2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.h>
42 #include "blk-mq-sched.h"
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
48 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
49 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
51 DEFINE_IDA(blk_queue_ida
);
54 * For the allocated request tables
56 struct kmem_cache
*request_cachep
;
59 * For queue allocation
61 struct kmem_cache
*blk_requestq_cachep
;
64 * Controlling structure to kblockd
66 static struct workqueue_struct
*kblockd_workqueue
;
68 static void blk_clear_congested(struct request_list
*rl
, int sync
)
70 #ifdef CONFIG_CGROUP_WRITEBACK
71 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
74 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
75 * flip its congestion state for events on other blkcgs.
77 if (rl
== &rl
->q
->root_rl
)
78 clear_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
82 static void blk_set_congested(struct request_list
*rl
, int sync
)
84 #ifdef CONFIG_CGROUP_WRITEBACK
85 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
87 /* see blk_clear_congested() */
88 if (rl
== &rl
->q
->root_rl
)
89 set_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
93 void blk_queue_congestion_threshold(struct request_queue
*q
)
97 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
98 if (nr
> q
->nr_requests
)
100 q
->nr_congestion_on
= nr
;
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
105 q
->nr_congestion_off
= nr
;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info. This function can only be called if @bdev is opened
114 * and the return value is never NULL.
116 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
118 struct request_queue
*q
= bdev_get_queue(bdev
);
120 return &q
->backing_dev_info
;
122 EXPORT_SYMBOL(blk_get_backing_dev_info
);
124 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
126 memset(rq
, 0, sizeof(*rq
));
128 INIT_LIST_HEAD(&rq
->queuelist
);
129 INIT_LIST_HEAD(&rq
->timeout_list
);
132 rq
->__sector
= (sector_t
) -1;
133 INIT_HLIST_NODE(&rq
->hash
);
134 RB_CLEAR_NODE(&rq
->rb_node
);
136 rq
->cmd_len
= BLK_MAX_CDB
;
138 rq
->internal_tag
= -1;
139 rq
->start_time
= jiffies
;
140 set_start_time_ns(rq
);
143 EXPORT_SYMBOL(blk_rq_init
);
145 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
146 unsigned int nbytes
, int error
)
149 bio
->bi_error
= error
;
151 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
152 bio_set_flag(bio
, BIO_QUIET
);
154 bio_advance(bio
, nbytes
);
156 /* don't actually finish bio if it's part of flush sequence */
157 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
161 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
165 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
166 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
167 (unsigned long long) rq
->cmd_flags
);
169 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
170 (unsigned long long)blk_rq_pos(rq
),
171 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
172 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
173 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
175 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
176 printk(KERN_INFO
" cdb: ");
177 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
178 printk("%02x ", rq
->cmd
[bit
]);
182 EXPORT_SYMBOL(blk_dump_rq_flags
);
184 static void blk_delay_work(struct work_struct
*work
)
186 struct request_queue
*q
;
188 q
= container_of(work
, struct request_queue
, delay_work
.work
);
189 spin_lock_irq(q
->queue_lock
);
191 spin_unlock_irq(q
->queue_lock
);
195 * blk_delay_queue - restart queueing after defined interval
196 * @q: The &struct request_queue in question
197 * @msecs: Delay in msecs
200 * Sometimes queueing needs to be postponed for a little while, to allow
201 * resources to come back. This function will make sure that queueing is
202 * restarted around the specified time. Queue lock must be held.
204 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
206 if (likely(!blk_queue_dead(q
)))
207 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
208 msecs_to_jiffies(msecs
));
210 EXPORT_SYMBOL(blk_delay_queue
);
213 * blk_start_queue_async - asynchronously restart a previously stopped queue
214 * @q: The &struct request_queue in question
217 * blk_start_queue_async() will clear the stop flag on the queue, and
218 * ensure that the request_fn for the queue is run from an async
221 void blk_start_queue_async(struct request_queue
*q
)
223 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
224 blk_run_queue_async(q
);
226 EXPORT_SYMBOL(blk_start_queue_async
);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue
*q
)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
244 EXPORT_SYMBOL(blk_start_queue
);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue
*q
)
262 cancel_delayed_work(&q
->delay_work
);
263 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
265 EXPORT_SYMBOL(blk_stop_queue
);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevator_exit()
282 * and blkcg_exit_queue() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue
*q
)
287 del_timer_sync(&q
->timeout
);
290 struct blk_mq_hw_ctx
*hctx
;
293 queue_for_each_hw_ctx(q
, hctx
, i
) {
294 cancel_work_sync(&hctx
->run_work
);
295 cancel_delayed_work_sync(&hctx
->delay_work
);
298 cancel_delayed_work_sync(&q
->delay_work
);
301 EXPORT_SYMBOL(blk_sync_queue
);
304 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
305 * @q: The queue to run
308 * Invoke request handling on a queue if there are any pending requests.
309 * May be used to restart request handling after a request has completed.
310 * This variant runs the queue whether or not the queue has been
311 * stopped. Must be called with the queue lock held and interrupts
312 * disabled. See also @blk_run_queue.
314 inline void __blk_run_queue_uncond(struct request_queue
*q
)
316 if (unlikely(blk_queue_dead(q
)))
320 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
321 * the queue lock internally. As a result multiple threads may be
322 * running such a request function concurrently. Keep track of the
323 * number of active request_fn invocations such that blk_drain_queue()
324 * can wait until all these request_fn calls have finished.
326 q
->request_fn_active
++;
328 q
->request_fn_active
--;
330 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
333 * __blk_run_queue - run a single device queue
334 * @q: The queue to run
337 * See @blk_run_queue. This variant must be called with the queue lock
338 * held and interrupts disabled.
340 void __blk_run_queue(struct request_queue
*q
)
342 if (unlikely(blk_queue_stopped(q
)))
345 __blk_run_queue_uncond(q
);
347 EXPORT_SYMBOL(__blk_run_queue
);
350 * blk_run_queue_async - run a single device queue in workqueue context
351 * @q: The queue to run
354 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
355 * of us. The caller must hold the queue lock.
357 void blk_run_queue_async(struct request_queue
*q
)
359 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
360 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
362 EXPORT_SYMBOL(blk_run_queue_async
);
365 * blk_run_queue - run a single device queue
366 * @q: The queue to run
369 * Invoke request handling on this queue, if it has pending work to do.
370 * May be used to restart queueing when a request has completed.
372 void blk_run_queue(struct request_queue
*q
)
376 spin_lock_irqsave(q
->queue_lock
, flags
);
378 spin_unlock_irqrestore(q
->queue_lock
, flags
);
380 EXPORT_SYMBOL(blk_run_queue
);
382 void blk_put_queue(struct request_queue
*q
)
384 kobject_put(&q
->kobj
);
386 EXPORT_SYMBOL(blk_put_queue
);
389 * __blk_drain_queue - drain requests from request_queue
391 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
393 * Drain requests from @q. If @drain_all is set, all requests are drained.
394 * If not, only ELVPRIV requests are drained. The caller is responsible
395 * for ensuring that no new requests which need to be drained are queued.
397 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
398 __releases(q
->queue_lock
)
399 __acquires(q
->queue_lock
)
403 lockdep_assert_held(q
->queue_lock
);
409 * The caller might be trying to drain @q before its
410 * elevator is initialized.
413 elv_drain_elevator(q
);
415 blkcg_drain_queue(q
);
418 * This function might be called on a queue which failed
419 * driver init after queue creation or is not yet fully
420 * active yet. Some drivers (e.g. fd and loop) get unhappy
421 * in such cases. Kick queue iff dispatch queue has
422 * something on it and @q has request_fn set.
424 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
427 drain
|= q
->nr_rqs_elvpriv
;
428 drain
|= q
->request_fn_active
;
431 * Unfortunately, requests are queued at and tracked from
432 * multiple places and there's no single counter which can
433 * be drained. Check all the queues and counters.
436 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
437 drain
|= !list_empty(&q
->queue_head
);
438 for (i
= 0; i
< 2; i
++) {
439 drain
|= q
->nr_rqs
[i
];
440 drain
|= q
->in_flight
[i
];
442 drain
|= !list_empty(&fq
->flush_queue
[i
]);
449 spin_unlock_irq(q
->queue_lock
);
453 spin_lock_irq(q
->queue_lock
);
457 * With queue marked dead, any woken up waiter will fail the
458 * allocation path, so the wakeup chaining is lost and we're
459 * left with hung waiters. We need to wake up those waiters.
462 struct request_list
*rl
;
464 blk_queue_for_each_rl(rl
, q
)
465 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
466 wake_up_all(&rl
->wait
[i
]);
471 * blk_queue_bypass_start - enter queue bypass mode
472 * @q: queue of interest
474 * In bypass mode, only the dispatch FIFO queue of @q is used. This
475 * function makes @q enter bypass mode and drains all requests which were
476 * throttled or issued before. On return, it's guaranteed that no request
477 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
478 * inside queue or RCU read lock.
480 void blk_queue_bypass_start(struct request_queue
*q
)
482 spin_lock_irq(q
->queue_lock
);
484 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
485 spin_unlock_irq(q
->queue_lock
);
488 * Queues start drained. Skip actual draining till init is
489 * complete. This avoids lenghty delays during queue init which
490 * can happen many times during boot.
492 if (blk_queue_init_done(q
)) {
493 spin_lock_irq(q
->queue_lock
);
494 __blk_drain_queue(q
, false);
495 spin_unlock_irq(q
->queue_lock
);
497 /* ensure blk_queue_bypass() is %true inside RCU read lock */
501 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
504 * blk_queue_bypass_end - leave queue bypass mode
505 * @q: queue of interest
507 * Leave bypass mode and restore the normal queueing behavior.
