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 #include <linux/math64.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
= NULL
;
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 while there is an
114 * active reference against the parent gendisk.
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
->start_time
= jiffies
;
139 set_start_time_ns(rq
);
142 EXPORT_SYMBOL(blk_rq_init
);
144 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
145 unsigned int nbytes
, int error
)
148 bio
->bi_error
= error
;
150 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
151 bio_set_flag(bio
, BIO_QUIET
);
153 bio_advance(bio
, nbytes
);
155 /* don't actually finish bio if it's part of flush sequence */
156 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
160 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
164 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
165 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
166 (unsigned long long) rq
->cmd_flags
);
168 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
169 (unsigned long long)blk_rq_pos(rq
),
170 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
171 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
172 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
174 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
175 printk(KERN_INFO
" cdb: ");
176 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
177 printk("%02x ", rq
->cmd
[bit
]);
181 EXPORT_SYMBOL(blk_dump_rq_flags
);
183 static void blk_delay_work(struct work_struct
*work
)
185 struct request_queue
*q
;
187 q
= container_of(work
, struct request_queue
, delay_work
.work
);
188 spin_lock_irq(q
->queue_lock
);
190 spin_unlock_irq(q
->queue_lock
);
194 * blk_delay_queue - restart queueing after defined interval
195 * @q: The &struct request_queue in question
196 * @msecs: Delay in msecs
199 * Sometimes queueing needs to be postponed for a little while, to allow
200 * resources to come back. This function will make sure that queueing is
201 * restarted around the specified time. Queue lock must be held.
203 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
205 if (likely(!blk_queue_dead(q
)))
206 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
207 msecs_to_jiffies(msecs
));
209 EXPORT_SYMBOL(blk_delay_queue
);
212 * blk_start_queue_async - asynchronously restart a previously stopped queue
213 * @q: The &struct request_queue in question
216 * blk_start_queue_async() will clear the stop flag on the queue, and
217 * ensure that the request_fn for the queue is run from an async
220 void blk_start_queue_async(struct request_queue
*q
)
222 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
223 blk_run_queue_async(q
);
225 EXPORT_SYMBOL(blk_start_queue_async
);
228 * blk_start_queue - restart a previously stopped queue
229 * @q: The &struct request_queue in question
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 void blk_start_queue(struct request_queue
*q
)
238 WARN_ON(!in_interrupt() && !irqs_disabled());
240 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
243 EXPORT_SYMBOL(blk_start_queue
);
246 * blk_stop_queue - stop a queue
247 * @q: The &struct request_queue in question
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 void blk_stop_queue(struct request_queue
*q
)
261 cancel_delayed_work(&q
->delay_work
);
262 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
264 EXPORT_SYMBOL(blk_stop_queue
);
267 * blk_sync_queue - cancel any pending callbacks on a queue
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
275 * that the callbacks might use. The caller must already have made sure
276 * that its ->make_request_fn will not re-add plugging prior to calling
279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevator_exit()
281 * and blkcg_exit_queue() to be called with queue lock initialized.
284 void blk_sync_queue(struct request_queue
*q
)
286 del_timer_sync(&q
->timeout
);
289 struct blk_mq_hw_ctx
*hctx
;
292 queue_for_each_hw_ctx(q
, hctx
, i
) {
293 cancel_delayed_work_sync(&hctx
->run_work
);
294 cancel_delayed_work_sync(&hctx
->delay_work
);
297 cancel_delayed_work_sync(&q
->delay_work
);
300 EXPORT_SYMBOL(blk_sync_queue
);
303 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
304 * @q: The queue to run
307 * Invoke request handling on a queue if there are any pending requests.
308 * May be used to restart request handling after a request has completed.
309 * This variant runs the queue whether or not the queue has been
310 * stopped. Must be called with the queue lock held and interrupts
311 * disabled. See also @blk_run_queue.
313 inline void __blk_run_queue_uncond(struct request_queue
*q
)
315 if (unlikely(blk_queue_dead(q
)))
319 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
320 * the queue lock internally. As a result multiple threads may be
321 * running such a request function concurrently. Keep track of the
322 * number of active request_fn invocations such that blk_drain_queue()
323 * can wait until all these request_fn calls have finished.
325 q
->request_fn_active
++;
327 q
->request_fn_active
--;
329 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
332 * __blk_run_queue - run a single device queue
333 * @q: The queue to run
336 * See @blk_run_queue. This variant must be called with the queue lock
337 * held and interrupts disabled.
339 void __blk_run_queue(struct request_queue
*q
)
341 if (unlikely(blk_queue_stopped(q
)))
344 __blk_run_queue_uncond(q
);
346 EXPORT_SYMBOL(__blk_run_queue
);
349 * blk_run_queue_async - run a single device queue in workqueue context
350 * @q: The queue to run
353 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
354 * of us. The caller must hold the queue lock.
356 void blk_run_queue_async(struct request_queue
*q
)
358 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
359 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
361 EXPORT_SYMBOL(blk_run_queue_async
);
364 * blk_run_queue - run a single device queue
365 * @q: The queue to run
368 * Invoke request handling on this queue, if it has pending work to do.
369 * May be used to restart queueing when a request has completed.
371 void blk_run_queue(struct request_queue
*q
)
375 spin_lock_irqsave(q
->queue_lock
, flags
);
377 spin_unlock_irqrestore(q
->queue_lock
, flags
);
379 EXPORT_SYMBOL(blk_run_queue
);
381 void blk_put_queue(struct request_queue
*q
)
383 kobject_put(&q
->kobj
);
385 EXPORT_SYMBOL(blk_put_queue
);
388 * __blk_drain_queue - drain requests from request_queue
390 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
392 * Drain requests from @q. If @drain_all is set, all requests are drained.
393 * If not, only ELVPRIV requests are drained. The caller is responsible
394 * for ensuring that no new requests which need to be drained are queued.
396 void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
397 __releases(q
->queue_lock
)
398 __acquires(q
->queue_lock
)
402 lockdep_assert_held(q
->queue_lock
);
408 * The caller might be trying to drain @q before its
409 * elevator is initialized.
412 elv_drain_elevator(q
);
414 blkcg_drain_queue(q
);
417 * This function might be called on a queue which failed
418 * driver init after queue creation or is not yet fully
419 * active yet. Some drivers (e.g. fd and loop) get unhappy
420 * in such cases. Kick queue iff dispatch queue has
421 * something on it and @q has request_fn set.
423 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
426 drain
|= q
->nr_rqs_elvpriv
;
427 drain
|= q
->request_fn_active
;
430 * Unfortunately, requests are queued at and tracked from
431 * multiple places and there's no single counter which can
432 * be drained. Check all the queues and counters.
435 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
436 drain
|= !list_empty(&q
->queue_head
);
437 for (i
= 0; i
< 2; i
++) {
438 drain
|= q
->nr_rqs
[i
];
439 drain
|= q
->in_flight
[i
];
441 drain
|= !list_empty(&fq
->flush_queue
[i
]);
448 spin_unlock_irq(q
->queue_lock
);
452 spin_lock_irq(q
->queue_lock
);
456 * With queue marked dead, any woken up waiter will fail the
457 * allocation path, so the wakeup chaining is lost and we're
458 * left with hung waiters. We need to wake up those waiters.
461 struct request_list
*rl
;
463 blk_queue_for_each_rl(rl
, q
)
464 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
465 wake_up_all(&rl
->wait
[i
]);
470 * blk_queue_bypass_start - enter queue bypass mode
471 * @q: queue of interest
473 * In bypass mode, only the dispatch FIFO queue of @q is used. This
474 * function makes @q enter bypass mode and drains all requests which were
475 * throttled or issued before. On return, it's guaranteed that no request
476 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
477 * inside queue or RCU read lock.
479 void blk_queue_bypass_start(struct request_queue
*q
)
481 spin_lock_irq(q
->queue_lock
);
483 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
484 spin_unlock_irq(q
->queue_lock
);
487 * Queues start drained. Skip actual draining till init is
488 * complete. This avoids lenghty delays during queue init which
489 * can happen many times during boot.
491 if (blk_queue_init_done(q
)) {
492 spin_lock_irq(q
->queue_lock
);
493 __blk_drain_queue(q
, false);
494 spin_unlock_irq(q
->queue_lock
);
496 /* ensure blk_queue_bypass() is %true inside RCU read lock */
500 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
503 * blk_queue_bypass_end - leave queue bypass mode
504 * @q: queue of interest
506 * Leave bypass mode and restore the normal queueing behavior.
508 void blk_queue_bypass_end(struct request_queue
*q
)
510 spin_lock_irq(q
->queue_lock
);
511 if (!--q
->bypass_depth
)
512 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
513 WARN_ON_ONCE(q
->bypass_depth
< 0);
514 spin_unlock_irq(q
->queue_lock
);
516 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
518 void blk_set_queue_dying(struct request_queue
*q
)
520 spin_lock_irq(q
->queue_lock
);
521 queue_flag_set(QUEUE_FLAG_DYING
, q
);
522 spin_unlock_irq(q
->queue_lock
);
525 blk_mq_wake_waiters(q
);
527 struct request_list
*rl
;
529 blk_queue_for_each_rl(rl
, q
) {
531 wake_up_all(&rl
->wait
[BLK_RW_SYNC
]);
532 wake_up_all(&rl
->wait
[BLK_RW_ASYNC
]);
537 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
540 * blk_cleanup_queue - shutdown a request queue
541 * @q: request queue to shutdown
543 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
544 * put it. All future requests will be failed immediately with -ENODEV.
546 void blk_cleanup_queue(struct request_queue
*q
)
548 spinlock_t
*lock
= q
->queue_lock
;
550 /* mark @q DYING, no new request or merges will be allowed afterwards */
551 mutex_lock(&q
->sysfs_lock
);
552 blk_set_queue_dying(q
);
556 * A dying queue is permanently in bypass mode till released. Note
557 * that, unlike blk_queue_bypass_start(), we aren't performing
558 * synchronize_rcu() after entering bypass mode to avoid the delay
559 * as some drivers create and destroy a lot of queues while
560 * probing. This is still safe because blk_release_queue() will be
561 * called only after the queue refcnt drops to zero and nothing,
562 * RCU or not, would be traversing the queue by then.
565 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
567 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
568 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
569 queue_flag_set(QUEUE_FLAG_DYING
, q
);
570 spin_unlock_irq(lock
);
571 mutex_unlock(&q
->sysfs_lock
);
574 * Drain all requests queued before DYING marking. Set DEAD flag to
575 * prevent that q->request_fn() gets invoked after draining finished.
