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/interrupt.h>
29 #include <linux/cpu.h>
30 #include <linux/blktrace_api.h>
35 #include <scsi/scsi_cmnd.h>
37 static void blk_unplug_work(void *data
);
38 static void blk_unplug_timeout(unsigned long data
);
39 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
40 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
41 static int __make_request(request_queue_t
*q
, struct bio
*bio
);
44 * For the allocated request tables
46 static kmem_cache_t
*request_cachep
;
49 * For queue allocation
51 static kmem_cache_t
*requestq_cachep
;
54 * For io context allocations
56 static kmem_cache_t
*iocontext_cachep
;
58 static wait_queue_head_t congestion_wqh
[2] = {
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
64 * Controlling structure to kblockd
66 static struct workqueue_struct
*kblockd_workqueue
;
68 unsigned long blk_max_low_pfn
, blk_max_pfn
;
70 EXPORT_SYMBOL(blk_max_low_pfn
);
71 EXPORT_SYMBOL(blk_max_pfn
);
73 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
75 /* Amount of time in which a process may batch requests */
76 #define BLK_BATCH_TIME (HZ/50UL)
78 /* Number of requests a "batching" process may submit */
79 #define BLK_BATCH_REQ 32
82 * Return the threshold (number of used requests) at which the queue is
83 * considered to be congested. It include a little hysteresis to keep the
84 * context switch rate down.
86 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
88 return q
->nr_congestion_on
;
92 * The threshold at which a queue is considered to be uncongested
94 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
96 return q
->nr_congestion_off
;
99 static void blk_queue_congestion_threshold(struct request_queue
*q
)
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
104 if (nr
> q
->nr_requests
)
106 q
->nr_congestion_on
= nr
;
108 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
111 q
->nr_congestion_off
= nr
;
115 * A queue has just exitted congestion. Note this in the global counter of
116 * congested queues, and wake up anyone who was waiting for requests to be
119 static void clear_queue_congested(request_queue_t
*q
, int rw
)
122 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
124 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
125 clear_bit(bit
, &q
->backing_dev_info
.state
);
126 smp_mb__after_clear_bit();
127 if (waitqueue_active(wqh
))
132 * A queue has just entered congestion. Flag that in the queue's VM-visible
133 * state flags and increment the global gounter of congested queues.
135 static void set_queue_congested(request_queue_t
*q
, int rw
)
139 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
140 set_bit(bit
, &q
->backing_dev_info
.state
);
144 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
147 * Locates the passed device's request queue and returns the address of its
150 * Will return NULL if the request queue cannot be located.
152 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
154 struct backing_dev_info
*ret
= NULL
;
155 request_queue_t
*q
= bdev_get_queue(bdev
);
158 ret
= &q
->backing_dev_info
;
162 EXPORT_SYMBOL(blk_get_backing_dev_info
);
164 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
167 q
->activity_data
= data
;
170 EXPORT_SYMBOL(blk_queue_activity_fn
);
173 * blk_queue_prep_rq - set a prepare_request function for queue
175 * @pfn: prepare_request function
177 * It's possible for a queue to register a prepare_request callback which
178 * is invoked before the request is handed to the request_fn. The goal of
179 * the function is to prepare a request for I/O, it can be used to build a
180 * cdb from the request data for instance.
183 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
188 EXPORT_SYMBOL(blk_queue_prep_rq
);
191 * blk_queue_merge_bvec - set a merge_bvec function for queue
193 * @mbfn: merge_bvec_fn
195 * Usually queues have static limitations on the max sectors or segments that
196 * we can put in a request. Stacking drivers may have some settings that
197 * are dynamic, and thus we have to query the queue whether it is ok to
198 * add a new bio_vec to a bio at a given offset or not. If the block device
199 * has such limitations, it needs to register a merge_bvec_fn to control
200 * the size of bio's sent to it. Note that a block device *must* allow a
201 * single page to be added to an empty bio. The block device driver may want
202 * to use the bio_split() function to deal with these bio's. By default
203 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
206 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
208 q
->merge_bvec_fn
= mbfn
;
211 EXPORT_SYMBOL(blk_queue_merge_bvec
);
213 void blk_queue_softirq_done(request_queue_t
*q
, softirq_done_fn
*fn
)
215 q
->softirq_done_fn
= fn
;
218 EXPORT_SYMBOL(blk_queue_softirq_done
);
221 * blk_queue_make_request - define an alternate make_request function for a device
222 * @q: the request queue for the device to be affected
223 * @mfn: the alternate make_request function
226 * The normal way for &struct bios to be passed to a device
227 * driver is for them to be collected into requests on a request
228 * queue, and then to allow the device driver to select requests
229 * off that queue when it is ready. This works well for many block
230 * devices. However some block devices (typically virtual devices
231 * such as md or lvm) do not benefit from the processing on the
232 * request queue, and are served best by having the requests passed
233 * directly to them. This can be achieved by providing a function
234 * to blk_queue_make_request().
237 * The driver that does this *must* be able to deal appropriately
238 * with buffers in "highmemory". This can be accomplished by either calling
239 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
240 * blk_queue_bounce() to create a buffer in normal memory.
242 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
247 q
->nr_requests
= BLKDEV_MAX_RQ
;
248 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
249 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
250 q
->make_request_fn
= mfn
;
251 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
252 q
->backing_dev_info
.state
= 0;
253 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
254 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
255 blk_queue_hardsect_size(q
, 512);
256 blk_queue_dma_alignment(q
, 511);
257 blk_queue_congestion_threshold(q
);
258 q
->nr_batching
= BLK_BATCH_REQ
;
260 q
->unplug_thresh
= 4; /* hmm */
261 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
262 if (q
->unplug_delay
== 0)
265 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
267 q
->unplug_timer
.function
= blk_unplug_timeout
;
268 q
->unplug_timer
.data
= (unsigned long)q
;
271 * by default assume old behaviour and bounce for any highmem page
273 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
275 blk_queue_activity_fn(q
, NULL
, NULL
);
278 EXPORT_SYMBOL(blk_queue_make_request
);
280 static inline void rq_init(request_queue_t
*q
, struct request
*rq
)
282 INIT_LIST_HEAD(&rq
->queuelist
);
283 INIT_LIST_HEAD(&rq
->donelist
);
286 rq
->rq_status
= RQ_ACTIVE
;
287 rq
->bio
= rq
->biotail
= NULL
;
296 rq
->nr_phys_segments
= 0;
299 rq
->end_io_data
= NULL
;
300 rq
->completion_data
= NULL
;
304 * blk_queue_ordered - does this queue support ordered writes
305 * @q: the request queue
306 * @ordered: one of QUEUE_ORDERED_*
307 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
310 * For journalled file systems, doing ordered writes on a commit
311 * block instead of explicitly doing wait_on_buffer (which is bad
312 * for performance) can be a big win. Block drivers supporting this
313 * feature should call this function and indicate so.
316 int blk_queue_ordered(request_queue_t
*q
, unsigned ordered
,
317 prepare_flush_fn
*prepare_flush_fn
)
319 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
320 prepare_flush_fn
== NULL
) {
321 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
325 if (ordered
!= QUEUE_ORDERED_NONE
&&
326 ordered
!= QUEUE_ORDERED_DRAIN
&&
327 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
328 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
329 ordered
!= QUEUE_ORDERED_TAG
&&
330 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
331 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
332 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
336 q
->ordered
= ordered
;
337 q
->next_ordered
= ordered
;
338 q
->prepare_flush_fn
= prepare_flush_fn
;
343 EXPORT_SYMBOL(blk_queue_ordered
);
346 * blk_queue_issue_flush_fn - set function for issuing a flush
347 * @q: the request queue
348 * @iff: the function to be called issuing the flush
351 * If a driver supports issuing a flush command, the support is notified
352 * to the block layer by defining it through this call.
355 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
357 q
->issue_flush_fn
= iff
;
360 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
363 * Cache flushing for ordered writes handling
365 inline unsigned blk_ordered_cur_seq(request_queue_t
*q
)
369 return 1 << ffz(q
->ordseq
);
372 unsigned blk_ordered_req_seq(struct request
*rq
)
374 request_queue_t
*q
= rq
->q
;
376 BUG_ON(q
->ordseq
== 0);
378 if (rq
== &q
->pre_flush_rq
)
379 return QUEUE_ORDSEQ_PREFLUSH
;
380 if (rq
== &q
->bar_rq
)
381 return QUEUE_ORDSEQ_BAR
;
382 if (rq
== &q
->post_flush_rq
)
383 return QUEUE_ORDSEQ_POSTFLUSH
;
385 if ((rq
->flags
& REQ_ORDERED_COLOR
) ==
386 (q
->orig_bar_rq
->flags
& REQ_ORDERED_COLOR
))
387 return QUEUE_ORDSEQ_DRAIN
;
389 return QUEUE_ORDSEQ_DONE
;
392 void blk_ordered_complete_seq(request_queue_t
*q
, unsigned seq
, int error
)
397 if (error
&& !q
->orderr
)
400 BUG_ON(q
->ordseq
& seq
);
403 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
407 * Okay, sequence complete.
410 uptodate
= q
->orderr
? q
->orderr
: 1;
414 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
415 end_that_request_last(rq
, uptodate
);
418 static void pre_flush_end_io(struct request
*rq
, int error
)
420 elv_completed_request(rq
->q
, rq
);
421 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
424 static void bar_end_io(struct request
*rq
, int error
)
426 elv_completed_request(rq
->q
, rq
);
427 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
430 static void post_flush_end_io(struct request
*rq
, int error
)
432 elv_completed_request(rq
->q
, rq
);
433 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
436 static void queue_flush(request_queue_t
*q
, unsigned which
)
439 rq_end_io_fn
*end_io
;
441 if (which
== QUEUE_ORDERED_PREFLUSH
) {
442 rq
= &q
->pre_flush_rq
;
443 end_io
= pre_flush_end_io
;
445 rq
= &q
->post_flush_rq
;
446 end_io
= post_flush_end_io
;
450 rq
->flags
= REQ_HARDBARRIER
;
451 rq
->elevator_private
= NULL
;
452 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
455 q
->prepare_flush_fn(q
, rq
);
457 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
460 static inline struct request
*start_ordered(request_queue_t
*q
,
465 q
->ordered
= q
->next_ordered
;
466 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
469 * Prep proxy barrier request.
471 blkdev_dequeue_request(rq
);
475 rq
->flags
= bio_data_dir(q
->orig_bar_rq
->bio
);
476 rq
->flags
|= q
->ordered
& QUEUE_ORDERED_FUA
? REQ_FUA
: 0;
477 rq
->elevator_private
= NULL
;
479 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
480 rq
->end_io
= bar_end_io
;
483 * Queue ordered sequence. As we stack them at the head, we
484 * need to queue in reverse order. Note that we rely on that
485 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
486 * request gets inbetween ordered sequence.