509 void blk_queue_bypass_end(struct request_queue
*q
)
511 spin_lock_irq(q
->queue_lock
);
512 if (!--q
->bypass_depth
)
513 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
514 WARN_ON_ONCE(q
->bypass_depth
< 0);
515 spin_unlock_irq(q
->queue_lock
);
517 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
519 void blk_set_queue_dying(struct request_queue
*q
)
521 spin_lock_irq(q
->queue_lock
);
522 queue_flag_set(QUEUE_FLAG_DYING
, q
);
523 spin_unlock_irq(q
->queue_lock
);
526 blk_mq_wake_waiters(q
);
528 struct request_list
*rl
;
530 blk_queue_for_each_rl(rl
, q
) {
532 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
533 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
538 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
541 * blk_cleanup_queue - shutdown a request queue
542 * @q: request queue to shutdown
544 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
545 * put it. All future requests will be failed immediately with -ENODEV.
547 void blk_cleanup_queue(struct request_queue
*q
)
549 spinlock_t
*lock
= q
->queue_lock
;
551 /* mark @q DYING, no new request or merges will be allowed afterwards */
552 mutex_lock(&q
->sysfs_lock
);
553 blk_set_queue_dying(q
);
557 * A dying queue is permanently in bypass mode till released. Note
558 * that, unlike blk_queue_bypass_start(), we aren't performing
559 * synchronize_rcu() after entering bypass mode to avoid the delay
560 * as some drivers create and destroy a lot of queues while
561 * probing. This is still safe because blk_release_queue() will be
562 * called only after the queue refcnt drops to zero and nothing,
563 * RCU or not, would be traversing the queue by then.
566 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
568 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
569 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
570 queue_flag_set(QUEUE_FLAG_DYING
, q
);
571 spin_unlock_irq(lock
);
572 mutex_unlock(&q
->sysfs_lock
);
575 * Drain all requests queued before DYING marking. Set DEAD flag to
576 * prevent that q->request_fn() gets invoked after draining finished.
581 __blk_drain_queue(q
, true);
582 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
583 spin_unlock_irq(lock
);
585 /* for synchronous bio-based driver finish in-flight integrity i/o */
586 blk_flush_integrity();
588 /* @q won't process any more request, flush async actions */
589 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
593 blk_mq_free_queue(q
);
594 percpu_ref_exit(&q
->q_usage_counter
);
597 if (q
->queue_lock
!= &q
->__queue_lock
)
598 q
->queue_lock
= &q
->__queue_lock
;
599 spin_unlock_irq(lock
);
601 bdi_unregister(&q
->backing_dev_info
);
603 /* @q is and will stay empty, shutdown and put */
606 EXPORT_SYMBOL(blk_cleanup_queue
);
608 /* Allocate memory local to the request queue */
609 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
611 int nid
= (int)(long)data
;
612 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
615 static void free_request_struct(void *element
, void *unused
)
617 kmem_cache_free(request_cachep
, element
);
620 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
623 if (unlikely(rl
->rq_pool
))
627 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
628 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
629 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
630 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
632 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
634 (void *)(long)q
->node
, gfp_mask
,
642 void blk_exit_rl(struct request_list
*rl
)
645 mempool_destroy(rl
->rq_pool
);
648 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
650 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
652 EXPORT_SYMBOL(blk_alloc_queue
);
654 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
659 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
665 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
666 !atomic_read(&q
->mq_freeze_depth
) ||
668 if (blk_queue_dying(q
))
675 void blk_queue_exit(struct request_queue
*q
)
677 percpu_ref_put(&q
->q_usage_counter
);
680 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
682 struct request_queue
*q
=
683 container_of(ref
, struct request_queue
, q_usage_counter
);
685 wake_up_all(&q
->mq_freeze_wq
);
688 static void blk_rq_timed_out_timer(unsigned long data
)
690 struct request_queue
*q
= (struct request_queue
*)data
;
692 kblockd_schedule_work(&q
->timeout_work
);
695 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
697 struct request_queue
*q
;
700 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
701 gfp_mask
| __GFP_ZERO
, node_id
);
705 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
709 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
713 q
->backing_dev_info
.ra_pages
=
714 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
715 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
716 q
->backing_dev_info
.name
= "block";
719 err
= bdi_init(&q
->backing_dev_info
);
723 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
724 laptop_mode_timer_fn
, (unsigned long) q
);
725 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
726 INIT_LIST_HEAD(&q
->queue_head
);
727 INIT_LIST_HEAD(&q
->timeout_list
);
728 INIT_LIST_HEAD(&q
->icq_list
);
729 #ifdef CONFIG_BLK_CGROUP
730 INIT_LIST_HEAD(&q
->blkg_list
);
732 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
734 kobject_init(&q
->kobj
, &blk_queue_ktype
);
736 mutex_init(&q
->sysfs_lock
);
737 spin_lock_init(&q
->__queue_lock
);
740 * By default initialize queue_lock to internal lock and driver can
741 * override it later if need be.
743 q
->queue_lock
= &q
->__queue_lock
;
746 * A queue starts its life with bypass turned on to avoid
747 * unnecessary bypass on/off overhead and nasty surprises during
748 * init. The initial bypass will be finished when the queue is
749 * registered by blk_register_queue().
752 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
754 init_waitqueue_head(&q
->mq_freeze_wq
);
757 * Init percpu_ref in atomic mode so that it's faster to shutdown.
758 * See blk_register_queue() for details.
760 if (percpu_ref_init(&q
->q_usage_counter
,
761 blk_queue_usage_counter_release
,
762 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
765 if (blkcg_init_queue(q
))
771 percpu_ref_exit(&q
->q_usage_counter
);
773 bdi_destroy(&q
->backing_dev_info
);
775 bioset_free(q
->bio_split
);
777 ida_simple_remove(&blk_queue_ida
, q
->id
);
779 kmem_cache_free(blk_requestq_cachep
, q
);
782 EXPORT_SYMBOL(blk_alloc_queue_node
);
785 * blk_init_queue - prepare a request queue for use with a block device
786 * @rfn: The function to be called to process requests that have been
787 * placed on the queue.
788 * @lock: Request queue spin lock
791 * If a block device wishes to use the standard request handling procedures,
792 * which sorts requests and coalesces adjacent requests, then it must
793 * call blk_init_queue(). The function @rfn will be called when there
794 * are requests on the queue that need to be processed. If the device
795 * supports plugging, then @rfn may not be called immediately when requests
796 * are available on the queue, but may be called at some time later instead.
797 * Plugged queues are generally unplugged when a buffer belonging to one
798 * of the requests on the queue is needed, or due to memory pressure.
800 * @rfn is not required, or even expected, to remove all requests off the
801 * queue, but only as many as it can handle at a time. If it does leave
802 * requests on the queue, it is responsible for arranging that the requests
803 * get dealt with eventually.
805 * The queue spin lock must be held while manipulating the requests on the
806 * request queue; this lock will be taken also from interrupt context, so irq
807 * disabling is needed for it.
809 * Function returns a pointer to the initialized request queue, or %NULL if
813 * blk_init_queue() must be paired with a blk_cleanup_queue() call
814 * when the block device is deactivated (such as at module unload).
817 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
819 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
821 EXPORT_SYMBOL(blk_init_queue
);
823 struct request_queue
*
824 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
826 struct request_queue
*q
;
828 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
834 q
->queue_lock
= lock
;
835 if (blk_init_allocated_queue(q
) < 0) {
836 blk_cleanup_queue(q
);
842 EXPORT_SYMBOL(blk_init_queue_node
);
844 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
847 int blk_init_allocated_queue(struct request_queue
*q
)
849 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
853 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
856 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
857 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
860 * This also sets hw/phys segments, boundary and size
862 blk_queue_make_request(q
, blk_queue_bio
);
864 q
->sg_reserved_size
= INT_MAX
;
866 /* Protect q->elevator from elevator_change */
867 mutex_lock(&q
->sysfs_lock
);
870 if (elevator_init(q
, NULL
)) {
871 mutex_unlock(&q
->sysfs_lock
);
875 mutex_unlock(&q
->sysfs_lock
);
879 blk_free_flush_queue(q
->fq
);
883 EXPORT_SYMBOL(blk_init_allocated_queue
);
885 bool blk_get_queue(struct request_queue
*q
)
887 if (likely(!blk_queue_dying(q
))) {
894 EXPORT_SYMBOL(blk_get_queue
);
896 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
898 if (rq
->rq_flags
& RQF_ELVPRIV
) {
899 elv_put_request(rl
->q
, rq
);
901 put_io_context(rq
->elv
.icq
->ioc
);
904 mempool_free(rq
, rl
->rq_pool
);
908 * ioc_batching returns true if the ioc is a valid batching request and
909 * should be given priority access to a request.
911 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
917 * Make sure the process is able to allocate at least 1 request
918 * even if the batch times out, otherwise we could theoretically
921 return ioc
->nr_batch_requests
== q
->nr_batching
||
922 (ioc
->nr_batch_requests
> 0
923 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
927 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
928 * will cause the process to be a "batcher" on all queues in the system. This
929 * is the behaviour we want though - once it gets a wakeup it should be given
932 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
934 if (!ioc
|| ioc_batching(q
, ioc
))
937 ioc
->nr_batch_requests
= q
->nr_batching
;
938 ioc
->last_waited
= jiffies
;
941 static void __freed_request(struct request_list
*rl
, int sync
)
943 struct request_queue
*q
= rl
->q
;
945 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
946 blk_clear_congested(rl
, sync
);
948 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
949 if (waitqueue_active(&rl
->wait
[sync
]))
950 wake_up(&rl
->wait
[sync
]);
952 blk_clear_rl_full(rl
, sync
);
957 * A request has just been released. Account for it, update the full and
958 * congestion status, wake up any waiters. Called under q->queue_lock.
