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> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
11 * This handles all read/write requests to block devices
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/highmem.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
37 #include <scsi/scsi_cmnd.h>
39 static void blk_unplug_work(struct work_struct
*work
);
40 static void blk_unplug_timeout(unsigned long data
);
41 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
42 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
43 static int __make_request(request_queue_t
*q
, struct bio
*bio
);
44 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
47 * For the allocated request tables
49 static struct kmem_cache
*request_cachep
;
52 * For queue allocation
54 static struct kmem_cache
*requestq_cachep
;
57 * For io context allocations
59 static struct kmem_cache
*iocontext_cachep
;
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 unsigned long blk_max_low_pfn
, blk_max_pfn
;
68 EXPORT_SYMBOL(blk_max_low_pfn
);
69 EXPORT_SYMBOL(blk_max_pfn
);
71 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
73 /* Amount of time in which a process may batch requests */
74 #define BLK_BATCH_TIME (HZ/50UL)
76 /* Number of requests a "batching" process may submit */
77 #define BLK_BATCH_REQ 32
80 * Return the threshold (number of used requests) at which the queue is
81 * considered to be congested. It include a little hysteresis to keep the
82 * context switch rate down.
84 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
86 return q
->nr_congestion_on
;
90 * The threshold at which a queue is considered to be uncongested
92 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
94 return q
->nr_congestion_off
;
97 static void blk_queue_congestion_threshold(struct request_queue
*q
)
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
102 if (nr
> q
->nr_requests
)
104 q
->nr_congestion_on
= nr
;
106 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
109 q
->nr_congestion_off
= nr
;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
123 struct backing_dev_info
*ret
= NULL
;
124 request_queue_t
*q
= bdev_get_queue(bdev
);
127 ret
= &q
->backing_dev_info
;
130 EXPORT_SYMBOL(blk_get_backing_dev_info
);
132 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
135 q
->activity_data
= data
;
137 EXPORT_SYMBOL(blk_queue_activity_fn
);
140 * blk_queue_prep_rq - set a prepare_request function for queue
142 * @pfn: prepare_request function
144 * It's possible for a queue to register a prepare_request callback which
145 * is invoked before the request is handed to the request_fn. The goal of
146 * the function is to prepare a request for I/O, it can be used to build a
147 * cdb from the request data for instance.
150 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
155 EXPORT_SYMBOL(blk_queue_prep_rq
);
158 * blk_queue_merge_bvec - set a merge_bvec function for queue
160 * @mbfn: merge_bvec_fn
162 * Usually queues have static limitations on the max sectors or segments that
163 * we can put in a request. Stacking drivers may have some settings that
164 * are dynamic, and thus we have to query the queue whether it is ok to
165 * add a new bio_vec to a bio at a given offset or not. If the block device
166 * has such limitations, it needs to register a merge_bvec_fn to control
167 * the size of bio's sent to it. Note that a block device *must* allow a
168 * single page to be added to an empty bio. The block device driver may want
169 * to use the bio_split() function to deal with these bio's. By default
170 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
173 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
175 q
->merge_bvec_fn
= mbfn
;
178 EXPORT_SYMBOL(blk_queue_merge_bvec
);
180 void blk_queue_softirq_done(request_queue_t
*q
, softirq_done_fn
*fn
)
182 q
->softirq_done_fn
= fn
;
185 EXPORT_SYMBOL(blk_queue_softirq_done
);
188 * blk_queue_make_request - define an alternate make_request function for a device
189 * @q: the request queue for the device to be affected
190 * @mfn: the alternate make_request function
193 * The normal way for &struct bios to be passed to a device
194 * driver is for them to be collected into requests on a request
195 * queue, and then to allow the device driver to select requests
196 * off that queue when it is ready. This works well for many block
197 * devices. However some block devices (typically virtual devices
198 * such as md or lvm) do not benefit from the processing on the
199 * request queue, and are served best by having the requests passed
200 * directly to them. This can be achieved by providing a function
201 * to blk_queue_make_request().
204 * The driver that does this *must* be able to deal appropriately
205 * with buffers in "highmemory". This can be accomplished by either calling
206 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
207 * blk_queue_bounce() to create a buffer in normal memory.
209 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
214 q
->nr_requests
= BLKDEV_MAX_RQ
;
215 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
216 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
217 q
->make_request_fn
= mfn
;
218 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
219 q
->backing_dev_info
.state
= 0;
220 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
221 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
222 blk_queue_hardsect_size(q
, 512);
223 blk_queue_dma_alignment(q
, 511);
224 blk_queue_congestion_threshold(q
);
225 q
->nr_batching
= BLK_BATCH_REQ
;
227 q
->unplug_thresh
= 4; /* hmm */
228 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
229 if (q
->unplug_delay
== 0)
232 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
234 q
->unplug_timer
.function
= blk_unplug_timeout
;
235 q
->unplug_timer
.data
= (unsigned long)q
;
238 * by default assume old behaviour and bounce for any highmem page
240 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
242 blk_queue_activity_fn(q
, NULL
, NULL
);
245 EXPORT_SYMBOL(blk_queue_make_request
);
247 static void rq_init(request_queue_t
*q
, struct request
*rq
)
249 INIT_LIST_HEAD(&rq
->queuelist
);
250 INIT_LIST_HEAD(&rq
->donelist
);
253 rq
->bio
= rq
->biotail
= NULL
;
254 INIT_HLIST_NODE(&rq
->hash
);
255 RB_CLEAR_NODE(&rq
->rb_node
);
263 rq
->nr_phys_segments
= 0;
266 rq
->end_io_data
= NULL
;
267 rq
->completion_data
= NULL
;
271 * blk_queue_ordered - does this queue support ordered writes
272 * @q: the request queue
273 * @ordered: one of QUEUE_ORDERED_*
274 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
277 * For journalled file systems, doing ordered writes on a commit
278 * block instead of explicitly doing wait_on_buffer (which is bad
279 * for performance) can be a big win. Block drivers supporting this
280 * feature should call this function and indicate so.
283 int blk_queue_ordered(request_queue_t
*q
, unsigned ordered
,
284 prepare_flush_fn
*prepare_flush_fn
)
286 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
287 prepare_flush_fn
== NULL
) {
288 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
292 if (ordered
!= QUEUE_ORDERED_NONE
&&
293 ordered
!= QUEUE_ORDERED_DRAIN
&&
294 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
295 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
296 ordered
!= QUEUE_ORDERED_TAG
&&
297 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
298 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
299 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
303 q
->ordered
= ordered
;
304 q
->next_ordered
= ordered
;
305 q
->prepare_flush_fn
= prepare_flush_fn
;
310 EXPORT_SYMBOL(blk_queue_ordered
);
313 * blk_queue_issue_flush_fn - set function for issuing a flush
314 * @q: the request queue
315 * @iff: the function to be called issuing the flush
318 * If a driver supports issuing a flush command, the support is notified
319 * to the block layer by defining it through this call.
322 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
324 q
->issue_flush_fn
= iff
;
327 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
330 * Cache flushing for ordered writes handling
332 inline unsigned blk_ordered_cur_seq(request_queue_t
*q
)
336 return 1 << ffz(q
->ordseq
);
339 unsigned blk_ordered_req_seq(struct request
*rq
)
341 request_queue_t
*q
= rq
->q
;
343 BUG_ON(q
->ordseq
== 0);
345 if (rq
== &q
->pre_flush_rq
)
346 return QUEUE_ORDSEQ_PREFLUSH
;
347 if (rq
== &q
->bar_rq
)
348 return QUEUE_ORDSEQ_BAR
;
349 if (rq
== &q
->post_flush_rq
)
350 return QUEUE_ORDSEQ_POSTFLUSH
;
352 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
353 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
354 return QUEUE_ORDSEQ_DRAIN
;
356 return QUEUE_ORDSEQ_DONE
;
359 void blk_ordered_complete_seq(request_queue_t
*q
, unsigned seq
, int error
)
364 if (error
&& !q
->orderr
)
367 BUG_ON(q
->ordseq
& seq
);
370 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
374 * Okay, sequence complete.
377 uptodate
= q
->orderr
? q
->orderr
: 1;
381 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
382 end_that_request_last(rq
, uptodate
);
385 static void pre_flush_end_io(struct request
*rq
, int error
)
387 elv_completed_request(rq
->q
, rq
);
388 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
391 static void bar_end_io(struct request
*rq
, int error
)
393 elv_completed_request(rq
->q
, rq
);
394 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
397 static void post_flush_end_io(struct request
*rq
, int error
)
399 elv_completed_request(rq
->q
, rq
);
400 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
403 static void queue_flush(request_queue_t
*q
, unsigned which
)
406 rq_end_io_fn
*end_io
;
408 if (which
== QUEUE_ORDERED_PREFLUSH
) {
409 rq
= &q
->pre_flush_rq
;
410 end_io
= pre_flush_end_io
;
412 rq
= &q
->post_flush_rq
;
413 end_io
= post_flush_end_io
;
416 rq
->cmd_flags
= REQ_HARDBARRIER
;
418 rq
->elevator_private
= NULL
;
419 rq
->elevator_private2
= NULL
;
420 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
422 q
->prepare_flush_fn(q
, rq
);
424 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
427 static inline struct request
*start_ordered(request_queue_t
*q
,
432 q
->ordered
= q
->next_ordered
;
433 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
436 * Prep proxy barrier request.
438 blkdev_dequeue_request(rq
);
443 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
444 rq
->cmd_flags
|= REQ_RW
;
445 rq
->cmd_flags
|= q
->ordered
& QUEUE_ORDERED_FUA
? REQ_FUA
: 0;
446 rq
->elevator_private
= NULL
;
447 rq
->elevator_private2
= NULL
;
448 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
449 rq
->end_io
= bar_end_io
;
452 * Queue ordered sequence. As we stack them at the head, we
453 * need to queue in reverse order. Note that we rely on that
454 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
455 * request gets inbetween ordered sequence.
457 if (q
->ordered
& QUEUE_ORDERED_POSTFLUSH
)
458 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
460 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
462 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
464 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
465 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
466 rq
= &q
->pre_flush_rq
;
468 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
470 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
471 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
478 int blk_do_ordered(request_queue_t
*q
, struct request
**rqp
)
480 struct request
*rq
= *rqp
;
481 int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
487 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
488 *rqp
= start_ordered(q
, rq
);
492 * This can happen when the queue switches to
493 * ORDERED_NONE while this request is on it.
495 blkdev_dequeue_request(rq
);
496 end_that_request_first(rq
, -EOPNOTSUPP
,
497 rq
->hard_nr_sectors
);
498 end_that_request_last(rq
, -EOPNOTSUPP
);
505 * Ordered sequence in progress
508 /* Special requests are not subject to ordering rules. */
509 if (!blk_fs_request(rq
) &&
510 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
513 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
514 /* Ordered by tag. Blocking the next barrier is enough. */
515 if (is_barrier
&& rq
!= &q
->bar_rq
)
518 /* Ordered by draining. Wait for turn. */
519 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
520 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
527 static int flush_dry_bio_endio(struct bio
*bio
, unsigned int bytes
, int error
)
529 request_queue_t
*q
= bio
->bi_private
;
530 struct bio_vec
*bvec
;
534 * This is dry run, restore bio_sector and size. We'll finish
535 * this request again with the original bi_end_io after an
536 * error occurs or post flush is complete.