488 if (q
->ordered
& QUEUE_ORDERED_POSTFLUSH
)
489 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
491 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
493 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
495 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
496 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
497 rq
= &q
->pre_flush_rq
;
499 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
501 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
502 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
509 int blk_do_ordered(request_queue_t
*q
, struct request
**rqp
)
511 struct request
*rq
= *rqp
;
512 int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
518 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
519 *rqp
= start_ordered(q
, rq
);
523 * This can happen when the queue switches to
524 * ORDERED_NONE while this request is on it.
526 blkdev_dequeue_request(rq
);
527 end_that_request_first(rq
, -EOPNOTSUPP
,
528 rq
->hard_nr_sectors
);
529 end_that_request_last(rq
, -EOPNOTSUPP
);
536 * Ordered sequence in progress
539 /* Special requests are not subject to ordering rules. */
540 if (!blk_fs_request(rq
) &&
541 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
544 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
545 /* Ordered by tag. Blocking the next barrier is enough. */
546 if (is_barrier
&& rq
!= &q
->bar_rq
)
549 /* Ordered by draining. Wait for turn. */
550 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
551 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
558 static int flush_dry_bio_endio(struct bio
*bio
, unsigned int bytes
, int error
)
560 request_queue_t
*q
= bio
->bi_private
;
561 struct bio_vec
*bvec
;
565 * This is dry run, restore bio_sector and size. We'll finish
566 * this request again with the original bi_end_io after an
567 * error occurs or post flush is complete.
576 bio_for_each_segment(bvec
, bio
, i
) {
577 bvec
->bv_len
+= bvec
->bv_offset
;
582 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
583 bio
->bi_size
= q
->bi_size
;
584 bio
->bi_sector
-= (q
->bi_size
>> 9);
590 static inline int ordered_bio_endio(struct request
*rq
, struct bio
*bio
,
591 unsigned int nbytes
, int error
)
593 request_queue_t
*q
= rq
->q
;
597 if (&q
->bar_rq
!= rq
)
601 * Okay, this is the barrier request in progress, dry finish it.
603 if (error
&& !q
->orderr
)
606 endio
= bio
->bi_end_io
;
607 private = bio
->bi_private
;
608 bio
->bi_end_io
= flush_dry_bio_endio
;
611 bio_endio(bio
, nbytes
, error
);
613 bio
->bi_end_io
= endio
;
614 bio
->bi_private
= private;
620 * blk_queue_bounce_limit - set bounce buffer limit for queue
621 * @q: the request queue for the device
622 * @dma_addr: bus address limit
625 * Different hardware can have different requirements as to what pages
626 * it can do I/O directly to. A low level driver can call
627 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
628 * buffers for doing I/O to pages residing above @page.
630 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
632 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
635 q
->bounce_gfp
= GFP_NOIO
;
636 #if BITS_PER_LONG == 64
637 /* Assume anything <= 4GB can be handled by IOMMU.
638 Actually some IOMMUs can handle everything, but I don't
639 know of a way to test this here. */
640 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
642 q
->bounce_pfn
= max_low_pfn
;
644 if (bounce_pfn
< blk_max_low_pfn
)
646 q
->bounce_pfn
= bounce_pfn
;
649 init_emergency_isa_pool();
650 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
651 q
->bounce_pfn
= bounce_pfn
;
655 EXPORT_SYMBOL(blk_queue_bounce_limit
);
658 * blk_queue_max_sectors - set max sectors for a request for this queue
659 * @q: the request queue for the device
660 * @max_sectors: max sectors in the usual 512b unit
663 * Enables a low level driver to set an upper limit on the size of
666 void blk_queue_max_sectors(request_queue_t
*q
, unsigned int max_sectors
)
668 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
669 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
670 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
673 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
674 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
676 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
677 q
->max_hw_sectors
= max_sectors
;
681 EXPORT_SYMBOL(blk_queue_max_sectors
);
684 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
685 * @q: the request queue for the device
686 * @max_segments: max number of segments
689 * Enables a low level driver to set an upper limit on the number of
690 * physical data segments in a request. This would be the largest sized
691 * scatter list the driver could handle.
693 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
697 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
700 q
->max_phys_segments
= max_segments
;
703 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
706 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
707 * @q: the request queue for the device
708 * @max_segments: max number of segments
711 * Enables a low level driver to set an upper limit on the number of
712 * hw data segments in a request. This would be the largest number of
713 * address/length pairs the host adapter can actually give as once
716 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
720 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
723 q
->max_hw_segments
= max_segments
;
726 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
729 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
730 * @q: the request queue for the device
731 * @max_size: max size of segment in bytes
734 * Enables a low level driver to set an upper limit on the size of a
737 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
739 if (max_size
< PAGE_CACHE_SIZE
) {
740 max_size
= PAGE_CACHE_SIZE
;
741 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
744 q
->max_segment_size
= max_size
;
747 EXPORT_SYMBOL(blk_queue_max_segment_size
);
750 * blk_queue_hardsect_size - set hardware sector size for the queue
751 * @q: the request queue for the device
752 * @size: the hardware sector size, in bytes
755 * This should typically be set to the lowest possible sector size
756 * that the hardware can operate on (possible without reverting to
757 * even internal read-modify-write operations). Usually the default
758 * of 512 covers most hardware.
760 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
762 q
->hardsect_size
= size
;
765 EXPORT_SYMBOL(blk_queue_hardsect_size
);
768 * Returns the minimum that is _not_ zero, unless both are zero.
770 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
773 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
774 * @t: the stacking driver (top)
775 * @b: the underlying device (bottom)
777 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
779 /* zero is "infinity" */
780 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
781 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
783 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
784 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
785 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
786 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
787 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
788 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
791 EXPORT_SYMBOL(blk_queue_stack_limits
);
794 * blk_queue_segment_boundary - set boundary rules for segment merging
795 * @q: the request queue for the device
796 * @mask: the memory boundary mask
798 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
800 if (mask
< PAGE_CACHE_SIZE
- 1) {
801 mask
= PAGE_CACHE_SIZE
- 1;
802 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
805 q
->seg_boundary_mask
= mask
;
808 EXPORT_SYMBOL(blk_queue_segment_boundary
);
811 * blk_queue_dma_alignment - set dma length and memory alignment
812 * @q: the request queue for the device
813 * @mask: alignment mask
816 * set required memory and length aligment for direct dma transactions.
817 * this is used when buiding direct io requests for the queue.
820 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
822 q
->dma_alignment
= mask
;
825 EXPORT_SYMBOL(blk_queue_dma_alignment
);
828 * blk_queue_find_tag - find a request by its tag and queue
829 * @q: The request queue for the device
830 * @tag: The tag of the request
833 * Should be used when a device returns a tag and you want to match
836 * no locks need be held.
838 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
840 struct blk_queue_tag
*bqt
= q
->queue_tags
;
842 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
845 return bqt
->tag_index
[tag
];
848 EXPORT_SYMBOL(blk_queue_find_tag
);
851 * __blk_free_tags - release a given set of tag maintenance info
852 * @bqt: the tag map to free
854 * Tries to free the specified @bqt@. Returns true if it was
855 * actually freed and false if there are still references using it
857 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
861 retval
= atomic_dec_and_test(&bqt
->refcnt
);
864 BUG_ON(!list_empty(&bqt
->busy_list
));
866 kfree(bqt
->tag_index
);
867 bqt
->tag_index
= NULL
;
880 * __blk_queue_free_tags - release tag maintenance info
881 * @q: the request queue for the device
884 * blk_cleanup_queue() will take care of calling this function, if tagging
885 * has been used. So there's no need to call this directly.
887 static void __blk_queue_free_tags(request_queue_t
*q
)
889 struct blk_queue_tag
*bqt
= q
->queue_tags
;
894 __blk_free_tags(bqt
);
896 q
->queue_tags
= NULL
;
897 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
902 * blk_free_tags - release a given set of tag maintenance info
903 * @bqt: the tag map to free
905 * For externally managed @bqt@ frees the map. Callers of this
906 * function must guarantee to have released all the queues that
907 * might have been using this tag map.
909 void blk_free_tags(struct blk_queue_tag
*bqt
)
911 if (unlikely(!__blk_free_tags(bqt
)))
914 EXPORT_SYMBOL(blk_free_tags
);
917 * blk_queue_free_tags - release tag maintenance info
918 * @q: the request queue for the device
921 * This is used to disabled tagged queuing to a device, yet leave
924 void blk_queue_free_tags(request_queue_t
*q
)
926 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
929 EXPORT_SYMBOL(blk_queue_free_tags
);
932 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
934 struct request
**tag_index
;
935 unsigned long *tag_map
;
938 if (q
&& depth
> q
->nr_requests
* 2) {
939 depth
= q
->nr_requests
* 2;
940 printk(KERN_ERR
"%s: adjusted depth to %d\n",
941 __FUNCTION__
, depth
);
944 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
948 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
949 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
953 tags
->real_max_depth
= depth
;
954 tags
->max_depth
= depth
;
955 tags
->tag_index
= tag_index
;
956 tags
->tag_map
= tag_map
;
964 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
967 struct blk_queue_tag
*tags
;
969 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
973 if (init_tag_map(q
, tags
, depth
))
976 INIT_LIST_HEAD(&tags
->busy_list
);
978 atomic_set(&tags
->refcnt
, 1);
986 * blk_init_tags - initialize the tag info for an external tag map
987 * @depth: the maximum queue depth supported
988 * @tags: the tag to use
990 struct blk_queue_tag
*blk_init_tags(int depth
)
992 return __blk_queue_init_tags(NULL
, depth
);
994 EXPORT_SYMBOL(blk_init_tags
);
997 * blk_queue_init_tags - initialize the queue tag info
998 * @q: the request queue for the device
999 * @depth: the maximum queue depth supported
1000 * @tags: the tag to use
1002 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
1003 struct blk_queue_tag
*tags
)
1007 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
1009 if (!tags
&& !q
->queue_tags
) {
1010 tags
= __blk_queue_init_tags(q
, depth
);
1014 } else if (q
->queue_tags
) {
1015 if ((rc
= blk_queue_resize_tags(q
, depth
)))
1017 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
1020 atomic_inc(&tags
->refcnt
);
1023 * assign it, all done
1025 q
->queue_tags
= tags
;
1026 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
1033 EXPORT_SYMBOL(blk_queue_init_tags
);
1036 * blk_queue_resize_tags - change the queueing depth
1037 * @q: the request queue for the device
1038 * @new_depth: the new max command queueing depth
1041 * Must be called with the queue lock held.
1043 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
1045 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1046 struct request
**tag_index
;
1047 unsigned long *tag_map
;
1048 int max_depth
, nr_ulongs
;
1054 * if we already have large enough real_max_depth. just
1055 * adjust max_depth. *NOTE* as requests with tag value
1056 * between new_depth and real_max_depth can be in-flight, tag
1057 * map can not be shrunk blindly here.