960 static void freed_request(struct request_list
*rl
, bool sync
,
961 req_flags_t rq_flags
)
963 struct request_queue
*q
= rl
->q
;
967 if (rq_flags
& RQF_ELVPRIV
)
970 __freed_request(rl
, sync
);
972 if (unlikely(rl
->starved
[sync
^ 1]))
973 __freed_request(rl
, sync
^ 1);
976 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
978 struct request_list
*rl
;
979 int on_thresh
, off_thresh
;
981 spin_lock_irq(q
->queue_lock
);
983 blk_queue_congestion_threshold(q
);
984 on_thresh
= queue_congestion_on_threshold(q
);
985 off_thresh
= queue_congestion_off_threshold(q
);
987 blk_queue_for_each_rl(rl
, q
) {
988 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
989 blk_set_congested(rl
, BLK_RW_SYNC
);
990 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
991 blk_clear_congested(rl
, BLK_RW_SYNC
);
993 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
994 blk_set_congested(rl
, BLK_RW_ASYNC
);
995 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
996 blk_clear_congested(rl
, BLK_RW_ASYNC
);
998 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
999 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1001 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1002 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1005 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1006 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1008 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1009 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1013 spin_unlock_irq(q
->queue_lock
);
1018 * __get_request - get a free request
1019 * @rl: request list to allocate from
1020 * @op: operation and flags
1021 * @bio: bio to allocate request for (can be %NULL)
1022 * @gfp_mask: allocation mask
1024 * Get a free request from @q. This function may fail under memory
1025 * pressure or if @q is dead.
1027 * Must be called with @q->queue_lock held and,
1028 * Returns ERR_PTR on failure, with @q->queue_lock held.
1029 * Returns request pointer on success, with @q->queue_lock *not held*.
1031 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1032 struct bio
*bio
, gfp_t gfp_mask
)
1034 struct request_queue
*q
= rl
->q
;
1036 struct elevator_type
*et
= q
->elevator
->type
;
1037 struct io_context
*ioc
= rq_ioc(bio
);
1038 struct io_cq
*icq
= NULL
;
1039 const bool is_sync
= op_is_sync(op
);
1041 req_flags_t rq_flags
= RQF_ALLOCED
;
1043 if (unlikely(blk_queue_dying(q
)))
1044 return ERR_PTR(-ENODEV
);
1046 may_queue
= elv_may_queue(q
, op
);
1047 if (may_queue
== ELV_MQUEUE_NO
)
1050 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1051 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1053 * The queue will fill after this allocation, so set
1054 * it as full, and mark this process as "batching".
1055 * This process will be allowed to complete a batch of
1056 * requests, others will be blocked.
1058 if (!blk_rl_full(rl
, is_sync
)) {
1059 ioc_set_batching(q
, ioc
);
1060 blk_set_rl_full(rl
, is_sync
);
1062 if (may_queue
!= ELV_MQUEUE_MUST
1063 && !ioc_batching(q
, ioc
)) {
1065 * The queue is full and the allocating
1066 * process is not a "batcher", and not
1067 * exempted by the IO scheduler
1069 return ERR_PTR(-ENOMEM
);
1073 blk_set_congested(rl
, is_sync
);
1077 * Only allow batching queuers to allocate up to 50% over the defined
1078 * limit of requests, otherwise we could have thousands of requests
1079 * allocated with any setting of ->nr_requests
1081 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1082 return ERR_PTR(-ENOMEM
);
1084 q
->nr_rqs
[is_sync
]++;
1085 rl
->count
[is_sync
]++;
1086 rl
->starved
[is_sync
] = 0;
1089 * Decide whether the new request will be managed by elevator. If
1090 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1091 * prevent the current elevator from being destroyed until the new
1092 * request is freed. This guarantees icq's won't be destroyed and
1093 * makes creating new ones safe.
1095 * Flush requests do not use the elevator so skip initialization.
1096 * This allows a request to share the flush and elevator data.
1098 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1099 * it will be created after releasing queue_lock.
1101 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1102 rq_flags
|= RQF_ELVPRIV
;
1103 q
->nr_rqs_elvpriv
++;
1104 if (et
->icq_cache
&& ioc
)
1105 icq
= ioc_lookup_icq(ioc
, q
);
1108 if (blk_queue_io_stat(q
))
1109 rq_flags
|= RQF_IO_STAT
;
1110 spin_unlock_irq(q
->queue_lock
);
1112 /* allocate and init request */
1113 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1118 blk_rq_set_rl(rq
, rl
);
1119 blk_rq_set_prio(rq
, ioc
);
1121 rq
->rq_flags
= rq_flags
;
1124 if (rq_flags
& RQF_ELVPRIV
) {
1125 if (unlikely(et
->icq_cache
&& !icq
)) {
1127 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1133 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1136 /* @rq->elv.icq holds io_context until @rq is freed */
1138 get_io_context(icq
->ioc
);
1142 * ioc may be NULL here, and ioc_batching will be false. That's
1143 * OK, if the queue is under the request limit then requests need
1144 * not count toward the nr_batch_requests limit. There will always
1145 * be some limit enforced by BLK_BATCH_TIME.
1147 if (ioc_batching(q
, ioc
))
1148 ioc
->nr_batch_requests
--;
1150 trace_block_getrq(q
, bio
, op
);
1155 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1156 * and may fail indefinitely under memory pressure and thus
1157 * shouldn't stall IO. Treat this request as !elvpriv. This will
1158 * disturb iosched and blkcg but weird is bettern than dead.
1160 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1161 __func__
, dev_name(q
->backing_dev_info
.dev
));
1163 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1166 spin_lock_irq(q
->queue_lock
);
1167 q
->nr_rqs_elvpriv
--;
1168 spin_unlock_irq(q
->queue_lock
);
1173 * Allocation failed presumably due to memory. Undo anything we
1174 * might have messed up.
1176 * Allocating task should really be put onto the front of the wait
1177 * queue, but this is pretty rare.
1179 spin_lock_irq(q
->queue_lock
);
1180 freed_request(rl
, is_sync
, rq_flags
);
1183 * in the very unlikely event that allocation failed and no
1184 * requests for this direction was pending, mark us starved so that
1185 * freeing of a request in the other direction will notice
1186 * us. another possible fix would be to split the rq mempool into
1190 if (unlikely(rl
->count
[is_sync
] == 0))
1191 rl
->starved
[is_sync
] = 1;
1192 return ERR_PTR(-ENOMEM
);
1196 * get_request - get a free request
1197 * @q: request_queue to allocate request from
1198 * @op: operation and flags
1199 * @bio: bio to allocate request for (can be %NULL)
1200 * @gfp_mask: allocation mask
1202 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1203 * this function keeps retrying under memory pressure and fails iff @q is dead.
1205 * Must be called with @q->queue_lock held and,
1206 * Returns ERR_PTR on failure, with @q->queue_lock held.
1207 * Returns request pointer on success, with @q->queue_lock *not held*.
1209 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1210 struct bio
*bio
, gfp_t gfp_mask
)
1212 const bool is_sync
= op_is_sync(op
);
1214 struct request_list
*rl
;
1217 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1219 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1223 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1228 /* wait on @rl and retry */
1229 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1230 TASK_UNINTERRUPTIBLE
);
1232 trace_block_sleeprq(q
, bio
, op
);
1234 spin_unlock_irq(q
->queue_lock
);
1238 * After sleeping, we become a "batching" process and will be able
1239 * to allocate at least one request, and up to a big batch of them
1240 * for a small period time. See ioc_batching, ioc_set_batching
1242 ioc_set_batching(q
, current
->io_context
);
1244 spin_lock_irq(q
->queue_lock
);
1245 finish_wait(&rl
->wait
[is_sync
], &wait
);
1250 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1255 /* create ioc upfront */
1256 create_io_context(gfp_mask
, q
->node
);
1258 spin_lock_irq(q
->queue_lock
);
1259 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1261 spin_unlock_irq(q
->queue_lock
);
1265 /* q->queue_lock is unlocked at this point */
1267 rq
->__sector
= (sector_t
) -1;
1268 rq
->bio
= rq
->biotail
= NULL
;
1272 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1275 return blk_mq_alloc_request(q
, rw
,
1276 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1277 0 : BLK_MQ_REQ_NOWAIT
);
1279 return blk_old_get_request(q
, rw
, gfp_mask
);
1281 EXPORT_SYMBOL(blk_get_request
);
1284 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1285 * @rq: request to be initialized
1288 void blk_rq_set_block_pc(struct request
*rq
)
1290 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1291 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1293 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1296 * blk_requeue_request - put a request back on queue
1297 * @q: request queue where request should be inserted
1298 * @rq: request to be inserted
1301 * Drivers often keep queueing requests until the hardware cannot accept
1302 * more, when that condition happens we need to put the request back
1303 * on the queue. Must be called with queue lock held.
1305 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1307 blk_delete_timer(rq
);
1308 blk_clear_rq_complete(rq
);
1309 trace_block_rq_requeue(q
, rq
);
1310 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1312 if (rq
->rq_flags
& RQF_QUEUED
)
1313 blk_queue_end_tag(q
, rq
);
1315 BUG_ON(blk_queued_rq(rq
));
1317 elv_requeue_request(q
, rq
);
1319 EXPORT_SYMBOL(blk_requeue_request
);
1321 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1324 blk_account_io_start(rq
, true);
1325 __elv_add_request(q
, rq
, where
);
1328 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1333 if (now
== part
->stamp
)
1336 inflight
= part_in_flight(part
);
1338 __part_stat_add(cpu
, part
, time_in_queue
,
1339 inflight
* (now
- part
->stamp
));
1340 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1346 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1347 * @cpu: cpu number for stats access
1348 * @part: target partition
1350 * The average IO queue length and utilisation statistics are maintained
1351 * by observing the current state of the queue length and the amount of
1352 * time it has been in this state for.