545 bio_for_each_segment(bvec
, bio
, i
) {
546 bvec
->bv_len
+= bvec
->bv_offset
;
551 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
552 bio
->bi_size
= q
->bi_size
;
553 bio
->bi_sector
-= (q
->bi_size
>> 9);
559 static int ordered_bio_endio(struct request
*rq
, struct bio
*bio
,
560 unsigned int nbytes
, int error
)
562 request_queue_t
*q
= rq
->q
;
566 if (&q
->bar_rq
!= rq
)
570 * Okay, this is the barrier request in progress, dry finish it.
572 if (error
&& !q
->orderr
)
575 endio
= bio
->bi_end_io
;
576 private = bio
->bi_private
;
577 bio
->bi_end_io
= flush_dry_bio_endio
;
580 bio_endio(bio
, nbytes
, error
);
582 bio
->bi_end_io
= endio
;
583 bio
->bi_private
= private;
589 * blk_queue_bounce_limit - set bounce buffer limit for queue
590 * @q: the request queue for the device
591 * @dma_addr: bus address limit
594 * Different hardware can have different requirements as to what pages
595 * it can do I/O directly to. A low level driver can call
596 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
597 * buffers for doing I/O to pages residing above @page.
599 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
601 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
604 q
->bounce_gfp
= GFP_NOIO
;
605 #if BITS_PER_LONG == 64
606 /* Assume anything <= 4GB can be handled by IOMMU.
607 Actually some IOMMUs can handle everything, but I don't
608 know of a way to test this here. */
609 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
611 q
->bounce_pfn
= max_low_pfn
;
613 if (bounce_pfn
< blk_max_low_pfn
)
615 q
->bounce_pfn
= bounce_pfn
;
618 init_emergency_isa_pool();
619 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
620 q
->bounce_pfn
= bounce_pfn
;
624 EXPORT_SYMBOL(blk_queue_bounce_limit
);
627 * blk_queue_max_sectors - set max sectors for a request for this queue
628 * @q: the request queue for the device
629 * @max_sectors: max sectors in the usual 512b unit
632 * Enables a low level driver to set an upper limit on the size of
635 void blk_queue_max_sectors(request_queue_t
*q
, unsigned int max_sectors
)
637 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
638 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
639 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
642 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
643 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
645 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
646 q
->max_hw_sectors
= max_sectors
;
650 EXPORT_SYMBOL(blk_queue_max_sectors
);
653 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
654 * @q: the request queue for the device
655 * @max_segments: max number of segments
658 * Enables a low level driver to set an upper limit on the number of
659 * physical data segments in a request. This would be the largest sized
660 * scatter list the driver could handle.
662 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
666 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
669 q
->max_phys_segments
= max_segments
;
672 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
675 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
676 * @q: the request queue for the device
677 * @max_segments: max number of segments
680 * Enables a low level driver to set an upper limit on the number of
681 * hw data segments in a request. This would be the largest number of
682 * address/length pairs the host adapter can actually give as once
685 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
689 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
692 q
->max_hw_segments
= max_segments
;
695 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
698 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
699 * @q: the request queue for the device
700 * @max_size: max size of segment in bytes
703 * Enables a low level driver to set an upper limit on the size of a
706 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
708 if (max_size
< PAGE_CACHE_SIZE
) {
709 max_size
= PAGE_CACHE_SIZE
;
710 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
713 q
->max_segment_size
= max_size
;
716 EXPORT_SYMBOL(blk_queue_max_segment_size
);
719 * blk_queue_hardsect_size - set hardware sector size for the queue
720 * @q: the request queue for the device
721 * @size: the hardware sector size, in bytes
724 * This should typically be set to the lowest possible sector size
725 * that the hardware can operate on (possible without reverting to
726 * even internal read-modify-write operations). Usually the default
727 * of 512 covers most hardware.
729 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
731 q
->hardsect_size
= size
;
734 EXPORT_SYMBOL(blk_queue_hardsect_size
);
737 * Returns the minimum that is _not_ zero, unless both are zero.
739 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
742 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
743 * @t: the stacking driver (top)
744 * @b: the underlying device (bottom)
746 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
748 /* zero is "infinity" */
749 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
750 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
752 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
753 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
754 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
755 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
756 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
757 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
760 EXPORT_SYMBOL(blk_queue_stack_limits
);
763 * blk_queue_segment_boundary - set boundary rules for segment merging
764 * @q: the request queue for the device
765 * @mask: the memory boundary mask
767 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
769 if (mask
< PAGE_CACHE_SIZE
- 1) {
770 mask
= PAGE_CACHE_SIZE
- 1;
771 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
774 q
->seg_boundary_mask
= mask
;
777 EXPORT_SYMBOL(blk_queue_segment_boundary
);
780 * blk_queue_dma_alignment - set dma length and memory alignment
781 * @q: the request queue for the device
782 * @mask: alignment mask
785 * set required memory and length aligment for direct dma transactions.
786 * this is used when buiding direct io requests for the queue.
789 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
791 q
->dma_alignment
= mask
;
794 EXPORT_SYMBOL(blk_queue_dma_alignment
);
797 * blk_queue_find_tag - find a request by its tag and queue
798 * @q: The request queue for the device
799 * @tag: The tag of the request
802 * Should be used when a device returns a tag and you want to match
805 * no locks need be held.
807 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
809 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
812 EXPORT_SYMBOL(blk_queue_find_tag
);
815 * __blk_free_tags - release a given set of tag maintenance info
816 * @bqt: the tag map to free
818 * Tries to free the specified @bqt@. Returns true if it was
819 * actually freed and false if there are still references using it
821 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
825 retval
= atomic_dec_and_test(&bqt
->refcnt
);
828 BUG_ON(!list_empty(&bqt
->busy_list
));
830 kfree(bqt
->tag_index
);
831 bqt
->tag_index
= NULL
;
844 * __blk_queue_free_tags - release tag maintenance info
845 * @q: the request queue for the device
848 * blk_cleanup_queue() will take care of calling this function, if tagging
849 * has been used. So there's no need to call this directly.
851 static void __blk_queue_free_tags(request_queue_t
*q
)
853 struct blk_queue_tag
*bqt
= q
->queue_tags
;
858 __blk_free_tags(bqt
);
860 q
->queue_tags
= NULL
;
861 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
866 * blk_free_tags - release a given set of tag maintenance info
867 * @bqt: the tag map to free
869 * For externally managed @bqt@ frees the map. Callers of this
870 * function must guarantee to have released all the queues that
871 * might have been using this tag map.
873 void blk_free_tags(struct blk_queue_tag
*bqt
)
875 if (unlikely(!__blk_free_tags(bqt
)))
878 EXPORT_SYMBOL(blk_free_tags
);
881 * blk_queue_free_tags - release tag maintenance info
882 * @q: the request queue for the device
885 * This is used to disabled tagged queuing to a device, yet leave
888 void blk_queue_free_tags(request_queue_t
*q
)
890 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
893 EXPORT_SYMBOL(blk_queue_free_tags
);
896 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
898 struct request
**tag_index
;
899 unsigned long *tag_map
;
902 if (q
&& depth
> q
->nr_requests
* 2) {
903 depth
= q
->nr_requests
* 2;
904 printk(KERN_ERR
"%s: adjusted depth to %d\n",
905 __FUNCTION__
, depth
);
908 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
912 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
913 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
917 tags
->real_max_depth
= depth
;
918 tags
->max_depth
= depth
;
919 tags
->tag_index
= tag_index
;
920 tags
->tag_map
= tag_map
;
928 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
931 struct blk_queue_tag
*tags
;
933 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
937 if (init_tag_map(q
, tags
, depth
))
940 INIT_LIST_HEAD(&tags
->busy_list
);
942 atomic_set(&tags
->refcnt
, 1);
950 * blk_init_tags - initialize the tag info for an external tag map
951 * @depth: the maximum queue depth supported
952 * @tags: the tag to use
954 struct blk_queue_tag
*blk_init_tags(int depth
)
956 return __blk_queue_init_tags(NULL
, depth
);
958 EXPORT_SYMBOL(blk_init_tags
);
961 * blk_queue_init_tags - initialize the queue tag info
962 * @q: the request queue for the device
963 * @depth: the maximum queue depth supported
964 * @tags: the tag to use
966 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
967 struct blk_queue_tag
*tags
)
971 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
973 if (!tags
&& !q
->queue_tags
) {
974 tags
= __blk_queue_init_tags(q
, depth
);
978 } else if (q
->queue_tags
) {
979 if ((rc
= blk_queue_resize_tags(q
, depth
)))
981 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
984 atomic_inc(&tags
->refcnt
);
987 * assign it, all done
989 q
->queue_tags
= tags
;
990 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
997 EXPORT_SYMBOL(blk_queue_init_tags
);
1000 * blk_queue_resize_tags - change the queueing depth
1001 * @q: the request queue for the device
1002 * @new_depth: the new max command queueing depth
1005 * Must be called with the queue lock held.
1007 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
1009 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1010 struct request
**tag_index
;
1011 unsigned long *tag_map
;
1012 int max_depth
, nr_ulongs
;
1018 * if we already have large enough real_max_depth. just
1019 * adjust max_depth. *NOTE* as requests with tag value
1020 * between new_depth and real_max_depth can be in-flight, tag
1021 * map can not be shrunk blindly here.
1023 if (new_depth
<= bqt
->real_max_depth
) {
1024 bqt
->max_depth
= new_depth
;
1029 * Currently cannot replace a shared tag map with a new
1030 * one, so error out if this is the case
1032 if (atomic_read(&bqt
->refcnt
) != 1)
1036 * save the old state info, so we can copy it back
1038 tag_index
= bqt
->tag_index
;
1039 tag_map
= bqt
->tag_map
;
1040 max_depth
= bqt
->real_max_depth
;
1042 if (init_tag_map(q
, bqt
, new_depth
))
1045 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1046 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1047 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1054 EXPORT_SYMBOL(blk_queue_resize_tags
);
1057 * blk_queue_end_tag - end tag operations for a request
1058 * @q: the request queue for the device
1059 * @rq: the request that has completed
1062 * Typically called when end_that_request_first() returns 0, meaning
1063 * all transfers have been done for a request. It's important to call
1064 * this function before end_that_request_last(), as that will put the
1065 * request back on the free list thus corrupting the internal tag list.
1068 * queue lock must be held.
1070 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
1072 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1077 if (unlikely(tag
>= bqt
->real_max_depth
))
1079 * This can happen after tag depth has been reduced.
1080 * FIXME: how about a warning or info message here?