1059 if (new_depth
<= bqt
->real_max_depth
) {
1060 bqt
->max_depth
= new_depth
;
1065 * Currently cannot replace a shared tag map with a new
1066 * one, so error out if this is the case
1068 if (atomic_read(&bqt
->refcnt
) != 1)
1072 * save the old state info, so we can copy it back
1074 tag_index
= bqt
->tag_index
;
1075 tag_map
= bqt
->tag_map
;
1076 max_depth
= bqt
->real_max_depth
;
1078 if (init_tag_map(q
, bqt
, new_depth
))
1081 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1082 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1083 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1090 EXPORT_SYMBOL(blk_queue_resize_tags
);
1093 * blk_queue_end_tag - end tag operations for a request
1094 * @q: the request queue for the device
1095 * @rq: the request that has completed
1098 * Typically called when end_that_request_first() returns 0, meaning
1099 * all transfers have been done for a request. It's important to call
1100 * this function before end_that_request_last(), as that will put the
1101 * request back on the free list thus corrupting the internal tag list.
1104 * queue lock must be held.
1106 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
1108 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1113 if (unlikely(tag
>= bqt
->real_max_depth
))
1115 * This can happen after tag depth has been reduced.
1116 * FIXME: how about a warning or info message here?
1120 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
1121 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1126 list_del_init(&rq
->queuelist
);
1127 rq
->flags
&= ~REQ_QUEUED
;
1130 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1131 printk(KERN_ERR
"%s: tag %d is missing\n",
1134 bqt
->tag_index
[tag
] = NULL
;
1138 EXPORT_SYMBOL(blk_queue_end_tag
);
1141 * blk_queue_start_tag - find a free tag and assign it
1142 * @q: the request queue for the device
1143 * @rq: the block request that needs tagging
1146 * This can either be used as a stand-alone helper, or possibly be
1147 * assigned as the queue &prep_rq_fn (in which case &struct request
1148 * automagically gets a tag assigned). Note that this function
1149 * assumes that any type of request can be queued! if this is not
1150 * true for your device, you must check the request type before
1151 * calling this function. The request will also be removed from
1152 * the request queue, so it's the drivers responsibility to readd
1153 * it if it should need to be restarted for some reason.
1156 * queue lock must be held.
1158 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
1160 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1163 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
1165 "%s: request %p for device [%s] already tagged %d",
1167 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1171 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1172 if (tag
>= bqt
->max_depth
)
1175 __set_bit(tag
, bqt
->tag_map
);
1177 rq
->flags
|= REQ_QUEUED
;
1179 bqt
->tag_index
[tag
] = rq
;
1180 blkdev_dequeue_request(rq
);
1181 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1186 EXPORT_SYMBOL(blk_queue_start_tag
);
1189 * blk_queue_invalidate_tags - invalidate all pending tags
1190 * @q: the request queue for the device
1193 * Hardware conditions may dictate a need to stop all pending requests.
1194 * In this case, we will safely clear the block side of the tag queue and
1195 * readd all requests to the request queue in the right order.
1198 * queue lock must be held.
1200 void blk_queue_invalidate_tags(request_queue_t
*q
)
1202 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1203 struct list_head
*tmp
, *n
;
1206 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1207 rq
= list_entry_rq(tmp
);
1209 if (rq
->tag
== -1) {
1211 "%s: bad tag found on list\n", __FUNCTION__
);
1212 list_del_init(&rq
->queuelist
);
1213 rq
->flags
&= ~REQ_QUEUED
;
1215 blk_queue_end_tag(q
, rq
);
1217 rq
->flags
&= ~REQ_STARTED
;
1218 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1222 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1224 static const char * const rq_flags
[] = {
1245 "REQ_DRIVE_TASKFILE",
1250 "REQ_ORDERED_COLOR",
1253 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1257 printk("%s: dev %s: flags = ", msg
,
1258 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1261 if (rq
->flags
& (1 << bit
))
1262 printk("%s ", rq_flags
[bit
]);
1264 } while (bit
< __REQ_NR_BITS
);
1266 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1268 rq
->current_nr_sectors
);
1269 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1271 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1273 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1274 printk("%02x ", rq
->cmd
[bit
]);
1279 EXPORT_SYMBOL(blk_dump_rq_flags
);
1281 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1283 struct bio_vec
*bv
, *bvprv
= NULL
;
1284 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1285 int high
, highprv
= 1;
1287 if (unlikely(!bio
->bi_io_vec
))
1290 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1291 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1292 bio_for_each_segment(bv
, bio
, i
) {
1294 * the trick here is making sure that a high page is never
1295 * considered part of another segment, since that might
1296 * change with the bounce page.
1298 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1299 if (high
|| highprv
)
1300 goto new_hw_segment
;
1302 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1304 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1306 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1308 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1309 goto new_hw_segment
;
1311 seg_size
+= bv
->bv_len
;
1312 hw_seg_size
+= bv
->bv_len
;
1317 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1318 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1319 hw_seg_size
+= bv
->bv_len
;
1322 if (hw_seg_size
> bio
->bi_hw_front_size
)
1323 bio
->bi_hw_front_size
= hw_seg_size
;
1324 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1330 seg_size
= bv
->bv_len
;
1333 if (hw_seg_size
> bio
->bi_hw_back_size
)
1334 bio
->bi_hw_back_size
= hw_seg_size
;
1335 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1336 bio
->bi_hw_front_size
= hw_seg_size
;
1337 bio
->bi_phys_segments
= nr_phys_segs
;
1338 bio
->bi_hw_segments
= nr_hw_segs
;
1339 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1343 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1346 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1349 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1351 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1355 * bio and nxt are contigous in memory, check if the queue allows
1356 * these two to be merged into one
1358 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1364 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1367 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1368 blk_recount_segments(q
, bio
);
1369 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1370 blk_recount_segments(q
, nxt
);
1371 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1372 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1374 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1381 * map a request to scatterlist, return number of sg entries setup. Caller
1382 * must make sure sg can hold rq->nr_phys_segments entries
1384 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1386 struct bio_vec
*bvec
, *bvprv
;
1388 int nsegs
, i
, cluster
;
1391 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1394 * for each bio in rq
1397 rq_for_each_bio(bio
, rq
) {
1399 * for each segment in bio
1401 bio_for_each_segment(bvec
, bio
, i
) {
1402 int nbytes
= bvec
->bv_len
;
1404 if (bvprv
&& cluster
) {
1405 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1408 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1410 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1413 sg
[nsegs
- 1].length
+= nbytes
;
1416 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1417 sg
[nsegs
].page
= bvec
->bv_page
;
1418 sg
[nsegs
].length
= nbytes
;
1419 sg
[nsegs
].offset
= bvec
->bv_offset
;
1424 } /* segments in bio */
1430 EXPORT_SYMBOL(blk_rq_map_sg
);
1433 * the standard queue merge functions, can be overridden with device
1434 * specific ones if so desired
1437 static inline int ll_new_mergeable(request_queue_t
*q
,
1438 struct request
*req
,
1441 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1443 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1444 req
->flags
|= REQ_NOMERGE
;
1445 if (req
== q
->last_merge
)
1446 q
->last_merge
= NULL
;
1451 * A hw segment is just getting larger, bump just the phys
1454 req
->nr_phys_segments
+= nr_phys_segs
;
1458 static inline int ll_new_hw_segment(request_queue_t
*q
,
1459 struct request
*req
,
1462 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1463 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1465 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1466 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1467 req
->flags
|= REQ_NOMERGE
;
1468 if (req
== q
->last_merge
)
1469 q
->last_merge
= NULL
;
1474 * This will form the start of a new hw segment. Bump both
1477 req
->nr_hw_segments
+= nr_hw_segs
;
1478 req
->nr_phys_segments
+= nr_phys_segs
;
1482 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1485 unsigned short max_sectors
;
1488 if (unlikely(blk_pc_request(req
)))
1489 max_sectors
= q
->max_hw_sectors
;
1491 max_sectors
= q
->max_sectors
;
1493 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1494 req
->flags
|= REQ_NOMERGE
;
1495 if (req
== q
->last_merge
)
1496 q
->last_merge
= NULL
;
1499 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1500 blk_recount_segments(q
, req
->biotail
);
1501 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1502 blk_recount_segments(q
, bio
);
1503 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1504 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1505 !BIOVEC_VIRT_OVERSIZE(len
)) {
1506 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1509 if (req
->nr_hw_segments
== 1)
1510 req
->bio
->bi_hw_front_size
= len
;
1511 if (bio
->bi_hw_segments
== 1)
1512 bio
->bi_hw_back_size
= len
;
1517 return ll_new_hw_segment(q
, req
, bio
);
1520 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1523 unsigned short max_sectors
;
1526 if (unlikely(blk_pc_request(req
)))
1527 max_sectors
= q
->max_hw_sectors
;
1529 max_sectors
= q
->max_sectors
;
1532 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1533 req
->flags
|= REQ_NOMERGE
;
1534 if (req
== q
->last_merge
)
1535 q
->last_merge
= NULL
;
1538 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1539 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1540 blk_recount_segments(q
, bio
);
1541 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1542 blk_recount_segments(q
, req
->bio
);
1543 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1544 !BIOVEC_VIRT_OVERSIZE(len
)) {
1545 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1548 if (bio
->bi_hw_segments
== 1)
1549 bio
->bi_hw_front_size
= len
;
1550 if (req
->nr_hw_segments
== 1)
1551 req
->biotail
->bi_hw_back_size
= len
;
1556 return ll_new_hw_segment(q
, req
, bio
);
1559 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1560 struct request
*next
)
1562 int total_phys_segments
;
1563 int total_hw_segments
;
1566 * First check if the either of the requests are re-queued
1567 * requests. Can't merge them if they are.
1569 if (req
->special
|| next
->special
)
1573 * Will it become too large?
1575 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1578 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1579 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1580 total_phys_segments
--;
1582 if (total_phys_segments
> q
->max_phys_segments
)
1585 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1586 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1587 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1589 * propagate the combined length to the end of the requests
1591 if (req
->nr_hw_segments
== 1)
1592 req
->bio
->bi_hw_front_size
= len
;
1593 if (next
->nr_hw_segments
== 1)
1594 next
->biotail
->bi_hw_back_size
= len
;
1595 total_hw_segments
--;
1598 if (total_hw_segments
> q
->max_hw_segments
)
1601 /* Merge is OK... */
1602 req
->nr_phys_segments
= total_phys_segments
;
1603 req
->nr_hw_segments
= total_hw_segments
;
1608 * "plug" the device if there are no outstanding requests: this will
1609 * force the transfer to start only after we have put all the requests
1612 * This is called with interrupts off and no requests on the queue and
1613 * with the queue lock held.
