2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
11 * This handles all read/write requests to block devices
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/highmem.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
37 #include <scsi/scsi_cmnd.h>
39 static void blk_unplug_work(struct work_struct
*work
);
40 static void blk_unplug_timeout(unsigned long data
);
41 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
42 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
43 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
47 * For the allocated request tables
49 static struct kmem_cache
*request_cachep
;
52 * For queue allocation
54 static struct kmem_cache
*requestq_cachep
;
57 * For io context allocations
59 static struct kmem_cache
*iocontext_cachep
;
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 unsigned long blk_max_low_pfn
, blk_max_pfn
;
68 EXPORT_SYMBOL(blk_max_low_pfn
);
69 EXPORT_SYMBOL(blk_max_pfn
);
71 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
73 /* Amount of time in which a process may batch requests */
74 #define BLK_BATCH_TIME (HZ/50UL)
76 /* Number of requests a "batching" process may submit */
77 #define BLK_BATCH_REQ 32
80 * Return the threshold (number of used requests) at which the queue is
81 * considered to be congested. It include a little hysteresis to keep the
82 * context switch rate down.
84 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
86 return q
->nr_congestion_on
;
90 * The threshold at which a queue is considered to be uncongested
92 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
94 return q
->nr_congestion_off
;
97 static void blk_queue_congestion_threshold(struct request_queue
*q
)
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
102 if (nr
> q
->nr_requests
)
104 q
->nr_congestion_on
= nr
;
106 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
109 q
->nr_congestion_off
= nr
;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
123 struct backing_dev_info
*ret
= NULL
;
124 struct request_queue
*q
= bdev_get_queue(bdev
);
127 ret
= &q
->backing_dev_info
;
130 EXPORT_SYMBOL(blk_get_backing_dev_info
);
133 * blk_queue_prep_rq - set a prepare_request function for queue
135 * @pfn: prepare_request function
137 * It's possible for a queue to register a prepare_request callback which
138 * is invoked before the request is handed to the request_fn. The goal of
139 * the function is to prepare a request for I/O, it can be used to build a
140 * cdb from the request data for instance.
143 void blk_queue_prep_rq(struct request_queue
*q
, prep_rq_fn
*pfn
)
148 EXPORT_SYMBOL(blk_queue_prep_rq
);
151 * blk_queue_merge_bvec - set a merge_bvec function for queue
153 * @mbfn: merge_bvec_fn
155 * Usually queues have static limitations on the max sectors or segments that
156 * we can put in a request. Stacking drivers may have some settings that
157 * are dynamic, and thus we have to query the queue whether it is ok to
158 * add a new bio_vec to a bio at a given offset or not. If the block device
159 * has such limitations, it needs to register a merge_bvec_fn to control
160 * the size of bio's sent to it. Note that a block device *must* allow a
161 * single page to be added to an empty bio. The block device driver may want
162 * to use the bio_split() function to deal with these bio's. By default
163 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
166 void blk_queue_merge_bvec(struct request_queue
*q
, merge_bvec_fn
*mbfn
)
168 q
->merge_bvec_fn
= mbfn
;
171 EXPORT_SYMBOL(blk_queue_merge_bvec
);
173 void blk_queue_softirq_done(struct request_queue
*q
, softirq_done_fn
*fn
)
175 q
->softirq_done_fn
= fn
;
178 EXPORT_SYMBOL(blk_queue_softirq_done
);
181 * blk_queue_make_request - define an alternate make_request function for a device
182 * @q: the request queue for the device to be affected
183 * @mfn: the alternate make_request function
186 * The normal way for &struct bios to be passed to a device
187 * driver is for them to be collected into requests on a request
188 * queue, and then to allow the device driver to select requests
189 * off that queue when it is ready. This works well for many block
190 * devices. However some block devices (typically virtual devices
191 * such as md or lvm) do not benefit from the processing on the
192 * request queue, and are served best by having the requests passed
193 * directly to them. This can be achieved by providing a function
194 * to blk_queue_make_request().
197 * The driver that does this *must* be able to deal appropriately
198 * with buffers in "highmemory". This can be accomplished by either calling
199 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
200 * blk_queue_bounce() to create a buffer in normal memory.
202 void blk_queue_make_request(struct request_queue
* q
, make_request_fn
* mfn
)
207 q
->nr_requests
= BLKDEV_MAX_RQ
;
208 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
209 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
210 q
->make_request_fn
= mfn
;
211 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
212 q
->backing_dev_info
.state
= 0;
213 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
214 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
215 blk_queue_hardsect_size(q
, 512);
216 blk_queue_dma_alignment(q
, 511);
217 blk_queue_congestion_threshold(q
);
218 q
->nr_batching
= BLK_BATCH_REQ
;
220 q
->unplug_thresh
= 4; /* hmm */
221 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
222 if (q
->unplug_delay
== 0)
225 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
227 q
->unplug_timer
.function
= blk_unplug_timeout
;
228 q
->unplug_timer
.data
= (unsigned long)q
;
231 * by default assume old behaviour and bounce for any highmem page
233 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
236 EXPORT_SYMBOL(blk_queue_make_request
);
238 static void rq_init(struct request_queue
*q
, struct request
*rq
)
240 INIT_LIST_HEAD(&rq
->queuelist
);
241 INIT_LIST_HEAD(&rq
->donelist
);
244 rq
->bio
= rq
->biotail
= NULL
;
245 INIT_HLIST_NODE(&rq
->hash
);
246 RB_CLEAR_NODE(&rq
->rb_node
);
254 rq
->nr_phys_segments
= 0;
257 rq
->end_io_data
= NULL
;
258 rq
->completion_data
= NULL
;
263 * blk_queue_ordered - does this queue support ordered writes
264 * @q: the request queue
265 * @ordered: one of QUEUE_ORDERED_*
266 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
269 * For journalled file systems, doing ordered writes on a commit
270 * block instead of explicitly doing wait_on_buffer (which is bad
271 * for performance) can be a big win. Block drivers supporting this
272 * feature should call this function and indicate so.
275 int blk_queue_ordered(struct request_queue
*q
, unsigned ordered
,
276 prepare_flush_fn
*prepare_flush_fn
)
278 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
279 prepare_flush_fn
== NULL
) {
280 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
284 if (ordered
!= QUEUE_ORDERED_NONE
&&
285 ordered
!= QUEUE_ORDERED_DRAIN
&&
286 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
287 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
288 ordered
!= QUEUE_ORDERED_TAG
&&
289 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
290 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
291 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
295 q
->ordered
= ordered
;
296 q
->next_ordered
= ordered
;
297 q
->prepare_flush_fn
= prepare_flush_fn
;
302 EXPORT_SYMBOL(blk_queue_ordered
);
305 * blk_queue_issue_flush_fn - set function for issuing a flush
306 * @q: the request queue
307 * @iff: the function to be called issuing the flush
310 * If a driver supports issuing a flush command, the support is notified
311 * to the block layer by defining it through this call.
314 void blk_queue_issue_flush_fn(struct request_queue
*q
, issue_flush_fn
*iff
)
316 q
->issue_flush_fn
= iff
;
319 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
322 * Cache flushing for ordered writes handling
324 inline unsigned blk_ordered_cur_seq(struct request_queue
*q
)
328 return 1 << ffz(q
->ordseq
);
331 unsigned blk_ordered_req_seq(struct request
*rq
)
333 struct request_queue
*q
= rq
->q
;
335 BUG_ON(q
->ordseq
== 0);
337 if (rq
== &q
->pre_flush_rq
)
338 return QUEUE_ORDSEQ_PREFLUSH
;
339 if (rq
== &q
->bar_rq
)
340 return QUEUE_ORDSEQ_BAR
;
341 if (rq
== &q
->post_flush_rq
)
342 return QUEUE_ORDSEQ_POSTFLUSH
;
345 * !fs requests don't need to follow barrier ordering. Always
346 * put them at the front. This fixes the following deadlock.
348 * http://thread.gmane.org/gmane.linux.kernel/537473
350 if (!blk_fs_request(rq
))
351 return QUEUE_ORDSEQ_DRAIN
;
353 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
354 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
355 return QUEUE_ORDSEQ_DRAIN
;
357 return QUEUE_ORDSEQ_DONE
;
360 void blk_ordered_complete_seq(struct request_queue
*q
, unsigned seq
, int error
)
365 if (error
&& !q
->orderr
)
368 BUG_ON(q
->ordseq
& seq
);
371 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
375 * Okay, sequence complete.
378 uptodate
= q
->orderr
? q
->orderr
: 1;
382 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
383 end_that_request_last(rq
, uptodate
);
386 static void pre_flush_end_io(struct request
*rq
, int error
)
388 elv_completed_request(rq
->q
, rq
);
389 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
392 static void bar_end_io(struct request
*rq
, int error
)
394 elv_completed_request(rq
->q
, rq
);
395 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
398 static void post_flush_end_io(struct request
*rq
, int error
)
400 elv_completed_request(rq
->q
, rq
);
401 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
404 static void queue_flush(struct request_queue
*q
, unsigned which
)
407 rq_end_io_fn
*end_io
;
409 if (which
== QUEUE_ORDERED_PREFLUSH
) {
410 rq
= &q
->pre_flush_rq
;
411 end_io
= pre_flush_end_io
;
413 rq
= &q
->post_flush_rq
;
414 end_io
= post_flush_end_io
;
417 rq
->cmd_flags
= REQ_HARDBARRIER
;
419 rq
->elevator_private
= NULL
;
420 rq
->elevator_private2
= NULL
;
421 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
423 q
->prepare_flush_fn(q
, rq
);
425 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
428 static inline struct request
*start_ordered(struct request_queue
*q
,
433 q
->ordered
= q
->next_ordered
;
434 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
437 * Prep proxy barrier request.
439 blkdev_dequeue_request(rq
);
444 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
445 rq
->cmd_flags
|= REQ_RW
;
446 rq
->cmd_flags
|= q
->ordered
& QUEUE_ORDERED_FUA
? REQ_FUA
: 0;
447 rq
->elevator_private
= NULL
;
448 rq
->elevator_private2
= NULL
;
449 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
450 rq
->end_io
= bar_end_io
;
453 * Queue ordered sequence. As we stack them at the head, we
454 * need to queue in reverse order. Note that we rely on that
455 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
456 * request gets inbetween ordered sequence.
458 if (q
->ordered
& QUEUE_ORDERED_POSTFLUSH
)
459 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
461 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
463 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
465 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
466 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
467 rq
= &q
->pre_flush_rq
;
469 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
471 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
472 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
479 int blk_do_ordered(struct request_queue
*q
, struct request
**rqp
)
481 struct request
*rq
= *rqp
;
482 int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
488 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
489 *rqp
= start_ordered(q
, rq
);
493 * This can happen when the queue switches to
494 * ORDERED_NONE while this request is on it.
496 blkdev_dequeue_request(rq
);
497 end_that_request_first(rq
, -EOPNOTSUPP
,
498 rq
->hard_nr_sectors
);
499 end_that_request_last(rq
, -EOPNOTSUPP
);
506 * Ordered sequence in progress
509 /* Special requests are not subject to ordering rules. */
510 if (!blk_fs_request(rq
) &&
511 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
514 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
515 /* Ordered by tag. Blocking the next barrier is enough. */
516 if (is_barrier
&& rq
!= &q
->bar_rq
)
519 /* Ordered by draining. Wait for turn. */
520 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
521 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
528 static int flush_dry_bio_endio(struct bio
*bio
, unsigned int bytes
, int error
)
530 struct request_queue
*q
= bio
->bi_private
;
533 * This is dry run, restore bio_sector and size. We'll finish
534 * this request again with the original bi_end_io after an
535 * error occurs or post flush is complete.