1354 * Normally, that accounting is done on IO completion, but that can result
1355 * in more than a second's worth of IO being accounted for within any one
1356 * second, leading to >100% utilisation. To deal with that, we call this
1357 * function to do a round-off before returning the results when reading
1358 * /proc/diskstats. This accounts immediately for all queue usage up to
1359 * the current jiffies and restarts the counters again.
1361 void part_round_stats(int cpu
, struct hd_struct
*part
)
1363 unsigned long now
= jiffies
;
1366 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1367 part_round_stats_single(cpu
, part
, now
);
1369 EXPORT_SYMBOL_GPL(part_round_stats
);
1372 static void blk_pm_put_request(struct request
*rq
)
1374 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1375 pm_runtime_mark_last_busy(rq
->q
->dev
);
1378 static inline void blk_pm_put_request(struct request
*rq
) {}
1382 * queue lock must be held
1384 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1386 req_flags_t rq_flags
= req
->rq_flags
;
1392 blk_mq_free_request(req
);
1396 blk_pm_put_request(req
);
1398 elv_completed_request(q
, req
);
1400 /* this is a bio leak */
1401 WARN_ON(req
->bio
!= NULL
);
1403 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1406 * Request may not have originated from ll_rw_blk. if not,
1407 * it didn't come out of our reserved rq pools
1409 if (rq_flags
& RQF_ALLOCED
) {
1410 struct request_list
*rl
= blk_rq_rl(req
);
1411 bool sync
= op_is_sync(req
->cmd_flags
);
1413 BUG_ON(!list_empty(&req
->queuelist
));
1414 BUG_ON(ELV_ON_HASH(req
));
1416 blk_free_request(rl
, req
);
1417 freed_request(rl
, sync
, rq_flags
);
1421 EXPORT_SYMBOL_GPL(__blk_put_request
);
1423 void blk_put_request(struct request
*req
)
1425 struct request_queue
*q
= req
->q
;
1428 blk_mq_free_request(req
);
1430 unsigned long flags
;
1432 spin_lock_irqsave(q
->queue_lock
, flags
);
1433 __blk_put_request(q
, req
);
1434 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1437 EXPORT_SYMBOL(blk_put_request
);
1439 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1442 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1444 if (!ll_back_merge_fn(q
, req
, bio
))
1447 trace_block_bio_backmerge(q
, req
, bio
);
1449 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1450 blk_rq_set_mixed_merge(req
);
1452 req
->biotail
->bi_next
= bio
;
1454 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1455 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1457 blk_account_io_start(req
, false);
1461 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1464 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1466 if (!ll_front_merge_fn(q
, req
, bio
))
1469 trace_block_bio_frontmerge(q
, req
, bio
);
1471 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1472 blk_rq_set_mixed_merge(req
);
1474 bio
->bi_next
= req
->bio
;
1477 req
->__sector
= bio
->bi_iter
.bi_sector
;
1478 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1479 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1481 blk_account_io_start(req
, false);
1486 * blk_attempt_plug_merge - try to merge with %current's plugged list
1487 * @q: request_queue new bio is being queued at
1488 * @bio: new bio being queued
1489 * @request_count: out parameter for number of traversed plugged requests
1490 * @same_queue_rq: pointer to &struct request that gets filled in when
1491 * another request associated with @q is found on the plug list
1492 * (optional, may be %NULL)
1494 * Determine whether @bio being queued on @q can be merged with a request
1495 * on %current's plugged list. Returns %true if merge was successful,
1498 * Plugging coalesces IOs from the same issuer for the same purpose without
1499 * going through @q->queue_lock. As such it's more of an issuing mechanism
1500 * than scheduling, and the request, while may have elvpriv data, is not
1501 * added on the elevator at this point. In addition, we don't have
1502 * reliable access to the elevator outside queue lock. Only check basic
1503 * merging parameters without querying the elevator.
1505 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1507 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1508 unsigned int *request_count
,
1509 struct request
**same_queue_rq
)
1511 struct blk_plug
*plug
;
1514 struct list_head
*plug_list
;
1516 plug
= current
->plug
;
1522 plug_list
= &plug
->mq_list
;
1524 plug_list
= &plug
->list
;
1526 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1532 * Only blk-mq multiple hardware queues case checks the
1533 * rq in the same queue, there should be only one such
1537 *same_queue_rq
= rq
;
1540 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1543 el_ret
= blk_try_merge(rq
, bio
);
1544 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1545 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1548 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1549 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1558 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1560 struct blk_plug
*plug
;
1562 struct list_head
*plug_list
;
1563 unsigned int ret
= 0;
1565 plug
= current
->plug
;
1570 plug_list
= &plug
->mq_list
;
1572 plug_list
= &plug
->list
;
1574 list_for_each_entry(rq
, plug_list
, queuelist
) {
1582 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1584 req
->cmd_type
= REQ_TYPE_FS
;
1585 if (bio
->bi_opf
& REQ_RAHEAD
)
1586 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1589 req
->__sector
= bio
->bi_iter
.bi_sector
;
1590 if (ioprio_valid(bio_prio(bio
)))
1591 req
->ioprio
= bio_prio(bio
);
1592 blk_rq_bio_prep(req
->q
, req
, bio
);
1595 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1597 struct blk_plug
*plug
;
1598 int el_ret
, where
= ELEVATOR_INSERT_SORT
;
1599 struct request
*req
;
1600 unsigned int request_count
= 0;
1601 unsigned int wb_acct
;
1604 * low level driver can indicate that it wants pages above a
1605 * certain limit bounced to low memory (ie for highmem, or even
1606 * ISA dma in theory)
1608 blk_queue_bounce(q
, &bio
);
1610 blk_queue_split(q
, &bio
, q
->bio_split
);
1612 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1613 bio
->bi_error
= -EIO
;
1615 return BLK_QC_T_NONE
;
1618 if (op_is_flush(bio
->bi_opf
)) {
1619 spin_lock_irq(q
->queue_lock
);
1620 where
= ELEVATOR_INSERT_FLUSH
;
1625 * Check if we can merge with the plugged list before grabbing
1628 if (!blk_queue_nomerges(q
)) {
1629 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1630 return BLK_QC_T_NONE
;
1632 request_count
= blk_plug_queued_count(q
);
1634 spin_lock_irq(q
->queue_lock
);
1636 el_ret
= elv_merge(q
, &req
, bio
);
1637 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1638 if (bio_attempt_back_merge(q
, req
, bio
)) {
1639 elv_bio_merged(q
, req
, bio
);
1640 if (!attempt_back_merge(q
, req
))
1641 elv_merged_request(q
, req
, el_ret
);
1644 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1645 if (bio_attempt_front_merge(q
, req
, bio
)) {
1646 elv_bio_merged(q
, req
, bio
);
1647 if (!attempt_front_merge(q
, req
))
1648 elv_merged_request(q
, req
, el_ret
);
1654 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1657 * Grab a free request. This is might sleep but can not fail.
1658 * Returns with the queue unlocked.
1660 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1662 __wbt_done(q
->rq_wb
, wb_acct
);
1663 bio
->bi_error
= PTR_ERR(req
);
1668 wbt_track(&req
->issue_stat
, wb_acct
);
1671 * After dropping the lock and possibly sleeping here, our request
1672 * may now be mergeable after it had proven unmergeable (above).
1673 * We don't worry about that case for efficiency. It won't happen
1674 * often, and the elevators are able to handle it.
1676 init_request_from_bio(req
, bio
);
1678 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1679 req
->cpu
= raw_smp_processor_id();
1681 plug
= current
->plug
;
1684 * If this is the first request added after a plug, fire
1687 * @request_count may become stale because of schedule
1688 * out, so check plug list again.
1690 if (!request_count
|| list_empty(&plug
->list
))
1691 trace_block_plug(q
);
1693 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1694 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1695 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1696 blk_flush_plug_list(plug
, false);
1697 trace_block_plug(q
);
1700 list_add_tail(&req
->queuelist
, &plug
->list
);
1701 blk_account_io_start(req
, true);
1703 spin_lock_irq(q
->queue_lock
);
1704 add_acct_request(q
, req
, where
);
1707 spin_unlock_irq(q
->queue_lock
);
1710 return BLK_QC_T_NONE
;
1714 * If bio->bi_dev is a partition, remap the location
1716 static inline void blk_partition_remap(struct bio
*bio
)
1718 struct block_device
*bdev
= bio
->bi_bdev
;
1721 * Zone reset does not include bi_size so bio_sectors() is always 0.
1722 * Include a test for the reset op code and perform the remap if needed.
1724 if (bdev
!= bdev
->bd_contains
&&
1725 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1726 struct hd_struct
*p
= bdev
->bd_part
;
1728 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1729 bio
->bi_bdev
= bdev
->bd_contains
;
1731 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1733 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1737 static void handle_bad_sector(struct bio
*bio
)
1739 char b
[BDEVNAME_SIZE
];
1741 printk(KERN_INFO
"attempt to access beyond end of device\n");
1742 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1743 bdevname(bio
->bi_bdev
, b
),
1745 (unsigned long long)bio_end_sector(bio
),
1746 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1749 #ifdef CONFIG_FAIL_MAKE_REQUEST
1751 static DECLARE_FAULT_ATTR(fail_make_request
);
1753 static int __init
setup_fail_make_request(char *str
)
1755 return setup_fault_attr(&fail_make_request
, str
);
1757 __setup("fail_make_request=", setup_fail_make_request
);
1759 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1761 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1764 static int __init
fail_make_request_debugfs(void)
1766 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1767 NULL
, &fail_make_request
);
1769 return PTR_ERR_OR_ZERO(dir
);
1772 late_initcall(fail_make_request_debugfs
);
1774 #else /* CONFIG_FAIL_MAKE_REQUEST */
1776 static inline bool should_fail_request(struct hd_struct
*part
,
1782 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1785 * Check whether this bio extends beyond the end of the device.