1084 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
1085 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1090 list_del_init(&rq
->queuelist
);
1091 rq
->cmd_flags
&= ~REQ_QUEUED
;
1094 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1095 printk(KERN_ERR
"%s: tag %d is missing\n",
1098 bqt
->tag_index
[tag
] = NULL
;
1102 EXPORT_SYMBOL(blk_queue_end_tag
);
1105 * blk_queue_start_tag - find a free tag and assign it
1106 * @q: the request queue for the device
1107 * @rq: the block request that needs tagging
1110 * This can either be used as a stand-alone helper, or possibly be
1111 * assigned as the queue &prep_rq_fn (in which case &struct request
1112 * automagically gets a tag assigned). Note that this function
1113 * assumes that any type of request can be queued! if this is not
1114 * true for your device, you must check the request type before
1115 * calling this function. The request will also be removed from
1116 * the request queue, so it's the drivers responsibility to readd
1117 * it if it should need to be restarted for some reason.
1120 * queue lock must be held.
1122 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
1124 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1127 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1129 "%s: request %p for device [%s] already tagged %d",
1131 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1136 * Protect against shared tag maps, as we may not have exclusive
1137 * access to the tag map.
1140 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1141 if (tag
>= bqt
->max_depth
)
1144 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1146 rq
->cmd_flags
|= REQ_QUEUED
;
1148 bqt
->tag_index
[tag
] = rq
;
1149 blkdev_dequeue_request(rq
);
1150 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1155 EXPORT_SYMBOL(blk_queue_start_tag
);
1158 * blk_queue_invalidate_tags - invalidate all pending tags
1159 * @q: the request queue for the device
1162 * Hardware conditions may dictate a need to stop all pending requests.
1163 * In this case, we will safely clear the block side of the tag queue and
1164 * readd all requests to the request queue in the right order.
1167 * queue lock must be held.
1169 void blk_queue_invalidate_tags(request_queue_t
*q
)
1171 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1172 struct list_head
*tmp
, *n
;
1175 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1176 rq
= list_entry_rq(tmp
);
1178 if (rq
->tag
== -1) {
1180 "%s: bad tag found on list\n", __FUNCTION__
);
1181 list_del_init(&rq
->queuelist
);
1182 rq
->cmd_flags
&= ~REQ_QUEUED
;
1184 blk_queue_end_tag(q
, rq
);
1186 rq
->cmd_flags
&= ~REQ_STARTED
;
1187 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1191 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1193 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1197 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1198 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1201 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1203 rq
->current_nr_sectors
);
1204 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1206 if (blk_pc_request(rq
)) {
1208 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1209 printk("%02x ", rq
->cmd
[bit
]);
1214 EXPORT_SYMBOL(blk_dump_rq_flags
);
1216 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1218 struct bio_vec
*bv
, *bvprv
= NULL
;
1219 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1220 int high
, highprv
= 1;
1222 if (unlikely(!bio
->bi_io_vec
))
1225 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1226 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1227 bio_for_each_segment(bv
, bio
, i
) {
1229 * the trick here is making sure that a high page is never
1230 * considered part of another segment, since that might
1231 * change with the bounce page.
1233 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1234 if (high
|| highprv
)
1235 goto new_hw_segment
;
1237 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1239 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1241 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1243 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1244 goto new_hw_segment
;
1246 seg_size
+= bv
->bv_len
;
1247 hw_seg_size
+= bv
->bv_len
;
1252 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1253 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1254 hw_seg_size
+= bv
->bv_len
;
1257 if (hw_seg_size
> bio
->bi_hw_front_size
)
1258 bio
->bi_hw_front_size
= hw_seg_size
;
1259 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1265 seg_size
= bv
->bv_len
;
1268 if (hw_seg_size
> bio
->bi_hw_back_size
)
1269 bio
->bi_hw_back_size
= hw_seg_size
;
1270 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1271 bio
->bi_hw_front_size
= hw_seg_size
;
1272 bio
->bi_phys_segments
= nr_phys_segs
;
1273 bio
->bi_hw_segments
= nr_hw_segs
;
1274 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1278 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1281 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1284 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1286 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1290 * bio and nxt are contigous in memory, check if the queue allows
1291 * these two to be merged into one
1293 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1299 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1302 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1303 blk_recount_segments(q
, bio
);
1304 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1305 blk_recount_segments(q
, nxt
);
1306 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1307 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1309 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1316 * map a request to scatterlist, return number of sg entries setup. Caller
1317 * must make sure sg can hold rq->nr_phys_segments entries
1319 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1321 struct bio_vec
*bvec
, *bvprv
;
1323 int nsegs
, i
, cluster
;
1326 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1329 * for each bio in rq
1332 rq_for_each_bio(bio
, rq
) {
1334 * for each segment in bio
1336 bio_for_each_segment(bvec
, bio
, i
) {
1337 int nbytes
= bvec
->bv_len
;
1339 if (bvprv
&& cluster
) {
1340 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1343 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1345 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1348 sg
[nsegs
- 1].length
+= nbytes
;
1351 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1352 sg
[nsegs
].page
= bvec
->bv_page
;
1353 sg
[nsegs
].length
= nbytes
;
1354 sg
[nsegs
].offset
= bvec
->bv_offset
;
1359 } /* segments in bio */
1365 EXPORT_SYMBOL(blk_rq_map_sg
);
1368 * the standard queue merge functions, can be overridden with device
1369 * specific ones if so desired
1372 static inline int ll_new_mergeable(request_queue_t
*q
,
1373 struct request
*req
,
1376 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1378 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1379 req
->cmd_flags
|= REQ_NOMERGE
;
1380 if (req
== q
->last_merge
)
1381 q
->last_merge
= NULL
;
1386 * A hw segment is just getting larger, bump just the phys
1389 req
->nr_phys_segments
+= nr_phys_segs
;
1393 static inline int ll_new_hw_segment(request_queue_t
*q
,
1394 struct request
*req
,
1397 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1398 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1400 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1401 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1402 req
->cmd_flags
|= REQ_NOMERGE
;
1403 if (req
== q
->last_merge
)
1404 q
->last_merge
= NULL
;
1409 * This will form the start of a new hw segment. Bump both
1412 req
->nr_hw_segments
+= nr_hw_segs
;
1413 req
->nr_phys_segments
+= nr_phys_segs
;
1417 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1420 unsigned short max_sectors
;
1423 if (unlikely(blk_pc_request(req
)))
1424 max_sectors
= q
->max_hw_sectors
;
1426 max_sectors
= q
->max_sectors
;
1428 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1429 req
->cmd_flags
|= REQ_NOMERGE
;
1430 if (req
== q
->last_merge
)
1431 q
->last_merge
= NULL
;
1434 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1435 blk_recount_segments(q
, req
->biotail
);
1436 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1437 blk_recount_segments(q
, bio
);
1438 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1439 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1440 !BIOVEC_VIRT_OVERSIZE(len
)) {
1441 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1444 if (req
->nr_hw_segments
== 1)
1445 req
->bio
->bi_hw_front_size
= len
;
1446 if (bio
->bi_hw_segments
== 1)
1447 bio
->bi_hw_back_size
= len
;
1452 return ll_new_hw_segment(q
, req
, bio
);
1455 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1458 unsigned short max_sectors
;
1461 if (unlikely(blk_pc_request(req
)))
1462 max_sectors
= q
->max_hw_sectors
;
1464 max_sectors
= q
->max_sectors
;
1467 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1468 req
->cmd_flags
|= REQ_NOMERGE
;
1469 if (req
== q
->last_merge
)
1470 q
->last_merge
= NULL
;
1473 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1474 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1475 blk_recount_segments(q
, bio
);
1476 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1477 blk_recount_segments(q
, req
->bio
);
1478 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1479 !BIOVEC_VIRT_OVERSIZE(len
)) {
1480 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1483 if (bio
->bi_hw_segments
== 1)
1484 bio
->bi_hw_front_size
= len
;
1485 if (req
->nr_hw_segments
== 1)
1486 req
->biotail
->bi_hw_back_size
= len
;
1491 return ll_new_hw_segment(q
, req
, bio
);
1494 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1495 struct request
*next
)
1497 int total_phys_segments
;
1498 int total_hw_segments
;
1501 * First check if the either of the requests are re-queued
1502 * requests. Can't merge them if they are.
1504 if (req
->special
|| next
->special
)
1508 * Will it become too large?
1510 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1513 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1514 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1515 total_phys_segments
--;
1517 if (total_phys_segments
> q
->max_phys_segments
)
1520 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1521 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1522 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1524 * propagate the combined length to the end of the requests
1526 if (req
->nr_hw_segments
== 1)
1527 req
->bio
->bi_hw_front_size
= len
;
1528 if (next
->nr_hw_segments
== 1)
1529 next
->biotail
->bi_hw_back_size
= len
;
1530 total_hw_segments
--;
1533 if (total_hw_segments
> q
->max_hw_segments
)
1536 /* Merge is OK... */
1537 req
->nr_phys_segments
= total_phys_segments
;
1538 req
->nr_hw_segments
= total_hw_segments
;
1543 * "plug" the device if there are no outstanding requests: this will
1544 * force the transfer to start only after we have put all the requests
1547 * This is called with interrupts off and no requests on the queue and
1548 * with the queue lock held.
1550 void blk_plug_device(request_queue_t
*q
)
1552 WARN_ON(!irqs_disabled());
1555 * don't plug a stopped queue, it must be paired with blk_start_queue()
1556 * which will restart the queueing
1558 if (blk_queue_stopped(q
))
1561 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1562 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1563 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1567 EXPORT_SYMBOL(blk_plug_device
);
1570 * remove the queue from the plugged list, if present. called with
1571 * queue lock held and interrupts disabled.
1573 int blk_remove_plug(request_queue_t
*q
)
1575 WARN_ON(!irqs_disabled());
1577 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1580 del_timer(&q
->unplug_timer
);
1584 EXPORT_SYMBOL(blk_remove_plug
);
1587 * remove the plug and let it rip..
1589 void __generic_unplug_device(request_queue_t
*q
)
1591 if (unlikely(blk_queue_stopped(q
)))
1594 if (!blk_remove_plug(q
))
1599 EXPORT_SYMBOL(__generic_unplug_device
);
1602 * generic_unplug_device - fire a request queue
1603 * @q: The &request_queue_t in question
1606 * Linux uses plugging to build bigger requests queues before letting
1607 * the device have at them. If a queue is plugged, the I/O scheduler
1608 * is still adding and merging requests on the queue. Once the queue
1609 * gets unplugged, the request_fn defined for the queue is invoked and
1610 * transfers started.
