXPORT_SYMBOL(blk_delay_queue
);
188 * blk_start_queue - restart a previously stopped queue
189 * @q: The &struct request_queue in question
192 * blk_start_queue() will clear the stop flag on the queue, and call
193 * the request_fn for the queue if it was in a stopped state when
194 * entered. Also see blk_stop_queue(). Queue lock must be held.
196 void blk_start_queue(struct request_queue
*q
)
198 WARN_ON(!irqs_disabled());
200 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
203 EXPORT_SYMBOL(blk_start_queue
);
206 * blk_stop_queue - stop a queue
207 * @q: The &struct request_queue in question
210 * The Linux block layer assumes that a block driver will consume all
211 * entries on the request queue when the request_fn strategy is called.
212 * Often this will not happen, because of hardware limitations (queue
213 * depth settings). If a device driver gets a 'queue full' response,
214 * or if it simply chooses not to queue more I/O at one point, it can
215 * call this function to prevent the request_fn from being called until
216 * the driver has signalled it's ready to go again. This happens by calling
217 * blk_start_queue() to restart queue operations. Queue lock must be held.
219 void blk_stop_queue(struct request_queue
*q
)
221 cancel_delayed_work(&q
->delay_work
);
222 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
224 EXPORT_SYMBOL(blk_stop_queue
);
227 * blk_sync_queue - cancel any pending callbacks on a queue
231 * The block layer may perform asynchronous callback activity
232 * on a queue, such as calling the unplug function after a timeout.
233 * A block device may call blk_sync_queue to ensure that any
234 * such activity is cancelled, thus allowing it to release resources
235 * that the callbacks might use. The caller must already have made sure
236 * that its ->make_request_fn will not re-add plugging prior to calling
239 * This function does not cancel any asynchronous activity arising
240 * out of elevator or throttling code. That would require elevator_exit()
241 * and blkcg_exit_queue() to be called with queue lock initialized.
244 void blk_sync_queue(struct request_queue
*q
)
246 del_timer_sync(&q
->timeout
);
249 struct blk_mq_hw_ctx
*hctx
;
252 queue_for_each_hw_ctx(q
, hctx
, i
) {
253 cancel_delayed_work_sync(&hctx
->run_work
);
254 cancel_delayed_work_sync(&hctx
->delay_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
--;
289 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
292 * __blk_run_queue - run a single device queue
293 * @q: The queue to run
296 * See @blk_run_queue. This variant must be called with the queue lock
297 * held and interrupts disabled.
299 void __blk_run_queue(struct request_queue
*q
)
301 if (unlikely(blk_queue_stopped(q
)))
304 __blk_run_queue_uncond(q
);
306 EXPORT_SYMBOL(__blk_run_queue
);
309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
314 * of us. The caller must hold the queue lock.
316 void blk_run_queue_async(struct request_queue
*q
)
318 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
319 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
321 EXPORT_SYMBOL(blk_run_queue_async
);
324 * blk_run_queue - run a single device queue
325 * @q: The queue to run
328 * Invoke request handling on this queue, if it has pending work to do.
329 * May be used to restart queueing when a request has completed.
331 void blk_run_queue(struct request_queue
*q
)
335 spin_lock_irqsave(q
->queue_lock
, flags
);
337 spin_unlock_irqrestore(q
->queue_lock
, flags
);
339 EXPORT_SYMBOL(blk_run_queue
);
341 void blk_put_queue(struct request_queue
*q
)
343 kobject_put(&q
->kobj
);
345 EXPORT_SYMBOL(blk_put_queue
);
348 * __blk_drain_queue - drain requests from request_queue
350 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
352 * Drain requests from @q. If @drain_all is set, all requests are drained.
353 * If not, only ELVPRIV requests are drained. The caller is responsible
354 * for ensuring that no new requests which need to be drained are queued.
356 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
357 __releases(q
->queue_lock
)
358 __acquires(q
->queue_lock
)
362 lockdep_assert_held(q
->queue_lock
);
368 * The caller might be trying to drain @q before its
369 * elevator is initialized.
372 elv_drain_elevator(q
);
374 blkcg_drain_queue(q
);
377 * This function might be called on a queue which failed
378 * driver init after queue creation or is not yet fully
379 * active yet. Some drivers (e.g. fd and loop) get unhappy
380 * in such cases. Kick queue iff dispatch queue has
381 * something on it and @q has request_fn set.
383 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
386 drain
|= q
->nr_rqs_elvpriv
;
387 drain
|= q
->request_fn_active
;
390 * Unfortunately, requests are queued at and tracked from
391 * multiple places and there's no single counter which can
392 * be drained. Check all the queues and counters.
395 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
396 drain
|= !list_empty(&q
->queue_head
);
397 for (i
= 0; i
< 2; i
++) {
398 drain
|= q
->nr_rqs
[i
];
399 drain
|= q
->in_flight
[i
];
401 drain
|= !list_empty(&fq
->flush_queue
[i
]);
408 spin_unlock_irq(q
->queue_lock
);
412 spin_lock_irq(q
->queue_lock
);
416 * With queue marked dead, any woken up waiter will fail the
417 * allocation path, so the wakeup chaining is lost and we're
418 * left with hung waiters. We need to wake up those waiters.
421 struct request_list
*rl
;
423 blk_queue_for_each_rl(rl
, q
)
424 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
425 wake_up_all(&rl
->wait
[i
]);
430 * blk_queue_bypass_start - enter queue bypass mode
431 * @q: queue of interest
433 * In bypass mode, only the dispatch FIFO queue of @q is used. This
434 * function makes @q enter bypass mode and drains all requests which were
435 * throttled or issued before. On return, it's guaranteed that no request
436 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
437 * inside queue or RCU read lock.
439 void blk_queue_bypass_start(struct request_queue
*q
)
441 spin_lock_irq(q
->queue_lock
);
443 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
444 spin_unlock_irq(q
->queue_lock
);
447 * Queues start drained. Skip actual draining till init is
448 * complete. This avoids lenghty delays during queue init which
449 * can happen many times during boot.
451 if (blk_queue_init_done(q
)) {
452 spin_lock_irq(q
->queue_lock
);
453 __blk_drain_queue(q
, false);
454 spin_unlock_irq(q
->queue_lock
);
456 /* ensure blk_queue_bypass() is %true inside RCU read lock */
460 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
463 * blk_queue_bypass_end - leave queue bypass mode
464 * @q: queue of interest
466 * Leave bypass mode and restore the normal queueing behavior.
468 void blk_queue_bypass_end(struct request_queue
*q
)
470 spin_lock_irq(q
->queue_lock
);
471 if (!--q
->bypass_depth
)
472 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
473 WARN_ON_ONCE(q
->bypass_depth
< 0);
474 spin_unlock_irq(q
->queue_lock
);
476 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
478 void blk_set_queue_dying(struct request_queue
*q
)
480 queue_flag_set_unlocked(QUEUE_FLAG_DYING
, q
);
483 blk_mq_wake_waiters(q
);
485 struct request_list
*rl
;
487 blk_queue_for_each_rl(rl
, q
) {
489 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
490 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
495 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
498 * blk_cleanup_queue - shutdown a request queue
499 * @q: request queue to shutdown
501 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
502 * put it. All future requests will be failed immediately with -ENODEV.
504 void blk_cleanup_queue(struct request_queue
*q
)
506 spinlock_t
*lock
= q
->queue_lock
;
508 /* mark @q DYING, no new request or merges will be allowed afterwards */
509 mutex_lock(&q
->sysfs_lock
);
510 blk_set_queue_dying(q
);
514 * A dying queue is permanently in bypass mode till released. Note
515 * that, unlike blk_queue_bypass_start(), we aren't performing
516 * synchronize_rcu() after entering bypass mode to avoid the delay
517 * as some drivers create and destroy a lot of queues while
518 * probing. This is still safe because blk_release_queue() will be
519 * called only after the queue refcnt drops to zero and nothing,
520 * RCU or not, would be traversing the queue by then.
523 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
525 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
526 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
527 queue_flag_set(QUEUE_FLAG_DYING
, q
);
528 spin_unlock_irq(lock
);
529 mutex_unlock(&q
->sysfs_lock
);
532 * Drain all requests queued before DYING marking. Set DEAD flag to
533 * prevent that q->request_fn() gets invoked after draining finished.