1787 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1794 /* Test device or partition size, when known. */
1795 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1797 sector_t sector
= bio
->bi_iter
.bi_sector
;
1799 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1801 * This may well happen - the kernel calls bread()
1802 * without checking the size of the device, e.g., when
1803 * mounting a device.
1805 handle_bad_sector(bio
);
1813 static noinline_for_stack
bool
1814 generic_make_request_checks(struct bio
*bio
)
1816 struct request_queue
*q
;
1817 int nr_sectors
= bio_sectors(bio
);
1819 char b
[BDEVNAME_SIZE
];
1820 struct hd_struct
*part
;
1824 if (bio_check_eod(bio
, nr_sectors
))
1827 q
= bdev_get_queue(bio
->bi_bdev
);
1830 "generic_make_request: Trying to access "
1831 "nonexistent block-device %s (%Lu)\n",
1832 bdevname(bio
->bi_bdev
, b
),
1833 (long long) bio
->bi_iter
.bi_sector
);
1837 part
= bio
->bi_bdev
->bd_part
;
1838 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1839 should_fail_request(&part_to_disk(part
)->part0
,
1840 bio
->bi_iter
.bi_size
))
1844 * If this device has partitions, remap block n
1845 * of partition p to block n+start(p) of the disk.
1847 blk_partition_remap(bio
);
1849 if (bio_check_eod(bio
, nr_sectors
))
1853 * Filter flush bio's early so that make_request based
1854 * drivers without flush support don't have to worry
1857 if (op_is_flush(bio
->bi_opf
) &&
1858 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1859 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1866 switch (bio_op(bio
)) {
1867 case REQ_OP_DISCARD
:
1868 if (!blk_queue_discard(q
))
1871 case REQ_OP_SECURE_ERASE
:
1872 if (!blk_queue_secure_erase(q
))
1875 case REQ_OP_WRITE_SAME
:
1876 if (!bdev_write_same(bio
->bi_bdev
))
1879 case REQ_OP_ZONE_REPORT
:
1880 case REQ_OP_ZONE_RESET
:
1881 if (!bdev_is_zoned(bio
->bi_bdev
))
1884 case REQ_OP_WRITE_ZEROES
:
1885 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1893 * Various block parts want %current->io_context and lazy ioc
1894 * allocation ends up trading a lot of pain for a small amount of
1895 * memory. Just allocate it upfront. This may fail and block
1896 * layer knows how to live with it.
1898 create_io_context(GFP_ATOMIC
, q
->node
);
1900 if (!blkcg_bio_issue_check(q
, bio
))
1903 trace_block_bio_queue(q
, bio
);
1909 bio
->bi_error
= err
;
1915 * generic_make_request - hand a buffer to its device driver for I/O
1916 * @bio: The bio describing the location in memory and on the device.
1918 * generic_make_request() is used to make I/O requests of block
1919 * devices. It is passed a &struct bio, which describes the I/O that needs
1922 * generic_make_request() does not return any status. The
1923 * success/failure status of the request, along with notification of
1924 * completion, is delivered asynchronously through the bio->bi_end_io
1925 * function described (one day) else where.
1927 * The caller of generic_make_request must make sure that bi_io_vec
1928 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1929 * set to describe the device address, and the
1930 * bi_end_io and optionally bi_private are set to describe how
1931 * completion notification should be signaled.
1933 * generic_make_request and the drivers it calls may use bi_next if this
1934 * bio happens to be merged with someone else, and may resubmit the bio to
1935 * a lower device by calling into generic_make_request recursively, which
1936 * means the bio should NOT be touched after the call to ->make_request_fn.
1938 blk_qc_t
generic_make_request(struct bio
*bio
)
1940 struct bio_list bio_list_on_stack
;
1941 blk_qc_t ret
= BLK_QC_T_NONE
;
1943 if (!generic_make_request_checks(bio
))
1947 * We only want one ->make_request_fn to be active at a time, else
1948 * stack usage with stacked devices could be a problem. So use
1949 * current->bio_list to keep a list of requests submited by a
1950 * make_request_fn function. current->bio_list is also used as a
1951 * flag to say if generic_make_request is currently active in this
1952 * task or not. If it is NULL, then no make_request is active. If
1953 * it is non-NULL, then a make_request is active, and new requests
1954 * should be added at the tail
1956 if (current
->bio_list
) {
1957 bio_list_add(current
->bio_list
, bio
);
1961 /* following loop may be a bit non-obvious, and so deserves some
1963 * Before entering the loop, bio->bi_next is NULL (as all callers
1964 * ensure that) so we have a list with a single bio.
1965 * We pretend that we have just taken it off a longer list, so
1966 * we assign bio_list to a pointer to the bio_list_on_stack,
1967 * thus initialising the bio_list of new bios to be
1968 * added. ->make_request() may indeed add some more bios
1969 * through a recursive call to generic_make_request. If it
1970 * did, we find a non-NULL value in bio_list and re-enter the loop
1971 * from the top. In this case we really did just take the bio
1972 * of the top of the list (no pretending) and so remove it from
1973 * bio_list, and call into ->make_request() again.
1975 BUG_ON(bio
->bi_next
);
1976 bio_list_init(&bio_list_on_stack
);
1977 current
->bio_list
= &bio_list_on_stack
;
1979 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1981 if (likely(blk_queue_enter(q
, false) == 0)) {
1982 ret
= q
->make_request_fn(q
, bio
);
1986 bio
= bio_list_pop(current
->bio_list
);
1988 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
1994 current
->bio_list
= NULL
; /* deactivate */
1999 EXPORT_SYMBOL(generic_make_request
);
2002 * submit_bio - submit a bio to the block device layer for I/O
2003 * @bio: The &struct bio which describes the I/O
2005 * submit_bio() is very similar in purpose to generic_make_request(), and
2006 * uses that function to do most of the work. Both are fairly rough
2007 * interfaces; @bio must be presetup and ready for I/O.
2010 blk_qc_t
submit_bio(struct bio
*bio
)
2013 * If it's a regular read/write or a barrier with data attached,
2014 * go through the normal accounting stuff before submission.
2016 if (bio_has_data(bio
)) {
2019 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2020 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2022 count
= bio_sectors(bio
);
2024 if (op_is_write(bio_op(bio
))) {
2025 count_vm_events(PGPGOUT
, count
);
2027 task_io_account_read(bio
->bi_iter
.bi_size
);
2028 count_vm_events(PGPGIN
, count
);
2031 if (unlikely(block_dump
)) {
2032 char b
[BDEVNAME_SIZE
];
2033 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2034 current
->comm
, task_pid_nr(current
),
2035 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2036 (unsigned long long)bio
->bi_iter
.bi_sector
,
2037 bdevname(bio
->bi_bdev
, b
),
2042 return generic_make_request(bio
);
2044 EXPORT_SYMBOL(submit_bio
);
2047 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2048 * for new the queue limits
2050 * @rq: the request being checked
2053 * @rq may have been made based on weaker limitations of upper-level queues
2054 * in request stacking drivers, and it may violate the limitation of @q.
2055 * Since the block layer and the underlying device driver trust @rq
2056 * after it is inserted to @q, it should be checked against @q before
2057 * the insertion using this generic function.
2059 * Request stacking drivers like request-based dm may change the queue
2060 * limits when retrying requests on other queues. Those requests need
2061 * to be checked against the new queue limits again during dispatch.
2063 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2066 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2067 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2072 * queue's settings related to segment counting like q->bounce_pfn
2073 * may differ from that of other stacking queues.
2074 * Recalculate it to check the request correctly on this queue's
2077 blk_recalc_rq_segments(rq
);
2078 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2079 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2087 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2088 * @q: the queue to submit the request
2089 * @rq: the request being queued
2091 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2093 unsigned long flags
;
2094 int where
= ELEVATOR_INSERT_BACK
;
2096 if (blk_cloned_rq_check_limits(q
, rq
))
2100 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2104 if (blk_queue_io_stat(q
))
2105 blk_account_io_start(rq
, true);
2106 blk_mq_sched_insert_request(rq
, false, true, false, false);
2110 spin_lock_irqsave(q
->queue_lock
, flags
);
2111 if (unlikely(blk_queue_dying(q
))) {
2112 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2117 * Submitting request must be dequeued before calling this function
2118 * because it will be linked to another request_queue
2120 BUG_ON(blk_queued_rq(rq
));
2122 if (op_is_flush(rq
->cmd_flags
))
2123 where
= ELEVATOR_INSERT_FLUSH
;
2125 add_acct_request(q
, rq
, where
);
2126 if (where
== ELEVATOR_INSERT_FLUSH
)
2128 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2132 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2135 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2136 * @rq: request to examine
2139 * A request could be merge of IOs which require different failure
2140 * handling. This function determines the number of bytes which
2141 * can be failed from the beginning of the request without
2142 * crossing into area which need to be retried further.
2145 * The number of bytes to fail.
2148 * queue_lock must be held.