1615 void blk_plug_device(request_queue_t
*q
)
1617 WARN_ON(!irqs_disabled());
1620 * don't plug a stopped queue, it must be paired with blk_start_queue()
1621 * which will restart the queueing
1623 if (blk_queue_stopped(q
))
1626 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1627 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1628 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1632 EXPORT_SYMBOL(blk_plug_device
);
1635 * remove the queue from the plugged list, if present. called with
1636 * queue lock held and interrupts disabled.
1638 int blk_remove_plug(request_queue_t
*q
)
1640 WARN_ON(!irqs_disabled());
1642 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1645 del_timer(&q
->unplug_timer
);
1649 EXPORT_SYMBOL(blk_remove_plug
);
1652 * remove the plug and let it rip..
1654 void __generic_unplug_device(request_queue_t
*q
)
1656 if (unlikely(blk_queue_stopped(q
)))
1659 if (!blk_remove_plug(q
))
1664 EXPORT_SYMBOL(__generic_unplug_device
);
1667 * generic_unplug_device - fire a request queue
1668 * @q: The &request_queue_t in question
1671 * Linux uses plugging to build bigger requests queues before letting
1672 * the device have at them. If a queue is plugged, the I/O scheduler
1673 * is still adding and merging requests on the queue. Once the queue
1674 * gets unplugged, the request_fn defined for the queue is invoked and
1675 * transfers started.
1677 void generic_unplug_device(request_queue_t
*q
)
1679 spin_lock_irq(q
->queue_lock
);
1680 __generic_unplug_device(q
);
1681 spin_unlock_irq(q
->queue_lock
);
1683 EXPORT_SYMBOL(generic_unplug_device
);
1685 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1688 request_queue_t
*q
= bdi
->unplug_io_data
;
1691 * devices don't necessarily have an ->unplug_fn defined
1694 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1695 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1701 static void blk_unplug_work(void *data
)
1703 request_queue_t
*q
= data
;
1705 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1706 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1711 static void blk_unplug_timeout(unsigned long data
)
1713 request_queue_t
*q
= (request_queue_t
*)data
;
1715 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1716 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1718 kblockd_schedule_work(&q
->unplug_work
);
1722 * blk_start_queue - restart a previously stopped queue
1723 * @q: The &request_queue_t in question
1726 * blk_start_queue() will clear the stop flag on the queue, and call
1727 * the request_fn for the queue if it was in a stopped state when
1728 * entered. Also see blk_stop_queue(). Queue lock must be held.
1730 void blk_start_queue(request_queue_t
*q
)
1732 WARN_ON(!irqs_disabled());
1734 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1737 * one level of recursion is ok and is much faster than kicking
1738 * the unplug handling
1740 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1742 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1745 kblockd_schedule_work(&q
->unplug_work
);
1749 EXPORT_SYMBOL(blk_start_queue
);
1752 * blk_stop_queue - stop a queue
1753 * @q: The &request_queue_t in question
1756 * The Linux block layer assumes that a block driver will consume all
1757 * entries on the request queue when the request_fn strategy is called.
1758 * Often this will not happen, because of hardware limitations (queue
1759 * depth settings). If a device driver gets a 'queue full' response,
1760 * or if it simply chooses not to queue more I/O at one point, it can
1761 * call this function to prevent the request_fn from being called until
1762 * the driver has signalled it's ready to go again. This happens by calling
1763 * blk_start_queue() to restart queue operations. Queue lock must be held.
1765 void blk_stop_queue(request_queue_t
*q
)
1768 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1770 EXPORT_SYMBOL(blk_stop_queue
);
1773 * blk_sync_queue - cancel any pending callbacks on a queue
1777 * The block layer may perform asynchronous callback activity
1778 * on a queue, such as calling the unplug function after a timeout.
1779 * A block device may call blk_sync_queue to ensure that any
1780 * such activity is cancelled, thus allowing it to release resources
1781 * the the callbacks might use. The caller must already have made sure
1782 * that its ->make_request_fn will not re-add plugging prior to calling
1786 void blk_sync_queue(struct request_queue
*q
)
1788 del_timer_sync(&q
->unplug_timer
);
1791 EXPORT_SYMBOL(blk_sync_queue
);
1794 * blk_run_queue - run a single device queue
1795 * @q: The queue to run
1797 void blk_run_queue(struct request_queue
*q
)
1799 unsigned long flags
;
1801 spin_lock_irqsave(q
->queue_lock
, flags
);
1805 * Only recurse once to avoid overrunning the stack, let the unplug
1806 * handling reinvoke the handler shortly if we already got there.
1808 if (!elv_queue_empty(q
)) {
1809 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1811 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1814 kblockd_schedule_work(&q
->unplug_work
);
1818 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1820 EXPORT_SYMBOL(blk_run_queue
);
1823 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1824 * @kobj: the kobj belonging of the request queue to be released
1827 * blk_cleanup_queue is the pair to blk_init_queue() or
1828 * blk_queue_make_request(). It should be called when a request queue is
1829 * being released; typically when a block device is being de-registered.
1830 * Currently, its primary task it to free all the &struct request
1831 * structures that were allocated to the queue and the queue itself.
1834 * Hopefully the low level driver will have finished any
1835 * outstanding requests first...
1837 static void blk_release_queue(struct kobject
*kobj
)
1839 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
1840 struct request_list
*rl
= &q
->rq
;
1845 mempool_destroy(rl
->rq_pool
);
1848 __blk_queue_free_tags(q
);
1850 blk_trace_shutdown(q
);
1852 kmem_cache_free(requestq_cachep
, q
);
1855 void blk_put_queue(request_queue_t
*q
)
1857 kobject_put(&q
->kobj
);
1859 EXPORT_SYMBOL(blk_put_queue
);
1861 void blk_cleanup_queue(request_queue_t
* q
)
1863 mutex_lock(&q
->sysfs_lock
);
1864 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1865 mutex_unlock(&q
->sysfs_lock
);
1868 elevator_exit(q
->elevator
);
1873 EXPORT_SYMBOL(blk_cleanup_queue
);
1875 static int blk_init_free_list(request_queue_t
*q
)
1877 struct request_list
*rl
= &q
->rq
;
1879 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1880 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1882 init_waitqueue_head(&rl
->wait
[READ
]);
1883 init_waitqueue_head(&rl
->wait
[WRITE
]);
1885 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1886 mempool_free_slab
, request_cachep
, q
->node
);
1894 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1896 return blk_alloc_queue_node(gfp_mask
, -1);
1898 EXPORT_SYMBOL(blk_alloc_queue
);
1900 static struct kobj_type queue_ktype
;
1902 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1906 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1910 memset(q
, 0, sizeof(*q
));
1911 init_timer(&q
->unplug_timer
);
1913 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
1914 q
->kobj
.ktype
= &queue_ktype
;
1915 kobject_init(&q
->kobj
);
1917 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1918 q
->backing_dev_info
.unplug_io_data
= q
;
1920 mutex_init(&q
->sysfs_lock
);
1924 EXPORT_SYMBOL(blk_alloc_queue_node
);
1927 * blk_init_queue - prepare a request queue for use with a block device
1928 * @rfn: The function to be called to process requests that have been
1929 * placed on the queue.
1930 * @lock: Request queue spin lock
1933 * If a block device wishes to use the standard request handling procedures,
1934 * which sorts requests and coalesces adjacent requests, then it must
1935 * call blk_init_queue(). The function @rfn will be called when there
1936 * are requests on the queue that need to be processed. If the device
1937 * supports plugging, then @rfn may not be called immediately when requests
1938 * are available on the queue, but may be called at some time later instead.
1939 * Plugged queues are generally unplugged when a buffer belonging to one
1940 * of the requests on the queue is needed, or due to memory pressure.
1942 * @rfn is not required, or even expected, to remove all requests off the
1943 * queue, but only as many as it can handle at a time. If it does leave
1944 * requests on the queue, it is responsible for arranging that the requests
1945 * get dealt with eventually.
1947 * The queue spin lock must be held while manipulating the requests on the
1948 * request queue; this lock will be taken also from interrupt context, so irq
1949 * disabling is needed for it.
1951 * Function returns a pointer to the initialized request queue, or NULL if
1952 * it didn't succeed.
1955 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1956 * when the block device is deactivated (such as at module unload).
1959 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1961 return blk_init_queue_node(rfn
, lock
, -1);
1963 EXPORT_SYMBOL(blk_init_queue
);
1966 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1968 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1974 if (blk_init_free_list(q
)) {
1975 kmem_cache_free(requestq_cachep
, q
);
1980 * if caller didn't supply a lock, they get per-queue locking with
1984 spin_lock_init(&q
->__queue_lock
);
1985 lock
= &q
->__queue_lock
;
1988 q
->request_fn
= rfn
;
1989 q
->back_merge_fn
= ll_back_merge_fn
;
1990 q
->front_merge_fn
= ll_front_merge_fn
;
1991 q
->merge_requests_fn
= ll_merge_requests_fn
;
1992 q
->prep_rq_fn
= NULL
;
1993 q
->unplug_fn
= generic_unplug_device
;
1994 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1995 q
->queue_lock
= lock
;
1997 blk_queue_segment_boundary(q
, 0xffffffff);
1999 blk_queue_make_request(q
, __make_request
);
2000 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
2002 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
2003 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
2008 if (!elevator_init(q
, NULL
)) {
2009 blk_queue_congestion_threshold(q
);
2016 EXPORT_SYMBOL(blk_init_queue_node
);
2018 int blk_get_queue(request_queue_t
*q
)
2020 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
2021 kobject_get(&q
->kobj
);
2028 EXPORT_SYMBOL(blk_get_queue
);
2030 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
2032 if (rq
->flags
& REQ_ELVPRIV
)
2033 elv_put_request(q
, rq
);
2034 mempool_free(rq
, q
->rq
.rq_pool
);
2037 static inline struct request
*
2038 blk_alloc_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
2039 int priv
, gfp_t gfp_mask
)
2041 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
2047 * first three bits are identical in rq->flags and bio->bi_rw,
2048 * see bio.h and blkdev.h
2053 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
))) {
2054 mempool_free(rq
, q
->rq
.rq_pool
);
2057 rq
->flags
|= REQ_ELVPRIV
;
2064 * ioc_batching returns true if the ioc is a valid batching request and
2065 * should be given priority access to a request.
2067 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
2073 * Make sure the process is able to allocate at least 1 request
2074 * even if the batch times out, otherwise we could theoretically
2077 return ioc
->nr_batch_requests
== q
->nr_batching
||
2078 (ioc
->nr_batch_requests
> 0
2079 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2083 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2084 * will cause the process to be a "batcher" on all queues in the system. This
2085 * is the behaviour we want though - once it gets a wakeup it should be given
2088 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
2090 if (!ioc
|| ioc_batching(q
, ioc
))
2093 ioc
->nr_batch_requests
= q
->nr_batching
;
2094 ioc
->last_waited
= jiffies
;
2097 static void __freed_request(request_queue_t
*q
, int rw
)
2099 struct request_list
*rl
= &q
->rq
;
2101 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2102 clear_queue_congested(q
, rw
);
2104 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2105 if (waitqueue_active(&rl
->wait
[rw
]))
2106 wake_up(&rl
->wait
[rw
]);
2108 blk_clear_queue_full(q
, rw
);
2113 * A request has just been released. Account for it, update the full and
2114 * congestion status, wake up any waiters. Called under q->queue_lock.