536 blk_mq_freeze_queue(q
);
540 __blk_drain_queue(q
, true);
542 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
543 spin_unlock_irq(lock
);
545 /* @q won't process any more request, flush async actions */
546 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
550 blk_mq_free_queue(q
);
553 if (q
->queue_lock
!= &q
->__queue_lock
)
554 q
->queue_lock
= &q
->__queue_lock
;
555 spin_unlock_irq(lock
);
557 /* @q is and will stay empty, shutdown and put */
560 EXPORT_SYMBOL(blk_cleanup_queue
);
562 /* Allocate memory local to the request queue */
563 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
565 int nid
= (int)(long)data
;
566 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
569 static void free_request_struct(void *element
, void *unused
)
571 kmem_cache_free(request_cachep
, element
);
574 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
577 if (unlikely(rl
->rq_pool
))
581 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
582 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
583 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
584 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
586 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
588 (void *)(long)q
->node
, gfp_mask
,
596 void blk_exit_rl(struct request_list
*rl
)
599 mempool_destroy(rl
->rq_pool
);
602 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
604 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
606 EXPORT_SYMBOL(blk_alloc_queue
);
608 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
610 struct request_queue
*q
;
613 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
614 gfp_mask
| __GFP_ZERO
, node_id
);
618 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
622 q
->backing_dev_info
.ra_pages
=
623 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
624 q
->backing_dev_info
.state
= 0;
625 q
->backing_dev_info
.capabilities
= 0;
626 q
->backing_dev_info
.name
= "block";
629 err
= bdi_init(&q
->backing_dev_info
);
633 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
634 laptop_mode_timer_fn
, (unsigned long) q
);
635 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
636 INIT_LIST_HEAD(&q
->queue_head
);
637 INIT_LIST_HEAD(&q
->timeout_list
);
638 INIT_LIST_HEAD(&q
->icq_list
);
639 #ifdef CONFIG_BLK_CGROUP
640 INIT_LIST_HEAD(&q
->blkg_list
);
642 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
644 kobject_init(&q
->kobj
, &blk_queue_ktype
);
646 mutex_init(&q
->sysfs_lock
);
647 spin_lock_init(&q
->__queue_lock
);
650 * By default initialize queue_lock to internal lock and driver can
651 * override it later if need be.
653 q
->queue_lock
= &q
->__queue_lock
;
656 * A queue starts its life with bypass turned on to avoid
657 * unnecessary bypass on/off overhead and nasty surprises during
658 * init. The initial bypass will be finished when the queue is
659 * registered by blk_register_queue().
662 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
664 init_waitqueue_head(&q
->mq_freeze_wq
);
666 if (blkcg_init_queue(q
))
672 bdi_destroy(&q
->backing_dev_info
);
674 ida_simple_remove(&blk_queue_ida
, q
->id
);
676 kmem_cache_free(blk_requestq_cachep
, q
);
679 EXPORT_SYMBOL(blk_alloc_queue_node
);
682 * blk_init_queue - prepare a request queue for use with a block device
683 * @rfn: The function to be called to process requests that have been
684 * placed on the queue.
685 * @lock: Request queue spin lock
688 * If a block device wishes to use the standard request handling procedures,
689 * which sorts requests and coalesces adjacent requests, then it must
690 * call blk_init_queue(). The function @rfn will be called when there
691 * are requests on the queue that need to be processed. If the device
692 * supports plugging, then @rfn may not be called immediately when requests
693 * are available on the queue, but may be called at some time later instead.
694 * Plugged queues are generally unplugged when a buffer belonging to one
695 * of the requests on the queue is needed, or due to memory pressure.
697 * @rfn is not required, or even expected, to remove all requests off the
698 * queue, but only as many as it can handle at a time. If it does leave
699 * requests on the queue, it is responsible for arranging that the requests
700 * get dealt with eventually.
702 * The queue spin lock must be held while manipulating the requests on the
703 * request queue; this lock will be taken also from interrupt context, so irq
704 * disabling is needed for it.
706 * Function returns a pointer to the initialized request queue, or %NULL if
710 * blk_init_queue() must be paired with a blk_cleanup_queue() call
711 * when the block device is deactivated (such as at module unload).
714 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
716 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
718 EXPORT_SYMBOL(blk_init_queue
);
720 struct request_queue
*
721 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
723 struct request_queue
*uninit_q
, *q
;
725 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
729 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
731 blk_cleanup_queue(uninit_q
);
735 EXPORT_SYMBOL(blk_init_queue_node
);
737 struct request_queue
*
738 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
744 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
748 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
752 q
->prep_rq_fn
= NULL
;
753 q
->unprep_rq_fn
= NULL
;
754 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
756 /* Override internal queue lock with supplied lock pointer */
758 q
->queue_lock
= lock
;
761 * This also sets hw/phys segments, boundary and size
763 blk_queue_make_request(q
, blk_queue_bio
);
765 q
->sg_reserved_size
= INT_MAX
;
767 /* Protect q->elevator from elevator_change */
768 mutex_lock(&q
->sysfs_lock
);
771 if (elevator_init(q
, NULL
)) {
772 mutex_unlock(&q
->sysfs_lock
);
776 mutex_unlock(&q
->sysfs_lock
);
781 blk_free_flush_queue(q
->fq
);
784 EXPORT_SYMBOL(blk_init_allocated_queue
);
786 bool blk_get_queue(struct request_queue
*q
)
788 if (likely(!blk_queue_dying(q
))) {
795 EXPORT_SYMBOL(blk_get_queue
);
797 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
799 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
800 elv_put_request(rl
->q
, rq
);
802 put_io_context(rq
->elv
.icq
->ioc
);
805 mempool_free(rq
, rl
->rq_pool
);
809 * ioc_batching returns true if the ioc is a valid batching request and
810 * should be given priority access to a request.
812 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
818 * Make sure the process is able to allocate at least 1 request
819 * even if the batch times out, otherwise we could theoretically
822 return ioc
->nr_batch_requests
== q
->nr_batching
||
823 (ioc
->nr_batch_requests
> 0
824 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
828 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
829 * will cause the process to be a "batcher" on all queues in the system. This
830 * is the behaviour we want though - once it gets a wakeup it should be given
833 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
835 if (!ioc
|| ioc_batching(q
, ioc
))
838 ioc
->nr_batch_requests
= q
->nr_batching
;
839 ioc
->last_waited
= jiffies
;
842 static void __freed_request(struct request_list
*rl
, int sync
)
844 struct request_queue
*q
= rl
->q
;
847 * bdi isn't aware of blkcg yet. As all async IOs end up root
848 * blkcg anyway, just use root blkcg state.
850 if (rl
== &q
->root_rl
&&
851 rl
->count
[sync
] < queue_congestion_off_threshold(q
))
852 blk_clear_queue_congested(q
, sync
);
854 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
855 if (waitqueue_active(&rl
->wait
[sync
]))
856 wake_up(&rl
->wait
[sync
]);
858 blk_clear_rl_full(rl
, sync
);
863 * A request has just been released. Account for it, update the full and
864 * congestion status, wake up any waiters. Called under q->queue_lock.
866 static void freed_request(struct request_list
*rl
, unsigned int flags
)
868 struct request_queue
*q
= rl
->q
;
869 int sync
= rw_is_sync(flags
);
873 if (flags
& REQ_ELVPRIV
)
876 __freed_request(rl
, sync
);
878 if (unlikely(rl
->starved
[sync
^ 1]))
879 __freed_request(rl
, sync
^ 1);
882 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
884 struct request_list
*rl
;
886 spin_lock_irq(q
->queue_lock
);
888 blk_queue_congestion_threshold(q
);
890 /* congestion isn't cgroup aware and follows root blkcg for now */
893 if (rl
->count
[BLK_RW_SYNC
] >= queue_congestion_on_threshold(q
))
894 blk_set_queue_congested(q
, BLK_RW_SYNC
);
895 else if (rl
->count
[BLK_RW_SYNC
] < queue_congestion_off_threshold(q
))
896 blk_clear_queue_congested(q
, BLK_RW_SYNC
);
898 if (rl
->count
[BLK_RW_ASYNC
] >= queue_congestion_on_threshold(q
))
899 blk_set_queue_congested(q
, BLK_RW_ASYNC
);
900 else if (rl
->count
[BLK_RW_ASYNC
] < queue_congestion_off_threshold(q
))
901 blk_clear_queue_congested(q
, BLK_RW_ASYNC
);
903 blk_queue_for_each_rl(rl
, q
) {
904 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
905 blk_set_rl_full(rl
, BLK_RW_SYNC
);
907 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
908 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
911 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
912 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
914 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
915 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
919 spin_unlock_irq(q
->queue_lock
);
924 * Determine if elevator data should be initialized when allocating the
925 * request associated with @bio.
927 static bool blk_rq_should_init_elevator(struct bio
*bio
)
933 * Flush requests do not use the elevator so skip initialization.
934 * This allows a request to share the flush and elevator data.
936 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
943 * rq_ioc - determine io_context for request allocation
944 * @bio: request being allocated is for this bio (can be %NULL)
946 * Determine io_context to use for request allocation for @bio. May return
947 * %NULL if %current->io_context doesn't exist.
949 static struct io_context
*rq_ioc(struct bio
*bio
)
951 #ifdef CONFIG_BLK_CGROUP
952 if (bio
&& bio
->bi_ioc
)
955 return current
->io_context
;
959 * __get_request - get a free request
960 * @rl: request list to allocate from
961 * @rw_flags: RW and SYNC flags
962 * @bio: bio to allocate request for (can be %NULL)
963 * @gfp_mask: allocation mask
965 * Get a free request from @q. This function may fail under memory
966 * pressure or if @q is dead.
968 * Must be called with @q->queue_lock held and,
969 * Returns ERR_PTR on failure, with @q->queue_lock held.
970 * Returns request pointer on success, with @q->queue_lock *not held*.