2150 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2152 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2153 unsigned int bytes
= 0;
2156 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2157 return blk_rq_bytes(rq
);
2160 * Currently the only 'mixing' which can happen is between
2161 * different fastfail types. We can safely fail portions
2162 * which have all the failfast bits that the first one has -
2163 * the ones which are at least as eager to fail as the first
2166 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2167 if ((bio
->bi_opf
& ff
) != ff
)
2169 bytes
+= bio
->bi_iter
.bi_size
;
2172 /* this could lead to infinite loop */
2173 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2176 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2178 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2180 if (blk_do_io_stat(req
)) {
2181 const int rw
= rq_data_dir(req
);
2182 struct hd_struct
*part
;
2185 cpu
= part_stat_lock();
2187 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2192 void blk_account_io_done(struct request
*req
)
2195 * Account IO completion. flush_rq isn't accounted as a
2196 * normal IO on queueing nor completion. Accounting the
2197 * containing request is enough.
2199 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2200 unsigned long duration
= jiffies
- req
->start_time
;
2201 const int rw
= rq_data_dir(req
);
2202 struct hd_struct
*part
;
2205 cpu
= part_stat_lock();
2208 part_stat_inc(cpu
, part
, ios
[rw
]);
2209 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2210 part_round_stats(cpu
, part
);
2211 part_dec_in_flight(part
, rw
);
2213 hd_struct_put(part
);
2220 * Don't process normal requests when queue is suspended
2221 * or in the process of suspending/resuming
2223 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2226 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2227 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2233 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2240 void blk_account_io_start(struct request
*rq
, bool new_io
)
2242 struct hd_struct
*part
;
2243 int rw
= rq_data_dir(rq
);
2246 if (!blk_do_io_stat(rq
))
2249 cpu
= part_stat_lock();
2253 part_stat_inc(cpu
, part
, merges
[rw
]);
2255 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2256 if (!hd_struct_try_get(part
)) {
2258 * The partition is already being removed,
2259 * the request will be accounted on the disk only
2261 * We take a reference on disk->part0 although that
2262 * partition will never be deleted, so we can treat
2263 * it as any other partition.
2265 part
= &rq
->rq_disk
->part0
;
2266 hd_struct_get(part
);
2268 part_round_stats(cpu
, part
);
2269 part_inc_in_flight(part
, rw
);
2277 * blk_peek_request - peek at the top of a request queue
2278 * @q: request queue to peek at
2281 * Return the request at the top of @q. The returned request
2282 * should be started using blk_start_request() before LLD starts
2286 * Pointer to the request at the top of @q if available. Null
2290 * queue_lock must be held.
2292 struct request
*blk_peek_request(struct request_queue
*q
)
2297 while ((rq
= __elv_next_request(q
)) != NULL
) {
2299 rq
= blk_pm_peek_request(q
, rq
);
2303 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2305 * This is the first time the device driver
2306 * sees this request (possibly after
2307 * requeueing). Notify IO scheduler.
2309 if (rq
->rq_flags
& RQF_SORTED
)
2310 elv_activate_rq(q
, rq
);
2313 * just mark as started even if we don't start
2314 * it, a request that has been delayed should
2315 * not be passed by new incoming requests
2317 rq
->rq_flags
|= RQF_STARTED
;
2318 trace_block_rq_issue(q
, rq
);
2321 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2322 q
->end_sector
= rq_end_sector(rq
);
2323 q
->boundary_rq
= NULL
;
2326 if (rq
->rq_flags
& RQF_DONTPREP
)
2329 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2331 * make sure space for the drain appears we
2332 * know we can do this because max_hw_segments
2333 * has been adjusted to be one fewer than the
2336 rq
->nr_phys_segments
++;
2342 ret
= q
->prep_rq_fn(q
, rq
);
2343 if (ret
== BLKPREP_OK
) {
2345 } else if (ret
== BLKPREP_DEFER
) {
2347 * the request may have been (partially) prepped.
2348 * we need to keep this request in the front to
2349 * avoid resource deadlock. RQF_STARTED will
2350 * prevent other fs requests from passing this one.
2352 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2353 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2355 * remove the space for the drain we added
2356 * so that we don't add it again
2358 --rq
->nr_phys_segments
;
2363 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2364 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2366 rq
->rq_flags
|= RQF_QUIET
;
2368 * Mark this request as started so we don't trigger
2369 * any debug logic in the end I/O path.
2371 blk_start_request(rq
);
2372 __blk_end_request_all(rq
, err
);
2374 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2381 EXPORT_SYMBOL(blk_peek_request
);
2383 void blk_dequeue_request(struct request
*rq
)
2385 struct request_queue
*q
= rq
->q
;
2387 BUG_ON(list_empty(&rq
->queuelist
));
2388 BUG_ON(ELV_ON_HASH(rq
));
2390 list_del_init(&rq
->queuelist
);
2393 * the time frame between a request being removed from the lists
2394 * and to it is freed is accounted as io that is in progress at
2397 if (blk_account_rq(rq
)) {
2398 q
->in_flight
[rq_is_sync(rq
)]++;
2399 set_io_start_time_ns(rq
);
2404 * blk_start_request - start request processing on the driver
2405 * @req: request to dequeue
2408 * Dequeue @req and start timeout timer on it. This hands off the
2409 * request to the driver.
2411 * Block internal functions which don't want to start timer should
2412 * call blk_dequeue_request().
2415 * queue_lock must be held.
2417 void blk_start_request(struct request
*req
)
2419 blk_dequeue_request(req
);
2421 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2422 blk_stat_set_issue_time(&req
->issue_stat
);
2423 req
->rq_flags
|= RQF_STATS
;
2424 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2428 * We are now handing the request to the hardware, initialize
2429 * resid_len to full count and add the timeout handler.
2431 req
->resid_len
= blk_rq_bytes(req
);
2432 if (unlikely(blk_bidi_rq(req
)))
2433 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2435 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2438 EXPORT_SYMBOL(blk_start_request
);
2441 * blk_fetch_request - fetch a request from a request queue
2442 * @q: request queue to fetch a request from
2445 * Return the request at the top of @q. The request is started on
2446 * return and LLD can start processing it immediately.
2449 * Pointer to the request at the top of @q if available. Null
2453 * queue_lock must be held.
2455 struct request
*blk_fetch_request(struct request_queue
*q
)
2459 rq
= blk_peek_request(q
);
2461 blk_start_request(rq
);
2464 EXPORT_SYMBOL(blk_fetch_request
);
2467 * blk_update_request - Special helper function for request stacking drivers
2468 * @req: the request being processed
2469 * @error: %0 for success, < %0 for error
2470 * @nr_bytes: number of bytes to complete @req
2473 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2474 * the request structure even if @req doesn't have leftover.
2475 * If @req has leftover, sets it up for the next range of segments.
2477 * This special helper function is only for request stacking drivers
2478 * (e.g. request-based dm) so that they can handle partial completion.
2479 * Actual device drivers should use blk_end_request instead.
2481 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2482 * %false return from this function.
2485 * %false - this request doesn't have any more data
2486 * %true - this request has more data
2488 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2492 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2498 * For fs requests, rq is just carrier of independent bio's
2499 * and each partial completion should be handled separately.
2500 * Reset per-request error on each partial completion.
2502 * TODO: tj: This is too subtle. It would be better to let
2503 * low level drivers do what they see fit.
2505 if (req
->cmd_type
== REQ_TYPE_FS
)
2508 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2509 !(req
->rq_flags
& RQF_QUIET
)) {
2514 error_type
= "recoverable transport";
2517 error_type
= "critical target";
2520 error_type
= "critical nexus";
2523 error_type
= "timeout";
2526 error_type
= "critical space allocation";
2529 error_type
= "critical medium";
2536 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2537 __func__
, error_type
, req
->rq_disk
?
2538 req
->rq_disk
->disk_name
: "?",
2539 (unsigned long long)blk_rq_pos(req
));
2543 blk_account_io_completion(req
, nr_bytes
);
2547 struct bio
*bio
= req
->bio
;
2548 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2550 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2551 req
->bio
= bio
->bi_next
;
2553 req_bio_endio(req
, bio
, bio_bytes
, error
);
2555 total_bytes
+= bio_bytes
;
2556 nr_bytes
-= bio_bytes
;
2567 * Reset counters so that the request stacking driver
2568 * can find how many bytes remain in the request
2571 req
->__data_len
= 0;
2575 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2577 req
->__data_len
-= total_bytes
;
2579 /* update sector only for requests with clear definition of sector */
2580 if (req
->cmd_type
== REQ_TYPE_FS
)
2581 req
->__sector
+= total_bytes
>> 9;
2583 /* mixed attributes always follow the first bio */
2584 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2585 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2586 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2590 * If total number of sectors is less than the first segment
2591 * size, something has gone terribly wrong.
2593 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2594 blk_dump_rq_flags(req
, "request botched");
2595 req
->__data_len
= blk_rq_cur_bytes(req
);
2598 /* recalculate the number of segments */
2599 blk_recalc_rq_segments(req
);
2603 EXPORT_SYMBOL_GPL(blk_update_request
);
2605 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2606 unsigned int nr_bytes
,
2607 unsigned int bidi_bytes
)
2609 if (blk_update_request(rq
, error
, nr_bytes
))
2612 /* Bidi request must be completed as a whole */
2613 if (unlikely(blk_bidi_rq(rq
)) &&
2614 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2617 if (blk_queue_add_random(rq
->q
))
2618 add_disk_randomness(rq
->rq_disk
);
2624 * blk_unprep_request - unprepare a request
2627 * This function makes a request ready for complete resubmission (or
2628 * completion). It happens only after all error handling is complete,
2629 * so represents the appropriate moment to deallocate any resources
2630 * that were allocated to the request in the prep_rq_fn. The queue
2631 * lock is held when calling this.