2116 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
2118 struct request_list
*rl
= &q
->rq
;
2124 __freed_request(q
, rw
);
2126 if (unlikely(rl
->starved
[rw
^ 1]))
2127 __freed_request(q
, rw
^ 1);
2130 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2132 * Get a free request, queue_lock must be held.
2133 * Returns NULL on failure, with queue_lock held.
2134 * Returns !NULL on success, with queue_lock *not held*.
2136 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
2139 struct request
*rq
= NULL
;
2140 struct request_list
*rl
= &q
->rq
;
2141 struct io_context
*ioc
= NULL
;
2142 int may_queue
, priv
;
2144 may_queue
= elv_may_queue(q
, rw
, bio
);
2145 if (may_queue
== ELV_MQUEUE_NO
)
2148 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2149 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2150 ioc
= current_io_context(GFP_ATOMIC
);
2152 * The queue will fill after this allocation, so set
2153 * it as full, and mark this process as "batching".
2154 * This process will be allowed to complete a batch of
2155 * requests, others will be blocked.
2157 if (!blk_queue_full(q
, rw
)) {
2158 ioc_set_batching(q
, ioc
);
2159 blk_set_queue_full(q
, rw
);
2161 if (may_queue
!= ELV_MQUEUE_MUST
2162 && !ioc_batching(q
, ioc
)) {
2164 * The queue is full and the allocating
2165 * process is not a "batcher", and not
2166 * exempted by the IO scheduler
2172 set_queue_congested(q
, rw
);
2176 * Only allow batching queuers to allocate up to 50% over the defined
2177 * limit of requests, otherwise we could have thousands of requests
2178 * allocated with any setting of ->nr_requests
2180 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2184 rl
->starved
[rw
] = 0;
2186 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2190 spin_unlock_irq(q
->queue_lock
);
2192 rq
= blk_alloc_request(q
, rw
, bio
, priv
, gfp_mask
);
2193 if (unlikely(!rq
)) {
2195 * Allocation failed presumably due to memory. Undo anything
2196 * we might have messed up.
2198 * Allocating task should really be put onto the front of the
2199 * wait queue, but this is pretty rare.
2201 spin_lock_irq(q
->queue_lock
);
2202 freed_request(q
, rw
, priv
);
2205 * in the very unlikely event that allocation failed and no
2206 * requests for this direction was pending, mark us starved
2207 * so that freeing of a request in the other direction will
2208 * notice us. another possible fix would be to split the
2209 * rq mempool into READ and WRITE
2212 if (unlikely(rl
->count
[rw
] == 0))
2213 rl
->starved
[rw
] = 1;
2219 * ioc may be NULL here, and ioc_batching will be false. That's
2220 * OK, if the queue is under the request limit then requests need
2221 * not count toward the nr_batch_requests limit. There will always
2222 * be some limit enforced by BLK_BATCH_TIME.
2224 if (ioc_batching(q
, ioc
))
2225 ioc
->nr_batch_requests
--;
2230 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2236 * No available requests for this queue, unplug the device and wait for some
2237 * requests to become available.
2239 * Called with q->queue_lock held, and returns with it unlocked.
2241 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
2246 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2249 struct request_list
*rl
= &q
->rq
;
2251 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2252 TASK_UNINTERRUPTIBLE
);
2254 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2257 struct io_context
*ioc
;
2259 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2261 __generic_unplug_device(q
);
2262 spin_unlock_irq(q
->queue_lock
);
2266 * After sleeping, we become a "batching" process and
2267 * will be able to allocate at least one request, and
2268 * up to a big batch of them for a small period time.
2269 * See ioc_batching, ioc_set_batching
2271 ioc
= current_io_context(GFP_NOIO
);
2272 ioc_set_batching(q
, ioc
);
2274 spin_lock_irq(q
->queue_lock
);
2276 finish_wait(&rl
->wait
[rw
], &wait
);
2282 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2286 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2288 spin_lock_irq(q
->queue_lock
);
2289 if (gfp_mask
& __GFP_WAIT
) {
2290 rq
= get_request_wait(q
, rw
, NULL
);
2292 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2294 spin_unlock_irq(q
->queue_lock
);
2296 /* q->queue_lock is unlocked at this point */
2300 EXPORT_SYMBOL(blk_get_request
);
2303 * blk_requeue_request - put a request back on queue
2304 * @q: request queue where request should be inserted
2305 * @rq: request to be inserted
2308 * Drivers often keep queueing requests until the hardware cannot accept
2309 * more, when that condition happens we need to put the request back
2310 * on the queue. Must be called with queue lock held.
2312 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2314 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2316 if (blk_rq_tagged(rq
))
2317 blk_queue_end_tag(q
, rq
);
2319 elv_requeue_request(q
, rq
);
2322 EXPORT_SYMBOL(blk_requeue_request
);
2325 * blk_insert_request - insert a special request in to a request queue
2326 * @q: request queue where request should be inserted
2327 * @rq: request to be inserted
2328 * @at_head: insert request at head or tail of queue
2329 * @data: private data
2332 * Many block devices need to execute commands asynchronously, so they don't
2333 * block the whole kernel from preemption during request execution. This is
2334 * accomplished normally by inserting aritficial requests tagged as
2335 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2336 * scheduled for actual execution by the request queue.
2338 * We have the option of inserting the head or the tail of the queue.
2339 * Typically we use the tail for new ioctls and so forth. We use the head
2340 * of the queue for things like a QUEUE_FULL message from a device, or a
2341 * host that is unable to accept a particular command.
2343 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2344 int at_head
, void *data
)
2346 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2347 unsigned long flags
;
2350 * tell I/O scheduler that this isn't a regular read/write (ie it
2351 * must not attempt merges on this) and that it acts as a soft
2354 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2358 spin_lock_irqsave(q
->queue_lock
, flags
);
2361 * If command is tagged, release the tag
2363 if (blk_rq_tagged(rq
))
2364 blk_queue_end_tag(q
, rq
);
2366 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2367 __elv_add_request(q
, rq
, where
, 0);
2369 if (blk_queue_plugged(q
))
2370 __generic_unplug_device(q
);
2373 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2376 EXPORT_SYMBOL(blk_insert_request
);
2379 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2380 * @q: request queue where request should be inserted
2381 * @rq: request structure to fill
2382 * @ubuf: the user buffer
2383 * @len: length of user data
2386 * Data will be mapped directly for zero copy io, if possible. Otherwise
2387 * a kernel bounce buffer is used.
2389 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2390 * still in process context.
2392 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2393 * before being submitted to the device, as pages mapped may be out of
2394 * reach. It's the callers responsibility to make sure this happens. The
2395 * original bio must be passed back in to blk_rq_unmap_user() for proper
2398 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2401 unsigned long uaddr
;
2405 if (len
> (q
->max_hw_sectors
<< 9))
2410 reading
= rq_data_dir(rq
) == READ
;
2413 * if alignment requirement is satisfied, map in user pages for
2414 * direct dma. else, set up kernel bounce buffers
2416 uaddr
= (unsigned long) ubuf
;
2417 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2418 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2420 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2423 rq
->bio
= rq
->biotail
= bio
;
2424 blk_rq_bio_prep(q
, rq
, bio
);
2426 rq
->buffer
= rq
->data
= NULL
;
2432 * bio is the err-ptr
2434 return PTR_ERR(bio
);
2437 EXPORT_SYMBOL(blk_rq_map_user
);
2440 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2441 * @q: request queue where request should be inserted
2442 * @rq: request to map data to
2443 * @iov: pointer to the iovec
2444 * @iov_count: number of elements in the iovec
2447 * Data will be mapped directly for zero copy io, if possible. Otherwise
2448 * a kernel bounce buffer is used.
2450 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2451 * still in process context.
2453 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2454 * before being submitted to the device, as pages mapped may be out of
2455 * reach. It's the callers responsibility to make sure this happens. The
2456 * original bio must be passed back in to blk_rq_unmap_user() for proper
2459 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2460 struct sg_iovec
*iov
, int iov_count
)
2464 if (!iov
|| iov_count
<= 0)
2467 /* we don't allow misaligned data like bio_map_user() does. If the
2468 * user is using sg, they're expected to know the alignment constraints
2469 * and respect them accordingly */
2470 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2472 return PTR_ERR(bio
);
2474 rq
->bio
= rq
->biotail
= bio
;
2475 blk_rq_bio_prep(q
, rq
, bio
);
2476 rq
->buffer
= rq
->data
= NULL
;
2477 rq
->data_len
= bio
->bi_size
;
2481 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2484 * blk_rq_unmap_user - unmap a request with user data
2485 * @bio: bio to be unmapped
2486 * @ulen: length of user buffer
2489 * Unmap a bio previously mapped by blk_rq_map_user().
2491 int blk_rq_unmap_user(struct bio
*bio
, unsigned int ulen
)
2496 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2497 bio_unmap_user(bio
);
2499 ret
= bio_uncopy_user(bio
);
2505 EXPORT_SYMBOL(blk_rq_unmap_user
);
2508 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2509 * @q: request queue where request should be inserted
2510 * @rq: request to fill
2511 * @kbuf: the kernel buffer
2512 * @len: length of user data
2513 * @gfp_mask: memory allocation flags
2515 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2516 unsigned int len
, gfp_t gfp_mask
)
2520 if (len
> (q
->max_hw_sectors
<< 9))
2525 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2527 return PTR_ERR(bio
);
2529 if (rq_data_dir(rq
) == WRITE
)
2530 bio
->bi_rw
|= (1 << BIO_RW
);
2532 rq
->bio
= rq
->biotail
= bio
;
2533 blk_rq_bio_prep(q
, rq
, bio
);
2535 rq
->buffer
= rq
->data
= NULL
;
2540 EXPORT_SYMBOL(blk_rq_map_kern
);
2543 * blk_execute_rq_nowait - insert a request into queue for execution
2544 * @q: queue to insert the request in
2545 * @bd_disk: matching gendisk
2546 * @rq: request to insert
2547 * @at_head: insert request at head or tail of queue
2548 * @done: I/O completion handler
2551 * Insert a fully prepared request at the back of the io scheduler queue
2552 * for execution. Don't wait for completion.