972 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
973 struct bio
*bio
, gfp_t gfp_mask
)
975 struct request_queue
*q
= rl
->q
;
977 struct elevator_type
*et
= q
->elevator
->type
;
978 struct io_context
*ioc
= rq_ioc(bio
);
979 struct io_cq
*icq
= NULL
;
980 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
983 if (unlikely(blk_queue_dying(q
)))
984 return ERR_PTR(-ENODEV
);
986 may_queue
= elv_may_queue(q
, rw_flags
);
987 if (may_queue
== ELV_MQUEUE_NO
)
990 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
991 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
993 * The queue will fill after this allocation, so set
994 * it as full, and mark this process as "batching".
995 * This process will be allowed to complete a batch of
996 * requests, others will be blocked.
998 if (!blk_rl_full(rl
, is_sync
)) {
999 ioc_set_batching(q
, ioc
);
1000 blk_set_rl_full(rl
, is_sync
);
1002 if (may_queue
!= ELV_MQUEUE_MUST
1003 && !ioc_batching(q
, ioc
)) {
1005 * The queue is full and the allocating
1006 * process is not a "batcher", and not
1007 * exempted by the IO scheduler
1009 return ERR_PTR(-ENOMEM
);
1014 * bdi isn't aware of blkcg yet. As all async IOs end up
1015 * root blkcg anyway, just use root blkcg state.
1017 if (rl
== &q
->root_rl
)
1018 blk_set_queue_congested(q
, is_sync
);
1022 * Only allow batching queuers to allocate up to 50% over the defined
1023 * limit of requests, otherwise we could have thousands of requests
1024 * allocated with any setting of ->nr_requests
1026 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1027 return ERR_PTR(-ENOMEM
);
1029 q
->nr_rqs
[is_sync
]++;
1030 rl
->count
[is_sync
]++;
1031 rl
->starved
[is_sync
] = 0;
1034 * Decide whether the new request will be managed by elevator. If
1035 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1036 * prevent the current elevator from being destroyed until the new
1037 * request is freed. This guarantees icq's won't be destroyed and
1038 * makes creating new ones safe.
1040 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1041 * it will be created after releasing queue_lock.
1043 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1044 rw_flags
|= REQ_ELVPRIV
;
1045 q
->nr_rqs_elvpriv
++;
1046 if (et
->icq_cache
&& ioc
)
1047 icq
= ioc_lookup_icq(ioc
, q
);
1050 if (blk_queue_io_stat(q
))
1051 rw_flags
|= REQ_IO_STAT
;
1052 spin_unlock_irq(q
->queue_lock
);
1054 /* allocate and init request */
1055 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1060 blk_rq_set_rl(rq
, rl
);
1061 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
1064 if (rw_flags
& REQ_ELVPRIV
) {
1065 if (unlikely(et
->icq_cache
&& !icq
)) {
1067 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1073 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1076 /* @rq->elv.icq holds io_context until @rq is freed */
1078 get_io_context(icq
->ioc
);
1082 * ioc may be NULL here, and ioc_batching will be false. That's
1083 * OK, if the queue is under the request limit then requests need
1084 * not count toward the nr_batch_requests limit. There will always
1085 * be some limit enforced by BLK_BATCH_TIME.
1087 if (ioc_batching(q
, ioc
))
1088 ioc
->nr_batch_requests
--;
1090 trace_block_getrq(q
, bio
, rw_flags
& 1);
1095 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1096 * and may fail indefinitely under memory pressure and thus
1097 * shouldn't stall IO. Treat this request as !elvpriv. This will
1098 * disturb iosched and blkcg but weird is bettern than dead.
1100 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1101 __func__
, dev_name(q
->backing_dev_info
.dev
));
1103 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1106 spin_lock_irq(q
->queue_lock
);
1107 q
->nr_rqs_elvpriv
--;
1108 spin_unlock_irq(q
->queue_lock
);
1113 * Allocation failed presumably due to memory. Undo anything we
1114 * might have messed up.
1116 * Allocating task should really be put onto the front of the wait
1117 * queue, but this is pretty rare.
1119 spin_lock_irq(q
->queue_lock
);
1120 freed_request(rl
, rw_flags
);
1123 * in the very unlikely event that allocation failed and no
1124 * requests for this direction was pending, mark us starved so that
1125 * freeing of a request in the other direction will notice
1126 * us. another possible fix would be to split the rq mempool into
1130 if (unlikely(rl
->count
[is_sync
] == 0))
1131 rl
->starved
[is_sync
] = 1;
1132 return ERR_PTR(-ENOMEM
);
1136 * get_request - get a free request
1137 * @q: request_queue to allocate request from
1138 * @rw_flags: RW and SYNC flags
1139 * @bio: bio to allocate request for (can be %NULL)
1140 * @gfp_mask: allocation mask
1142 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1143 * function keeps retrying under memory pressure and fails iff @q is dead.
1145 * Must be called with @q->queue_lock held and,
1146 * Returns ERR_PTR on failure, with @q->queue_lock held.
1147 * Returns request pointer on success, with @q->queue_lock *not held*.
1149 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1150 struct bio
*bio
, gfp_t gfp_mask
)
1152 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1154 struct request_list
*rl
;
1157 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1159 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1163 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dying(q
))) {
1168 /* wait on @rl and retry */
1169 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1170 TASK_UNINTERRUPTIBLE
);
1172 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1174 spin_unlock_irq(q
->queue_lock
);
1178 * After sleeping, we become a "batching" process and will be able
1179 * to allocate at least one request, and up to a big batch of them
1180 * for a small period time. See ioc_batching, ioc_set_batching
1182 ioc_set_batching(q
, current
->io_context
);
1184 spin_lock_irq(q
->queue_lock
);
1185 finish_wait(&rl
->wait
[is_sync
], &wait
);
1190 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1195 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1197 /* create ioc upfront */
1198 create_io_context(gfp_mask
, q
->node
);
1200 spin_lock_irq(q
->queue_lock
);
1201 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1203 spin_unlock_irq(q
->queue_lock
);
1204 /* q->queue_lock is unlocked at this point */
1209 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1212 return blk_mq_alloc_request(q
, rw
, gfp_mask
, false);
1214 return blk_old_get_request(q
, rw
, gfp_mask
);
1216 EXPORT_SYMBOL(blk_get_request
);
1219 * blk_make_request - given a bio, allocate a corresponding struct request.
1220 * @q: target request queue
1221 * @bio: The bio describing the memory mappings that will be submitted for IO.
1222 * It may be a chained-bio properly constructed by block/bio layer.
1223 * @gfp_mask: gfp flags to be used for memory allocation
1225 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1226 * type commands. Where the struct request needs to be farther initialized by
1227 * the caller. It is passed a &struct bio, which describes the memory info of
1230 * The caller of blk_make_request must make sure that bi_io_vec
1231 * are set to describe the memory buffers. That bio_data_dir() will return
1232 * the needed direction of the request. (And all bio's in the passed bio-chain
1233 * are properly set accordingly)
1235 * If called under none-sleepable conditions, mapped bio buffers must not
1236 * need bouncing, by calling the appropriate masked or flagged allocator,
1237 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1240 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1241 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1242 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1243 * completion of a bio that hasn't been submitted yet, thus resulting in a
1244 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1245 * of bio_alloc(), as that avoids the mempool deadlock.
1246 * If possible a big IO should be split into smaller parts when allocation
1247 * fails. Partial allocation should not be an error, or you risk a live-lock.
1249 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1252 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1257 blk_rq_set_block_pc(rq
);
1260 struct bio
*bounce_bio
= bio
;
1263 blk_queue_bounce(q
, &bounce_bio
);
1264 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1265 if (unlikely(ret
)) {
1266 blk_put_request(rq
);
1267 return ERR_PTR(ret
);
1273 EXPORT_SYMBOL(blk_make_request
);
1276 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1277 * @rq: request to be initialized
1280 void blk_rq_set_block_pc(struct request
*rq
)
1282 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1284 rq
->__sector
= (sector_t
) -1;
1285 rq
->bio
= rq
->biotail
= NULL
;
1286 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1288 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1291 * blk_requeue_request - put a request back on queue
1292 * @q: request queue where request should be inserted
1293 * @rq: request to be inserted
1296 * Drivers often keep queueing requests until the hardware cannot accept
1297 * more, when that condition happens we need to put the request back
1298 * on the queue. Must be called with queue lock held.
1300 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1302 blk_delete_timer(rq
);
1303 blk_clear_rq_complete(rq
);
1304 trace_block_rq_requeue(q
, rq
);
1306 if (rq
->cmd_flags
& REQ_QUEUED
)
1307 blk_queue_end_tag(q
, rq
);
1309 BUG_ON(blk_queued_rq(rq
));
1311 elv_requeue_request(q
, rq
);
1313 EXPORT_SYMBOL(blk_requeue_request
);
1315 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1318 blk_account_io_start(rq
, true);
1319 __elv_add_request(q
, rq
, where
);
1322 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1327 if (now
== part
->stamp
)
1330 inflight
= part_in_flight(part
);
1332 __part_stat_add(cpu
, part
, time_in_queue
,
1333 inflight
* (now
- part
->stamp
));
1334 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1340 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1341 * @cpu: cpu number for stats access
1342 * @part: target partition
1344 * The average IO queue length and utilisation statistics are maintained
1345 * by observing the current state of the queue length and the amount of
1346 * time it has been in this state for.