1612 void generic_unplug_device(request_queue_t
*q
)
1614 spin_lock_irq(q
->queue_lock
);
1615 __generic_unplug_device(q
);
1616 spin_unlock_irq(q
->queue_lock
);
1618 EXPORT_SYMBOL(generic_unplug_device
);
1620 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1623 request_queue_t
*q
= bdi
->unplug_io_data
;
1626 * devices don't necessarily have an ->unplug_fn defined
1629 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1630 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1636 static void blk_unplug_work(struct work_struct
*work
)
1638 request_queue_t
*q
= container_of(work
, request_queue_t
, unplug_work
);
1640 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1641 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1646 static void blk_unplug_timeout(unsigned long data
)
1648 request_queue_t
*q
= (request_queue_t
*)data
;
1650 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1651 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1653 kblockd_schedule_work(&q
->unplug_work
);
1657 * blk_start_queue - restart a previously stopped queue
1658 * @q: The &request_queue_t in question
1661 * blk_start_queue() will clear the stop flag on the queue, and call
1662 * the request_fn for the queue if it was in a stopped state when
1663 * entered. Also see blk_stop_queue(). Queue lock must be held.
1665 void blk_start_queue(request_queue_t
*q
)
1667 WARN_ON(!irqs_disabled());
1669 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1672 * one level of recursion is ok and is much faster than kicking
1673 * the unplug handling
1675 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1677 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1680 kblockd_schedule_work(&q
->unplug_work
);
1684 EXPORT_SYMBOL(blk_start_queue
);
1687 * blk_stop_queue - stop a queue
1688 * @q: The &request_queue_t in question
1691 * The Linux block layer assumes that a block driver will consume all
1692 * entries on the request queue when the request_fn strategy is called.
1693 * Often this will not happen, because of hardware limitations (queue
1694 * depth settings). If a device driver gets a 'queue full' response,
1695 * or if it simply chooses not to queue more I/O at one point, it can
1696 * call this function to prevent the request_fn from being called until
1697 * the driver has signalled it's ready to go again. This happens by calling
1698 * blk_start_queue() to restart queue operations. Queue lock must be held.
1700 void blk_stop_queue(request_queue_t
*q
)
1703 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1705 EXPORT_SYMBOL(blk_stop_queue
);
1708 * blk_sync_queue - cancel any pending callbacks on a queue
1712 * The block layer may perform asynchronous callback activity
1713 * on a queue, such as calling the unplug function after a timeout.
1714 * A block device may call blk_sync_queue to ensure that any
1715 * such activity is cancelled, thus allowing it to release resources
1716 * the the callbacks might use. The caller must already have made sure
1717 * that its ->make_request_fn will not re-add plugging prior to calling
1721 void blk_sync_queue(struct request_queue
*q
)
1723 del_timer_sync(&q
->unplug_timer
);
1726 EXPORT_SYMBOL(blk_sync_queue
);
1729 * blk_run_queue - run a single device queue
1730 * @q: The queue to run
1732 void blk_run_queue(struct request_queue
*q
)
1734 unsigned long flags
;
1736 spin_lock_irqsave(q
->queue_lock
, flags
);
1740 * Only recurse once to avoid overrunning the stack, let the unplug
1741 * handling reinvoke the handler shortly if we already got there.
1743 if (!elv_queue_empty(q
)) {
1744 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1746 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1749 kblockd_schedule_work(&q
->unplug_work
);
1753 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1755 EXPORT_SYMBOL(blk_run_queue
);
1758 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1759 * @kobj: the kobj belonging of the request queue to be released
1762 * blk_cleanup_queue is the pair to blk_init_queue() or
1763 * blk_queue_make_request(). It should be called when a request queue is
1764 * being released; typically when a block device is being de-registered.
1765 * Currently, its primary task it to free all the &struct request
1766 * structures that were allocated to the queue and the queue itself.
1769 * Hopefully the low level driver will have finished any
1770 * outstanding requests first...
1772 static void blk_release_queue(struct kobject
*kobj
)
1774 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
1775 struct request_list
*rl
= &q
->rq
;
1780 mempool_destroy(rl
->rq_pool
);
1783 __blk_queue_free_tags(q
);
1785 blk_trace_shutdown(q
);
1787 kmem_cache_free(requestq_cachep
, q
);
1790 void blk_put_queue(request_queue_t
*q
)
1792 kobject_put(&q
->kobj
);
1794 EXPORT_SYMBOL(blk_put_queue
);
1796 void blk_cleanup_queue(request_queue_t
* q
)
1798 mutex_lock(&q
->sysfs_lock
);
1799 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1800 mutex_unlock(&q
->sysfs_lock
);
1803 elevator_exit(q
->elevator
);
1808 EXPORT_SYMBOL(blk_cleanup_queue
);
1810 static int blk_init_free_list(request_queue_t
*q
)
1812 struct request_list
*rl
= &q
->rq
;
1814 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1815 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1817 init_waitqueue_head(&rl
->wait
[READ
]);
1818 init_waitqueue_head(&rl
->wait
[WRITE
]);
1820 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1821 mempool_free_slab
, request_cachep
, q
->node
);
1829 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1831 return blk_alloc_queue_node(gfp_mask
, -1);
1833 EXPORT_SYMBOL(blk_alloc_queue
);
1835 static struct kobj_type queue_ktype
;
1837 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1841 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1845 memset(q
, 0, sizeof(*q
));
1846 init_timer(&q
->unplug_timer
);
1848 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
1849 q
->kobj
.ktype
= &queue_ktype
;
1850 kobject_init(&q
->kobj
);
1852 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1853 q
->backing_dev_info
.unplug_io_data
= q
;
1855 mutex_init(&q
->sysfs_lock
);
1859 EXPORT_SYMBOL(blk_alloc_queue_node
);
1862 * blk_init_queue - prepare a request queue for use with a block device
1863 * @rfn: The function to be called to process requests that have been
1864 * placed on the queue.
1865 * @lock: Request queue spin lock
1868 * If a block device wishes to use the standard request handling procedures,
1869 * which sorts requests and coalesces adjacent requests, then it must
1870 * call blk_init_queue(). The function @rfn will be called when there
1871 * are requests on the queue that need to be processed. If the device
1872 * supports plugging, then @rfn may not be called immediately when requests
1873 * are available on the queue, but may be called at some time later instead.
1874 * Plugged queues are generally unplugged when a buffer belonging to one
1875 * of the requests on the queue is needed, or due to memory pressure.
1877 * @rfn is not required, or even expected, to remove all requests off the
1878 * queue, but only as many as it can handle at a time. If it does leave
1879 * requests on the queue, it is responsible for arranging that the requests
1880 * get dealt with eventually.
1882 * The queue spin lock must be held while manipulating the requests on the
1883 * request queue; this lock will be taken also from interrupt context, so irq
1884 * disabling is needed for it.
1886 * Function returns a pointer to the initialized request queue, or NULL if
1887 * it didn't succeed.
1890 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1891 * when the block device is deactivated (such as at module unload).
1894 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1896 return blk_init_queue_node(rfn
, lock
, -1);
1898 EXPORT_SYMBOL(blk_init_queue
);
1901 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1903 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1909 if (blk_init_free_list(q
)) {
1910 kmem_cache_free(requestq_cachep
, q
);
1915 * if caller didn't supply a lock, they get per-queue locking with
1919 spin_lock_init(&q
->__queue_lock
);
1920 lock
= &q
->__queue_lock
;
1923 q
->request_fn
= rfn
;
1924 q
->back_merge_fn
= ll_back_merge_fn
;
1925 q
->front_merge_fn
= ll_front_merge_fn
;
1926 q
->merge_requests_fn
= ll_merge_requests_fn
;
1927 q
->prep_rq_fn
= NULL
;
1928 q
->unplug_fn
= generic_unplug_device
;
1929 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1930 q
->queue_lock
= lock
;
1932 blk_queue_segment_boundary(q
, 0xffffffff);
1934 blk_queue_make_request(q
, __make_request
);
1935 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1937 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1938 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1943 if (!elevator_init(q
, NULL
)) {
1944 blk_queue_congestion_threshold(q
);
1951 EXPORT_SYMBOL(blk_init_queue_node
);
1953 int blk_get_queue(request_queue_t
*q
)
1955 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1956 kobject_get(&q
->kobj
);
1963 EXPORT_SYMBOL(blk_get_queue
);
1965 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1967 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1968 elv_put_request(q
, rq
);
1969 mempool_free(rq
, q
->rq
.rq_pool
);
1972 static struct request
*
1973 blk_alloc_request(request_queue_t
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1975 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1981 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1982 * see bio.h and blkdev.h
1984 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
1987 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
1988 mempool_free(rq
, q
->rq
.rq_pool
);
1991 rq
->cmd_flags
|= REQ_ELVPRIV
;
1998 * ioc_batching returns true if the ioc is a valid batching request and
1999 * should be given priority access to a request.
2001 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
2007 * Make sure the process is able to allocate at least 1 request
2008 * even if the batch times out, otherwise we could theoretically
2011 return ioc
->nr_batch_requests
== q
->nr_batching
||
2012 (ioc
->nr_batch_requests
> 0
2013 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2017 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2018 * will cause the process to be a "batcher" on all queues in the system. This
2019 * is the behaviour we want though - once it gets a wakeup it should be given
2022 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
2024 if (!ioc
|| ioc_batching(q
, ioc
))
2027 ioc
->nr_batch_requests
= q
->nr_batching
;
2028 ioc
->last_waited
= jiffies
;
2031 static void __freed_request(request_queue_t
*q
, int rw
)
2033 struct request_list
*rl
= &q
->rq
;
2035 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2036 blk_clear_queue_congested(q
, rw
);
2038 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2039 if (waitqueue_active(&rl
->wait
[rw
]))
2040 wake_up(&rl
->wait
[rw
]);
2042 blk_clear_queue_full(q
, rw
);
2047 * A request has just been released. Account for it, update the full and
2048 * congestion status, wake up any waiters. Called under q->queue_lock.
2050 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
2052 struct request_list
*rl
= &q
->rq
;
2058 __freed_request(q
, rw
);
2060 if (unlikely(rl
->starved
[rw
^ 1]))
2061 __freed_request(q
, rw
^ 1);
2064 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2066 * Get a free request, queue_lock must be held.
2067 * Returns NULL on failure, with queue_lock held.
2068 * Returns !NULL on success, with queue_lock *not held*.
2070 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
2073 struct request
*rq
= NULL
;
2074 struct request_list
*rl
= &q
->rq
;
2075 struct io_context
*ioc
= NULL
;
2076 int may_queue
, priv
;
2078 may_queue
= elv_may_queue(q
, rw
);
2079 if (may_queue
== ELV_MQUEUE_NO
)
2082 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2083 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2084 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2086 * The queue will fill after this allocation, so set
2087 * it as full, and mark this process as "batching".
2088 * This process will be allowed to complete a batch of
2089 * requests, others will be blocked.