580 __blk_drain_queue(q
, true);
581 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
582 spin_unlock_irq(lock
);
584 /* for synchronous bio-based driver finish in-flight integrity i/o */
585 blk_flush_integrity();
587 /* @q won't process any more request, flush async actions */
588 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
592 blk_mq_free_queue(q
);
593 percpu_ref_exit(&q
->q_usage_counter
);
596 if (q
->queue_lock
!= &q
->__queue_lock
)
597 q
->queue_lock
= &q
->__queue_lock
;
598 spin_unlock_irq(lock
);
600 bdi_unregister(&q
->backing_dev_info
);
602 /* @q is and will stay empty, shutdown and put */
605 EXPORT_SYMBOL(blk_cleanup_queue
);
607 /* Allocate memory local to the request queue */
608 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
610 int nid
= (int)(long)data
;
611 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
614 static void free_request_struct(void *element
, void *unused
)
616 kmem_cache_free(request_cachep
, element
);
619 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
622 if (unlikely(rl
->rq_pool
))
626 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
627 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
628 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
629 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
631 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
633 (void *)(long)q
->node
, gfp_mask
,
641 void blk_exit_rl(struct request_list
*rl
)
644 mempool_destroy(rl
->rq_pool
);
647 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
649 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
651 EXPORT_SYMBOL(blk_alloc_queue
);
653 int blk_queue_enter(struct request_queue
*q
, gfp_t gfp
)
658 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
661 if (!gfpflags_allow_blocking(gfp
))
664 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
665 !atomic_read(&q
->mq_freeze_depth
) ||
667 if (blk_queue_dying(q
))
674 void blk_queue_exit(struct request_queue
*q
)
676 percpu_ref_put(&q
->q_usage_counter
);
679 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
681 struct request_queue
*q
=
682 container_of(ref
, struct request_queue
, q_usage_counter
);
684 wake_up_all(&q
->mq_freeze_wq
);
687 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
689 struct request_queue
*q
;
692 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
693 gfp_mask
| __GFP_ZERO
, node_id
);
697 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
701 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
705 q
->backing_dev_info
.ra_pages
=
706 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
707 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
708 q
->backing_dev_info
.name
= "block";
711 err
= bdi_init(&q
->backing_dev_info
);
715 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
716 laptop_mode_timer_fn
, (unsigned long) q
);
717 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
718 INIT_LIST_HEAD(&q
->queue_head
);
719 INIT_LIST_HEAD(&q
->timeout_list
);
720 INIT_LIST_HEAD(&q
->icq_list
);
721 #ifdef CONFIG_BLK_CGROUP
722 INIT_LIST_HEAD(&q
->blkg_list
);
724 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
726 kobject_init(&q
->kobj
, &blk_queue_ktype
);
728 mutex_init(&q
->sysfs_lock
);
729 spin_lock_init(&q
->__queue_lock
);
732 * By default initialize queue_lock to internal lock and driver can
733 * override it later if need be.
735 q
->queue_lock
= &q
->__queue_lock
;
738 * A queue starts its life with bypass turned on to avoid
739 * unnecessary bypass on/off overhead and nasty surprises during
740 * init. The initial bypass will be finished when the queue is
741 * registered by blk_register_queue().
744 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
746 init_waitqueue_head(&q
->mq_freeze_wq
);
749 * Init percpu_ref in atomic mode so that it's faster to shutdown.
750 * See blk_register_queue() for details.
752 if (percpu_ref_init(&q
->q_usage_counter
,
753 blk_queue_usage_counter_release
,
754 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
757 if (blkcg_init_queue(q
))
763 percpu_ref_exit(&q
->q_usage_counter
);
765 bdi_destroy(&q
->backing_dev_info
);
767 bioset_free(q
->bio_split
);
769 ida_simple_remove(&blk_queue_ida
, q
->id
);
771 kmem_cache_free(blk_requestq_cachep
, q
);
774 EXPORT_SYMBOL(blk_alloc_queue_node
);
777 * blk_init_queue - prepare a request queue for use with a block device
778 * @rfn: The function to be called to process requests that have been
779 * placed on the queue.
780 * @lock: Request queue spin lock
783 * If a block device wishes to use the standard request handling procedures,
784 * which sorts requests and coalesces adjacent requests, then it must
785 * call blk_init_queue(). The function @rfn will be called when there
786 * are requests on the queue that need to be processed. If the device
787 * supports plugging, then @rfn may not be called immediately when requests
788 * are available on the queue, but may be called at some time later instead.
789 * Plugged queues are generally unplugged when a buffer belonging to one
790 * of the requests on the queue is needed, or due to memory pressure.
792 * @rfn is not required, or even expected, to remove all requests off the
793 * queue, but only as many as it can handle at a time. If it does leave
794 * requests on the queue, it is responsible for arranging that the requests
795 * get dealt with eventually.
797 * The queue spin lock must be held while manipulating the requests on the
798 * request queue; this lock will be taken also from interrupt context, so irq
799 * disabling is needed for it.
801 * Function returns a pointer to the initialized request queue, or %NULL if
805 * blk_init_queue() must be paired with a blk_cleanup_queue() call
806 * when the block device is deactivated (such as at module unload).
809 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
811 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
813 EXPORT_SYMBOL(blk_init_queue
);
815 struct request_queue
*
816 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
818 struct request_queue
*uninit_q
, *q
;
820 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
824 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
826 blk_cleanup_queue(uninit_q
);
830 EXPORT_SYMBOL(blk_init_queue_node
);
832 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
834 struct request_queue
*
835 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
841 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
845 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
849 q
->prep_rq_fn
= NULL
;
850 q
->unprep_rq_fn
= NULL
;
851 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
853 /* Override internal queue lock with supplied lock pointer */
855 q
->queue_lock
= lock
;
858 * This also sets hw/phys segments, boundary and size
860 blk_queue_make_request(q
, blk_queue_bio
);
862 q
->sg_reserved_size
= INT_MAX
;
864 /* Protect q->elevator from elevator_change */
865 mutex_lock(&q
->sysfs_lock
);
868 if (elevator_init(q
, NULL
)) {
869 mutex_unlock(&q
->sysfs_lock
);
873 mutex_unlock(&q
->sysfs_lock
);
878 blk_free_flush_queue(q
->fq
);
881 EXPORT_SYMBOL(blk_init_allocated_queue
);
883 bool blk_get_queue(struct request_queue
*q
)
885 if (likely(!blk_queue_dying(q
))) {
892 EXPORT_SYMBOL(blk_get_queue
);
894 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
896 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
897 elv_put_request(rl
->q
, rq
);
899 put_io_context(rq
->elv
.icq
->ioc
);
902 mempool_free(rq
, rl
->rq_pool
);
906 * ioc_batching returns true if the ioc is a valid batching request and
907 * should be given priority access to a request.
909 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
915 * Make sure the process is able to allocate at least 1 request
916 * even if the batch times out, otherwise we could theoretically
919 return ioc
->nr_batch_requests
== q
->nr_batching
||
920 (ioc
->nr_batch_requests
> 0
921 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
925 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
926 * will cause the process to be a "batcher" on all queues in the system. This
927 * is the behaviour we want though - once it gets a wakeup it should be given
930 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
932 if (!ioc
|| ioc_batching(q
, ioc
))
935 ioc
->nr_batch_requests
= q
->nr_batching
;
936 ioc
->last_waited
= jiffies
;
939 static void __freed_request(struct request_list
*rl
, int sync
)
941 struct request_queue
*q
= rl
->q
;
943 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
944 blk_clear_congested(rl
, sync
);
946 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
947 if (waitqueue_active(&rl
->wait
[sync
]))
948 wake_up(&rl
->wait
[sync
]);
950 blk_clear_rl_full(rl
, sync
);
955 * A request has just been released. Account for it, update the full and
956 * congestion status, wake up any waiters. Called under q->queue_lock.
958 static void freed_request(struct request_list
*rl
, unsigned int flags
)
960 struct request_queue
*q
= rl
->q
;
961 int sync
= rw_is_sync(flags
);
965 if (flags
& REQ_ELVPRIV
)
968 __freed_request(rl
, sync
);
970 if (unlikely(rl
->starved
[sync
^ 1]))
971 __freed_request(rl
, sync
^ 1);
974 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
976 struct request_list
*rl
;
977 int on_thresh
, off_thresh
;
979 spin_lock_irq(q
->queue_lock
);
981 blk_queue_congestion_threshold(q
);
982 on_thresh
= queue_congestion_on_threshold(q
);
983 off_thresh
= queue_congestion_off_threshold(q
);
985 blk_queue_for_each_rl(rl
, q
) {
986 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
987 blk_set_congested(rl
, BLK_RW_SYNC
);
988 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
989 blk_clear_congested(rl
, BLK_RW_SYNC
);
991 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
992 blk_set_congested(rl
, BLK_RW_ASYNC
);
993 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
994 blk_clear_congested(rl
, BLK_RW_ASYNC
);
996 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
997 blk_set_rl_full(rl
, BLK_RW_SYNC
);
999 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1000 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1003 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1004 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1006 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1007 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1011 spin_unlock_irq(q
->queue_lock
);
1016 * Determine if elevator data should be initialized when allocating the
1017 * request associated with @bio.
1019 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1025 * Flush requests do not use the elevator so skip initialization.
1026 * This allows a request to share the flush and elevator data.
1028 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
1035 * rq_ioc - determine io_context for request allocation
1036 * @bio: request being allocated is for this bio (can be %NULL)
1038 * Determine io_context to use for request allocation for @bio. May return
1039 * %NULL if %current->io_context doesn't exist.
1041 static struct io_context
*rq_ioc(struct bio
*bio
)
1043 #ifdef CONFIG_BLK_CGROUP
1044 if (bio
&& bio
->bi_ioc
)
1047 return current
->io_context
;
1051 * __get_request - get a free request
1052 * @rl: request list to allocate from
1053 * @rw_flags: RW and SYNC flags
1054 * @bio: bio to allocate request for (can be %NULL)
1055 * @gfp_mask: allocation mask
1057 * Get a free request from @q. This function may fail under memory
1058 * pressure or if @q is dead.
1060 * Must be called with @q->queue_lock held and,
1061 * Returns ERR_PTR on failure, with @q->queue_lock held.
1062 * Returns request pointer on success, with @q->queue_lock *not held*.