2633 void blk_unprep_request(struct request
*req
)
2635 struct request_queue
*q
= req
->q
;
2637 req
->rq_flags
&= ~RQF_DONTPREP
;
2638 if (q
->unprep_rq_fn
)
2639 q
->unprep_rq_fn(q
, req
);
2641 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2644 * queue lock must be held
2646 void blk_finish_request(struct request
*req
, int error
)
2648 struct request_queue
*q
= req
->q
;
2650 if (req
->rq_flags
& RQF_STATS
)
2651 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2653 if (req
->rq_flags
& RQF_QUEUED
)
2654 blk_queue_end_tag(q
, req
);
2656 BUG_ON(blk_queued_rq(req
));
2658 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2659 laptop_io_completion(&req
->q
->backing_dev_info
);
2661 blk_delete_timer(req
);
2663 if (req
->rq_flags
& RQF_DONTPREP
)
2664 blk_unprep_request(req
);
2666 blk_account_io_done(req
);
2669 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2670 req
->end_io(req
, error
);
2672 if (blk_bidi_rq(req
))
2673 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2675 __blk_put_request(q
, req
);
2678 EXPORT_SYMBOL(blk_finish_request
);
2681 * blk_end_bidi_request - Complete a bidi request
2682 * @rq: the request to complete
2683 * @error: %0 for success, < %0 for error
2684 * @nr_bytes: number of bytes to complete @rq
2685 * @bidi_bytes: number of bytes to complete @rq->next_rq
2688 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2689 * Drivers that supports bidi can safely call this member for any
2690 * type of request, bidi or uni. In the later case @bidi_bytes is
2694 * %false - we are done with this request
2695 * %true - still buffers pending for this request
2697 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2698 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2700 struct request_queue
*q
= rq
->q
;
2701 unsigned long flags
;
2703 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2706 spin_lock_irqsave(q
->queue_lock
, flags
);
2707 blk_finish_request(rq
, error
);
2708 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2714 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2715 * @rq: the request to complete
2716 * @error: %0 for success, < %0 for error
2717 * @nr_bytes: number of bytes to complete @rq
2718 * @bidi_bytes: number of bytes to complete @rq->next_rq
2721 * Identical to blk_end_bidi_request() except that queue lock is
2722 * assumed to be locked on entry and remains so on return.
2725 * %false - we are done with this request
2726 * %true - still buffers pending for this request
2728 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2729 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2731 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2734 blk_finish_request(rq
, error
);
2740 * blk_end_request - Helper function for drivers to complete the request.
2741 * @rq: the request being processed
2742 * @error: %0 for success, < %0 for error
2743 * @nr_bytes: number of bytes to complete
2746 * Ends I/O on a number of bytes attached to @rq.
2747 * If @rq has leftover, sets it up for the next range of segments.
2750 * %false - we are done with this request
2751 * %true - still buffers pending for this request
2753 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2755 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2757 EXPORT_SYMBOL(blk_end_request
);
2760 * blk_end_request_all - Helper function for drives to finish the request.
2761 * @rq: the request to finish
2762 * @error: %0 for success, < %0 for error
2765 * Completely finish @rq.
2767 void blk_end_request_all(struct request
*rq
, int error
)
2770 unsigned int bidi_bytes
= 0;
2772 if (unlikely(blk_bidi_rq(rq
)))
2773 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2775 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2778 EXPORT_SYMBOL(blk_end_request_all
);
2781 * blk_end_request_cur - Helper function to finish the current request chunk.
2782 * @rq: the request to finish the current chunk for
2783 * @error: %0 for success, < %0 for error
2786 * Complete the current consecutively mapped chunk from @rq.
2789 * %false - we are done with this request
2790 * %true - still buffers pending for this request
2792 bool blk_end_request_cur(struct request
*rq
, int error
)
2794 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2796 EXPORT_SYMBOL(blk_end_request_cur
);
2799 * blk_end_request_err - Finish a request till the next failure boundary.
2800 * @rq: the request to finish till the next failure boundary for
2801 * @error: must be negative errno
2804 * Complete @rq till the next failure boundary.
2807 * %false - we are done with this request
2808 * %true - still buffers pending for this request
2810 bool blk_end_request_err(struct request
*rq
, int error
)
2812 WARN_ON(error
>= 0);
2813 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2815 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2818 * __blk_end_request - Helper function for drivers to complete the request.
2819 * @rq: the request being processed
2820 * @error: %0 for success, < %0 for error
2821 * @nr_bytes: number of bytes to complete
2824 * Must be called with queue lock held unlike blk_end_request().
2827 * %false - we are done with this request
2828 * %true - still buffers pending for this request
2830 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2832 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2834 EXPORT_SYMBOL(__blk_end_request
);
2837 * __blk_end_request_all - Helper function for drives to finish the request.
2838 * @rq: the request to finish
2839 * @error: %0 for success, < %0 for error
2842 * Completely finish @rq. Must be called with queue lock held.
2844 void __blk_end_request_all(struct request
*rq
, int error
)
2847 unsigned int bidi_bytes
= 0;
2849 if (unlikely(blk_bidi_rq(rq
)))
2850 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2852 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2855 EXPORT_SYMBOL(__blk_end_request_all
);
2858 * __blk_end_request_cur - Helper function to finish the current request chunk.
2859 * @rq: the request to finish the current chunk for
2860 * @error: %0 for success, < %0 for error
2863 * Complete the current consecutively mapped chunk from @rq. Must
2864 * be called with queue lock held.
2867 * %false - we are done with this request
2868 * %true - still buffers pending for this request
2870 bool __blk_end_request_cur(struct request
*rq
, int error
)
2872 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2874 EXPORT_SYMBOL(__blk_end_request_cur
);
2877 * __blk_end_request_err - Finish a request till the next failure boundary.
2878 * @rq: the request to finish till the next failure boundary for
2879 * @error: must be negative errno
2882 * Complete @rq till the next failure boundary. Must be called
2883 * with queue lock held.
2886 * %false - we are done with this request
2887 * %true - still buffers pending for this request
2889 bool __blk_end_request_err(struct request
*rq
, int error
)
2891 WARN_ON(error
>= 0);
2892 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2894 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2896 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2899 if (bio_has_data(bio
))
2900 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2902 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2903 rq
->bio
= rq
->biotail
= bio
;
2906 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2909 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2911 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2912 * @rq: the request to be flushed
2915 * Flush all pages in @rq.
2917 void rq_flush_dcache_pages(struct request
*rq
)
2919 struct req_iterator iter
;
2920 struct bio_vec bvec
;
2922 rq_for_each_segment(bvec
, rq
, iter
)
2923 flush_dcache_page(bvec
.bv_page
);
2925 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2929 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2930 * @q : the queue of the device being checked
2933 * Check if underlying low-level drivers of a device are busy.
2934 * If the drivers want to export their busy state, they must set own
2935 * exporting function using blk_queue_lld_busy() first.
2937 * Basically, this function is used only by request stacking drivers
2938 * to stop dispatching requests to underlying devices when underlying
2939 * devices are busy. This behavior helps more I/O merging on the queue
2940 * of the request stacking driver and prevents I/O throughput regression
2941 * on burst I/O load.
2944 * 0 - Not busy (The request stacking driver should dispatch request)
2945 * 1 - Busy (The request stacking driver should stop dispatching request)
2947 int blk_lld_busy(struct request_queue
*q
)
2950 return q
->lld_busy_fn(q
);
2954 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2957 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2958 * @rq: the clone request to be cleaned up
2961 * Free all bios in @rq for a cloned request.
2963 void blk_rq_unprep_clone(struct request
*rq
)
2967 while ((bio
= rq
->bio
) != NULL
) {
2968 rq
->bio
= bio
->bi_next
;
2973 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2976 * Copy attributes of the original request to the clone request.
2977 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2979 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2981 dst
->cpu
= src
->cpu
;
2982 dst
->cmd_flags
= src
->cmd_flags
| REQ_NOMERGE
;
2983 dst
->cmd_type
= src
->cmd_type
;
2984 dst
->__sector
= blk_rq_pos(src
);
2985 dst
->__data_len
= blk_rq_bytes(src
);
2986 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2987 dst
->ioprio
= src
->ioprio
;
2988 dst
->extra_len
= src
->extra_len
;
2992 * blk_rq_prep_clone - Helper function to setup clone request
2993 * @rq: the request to be setup
2994 * @rq_src: original request to be cloned
2995 * @bs: bio_set that bios for clone are allocated from
2996 * @gfp_mask: memory allocation mask for bio
2997 * @bio_ctr: setup function to be called for each clone bio.
2998 * Returns %0 for success, non %0 for failure.
2999 * @data: private data to be passed to @bio_ctr
3002 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3003 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3004 * are not copied, and copying such parts is the caller's responsibility.
3005 * Also, pages which the original bios are pointing to are not copied
3006 * and the cloned bios just point same pages.
3007 * So cloned bios must be completed before original bios, which means
3008 * the caller must complete @rq before @rq_src.
3010 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3011 struct bio_set
*bs
, gfp_t gfp_mask
,
3012 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3015 struct bio
*bio
, *bio_src
;
3020 __rq_for_each_bio(bio_src
, rq_src
) {
3021 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3025 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3029 rq
->biotail
->bi_next
= bio
;
3032 rq
->bio
= rq
->biotail
= bio
;
3035 __blk_rq_prep_clone(rq
, rq_src
);
3042 blk_rq_unprep_clone(rq
);
3046 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3048 int kblockd_schedule_work(struct work_struct
*work
)
3050 return queue_work(kblockd_workqueue
, work
);
3052 EXPORT_SYMBOL(kblockd_schedule_work
);
3054 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3056 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3058 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3060 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3061 unsigned long delay
)
3063 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3065 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3067 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3068 unsigned long delay
)
3070 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3072 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3075 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3076 * @plug: The &struct blk_plug that needs to be initialized
3079 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3080 * pending I/O should the task end up blocking between blk_start_plug() and
3081 * blk_finish_plug(). This is important from a performance perspective, but
3082 * also ensures that we don't deadlock. For instance, if the task is blocking
3083 * for a memory allocation, memory reclaim could end up wanting to free a
3084 * page belonging to that request that is currently residing in our private
3085 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3086 * this kind of deadlock.