543 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
544 bio
->bi_size
= q
->bi_size
;
545 bio
->bi_sector
-= (q
->bi_size
>> 9);
551 static int ordered_bio_endio(struct request
*rq
, struct bio
*bio
,
552 unsigned int nbytes
, int error
)
554 struct request_queue
*q
= rq
->q
;
558 if (&q
->bar_rq
!= rq
)
562 * Okay, this is the barrier request in progress, dry finish it.
564 if (error
&& !q
->orderr
)
567 endio
= bio
->bi_end_io
;
568 private = bio
->bi_private
;
569 bio
->bi_end_io
= flush_dry_bio_endio
;
572 bio_endio(bio
, nbytes
, error
);
574 bio
->bi_end_io
= endio
;
575 bio
->bi_private
= private;
581 * blk_queue_bounce_limit - set bounce buffer limit for queue
582 * @q: the request queue for the device
583 * @dma_addr: bus address limit
586 * Different hardware can have different requirements as to what pages
587 * it can do I/O directly to. A low level driver can call
588 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
589 * buffers for doing I/O to pages residing above @page.
591 void blk_queue_bounce_limit(struct request_queue
*q
, u64 dma_addr
)
593 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
596 q
->bounce_gfp
= GFP_NOIO
;
597 #if BITS_PER_LONG == 64
598 /* Assume anything <= 4GB can be handled by IOMMU.
599 Actually some IOMMUs can handle everything, but I don't
600 know of a way to test this here. */
601 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
603 q
->bounce_pfn
= max_low_pfn
;
605 if (bounce_pfn
< blk_max_low_pfn
)
607 q
->bounce_pfn
= bounce_pfn
;
610 init_emergency_isa_pool();
611 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
612 q
->bounce_pfn
= bounce_pfn
;
616 EXPORT_SYMBOL(blk_queue_bounce_limit
);
619 * blk_queue_max_sectors - set max sectors for a request for this queue
620 * @q: the request queue for the device
621 * @max_sectors: max sectors in the usual 512b unit
624 * Enables a low level driver to set an upper limit on the size of
627 void blk_queue_max_sectors(struct request_queue
*q
, unsigned int max_sectors
)
629 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
630 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
631 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
634 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
635 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
637 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
638 q
->max_hw_sectors
= max_sectors
;
642 EXPORT_SYMBOL(blk_queue_max_sectors
);
645 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
646 * @q: the request queue for the device
647 * @max_segments: max number of segments
650 * Enables a low level driver to set an upper limit on the number of
651 * physical data segments in a request. This would be the largest sized
652 * scatter list the driver could handle.
654 void blk_queue_max_phys_segments(struct request_queue
*q
,
655 unsigned short max_segments
)
659 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
662 q
->max_phys_segments
= max_segments
;
665 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
668 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
669 * @q: the request queue for the device
670 * @max_segments: max number of segments
673 * Enables a low level driver to set an upper limit on the number of
674 * hw data segments in a request. This would be the largest number of
675 * address/length pairs the host adapter can actually give as once
678 void blk_queue_max_hw_segments(struct request_queue
*q
,
679 unsigned short max_segments
)
683 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
686 q
->max_hw_segments
= max_segments
;
689 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
692 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
693 * @q: the request queue for the device
694 * @max_size: max size of segment in bytes
697 * Enables a low level driver to set an upper limit on the size of a
700 void blk_queue_max_segment_size(struct request_queue
*q
, unsigned int max_size
)
702 if (max_size
< PAGE_CACHE_SIZE
) {
703 max_size
= PAGE_CACHE_SIZE
;
704 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
707 q
->max_segment_size
= max_size
;
710 EXPORT_SYMBOL(blk_queue_max_segment_size
);
713 * blk_queue_hardsect_size - set hardware sector size for the queue
714 * @q: the request queue for the device
715 * @size: the hardware sector size, in bytes
718 * This should typically be set to the lowest possible sector size
719 * that the hardware can operate on (possible without reverting to
720 * even internal read-modify-write operations). Usually the default
721 * of 512 covers most hardware.
723 void blk_queue_hardsect_size(struct request_queue
*q
, unsigned short size
)
725 q
->hardsect_size
= size
;
728 EXPORT_SYMBOL(blk_queue_hardsect_size
);
731 * Returns the minimum that is _not_ zero, unless both are zero.
733 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
736 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
737 * @t: the stacking driver (top)
738 * @b: the underlying device (bottom)
740 void blk_queue_stack_limits(struct request_queue
*t
, struct request_queue
*b
)
742 /* zero is "infinity" */
743 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
744 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
746 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
747 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
748 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
749 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
750 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
751 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
754 EXPORT_SYMBOL(blk_queue_stack_limits
);
757 * blk_queue_segment_boundary - set boundary rules for segment merging
758 * @q: the request queue for the device
759 * @mask: the memory boundary mask
761 void blk_queue_segment_boundary(struct request_queue
*q
, unsigned long mask
)
763 if (mask
< PAGE_CACHE_SIZE
- 1) {
764 mask
= PAGE_CACHE_SIZE
- 1;
765 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
768 q
->seg_boundary_mask
= mask
;
771 EXPORT_SYMBOL(blk_queue_segment_boundary
);
774 * blk_queue_dma_alignment - set dma length and memory alignment
775 * @q: the request queue for the device
776 * @mask: alignment mask
779 * set required memory and length aligment for direct dma transactions.
780 * this is used when buiding direct io requests for the queue.
783 void blk_queue_dma_alignment(struct request_queue
*q
, int mask
)
785 q
->dma_alignment
= mask
;
788 EXPORT_SYMBOL(blk_queue_dma_alignment
);
791 * blk_queue_find_tag - find a request by its tag and queue
792 * @q: The request queue for the device
793 * @tag: The tag of the request
796 * Should be used when a device returns a tag and you want to match
799 * no locks need be held.
801 struct request
*blk_queue_find_tag(struct request_queue
*q
, int tag
)
803 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
806 EXPORT_SYMBOL(blk_queue_find_tag
);
809 * __blk_free_tags - release a given set of tag maintenance info
810 * @bqt: the tag map to free
812 * Tries to free the specified @bqt@. Returns true if it was
813 * actually freed and false if there are still references using it
815 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
819 retval
= atomic_dec_and_test(&bqt
->refcnt
);
823 kfree(bqt
->tag_index
);
824 bqt
->tag_index
= NULL
;
837 * __blk_queue_free_tags - release tag maintenance info
838 * @q: the request queue for the device
841 * blk_cleanup_queue() will take care of calling this function, if tagging
842 * has been used. So there's no need to call this directly.
844 static void __blk_queue_free_tags(struct request_queue
*q
)
846 struct blk_queue_tag
*bqt
= q
->queue_tags
;
851 __blk_free_tags(bqt
);
853 q
->queue_tags
= NULL
;
854 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
859 * blk_free_tags - release a given set of tag maintenance info
860 * @bqt: the tag map to free
862 * For externally managed @bqt@ frees the map. Callers of this
863 * function must guarantee to have released all the queues that
864 * might have been using this tag map.
866 void blk_free_tags(struct blk_queue_tag
*bqt
)
868 if (unlikely(!__blk_free_tags(bqt
)))
871 EXPORT_SYMBOL(blk_free_tags
);
874 * blk_queue_free_tags - release tag maintenance info
875 * @q: the request queue for the device
878 * This is used to disabled tagged queuing to a device, yet leave
881 void blk_queue_free_tags(struct request_queue
*q
)
883 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
886 EXPORT_SYMBOL(blk_queue_free_tags
);
889 init_tag_map(struct request_queue
*q
, struct blk_queue_tag
*tags
, int depth
)
891 struct request
**tag_index
;
892 unsigned long *tag_map
;
895 if (q
&& depth
> q
->nr_requests
* 2) {
896 depth
= q
->nr_requests
* 2;
897 printk(KERN_ERR
"%s: adjusted depth to %d\n",
898 __FUNCTION__
, depth
);
901 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
905 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
906 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
910 tags
->real_max_depth
= depth
;
911 tags
->max_depth
= depth
;
912 tags
->tag_index
= tag_index
;
913 tags
->tag_map
= tag_map
;
921 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
924 struct blk_queue_tag
*tags
;
926 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
930 if (init_tag_map(q
, tags
, depth
))
934 atomic_set(&tags
->refcnt
, 1);
942 * blk_init_tags - initialize the tag info for an external tag map
943 * @depth: the maximum queue depth supported
944 * @tags: the tag to use
946 struct blk_queue_tag
*blk_init_tags(int depth
)
948 return __blk_queue_init_tags(NULL
, depth
);
950 EXPORT_SYMBOL(blk_init_tags
);
953 * blk_queue_init_tags - initialize the queue tag info
954 * @q: the request queue for the device
955 * @depth: the maximum queue depth supported
956 * @tags: the tag to use
958 int blk_queue_init_tags(struct request_queue
*q
, int depth
,
959 struct blk_queue_tag
*tags
)
963 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
965 if (!tags
&& !q
->queue_tags
) {
966 tags
= __blk_queue_init_tags(q
, depth
);
970 } else if (q
->queue_tags
) {
971 if ((rc
= blk_queue_resize_tags(q
, depth
)))
973 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
976 atomic_inc(&tags
->refcnt
);
979 * assign it, all done
981 q
->queue_tags
= tags
;
982 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
983 INIT_LIST_HEAD(&q
->tag_busy_list
);
990 EXPORT_SYMBOL(blk_queue_init_tags
);
993 * blk_queue_resize_tags - change the queueing depth
994 * @q: the request queue for the device
995 * @new_depth: the new max command queueing depth
998 * Must be called with the queue lock held.
1000 int blk_queue_resize_tags(struct request_queue
*q
, int new_depth
)
1002 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1003 struct request
**tag_index
;
1004 unsigned long *tag_map
;
1005 int max_depth
, nr_ulongs
;
1011 * if we already have large enough real_max_depth. just
1012 * adjust max_depth. *NOTE* as requests with tag value
1013 * between new_depth and real_max_depth can be in-flight, tag
1014 * map can not be shrunk blindly here.
1016 if (new_depth
<= bqt
->real_max_depth
) {
1017 bqt
->max_depth
= new_depth
;
1022 * Currently cannot replace a shared tag map with a new
1023 * one, so error out if this is the case
1025 if (atomic_read(&bqt
->refcnt
) != 1)
1029 * save the old state info, so we can copy it back
1031 tag_index
= bqt
->tag_index
;
1032 tag_map
= bqt
->tag_map
;
1033 max_depth
= bqt
->real_max_depth
;
1035 if (init_tag_map(q
, bqt
, new_depth
))
1038 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1039 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1040 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1047 EXPORT_SYMBOL(blk_queue_resize_tags
);
1050 * blk_queue_end_tag - end tag operations for a request
1051 * @q: the request queue for the device
1052 * @rq: the request that has completed
1055 * Typically called when end_that_request_first() returns 0, meaning
1056 * all transfers have been done for a request. It's important to call
1057 * this function before end_that_request_last(), as that will put the
1058 * request back on the free list thus corrupting the internal tag list.
1061 * queue lock must be held.
1063 void blk_queue_end_tag(struct request_queue
*q
, struct request
*rq
)
1065 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1070 if (unlikely(tag
>= bqt
->real_max_depth
))
1072 * This can happen after tag depth has been reduced.
1073 * FIXME: how about a warning or info message here?