2554 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2555 struct request
*rq
, int at_head
,
2558 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2560 rq
->rq_disk
= bd_disk
;
2561 rq
->flags
|= REQ_NOMERGE
;
2563 WARN_ON(irqs_disabled());
2564 spin_lock_irq(q
->queue_lock
);
2565 __elv_add_request(q
, rq
, where
, 1);
2566 __generic_unplug_device(q
);
2567 spin_unlock_irq(q
->queue_lock
);
2569 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2572 * blk_execute_rq - insert a request into queue for execution
2573 * @q: queue to insert the request in
2574 * @bd_disk: matching gendisk
2575 * @rq: request to insert
2576 * @at_head: insert request at head or tail of queue
2579 * Insert a fully prepared request at the back of the io scheduler queue
2580 * for execution and wait for completion.
2582 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2583 struct request
*rq
, int at_head
)
2585 DECLARE_COMPLETION_ONSTACK(wait
);
2586 char sense
[SCSI_SENSE_BUFFERSIZE
];
2590 * we need an extra reference to the request, so we can look at
2591 * it after io completion
2596 memset(sense
, 0, sizeof(sense
));
2601 rq
->waiting
= &wait
;
2602 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2603 wait_for_completion(&wait
);
2612 EXPORT_SYMBOL(blk_execute_rq
);
2615 * blkdev_issue_flush - queue a flush
2616 * @bdev: blockdev to issue flush for
2617 * @error_sector: error sector
2620 * Issue a flush for the block device in question. Caller can supply
2621 * room for storing the error offset in case of a flush error, if they
2622 * wish to. Caller must run wait_for_completion() on its own.
2624 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2628 if (bdev
->bd_disk
== NULL
)
2631 q
= bdev_get_queue(bdev
);
2634 if (!q
->issue_flush_fn
)
2637 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2640 EXPORT_SYMBOL(blkdev_issue_flush
);
2642 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2644 int rw
= rq_data_dir(rq
);
2646 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2650 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2652 disk_round_stats(rq
->rq_disk
);
2653 rq
->rq_disk
->in_flight
++;
2658 * add-request adds a request to the linked list.
2659 * queue lock is held and interrupts disabled, as we muck with the
2660 * request queue list.
2662 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2664 drive_stat_acct(req
, req
->nr_sectors
, 1);
2667 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2670 * elevator indicated where it wants this request to be
2671 * inserted at elevator_merge time
2673 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2677 * disk_round_stats() - Round off the performance stats on a struct
2680 * The average IO queue length and utilisation statistics are maintained
2681 * by observing the current state of the queue length and the amount of
2682 * time it has been in this state for.
2684 * Normally, that accounting is done on IO completion, but that can result
2685 * in more than a second's worth of IO being accounted for within any one
2686 * second, leading to >100% utilisation. To deal with that, we call this
2687 * function to do a round-off before returning the results when reading
2688 * /proc/diskstats. This accounts immediately for all queue usage up to
2689 * the current jiffies and restarts the counters again.
2691 void disk_round_stats(struct gendisk
*disk
)
2693 unsigned long now
= jiffies
;
2695 if (now
== disk
->stamp
)
2698 if (disk
->in_flight
) {
2699 __disk_stat_add(disk
, time_in_queue
,
2700 disk
->in_flight
* (now
- disk
->stamp
));
2701 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2706 EXPORT_SYMBOL_GPL(disk_round_stats
);
2709 * queue lock must be held
2711 void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2713 struct request_list
*rl
= req
->rl
;
2717 if (unlikely(--req
->ref_count
))
2720 elv_completed_request(q
, req
);
2722 req
->rq_status
= RQ_INACTIVE
;
2726 * Request may not have originated from ll_rw_blk. if not,
2727 * it didn't come out of our reserved rq pools
2730 int rw
= rq_data_dir(req
);
2731 int priv
= req
->flags
& REQ_ELVPRIV
;
2733 BUG_ON(!list_empty(&req
->queuelist
));
2735 blk_free_request(q
, req
);
2736 freed_request(q
, rw
, priv
);
2740 EXPORT_SYMBOL_GPL(__blk_put_request
);
2742 void blk_put_request(struct request
*req
)
2744 unsigned long flags
;
2745 request_queue_t
*q
= req
->q
;
2748 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2749 * following if (q) test.
2752 spin_lock_irqsave(q
->queue_lock
, flags
);
2753 __blk_put_request(q
, req
);
2754 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2758 EXPORT_SYMBOL(blk_put_request
);
2761 * blk_end_sync_rq - executes a completion event on a request
2762 * @rq: request to complete
2763 * @error: end io status of the request
2765 void blk_end_sync_rq(struct request
*rq
, int error
)
2767 struct completion
*waiting
= rq
->waiting
;
2770 __blk_put_request(rq
->q
, rq
);
2773 * complete last, if this is a stack request the process (and thus
2774 * the rq pointer) could be invalid right after this complete()
2778 EXPORT_SYMBOL(blk_end_sync_rq
);
2781 * blk_congestion_wait - wait for a queue to become uncongested
2782 * @rw: READ or WRITE
2783 * @timeout: timeout in jiffies
2785 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2786 * If no queues are congested then just wait for the next request to be
2789 long blk_congestion_wait(int rw
, long timeout
)
2793 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2795 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2796 ret
= io_schedule_timeout(timeout
);
2797 finish_wait(wqh
, &wait
);
2801 EXPORT_SYMBOL(blk_congestion_wait
);
2804 * blk_congestion_end - wake up sleepers on a congestion queue
2805 * @rw: READ or WRITE
2807 void blk_congestion_end(int rw
)
2809 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2811 if (waitqueue_active(wqh
))
2816 * Has to be called with the request spinlock acquired
2818 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2819 struct request
*next
)
2821 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2827 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2830 if (rq_data_dir(req
) != rq_data_dir(next
)
2831 || req
->rq_disk
!= next
->rq_disk
2832 || next
->waiting
|| next
->special
)
2836 * If we are allowed to merge, then append bio list
2837 * from next to rq and release next. merge_requests_fn
2838 * will have updated segment counts, update sector
2841 if (!q
->merge_requests_fn(q
, req
, next
))
2845 * At this point we have either done a back merge
2846 * or front merge. We need the smaller start_time of
2847 * the merged requests to be the current request
2848 * for accounting purposes.
2850 if (time_after(req
->start_time
, next
->start_time
))
2851 req
->start_time
= next
->start_time
;
2853 req
->biotail
->bi_next
= next
->bio
;
2854 req
->biotail
= next
->biotail
;
2856 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2858 elv_merge_requests(q
, req
, next
);
2861 disk_round_stats(req
->rq_disk
);
2862 req
->rq_disk
->in_flight
--;
2865 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2867 __blk_put_request(q
, next
);
2871 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2873 struct request
*next
= elv_latter_request(q
, rq
);
2876 return attempt_merge(q
, rq
, next
);
2881 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2883 struct request
*prev
= elv_former_request(q
, rq
);
2886 return attempt_merge(q
, prev
, rq
);
2891 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2893 req
->flags
|= REQ_CMD
;
2896 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2898 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2899 req
->flags
|= REQ_FAILFAST
;
2902 * REQ_BARRIER implies no merging, but lets make it explicit
2904 if (unlikely(bio_barrier(bio
)))
2905 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2908 req
->flags
|= REQ_RW_SYNC
;
2911 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2912 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2913 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2914 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2915 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2916 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2917 req
->waiting
= NULL
;
2918 req
->bio
= req
->biotail
= bio
;
2919 req
->ioprio
= bio_prio(bio
);
2920 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2921 req
->start_time
= jiffies
;
2924 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2926 struct request
*req
;
2927 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2928 unsigned short prio
;
2931 sector
= bio
->bi_sector
;
2932 nr_sectors
= bio_sectors(bio
);
2933 cur_nr_sectors
= bio_cur_sectors(bio
);
2934 prio
= bio_prio(bio
);
2936 rw
= bio_data_dir(bio
);
2937 sync
= bio_sync(bio
);
2940 * low level driver can indicate that it wants pages above a
2941 * certain limit bounced to low memory (ie for highmem, or even
2942 * ISA dma in theory)
2944 blk_queue_bounce(q
, &bio
);
2946 spin_lock_prefetch(q
->queue_lock
);
2948 barrier
= bio_barrier(bio
);
2949 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2954 spin_lock_irq(q
->queue_lock
);
2956 if (unlikely(barrier
) || elv_queue_empty(q
))
2959 el_ret
= elv_merge(q
, &req
, bio
);
2961 case ELEVATOR_BACK_MERGE
:
2962 BUG_ON(!rq_mergeable(req
));
2964 if (!q
->back_merge_fn(q
, req
, bio
))
2967 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2969 req
->biotail
->bi_next
= bio
;
2971 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2972 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2973 drive_stat_acct(req
, nr_sectors
, 0);
2974 if (!attempt_back_merge(q
, req
))
2975 elv_merged_request(q
, req
);
2978 case ELEVATOR_FRONT_MERGE
:
2979 BUG_ON(!rq_mergeable(req
));
2981 if (!q
->front_merge_fn(q
, req
, bio
))
2984 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
2986 bio
->bi_next
= req
->bio
;
2990 * may not be valid. if the low level driver said
2991 * it didn't need a bounce buffer then it better
2992 * not touch req->buffer either...
2994 req
->buffer
= bio_data(bio
);
2995 req
->current_nr_sectors
= cur_nr_sectors
;
2996 req
->hard_cur_sectors
= cur_nr_sectors
;
2997 req
->sector
= req
->hard_sector
= sector
;
2998 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2999 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3000 drive_stat_acct(req
, nr_sectors
, 0);
3001 if (!attempt_front_merge(q
, req
))
3002 elv_merged_request(q
, req
);
3005 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3012 * Grab a free request. This is might sleep but can not fail.
3013 * Returns with the queue unlocked.
3015 req
= get_request_wait(q
, rw
, bio
);
3018 * After dropping the lock and possibly sleeping here, our request
3019 * may now be mergeable after it had proven unmergeable (above).
3020 * We don't worry about that case for efficiency. It won't happen
3021 * often, and the elevators are able to handle it.
3023 init_request_from_bio(req
, bio
);
3025 spin_lock_irq(q
->queue_lock
);
3026 if (elv_queue_empty(q
))
3028 add_request(q
, req
);
3031 __generic_unplug_device(q
);
3033 spin_unlock_irq(q
->queue_lock
);
3037 bio_endio(bio
, nr_sectors
<< 9, err
);
3042 * If bio->bi_dev is a partition, remap the location
3044 static inline void blk_partition_remap(struct bio
*bio
)
3046 struct block_device
*bdev
= bio
->bi_bdev
;
3048 if (bdev
!= bdev
->bd_contains
) {
3049 struct hd_struct
*p
= bdev
->bd_part
;
3050 const int rw
= bio_data_dir(bio
);
3052 p
->sectors
[rw
] += bio_sectors(bio
);
3055 bio
->bi_sector
+= p
->start_sect
;
3056 bio
->bi_bdev
= bdev
->bd_contains
;
3060 static void handle_bad_sector(struct bio
*bio
)
3062 char b
[BDEVNAME_SIZE
];
3064 printk(KERN_INFO
"attempt to access beyond end of device\n");
3065 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3066 bdevname(bio
->bi_bdev
, b
),
3068 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3069 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3071 set_bit(BIO_EOF
, &bio
->bi_flags
);
3075 * generic_make_request: hand a buffer to its device driver for I/O
3076 * @bio: The bio describing the location in memory and on the device.