1348 * Normally, that accounting is done on IO completion, but that can result
1349 * in more than a second's worth of IO being accounted for within any one
1350 * second, leading to >100% utilisation. To deal with that, we call this
1351 * function to do a round-off before returning the results when reading
1352 * /proc/diskstats. This accounts immediately for all queue usage up to
1353 * the current jiffies and restarts the counters again.
1355 void part_round_stats(int cpu
, struct hd_struct
*part
)
1357 unsigned long now
= jiffies
;
1360 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1361 part_round_stats_single(cpu
, part
, now
);
1363 EXPORT_SYMBOL_GPL(part_round_stats
);
1366 static void blk_pm_put_request(struct request
*rq
)
1368 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1369 pm_runtime_mark_last_busy(rq
->q
->dev
);
1372 static inline void blk_pm_put_request(struct request
*rq
) {}
1376 * queue lock must be held
1378 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1384 blk_mq_free_request(req
);
1388 blk_pm_put_request(req
);
1390 elv_completed_request(q
, req
);
1392 /* this is a bio leak */
1393 WARN_ON(req
->bio
!= NULL
);
1396 * Request may not have originated from ll_rw_blk. if not,
1397 * it didn't come out of our reserved rq pools
1399 if (req
->cmd_flags
& REQ_ALLOCED
) {
1400 unsigned int flags
= req
->cmd_flags
;
1401 struct request_list
*rl
= blk_rq_rl(req
);
1403 BUG_ON(!list_empty(&req
->queuelist
));
1404 BUG_ON(ELV_ON_HASH(req
));
1406 blk_free_request(rl
, req
);
1407 freed_request(rl
, flags
);
1411 EXPORT_SYMBOL_GPL(__blk_put_request
);
1413 void blk_put_request(struct request
*req
)
1415 struct request_queue
*q
= req
->q
;
1418 blk_mq_free_request(req
);
1420 unsigned long flags
;
1422 spin_lock_irqsave(q
->queue_lock
, flags
);
1423 __blk_put_request(q
, req
);
1424 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1427 EXPORT_SYMBOL(blk_put_request
);
1430 * blk_add_request_payload - add a payload to a request
1431 * @rq: request to update
1432 * @page: page backing the payload
1433 * @len: length of the payload.
1435 * This allows to later add a payload to an already submitted request by
1436 * a block driver. The driver needs to take care of freeing the payload
1439 * Note that this is a quite horrible hack and nothing but handling of
1440 * discard requests should ever use it.
1442 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1445 struct bio
*bio
= rq
->bio
;
1447 bio
->bi_io_vec
->bv_page
= page
;
1448 bio
->bi_io_vec
->bv_offset
= 0;
1449 bio
->bi_io_vec
->bv_len
= len
;
1451 bio
->bi_iter
.bi_size
= len
;
1453 bio
->bi_phys_segments
= 1;
1455 rq
->__data_len
= rq
->resid_len
= len
;
1456 rq
->nr_phys_segments
= 1;
1458 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1460 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1463 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1465 if (!ll_back_merge_fn(q
, req
, bio
))
1468 trace_block_bio_backmerge(q
, req
, bio
);
1470 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1471 blk_rq_set_mixed_merge(req
);
1473 req
->biotail
->bi_next
= bio
;
1475 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1476 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1478 blk_account_io_start(req
, false);
1482 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1485 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1487 if (!ll_front_merge_fn(q
, req
, bio
))
1490 trace_block_bio_frontmerge(q
, req
, bio
);
1492 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1493 blk_rq_set_mixed_merge(req
);
1495 bio
->bi_next
= req
->bio
;
1498 req
->__sector
= bio
->bi_iter
.bi_sector
;
1499 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1500 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1502 blk_account_io_start(req
, false);
1507 * blk_attempt_plug_merge - try to merge with %current's plugged list
1508 * @q: request_queue new bio is being queued at
1509 * @bio: new bio being queued
1510 * @request_count: out parameter for number of traversed plugged requests
1512 * Determine whether @bio being queued on @q can be merged with a request
1513 * on %current's plugged list. Returns %true if merge was successful,
1516 * Plugging coalesces IOs from the same issuer for the same purpose without
1517 * going through @q->queue_lock. As such it's more of an issuing mechanism
1518 * than scheduling, and the request, while may have elvpriv data, is not
1519 * added on the elevator at this point. In addition, we don't have
1520 * reliable access to the elevator outside queue lock. Only check basic
1521 * merging parameters without querying the elevator.
1523 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1525 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1526 unsigned int *request_count
,
1527 struct request
**same_queue_rq
)
1529 struct blk_plug
*plug
;
1532 struct list_head
*plug_list
;
1534 plug
= current
->plug
;
1540 plug_list
= &plug
->mq_list
;
1542 plug_list
= &plug
->list
;
1544 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1550 * Only blk-mq multiple hardware queues case checks the
1551 * rq in the same queue, there should be only one such
1555 *same_queue_rq
= rq
;
1558 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1561 el_ret
= blk_try_merge(rq
, bio
);
1562 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1563 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1566 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1567 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1576 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1578 req
->cmd_type
= REQ_TYPE_FS
;
1580 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1581 if (bio
->bi_rw
& REQ_RAHEAD
)
1582 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1585 req
->__sector
= bio
->bi_iter
.bi_sector
;
1586 req
->ioprio
= bio_prio(bio
);
1587 blk_rq_bio_prep(req
->q
, req
, bio
);
1590 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1592 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1593 struct blk_plug
*plug
;
1594 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1595 struct request
*req
;
1596 unsigned int request_count
= 0;
1599 * low level driver can indicate that it wants pages above a
1600 * certain limit bounced to low memory (ie for highmem, or even
1601 * ISA dma in theory)
1603 blk_queue_bounce(q
, &bio
);
1605 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1606 bio_endio(bio
, -EIO
);
1610 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1611 spin_lock_irq(q
->queue_lock
);
1612 where
= ELEVATOR_INSERT_FLUSH
;
1617 * Check if we can merge with the plugged list before grabbing
1620 if (!blk_queue_nomerges(q
) &&
1621 blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1624 spin_lock_irq(q
->queue_lock
);
1626 el_ret
= elv_merge(q
, &req
, bio
);
1627 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1628 if (bio_attempt_back_merge(q
, req
, bio
)) {
1629 elv_bio_merged(q
, req
, bio
);
1630 if (!attempt_back_merge(q
, req
))
1631 elv_merged_request(q
, req
, el_ret
);
1634 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1635 if (bio_attempt_front_merge(q
, req
, bio
)) {
1636 elv_bio_merged(q
, req
, bio
);
1637 if (!attempt_front_merge(q
, req
))
1638 elv_merged_request(q
, req
, el_ret
);
1645 * This sync check and mask will be re-done in init_request_from_bio(),
1646 * but we need to set it earlier to expose the sync flag to the
1647 * rq allocator and io schedulers.
1649 rw_flags
= bio_data_dir(bio
);
1651 rw_flags
|= REQ_SYNC
;
1654 * Grab a free request. This is might sleep but can not fail.
1655 * Returns with the queue unlocked.
1657 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1659 bio_endio(bio
, PTR_ERR(req
)); /* @q is dead */
1664 * After dropping the lock and possibly sleeping here, our request
1665 * may now be mergeable after it had proven unmergeable (above).
1666 * We don't worry about that case for efficiency. It won't happen
1667 * often, and the elevators are able to handle it.
1669 init_request_from_bio(req
, bio
);
1671 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1672 req
->cpu
= raw_smp_processor_id();
1674 plug
= current
->plug
;
1677 * If this is the first request added after a plug, fire
1681 trace_block_plug(q
);
1683 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1684 blk_flush_plug_list(plug
, false);
1685 trace_block_plug(q
);
1688 list_add_tail(&req
->queuelist
, &plug
->list
);
1689 blk_account_io_start(req
, true);
1691 spin_lock_irq(q
->queue_lock
);
1692 add_acct_request(q
, req
, where
);
1695 spin_unlock_irq(q
->queue_lock
);
1698 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1701 * If bio->bi_dev is a partition, remap the location
1703 static inline void blk_partition_remap(struct bio
*bio
)
1705 struct block_device
*bdev
= bio
->bi_bdev
;
1707 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1708 struct hd_struct
*p
= bdev
->bd_part
;
1710 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1711 bio
->bi_bdev
= bdev
->bd_contains
;
1713 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1715 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1719 static void handle_bad_sector(struct bio
*bio
)
1721 char b
[BDEVNAME_SIZE
];
1723 printk(KERN_INFO
"attempt to access beyond end of device\n");
1724 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1725 bdevname(bio
->bi_bdev
, b
),
1727 (unsigned long long)bio_end_sector(bio
),
1728 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1731 #ifdef CONFIG_FAIL_MAKE_REQUEST
1733 static DECLARE_FAULT_ATTR(fail_make_request
);
1735 static int __init
setup_fail_make_request(char *str
)
1737 return setup_fault_attr(&fail_make_request
, str
);
1739 __setup("fail_make_request=", setup_fail_make_request
);
1741 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1743 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1746 static int __init
fail_make_request_debugfs(void)
1748 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1749 NULL
, &fail_make_request
);
1751 return PTR_ERR_OR_ZERO(dir
);
1754 late_initcall(fail_make_request_debugfs
);
1756 #else /* CONFIG_FAIL_MAKE_REQUEST */
1758 static inline bool should_fail_request(struct hd_struct
*part
,
1764 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1767 * Check whether this bio extends beyond the end of the device.