1064 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
1065 struct bio
*bio
, gfp_t gfp_mask
)
1067 struct request_queue
*q
= rl
->q
;
1069 struct elevator_type
*et
= q
->elevator
->type
;
1070 struct io_context
*ioc
= rq_ioc(bio
);
1071 struct io_cq
*icq
= NULL
;
1072 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1075 if (unlikely(blk_queue_dying(q
)))
1076 return ERR_PTR(-ENODEV
);
1078 may_queue
= elv_may_queue(q
, rw_flags
);
1079 if (may_queue
== ELV_MQUEUE_NO
)
1082 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1083 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1085 * The queue will fill after this allocation, so set
1086 * it as full, and mark this process as "batching".
1087 * This process will be allowed to complete a batch of
1088 * requests, others will be blocked.
1090 if (!blk_rl_full(rl
, is_sync
)) {
1091 ioc_set_batching(q
, ioc
);
1092 blk_set_rl_full(rl
, is_sync
);
1094 if (may_queue
!= ELV_MQUEUE_MUST
1095 && !ioc_batching(q
, ioc
)) {
1097 * The queue is full and the allocating
1098 * process is not a "batcher", and not
1099 * exempted by the IO scheduler
1101 return ERR_PTR(-ENOMEM
);
1105 blk_set_congested(rl
, is_sync
);
1109 * Only allow batching queuers to allocate up to 50% over the defined
1110 * limit of requests, otherwise we could have thousands of requests
1111 * allocated with any setting of ->nr_requests
1113 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1114 return ERR_PTR(-ENOMEM
);
1116 q
->nr_rqs
[is_sync
]++;
1117 rl
->count
[is_sync
]++;
1118 rl
->starved
[is_sync
] = 0;
1121 * Decide whether the new request will be managed by elevator. If
1122 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1123 * prevent the current elevator from being destroyed until the new
1124 * request is freed. This guarantees icq's won't be destroyed and
1125 * makes creating new ones safe.
1127 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1128 * it will be created after releasing queue_lock.
1130 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1131 rw_flags
|= REQ_ELVPRIV
;
1132 q
->nr_rqs_elvpriv
++;
1133 if (et
->icq_cache
&& ioc
)
1134 icq
= ioc_lookup_icq(ioc
, q
);
1137 if (blk_queue_io_stat(q
))
1138 rw_flags
|= REQ_IO_STAT
;
1139 spin_unlock_irq(q
->queue_lock
);
1141 /* allocate and init request */
1142 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1147 blk_rq_set_rl(rq
, rl
);
1148 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
1151 if (rw_flags
& REQ_ELVPRIV
) {
1152 if (unlikely(et
->icq_cache
&& !icq
)) {
1154 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1160 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1163 /* @rq->elv.icq holds io_context until @rq is freed */
1165 get_io_context(icq
->ioc
);
1169 * ioc may be NULL here, and ioc_batching will be false. That's
1170 * OK, if the queue is under the request limit then requests need
1171 * not count toward the nr_batch_requests limit. There will always
1172 * be some limit enforced by BLK_BATCH_TIME.
1174 if (ioc_batching(q
, ioc
))
1175 ioc
->nr_batch_requests
--;
1177 trace_block_getrq(q
, bio
, rw_flags
& 1);
1182 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1183 * and may fail indefinitely under memory pressure and thus
1184 * shouldn't stall IO. Treat this request as !elvpriv. This will
1185 * disturb iosched and blkcg but weird is bettern than dead.
1187 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1188 __func__
, dev_name(q
->backing_dev_info
.dev
));
1190 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1193 spin_lock_irq(q
->queue_lock
);
1194 q
->nr_rqs_elvpriv
--;
1195 spin_unlock_irq(q
->queue_lock
);
1200 * Allocation failed presumably due to memory. Undo anything we
1201 * might have messed up.
1203 * Allocating task should really be put onto the front of the wait
1204 * queue, but this is pretty rare.
1206 spin_lock_irq(q
->queue_lock
);
1207 freed_request(rl
, rw_flags
);
1210 * in the very unlikely event that allocation failed and no
1211 * requests for this direction was pending, mark us starved so that
1212 * freeing of a request in the other direction will notice
1213 * us. another possible fix would be to split the rq mempool into
1217 if (unlikely(rl
->count
[is_sync
] == 0))
1218 rl
->starved
[is_sync
] = 1;
1219 return ERR_PTR(-ENOMEM
);
1223 * get_request - get a free request
1224 * @q: request_queue to allocate request from
1225 * @rw_flags: RW and SYNC flags
1226 * @bio: bio to allocate request for (can be %NULL)
1227 * @gfp_mask: allocation mask
1229 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1230 * this function keeps retrying under memory pressure and fails iff @q is dead.
1232 * Must be called with @q->queue_lock held and,
1233 * Returns ERR_PTR on failure, with @q->queue_lock held.
1234 * Returns request pointer on success, with @q->queue_lock *not held*.
1236 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1237 struct bio
*bio
, gfp_t gfp_mask
)
1239 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1241 struct request_list
*rl
;
1244 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1246 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1250 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1255 /* wait on @rl and retry */
1256 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1257 TASK_UNINTERRUPTIBLE
);
1259 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1261 spin_unlock_irq(q
->queue_lock
);
1265 * After sleeping, we become a "batching" process and will be able
1266 * to allocate at least one request, and up to a big batch of them
1267 * for a small period time. See ioc_batching, ioc_set_batching
1269 ioc_set_batching(q
, current
->io_context
);
1271 spin_lock_irq(q
->queue_lock
);
1272 finish_wait(&rl
->wait
[is_sync
], &wait
);
1277 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1282 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1284 /* create ioc upfront */
1285 create_io_context(gfp_mask
, q
->node
);
1287 spin_lock_irq(q
->queue_lock
);
1288 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1290 spin_unlock_irq(q
->queue_lock
);
1291 /* q->queue_lock is unlocked at this point */
1296 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1299 return blk_mq_alloc_request(q
, rw
, gfp_mask
, false);
1301 return blk_old_get_request(q
, rw
, gfp_mask
);
1303 EXPORT_SYMBOL(blk_get_request
);
1306 * blk_make_request - given a bio, allocate a corresponding struct request.
1307 * @q: target request queue
1308 * @bio: The bio describing the memory mappings that will be submitted for IO.
1309 * It may be a chained-bio properly constructed by block/bio layer.
1310 * @gfp_mask: gfp flags to be used for memory allocation
1312 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1313 * type commands. Where the struct request needs to be farther initialized by
1314 * the caller. It is passed a &struct bio, which describes the memory info of
1317 * The caller of blk_make_request must make sure that bi_io_vec
1318 * are set to describe the memory buffers. That bio_data_dir() will return
1319 * the needed direction of the request. (And all bio's in the passed bio-chain
1320 * are properly set accordingly)
1322 * If called under none-sleepable conditions, mapped bio buffers must not
1323 * need bouncing, by calling the appropriate masked or flagged allocator,
1324 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1327 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1328 * given to how you allocate bios. In particular, you cannot use
1329 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1330 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1331 * thus resulting in a deadlock. Alternatively bios should be allocated using
1332 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1333 * If possible a big IO should be split into smaller parts when allocation
1334 * fails. Partial allocation should not be an error, or you risk a live-lock.
1336 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1339 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1344 blk_rq_set_block_pc(rq
);
1347 struct bio
*bounce_bio
= bio
;
1350 blk_queue_bounce(q
, &bounce_bio
);
1351 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1352 if (unlikely(ret
)) {
1353 blk_put_request(rq
);
1354 return ERR_PTR(ret
);
1360 EXPORT_SYMBOL(blk_make_request
);
1363 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1364 * @rq: request to be initialized
1367 void blk_rq_set_block_pc(struct request
*rq
)
1369 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1371 rq
->__sector
= (sector_t
) -1;
1372 rq
->bio
= rq
->biotail
= NULL
;
1373 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1375 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1378 * blk_requeue_request - put a request back on queue
1379 * @q: request queue where request should be inserted
1380 * @rq: request to be inserted
1383 * Drivers often keep queueing requests until the hardware cannot accept
1384 * more, when that condition happens we need to put the request back
1385 * on the queue. Must be called with queue lock held.
1387 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1389 blk_delete_timer(rq
);
1390 blk_clear_rq_complete(rq
);
1391 trace_block_rq_requeue(q
, rq
);
1393 if (rq
->cmd_flags
& REQ_QUEUED
)
1394 blk_queue_end_tag(q
, rq
);
1396 BUG_ON(blk_queued_rq(rq
));
1398 elv_requeue_request(q
, rq
);
1400 EXPORT_SYMBOL(blk_requeue_request
);
1402 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1405 blk_account_io_start(rq
, true);
1406 __elv_add_request(q
, rq
, where
);
1409 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1414 if (now
== part
->stamp
)
1417 inflight
= part_in_flight(part
);
1419 __part_stat_add(cpu
, part
, time_in_queue
,
1420 inflight
* (now
- part
->stamp
));
1421 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1427 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1428 * @cpu: cpu number for stats access
1429 * @part: target partition
1431 * The average IO queue length and utilisation statistics are maintained
1432 * by observing the current state of the queue length and the amount of
1433 * time it has been in this state for.
1435 * Normally, that accounting is done on IO completion, but that can result
1436 * in more than a second's worth of IO being accounted for within any one
1437 * second, leading to >100% utilisation. To deal with that, we call this
1438 * function to do a round-off before returning the results when reading
1439 * /proc/diskstats. This accounts immediately for all queue usage up to
1440 * the current jiffies and restarts the counters again.
1442 void part_round_stats(int cpu
, struct hd_struct
*part
)
1444 unsigned long now
= jiffies
;
1447 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1448 part_round_stats_single(cpu
, part
, now
);
1450 EXPORT_SYMBOL_GPL(part_round_stats
);
1453 static void blk_pm_put_request(struct request
*rq
)
1455 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1456 pm_runtime_mark_last_busy(rq
->q
->dev
);
1459 static inline void blk_pm_put_request(struct request
*rq
) {}
1463 * queue lock must be held
1465 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1471 blk_mq_free_request(req
);
1475 blk_pm_put_request(req
);
1477 elv_completed_request(q
, req
);
1479 /* this is a bio leak */
1480 WARN_ON(req
->bio
!= NULL
);
1483 * Request may not have originated from ll_rw_blk. if not,
1484 * it didn't come out of our reserved rq pools
1486 if (req
->cmd_flags
& REQ_ALLOCED
) {
1487 unsigned int flags
= req
->cmd_flags
;
1488 struct request_list
*rl
= blk_rq_rl(req
);
1490 BUG_ON(!list_empty(&req
->queuelist
));
1491 BUG_ON(ELV_ON_HASH(req
));
1493 blk_free_request(rl
, req
);
1494 freed_request(rl
, flags
);
1498 EXPORT_SYMBOL_GPL(__blk_put_request
);
1500 void blk_put_request(struct request
*req
)
1502 struct request_queue
*q
= req
->q
;
1505 blk_mq_free_request(req
);
1507 unsigned long flags
;
1509 spin_lock_irqsave(q
->queue_lock
, flags
);
1510 __blk_put_request(q
, req
);
1511 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1514 EXPORT_SYMBOL(blk_put_request
);
1517 * blk_add_request_payload - add a payload to a request
1518 * @rq: request to update
1519 * @page: page backing the payload
1520 * @len: length of the payload.