3078 * generic_make_request() is used to make I/O requests of block
3079 * devices. It is passed a &struct bio, which describes the I/O that needs
3082 * generic_make_request() does not return any status. The
3083 * success/failure status of the request, along with notification of
3084 * completion, is delivered asynchronously through the bio->bi_end_io
3085 * function described (one day) else where.
3087 * The caller of generic_make_request must make sure that bi_io_vec
3088 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3089 * set to describe the device address, and the
3090 * bi_end_io and optionally bi_private are set to describe how
3091 * completion notification should be signaled.
3093 * generic_make_request and the drivers it calls may use bi_next if this
3094 * bio happens to be merged with someone else, and may change bi_dev and
3095 * bi_sector for remaps as it sees fit. So the values of these fields
3096 * should NOT be depended on after the call to generic_make_request.
3098 void generic_make_request(struct bio
*bio
)
3102 int ret
, nr_sectors
= bio_sectors(bio
);
3106 /* Test device or partition size, when known. */
3107 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3109 sector_t sector
= bio
->bi_sector
;
3111 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3113 * This may well happen - the kernel calls bread()
3114 * without checking the size of the device, e.g., when
3115 * mounting a device.
3117 handle_bad_sector(bio
);
3123 * Resolve the mapping until finished. (drivers are
3124 * still free to implement/resolve their own stacking
3125 * by explicitly returning 0)
3127 * NOTE: we don't repeat the blk_size check for each new device.
3128 * Stacking drivers are expected to know what they are doing.
3133 char b
[BDEVNAME_SIZE
];
3135 q
= bdev_get_queue(bio
->bi_bdev
);
3138 "generic_make_request: Trying to access "
3139 "nonexistent block-device %s (%Lu)\n",
3140 bdevname(bio
->bi_bdev
, b
),
3141 (long long) bio
->bi_sector
);
3143 bio_endio(bio
, bio
->bi_size
, -EIO
);
3147 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
3148 printk("bio too big device %s (%u > %u)\n",
3149 bdevname(bio
->bi_bdev
, b
),
3155 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3159 * If this device has partitions, remap block n
3160 * of partition p to block n+start(p) of the disk.
3162 blk_partition_remap(bio
);
3164 if (maxsector
!= -1)
3165 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3168 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3170 maxsector
= bio
->bi_sector
;
3171 old_dev
= bio
->bi_bdev
->bd_dev
;
3173 ret
= q
->make_request_fn(q
, bio
);
3177 EXPORT_SYMBOL(generic_make_request
);
3180 * submit_bio: submit a bio to the block device layer for I/O
3181 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3182 * @bio: The &struct bio which describes the I/O
3184 * submit_bio() is very similar in purpose to generic_make_request(), and
3185 * uses that function to do most of the work. Both are fairly rough
3186 * interfaces, @bio must be presetup and ready for I/O.
3189 void submit_bio(int rw
, struct bio
*bio
)
3191 int count
= bio_sectors(bio
);
3193 BIO_BUG_ON(!bio
->bi_size
);
3194 BIO_BUG_ON(!bio
->bi_io_vec
);
3197 count_vm_events(PGPGOUT
, count
);
3199 count_vm_events(PGPGIN
, count
);
3201 if (unlikely(block_dump
)) {
3202 char b
[BDEVNAME_SIZE
];
3203 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3204 current
->comm
, current
->pid
,
3205 (rw
& WRITE
) ? "WRITE" : "READ",
3206 (unsigned long long)bio
->bi_sector
,
3207 bdevname(bio
->bi_bdev
,b
));
3210 generic_make_request(bio
);
3213 EXPORT_SYMBOL(submit_bio
);
3215 static void blk_recalc_rq_segments(struct request
*rq
)
3217 struct bio
*bio
, *prevbio
= NULL
;
3218 int nr_phys_segs
, nr_hw_segs
;
3219 unsigned int phys_size
, hw_size
;
3220 request_queue_t
*q
= rq
->q
;
3225 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3226 rq_for_each_bio(bio
, rq
) {
3227 /* Force bio hw/phys segs to be recalculated. */
3228 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3230 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3231 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3233 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3234 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3236 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3237 pseg
<= q
->max_segment_size
) {
3239 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3243 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3244 hseg
<= q
->max_segment_size
) {
3246 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3253 rq
->nr_phys_segments
= nr_phys_segs
;
3254 rq
->nr_hw_segments
= nr_hw_segs
;
3257 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3259 if (blk_fs_request(rq
)) {
3260 rq
->hard_sector
+= nsect
;
3261 rq
->hard_nr_sectors
-= nsect
;
3264 * Move the I/O submission pointers ahead if required.
3266 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3267 (rq
->sector
<= rq
->hard_sector
)) {
3268 rq
->sector
= rq
->hard_sector
;
3269 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3270 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3271 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3272 rq
->buffer
= bio_data(rq
->bio
);
3276 * if total number of sectors is less than the first segment
3277 * size, something has gone terribly wrong
3279 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3280 printk("blk: request botched\n");
3281 rq
->nr_sectors
= rq
->current_nr_sectors
;
3286 static int __end_that_request_first(struct request
*req
, int uptodate
,
3289 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3292 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3295 * extend uptodate bool to allow < 0 value to be direct io error
3298 if (end_io_error(uptodate
))
3299 error
= !uptodate
? -EIO
: uptodate
;
3302 * for a REQ_BLOCK_PC request, we want to carry any eventual
3303 * sense key with us all the way through
3305 if (!blk_pc_request(req
))
3309 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3310 printk("end_request: I/O error, dev %s, sector %llu\n",
3311 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3312 (unsigned long long)req
->sector
);
3315 if (blk_fs_request(req
) && req
->rq_disk
) {
3316 const int rw
= rq_data_dir(req
);
3318 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3321 total_bytes
= bio_nbytes
= 0;
3322 while ((bio
= req
->bio
) != NULL
) {
3325 if (nr_bytes
>= bio
->bi_size
) {
3326 req
->bio
= bio
->bi_next
;
3327 nbytes
= bio
->bi_size
;
3328 if (!ordered_bio_endio(req
, bio
, nbytes
, error
))
3329 bio_endio(bio
, nbytes
, error
);
3333 int idx
= bio
->bi_idx
+ next_idx
;
3335 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3336 blk_dump_rq_flags(req
, "__end_that");
3337 printk("%s: bio idx %d >= vcnt %d\n",
3339 bio
->bi_idx
, bio
->bi_vcnt
);
3343 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3344 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3347 * not a complete bvec done
3349 if (unlikely(nbytes
> nr_bytes
)) {
3350 bio_nbytes
+= nr_bytes
;
3351 total_bytes
+= nr_bytes
;
3356 * advance to the next vector
3359 bio_nbytes
+= nbytes
;
3362 total_bytes
+= nbytes
;
3365 if ((bio
= req
->bio
)) {
3367 * end more in this run, or just return 'not-done'
3369 if (unlikely(nr_bytes
<= 0))
3381 * if the request wasn't completed, update state
3384 if (!ordered_bio_endio(req
, bio
, bio_nbytes
, error
))
3385 bio_endio(bio
, bio_nbytes
, error
);
3386 bio
->bi_idx
+= next_idx
;
3387 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3388 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3391 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3392 blk_recalc_rq_segments(req
);
3397 * end_that_request_first - end I/O on a request
3398 * @req: the request being processed
3399 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3400 * @nr_sectors: number of sectors to end I/O on
3403 * Ends I/O on a number of sectors attached to @req, and sets it up
3404 * for the next range of segments (if any) in the cluster.
3407 * 0 - we are done with this request, call end_that_request_last()
3408 * 1 - still buffers pending for this request
3410 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3412 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3415 EXPORT_SYMBOL(end_that_request_first
);
3418 * end_that_request_chunk - end I/O on a request
3419 * @req: the request being processed
3420 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3421 * @nr_bytes: number of bytes to complete
3424 * Ends I/O on a number of bytes attached to @req, and sets it up
3425 * for the next range of segments (if any). Like end_that_request_first(),
3426 * but deals with bytes instead of sectors.
3429 * 0 - we are done with this request, call end_that_request_last()
3430 * 1 - still buffers pending for this request
3432 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3434 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3437 EXPORT_SYMBOL(end_that_request_chunk
);
3440 * splice the completion data to a local structure and hand off to
3441 * process_completion_queue() to complete the requests
3443 static void blk_done_softirq(struct softirq_action
*h
)
3445 struct list_head
*cpu_list
, local_list
;
3447 local_irq_disable();
3448 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3449 list_replace_init(cpu_list
, &local_list
);
3452 while (!list_empty(&local_list
)) {
3453 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3455 list_del_init(&rq
->donelist
);
3456 rq
->q
->softirq_done_fn(rq
);
3460 #ifdef CONFIG_HOTPLUG_CPU
3462 static int blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3466 * If a CPU goes away, splice its entries to the current CPU
3467 * and trigger a run of the softirq
3469 if (action
== CPU_DEAD
) {
3470 int cpu
= (unsigned long) hcpu
;
3472 local_irq_disable();
3473 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3474 &__get_cpu_var(blk_cpu_done
));
3475 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3483 static struct notifier_block __devinitdata blk_cpu_notifier
= {
3484 .notifier_call
= blk_cpu_notify
,
3487 #endif /* CONFIG_HOTPLUG_CPU */
3490 * blk_complete_request - end I/O on a request
3491 * @req: the request being processed
3494 * Ends all I/O on a request. It does not handle partial completions,
3495 * unless the driver actually implements this in its completion callback
3496 * through requeueing. Theh actual completion happens out-of-order,
3497 * through a softirq handler. The user must have registered a completion
3498 * callback through blk_queue_softirq_done().
3501 void blk_complete_request(struct request
*req
)
3503 struct list_head
*cpu_list
;
3504 unsigned long flags
;
3506 BUG_ON(!req
->q
->softirq_done_fn
);
3508 local_irq_save(flags
);
3510 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3511 list_add_tail(&req
->donelist
, cpu_list
);
3512 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3514 local_irq_restore(flags
);
3517 EXPORT_SYMBOL(blk_complete_request
);
3520 * queue lock must be held
3522 void end_that_request_last(struct request
*req
, int uptodate
)
3524 struct gendisk
*disk
= req
->rq_disk
;
3528 * extend uptodate bool to allow < 0 value to be direct io error
3531 if (end_io_error(uptodate
))
3532 error
= !uptodate
? -EIO
: uptodate
;
3534 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3535 laptop_io_completion();
3538 * Account IO completion. bar_rq isn't accounted as a normal
3539 * IO on queueing nor completion. Accounting the containing
3540 * request is enough.