1769 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1776 /* Test device or partition size, when known. */
1777 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1779 sector_t sector
= bio
->bi_iter
.bi_sector
;
1781 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1783 * This may well happen - the kernel calls bread()
1784 * without checking the size of the device, e.g., when
1785 * mounting a device.
1787 handle_bad_sector(bio
);
1795 static noinline_for_stack
bool
1796 generic_make_request_checks(struct bio
*bio
)
1798 struct request_queue
*q
;
1799 int nr_sectors
= bio_sectors(bio
);
1801 char b
[BDEVNAME_SIZE
];
1802 struct hd_struct
*part
;
1806 if (bio_check_eod(bio
, nr_sectors
))
1809 q
= bdev_get_queue(bio
->bi_bdev
);
1812 "generic_make_request: Trying to access "
1813 "nonexistent block-device %s (%Lu)\n",
1814 bdevname(bio
->bi_bdev
, b
),
1815 (long long) bio
->bi_iter
.bi_sector
);
1819 if (likely(bio_is_rw(bio
) &&
1820 nr_sectors
> queue_max_hw_sectors(q
))) {
1821 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1822 bdevname(bio
->bi_bdev
, b
),
1824 queue_max_hw_sectors(q
));
1828 part
= bio
->bi_bdev
->bd_part
;
1829 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1830 should_fail_request(&part_to_disk(part
)->part0
,
1831 bio
->bi_iter
.bi_size
))
1835 * If this device has partitions, remap block n
1836 * of partition p to block n+start(p) of the disk.
1838 blk_partition_remap(bio
);
1840 if (bio_check_eod(bio
, nr_sectors
))
1844 * Filter flush bio's early so that make_request based
1845 * drivers without flush support don't have to worry
1848 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1849 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1856 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1857 (!blk_queue_discard(q
) ||
1858 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1863 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1869 * Various block parts want %current->io_context and lazy ioc
1870 * allocation ends up trading a lot of pain for a small amount of
1871 * memory. Just allocate it upfront. This may fail and block
1872 * layer knows how to live with it.
1874 create_io_context(GFP_ATOMIC
, q
->node
);
1876 if (blk_throtl_bio(q
, bio
))
1877 return false; /* throttled, will be resubmitted later */
1879 trace_block_bio_queue(q
, bio
);
1883 bio_endio(bio
, err
);
1888 * generic_make_request - hand a buffer to its device driver for I/O
1889 * @bio: The bio describing the location in memory and on the device.
1891 * generic_make_request() is used to make I/O requests of block
1892 * devices. It is passed a &struct bio, which describes the I/O that needs
1895 * generic_make_request() does not return any status. The
1896 * success/failure status of the request, along with notification of
1897 * completion, is delivered asynchronously through the bio->bi_end_io
1898 * function described (one day) else where.
1900 * The caller of generic_make_request must make sure that bi_io_vec
1901 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1902 * set to describe the device address, and the
1903 * bi_end_io and optionally bi_private are set to describe how
1904 * completion notification should be signaled.
1906 * generic_make_request and the drivers it calls may use bi_next if this
1907 * bio happens to be merged with someone else, and may resubmit the bio to
1908 * a lower device by calling into generic_make_request recursively, which
1909 * means the bio should NOT be touched after the call to ->make_request_fn.
1911 void generic_make_request(struct bio
*bio
)
1913 struct bio_list bio_list_on_stack
;
1915 if (!generic_make_request_checks(bio
))
1919 * We only want one ->make_request_fn to be active at a time, else
1920 * stack usage with stacked devices could be a problem. So use
1921 * current->bio_list to keep a list of requests submited by a
1922 * make_request_fn function. current->bio_list is also used as a
1923 * flag to say if generic_make_request is currently active in this
1924 * task or not. If it is NULL, then no make_request is active. If
1925 * it is non-NULL, then a make_request is active, and new requests
1926 * should be added at the tail
1928 if (current
->bio_list
) {
1929 bio_list_add(current
->bio_list
, bio
);
1933 /* following loop may be a bit non-obvious, and so deserves some
1935 * Before entering the loop, bio->bi_next is NULL (as all callers
1936 * ensure that) so we have a list with a single bio.
1937 * We pretend that we have just taken it off a longer list, so
1938 * we assign bio_list to a pointer to the bio_list_on_stack,
1939 * thus initialising the bio_list of new bios to be
1940 * added. ->make_request() may indeed add some more bios
1941 * through a recursive call to generic_make_request. If it
1942 * did, we find a non-NULL value in bio_list and re-enter the loop
1943 * from the top. In this case we really did just take the bio
1944 * of the top of the list (no pretending) and so remove it from
1945 * bio_list, and call into ->make_request() again.
1947 BUG_ON(bio
->bi_next
);
1948 bio_list_init(&bio_list_on_stack
);
1949 current
->bio_list
= &bio_list_on_stack
;
1951 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1953 q
->make_request_fn(q
, bio
);
1955 bio
= bio_list_pop(current
->bio_list
);
1957 current
->bio_list
= NULL
; /* deactivate */
1959 EXPORT_SYMBOL(generic_make_request
);
1962 * submit_bio - submit a bio to the block device layer for I/O
1963 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1964 * @bio: The &struct bio which describes the I/O
1966 * submit_bio() is very similar in purpose to generic_make_request(), and
1967 * uses that function to do most of the work. Both are fairly rough
1968 * interfaces; @bio must be presetup and ready for I/O.
1971 void submit_bio(int rw
, struct bio
*bio
)
1976 * If it's a regular read/write or a barrier with data attached,
1977 * go through the normal accounting stuff before submission.
1979 if (bio_has_data(bio
)) {
1982 if (unlikely(rw
& REQ_WRITE_SAME
))
1983 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
1985 count
= bio_sectors(bio
);
1988 count_vm_events(PGPGOUT
, count
);
1990 task_io_account_read(bio
->bi_iter
.bi_size
);
1991 count_vm_events(PGPGIN
, count
);
1994 if (unlikely(block_dump
)) {
1995 char b
[BDEVNAME_SIZE
];
1996 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1997 current
->comm
, task_pid_nr(current
),
1998 (rw
& WRITE
) ? "WRITE" : "READ",
1999 (unsigned long long)bio
->bi_iter
.bi_sector
,
2000 bdevname(bio
->bi_bdev
, b
),
2005 generic_make_request(bio
);
2007 EXPORT_SYMBOL(submit_bio
);
2010 * blk_rq_check_limits - Helper function to check a request for the queue limit
2012 * @rq: the request being checked
2015 * @rq may have been made based on weaker limitations of upper-level queues
2016 * in request stacking drivers, and it may violate the limitation of @q.
2017 * Since the block layer and the underlying device driver trust @rq
2018 * after it is inserted to @q, it should be checked against @q before
2019 * the insertion using this generic function.
2021 * This function should also be useful for request stacking drivers
2022 * in some cases below, so export this function.
2023 * Request stacking drivers like request-based dm may change the queue
2024 * limits while requests are in the queue (e.g. dm's table swapping).
2025 * Such request stacking drivers should check those requests against
2026 * the new queue limits again when they dispatch those requests,
2027 * although such checkings are also done against the old queue limits
2028 * when submitting requests.
2030 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
2032 if (!rq_mergeable(rq
))
2035 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
2036 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2041 * queue's settings related to segment counting like q->bounce_pfn
2042 * may differ from that of other stacking queues.
2043 * Recalculate it to check the request correctly on this queue's
2046 blk_recalc_rq_segments(rq
);
2047 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2048 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2054 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
2057 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2058 * @q: the queue to submit the request
2059 * @rq: the request being queued
2061 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2063 unsigned long flags
;
2064 int where
= ELEVATOR_INSERT_BACK
;
2066 if (blk_rq_check_limits(q
, rq
))
2070 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2074 if (blk_queue_io_stat(q
))
2075 blk_account_io_start(rq
, true);
2076 blk_mq_insert_request(rq
, false, true, true);
2080 spin_lock_irqsave(q
->queue_lock
, flags
);
2081 if (unlikely(blk_queue_dying(q
))) {
2082 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2087 * Submitting request must be dequeued before calling this function
2088 * because it will be linked to another request_queue
2090 BUG_ON(blk_queued_rq(rq
));
2092 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
2093 where
= ELEVATOR_INSERT_FLUSH
;
2095 add_acct_request(q
, rq
, where
);
2096 if (where
== ELEVATOR_INSERT_FLUSH
)
2098 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2102 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2105 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2106 * @rq: request to examine
2109 * A request could be merge of IOs which require different failure
2110 * handling. This function determines the number of bytes which
2111 * can be failed from the beginning of the request without
2112 * crossing into area which need to be retried further.