2091 if (!blk_queue_full(q
, rw
)) {
2092 ioc_set_batching(q
, ioc
);
2093 blk_set_queue_full(q
, rw
);
2095 if (may_queue
!= ELV_MQUEUE_MUST
2096 && !ioc_batching(q
, ioc
)) {
2098 * The queue is full and the allocating
2099 * process is not a "batcher", and not
2100 * exempted by the IO scheduler
2106 blk_set_queue_congested(q
, rw
);
2110 * Only allow batching queuers to allocate up to 50% over the defined
2111 * limit of requests, otherwise we could have thousands of requests
2112 * allocated with any setting of ->nr_requests
2114 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2118 rl
->starved
[rw
] = 0;
2120 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2124 spin_unlock_irq(q
->queue_lock
);
2126 rq
= blk_alloc_request(q
, rw
, priv
, gfp_mask
);
2127 if (unlikely(!rq
)) {
2129 * Allocation failed presumably due to memory. Undo anything
2130 * we might have messed up.
2132 * Allocating task should really be put onto the front of the
2133 * wait queue, but this is pretty rare.
2135 spin_lock_irq(q
->queue_lock
);
2136 freed_request(q
, rw
, priv
);
2139 * in the very unlikely event that allocation failed and no
2140 * requests for this direction was pending, mark us starved
2141 * so that freeing of a request in the other direction will
2142 * notice us. another possible fix would be to split the
2143 * rq mempool into READ and WRITE
2146 if (unlikely(rl
->count
[rw
] == 0))
2147 rl
->starved
[rw
] = 1;
2153 * ioc may be NULL here, and ioc_batching will be false. That's
2154 * OK, if the queue is under the request limit then requests need
2155 * not count toward the nr_batch_requests limit. There will always
2156 * be some limit enforced by BLK_BATCH_TIME.
2158 if (ioc_batching(q
, ioc
))
2159 ioc
->nr_batch_requests
--;
2163 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2169 * No available requests for this queue, unplug the device and wait for some
2170 * requests to become available.
2172 * Called with q->queue_lock held, and returns with it unlocked.
2174 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
2179 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2182 struct request_list
*rl
= &q
->rq
;
2184 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2185 TASK_UNINTERRUPTIBLE
);
2187 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2190 struct io_context
*ioc
;
2192 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2194 __generic_unplug_device(q
);
2195 spin_unlock_irq(q
->queue_lock
);
2199 * After sleeping, we become a "batching" process and
2200 * will be able to allocate at least one request, and
2201 * up to a big batch of them for a small period time.
2202 * See ioc_batching, ioc_set_batching
2204 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2205 ioc_set_batching(q
, ioc
);
2207 spin_lock_irq(q
->queue_lock
);
2209 finish_wait(&rl
->wait
[rw
], &wait
);
2215 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2219 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2221 spin_lock_irq(q
->queue_lock
);
2222 if (gfp_mask
& __GFP_WAIT
) {
2223 rq
= get_request_wait(q
, rw
, NULL
);
2225 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2227 spin_unlock_irq(q
->queue_lock
);
2229 /* q->queue_lock is unlocked at this point */
2233 EXPORT_SYMBOL(blk_get_request
);
2236 * blk_start_queueing - initiate dispatch of requests to device
2237 * @q: request queue to kick into gear
2239 * This is basically a helper to remove the need to know whether a queue
2240 * is plugged or not if someone just wants to initiate dispatch of requests
2243 * The queue lock must be held with interrupts disabled.
2245 void blk_start_queueing(request_queue_t
*q
)
2247 if (!blk_queue_plugged(q
))
2250 __generic_unplug_device(q
);
2252 EXPORT_SYMBOL(blk_start_queueing
);
2255 * blk_requeue_request - put a request back on queue
2256 * @q: request queue where request should be inserted
2257 * @rq: request to be inserted
2260 * Drivers often keep queueing requests until the hardware cannot accept
2261 * more, when that condition happens we need to put the request back
2262 * on the queue. Must be called with queue lock held.
2264 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2266 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2268 if (blk_rq_tagged(rq
))
2269 blk_queue_end_tag(q
, rq
);
2271 elv_requeue_request(q
, rq
);
2274 EXPORT_SYMBOL(blk_requeue_request
);
2277 * blk_insert_request - insert a special request in to a request queue
2278 * @q: request queue where request should be inserted
2279 * @rq: request to be inserted
2280 * @at_head: insert request at head or tail of queue
2281 * @data: private data
2284 * Many block devices need to execute commands asynchronously, so they don't
2285 * block the whole kernel from preemption during request execution. This is
2286 * accomplished normally by inserting aritficial requests tagged as
2287 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2288 * scheduled for actual execution by the request queue.
2290 * We have the option of inserting the head or the tail of the queue.
2291 * Typically we use the tail for new ioctls and so forth. We use the head
2292 * of the queue for things like a QUEUE_FULL message from a device, or a
2293 * host that is unable to accept a particular command.
2295 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2296 int at_head
, void *data
)
2298 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2299 unsigned long flags
;
2302 * tell I/O scheduler that this isn't a regular read/write (ie it
2303 * must not attempt merges on this) and that it acts as a soft
2306 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2307 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2311 spin_lock_irqsave(q
->queue_lock
, flags
);
2314 * If command is tagged, release the tag
2316 if (blk_rq_tagged(rq
))
2317 blk_queue_end_tag(q
, rq
);
2319 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2320 __elv_add_request(q
, rq
, where
, 0);
2321 blk_start_queueing(q
);
2322 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2325 EXPORT_SYMBOL(blk_insert_request
);
2327 static int __blk_rq_unmap_user(struct bio
*bio
)
2332 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2333 bio_unmap_user(bio
);
2335 ret
= bio_uncopy_user(bio
);
2341 static int __blk_rq_map_user(request_queue_t
*q
, struct request
*rq
,
2342 void __user
*ubuf
, unsigned int len
)
2344 unsigned long uaddr
;
2345 struct bio
*bio
, *orig_bio
;
2348 reading
= rq_data_dir(rq
) == READ
;
2351 * if alignment requirement is satisfied, map in user pages for
2352 * direct dma. else, set up kernel bounce buffers
2354 uaddr
= (unsigned long) ubuf
;
2355 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2356 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2358 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2361 return PTR_ERR(bio
);
2365 blk_queue_bounce(q
, &bio
);
2367 * We link the bounce buffer in and could have to traverse it
2368 * later so we have to get a ref to prevent it from being freed
2373 * for most (all? don't know of any) queues we could
2374 * skip grabbing the queue lock here. only drivers with
2375 * funky private ->back_merge_fn() function could be
2378 spin_lock_irq(q
->queue_lock
);
2380 blk_rq_bio_prep(q
, rq
, bio
);
2381 else if (!q
->back_merge_fn(q
, rq
, bio
)) {
2383 spin_unlock_irq(q
->queue_lock
);
2386 rq
->biotail
->bi_next
= bio
;
2389 rq
->nr_sectors
+= bio_sectors(bio
);
2390 rq
->hard_nr_sectors
= rq
->nr_sectors
;
2391 rq
->data_len
+= bio
->bi_size
;
2393 spin_unlock_irq(q
->queue_lock
);
2395 return bio
->bi_size
;
2398 /* if it was boucned we must call the end io function */
2399 bio_endio(bio
, bio
->bi_size
, 0);
2400 __blk_rq_unmap_user(orig_bio
);
2406 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2407 * @q: request queue where request should be inserted
2408 * @rq: request structure to fill
2409 * @ubuf: the user buffer
2410 * @len: length of user data
2413 * Data will be mapped directly for zero copy io, if possible. Otherwise
2414 * a kernel bounce buffer is used.
2416 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2417 * still in process context.
2419 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2420 * before being submitted to the device, as pages mapped may be out of
2421 * reach. It's the callers responsibility to make sure this happens. The
2422 * original bio must be passed back in to blk_rq_unmap_user() for proper
2425 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2428 unsigned long bytes_read
= 0;
2431 if (len
> (q
->max_hw_sectors
<< 9))
2436 while (bytes_read
!= len
) {
2437 unsigned long map_len
, end
, start
;
2439 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2440 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2442 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2445 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2446 * pages. If this happens we just lower the requested
2447 * mapping len by a page so that we can fit
2449 if (end
- start
> BIO_MAX_PAGES
)
2450 map_len
-= PAGE_SIZE
;
2452 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2459 rq
->buffer
= rq
->data
= NULL
;
2462 blk_rq_unmap_user(rq
);
2466 EXPORT_SYMBOL(blk_rq_map_user
);
2469 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2470 * @q: request queue where request should be inserted
2471 * @rq: request to map data to
2472 * @iov: pointer to the iovec
2473 * @iov_count: number of elements in the iovec
2476 * Data will be mapped directly for zero copy io, if possible. Otherwise
2477 * a kernel bounce buffer is used.
2479 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2480 * still in process context.
2482 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2483 * before being submitted to the device, as pages mapped may be out of
2484 * reach. It's the callers responsibility to make sure this happens. The
2485 * original bio must be passed back in to blk_rq_unmap_user() for proper
2488 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2489 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2493 if (!iov
|| iov_count
<= 0)
2496 /* we don't allow misaligned data like bio_map_user() does. If the
2497 * user is using sg, they're expected to know the alignment constraints
2498 * and respect them accordingly */
2499 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2501 return PTR_ERR(bio
);
2503 if (bio
->bi_size
!= len
) {
2504 bio_endio(bio
, bio
->bi_size
, 0);
2505 bio_unmap_user(bio
);
2510 blk_rq_bio_prep(q
, rq
, bio
);
2511 rq
->buffer
= rq
->data
= NULL
;
2515 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2518 * blk_rq_unmap_user - unmap a request with user data
2519 * @rq: rq to be unmapped
2522 * Unmap a rq previously mapped by blk_rq_map_user().
2523 * rq->bio must be set to the original head of the request.
2525 int blk_rq_unmap_user(struct request
*rq
)
2527 struct bio
*bio
, *mapped_bio
;
2529 while ((bio
= rq
->bio
)) {
2530 if (bio_flagged(bio
, BIO_BOUNCED
))
2531 mapped_bio
= bio
->bi_private
;
2535 __blk_rq_unmap_user(mapped_bio
);
2536 rq
->bio
= bio
->bi_next
;
2542 EXPORT_SYMBOL(blk_rq_unmap_user
);
2545 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2546 * @q: request queue where request should be inserted
2547 * @rq: request to fill
2548 * @kbuf: the kernel buffer
2549 * @len: length of user data
2550 * @gfp_mask: memory allocation flags
2552 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2553 unsigned int len
, gfp_t gfp_mask
)
2557 if (len
> (q
->max_hw_sectors
<< 9))
2562 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2564 return PTR_ERR(bio
);
2566 if (rq_data_dir(rq
) == WRITE
)
2567 bio
->bi_rw
|= (1 << BIO_RW
);
2569 blk_rq_bio_prep(q
, rq
, bio
);
2570 rq
->buffer
= rq
->data
= NULL
;
2574 EXPORT_SYMBOL(blk_rq_map_kern
);
2577 * blk_execute_rq_nowait - insert a request into queue for execution
2578 * @q: queue to insert the request in
2579 * @bd_disk: matching gendisk
2580 * @rq: request to insert
2581 * @at_head: insert request at head or tail of queue
2582 * @done: I/O completion handler
2585 * Insert a fully prepared request at the back of the io scheduler queue
2586 * for execution. Don't wait for completion.