1077 list_del_init(&rq
->queuelist
);
1078 rq
->cmd_flags
&= ~REQ_QUEUED
;
1081 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1082 printk(KERN_ERR
"%s: tag %d is missing\n",
1085 bqt
->tag_index
[tag
] = NULL
;
1088 * We use test_and_clear_bit's memory ordering properties here.
1089 * The tag_map bit acts as a lock for tag_index[bit], so we need
1090 * a barrer before clearing the bit (precisely: release semantics).
1091 * Could use clear_bit_unlock when it is merged.
1093 if (unlikely(!test_and_clear_bit(tag
, bqt
->tag_map
))) {
1094 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1102 EXPORT_SYMBOL(blk_queue_end_tag
);
1105 * blk_queue_start_tag - find a free tag and assign it
1106 * @q: the request queue for the device
1107 * @rq: the block request that needs tagging
1110 * This can either be used as a stand-alone helper, or possibly be
1111 * assigned as the queue &prep_rq_fn (in which case &struct request
1112 * automagically gets a tag assigned). Note that this function
1113 * assumes that any type of request can be queued! if this is not
1114 * true for your device, you must check the request type before
1115 * calling this function. The request will also be removed from
1116 * the request queue, so it's the drivers responsibility to readd
1117 * it if it should need to be restarted for some reason.
1120 * queue lock must be held.
1122 int blk_queue_start_tag(struct request_queue
*q
, struct request
*rq
)
1124 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1127 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1129 "%s: request %p for device [%s] already tagged %d",
1131 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1136 * Protect against shared tag maps, as we may not have exclusive
1137 * access to the tag map.
1140 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1141 if (tag
>= bqt
->max_depth
)
1144 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1146 * We rely on test_and_set_bit providing lock memory ordering semantics
1147 * (could use test_and_set_bit_lock when it is merged).
1150 rq
->cmd_flags
|= REQ_QUEUED
;
1152 bqt
->tag_index
[tag
] = rq
;
1153 blkdev_dequeue_request(rq
);
1154 list_add(&rq
->queuelist
, &q
->tag_busy_list
);
1159 EXPORT_SYMBOL(blk_queue_start_tag
);
1162 * blk_queue_invalidate_tags - invalidate all pending tags
1163 * @q: the request queue for the device
1166 * Hardware conditions may dictate a need to stop all pending requests.
1167 * In this case, we will safely clear the block side of the tag queue and
1168 * readd all requests to the request queue in the right order.
1171 * queue lock must be held.
1173 void blk_queue_invalidate_tags(struct request_queue
*q
)
1175 struct list_head
*tmp
, *n
;
1178 list_for_each_safe(tmp
, n
, &q
->tag_busy_list
) {
1179 rq
= list_entry_rq(tmp
);
1181 if (rq
->tag
== -1) {
1183 "%s: bad tag found on list\n", __FUNCTION__
);
1184 list_del_init(&rq
->queuelist
);
1185 rq
->cmd_flags
&= ~REQ_QUEUED
;
1187 blk_queue_end_tag(q
, rq
);
1189 rq
->cmd_flags
&= ~REQ_STARTED
;
1190 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1194 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1196 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1200 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1201 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1204 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1206 rq
->current_nr_sectors
);
1207 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1209 if (blk_pc_request(rq
)) {
1211 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1212 printk("%02x ", rq
->cmd
[bit
]);
1217 EXPORT_SYMBOL(blk_dump_rq_flags
);
1219 void blk_recount_segments(struct request_queue
*q
, struct bio
*bio
)
1221 struct bio_vec
*bv
, *bvprv
= NULL
;
1222 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1223 int high
, highprv
= 1;
1225 if (unlikely(!bio
->bi_io_vec
))
1228 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1229 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1230 bio_for_each_segment(bv
, bio
, i
) {
1232 * the trick here is making sure that a high page is never
1233 * considered part of another segment, since that might
1234 * change with the bounce page.
1236 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1237 if (high
|| highprv
)
1238 goto new_hw_segment
;
1240 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1242 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1244 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1246 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1247 goto new_hw_segment
;
1249 seg_size
+= bv
->bv_len
;
1250 hw_seg_size
+= bv
->bv_len
;
1255 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1256 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1257 hw_seg_size
+= bv
->bv_len
;
1260 if (hw_seg_size
> bio
->bi_hw_front_size
)
1261 bio
->bi_hw_front_size
= hw_seg_size
;
1262 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1268 seg_size
= bv
->bv_len
;
1271 if (hw_seg_size
> bio
->bi_hw_back_size
)
1272 bio
->bi_hw_back_size
= hw_seg_size
;
1273 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1274 bio
->bi_hw_front_size
= hw_seg_size
;
1275 bio
->bi_phys_segments
= nr_phys_segs
;
1276 bio
->bi_hw_segments
= nr_hw_segs
;
1277 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1279 EXPORT_SYMBOL(blk_recount_segments
);
1281 static int blk_phys_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1284 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1287 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1289 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1293 * bio and nxt are contigous in memory, check if the queue allows
1294 * these two to be merged into one
1296 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1302 static int blk_hw_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1305 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1306 blk_recount_segments(q
, bio
);
1307 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1308 blk_recount_segments(q
, nxt
);
1309 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1310 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
))
1312 if (bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
> q
->max_segment_size
)
1319 * map a request to scatterlist, return number of sg entries setup. Caller
1320 * must make sure sg can hold rq->nr_phys_segments entries
1322 int blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
1323 struct scatterlist
*sg
)
1325 struct bio_vec
*bvec
, *bvprv
;
1327 int nsegs
, i
, cluster
;
1330 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1333 * for each bio in rq
1336 rq_for_each_bio(bio
, rq
) {
1338 * for each segment in bio
1340 bio_for_each_segment(bvec
, bio
, i
) {
1341 int nbytes
= bvec
->bv_len
;
1343 if (bvprv
&& cluster
) {
1344 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1347 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1349 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1352 sg
[nsegs
- 1].length
+= nbytes
;
1355 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1356 sg
[nsegs
].page
= bvec
->bv_page
;
1357 sg
[nsegs
].length
= nbytes
;
1358 sg
[nsegs
].offset
= bvec
->bv_offset
;
1363 } /* segments in bio */
1369 EXPORT_SYMBOL(blk_rq_map_sg
);
1372 * the standard queue merge functions, can be overridden with device
1373 * specific ones if so desired
1376 static inline int ll_new_mergeable(struct request_queue
*q
,
1377 struct request
*req
,
1380 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1382 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1383 req
->cmd_flags
|= REQ_NOMERGE
;
1384 if (req
== q
->last_merge
)
1385 q
->last_merge
= NULL
;
1390 * A hw segment is just getting larger, bump just the phys
1393 req
->nr_phys_segments
+= nr_phys_segs
;
1397 static inline int ll_new_hw_segment(struct request_queue
*q
,
1398 struct request
*req
,
1401 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1402 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1404 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1405 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1406 req
->cmd_flags
|= REQ_NOMERGE
;
1407 if (req
== q
->last_merge
)
1408 q
->last_merge
= NULL
;
1413 * This will form the start of a new hw segment. Bump both
1416 req
->nr_hw_segments
+= nr_hw_segs
;
1417 req
->nr_phys_segments
+= nr_phys_segs
;
1421 int ll_back_merge_fn(struct request_queue
*q
, struct request
*req
, struct bio
*bio
)
1423 unsigned short max_sectors
;
1426 if (unlikely(blk_pc_request(req
)))
1427 max_sectors
= q
->max_hw_sectors
;
1429 max_sectors
= q
->max_sectors
;
1431 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1432 req
->cmd_flags
|= REQ_NOMERGE
;
1433 if (req
== q
->last_merge
)
1434 q
->last_merge
= NULL
;
1437 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1438 blk_recount_segments(q
, req
->biotail
);
1439 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1440 blk_recount_segments(q
, bio
);
1441 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1442 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1443 !BIOVEC_VIRT_OVERSIZE(len
)) {
1444 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1447 if (req
->nr_hw_segments
== 1)
1448 req
->bio
->bi_hw_front_size
= len
;
1449 if (bio
->bi_hw_segments
== 1)
1450 bio
->bi_hw_back_size
= len
;
1455 return ll_new_hw_segment(q
, req
, bio
);
1457 EXPORT_SYMBOL(ll_back_merge_fn
);
1459 static int ll_front_merge_fn(struct request_queue
*q
, struct request
*req
,
1462 unsigned short max_sectors
;
1465 if (unlikely(blk_pc_request(req
)))
1466 max_sectors
= q
->max_hw_sectors
;
1468 max_sectors
= q
->max_sectors
;
1471 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1472 req
->cmd_flags
|= REQ_NOMERGE
;
1473 if (req
== q
->last_merge
)
1474 q
->last_merge
= NULL
;
1477 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1478 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1479 blk_recount_segments(q
, bio
);
1480 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1481 blk_recount_segments(q
, req
->bio
);
1482 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1483 !BIOVEC_VIRT_OVERSIZE(len
)) {
1484 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1487 if (bio
->bi_hw_segments
== 1)
1488 bio
->bi_hw_front_size
= len
;
1489 if (req
->nr_hw_segments
== 1)
1490 req
->biotail
->bi_hw_back_size
= len
;
1495 return ll_new_hw_segment(q
, req
, bio
);
1498 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
1499 struct request
*next
)
1501 int total_phys_segments
;
1502 int total_hw_segments
;
1505 * First check if the either of the requests are re-queued
1506 * requests. Can't merge them if they are.
1508 if (req
->special
|| next
->special
)
1512 * Will it become too large?
1514 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1517 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1518 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1519 total_phys_segments
--;
1521 if (total_phys_segments
> q
->max_phys_segments
)
1524 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1525 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1526 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1528 * propagate the combined length to the end of the requests
1530 if (req
->nr_hw_segments
== 1)
1531 req
->bio
->bi_hw_front_size
= len
;
1532 if (next
->nr_hw_segments
== 1)
1533 next
->biotail
->bi_hw_back_size
= len
;
1534 total_hw_segments
--;
1537 if (total_hw_segments
> q
->max_hw_segments
)
1540 /* Merge is OK... */
1541 req
->nr_phys_segments
= total_phys_segments
;
1542 req
->nr_hw_segments
= total_hw_segments
;
1547 * "plug" the device if there are no outstanding requests: this will
1548 * force the transfer to start only after we have put all the requests
1551 * This is called with interrupts off and no requests on the queue and
1552 * with the queue lock held.
1554 void blk_plug_device(struct request_queue
*q
)
1556 WARN_ON(!irqs_disabled());
1559 * don't plug a stopped queue, it must be paired with blk_start_queue()
1560 * which will restart the queueing
1562 if (blk_queue_stopped(q
))
1565 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1566 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1567 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1571 EXPORT_SYMBOL(blk_plug_device
);
1574 * remove the queue from the plugged list, if present. called with
1575 * queue lock held and interrupts disabled.
1577 int blk_remove_plug(struct request_queue
*q
)
1579 WARN_ON(!irqs_disabled());
1581 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1584 del_timer(&q
->unplug_timer
);
1588 EXPORT_SYMBOL(blk_remove_plug
);
1591 * remove the plug and let it rip..
1593 void __generic_unplug_device(struct request_queue
*q
)
1595 if (unlikely(blk_queue_stopped(q
)))
1598 if (!blk_remove_plug(q
))
1603 EXPORT_SYMBOL(__generic_unplug_device
);
1606 * generic_unplug_device - fire a request queue
1607 * @q: The &struct request_queue in question
1610 * Linux uses plugging to build bigger requests queues before letting
1611 * the device have at them. If a queue is plugged, the I/O scheduler
1612 * is still adding and merging requests on the queue. Once the queue
1613 * gets unplugged, the request_fn defined for the queue is invoked and
1614 * transfers started.