2115 * The number of bytes to fail.
2118 * queue_lock must be held.
2120 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2122 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2123 unsigned int bytes
= 0;
2126 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2127 return blk_rq_bytes(rq
);
2130 * Currently the only 'mixing' which can happen is between
2131 * different fastfail types. We can safely fail portions
2132 * which have all the failfast bits that the first one has -
2133 * the ones which are at least as eager to fail as the first
2136 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2137 if ((bio
->bi_rw
& ff
) != ff
)
2139 bytes
+= bio
->bi_iter
.bi_size
;
2142 /* this could lead to infinite loop */
2143 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2146 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2148 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2150 if (blk_do_io_stat(req
)) {
2151 const int rw
= rq_data_dir(req
);
2152 struct hd_struct
*part
;
2155 cpu
= part_stat_lock();
2157 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2162 void blk_account_io_done(struct request
*req
)
2165 * Account IO completion. flush_rq isn't accounted as a
2166 * normal IO on queueing nor completion. Accounting the
2167 * containing request is enough.
2169 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2170 unsigned long duration
= jiffies
- req
->start_time
;
2171 const int rw
= rq_data_dir(req
);
2172 struct hd_struct
*part
;
2175 cpu
= part_stat_lock();
2178 part_stat_inc(cpu
, part
, ios
[rw
]);
2179 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2180 part_round_stats(cpu
, part
);
2181 part_dec_in_flight(part
, rw
);
2183 hd_struct_put(part
);
2190 * Don't process normal requests when queue is suspended
2191 * or in the process of suspending/resuming
2193 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2196 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2197 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2203 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2210 void blk_account_io_start(struct request
*rq
, bool new_io
)
2212 struct hd_struct
*part
;
2213 int rw
= rq_data_dir(rq
);
2216 if (!blk_do_io_stat(rq
))
2219 cpu
= part_stat_lock();
2223 part_stat_inc(cpu
, part
, merges
[rw
]);
2225 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2226 if (!hd_struct_try_get(part
)) {
2228 * The partition is already being removed,
2229 * the request will be accounted on the disk only
2231 * We take a reference on disk->part0 although that
2232 * partition will never be deleted, so we can treat
2233 * it as any other partition.
2235 part
= &rq
->rq_disk
->part0
;
2236 hd_struct_get(part
);
2238 part_round_stats(cpu
, part
);
2239 part_inc_in_flight(part
, rw
);
2247 * blk_peek_request - peek at the top of a request queue
2248 * @q: request queue to peek at
2251 * Return the request at the top of @q. The returned request
2252 * should be started using blk_start_request() before LLD starts
2256 * Pointer to the request at the top of @q if available. Null
2260 * queue_lock must be held.
2262 struct request
*blk_peek_request(struct request_queue
*q
)
2267 while ((rq
= __elv_next_request(q
)) != NULL
) {
2269 rq
= blk_pm_peek_request(q
, rq
);
2273 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2275 * This is the first time the device driver
2276 * sees this request (possibly after
2277 * requeueing). Notify IO scheduler.
2279 if (rq
->cmd_flags
& REQ_SORTED
)
2280 elv_activate_rq(q
, rq
);
2283 * just mark as started even if we don't start
2284 * it, a request that has been delayed should
2285 * not be passed by new incoming requests
2287 rq
->cmd_flags
|= REQ_STARTED
;
2288 trace_block_rq_issue(q
, rq
);
2291 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2292 q
->end_sector
= rq_end_sector(rq
);
2293 q
->boundary_rq
= NULL
;
2296 if (rq
->cmd_flags
& REQ_DONTPREP
)
2299 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2301 * make sure space for the drain appears we
2302 * know we can do this because max_hw_segments
2303 * has been adjusted to be one fewer than the
2306 rq
->nr_phys_segments
++;
2312 ret
= q
->prep_rq_fn(q
, rq
);
2313 if (ret
== BLKPREP_OK
) {
2315 } else if (ret
== BLKPREP_DEFER
) {
2317 * the request may have been (partially) prepped.
2318 * we need to keep this request in the front to
2319 * avoid resource deadlock. REQ_STARTED will
2320 * prevent other fs requests from passing this one.
2322 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2323 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2325 * remove the space for the drain we added
2326 * so that we don't add it again
2328 --rq
->nr_phys_segments
;
2333 } else if (ret
== BLKPREP_KILL
) {
2334 rq
->cmd_flags
|= REQ_QUIET
;
2336 * Mark this request as started so we don't trigger
2337 * any debug logic in the end I/O path.
2339 blk_start_request(rq
);
2340 __blk_end_request_all(rq
, -EIO
);
2342 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2349 EXPORT_SYMBOL(blk_peek_request
);
2351 void blk_dequeue_request(struct request
*rq
)
2353 struct request_queue
*q
= rq
->q
;
2355 BUG_ON(list_empty(&rq
->queuelist
));
2356 BUG_ON(ELV_ON_HASH(rq
));
2358 list_del_init(&rq
->queuelist
);
2361 * the time frame between a request being removed from the lists
2362 * and to it is freed is accounted as io that is in progress at
2365 if (blk_account_rq(rq
)) {
2366 q
->in_flight
[rq_is_sync(rq
)]++;
2367 set_io_start_time_ns(rq
);
2372 * blk_start_request - start request processing on the driver
2373 * @req: request to dequeue
2376 * Dequeue @req and start timeout timer on it. This hands off the
2377 * request to the driver.
2379 * Block internal functions which don't want to start timer should
2380 * call blk_dequeue_request().
2383 * queue_lock must be held.
2385 void blk_start_request(struct request
*req
)
2387 blk_dequeue_request(req
);
2390 * We are now handing the request to the hardware, initialize
2391 * resid_len to full count and add the timeout handler.
2393 req
->resid_len
= blk_rq_bytes(req
);
2394 if (unlikely(blk_bidi_rq(req
)))
2395 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2397 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2400 EXPORT_SYMBOL(blk_start_request
);
2403 * blk_fetch_request - fetch a request from a request queue
2404 * @q: request queue to fetch a request from
2407 * Return the request at the top of @q. The request is started on
2408 * return and LLD can start processing it immediately.
2411 * Pointer to the request at the top of @q if available. Null
2415 * queue_lock must be held.
2417 struct request
*blk_fetch_request(struct request_queue
*q
)
2421 rq
= blk_peek_request(q
);
2423 blk_start_request(rq
);
2426 EXPORT_SYMBOL(blk_fetch_request
);
2429 * blk_update_request - Special helper function for request stacking drivers
2430 * @req: the request being processed
2431 * @error: %0 for success, < %0 for error
2432 * @nr_bytes: number of bytes to complete @req
2435 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2436 * the request structure even if @req doesn't have leftover.
2437 * If @req has leftover, sets it up for the next range of segments.
2439 * This special helper function is only for request stacking drivers
2440 * (e.g. request-based dm) so that they can handle partial completion.
2441 * Actual device drivers should use blk_end_request instead.
2443 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2444 * %false return from this function.
2447 * %false - this request doesn't have any more data
2448 * %true - this request has more data
2450 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2454 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2460 * For fs requests, rq is just carrier of independent bio's
2461 * and each partial completion should be handled separately.
2462 * Reset per-request error on each partial completion.
2464 * TODO: tj: This is too subtle. It would be better to let
2465 * low level drivers do what they see fit.
2467 if (req
->cmd_type
== REQ_TYPE_FS
)
2470 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2471 !(req
->cmd_flags
& REQ_QUIET
)) {
2476 error_type
= "recoverable transport";
2479 error_type
= "critical target";
2482 error_type
= "critical nexus";
2485 error_type
= "timeout";
2488 error_type
= "critical space allocation";
2491 error_type
= "critical medium";
2498 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2499 __func__
, error_type
, req
->rq_disk
?
2500 req
->rq_disk
->disk_name
: "?",
2501 (unsigned long long)blk_rq_pos(req
));
2505 blk_account_io_completion(req
, nr_bytes
);
2509 struct bio
*bio
= req
->bio
;
2510 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2512 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2513 req
->bio
= bio
->bi_next
;
2515 req_bio_endio(req
, bio
, bio_bytes
, error
);
2517 total_bytes
+= bio_bytes
;
2518 nr_bytes
-= bio_bytes
;
2529 * Reset counters so that the request stacking driver
2530 * can find how many bytes remain in the request
2533 req
->__data_len
= 0;
2537 req
->__data_len
-= total_bytes
;
2539 /* update sector only for requests with clear definition of sector */
2540 if (req
->cmd_type
== REQ_TYPE_FS
)
2541 req
->__sector
+= total_bytes
>> 9;
2543 /* mixed attributes always follow the first bio */
2544 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2545 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2546 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2550 * If total number of sectors is less than the first segment
2551 * size, something has gone terribly wrong.