1522 * This allows to later add a payload to an already submitted request by
1523 * a block driver. The driver needs to take care of freeing the payload
1526 * Note that this is a quite horrible hack and nothing but handling of
1527 * discard requests should ever use it.
1529 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1532 struct bio
*bio
= rq
->bio
;
1534 bio
->bi_io_vec
->bv_page
= page
;
1535 bio
->bi_io_vec
->bv_offset
= 0;
1536 bio
->bi_io_vec
->bv_len
= len
;
1538 bio
->bi_iter
.bi_size
= len
;
1540 bio
->bi_phys_segments
= 1;
1542 rq
->__data_len
= rq
->resid_len
= len
;
1543 rq
->nr_phys_segments
= 1;
1545 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1547 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1550 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1552 if (!ll_back_merge_fn(q
, req
, bio
))
1555 trace_block_bio_backmerge(q
, req
, bio
);
1557 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1558 blk_rq_set_mixed_merge(req
);
1560 req
->biotail
->bi_next
= bio
;
1562 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1563 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1565 blk_account_io_start(req
, false);
1569 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1572 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1574 if (!ll_front_merge_fn(q
, req
, bio
))
1577 trace_block_bio_frontmerge(q
, req
, bio
);
1579 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1580 blk_rq_set_mixed_merge(req
);
1582 bio
->bi_next
= req
->bio
;
1585 req
->__sector
= bio
->bi_iter
.bi_sector
;
1586 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1587 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1589 blk_account_io_start(req
, false);
1594 * blk_attempt_plug_merge - try to merge with %current's plugged list
1595 * @q: request_queue new bio is being queued at
1596 * @bio: new bio being queued
1597 * @request_count: out parameter for number of traversed plugged requests
1598 * @same_queue_rq: pointer to &struct request that gets filled in when
1599 * another request associated with @q is found on the plug list
1600 * (optional, may be %NULL)
1602 * Determine whether @bio being queued on @q can be merged with a request
1603 * on %current's plugged list. Returns %true if merge was successful,
1606 * Plugging coalesces IOs from the same issuer for the same purpose without
1607 * going through @q->queue_lock. As such it's more of an issuing mechanism
1608 * than scheduling, and the request, while may have elvpriv data, is not
1609 * added on the elevator at this point. In addition, we don't have
1610 * reliable access to the elevator outside queue lock. Only check basic
1611 * merging parameters without querying the elevator.
1613 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1615 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1616 unsigned int *request_count
,
1617 struct request
**same_queue_rq
)
1619 struct blk_plug
*plug
;
1622 struct list_head
*plug_list
;
1624 plug
= current
->plug
;
1630 plug_list
= &plug
->mq_list
;
1632 plug_list
= &plug
->list
;
1634 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1640 * Only blk-mq multiple hardware queues case checks the
1641 * rq in the same queue, there should be only one such
1645 *same_queue_rq
= rq
;
1648 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1651 el_ret
= blk_try_merge(rq
, bio
);
1652 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1653 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1656 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1657 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1666 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1668 struct blk_plug
*plug
;
1670 struct list_head
*plug_list
;
1671 unsigned int ret
= 0;
1673 plug
= current
->plug
;
1678 plug_list
= &plug
->mq_list
;
1680 plug_list
= &plug
->list
;
1682 list_for_each_entry(rq
, plug_list
, queuelist
) {
1690 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1692 req
->cmd_type
= REQ_TYPE_FS
;
1694 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1695 if (bio
->bi_rw
& REQ_RAHEAD
)
1696 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1698 #ifdef CONFIG_JOURNAL_DATA_TAG
1699 if (bio_flagged(bio
, BIO_JMETA
) || bio_flagged(bio
, BIO_JOURNAL
))
1700 req
->cmd_flags
|= REQ_META
;
1704 req
->__sector
= bio
->bi_iter
.bi_sector
;
1705 req
->ioprio
= bio_prio(bio
);
1706 blk_rq_bio_prep(req
->q
, req
, bio
);
1709 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1711 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1712 struct blk_plug
*plug
;
1713 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1714 struct request
*req
;
1715 unsigned int request_count
= 0;
1718 * low level driver can indicate that it wants pages above a
1719 * certain limit bounced to low memory (ie for highmem, or even
1720 * ISA dma in theory)
1722 blk_queue_bounce(q
, &bio
);
1724 blk_queue_split(q
, &bio
, q
->bio_split
);
1726 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1727 bio
->bi_error
= -EIO
;
1729 return BLK_QC_T_NONE
;
1732 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1733 spin_lock_irq(q
->queue_lock
);
1734 where
= ELEVATOR_INSERT_FLUSH
;
1739 * Check if we can merge with the plugged list before grabbing
1742 if (!blk_queue_nomerges(q
)) {
1743 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1744 return BLK_QC_T_NONE
;
1746 request_count
= blk_plug_queued_count(q
);
1748 spin_lock_irq(q
->queue_lock
);
1750 el_ret
= elv_merge(q
, &req
, bio
);
1751 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1752 if (bio_attempt_back_merge(q
, req
, bio
)) {
1753 elv_bio_merged(q
, req
, bio
);
1754 if (!attempt_back_merge(q
, req
))
1755 elv_merged_request(q
, req
, el_ret
);
1758 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1759 if (bio_attempt_front_merge(q
, req
, bio
)) {
1760 elv_bio_merged(q
, req
, bio
);
1761 if (!attempt_front_merge(q
, req
))
1762 elv_merged_request(q
, req
, el_ret
);
1769 * This sync check and mask will be re-done in init_request_from_bio(),
1770 * but we need to set it earlier to expose the sync flag to the
1771 * rq allocator and io schedulers.
1773 rw_flags
= bio_data_dir(bio
);
1775 rw_flags
|= REQ_SYNC
;
1778 * Grab a free request. This is might sleep but can not fail.
1779 * Returns with the queue unlocked.
1781 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1783 bio
->bi_error
= PTR_ERR(req
);
1789 * After dropping the lock and possibly sleeping here, our request
1790 * may now be mergeable after it had proven unmergeable (above).
1791 * We don't worry about that case for efficiency. It won't happen
1792 * often, and the elevators are able to handle it.
1794 init_request_from_bio(req
, bio
);
1796 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1797 req
->cpu
= raw_smp_processor_id();
1799 plug
= current
->plug
;
1802 * If this is the first request added after a plug, fire
1806 trace_block_plug(q
);
1808 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1809 blk_flush_plug_list(plug
, false);
1810 trace_block_plug(q
);
1813 list_add_tail(&req
->queuelist
, &plug
->list
);
1814 blk_account_io_start(req
, true);
1816 spin_lock_irq(q
->queue_lock
);
1817 add_acct_request(q
, req
, where
);
1820 spin_unlock_irq(q
->queue_lock
);
1823 return BLK_QC_T_NONE
;
1827 * If bio->bi_dev is a partition, remap the location
1829 static inline void blk_partition_remap(struct bio
*bio
)
1831 struct block_device
*bdev
= bio
->bi_bdev
;
1833 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1834 struct hd_struct
*p
= bdev
->bd_part
;
1836 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1837 bio
->bi_bdev
= bdev
->bd_contains
;
1839 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1841 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1845 static void handle_bad_sector(struct bio
*bio
)
1847 char b
[BDEVNAME_SIZE
];
1849 printk(KERN_INFO
"attempt to access beyond end of device\n");
1850 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1851 bdevname(bio
->bi_bdev
, b
),
1853 (unsigned long long)bio_end_sector(bio
),
1854 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1857 #ifdef CONFIG_FAIL_MAKE_REQUEST
1859 static DECLARE_FAULT_ATTR(fail_make_request
);
1861 static int __init
setup_fail_make_request(char *str
)
1863 return setup_fault_attr(&fail_make_request
, str
);
1865 __setup("fail_make_request=", setup_fail_make_request
);
1867 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1869 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1872 static int __init
fail_make_request_debugfs(void)
1874 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1875 NULL
, &fail_make_request
);
1877 return PTR_ERR_OR_ZERO(dir
);
1880 late_initcall(fail_make_request_debugfs
);
1882 #else /* CONFIG_FAIL_MAKE_REQUEST */
1884 static inline bool should_fail_request(struct hd_struct
*part
,
1890 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1893 * Check whether this bio extends beyond the end of the device.
1895 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1902 /* Test device or partition size, when known. */
1903 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1905 sector_t sector
= bio
->bi_iter
.bi_sector
;
1907 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1909 * This may well happen - the kernel calls bread()
1910 * without checking the size of the device, e.g., when
1911 * mounting a device.
1913 handle_bad_sector(bio
);
1921 static noinline_for_stack
bool
1922 generic_make_request_checks(struct bio
*bio
)
1924 struct request_queue
*q
;
1925 int nr_sectors
= bio_sectors(bio
);
1927 char b
[BDEVNAME_SIZE
];
1928 struct hd_struct
*part
;
1932 if (bio_check_eod(bio
, nr_sectors
))
1935 q
= bdev_get_queue(bio
->bi_bdev
);
1938 "generic_make_request: Trying to access "
1939 "nonexistent block-device %s (%Lu)\n",
1940 bdevname(bio
->bi_bdev
, b
),
1941 (long long) bio
->bi_iter
.bi_sector
);
1945 part
= bio
->bi_bdev
->bd_part
;
1946 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1947 should_fail_request(&part_to_disk(part
)->part0
,
1948 bio
->bi_iter
.bi_size
))
1952 * If this device has partitions, remap block n
1953 * of partition p to block n+start(p) of the disk.
1955 blk_partition_remap(bio
);
1957 if (bio_check_eod(bio
, nr_sectors
))
1961 * Filter flush bio's early so that make_request based
1962 * drivers without flush support don't have to worry
1965 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1966 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1973 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1974 (!blk_queue_discard(q
) ||
1975 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1980 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1986 * Various block parts want %current->io_context and lazy ioc
1987 * allocation ends up trading a lot of pain for a small amount of
1988 * memory. Just allocate it upfront. This may fail and block
1989 * layer knows how to live with it.