1616 void generic_unplug_device(struct request_queue
*q
)
1618 spin_lock_irq(q
->queue_lock
);
1619 __generic_unplug_device(q
);
1620 spin_unlock_irq(q
->queue_lock
);
1622 EXPORT_SYMBOL(generic_unplug_device
);
1624 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1627 struct request_queue
*q
= bdi
->unplug_io_data
;
1630 * devices don't necessarily have an ->unplug_fn defined
1633 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1634 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1640 static void blk_unplug_work(struct work_struct
*work
)
1642 struct request_queue
*q
=
1643 container_of(work
, struct request_queue
, unplug_work
);
1645 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1646 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1651 static void blk_unplug_timeout(unsigned long data
)
1653 struct request_queue
*q
= (struct request_queue
*)data
;
1655 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1656 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1658 kblockd_schedule_work(&q
->unplug_work
);
1662 * blk_start_queue - restart a previously stopped queue
1663 * @q: The &struct request_queue in question
1666 * blk_start_queue() will clear the stop flag on the queue, and call
1667 * the request_fn for the queue if it was in a stopped state when
1668 * entered. Also see blk_stop_queue(). Queue lock must be held.
1670 void blk_start_queue(struct request_queue
*q
)
1672 WARN_ON(!irqs_disabled());
1674 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1677 * one level of recursion is ok and is much faster than kicking
1678 * the unplug handling
1680 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1682 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1685 kblockd_schedule_work(&q
->unplug_work
);
1689 EXPORT_SYMBOL(blk_start_queue
);
1692 * blk_stop_queue - stop a queue
1693 * @q: The &struct request_queue in question
1696 * The Linux block layer assumes that a block driver will consume all
1697 * entries on the request queue when the request_fn strategy is called.
1698 * Often this will not happen, because of hardware limitations (queue
1699 * depth settings). If a device driver gets a 'queue full' response,
1700 * or if it simply chooses not to queue more I/O at one point, it can
1701 * call this function to prevent the request_fn from being called until
1702 * the driver has signalled it's ready to go again. This happens by calling
1703 * blk_start_queue() to restart queue operations. Queue lock must be held.
1705 void blk_stop_queue(struct request_queue
*q
)
1708 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1710 EXPORT_SYMBOL(blk_stop_queue
);
1713 * blk_sync_queue - cancel any pending callbacks on a queue
1717 * The block layer may perform asynchronous callback activity
1718 * on a queue, such as calling the unplug function after a timeout.
1719 * A block device may call blk_sync_queue to ensure that any
1720 * such activity is cancelled, thus allowing it to release resources
1721 * that the callbacks might use. The caller must already have made sure
1722 * that its ->make_request_fn will not re-add plugging prior to calling
1726 void blk_sync_queue(struct request_queue
*q
)
1728 del_timer_sync(&q
->unplug_timer
);
1730 EXPORT_SYMBOL(blk_sync_queue
);
1733 * blk_run_queue - run a single device queue
1734 * @q: The queue to run
1736 void blk_run_queue(struct request_queue
*q
)
1738 unsigned long flags
;
1740 spin_lock_irqsave(q
->queue_lock
, flags
);
1744 * Only recurse once to avoid overrunning the stack, let the unplug
1745 * handling reinvoke the handler shortly if we already got there.
1747 if (!elv_queue_empty(q
)) {
1748 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1750 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1753 kblockd_schedule_work(&q
->unplug_work
);
1757 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1759 EXPORT_SYMBOL(blk_run_queue
);
1762 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
1763 * @kobj: the kobj belonging of the request queue to be released
1766 * blk_cleanup_queue is the pair to blk_init_queue() or
1767 * blk_queue_make_request(). It should be called when a request queue is
1768 * being released; typically when a block device is being de-registered.
1769 * Currently, its primary task it to free all the &struct request
1770 * structures that were allocated to the queue and the queue itself.
1773 * Hopefully the low level driver will have finished any
1774 * outstanding requests first...
1776 static void blk_release_queue(struct kobject
*kobj
)
1778 struct request_queue
*q
=
1779 container_of(kobj
, struct request_queue
, kobj
);
1780 struct request_list
*rl
= &q
->rq
;
1785 mempool_destroy(rl
->rq_pool
);
1788 __blk_queue_free_tags(q
);
1790 blk_trace_shutdown(q
);
1792 kmem_cache_free(requestq_cachep
, q
);
1795 void blk_put_queue(struct request_queue
*q
)
1797 kobject_put(&q
->kobj
);
1799 EXPORT_SYMBOL(blk_put_queue
);
1801 void blk_cleanup_queue(struct request_queue
* q
)
1803 mutex_lock(&q
->sysfs_lock
);
1804 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1805 mutex_unlock(&q
->sysfs_lock
);
1808 elevator_exit(q
->elevator
);
1813 EXPORT_SYMBOL(blk_cleanup_queue
);
1815 static int blk_init_free_list(struct request_queue
*q
)
1817 struct request_list
*rl
= &q
->rq
;
1819 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1820 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1822 init_waitqueue_head(&rl
->wait
[READ
]);
1823 init_waitqueue_head(&rl
->wait
[WRITE
]);
1825 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1826 mempool_free_slab
, request_cachep
, q
->node
);
1834 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
1836 return blk_alloc_queue_node(gfp_mask
, -1);
1838 EXPORT_SYMBOL(blk_alloc_queue
);
1840 static struct kobj_type queue_ktype
;
1842 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1844 struct request_queue
*q
;
1846 q
= kmem_cache_alloc_node(requestq_cachep
,
1847 gfp_mask
| __GFP_ZERO
, node_id
);
1851 init_timer(&q
->unplug_timer
);
1853 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
1854 q
->kobj
.ktype
= &queue_ktype
;
1855 kobject_init(&q
->kobj
);
1857 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1858 q
->backing_dev_info
.unplug_io_data
= q
;
1860 mutex_init(&q
->sysfs_lock
);
1864 EXPORT_SYMBOL(blk_alloc_queue_node
);
1867 * blk_init_queue - prepare a request queue for use with a block device
1868 * @rfn: The function to be called to process requests that have been
1869 * placed on the queue.
1870 * @lock: Request queue spin lock
1873 * If a block device wishes to use the standard request handling procedures,
1874 * which sorts requests and coalesces adjacent requests, then it must
1875 * call blk_init_queue(). The function @rfn will be called when there
1876 * are requests on the queue that need to be processed. If the device
1877 * supports plugging, then @rfn may not be called immediately when requests
1878 * are available on the queue, but may be called at some time later instead.
1879 * Plugged queues are generally unplugged when a buffer belonging to one
1880 * of the requests on the queue is needed, or due to memory pressure.
1882 * @rfn is not required, or even expected, to remove all requests off the
1883 * queue, but only as many as it can handle at a time. If it does leave
1884 * requests on the queue, it is responsible for arranging that the requests
1885 * get dealt with eventually.
1887 * The queue spin lock must be held while manipulating the requests on the
1888 * request queue; this lock will be taken also from interrupt context, so irq
1889 * disabling is needed for it.
1891 * Function returns a pointer to the initialized request queue, or NULL if
1892 * it didn't succeed.
1895 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1896 * when the block device is deactivated (such as at module unload).
1899 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1901 return blk_init_queue_node(rfn
, lock
, -1);
1903 EXPORT_SYMBOL(blk_init_queue
);
1905 struct request_queue
*
1906 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1908 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1914 if (blk_init_free_list(q
)) {
1915 kmem_cache_free(requestq_cachep
, q
);
1920 * if caller didn't supply a lock, they get per-queue locking with
1924 spin_lock_init(&q
->__queue_lock
);
1925 lock
= &q
->__queue_lock
;
1928 q
->request_fn
= rfn
;
1929 q
->prep_rq_fn
= NULL
;
1930 q
->unplug_fn
= generic_unplug_device
;
1931 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1932 q
->queue_lock
= lock
;
1934 blk_queue_segment_boundary(q
, 0xffffffff);
1936 blk_queue_make_request(q
, __make_request
);
1937 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1939 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1940 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1942 q
->sg_reserved_size
= INT_MAX
;
1947 if (!elevator_init(q
, NULL
)) {
1948 blk_queue_congestion_threshold(q
);
1955 EXPORT_SYMBOL(blk_init_queue_node
);
1957 int blk_get_queue(struct request_queue
*q
)
1959 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1960 kobject_get(&q
->kobj
);
1967 EXPORT_SYMBOL(blk_get_queue
);
1969 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
1971 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1972 elv_put_request(q
, rq
);
1973 mempool_free(rq
, q
->rq
.rq_pool
);
1976 static struct request
*
1977 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1979 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1985 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1986 * see bio.h and blkdev.h
1988 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
1991 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
1992 mempool_free(rq
, q
->rq
.rq_pool
);
1995 rq
->cmd_flags
|= REQ_ELVPRIV
;
2002 * ioc_batching returns true if the ioc is a valid batching request and
2003 * should be given priority access to a request.
2005 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
2011 * Make sure the process is able to allocate at least 1 request
2012 * even if the batch times out, otherwise we could theoretically
2015 return ioc
->nr_batch_requests
== q
->nr_batching
||
2016 (ioc
->nr_batch_requests
> 0
2017 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2021 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2022 * will cause the process to be a "batcher" on all queues in the system. This
2023 * is the behaviour we want though - once it gets a wakeup it should be given
2026 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
2028 if (!ioc
|| ioc_batching(q
, ioc
))
2031 ioc
->nr_batch_requests
= q
->nr_batching
;
2032 ioc
->last_waited
= jiffies
;
2035 static void __freed_request(struct request_queue
*q
, int rw
)
2037 struct request_list
*rl
= &q
->rq
;
2039 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2040 blk_clear_queue_congested(q
, rw
);
2042 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2043 if (waitqueue_active(&rl
->wait
[rw
]))
2044 wake_up(&rl
->wait
[rw
]);
2046 blk_clear_queue_full(q
, rw
);
2051 * A request has just been released. Account for it, update the full and
2052 * congestion status, wake up any waiters. Called under q->queue_lock.
2054 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
2056 struct request_list
*rl
= &q
->rq
;
2062 __freed_request(q
, rw
);
2064 if (unlikely(rl
->starved
[rw
^ 1]))
2065 __freed_request(q
, rw
^ 1);
2068 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2070 * Get a free request, queue_lock must be held.
2071 * Returns NULL on failure, with queue_lock held.
2072 * Returns !NULL on success, with queue_lock *not held*.
2074 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
2075 struct bio
*bio
, gfp_t gfp_mask
)
2077 struct request
*rq
= NULL
;
2078 struct request_list
*rl
= &q
->rq
;
2079 struct io_context
*ioc
= NULL
;
2080 const int rw
= rw_flags
& 0x01;
2081 int may_queue
, priv
;
2083 may_queue
= elv_may_queue(q
, rw_flags
);
2084 if (may_queue
== ELV_MQUEUE_NO
)
2087 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2088 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2089 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2091 * The queue will fill after this allocation, so set
2092 * it as full, and mark this process as "batching".
2093 * This process will be allowed to complete a batch of
2094 * requests, others will be blocked.