2588 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2589 struct request
*rq
, int at_head
,
2592 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2594 rq
->rq_disk
= bd_disk
;
2595 rq
->cmd_flags
|= REQ_NOMERGE
;
2597 WARN_ON(irqs_disabled());
2598 spin_lock_irq(q
->queue_lock
);
2599 __elv_add_request(q
, rq
, where
, 1);
2600 __generic_unplug_device(q
);
2601 spin_unlock_irq(q
->queue_lock
);
2603 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2606 * blk_execute_rq - insert a request into queue for execution
2607 * @q: queue to insert the request in
2608 * @bd_disk: matching gendisk
2609 * @rq: request to insert
2610 * @at_head: insert request at head or tail of queue
2613 * Insert a fully prepared request at the back of the io scheduler queue
2614 * for execution and wait for completion.
2616 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2617 struct request
*rq
, int at_head
)
2619 DECLARE_COMPLETION_ONSTACK(wait
);
2620 char sense
[SCSI_SENSE_BUFFERSIZE
];
2624 * we need an extra reference to the request, so we can look at
2625 * it after io completion
2630 memset(sense
, 0, sizeof(sense
));
2635 rq
->end_io_data
= &wait
;
2636 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2637 wait_for_completion(&wait
);
2645 EXPORT_SYMBOL(blk_execute_rq
);
2648 * blkdev_issue_flush - queue a flush
2649 * @bdev: blockdev to issue flush for
2650 * @error_sector: error sector
2653 * Issue a flush for the block device in question. Caller can supply
2654 * room for storing the error offset in case of a flush error, if they
2655 * wish to. Caller must run wait_for_completion() on its own.
2657 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2661 if (bdev
->bd_disk
== NULL
)
2664 q
= bdev_get_queue(bdev
);
2667 if (!q
->issue_flush_fn
)
2670 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2673 EXPORT_SYMBOL(blkdev_issue_flush
);
2675 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2677 int rw
= rq_data_dir(rq
);
2679 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2683 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2685 disk_round_stats(rq
->rq_disk
);
2686 rq
->rq_disk
->in_flight
++;
2691 * add-request adds a request to the linked list.
2692 * queue lock is held and interrupts disabled, as we muck with the
2693 * request queue list.
2695 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2697 drive_stat_acct(req
, req
->nr_sectors
, 1);
2700 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2703 * elevator indicated where it wants this request to be
2704 * inserted at elevator_merge time
2706 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2710 * disk_round_stats() - Round off the performance stats on a struct
2713 * The average IO queue length and utilisation statistics are maintained
2714 * by observing the current state of the queue length and the amount of
2715 * time it has been in this state for.
2717 * Normally, that accounting is done on IO completion, but that can result
2718 * in more than a second's worth of IO being accounted for within any one
2719 * second, leading to >100% utilisation. To deal with that, we call this
2720 * function to do a round-off before returning the results when reading
2721 * /proc/diskstats. This accounts immediately for all queue usage up to
2722 * the current jiffies and restarts the counters again.
2724 void disk_round_stats(struct gendisk
*disk
)
2726 unsigned long now
= jiffies
;
2728 if (now
== disk
->stamp
)
2731 if (disk
->in_flight
) {
2732 __disk_stat_add(disk
, time_in_queue
,
2733 disk
->in_flight
* (now
- disk
->stamp
));
2734 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2739 EXPORT_SYMBOL_GPL(disk_round_stats
);
2742 * queue lock must be held
2744 void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2748 if (unlikely(--req
->ref_count
))
2751 elv_completed_request(q
, req
);
2754 * Request may not have originated from ll_rw_blk. if not,
2755 * it didn't come out of our reserved rq pools
2757 if (req
->cmd_flags
& REQ_ALLOCED
) {
2758 int rw
= rq_data_dir(req
);
2759 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2761 BUG_ON(!list_empty(&req
->queuelist
));
2762 BUG_ON(!hlist_unhashed(&req
->hash
));
2764 blk_free_request(q
, req
);
2765 freed_request(q
, rw
, priv
);
2769 EXPORT_SYMBOL_GPL(__blk_put_request
);
2771 void blk_put_request(struct request
*req
)
2773 unsigned long flags
;
2774 request_queue_t
*q
= req
->q
;
2777 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2778 * following if (q) test.
2781 spin_lock_irqsave(q
->queue_lock
, flags
);
2782 __blk_put_request(q
, req
);
2783 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2787 EXPORT_SYMBOL(blk_put_request
);
2790 * blk_end_sync_rq - executes a completion event on a request
2791 * @rq: request to complete
2792 * @error: end io status of the request
2794 void blk_end_sync_rq(struct request
*rq
, int error
)
2796 struct completion
*waiting
= rq
->end_io_data
;
2798 rq
->end_io_data
= NULL
;
2799 __blk_put_request(rq
->q
, rq
);
2802 * complete last, if this is a stack request the process (and thus
2803 * the rq pointer) could be invalid right after this complete()
2807 EXPORT_SYMBOL(blk_end_sync_rq
);
2810 * Has to be called with the request spinlock acquired
2812 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2813 struct request
*next
)
2815 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2821 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2824 if (rq_data_dir(req
) != rq_data_dir(next
)
2825 || req
->rq_disk
!= next
->rq_disk
2830 * If we are allowed to merge, then append bio list
2831 * from next to rq and release next. merge_requests_fn
2832 * will have updated segment counts, update sector
2835 if (!q
->merge_requests_fn(q
, req
, next
))
2839 * At this point we have either done a back merge
2840 * or front merge. We need the smaller start_time of
2841 * the merged requests to be the current request
2842 * for accounting purposes.
2844 if (time_after(req
->start_time
, next
->start_time
))
2845 req
->start_time
= next
->start_time
;
2847 req
->biotail
->bi_next
= next
->bio
;
2848 req
->biotail
= next
->biotail
;
2850 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2852 elv_merge_requests(q
, req
, next
);
2855 disk_round_stats(req
->rq_disk
);
2856 req
->rq_disk
->in_flight
--;
2859 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2861 __blk_put_request(q
, next
);
2865 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2867 struct request
*next
= elv_latter_request(q
, rq
);
2870 return attempt_merge(q
, rq
, next
);
2875 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2877 struct request
*prev
= elv_former_request(q
, rq
);
2880 return attempt_merge(q
, prev
, rq
);
2885 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2887 req
->cmd_type
= REQ_TYPE_FS
;
2890 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2892 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2893 req
->cmd_flags
|= REQ_FAILFAST
;
2896 * REQ_BARRIER implies no merging, but lets make it explicit
2898 if (unlikely(bio_barrier(bio
)))
2899 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2902 req
->cmd_flags
|= REQ_RW_SYNC
;
2903 if (bio_rw_meta(bio
))
2904 req
->cmd_flags
|= REQ_RW_META
;
2907 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2908 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2909 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2910 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2911 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2912 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2913 req
->bio
= req
->biotail
= bio
;
2914 req
->ioprio
= bio_prio(bio
);
2915 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2916 req
->start_time
= jiffies
;
2919 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2921 struct request
*req
;
2922 int el_ret
, nr_sectors
, barrier
, err
;
2923 const unsigned short prio
= bio_prio(bio
);
2924 const int sync
= bio_sync(bio
);
2926 nr_sectors
= bio_sectors(bio
);
2929 * low level driver can indicate that it wants pages above a
2930 * certain limit bounced to low memory (ie for highmem, or even
2931 * ISA dma in theory)
2933 blk_queue_bounce(q
, &bio
);
2935 barrier
= bio_barrier(bio
);
2936 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2941 spin_lock_irq(q
->queue_lock
);
2943 if (unlikely(barrier
) || elv_queue_empty(q
))
2946 el_ret
= elv_merge(q
, &req
, bio
);
2948 case ELEVATOR_BACK_MERGE
:
2949 BUG_ON(!rq_mergeable(req
));
2951 if (!q
->back_merge_fn(q
, req
, bio
))
2954 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2956 req
->biotail
->bi_next
= bio
;
2958 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2959 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2960 drive_stat_acct(req
, nr_sectors
, 0);
2961 if (!attempt_back_merge(q
, req
))
2962 elv_merged_request(q
, req
, el_ret
);
2965 case ELEVATOR_FRONT_MERGE
:
2966 BUG_ON(!rq_mergeable(req
));
2968 if (!q
->front_merge_fn(q
, req
, bio
))
2971 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
2973 bio
->bi_next
= req
->bio
;
2977 * may not be valid. if the low level driver said
2978 * it didn't need a bounce buffer then it better
2979 * not touch req->buffer either...
2981 req
->buffer
= bio_data(bio
);
2982 req
->current_nr_sectors
= bio_cur_sectors(bio
);
2983 req
->hard_cur_sectors
= req
->current_nr_sectors
;
2984 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
2985 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2986 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2987 drive_stat_acct(req
, nr_sectors
, 0);
2988 if (!attempt_front_merge(q
, req
))
2989 elv_merged_request(q
, req
, el_ret
);
2992 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2999 * Grab a free request. This is might sleep but can not fail.
3000 * Returns with the queue unlocked.
3002 req
= get_request_wait(q
, bio_data_dir(bio
), bio
);
3005 * After dropping the lock and possibly sleeping here, our request
3006 * may now be mergeable after it had proven unmergeable (above).
3007 * We don't worry about that case for efficiency. It won't happen
3008 * often, and the elevators are able to handle it.
3010 init_request_from_bio(req
, bio
);
3012 spin_lock_irq(q
->queue_lock
);
3013 if (elv_queue_empty(q
))
3015 add_request(q
, req
);
3018 __generic_unplug_device(q
);
3020 spin_unlock_irq(q
->queue_lock
);
3024 bio_endio(bio
, nr_sectors
<< 9, err
);
3029 * If bio->bi_dev is a partition, remap the location
3031 static inline void blk_partition_remap(struct bio
*bio
)
3033 struct block_device
*bdev
= bio
->bi_bdev
;
3035 if (bdev
!= bdev
->bd_contains
) {
3036 struct hd_struct
*p
= bdev
->bd_part
;
3037 const int rw
= bio_data_dir(bio
);
3039 p
->sectors
[rw
] += bio_sectors(bio
);
3042 bio
->bi_sector
+= p
->start_sect
;
3043 bio
->bi_bdev
= bdev
->bd_contains
;
3047 static void handle_bad_sector(struct bio
*bio
)
3049 char b
[BDEVNAME_SIZE
];
3051 printk(KERN_INFO
"attempt to access beyond end of device\n");
3052 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3053 bdevname(bio
->bi_bdev
, b
),
3055 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3056 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3058 set_bit(BIO_EOF
, &bio
->bi_flags
);
3061 #ifdef CONFIG_FAIL_MAKE_REQUEST
3063 static DECLARE_FAULT_ATTR(fail_make_request
);
3065 static int __init
setup_fail_make_request(char *str
)
3067 return setup_fault_attr(&fail_make_request
, str
);
3069 __setup("fail_make_request=", setup_fail_make_request
);
3071 static int should_fail_request(struct bio
*bio
)
3073 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3074 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3075 return should_fail(&fail_make_request
, bio
->bi_size
);
3080 static int __init
fail_make_request_debugfs(void)
3082 return init_fault_attr_dentries(&fail_make_request
,
3083 "fail_make_request");
3086 late_initcall(fail_make_request_debugfs
);
3088 #else /* CONFIG_FAIL_MAKE_REQUEST */
3090 static inline int should_fail_request(struct bio
*bio
)
3095 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3098 * generic_make_request: hand a buffer to its device driver for I/O
3099 * @bio: The bio describing the location in memory and on the device.