1991 create_io_context(GFP_ATOMIC
, q
->node
);
1993 if (!blkcg_bio_issue_check(q
, bio
))
1996 trace_block_bio_queue(q
, bio
);
2000 bio
->bi_error
= err
;
2006 * generic_make_request - hand a buffer to its device driver for I/O
2007 * @bio: The bio describing the location in memory and on the device.
2009 * generic_make_request() is used to make I/O requests of block
2010 * devices. It is passed a &struct bio, which describes the I/O that needs
2013 * generic_make_request() does not return any status. The
2014 * success/failure status of the request, along with notification of
2015 * completion, is delivered asynchronously through the bio->bi_end_io
2016 * function described (one day) else where.
2018 * The caller of generic_make_request must make sure that bi_io_vec
2019 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2020 * set to describe the device address, and the
2021 * bi_end_io and optionally bi_private are set to describe how
2022 * completion notification should be signaled.
2024 * generic_make_request and the drivers it calls may use bi_next if this
2025 * bio happens to be merged with someone else, and may resubmit the bio to
2026 * a lower device by calling into generic_make_request recursively, which
2027 * means the bio should NOT be touched after the call to ->make_request_fn.
2029 blk_qc_t
generic_make_request(struct bio
*bio
)
2032 * bio_list_on_stack[0] contains bios submitted by the current
2034 * bio_list_on_stack[1] contains bios that were submitted before
2035 * the current make_request_fn, but that haven't been processed
2038 struct bio_list bio_list_on_stack
[2];
2039 blk_qc_t ret
= BLK_QC_T_NONE
;
2041 if (!generic_make_request_checks(bio
))
2045 * We only want one ->make_request_fn to be active at a time, else
2046 * stack usage with stacked devices could be a problem. So use
2047 * current->bio_list to keep a list of requests submited by a
2048 * make_request_fn function. current->bio_list is also used as a
2049 * flag to say if generic_make_request is currently active in this
2050 * task or not. If it is NULL, then no make_request is active. If
2051 * it is non-NULL, then a make_request is active, and new requests
2052 * should be added at the tail
2054 if (current
->bio_list
) {
2055 bio_list_add(¤t
->bio_list
[0], bio
);
2059 /* following loop may be a bit non-obvious, and so deserves some
2061 * Before entering the loop, bio->bi_next is NULL (as all callers
2062 * ensure that) so we have a list with a single bio.
2063 * We pretend that we have just taken it off a longer list, so
2064 * we assign bio_list to a pointer to the bio_list_on_stack,
2065 * thus initialising the bio_list of new bios to be
2066 * added. ->make_request() may indeed add some more bios
2067 * through a recursive call to generic_make_request. If it
2068 * did, we find a non-NULL value in bio_list and re-enter the loop
2069 * from the top. In this case we really did just take the bio
2070 * of the top of the list (no pretending) and so remove it from
2071 * bio_list, and call into ->make_request() again.
2073 BUG_ON(bio
->bi_next
);
2074 bio_list_init(&bio_list_on_stack
[0]);
2075 current
->bio_list
= bio_list_on_stack
;
2077 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2079 if (likely(blk_queue_enter(q
, __GFP_DIRECT_RECLAIM
) == 0)) {
2080 struct bio_list lower
, same
;
2082 /* Create a fresh bio_list for all subordinate requests */
2083 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2084 bio_list_init(&bio_list_on_stack
[0]);
2086 ret
= q
->make_request_fn(q
, bio
);
2089 /* sort new bios into those for a lower level
2090 * and those for the same level
2092 bio_list_init(&lower
);
2093 bio_list_init(&same
);
2094 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2095 if (q
== bdev_get_queue(bio
->bi_bdev
))
2096 bio_list_add(&same
, bio
);
2098 bio_list_add(&lower
, bio
);
2099 /* now assemble so we handle the lowest level first */
2100 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2101 bio_list_merge(&bio_list_on_stack
[0], &same
);
2102 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2106 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2108 current
->bio_list
= NULL
; /* deactivate */
2113 EXPORT_SYMBOL(generic_make_request
);
2116 * submit_bio - submit a bio to the block device layer for I/O
2117 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2118 * @bio: The &struct bio which describes the I/O
2120 * submit_bio() is very similar in purpose to generic_make_request(), and
2121 * uses that function to do most of the work. Both are fairly rough
2122 * interfaces; @bio must be presetup and ready for I/O.
2125 blk_qc_t
submit_bio(int rw
, struct bio
*bio
)
2130 * If it's a regular read/write or a barrier with data attached,
2131 * go through the normal accounting stuff before submission.
2133 if (bio_has_data(bio
)) {
2136 if (unlikely(rw
& REQ_WRITE_SAME
))
2137 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2139 count
= bio_sectors(bio
);
2142 count_vm_events(PGPGOUT
, count
);
2144 task_io_account_read(bio
->bi_iter
.bi_size
);
2145 count_vm_events(PGPGIN
, count
);
2148 if (unlikely(block_dump
)) {
2149 char b
[BDEVNAME_SIZE
];
2150 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2151 current
->comm
, task_pid_nr(current
),
2152 (rw
& WRITE
) ? "WRITE" : "READ",
2153 (unsigned long long)bio
->bi_iter
.bi_sector
,
2154 bdevname(bio
->bi_bdev
, b
),
2159 return generic_make_request(bio
);
2161 EXPORT_SYMBOL(submit_bio
);
2164 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2165 * for new the queue limits
2167 * @rq: the request being checked
2170 * @rq may have been made based on weaker limitations of upper-level queues
2171 * in request stacking drivers, and it may violate the limitation of @q.
2172 * Since the block layer and the underlying device driver trust @rq
2173 * after it is inserted to @q, it should be checked against @q before
2174 * the insertion using this generic function.
2176 * Request stacking drivers like request-based dm may change the queue
2177 * limits when retrying requests on other queues. Those requests need
2178 * to be checked against the new queue limits again during dispatch.
2180 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2183 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
2184 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2189 * queue's settings related to segment counting like q->bounce_pfn
2190 * may differ from that of other stacking queues.
2191 * Recalculate it to check the request correctly on this queue's
2194 blk_recalc_rq_segments(rq
);
2195 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2196 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2204 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2205 * @q: the queue to submit the request
2206 * @rq: the request being queued
2208 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2210 unsigned long flags
;
2211 int where
= ELEVATOR_INSERT_BACK
;
2213 if (blk_cloned_rq_check_limits(q
, rq
))
2217 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2221 if (blk_queue_io_stat(q
))
2222 blk_account_io_start(rq
, true);
2223 blk_mq_insert_request(rq
, false, true, false);
2227 spin_lock_irqsave(q
->queue_lock
, flags
);
2228 if (unlikely(blk_queue_dying(q
))) {
2229 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2234 * Submitting request must be dequeued before calling this function
2235 * because it will be linked to another request_queue
2237 BUG_ON(blk_queued_rq(rq
));
2239 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
2240 where
= ELEVATOR_INSERT_FLUSH
;
2242 add_acct_request(q
, rq
, where
);
2243 if (where
== ELEVATOR_INSERT_FLUSH
)
2245 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2249 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2252 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2253 * @rq: request to examine
2256 * A request could be merge of IOs which require different failure
2257 * handling. This function determines the number of bytes which
2258 * can be failed from the beginning of the request without
2259 * crossing into area which need to be retried further.
2262 * The number of bytes to fail.
2265 * queue_lock must be held.
2267 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2269 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2270 unsigned int bytes
= 0;
2273 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2274 return blk_rq_bytes(rq
);
2277 * Currently the only 'mixing' which can happen is between
2278 * different fastfail types. We can safely fail portions
2279 * which have all the failfast bits that the first one has -
2280 * the ones which are at least as eager to fail as the first
2283 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2284 if ((bio
->bi_rw
& ff
) != ff
)
2286 bytes
+= bio
->bi_iter
.bi_size
;
2289 /* this could lead to infinite loop */
2290 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2293 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2295 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2297 if (blk_do_io_stat(req
)) {
2298 const int rw
= rq_data_dir(req
);
2299 struct hd_struct
*part
;
2302 cpu
= part_stat_lock();
2304 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2305 if (req
->cmd_flags
& REQ_DISCARD
)
2306 part_stat_add(cpu
, part
, discard_sectors
, bytes
>> 9);
2311 void blk_account_io_done(struct request
*req
)
2314 * Account IO completion. flush_rq isn't accounted as a
2315 * normal IO on queueing nor completion. Accounting the
2316 * containing request is enough.
2318 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2319 unsigned long duration
= jiffies
- req
->start_time
;
2320 const int rw
= rq_data_dir(req
);
2321 struct hd_struct
*part
;
2324 cpu
= part_stat_lock();
2327 part_stat_inc(cpu
, part
, ios
[rw
]);
2328 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2329 part_round_stats(cpu
, part
);
2330 part_dec_in_flight(part
, rw
);
2331 if (req
->cmd_flags
& REQ_DISCARD
)
2332 part_stat_inc(cpu
, part
, discard_ios
);
2333 if (!(req
->cmd_flags
& REQ_STARTED
))
2334 part_stat_inc(cpu
, part
, flush_ios
);
2336 hd_struct_put(part
);
2340 if (req
->cmd_flags
& REQ_FLUSH_SEQ
)
2341 req
->q
->flush_ios
++;
2346 * Don't process normal requests when queue is suspended
2347 * or in the process of suspending/resuming
2349 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2352 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2353 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2359 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2366 void blk_account_io_start(struct request
*rq
, bool new_io
)
2368 struct hd_struct
*part
;
2369 int rw
= rq_data_dir(rq
);
2372 if (!blk_do_io_stat(rq
))
2375 cpu
= part_stat_lock();
2379 part_stat_inc(cpu
, part
, merges
[rw
]);
2381 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2382 if (!hd_struct_try_get(part
)) {
2384 * The partition is already being removed,
2385 * the request will be accounted on the disk only
2387 * We take a reference on disk->part0 although that
2388 * partition will never be deleted, so we can treat
2389 * it as any other partition.
2391 part
= &rq
->rq_disk
->part0
;
2392 hd_struct_get(part
);
2394 part_round_stats(cpu
, part
);
2395 part_inc_in_flight(part
, rw
);
2403 * blk_peek_request - peek at the top of a request queue
2404 * @q: request queue to peek at
2407 * Return the request at the top of @q. The returned request
2408 * should be started using blk_start_request() before LLD starts
2412 * Pointer to the request at the top of @q if available. Null
2416 * queue_lock must be held.