2096 if (!blk_queue_full(q
, rw
)) {
2097 ioc_set_batching(q
, ioc
);
2098 blk_set_queue_full(q
, rw
);
2100 if (may_queue
!= ELV_MQUEUE_MUST
2101 && !ioc_batching(q
, ioc
)) {
2103 * The queue is full and the allocating
2104 * process is not a "batcher", and not
2105 * exempted by the IO scheduler
2111 blk_set_queue_congested(q
, rw
);
2115 * Only allow batching queuers to allocate up to 50% over the defined
2116 * limit of requests, otherwise we could have thousands of requests
2117 * allocated with any setting of ->nr_requests
2119 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2123 rl
->starved
[rw
] = 0;
2125 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2129 spin_unlock_irq(q
->queue_lock
);
2131 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2132 if (unlikely(!rq
)) {
2134 * Allocation failed presumably due to memory. Undo anything
2135 * we might have messed up.
2137 * Allocating task should really be put onto the front of the
2138 * wait queue, but this is pretty rare.
2140 spin_lock_irq(q
->queue_lock
);
2141 freed_request(q
, rw
, priv
);
2144 * in the very unlikely event that allocation failed and no
2145 * requests for this direction was pending, mark us starved
2146 * so that freeing of a request in the other direction will
2147 * notice us. another possible fix would be to split the
2148 * rq mempool into READ and WRITE
2151 if (unlikely(rl
->count
[rw
] == 0))
2152 rl
->starved
[rw
] = 1;
2158 * ioc may be NULL here, and ioc_batching will be false. That's
2159 * OK, if the queue is under the request limit then requests need
2160 * not count toward the nr_batch_requests limit. There will always
2161 * be some limit enforced by BLK_BATCH_TIME.
2163 if (ioc_batching(q
, ioc
))
2164 ioc
->nr_batch_requests
--;
2168 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2174 * No available requests for this queue, unplug the device and wait for some
2175 * requests to become available.
2177 * Called with q->queue_lock held, and returns with it unlocked.
2179 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
2182 const int rw
= rw_flags
& 0x01;
2185 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2188 struct request_list
*rl
= &q
->rq
;
2190 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2191 TASK_UNINTERRUPTIBLE
);
2193 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2196 struct io_context
*ioc
;
2198 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2200 __generic_unplug_device(q
);
2201 spin_unlock_irq(q
->queue_lock
);
2205 * After sleeping, we become a "batching" process and
2206 * will be able to allocate at least one request, and
2207 * up to a big batch of them for a small period time.
2208 * See ioc_batching, ioc_set_batching
2210 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2211 ioc_set_batching(q
, ioc
);
2213 spin_lock_irq(q
->queue_lock
);
2215 finish_wait(&rl
->wait
[rw
], &wait
);
2221 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
2225 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2227 spin_lock_irq(q
->queue_lock
);
2228 if (gfp_mask
& __GFP_WAIT
) {
2229 rq
= get_request_wait(q
, rw
, NULL
);
2231 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2233 spin_unlock_irq(q
->queue_lock
);
2235 /* q->queue_lock is unlocked at this point */
2239 EXPORT_SYMBOL(blk_get_request
);
2242 * blk_start_queueing - initiate dispatch of requests to device
2243 * @q: request queue to kick into gear
2245 * This is basically a helper to remove the need to know whether a queue
2246 * is plugged or not if someone just wants to initiate dispatch of requests
2249 * The queue lock must be held with interrupts disabled.
2251 void blk_start_queueing(struct request_queue
*q
)
2253 if (!blk_queue_plugged(q
))
2256 __generic_unplug_device(q
);
2258 EXPORT_SYMBOL(blk_start_queueing
);
2261 * blk_requeue_request - put a request back on queue
2262 * @q: request queue where request should be inserted
2263 * @rq: request to be inserted
2266 * Drivers often keep queueing requests until the hardware cannot accept
2267 * more, when that condition happens we need to put the request back
2268 * on the queue. Must be called with queue lock held.
2270 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
2272 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2274 if (blk_rq_tagged(rq
))
2275 blk_queue_end_tag(q
, rq
);
2277 elv_requeue_request(q
, rq
);
2280 EXPORT_SYMBOL(blk_requeue_request
);
2283 * blk_insert_request - insert a special request in to a request queue
2284 * @q: request queue where request should be inserted
2285 * @rq: request to be inserted
2286 * @at_head: insert request at head or tail of queue
2287 * @data: private data
2290 * Many block devices need to execute commands asynchronously, so they don't
2291 * block the whole kernel from preemption during request execution. This is
2292 * accomplished normally by inserting aritficial requests tagged as
2293 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2294 * scheduled for actual execution by the request queue.
2296 * We have the option of inserting the head or the tail of the queue.
2297 * Typically we use the tail for new ioctls and so forth. We use the head
2298 * of the queue for things like a QUEUE_FULL message from a device, or a
2299 * host that is unable to accept a particular command.
2301 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
2302 int at_head
, void *data
)
2304 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2305 unsigned long flags
;
2308 * tell I/O scheduler that this isn't a regular read/write (ie it
2309 * must not attempt merges on this) and that it acts as a soft
2312 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2313 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2317 spin_lock_irqsave(q
->queue_lock
, flags
);
2320 * If command is tagged, release the tag
2322 if (blk_rq_tagged(rq
))
2323 blk_queue_end_tag(q
, rq
);
2325 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2326 __elv_add_request(q
, rq
, where
, 0);
2327 blk_start_queueing(q
);
2328 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2331 EXPORT_SYMBOL(blk_insert_request
);
2333 static int __blk_rq_unmap_user(struct bio
*bio
)
2338 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2339 bio_unmap_user(bio
);
2341 ret
= bio_uncopy_user(bio
);
2347 static int __blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2348 void __user
*ubuf
, unsigned int len
)
2350 unsigned long uaddr
;
2351 struct bio
*bio
, *orig_bio
;
2354 reading
= rq_data_dir(rq
) == READ
;
2357 * if alignment requirement is satisfied, map in user pages for
2358 * direct dma. else, set up kernel bounce buffers
2360 uaddr
= (unsigned long) ubuf
;
2361 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2362 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2364 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2367 return PTR_ERR(bio
);
2370 blk_queue_bounce(q
, &bio
);
2373 * We link the bounce buffer in and could have to traverse it
2374 * later so we have to get a ref to prevent it from being freed
2379 blk_rq_bio_prep(q
, rq
, bio
);
2380 else if (!ll_back_merge_fn(q
, rq
, bio
)) {
2384 rq
->biotail
->bi_next
= bio
;
2387 rq
->data_len
+= bio
->bi_size
;
2390 return bio
->bi_size
;
2393 /* if it was boucned we must call the end io function */
2394 bio_endio(bio
, bio
->bi_size
, 0);
2395 __blk_rq_unmap_user(orig_bio
);
2401 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2402 * @q: request queue where request should be inserted
2403 * @rq: request structure to fill
2404 * @ubuf: the user buffer
2405 * @len: length of user data
2408 * Data will be mapped directly for zero copy io, if possible. Otherwise
2409 * a kernel bounce buffer is used.
2411 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2412 * still in process context.
2414 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2415 * before being submitted to the device, as pages mapped may be out of
2416 * reach. It's the callers responsibility to make sure this happens. The
2417 * original bio must be passed back in to blk_rq_unmap_user() for proper
2420 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2421 void __user
*ubuf
, unsigned long len
)
2423 unsigned long bytes_read
= 0;
2424 struct bio
*bio
= NULL
;
2427 if (len
> (q
->max_hw_sectors
<< 9))
2432 while (bytes_read
!= len
) {
2433 unsigned long map_len
, end
, start
;
2435 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2436 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2438 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2441 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2442 * pages. If this happens we just lower the requested
2443 * mapping len by a page so that we can fit
2445 if (end
- start
> BIO_MAX_PAGES
)
2446 map_len
-= PAGE_SIZE
;
2448 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2457 rq
->buffer
= rq
->data
= NULL
;
2460 blk_rq_unmap_user(bio
);
2464 EXPORT_SYMBOL(blk_rq_map_user
);
2467 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2468 * @q: request queue where request should be inserted
2469 * @rq: request to map data to
2470 * @iov: pointer to the iovec
2471 * @iov_count: number of elements in the iovec
2472 * @len: I/O byte count
2475 * Data will be mapped directly for zero copy io, if possible. Otherwise
2476 * a kernel bounce buffer is used.
2478 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2479 * still in process context.
2481 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2482 * before being submitted to the device, as pages mapped may be out of
2483 * reach. It's the callers responsibility to make sure this happens. The
2484 * original bio must be passed back in to blk_rq_unmap_user() for proper
2487 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
2488 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2492 if (!iov
|| iov_count
<= 0)
2495 /* we don't allow misaligned data like bio_map_user() does. If the
2496 * user is using sg, they're expected to know the alignment constraints
2497 * and respect them accordingly */
2498 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2500 return PTR_ERR(bio
);
2502 if (bio
->bi_size
!= len
) {
2503 bio_endio(bio
, bio
->bi_size
, 0);
2504 bio_unmap_user(bio
);
2509 blk_rq_bio_prep(q
, rq
, bio
);
2510 rq
->buffer
= rq
->data
= NULL
;
2514 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2517 * blk_rq_unmap_user - unmap a request with user data
2518 * @bio: start of bio list
2521 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2522 * supply the original rq->bio from the blk_rq_map_user() return, since
2523 * the io completion may have changed rq->bio.
2525 int blk_rq_unmap_user(struct bio
*bio
)
2527 struct bio
*mapped_bio
;
2532 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2533 mapped_bio
= bio
->bi_private
;
2535 ret2
= __blk_rq_unmap_user(mapped_bio
);
2541 bio_put(mapped_bio
);
2547 EXPORT_SYMBOL(blk_rq_unmap_user
);
2550 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2551 * @q: request queue where request should be inserted
2552 * @rq: request to fill
2553 * @kbuf: the kernel buffer
2554 * @len: length of user data
2555 * @gfp_mask: memory allocation flags
2557 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
2558 unsigned int len
, gfp_t gfp_mask
)
2562 if (len
> (q
->max_hw_sectors
<< 9))
2567 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2569 return PTR_ERR(bio
);
2571 if (rq_data_dir(rq
) == WRITE
)
2572 bio
->bi_rw
|= (1 << BIO_RW
);
2574 blk_rq_bio_prep(q
, rq
, bio
);
2575 blk_queue_bounce(q
, &rq
->bio
);
2576 rq
->buffer
= rq
->data
= NULL
;
2580 EXPORT_SYMBOL(blk_rq_map_kern
);
2583 * blk_execute_rq_nowait - insert a request into queue for execution
2584 * @q: queue to insert the request in
2585 * @bd_disk: matching gendisk
2586 * @rq: request to insert
2587 * @at_head: insert request at head or tail of queue
2588 * @done: I/O completion handler
2591 * Insert a fully prepared request at the back of the io scheduler queue
2592 * for execution. Don't wait for completion.
2594 void blk_execute_rq_nowait(struct request_queue
*q
, struct gendisk
*bd_disk
,
2595 struct request
*rq
, int at_head
,
2598 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2600 rq
->rq_disk
= bd_disk
;
2601 rq
->cmd_flags
|= REQ_NOMERGE
;
2603 WARN_ON(irqs_disabled());
2604 spin_lock_irq(q
->queue_lock
);
2605 __elv_add_request(q
, rq
, where
, 1);
2606 __generic_unplug_device(q
);
2607 spin_unlock_irq(q
->queue_lock
);
2609 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2612 * blk_execute_rq - insert a request into queue for execution
2613 * @q: queue to insert the request in
2614 * @bd_disk: matching gendisk
2615 * @rq: request to insert
2616 * @at_head: insert request at head or tail of queue
2619 * Insert a fully prepared request at the back of the io scheduler queue
2620 * for execution and wait for completion.