3101 * generic_make_request() is used to make I/O requests of block
3102 * devices. It is passed a &struct bio, which describes the I/O that needs
3105 * generic_make_request() does not return any status. The
3106 * success/failure status of the request, along with notification of
3107 * completion, is delivered asynchronously through the bio->bi_end_io
3108 * function described (one day) else where.
3110 * The caller of generic_make_request must make sure that bi_io_vec
3111 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3112 * set to describe the device address, and the
3113 * bi_end_io and optionally bi_private are set to describe how
3114 * completion notification should be signaled.
3116 * generic_make_request and the drivers it calls may use bi_next if this
3117 * bio happens to be merged with someone else, and may change bi_dev and
3118 * bi_sector for remaps as it sees fit. So the values of these fields
3119 * should NOT be depended on after the call to generic_make_request.
3121 void generic_make_request(struct bio
*bio
)
3125 sector_t old_sector
;
3126 int ret
, nr_sectors
= bio_sectors(bio
);
3130 /* Test device or partition size, when known. */
3131 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3133 sector_t sector
= bio
->bi_sector
;
3135 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3137 * This may well happen - the kernel calls bread()
3138 * without checking the size of the device, e.g., when
3139 * mounting a device.
3141 handle_bad_sector(bio
);
3147 * Resolve the mapping until finished. (drivers are
3148 * still free to implement/resolve their own stacking
3149 * by explicitly returning 0)
3151 * NOTE: we don't repeat the blk_size check for each new device.
3152 * Stacking drivers are expected to know what they are doing.
3157 char b
[BDEVNAME_SIZE
];
3159 q
= bdev_get_queue(bio
->bi_bdev
);
3162 "generic_make_request: Trying to access "
3163 "nonexistent block-device %s (%Lu)\n",
3164 bdevname(bio
->bi_bdev
, b
),
3165 (long long) bio
->bi_sector
);
3167 bio_endio(bio
, bio
->bi_size
, -EIO
);
3171 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
3172 printk("bio too big device %s (%u > %u)\n",
3173 bdevname(bio
->bi_bdev
, b
),
3179 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3182 if (should_fail_request(bio
))
3186 * If this device has partitions, remap block n
3187 * of partition p to block n+start(p) of the disk.
3189 blk_partition_remap(bio
);
3191 if (old_sector
!= -1)
3192 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3195 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3197 old_sector
= bio
->bi_sector
;
3198 old_dev
= bio
->bi_bdev
->bd_dev
;
3200 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3202 sector_t sector
= bio
->bi_sector
;
3204 if (maxsector
< nr_sectors
||
3205 maxsector
- nr_sectors
< sector
) {
3207 * This may well happen - partitions are not
3208 * checked to make sure they are within the size
3209 * of the whole device.
3211 handle_bad_sector(bio
);
3216 ret
= q
->make_request_fn(q
, bio
);
3220 EXPORT_SYMBOL(generic_make_request
);
3223 * submit_bio: submit a bio to the block device layer for I/O
3224 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3225 * @bio: The &struct bio which describes the I/O
3227 * submit_bio() is very similar in purpose to generic_make_request(), and
3228 * uses that function to do most of the work. Both are fairly rough
3229 * interfaces, @bio must be presetup and ready for I/O.
3232 void submit_bio(int rw
, struct bio
*bio
)
3234 int count
= bio_sectors(bio
);
3236 BIO_BUG_ON(!bio
->bi_size
);
3237 BIO_BUG_ON(!bio
->bi_io_vec
);
3240 count_vm_events(PGPGOUT
, count
);
3242 task_io_account_read(bio
->bi_size
);
3243 count_vm_events(PGPGIN
, count
);
3246 if (unlikely(block_dump
)) {
3247 char b
[BDEVNAME_SIZE
];
3248 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3249 current
->comm
, current
->pid
,
3250 (rw
& WRITE
) ? "WRITE" : "READ",
3251 (unsigned long long)bio
->bi_sector
,
3252 bdevname(bio
->bi_bdev
,b
));
3255 generic_make_request(bio
);
3258 EXPORT_SYMBOL(submit_bio
);
3260 static void blk_recalc_rq_segments(struct request
*rq
)
3262 struct bio
*bio
, *prevbio
= NULL
;
3263 int nr_phys_segs
, nr_hw_segs
;
3264 unsigned int phys_size
, hw_size
;
3265 request_queue_t
*q
= rq
->q
;
3270 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3271 rq_for_each_bio(bio
, rq
) {
3272 /* Force bio hw/phys segs to be recalculated. */
3273 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3275 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3276 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3278 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3279 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3281 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3282 pseg
<= q
->max_segment_size
) {
3284 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3288 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3289 hseg
<= q
->max_segment_size
) {
3291 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3298 rq
->nr_phys_segments
= nr_phys_segs
;
3299 rq
->nr_hw_segments
= nr_hw_segs
;
3302 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3304 if (blk_fs_request(rq
)) {
3305 rq
->hard_sector
+= nsect
;
3306 rq
->hard_nr_sectors
-= nsect
;
3309 * Move the I/O submission pointers ahead if required.
3311 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3312 (rq
->sector
<= rq
->hard_sector
)) {
3313 rq
->sector
= rq
->hard_sector
;
3314 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3315 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3316 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3317 rq
->buffer
= bio_data(rq
->bio
);
3321 * if total number of sectors is less than the first segment
3322 * size, something has gone terribly wrong
3324 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3325 printk("blk: request botched\n");
3326 rq
->nr_sectors
= rq
->current_nr_sectors
;
3331 static int __end_that_request_first(struct request
*req
, int uptodate
,
3334 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3337 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3340 * extend uptodate bool to allow < 0 value to be direct io error
3343 if (end_io_error(uptodate
))
3344 error
= !uptodate
? -EIO
: uptodate
;
3347 * for a REQ_BLOCK_PC request, we want to carry any eventual
3348 * sense key with us all the way through
3350 if (!blk_pc_request(req
))
3354 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3355 printk("end_request: I/O error, dev %s, sector %llu\n",
3356 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3357 (unsigned long long)req
->sector
);
3360 if (blk_fs_request(req
) && req
->rq_disk
) {
3361 const int rw
= rq_data_dir(req
);
3363 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3366 total_bytes
= bio_nbytes
= 0;
3367 while ((bio
= req
->bio
) != NULL
) {
3370 if (nr_bytes
>= bio
->bi_size
) {
3371 req
->bio
= bio
->bi_next
;
3372 nbytes
= bio
->bi_size
;
3373 if (!ordered_bio_endio(req
, bio
, nbytes
, error
))
3374 bio_endio(bio
, nbytes
, error
);
3378 int idx
= bio
->bi_idx
+ next_idx
;
3380 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3381 blk_dump_rq_flags(req
, "__end_that");
3382 printk("%s: bio idx %d >= vcnt %d\n",
3384 bio
->bi_idx
, bio
->bi_vcnt
);
3388 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3389 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3392 * not a complete bvec done
3394 if (unlikely(nbytes
> nr_bytes
)) {
3395 bio_nbytes
+= nr_bytes
;
3396 total_bytes
+= nr_bytes
;
3401 * advance to the next vector
3404 bio_nbytes
+= nbytes
;
3407 total_bytes
+= nbytes
;
3410 if ((bio
= req
->bio
)) {
3412 * end more in this run, or just return 'not-done'
3414 if (unlikely(nr_bytes
<= 0))
3426 * if the request wasn't completed, update state
3429 if (!ordered_bio_endio(req
, bio
, bio_nbytes
, error
))
3430 bio_endio(bio
, bio_nbytes
, error
);
3431 bio
->bi_idx
+= next_idx
;
3432 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3433 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3436 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3437 blk_recalc_rq_segments(req
);
3442 * end_that_request_first - end I/O on a request
3443 * @req: the request being processed
3444 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3445 * @nr_sectors: number of sectors to end I/O on
3448 * Ends I/O on a number of sectors attached to @req, and sets it up
3449 * for the next range of segments (if any) in the cluster.
3452 * 0 - we are done with this request, call end_that_request_last()
3453 * 1 - still buffers pending for this request
3455 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3457 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3460 EXPORT_SYMBOL(end_that_request_first
);
3463 * end_that_request_chunk - end I/O on a request
3464 * @req: the request being processed
3465 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3466 * @nr_bytes: number of bytes to complete
3469 * Ends I/O on a number of bytes attached to @req, and sets it up
3470 * for the next range of segments (if any). Like end_that_request_first(),
3471 * but deals with bytes instead of sectors.
3474 * 0 - we are done with this request, call end_that_request_last()
3475 * 1 - still buffers pending for this request
3477 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3479 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3482 EXPORT_SYMBOL(end_that_request_chunk
);
3485 * splice the completion data to a local structure and hand off to
3486 * process_completion_queue() to complete the requests
3488 static void blk_done_softirq(struct softirq_action
*h
)
3490 struct list_head
*cpu_list
, local_list
;
3492 local_irq_disable();
3493 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3494 list_replace_init(cpu_list
, &local_list
);
3497 while (!list_empty(&local_list
)) {
3498 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3500 list_del_init(&rq
->donelist
);
3501 rq
->q
->softirq_done_fn(rq
);
3505 static int blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3509 * If a CPU goes away, splice its entries to the current CPU
3510 * and trigger a run of the softirq
3512 if (action
== CPU_DEAD
) {
3513 int cpu
= (unsigned long) hcpu
;
3515 local_irq_disable();
3516 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3517 &__get_cpu_var(blk_cpu_done
));
3518 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3526 static struct notifier_block __devinitdata blk_cpu_notifier
= {
3527 .notifier_call
= blk_cpu_notify
,
3531 * blk_complete_request - end I/O on a request
3532 * @req: the request being processed
3535 * Ends all I/O on a request. It does not handle partial completions,
3536 * unless the driver actually implements this in its completion callback
3537 * through requeueing. Theh actual completion happens out-of-order,
3538 * through a softirq handler. The user must have registered a completion
3539 * callback through blk_queue_softirq_done().