3542 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3543 unsigned long duration
= jiffies
- req
->start_time
;
3544 const int rw
= rq_data_dir(req
);
3546 __disk_stat_inc(disk
, ios
[rw
]);
3547 __disk_stat_add(disk
, ticks
[rw
], duration
);
3548 disk_round_stats(disk
);
3552 req
->end_io(req
, error
);
3554 __blk_put_request(req
->q
, req
);
3557 EXPORT_SYMBOL(end_that_request_last
);
3559 void end_request(struct request
*req
, int uptodate
)
3561 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3562 add_disk_randomness(req
->rq_disk
);
3563 blkdev_dequeue_request(req
);
3564 end_that_request_last(req
, uptodate
);
3568 EXPORT_SYMBOL(end_request
);
3570 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3572 /* first two bits are identical in rq->flags and bio->bi_rw */
3573 rq
->flags
|= (bio
->bi_rw
& 3);
3575 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3576 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3577 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3578 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3579 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3580 rq
->buffer
= bio_data(bio
);
3582 rq
->bio
= rq
->biotail
= bio
;
3585 EXPORT_SYMBOL(blk_rq_bio_prep
);
3587 int kblockd_schedule_work(struct work_struct
*work
)
3589 return queue_work(kblockd_workqueue
, work
);
3592 EXPORT_SYMBOL(kblockd_schedule_work
);
3594 void kblockd_flush(void)
3596 flush_workqueue(kblockd_workqueue
);
3598 EXPORT_SYMBOL(kblockd_flush
);
3600 int __init
blk_dev_init(void)
3604 kblockd_workqueue
= create_workqueue("kblockd");
3605 if (!kblockd_workqueue
)
3606 panic("Failed to create kblockd\n");
3608 request_cachep
= kmem_cache_create("blkdev_requests",
3609 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3611 requestq_cachep
= kmem_cache_create("blkdev_queue",
3612 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3614 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3615 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3617 for_each_possible_cpu(i
)
3618 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3620 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3621 register_hotcpu_notifier(&blk_cpu_notifier
);
3623 blk_max_low_pfn
= max_low_pfn
;
3624 blk_max_pfn
= max_pfn
;
3630 * IO Context helper functions
3632 void put_io_context(struct io_context
*ioc
)
3637 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3639 if (atomic_dec_and_test(&ioc
->refcount
)) {
3640 struct cfq_io_context
*cic
;
3643 if (ioc
->aic
&& ioc
->aic
->dtor
)
3644 ioc
->aic
->dtor(ioc
->aic
);
3645 if (ioc
->cic_root
.rb_node
!= NULL
) {
3646 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3648 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3653 kmem_cache_free(iocontext_cachep
, ioc
);
3656 EXPORT_SYMBOL(put_io_context
);
3658 /* Called by the exitting task */
3659 void exit_io_context(void)
3661 unsigned long flags
;
3662 struct io_context
*ioc
;
3663 struct cfq_io_context
*cic
;
3665 local_irq_save(flags
);
3667 ioc
= current
->io_context
;
3668 current
->io_context
= NULL
;
3670 task_unlock(current
);
3671 local_irq_restore(flags
);
3673 if (ioc
->aic
&& ioc
->aic
->exit
)
3674 ioc
->aic
->exit(ioc
->aic
);
3675 if (ioc
->cic_root
.rb_node
!= NULL
) {
3676 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3680 put_io_context(ioc
);
3684 * If the current task has no IO context then create one and initialise it.
3685 * Otherwise, return its existing IO context.
3687 * This returned IO context doesn't have a specifically elevated refcount,
3688 * but since the current task itself holds a reference, the context can be
3689 * used in general code, so long as it stays within `current` context.
3691 struct io_context
*current_io_context(gfp_t gfp_flags
)
3693 struct task_struct
*tsk
= current
;
3694 struct io_context
*ret
;
3696 ret
= tsk
->io_context
;
3700 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3702 atomic_set(&ret
->refcount
, 1);
3703 ret
->task
= current
;
3704 ret
->set_ioprio
= NULL
;
3705 ret
->last_waited
= jiffies
; /* doesn't matter... */
3706 ret
->nr_batch_requests
= 0; /* because this is 0 */
3708 ret
->cic_root
.rb_node
= NULL
;
3709 /* make sure set_task_ioprio() sees the settings above */
3711 tsk
->io_context
= ret
;
3716 EXPORT_SYMBOL(current_io_context
);
3719 * If the current task has no IO context then create one and initialise it.
3720 * If it does have a context, take a ref on it.
3722 * This is always called in the context of the task which submitted the I/O.
3724 struct io_context
*get_io_context(gfp_t gfp_flags
)
3726 struct io_context
*ret
;
3727 ret
= current_io_context(gfp_flags
);
3729 atomic_inc(&ret
->refcount
);
3732 EXPORT_SYMBOL(get_io_context
);
3734 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3736 struct io_context
*src
= *psrc
;
3737 struct io_context
*dst
= *pdst
;
3740 BUG_ON(atomic_read(&src
->refcount
) == 0);
3741 atomic_inc(&src
->refcount
);
3742 put_io_context(dst
);
3746 EXPORT_SYMBOL(copy_io_context
);
3748 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3750 struct io_context
*temp
;
3755 EXPORT_SYMBOL(swap_io_context
);
3760 struct queue_sysfs_entry
{
3761 struct attribute attr
;
3762 ssize_t (*show
)(struct request_queue
*, char *);
3763 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3767 queue_var_show(unsigned int var
, char *page
)
3769 return sprintf(page
, "%d\n", var
);
3773 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3775 char *p
= (char *) page
;
3777 *var
= simple_strtoul(p
, &p
, 10);
3781 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3783 return queue_var_show(q
->nr_requests
, (page
));
3787 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3789 struct request_list
*rl
= &q
->rq
;
3791 int ret
= queue_var_store(&nr
, page
, count
);
3792 if (nr
< BLKDEV_MIN_RQ
)
3795 spin_lock_irq(q
->queue_lock
);
3796 q
->nr_requests
= nr
;
3797 blk_queue_congestion_threshold(q
);
3799 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3800 set_queue_congested(q
, READ
);
3801 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3802 clear_queue_congested(q
, READ
);
3804 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3805 set_queue_congested(q
, WRITE
);
3806 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3807 clear_queue_congested(q
, WRITE
);
3809 if (rl
->count
[READ
] >= q
->nr_requests
) {
3810 blk_set_queue_full(q
, READ
);
3811 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3812 blk_clear_queue_full(q
, READ
);
3813 wake_up(&rl
->wait
[READ
]);
3816 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3817 blk_set_queue_full(q
, WRITE
);
3818 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3819 blk_clear_queue_full(q
, WRITE
);
3820 wake_up(&rl
->wait
[WRITE
]);
3822 spin_unlock_irq(q
->queue_lock
);
3826 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3828 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3830 return queue_var_show(ra_kb
, (page
));
3834 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3836 unsigned long ra_kb
;
3837 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3839 spin_lock_irq(q
->queue_lock
);
3840 if (ra_kb
> (q
->max_sectors
>> 1))
3841 ra_kb
= (q
->max_sectors
>> 1);
3843 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3844 spin_unlock_irq(q
->queue_lock
);
3849 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3851 int max_sectors_kb
= q
->max_sectors
>> 1;
3853 return queue_var_show(max_sectors_kb
, (page
));
3857 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3859 unsigned long max_sectors_kb
,
3860 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3861 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3862 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3865 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3868 * Take the queue lock to update the readahead and max_sectors
3869 * values synchronously:
3871 spin_lock_irq(q
->queue_lock
);
3873 * Trim readahead window as well, if necessary:
3875 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3876 if (ra_kb
> max_sectors_kb
)
3877 q
->backing_dev_info
.ra_pages
=
3878 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3880 q
->max_sectors
= max_sectors_kb
<< 1;
3881 spin_unlock_irq(q
->queue_lock
);
3886 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3888 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3890 return queue_var_show(max_hw_sectors_kb
, (page
));
3894 static struct queue_sysfs_entry queue_requests_entry
= {
3895 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3896 .show
= queue_requests_show
,
3897 .store
= queue_requests_store
,
3900 static struct queue_sysfs_entry queue_ra_entry
= {
3901 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3902 .show
= queue_ra_show
,
3903 .store
= queue_ra_store
,
3906 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3907 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3908 .show
= queue_max_sectors_show
,
3909 .store
= queue_max_sectors_store
,
3912 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3913 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3914 .show
= queue_max_hw_sectors_show
,
3917 static struct queue_sysfs_entry queue_iosched_entry
= {
3918 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3919 .show
= elv_iosched_show
,
3920 .store
= elv_iosched_store
,
3923 static struct attribute
*default_attrs
[] = {
3924 &queue_requests_entry
.attr
,
3925 &queue_ra_entry
.attr
,
3926 &queue_max_hw_sectors_entry
.attr
,
3927 &queue_max_sectors_entry
.attr
,
3928 &queue_iosched_entry
.attr
,
3932 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3935 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3937 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3938 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
3943 mutex_lock(&q
->sysfs_lock
);
3944 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
3945 mutex_unlock(&q
->sysfs_lock
);
3948 res
= entry
->show(q
, page
);
3949 mutex_unlock(&q
->sysfs_lock
);
3954 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3955 const char *page
, size_t length
)
3957 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3958 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
3964 mutex_lock(&q
->sysfs_lock
);
3965 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
3966 mutex_unlock(&q
->sysfs_lock
);
3969 res
= entry
->store(q
, page
, length
);
3970 mutex_unlock(&q
->sysfs_lock
);
3974 static struct sysfs_ops queue_sysfs_ops
= {
3975 .show
= queue_attr_show
,
3976 .store
= queue_attr_store
,
3979 static struct kobj_type queue_ktype
= {
3980 .sysfs_ops
= &queue_sysfs_ops
,
3981 .default_attrs
= default_attrs
,
3982 .release
= blk_release_queue
,
3985 int blk_register_queue(struct gendisk
*disk
)
3989 request_queue_t
*q
= disk
->queue
;
3991 if (!q
|| !q
->request_fn
)
3994 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3996 ret
= kobject_add(&q
->kobj
);
4000 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4002 ret
= elv_register_queue(q
);
4004 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4005 kobject_del(&q
->kobj
);
4012 void blk_unregister_queue(struct gendisk
*disk
)
4014 request_queue_t
*q
= disk
->queue
;
4016 if (q
&& q
->request_fn
) {
4017 elv_unregister_queue(q
);
4019 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4020 kobject_del(&q
->kobj
);
4021 kobject_put(&disk
->kobj
);