2553 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2554 blk_dump_rq_flags(req
, "request botched");
2555 req
->__data_len
= blk_rq_cur_bytes(req
);
2558 /* recalculate the number of segments */
2559 blk_recalc_rq_segments(req
);
2563 EXPORT_SYMBOL_GPL(blk_update_request
);
2565 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2566 unsigned int nr_bytes
,
2567 unsigned int bidi_bytes
)
2569 if (blk_update_request(rq
, error
, nr_bytes
))
2572 /* Bidi request must be completed as a whole */
2573 if (unlikely(blk_bidi_rq(rq
)) &&
2574 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2577 if (blk_queue_add_random(rq
->q
))
2578 add_disk_randomness(rq
->rq_disk
);
2584 * blk_unprep_request - unprepare a request
2587 * This function makes a request ready for complete resubmission (or
2588 * completion). It happens only after all error handling is complete,
2589 * so represents the appropriate moment to deallocate any resources
2590 * that were allocated to the request in the prep_rq_fn. The queue
2591 * lock is held when calling this.
2593 void blk_unprep_request(struct request
*req
)
2595 struct request_queue
*q
= req
->q
;
2597 req
->cmd_flags
&= ~REQ_DONTPREP
;
2598 if (q
->unprep_rq_fn
)
2599 q
->unprep_rq_fn(q
, req
);
2601 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2604 * queue lock must be held
2606 void blk_finish_request(struct request
*req
, int error
)
2608 if (req
->cmd_flags
& REQ_QUEUED
)
2609 blk_queue_end_tag(req
->q
, req
);
2611 BUG_ON(blk_queued_rq(req
));
2613 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2614 laptop_io_completion(&req
->q
->backing_dev_info
);
2616 blk_delete_timer(req
);
2618 if (req
->cmd_flags
& REQ_DONTPREP
)
2619 blk_unprep_request(req
);
2621 blk_account_io_done(req
);
2624 req
->end_io(req
, error
);
2626 if (blk_bidi_rq(req
))
2627 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2629 __blk_put_request(req
->q
, req
);
2632 EXPORT_SYMBOL(blk_finish_request
);
2635 * blk_end_bidi_request - Complete a bidi request
2636 * @rq: the request to complete
2637 * @error: %0 for success, < %0 for error
2638 * @nr_bytes: number of bytes to complete @rq
2639 * @bidi_bytes: number of bytes to complete @rq->next_rq
2642 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2643 * Drivers that supports bidi can safely call this member for any
2644 * type of request, bidi or uni. In the later case @bidi_bytes is
2648 * %false - we are done with this request
2649 * %true - still buffers pending for this request
2651 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2652 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2654 struct request_queue
*q
= rq
->q
;
2655 unsigned long flags
;
2657 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2660 spin_lock_irqsave(q
->queue_lock
, flags
);
2661 blk_finish_request(rq
, error
);
2662 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2668 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2669 * @rq: the request to complete
2670 * @error: %0 for success, < %0 for error
2671 * @nr_bytes: number of bytes to complete @rq
2672 * @bidi_bytes: number of bytes to complete @rq->next_rq
2675 * Identical to blk_end_bidi_request() except that queue lock is
2676 * assumed to be locked on entry and remains so on return.
2679 * %false - we are done with this request
2680 * %true - still buffers pending for this request
2682 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2683 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2685 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2688 blk_finish_request(rq
, error
);
2694 * blk_end_request - Helper function for drivers to complete the request.
2695 * @rq: the request being processed
2696 * @error: %0 for success, < %0 for error
2697 * @nr_bytes: number of bytes to complete
2700 * Ends I/O on a number of bytes attached to @rq.
2701 * If @rq has leftover, sets it up for the next range of segments.
2704 * %false - we are done with this request
2705 * %true - still buffers pending for this request
2707 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2709 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2711 EXPORT_SYMBOL(blk_end_request
);
2714 * blk_end_request_all - Helper function for drives to finish the request.
2715 * @rq: the request to finish
2716 * @error: %0 for success, < %0 for error
2719 * Completely finish @rq.
2721 void blk_end_request_all(struct request
*rq
, int error
)
2724 unsigned int bidi_bytes
= 0;
2726 if (unlikely(blk_bidi_rq(rq
)))
2727 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2729 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2732 EXPORT_SYMBOL(blk_end_request_all
);
2735 * blk_end_request_cur - Helper function to finish the current request chunk.
2736 * @rq: the request to finish the current chunk for
2737 * @error: %0 for success, < %0 for error
2740 * Complete the current consecutively mapped chunk from @rq.
2743 * %false - we are done with this request
2744 * %true - still buffers pending for this request
2746 bool blk_end_request_cur(struct request
*rq
, int error
)
2748 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2750 EXPORT_SYMBOL(blk_end_request_cur
);
2753 * blk_end_request_err - Finish a request till the next failure boundary.
2754 * @rq: the request to finish till the next failure boundary for
2755 * @error: must be negative errno
2758 * Complete @rq till the next failure boundary.
2761 * %false - we are done with this request
2762 * %true - still buffers pending for this request
2764 bool blk_end_request_err(struct request
*rq
, int error
)
2766 WARN_ON(error
>= 0);
2767 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2769 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2772 * __blk_end_request - Helper function for drivers to complete the request.
2773 * @rq: the request being processed
2774 * @error: %0 for success, < %0 for error
2775 * @nr_bytes: number of bytes to complete
2778 * Must be called with queue lock held unlike blk_end_request().
2781 * %false - we are done with this request
2782 * %true - still buffers pending for this request
2784 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2786 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2788 EXPORT_SYMBOL(__blk_end_request
);
2791 * __blk_end_request_all - Helper function for drives to finish the request.
2792 * @rq: the request to finish
2793 * @error: %0 for success, < %0 for error
2796 * Completely finish @rq. Must be called with queue lock held.
2798 void __blk_end_request_all(struct request
*rq
, int error
)
2801 unsigned int bidi_bytes
= 0;
2803 if (unlikely(blk_bidi_rq(rq
)))
2804 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2806 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2809 EXPORT_SYMBOL(__blk_end_request_all
);
2812 * __blk_end_request_cur - Helper function to finish the current request chunk.
2813 * @rq: the request to finish the current chunk for
2814 * @error: %0 for success, < %0 for error
2817 * Complete the current consecutively mapped chunk from @rq. Must
2818 * be called with queue lock held.
2821 * %false - we are done with this request
2822 * %true - still buffers pending for this request
2824 bool __blk_end_request_cur(struct request
*rq
, int error
)
2826 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2828 EXPORT_SYMBOL(__blk_end_request_cur
);
2831 * __blk_end_request_err - Finish a request till the next failure boundary.
2832 * @rq: the request to finish till the next failure boundary for
2833 * @error: must be negative errno
2836 * Complete @rq till the next failure boundary. Must be called
2837 * with queue lock held.
2840 * %false - we are done with this request
2841 * %true - still buffers pending for this request
2843 bool __blk_end_request_err(struct request
*rq
, int error
)
2845 WARN_ON(error
>= 0);
2846 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2848 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2850 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2853 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2854 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2856 if (bio_has_data(bio
))
2857 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2859 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2860 rq
->bio
= rq
->biotail
= bio
;
2863 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2866 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2868 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2869 * @rq: the request to be flushed
2872 * Flush all pages in @rq.
2874 void rq_flush_dcache_pages(struct request
*rq
)
2876 struct req_iterator iter
;
2877 struct bio_vec bvec
;
2879 rq_for_each_segment(bvec
, rq
, iter
)
2880 flush_dcache_page(bvec
.bv_page
);
2882 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2886 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2887 * @q : the queue of the device being checked
2890 * Check if underlying low-level drivers of a device are busy.
2891 * If the drivers want to export their busy state, they must set own
2892 * exporting function using blk_queue_lld_busy() first.
2894 * Basically, this function is used only by request stacking drivers
2895 * to stop dispatching requests to underlying devices when underlying
2896 * devices are busy. This behavior helps more I/O merging on the queue
2897 * of the request stacking driver and prevents I/O throughput regression
2898 * on burst I/O load.
2901 * 0 - Not busy (The request stacking driver should dispatch request)
2902 * 1 - Busy (The request stacking driver should stop dispatching request)
2904 int blk_lld_busy(struct request_queue
*q
)
2907 return q
->lld_busy_fn(q
);
2911 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2913 void blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2915 dst
->cpu
= src
->cpu
;
2916 dst
->cmd_flags
|= (src
->cmd_flags
& REQ_CLONE_MASK
);
2917 dst
->cmd_flags
|= REQ_NOMERGE
| REQ_CLONE
;
2918 dst
->cmd_type
= src
->cmd_type
;
2919 dst
->__sector
= blk_rq_pos(src
);
2920 dst
->__data_len
= blk_rq_bytes(src
);
2921 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2922 dst
->ioprio
= src
->ioprio
;
2923 dst
->extra_len
= src
->extra_len
;
2924 dst
->bio
= src
->bio
;
2925 dst
->biotail
= src
->biotail
;
2926 dst
->cmd
= src
->cmd
;
2927 dst
->cmd_len
= src
->cmd_len
;
2928 dst
->sense
= src
->sense
;
2930 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2932 int kblockd_schedule_work(struct work_struct
*work
)
2934 return queue_work(kblockd_workqueue
, work
);
2936 EXPORT_SYMBOL(kblockd_schedule_work
);
2938 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
2939 unsigned long delay
)
2941 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2943 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2945 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
2946 unsigned long delay
)
2948 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
2950 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
2953 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2954 * @plug: The &struct blk_plug that needs to be initialized
2957 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2958 * pending I/O should the task end up blocking between blk_start_plug() and
2959 * blk_finish_plug(). This is important from a performance perspective, but
2960 * also ensures that we don't deadlock. For instance, if the task is blocking
2961 * for a memory allocation, memory reclaim could end up wanting to free a
2962 * page belonging to that request that is currently residing in our private
2963 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2964 * this kind of deadlock.