2418 struct request
*blk_peek_request(struct request_queue
*q
)
2423 while ((rq
= __elv_next_request(q
)) != NULL
) {
2425 rq
= blk_pm_peek_request(q
, rq
);
2429 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2431 * This is the first time the device driver
2432 * sees this request (possibly after
2433 * requeueing). Notify IO scheduler.
2435 if (rq
->cmd_flags
& REQ_SORTED
)
2436 elv_activate_rq(q
, rq
);
2439 * just mark as started even if we don't start
2440 * it, a request that has been delayed should
2441 * not be passed by new incoming requests
2443 rq
->cmd_flags
|= REQ_STARTED
;
2444 trace_block_rq_issue(q
, rq
);
2447 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2448 q
->end_sector
= rq_end_sector(rq
);
2449 q
->boundary_rq
= NULL
;
2452 if (rq
->cmd_flags
& REQ_DONTPREP
)
2455 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2457 * make sure space for the drain appears we
2458 * know we can do this because max_hw_segments
2459 * has been adjusted to be one fewer than the
2462 rq
->nr_phys_segments
++;
2468 ret
= q
->prep_rq_fn(q
, rq
);
2469 if (ret
== BLKPREP_OK
) {
2471 } else if (ret
== BLKPREP_DEFER
) {
2473 * the request may have been (partially) prepped.
2474 * we need to keep this request in the front to
2475 * avoid resource deadlock. REQ_STARTED will
2476 * prevent other fs requests from passing this one.
2478 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2479 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2481 * remove the space for the drain we added
2482 * so that we don't add it again
2484 --rq
->nr_phys_segments
;
2489 } else if (ret
== BLKPREP_KILL
) {
2490 rq
->cmd_flags
|= REQ_QUIET
;
2492 * Mark this request as started so we don't trigger
2493 * any debug logic in the end I/O path.
2495 blk_start_request(rq
);
2496 __blk_end_request_all(rq
, -EIO
);
2498 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2505 EXPORT_SYMBOL(blk_peek_request
);
2507 void blk_dequeue_request(struct request
*rq
)
2509 struct request_queue
*q
= rq
->q
;
2511 BUG_ON(list_empty(&rq
->queuelist
));
2512 BUG_ON(ELV_ON_HASH(rq
));
2514 list_del_init(&rq
->queuelist
);
2517 * the time frame between a request being removed from the lists
2518 * and to it is freed is accounted as io that is in progress at
2521 if (blk_account_rq(rq
)) {
2522 if (!queue_in_flight(q
))
2523 q
->in_flight_stamp
= ktime_get();
2524 q
->in_flight
[rq_is_sync(rq
)]++;
2525 set_io_start_time_ns(rq
);
2530 * blk_start_request - start request processing on the driver
2531 * @req: request to dequeue
2534 * Dequeue @req and start timeout timer on it. This hands off the
2535 * request to the driver.
2537 * Block internal functions which don't want to start timer should
2538 * call blk_dequeue_request().
2541 * queue_lock must be held.
2543 void blk_start_request(struct request
*req
)
2545 blk_dequeue_request(req
);
2548 * We are now handing the request to the hardware, initialize
2549 * resid_len to full count and add the timeout handler.
2551 req
->resid_len
= blk_rq_bytes(req
);
2552 if (unlikely(blk_bidi_rq(req
)))
2553 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2555 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2558 EXPORT_SYMBOL(blk_start_request
);
2561 * blk_fetch_request - fetch a request from a request queue
2562 * @q: request queue to fetch a request from
2565 * Return the request at the top of @q. The request is started on
2566 * return and LLD can start processing it immediately.
2569 * Pointer to the request at the top of @q if available. Null
2573 * queue_lock must be held.
2575 struct request
*blk_fetch_request(struct request_queue
*q
)
2579 rq
= blk_peek_request(q
);
2581 blk_start_request(rq
);
2584 EXPORT_SYMBOL(blk_fetch_request
);
2587 * blk_update_request - Special helper function for request stacking drivers
2588 * @req: the request being processed
2589 * @error: %0 for success, < %0 for error
2590 * @nr_bytes: number of bytes to complete @req
2593 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2594 * the request structure even if @req doesn't have leftover.
2595 * If @req has leftover, sets it up for the next range of segments.
2597 * This special helper function is only for request stacking drivers
2598 * (e.g. request-based dm) so that they can handle partial completion.
2599 * Actual device drivers should use blk_end_request instead.
2601 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2602 * %false return from this function.
2605 * %false - this request doesn't have any more data
2606 * %true - this request has more data
2608 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2612 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2618 * For fs requests, rq is just carrier of independent bio's
2619 * and each partial completion should be handled separately.
2620 * Reset per-request error on each partial completion.
2622 * TODO: tj: This is too subtle. It would be better to let
2623 * low level drivers do what they see fit.
2625 if (req
->cmd_type
== REQ_TYPE_FS
)
2628 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2629 !(req
->cmd_flags
& REQ_QUIET
)) {
2634 error_type
= "recoverable transport";
2637 error_type
= "critical target";
2640 error_type
= "critical nexus";
2643 error_type
= "timeout";
2646 error_type
= "critical space allocation";
2649 error_type
= "critical medium";
2656 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2657 __func__
, error_type
, req
->rq_disk
?
2658 req
->rq_disk
->disk_name
: "?",
2659 (unsigned long long)blk_rq_pos(req
));
2663 blk_account_io_completion(req
, nr_bytes
);
2667 struct bio
*bio
= req
->bio
;
2668 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2670 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2671 req
->bio
= bio
->bi_next
;
2673 req_bio_endio(req
, bio
, bio_bytes
, error
);
2675 total_bytes
+= bio_bytes
;
2676 nr_bytes
-= bio_bytes
;
2687 * Reset counters so that the request stacking driver
2688 * can find how many bytes remain in the request
2691 req
->__data_len
= 0;
2695 req
->__data_len
-= total_bytes
;
2697 /* update sector only for requests with clear definition of sector */
2698 if (req
->cmd_type
== REQ_TYPE_FS
)
2699 req
->__sector
+= total_bytes
>> 9;
2701 /* mixed attributes always follow the first bio */
2702 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2703 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2704 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2708 * If total number of sectors is less than the first segment
2709 * size, something has gone terribly wrong.
2711 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2712 blk_dump_rq_flags(req
, "request botched");
2713 req
->__data_len
= blk_rq_cur_bytes(req
);
2716 /* recalculate the number of segments */
2717 blk_recalc_rq_segments(req
);
2721 EXPORT_SYMBOL_GPL(blk_update_request
);
2723 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2724 unsigned int nr_bytes
,
2725 unsigned int bidi_bytes
)
2727 if (blk_update_request(rq
, error
, nr_bytes
))
2730 /* Bidi request must be completed as a whole */
2731 if (unlikely(blk_bidi_rq(rq
)) &&
2732 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2735 if (blk_queue_add_random(rq
->q
))
2736 add_disk_randomness(rq
->rq_disk
);
2742 * blk_unprep_request - unprepare a request
2745 * This function makes a request ready for complete resubmission (or
2746 * completion). It happens only after all error handling is complete,
2747 * so represents the appropriate moment to deallocate any resources
2748 * that were allocated to the request in the prep_rq_fn. The queue
2749 * lock is held when calling this.
2751 void blk_unprep_request(struct request
*req
)
2753 struct request_queue
*q
= req
->q
;
2755 req
->cmd_flags
&= ~REQ_DONTPREP
;
2756 if (q
->unprep_rq_fn
)
2757 q
->unprep_rq_fn(q
, req
);
2759 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2762 * queue lock must be held
2764 void blk_finish_request(struct request
*req
, int error
)
2766 if (req
->cmd_flags
& REQ_QUEUED
)
2767 blk_queue_end_tag(req
->q
, req
);
2769 BUG_ON(blk_queued_rq(req
));
2771 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2772 laptop_io_completion(&req
->q
->backing_dev_info
);
2774 blk_delete_timer(req
);
2776 if (req
->cmd_flags
& REQ_DONTPREP
)
2777 blk_unprep_request(req
);
2779 blk_account_io_done(req
);
2782 req
->end_io(req
, error
);
2784 if (blk_bidi_rq(req
))
2785 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2787 __blk_put_request(req
->q
, req
);
2790 EXPORT_SYMBOL(blk_finish_request
);
2793 * blk_end_bidi_request - Complete a bidi request
2794 * @rq: the request to complete
2795 * @error: %0 for success, < %0 for error
2796 * @nr_bytes: number of bytes to complete @rq
2797 * @bidi_bytes: number of bytes to complete @rq->next_rq
2800 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2801 * Drivers that supports bidi can safely call this member for any
2802 * type of request, bidi or uni. In the later case @bidi_bytes is
2806 * %false - we are done with this request
2807 * %true - still buffers pending for this request
2809 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2810 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2812 struct request_queue
*q
= rq
->q
;
2813 unsigned long flags
;
2815 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2818 spin_lock_irqsave(q
->queue_lock
, flags
);
2819 blk_finish_request(rq
, error
);
2820 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2826 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2827 * @rq: the request to complete
2828 * @error: %0 for success, < %0 for error
2829 * @nr_bytes: number of bytes to complete @rq
2830 * @bidi_bytes: number of bytes to complete @rq->next_rq
2833 * Identical to blk_end_bidi_request() except that queue lock is
2834 * assumed to be locked on entry and remains so on return.
2837 * %false - we are done with this request
2838 * %true - still buffers pending for this request
2840 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2841 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2843 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2846 blk_finish_request(rq
, error
);
2852 * blk_end_request - Helper function for drivers to complete the request.
2853 * @rq: the request being processed
2854 * @error: %0 for success, < %0 for error
2855 * @nr_bytes: number of bytes to complete
2858 * Ends I/O on a number of bytes attached to @rq.
2859 * If @rq has leftover, sets it up for the next range of segments.
2862 * %false - we are done with this request
2863 * %true - still buffers pending for this request
2865 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2867 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2869 EXPORT_SYMBOL(blk_end_request
);
2872 * blk_end_request_all - Helper function for drives to finish the request.
2873 * @rq: the request to finish
2874 * @error: %0 for success, < %0 for error
2877 * Completely finish @rq.
2879 void blk_end_request_all(struct request
*rq
, int error
)
2882 unsigned int bidi_bytes
= 0;
2884 if (unlikely(blk_bidi_rq(rq
)))
2885 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2887 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2890 EXPORT_SYMBOL(blk_end_request_all
);
2893 * blk_end_request_cur - Helper function to finish the current request chunk.