2622 int blk_execute_rq(struct request_queue
*q
, struct gendisk
*bd_disk
,
2623 struct request
*rq
, int at_head
)
2625 DECLARE_COMPLETION_ONSTACK(wait
);
2626 char sense
[SCSI_SENSE_BUFFERSIZE
];
2630 * we need an extra reference to the request, so we can look at
2631 * it after io completion
2636 memset(sense
, 0, sizeof(sense
));
2641 rq
->end_io_data
= &wait
;
2642 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2643 wait_for_completion(&wait
);
2651 EXPORT_SYMBOL(blk_execute_rq
);
2654 * blkdev_issue_flush - queue a flush
2655 * @bdev: blockdev to issue flush for
2656 * @error_sector: error sector
2659 * Issue a flush for the block device in question. Caller can supply
2660 * room for storing the error offset in case of a flush error, if they
2661 * wish to. Caller must run wait_for_completion() on its own.
2663 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2665 struct request_queue
*q
;
2667 if (bdev
->bd_disk
== NULL
)
2670 q
= bdev_get_queue(bdev
);
2673 if (!q
->issue_flush_fn
)
2676 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2679 EXPORT_SYMBOL(blkdev_issue_flush
);
2681 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2683 int rw
= rq_data_dir(rq
);
2685 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2689 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2691 disk_round_stats(rq
->rq_disk
);
2692 rq
->rq_disk
->in_flight
++;
2697 * add-request adds a request to the linked list.
2698 * queue lock is held and interrupts disabled, as we muck with the
2699 * request queue list.
2701 static inline void add_request(struct request_queue
* q
, struct request
* req
)
2703 drive_stat_acct(req
, req
->nr_sectors
, 1);
2706 * elevator indicated where it wants this request to be
2707 * inserted at elevator_merge time
2709 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2713 * disk_round_stats() - Round off the performance stats on a struct
2716 * The average IO queue length and utilisation statistics are maintained
2717 * by observing the current state of the queue length and the amount of
2718 * time it has been in this state for.
2720 * Normally, that accounting is done on IO completion, but that can result
2721 * in more than a second's worth of IO being accounted for within any one
2722 * second, leading to >100% utilisation. To deal with that, we call this
2723 * function to do a round-off before returning the results when reading
2724 * /proc/diskstats. This accounts immediately for all queue usage up to
2725 * the current jiffies and restarts the counters again.
2727 void disk_round_stats(struct gendisk
*disk
)
2729 unsigned long now
= jiffies
;
2731 if (now
== disk
->stamp
)
2734 if (disk
->in_flight
) {
2735 __disk_stat_add(disk
, time_in_queue
,
2736 disk
->in_flight
* (now
- disk
->stamp
));
2737 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2742 EXPORT_SYMBOL_GPL(disk_round_stats
);
2745 * queue lock must be held
2747 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
2751 if (unlikely(--req
->ref_count
))
2754 elv_completed_request(q
, req
);
2757 * Request may not have originated from ll_rw_blk. if not,
2758 * it didn't come out of our reserved rq pools
2760 if (req
->cmd_flags
& REQ_ALLOCED
) {
2761 int rw
= rq_data_dir(req
);
2762 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2764 BUG_ON(!list_empty(&req
->queuelist
));
2765 BUG_ON(!hlist_unhashed(&req
->hash
));
2767 blk_free_request(q
, req
);
2768 freed_request(q
, rw
, priv
);
2772 EXPORT_SYMBOL_GPL(__blk_put_request
);
2774 void blk_put_request(struct request
*req
)
2776 unsigned long flags
;
2777 struct request_queue
*q
= req
->q
;
2780 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2781 * following if (q) test.
2784 spin_lock_irqsave(q
->queue_lock
, flags
);
2785 __blk_put_request(q
, req
);
2786 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2790 EXPORT_SYMBOL(blk_put_request
);
2793 * blk_end_sync_rq - executes a completion event on a request
2794 * @rq: request to complete
2795 * @error: end io status of the request
2797 void blk_end_sync_rq(struct request
*rq
, int error
)
2799 struct completion
*waiting
= rq
->end_io_data
;
2801 rq
->end_io_data
= NULL
;
2802 __blk_put_request(rq
->q
, rq
);
2805 * complete last, if this is a stack request the process (and thus
2806 * the rq pointer) could be invalid right after this complete()
2810 EXPORT_SYMBOL(blk_end_sync_rq
);
2813 * Has to be called with the request spinlock acquired
2815 static int attempt_merge(struct request_queue
*q
, struct request
*req
,
2816 struct request
*next
)
2818 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2824 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2827 if (rq_data_dir(req
) != rq_data_dir(next
)
2828 || req
->rq_disk
!= next
->rq_disk
2833 * If we are allowed to merge, then append bio list
2834 * from next to rq and release next. merge_requests_fn
2835 * will have updated segment counts, update sector
2838 if (!ll_merge_requests_fn(q
, req
, next
))
2842 * At this point we have either done a back merge
2843 * or front merge. We need the smaller start_time of
2844 * the merged requests to be the current request
2845 * for accounting purposes.
2847 if (time_after(req
->start_time
, next
->start_time
))
2848 req
->start_time
= next
->start_time
;
2850 req
->biotail
->bi_next
= next
->bio
;
2851 req
->biotail
= next
->biotail
;
2853 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2855 elv_merge_requests(q
, req
, next
);
2858 disk_round_stats(req
->rq_disk
);
2859 req
->rq_disk
->in_flight
--;
2862 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2864 __blk_put_request(q
, next
);
2868 static inline int attempt_back_merge(struct request_queue
*q
,
2871 struct request
*next
= elv_latter_request(q
, rq
);
2874 return attempt_merge(q
, rq
, next
);
2879 static inline int attempt_front_merge(struct request_queue
*q
,
2882 struct request
*prev
= elv_former_request(q
, rq
);
2885 return attempt_merge(q
, prev
, rq
);
2890 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2892 req
->cmd_type
= REQ_TYPE_FS
;
2895 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2897 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2898 req
->cmd_flags
|= REQ_FAILFAST
;
2901 * REQ_BARRIER implies no merging, but lets make it explicit
2903 if (unlikely(bio_barrier(bio
)))
2904 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2907 req
->cmd_flags
|= REQ_RW_SYNC
;
2908 if (bio_rw_meta(bio
))
2909 req
->cmd_flags
|= REQ_RW_META
;
2912 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2913 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2914 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2915 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2916 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2917 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
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(struct request_queue
*q
, struct bio
*bio
)
2926 struct request
*req
;
2927 int el_ret
, nr_sectors
, barrier
, err
;
2928 const unsigned short prio
= bio_prio(bio
);
2929 const int sync
= bio_sync(bio
);
2932 nr_sectors
= bio_sectors(bio
);
2935 * low level driver can indicate that it wants pages above a
2936 * certain limit bounced to low memory (ie for highmem, or even
2937 * ISA dma in theory)
2939 blk_queue_bounce(q
, &bio
);
2941 barrier
= bio_barrier(bio
);
2942 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2947 spin_lock_irq(q
->queue_lock
);
2949 if (unlikely(barrier
) || elv_queue_empty(q
))
2952 el_ret
= elv_merge(q
, &req
, bio
);
2954 case ELEVATOR_BACK_MERGE
:
2955 BUG_ON(!rq_mergeable(req
));
2957 if (!ll_back_merge_fn(q
, req
, bio
))
2960 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2962 req
->biotail
->bi_next
= bio
;
2964 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2965 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2966 drive_stat_acct(req
, nr_sectors
, 0);
2967 if (!attempt_back_merge(q
, req
))
2968 elv_merged_request(q
, req
, el_ret
);
2971 case ELEVATOR_FRONT_MERGE
:
2972 BUG_ON(!rq_mergeable(req
));
2974 if (!ll_front_merge_fn(q
, req
, bio
))
2977 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
2979 bio
->bi_next
= req
->bio
;
2983 * may not be valid. if the low level driver said
2984 * it didn't need a bounce buffer then it better
2985 * not touch req->buffer either...
2987 req
->buffer
= bio_data(bio
);
2988 req
->current_nr_sectors
= bio_cur_sectors(bio
);
2989 req
->hard_cur_sectors
= req
->current_nr_sectors
;
2990 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
2991 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2992 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2993 drive_stat_acct(req
, nr_sectors
, 0);
2994 if (!attempt_front_merge(q
, req
))
2995 elv_merged_request(q
, req
, el_ret
);
2998 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3005 * This sync check and mask will be re-done in init_request_from_bio(),
3006 * but we need to set it earlier to expose the sync flag to the
3007 * rq allocator and io schedulers.
3009 rw_flags
= bio_data_dir(bio
);
3011 rw_flags
|= REQ_RW_SYNC
;
3014 * Grab a free request. This is might sleep but can not fail.
3015 * Returns with the queue unlocked.
3017 req
= get_request_wait(q
, rw_flags
, bio
);
3020 * After dropping the lock and possibly sleeping here, our request
3021 * may now be mergeable after it had proven unmergeable (above).
3022 * We don't worry about that case for efficiency. It won't happen
3023 * often, and the elevators are able to handle it.
3025 init_request_from_bio(req
, bio
);
3027 spin_lock_irq(q
->queue_lock
);
3028 if (elv_queue_empty(q
))
3030 add_request(q
, req
);
3033 __generic_unplug_device(q
);
3035 spin_unlock_irq(q
->queue_lock
);
3039 bio_endio(bio
, nr_sectors
<< 9, err
);
3044 * If bio->bi_dev is a partition, remap the location
3046 static inline void blk_partition_remap(struct bio
*bio
)
3048 struct block_device
*bdev
= bio
->bi_bdev
;
3050 if (bdev
!= bdev
->bd_contains
) {
3051 struct hd_struct
*p
= bdev
->bd_part
;
3052 const int rw
= bio_data_dir(bio
);
3054 p
->sectors
[rw
] += bio_sectors(bio
);
3057 bio
->bi_sector
+= p
->start_sect
;
3058 bio
->bi_bdev
= bdev
->bd_contains
;
3060 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
3061 bdev
->bd_dev
, bio
->bi_sector
,
3062 bio
->bi_sector
- p
->start_sect
);
3066 static void handle_bad_sector(struct bio
*bio
)
3068 char b
[BDEVNAME_SIZE
];
3070 printk(KERN_INFO
"attempt to access beyond end of device\n");
3071 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3072 bdevname(bio
->bi_bdev
, b
),
3074 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3075 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3077 set_bit(BIO_EOF
, &bio
->bi_flags
);
3080 #ifdef CONFIG_FAIL_MAKE_REQUEST
3082 static DECLARE_FAULT_ATTR(fail_make_request
);
3084 static int __init
setup_fail_make_request(char *str
)
3086 return setup_fault_attr(&fail_make_request
, str
);
3088 __setup("fail_make_request=", setup_fail_make_request
);
3090 static int should_fail_request(struct bio
*bio
)
3092 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3093 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3094 return should_fail(&fail_make_request
, bio
->bi_size
);
3099 static int __init
fail_make_request_debugfs(void)
3101 return init_fault_attr_dentries(&fail_make_request
,
3102 "fail_make_request");
3105 late_initcall(fail_make_request_debugfs
);
3107 #else /* CONFIG_FAIL_MAKE_REQUEST */
3109 static inline int should_fail_request(struct bio
*bio
)
3114 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3117 * generic_make_request: hand a buffer to its device driver for I/O
3118 * @bio: The bio describing the location in memory and on the device.
3120 * generic_make_request() is used to make I/O requests of block
3121 * devices. It is passed a &struct bio, which describes the I/O that needs
3124 * generic_make_request() does not return any status. The
3125 * success/failure status of the request, along with notification of
3126 * completion, is delivered asynchronously through the bio->bi_end_io
3127 * function described (one day) else where.