2966 void blk_start_plug(struct blk_plug
*plug
)
2968 struct task_struct
*tsk
= current
;
2971 * If this is a nested plug, don't actually assign it.
2976 INIT_LIST_HEAD(&plug
->list
);
2977 INIT_LIST_HEAD(&plug
->mq_list
);
2978 INIT_LIST_HEAD(&plug
->cb_list
);
2980 * Store ordering should not be needed here, since a potential
2981 * preempt will imply a full memory barrier
2985 EXPORT_SYMBOL(blk_start_plug
);
2987 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2989 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2990 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2992 return !(rqa
->q
< rqb
->q
||
2993 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
2997 * If 'from_schedule' is true, then postpone the dispatch of requests
2998 * until a safe kblockd context. We due this to avoid accidental big
2999 * additional stack usage in driver dispatch, in places where the originally
3000 * plugger did not intend it.
3002 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3004 __releases(q
->queue_lock
)
3006 trace_block_unplug(q
, depth
, !from_schedule
);
3009 blk_run_queue_async(q
);
3012 spin_unlock(q
->queue_lock
);
3015 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3017 LIST_HEAD(callbacks
);
3019 while (!list_empty(&plug
->cb_list
)) {
3020 list_splice_init(&plug
->cb_list
, &callbacks
);
3022 while (!list_empty(&callbacks
)) {
3023 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3026 list_del(&cb
->list
);
3027 cb
->callback(cb
, from_schedule
);
3032 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3035 struct blk_plug
*plug
= current
->plug
;
3036 struct blk_plug_cb
*cb
;
3041 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3042 if (cb
->callback
== unplug
&& cb
->data
== data
)
3045 /* Not currently on the callback list */
3046 BUG_ON(size
< sizeof(*cb
));
3047 cb
= kzalloc(size
, GFP_ATOMIC
);
3050 cb
->callback
= unplug
;
3051 list_add(&cb
->list
, &plug
->cb_list
);
3055 EXPORT_SYMBOL(blk_check_plugged
);
3057 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3059 struct request_queue
*q
;
3060 unsigned long flags
;
3065 flush_plug_callbacks(plug
, from_schedule
);
3067 if (!list_empty(&plug
->mq_list
))
3068 blk_mq_flush_plug_list(plug
, from_schedule
);
3070 if (list_empty(&plug
->list
))
3073 list_splice_init(&plug
->list
, &list
);
3075 list_sort(NULL
, &list
, plug_rq_cmp
);
3081 * Save and disable interrupts here, to avoid doing it for every
3082 * queue lock we have to take.
3084 local_irq_save(flags
);
3085 while (!list_empty(&list
)) {
3086 rq
= list_entry_rq(list
.next
);
3087 list_del_init(&rq
->queuelist
);
3091 * This drops the queue lock
3094 queue_unplugged(q
, depth
, from_schedule
);
3097 spin_lock(q
->queue_lock
);
3101 * Short-circuit if @q is dead
3103 if (unlikely(blk_queue_dying(q
))) {
3104 __blk_end_request_all(rq
, -ENODEV
);
3109 * rq is already accounted, so use raw insert
3111 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3112 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3114 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3120 * This drops the queue lock
3123 queue_unplugged(q
, depth
, from_schedule
);
3125 local_irq_restore(flags
);
3128 void blk_finish_plug(struct blk_plug
*plug
)
3130 if (plug
!= current
->plug
)
3132 blk_flush_plug_list(plug
, false);
3134 current
->plug
= NULL
;
3136 EXPORT_SYMBOL(blk_finish_plug
);
3140 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3141 * @q: the queue of the device
3142 * @dev: the device the queue belongs to
3145 * Initialize runtime-PM-related fields for @q and start auto suspend for
3146 * @dev. Drivers that want to take advantage of request-based runtime PM
3147 * should call this function after @dev has been initialized, and its
3148 * request queue @q has been allocated, and runtime PM for it can not happen
3149 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3150 * cases, driver should call this function before any I/O has taken place.
3152 * This function takes care of setting up using auto suspend for the device,
3153 * the autosuspend delay is set to -1 to make runtime suspend impossible
3154 * until an updated value is either set by user or by driver. Drivers do
3155 * not need to touch other autosuspend settings.
3157 * The block layer runtime PM is request based, so only works for drivers
3158 * that use request as their IO unit instead of those directly use bio's.
3160 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3163 q
->rpm_status
= RPM_ACTIVE
;
3164 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3165 pm_runtime_use_autosuspend(q
->dev
);
3167 EXPORT_SYMBOL(blk_pm_runtime_init
);
3170 * blk_pre_runtime_suspend - Pre runtime suspend check
3171 * @q: the queue of the device
3174 * This function will check if runtime suspend is allowed for the device
3175 * by examining if there are any requests pending in the queue. If there
3176 * are requests pending, the device can not be runtime suspended; otherwise,
3177 * the queue's status will be updated to SUSPENDING and the driver can
3178 * proceed to suspend the device.
3180 * For the not allowed case, we mark last busy for the device so that
3181 * runtime PM core will try to autosuspend it some time later.
3183 * This function should be called near the start of the device's
3184 * runtime_suspend callback.
3187 * 0 - OK to runtime suspend the device
3188 * -EBUSY - Device should not be runtime suspended
3190 int blk_pre_runtime_suspend(struct request_queue
*q
)
3194 spin_lock_irq(q
->queue_lock
);
3195 if (q
->nr_pending
) {
3197 pm_runtime_mark_last_busy(q
->dev
);
3199 q
->rpm_status
= RPM_SUSPENDING
;
3201 spin_unlock_irq(q
->queue_lock
);
3204 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3207 * blk_post_runtime_suspend - Post runtime suspend processing
3208 * @q: the queue of the device
3209 * @err: return value of the device's runtime_suspend function
3212 * Update the queue's runtime status according to the return value of the
3213 * device's runtime suspend function and mark last busy for the device so
3214 * that PM core will try to auto suspend the device at a later time.
3216 * This function should be called near the end of the device's
3217 * runtime_suspend callback.
3219 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3221 spin_lock_irq(q
->queue_lock
);
3223 q
->rpm_status
= RPM_SUSPENDED
;
3225 q
->rpm_status
= RPM_ACTIVE
;
3226 pm_runtime_mark_last_busy(q
->dev
);
3228 spin_unlock_irq(q
->queue_lock
);
3230 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3233 * blk_pre_runtime_resume - Pre runtime resume processing
3234 * @q: the queue of the device
3237 * Update the queue's runtime status to RESUMING in preparation for the
3238 * runtime resume of the device.
3240 * This function should be called near the start of the device's
3241 * runtime_resume callback.
3243 void blk_pre_runtime_resume(struct request_queue
*q
)
3245 spin_lock_irq(q
->queue_lock
);
3246 q
->rpm_status
= RPM_RESUMING
;
3247 spin_unlock_irq(q
->queue_lock
);
3249 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3252 * blk_post_runtime_resume - Post runtime resume processing
3253 * @q: the queue of the device
3254 * @err: return value of the device's runtime_resume function
3257 * Update the queue's runtime status according to the return value of the
3258 * device's runtime_resume function. If it is successfully resumed, process
3259 * the requests that are queued into the device's queue when it is resuming
3260 * and then mark last busy and initiate autosuspend for it.
3262 * This function should be called near the end of the device's
3263 * runtime_resume callback.
3265 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3267 spin_lock_irq(q
->queue_lock
);
3269 q
->rpm_status
= RPM_ACTIVE
;
3271 pm_runtime_mark_last_busy(q
->dev
);
3272 pm_request_autosuspend(q
->dev
);
3274 q
->rpm_status
= RPM_SUSPENDED
;
3276 spin_unlock_irq(q
->queue_lock
);
3278 EXPORT_SYMBOL(blk_post_runtime_resume
);
3281 int __init
blk_dev_init(void)
3283 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3284 sizeof(((struct request
*)0)->cmd_flags
));
3286 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3287 kblockd_workqueue
= alloc_workqueue("kblockd",
3288 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3289 if (!kblockd_workqueue
)
3290 panic("Failed to create kblockd\n");
3292 request_cachep
= kmem_cache_create("blkdev_requests",
3293 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3295 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3296 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);