2894 * @rq: the request to finish the current chunk for
2895 * @error: %0 for success, < %0 for error
2898 * Complete the current consecutively mapped chunk from @rq.
2901 * %false - we are done with this request
2902 * %true - still buffers pending for this request
2904 bool blk_end_request_cur(struct request
*rq
, int error
)
2906 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2908 EXPORT_SYMBOL(blk_end_request_cur
);
2911 * blk_end_request_err - Finish a request till the next failure boundary.
2912 * @rq: the request to finish till the next failure boundary for
2913 * @error: must be negative errno
2916 * Complete @rq till the next failure boundary.
2919 * %false - we are done with this request
2920 * %true - still buffers pending for this request
2922 bool blk_end_request_err(struct request
*rq
, int error
)
2924 WARN_ON(error
>= 0);
2925 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2927 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2930 * __blk_end_request - Helper function for drivers to complete the request.
2931 * @rq: the request being processed
2932 * @error: %0 for success, < %0 for error
2933 * @nr_bytes: number of bytes to complete
2936 * Must be called with queue lock held unlike blk_end_request().
2939 * %false - we are done with this request
2940 * %true - still buffers pending for this request
2942 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2944 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2946 EXPORT_SYMBOL(__blk_end_request
);
2949 * __blk_end_request_all - Helper function for drives to finish the request.
2950 * @rq: the request to finish
2951 * @error: %0 for success, < %0 for error
2954 * Completely finish @rq. Must be called with queue lock held.
2956 void __blk_end_request_all(struct request
*rq
, int error
)
2959 unsigned int bidi_bytes
= 0;
2961 if (unlikely(blk_bidi_rq(rq
)))
2962 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2964 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2967 EXPORT_SYMBOL(__blk_end_request_all
);
2970 * __blk_end_request_cur - Helper function to finish the current request chunk.
2971 * @rq: the request to finish the current chunk for
2972 * @error: %0 for success, < %0 for error
2975 * Complete the current consecutively mapped chunk from @rq. Must
2976 * be called with queue lock held.
2979 * %false - we are done with this request
2980 * %true - still buffers pending for this request
2982 bool __blk_end_request_cur(struct request
*rq
, int error
)
2984 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2986 EXPORT_SYMBOL(__blk_end_request_cur
);
2989 * __blk_end_request_err - Finish a request till the next failure boundary.
2990 * @rq: the request to finish till the next failure boundary for
2991 * @error: must be negative errno
2994 * Complete @rq till the next failure boundary. Must be called
2995 * with queue lock held.
2998 * %false - we are done with this request
2999 * %true - still buffers pending for this request
3001 bool __blk_end_request_err(struct request
*rq
, int error
)
3003 WARN_ON(error
>= 0);
3004 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
3006 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
3008 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3011 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
3012 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
3014 if (bio_has_data(bio
))
3015 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3017 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3018 rq
->bio
= rq
->biotail
= bio
;
3021 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
3024 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3026 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3027 * @rq: the request to be flushed
3030 * Flush all pages in @rq.
3032 void rq_flush_dcache_pages(struct request
*rq
)
3034 struct req_iterator iter
;
3035 struct bio_vec bvec
;
3037 rq_for_each_segment(bvec
, rq
, iter
)
3038 flush_dcache_page(bvec
.bv_page
);
3040 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3044 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3045 * @q : the queue of the device being checked
3048 * Check if underlying low-level drivers of a device are busy.
3049 * If the drivers want to export their busy state, they must set own
3050 * exporting function using blk_queue_lld_busy() first.
3052 * Basically, this function is used only by request stacking drivers
3053 * to stop dispatching requests to underlying devices when underlying
3054 * devices are busy. This behavior helps more I/O merging on the queue
3055 * of the request stacking driver and prevents I/O throughput regression
3056 * on burst I/O load.
3059 * 0 - Not busy (The request stacking driver should dispatch request)
3060 * 1 - Busy (The request stacking driver should stop dispatching request)
3062 int blk_lld_busy(struct request_queue
*q
)
3065 return q
->lld_busy_fn(q
);
3069 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3072 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3073 * @rq: the clone request to be cleaned up
3076 * Free all bios in @rq for a cloned request.
3078 void blk_rq_unprep_clone(struct request
*rq
)
3082 while ((bio
= rq
->bio
) != NULL
) {
3083 rq
->bio
= bio
->bi_next
;
3088 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3091 * Copy attributes of the original request to the clone request.
3092 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3094 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3096 dst
->cpu
= src
->cpu
;
3097 dst
->cmd_flags
|= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
3098 dst
->cmd_type
= src
->cmd_type
;
3099 dst
->__sector
= blk_rq_pos(src
);
3100 dst
->__data_len
= blk_rq_bytes(src
);
3101 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3102 dst
->ioprio
= src
->ioprio
;
3103 dst
->extra_len
= src
->extra_len
;
3107 * blk_rq_prep_clone - Helper function to setup clone request
3108 * @rq: the request to be setup
3109 * @rq_src: original request to be cloned
3110 * @bs: bio_set that bios for clone are allocated from
3111 * @gfp_mask: memory allocation mask for bio
3112 * @bio_ctr: setup function to be called for each clone bio.
3113 * Returns %0 for success, non %0 for failure.
3114 * @data: private data to be passed to @bio_ctr
3117 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3118 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3119 * are not copied, and copying such parts is the caller's responsibility.
3120 * Also, pages which the original bios are pointing to are not copied
3121 * and the cloned bios just point same pages.
3122 * So cloned bios must be completed before original bios, which means
3123 * the caller must complete @rq before @rq_src.
3125 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3126 struct bio_set
*bs
, gfp_t gfp_mask
,
3127 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3130 struct bio
*bio
, *bio_src
;
3135 __rq_for_each_bio(bio_src
, rq_src
) {
3136 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3140 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3144 rq
->biotail
->bi_next
= bio
;
3147 rq
->bio
= rq
->biotail
= bio
;
3150 __blk_rq_prep_clone(rq
, rq_src
);
3157 blk_rq_unprep_clone(rq
);
3161 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3163 int kblockd_schedule_work(struct work_struct
*work
)
3165 return queue_work(kblockd_workqueue
, work
);
3167 EXPORT_SYMBOL(kblockd_schedule_work
);
3169 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3170 unsigned long delay
)
3172 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3174 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3176 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3177 unsigned long delay
)
3179 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3181 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3184 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3185 * @plug: The &struct blk_plug that needs to be initialized
3188 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3189 * pending I/O should the task end up blocking between blk_start_plug() and
3190 * blk_finish_plug(). This is important from a performance perspective, but
3191 * also ensures that we don't deadlock. For instance, if the task is blocking
3192 * for a memory allocation, memory reclaim could end up wanting to free a
3193 * page belonging to that request that is currently residing in our private
3194 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3195 * this kind of deadlock.
3197 void blk_start_plug(struct blk_plug
*plug
)
3199 struct task_struct
*tsk
= current
;
3202 * If this is a nested plug, don't actually assign it.
3207 INIT_LIST_HEAD(&plug
->list
);
3208 INIT_LIST_HEAD(&plug
->mq_list
);
3209 INIT_LIST_HEAD(&plug
->cb_list
);
3211 * Store ordering should not be needed here, since a potential
3212 * preempt will imply a full memory barrier
3216 EXPORT_SYMBOL(blk_start_plug
);
3218 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3220 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3221 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3223 return !(rqa
->q
< rqb
->q
||
3224 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3228 * If 'from_schedule' is true, then postpone the dispatch of requests
3229 * until a safe kblockd context. We due this to avoid accidental big
3230 * additional stack usage in driver dispatch, in places where the originally
3231 * plugger did not intend it.
3233 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3235 __releases(q
->queue_lock
)
3237 trace_block_unplug(q
, depth
, !from_schedule
);
3240 blk_run_queue_async(q
);
3243 spin_unlock(q
->queue_lock
);
3246 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3248 LIST_HEAD(callbacks
);
3250 while (!list_empty(&plug
->cb_list
)) {
3251 list_splice_init(&plug
->cb_list
, &callbacks
);
3253 while (!list_empty(&callbacks
)) {
3254 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3257 list_del(&cb
->list
);
3258 cb
->callback(cb
, from_schedule
);
3263 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3266 struct blk_plug
*plug
= current
->plug
;
3267 struct blk_plug_cb
*cb
;
3272 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3273 if (cb
->callback
== unplug
&& cb
->data
== data
)
3276 /* Not currently on the callback list */
3277 BUG_ON(size
< sizeof(*cb
));
3278 cb
= kzalloc(size
, GFP_ATOMIC
);
3281 cb
->callback
= unplug
;
3282 list_add(&cb
->list
, &plug
->cb_list
);
3286 EXPORT_SYMBOL(blk_check_plugged
);
3288 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3290 struct request_queue
*q
;
3291 unsigned long flags
;
3296 flush_plug_callbacks(plug
, from_schedule
);
3298 if (!list_empty(&plug
->mq_list
))
3299 blk_mq_flush_plug_list(plug
, from_schedule
);
3301 if (list_empty(&plug
->list
))
3304 list_splice_init(&plug
->list
, &list
);
3306 list_sort(NULL
, &list
, plug_rq_cmp
);
3312 * Save and disable interrupts here, to avoid doing it for every
3313 * queue lock we have to take.