3129 * The caller of generic_make_request must make sure that bi_io_vec
3130 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3131 * set to describe the device address, and the
3132 * bi_end_io and optionally bi_private are set to describe how
3133 * completion notification should be signaled.
3135 * generic_make_request and the drivers it calls may use bi_next if this
3136 * bio happens to be merged with someone else, and may change bi_dev and
3137 * bi_sector for remaps as it sees fit. So the values of these fields
3138 * should NOT be depended on after the call to generic_make_request.
3140 static inline void __generic_make_request(struct bio
*bio
)
3142 struct request_queue
*q
;
3144 sector_t old_sector
;
3145 int ret
, nr_sectors
= bio_sectors(bio
);
3149 /* Test device or partition size, when known. */
3150 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3152 sector_t sector
= bio
->bi_sector
;
3154 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3156 * This may well happen - the kernel calls bread()
3157 * without checking the size of the device, e.g., when
3158 * mounting a device.
3160 handle_bad_sector(bio
);
3166 * Resolve the mapping until finished. (drivers are
3167 * still free to implement/resolve their own stacking
3168 * by explicitly returning 0)
3170 * NOTE: we don't repeat the blk_size check for each new device.
3171 * Stacking drivers are expected to know what they are doing.
3176 char b
[BDEVNAME_SIZE
];
3178 q
= bdev_get_queue(bio
->bi_bdev
);
3181 "generic_make_request: Trying to access "
3182 "nonexistent block-device %s (%Lu)\n",
3183 bdevname(bio
->bi_bdev
, b
),
3184 (long long) bio
->bi_sector
);
3186 bio_endio(bio
, bio
->bi_size
, -EIO
);
3190 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
3191 printk("bio too big device %s (%u > %u)\n",
3192 bdevname(bio
->bi_bdev
, b
),
3198 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3201 if (should_fail_request(bio
))
3205 * If this device has partitions, remap block n
3206 * of partition p to block n+start(p) of the disk.
3208 blk_partition_remap(bio
);
3210 if (old_sector
!= -1)
3211 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3214 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3216 old_sector
= bio
->bi_sector
;
3217 old_dev
= bio
->bi_bdev
->bd_dev
;
3219 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3221 sector_t sector
= bio
->bi_sector
;
3223 if (maxsector
< nr_sectors
||
3224 maxsector
- nr_sectors
< sector
) {
3226 * This may well happen - partitions are not
3227 * checked to make sure they are within the size
3228 * of the whole device.
3230 handle_bad_sector(bio
);
3235 ret
= q
->make_request_fn(q
, bio
);
3240 * We only want one ->make_request_fn to be active at a time,
3241 * else stack usage with stacked devices could be a problem.
3242 * So use current->bio_{list,tail} to keep a list of requests
3243 * submited by a make_request_fn function.
3244 * current->bio_tail is also used as a flag to say if
3245 * generic_make_request is currently active in this task or not.
3246 * If it is NULL, then no make_request is active. If it is non-NULL,
3247 * then a make_request is active, and new requests should be added
3250 void generic_make_request(struct bio
*bio
)
3252 if (current
->bio_tail
) {
3253 /* make_request is active */
3254 *(current
->bio_tail
) = bio
;
3255 bio
->bi_next
= NULL
;
3256 current
->bio_tail
= &bio
->bi_next
;
3259 /* following loop may be a bit non-obvious, and so deserves some
3261 * Before entering the loop, bio->bi_next is NULL (as all callers
3262 * ensure that) so we have a list with a single bio.
3263 * We pretend that we have just taken it off a longer list, so
3264 * we assign bio_list to the next (which is NULL) and bio_tail
3265 * to &bio_list, thus initialising the bio_list of new bios to be
3266 * added. __generic_make_request may indeed add some more bios
3267 * through a recursive call to generic_make_request. If it
3268 * did, we find a non-NULL value in bio_list and re-enter the loop
3269 * from the top. In this case we really did just take the bio
3270 * of the top of the list (no pretending) and so fixup bio_list and
3271 * bio_tail or bi_next, and call into __generic_make_request again.
3273 * The loop was structured like this to make only one call to
3274 * __generic_make_request (which is important as it is large and
3275 * inlined) and to keep the structure simple.
3277 BUG_ON(bio
->bi_next
);
3279 current
->bio_list
= bio
->bi_next
;
3280 if (bio
->bi_next
== NULL
)
3281 current
->bio_tail
= ¤t
->bio_list
;
3283 bio
->bi_next
= NULL
;
3284 __generic_make_request(bio
);
3285 bio
= current
->bio_list
;
3287 current
->bio_tail
= NULL
; /* deactivate */
3290 EXPORT_SYMBOL(generic_make_request
);
3293 * submit_bio: submit a bio to the block device layer for I/O
3294 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3295 * @bio: The &struct bio which describes the I/O
3297 * submit_bio() is very similar in purpose to generic_make_request(), and
3298 * uses that function to do most of the work. Both are fairly rough
3299 * interfaces, @bio must be presetup and ready for I/O.
3302 void submit_bio(int rw
, struct bio
*bio
)
3304 int count
= bio_sectors(bio
);
3306 BIO_BUG_ON(!bio
->bi_size
);
3307 BIO_BUG_ON(!bio
->bi_io_vec
);
3310 count_vm_events(PGPGOUT
, count
);
3312 task_io_account_read(bio
->bi_size
);
3313 count_vm_events(PGPGIN
, count
);
3316 if (unlikely(block_dump
)) {
3317 char b
[BDEVNAME_SIZE
];
3318 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3319 current
->comm
, current
->pid
,
3320 (rw
& WRITE
) ? "WRITE" : "READ",
3321 (unsigned long long)bio
->bi_sector
,
3322 bdevname(bio
->bi_bdev
,b
));
3325 generic_make_request(bio
);
3328 EXPORT_SYMBOL(submit_bio
);
3330 static void blk_recalc_rq_segments(struct request
*rq
)
3332 struct bio
*bio
, *prevbio
= NULL
;
3333 int nr_phys_segs
, nr_hw_segs
;
3334 unsigned int phys_size
, hw_size
;
3335 struct request_queue
*q
= rq
->q
;
3340 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3341 rq_for_each_bio(bio
, rq
) {
3342 /* Force bio hw/phys segs to be recalculated. */
3343 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3345 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3346 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3348 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3349 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3351 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3352 pseg
<= q
->max_segment_size
) {
3354 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3358 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3359 hseg
<= q
->max_segment_size
) {
3361 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3368 rq
->nr_phys_segments
= nr_phys_segs
;
3369 rq
->nr_hw_segments
= nr_hw_segs
;
3372 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3374 if (blk_fs_request(rq
)) {
3375 rq
->hard_sector
+= nsect
;
3376 rq
->hard_nr_sectors
-= nsect
;
3379 * Move the I/O submission pointers ahead if required.
3381 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3382 (rq
->sector
<= rq
->hard_sector
)) {
3383 rq
->sector
= rq
->hard_sector
;
3384 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3385 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3386 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3387 rq
->buffer
= bio_data(rq
->bio
);
3391 * if total number of sectors is less than the first segment
3392 * size, something has gone terribly wrong
3394 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3395 printk("blk: request botched\n");
3396 rq
->nr_sectors
= rq
->current_nr_sectors
;
3401 static int __end_that_request_first(struct request
*req
, int uptodate
,
3404 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3407 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3410 * extend uptodate bool to allow < 0 value to be direct io error
3413 if (end_io_error(uptodate
))
3414 error
= !uptodate
? -EIO
: uptodate
;
3417 * for a REQ_BLOCK_PC request, we want to carry any eventual
3418 * sense key with us all the way through
3420 if (!blk_pc_request(req
))
3424 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3425 printk("end_request: I/O error, dev %s, sector %llu\n",
3426 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3427 (unsigned long long)req
->sector
);
3430 if (blk_fs_request(req
) && req
->rq_disk
) {
3431 const int rw
= rq_data_dir(req
);
3433 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3436 total_bytes
= bio_nbytes
= 0;
3437 while ((bio
= req
->bio
) != NULL
) {
3440 if (nr_bytes
>= bio
->bi_size
) {
3441 req
->bio
= bio
->bi_next
;
3442 nbytes
= bio
->bi_size
;
3443 if (!ordered_bio_endio(req
, bio
, nbytes
, error
))
3444 bio_endio(bio
, nbytes
, error
);
3448 int idx
= bio
->bi_idx
+ next_idx
;
3450 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3451 blk_dump_rq_flags(req
, "__end_that");
3452 printk("%s: bio idx %d >= vcnt %d\n",
3454 bio
->bi_idx
, bio
->bi_vcnt
);
3458 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3459 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3462 * not a complete bvec done
3464 if (unlikely(nbytes
> nr_bytes
)) {
3465 bio_nbytes
+= nr_bytes
;
3466 total_bytes
+= nr_bytes
;
3471 * advance to the next vector
3474 bio_nbytes
+= nbytes
;
3477 total_bytes
+= nbytes
;
3480 if ((bio
= req
->bio
)) {
3482 * end more in this run, or just return 'not-done'
3484 if (unlikely(nr_bytes
<= 0))
3496 * if the request wasn't completed, update state
3499 if (!ordered_bio_endio(req
, bio
, bio_nbytes
, error
))
3500 bio_endio(bio
, bio_nbytes
, error
);
3501 bio
->bi_idx
+= next_idx
;
3502 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3503 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3506 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3507 blk_recalc_rq_segments(req
);
3512 * end_that_request_first - end I/O on a request
3513 * @req: the request being processed
3514 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3515 * @nr_sectors: number of sectors to end I/O on
3518 * Ends I/O on a number of sectors attached to @req, and sets it up
3519 * for the next range of segments (if any) in the cluster.
3522 * 0 - we are done with this request, call end_that_request_last()
3523 * 1 - still buffers pending for this request
3525 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3527 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3530 EXPORT_SYMBOL(end_that_request_first
);
3533 * end_that_request_chunk - end I/O on a request
3534 * @req: the request being processed
3535 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3536 * @nr_bytes: number of bytes to complete
3539 * Ends I/O on a number of bytes attached to @req, and sets it up
3540 * for the next range of segments (if any). Like end_that_request_first(),
3541 * but deals with bytes instead of sectors.
3544 * 0 - we are done with this request, call end_that_request_last()
3545 * 1 - still buffers pending for this request
3547 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3549 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3552 EXPORT_SYMBOL(end_that_request_chunk
);
3555 * splice the completion data to a local structure and hand off to
3556 * process_completion_queue() to complete the requests
3558 static void blk_done_softirq(struct softirq_action
*h
)
3560 struct list_head
*cpu_list
, local_list
;
3562 local_irq_disable();
3563 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3564 list_replace_init(cpu_list
, &local_list
);
3567 while (!list_empty(&local_list
)) {
3568 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3570 list_del_init(&rq
->donelist
);
3571 rq
->q
->softirq_done_fn(rq
);
3575 static int blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3579 * If a CPU goes away, splice its entries to the current CPU
3580 * and trigger a run of the softirq
3582 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3583 int cpu
= (unsigned long) hcpu
;
3585 local_irq_disable();
3586 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3587 &__get_cpu_var(blk_cpu_done
));
3588 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3596 static struct notifier_block __devinitdata blk_cpu_notifier
= {
3597 .notifier_call
= blk_cpu_notify
,
3601 * blk_complete_request - end I/O on a request
3602 * @req: the request being processed
3605 * Ends all I/O on a request. It does not handle partial completions,
3606 * unless the driver actually implements this in its completion callback
3607 * through requeueing. Theh actual completion happens out-of-order,
3608 * through a softirq handler. The user must have registered a completion
3609 * callback through blk_queue_softirq_done().