3088 void blk_start_plug(struct blk_plug
*plug
)
3090 struct task_struct
*tsk
= current
;
3093 * If this is a nested plug, don't actually assign it.
3098 INIT_LIST_HEAD(&plug
->list
);
3099 INIT_LIST_HEAD(&plug
->mq_list
);
3100 INIT_LIST_HEAD(&plug
->cb_list
);
3102 * Store ordering should not be needed here, since a potential
3103 * preempt will imply a full memory barrier
3107 EXPORT_SYMBOL(blk_start_plug
);
3109 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3111 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3112 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3114 return !(rqa
->q
< rqb
->q
||
3115 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3119 * If 'from_schedule' is true, then postpone the dispatch of requests
3120 * until a safe kblockd context. We due this to avoid accidental big
3121 * additional stack usage in driver dispatch, in places where the originally
3122 * plugger did not intend it.
3124 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3126 __releases(q
->queue_lock
)
3128 trace_block_unplug(q
, depth
, !from_schedule
);
3131 blk_run_queue_async(q
);
3134 spin_unlock(q
->queue_lock
);
3137 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3139 LIST_HEAD(callbacks
);
3141 while (!list_empty(&plug
->cb_list
)) {
3142 list_splice_init(&plug
->cb_list
, &callbacks
);
3144 while (!list_empty(&callbacks
)) {
3145 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3148 list_del(&cb
->list
);
3149 cb
->callback(cb
, from_schedule
);
3154 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3157 struct blk_plug
*plug
= current
->plug
;
3158 struct blk_plug_cb
*cb
;
3163 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3164 if (cb
->callback
== unplug
&& cb
->data
== data
)
3167 /* Not currently on the callback list */
3168 BUG_ON(size
< sizeof(*cb
));
3169 cb
= kzalloc(size
, GFP_ATOMIC
);
3172 cb
->callback
= unplug
;
3173 list_add(&cb
->list
, &plug
->cb_list
);
3177 EXPORT_SYMBOL(blk_check_plugged
);
3179 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3181 struct request_queue
*q
;
3182 unsigned long flags
;
3187 flush_plug_callbacks(plug
, from_schedule
);
3189 if (!list_empty(&plug
->mq_list
))
3190 blk_mq_flush_plug_list(plug
, from_schedule
);
3192 if (list_empty(&plug
->list
))
3195 list_splice_init(&plug
->list
, &list
);
3197 list_sort(NULL
, &list
, plug_rq_cmp
);
3203 * Save and disable interrupts here, to avoid doing it for every
3204 * queue lock we have to take.
3206 local_irq_save(flags
);
3207 while (!list_empty(&list
)) {
3208 rq
= list_entry_rq(list
.next
);
3209 list_del_init(&rq
->queuelist
);
3213 * This drops the queue lock
3216 queue_unplugged(q
, depth
, from_schedule
);
3219 spin_lock(q
->queue_lock
);
3223 * Short-circuit if @q is dead
3225 if (unlikely(blk_queue_dying(q
))) {
3226 __blk_end_request_all(rq
, -ENODEV
);
3231 * rq is already accounted, so use raw insert
3233 if (op_is_flush(rq
->cmd_flags
))
3234 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3236 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3242 * This drops the queue lock
3245 queue_unplugged(q
, depth
, from_schedule
);
3247 local_irq_restore(flags
);
3250 void blk_finish_plug(struct blk_plug
*plug
)
3252 if (plug
!= current
->plug
)
3254 blk_flush_plug_list(plug
, false);
3256 current
->plug
= NULL
;
3258 EXPORT_SYMBOL(blk_finish_plug
);
3262 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3263 * @q: the queue of the device
3264 * @dev: the device the queue belongs to
3267 * Initialize runtime-PM-related fields for @q and start auto suspend for
3268 * @dev. Drivers that want to take advantage of request-based runtime PM
3269 * should call this function after @dev has been initialized, and its
3270 * request queue @q has been allocated, and runtime PM for it can not happen
3271 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3272 * cases, driver should call this function before any I/O has taken place.
3274 * This function takes care of setting up using auto suspend for the device,
3275 * the autosuspend delay is set to -1 to make runtime suspend impossible
3276 * until an updated value is either set by user or by driver. Drivers do
3277 * not need to touch other autosuspend settings.
3279 * The block layer runtime PM is request based, so only works for drivers
3280 * that use request as their IO unit instead of those directly use bio's.
3282 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3285 q
->rpm_status
= RPM_ACTIVE
;
3286 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3287 pm_runtime_use_autosuspend(q
->dev
);
3289 EXPORT_SYMBOL(blk_pm_runtime_init
);
3292 * blk_pre_runtime_suspend - Pre runtime suspend check
3293 * @q: the queue of the device
3296 * This function will check if runtime suspend is allowed for the device
3297 * by examining if there are any requests pending in the queue. If there
3298 * are requests pending, the device can not be runtime suspended; otherwise,
3299 * the queue's status will be updated to SUSPENDING and the driver can
3300 * proceed to suspend the device.
3302 * For the not allowed case, we mark last busy for the device so that
3303 * runtime PM core will try to autosuspend it some time later.
3305 * This function should be called near the start of the device's
3306 * runtime_suspend callback.
3309 * 0 - OK to runtime suspend the device
3310 * -EBUSY - Device should not be runtime suspended
3312 int blk_pre_runtime_suspend(struct request_queue
*q
)
3319 spin_lock_irq(q
->queue_lock
);
3320 if (q
->nr_pending
) {
3322 pm_runtime_mark_last_busy(q
->dev
);
3324 q
->rpm_status
= RPM_SUSPENDING
;
3326 spin_unlock_irq(q
->queue_lock
);
3329 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3332 * blk_post_runtime_suspend - Post runtime suspend processing
3333 * @q: the queue of the device
3334 * @err: return value of the device's runtime_suspend function
3337 * Update the queue's runtime status according to the return value of the
3338 * device's runtime suspend function and mark last busy for the device so
3339 * that PM core will try to auto suspend the device at a later time.
3341 * This function should be called near the end of the device's
3342 * runtime_suspend callback.
3344 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3349 spin_lock_irq(q
->queue_lock
);
3351 q
->rpm_status
= RPM_SUSPENDED
;
3353 q
->rpm_status
= RPM_ACTIVE
;
3354 pm_runtime_mark_last_busy(q
->dev
);
3356 spin_unlock_irq(q
->queue_lock
);
3358 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3361 * blk_pre_runtime_resume - Pre runtime resume processing
3362 * @q: the queue of the device
3365 * Update the queue's runtime status to RESUMING in preparation for the
3366 * runtime resume of the device.
3368 * This function should be called near the start of the device's
3369 * runtime_resume callback.
3371 void blk_pre_runtime_resume(struct request_queue
*q
)
3376 spin_lock_irq(q
->queue_lock
);
3377 q
->rpm_status
= RPM_RESUMING
;
3378 spin_unlock_irq(q
->queue_lock
);
3380 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3383 * blk_post_runtime_resume - Post runtime resume processing
3384 * @q: the queue of the device
3385 * @err: return value of the device's runtime_resume function
3388 * Update the queue's runtime status according to the return value of the
3389 * device's runtime_resume function. If it is successfully resumed, process
3390 * the requests that are queued into the device's queue when it is resuming
3391 * and then mark last busy and initiate autosuspend for it.
3393 * This function should be called near the end of the device's
3394 * runtime_resume callback.
3396 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3401 spin_lock_irq(q
->queue_lock
);
3403 q
->rpm_status
= RPM_ACTIVE
;
3405 pm_runtime_mark_last_busy(q
->dev
);
3406 pm_request_autosuspend(q
->dev
);
3408 q
->rpm_status
= RPM_SUSPENDED
;
3410 spin_unlock_irq(q
->queue_lock
);
3412 EXPORT_SYMBOL(blk_post_runtime_resume
);
3415 * blk_set_runtime_active - Force runtime status of the queue to be active
3416 * @q: the queue of the device
3418 * If the device is left runtime suspended during system suspend the resume
3419 * hook typically resumes the device and corrects runtime status
3420 * accordingly. However, that does not affect the queue runtime PM status
3421 * which is still "suspended". This prevents processing requests from the
3424 * This function can be used in driver's resume hook to correct queue
3425 * runtime PM status and re-enable peeking requests from the queue. It
3426 * should be called before first request is added to the queue.
3428 void blk_set_runtime_active(struct request_queue
*q
)
3430 spin_lock_irq(q
->queue_lock
);
3431 q
->rpm_status
= RPM_ACTIVE
;
3432 pm_runtime_mark_last_busy(q
->dev
);
3433 pm_request_autosuspend(q
->dev
);
3434 spin_unlock_irq(q
->queue_lock
);
3436 EXPORT_SYMBOL(blk_set_runtime_active
);
3439 int __init
blk_dev_init(void)
3441 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3442 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3443 FIELD_SIZEOF(struct request
, cmd_flags
));
3444 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3445 FIELD_SIZEOF(struct bio
, bi_opf
));
3447 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3448 kblockd_workqueue
= alloc_workqueue("kblockd",
3449 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3450 if (!kblockd_workqueue
)
3451 panic("Failed to create kblockd\n");
3453 request_cachep
= kmem_cache_create("blkdev_requests",
3454 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3456 blk_requestq_cachep
= kmem_cache_create("request_queue",
3457 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);