3315 local_irq_save(flags
);
3316 while (!list_empty(&list
)) {
3317 rq
= list_entry_rq(list
.next
);
3318 list_del_init(&rq
->queuelist
);
3322 * This drops the queue lock
3325 queue_unplugged(q
, depth
, from_schedule
);
3328 spin_lock(q
->queue_lock
);
3332 * Short-circuit if @q is dead
3334 if (unlikely(blk_queue_dying(q
))) {
3335 __blk_end_request_all(rq
, -ENODEV
);
3340 * rq is already accounted, so use raw insert
3342 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3343 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3345 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3351 * This drops the queue lock
3354 queue_unplugged(q
, depth
, from_schedule
);
3356 local_irq_restore(flags
);
3359 void blk_finish_plug(struct blk_plug
*plug
)
3361 if (plug
!= current
->plug
)
3363 blk_flush_plug_list(plug
, false);
3365 current
->plug
= NULL
;
3367 EXPORT_SYMBOL(blk_finish_plug
);
3369 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
3371 struct blk_plug
*plug
;
3374 if (!q
->mq_ops
|| !q
->mq_ops
->poll
|| !blk_qc_t_valid(cookie
) ||
3375 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
3378 plug
= current
->plug
;
3380 blk_flush_plug_list(plug
, false);
3382 state
= current
->state
;
3383 while (!need_resched()) {
3384 unsigned int queue_num
= blk_qc_t_to_queue_num(cookie
);
3385 struct blk_mq_hw_ctx
*hctx
= q
->queue_hw_ctx
[queue_num
];
3388 hctx
->poll_invoked
++;
3390 ret
= q
->mq_ops
->poll(hctx
, blk_qc_t_to_tag(cookie
));
3392 hctx
->poll_success
++;
3393 set_current_state(TASK_RUNNING
);
3397 if (signal_pending_state(state
, current
))
3398 set_current_state(TASK_RUNNING
);
3400 if (current
->state
== TASK_RUNNING
)
3412 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3413 * @q: the queue of the device
3414 * @dev: the device the queue belongs to
3417 * Initialize runtime-PM-related fields for @q and start auto suspend for
3418 * @dev. Drivers that want to take advantage of request-based runtime PM
3419 * should call this function after @dev has been initialized, and its
3420 * request queue @q has been allocated, and runtime PM for it can not happen
3421 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3422 * cases, driver should call this function before any I/O has taken place.
3424 * This function takes care of setting up using auto suspend for the device,
3425 * the autosuspend delay is set to -1 to make runtime suspend impossible
3426 * until an updated value is either set by user or by driver. Drivers do
3427 * not need to touch other autosuspend settings.
3429 * The block layer runtime PM is request based, so only works for drivers
3430 * that use request as their IO unit instead of those directly use bio's.
3432 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3435 q
->rpm_status
= RPM_ACTIVE
;
3436 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3437 pm_runtime_use_autosuspend(q
->dev
);
3439 EXPORT_SYMBOL(blk_pm_runtime_init
);
3442 * blk_pre_runtime_suspend - Pre runtime suspend check
3443 * @q: the queue of the device
3446 * This function will check if runtime suspend is allowed for the device
3447 * by examining if there are any requests pending in the queue. If there
3448 * are requests pending, the device can not be runtime suspended; otherwise,
3449 * the queue's status will be updated to SUSPENDING and the driver can
3450 * proceed to suspend the device.
3452 * For the not allowed case, we mark last busy for the device so that
3453 * runtime PM core will try to autosuspend it some time later.
3455 * This function should be called near the start of the device's
3456 * runtime_suspend callback.
3459 * 0 - OK to runtime suspend the device
3460 * -EBUSY - Device should not be runtime suspended
3462 int blk_pre_runtime_suspend(struct request_queue
*q
)
3469 spin_lock_irq(q
->queue_lock
);
3470 if (q
->nr_pending
) {
3472 pm_runtime_mark_last_busy(q
->dev
);
3474 q
->rpm_status
= RPM_SUSPENDING
;
3476 spin_unlock_irq(q
->queue_lock
);
3479 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3482 * blk_post_runtime_suspend - Post runtime suspend processing
3483 * @q: the queue of the device
3484 * @err: return value of the device's runtime_suspend function
3487 * Update the queue's runtime status according to the return value of the
3488 * device's runtime suspend function and mark last busy for the device so
3489 * that PM core will try to auto suspend the device at a later time.
3491 * This function should be called near the end of the device's
3492 * runtime_suspend callback.
3494 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3499 spin_lock_irq(q
->queue_lock
);
3501 q
->rpm_status
= RPM_SUSPENDED
;
3503 q
->rpm_status
= RPM_ACTIVE
;
3504 pm_runtime_mark_last_busy(q
->dev
);
3506 spin_unlock_irq(q
->queue_lock
);
3508 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3511 * blk_pre_runtime_resume - Pre runtime resume processing
3512 * @q: the queue of the device
3515 * Update the queue's runtime status to RESUMING in preparation for the
3516 * runtime resume of the device.
3518 * This function should be called near the start of the device's
3519 * runtime_resume callback.
3521 void blk_pre_runtime_resume(struct request_queue
*q
)
3526 spin_lock_irq(q
->queue_lock
);
3527 q
->rpm_status
= RPM_RESUMING
;
3528 spin_unlock_irq(q
->queue_lock
);
3530 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3533 * blk_post_runtime_resume - Post runtime resume processing
3534 * @q: the queue of the device
3535 * @err: return value of the device's runtime_resume function
3538 * Update the queue's runtime status according to the return value of the
3539 * device's runtime_resume function. If it is successfully resumed, process
3540 * the requests that are queued into the device's queue when it is resuming
3541 * and then mark last busy and initiate autosuspend for it.
3543 * This function should be called near the end of the device's
3544 * runtime_resume callback.
3546 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3551 spin_lock_irq(q
->queue_lock
);
3553 q
->rpm_status
= RPM_ACTIVE
;
3555 pm_runtime_mark_last_busy(q
->dev
);
3556 pm_request_autosuspend(q
->dev
);
3558 q
->rpm_status
= RPM_SUSPENDED
;
3560 spin_unlock_irq(q
->queue_lock
);
3562 EXPORT_SYMBOL(blk_post_runtime_resume
);
3565 #if !defined(CONFIG_SAMSUNG_PRODUCT_SHIP)
3566 /*********************************
3568 **********************************/
3569 #ifdef CONFIG_DEBUG_FS
3570 #include <linux/debugfs.h>
3572 #define DBGFS_FUNC_DECL(name) \
3573 static int sio_open_##name(struct inode *inode, struct file *file) \
3575 return single_open(file, sio_show_##name, inode->i_private); \
3577 static const struct file_operations sio_fops_##name = { \
3578 .owner = THIS_MODULE, \
3579 .open = sio_open_##name, \
3580 .llseek = seq_lseek, \
3582 .release = single_release, \
3585 static int sio_show_patches(struct seq_file
*s
, void *p
)
3587 extern char *__start_sio_patches
;
3588 extern char *__stop_sio_patches
;
3589 char **p_version_str
;
3591 for (p_version_str
= &__start_sio_patches
; p_version_str
< &__stop_sio_patches
; ++p_version_str
)
3592 seq_printf(s
, "%s\n", *p_version_str
);
3597 static struct dentry
*sio_debugfs_root
;
3599 DBGFS_FUNC_DECL(patches
);
3601 SIO_PATCH_VERSION(SIO_patch_manager
, 1, 0, "");
3603 static int __init
sio_debugfs_init(void)
3605 if (!debugfs_initialized())
3608 sio_debugfs_root
= debugfs_create_dir("sio", NULL
);
3609 if (!sio_debugfs_root
)
3612 debugfs_create_file("patches", 0400, sio_debugfs_root
, NULL
, &sio_fops_patches
);
3617 static void __exit
sio_debugfs_exit(void)
3619 debugfs_remove_recursive(sio_debugfs_root
);
3622 static int __init
sio_debugfs_init(void)
3627 static void __exit
sio_debugfs_exit(void) { }
3631 int __init
blk_dev_init(void)
3633 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3634 FIELD_SIZEOF(struct request
, cmd_flags
));
3636 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3637 kblockd_workqueue
= alloc_workqueue("kblockd",
3638 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3639 if (!kblockd_workqueue
)
3640 panic("Failed to create kblockd\n");
3642 request_cachep
= kmem_cache_create("blkdev_requests",
3643 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3645 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3646 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3648 #if !defined(CONFIG_SAMSUNG_PRODUCT_SHIP)
3656 * Blk IO latency support. We want this to be as cheap as possible, so doing
3657 * this lockless (and avoiding atomics), a few off by a few errors in this
3658 * code is not harmful, and we don't want to do anything that is
3660 * TODO : If necessary, we can make the histograms per-cpu and aggregate
3661 * them when printing them out.
3664 blk_zero_latency_hist(struct io_latency_state
*s
)
3666 memset(s
->latency_y_axis_read
, 0,
3667 sizeof(s
->latency_y_axis_read
));
3668 memset(s
->latency_y_axis_write
, 0,
3669 sizeof(s
->latency_y_axis_write
));
3670 s
->latency_reads_elems
= 0;
3671 s
->latency_writes_elems
= 0;
3673 EXPORT_SYMBOL(blk_zero_latency_hist
);
3676 blk_latency_hist_show(struct io_latency_state
*s
, char *buf
)
3679 int bytes_written
= 0;
3680 u_int64_t num_elem
, elem
;
3683 num_elem
= s
->latency_reads_elems
;
3685 bytes_written
+= scnprintf(buf
+ bytes_written
,
3686 PAGE_SIZE
- bytes_written
,
3687 "IO svc_time Read Latency Histogram (n = %llu):\n",
3690 i
< ARRAY_SIZE(latency_x_axis_us
);
3692 elem
= s
->latency_y_axis_read
[i
];
3693 pct
= div64_u64(elem
* 100, num_elem
);
3694 bytes_written
+= scnprintf(buf
+ bytes_written
,
3695 PAGE_SIZE
- bytes_written
,
3696 "\t< %5lluus%15llu%15d%%\n",
3697 latency_x_axis_us
[i
],
3700 /* Last element in y-axis table is overflow */
3701 elem
= s
->latency_y_axis_read
[i
];
3702 pct
= div64_u64(elem
* 100, num_elem
);
3703 bytes_written
+= scnprintf(buf
+ bytes_written
,
3704 PAGE_SIZE
- bytes_written
,
3705 "\t> %5dms%15llu%15d%%\n", 10,
3708 num_elem
= s
->latency_writes_elems
;
3710 bytes_written
+= scnprintf(buf
+ bytes_written
,
3711 PAGE_SIZE
- bytes_written
,
3712 "IO svc_time Write Latency Histogram (n = %llu):\n",
3715 i
< ARRAY_SIZE(latency_x_axis_us
);
3717 elem
= s
->latency_y_axis_write
[i
];
3718 pct
= div64_u64(elem
* 100, num_elem
);
3719 bytes_written
+= scnprintf(buf
+ bytes_written
,
3720 PAGE_SIZE
- bytes_written
,
3721 "\t< %5lluus%15llu%15d%%\n",
3722 latency_x_axis_us
[i
],
3725 /* Last element in y-axis table is overflow */
3726 elem
= s
->latency_y_axis_write
[i
];
3727 pct
= div64_u64(elem
* 100, num_elem
);
3728 bytes_written
+= scnprintf(buf
+ bytes_written
,
3729 PAGE_SIZE
- bytes_written
,
3730 "\t> %5dms%15llu%15d%%\n", 10,
3733 return bytes_written
;
3735 EXPORT_SYMBOL(blk_latency_hist_show
);