3612 void blk_complete_request(struct request
*req
)
3614 struct list_head
*cpu_list
;
3615 unsigned long flags
;
3617 BUG_ON(!req
->q
->softirq_done_fn
);
3619 local_irq_save(flags
);
3621 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3622 list_add_tail(&req
->donelist
, cpu_list
);
3623 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3625 local_irq_restore(flags
);
3628 EXPORT_SYMBOL(blk_complete_request
);
3631 * queue lock must be held
3633 void end_that_request_last(struct request
*req
, int uptodate
)
3635 struct gendisk
*disk
= req
->rq_disk
;
3639 * extend uptodate bool to allow < 0 value to be direct io error
3642 if (end_io_error(uptodate
))
3643 error
= !uptodate
? -EIO
: uptodate
;
3645 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3646 laptop_io_completion();
3649 * Account IO completion. bar_rq isn't accounted as a normal
3650 * IO on queueing nor completion. Accounting the containing
3651 * request is enough.
3653 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3654 unsigned long duration
= jiffies
- req
->start_time
;
3655 const int rw
= rq_data_dir(req
);
3657 __disk_stat_inc(disk
, ios
[rw
]);
3658 __disk_stat_add(disk
, ticks
[rw
], duration
);
3659 disk_round_stats(disk
);
3663 req
->end_io(req
, error
);
3665 __blk_put_request(req
->q
, req
);
3668 EXPORT_SYMBOL(end_that_request_last
);
3670 void end_request(struct request
*req
, int uptodate
)
3672 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3673 add_disk_randomness(req
->rq_disk
);
3674 blkdev_dequeue_request(req
);
3675 end_that_request_last(req
, uptodate
);
3679 EXPORT_SYMBOL(end_request
);
3681 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3684 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3685 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3687 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3688 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3689 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3690 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3691 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3692 rq
->buffer
= bio_data(bio
);
3693 rq
->data_len
= bio
->bi_size
;
3695 rq
->bio
= rq
->biotail
= bio
;
3698 EXPORT_SYMBOL(blk_rq_bio_prep
);
3700 int kblockd_schedule_work(struct work_struct
*work
)
3702 return queue_work(kblockd_workqueue
, work
);
3705 EXPORT_SYMBOL(kblockd_schedule_work
);
3707 void kblockd_flush_work(struct work_struct
*work
)
3709 cancel_work_sync(work
);
3711 EXPORT_SYMBOL(kblockd_flush_work
);
3713 int __init
blk_dev_init(void)
3717 kblockd_workqueue
= create_workqueue("kblockd");
3718 if (!kblockd_workqueue
)
3719 panic("Failed to create kblockd\n");
3721 request_cachep
= kmem_cache_create("blkdev_requests",
3722 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3724 requestq_cachep
= kmem_cache_create("blkdev_queue",
3725 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3727 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3728 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
3730 for_each_possible_cpu(i
)
3731 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3733 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3734 register_hotcpu_notifier(&blk_cpu_notifier
);
3736 blk_max_low_pfn
= max_low_pfn
- 1;
3737 blk_max_pfn
= max_pfn
- 1;
3743 * IO Context helper functions
3745 void put_io_context(struct io_context
*ioc
)
3750 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3752 if (atomic_dec_and_test(&ioc
->refcount
)) {
3753 struct cfq_io_context
*cic
;
3756 if (ioc
->aic
&& ioc
->aic
->dtor
)
3757 ioc
->aic
->dtor(ioc
->aic
);
3758 if (ioc
->cic_root
.rb_node
!= NULL
) {
3759 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3761 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3766 kmem_cache_free(iocontext_cachep
, ioc
);
3769 EXPORT_SYMBOL(put_io_context
);
3771 /* Called by the exitting task */
3772 void exit_io_context(void)
3774 struct io_context
*ioc
;
3775 struct cfq_io_context
*cic
;
3778 ioc
= current
->io_context
;
3779 current
->io_context
= NULL
;
3780 task_unlock(current
);
3783 if (ioc
->aic
&& ioc
->aic
->exit
)
3784 ioc
->aic
->exit(ioc
->aic
);
3785 if (ioc
->cic_root
.rb_node
!= NULL
) {
3786 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3790 put_io_context(ioc
);
3794 * If the current task has no IO context then create one and initialise it.
3795 * Otherwise, return its existing IO context.
3797 * This returned IO context doesn't have a specifically elevated refcount,
3798 * but since the current task itself holds a reference, the context can be
3799 * used in general code, so long as it stays within `current` context.
3801 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3803 struct task_struct
*tsk
= current
;
3804 struct io_context
*ret
;
3806 ret
= tsk
->io_context
;
3810 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3812 atomic_set(&ret
->refcount
, 1);
3813 ret
->task
= current
;
3814 ret
->ioprio_changed
= 0;
3815 ret
->last_waited
= jiffies
; /* doesn't matter... */
3816 ret
->nr_batch_requests
= 0; /* because this is 0 */
3818 ret
->cic_root
.rb_node
= NULL
;
3819 ret
->ioc_data
= NULL
;
3820 /* make sure set_task_ioprio() sees the settings above */
3822 tsk
->io_context
= ret
;
3829 * If the current task has no IO context then create one and initialise it.
3830 * If it does have a context, take a ref on it.
3832 * This is always called in the context of the task which submitted the I/O.
3834 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3836 struct io_context
*ret
;
3837 ret
= current_io_context(gfp_flags
, node
);
3839 atomic_inc(&ret
->refcount
);
3842 EXPORT_SYMBOL(get_io_context
);
3844 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3846 struct io_context
*src
= *psrc
;
3847 struct io_context
*dst
= *pdst
;
3850 BUG_ON(atomic_read(&src
->refcount
) == 0);
3851 atomic_inc(&src
->refcount
);
3852 put_io_context(dst
);
3856 EXPORT_SYMBOL(copy_io_context
);
3858 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3860 struct io_context
*temp
;
3865 EXPORT_SYMBOL(swap_io_context
);
3870 struct queue_sysfs_entry
{
3871 struct attribute attr
;
3872 ssize_t (*show
)(struct request_queue
*, char *);
3873 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3877 queue_var_show(unsigned int var
, char *page
)
3879 return sprintf(page
, "%d\n", var
);
3883 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3885 char *p
= (char *) page
;
3887 *var
= simple_strtoul(p
, &p
, 10);
3891 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3893 return queue_var_show(q
->nr_requests
, (page
));
3897 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3899 struct request_list
*rl
= &q
->rq
;
3901 int ret
= queue_var_store(&nr
, page
, count
);
3902 if (nr
< BLKDEV_MIN_RQ
)
3905 spin_lock_irq(q
->queue_lock
);
3906 q
->nr_requests
= nr
;
3907 blk_queue_congestion_threshold(q
);
3909 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3910 blk_set_queue_congested(q
, READ
);
3911 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3912 blk_clear_queue_congested(q
, READ
);
3914 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3915 blk_set_queue_congested(q
, WRITE
);
3916 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3917 blk_clear_queue_congested(q
, WRITE
);
3919 if (rl
->count
[READ
] >= q
->nr_requests
) {
3920 blk_set_queue_full(q
, READ
);
3921 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3922 blk_clear_queue_full(q
, READ
);
3923 wake_up(&rl
->wait
[READ
]);
3926 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3927 blk_set_queue_full(q
, WRITE
);
3928 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3929 blk_clear_queue_full(q
, WRITE
);
3930 wake_up(&rl
->wait
[WRITE
]);
3932 spin_unlock_irq(q
->queue_lock
);
3936 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3938 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3940 return queue_var_show(ra_kb
, (page
));
3944 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3946 unsigned long ra_kb
;
3947 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3949 spin_lock_irq(q
->queue_lock
);
3950 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3951 spin_unlock_irq(q
->queue_lock
);
3956 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3958 int max_sectors_kb
= q
->max_sectors
>> 1;
3960 return queue_var_show(max_sectors_kb
, (page
));
3964 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3966 unsigned long max_sectors_kb
,
3967 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3968 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3969 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3972 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3975 * Take the queue lock to update the readahead and max_sectors
3976 * values synchronously:
3978 spin_lock_irq(q
->queue_lock
);
3980 * Trim readahead window as well, if necessary:
3982 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3983 if (ra_kb
> max_sectors_kb
)
3984 q
->backing_dev_info
.ra_pages
=
3985 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3987 q
->max_sectors
= max_sectors_kb
<< 1;
3988 spin_unlock_irq(q
->queue_lock
);
3993 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3995 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3997 return queue_var_show(max_hw_sectors_kb
, (page
));
4001 static struct queue_sysfs_entry queue_requests_entry
= {
4002 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4003 .show
= queue_requests_show
,
4004 .store
= queue_requests_store
,
4007 static struct queue_sysfs_entry queue_ra_entry
= {
4008 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4009 .show
= queue_ra_show
,
4010 .store
= queue_ra_store
,
4013 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4014 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4015 .show
= queue_max_sectors_show
,
4016 .store
= queue_max_sectors_store
,
4019 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4020 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4021 .show
= queue_max_hw_sectors_show
,
4024 static struct queue_sysfs_entry queue_iosched_entry
= {
4025 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4026 .show
= elv_iosched_show
,
4027 .store
= elv_iosched_store
,
4030 static struct attribute
*default_attrs
[] = {
4031 &queue_requests_entry
.attr
,
4032 &queue_ra_entry
.attr
,
4033 &queue_max_hw_sectors_entry
.attr
,
4034 &queue_max_sectors_entry
.attr
,
4035 &queue_iosched_entry
.attr
,
4039 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4042 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4044 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4045 struct request_queue
*q
=
4046 container_of(kobj
, struct request_queue
, kobj
);
4051 mutex_lock(&q
->sysfs_lock
);
4052 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4053 mutex_unlock(&q
->sysfs_lock
);
4056 res
= entry
->show(q
, page
);
4057 mutex_unlock(&q
->sysfs_lock
);
4062 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4063 const char *page
, size_t length
)
4065 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4066 struct request_queue
*q
= container_of(kobj
, struct request_queue
, kobj
);
4072 mutex_lock(&q
->sysfs_lock
);
4073 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4074 mutex_unlock(&q
->sysfs_lock
);
4077 res
= entry
->store(q
, page
, length
);
4078 mutex_unlock(&q
->sysfs_lock
);
4082 static struct sysfs_ops queue_sysfs_ops
= {
4083 .show
= queue_attr_show
,
4084 .store
= queue_attr_store
,
4087 static struct kobj_type queue_ktype
= {
4088 .sysfs_ops
= &queue_sysfs_ops
,
4089 .default_attrs
= default_attrs
,
4090 .release
= blk_release_queue
,
4093 int blk_register_queue(struct gendisk
*disk
)
4097 struct request_queue
*q
= disk
->queue
;
4099 if (!q
|| !q
->request_fn
)
4102 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4104 ret
= kobject_add(&q
->kobj
);
4108 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4110 ret
= elv_register_queue(q
);
4112 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4113 kobject_del(&q
->kobj
);
4120 void blk_unregister_queue(struct gendisk
*disk
)
4122 struct request_queue
*q
= disk
->queue
;
4124 if (q
&& q
->request_fn
) {
4125 elv_unregister_queue(q
);
4127 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4128 kobject_del(&q
->kobj
);
4129 kobject_put(&disk
->kobj
);