3542 void blk_complete_request(struct request
*req
)
3544 struct list_head
*cpu_list
;
3545 unsigned long flags
;
3547 BUG_ON(!req
->q
->softirq_done_fn
);
3549 local_irq_save(flags
);
3551 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3552 list_add_tail(&req
->donelist
, cpu_list
);
3553 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3555 local_irq_restore(flags
);
3558 EXPORT_SYMBOL(blk_complete_request
);
3561 * queue lock must be held
3563 void end_that_request_last(struct request
*req
, int uptodate
)
3565 struct gendisk
*disk
= req
->rq_disk
;
3569 * extend uptodate bool to allow < 0 value to be direct io error
3572 if (end_io_error(uptodate
))
3573 error
= !uptodate
? -EIO
: uptodate
;
3575 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3576 laptop_io_completion();
3579 * Account IO completion. bar_rq isn't accounted as a normal
3580 * IO on queueing nor completion. Accounting the containing
3581 * request is enough.
3583 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3584 unsigned long duration
= jiffies
- req
->start_time
;
3585 const int rw
= rq_data_dir(req
);
3587 __disk_stat_inc(disk
, ios
[rw
]);
3588 __disk_stat_add(disk
, ticks
[rw
], duration
);
3589 disk_round_stats(disk
);
3593 req
->end_io(req
, error
);
3595 __blk_put_request(req
->q
, req
);
3598 EXPORT_SYMBOL(end_that_request_last
);
3600 void end_request(struct request
*req
, int uptodate
)
3602 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3603 add_disk_randomness(req
->rq_disk
);
3604 blkdev_dequeue_request(req
);
3605 end_that_request_last(req
, uptodate
);
3609 EXPORT_SYMBOL(end_request
);
3611 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3613 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3614 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3616 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3617 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3618 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3619 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3620 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3621 rq
->buffer
= bio_data(bio
);
3622 rq
->data_len
= bio
->bi_size
;
3624 rq
->bio
= rq
->biotail
= bio
;
3627 EXPORT_SYMBOL(blk_rq_bio_prep
);
3629 int kblockd_schedule_work(struct work_struct
*work
)
3631 return queue_work(kblockd_workqueue
, work
);
3634 EXPORT_SYMBOL(kblockd_schedule_work
);
3636 void kblockd_flush(void)
3638 flush_workqueue(kblockd_workqueue
);
3640 EXPORT_SYMBOL(kblockd_flush
);
3642 int __init
blk_dev_init(void)
3646 kblockd_workqueue
= create_workqueue("kblockd");
3647 if (!kblockd_workqueue
)
3648 panic("Failed to create kblockd\n");
3650 request_cachep
= kmem_cache_create("blkdev_requests",
3651 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3653 requestq_cachep
= kmem_cache_create("blkdev_queue",
3654 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3656 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3657 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3659 for_each_possible_cpu(i
)
3660 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3662 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3663 register_hotcpu_notifier(&blk_cpu_notifier
);
3665 blk_max_low_pfn
= max_low_pfn
;
3666 blk_max_pfn
= max_pfn
;
3672 * IO Context helper functions
3674 void put_io_context(struct io_context
*ioc
)
3679 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3681 if (atomic_dec_and_test(&ioc
->refcount
)) {
3682 struct cfq_io_context
*cic
;
3685 if (ioc
->aic
&& ioc
->aic
->dtor
)
3686 ioc
->aic
->dtor(ioc
->aic
);
3687 if (ioc
->cic_root
.rb_node
!= NULL
) {
3688 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3690 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3695 kmem_cache_free(iocontext_cachep
, ioc
);
3698 EXPORT_SYMBOL(put_io_context
);
3700 /* Called by the exitting task */
3701 void exit_io_context(void)
3703 struct io_context
*ioc
;
3704 struct cfq_io_context
*cic
;
3707 ioc
= current
->io_context
;
3708 current
->io_context
= NULL
;
3709 task_unlock(current
);
3712 if (ioc
->aic
&& ioc
->aic
->exit
)
3713 ioc
->aic
->exit(ioc
->aic
);
3714 if (ioc
->cic_root
.rb_node
!= NULL
) {
3715 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3719 put_io_context(ioc
);
3723 * If the current task has no IO context then create one and initialise it.
3724 * Otherwise, return its existing IO context.
3726 * This returned IO context doesn't have a specifically elevated refcount,
3727 * but since the current task itself holds a reference, the context can be
3728 * used in general code, so long as it stays within `current` context.
3730 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3732 struct task_struct
*tsk
= current
;
3733 struct io_context
*ret
;
3735 ret
= tsk
->io_context
;
3739 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3741 atomic_set(&ret
->refcount
, 1);
3742 ret
->task
= current
;
3743 ret
->ioprio_changed
= 0;
3744 ret
->last_waited
= jiffies
; /* doesn't matter... */
3745 ret
->nr_batch_requests
= 0; /* because this is 0 */
3747 ret
->cic_root
.rb_node
= NULL
;
3748 /* make sure set_task_ioprio() sees the settings above */
3750 tsk
->io_context
= ret
;
3755 EXPORT_SYMBOL(current_io_context
);
3758 * If the current task has no IO context then create one and initialise it.
3759 * If it does have a context, take a ref on it.
3761 * This is always called in the context of the task which submitted the I/O.
3763 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3765 struct io_context
*ret
;
3766 ret
= current_io_context(gfp_flags
, node
);
3768 atomic_inc(&ret
->refcount
);
3771 EXPORT_SYMBOL(get_io_context
);
3773 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3775 struct io_context
*src
= *psrc
;
3776 struct io_context
*dst
= *pdst
;
3779 BUG_ON(atomic_read(&src
->refcount
) == 0);
3780 atomic_inc(&src
->refcount
);
3781 put_io_context(dst
);
3785 EXPORT_SYMBOL(copy_io_context
);
3787 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3789 struct io_context
*temp
;
3794 EXPORT_SYMBOL(swap_io_context
);
3799 struct queue_sysfs_entry
{
3800 struct attribute attr
;
3801 ssize_t (*show
)(struct request_queue
*, char *);
3802 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3806 queue_var_show(unsigned int var
, char *page
)
3808 return sprintf(page
, "%d\n", var
);
3812 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3814 char *p
= (char *) page
;
3816 *var
= simple_strtoul(p
, &p
, 10);
3820 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3822 return queue_var_show(q
->nr_requests
, (page
));
3826 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3828 struct request_list
*rl
= &q
->rq
;
3830 int ret
= queue_var_store(&nr
, page
, count
);
3831 if (nr
< BLKDEV_MIN_RQ
)
3834 spin_lock_irq(q
->queue_lock
);
3835 q
->nr_requests
= nr
;
3836 blk_queue_congestion_threshold(q
);
3838 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3839 blk_set_queue_congested(q
, READ
);
3840 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3841 blk_clear_queue_congested(q
, READ
);
3843 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3844 blk_set_queue_congested(q
, WRITE
);
3845 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3846 blk_clear_queue_congested(q
, WRITE
);
3848 if (rl
->count
[READ
] >= q
->nr_requests
) {
3849 blk_set_queue_full(q
, READ
);
3850 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3851 blk_clear_queue_full(q
, READ
);
3852 wake_up(&rl
->wait
[READ
]);
3855 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3856 blk_set_queue_full(q
, WRITE
);
3857 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3858 blk_clear_queue_full(q
, WRITE
);
3859 wake_up(&rl
->wait
[WRITE
]);
3861 spin_unlock_irq(q
->queue_lock
);
3865 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3867 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3869 return queue_var_show(ra_kb
, (page
));
3873 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3875 unsigned long ra_kb
;
3876 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3878 spin_lock_irq(q
->queue_lock
);
3879 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3880 spin_unlock_irq(q
->queue_lock
);
3885 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3887 int max_sectors_kb
= q
->max_sectors
>> 1;
3889 return queue_var_show(max_sectors_kb
, (page
));
3893 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3895 unsigned long max_sectors_kb
,
3896 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3897 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3898 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3901 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3904 * Take the queue lock to update the readahead and max_sectors
3905 * values synchronously:
3907 spin_lock_irq(q
->queue_lock
);
3909 * Trim readahead window as well, if necessary:
3911 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3912 if (ra_kb
> max_sectors_kb
)
3913 q
->backing_dev_info
.ra_pages
=
3914 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3916 q
->max_sectors
= max_sectors_kb
<< 1;
3917 spin_unlock_irq(q
->queue_lock
);
3922 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3924 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3926 return queue_var_show(max_hw_sectors_kb
, (page
));
3930 static struct queue_sysfs_entry queue_requests_entry
= {
3931 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3932 .show
= queue_requests_show
,
3933 .store
= queue_requests_store
,
3936 static struct queue_sysfs_entry queue_ra_entry
= {
3937 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3938 .show
= queue_ra_show
,
3939 .store
= queue_ra_store
,
3942 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3943 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3944 .show
= queue_max_sectors_show
,
3945 .store
= queue_max_sectors_store
,
3948 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3949 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3950 .show
= queue_max_hw_sectors_show
,
3953 static struct queue_sysfs_entry queue_iosched_entry
= {
3954 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3955 .show
= elv_iosched_show
,
3956 .store
= elv_iosched_store
,
3959 static struct attribute
*default_attrs
[] = {
3960 &queue_requests_entry
.attr
,
3961 &queue_ra_entry
.attr
,
3962 &queue_max_hw_sectors_entry
.attr
,
3963 &queue_max_sectors_entry
.attr
,
3964 &queue_iosched_entry
.attr
,
3968 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3971 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3973 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3974 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
3979 mutex_lock(&q
->sysfs_lock
);
3980 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
3981 mutex_unlock(&q
->sysfs_lock
);
3984 res
= entry
->show(q
, page
);
3985 mutex_unlock(&q
->sysfs_lock
);
3990 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3991 const char *page
, size_t length
)
3993 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3994 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
4000 mutex_lock(&q
->sysfs_lock
);
4001 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4002 mutex_unlock(&q
->sysfs_lock
);
4005 res
= entry
->store(q
, page
, length
);
4006 mutex_unlock(&q
->sysfs_lock
);
4010 static struct sysfs_ops queue_sysfs_ops
= {
4011 .show
= queue_attr_show
,
4012 .store
= queue_attr_store
,
4015 static struct kobj_type queue_ktype
= {
4016 .sysfs_ops
= &queue_sysfs_ops
,
4017 .default_attrs
= default_attrs
,
4018 .release
= blk_release_queue
,
4021 int blk_register_queue(struct gendisk
*disk
)
4025 request_queue_t
*q
= disk
->queue
;
4027 if (!q
|| !q
->request_fn
)
4030 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4032 ret
= kobject_add(&q
->kobj
);
4036 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4038 ret
= elv_register_queue(q
);
4040 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4041 kobject_del(&q
->kobj
);
4048 void blk_unregister_queue(struct gendisk
*disk
)
4050 request_queue_t
*q
= disk
->queue
;
4052 if (q
&& q
->request_fn
) {
4053 elv_unregister_queue(q
);
4055 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4056 kobject_del(&q
->kobj
);
4057 kobject_put(&disk
->kobj
);